This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add regex nodes for locale
[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
7fefc6c1 42These are various utility functions for manipulating UTF8-encoded
72d33970 43strings. For the uninitiated, this is a method of representing arbitrary
61296642 44Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
45characters in the ASCII range are unmodified, and a zero byte never appears
46within non-zero characters.
166f8a29 47
eaf7a4d2
CS
48=cut
49*/
50
51/*
7bbfa158 52=for apidoc is_invariant_string
eaf7a4d2 53
b8ed1ca9
KW
54Returns true iff the first C<len> bytes of the string C<s> are the same
55regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
56EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
57machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 58definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
59so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
60EBCDIC).
eaf7a4d2 61
75200dff
KW
62If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
63use this option, that C<s> can't have embedded C<NUL> characters and has to
64have a terminating C<NUL> byte).
9f7e3d64 65
a1433954 66See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
67
68=cut
69*/
70
71bool
7bbfa158 72Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
73{
74 const U8* const send = s + (len ? len : strlen((const char *)s));
75 const U8* x = s;
76
7bbfa158 77 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
78
79 for (; x < send; ++x) {
80 if (!UTF8_IS_INVARIANT(*x))
81 break;
82 }
83
84 return x == send;
85}
86
87/*
378516de 88=for apidoc uvoffuni_to_utf8_flags
eebe1485 89
a27992cc 90THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
91Instead, B<Almost all code should use L</uvchr_to_utf8> or
92L</uvchr_to_utf8_flags>>.
a27992cc 93
de69f3af
KW
94This function is like them, but the input is a strict Unicode
95(as opposed to native) code point. Only in very rare circumstances should code
96not be using the native code point.
949cf498 97
de69f3af 98For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 99
eebe1485
SC
100=cut
101*/
102
dfe13c55 103U8 *
378516de 104Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 105{
378516de 106 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 107
d9432125
KW
108 if (UNI_IS_INVARIANT(uv)) {
109 *d++ = (U8) LATIN1_TO_NATIVE(uv);
110 return d;
111 }
112
9415f659
KW
113#ifdef EBCDIC
114 /* Not representable in UTF-EBCDIC */
115 flags |= UNICODE_DISALLOW_FE_FF;
116#endif
117
979f77b6
KW
118 /* The first problematic code point is the first surrogate */
119 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 120 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 121 {
949cf498
KW
122 if (UNICODE_IS_SURROGATE(uv)) {
123 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 124 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
125 "UTF-16 surrogate U+%04"UVXf, uv);
126 }
127 if (flags & UNICODE_DISALLOW_SURROGATE) {
128 return NULL;
129 }
130 }
131 else if (UNICODE_IS_SUPER(uv)) {
132 if (flags & UNICODE_WARN_SUPER
133 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
134 {
8457b38f 135 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
136 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
137 }
138 if (flags & UNICODE_DISALLOW_SUPER
139 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
140 {
9415f659
KW
141#ifdef EBCDIC
142 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
e5964223 143 NOT_REACHED;
9415f659 144#endif
949cf498
KW
145 return NULL;
146 }
147 }
148 else if (UNICODE_IS_NONCHAR(uv)) {
149 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 150 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
151 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
152 uv);
153 }
154 if (flags & UNICODE_DISALLOW_NONCHAR) {
155 return NULL;
156 }
157 }
507b9800 158 }
d9432125 159
2d331972 160#if defined(EBCDIC)
d9432125 161 {
5aaebcb3 162 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
163 U8 *p = d+len-1;
164 while (p > d) {
bc3632a8 165 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
166 uv >>= UTF_ACCUMULATION_SHIFT;
167 }
bc3632a8 168 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
169 return d+len;
170 }
171#else /* Non loop style */
a0ed51b3 172 if (uv < 0x800) {
eb160463
GS
173 *d++ = (U8)(( uv >> 6) | 0xc0);
174 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
175 return d;
176 }
177 if (uv < 0x10000) {
eb160463
GS
178 *d++ = (U8)(( uv >> 12) | 0xe0);
179 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
180 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
181 return d;
182 }
183 if (uv < 0x200000) {
eb160463
GS
184 *d++ = (U8)(( uv >> 18) | 0xf0);
185 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
186 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
187 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
188 return d;
189 }
190 if (uv < 0x4000000) {
eb160463
GS
191 *d++ = (U8)(( uv >> 24) | 0xf8);
192 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
194 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
195 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
196 return d;
197 }
198 if (uv < 0x80000000) {
eb160463
GS
199 *d++ = (U8)(( uv >> 30) | 0xfc);
200 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
204 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
205 return d;
206 }
6588300d 207#ifdef UTF8_QUAD_MAX
d7578b48 208 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
209#endif
210 {
eb160463
GS
211 *d++ = 0xfe; /* Can't match U+FEFF! */
212 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
217 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
218 return d;
219 }
6588300d 220#ifdef UTF8_QUAD_MAX
a0ed51b3 221 {
eb160463
GS
222 *d++ = 0xff; /* Can't match U+FFFE! */
223 *d++ = 0x80; /* 6 Reserved bits */
224 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
225 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
233 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
234 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
235 return d;
236 }
237#endif
537124e4 238#endif /* Non loop style */
a0ed51b3 239}
646ca15d 240/*
07693fe6
KW
241=for apidoc uvchr_to_utf8
242
bcb1a2d4 243Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
244of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
245C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
246the byte after the end of the new character. In other words,
07693fe6
KW
247
248 d = uvchr_to_utf8(d, uv);
249
250is the recommended wide native character-aware way of saying
251
252 *(d++) = uv;
253
de69f3af
KW
254This function accepts any UV as input. To forbid or warn on non-Unicode code
255points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
256
07693fe6
KW
257=cut
258*/
259
de69f3af
KW
260/* This is also a macro */
261PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
262
07693fe6
KW
263U8 *
264Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
265{
de69f3af 266 return uvchr_to_utf8(d, uv);
07693fe6
KW
267}
268
de69f3af
KW
269/*
270=for apidoc uvchr_to_utf8_flags
271
272Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
273of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
274C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
275the byte after the end of the new character. In other words,
de69f3af
KW
276
277 d = uvchr_to_utf8_flags(d, uv, flags);
278
279or, in most cases,
280
281 d = uvchr_to_utf8_flags(d, uv, 0);
282
283This is the Unicode-aware way of saying
284
285 *(d++) = uv;
286
287This function will convert to UTF-8 (and not warn) even code points that aren't
288legal Unicode or are problematic, unless C<flags> contains one or more of the
289following flags:
290
291If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
292the function will raise a warning, provided UTF8 warnings are enabled. If instead
293UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
294If both flags are set, the function will both warn and return NULL.
295
4c3cfd5d 296The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 297affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 298UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
299code points that are
300above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
301even less portable) can be warned and/or disallowed even if other above-Unicode
302code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
303flags.
304
305And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
306above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
307DISALLOW flags.
308
309=cut
310*/
311
312/* This is also a macro */
313PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
314
07693fe6
KW
315U8 *
316Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
317{
de69f3af 318 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
319}
320
321/*
492a624f
KW
322=for apidoc is_utf8_char_buf
323
3cedd9d9 324This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
325
326=cut */
327
328STRLEN
329Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
330{
331
492a624f
KW
332 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
333
6302f837 334 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
335}
336
337/*
87cea99e 338=for apidoc is_utf8_string
6662521e 339
a1433954 340Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 341UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
342using C<strlen(s)> (which means if you use this option, that C<s> can't have
343embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
344that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 345
7bbfa158 346See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 347
6662521e
GS
348=cut
349*/
350
8e84507e 351bool
668b6d8d 352Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 353{
35da51f7 354 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 355 const U8* x = s;
067a85ef 356
7918f24d 357 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 358
6662521e 359 while (x < send) {
6302f837
KW
360 STRLEN len = isUTF8_CHAR(x, send);
361 if (UNLIKELY(! len)) {
362 return FALSE;
363 }
364 x += len;
6662521e 365 }
768c67ee 366
067a85ef 367 return TRUE;
6662521e
GS
368}
369
67e989fb 370/*
814fafa7
NC
371Implemented as a macro in utf8.h
372
87cea99e 373=for apidoc is_utf8_string_loc
814fafa7 374
a1433954
KW
375Like L</is_utf8_string> but stores the location of the failure (in the
376case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
377"utf8ness success") in the C<ep>.
378
a1433954 379See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 380
87cea99e 381=for apidoc is_utf8_string_loclen
81cd54e3 382
a1433954
KW
383Like L</is_utf8_string>() but stores the location of the failure (in the
384case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
385"utf8ness success") in the C<ep>, and the number of UTF-8
386encoded characters in the C<el>.
387
a1433954 388See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
389
390=cut
391*/
392
393bool
668b6d8d 394Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 395{
35da51f7 396 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 397 const U8* x = s;
3ebfea28 398 STRLEN outlen = 0;
7918f24d
NC
399
400 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 401
81cd54e3 402 while (x < send) {
6302f837
KW
403 STRLEN len = isUTF8_CHAR(x, send);
404 if (UNLIKELY(! len)) {
405 goto out;
406 }
407 x += len;
408 outlen++;
81cd54e3 409 }
768c67ee
JH
410
411 out:
3ebfea28
AL
412 if (el)
413 *el = outlen;
414
768c67ee
JH
415 if (ep)
416 *ep = x;
3ebfea28 417 return (x == send);
81cd54e3
JH
418}
419
420/*
768c67ee 421
de69f3af 422=for apidoc utf8n_to_uvchr
378516de
KW
423
424THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 425Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 426
9041c2e3 427Bottom level UTF-8 decode routine.
de69f3af 428Returns the native code point value of the first character in the string C<s>,
746afd53
KW
429which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
430C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
431the length, in bytes, of that character.
949cf498
KW
432
433The value of C<flags> determines the behavior when C<s> does not point to a
434well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
435zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
436next possible position in C<s> that could begin a non-malformed character.
437Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
438
439Various ALLOW flags can be set in C<flags> to allow (and not warn on)
440individual types of malformations, such as the sequence being overlong (that
441is, when there is a shorter sequence that can express the same code point;
442overlong sequences are expressly forbidden in the UTF-8 standard due to
443potential security issues). Another malformation example is the first byte of
444a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
445flags. For allowed 0 length strings, this function returns 0; for allowed
446overlong sequences, the computed code point is returned; for all other allowed
447malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
448determinable reasonable value.
949cf498
KW
449
450The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
451flags) malformation is found. If this flag is set, the routine assumes that
452the caller will raise a warning, and this function will silently just set
d088425d
KW
453C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
454
75200dff 455Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
456character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
457in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
458first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
459input had an error.
949cf498
KW
460
461Certain code points are considered problematic. These are Unicode surrogates,
746afd53 462Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 463By default these are considered regular code points, but certain situations
5eafe189 464warrant special handling for them. If C<flags> contains
949cf498
KW
465UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
466malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
467UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
468maximum) can be set to disallow these categories individually.
469
470The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
471UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
472for their respective categories, but otherwise the code points are considered
473valid (not malformations). To get a category to both be treated as a
474malformation and raise a warning, specify both the WARN and DISALLOW flags.
475(But note that warnings are not raised if lexically disabled nor if
476UTF8_CHECK_ONLY is also specified.)
477
478Very large code points (above 0x7FFF_FFFF) are considered more problematic than
479the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
480reasons: they requre at least 32 bits to represent them on ASCII platforms, are
481not representable at all on EBCDIC platforms, and the original UTF-8
482specification never went above this number (the current 0x10FFFF limit was
483imposed later). (The smaller ones, those that fit into 32 bits, are
484representable by a UV on ASCII platforms, but not by an IV, which means that
485the number of operations that can be performed on them is quite restricted.)
486The UTF-8 encoding on ASCII platforms for these large code points begins with a
487byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
488be treated as malformations, while allowing smaller above-Unicode code points.
489(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
490including these, as malformations.)
491Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 492the other WARN flags, but applies just to these code points.
949cf498
KW
493
494All other code points corresponding to Unicode characters, including private
495use and those yet to be assigned, are never considered malformed and never
496warn.
67e989fb 497
37607a96
PK
498=cut
499*/
67e989fb 500
a0ed51b3 501UV
de69f3af 502Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 503{
d4c19fe8 504 const U8 * const s0 = s;
eb83ed87 505 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 506 U8 * send;
eb83ed87
KW
507 UV uv = *s;
508 STRLEN expectlen;
949cf498 509 SV* sv = NULL;
eb83ed87
KW
510 UV outlier_ret = 0; /* return value when input is in error or problematic
511 */
512 UV pack_warn = 0; /* Save result of packWARN() for later */
513 bool unexpected_non_continuation = FALSE;
514 bool overflowed = FALSE;
2f8f112e 515 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 516
eb83ed87 517 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 518
de69f3af 519 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 520
eb83ed87
KW
521 /* The order of malformation tests here is important. We should consume as
522 * few bytes as possible in order to not skip any valid character. This is
523 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
524 * http://unicode.org/reports/tr36 for more discussion as to why. For
525 * example, once we've done a UTF8SKIP, we can tell the expected number of
526 * bytes, and could fail right off the bat if the input parameters indicate
527 * that there are too few available. But it could be that just that first
528 * byte is garbled, and the intended character occupies fewer bytes. If we
529 * blindly assumed that the first byte is correct, and skipped based on
530 * that number, we could skip over a valid input character. So instead, we
531 * always examine the sequence byte-by-byte.
