This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Remove x2p
[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"
b24b43f7 35#include "charclass_invlists.h"
a0ed51b3 36
27da23d5
JH
37static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
a0ed51b3 42
166f8a29 43This file contains various utility functions for manipulating UTF8-encoded
72d33970 44strings. For the uninitiated, this is a method of representing arbitrary
61296642 45Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
46characters in the ASCII range are unmodified, and a zero byte never appears
47within non-zero characters.
166f8a29 48
eaf7a4d2
CS
49=cut
50*/
51
52/*
53=for apidoc is_ascii_string
54
a1433954 55Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
56or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
57is, if they are invariant. On ASCII-ish machines, only ASCII characters
58fit this definition, hence the function's name.
eaf7a4d2 59
75200dff
KW
60If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
61use this option, that C<s> can't have embedded C<NUL> characters and has to
62have a terminating C<NUL> byte).
9f7e3d64 63
a1433954 64See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
65
66=cut
67*/
68
69bool
668b6d8d 70Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
71{
72 const U8* const send = s + (len ? len : strlen((const char *)s));
73 const U8* x = s;
74
75 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
76
77 for (; x < send; ++x) {
78 if (!UTF8_IS_INVARIANT(*x))
79 break;
80 }
81
82 return x == send;
83}
84
85/*
378516de 86=for apidoc uvoffuni_to_utf8_flags
eebe1485 87
a27992cc 88THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
89Instead, B<Almost all code should use L</uvchr_to_utf8> or
90L</uvchr_to_utf8_flags>>.
a27992cc 91
de69f3af
KW
92This function is like them, but the input is a strict Unicode
93(as opposed to native) code point. Only in very rare circumstances should code
94not be using the native code point.
949cf498 95
de69f3af 96For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 97
eebe1485
SC
98=cut
99*/
100
dfe13c55 101U8 *
378516de 102Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 103{
378516de 104 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 105
d9432125
KW
106 if (UNI_IS_INVARIANT(uv)) {
107 *d++ = (U8) LATIN1_TO_NATIVE(uv);
108 return d;
109 }
110
979f77b6
KW
111 /* The first problematic code point is the first surrogate */
112 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 113 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 114 {
949cf498
KW
115 if (UNICODE_IS_SURROGATE(uv)) {
116 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 117 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
118 "UTF-16 surrogate U+%04"UVXf, uv);
119 }
120 if (flags & UNICODE_DISALLOW_SURROGATE) {
121 return NULL;
122 }
123 }
124 else if (UNICODE_IS_SUPER(uv)) {
125 if (flags & UNICODE_WARN_SUPER
126 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
127 {
8457b38f 128 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
129 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
130 }
131 if (flags & UNICODE_DISALLOW_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
133 {
134 return NULL;
135 }
136 }
137 else if (UNICODE_IS_NONCHAR(uv)) {
138 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 139 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
140 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
141 uv);
142 }
143 if (flags & UNICODE_DISALLOW_NONCHAR) {
144 return NULL;
145 }
146 }
507b9800 147 }
d9432125 148
2d331972 149#if defined(EBCDIC)
d9432125 150 {
5aaebcb3 151 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
152 U8 *p = d+len-1;
153 while (p > d) {
bc3632a8 154 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
155 uv >>= UTF_ACCUMULATION_SHIFT;
156 }
bc3632a8 157 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
158 return d+len;
159 }
160#else /* Non loop style */
a0ed51b3 161 if (uv < 0x800) {
eb160463
GS
162 *d++ = (U8)(( uv >> 6) | 0xc0);
163 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
164 return d;
165 }
166 if (uv < 0x10000) {
eb160463
GS
167 *d++ = (U8)(( uv >> 12) | 0xe0);
168 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
169 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
170 return d;
171 }
172 if (uv < 0x200000) {
eb160463
GS
173 *d++ = (U8)(( uv >> 18) | 0xf0);
174 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
175 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
176 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
177 return d;
178 }
179 if (uv < 0x4000000) {
eb160463
GS
180 *d++ = (U8)(( uv >> 24) | 0xf8);
181 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
185 return d;
186 }
187 if (uv < 0x80000000) {
eb160463
GS
188 *d++ = (U8)(( uv >> 30) | 0xfc);
189 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
193 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
194 return d;
195 }
6588300d 196#ifdef UTF8_QUAD_MAX
d7578b48 197 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
198#endif
199 {
eb160463
GS
200 *d++ = 0xfe; /* Can't match U+FEFF! */
201 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
206 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
207 return d;
208 }
6588300d 209#ifdef UTF8_QUAD_MAX
a0ed51b3 210 {
eb160463
GS
211 *d++ = 0xff; /* Can't match U+FFFE! */
212 *d++ = 0x80; /* 6 Reserved bits */
213 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
214 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
223 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
224 return d;
225 }
226#endif
537124e4 227#endif /* Non loop style */
a0ed51b3 228}
646ca15d 229/*
07693fe6
KW
230=for apidoc uvchr_to_utf8
231
bcb1a2d4 232Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
233of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
234C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
235the byte after the end of the new character. In other words,
07693fe6
KW
236
237 d = uvchr_to_utf8(d, uv);
238
239is the recommended wide native character-aware way of saying
240
241 *(d++) = uv;
242
de69f3af
KW
243This function accepts any UV as input. To forbid or warn on non-Unicode code
244points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
245
07693fe6
KW
246=cut
247*/
248
de69f3af
KW
249/* This is also a macro */
250PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
251
07693fe6
KW
252U8 *
253Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
254{
de69f3af 255 return uvchr_to_utf8(d, uv);
07693fe6
KW
256}
257
de69f3af
KW
258/*
259=for apidoc uvchr_to_utf8_flags
260
261Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
262of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
263C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
264the byte after the end of the new character. In other words,
de69f3af
KW
265
266 d = uvchr_to_utf8_flags(d, uv, flags);
267
268or, in most cases,
269
270 d = uvchr_to_utf8_flags(d, uv, 0);
271
272This is the Unicode-aware way of saying
273
274 *(d++) = uv;
275
276This function will convert to UTF-8 (and not warn) even code points that aren't
277legal Unicode or are problematic, unless C<flags> contains one or more of the
278following flags:
279
280If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
281the function will raise a warning, provided UTF8 warnings are enabled. If instead
282UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
283If both flags are set, the function will both warn and return NULL.
284
4c3cfd5d 285The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 286affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 287UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
288code points that are
289above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
290even less portable) can be warned and/or disallowed even if other above-Unicode
291code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
292flags.
293
294And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
295above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
296DISALLOW flags.
297
298=cut
299*/
300
301/* This is also a macro */
302PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
303
07693fe6
KW
304U8 *
305Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
306{
de69f3af 307 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
308}
309
310/*
646ca15d 311
f7d739d1 312Tests if the first C<len> bytes of string C<s> form a valid UTF-8
bcb1a2d4
KW
313character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC) character is a
314valid UTF-8 character. The number of bytes in the UTF-8 character
646ca15d
JH
315will be returned if it is valid, otherwise 0.
316
317This is the "slow" version as opposed to the "fast" version which is
318the "unrolled" IS_UTF8_CHAR(). E.g. for t/uni/class.t the speed
319difference is a factor of 2 to 3. For lengths (UTF8SKIP(s)) of four
320or less you should use the IS_UTF8_CHAR(), for lengths of five or more
321you should use the _slow(). In practice this means that the _slow()
322will be used very rarely, since the maximum Unicode code point (as of
323Unicode 4.1) is U+10FFFF, which encodes in UTF-8 to four bytes. Only
537124e4 324the "Perl extended UTF-8" (e.g, the infamous 'v-strings') will encode into
646ca15d
JH
325five bytes or more.
326
327=cut */
7af276bc 328PERL_STATIC_INLINE STRLEN
5f66b61c 329S_is_utf8_char_slow(const U8 *s, const STRLEN len)
646ca15d 330{
cd7e6c88 331 dTHX; /* The function called below requires thread context */
646ca15d 332
cd7e6c88 333 STRLEN actual_len;
646ca15d 334
cd7e6c88 335 PERL_ARGS_ASSERT_IS_UTF8_CHAR_SLOW;
646ca15d 336
18712bce 337 utf8n_to_uvchr(s, len, &actual_len, UTF8_CHECK_ONLY);
646ca15d 338
cd7e6c88 339 return (actual_len == (STRLEN) -1) ? 0 : actual_len;
646ca15d 340}
9041c2e3
NIS
341
342/*
492a624f
KW
343=for apidoc is_utf8_char_buf
344
345Returns the number of bytes that comprise the first UTF-8 encoded character in
346buffer C<buf>. C<buf_end> should point to one position beyond the end of the
347buffer. 0 is returned if C<buf> does not point to a complete, valid UTF-8
348encoded character.
349
350Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
351machines) is a valid UTF-8 character.
352
353=cut */
354
355STRLEN
356Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
357{
358
359 STRLEN len;
360
361 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
362
363 if (buf_end <= buf) {
364 return 0;
365 }
366
367 len = buf_end - buf;
368 if (len > UTF8SKIP(buf)) {
369 len = UTF8SKIP(buf);
370 }
371
492a624f
KW
372 if (IS_UTF8_CHAR_FAST(len))
373 return IS_UTF8_CHAR(buf, len) ? len : 0;
492a624f
KW
374 return is_utf8_char_slow(buf, len);
375}
376
377/*
87cea99e 378=for apidoc is_utf8_char
eebe1485 379
5da9da9e 380Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
381character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
382character is a valid UTF-8 character. The actual number of bytes in the UTF-8
383character will be returned if it is valid, otherwise 0.
9041c2e3 384
76848387 385This function is deprecated due to the possibility that malformed input could
a1433954 386cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 387instead.
e0328548 388
82686b01 389=cut */
76848387 390
067a85ef 391STRLEN
668b6d8d 392Perl_is_utf8_char(const U8 *s)
386d01d6 393{
7918f24d 394 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 395
76848387 396 /* Assumes we have enough space, which is why this is deprecated */
492a624f 397 return is_utf8_char_buf(s, s + UTF8SKIP(s));
386d01d6
GS
398}
399
eaf7a4d2 400
6662521e 401/*
87cea99e 402=for apidoc is_utf8_string
6662521e 403
a1433954 404Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 405UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
406using C<strlen(s)> (which means if you use this option, that C<s> can't have
407embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
408that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 409
a1433954 410See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 411
6662521e
GS
412=cut
413*/
414
8e84507e 415bool
668b6d8d 416Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 417{
35da51f7 418 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 419 const U8* x = s;
067a85ef 420
7918f24d 421 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 422
6662521e 423 while (x < send) {
1acdb0da 424 /* Inline the easy bits of is_utf8_char() here for speed... */
e0328548
KW
425 if (UTF8_IS_INVARIANT(*x)) {
426 x++;
427 }
1acdb0da
JH
428 else {
429 /* ... and call is_utf8_char() only if really needed. */
e0328548
KW
430 const STRLEN c = UTF8SKIP(x);
431 const U8* const next_char_ptr = x + c;
432
433 if (next_char_ptr > send) {
434 return FALSE;
435 }
436
768c67ee
JH
437 if (IS_UTF8_CHAR_FAST(c)) {
438 if (!IS_UTF8_CHAR(x, c))
e0328548 439 return FALSE;
3c614e38 440 }
e0328548
KW
441 else if (! is_utf8_char_slow(x, c)) {
442 return FALSE;
443 }
444 x = next_char_ptr;
1acdb0da 445 }
6662521e 446 }
768c67ee 447
067a85ef 448 return TRUE;
6662521e
GS
449}
450
67e989fb 451/*
814fafa7
NC
452Implemented as a macro in utf8.h
453
87cea99e 454=for apidoc is_utf8_string_loc
814fafa7 455
a1433954
KW
456Like L</is_utf8_string> but stores the location of the failure (in the
457case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
458"utf8ness success") in the C<ep>.
459
a1433954 460See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 461
87cea99e 462=for apidoc is_utf8_string_loclen
81cd54e3 463
a1433954
KW
464Like L</is_utf8_string>() but stores the location of the failure (in the
465case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
466"utf8ness success") in the C<ep>, and the number of UTF-8
467encoded characters in the C<el>.
468
a1433954 469See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
470
471=cut
472*/
473
474bool
668b6d8d 475Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 476{
35da51f7 477 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 478 const U8* x = s;
81cd54e3 479 STRLEN c;
3ebfea28 480 STRLEN outlen = 0;
7918f24d
NC
481
482 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 483
81cd54e3 484 while (x < send) {
e0328548
KW
485 const U8* next_char_ptr;
486
81cd54e3
JH
487 /* Inline the easy bits of is_utf8_char() here for speed... */
488 if (UTF8_IS_INVARIANT(*x))
e0328548 489 next_char_ptr = x + 1;
81cd54e3 490 else {
768c67ee 491 /* ... and call is_utf8_char() only if really needed. */
768c67ee 492 c = UTF8SKIP(x);
e0328548
KW
493 next_char_ptr = c + x;
494 if (next_char_ptr > send) {
495 goto out;
496 }
768c67ee
JH
497 if (IS_UTF8_CHAR_FAST(c)) {
498 if (!IS_UTF8_CHAR(x, c))
499 c = 0;
500 } else
501 c = is_utf8_char_slow(x, c);
768c67ee
JH
502 if (!c)
503 goto out;
81cd54e3 504 }
e0328548 505 x = next_char_ptr;
3ebfea28 506 outlen++;
81cd54e3 507 }
768c67ee
JH
508
509 out:
3ebfea28
AL
510 if (el)
511 *el = outlen;
512
768c67ee
JH
513 if (ep)
514 *ep = x;
3ebfea28 515 return (x == send);
81cd54e3
JH
516}
517
518/*
768c67ee 519
de69f3af 520=for apidoc utf8n_to_uvchr
378516de
KW
521
522THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 523Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 524
9041c2e3 525Bottom level UTF-8 decode routine.
de69f3af 526Returns the native code point value of the first character in the string C<s>,
746afd53
KW
527which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
528C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
529the length, in bytes, of that character.
