This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Was taking sizeof() wrong thing
[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/*
52=for apidoc is_ascii_string
53
a1433954 54Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
55or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
56is, if they are invariant. On ASCII-ish machines, only ASCII characters
57fit this definition, hence the function's name.
eaf7a4d2 58
75200dff
KW
59If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
60use this option, that C<s> can't have embedded C<NUL> characters and has to
61have a terminating C<NUL> byte).
9f7e3d64 62
a1433954 63See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
64
65=cut
66*/
67
68bool
668b6d8d 69Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
70{
71 const U8* const send = s + (len ? len : strlen((const char *)s));
72 const U8* x = s;
73
74 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
75
76 for (; x < send; ++x) {
77 if (!UTF8_IS_INVARIANT(*x))
78 break;
79 }
80
81 return x == send;
82}
83
84/*
378516de 85=for apidoc uvoffuni_to_utf8_flags
eebe1485 86
a27992cc 87THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
88Instead, B<Almost all code should use L</uvchr_to_utf8> or
89L</uvchr_to_utf8_flags>>.
a27992cc 90
de69f3af
KW
91This function is like them, but the input is a strict Unicode
92(as opposed to native) code point. Only in very rare circumstances should code
93not be using the native code point.
949cf498 94
de69f3af 95For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 96
eebe1485
SC
97=cut
98*/
99
dfe13c55 100U8 *
378516de 101Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 102{
378516de 103 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 104
d9432125
KW
105 if (UNI_IS_INVARIANT(uv)) {
106 *d++ = (U8) LATIN1_TO_NATIVE(uv);
107 return d;
108 }
109
9415f659
KW
110#ifdef EBCDIC
111 /* Not representable in UTF-EBCDIC */
112 flags |= UNICODE_DISALLOW_FE_FF;
113#endif
114
979f77b6
KW
115 /* The first problematic code point is the first surrogate */
116 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 117 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 118 {
949cf498
KW
119 if (UNICODE_IS_SURROGATE(uv)) {
120 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 121 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
122 "UTF-16 surrogate U+%04"UVXf, uv);
123 }
124 if (flags & UNICODE_DISALLOW_SURROGATE) {
125 return NULL;
126 }
127 }
128 else if (UNICODE_IS_SUPER(uv)) {
129 if (flags & UNICODE_WARN_SUPER
130 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
131 {
8457b38f 132 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
133 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
134 }
135 if (flags & UNICODE_DISALLOW_SUPER
136 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
137 {
9415f659
KW
138#ifdef EBCDIC
139 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
140 assert(0);
141#endif
949cf498
KW
142 return NULL;
143 }
144 }
145 else if (UNICODE_IS_NONCHAR(uv)) {
146 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 147 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
148 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
149 uv);
150 }
151 if (flags & UNICODE_DISALLOW_NONCHAR) {
152 return NULL;
153 }
154 }
507b9800 155 }
d9432125 156
2d331972 157#if defined(EBCDIC)
d9432125 158 {
5aaebcb3 159 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
160 U8 *p = d+len-1;
161 while (p > d) {
bc3632a8 162 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
163 uv >>= UTF_ACCUMULATION_SHIFT;
164 }
bc3632a8 165 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
166 return d+len;
167 }
168#else /* Non loop style */
a0ed51b3 169 if (uv < 0x800) {
eb160463
GS
170 *d++ = (U8)(( uv >> 6) | 0xc0);
171 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
172 return d;
173 }
174 if (uv < 0x10000) {
eb160463
GS
175 *d++ = (U8)(( uv >> 12) | 0xe0);
176 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
177 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
178 return d;
179 }
180 if (uv < 0x200000) {
eb160463
GS
181 *d++ = (U8)(( uv >> 18) | 0xf0);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
185 return d;
186 }
187 if (uv < 0x4000000) {
eb160463
GS
188 *d++ = (U8)(( uv >> 24) | 0xf8);
189 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
192 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
193 return d;
194 }
195 if (uv < 0x80000000) {
eb160463
GS
196 *d++ = (U8)(( uv >> 30) | 0xfc);
197 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
198 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
199 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
201 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
202 return d;
203 }
6588300d 204#ifdef UTF8_QUAD_MAX
d7578b48 205 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
206#endif
207 {
eb160463
GS
208 *d++ = 0xfe; /* Can't match U+FEFF! */
209 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
210 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
211 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
214 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
215 return d;
216 }
6588300d 217#ifdef UTF8_QUAD_MAX
a0ed51b3 218 {
eb160463
GS
219 *d++ = 0xff; /* Can't match U+FFFE! */
220 *d++ = 0x80; /* 6 Reserved bits */
221 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
222 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
223 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
224 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
231 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
232 return d;
233 }
234#endif
537124e4 235#endif /* Non loop style */
a0ed51b3 236}
646ca15d 237/*
07693fe6
KW
238=for apidoc uvchr_to_utf8
239
bcb1a2d4 240Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
241of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
242C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
243the byte after the end of the new character. In other words,
07693fe6
KW
244
245 d = uvchr_to_utf8(d, uv);
246
247is the recommended wide native character-aware way of saying
248
249 *(d++) = uv;
250
de69f3af
KW
251This function accepts any UV as input. To forbid or warn on non-Unicode code
252points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
253
07693fe6
KW
254=cut
255*/
256
de69f3af
KW
257/* This is also a macro */
258PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
259
07693fe6
KW
260U8 *
261Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
262{
de69f3af 263 return uvchr_to_utf8(d, uv);
07693fe6
KW
264}
265
de69f3af
KW
266/*
267=for apidoc uvchr_to_utf8_flags
268
269Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
270of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
271C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
272the byte after the end of the new character. In other words,
de69f3af
KW
273
274 d = uvchr_to_utf8_flags(d, uv, flags);
275
276or, in most cases,
277
278 d = uvchr_to_utf8_flags(d, uv, 0);
279
280This is the Unicode-aware way of saying
281
282 *(d++) = uv;
283
284This function will convert to UTF-8 (and not warn) even code points that aren't
285legal Unicode or are problematic, unless C<flags> contains one or more of the
286following flags:
287
288If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
289the function will raise a warning, provided UTF8 warnings are enabled. If instead
290UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
291If both flags are set, the function will both warn and return NULL.
292
4c3cfd5d 293The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 294affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 295UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
296code points that are
297above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
298even less portable) can be warned and/or disallowed even if other above-Unicode
299code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
300flags.
301
302And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
303above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
304DISALLOW flags.
305
306=cut
307*/
308
309/* This is also a macro */
310PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
311
07693fe6
KW
312U8 *
313Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
314{
de69f3af 315 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
316}
317
318/*
492a624f
KW
319=for apidoc is_utf8_char_buf
320
3cedd9d9 321This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
322
323=cut */
324
325STRLEN
326Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
327{
328
492a624f
KW
329 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
330
6302f837 331 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
332}
333
334/*
87cea99e 335=for apidoc is_utf8_string
6662521e 336
a1433954 337Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 338UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
339using C<strlen(s)> (which means if you use this option, that C<s> can't have
340embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
341that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 342
a1433954 343See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 344
6662521e
GS
345=cut
346*/
347
8e84507e 348bool
668b6d8d 349Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 350{
35da51f7 351 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 352 const U8* x = s;
067a85ef 353
7918f24d 354 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 355
6662521e 356 while (x < send) {
6302f837
KW
357 STRLEN len = isUTF8_CHAR(x, send);
358 if (UNLIKELY(! len)) {
359 return FALSE;
360 }
361 x += len;
6662521e 362 }
768c67ee 363
067a85ef 364 return TRUE;
6662521e
GS
365}
366
67e989fb 367/*
814fafa7
NC
368Implemented as a macro in utf8.h
369
87cea99e 370=for apidoc is_utf8_string_loc
814fafa7 371
a1433954
KW
372Like L</is_utf8_string> but stores the location of the failure (in the
373case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
374"utf8ness success") in the C<ep>.
375
a1433954 376See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 377
87cea99e 378=for apidoc is_utf8_string_loclen
81cd54e3 379
a1433954
KW
380Like L</is_utf8_string>() but stores the location of the failure (in the
381case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
382"utf8ness success") in the C<ep>, and the number of UTF-8
383encoded characters in the C<el>.