532 *
533 * We also should not consume too few bytes, otherwise someone could inject
534 * things. For example, an input could be deliberately designed to
535 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 536 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
537 * which is actually part of of the overflowing sequence), that could look
538 * legitimate to the caller, which could discard the initial partial
539 * sequence and process the rest, inappropriately */
540
541 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 542 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
543 if (retlen) {
544 *retlen = 0;
545 }
a0dbb045 546
eb83ed87
KW
547 if (flags & UTF8_ALLOW_EMPTY) {
548 return 0;
549 }
550 if (! (flags & UTF8_CHECK_ONLY)) {
551 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
552 }
0c443dc2
JH
553 goto malformed;
554 }
555
eb83ed87
KW
556 expectlen = UTF8SKIP(s);
557
558 /* A well-formed UTF-8 character, as the vast majority of calls to this
559 * function will be for, has this expected length. For efficiency, set
560 * things up here to return it. It will be overriden only in those rare
561 * cases where a malformation is found */
562 if (retlen) {
563 *retlen = expectlen;
564 }
565
566 /* An invariant is trivially well-formed */
1d72bdf6 567 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 568 return uv;
a0ed51b3 569 }
67e989fb 570
eb83ed87 571 /* A continuation character can't start a valid sequence */
b5b9af04 572 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
573 if (flags & UTF8_ALLOW_CONTINUATION) {
574 if (retlen) {
575 *retlen = 1;
576 }
577 return UNICODE_REPLACEMENT;
578 }
ba210ebe 579
eb83ed87
KW
580 if (! (flags & UTF8_CHECK_ONLY)) {
581 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
582 }
583 curlen = 1;
ba210ebe
JH
584 goto malformed;
585 }
9041c2e3 586
dcd27b3c
KW
587 /* Here is not a continuation byte, nor an invariant. The only thing left
588 * is a start byte (possibly for an overlong) */
589
1d72bdf6 590#ifdef EBCDIC
bc3632a8 591 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
592#endif
593
eb83ed87
KW
594 /* Remove the leading bits that indicate the number of bytes in the
595 * character's whole UTF-8 sequence, leaving just the bits that are part of
596 * the value */
597 uv &= UTF_START_MASK(expectlen);
ba210ebe 598
eb83ed87
KW
599 /* Now, loop through the remaining bytes in the character's sequence,
600 * accumulating each into the working value as we go. Be sure to not look
601 * past the end of the input string */
0b8d30e8
KW
602 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
603
eb83ed87 604 for (s = s0 + 1; s < send; s++) {
b5b9af04 605 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
606#ifndef EBCDIC /* Can't overflow in EBCDIC */
607 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
608
609 /* The original implementors viewed this malformation as more
610 * serious than the others (though I, khw, don't understand
611 * why, since other malformations also give very very wrong
612 * results), so there is no way to turn off checking for it.
613 * Set a flag, but keep going in the loop, so that we absorb
614 * the rest of the bytes that comprise the character. */
615 overflowed = TRUE;
616 overflow_byte = *s; /* Save for warning message's use */
617 }
618#endif
8850bf83 619 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
620 }
621 else {
622 /* Here, found a non-continuation before processing all expected
623 * bytes. This byte begins a new character, so quit, even if
624 * allowing this malformation. */
625 unexpected_non_continuation = TRUE;
626 break;
627 }
628 } /* End of loop through the character's bytes */
629
630 /* Save how many bytes were actually in the character */
631 curlen = s - s0;
632
633 /* The loop above finds two types of malformations: non-continuation and/or
634 * overflow. The non-continuation malformation is really a too-short
635 * malformation, as it means that the current character ended before it was
636 * expected to (being terminated prematurely by the beginning of the next
637 * character, whereas in the too-short malformation there just are too few
638 * bytes available to hold the character. In both cases, the check below
639 * that we have found the expected number of bytes would fail if executed.)
640 * Thus the non-continuation malformation is really unnecessary, being a
641 * subset of the too-short malformation. But there may be existing
642 * applications that are expecting the non-continuation type, so we retain
643 * it, and return it in preference to the too-short malformation. (If this
644 * code were being written from scratch, the two types might be collapsed
645 * into one.) I, khw, am also giving priority to returning the
646 * non-continuation and too-short malformations over overflow when multiple
647 * ones are present. I don't know of any real reason to prefer one over
648 * the other, except that it seems to me that multiple-byte errors trumps
649 * errors from a single byte */
b5b9af04 650 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
651 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
652 if (! (flags & UTF8_CHECK_ONLY)) {
653 if (curlen == 1) {
654 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
655 }
656 else {
657 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
658 }
659 }
eb83ed87
KW
660 goto malformed;
661 }
662 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
663
664 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
665 * as what the original expectations were. */
666 do_overlong_test = FALSE;
eb83ed87
KW
667 if (retlen) {
668 *retlen = curlen;
669 }
670 }
b5b9af04 671 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
672 if (! (flags & UTF8_ALLOW_SHORT)) {
673 if (! (flags & UTF8_CHECK_ONLY)) {
674 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 675 }
eb83ed87
KW
676 goto malformed;
677 }
678 uv = UNICODE_REPLACEMENT;
2f8f112e 679 do_overlong_test = FALSE;
eb83ed87
KW
680 if (retlen) {
681 *retlen = curlen;
682 }
683 }
684
ea5ced44 685#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 686 if (UNLIKELY(overflowed)) {
eb83ed87 687 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 688 goto malformed;
eb83ed87
KW
689 }
690#endif
691
2f8f112e 692 if (do_overlong_test
5aaebcb3 693 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
694 && ! (flags & UTF8_ALLOW_LONG))
695 {
eb83ed87
KW
696 /* The overlong malformation has lower precedence than the others.
697 * Note that if this malformation is allowed, we return the actual
698 * value, instead of the replacement character. This is because this
699 * value is actually well-defined. */
700 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 701 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
702 }
703 goto malformed;
704 }
705
1a89bb6c 706 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
707 * problematic code point that is not allowed by the input parameters. */
708 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
709 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
710 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
711 {
949cf498 712 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
713
714 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
715 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 716 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 717 && ckWARN_d(WARN_SURROGATE))
eb83ed87 718 {
111d382d 719 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 720 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
721 }
722 if (flags & UTF8_DISALLOW_SURROGATE) {
723 goto disallowed;
724 }
725 }
949cf498 726 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 727 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 728 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 729 {
111d382d 730 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 731 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 732 }
ea5ced44
KW
733#ifndef EBCDIC /* EBCDIC always allows FE, FF */
734
735 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
736 * points. We test for these after the regular SUPER ones, and
737 * before possibly bailing out, so that the more dire warning
738 * overrides the regular one, if applicable */
739 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
740 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
741 {
742 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
743 == UTF8_WARN_FE_FF
744 && ckWARN_d(WARN_UTF8))
745 {
746 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
747 pack_warn = packWARN(WARN_UTF8);
748 }
749 if (flags & UTF8_DISALLOW_FE_FF) {
750 goto disallowed;
751 }
752 }
753#endif
949cf498
KW
754 if (flags & UTF8_DISALLOW_SUPER) {
755 goto disallowed;
756 }
757 }
4190d317
KW
758 else if (UNICODE_IS_NONCHAR(uv)) {
759 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 760 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
761 {
762 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 763 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
764 }
765 if (flags & UTF8_DISALLOW_NONCHAR) {
766 goto disallowed;
767 }
768 }
949cf498 769
eb83ed87 770 if (sv) {
de69f3af
KW
771 outlier_ret = uv; /* Note we don't bother to convert to native,
772 as all the outlier code points are the same
773 in both ASCII and EBCDIC */
eb83ed87
KW
774 goto do_warn;
775 }
776
949cf498
KW
777 /* Here, this is not considered a malformed character, so drop through
778 * to return it */
a0ed51b3 779 }
ba210ebe 780
de69f3af 781 return UNI_TO_NATIVE(uv);
ba210ebe 782
eb83ed87
KW
783 /* There are three cases which get to beyond this point. In all 3 cases:
784 * <sv> if not null points to a string to print as a warning.
785 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
786 * set.
787 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
788 * This is done by initializing it to 0, and changing it only
789 * for case 1).
790 * The 3 cases are:
791 * 1) The input is valid but problematic, and to be warned about. The
792 * return value is the resultant code point; <*retlen> is set to
793 * <curlen>, the number of bytes that comprise the code point.
794 * <pack_warn> contains the result of packWARN() for the warning
795 * types. The entry point for this case is the label <do_warn>;
796 * 2) The input is a valid code point but disallowed by the parameters to
797 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
798 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
799 * comprise the code point. <pack_warn> contains the result of
800 * packWARN() for the warning types. The entry point for this case is
801 * the label <disallowed>.
802 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
803 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
804 * bytes that comprise the malformation. All such malformations are
805 * assumed to be warning type <utf8>. The entry point for this case
806 * is the label <malformed>.
807 */
949cf498 808
ba210ebe
JH
809malformed:
810
eb83ed87
KW
811 if (sv && ckWARN_d(WARN_UTF8)) {
812 pack_warn = packWARN(WARN_UTF8);
813 }
814
815disallowed:
816
fcc8fcf6 817 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 818 if (retlen)
10edeb5d 819 *retlen = ((STRLEN) -1);
ba210ebe
JH
820 return 0;
821 }
822
eb83ed87 823do_warn:
5b311467 824
eb83ed87
KW
825 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
826 if warnings are to be raised. */
f555bc63 827 const char * const string = SvPVX_const(sv);
a0dbb045 828
f555bc63
KW
829 if (PL_op)
830 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
831 else
832 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
833 }
834
eb83ed87
KW
835 if (retlen) {
836 *retlen = curlen;
837 }
ba210ebe 838
eb83ed87 839 return outlier_ret;
a0ed51b3
LW
840}
841
8e84507e 842/*
ec5f19d0
KW
843=for apidoc utf8_to_uvchr_buf
844
845Returns the native code point of the first character in the string C<s> which
846is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 847C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 848
524080c4
KW
849If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
850enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
851NULL) to -1. If those warnings are off, the computed value, if well-defined
852(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
853C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
854the next possible position in C<s> that could begin a non-malformed character.
de69f3af 855See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 856returned.
ec5f19d0
KW
857
858=cut
859*/
860
861
862UV
863Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
864{
ec5f19d0
KW
865 assert(s < send);
866
867 return utf8n_to_uvchr(s, send - s, retlen,
868 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
869}
870
27d6c58a 871/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 872 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 873 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
874
875UV
876Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
877{
010ab96b
KW
878 UV expectlen = UTF8SKIP(s);
879 const U8* send = s + expectlen;
9ff2f0f7 880 UV uv = *s;
3986bb7c 881
27d6c58a 882 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 883 PERL_UNUSED_CONTEXT;
27d6c58a 884
010ab96b
KW
885 if (retlen) {
886 *retlen = expectlen;
887 }
888
889 /* An invariant is trivially returned */
890 if (expectlen == 1) {
9ff2f0f7 891 return uv;
010ab96b
KW
892 }
893
9ff2f0f7
KW
894#ifdef EBCDIC
895 uv = NATIVE_UTF8_TO_I8(uv);
896#endif
897
010ab96b
KW
898 /* Remove the leading bits that indicate the number of bytes, leaving just
899 * the bits that are part of the value */
900 uv &= UTF_START_MASK(expectlen);
901
902 /* Now, loop through the remaining bytes, accumulating each into the
903 * working total as we go. (I khw tried unrolling the loop for up to 4
904 * bytes, but there was no performance improvement) */
905 for (++s; s < send; s++) {
906 uv = UTF8_ACCUMULATE(uv, *s);
907 }
908
3986bb7c 909 return UNI_TO_NATIVE(uv);
010ab96b 910
27d6c58a
KW
911}
912
ec5f19d0 913/*
ec5f19d0
KW
914=for apidoc utf8_to_uvuni_buf
915
de69f3af
KW
916Only in very rare circumstances should code need to be dealing in Unicode
917(as opposed to native) code points. In those few cases, use
918C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
919
920Returns the Unicode (not-native) code point of the first character in the
921string C<s> which
ec5f19d0
KW
922is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
923C<retlen> will be set to the length, in bytes, of that character.
924
524080c4
KW
925If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
926enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
927NULL) to -1. If those warnings are off, the computed value if well-defined (or
928the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
929is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
930next possible position in C<s> that could begin a non-malformed character.
de69f3af 931See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
932
933=cut
934*/
935
936UV
937Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
938{
939 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
940
941 assert(send > s);
942
943 /* Call the low level routine asking for checks */
de69f3af
KW
944 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
945 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
946}
947
b76347f2 948/*
87cea99e 949=for apidoc utf8_length
b76347f2
JH
950
951Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
952Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
953up past C<e>, croaks.
b76347f2
JH
954
955=cut
956*/
957
958STRLEN
35a4481c 959Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
960{
961 STRLEN len = 0;
962
7918f24d
NC
963 PERL_ARGS_ASSERT_UTF8_LENGTH;
964
8850bf83
JH
965 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
966 * the bitops (especially ~) can create illegal UTF-8.
967 * In other words: in Perl UTF-8 is not just for Unicode. */
968
a3b680e6
AL
969 if (e < s)
970 goto warn_and_return;
b76347f2 971 while (s < e) {
4cbf4130 972 s += UTF8SKIP(s);
8e91ec7f
AV
973 len++;
974 }
975
976 if (e != s) {
977 len--;
978 warn_and_return:
9b387841
NC
979 if (PL_op)
980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
981 "%s in %s", unees, OP_DESC(PL_op));
982 else
61a12c31 983 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
984 }
985
986 return len;
987}
988
b06226ff 989/*
87cea99e 990=for apidoc utf8_distance
b06226ff 991
1e54db1a 992Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
993and C<b>.
994
995WARNING: use only if you *know* that the pointers point inside the
996same UTF-8 buffer.