949cf498
KW
530
531The value of C<flags> determines the behavior when C<s> does not point to a
532well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
533zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
534next possible position in C<s> that could begin a non-malformed character.
535Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
536
537Various ALLOW flags can be set in C<flags> to allow (and not warn on)
538individual types of malformations, such as the sequence being overlong (that
539is, when there is a shorter sequence that can express the same code point;
540overlong sequences are expressly forbidden in the UTF-8 standard due to
541potential security issues). Another malformation example is the first byte of
542a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
543flags. For allowed 0 length strings, this function returns 0; for allowed
544overlong sequences, the computed code point is returned; for all other allowed
545malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
546determinable reasonable value.
949cf498
KW
547
548The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
549flags) malformation is found. If this flag is set, the routine assumes that
550the caller will raise a warning, and this function will silently just set
d088425d
KW
551C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
552
75200dff 553Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
554character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
555in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
556first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
557input had an error.
949cf498
KW
558
559Certain code points are considered problematic. These are Unicode surrogates,
746afd53 560Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 561By default these are considered regular code points, but certain situations
5eafe189 562warrant special handling for them. If C<flags> contains
949cf498
KW
563UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
564malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
565UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
566maximum) can be set to disallow these categories individually.
567
568The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
569UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
570for their respective categories, but otherwise the code points are considered
571valid (not malformations). To get a category to both be treated as a
572malformation and raise a warning, specify both the WARN and DISALLOW flags.
573(But note that warnings are not raised if lexically disabled nor if
574UTF8_CHECK_ONLY is also specified.)
575
576Very large code points (above 0x7FFF_FFFF) are considered more problematic than
577the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
578reasons: they requre at least 32 bits to represent them on ASCII platforms, are
579not representable at all on EBCDIC platforms, and the original UTF-8
580specification never went above this number (the current 0x10FFFF limit was
581imposed later). (The smaller ones, those that fit into 32 bits, are
582representable by a UV on ASCII platforms, but not by an IV, which means that
583the number of operations that can be performed on them is quite restricted.)
584The UTF-8 encoding on ASCII platforms for these large code points begins with a
585byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
586be treated as malformations, while allowing smaller above-Unicode code points.
587(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
588including these, as malformations.)
589Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 590the other WARN flags, but applies just to these code points.
949cf498
KW
591
592All other code points corresponding to Unicode characters, including private
593use and those yet to be assigned, are never considered malformed and never
594warn.
67e989fb 595
37607a96
PK
596=cut
597*/
67e989fb 598
a0ed51b3 599UV
de69f3af 600Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 601{
97aff369 602 dVAR;
d4c19fe8 603 const U8 * const s0 = s;
eb83ed87 604 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 605 U8 * send;
eb83ed87
KW
606 UV uv = *s;
607 STRLEN expectlen;
949cf498 608 SV* sv = NULL;
eb83ed87
KW
609 UV outlier_ret = 0; /* return value when input is in error or problematic
610 */
611 UV pack_warn = 0; /* Save result of packWARN() for later */
612 bool unexpected_non_continuation = FALSE;
613 bool overflowed = FALSE;
2f8f112e 614 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 615
eb83ed87 616 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 617
de69f3af 618 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 619
eb83ed87
KW
620 /* The order of malformation tests here is important. We should consume as
621 * few bytes as possible in order to not skip any valid character. This is
622 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
623 * http://unicode.org/reports/tr36 for more discussion as to why. For
624 * example, once we've done a UTF8SKIP, we can tell the expected number of
625 * bytes, and could fail right off the bat if the input parameters indicate
626 * that there are too few available. But it could be that just that first
627 * byte is garbled, and the intended character occupies fewer bytes. If we
628 * blindly assumed that the first byte is correct, and skipped based on
629 * that number, we could skip over a valid input character. So instead, we
630 * always examine the sequence byte-by-byte.
631 *
632 * We also should not consume too few bytes, otherwise someone could inject
633 * things. For example, an input could be deliberately designed to
634 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 635 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
636 * which is actually part of of the overflowing sequence), that could look
637 * legitimate to the caller, which could discard the initial partial
638 * sequence and process the rest, inappropriately */
639
640 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 641 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
642 if (retlen) {
643 *retlen = 0;
644 }
a0dbb045 645
eb83ed87
KW
646 if (flags & UTF8_ALLOW_EMPTY) {
647 return 0;
648 }
649 if (! (flags & UTF8_CHECK_ONLY)) {
650 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
651 }
0c443dc2
JH
652 goto malformed;
653 }
654
eb83ed87
KW
655 expectlen = UTF8SKIP(s);
656
657 /* A well-formed UTF-8 character, as the vast majority of calls to this
658 * function will be for, has this expected length. For efficiency, set
659 * things up here to return it. It will be overriden only in those rare
660 * cases where a malformation is found */
661 if (retlen) {
662 *retlen = expectlen;
663 }
664
665 /* An invariant is trivially well-formed */
1d72bdf6 666 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 667 return uv;
a0ed51b3 668 }
67e989fb 669
eb83ed87 670 /* A continuation character can't start a valid sequence */
b5b9af04 671 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
672 if (flags & UTF8_ALLOW_CONTINUATION) {
673 if (retlen) {
674 *retlen = 1;
675 }
676 return UNICODE_REPLACEMENT;
677 }
ba210ebe 678
eb83ed87
KW
679 if (! (flags & UTF8_CHECK_ONLY)) {
680 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
681 }
682 curlen = 1;
ba210ebe
JH
683 goto malformed;
684 }
9041c2e3 685
dcd27b3c
KW
686 /* Here is not a continuation byte, nor an invariant. The only thing left
687 * is a start byte (possibly for an overlong) */
688
1d72bdf6 689#ifdef EBCDIC
bc3632a8 690 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
691#endif
692
eb83ed87
KW
693 /* Remove the leading bits that indicate the number of bytes in the
694 * character's whole UTF-8 sequence, leaving just the bits that are part of
695 * the value */
696 uv &= UTF_START_MASK(expectlen);
ba210ebe 697
eb83ed87
KW
698 /* Now, loop through the remaining bytes in the character's sequence,
699 * accumulating each into the working value as we go. Be sure to not look
700 * past the end of the input string */
0b8d30e8
KW
701 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
702
eb83ed87 703 for (s = s0 + 1; s < send; s++) {
b5b9af04 704 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
705#ifndef EBCDIC /* Can't overflow in EBCDIC */
706 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
707
708 /* The original implementors viewed this malformation as more
709 * serious than the others (though I, khw, don't understand
710 * why, since other malformations also give very very wrong
711 * results), so there is no way to turn off checking for it.
712 * Set a flag, but keep going in the loop, so that we absorb
713 * the rest of the bytes that comprise the character. */
714 overflowed = TRUE;
715 overflow_byte = *s; /* Save for warning message's use */
716 }
717#endif
8850bf83 718 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
719 }
720 else {
721 /* Here, found a non-continuation before processing all expected
722 * bytes. This byte begins a new character, so quit, even if
723 * allowing this malformation. */
724 unexpected_non_continuation = TRUE;
725 break;
726 }
727 } /* End of loop through the character's bytes */
728
729 /* Save how many bytes were actually in the character */
730 curlen = s - s0;
731
732 /* The loop above finds two types of malformations: non-continuation and/or
733 * overflow. The non-continuation malformation is really a too-short
734 * malformation, as it means that the current character ended before it was
735 * expected to (being terminated prematurely by the beginning of the next
736 * character, whereas in the too-short malformation there just are too few
737 * bytes available to hold the character. In both cases, the check below
738 * that we have found the expected number of bytes would fail if executed.)
739 * Thus the non-continuation malformation is really unnecessary, being a
740 * subset of the too-short malformation. But there may be existing
741 * applications that are expecting the non-continuation type, so we retain
742 * it, and return it in preference to the too-short malformation. (If this
743 * code were being written from scratch, the two types might be collapsed
744 * into one.) I, khw, am also giving priority to returning the
745 * non-continuation and too-short malformations over overflow when multiple
746 * ones are present. I don't know of any real reason to prefer one over
747 * the other, except that it seems to me that multiple-byte errors trumps
748 * errors from a single byte */
b5b9af04 749 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
750 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
751 if (! (flags & UTF8_CHECK_ONLY)) {
752 if (curlen == 1) {
753 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
754 }
755 else {
756 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
757 }
758 }
eb83ed87
KW
759 goto malformed;
760 }
761 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
762
763 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
764 * as what the original expectations were. */
765 do_overlong_test = FALSE;
eb83ed87
KW
766 if (retlen) {
767 *retlen = curlen;
768 }
769 }
b5b9af04 770 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
771 if (! (flags & UTF8_ALLOW_SHORT)) {
772 if (! (flags & UTF8_CHECK_ONLY)) {
773 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 774 }
eb83ed87
KW
775 goto malformed;
776 }
777 uv = UNICODE_REPLACEMENT;
2f8f112e 778 do_overlong_test = FALSE;
eb83ed87
KW
779 if (retlen) {
780 *retlen = curlen;
781 }
782 }
783
ea5ced44 784#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 785 if (UNLIKELY(overflowed)) {
eb83ed87 786 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 787 goto malformed;
eb83ed87
KW
788 }
789#endif
790
2f8f112e 791 if (do_overlong_test
5aaebcb3 792 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
793 && ! (flags & UTF8_ALLOW_LONG))
794 {
eb83ed87
KW
795 /* The overlong malformation has lower precedence than the others.
796 * Note that if this malformation is allowed, we return the actual
797 * value, instead of the replacement character. This is because this
798 * value is actually well-defined. */
799 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 800 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
801 }
802 goto malformed;
803 }
804
1a89bb6c 805 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
806 * problematic code point that is not allowed by the input parameters. */
807 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
808 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
809 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
810 {
949cf498 811 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
812
813 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
814 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 815 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 816 && ckWARN_d(WARN_SURROGATE))
eb83ed87 817 {
111d382d 818 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 819 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
820 }
821 if (flags & UTF8_DISALLOW_SURROGATE) {
822 goto disallowed;
823 }
824 }
949cf498 825 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 826 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 827 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 828 {
111d382d 829 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 830 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 831 }
ea5ced44
KW
832#ifndef EBCDIC /* EBCDIC always allows FE, FF */
833
834 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
835 * points. We test for these after the regular SUPER ones, and
836 * before possibly bailing out, so that the more dire warning
837 * overrides the regular one, if applicable */
838 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
839 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
840 {
841 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
842 == UTF8_WARN_FE_FF
843 && ckWARN_d(WARN_UTF8))
844 {
845 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
846 pack_warn = packWARN(WARN_UTF8);
847 }
848 if (flags & UTF8_DISALLOW_FE_FF) {
849 goto disallowed;
850 }
851 }
852#endif
949cf498
KW
853 if (flags & UTF8_DISALLOW_SUPER) {
854 goto disallowed;
855 }
856 }
4190d317
KW
857 else if (UNICODE_IS_NONCHAR(uv)) {
858 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 859 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
860 {
861 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 862 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
863 }
864 if (flags & UTF8_DISALLOW_NONCHAR) {
865 goto disallowed;
866 }
867 }
949cf498 868
eb83ed87 869 if (sv) {
de69f3af
KW
870 outlier_ret = uv; /* Note we don't bother to convert to native,
871 as all the outlier code points are the same
872 in both ASCII and EBCDIC */
eb83ed87
KW
873 goto do_warn;
874 }
875
949cf498
KW
876 /* Here, this is not considered a malformed character, so drop through
877 * to return it */
a0ed51b3 878 }
ba210ebe 879
de69f3af 880 return UNI_TO_NATIVE(uv);
ba210ebe 881
eb83ed87
KW
882 /* There are three cases which get to beyond this point. In all 3 cases:
883 * <sv> if not null points to a string to print as a warning.
884 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
885 * set.
886 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
887 * This is done by initializing it to 0, and changing it only
888 * for case 1).
889 * The 3 cases are:
890 * 1) The input is valid but problematic, and to be warned about. The
891 * return value is the resultant code point; <*retlen> is set to
892 * <curlen>, the number of bytes that comprise the code point.
893 * <pack_warn> contains the result of packWARN() for the warning
894 * types. The entry point for this case is the label <do_warn>;
895 * 2) The input is a valid code point but disallowed by the parameters to
896 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
897 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
898 * comprise the code point. <pack_warn> contains the result of
899 * packWARN() for the warning types. The entry point for this case is
900 * the label <disallowed>.
901 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
902 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
903 * bytes that comprise the malformation. All such malformations are
904 * assumed to be warning type <utf8>. The entry point for this case
905 * is the label <malformed>.