384
a1433954 385See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
386
387=cut
388*/
389
390bool
668b6d8d 391Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 392{
35da51f7 393 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 394 const U8* x = s;
3ebfea28 395 STRLEN outlen = 0;
7918f24d
NC
396
397 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 398
81cd54e3 399 while (x < send) {
6302f837
KW
400 STRLEN len = isUTF8_CHAR(x, send);
401 if (UNLIKELY(! len)) {
402 goto out;
403 }
404 x += len;
405 outlen++;
81cd54e3 406 }
768c67ee
JH
407
408 out:
3ebfea28
AL
409 if (el)
410 *el = outlen;
411
768c67ee
JH
412 if (ep)
413 *ep = x;
3ebfea28 414 return (x == send);
81cd54e3
JH
415}
416
417/*
768c67ee 418
de69f3af 419=for apidoc utf8n_to_uvchr
378516de
KW
420
421THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 422Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 423
9041c2e3 424Bottom level UTF-8 decode routine.
de69f3af 425Returns the native code point value of the first character in the string C<s>,
746afd53
KW
426which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
427C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
428the length, in bytes, of that character.
949cf498
KW
429
430The value of C<flags> determines the behavior when C<s> does not point to a
431well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
432zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
433next possible position in C<s> that could begin a non-malformed character.
434Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
435
436Various ALLOW flags can be set in C<flags> to allow (and not warn on)
437individual types of malformations, such as the sequence being overlong (that
438is, when there is a shorter sequence that can express the same code point;
439overlong sequences are expressly forbidden in the UTF-8 standard due to
440potential security issues). Another malformation example is the first byte of
441a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
442flags. For allowed 0 length strings, this function returns 0; for allowed
443overlong sequences, the computed code point is returned; for all other allowed
444malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
445determinable reasonable value.
949cf498
KW
446
447The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
448flags) malformation is found. If this flag is set, the routine assumes that
449the caller will raise a warning, and this function will silently just set
d088425d
KW
450C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
451
75200dff 452Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
453character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
454in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
455first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
456input had an error.
949cf498
KW
457
458Certain code points are considered problematic. These are Unicode surrogates,
746afd53 459Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 460By default these are considered regular code points, but certain situations
5eafe189 461warrant special handling for them. If C<flags> contains
949cf498
KW
462UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
463malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
464UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
465maximum) can be set to disallow these categories individually.
466
467The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
468UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
469for their respective categories, but otherwise the code points are considered
470valid (not malformations). To get a category to both be treated as a
471malformation and raise a warning, specify both the WARN and DISALLOW flags.
472(But note that warnings are not raised if lexically disabled nor if
473UTF8_CHECK_ONLY is also specified.)
474
475Very large code points (above 0x7FFF_FFFF) are considered more problematic than
476the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
477reasons: they requre at least 32 bits to represent them on ASCII platforms, are
478not representable at all on EBCDIC platforms, and the original UTF-8
479specification never went above this number (the current 0x10FFFF limit was
480imposed later). (The smaller ones, those that fit into 32 bits, are
481representable by a UV on ASCII platforms, but not by an IV, which means that
482the number of operations that can be performed on them is quite restricted.)
483The UTF-8 encoding on ASCII platforms for these large code points begins with a
484byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
485be treated as malformations, while allowing smaller above-Unicode code points.
486(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
487including these, as malformations.)
488Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 489the other WARN flags, but applies just to these code points.
949cf498
KW
490
491All other code points corresponding to Unicode characters, including private
492use and those yet to be assigned, are never considered malformed and never
493warn.
67e989fb 494
37607a96
PK
495=cut
496*/
67e989fb 497
a0ed51b3 498UV
de69f3af 499Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 500{
d4c19fe8 501 const U8 * const s0 = s;
eb83ed87 502 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 503 U8 * send;
eb83ed87
KW
504 UV uv = *s;
505 STRLEN expectlen;
949cf498 506 SV* sv = NULL;
eb83ed87
KW
507 UV outlier_ret = 0; /* return value when input is in error or problematic
508 */
509 UV pack_warn = 0; /* Save result of packWARN() for later */
510 bool unexpected_non_continuation = FALSE;
511 bool overflowed = FALSE;
2f8f112e 512 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 513
eb83ed87 514 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 515
de69f3af 516 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 517
eb83ed87
KW
518 /* The order of malformation tests here is important. We should consume as
519 * few bytes as possible in order to not skip any valid character. This is
520 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
521 * http://unicode.org/reports/tr36 for more discussion as to why. For
522 * example, once we've done a UTF8SKIP, we can tell the expected number of
523 * bytes, and could fail right off the bat if the input parameters indicate
524 * that there are too few available. But it could be that just that first
525 * byte is garbled, and the intended character occupies fewer bytes. If we
526 * blindly assumed that the first byte is correct, and skipped based on
527 * that number, we could skip over a valid input character. So instead, we
528 * always examine the sequence byte-by-byte.
529 *
530 * We also should not consume too few bytes, otherwise someone could inject
531 * things. For example, an input could be deliberately designed to
532 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 533 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
534 * which is actually part of of the overflowing sequence), that could look
535 * legitimate to the caller, which could discard the initial partial
536 * sequence and process the rest, inappropriately */
537
538 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 539 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
540 if (retlen) {
541 *retlen = 0;
542 }
a0dbb045 543
eb83ed87
KW
544 if (flags & UTF8_ALLOW_EMPTY) {
545 return 0;
546 }
547 if (! (flags & UTF8_CHECK_ONLY)) {
548 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
549 }
0c443dc2
JH
550 goto malformed;
551 }
552
eb83ed87
KW
553 expectlen = UTF8SKIP(s);
554
555 /* A well-formed UTF-8 character, as the vast majority of calls to this
556 * function will be for, has this expected length. For efficiency, set
557 * things up here to return it. It will be overriden only in those rare
558 * cases where a malformation is found */
559 if (retlen) {
560 *retlen = expectlen;
561 }
562
563 /* An invariant is trivially well-formed */
1d72bdf6 564 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 565 return uv;
a0ed51b3 566 }
67e989fb 567
eb83ed87 568 /* A continuation character can't start a valid sequence */
b5b9af04 569 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
570 if (flags & UTF8_ALLOW_CONTINUATION) {
571 if (retlen) {
572 *retlen = 1;
573 }
574 return UNICODE_REPLACEMENT;
575 }
ba210ebe 576
eb83ed87
KW
577 if (! (flags & UTF8_CHECK_ONLY)) {
578 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
579 }
580 curlen = 1;
ba210ebe
JH
581 goto malformed;
582 }
9041c2e3 583
dcd27b3c
KW
584 /* Here is not a continuation byte, nor an invariant. The only thing left
585 * is a start byte (possibly for an overlong) */
586
1d72bdf6 587#ifdef EBCDIC
bc3632a8 588 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
589#endif
590
eb83ed87
KW
591 /* Remove the leading bits that indicate the number of bytes in the
592 * character's whole UTF-8 sequence, leaving just the bits that are part of
593 * the value */
594 uv &= UTF_START_MASK(expectlen);
ba210ebe 595
eb83ed87
KW
596 /* Now, loop through the remaining bytes in the character's sequence,
597 * accumulating each into the working value as we go. Be sure to not look
598 * past the end of the input string */
0b8d30e8
KW
599 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
600
eb83ed87 601 for (s = s0 + 1; s < send; s++) {
b5b9af04 602 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
603#ifndef EBCDIC /* Can't overflow in EBCDIC */
604 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
605
606 /* The original implementors viewed this malformation as more
607 * serious than the others (though I, khw, don't understand
608 * why, since other malformations also give very very wrong
609 * results), so there is no way to turn off checking for it.