997
37607a96
PK
998=cut
999*/
a0ed51b3 1000
02eb7b47 1001IV
35a4481c 1002Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1003{
7918f24d
NC
1004 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1005
bf1665bc 1006 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1007}
1008
b06226ff 1009/*
87cea99e 1010=for apidoc utf8_hop
b06226ff 1011
8850bf83
JH
1012Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1013forward or backward.
b06226ff
JH
1014
1015WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1016the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1017on the first byte of character or just after the last byte of a character.
b06226ff 1018
37607a96
PK
1019=cut
1020*/
a0ed51b3
LW
1021
1022U8 *
ddeaf645 1023Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1024{
7918f24d
NC
1025 PERL_ARGS_ASSERT_UTF8_HOP;
1026
8850bf83
JH
1027 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1028 * the bitops (especially ~) can create illegal UTF-8.
1029 * In other words: in Perl UTF-8 is not just for Unicode. */
1030
a0ed51b3
LW
1031 if (off >= 0) {
1032 while (off--)
1033 s += UTF8SKIP(s);
1034 }
1035 else {
1036 while (off++) {
1037 s--;
8850bf83
JH
1038 while (UTF8_IS_CONTINUATION(*s))
1039 s--;
a0ed51b3
LW
1040 }
1041 }
4373e329 1042 return (U8 *)s;
a0ed51b3
LW
1043}
1044
6940069f 1045/*
fed3ba5d
NC
1046=for apidoc bytes_cmp_utf8
1047
a1433954 1048Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1049sequence of characters (stored as UTF-8)
1050in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1051equal, -1 or -2 if the first string is less than the second string, +1 or +2
1052if the first string is greater than the second string.
1053
1054-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1055longer string. -2 or +2 is returned if
1056there was a difference between characters
fed3ba5d
NC
1057within the strings.
1058
1059=cut
1060*/
1061
1062int
1063Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1064{
1065 const U8 *const bend = b + blen;
1066 const U8 *const uend = u + ulen;
1067
1068 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1069
1070 while (b < bend && u < uend) {
1071 U8 c = *u++;
1072 if (!UTF8_IS_INVARIANT(c)) {
1073 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1074 if (u < uend) {
1075 U8 c1 = *u++;
1076 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1077 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1078 } else {
1079 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1080 "Malformed UTF-8 character "
1081 "(unexpected non-continuation byte 0x%02x"
1082 ", immediately after start byte 0x%02x)"
1083 /* Dear diag.t, it's in the pod. */
1084 "%s%s", c1, c,
1085 PL_op ? " in " : "",
1086 PL_op ? OP_DESC(PL_op) : "");
1087 return -2;
1088 }
1089 } else {
1090 if (PL_op)
1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1092 "%s in %s", unees, OP_DESC(PL_op));
1093 else
61a12c31 1094 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1095 return -2; /* Really want to return undef :-) */
1096 }
1097 } else {
1098 return -2;
1099 }
1100 }
1101 if (*b != c) {
1102 return *b < c ? -2 : +2;
1103 }
1104 ++b;
1105 }
1106
1107 if (b == bend && u == uend)
1108 return 0;
1109
1110 return b < bend ? +1 : -1;
1111}
1112
1113/*
87cea99e 1114=for apidoc utf8_to_bytes
6940069f 1115
2bbc8d55 1116Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1117Unlike L</bytes_to_utf8>, this over-writes the original string, and
1118updates C<len> to contain the new length.
67e989fb 1119Returns zero on failure, setting C<len> to -1.
6940069f 1120
a1433954 1121If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1122
6940069f
GS
1123=cut
1124*/
1125
1126U8 *
37607a96 1127Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1128{
d4c19fe8
AL
1129 U8 * const save = s;
1130 U8 * const send = s + *len;
6940069f 1131 U8 *d;
246fae53 1132
7918f24d 1133 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1134 PERL_UNUSED_CONTEXT;
7918f24d 1135
1e54db1a 1136 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1137 while (s < send) {
d59937ca
KW
1138 if (! UTF8_IS_INVARIANT(*s)) {
1139 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1140 *len = ((STRLEN) -1);
1141 return 0;
1142 }
1143 s++;
dcad2880 1144 }
d59937ca 1145 s++;
246fae53 1146 }
dcad2880
JH
1147
1148 d = s = save;
6940069f 1149 while (s < send) {
80e0b38f
KW
1150 U8 c = *s++;
1151 if (! UTF8_IS_INVARIANT(c)) {
1152 /* Then it is two-byte encoded */
1153 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1154 s++;
1155 }
1156 *d++ = c;
6940069f
GS
1157 }
1158 *d = '\0';
246fae53 1159 *len = d - save;
6940069f
GS
1160 return save;
1161}
1162
1163/*
87cea99e 1164=for apidoc bytes_from_utf8
f9a63242 1165
2bbc8d55 1166Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1167Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1168the newly-created string, and updates C<len> to contain the new
1169length. Returns the original string if no conversion occurs, C<len>
72d33970 1170is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11710 if C<s> is converted or consisted entirely of characters that are invariant
1172in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1173
37607a96
PK
1174=cut
1175*/
f9a63242
JH
1176
1177U8 *
e1ec3a88 1178Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1179{
f9a63242 1180 U8 *d;
e1ec3a88
AL
1181 const U8 *start = s;
1182 const U8 *send;
f9a63242
JH
1183 I32 count = 0;
1184
7918f24d 1185 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1186 PERL_UNUSED_CONTEXT;
f9a63242 1187 if (!*is_utf8)
73d840c0 1188 return (U8 *)start;
f9a63242 1189
1e54db1a 1190 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1191 for (send = s + *len; s < send;) {
d59937ca
KW
1192 if (! UTF8_IS_INVARIANT(*s)) {
1193 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1194 return (U8 *)start;
d59937ca
KW
1195 }
1196 count++;
1197 s++;
db42d148 1198 }
d59937ca 1199 s++;
f9a63242
JH
1200 }
1201
35da51f7 1202 *is_utf8 = FALSE;
f9a63242 1203
212542aa 1204 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1205 s = start; start = d;
f9a63242
JH
1206 while (s < send) {
1207 U8 c = *s++;
1a91c45d 1208 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1209 /* Then it is two-byte encoded */
1a91c45d
KW
1210 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1211 s++;
c4d5f83a
NIS
1212 }
1213 *d++ = c;
f9a63242
JH
1214 }
1215 *d = '\0';
1216 *len = d - start;
73d840c0 1217 return (U8 *)start;
f9a63242
JH
1218}
1219
1220/*
87cea99e 1221=for apidoc bytes_to_utf8
6940069f 1222
ff97e5cf
KW
1223Converts a string C<s> of length C<len> bytes from the native encoding into
1224UTF-8.
6662521e 1225Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1226reflect the new length in bytes.
6940069f 1227
75200dff 1228A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1229
1230If you want to convert to UTF-8 from encodings other than
1231the native (Latin1 or EBCDIC),
a1433954 1232see L</sv_recode_to_utf8>().
c9ada85f 1233
497711e7 1234=cut
6940069f
GS
1235*/
1236
c682ebef
FC
1237/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1238 likewise need duplication. */
1239
6940069f 1240U8*
35a4481c 1241Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1242{
35a4481c 1243 const U8 * const send = s + (*len);
6940069f
GS
1244 U8 *d;
1245 U8 *dst;
7918f24d
NC
1246
1247 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1248 PERL_UNUSED_CONTEXT;
6940069f 1249
212542aa 1250 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1251 dst = d;
1252
1253 while (s < send) {
55d09dc8
KW
1254 append_utf8_from_native_byte(*s, &d);
1255 s++;
6940069f
GS
1256 }
1257 *d = '\0';
6662521e 1258 *len = d-dst;
6940069f
GS
1259 return dst;
1260}
1261
a0ed51b3 1262/*
dea0fc0b 1263 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1264 *
1265 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1266 * We optimize for native, for obvious reasons. */
1267
1268U8*
dea0fc0b 1269Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1270{
dea0fc0b
JH
1271 U8* pend;
1272 U8* dstart = d;
1273
7918f24d
NC
1274 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1275
dea0fc0b 1276 if (bytelen & 1)
f5992bc4 1277 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1278
1279 pend = p + bytelen;
1280
a0ed51b3 1281 while (p < pend) {
dea0fc0b
JH
1282 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1283 p += 2;
56d37426
KW
1284 if (UNI_IS_INVARIANT(uv)) {
1285 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1286 continue;
1287 }
56d37426
KW
1288 if (uv <= MAX_UTF8_TWO_BYTE) {
1289 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1290 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1291 continue;
1292 }
46956fad
KW
1293#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1294#define LAST_HIGH_SURROGATE 0xDBFF
1295#define FIRST_LOW_SURROGATE 0xDC00
1296#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1297 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1298 if (p >= pend) {
dea0fc0b 1299 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1300 } else {
1301 UV low = (p[0] << 8) + p[1];
1302 p += 2;
46956fad 1303 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1304 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1305 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1306 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1307 }
46956fad 1308 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1309 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1310 }
56d37426
KW
1311#ifdef EBCDIC
1312 d = uvoffuni_to_utf8_flags(d, uv, 0);
1313#else
a0ed51b3 1314 if (uv < 0x10000) {
eb160463
GS
1315 *d++ = (U8)(( uv >> 12) | 0xe0);
1316 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1317 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1318 continue;
1319 }
1320 else {
eb160463
GS
1321 *d++ = (U8)(( uv >> 18) | 0xf0);
1322 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1323 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1324 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1325 continue;
1326 }
56d37426 1327#endif
a0ed51b3 1328 }
dea0fc0b 1329 *newlen = d - dstart;
a0ed51b3
LW
1330 return d;
1331}
1332
1333/* Note: this one is slightly destructive of the source. */
1334
1335U8*
dea0fc0b 1336Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1337{
1338 U8* s = (U8*)p;
d4c19fe8 1339 U8* const send = s + bytelen;
7918f24d
NC
1340
1341 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1342
e0ea5e2d
NC
1343 if (bytelen & 1)
1344 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1345 (UV)bytelen);
1346
a0ed51b3 1347 while (s < send) {
d4c19fe8 1348 const U8 tmp = s[0];
a0ed51b3
LW
1349 s[0] = s[1];
1350 s[1] = tmp;
1351 s += 2;
1352 }
dea0fc0b 1353 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1354}
1355
922e8cb4
KW
1356bool
1357Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1358{
1359 U8 tmpbuf[UTF8_MAXBYTES+1];
1360 uvchr_to_utf8(tmpbuf, c);
1361 return _is_utf8_FOO(classnum, tmpbuf);
1362}
1363
f9ae8fb6
JD
1364/* Internal function so we can deprecate the external one, and call
1365 this one from other deprecated functions in this file */
1366
f2645549
KW
1367bool
1368Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1369{
f2645549 1370 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1371
1372 if (*p == '_')
1373 return TRUE;
f25ce844 1374 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1375}
1376
5092f92a 1377bool
eba68aa0
KW
1378Perl__is_uni_perl_idcont(pTHX_ UV c)
1379{
1380 U8 tmpbuf[UTF8_MAXBYTES+1];
1381 uvchr_to_utf8(tmpbuf, c);
1382 return _is_utf8_perl_idcont(tmpbuf);
1383}
1384
1385bool
f91dcd13
KW
1386Perl__is_uni_perl_idstart(pTHX_ UV c)
1387{
1388 U8 tmpbuf[UTF8_MAXBYTES+1];
1389 uvchr_to_utf8(tmpbuf, c);
1390 return _is_utf8_perl_idstart(tmpbuf);
1391}
1392
3a4c58c9
KW
1393UV
1394Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1395{
1396 /* We have the latin1-range values compiled into the core, so just use
1397 * those, converting the result to utf8. The only difference between upper
1398 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1399 * either "SS" or "Ss". Which one to use is passed into the routine in
1400 * 'S_or_s' to avoid a test */
1401
1402 UV converted = toUPPER_LATIN1_MOD(c);
1403
1404 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1405
1406 assert(S_or_s == 'S' || S_or_s == 's');
1407
6f2d5cbc 1408 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1409 characters in this range */
3a4c58c9
KW
1410 *p = (U8) converted;
1411 *lenp = 1;
1412 return converted;
1413 }
1414
1415 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1416 * which it maps to one of them, so as to only have to have one check for
1417 * it in the main case */
1418 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1419 switch (c) {
1420 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1421 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1422 break;
1423 case MICRO_SIGN:
1424 converted = GREEK_CAPITAL_LETTER_MU;
1425 break;
1426 case LATIN_SMALL_LETTER_SHARP_S:
1427 *(p)++ = 'S';
1428 *p = S_or_s;
1429 *lenp = 2;
1430 return 'S';
1431 default:
1432 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1433 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1434 }
1435 }
1436
1437 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1438 *p = UTF8_TWO_BYTE_LO(converted);
1439 *lenp = 2;
1440
1441 return converted;
1442}
1443
50bda2c3
KW
1444/* Call the function to convert a UTF-8 encoded character to the specified case.
1445 * Note that there may be more than one character in the result.
1446 * INP is a pointer to the first byte of the input character
1447 * OUTP will be set to the first byte of the string of changed characters. It
1448 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1449 * LENP will be set to the length in bytes of the string of changed characters
1450 *
1451 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1452#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1453#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1454#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1455
1456/* This additionally has the input parameter SPECIALS, which if non-zero will
1457 * cause this to use the SPECIALS hash for folding (meaning get full case
1458 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1459#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1460
84afefe6
JH
1461UV
1462Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1463{
a1433954
KW
1464 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1465 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1466 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1467 * the changed version may be longer than the original character.