906 */
949cf498 907
ba210ebe
JH
908malformed:
909
eb83ed87
KW
910 if (sv && ckWARN_d(WARN_UTF8)) {
911 pack_warn = packWARN(WARN_UTF8);
912 }
913
914disallowed:
915
fcc8fcf6 916 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 917 if (retlen)
10edeb5d 918 *retlen = ((STRLEN) -1);
ba210ebe
JH
919 return 0;
920 }
921
eb83ed87 922do_warn:
5b311467 923
eb83ed87
KW
924 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
925 if warnings are to be raised. */
f555bc63 926 const char * const string = SvPVX_const(sv);
a0dbb045 927
f555bc63
KW
928 if (PL_op)
929 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
930 else
931 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
932 }
933
eb83ed87
KW
934 if (retlen) {
935 *retlen = curlen;
936 }
ba210ebe 937
eb83ed87 938 return outlier_ret;
a0ed51b3
LW
939}
940
8e84507e 941/*
ec5f19d0
KW
942=for apidoc utf8_to_uvchr_buf
943
944Returns the native code point of the first character in the string C<s> which
945is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 946C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 947
524080c4
KW
948If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
949enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
950NULL) to -1. If those warnings are off, the computed value, if well-defined
951(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
952C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
953the next possible position in C<s> that could begin a non-malformed character.
de69f3af 954See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 955returned.
ec5f19d0
KW
956
957=cut
958*/
959
960
961UV
962Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
963{
ec5f19d0
KW
964 assert(s < send);
965
966 return utf8n_to_uvchr(s, send - s, retlen,
967 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
968}
969
27d6c58a 970/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 971 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 972 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
973
974UV
975Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
976{
010ab96b
KW
977 UV expectlen = UTF8SKIP(s);
978 const U8* send = s + expectlen;
9ff2f0f7 979 UV uv = *s;
3986bb7c 980
27d6c58a
KW
981 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
982
010ab96b
KW
983 if (retlen) {
984 *retlen = expectlen;
985 }
986
987 /* An invariant is trivially returned */
988 if (expectlen == 1) {
9ff2f0f7 989 return uv;
010ab96b
KW
990 }
991
9ff2f0f7
KW
992#ifdef EBCDIC
993 uv = NATIVE_UTF8_TO_I8(uv);
994#endif
995
010ab96b
KW
996 /* Remove the leading bits that indicate the number of bytes, leaving just
997 * the bits that are part of the value */
998 uv &= UTF_START_MASK(expectlen);
999
1000 /* Now, loop through the remaining bytes, accumulating each into the
1001 * working total as we go. (I khw tried unrolling the loop for up to 4
1002 * bytes, but there was no performance improvement) */
1003 for (++s; s < send; s++) {
1004 uv = UTF8_ACCUMULATE(uv, *s);
1005 }
1006
3986bb7c 1007 return UNI_TO_NATIVE(uv);
010ab96b 1008
27d6c58a
KW
1009}
1010
ec5f19d0 1011/*
87cea99e 1012=for apidoc utf8_to_uvchr
9041c2e3 1013
6ee84de2 1014Returns the native code point of the first character in the string C<s>
1e54db1a 1015which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1016length, in bytes, of that character.
1017
4b88fb76 1018Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
1019malformed input could cause reading beyond the end of the input buffer, which
1020is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 1021
524080c4
KW
1022If C<s> points to one of the detected malformations, and UTF8 warnings are
1023enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1024NULL) to -1. If those warnings are off, the computed value if well-defined (or
1025the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1026is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1027next possible position in C<s> that could begin a non-malformed character.
de69f3af 1028See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
1029
1030=cut
1031*/
1032
1033UV
7fc63493 1034Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 1035{
7918f24d
NC
1036 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
1037
2ff6c191 1038 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
1039}
1040
1041/*
ec5f19d0
KW
1042=for apidoc utf8_to_uvuni_buf
1043
de69f3af
KW
1044Only in very rare circumstances should code need to be dealing in Unicode
1045(as opposed to native) code points. In those few cases, use
1046C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1047
1048Returns the Unicode (not-native) code point of the first character in the
1049string C<s> which
ec5f19d0
KW
1050is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1051C<retlen> will be set to the length, in bytes, of that character.
1052
524080c4
KW
1053If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1054enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1055NULL) to -1. If those warnings are off, the computed value if well-defined (or
1056the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1057is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1058next possible position in C<s> that could begin a non-malformed character.
de69f3af 1059See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
1060
1061=cut
1062*/
1063
1064UV
1065Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1066{
1067 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1068
1069 assert(send > s);
1070
1071 /* Call the low level routine asking for checks */
de69f3af
KW
1072 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1073 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1074}
1075
5495102a
KW
1076/* DEPRECATED!
1077 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 1078 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 1079 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
1080
1081UV
1082Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
1083{
1084 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1085
010ab96b 1086 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
27d6c58a
KW
1087}
1088
ec5f19d0 1089/*
87cea99e 1090=for apidoc utf8_to_uvuni
9041c2e3
NIS
1091
1092Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1093which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1094length, in bytes, of that character.
1095
4b88fb76 1096Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31 1097malformed input could cause reading beyond the end of the input buffer, which
4f83cdcd
KW
1098is one reason why this function is deprecated. The other is that only in
1099extremely limited circumstances should the Unicode versus native code point be
de69f3af 1100of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
9041c2e3 1101
524080c4
KW
1102If C<s> points to one of the detected malformations, and UTF8 warnings are
1103enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1104NULL) to -1. If those warnings are off, the computed value if well-defined (or
1105the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1106is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1107next possible position in C<s> that could begin a non-malformed character.
de69f3af 1108See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1109
1110=cut
1111*/
1112
1113UV
7fc63493 1114Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1115{
7918f24d
NC
1116 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1117
5495102a 1118 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
8e84507e
NIS
1119}
1120
b76347f2 1121/*
87cea99e 1122=for apidoc utf8_length
b76347f2
JH
1123
1124Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1125Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1126up past C<e>, croaks.
b76347f2
JH
1127
1128=cut
1129*/
1130
1131STRLEN
35a4481c 1132Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1133{
97aff369 1134 dVAR;
b76347f2
JH
1135 STRLEN len = 0;
1136
7918f24d
NC
1137 PERL_ARGS_ASSERT_UTF8_LENGTH;
1138
8850bf83
JH
1139 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1140 * the bitops (especially ~) can create illegal UTF-8.
1141 * In other words: in Perl UTF-8 is not just for Unicode. */
1142
a3b680e6
AL
1143 if (e < s)
1144 goto warn_and_return;
b76347f2 1145 while (s < e) {
4cbf4130 1146 s += UTF8SKIP(s);
8e91ec7f
AV
1147 len++;
1148 }
1149
1150 if (e != s) {
1151 len--;
1152 warn_and_return:
9b387841
NC
1153 if (PL_op)
1154 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1155 "%s in %s", unees, OP_DESC(PL_op));
1156 else
61a12c31 1157 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1158 }
1159
1160 return len;
1161}
1162
b06226ff 1163/*
87cea99e 1164=for apidoc utf8_distance
b06226ff 1165
1e54db1a 1166Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1167and C<b>.
1168
1169WARNING: use only if you *know* that the pointers point inside the
1170same UTF-8 buffer.
1171
37607a96
PK
1172=cut
1173*/
a0ed51b3 1174
02eb7b47 1175IV
35a4481c 1176Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1177{
7918f24d
NC
1178 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1179
bf1665bc 1180 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1181}
1182
b06226ff 1183/*
87cea99e 1184=for apidoc utf8_hop
b06226ff 1185
8850bf83
JH
1186Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1187forward or backward.
b06226ff
JH
1188
1189WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1190the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1191on the first byte of character or just after the last byte of a character.
b06226ff 1192
37607a96
PK
1193=cut
1194*/
a0ed51b3
LW
1195
1196U8 *
4373e329 1197Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1198{
7918f24d
NC
1199 PERL_ARGS_ASSERT_UTF8_HOP;
1200
96a5add6 1201 PERL_UNUSED_CONTEXT;
8850bf83
JH
1202 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1203 * the bitops (especially ~) can create illegal UTF-8.
1204 * In other words: in Perl UTF-8 is not just for Unicode. */
1205
a0ed51b3
LW
1206 if (off >= 0) {
1207 while (off--)
1208 s += UTF8SKIP(s);
1209 }
1210 else {
1211 while (off++) {
1212 s--;
8850bf83
JH
1213 while (UTF8_IS_CONTINUATION(*s))
1214 s--;
a0ed51b3
LW
1215 }
1216 }
4373e329 1217 return (U8 *)s;
a0ed51b3
LW
1218}
1219
6940069f 1220/*
fed3ba5d
NC
1221=for apidoc bytes_cmp_utf8
1222
a1433954 1223Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1224sequence of characters (stored as UTF-8)
1225in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1226equal, -1 or -2 if the first string is less than the second string, +1 or +2
1227if the first string is greater than the second string.
1228
1229-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1230longer string. -2 or +2 is returned if
1231there was a difference between characters
fed3ba5d
NC
1232within the strings.
1233
1234=cut
1235*/
1236
1237int
1238Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1239{
1240 const U8 *const bend = b + blen;
1241 const U8 *const uend = u + ulen;
1242
1243 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1244
1245 PERL_UNUSED_CONTEXT;
1246
1247 while (b < bend && u < uend) {
1248 U8 c = *u++;
1249 if (!UTF8_IS_INVARIANT(c)) {
1250 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1251 if (u < uend) {
1252 U8 c1 = *u++;
1253 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1254 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1255 } else {
1256 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1257 "Malformed UTF-8 character "
1258 "(unexpected non-continuation byte 0x%02x"
1259 ", immediately after start byte 0x%02x)"
1260 /* Dear diag.t, it's in the pod. */
1261 "%s%s", c1, c,
1262 PL_op ? " in " : "",
1263 PL_op ? OP_DESC(PL_op) : "");
1264 return -2;
1265 }
1266 } else {
1267 if (PL_op)
1268 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1269 "%s in %s", unees, OP_DESC(PL_op));
1270 else
61a12c31 1271 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1272 return -2; /* Really want to return undef :-) */
1273 }
1274 } else {
1275 return -2;
1276 }
1277 }
1278 if (*b != c) {
1279 return *b < c ? -2 : +2;
1280 }
1281 ++b;
1282 }
1283
1284 if (b == bend && u == uend)
1285 return 0;
1286
1287 return b < bend ? +1 : -1;
1288}
1289
1290/*
87cea99e 1291=for apidoc utf8_to_bytes
6940069f 1292
2bbc8d55 1293Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1294Unlike L</bytes_to_utf8>, this over-writes the original string, and
1295updates C<len> to contain the new length.
67e989fb 1296Returns zero on failure, setting C<len> to -1.
6940069f 1297
a1433954 1298If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1299
6940069f
GS
1300=cut
1301*/
1302
1303U8 *
37607a96 1304Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1305{
d4c19fe8
AL
1306 U8 * const save = s;
1307 U8 * const send = s + *len;
6940069f 1308 U8 *d;
246fae53 1309
7918f24d
NC
1310 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1311
1e54db1a 1312 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1313 while (s < send) {
d59937ca
KW
1314 if (! UTF8_IS_INVARIANT(*s)) {
1315 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1316 *len = ((STRLEN) -1);
1317 return 0;
1318 }
1319 s++;
dcad2880 1320 }
d59937ca 1321 s++;
246fae53 1322 }
dcad2880
JH
1323
1324 d = s = save;
6940069f 1325 while (s < send) {
80e0b38f
KW
1326 U8 c = *s++;
1327 if (! UTF8_IS_INVARIANT(c)) {
1328 /* Then it is two-byte encoded */
1329 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1330 s++;
1331 }
1332 *d++ = c;
6940069f
GS
1333 }
1334 *d = '\0';
246fae53 1335 *len = d - save;
6940069f
GS
1336 return save;
1337}
1338
1339/*
87cea99e 1340=for apidoc bytes_from_utf8
f9a63242 1341
2bbc8d55 1342Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1343Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1344the newly-created string, and updates C<len> to contain the new
1345length. Returns the original string if no conversion occurs, C<len>
72d33970 1346is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
13470 if C<s> is converted or consisted entirely of characters that are invariant
1348in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1349
37607a96
PK
1350=cut
1351*/
f9a63242
JH
1352
1353U8 *
e1ec3a88 1354Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1355{
f9a63242 1356 U8 *d;
e1ec3a88
AL
1357 const U8 *start = s;
1358 const U8 *send;
f9a63242
JH
1359 I32 count = 0;
1360
7918f24d
NC
1361 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1362
96a5add6 1363 PERL_UNUSED_CONTEXT;
f9a63242 1364 if (!*is_utf8)
73d840c0 1365 return (U8 *)start;
f9a63242 1366
1e54db1a 1367 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1368 for (send = s + *len; s < send;) {
d59937ca
KW
1369 if (! UTF8_IS_INVARIANT(*s)) {
1370 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1371 return (U8 *)start;
d59937ca
KW
1372 }
1373 count++;
1374 s++;
db42d148 1375 }
d59937ca 1376 s++;
f9a63242
JH
1377 }
1378
35da51f7 1379 *is_utf8 = FALSE;
f9a63242 1380
212542aa 1381 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1382 s = start; start = d;
f9a63242
JH
1383 while (s < send) {
1384 U8 c = *s++;
1a91c45d 1385 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1386 /* Then it is two-byte encoded */
1a91c45d
KW
1387 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1388 s++;
c4d5f83a
NIS
1389 }
1390 *d++ = c;
f9a63242
JH
1391 }
1392 *d = '\0';
1393 *len = d - start;
73d840c0 1394 return (U8 *)start;
f9a63242
JH
1395}
1396
1397/*
87cea99e 1398=for apidoc bytes_to_utf8
6940069f 1399
ff97e5cf
KW
1400Converts a string C<s> of length C<len> bytes from the native encoding into
1401UTF-8.
6662521e 1402Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1403reflect the new length in bytes.