610 * Set a flag, but keep going in the loop, so that we absorb
611 * the rest of the bytes that comprise the character. */
612 overflowed = TRUE;
613 overflow_byte = *s; /* Save for warning message's use */
614 }
615#endif
8850bf83 616 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
617 }
618 else {
619 /* Here, found a non-continuation before processing all expected
620 * bytes. This byte begins a new character, so quit, even if
621 * allowing this malformation. */
622 unexpected_non_continuation = TRUE;
623 break;
624 }
625 } /* End of loop through the character's bytes */
626
627 /* Save how many bytes were actually in the character */
628 curlen = s - s0;
629
630 /* The loop above finds two types of malformations: non-continuation and/or
631 * overflow. The non-continuation malformation is really a too-short
632 * malformation, as it means that the current character ended before it was
633 * expected to (being terminated prematurely by the beginning of the next
634 * character, whereas in the too-short malformation there just are too few
635 * bytes available to hold the character. In both cases, the check below
636 * that we have found the expected number of bytes would fail if executed.)
637 * Thus the non-continuation malformation is really unnecessary, being a
638 * subset of the too-short malformation. But there may be existing
639 * applications that are expecting the non-continuation type, so we retain
640 * it, and return it in preference to the too-short malformation. (If this
641 * code were being written from scratch, the two types might be collapsed
642 * into one.) I, khw, am also giving priority to returning the
643 * non-continuation and too-short malformations over overflow when multiple
644 * ones are present. I don't know of any real reason to prefer one over
645 * the other, except that it seems to me that multiple-byte errors trumps
646 * errors from a single byte */
b5b9af04 647 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
648 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
649 if (! (flags & UTF8_CHECK_ONLY)) {
650 if (curlen == 1) {
651 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
652 }
653 else {
654 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
655 }
656 }
eb83ed87
KW
657 goto malformed;
658 }
659 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
660
661 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
662 * as what the original expectations were. */
663 do_overlong_test = FALSE;
eb83ed87
KW
664 if (retlen) {
665 *retlen = curlen;
666 }
667 }
b5b9af04 668 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
669 if (! (flags & UTF8_ALLOW_SHORT)) {
670 if (! (flags & UTF8_CHECK_ONLY)) {
671 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 672 }
eb83ed87
KW
673 goto malformed;
674 }
675 uv = UNICODE_REPLACEMENT;
2f8f112e 676 do_overlong_test = FALSE;
eb83ed87
KW
677 if (retlen) {
678 *retlen = curlen;
679 }
680 }
681
ea5ced44 682#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 683 if (UNLIKELY(overflowed)) {
eb83ed87 684 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 685 goto malformed;
eb83ed87
KW
686 }
687#endif
688
2f8f112e 689 if (do_overlong_test
5aaebcb3 690 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
691 && ! (flags & UTF8_ALLOW_LONG))
692 {
eb83ed87
KW
693 /* The overlong malformation has lower precedence than the others.
694 * Note that if this malformation is allowed, we return the actual
695 * value, instead of the replacement character. This is because this
696 * value is actually well-defined. */
697 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 698 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
699 }
700 goto malformed;
701 }
702
1a89bb6c 703 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
704 * problematic code point that is not allowed by the input parameters. */
705 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
706 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
707 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
708 {
949cf498 709 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
710
711 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
712 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 713 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 714 && ckWARN_d(WARN_SURROGATE))
eb83ed87 715 {
111d382d 716 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 717 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
718 }
719 if (flags & UTF8_DISALLOW_SURROGATE) {
720 goto disallowed;
721 }
722 }
949cf498 723 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 724 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 725 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 726 {
111d382d 727 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 728 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 729 }
ea5ced44
KW
730#ifndef EBCDIC /* EBCDIC always allows FE, FF */
731
732 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
733 * points. We test for these after the regular SUPER ones, and
734 * before possibly bailing out, so that the more dire warning
735 * overrides the regular one, if applicable */
736 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
737 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
738 {
739 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
740 == UTF8_WARN_FE_FF
741 && ckWARN_d(WARN_UTF8))
742 {
743 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
744 pack_warn = packWARN(WARN_UTF8);
745 }
746 if (flags & UTF8_DISALLOW_FE_FF) {
747 goto disallowed;
748 }
749 }
750#endif
949cf498
KW
751 if (flags & UTF8_DISALLOW_SUPER) {
752 goto disallowed;
753 }
754 }
4190d317
KW
755 else if (UNICODE_IS_NONCHAR(uv)) {
756 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 757 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
758 {
759 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 760 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
761 }
762 if (flags & UTF8_DISALLOW_NONCHAR) {
763 goto disallowed;
764 }
765 }
949cf498 766
eb83ed87 767 if (sv) {
de69f3af
KW
768 outlier_ret = uv; /* Note we don't bother to convert to native,
769 as all the outlier code points are the same
770 in both ASCII and EBCDIC */
eb83ed87
KW
771 goto do_warn;
772 }
773
949cf498
KW
774 /* Here, this is not considered a malformed character, so drop through
775 * to return it */
a0ed51b3 776 }
ba210ebe 777
de69f3af 778 return UNI_TO_NATIVE(uv);
ba210ebe 779
eb83ed87
KW
780 /* There are three cases which get to beyond this point. In all 3 cases:
781 * <sv> if not null points to a string to print as a warning.
782 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
783 * set.
784 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
785 * This is done by initializing it to 0, and changing it only
786 * for case 1).
787 * The 3 cases are:
788 * 1) The input is valid but problematic, and to be warned about. The
789 * return value is the resultant code point; <*retlen> is set to
790 * <curlen>, the number of bytes that comprise the code point.
791 * <pack_warn> contains the result of packWARN() for the warning
792 * types. The entry point for this case is the label <do_warn>;
793 * 2) The input is a valid code point but disallowed by the parameters to
794 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
795 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
796 * comprise the code point. <pack_warn> contains the result of
797 * packWARN() for the warning types. The entry point for this case is
798 * the label <disallowed>.
799 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
800 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
801 * bytes that comprise the malformation. All such malformations are
802 * assumed to be warning type <utf8>. The entry point for this case
803 * is the label <malformed>.
804 */
949cf498 805
ba210ebe
JH
806malformed:
807
eb83ed87
KW
808 if (sv && ckWARN_d(WARN_UTF8)) {
809 pack_warn = packWARN(WARN_UTF8);
810 }
811
812disallowed:
813
fcc8fcf6 814 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 815 if (retlen)
10edeb5d 816 *retlen = ((STRLEN) -1);
ba210ebe
JH
817 return 0;
818 }
819
eb83ed87 820do_warn:
5b311467 821
eb83ed87
KW
822 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
823 if warnings are to be raised. */
f555bc63 824 const char * const string = SvPVX_const(sv);
a0dbb045 825
f555bc63
KW
826 if (PL_op)
827 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
828 else
829 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
830 }
831
eb83ed87
KW
832 if (retlen) {
833 *retlen = curlen;
834 }
ba210ebe 835
eb83ed87 836 return outlier_ret;
a0ed51b3
LW
837}
838
8e84507e 839/*
ec5f19d0
KW
840=for apidoc utf8_to_uvchr_buf
841
842Returns the native code point of the first character in the string C<s> which
843is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 844C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 845
524080c4
KW
846If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
847enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
848NULL) to -1. If those warnings are off, the computed value, if well-defined
849(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
850C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
851the next possible position in C<s> that could begin a non-malformed character.
de69f3af 852See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 853returned.
ec5f19d0
KW
854
855=cut
856*/
857
858
859UV
860Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
861{
ec5f19d0
KW
862 assert(s < send);
863
864 return utf8n_to_uvchr(s, send - s, retlen,
865 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
866}
867
27d6c58a 868/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 869 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 870 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
871
872UV
873Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
874{
010ab96b
KW
875 UV expectlen = UTF8SKIP(s);
876 const U8* send = s + expectlen;
9ff2f0f7 877 UV uv = *s;
3986bb7c 878
27d6c58a 879 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 880 PERL_UNUSED_CONTEXT;
27d6c58a 881
010ab96b
KW
882 if (retlen) {
883 *retlen = expectlen;
884 }
885
886 /* An invariant is trivially returned */
887 if (expectlen == 1) {
9ff2f0f7 888 return uv;
010ab96b
KW
889 }
890
9ff2f0f7
KW
891#ifdef EBCDIC
892 uv = NATIVE_UTF8_TO_I8(uv);
893#endif
894
010ab96b
KW
895 /* Remove the leading bits that indicate the number of bytes, leaving just
896 * the bits that are part of the value */
897 uv &= UTF_START_MASK(expectlen);
898
899 /* Now, loop through the remaining bytes, accumulating each into the
900 * working total as we go. (I khw tried unrolling the loop for up to 4
901 * bytes, but there was no performance improvement) */
902 for (++s; s < send; s++) {
903 uv = UTF8_ACCUMULATE(uv, *s);
904 }
905
3986bb7c 906 return UNI_TO_NATIVE(uv);
010ab96b 907
27d6c58a
KW
908}
909
ec5f19d0 910/*
ec5f19d0
KW
911=for apidoc utf8_to_uvuni_buf
912
de69f3af
KW
913Only in very rare circumstances should code need to be dealing in Unicode
914(as opposed to native) code points. In those few cases, use
915C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
916
917Returns the Unicode (not-native) code point of the first character in the
918string C<s> which
ec5f19d0
KW
919is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
920C<retlen> will be set to the length, in bytes, of that character.