1468 *
1469 * The ordinal of the first character of the changed version is returned
1470 * (but note, as explained above, that there may be more.) */
1471
7918f24d
NC
1472 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1473
3a4c58c9
KW
1474 if (c < 256) {
1475 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1476 }
1477
0ebc6274 1478 uvchr_to_utf8(p, c);
3a4c58c9 1479 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1480}
1481
84afefe6
JH
1482UV
1483Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1484{
7918f24d
NC
1485 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1486
3a4c58c9
KW
1487 if (c < 256) {
1488 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1489 }
1490
0ebc6274 1491 uvchr_to_utf8(p, c);
3a4c58c9 1492 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1493}
1494
afc16117 1495STATIC U8
81611534 1496S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1497{
1498 /* We have the latin1-range values compiled into the core, so just use
1499 * those, converting the result to utf8. Since the result is always just
a1433954 1500 * one character, we allow <p> to be NULL */
afc16117
KW
1501
1502 U8 converted = toLOWER_LATIN1(c);
1503
1504 if (p != NULL) {
6f2d5cbc 1505 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1506 *p = converted;
1507 *lenp = 1;
1508 }
1509 else {
430c9760
KW
1510 /* Result is known to always be < 256, so can use the EIGHT_BIT
1511 * macros */
1512 *p = UTF8_EIGHT_BIT_HI(converted);
1513 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1514 *lenp = 2;
1515 }
1516 }
1517 return converted;
1518}
1519
84afefe6
JH
1520UV
1521Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1522{
7918f24d
NC
1523 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1524
afc16117
KW
1525 if (c < 256) {
1526 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1527 }
1528
afc16117 1529 uvchr_to_utf8(p, c);
968c5e6a 1530 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1531}
1532
84afefe6 1533UV
51910141 1534Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1535{
51910141 1536 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1537 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1538 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1539 *
1540 * Not to be used for locale folds
51910141 1541 */
f673fad4 1542
a1dde8de
KW
1543 UV converted;
1544
1545 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1546 PERL_UNUSED_CONTEXT;
a1dde8de 1547
1ca267a5
KW
1548 assert (! (flags & FOLD_FLAGS_LOCALE));
1549
a1dde8de
KW
1550 if (c == MICRO_SIGN) {
1551 converted = GREEK_SMALL_LETTER_MU;
1552 }
51910141 1553 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1554
1555 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1556 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1557 * under those circumstances. */
1558 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1559 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1560 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1561 p, *lenp, U8);
1562 return LATIN_SMALL_LETTER_LONG_S;
1563 }
1564 else {
4f489194
KW
1565 *(p)++ = 's';
1566 *p = 's';
1567 *lenp = 2;
1568 return 's';
1ca267a5 1569 }
a1dde8de
KW
1570 }
1571 else { /* In this range the fold of all other characters is their lower
1572 case */
1573 converted = toLOWER_LATIN1(c);
1574 }
1575
6f2d5cbc 1576 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1577 *p = (U8) converted;
1578 *lenp = 1;
1579 }
1580 else {
1581 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1582 *p = UTF8_TWO_BYTE_LO(converted);
1583 *lenp = 2;
1584 }
1585
1586 return converted;
1587}
1588
1589UV
31f05a37 1590Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1591{
4b593389 1592
a0270393
KW
1593 /* Not currently externally documented, and subject to change
1594 * <flags> bits meanings:
1595 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1596 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1597 * locale are to be used.
a0270393
KW
1598 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1599 */
4b593389 1600
36bb2ab6 1601 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1602
2f306ab9 1603 /* Treat a UTF-8 locale as not being in locale at all */
31f05a37
KW
1604 if (IN_UTF8_CTYPE_LOCALE) {
1605 flags &= ~FOLD_FLAGS_LOCALE;
1606 }
1607
a1dde8de 1608 if (c < 256) {
a0270393 1609 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1610 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1611 /* It is illegal for the fold to cross the 255/256 boundary under
1612 * locale; in this case return the original */
1613 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1614 ? c
1615 : result;
a1dde8de
KW
1616 }
1617
2f306ab9 1618 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1619 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1620 uvchr_to_utf8(p, c);
1621 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1622 }
1623 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1624 the special flags. */
1625 U8 utf8_c[UTF8_MAXBYTES + 1];
1626 uvchr_to_utf8(utf8_c, c);
445bf929 1627 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1628 }
84afefe6
JH
1629}
1630
26483009 1631PERL_STATIC_INLINE bool
5141f98e 1632S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1633 const char *const swashname, SV* const invlist)
bde6a22d 1634{
ea317ccb
KW
1635 /* returns a boolean giving whether or not the UTF8-encoded character that
1636 * starts at <p> is in the swash indicated by <swashname>. <swash>
1637 * contains a pointer to where the swash indicated by <swashname>
1638 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1639 * look at <*swash> and only generate a swash if it is not null. <invlist>
1640 * is NULL or an inversion list that defines the swash. If not null, it
1641 * saves time during initialization of the swash.
ea317ccb
KW
1642 *
1643 * Note that it is assumed that the buffer length of <p> is enough to
1644 * contain all the bytes that comprise the character. Thus, <*p> should
1645 * have been checked before this call for mal-formedness enough to assure
1646 * that. */
1647
7918f24d
NC
1648 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1649
492a624f 1650 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1651 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1652 * as far as there being enough bytes available in it to accommodate the
1653 * character without reading beyond the end, and pass that number on to the
1654 * validating routine */
6302f837 1655 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1656 if (ckWARN_d(WARN_UTF8)) {
1657 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1658 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1659 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1660 what the malformation is */
1661 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1662 }
1663 }
1664 return FALSE;
1665 }
87367d5f
KW
1666 if (!*swash) {
1667 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1668 *swash = _core_swash_init("utf8",
1669
1670 /* Only use the name if there is no inversion
1671 * list; otherwise will go out to disk */
1672 (invlist) ? "" : swashname,
1673
1674 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1675 }
28123549 1676
bde6a22d
NC
1677 return swash_fetch(*swash, p, TRUE) != 0;
1678}
1679
1680bool
922e8cb4
KW
1681Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1682{
922e8cb4
KW
1683 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1684
1685 assert(classnum < _FIRST_NON_SWASH_CC);
1686
f25ce844
KW
1687 return is_utf8_common(p,
1688 &PL_utf8_swash_ptrs[classnum],
1689 swash_property_names[classnum],
1690 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1691}
1692
1693bool
f2645549 1694Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1695{
f2645549 1696 SV* invlist = NULL;
7918f24d 1697
f2645549 1698 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1699
f2645549
KW
1700 if (! PL_utf8_perl_idstart) {
1701 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1702 }
60071a22 1703 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1704}
1705
1706bool
f2645549 1707Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1708{
f2645549 1709 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1710
1711 if (*p == '_')
1712 return TRUE;
f25ce844 1713 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1714}
1715
1716bool
eba68aa0
KW
1717Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1718{
b24b43f7 1719 SV* invlist = NULL;
eba68aa0
KW
1720
1721 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1722
b24b43f7
KW
1723 if (! PL_utf8_perl_idcont) {
1724 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1725 }
60071a22 1726 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1727}
1728
eba68aa0 1729bool
f2645549 1730Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1731{
f2645549 1732 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1733
f25ce844 1734 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1735}
1736
1737bool
f2645549 1738Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1739{
f2645549 1740 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1741
f25ce844 1742 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1743}
1744
1745bool
7dbf68d2
KW
1746Perl__is_utf8_mark(pTHX_ const U8 *p)
1747{
7dbf68d2
KW
1748 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1749
f25ce844 1750 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1751}
1752
6b5c0936 1753/*
87cea99e 1754=for apidoc to_utf8_case
6b5c0936 1755
6fae5207 1756C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1757the character that is being converted. This routine assumes that the character
1758at C<p> is well-formed.
6b5c0936 1759
6fae5207
KW
1760C<ustrp> is a pointer to the character buffer to put the
1761conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1762of the result.
1763
6fae5207 1764C<swashp> is a pointer to the swash to use.
6b5c0936 1765
a1433954 1766Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1767and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1768but not always, a multicharacter mapping), is tried first.
6b5c0936 1769
4a8240a3
KW
1770C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1771any special mappings; C<""> means to use the special mappings. Values other
1772than these two are treated as the name of the hash containing the special
1773mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1774
6fae5207 1775C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1776%utf8::ToLower.
1777
1778=cut */
6b5c0936 1779
2104c8d9 1780UV
9a957fbc
AL
1781Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1782 SV **swashp, const char *normal, const char *special)
a0ed51b3 1783{
0134edef 1784 STRLEN len = 0;
f4cd282c 1785 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1786
1787 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1788
9ae3ac1a
KW
1789 /* Note that swash_fetch() doesn't output warnings for these because it
1790 * assumes we will */
8457b38f 1791 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1792 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1793 if (ckWARN_d(WARN_SURROGATE)) {
1794 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1795 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1796 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1797 }
9ae3ac1a
KW
1798 }
1799 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1800 if (ckWARN_d(WARN_NON_UNICODE)) {
1801 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1802 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1803 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1804 }
9ae3ac1a
KW
1805 }
1806
1807 /* Note that non-characters are perfectly legal, so no warning should
1808 * be given */
1809 }
1810
0134edef 1811 if (!*swashp) /* load on-demand */
5ab9d2ef 1812 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1813
a6f87d8c 1814 if (special) {
0134edef 1815 /* It might be "special" (sometimes, but not always,
2a37f04d 1816 * a multicharacter mapping) */
4a8240a3 1817 HV *hv = NULL;
b08cf34e
JH
1818 SV **svp;
1819
4a8240a3
KW
1820 /* If passed in the specials name, use that; otherwise use any
1821 * given in the swash */
1822 if (*special != '\0') {
1823 hv = get_hv(special, 0);
1824 }
1825 else {
1826 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1827 if (svp) {
1828 hv = MUTABLE_HV(SvRV(*svp));
1829 }
1830 }
1831
176fe009
KW
1832 if (hv
1833 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1834 && (*svp))
1835 {
cfd0369c 1836 const char *s;
47654450 1837
cfd0369c 1838 s = SvPV_const(*svp, len);
47654450 1839 if (len == 1)
f4cd282c 1840 /* EIGHTBIT */
c80e42f3 1841 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1842 else {
d2dcd0fb 1843 Copy(s, ustrp, len, U8);
29e98929 1844 }
983ffd37 1845 }
0134edef
JH
1846 }
1847
1848 if (!len && *swashp) {
f4cd282c 1849 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1850
0134edef
JH
1851 if (uv2) {
1852 /* It was "normal" (a single character mapping). */
f4cd282c 1853 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1854 }
1855 }
1feea2c7 1856
cbe07460
KW
1857 if (len) {
1858 if (lenp) {
1859 *lenp = len;
1860 }
1861 return valid_utf8_to_uvchr(ustrp, 0);
1862 }
1863
1864 /* Here, there was no mapping defined, which means that the code point maps
1865 * to itself. Return the inputs */
bfdf22ec 1866 len = UTF8SKIP(p);
ca9fab46
KW
1867 if (p != ustrp) { /* Don't copy onto itself */
1868 Copy(p, ustrp, len, U8);
1869 }
0134edef 1870
2a37f04d
JH
1871 if (lenp)
1872 *lenp = len;
1873
f4cd282c 1874 return uv1;
cbe07460 1875
a0ed51b3
LW
1876}
1877
051a06d4 1878STATIC UV
ab0b796c 1879S_check_locale_boundary_crossing(pTHX_ const char * const func_name, const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4
KW
1880{
1881 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1882 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1883 * result contains a character that crosses the 255/256 boundary, disallow
1884 * the change, and return the original code point. See L<perlfunc/lc> for
1885 * why;
051a06d4 1886 *
a1433954
KW
1887 * p points to the original string whose case was changed; assumed
1888 * by this routine to be well-formed
051a06d4
KW
1889 * result the code point of the first character in the changed-case string
1890 * ustrp points to the changed-case string (<result> represents its first char)
1891 * lenp points to the length of <ustrp> */
1892
1893 UV original; /* To store the first code point of <p> */
1894
1895 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1896
a4f12ed7 1897 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1898
1899 /* We know immediately if the first character in the string crosses the
1900 * boundary, so can skip */
1901 if (result > 255) {
1902
1903 /* Look at every character in the result; if any cross the
1904 * boundary, the whole thing is disallowed */
1905 U8* s = ustrp + UTF8SKIP(ustrp);
1906 U8* e = ustrp + *lenp;
1907 while (s < e) {
a4f12ed7 1908 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1909 goto bad_crossing;
1910 }
1911 s += UTF8SKIP(s);
1912 }
1913
1914 /* Here, no characters crossed, result is ok as-is */
1915 return result;
1916 }
1917
1918bad_crossing:
1919
1920 /* Failed, have to return the original */
4b88fb76 1921 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1922
1923 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1924 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1925 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1926 "resolved to \"\\x{%"UVXf"}\".",
1927 func_name,
1928 original,
1929 original);
051a06d4
KW
1930 Copy(p, ustrp, *lenp, char);
1931 return original;
1932}
1933
d3e79532 1934/*
87cea99e 1935=for apidoc to_utf8_upper
d3e79532 1936
1f607577 1937Instead use L</toUPPER_utf8>.
a1433954 1938
d3e79532
JH
1939=cut */
1940
051a06d4 1941/* Not currently externally documented, and subject to change:
31f05a37
KW
1942 * <flags> is set iff iff the rules from the current underlying locale are to
1943 * be used. */
051a06d4 1944
2104c8d9 1945UV
31f05a37 1946Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1947{
051a06d4
KW
1948 UV result;
1949
1950 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1951
31f05a37
KW
1952 if (flags && IN_UTF8_CTYPE_LOCALE) {
1953 flags = FALSE;
1954 }
1955
3a4c58c9 1956 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1957 if (flags) {
1958 result = toUPPER_LC(*p);
1959 }
1960 else {
81c6c7ce 1961 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1962 }
3a4c58c9
KW
1963 }
1964 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1965 if (flags) {
a6d8b88b 1966 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1967 result = toUPPER_LC(c);
051a06d4
KW
1968 }
1969 else {
94bb8c36 1970 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1971 ustrp, lenp, 'S');
051a06d4
KW
1972 }
1973 }
1974 else { /* utf8, ord above 255 */
1975 result = CALL_UPPER_CASE(p, ustrp, lenp);
1976
1977 if (flags) {
ab0b796c 1978 result = check_locale_boundary_crossing("uc", p, result, ustrp, lenp);
051a06d4
KW
1979 }
1980 return result;
1981 }
1982
1983 /* Here, used locale rules. Convert back to utf8 */
1984 if (UTF8_IS_INVARIANT(result)) {
1985 *ustrp = (U8) result;
1986 *lenp = 1;
1987 }
1988 else {
62cb07ea
KW
1989 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1990 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1991 *lenp = 2;
3a4c58c9 1992 }
baa60164 1993
051a06d4 1994 return result;
983ffd37 1995}
a0ed51b3 1996
d3e79532 1997/*
87cea99e 1998=for apidoc to_utf8_title
d3e79532 1999
1f607577 2000Instead use L</toTITLE_utf8>.