6940069f 1404
75200dff 1405A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1406
1407If you want to convert to UTF-8 from encodings other than
1408the native (Latin1 or EBCDIC),
a1433954 1409see L</sv_recode_to_utf8>().
c9ada85f 1410
497711e7 1411=cut
6940069f
GS
1412*/
1413
c682ebef
FC
1414/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1415 likewise need duplication. */
1416
6940069f 1417U8*
35a4481c 1418Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1419{
35a4481c 1420 const U8 * const send = s + (*len);
6940069f
GS
1421 U8 *d;
1422 U8 *dst;
7918f24d
NC
1423
1424 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1425 PERL_UNUSED_CONTEXT;
6940069f 1426
212542aa 1427 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1428 dst = d;
1429
1430 while (s < send) {
55d09dc8
KW
1431 append_utf8_from_native_byte(*s, &d);
1432 s++;
6940069f
GS
1433 }
1434 *d = '\0';
6662521e 1435 *len = d-dst;
6940069f
GS
1436 return dst;
1437}
1438
a0ed51b3 1439/*
dea0fc0b 1440 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1441 *
1442 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1443 * We optimize for native, for obvious reasons. */
1444
1445U8*
dea0fc0b 1446Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1447{
dea0fc0b
JH
1448 U8* pend;
1449 U8* dstart = d;
1450
7918f24d
NC
1451 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1452
dea0fc0b 1453 if (bytelen & 1)
f5992bc4 1454 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1455
1456 pend = p + bytelen;
1457
a0ed51b3 1458 while (p < pend) {
dea0fc0b
JH
1459 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1460 p += 2;
56d37426
KW
1461 if (UNI_IS_INVARIANT(uv)) {
1462 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1463 continue;
1464 }
56d37426
KW
1465 if (uv <= MAX_UTF8_TWO_BYTE) {
1466 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1467 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1468 continue;
1469 }
46956fad
KW
1470#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1471#define LAST_HIGH_SURROGATE 0xDBFF
1472#define FIRST_LOW_SURROGATE 0xDC00
1473#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1474 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1475 if (p >= pend) {
dea0fc0b 1476 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1477 } else {
1478 UV low = (p[0] << 8) + p[1];
1479 p += 2;
46956fad 1480 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1481 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1482 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1483 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1484 }
46956fad 1485 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1486 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1487 }
56d37426
KW
1488#ifdef EBCDIC
1489 d = uvoffuni_to_utf8_flags(d, uv, 0);
1490#else
a0ed51b3 1491 if (uv < 0x10000) {
eb160463
GS
1492 *d++ = (U8)(( uv >> 12) | 0xe0);
1493 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1494 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1495 continue;
1496 }
1497 else {
eb160463
GS
1498 *d++ = (U8)(( uv >> 18) | 0xf0);
1499 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1500 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1501 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1502 continue;
1503 }
56d37426 1504#endif
a0ed51b3 1505 }
dea0fc0b 1506 *newlen = d - dstart;
a0ed51b3
LW
1507 return d;
1508}
1509
1510/* Note: this one is slightly destructive of the source. */
1511
1512U8*
dea0fc0b 1513Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1514{
1515 U8* s = (U8*)p;
d4c19fe8 1516 U8* const send = s + bytelen;
7918f24d
NC
1517
1518 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1519
e0ea5e2d
NC
1520 if (bytelen & 1)
1521 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1522 (UV)bytelen);
1523
a0ed51b3 1524 while (s < send) {
d4c19fe8 1525 const U8 tmp = s[0];
a0ed51b3
LW
1526 s[0] = s[1];
1527 s[1] = tmp;
1528 s += 2;
1529 }
dea0fc0b 1530 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1531}
1532
922e8cb4
KW
1533bool
1534Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1535{
1536 U8 tmpbuf[UTF8_MAXBYTES+1];
1537 uvchr_to_utf8(tmpbuf, c);
1538 return _is_utf8_FOO(classnum, tmpbuf);
1539}
1540
f9ae8fb6
JD
1541/* Internal function so we can deprecate the external one, and call
1542 this one from other deprecated functions in this file */
1543
1544PERL_STATIC_INLINE bool
61b19385
KW
1545S_is_utf8_idfirst(pTHX_ const U8 *p)
1546{
1547 dVAR;
1548
1549 if (*p == '_')
1550 return TRUE;
1551 /* is_utf8_idstart would be more logical. */
f25ce844 1552 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1553}
1554
5092f92a 1555bool
84afefe6 1556Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1557{
89ebb4a3 1558 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1559 uvchr_to_utf8(tmpbuf, c);
61b19385 1560 return S_is_utf8_idfirst(aTHX_ tmpbuf);
a0ed51b3
LW
1561}
1562
1563bool
eba68aa0
KW
1564Perl__is_uni_perl_idcont(pTHX_ UV c)
1565{
1566 U8 tmpbuf[UTF8_MAXBYTES+1];
1567 uvchr_to_utf8(tmpbuf, c);
1568 return _is_utf8_perl_idcont(tmpbuf);
1569}
1570
1571bool
f91dcd13
KW
1572Perl__is_uni_perl_idstart(pTHX_ UV c)
1573{
1574 U8 tmpbuf[UTF8_MAXBYTES+1];
1575 uvchr_to_utf8(tmpbuf, c);
1576 return _is_utf8_perl_idstart(tmpbuf);
1577}
1578
3a4c58c9
KW
1579UV
1580Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1581{
1582 /* We have the latin1-range values compiled into the core, so just use
1583 * those, converting the result to utf8. The only difference between upper
1584 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1585 * either "SS" or "Ss". Which one to use is passed into the routine in
1586 * 'S_or_s' to avoid a test */
1587
1588 UV converted = toUPPER_LATIN1_MOD(c);
1589
1590 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1591
1592 assert(S_or_s == 'S' || S_or_s == 's');
1593
6f2d5cbc 1594 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1595 characters in this range */
3a4c58c9
KW
1596 *p = (U8) converted;
1597 *lenp = 1;
1598 return converted;
1599 }
1600
1601 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1602 * which it maps to one of them, so as to only have to have one check for
1603 * it in the main case */
1604 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1605 switch (c) {
1606 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1607 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1608 break;
1609 case MICRO_SIGN:
1610 converted = GREEK_CAPITAL_LETTER_MU;
1611 break;
1612 case LATIN_SMALL_LETTER_SHARP_S:
1613 *(p)++ = 'S';
1614 *p = S_or_s;
1615 *lenp = 2;
1616 return 'S';
1617 default:
1618 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1619 assert(0); /* NOTREACHED */
3a4c58c9
KW
1620 }
1621 }
1622
1623 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1624 *p = UTF8_TWO_BYTE_LO(converted);
1625 *lenp = 2;
1626
1627 return converted;
1628}
1629
50bda2c3
KW
1630/* Call the function to convert a UTF-8 encoded character to the specified case.
1631 * Note that there may be more than one character in the result.
1632 * INP is a pointer to the first byte of the input character
1633 * OUTP will be set to the first byte of the string of changed characters. It
1634 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1635 * LENP will be set to the length in bytes of the string of changed characters
1636 *
1637 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1638#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1639#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1640#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1641
1642/* This additionally has the input parameter SPECIALS, which if non-zero will
1643 * cause this to use the SPECIALS hash for folding (meaning get full case
1644 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1645#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1646
84afefe6
JH
1647UV
1648Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1649{
3a4c58c9
KW
1650 dVAR;
1651
a1433954
KW
1652 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1653 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1654 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1655 * the changed version may be longer than the original character.
1656 *
1657 * The ordinal of the first character of the changed version is returned
1658 * (but note, as explained above, that there may be more.) */
1659
7918f24d
NC
1660 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1661
3a4c58c9
KW
1662 if (c < 256) {
1663 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1664 }
1665
0ebc6274 1666 uvchr_to_utf8(p, c);
3a4c58c9 1667 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1668}
1669
84afefe6
JH
1670UV
1671Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1672{
3a4c58c9
KW
1673 dVAR;
1674
7918f24d
NC
1675 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1676
3a4c58c9
KW
1677 if (c < 256) {
1678 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1679 }
1680
0ebc6274 1681 uvchr_to_utf8(p, c);
3a4c58c9 1682 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1683}
1684
afc16117
KW
1685STATIC U8
1686S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1687{
1688 /* We have the latin1-range values compiled into the core, so just use
1689 * those, converting the result to utf8. Since the result is always just
a1433954 1690 * one character, we allow <p> to be NULL */
afc16117
KW
1691
1692 U8 converted = toLOWER_LATIN1(c);
1693
1694 if (p != NULL) {
6f2d5cbc 1695 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1696 *p = converted;
1697 *lenp = 1;
1698 }
1699 else {
1700 *p = UTF8_TWO_BYTE_HI(converted);
1701 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1702 *lenp = 2;
1703 }
1704 }
1705 return converted;
1706}
1707
84afefe6
JH
1708UV
1709Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1710{
968c5e6a
KW
1711 dVAR;
1712
7918f24d
NC
1713 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1714
afc16117
KW
1715 if (c < 256) {
1716 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1717 }
1718
afc16117 1719 uvchr_to_utf8(p, c);
968c5e6a 1720 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1721}
1722
84afefe6 1723UV
51910141 1724Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1725{
51910141 1726 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1727 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1728 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1729 *
1730 * Not to be used for locale folds
51910141 1731 */
f673fad4 1732
a1dde8de
KW
1733 UV converted;
1734
1735 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1736
1ca267a5
KW
1737 assert (! (flags & FOLD_FLAGS_LOCALE));
1738
a1dde8de
KW
1739 if (c == MICRO_SIGN) {
1740 converted = GREEK_SMALL_LETTER_MU;
1741 }
51910141 1742 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1743
1744 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1745 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1746 * under those circumstances. */
1747 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1748 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1749 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1750 p, *lenp, U8);
1751 return LATIN_SMALL_LETTER_LONG_S;
1752 }
1753 else {
4f489194
KW
1754 *(p)++ = 's';
1755 *p = 's';
1756 *lenp = 2;
1757 return 's';
1ca267a5 1758 }
a1dde8de
KW
1759 }
1760 else { /* In this range the fold of all other characters is their lower
1761 case */
1762 converted = toLOWER_LATIN1(c);
1763 }
1764
6f2d5cbc 1765 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1766 *p = (U8) converted;
1767 *lenp = 1;
1768 }
1769 else {
1770 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1771 *p = UTF8_TWO_BYTE_LO(converted);
1772 *lenp = 2;
1773 }
1774
1775 return converted;
1776}
1777
1778UV
31f05a37 1779Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1780{
4b593389 1781
a0270393
KW
1782 /* Not currently externally documented, and subject to change
1783 * <flags> bits meanings:
1784 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1785 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1786 * locale are to be used.
a0270393
KW
1787 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1788 */
4b593389 1789
36bb2ab6 1790 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1791
31f05a37
KW
1792 /* Tread a UTF-8 locale as not being in locale at all */
1793 if (IN_UTF8_CTYPE_LOCALE) {
1794 flags &= ~FOLD_FLAGS_LOCALE;
1795 }
1796
a1dde8de 1797 if (c < 256) {
a0270393 1798 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1799 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1800 /* It is illegal for the fold to cross the 255/256 boundary under
1801 * locale; in this case return the original */
1802 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1803 ? c
1804 : result;
a1dde8de
KW
1805 }
1806
a0270393
KW
1807 /* If no special needs, just use the macro */
1808 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1809 uvchr_to_utf8(p, c);
1810 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1811 }
1812 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1813 the special flags. */
1814 U8 utf8_c[UTF8_MAXBYTES + 1];
1815 uvchr_to_utf8(utf8_c, c);
445bf929 1816 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1817 }
84afefe6
JH
1818}
1819
26483009 1820PERL_STATIC_INLINE bool
5141f98e 1821S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1822 const char *const swashname, SV* const invlist)
bde6a22d 1823{
ea317ccb
KW
1824 /* returns a boolean giving whether or not the UTF8-encoded character that
1825 * starts at <p> is in the swash indicated by <swashname>. <swash>
1826 * contains a pointer to where the swash indicated by <swashname>
1827 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1828 * look at <*swash> and only generate a swash if it is not null. <invlist>
1829 * is NULL or an inversion list that defines the swash. If not null, it
1830 * saves time during initialization of the swash.
ea317ccb
KW
1831 *
1832 * Note that it is assumed that the buffer length of <p> is enough to
1833 * contain all the bytes that comprise the character. Thus, <*p> should
1834 * have been checked before this call for mal-formedness enough to assure
1835 * that. */
1836
97aff369 1837 dVAR;
7918f24d
NC
1838
1839 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1840
492a624f 1841 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1842 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1843 * as far as there being enough bytes available in it to accommodate the
1844 * character without reading beyond the end, and pass that number on to the
1845 * validating routine */
28123549
KW
1846 if (! is_utf8_char_buf(p, p + UTF8SKIP(p))) {
1847 if (ckWARN_d(WARN_UTF8)) {
1848 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1849 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1850 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1851 what the malformation is */
1852 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1853 }
1854 }
1855 return FALSE;
1856 }
87367d5f
KW
1857 if (!*swash) {
1858 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1859 *swash = _core_swash_init("utf8",
1860
1861 /* Only use the name if there is no inversion
1862 * list; otherwise will go out to disk */
1863 (invlist) ? "" : swashname,
1864
1865 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1866 }
28123549 1867
bde6a22d
NC
1868 return swash_fetch(*swash, p, TRUE) != 0;
1869}
1870
1871bool
922e8cb4
KW
1872Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1873{
1874 dVAR;
1875
1876 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1877
1878 assert(classnum < _FIRST_NON_SWASH_CC);
1879
f25ce844
KW
1880 return is_utf8_common(p,
1881 &PL_utf8_swash_ptrs[classnum],
1882 swash_property_names[classnum],
1883 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1884}
1885
1886bool
7fc63493 1887Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1888{
97aff369 1889 dVAR;
7918f24d
NC
1890
1891 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1892
61b19385 1893 return S_is_utf8_idfirst(aTHX_ p);
82686b01
JH
1894}
1895
1896bool
c11ff943
KW
1897Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1898{
1899 dVAR;
1900
1901 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1902
1903 if (*p == '_')
1904 return TRUE;
1905 /* is_utf8_idstart would be more logical. */
f25ce844 1906 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1907}
1908
1909bool
d65654cb 1910Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
b6912c02
KW
1911{
1912 dVAR;
b24b43f7 1913 SV* invlist = NULL;
b6912c02 1914
d65654cb 1915 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
b6912c02 1916
b24b43f7
KW
1917 if (! PL_utf8_perl_idstart) {
1918 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1919 }
1920 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
b6912c02
KW
1921}
1922
1923bool
eba68aa0
KW
1924Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1925{
1926 dVAR;
b24b43f7 1927 SV* invlist = NULL;
eba68aa0
KW
1928
1929 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1930
b24b43f7
KW
1931 if (! PL_utf8_perl_idcont) {
1932 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1933 }
1934 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
eba68aa0
KW
1935}
1936
1937
1938bool
7fc63493 1939Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1940{
97aff369 1941 dVAR;
7918f24d
NC
1942
1943 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1944
f25ce844 1945 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1946}
1947
1948bool
c11ff943
KW
1949Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1950{
1951 dVAR;
1952
1953 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1954
f25ce844 1955 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1956}
1957
1958bool
7dbf68d2
KW
1959Perl__is_utf8_mark(pTHX_ const U8 *p)
1960{
1961 dVAR;
1962
1963 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1964
f25ce844 1965 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1966}
1967
6b5c0936 1968/*
87cea99e 1969=for apidoc to_utf8_case
6b5c0936 1970
6fae5207 1971C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1972the character that is being converted. This routine assumes that the character
1973at C<p> is well-formed.