921
524080c4
KW
922If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
923enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
924NULL) to -1. If those warnings are off, the computed value if well-defined (or
925the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
926is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
927next possible position in C<s> that could begin a non-malformed character.
de69f3af 928See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
929
930=cut
931*/
932
933UV
934Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
935{
936 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
937
938 assert(send > s);
939
940 /* Call the low level routine asking for checks */
de69f3af
KW
941 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
942 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
943}
944
b76347f2 945/*
87cea99e 946=for apidoc utf8_length
b76347f2
JH
947
948Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
949Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
950up past C<e>, croaks.
b76347f2
JH
951
952=cut
953*/
954
955STRLEN
35a4481c 956Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
957{
958 STRLEN len = 0;
959
7918f24d
NC
960 PERL_ARGS_ASSERT_UTF8_LENGTH;
961
8850bf83
JH
962 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
963 * the bitops (especially ~) can create illegal UTF-8.
964 * In other words: in Perl UTF-8 is not just for Unicode. */
965
a3b680e6
AL
966 if (e < s)
967 goto warn_and_return;
b76347f2 968 while (s < e) {
4cbf4130 969 s += UTF8SKIP(s);
8e91ec7f
AV
970 len++;
971 }
972
973 if (e != s) {
974 len--;
975 warn_and_return:
9b387841
NC
976 if (PL_op)
977 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
978 "%s in %s", unees, OP_DESC(PL_op));
979 else
61a12c31 980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
981 }
982
983 return len;
984}
985
b06226ff 986/*
87cea99e 987=for apidoc utf8_distance
b06226ff 988
1e54db1a 989Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
990and C<b>.
991
992WARNING: use only if you *know* that the pointers point inside the
993same UTF-8 buffer.
994
37607a96
PK
995=cut
996*/
a0ed51b3 997
02eb7b47 998IV
35a4481c 999Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1000{
7918f24d
NC
1001 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1002
bf1665bc 1003 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1004}
1005
b06226ff 1006/*
87cea99e 1007=for apidoc utf8_hop
b06226ff 1008
8850bf83
JH
1009Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1010forward or backward.
b06226ff
JH
1011
1012WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1013the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1014on the first byte of character or just after the last byte of a character.
b06226ff 1015
37607a96
PK
1016=cut
1017*/
a0ed51b3
LW
1018
1019U8 *
ddeaf645 1020Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1021{
7918f24d
NC
1022 PERL_ARGS_ASSERT_UTF8_HOP;
1023
8850bf83
JH
1024 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1025 * the bitops (especially ~) can create illegal UTF-8.
1026 * In other words: in Perl UTF-8 is not just for Unicode. */
1027
a0ed51b3
LW
1028 if (off >= 0) {
1029 while (off--)
1030 s += UTF8SKIP(s);
1031 }
1032 else {
1033 while (off++) {
1034 s--;
8850bf83
JH
1035 while (UTF8_IS_CONTINUATION(*s))
1036 s--;
a0ed51b3
LW
1037 }
1038 }
4373e329 1039 return (U8 *)s;
a0ed51b3
LW
1040}
1041
6940069f 1042/*
fed3ba5d
NC
1043=for apidoc bytes_cmp_utf8
1044
a1433954 1045Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1046sequence of characters (stored as UTF-8)
1047in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1048equal, -1 or -2 if the first string is less than the second string, +1 or +2
1049if the first string is greater than the second string.
1050
1051-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1052longer string. -2 or +2 is returned if
1053there was a difference between characters
fed3ba5d
NC
1054within the strings.
1055
1056=cut
1057*/
1058
1059int
1060Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1061{
1062 const U8 *const bend = b + blen;
1063 const U8 *const uend = u + ulen;
1064
1065 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1066
1067 while (b < bend && u < uend) {
1068 U8 c = *u++;
1069 if (!UTF8_IS_INVARIANT(c)) {
1070 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1071 if (u < uend) {
1072 U8 c1 = *u++;
1073 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1074 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1075 } else {
1076 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1077 "Malformed UTF-8 character "
1078 "(unexpected non-continuation byte 0x%02x"
1079 ", immediately after start byte 0x%02x)"
1080 /* Dear diag.t, it's in the pod. */
1081 "%s%s", c1, c,
1082 PL_op ? " in " : "",
1083 PL_op ? OP_DESC(PL_op) : "");
1084 return -2;
1085 }
1086 } else {
1087 if (PL_op)
1088 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1089 "%s in %s", unees, OP_DESC(PL_op));
1090 else
61a12c31 1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1092 return -2; /* Really want to return undef :-) */
1093 }
1094 } else {
1095 return -2;
1096 }
1097 }
1098 if (*b != c) {
1099 return *b < c ? -2 : +2;
1100 }
1101 ++b;
1102 }
1103
1104 if (b == bend && u == uend)
1105 return 0;
1106
1107 return b < bend ? +1 : -1;
1108}
1109
1110/*
87cea99e 1111=for apidoc utf8_to_bytes
6940069f 1112
2bbc8d55 1113Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1114Unlike L</bytes_to_utf8>, this over-writes the original string, and
1115updates C<len> to contain the new length.
67e989fb 1116Returns zero on failure, setting C<len> to -1.
6940069f 1117
a1433954 1118If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1119
6940069f
GS
1120=cut
1121*/
1122
1123U8 *
37607a96 1124Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1125{
d4c19fe8
AL
1126 U8 * const save = s;
1127 U8 * const send = s + *len;
6940069f 1128 U8 *d;
246fae53 1129
7918f24d 1130 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1131 PERL_UNUSED_CONTEXT;
7918f24d 1132
1e54db1a 1133 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1134 while (s < send) {
d59937ca
KW
1135 if (! UTF8_IS_INVARIANT(*s)) {
1136 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1137 *len = ((STRLEN) -1);
1138 return 0;
1139 }
1140 s++;
dcad2880 1141 }
d59937ca 1142 s++;
246fae53 1143 }
dcad2880
JH
1144
1145 d = s = save;
6940069f 1146 while (s < send) {
80e0b38f
KW
1147 U8 c = *s++;
1148 if (! UTF8_IS_INVARIANT(c)) {
1149 /* Then it is two-byte encoded */
1150 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1151 s++;
1152 }
1153 *d++ = c;
6940069f
GS
1154 }
1155 *d = '\0';
246fae53 1156 *len = d - save;
6940069f
GS
1157 return save;
1158}
1159
1160/*
87cea99e 1161=for apidoc bytes_from_utf8
f9a63242 1162
2bbc8d55 1163Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1164Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1165the newly-created string, and updates C<len> to contain the new
1166length. Returns the original string if no conversion occurs, C<len>
72d33970 1167is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
11680 if C<s> is converted or consisted entirely of characters that are invariant
1169in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1170
37607a96
PK
1171=cut
1172*/
f9a63242
JH
1173
1174U8 *
e1ec3a88 1175Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1176{
f9a63242 1177 U8 *d;
e1ec3a88
AL
1178 const U8 *start = s;
1179 const U8 *send;
f9a63242
JH
1180 I32 count = 0;
1181
7918f24d 1182 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1183 PERL_UNUSED_CONTEXT;
f9a63242 1184 if (!*is_utf8)
73d840c0 1185 return (U8 *)start;
f9a63242 1186
1e54db1a 1187 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1188 for (send = s + *len; s < send;) {
d59937ca
KW
1189 if (! UTF8_IS_INVARIANT(*s)) {
1190 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1191 return (U8 *)start;
d59937ca
KW
1192 }
1193 count++;
1194 s++;
db42d148 1195 }
d59937ca 1196 s++;
f9a63242
JH
1197 }
1198
35da51f7 1199 *is_utf8 = FALSE;
f9a63242 1200
212542aa 1201 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1202 s = start; start = d;
f9a63242
JH
1203 while (s < send) {
1204 U8 c = *s++;
1a91c45d 1205 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1206 /* Then it is two-byte encoded */
1a91c45d
KW
1207 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1208 s++;
c4d5f83a
NIS
1209 }
1210 *d++ = c;
f9a63242
JH
1211 }
1212 *d = '\0';
1213 *len = d - start;
73d840c0 1214 return (U8 *)start;
f9a63242
JH
1215}
1216
1217/*
87cea99e 1218=for apidoc bytes_to_utf8
6940069f 1219
ff97e5cf
KW
1220Converts a string C<s> of length C<len> bytes from the native encoding into
1221UTF-8.