a1433954 2001
d3e79532
JH
2002=cut */
2003
051a06d4 2004/* Not currently externally documented, and subject to change:
31f05a37
KW
2005 * <flags> is set iff the rules from the current underlying locale are to be
2006 * used. Since titlecase is not defined in POSIX, for other than a
2007 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2008 */
051a06d4 2009
983ffd37 2010UV
31f05a37 2011Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2012{
051a06d4
KW
2013 UV result;
2014
2015 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2016
31f05a37
KW
2017 if (flags && IN_UTF8_CTYPE_LOCALE) {
2018 flags = FALSE;
2019 }
2020
3a4c58c9 2021 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2022 if (flags) {
2023 result = toUPPER_LC(*p);
2024 }
2025 else {
81c6c7ce 2026 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2027 }
3a4c58c9
KW
2028 }
2029 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2030 if (flags) {
a6d8b88b 2031 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2032 result = toUPPER_LC(c);
051a06d4
KW
2033 }
2034 else {
94bb8c36 2035 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2036 ustrp, lenp, 's');
051a06d4
KW
2037 }
2038 }
2039 else { /* utf8, ord above 255 */
2040 result = CALL_TITLE_CASE(p, ustrp, lenp);
2041
2042 if (flags) {
ab0b796c 2043 result = check_locale_boundary_crossing("ucfirst", p, result, ustrp, lenp);
051a06d4
KW
2044 }
2045 return result;
2046 }
2047
2048 /* Here, used locale rules. Convert back to utf8 */
2049 if (UTF8_IS_INVARIANT(result)) {
2050 *ustrp = (U8) result;
2051 *lenp = 1;
2052 }
2053 else {
62cb07ea
KW
2054 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2055 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2056 *lenp = 2;
3a4c58c9
KW
2057 }
2058
051a06d4 2059 return result;
a0ed51b3
LW
2060}
2061
d3e79532 2062/*
87cea99e 2063=for apidoc to_utf8_lower
d3e79532 2064
1f607577 2065Instead use L</toLOWER_utf8>.
a1433954 2066
d3e79532
JH
2067=cut */
2068
051a06d4 2069/* Not currently externally documented, and subject to change:
31f05a37
KW
2070 * <flags> is set iff iff the rules from the current underlying locale are to
2071 * be used.
2072 */
051a06d4 2073
2104c8d9 2074UV
31f05a37 2075Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2076{
051a06d4
KW
2077 UV result;
2078
051a06d4 2079 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2080
31f05a37
KW
2081 if (flags && IN_UTF8_CTYPE_LOCALE) {
2082 flags = FALSE;
2083 }
2084
968c5e6a 2085 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2086 if (flags) {
2087 result = toLOWER_LC(*p);
2088 }
2089 else {
81c6c7ce 2090 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2091 }
968c5e6a
KW
2092 }
2093 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2094 if (flags) {
a6d8b88b 2095 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2096 result = toLOWER_LC(c);
051a06d4
KW
2097 }
2098 else {
94bb8c36 2099 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2100 ustrp, lenp);
051a06d4 2101 }
968c5e6a 2102 }
051a06d4
KW
2103 else { /* utf8, ord above 255 */
2104 result = CALL_LOWER_CASE(p, ustrp, lenp);
2105
2106 if (flags) {
ab0b796c 2107 result = check_locale_boundary_crossing("lc", p, result, ustrp, lenp);
051a06d4 2108 }
968c5e6a 2109
051a06d4
KW
2110 return result;
2111 }
2112
2113 /* Here, used locale rules. Convert back to utf8 */
2114 if (UTF8_IS_INVARIANT(result)) {
2115 *ustrp = (U8) result;
2116 *lenp = 1;
2117 }
2118 else {
62cb07ea
KW
2119 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2120 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2121 *lenp = 2;
2122 }
2123
051a06d4 2124 return result;
b4e400f9
JH
2125}
2126
d3e79532 2127/*
87cea99e 2128=for apidoc to_utf8_fold
d3e79532 2129
1f607577 2130Instead use L</toFOLD_utf8>.
a1433954 2131
d3e79532
JH
2132=cut */
2133
051a06d4
KW
2134/* Not currently externally documented, and subject to change,
2135 * in <flags>
31f05a37
KW
2136 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2137 * locale are to be used.
051a06d4
KW
2138 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2139 * otherwise simple folds
a0270393
KW
2140 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2141 * prohibited
445bf929 2142 */
36bb2ab6 2143
b4e400f9 2144UV
445bf929 2145Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2146{
051a06d4
KW
2147 UV result;
2148
36bb2ab6 2149 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2150
a0270393
KW
2151 /* These are mutually exclusive */
2152 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2153
50ba90ff
KW
2154 assert(p != ustrp); /* Otherwise overwrites */
2155
31f05a37
KW
2156 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2157 flags &= ~FOLD_FLAGS_LOCALE;
2158 }
2159
a1dde8de 2160 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2161 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2162 result = toFOLD_LC(*p);
051a06d4
KW
2163 }
2164 else {
81c6c7ce 2165 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2166 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2167 }
a1dde8de
KW
2168 }
2169 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2170 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2171 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2172 result = toFOLD_LC(c);
051a06d4
KW
2173 }
2174 else {
94bb8c36 2175 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2176 ustrp, lenp,
1ca267a5 2177 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2178 }
a1dde8de 2179 }
051a06d4 2180 else { /* utf8, ord above 255 */
a0270393 2181 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2182
1ca267a5
KW
2183 if (flags & FOLD_FLAGS_LOCALE) {
2184
76f2ffcd
KW
2185# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2186# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
2187
2188 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
2189 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2190
538e84ed
KW
2191 /* Special case these two characters, as what normally gets
2192 * returned under locale doesn't work */
76f2ffcd
KW
2193 if (UTF8SKIP(p) == cap_sharp_s_len
2194 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2195 {
ab0b796c
KW
2196 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2197 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2198 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2199 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2200 goto return_long_s;
2201 }
76f2ffcd
KW
2202 else if (UTF8SKIP(p) == long_s_t_len
2203 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2204 {
ab0b796c
KW
2205 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2206 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2207 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2208 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2209 goto return_ligature_st;
2210 }
ab0b796c 2211 return check_locale_boundary_crossing("fc", p, result, ustrp, lenp);
051a06d4 2212 }
a0270393
KW
2213 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2214 return result;
2215 }
2216 else {
2217 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2218 * character above the ASCII range, and the result should not
2219 * contain an ASCII character. */
a0270393
KW
2220
2221 UV original; /* To store the first code point of <p> */
2222
2223 /* Look at every character in the result; if any cross the
2224 * boundary, the whole thing is disallowed */
2225 U8* s = ustrp;
2226 U8* e = ustrp + *lenp;
2227 while (s < e) {
2228 if (isASCII(*s)) {
2229 /* Crossed, have to return the original */
2230 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2231
9fc2026f 2232 /* But in these instances, there is an alternative we can
1ca267a5 2233 * return that is valid */
9fc2026f
KW
2234 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2235 || original == LATIN_SMALL_LETTER_SHARP_S)
2236 {
1ca267a5
KW
2237 goto return_long_s;
2238 }
9fc2026f
KW
2239 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2240 goto return_ligature_st;
2241 }
a0270393
KW
2242 Copy(p, ustrp, *lenp, char);
2243 return original;
2244 }
2245 s += UTF8SKIP(s);
2246 }
051a06d4 2247
a0270393
KW
2248 /* Here, no characters crossed, result is ok as-is */
2249 return result;
2250 }
051a06d4
KW
2251 }
2252
2253 /* Here, used locale rules. Convert back to utf8 */
2254 if (UTF8_IS_INVARIANT(result)) {
2255 *ustrp = (U8) result;
2256 *lenp = 1;
2257 }
2258 else {
62cb07ea
KW
2259 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2260 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2261 *lenp = 2;
2262 }
2263
051a06d4 2264 return result;
1ca267a5
KW
2265
2266 return_long_s:
2267 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2268 * folds to a string of two of these characters. By returning this
2269 * instead, then, e.g.,
2270 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2271 * works. */
2272
2273 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2274 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2275 ustrp, *lenp, U8);
2276 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2277
2278 return_ligature_st:
2279 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2280 * have the other one fold to it */
2281
2282 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2283 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2284 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2285}
2286
711a919c 2287/* Note:
f90a9a02 2288 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2289 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2290 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2291 */
c4a5db0c 2292
a0ed51b3 2293SV*
7fc63493 2294Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2295{
c4a5db0c
KW
2296 PERL_ARGS_ASSERT_SWASH_INIT;
2297
2298 /* Returns a copy of a swash initiated by the called function. This is the
2299 * public interface, and returning a copy prevents others from doing
2300 * mischief on the original */
2301
5d3d13d1 2302 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2303}
2304
2305SV*
5d3d13d1 2306Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2307{
2c1f00b9
YO
2308
2309 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2310 * use the following define */
2311
2312#define CORE_SWASH_INIT_RETURN(x) \
2313 PL_curpm= old_PL_curpm; \
2314 return x
2315
c4a5db0c 2316 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2317 * by calling utf8_heavy.pl in the general case. The returned value may be
2318 * the swash's inversion list instead if the input parameters allow it.
2319 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2320 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2321 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2322 *
2323 * This interface should only be used by functions that won't destroy or
2324 * adversely change the swash, as doing so affects all other uses of the
2325 * swash in the program; the general public should use 'Perl_swash_init'
2326 * instead.
2327 *
2328 * pkg is the name of the package that <name> should be in.
2329 * name is the name of the swash to find. Typically it is a Unicode
2330 * property name, including user-defined ones
2331 * listsv is a string to initialize the swash with. It must be of the form
2332 * documented as the subroutine return value in
2333 * L<perlunicode/User-Defined Character Properties>
2334 * minbits is the number of bits required to represent each data element.
2335 * It is '1' for binary properties.
2336 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2337 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2338 * flags_p if non-NULL is the address of various input and output flag bits
2339 * to the routine, as follows: ('I' means is input to the routine;
2340 * 'O' means output from the routine. Only flags marked O are
2341 * meaningful on return.)
2342 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2343 * came from a user-defined property. (I O)
5d3d13d1
KW
2344 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2345 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2346 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2347 * return of an inversion list instead of a swash hash if this routine
2348 * thinks that would result in faster execution of swash_fetch() later
2349 * on. (I)
9a53f6cf
KW
2350 *
2351 * Thus there are three possible inputs to find the swash: <name>,
2352 * <listsv>, and <invlist>. At least one must be specified. The result
2353 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2354 * actions can intersect, etc. what <name> gives. To avoid going out to
2355 * disk at all, <invlist> should specify completely what the swash should
2356 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2357 *
2358 * <invlist> is only valid for binary properties */
c4a5db0c 2359
2c1f00b9
YO
2360 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2361
c4a5db0c 2362 SV* retval = &PL_sv_undef;
83199d38 2363 HV* swash_hv = NULL;
87367d5f
KW
2364 const int invlist_swash_boundary =
2365 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2366 ? 512 /* Based on some benchmarking, but not extensive, see commit
2367 message */
2368 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2369
2370 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2371 assert(! invlist || minbits == 1);
2372
2c1f00b9
YO
2373 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2374 that triggered the swash init and the swash init perl logic itself.