6b5c0936 1974
6fae5207
KW
1975C<ustrp> is a pointer to the character buffer to put the
1976conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1977of the result.
1978
6fae5207 1979C<swashp> is a pointer to the swash to use.
6b5c0936 1980
a1433954 1981Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1982and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1983but not always, a multicharacter mapping), is tried first.
6b5c0936 1984
4a8240a3
KW
1985C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1986any special mappings; C<""> means to use the special mappings. Values other
1987than these two are treated as the name of the hash containing the special
1988mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1989
6fae5207 1990C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1991%utf8::ToLower.
1992
1993=cut */
6b5c0936 1994
2104c8d9 1995UV
9a957fbc
AL
1996Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1997 SV **swashp, const char *normal, const char *special)
a0ed51b3 1998{
97aff369 1999 dVAR;
0134edef 2000 STRLEN len = 0;
f4cd282c 2001 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
2002
2003 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2004
9ae3ac1a
KW
2005 /* Note that swash_fetch() doesn't output warnings for these because it
2006 * assumes we will */
8457b38f 2007 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 2008 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
2009 if (ckWARN_d(WARN_SURROGATE)) {
2010 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2011 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2012 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
2013 }
9ae3ac1a
KW
2014 }
2015 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
2016 if (ckWARN_d(WARN_NON_UNICODE)) {
2017 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2018 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2019 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
2020 }
9ae3ac1a
KW
2021 }
2022
2023 /* Note that non-characters are perfectly legal, so no warning should
2024 * be given */
2025 }
2026
0134edef 2027 if (!*swashp) /* load on-demand */
5ab9d2ef 2028 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2029
a6f87d8c 2030 if (special) {
0134edef 2031 /* It might be "special" (sometimes, but not always,
2a37f04d 2032 * a multicharacter mapping) */
4a8240a3 2033 HV *hv = NULL;
b08cf34e
JH
2034 SV **svp;
2035
4a8240a3
KW
2036 /* If passed in the specials name, use that; otherwise use any
2037 * given in the swash */
2038 if (*special != '\0') {
2039 hv = get_hv(special, 0);
2040 }
2041 else {
2042 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2043 if (svp) {
2044 hv = MUTABLE_HV(SvRV(*svp));
2045 }
2046 }
2047
176fe009
KW
2048 if (hv
2049 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
2050 && (*svp))
2051 {
cfd0369c 2052 const char *s;
47654450 2053
cfd0369c 2054 s = SvPV_const(*svp, len);
47654450 2055 if (len == 1)
f4cd282c 2056 /* EIGHTBIT */
c80e42f3 2057 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2058 else {
d2dcd0fb 2059 Copy(s, ustrp, len, U8);
29e98929 2060 }
983ffd37 2061 }
0134edef
JH
2062 }
2063
2064 if (!len && *swashp) {
f4cd282c 2065 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 2066
0134edef
JH
2067 if (uv2) {
2068 /* It was "normal" (a single character mapping). */
f4cd282c 2069 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2070 }
2071 }
1feea2c7 2072
cbe07460
KW
2073 if (len) {
2074 if (lenp) {
2075 *lenp = len;
2076 }
2077 return valid_utf8_to_uvchr(ustrp, 0);
2078 }
2079
2080 /* Here, there was no mapping defined, which means that the code point maps
2081 * to itself. Return the inputs */
bfdf22ec 2082 len = UTF8SKIP(p);
ca9fab46
KW
2083 if (p != ustrp) { /* Don't copy onto itself */
2084 Copy(p, ustrp, len, U8);
2085 }
0134edef 2086
2a37f04d
JH
2087 if (lenp)
2088 *lenp = len;
2089
f4cd282c 2090 return uv1;
cbe07460 2091
a0ed51b3
LW
2092}
2093
051a06d4
KW
2094STATIC UV
2095S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2096{
2097 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
2098 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2099 * result contains a character that crosses the 255/256 boundary, disallow
2100 * the change, and return the original code point. See L<perlfunc/lc> for
2101 * why;
051a06d4 2102 *
a1433954
KW
2103 * p points to the original string whose case was changed; assumed
2104 * by this routine to be well-formed
051a06d4
KW
2105 * result the code point of the first character in the changed-case string
2106 * ustrp points to the changed-case string (<result> represents its first char)
2107 * lenp points to the length of <ustrp> */
2108
2109 UV original; /* To store the first code point of <p> */
2110
2111 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2112
a4f12ed7 2113 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2114
2115 /* We know immediately if the first character in the string crosses the
2116 * boundary, so can skip */
2117 if (result > 255) {
2118
2119 /* Look at every character in the result; if any cross the
2120 * boundary, the whole thing is disallowed */
2121 U8* s = ustrp + UTF8SKIP(ustrp);
2122 U8* e = ustrp + *lenp;
2123 while (s < e) {
a4f12ed7 2124 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2125 goto bad_crossing;
2126 }
2127 s += UTF8SKIP(s);
2128 }
2129
2130 /* Here, no characters crossed, result is ok as-is */
2131 return result;
2132 }
2133
2134bad_crossing:
2135
2136 /* Failed, have to return the original */
4b88fb76 2137 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2138 Copy(p, ustrp, *lenp, char);
2139 return original;
2140}
2141
d3e79532 2142/*
87cea99e 2143=for apidoc to_utf8_upper
d3e79532 2144
1f607577 2145Instead use L</toUPPER_utf8>.
a1433954 2146
d3e79532
JH
2147=cut */
2148
051a06d4 2149/* Not currently externally documented, and subject to change:
31f05a37
KW
2150 * <flags> is set iff iff the rules from the current underlying locale are to
2151 * be used. */
051a06d4 2152
2104c8d9 2153UV
31f05a37 2154Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2155{
97aff369 2156 dVAR;
7918f24d 2157
051a06d4
KW
2158 UV result;
2159
2160 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2161
31f05a37
KW
2162 if (flags && IN_UTF8_CTYPE_LOCALE) {
2163 flags = FALSE;
2164 }
2165
3a4c58c9 2166 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2167 if (flags) {
2168 result = toUPPER_LC(*p);
2169 }
2170 else {
81c6c7ce 2171 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2172 }
3a4c58c9
KW
2173 }
2174 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2175 if (flags) {
a6d8b88b 2176 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2177 result = toUPPER_LC(c);
051a06d4
KW
2178 }
2179 else {
94bb8c36 2180 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2181 ustrp, lenp, 'S');
051a06d4
KW
2182 }
2183 }
2184 else { /* utf8, ord above 255 */
2185 result = CALL_UPPER_CASE(p, ustrp, lenp);
2186
2187 if (flags) {
2188 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2189 }
2190 return result;
2191 }
2192
2193 /* Here, used locale rules. Convert back to utf8 */
2194 if (UTF8_IS_INVARIANT(result)) {
2195 *ustrp = (U8) result;
2196 *lenp = 1;
2197 }
2198 else {
62cb07ea
KW
2199 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2200 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2201 *lenp = 2;
3a4c58c9 2202 }
baa60164 2203
051a06d4 2204 return result;
983ffd37 2205}
a0ed51b3 2206
d3e79532 2207/*
87cea99e 2208=for apidoc to_utf8_title
d3e79532 2209
1f607577 2210Instead use L</toTITLE_utf8>.
a1433954 2211
d3e79532
JH
2212=cut */
2213
051a06d4 2214/* Not currently externally documented, and subject to change:
31f05a37
KW
2215 * <flags> is set iff the rules from the current underlying locale are to be
2216 * used. Since titlecase is not defined in POSIX, for other than a
2217 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2218 */
051a06d4 2219
983ffd37 2220UV
31f05a37 2221Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2222{
97aff369 2223 dVAR;
7918f24d 2224
051a06d4
KW
2225 UV result;
2226
2227 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2228
31f05a37
KW
2229 if (flags && IN_UTF8_CTYPE_LOCALE) {
2230 flags = FALSE;
2231 }
2232
3a4c58c9 2233 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2234 if (flags) {
2235 result = toUPPER_LC(*p);
2236 }
2237 else {
81c6c7ce 2238 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2239 }
3a4c58c9
KW
2240 }
2241 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2242 if (flags) {
a6d8b88b 2243 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2244 result = toUPPER_LC(c);
051a06d4
KW
2245 }
2246 else {
94bb8c36 2247 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2248 ustrp, lenp, 's');
051a06d4
KW
2249 }
2250 }
2251 else { /* utf8, ord above 255 */
2252 result = CALL_TITLE_CASE(p, ustrp, lenp);
2253
2254 if (flags) {
2255 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2256 }
2257 return result;
2258 }
2259
2260 /* Here, used locale rules. Convert back to utf8 */
2261 if (UTF8_IS_INVARIANT(result)) {
2262 *ustrp = (U8) result;
2263 *lenp = 1;
2264 }
2265 else {
62cb07ea
KW
2266 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2267 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2268 *lenp = 2;
3a4c58c9
KW
2269 }
2270
051a06d4 2271 return result;
a0ed51b3
LW
2272}
2273
d3e79532 2274/*
87cea99e 2275=for apidoc to_utf8_lower
d3e79532 2276
1f607577 2277Instead use L</toLOWER_utf8>.
a1433954 2278
d3e79532
JH
2279=cut */
2280
051a06d4 2281/* Not currently externally documented, and subject to change:
31f05a37
KW
2282 * <flags> is set iff iff the rules from the current underlying locale are to
2283 * be used.
2284 */
051a06d4 2285
2104c8d9 2286UV
31f05a37 2287Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2288{
051a06d4
KW
2289 UV result;
2290
97aff369 2291 dVAR;
7918f24d 2292
051a06d4 2293 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2294
31f05a37
KW
2295 if (flags && IN_UTF8_CTYPE_LOCALE) {
2296 flags = FALSE;
2297 }
2298
968c5e6a 2299 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2300 if (flags) {
2301 result = toLOWER_LC(*p);
2302 }
2303 else {
81c6c7ce 2304 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2305 }
968c5e6a
KW
2306 }
2307 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2308 if (flags) {
a6d8b88b 2309 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2310 result = toLOWER_LC(c);
051a06d4
KW
2311 }
2312 else {
94bb8c36 2313 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2314 ustrp, lenp);
051a06d4 2315 }
968c5e6a 2316 }
051a06d4
KW
2317 else { /* utf8, ord above 255 */
2318 result = CALL_LOWER_CASE(p, ustrp, lenp);
2319
2320 if (flags) {
2321 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2322 }
968c5e6a 2323
051a06d4
KW
2324 return result;
2325 }
2326
2327 /* Here, used locale rules. Convert back to utf8 */
2328 if (UTF8_IS_INVARIANT(result)) {
2329 *ustrp = (U8) result;
2330 *lenp = 1;
2331 }
2332 else {
62cb07ea
KW
2333 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2334 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2335 *lenp = 2;
2336 }
2337
051a06d4 2338 return result;
b4e400f9
JH
2339}
2340
d3e79532 2341/*
87cea99e 2342=for apidoc to_utf8_fold
d3e79532 2343
1f607577 2344Instead use L</toFOLD_utf8>.
a1433954 2345
d3e79532
JH
2346=cut */
2347
051a06d4
KW
2348/* Not currently externally documented, and subject to change,
2349 * in <flags>
31f05a37
KW
2350 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2351 * locale are to be used.