6662521e 1222Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1223reflect the new length in bytes.
6940069f 1224
75200dff 1225A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1226
1227If you want to convert to UTF-8 from encodings other than
1228the native (Latin1 or EBCDIC),
a1433954 1229see L</sv_recode_to_utf8>().
c9ada85f 1230
497711e7 1231=cut
6940069f
GS
1232*/
1233
c682ebef
FC
1234/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1235 likewise need duplication. */
1236
6940069f 1237U8*
35a4481c 1238Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1239{
35a4481c 1240 const U8 * const send = s + (*len);
6940069f
GS
1241 U8 *d;
1242 U8 *dst;
7918f24d
NC
1243
1244 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1245 PERL_UNUSED_CONTEXT;
6940069f 1246
212542aa 1247 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1248 dst = d;
1249
1250 while (s < send) {
55d09dc8
KW
1251 append_utf8_from_native_byte(*s, &d);
1252 s++;
6940069f
GS
1253 }
1254 *d = '\0';
6662521e 1255 *len = d-dst;
6940069f
GS
1256 return dst;
1257}
1258
a0ed51b3 1259/*
dea0fc0b 1260 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1261 *
1262 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1263 * We optimize for native, for obvious reasons. */
1264
1265U8*
dea0fc0b 1266Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1267{
dea0fc0b
JH
1268 U8* pend;
1269 U8* dstart = d;
1270
7918f24d
NC
1271 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1272
dea0fc0b 1273 if (bytelen & 1)
f5992bc4 1274 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1275
1276 pend = p + bytelen;
1277
a0ed51b3 1278 while (p < pend) {
dea0fc0b
JH
1279 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1280 p += 2;
56d37426
KW
1281 if (UNI_IS_INVARIANT(uv)) {
1282 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1283 continue;
1284 }
56d37426
KW
1285 if (uv <= MAX_UTF8_TWO_BYTE) {
1286 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1287 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1288 continue;
1289 }
46956fad
KW
1290#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1291#define LAST_HIGH_SURROGATE 0xDBFF
1292#define FIRST_LOW_SURROGATE 0xDC00
1293#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1294 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1295 if (p >= pend) {
dea0fc0b 1296 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1297 } else {
1298 UV low = (p[0] << 8) + p[1];
1299 p += 2;
46956fad 1300 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1301 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1302 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1303 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1304 }
46956fad 1305 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1306 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1307 }
56d37426
KW
1308#ifdef EBCDIC
1309 d = uvoffuni_to_utf8_flags(d, uv, 0);
1310#else
a0ed51b3 1311 if (uv < 0x10000) {
eb160463
GS
1312 *d++ = (U8)(( uv >> 12) | 0xe0);
1313 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1314 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1315 continue;
1316 }
1317 else {
eb160463
GS
1318 *d++ = (U8)(( uv >> 18) | 0xf0);
1319 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1320 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1321 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1322 continue;
1323 }
56d37426 1324#endif
a0ed51b3 1325 }
dea0fc0b 1326 *newlen = d - dstart;
a0ed51b3
LW
1327 return d;
1328}
1329
1330/* Note: this one is slightly destructive of the source. */
1331
1332U8*
dea0fc0b 1333Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1334{
1335 U8* s = (U8*)p;
d4c19fe8 1336 U8* const send = s + bytelen;
7918f24d
NC
1337
1338 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1339
e0ea5e2d
NC
1340 if (bytelen & 1)
1341 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1342 (UV)bytelen);
1343
a0ed51b3 1344 while (s < send) {
d4c19fe8 1345 const U8 tmp = s[0];
a0ed51b3
LW
1346 s[0] = s[1];
1347 s[1] = tmp;
1348 s += 2;
1349 }
dea0fc0b 1350 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1351}
1352
922e8cb4
KW
1353bool
1354Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1355{
1356 U8 tmpbuf[UTF8_MAXBYTES+1];
1357 uvchr_to_utf8(tmpbuf, c);
1358 return _is_utf8_FOO(classnum, tmpbuf);
1359}
1360
f9ae8fb6
JD
1361/* Internal function so we can deprecate the external one, and call
1362 this one from other deprecated functions in this file */
1363
f2645549
KW
1364bool
1365Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1366{
f2645549 1367 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1368
1369 if (*p == '_')
1370 return TRUE;
f25ce844 1371 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1372}
1373
5092f92a 1374bool
eba68aa0
KW
1375Perl__is_uni_perl_idcont(pTHX_ UV c)
1376{
1377 U8 tmpbuf[UTF8_MAXBYTES+1];
1378 uvchr_to_utf8(tmpbuf, c);
1379 return _is_utf8_perl_idcont(tmpbuf);
1380}
1381
1382bool
f91dcd13
KW
1383Perl__is_uni_perl_idstart(pTHX_ UV c)
1384{
1385 U8 tmpbuf[UTF8_MAXBYTES+1];
1386 uvchr_to_utf8(tmpbuf, c);
1387 return _is_utf8_perl_idstart(tmpbuf);
1388}
1389
3a4c58c9
KW
1390UV
1391Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1392{
1393 /* We have the latin1-range values compiled into the core, so just use
1394 * those, converting the result to utf8. The only difference between upper
1395 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1396 * either "SS" or "Ss". Which one to use is passed into the routine in
1397 * 'S_or_s' to avoid a test */
1398
1399 UV converted = toUPPER_LATIN1_MOD(c);
1400
1401 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1402
1403 assert(S_or_s == 'S' || S_or_s == 's');
1404
6f2d5cbc 1405 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1406 characters in this range */
3a4c58c9
KW
1407 *p = (U8) converted;
1408 *lenp = 1;
1409 return converted;
1410 }
1411
1412 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1413 * which it maps to one of them, so as to only have to have one check for
1414 * it in the main case */
1415 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1416 switch (c) {
1417 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1418 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1419 break;
1420 case MICRO_SIGN:
1421 converted = GREEK_CAPITAL_LETTER_MU;
1422 break;
1423 case LATIN_SMALL_LETTER_SHARP_S:
1424 *(p)++ = 'S';
1425 *p = S_or_s;
1426 *lenp = 2;
1427 return 'S';
1428 default:
1429 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
a25b5927 1430 assert(0); /* NOTREACHED */
3a4c58c9
KW
1431 }
1432 }
1433
1434 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1435 *p = UTF8_TWO_BYTE_LO(converted);
1436 *lenp = 2;
1437
1438 return converted;
1439}
1440
50bda2c3
KW
1441/* Call the function to convert a UTF-8 encoded character to the specified case.
1442 * Note that there may be more than one character in the result.
1443 * INP is a pointer to the first byte of the input character
1444 * OUTP will be set to the first byte of the string of changed characters. It
1445 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1446 * LENP will be set to the length in bytes of the string of changed characters
1447 *
1448 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1449#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1450#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1451#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1452
1453/* This additionally has the input parameter SPECIALS, which if non-zero will
1454 * cause this to use the SPECIALS hash for folding (meaning get full case
1455 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1456#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1457
84afefe6
JH
1458UV
1459Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1460{
a1433954
KW
1461 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1462 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1463 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1464 * the changed version may be longer than the original character.