2375 See perl #122747 */
2376
9a53f6cf
KW
2377 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2378 * so */
2379 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2380 dSP;
2381 const size_t pkg_len = strlen(pkg);
2382 const size_t name_len = strlen(name);
2383 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2384 SV* errsv_save;
2385 GV *method;
2386
2387 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2388
2389 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2390 ENTER;
69794297 2391 SAVEHINTS();
650f067c
JL
2392 /* We might get here via a subroutine signature which uses a utf8
2393 * parameter name, at which point PL_subname will have been set
2394 * but not yet used. */
2395 save_item(PL_subname);
69794297
KW
2396 if (PL_parser && PL_parser->error_count)
2397 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2398 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2399 if (!method) { /* demand load utf8 */
2400 ENTER;
db2c6cb3
FC
2401 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2402 GvSV(PL_errgv) = NULL;
1a419e6b 2403#ifndef NO_TAINT_SUPPORT
69794297
KW
2404 /* It is assumed that callers of this routine are not passing in
2405 * any user derived data. */
284167a5
SM
2406 SAVEBOOL(TAINT_get);
2407 TAINT_NOT;
2408#endif
69794297
KW
2409 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2410 NULL);
eed484f9 2411 {
db2c6cb3
FC
2412 /* Not ERRSV, as there is no need to vivify a scalar we are
2413 about to discard. */
2414 SV * const errsv = GvSV(PL_errgv);
2415 if (!SvTRUE(errsv)) {
2416 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2417 SvREFCNT_dec(errsv);
2418 }
eed484f9 2419 }
69794297
KW
2420 LEAVE;
2421 }
2422 SPAGAIN;
2423 PUSHMARK(SP);
2424 EXTEND(SP,5);
2425 mPUSHp(pkg, pkg_len);
2426 mPUSHp(name, name_len);
2427 PUSHs(listsv);
2428 mPUSHi(minbits);
2429 mPUSHi(none);
2430 PUTBACK;
db2c6cb3
FC
2431 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2432 GvSV(PL_errgv) = NULL;
69794297
KW
2433 /* If we already have a pointer to the method, no need to use
2434 * call_method() to repeat the lookup. */
c41800a8
KW
2435 if (method
2436 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2437 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2438 {
2439 retval = *PL_stack_sp--;
2440 SvREFCNT_inc(retval);
2441 }
eed484f9 2442 {
db2c6cb3
FC
2443 /* Not ERRSV. See above. */
2444 SV * const errsv = GvSV(PL_errgv);
2445 if (!SvTRUE(errsv)) {
2446 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2447 SvREFCNT_dec(errsv);
2448 }
eed484f9 2449 }
ce3b816e 2450 LEAVE;
69794297
KW
2451 POPSTACK;
2452 if (IN_PERL_COMPILETIME) {
2453 CopHINTS_set(PL_curcop, PL_hints);
2454 }
2455 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2456 if (SvPOK(retval))
2457
2458 /* If caller wants to handle missing properties, let them */
5d3d13d1 2459 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2460 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2461 }
2462 Perl_croak(aTHX_
2463 "Can't find Unicode property definition \"%"SVf"\"",
2464 SVfARG(retval));
a25b5927 2465 NOT_REACHED; /* NOTREACHED */
69794297 2466 }
9a53f6cf 2467 } /* End of calling the module to find the swash */
36eb48b4 2468
83199d38
KW
2469 /* If this operation fetched a swash, and we will need it later, get it */
2470 if (retval != &PL_sv_undef
2471 && (minbits == 1 || (flags_p
2472 && ! (*flags_p
2473 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2474 {
2475 swash_hv = MUTABLE_HV(SvRV(retval));
2476
2477 /* If we don't already know that there is a user-defined component to
2478 * this swash, and the user has indicated they wish to know if there is
2479 * one (by passing <flags_p>), find out */
2480 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2481 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2482 if (user_defined && SvUV(*user_defined)) {
2483 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2484 }
2485 }
2486 }
2487
36eb48b4
KW
2488 /* Make sure there is an inversion list for binary properties */
2489 if (minbits == 1) {
2490 SV** swash_invlistsvp = NULL;
2491 SV* swash_invlist = NULL;
9a53f6cf 2492 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2493 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2494 an unclaimed reference count */
36eb48b4 2495
9a53f6cf 2496 /* If this operation fetched a swash, get its already existing
83199d38 2497 * inversion list, or create one for it */
36eb48b4 2498
83199d38 2499 if (swash_hv) {
5c9f4bd2 2500 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2501 if (swash_invlistsvp) {
2502 swash_invlist = *swash_invlistsvp;
2503 invlist_in_swash_is_valid = TRUE;
2504 }
2505 else {
36eb48b4 2506 swash_invlist = _swash_to_invlist(retval);
02c85471 2507 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2508 }
2509 }
2510
2511 /* If an inversion list was passed in, have to include it */
2512 if (invlist) {
2513
2514 /* Any fetched swash will by now have an inversion list in it;
2515 * otherwise <swash_invlist> will be NULL, indicating that we
2516 * didn't fetch a swash */
2517 if (swash_invlist) {
2518
2519 /* Add the passed-in inversion list, which invalidates the one
2520 * already stored in the swash */
2521 invlist_in_swash_is_valid = FALSE;
2522 _invlist_union(invlist, swash_invlist, &swash_invlist);
2523 }
2524 else {
2525
87367d5f
KW
2526 /* Here, there is no swash already. Set up a minimal one, if
2527 * we are going to return a swash */
2528 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2529 swash_hv = newHV();
4aca0fe6 2530 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2531 }
9a53f6cf
KW
2532 swash_invlist = invlist;
2533 }
9a53f6cf
KW
2534 }
2535
2536 /* Here, we have computed the union of all the passed-in data. It may
2537 * be that there was an inversion list in the swash which didn't get
538e84ed 2538 * touched; otherwise save the computed one */
87367d5f
KW
2539 if (! invlist_in_swash_is_valid
2540 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2541 {
5c9f4bd2 2542 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2543 {
2544 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2545 }
cc34d8c5
FC
2546 /* We just stole a reference count. */
2547 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2548 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2549 }
87367d5f 2550
dbfdbd26
KW
2551 SvREADONLY_on(swash_invlist);
2552
c41800a8 2553 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2554 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2555 SvREFCNT_dec(retval);
02c85471
FC
2556 if (!swash_invlist_unclaimed)
2557 SvREFCNT_inc_simple_void_NN(swash_invlist);
2558 retval = newRV_noinc(swash_invlist);
87367d5f 2559 }
36eb48b4
KW
2560 }
2561
2c1f00b9
YO
2562 CORE_SWASH_INIT_RETURN(retval);
2563#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2564}
2565
035d37be
JH
2566
2567/* This API is wrong for special case conversions since we may need to
2568 * return several Unicode characters for a single Unicode character
2569 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2570 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2571 * multiple values. --jhi
2572 * For those, you should use to_utf8_case() instead */
b0e3252e 2573/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2574
2575/* Note:
2576 * Returns the value of property/mapping C<swash> for the first character
2577 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2578 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2579 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2580 *
2581 * A "swash" is a hash which contains initially the keys/values set up by
2582 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2583 * property for all possible code points. Things are stored in a compact form
2584 * (see utf8_heavy.pl) so that calculation is required to find the actual
2585 * property value for a given code point. As code points are looked up, new
2586 * key/value pairs are added to the hash, so that the calculation doesn't have
2587 * to ever be re-done. Further, each calculation is done, not just for the
2588 * desired one, but for a whole block of code points adjacent to that one.
2589 * For binary properties on ASCII machines, the block is usually for 64 code
2590 * points, starting with a code point evenly divisible by 64. Thus if the
2591 * property value for code point 257 is requested, the code goes out and
2592 * calculates the property values for all 64 code points between 256 and 319,
2593 * and stores these as a single 64-bit long bit vector, called a "swatch",
2594 * under the key for code point 256. The key is the UTF-8 encoding for code
2595 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2596 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2597 * for code point 258 is then requested, this code realizes that it would be
2598 * stored under the key for 256, and would find that value and extract the
2599 * relevant bit, offset from 256.
2600 *
2601 * Non-binary properties are stored in as many bits as necessary to represent
2602 * their values (32 currently, though the code is more general than that), not
2603 * as single bits, but the principal is the same: the value for each key is a
2604 * vector that encompasses the property values for all code points whose UTF-8
2605 * representations are represented by the key. That is, for all code points
2606 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2607 * bytes of that.
680c470c 2608 */
a0ed51b3 2609UV
680c470c 2610Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2611{
ef8f7699 2612 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2613 U32 klen;
2614 U32 off;
9b56a019 2615 STRLEN slen = 0;
7d85a32c 2616 STRLEN needents;
cfd0369c 2617 const U8 *tmps = NULL;
979f2922 2618 SV *swatch;
08fb1ac5 2619 const U8 c = *ptr;
3568d838 2620
7918f24d
NC
2621 PERL_ARGS_ASSERT_SWASH_FETCH;
2622
87367d5f
KW
2623 /* If it really isn't a hash, it isn't really swash; must be an inversion
2624 * list */
2625 if (SvTYPE(hv) != SVt_PVHV) {
2626 return _invlist_contains_cp((SV*)hv,
2627 (do_utf8)
2628 ? valid_utf8_to_uvchr(ptr, NULL)
2629 : c);
2630 }
2631
08fb1ac5
KW
2632 /* We store the values in a "swatch" which is a vec() value in a swash
2633 * hash. Code points 0-255 are a single vec() stored with key length
2634 * (klen) 0. All other code points have a UTF-8 representation
2635 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2636 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2637 * length for them is the length of the encoded char - 1. ptr[klen] is the
2638 * final byte in the sequence representing the character */
2639 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2640 klen = 0;
2641 needents = 256;
2642 off = c;
3568d838 2643 }
08fb1ac5
KW
2644 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2645 klen = 0;
2646 needents = 256;
2647 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2648 }
2649 else {
08fb1ac5
KW
2650 klen = UTF8SKIP(ptr) - 1;
2651
2652 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2653 * the vec is the final byte in the sequence. (In EBCDIC this is
2654 * converted to I8 to get consecutive values.) To help you visualize
2655 * all this:
2656 * Straight 1047 After final byte
2657 * UTF-8 UTF-EBCDIC I8 transform
2658 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2659 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2660 * ...
2661 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2662 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2663 * ...
2664 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2665 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2666 * ...
2667 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2668 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2669 * ...
2670 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2671 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2672 *
2673 * (There are no discontinuities in the elided (...) entries.)
2674 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2675 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2676 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2677 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2678 * index into the vec() swatch (after subtracting 0x80, which we
2679 * actually do with an '&').
2680 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2681 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2682 * dicontinuities which go away by transforming it into I8, and we
2683 * effectively subtract 0xA0 to get the index. */
979f2922 2684 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2685 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2686 }
7d85a32c 2687
a0ed51b3
LW
2688 /*
2689 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2690 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2691 * it's nothing to sniff at.) Pity we usually come through at least
2692 * two function calls to get here...
2693 *
2694 * NB: this code assumes that swatches are never modified, once generated!
2695 */
2696
3568d838 2697 if (hv == PL_last_swash_hv &&
a0ed51b3 2698 klen == PL_last_swash_klen &&
27da23d5 2699 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2700 {
2701 tmps = PL_last_swash_tmps;
2702 slen = PL_last_swash_slen;
2703 }
2704 else {
2705 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2706 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2707
b0e3252e 2708 /* If not cached, generate it via swatch_get */
979f2922 2709 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2710 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2711 {
2712 if (klen) {
2713 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2714 swatch = swatch_get(swash,
2715 code_point & ~((UV)needents - 1),
2716 needents);
2717 }
2718 else { /* For the first 256 code points, the swatch has a key of
2719 length 0 */
2720 swatch = swatch_get(swash, 0, needents);
2721 }
979f2922 2722
923e4eb5 2723 if (IN_PERL_COMPILETIME)
623e6609 2724 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2725
979f2922 2726 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2727
979f2922
ST
2728 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2729 || (slen << 3) < needents)
5637ef5b
NC
2730 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2731 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2732 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2733 }
2734
2735 PL_last_swash_hv = hv;
16d8f38a 2736 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2737 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2738 /* FIXME change interpvar.h? */
2739 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2740 PL_last_swash_slen = slen;
2741 if (klen)
2742 Copy(ptr, PL_last_swash_key, klen, U8);
2743 }
2744
9faf8d75 2745 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2746 case 1:
e7aca353 2747 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2748 case 8:
e7aca353 2749 return ((UV) tmps[off]);
a0ed51b3
LW
2750 case 16:
2751 off <<= 1;
e7aca353
JH
2752 return
2753 ((UV) tmps[off ] << 8) +
2754 ((UV) tmps[off + 1]);
a0ed51b3
LW
2755 case 32:
2756 off <<= 2;
e7aca353
JH
2757 return
2758 ((UV) tmps[off ] << 24) +
2759 ((UV) tmps[off + 1] << 16) +
2760 ((UV) tmps[off + 2] << 8) +
2761 ((UV) tmps[off + 3]);
a0ed51b3 2762 }
5637ef5b
NC
2763 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2764 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2765 NORETURN_FUNCTION_END;
a0ed51b3 2766}
2b9d42f0 2767
319009ee
KW
2768/* Read a single line of the main body of the swash input text. These are of
2769 * the form:
2770 * 0053 0056 0073
2771 * where each number is hex. The first two numbers form the minimum and
2772 * maximum of a range, and the third is the value associated with the range.
2773 * Not all swashes should have a third number
2774 *
2775 * On input: l points to the beginning of the line to be examined; it points
2776 * to somewhere in the string of the whole input text, and is
2777 * terminated by a \n or the null string terminator.
2778 * lend points to the null terminator of that string
2779 * wants_value is non-zero if the swash expects a third number
2780 * typestr is the name of the swash's mapping, like 'ToLower'
2781 * On output: *min, *max, and *val are set to the values read from the line.
2782 * returns a pointer just beyond the line examined. If there was no
2783 * valid min number on the line, returns lend+1
2784 */
2785
2786STATIC U8*
2787S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2788 const bool wants_value, const U8* const typestr)
2789{
2790 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2791 STRLEN numlen; /* Length of the number */
02470786
KW
2792 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2793 | PERL_SCAN_DISALLOW_PREFIX
2794 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2795
2796 /* nl points to the next \n in the scan */
2797 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2798
95543e92
KW
2799 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2800
319009ee
KW
2801 /* Get the first number on the line: the range minimum */
2802 numlen = lend - l;
2803 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2804 *max = *min; /* So can never return without setting max */
319009ee
KW
2805 if (numlen) /* If found a hex number, position past it */
2806 l += numlen;
2807 else if (nl) { /* Else, go handle next line, if any */
2808 return nl + 1; /* 1 is length of "\n" */
2809 }
2810 else { /* Else, no next line */
2811 return lend + 1; /* to LIST's end at which \n is not found */
2812 }
2813
2814 /* The max range value follows, separated by a BLANK */
2815 if (isBLANK(*l)) {
2816 ++l;
02470786
KW
2817 flags = PERL_SCAN_SILENT_ILLDIGIT
2818 | PERL_SCAN_DISALLOW_PREFIX
2819 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2820 numlen = lend - l;
2821 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2822 if (numlen)
2823 l += numlen;
2824 else /* If no value here, it is a single element range */
2825 *max = *min;
2826
2827 /* Non-binary tables have a third entry: what the first element of the
24303724 2828 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2829 if (wants_value) {
2830 if (isBLANK(*l)) {
2831 ++l;
f2a7d0fc
KW
2832 flags = PERL_SCAN_SILENT_ILLDIGIT
2833 | PERL_SCAN_DISALLOW_PREFIX
2834 | PERL_SCAN_SILENT_NON_PORTABLE;
2835 numlen = lend - l;
2836 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2837 if (numlen)
2838 l += numlen;
2839 else
2840 *val = 0;
319009ee
KW
2841 }
2842 else {
2843 *val = 0;
2844 if (typeto) {
dcbac5bb 2845 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2846 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2847 typestr, l);
2848 }
2849 }
2850 }
2851 else
2852 *val = 0; /* bits == 1, then any val should be ignored */
2853 }
2854 else { /* Nothing following range min, should be single element with no
2855 mapping expected */
319009ee
KW
2856 if (wants_value) {
2857 *val = 0;
2858 if (typeto) {
dcbac5bb 2859 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2860 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2861 }
2862 }
2863 else
2864 *val = 0; /* bits == 1, then val should be ignored */
2865 }
2866
2867 /* Position to next line if any, or EOF */
2868 if (nl)
2869 l = nl + 1;
2870 else
2871 l = lend;
2872
2873 return l;
2874}
2875
979f2922
ST
2876/* Note:
2877 * Returns a swatch (a bit vector string) for a code point sequence
2878 * that starts from the value C<start> and comprises the number C<span>.