051a06d4
KW
2352 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2353 * otherwise simple folds
a0270393
KW
2354 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2355 * prohibited
445bf929 2356 */
36bb2ab6 2357
b4e400f9 2358UV
445bf929 2359Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2360{
97aff369 2361 dVAR;
7918f24d 2362
051a06d4
KW
2363 UV result;
2364
36bb2ab6 2365 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2366
a0270393
KW
2367 /* These are mutually exclusive */
2368 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2369
50ba90ff
KW
2370 assert(p != ustrp); /* Otherwise overwrites */
2371
31f05a37
KW
2372 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2373 flags &= ~FOLD_FLAGS_LOCALE;
2374 }
2375
a1dde8de 2376 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2377 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2378 result = toFOLD_LC(*p);
051a06d4
KW
2379 }
2380 else {
81c6c7ce 2381 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2382 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2383 }
a1dde8de
KW
2384 }
2385 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2386 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2387 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2388 result = toFOLD_LC(c);
051a06d4
KW
2389 }
2390 else {
94bb8c36 2391 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2392 ustrp, lenp,
1ca267a5 2393 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2394 }
a1dde8de 2395 }
051a06d4 2396 else { /* utf8, ord above 255 */
a0270393 2397 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2398
1ca267a5
KW
2399 if (flags & FOLD_FLAGS_LOCALE) {
2400
538e84ed
KW
2401 /* Special case these two characters, as what normally gets
2402 * returned under locale doesn't work */
1ca267a5
KW
2403 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2404 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2405 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2406 {
2407 goto return_long_s;
2408 }
9fc2026f
KW
2409 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2410 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2411 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2412 {
2413 goto return_ligature_st;
2414 }
a0270393 2415 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2416 }
a0270393
KW
2417 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2418 return result;
2419 }
2420 else {
2421 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2422 * character above the ASCII range, and the result should not
2423 * contain an ASCII character. */
a0270393
KW
2424
2425 UV original; /* To store the first code point of <p> */
2426
2427 /* Look at every character in the result; if any cross the
2428 * boundary, the whole thing is disallowed */
2429 U8* s = ustrp;
2430 U8* e = ustrp + *lenp;
2431 while (s < e) {
2432 if (isASCII(*s)) {
2433 /* Crossed, have to return the original */
2434 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2435
9fc2026f 2436 /* But in these instances, there is an alternative we can
1ca267a5 2437 * return that is valid */
9fc2026f
KW
2438 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2439 || original == LATIN_SMALL_LETTER_SHARP_S)
2440 {
1ca267a5
KW
2441 goto return_long_s;
2442 }
9fc2026f
KW
2443 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2444 goto return_ligature_st;
2445 }
a0270393
KW
2446 Copy(p, ustrp, *lenp, char);
2447 return original;
2448 }
2449 s += UTF8SKIP(s);
2450 }
051a06d4 2451
a0270393
KW
2452 /* Here, no characters crossed, result is ok as-is */
2453 return result;
2454 }
051a06d4
KW
2455 }
2456
2457 /* Here, used locale rules. Convert back to utf8 */
2458 if (UTF8_IS_INVARIANT(result)) {
2459 *ustrp = (U8) result;
2460 *lenp = 1;
2461 }
2462 else {
62cb07ea
KW
2463 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2464 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2465 *lenp = 2;
2466 }
2467
051a06d4 2468 return result;
1ca267a5
KW
2469
2470 return_long_s:
2471 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2472 * folds to a string of two of these characters. By returning this
2473 * instead, then, e.g.,
2474 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2475 * works. */
2476
2477 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2478 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2479 ustrp, *lenp, U8);
2480 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2481
2482 return_ligature_st:
2483 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2484 * have the other one fold to it */
2485
2486 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2487 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2488 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2489}
2490
711a919c 2491/* Note:
f90a9a02 2492 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2493 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2494 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2495 */
c4a5db0c 2496
a0ed51b3 2497SV*
7fc63493 2498Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2499{
c4a5db0c
KW
2500 PERL_ARGS_ASSERT_SWASH_INIT;
2501
2502 /* Returns a copy of a swash initiated by the called function. This is the
2503 * public interface, and returning a copy prevents others from doing
2504 * mischief on the original */
2505
5d3d13d1 2506 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2507}
2508
2509SV*
5d3d13d1 2510Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2511{
2512 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2513 * by calling utf8_heavy.pl in the general case. The returned value may be
2514 * the swash's inversion list instead if the input parameters allow it.
2515 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2516 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2517 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2518 *
2519 * This interface should only be used by functions that won't destroy or
2520 * adversely change the swash, as doing so affects all other uses of the
2521 * swash in the program; the general public should use 'Perl_swash_init'
2522 * instead.
2523 *
2524 * pkg is the name of the package that <name> should be in.
2525 * name is the name of the swash to find. Typically it is a Unicode
2526 * property name, including user-defined ones
2527 * listsv is a string to initialize the swash with. It must be of the form
2528 * documented as the subroutine return value in
2529 * L<perlunicode/User-Defined Character Properties>
2530 * minbits is the number of bits required to represent each data element.
2531 * It is '1' for binary properties.
2532 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2533 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2534 * flags_p if non-NULL is the address of various input and output flag bits
2535 * to the routine, as follows: ('I' means is input to the routine;
2536 * 'O' means output from the routine. Only flags marked O are
2537 * meaningful on return.)
2538 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2539 * came from a user-defined property. (I O)
5d3d13d1
KW
2540 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2541 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2542 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2543 * return of an inversion list instead of a swash hash if this routine
2544 * thinks that would result in faster execution of swash_fetch() later
2545 * on. (I)
9a53f6cf
KW
2546 *
2547 * Thus there are three possible inputs to find the swash: <name>,
2548 * <listsv>, and <invlist>. At least one must be specified. The result
2549 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2550 * actions can intersect, etc. what <name> gives. To avoid going out to
2551 * disk at all, <invlist> should specify completely what the swash should
2552 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2553 *
2554 * <invlist> is only valid for binary properties */
c4a5db0c 2555
27da23d5 2556 dVAR;
c4a5db0c 2557 SV* retval = &PL_sv_undef;
83199d38 2558 HV* swash_hv = NULL;
87367d5f
KW
2559 const int invlist_swash_boundary =
2560 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2561 ? 512 /* Based on some benchmarking, but not extensive, see commit
2562 message */
2563 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2564
2565 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2566 assert(! invlist || minbits == 1);
2567
2568 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2569 * so */
2570 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2571 dSP;
2572 const size_t pkg_len = strlen(pkg);
2573 const size_t name_len = strlen(name);
2574 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2575 SV* errsv_save;
2576 GV *method;
2577
2578 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2579
2580 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2581 ENTER;
69794297
KW
2582 SAVEHINTS();
2583 save_re_context();
650f067c
JL
2584 /* We might get here via a subroutine signature which uses a utf8
2585 * parameter name, at which point PL_subname will have been set
2586 * but not yet used. */
2587 save_item(PL_subname);
69794297
KW
2588 if (PL_parser && PL_parser->error_count)
2589 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2590 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2591 if (!method) { /* demand load utf8 */
2592 ENTER;
db2c6cb3
FC
2593 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2594 GvSV(PL_errgv) = NULL;
69794297
KW
2595 /* It is assumed that callers of this routine are not passing in
2596 * any user derived data. */
2597 /* Need to do this after save_re_context() as it will set
2598 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2599 * in Perl_magic_get). Even line to create errsv_save can turn on
2600 * PL_tainted. */
284167a5
SM
2601#ifndef NO_TAINT_SUPPORT
2602 SAVEBOOL(TAINT_get);
2603 TAINT_NOT;
2604#endif
69794297
KW
2605 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2606 NULL);
eed484f9 2607 {
db2c6cb3
FC
2608 /* Not ERRSV, as there is no need to vivify a scalar we are
2609 about to discard. */
2610 SV * const errsv = GvSV(PL_errgv);
2611 if (!SvTRUE(errsv)) {
2612 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2613 SvREFCNT_dec(errsv);
2614 }
eed484f9 2615 }
69794297
KW
2616 LEAVE;
2617 }
2618 SPAGAIN;
2619 PUSHMARK(SP);
2620 EXTEND(SP,5);
2621 mPUSHp(pkg, pkg_len);
2622 mPUSHp(name, name_len);
2623 PUSHs(listsv);
2624 mPUSHi(minbits);
2625 mPUSHi(none);
2626 PUTBACK;
db2c6cb3
FC
2627 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2628 GvSV(PL_errgv) = NULL;
69794297
KW
2629 /* If we already have a pointer to the method, no need to use
2630 * call_method() to repeat the lookup. */
c41800a8
KW
2631 if (method
2632 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2633 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2634 {
2635 retval = *PL_stack_sp--;
2636 SvREFCNT_inc(retval);
2637 }
eed484f9 2638 {
db2c6cb3
FC
2639 /* Not ERRSV. See above. */
2640 SV * const errsv = GvSV(PL_errgv);
2641 if (!SvTRUE(errsv)) {
2642 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2643 SvREFCNT_dec(errsv);
2644 }
eed484f9 2645 }
ce3b816e 2646 LEAVE;
69794297
KW
2647 POPSTACK;
2648 if (IN_PERL_COMPILETIME) {
2649 CopHINTS_set(PL_curcop, PL_hints);
2650 }
2651 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2652 if (SvPOK(retval))
2653
2654 /* If caller wants to handle missing properties, let them */
5d3d13d1 2655 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2656 return NULL;
2657 }
2658 Perl_croak(aTHX_
2659 "Can't find Unicode property definition \"%"SVf"\"",
2660 SVfARG(retval));
2661 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2662 }
9a53f6cf 2663 } /* End of calling the module to find the swash */
36eb48b4 2664
83199d38
KW
2665 /* If this operation fetched a swash, and we will need it later, get it */
2666 if (retval != &PL_sv_undef
2667 && (minbits == 1 || (flags_p
2668 && ! (*flags_p
2669 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2670 {
2671 swash_hv = MUTABLE_HV(SvRV(retval));
2672
2673 /* If we don't already know that there is a user-defined component to
2674 * this swash, and the user has indicated they wish to know if there is
2675 * one (by passing <flags_p>), find out */
2676 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2677 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2678 if (user_defined && SvUV(*user_defined)) {
2679 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2680 }
2681 }
2682 }
2683
36eb48b4
KW
2684 /* Make sure there is an inversion list for binary properties */
2685 if (minbits == 1) {
2686 SV** swash_invlistsvp = NULL;
2687 SV* swash_invlist = NULL;
9a53f6cf 2688 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2689 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2690 an unclaimed reference count */
36eb48b4 2691
9a53f6cf 2692 /* If this operation fetched a swash, get its already existing
83199d38 2693 * inversion list, or create one for it */
36eb48b4 2694
83199d38 2695 if (swash_hv) {
5c9f4bd2 2696 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2697 if (swash_invlistsvp) {
2698 swash_invlist = *swash_invlistsvp;
2699 invlist_in_swash_is_valid = TRUE;
2700 }
2701 else {
36eb48b4 2702 swash_invlist = _swash_to_invlist(retval);
02c85471 2703 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2704 }
2705 }
2706
2707 /* If an inversion list was passed in, have to include it */
2708 if (invlist) {
2709
2710 /* Any fetched swash will by now have an inversion list in it;
2711 * otherwise <swash_invlist> will be NULL, indicating that we
2712 * didn't fetch a swash */
2713 if (swash_invlist) {
2714
2715 /* Add the passed-in inversion list, which invalidates the one
2716 * already stored in the swash */
2717 invlist_in_swash_is_valid = FALSE;
2718 _invlist_union(invlist, swash_invlist, &swash_invlist);
2719 }
2720 else {
2721
87367d5f
KW
2722 /* Here, there is no swash already. Set up a minimal one, if
2723 * we are going to return a swash */
2724 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2725 swash_hv = newHV();
4aca0fe6 2726 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2727 }
9a53f6cf
KW
2728 swash_invlist = invlist;
2729 }
9a53f6cf
KW
2730 }
2731
2732 /* Here, we have computed the union of all the passed-in data. It may
2733 * be that there was an inversion list in the swash which didn't get
538e84ed 2734 * touched; otherwise save the computed one */
87367d5f
KW
2735 if (! invlist_in_swash_is_valid
2736 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2737 {
5c9f4bd2 2738 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2739 {
2740 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2741 }
cc34d8c5
FC
2742 /* We just stole a reference count. */
2743 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2744 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2745 }
87367d5f 2746
dbfdbd26
KW
2747 SvREADONLY_on(swash_invlist);
2748
c41800a8 2749 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2750 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2751 SvREFCNT_dec(retval);
02c85471
FC
2752 if (!swash_invlist_unclaimed)
2753 SvREFCNT_inc_simple_void_NN(swash_invlist);
2754 retval = newRV_noinc(swash_invlist);
87367d5f 2755 }
36eb48b4
KW
2756 }
2757
a0ed51b3
LW
2758 return retval;
2759}
2760
035d37be
JH
2761
2762/* This API is wrong for special case conversions since we may need to
2763 * return several Unicode characters for a single Unicode character
2764 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2765 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2766 * multiple values. --jhi
2767 * For those, you should use to_utf8_case() instead */
b0e3252e 2768/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2769
2770/* Note:
2771 * Returns the value of property/mapping C<swash> for the first character
2772 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2773 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2774 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2775 *
2776 * A "swash" is a hash which contains initially the keys/values set up by
2777 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2778 * property for all possible code points. Things are stored in a compact form
2779 * (see utf8_heavy.pl) so that calculation is required to find the actual
2780 * property value for a given code point. As code points are looked up, new
2781 * key/value pairs are added to the hash, so that the calculation doesn't have
2782 * to ever be re-done. Further, each calculation is done, not just for the
2783 * desired one, but for a whole block of code points adjacent to that one.
2784 * For binary properties on ASCII machines, the block is usually for 64 code
2785 * points, starting with a code point evenly divisible by 64. Thus if the
2786 * property value for code point 257 is requested, the code goes out and
2787 * calculates the property values for all 64 code points between 256 and 319,
2788 * and stores these as a single 64-bit long bit vector, called a "swatch",
2789 * under the key for code point 256. The key is the UTF-8 encoding for code
2790 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2791 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2792 * for code point 258 is then requested, this code realizes that it would be
2793 * stored under the key for 256, and would find that value and extract the
2794 * relevant bit, offset from 256.
2795 *
2796 * Non-binary properties are stored in as many bits as necessary to represent
2797 * their values (32 currently, though the code is more general than that), not
2798 * as single bits, but the principal is the same: the value for each key is a
2799 * vector that encompasses the property values for all code points whose UTF-8
2800 * representations are represented by the key. That is, for all code points
2801 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2802 * bytes of that.