1465 *
1466 * The ordinal of the first character of the changed version is returned
1467 * (but note, as explained above, that there may be more.) */
1468
7918f24d
NC
1469 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1470
3a4c58c9
KW
1471 if (c < 256) {
1472 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1473 }
1474
0ebc6274 1475 uvchr_to_utf8(p, c);
3a4c58c9 1476 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1477}
1478
84afefe6
JH
1479UV
1480Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1481{
7918f24d
NC
1482 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1483
3a4c58c9
KW
1484 if (c < 256) {
1485 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1486 }
1487
0ebc6274 1488 uvchr_to_utf8(p, c);
3a4c58c9 1489 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1490}
1491
afc16117 1492STATIC U8
81611534 1493S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1494{
1495 /* We have the latin1-range values compiled into the core, so just use
1496 * those, converting the result to utf8. Since the result is always just
a1433954 1497 * one character, we allow <p> to be NULL */
afc16117
KW
1498
1499 U8 converted = toLOWER_LATIN1(c);
1500
1501 if (p != NULL) {
6f2d5cbc 1502 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1503 *p = converted;
1504 *lenp = 1;
1505 }
1506 else {
430c9760
KW
1507 /* Result is known to always be < 256, so can use the EIGHT_BIT
1508 * macros */
1509 *p = UTF8_EIGHT_BIT_HI(converted);
1510 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1511 *lenp = 2;
1512 }
1513 }
1514 return converted;
1515}
1516
84afefe6
JH
1517UV
1518Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1519{
7918f24d
NC
1520 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1521
afc16117
KW
1522 if (c < 256) {
1523 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1524 }
1525
afc16117 1526 uvchr_to_utf8(p, c);
968c5e6a 1527 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1528}
1529
84afefe6 1530UV
51910141 1531Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1532{
51910141 1533 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1534 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1535 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1536 *
1537 * Not to be used for locale folds
51910141 1538 */
f673fad4 1539
a1dde8de
KW
1540 UV converted;
1541
1542 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1543 PERL_UNUSED_CONTEXT;
a1dde8de 1544
1ca267a5
KW
1545 assert (! (flags & FOLD_FLAGS_LOCALE));
1546
a1dde8de
KW
1547 if (c == MICRO_SIGN) {
1548 converted = GREEK_SMALL_LETTER_MU;
1549 }
51910141 1550 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1551
1552 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1553 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1554 * under those circumstances. */
1555 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1556 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1557 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1558 p, *lenp, U8);
1559 return LATIN_SMALL_LETTER_LONG_S;
1560 }
1561 else {
4f489194
KW
1562 *(p)++ = 's';
1563 *p = 's';
1564 *lenp = 2;
1565 return 's';
1ca267a5 1566 }
a1dde8de
KW
1567 }
1568 else { /* In this range the fold of all other characters is their lower
1569 case */
1570 converted = toLOWER_LATIN1(c);
1571 }
1572
6f2d5cbc 1573 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1574 *p = (U8) converted;
1575 *lenp = 1;
1576 }
1577 else {
1578 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1579 *p = UTF8_TWO_BYTE_LO(converted);
1580 *lenp = 2;
1581 }
1582
1583 return converted;
1584}
1585
1586UV
31f05a37 1587Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1588{
4b593389 1589
a0270393
KW
1590 /* Not currently externally documented, and subject to change
1591 * <flags> bits meanings:
1592 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1593 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1594 * locale are to be used.
a0270393
KW
1595 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1596 */
4b593389 1597
36bb2ab6 1598 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1599
31f05a37
KW
1600 /* Tread a UTF-8 locale as not being in locale at all */
1601 if (IN_UTF8_CTYPE_LOCALE) {
1602 flags &= ~FOLD_FLAGS_LOCALE;
1603 }
1604
a1dde8de 1605 if (c < 256) {
a0270393 1606 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1607 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1608 /* It is illegal for the fold to cross the 255/256 boundary under
1609 * locale; in this case return the original */
1610 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1611 ? c
1612 : result;
a1dde8de
KW
1613 }
1614
a0270393
KW
1615 /* If no special needs, just use the macro */
1616 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1617 uvchr_to_utf8(p, c);
1618 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1619 }
1620 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1621 the special flags. */
1622 U8 utf8_c[UTF8_MAXBYTES + 1];
1623 uvchr_to_utf8(utf8_c, c);
445bf929 1624 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1625 }
84afefe6
JH
1626}
1627
26483009 1628PERL_STATIC_INLINE bool
5141f98e 1629S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1630 const char *const swashname, SV* const invlist)
bde6a22d 1631{
ea317ccb
KW
1632 /* returns a boolean giving whether or not the UTF8-encoded character that
1633 * starts at <p> is in the swash indicated by <swashname>. <swash>
1634 * contains a pointer to where the swash indicated by <swashname>
1635 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1636 * look at <*swash> and only generate a swash if it is not null. <invlist>
1637 * is NULL or an inversion list that defines the swash. If not null, it
1638 * saves time during initialization of the swash.
ea317ccb
KW
1639 *
1640 * Note that it is assumed that the buffer length of <p> is enough to
1641 * contain all the bytes that comprise the character. Thus, <*p> should
1642 * have been checked before this call for mal-formedness enough to assure
1643 * that. */
1644
7918f24d
NC
1645 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1646
492a624f 1647 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1648 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1649 * as far as there being enough bytes available in it to accommodate the
1650 * character without reading beyond the end, and pass that number on to the
1651 * validating routine */
6302f837 1652 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1653 if (ckWARN_d(WARN_UTF8)) {
1654 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1655 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1656 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1657 what the malformation is */
1658 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1659 }
1660 }
1661 return FALSE;
1662 }
87367d5f
KW
1663 if (!*swash) {
1664 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1665 *swash = _core_swash_init("utf8",
1666
1667 /* Only use the name if there is no inversion
1668 * list; otherwise will go out to disk */
1669 (invlist) ? "" : swashname,
1670
1671 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1672 }
28123549 1673
bde6a22d
NC
1674 return swash_fetch(*swash, p, TRUE) != 0;
1675}
1676
1677bool
922e8cb4
KW
1678Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1679{
922e8cb4
KW
1680 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1681
1682 assert(classnum < _FIRST_NON_SWASH_CC);
1683
f25ce844
KW
1684 return is_utf8_common(p,
1685 &PL_utf8_swash_ptrs[classnum],
1686 swash_property_names[classnum],
1687 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1688}
1689
1690bool
f2645549 1691Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1692{
f2645549 1693 SV* invlist = NULL;
7918f24d 1694
f2645549 1695 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1696
f2645549
KW
1697 if (! PL_utf8_perl_idstart) {
1698 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1699 }
60071a22 1700 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1701}
1702
1703bool
f2645549 1704Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1705{
f2645549 1706 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1707
1708 if (*p == '_')
1709 return TRUE;
f25ce844 1710 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1711}
1712
1713bool
eba68aa0
KW
1714Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1715{
b24b43f7 1716 SV* invlist = NULL;
eba68aa0
KW
1717
1718 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1719
b24b43f7
KW
1720 if (! PL_utf8_perl_idcont) {
1721 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1722 }
60071a22 1723 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1724}
1725
eba68aa0 1726bool
f2645549 1727Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1728{
f2645549 1729 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1730
f25ce844 1731 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1732}
1733
1734bool
f2645549 1735Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1736{
f2645549 1737 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1738
f25ce844 1739 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1740}
1741
1742bool
7dbf68d2
KW
1743Perl__is_utf8_mark(pTHX_ const U8 *p)
1744{
7dbf68d2
KW
1745 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1746
f25ce844 1747 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1748}
1749
6b5c0936 1750/*
87cea99e 1751=for apidoc to_utf8_case
6b5c0936 1752
6fae5207 1753C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1754the character that is being converted. This routine assumes that the character
1755at C<p> is well-formed.
6b5c0936 1756
6fae5207
KW
1757C<ustrp> is a pointer to the character buffer to put the
1758conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1759of the result.