2879 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2880 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2881 */
2882STATIC SV*
b0e3252e 2883S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2884{
2885 SV *swatch;
77f9f126 2886 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2887 STRLEN lcur, xcur, scur;
ef8f7699 2888 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2889 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2890
88d45d28
KW
2891 SV** listsvp = NULL; /* The string containing the main body of the table */
2892 SV** extssvp = NULL;
2893 SV** invert_it_svp = NULL;
2894 U8* typestr = NULL;
786861f5
KW
2895 STRLEN bits;
2896 STRLEN octets; /* if bits == 1, then octets == 0 */
2897 UV none;
2898 UV end = start + span;
972dd592 2899
36eb48b4 2900 if (invlistsvp == NULL) {
786861f5
KW
2901 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2902 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2903 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2904 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2905 listsvp = hv_fetchs(hv, "LIST", FALSE);
2906 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2907
2908 bits = SvUV(*bitssvp);
2909 none = SvUV(*nonesvp);
2910 typestr = (U8*)SvPV_nolen(*typesvp);
2911 }
36eb48b4
KW
2912 else {
2913 bits = 1;
2914 none = 0;
2915 }
786861f5 2916 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2917
b0e3252e 2918 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2919
979f2922 2920 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2921 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2922 (UV)bits);
979f2922
ST
2923 }
2924
84ea5ef6
KW
2925 /* If overflowed, use the max possible */
2926 if (end < start) {
2927 end = UV_MAX;
2928 span = end - start;
2929 }
2930
979f2922 2931 /* create and initialize $swatch */
979f2922 2932 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2933 swatch = newSV(scur);
2934 SvPOK_on(swatch);
979f2922
ST
2935 s = (U8*)SvPVX(swatch);
2936 if (octets && none) {
0bd48802 2937 const U8* const e = s + scur;
979f2922
ST
2938 while (s < e) {
2939 if (bits == 8)
2940 *s++ = (U8)(none & 0xff);
2941 else if (bits == 16) {
2942 *s++ = (U8)((none >> 8) & 0xff);
2943 *s++ = (U8)( none & 0xff);
2944 }
2945 else if (bits == 32) {
2946 *s++ = (U8)((none >> 24) & 0xff);
2947 *s++ = (U8)((none >> 16) & 0xff);
2948 *s++ = (U8)((none >> 8) & 0xff);
2949 *s++ = (U8)( none & 0xff);
2950 }
2951 }
2952 *s = '\0';
2953 }
2954 else {
2955 (void)memzero((U8*)s, scur + 1);
2956 }
2957 SvCUR_set(swatch, scur);
2958 s = (U8*)SvPVX(swatch);
2959
36eb48b4
KW
2960 if (invlistsvp) { /* If has an inversion list set up use that */
2961 _invlist_populate_swatch(*invlistsvp, start, end, s);
2962 return swatch;
2963 }
2964
2965 /* read $swash->{LIST} */
979f2922
ST
2966 l = (U8*)SvPV(*listsvp, lcur);
2967 lend = l + lcur;
2968 while (l < lend) {
8ed25d53 2969 UV min, max, val, upper;
95543e92
KW
2970 l = swash_scan_list_line(l, lend, &min, &max, &val,
2971 cBOOL(octets), typestr);
319009ee 2972 if (l > lend) {
979f2922
ST
2973 break;
2974 }
2975
972dd592 2976 /* If looking for something beyond this range, go try the next one */
979f2922
ST
2977 if (max < start)
2978 continue;
2979
8ed25d53
KW
2980 /* <end> is generally 1 beyond where we want to set things, but at the
2981 * platform's infinity, where we can't go any higher, we want to
2982 * include the code point at <end> */
2983 upper = (max < end)
2984 ? max
2985 : (max != UV_MAX || end != UV_MAX)
2986 ? end - 1
2987 : end;
2988
979f2922 2989 if (octets) {
35da51f7 2990 UV key;
979f2922
ST
2991 if (min < start) {
2992 if (!none || val < none) {
2993 val += start - min;
2994 }
2995 min = start;
2996 }
8ed25d53 2997 for (key = min; key <= upper; key++) {
979f2922 2998 STRLEN offset;
979f2922
ST
2999 /* offset must be non-negative (start <= min <= key < end) */
3000 offset = octets * (key - start);
3001 if (bits == 8)
3002 s[offset] = (U8)(val & 0xff);
3003 else if (bits == 16) {
3004 s[offset ] = (U8)((val >> 8) & 0xff);
3005 s[offset + 1] = (U8)( val & 0xff);
3006 }
3007 else if (bits == 32) {
3008 s[offset ] = (U8)((val >> 24) & 0xff);
3009 s[offset + 1] = (U8)((val >> 16) & 0xff);
3010 s[offset + 2] = (U8)((val >> 8) & 0xff);
3011 s[offset + 3] = (U8)( val & 0xff);
3012 }
3013
3014 if (!none || val < none)
3015 ++val;
3016 }
3017 }
711a919c 3018 else { /* bits == 1, then val should be ignored */
35da51f7 3019 UV key;
979f2922
ST
3020 if (min < start)
3021 min = start;
6cb05c12 3022
8ed25d53 3023 for (key = min; key <= upper; key++) {
0bd48802 3024 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3025 s[offset >> 3] |= 1 << (offset & 7);
3026 }
3027 }
3028 } /* while */
979f2922 3029
9479a769 3030 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3031 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3032
3033 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3034 * be 0, and their inversion should also be 0, as we don't succeed any
3035 * Unicode property matches for non-Unicode code points */
3036 if (start <= PERL_UNICODE_MAX) {
3037
3038 /* The code below assumes that we never cross the
3039 * Unicode/above-Unicode boundary in a range, as otherwise we would
3040 * have to figure out where to stop flipping the bits. Since this
3041 * boundary is divisible by a large power of 2, and swatches comes
3042 * in small powers of 2, this should be a valid assumption */
3043 assert(start + span - 1 <= PERL_UNICODE_MAX);
3044
507a8485
KW
3045 send = s + scur;
3046 while (s < send) {
3047 *s = ~(*s);
3048 s++;
3049 }
0bda3001 3050 }
77f9f126
KW
3051 }
3052
d73c39c5
KW
3053 /* read $swash->{EXTRAS}
3054 * This code also copied to swash_to_invlist() below */
979f2922
ST
3055 x = (U8*)SvPV(*extssvp, xcur);
3056 xend = x + xcur;
3057 while (x < xend) {
3058 STRLEN namelen;
3059 U8 *namestr;
3060 SV** othersvp;
3061 HV* otherhv;
3062 STRLEN otherbits;
3063 SV **otherbitssvp, *other;
711a919c 3064 U8 *s, *o, *nl;
979f2922
ST
3065 STRLEN slen, olen;
3066
35da51f7 3067 const U8 opc = *x++;
979f2922
ST
3068 if (opc == '\n')
3069 continue;
3070
3071 nl = (U8*)memchr(x, '\n', xend - x);
3072
3073 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3074 if (nl) {
3075 x = nl + 1; /* 1 is length of "\n" */
3076 continue;
3077 }
3078 else {
3079 x = xend; /* to EXTRAS' end at which \n is not found */
3080 break;
3081 }
3082 }
3083
3084 namestr = x;
3085 if (nl) {
3086 namelen = nl - namestr;
3087 x = nl + 1;
3088 }
3089 else {
3090 namelen = xend - namestr;
3091 x = xend;
3092 }
3093
3094 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3095 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3096 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3097 otherbits = (STRLEN)SvUV(*otherbitssvp);
3098 if (bits < otherbits)
5637ef5b
NC
3099 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3100 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3101
3102 /* The "other" swatch must be destroyed after. */
b0e3252e 3103 other = swatch_get(*othersvp, start, span);
979f2922
ST
3104 o = (U8*)SvPV(other, olen);
3105
3106 if (!olen)
b0e3252e 3107 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3108
3109 s = (U8*)SvPV(swatch, slen);
3110 if (bits == 1 && otherbits == 1) {
3111 if (slen != olen)
5637ef5b
NC
3112 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3113 "mismatch, slen=%"UVuf", olen=%"UVuf,
3114 (UV)slen, (UV)olen);
979f2922
ST
3115
3116 switch (opc) {
3117 case '+':
3118 while (slen--)
3119 *s++ |= *o++;
3120 break;
3121 case '!':
3122 while (slen--)
3123 *s++ |= ~*o++;
3124 break;
3125 case '-':
3126 while (slen--)
3127 *s++ &= ~*o++;
3128 break;
3129 case '&':
3130 while (slen--)
3131 *s++ &= *o++;
3132 break;
3133 default:
3134 break;
3135 }
3136 }
711a919c 3137 else {
979f2922
ST
3138 STRLEN otheroctets = otherbits >> 3;
3139 STRLEN offset = 0;
35da51f7 3140 U8* const send = s + slen;
979f2922
ST
3141
3142 while (s < send) {
3143 UV otherval = 0;
3144
3145 if (otherbits == 1) {
3146 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3147 ++offset;
3148 }
3149 else {
3150 STRLEN vlen = otheroctets;
3151 otherval = *o++;
3152 while (--vlen) {
3153 otherval <<= 8;
3154 otherval |= *o++;
3155 }
3156 }
3157
711a919c 3158 if (opc == '+' && otherval)
6f207bd3 3159 NOOP; /* replace with otherval */
979f2922
ST
3160 else if (opc == '!' && !otherval)
3161 otherval = 1;
3162 else if (opc == '-' && otherval)
3163 otherval = 0;
3164 else if (opc == '&' && !otherval)
3165 otherval = 0;
3166 else {
711a919c 3167 s += octets; /* no replacement */
979f2922
ST
3168 continue;
3169 }
3170
3171 if (bits == 8)
3172 *s++ = (U8)( otherval & 0xff);
3173 else if (bits == 16) {
3174 *s++ = (U8)((otherval >> 8) & 0xff);
3175 *s++ = (U8)( otherval & 0xff);
3176 }
3177 else if (bits == 32) {
3178 *s++ = (U8)((otherval >> 24) & 0xff);
3179 *s++ = (U8)((otherval >> 16) & 0xff);
3180 *s++ = (U8)((otherval >> 8) & 0xff);
3181 *s++ = (U8)( otherval & 0xff);
3182 }
3183 }
3184 }
3185 sv_free(other); /* through with it! */
3186 } /* while */
3187 return swatch;
3188}
3189
064c021d 3190HV*
4c2e1131 3191Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3192{
3193
79a2a0e8 3194 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3195 * Can't be used on a property that is subject to user override, as it
3196 * relies on the value of SPECIALS in the swash which would be set by
3197 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3198 * for overridden properties
064c021d
KW
3199 *
3200 * Returns a hash which is the inversion and closure of a swash mapping.
3201 * For example, consider the input lines:
3202 * 004B 006B
3203 * 004C 006C
3204 * 212A 006B
3205 *
3206 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3207 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3208 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3209 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3210 *
538e84ed
KW
3211 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3212 * keys are only code points that are folded-to, so it isn't a full closure.
3213 *
064c021d
KW
3214 * Essentially, for any code point, it gives all the code points that map to
3215 * it, or the list of 'froms' for that point.
3216 *
5662e334
KW
3217 * Currently it ignores any additions or deletions from other swashes,
3218 * looking at just the main body of the swash, and if there are SPECIALS
3219 * in the swash, at that hash
3220 *
3221 * The specials hash can be extra code points, and most likely consists of
3222 * maps from single code points to multiple ones (each expressed as a string
3223 * of utf8 characters). This function currently returns only 1-1 mappings.