680c470c 2803 */
a0ed51b3 2804UV
680c470c 2805Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2806{
27da23d5 2807 dVAR;
ef8f7699 2808 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2809 U32 klen;
2810 U32 off;
9b56a019 2811 STRLEN slen = 0;
7d85a32c 2812 STRLEN needents;
cfd0369c 2813 const U8 *tmps = NULL;
a0ed51b3 2814 U32 bit;
979f2922 2815 SV *swatch;
08fb1ac5 2816 const U8 c = *ptr;
3568d838 2817
7918f24d
NC
2818 PERL_ARGS_ASSERT_SWASH_FETCH;
2819
87367d5f
KW
2820 /* If it really isn't a hash, it isn't really swash; must be an inversion
2821 * list */
2822 if (SvTYPE(hv) != SVt_PVHV) {
2823 return _invlist_contains_cp((SV*)hv,
2824 (do_utf8)
2825 ? valid_utf8_to_uvchr(ptr, NULL)
2826 : c);
2827 }
2828
08fb1ac5
KW
2829 /* We store the values in a "swatch" which is a vec() value in a swash
2830 * hash. Code points 0-255 are a single vec() stored with key length
2831 * (klen) 0. All other code points have a UTF-8 representation
2832 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2833 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2834 * length for them is the length of the encoded char - 1. ptr[klen] is the
2835 * final byte in the sequence representing the character */
2836 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2837 klen = 0;
2838 needents = 256;
2839 off = c;
3568d838 2840 }
08fb1ac5
KW
2841 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2842 klen = 0;
2843 needents = 256;
2844 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2845 }
2846 else {
08fb1ac5
KW
2847 klen = UTF8SKIP(ptr) - 1;
2848
2849 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2850 * the vec is the final byte in the sequence. (In EBCDIC this is
2851 * converted to I8 to get consecutive values.) To help you visualize
2852 * all this:
2853 * Straight 1047 After final byte
2854 * UTF-8 UTF-EBCDIC I8 transform
2855 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2856 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2857 * ...
2858 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2859 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2860 * ...
2861 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2862 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2863 * ...
2864 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2865 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2866 * ...
2867 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2868 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2869 *
2870 * (There are no discontinuities in the elided (...) entries.)
2871 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2872 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2873 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2874 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2875 * index into the vec() swatch (after subtracting 0x80, which we
2876 * actually do with an '&').
2877 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2878 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2879 * dicontinuities which go away by transforming it into I8, and we
2880 * effectively subtract 0xA0 to get the index. */
979f2922 2881 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2882 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2883 }
7d85a32c 2884
a0ed51b3
LW
2885 /*
2886 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2887 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2888 * it's nothing to sniff at.) Pity we usually come through at least
2889 * two function calls to get here...
2890 *
2891 * NB: this code assumes that swatches are never modified, once generated!
2892 */
2893
3568d838 2894 if (hv == PL_last_swash_hv &&
a0ed51b3 2895 klen == PL_last_swash_klen &&
27da23d5 2896 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2897 {
2898 tmps = PL_last_swash_tmps;
2899 slen = PL_last_swash_slen;
2900 }
2901 else {
2902 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2903 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2904
b0e3252e 2905 /* If not cached, generate it via swatch_get */
979f2922 2906 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2907 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2908 {
2909 if (klen) {
2910 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2911 swatch = swatch_get(swash,
2912 code_point & ~((UV)needents - 1),
2913 needents);
2914 }
2915 else { /* For the first 256 code points, the swatch has a key of
2916 length 0 */
2917 swatch = swatch_get(swash, 0, needents);
2918 }
979f2922 2919
923e4eb5 2920 if (IN_PERL_COMPILETIME)
623e6609 2921 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2922
979f2922 2923 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2924
979f2922
ST
2925 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2926 || (slen << 3) < needents)
5637ef5b
NC
2927 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2928 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2929 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2930 }
2931
2932 PL_last_swash_hv = hv;
16d8f38a 2933 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2934 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2935 /* FIXME change interpvar.h? */
2936 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2937 PL_last_swash_slen = slen;
2938 if (klen)
2939 Copy(ptr, PL_last_swash_key, klen, U8);
2940 }
2941
9faf8d75 2942 switch ((int)((slen << 3) / needents)) {
a0ed51b3
LW
2943 case 1:
2944 bit = 1 << (off & 7);
2945 off >>= 3;
2946 return (tmps[off] & bit) != 0;
2947 case 8:
2948 return tmps[off];
2949 case 16:
2950 off <<= 1;
2951 return (tmps[off] << 8) + tmps[off + 1] ;
2952 case 32:
2953 off <<= 2;
2954 return (tmps[off] << 24) + (tmps[off+1] << 16) + (tmps[off+2] << 8) + tmps[off + 3] ;
2955 }
5637ef5b
NC
2956 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2957 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2958 NORETURN_FUNCTION_END;
a0ed51b3 2959}
2b9d42f0 2960
319009ee
KW
2961/* Read a single line of the main body of the swash input text. These are of
2962 * the form:
2963 * 0053 0056 0073
2964 * where each number is hex. The first two numbers form the minimum and
2965 * maximum of a range, and the third is the value associated with the range.
2966 * Not all swashes should have a third number
2967 *
2968 * On input: l points to the beginning of the line to be examined; it points
2969 * to somewhere in the string of the whole input text, and is
2970 * terminated by a \n or the null string terminator.
2971 * lend points to the null terminator of that string
2972 * wants_value is non-zero if the swash expects a third number
2973 * typestr is the name of the swash's mapping, like 'ToLower'
2974 * On output: *min, *max, and *val are set to the values read from the line.
2975 * returns a pointer just beyond the line examined. If there was no
2976 * valid min number on the line, returns lend+1
2977 */
2978
2979STATIC U8*
2980S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2981 const bool wants_value, const U8* const typestr)
2982{
2983 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2984 STRLEN numlen; /* Length of the number */
02470786
KW
2985 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2986 | PERL_SCAN_DISALLOW_PREFIX
2987 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2988
2989 /* nl points to the next \n in the scan */
2990 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2991
2992 /* Get the first number on the line: the range minimum */
2993 numlen = lend - l;
2994 *min = grok_hex((char *)l, &numlen, &flags, NULL);
2995 if (numlen) /* If found a hex number, position past it */
2996 l += numlen;
2997 else if (nl) { /* Else, go handle next line, if any */
2998 return nl + 1; /* 1 is length of "\n" */
2999 }
3000 else { /* Else, no next line */
3001 return lend + 1; /* to LIST's end at which \n is not found */
3002 }
3003
3004 /* The max range value follows, separated by a BLANK */
3005 if (isBLANK(*l)) {
3006 ++l;
02470786
KW
3007 flags = PERL_SCAN_SILENT_ILLDIGIT
3008 | PERL_SCAN_DISALLOW_PREFIX
3009 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3010 numlen = lend - l;
3011 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3012 if (numlen)
3013 l += numlen;
3014 else /* If no value here, it is a single element range */
3015 *max = *min;
3016
3017 /* Non-binary tables have a third entry: what the first element of the
24303724 3018 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3019 if (wants_value) {
3020 if (isBLANK(*l)) {
3021 ++l;
f2a7d0fc
KW
3022 flags = PERL_SCAN_SILENT_ILLDIGIT
3023 | PERL_SCAN_DISALLOW_PREFIX
3024 | PERL_SCAN_SILENT_NON_PORTABLE;
3025 numlen = lend - l;
3026 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3027 if (numlen)
3028 l += numlen;
3029 else
3030 *val = 0;
319009ee
KW
3031 }
3032 else {
3033 *val = 0;
3034 if (typeto) {
dcbac5bb 3035 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3036 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3037 typestr, l);
3038 }
3039 }
3040 }
3041 else
3042 *val = 0; /* bits == 1, then any val should be ignored */
3043 }
3044 else { /* Nothing following range min, should be single element with no
3045 mapping expected */
3046 *max = *min;
3047 if (wants_value) {
3048 *val = 0;
3049 if (typeto) {
dcbac5bb 3050 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3051 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3052 }
3053 }
3054 else
3055 *val = 0; /* bits == 1, then val should be ignored */
3056 }
3057
3058 /* Position to next line if any, or EOF */
3059 if (nl)
3060 l = nl + 1;
3061 else
3062 l = lend;
3063
3064 return l;
3065}
3066
979f2922
ST
3067/* Note:
3068 * Returns a swatch (a bit vector string) for a code point sequence
3069 * that starts from the value C<start> and comprises the number C<span>.
3070 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3071 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3072 */
3073STATIC SV*
b0e3252e 3074S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3075{
3076 SV *swatch;
77f9f126 3077 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3078 STRLEN lcur, xcur, scur;
ef8f7699 3079 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3080 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3081
88d45d28
KW
3082 SV** listsvp = NULL; /* The string containing the main body of the table */
3083 SV** extssvp = NULL;
3084 SV** invert_it_svp = NULL;
3085 U8* typestr = NULL;
786861f5
KW
3086 STRLEN bits;
3087 STRLEN octets; /* if bits == 1, then octets == 0 */
3088 UV none;
3089 UV end = start + span;
972dd592 3090
36eb48b4 3091 if (invlistsvp == NULL) {
786861f5
KW
3092 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3093 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3094 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3095 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3096 listsvp = hv_fetchs(hv, "LIST", FALSE);
3097 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3098
3099 bits = SvUV(*bitssvp);
3100 none = SvUV(*nonesvp);
3101 typestr = (U8*)SvPV_nolen(*typesvp);
3102 }
36eb48b4
KW
3103 else {
3104 bits = 1;
3105 none = 0;
3106 }
786861f5 3107 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3108
b0e3252e 3109 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3110
979f2922 3111 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3112 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3113 (UV)bits);
979f2922
ST
3114 }
3115
84ea5ef6
KW
3116 /* If overflowed, use the max possible */
3117 if (end < start) {
3118 end = UV_MAX;
3119 span = end - start;
3120 }
3121
979f2922 3122 /* create and initialize $swatch */
979f2922 3123 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3124 swatch = newSV(scur);
3125 SvPOK_on(swatch);
979f2922
ST
3126 s = (U8*)SvPVX(swatch);
3127 if (octets && none) {
0bd48802 3128 const U8* const e = s + scur;
979f2922
ST
3129 while (s < e) {
3130 if (bits == 8)
3131 *s++ = (U8)(none & 0xff);
3132 else if (bits == 16) {
3133 *s++ = (U8)((none >> 8) & 0xff);
3134 *s++ = (U8)( none & 0xff);
3135 }
3136 else if (bits == 32) {
3137 *s++ = (U8)((none >> 24) & 0xff);
3138 *s++ = (U8)((none >> 16) & 0xff);
3139 *s++ = (U8)((none >> 8) & 0xff);
3140 *s++ = (U8)( none & 0xff);
3141 }
3142 }
3143 *s = '\0';
3144 }
3145 else {
3146 (void)memzero((U8*)s, scur + 1);
3147 }
3148 SvCUR_set(swatch, scur);
3149 s = (U8*)SvPVX(swatch);
3150
36eb48b4
KW
3151 if (invlistsvp) { /* If has an inversion list set up use that */
3152 _invlist_populate_swatch(*invlistsvp, start, end, s);
3153 return swatch;
3154 }
3155
3156 /* read $swash->{LIST} */
979f2922
ST
3157 l = (U8*)SvPV(*listsvp, lcur);
3158 lend = l + lcur;
3159 while (l < lend) {
8ed25d53 3160 UV min, max, val, upper;
319009ee
KW
3161 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3162 cBOOL(octets), typestr);
3163 if (l > lend) {
979f2922
ST
3164 break;
3165 }
3166
972dd592 3167 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3168 if (max < start)
3169 continue;
3170
8ed25d53
KW
3171 /* <end> is generally 1 beyond where we want to set things, but at the
3172 * platform's infinity, where we can't go any higher, we want to
3173 * include the code point at <end> */
3174 upper = (max < end)
3175 ? max
3176 : (max != UV_MAX || end != UV_MAX)
3177 ? end - 1
3178 : end;
3179
979f2922 3180 if (octets) {
35da51f7 3181 UV key;
979f2922
ST
3182 if (min < start) {
3183 if (!none || val < none) {
3184 val += start - min;
3185 }
3186 min = start;
3187 }
8ed25d53 3188 for (key = min; key <= upper; key++) {
979f2922 3189 STRLEN offset;
979f2922
ST
3190 /* offset must be non-negative (start <= min <= key < end) */
3191 offset = octets * (key - start);
3192 if (bits == 8)
3193 s[offset] = (U8)(val & 0xff);
3194 else if (bits == 16) {
3195 s[offset ] = (U8)((val >> 8) & 0xff);
3196 s[offset + 1] = (U8)( val & 0xff);
3197 }
3198 else if (bits == 32) {
3199 s[offset ] = (U8)((val >> 24) & 0xff);
3200 s[offset + 1] = (U8)((val >> 16) & 0xff);
3201 s[offset + 2] = (U8)((val >> 8) & 0xff);
3202 s[offset + 3] = (U8)( val & 0xff);
3203 }
3204
3205 if (!none || val < none)
3206 ++val;
3207 }
3208 }
711a919c 3209 else { /* bits == 1, then val should be ignored */
35da51f7 3210 UV key;
979f2922
ST
3211 if (min < start)
3212 min = start;
6cb05c12 3213
8ed25d53 3214 for (key = min; key <= upper; key++) {
0bd48802 3215 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3216 s[offset >> 3] |= 1 << (offset & 7);
3217 }
3218 }
3219 } /* while */
979f2922 3220
9479a769 3221 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3222 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3223
3224 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3225 * be 0, and their inversion should also be 0, as we don't succeed any
3226 * Unicode property matches for non-Unicode code points */
3227 if (start <= PERL_UNICODE_MAX) {
3228
3229 /* The code below assumes that we never cross the
3230 * Unicode/above-Unicode boundary in a range, as otherwise we would
3231 * have to figure out where to stop flipping the bits. Since this