1760
6fae5207 1761C<swashp> is a pointer to the swash to use.
6b5c0936 1762
a1433954 1763Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1764and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1765but not always, a multicharacter mapping), is tried first.
6b5c0936 1766
4a8240a3
KW
1767C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1768any special mappings; C<""> means to use the special mappings. Values other
1769than these two are treated as the name of the hash containing the special
1770mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1771
6fae5207 1772C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1773%utf8::ToLower.
1774
1775=cut */
6b5c0936 1776
2104c8d9 1777UV
9a957fbc
AL
1778Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1779 SV **swashp, const char *normal, const char *special)
a0ed51b3 1780{
0134edef 1781 STRLEN len = 0;
f4cd282c 1782 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1783
1784 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1785
9ae3ac1a
KW
1786 /* Note that swash_fetch() doesn't output warnings for these because it
1787 * assumes we will */
8457b38f 1788 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1789 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1790 if (ckWARN_d(WARN_SURROGATE)) {
1791 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1792 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1793 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1794 }
9ae3ac1a
KW
1795 }
1796 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1797 if (ckWARN_d(WARN_NON_UNICODE)) {
1798 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1799 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1800 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1801 }
9ae3ac1a
KW
1802 }
1803
1804 /* Note that non-characters are perfectly legal, so no warning should
1805 * be given */
1806 }
1807
0134edef 1808 if (!*swashp) /* load on-demand */
5ab9d2ef 1809 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1810
a6f87d8c 1811 if (special) {
0134edef 1812 /* It might be "special" (sometimes, but not always,
2a37f04d 1813 * a multicharacter mapping) */
4a8240a3 1814 HV *hv = NULL;
b08cf34e
JH
1815 SV **svp;
1816
4a8240a3
KW
1817 /* If passed in the specials name, use that; otherwise use any
1818 * given in the swash */
1819 if (*special != '\0') {
1820 hv = get_hv(special, 0);
1821 }
1822 else {
1823 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1824 if (svp) {
1825 hv = MUTABLE_HV(SvRV(*svp));
1826 }
1827 }
1828
176fe009
KW
1829 if (hv
1830 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1831 && (*svp))
1832 {
cfd0369c 1833 const char *s;
47654450 1834
cfd0369c 1835 s = SvPV_const(*svp, len);
47654450 1836 if (len == 1)
f4cd282c 1837 /* EIGHTBIT */
c80e42f3 1838 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1839 else {
d2dcd0fb 1840 Copy(s, ustrp, len, U8);
29e98929 1841 }
983ffd37 1842 }
0134edef
JH
1843 }
1844
1845 if (!len && *swashp) {
f4cd282c 1846 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1847
0134edef
JH
1848 if (uv2) {
1849 /* It was "normal" (a single character mapping). */
f4cd282c 1850 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1851 }
1852 }
1feea2c7 1853
cbe07460
KW
1854 if (len) {
1855 if (lenp) {
1856 *lenp = len;
1857 }
1858 return valid_utf8_to_uvchr(ustrp, 0);
1859 }
1860
1861 /* Here, there was no mapping defined, which means that the code point maps
1862 * to itself. Return the inputs */
bfdf22ec 1863 len = UTF8SKIP(p);
ca9fab46
KW
1864 if (p != ustrp) { /* Don't copy onto itself */
1865 Copy(p, ustrp, len, U8);
1866 }
0134edef 1867
2a37f04d
JH
1868 if (lenp)
1869 *lenp = len;
1870
f4cd282c 1871 return uv1;
cbe07460 1872
a0ed51b3
LW
1873}
1874
051a06d4 1875STATIC UV
ab0b796c 1876S_check_locale_boundary_crossing(pTHX_ const char * const func_name, const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4
KW
1877{
1878 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
1879 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1880 * result contains a character that crosses the 255/256 boundary, disallow
1881 * the change, and return the original code point. See L<perlfunc/lc> for
1882 * why;
051a06d4 1883 *
a1433954
KW
1884 * p points to the original string whose case was changed; assumed
1885 * by this routine to be well-formed
051a06d4
KW
1886 * result the code point of the first character in the changed-case string
1887 * ustrp points to the changed-case string (<result> represents its first char)
1888 * lenp points to the length of <ustrp> */
1889
1890 UV original; /* To store the first code point of <p> */
1891
1892 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1893
a4f12ed7 1894 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1895
1896 /* We know immediately if the first character in the string crosses the
1897 * boundary, so can skip */
1898 if (result > 255) {
1899
1900 /* Look at every character in the result; if any cross the
1901 * boundary, the whole thing is disallowed */
1902 U8* s = ustrp + UTF8SKIP(ustrp);
1903 U8* e = ustrp + *lenp;
1904 while (s < e) {
a4f12ed7 1905 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1906 goto bad_crossing;
1907 }
1908 s += UTF8SKIP(s);
1909 }
1910
1911 /* Here, no characters crossed, result is ok as-is */
1912 return result;
1913 }
1914
1915bad_crossing:
1916
1917 /* Failed, have to return the original */
4b88fb76 1918 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1919
1920 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1921 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1922 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1923 "resolved to \"\\x{%"UVXf"}\".",
1924 func_name,
1925 original,
1926 original);
051a06d4
KW
1927 Copy(p, ustrp, *lenp, char);
1928 return original;
1929}
1930
d3e79532 1931/*
87cea99e 1932=for apidoc to_utf8_upper
d3e79532 1933
1f607577 1934Instead use L</toUPPER_utf8>.
a1433954 1935
d3e79532
JH
1936=cut */
1937
051a06d4 1938/* Not currently externally documented, and subject to change:
31f05a37
KW
1939 * <flags> is set iff iff the rules from the current underlying locale are to
1940 * be used. */
051a06d4 1941
2104c8d9 1942UV
31f05a37 1943Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1944{
051a06d4
KW
1945 UV result;
1946
1947 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1948
31f05a37
KW
1949 if (flags && IN_UTF8_CTYPE_LOCALE) {
1950 flags = FALSE;
1951 }
1952
3a4c58c9 1953 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1954 if (flags) {
1955 result = toUPPER_LC(*p);
1956 }
1957 else {
81c6c7ce 1958 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1959 }
3a4c58c9
KW
1960 }
1961 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1962 if (flags) {
a6d8b88b 1963 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1964 result = toUPPER_LC(c);
051a06d4
KW
1965 }
1966 else {
94bb8c36 1967 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1968 ustrp, lenp, 'S');
051a06d4
KW
1969 }
1970 }
1971 else { /* utf8, ord above 255 */
1972 result = CALL_UPPER_CASE(p, ustrp, lenp);
1973
1974 if (flags) {
ab0b796c 1975 result = check_locale_boundary_crossing("uc", p, result, ustrp, lenp);
051a06d4
KW
1976 }
1977 return result;
1978 }
1979
1980 /* Here, used locale rules. Convert back to utf8 */
1981 if (UTF8_IS_INVARIANT(result)) {
1982 *ustrp = (U8) result;
1983 *lenp = 1;
1984 }
1985 else {
62cb07ea
KW
1986 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1987 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1988 *lenp = 2;
3a4c58c9 1989 }
baa60164 1990
051a06d4 1991 return result;
983ffd37 1992}
a0ed51b3 1993
d3e79532 1994/*
87cea99e 1995=for apidoc to_utf8_title
d3e79532 1996
1f607577 1997Instead use L</toTITLE_utf8>.