3224 * However consider this possible input in the specials hash:
3225 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3226 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3227 *
3228 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3229 * currently handle. But it also means that FB05 and FB06 are equivalent in
3230 * a 1-1 mapping which we should handle, and this relationship may not be in
3231 * the main table. Therefore this function examines all the multi-char
3232 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3233
3234 U8 *l, *lend;
3235 STRLEN lcur;
3236 HV *const hv = MUTABLE_HV(SvRV(swash));
3237
923b6d4e
KW
3238 /* The string containing the main body of the table. This will have its
3239 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3240 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3241
3242 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3243 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3244 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3245 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3246 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3247 const STRLEN bits = SvUV(*bitssvp);
3248 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3249 const UV none = SvUV(*nonesvp);
5662e334 3250 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3251
3252 HV* ret = newHV();
3253
3254 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3255
3256 /* Must have at least 8 bits to get the mappings */
3257 if (bits != 8 && bits != 16 && bits != 32) {
3258 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3259 (UV)bits);
3260 }
3261
5662e334
KW
3262 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3263 mapping to more than one character */
3264
3265 /* Construct an inverse mapping hash for the specials */
3266 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3267 HV * specials_inverse = newHV();
3268 char *char_from; /* the lhs of the map */
3269 I32 from_len; /* its byte length */
3270 char *char_to; /* the rhs of the map */
3271 I32 to_len; /* its byte length */
3272 SV *sv_to; /* and in a sv */
3273 AV* from_list; /* list of things that map to each 'to' */
3274
3275 hv_iterinit(specials_hv);
3276
3277 /* The keys are the characters (in utf8) that map to the corresponding
3278 * utf8 string value. Iterate through the list creating the inverse
3279 * list. */
3280 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3281 SV** listp;
3282 if (! SvPOK(sv_to)) {
5637ef5b
NC
3283 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3284 "unexpectedly is not a string, flags=%lu",
3285 (unsigned long)SvFLAGS(sv_to));
5662e334 3286 }
4b88fb76 3287 /*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
3288
3289 /* Each key in the inverse list is a mapped-to value, and the key's
3290 * hash value is a list of the strings (each in utf8) that map to
3291 * it. Those strings are all one character long */
3292 if ((listp = hv_fetch(specials_inverse,
3293 SvPVX(sv_to),
3294 SvCUR(sv_to), 0)))
3295 {
3296 from_list = (AV*) *listp;
3297 }
3298 else { /* No entry yet for it: create one */
3299 from_list = newAV();
3300 if (! hv_store(specials_inverse,
3301 SvPVX(sv_to),
3302 SvCUR(sv_to),
3303 (SV*) from_list, 0))
3304 {
3305 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3306 }
3307 }
3308
3309 /* Here have the list associated with this 'to' (perhaps newly
3310 * created and empty). Just add to it. Note that we ASSUME that
3311 * the input is guaranteed to not have duplications, so we don't
3312 * check for that. Duplications just slow down execution time. */
3313 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3314 }
3315
3316 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3317 * it looking for cases like the FB05/FB06 examples above. There would
3318 * be an entry in the hash like
3319 * 'st' => [ FB05, FB06 ]
3320 * In this example we will create two lists that get stored in the
3321 * returned hash, 'ret':
3322 * FB05 => [ FB05, FB06 ]
3323 * FB06 => [ FB05, FB06 ]
3324 *
3325 * Note that there is nothing to do if the array only has one element.
3326 * (In the normal 1-1 case handled below, we don't have to worry about
3327 * two lists, as everything gets tied to the single list that is
3328 * generated for the single character 'to'. But here, we are omitting
3329 * that list, ('st' in the example), so must have multiple lists.) */
3330 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3331 &char_to, &to_len)))
3332 {
b9f2b683 3333 if (av_tindex(from_list) > 0) {
c70927a6 3334 SSize_t i;
5662e334
KW
3335
3336 /* We iterate over all combinations of i,j to place each code
3337 * point on each list */
b9f2b683 3338 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3339 SSize_t j;
5662e334
KW
3340 AV* i_list = newAV();
3341 SV** entryp = av_fetch(from_list, i, FALSE);
3342 if (entryp == NULL) {
3343 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3344 }
3345 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3346 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3347 }
3348 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3349 (SV*) i_list, FALSE))
3350 {
3351 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3352 }
3353
538e84ed 3354 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3355 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3356 entryp = av_fetch(from_list, j, FALSE);
3357 if (entryp == NULL) {
3358 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3359 }
3360
3361 /* When i==j this adds itself to the list */
4b88fb76
KW
3362 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3363 (U8*) SvPVX(*entryp),
3364 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3365 0)));
4637d003 3366 /*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
3367 }
3368 }
3369 }
3370 }
3371 SvREFCNT_dec(specials_inverse); /* done with it */
3372 } /* End of specials */
3373
064c021d
KW
3374 /* read $swash->{LIST} */
3375 l = (U8*)SvPV(*listsvp, lcur);
3376 lend = l + lcur;
3377
3378 /* Go through each input line */
3379 while (l < lend) {
3380 UV min, max, val;
3381 UV inverse;
95543e92
KW
3382 l = swash_scan_list_line(l, lend, &min, &max, &val,
3383 cBOOL(octets), typestr);
064c021d
KW
3384 if (l > lend) {
3385 break;
3386 }
3387
3388 /* Each element in the range is to be inverted */
3389 for (inverse = min; inverse <= max; inverse++) {
3390 AV* list;
064c021d
KW
3391 SV** listp;
3392 IV i;
3393 bool found_key = FALSE;
5662e334 3394 bool found_inverse = FALSE;
064c021d
KW
3395
3396 /* The key is the inverse mapping */
3397 char key[UTF8_MAXBYTES+1];
c80e42f3 3398 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3399 STRLEN key_len = key_end - key;
3400
064c021d
KW
3401 /* Get the list for the map */
3402 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3403 list = (AV*) *listp;
3404 }
3405 else { /* No entry yet for it: create one */
3406 list = newAV();
3407 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3408 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3409 }
3410 }
3411
5662e334
KW
3412 /* Look through list to see if this inverse mapping already is
3413 * listed, or if there is a mapping to itself already */
b9f2b683 3414 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3415 SV** entryp = av_fetch(list, i, FALSE);
3416 SV* entry;
414db8a1 3417 UV uv;
064c021d
KW
3418 if (entryp == NULL) {
3419 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3420 }
3421 entry = *entryp;
414db8a1
DM
3422 uv = SvUV(entry);
3423 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3424 if (uv == val) {
064c021d 3425 found_key = TRUE;
5662e334 3426 }
414db8a1 3427 if (uv == inverse) {
5662e334
KW
3428 found_inverse = TRUE;
3429 }
3430
3431 /* No need to continue searching if found everything we are
3432 * looking for */
3433 if (found_key && found_inverse) {
064c021d
KW
3434 break;
3435 }
3436 }
56ca34ca
KW
3437
3438 /* Make sure there is a mapping to itself on the list */
064c021d 3439 if (! found_key) {
d397ff6a 3440 av_push(list, newSVuv(val));
4637d003 3441 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3442 }
3443
3444
3445 /* Simply add the value to the list */
5662e334
KW
3446 if (! found_inverse) {
3447 av_push(list, newSVuv(inverse));
4637d003 3448 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3449 }
064c021d 3450
b0e3252e 3451 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3452 * range. That makes more compact tables possible. You can
3453 * express the capitalization, for example, of all consecutive
3454 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3455 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3456 * and it's not documented; it appears to be used only in
b0e3252e 3457 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3458 * in case */
064c021d
KW
3459 if (!none || val < none) {
3460 ++val;
3461 }
3462 }
3463 }
3464
3465 return ret;
3466}
3467
a25abddc 3468SV*
d764b54e
KW
3469Perl__swash_to_invlist(pTHX_ SV* const swash)
3470{
3471
ed92f1b3
KW
3472 /* Subject to change or removal. For use only in one place in regcomp.c.
3473 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3474
3475 U8 *l, *lend;
3476 char *loc;
3477 STRLEN lcur;
3478 HV *const hv = MUTABLE_HV(SvRV(swash));
3479 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3480 U8 empty[] = "";
923b6d4e
KW
3481 SV** listsvp;
3482 SV** typesvp;
3483 SV** bitssvp;
3484 SV** extssvp;
3485 SV** invert_it_svp;
d764b54e 3486
923b6d4e
KW
3487 U8* typestr;
3488 STRLEN bits;
3489 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3490 U8 *x, *xend;
3491 STRLEN xcur;
d764b54e 3492
a25abddc 3493 SV* invlist;
d764b54e 3494
b81740c0
KW
3495 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3496
923b6d4e
KW
3497 /* If not a hash, it must be the swash's inversion list instead */
3498 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3499 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3500 }
3501
3502 /* The string containing the main body of the table */
3503 listsvp = hv_fetchs(hv, "LIST", FALSE);
3504 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3505 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3506 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3507 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3508
3509 typestr = (U8*)SvPV_nolen(*typesvp);
3510 bits = SvUV(*bitssvp);
3511 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3512
d764b54e 3513 /* read $swash->{LIST} */
b443038a
KW
3514 if (SvPOK(*listsvp)) {
3515 l = (U8*)SvPV(*listsvp, lcur);
3516 }
3517 else {
3518 /* LIST legitimately doesn't contain a string during compilation phases
3519 * of Perl itself, before the Unicode tables are generated. In this
3520 * case, just fake things up by creating an empty list */
3521 l = empty;
3522 lcur = 0;
3523 }
d764b54e
KW
3524 loc = (char *) l;
3525 lend = l + lcur;
3526
31aa6e0b 3527 if (*l == 'V') { /* Inversion list format */
99944878 3528 const char *after_atou = (char *) lend;
31aa6e0b
KW
3529 UV element0;
3530 UV* other_elements_ptr;
3531
3532 /* The first number is a count of the rest */
3533 l++;
99944878 3534 elements = grok_atou((const char *)l, &after_atou);
eb092534
KW
3535 if (elements == 0) {
3536 invlist = _new_invlist(0);
3537 }
3538 else {
99944878
JH
3539 while (isSPACE(*l)) l++;
3540 l = (U8 *) after_atou;
1f9f7d4c
KW
3541
3542 /* Get the 0th element, which is needed to setup the inversion list */
99944878
JH
3543 while (isSPACE(*l)) l++;
3544 element0 = (UV) grok_atou((const char *)l, &after_atou);
3545 l = (U8 *) after_atou;
1f9f7d4c
KW
3546 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3547 elements--;
3548
3549 /* Then just populate the rest of the input */
3550 while (elements-- > 0) {
3551 if (l > lend) {
3552 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3553 }
99944878
JH
3554 while (isSPACE(*l)) l++;
3555 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3556 l = (U8 *) after_atou;
1f9f7d4c 3557 }
eb092534 3558 }
31aa6e0b
KW
3559 }
3560 else {
3561
1784d2f9
KW
3562 /* Scan the input to count the number of lines to preallocate array
3563 * size based on worst possible case, which is each line in the input
3564 * creates 2 elements in the inversion list: 1) the beginning of a
3565 * range in the list; 2) the beginning of a range not in the list. */
3566 while ((loc = (strchr(loc, '\n'))) != NULL) {
3567 elements += 2;
3568 loc++;
3569 }
d764b54e 3570
1784d2f9
KW
3571 /* If the ending is somehow corrupt and isn't a new line, add another
3572 * element for the final range that isn't in the inversion list */
3573 if (! (*lend == '\n'
3574 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3575 {
3576 elements++;
3577 }
d764b54e 3578
1784d2f9 3579 invlist = _new_invlist(elements);
d764b54e 3580
1784d2f9
KW
3581 /* Now go through the input again, adding each range to the list */
3582 while (l < lend) {
3583 UV start, end;
3584 UV val; /* Not used by this function */
d764b54e 3585
95543e92
KW
3586 l = swash_scan_list_line(l, lend, &start, &end, &val,
3587 cBOOL(octets), typestr);
d764b54e 3588
1784d2f9
KW
3589 if (l > lend) {
3590 break;
3591 }
3592
3593 invlist = _add_range_to_invlist(invlist, start, end);
3594 }
31aa6e0b 3595 }
d764b54e 3596
77f9f126
KW
3597 /* Invert if the data says it should be */
3598 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3599 _invlist_invert(invlist);
77f9f126
KW
3600 }
3601
b0e3252e 3602 /* This code is copied from swatch_get()
d73c39c5
KW
3603 * read $swash->{EXTRAS} */
3604 x = (U8*)SvPV(*extssvp, xcur);
3605 xend = x + xcur;
3606 while (x < xend) {
3607 STRLEN namelen;
3608 U8 *namestr;
3609 SV** othersvp;
3610 HV* otherhv;
3611 STRLEN otherbits;
3612 SV **otherbitssvp, *other;
3613 U8 *nl;
3614
3615 const U8 opc = *x++;
3616 if (opc == '\n')
3617 continue;
3618
3619 nl = (U8*)memchr(x, '\n', xend - x);
3620
3621 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3622 if (nl) {
3623 x = nl + 1; /* 1 is length of "\n" */
3624 continue;
3625 }
3626 else {
3627 x = xend; /* to EXTRAS' end at which \n is not found */
3628 break;
3629 }
3630 }
3631
3632 namestr = x;
3633 if (nl) {
3634 namelen = nl - namestr;
3635 x = nl + 1;
3636 }
3637 else {
3638 namelen = xend - namestr;
3639 x = xend;
3640 }
3641
3642 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3643 otherhv = MUTABLE_HV(SvRV(*othersvp));
3644 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3645 otherbits = (STRLEN)SvUV(*otherbitssvp);
3646
3647 if (bits != otherbits || bits != 1) {
5637ef5b
NC
3648 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3649 "properties, bits=%"UVuf", otherbits=%"UVuf,
3650 (UV)bits, (UV)otherbits);
d73c39c5
KW
3651 }
3652
3653 /* The "other" swatch must be destroyed after. */
3654 other = _swash_to_invlist((SV *)*othersvp);
3655
b0e3252e 3656 /* End of code copied from swatch_get() */
d73c39c5
KW
3657 switch (opc) {
3658 case '+':
3659 _invlist_union(invlist, other, &invlist);
3660 break;
3661 case '!':
6c46377d 3662 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
d73c39c5
KW
3663 break;
3664 case '-':
3665 _invlist_subtract(invlist, other, &invlist);
3666 break;
3667 case '&':
3668 _invlist_intersection(invlist, other, &invlist);
3669 break;
3670 default:
3671 break;
3672 }
3673 sv_free(other); /* through with it! */
3674 }
3675
dbfdbd26 3676 SvREADONLY_on(invlist);
d764b54e
KW
3677 return invlist;
3678}
3679
3fdfee00
KW