3232 * boundary is divisible by a large power of 2, and swatches comes
3233 * in small powers of 2, this should be a valid assumption */
3234 assert(start + span - 1 <= PERL_UNICODE_MAX);
3235
507a8485
KW
3236 send = s + scur;
3237 while (s < send) {
3238 *s = ~(*s);
3239 s++;
3240 }
0bda3001 3241 }
77f9f126
KW
3242 }
3243
d73c39c5
KW
3244 /* read $swash->{EXTRAS}
3245 * This code also copied to swash_to_invlist() below */
979f2922
ST
3246 x = (U8*)SvPV(*extssvp, xcur);
3247 xend = x + xcur;
3248 while (x < xend) {
3249 STRLEN namelen;
3250 U8 *namestr;
3251 SV** othersvp;
3252 HV* otherhv;
3253 STRLEN otherbits;
3254 SV **otherbitssvp, *other;
711a919c 3255 U8 *s, *o, *nl;
979f2922
ST
3256 STRLEN slen, olen;
3257
35da51f7 3258 const U8 opc = *x++;
979f2922
ST
3259 if (opc == '\n')
3260 continue;
3261
3262 nl = (U8*)memchr(x, '\n', xend - x);
3263
3264 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3265 if (nl) {
3266 x = nl + 1; /* 1 is length of "\n" */
3267 continue;
3268 }
3269 else {
3270 x = xend; /* to EXTRAS' end at which \n is not found */
3271 break;
3272 }
3273 }
3274
3275 namestr = x;
3276 if (nl) {
3277 namelen = nl - namestr;
3278 x = nl + 1;
3279 }
3280 else {
3281 namelen = xend - namestr;
3282 x = xend;
3283 }
3284
3285 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3286 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3287 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3288 otherbits = (STRLEN)SvUV(*otherbitssvp);
3289 if (bits < otherbits)
5637ef5b
NC
3290 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3291 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3292
3293 /* The "other" swatch must be destroyed after. */
b0e3252e 3294 other = swatch_get(*othersvp, start, span);
979f2922
ST
3295 o = (U8*)SvPV(other, olen);
3296
3297 if (!olen)
b0e3252e 3298 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3299
3300 s = (U8*)SvPV(swatch, slen);
3301 if (bits == 1 && otherbits == 1) {
3302 if (slen != olen)
5637ef5b
NC
3303 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3304 "mismatch, slen=%"UVuf", olen=%"UVuf,
3305 (UV)slen, (UV)olen);
979f2922
ST
3306
3307 switch (opc) {
3308 case '+':
3309 while (slen--)
3310 *s++ |= *o++;
3311 break;
3312 case '!':
3313 while (slen--)
3314 *s++ |= ~*o++;
3315 break;
3316 case '-':
3317 while (slen--)
3318 *s++ &= ~*o++;
3319 break;
3320 case '&':
3321 while (slen--)
3322 *s++ &= *o++;
3323 break;
3324 default:
3325 break;
3326 }
3327 }
711a919c 3328 else {
979f2922
ST
3329 STRLEN otheroctets = otherbits >> 3;
3330 STRLEN offset = 0;
35da51f7 3331 U8* const send = s + slen;
979f2922
ST
3332
3333 while (s < send) {
3334 UV otherval = 0;
3335
3336 if (otherbits == 1) {
3337 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3338 ++offset;
3339 }
3340 else {
3341 STRLEN vlen = otheroctets;
3342 otherval = *o++;
3343 while (--vlen) {
3344 otherval <<= 8;
3345 otherval |= *o++;
3346 }
3347 }
3348
711a919c 3349 if (opc == '+' && otherval)
6f207bd3 3350 NOOP; /* replace with otherval */
979f2922
ST
3351 else if (opc == '!' && !otherval)
3352 otherval = 1;
3353 else if (opc == '-' && otherval)
3354 otherval = 0;
3355 else if (opc == '&' && !otherval)
3356 otherval = 0;
3357 else {
711a919c 3358 s += octets; /* no replacement */
979f2922
ST
3359 continue;
3360 }
3361
3362 if (bits == 8)
3363 *s++ = (U8)( otherval & 0xff);
3364 else if (bits == 16) {
3365 *s++ = (U8)((otherval >> 8) & 0xff);
3366 *s++ = (U8)( otherval & 0xff);
3367 }
3368 else if (bits == 32) {
3369 *s++ = (U8)((otherval >> 24) & 0xff);
3370 *s++ = (U8)((otherval >> 16) & 0xff);
3371 *s++ = (U8)((otherval >> 8) & 0xff);
3372 *s++ = (U8)( otherval & 0xff);
3373 }
3374 }
3375 }
3376 sv_free(other); /* through with it! */
3377 } /* while */
3378 return swatch;
3379}
3380
064c021d 3381HV*
4c2e1131 3382Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3383{
3384
79a2a0e8 3385 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3386 * Can't be used on a property that is subject to user override, as it
3387 * relies on the value of SPECIALS in the swash which would be set by
3388 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3389 * for overridden properties
064c021d
KW
3390 *
3391 * Returns a hash which is the inversion and closure of a swash mapping.
3392 * For example, consider the input lines:
3393 * 004B 006B
3394 * 004C 006C
3395 * 212A 006B
3396 *
3397 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3398 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3399 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3400 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3401 *
538e84ed
KW
3402 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3403 * keys are only code points that are folded-to, so it isn't a full closure.
3404 *
064c021d
KW
3405 * Essentially, for any code point, it gives all the code points that map to
3406 * it, or the list of 'froms' for that point.
3407 *
5662e334
KW
3408 * Currently it ignores any additions or deletions from other swashes,
3409 * looking at just the main body of the swash, and if there are SPECIALS
3410 * in the swash, at that hash
3411 *
3412 * The specials hash can be extra code points, and most likely consists of
3413 * maps from single code points to multiple ones (each expressed as a string
3414 * of utf8 characters). This function currently returns only 1-1 mappings.
3415 * However consider this possible input in the specials hash:
3416 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3417 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3418 *
3419 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3420 * currently handle. But it also means that FB05 and FB06 are equivalent in
3421 * a 1-1 mapping which we should handle, and this relationship may not be in
3422 * the main table. Therefore this function examines all the multi-char
3423 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3424
3425 U8 *l, *lend;
3426 STRLEN lcur;
3427 HV *const hv = MUTABLE_HV(SvRV(swash));
3428
923b6d4e
KW
3429 /* The string containing the main body of the table. This will have its
3430 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3431 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3432
3433 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3434 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3435 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3436 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3437 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3438 const STRLEN bits = SvUV(*bitssvp);
3439 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3440 const UV none = SvUV(*nonesvp);
5662e334 3441 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3442
3443 HV* ret = newHV();
3444
3445 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3446
3447 /* Must have at least 8 bits to get the mappings */
3448 if (bits != 8 && bits != 16 && bits != 32) {
3449 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3450 (UV)bits);
3451 }
3452
5662e334
KW
3453 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3454 mapping to more than one character */
3455
3456 /* Construct an inverse mapping hash for the specials */
3457 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3458 HV * specials_inverse = newHV();
3459 char *char_from; /* the lhs of the map */
3460 I32 from_len; /* its byte length */
3461 char *char_to; /* the rhs of the map */
3462 I32 to_len; /* its byte length */
3463 SV *sv_to; /* and in a sv */
3464 AV* from_list; /* list of things that map to each 'to' */
3465
3466 hv_iterinit(specials_hv);
3467
3468 /* The keys are the characters (in utf8) that map to the corresponding
3469 * utf8 string value. Iterate through the list creating the inverse
3470 * list. */
3471 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3472 SV** listp;
3473 if (! SvPOK(sv_to)) {
5637ef5b
NC
3474 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3475 "unexpectedly is not a string, flags=%lu",
3476 (unsigned long)SvFLAGS(sv_to));
5662e334 3477 }
4b88fb76 3478 /*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
3479
3480 /* Each key in the inverse list is a mapped-to value, and the key's
3481 * hash value is a list of the strings (each in utf8) that map to
3482 * it. Those strings are all one character long */
3483 if ((listp = hv_fetch(specials_inverse,
3484 SvPVX(sv_to),
3485 SvCUR(sv_to), 0)))
3486 {
3487 from_list = (AV*) *listp;
3488 }
3489 else { /* No entry yet for it: create one */
3490 from_list = newAV();
3491 if (! hv_store(specials_inverse,
3492 SvPVX(sv_to),
3493 SvCUR(sv_to),
3494 (SV*) from_list, 0))
3495 {
3496 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3497 }
3498 }
3499
3500 /* Here have the list associated with this 'to' (perhaps newly
3501 * created and empty). Just add to it. Note that we ASSUME that
3502 * the input is guaranteed to not have duplications, so we don't
3503 * check for that. Duplications just slow down execution time. */
3504 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3505 }
3506
3507 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3508 * it looking for cases like the FB05/FB06 examples above. There would
3509 * be an entry in the hash like
3510 * 'st' => [ FB05, FB06 ]
3511 * In this example we will create two lists that get stored in the
3512 * returned hash, 'ret':
3513 * FB05 => [ FB05, FB06 ]
3514 * FB06 => [ FB05, FB06 ]
3515 *
3516 * Note that there is nothing to do if the array only has one element.
3517 * (In the normal 1-1 case handled below, we don't have to worry about
3518 * two lists, as everything gets tied to the single list that is
3519 * generated for the single character 'to'. But here, we are omitting
3520 * that list, ('st' in the example), so must have multiple lists.) */
3521 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3522 &char_to, &to_len)))
3523 {
b9f2b683 3524 if (av_tindex(from_list) > 0) {
c70927a6 3525 SSize_t i;
5662e334
KW
3526
3527 /* We iterate over all combinations of i,j to place each code
3528 * point on each list */
b9f2b683 3529 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3530 SSize_t j;
5662e334
KW
3531 AV* i_list = newAV();
3532 SV** entryp = av_fetch(from_list, i, FALSE);
3533 if (entryp == NULL) {
3534 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3535 }
3536 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3537 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3538 }
3539 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3540 (SV*) i_list, FALSE))
3541 {
3542 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3543 }
3544
538e84ed 3545 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3546 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3547 entryp = av_fetch(from_list, j, FALSE);
3548 if (entryp == NULL) {
3549 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3550 }
3551
3552 /* When i==j this adds itself to the list */
4b88fb76
KW
3553 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3554 (U8*) SvPVX(*entryp),
3555 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3556 0)));
4637d003 3557 /*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
3558 }
3559 }
3560 }
3561 }
3562 SvREFCNT_dec(specials_inverse); /* done with it */
3563 } /* End of specials */
3564
064c021d
KW
3565 /* read $swash->{LIST} */
3566 l = (U8*)SvPV(*listsvp, lcur);
3567 lend = l + lcur;
3568
3569 /* Go through each input line */
3570 while (l < lend) {
3571 UV min, max, val;
3572 UV inverse;
3573 l = S_swash_scan_list_line(aTHX_ l, lend, &min, &max, &val,
3574 cBOOL(octets), typestr);
3575 if (l > lend) {
3576 break;
3577 }
3578
3579 /* Each element in the range is to be inverted */
3580 for (inverse = min; inverse <= max; inverse++) {
3581 AV* list;
064c021d
KW
3582 SV** listp;
3583 IV i;
3584 bool found_key = FALSE;
5662e334 3585 bool found_inverse = FALSE;
064c021d
KW
3586
3587 /* The key is the inverse mapping */
3588 char key[UTF8_MAXBYTES+1];
c80e42f3 3589 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3590 STRLEN key_len = key_end - key;
3591
064c021d
KW
3592 /* Get the list for the map */
3593 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3594 list = (AV*) *listp;
3595 }
3596 else { /* No entry yet for it: create one */
3597 list = newAV();
3598 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3599 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3600 }
3601 }
3602
5662e334
KW
3603 /* Look through list to see if this inverse mapping already is
3604 * listed, or if there is a mapping to itself already */
b9f2b683 3605 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3606 SV** entryp = av_fetch(list, i, FALSE);
3607 SV* entry;
3608 if (entryp == NULL) {
3609 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3610 }
3611 entry = *entryp;
5662e334 3612 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
56ca34ca 3613 if (SvUV(entry) == val) {
064c021d 3614 found_key = TRUE;
5662e334
KW
3615 }
3616 if (SvUV(entry) == inverse) {
3617 found_inverse = TRUE;
3618 }
3619
3620 /* No need to continue searching if found everything we are
3621 * looking for */
3622 if (found_key && found_inverse) {
064c021d
KW
3623 break;
3624 }
3625 }
56ca34ca
KW
3626
3627 /* Make sure there is a mapping to itself on the list */
064c021d 3628 if (! found_key) {
d397ff6a 3629 av_push(list, newSVuv(val));
4637d003 3630 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3631 }
3632
3633
3634 /* Simply add the value to the list */
5662e334
KW
3635 if (! found_inverse) {
3636 av_push(list, newSVuv(inverse));
4637d003 3637 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3638 }
064c021d 3639
b0e3252e 3640 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3641 * range. That makes more compact tables possible. You can
3642 * express the capitalization, for example, of all consecutive
3643 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3644 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3645 * and it's not documented; it appears to be used only in
b0e3252e 3646 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3647 * in case */
064c021d
KW
3648 if (!none || val < none) {
3649 ++val;
3650 }
3651 }
3652 }
3653
3654 return ret;
3655}
3656
a25abddc 3657SV*
d764b54e
KW
3658Perl__swash_to_invlist(pTHX_ SV* const swash)
3659{
3660
ed92f1b3
KW
3661 /* Subject to change or removal. For use only in one place in regcomp.c.
3662 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3663
3664 U8 *l, *lend;
3665 char *loc;
3666 STRLEN lcur;
3667 HV *const hv = MUTABLE_HV(SvRV(swash));
3668 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3669 U8 empty[] = "";
923b6d4e
KW
3670 SV** listsvp;
3671 SV** typesvp;
3672 SV** bitssvp;
3673 SV** extssvp;
3674 SV** invert_it_svp;
d764b54e 3675
923b6d4e
KW
3676 U8* typestr;
3677 STRLEN bits;
3678 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3679 U8 *x, *xend;
3680 STRLEN xcur;
d764b54e 3681
a25abddc 3682 SV* invlist;
d764b54e 3683
b81740c0
KW
3684 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;