a1433954 1998
d3e79532
JH
1999=cut */
2000
051a06d4 2001/* Not currently externally documented, and subject to change:
31f05a37
KW
2002 * <flags> is set iff the rules from the current underlying locale are to be
2003 * used. Since titlecase is not defined in POSIX, for other than a
2004 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2005 */
051a06d4 2006
983ffd37 2007UV
31f05a37 2008Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2009{
051a06d4
KW
2010 UV result;
2011
2012 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2013
31f05a37
KW
2014 if (flags && IN_UTF8_CTYPE_LOCALE) {
2015 flags = FALSE;
2016 }
2017
3a4c58c9 2018 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2019 if (flags) {
2020 result = toUPPER_LC(*p);
2021 }
2022 else {
81c6c7ce 2023 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2024 }
3a4c58c9
KW
2025 }
2026 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2027 if (flags) {
a6d8b88b 2028 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2029 result = toUPPER_LC(c);
051a06d4
KW
2030 }
2031 else {
94bb8c36 2032 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2033 ustrp, lenp, 's');
051a06d4
KW
2034 }
2035 }
2036 else { /* utf8, ord above 255 */
2037 result = CALL_TITLE_CASE(p, ustrp, lenp);
2038
2039 if (flags) {
ab0b796c 2040 result = check_locale_boundary_crossing("ucfirst", p, result, ustrp, lenp);
051a06d4
KW
2041 }
2042 return result;
2043 }
2044
2045 /* Here, used locale rules. Convert back to utf8 */
2046 if (UTF8_IS_INVARIANT(result)) {
2047 *ustrp = (U8) result;
2048 *lenp = 1;
2049 }
2050 else {
62cb07ea
KW
2051 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2052 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2053 *lenp = 2;
3a4c58c9
KW
2054 }
2055
051a06d4 2056 return result;
a0ed51b3
LW
2057}
2058
d3e79532 2059/*
87cea99e 2060=for apidoc to_utf8_lower
d3e79532 2061
1f607577 2062Instead use L</toLOWER_utf8>.
a1433954 2063
d3e79532
JH
2064=cut */
2065
051a06d4 2066/* Not currently externally documented, and subject to change:
31f05a37
KW
2067 * <flags> is set iff iff the rules from the current underlying locale are to
2068 * be used.
2069 */
051a06d4 2070
2104c8d9 2071UV
31f05a37 2072Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2073{
051a06d4
KW
2074 UV result;
2075
051a06d4 2076 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2077
31f05a37
KW
2078 if (flags && IN_UTF8_CTYPE_LOCALE) {
2079 flags = FALSE;
2080 }
2081
968c5e6a 2082 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2083 if (flags) {
2084 result = toLOWER_LC(*p);
2085 }
2086 else {
81c6c7ce 2087 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2088 }
968c5e6a
KW
2089 }
2090 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2091 if (flags) {
a6d8b88b 2092 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2093 result = toLOWER_LC(c);
051a06d4
KW
2094 }
2095 else {
94bb8c36 2096 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2097 ustrp, lenp);
051a06d4 2098 }
968c5e6a 2099 }
051a06d4
KW
2100 else { /* utf8, ord above 255 */
2101 result = CALL_LOWER_CASE(p, ustrp, lenp);
2102
2103 if (flags) {
ab0b796c 2104 result = check_locale_boundary_crossing("lc", p, result, ustrp, lenp);
051a06d4 2105 }
968c5e6a 2106
051a06d4
KW
2107 return result;
2108 }
2109
2110 /* Here, used locale rules. Convert back to utf8 */
2111 if (UTF8_IS_INVARIANT(result)) {
2112 *ustrp = (U8) result;
2113 *lenp = 1;
2114 }
2115 else {
62cb07ea
KW
2116 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2117 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2118 *lenp = 2;
2119 }
2120
051a06d4 2121 return result;
b4e400f9
JH
2122}
2123
d3e79532 2124/*
87cea99e 2125=for apidoc to_utf8_fold
d3e79532 2126
1f607577 2127Instead use L</toFOLD_utf8>.
a1433954 2128
d3e79532
JH
2129=cut */
2130
051a06d4
KW
2131/* Not currently externally documented, and subject to change,
2132 * in <flags>
31f05a37
KW
2133 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2134 * locale are to be used.
051a06d4
KW
2135 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2136 * otherwise simple folds
a0270393
KW
2137 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2138 * prohibited
445bf929 2139 */
36bb2ab6 2140
b4e400f9 2141UV
445bf929 2142Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2143{
051a06d4
KW
2144 UV result;
2145
36bb2ab6 2146 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2147
a0270393
KW
2148 /* These are mutually exclusive */
2149 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2150
50ba90ff
KW
2151 assert(p != ustrp); /* Otherwise overwrites */
2152
31f05a37
KW
2153 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2154 flags &= ~FOLD_FLAGS_LOCALE;
2155 }
2156
a1dde8de 2157 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2158 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2159 result = toFOLD_LC(*p);
051a06d4
KW
2160 }
2161 else {
81c6c7ce 2162 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2163 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2164 }
a1dde8de
KW
2165 }
2166 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2167 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2168 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2169 result = toFOLD_LC(c);
051a06d4
KW
2170 }
2171 else {
94bb8c36 2172 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2173 ustrp, lenp,
1ca267a5 2174 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2175 }
a1dde8de 2176 }
051a06d4 2177 else { /* utf8, ord above 255 */
a0270393 2178 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2179
1ca267a5
KW
2180 if (flags & FOLD_FLAGS_LOCALE) {
2181
538e84ed
KW
2182 /* Special case these two characters, as what normally gets
2183 * returned under locale doesn't work */
1ca267a5
KW
2184 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2185 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2186 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2187 {
ab0b796c
KW
2188 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2189 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2190 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2191 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2192 goto return_long_s;
2193 }
2721f3e1 2194 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1
9fc2026f
KW
2195 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2196 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2197 {
ab0b796c
KW
2198 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2199 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2200 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2201 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2202 goto return_ligature_st;
2203 }
ab0b796c 2204 return check_locale_boundary_crossing("fc", p, result, ustrp, lenp);
051a06d4 2205 }
a0270393
KW
2206 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2207 return result;
2208 }
2209 else {
2210 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2211 * character above the ASCII range, and the result should not
2212 * contain an ASCII character. */
a0270393
KW
2213
2214 UV original; /* To store the first code point of <p> */
2215
2216 /* Look at every character in the result; if any cross the
2217 * boundary, the whole thing is disallowed */
2218 U8* s = ustrp;
2219 U8* e = ustrp + *lenp;
2220 while (s < e) {
2221 if (isASCII(*s)) {
2222 /* Crossed, have to return the original */
2223 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2224
9fc2026f 2225 /* But in these instances, there is an alternative we can
1ca267a5 2226 * return that is valid */
9fc2026f
KW
2227 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2228 || original == LATIN_SMALL_LETTER_SHARP_S)
2229 {
1ca267a5
KW
2230 goto return_long_s;
2231 }
9fc2026f
KW
2232 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2233 goto return_ligature_st;
2234 }
a0270393
KW
2235 Copy(p, ustrp, *lenp, char);
2236 return original;
2237 }
2238 s += UTF8SKIP(s);
2239 }
051a06d4 2240
a0270393
KW
2241 /* Here, no characters crossed, result is ok as-is */
2242 return result;
2243 }
051a06d4
KW
2244 }
2245
2246 /* Here, used locale rules. Convert back to utf8 */
2247 if (UTF8_IS_INVARIANT(result)) {
2248 *ustrp = (U8) result;
2249 *lenp = 1;
2250 }
2251 else {
62cb07ea
KW
2252 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2253 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2254 *lenp = 2;
2255 }
2256
051a06d4 2257 return result;
1ca267a5
KW
2258
2259 return_long_s:
2260 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2261 * folds to a string of two of these characters. By returning this
2262 * instead, then, e.g.,
2263 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2264 * works. */
2265
2266 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2267 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2268 ustrp, *lenp, U8);
2269 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2270
2271 return_ligature_st:
2272 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2273 * have the other one fold to it */
2274
2275 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2276 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2277 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2278}
2279
711a919c 2280/* Note:
f90a9a02 2281 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2282 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2283 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2284 */
c4a5db0c 2285
a0ed51b3 2286SV*
7fc63493 2287Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2288{
c4a5db0c
KW
2289 PERL_ARGS_ASSERT_SWASH_INIT;
2290
2291 /* Returns a copy of a swash initiated by the called function. This is the
2292 * public interface, and returning a copy prevents others from doing
2293 * mischief on the original */
2294
5d3d13d1 2295 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2296}
2297
2298SV*
5d3d13d1 2299Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2300{
2c1f00b9
YO
2301
2302 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2303 * use the following define */
2304
2305#define CORE_SWASH_INIT_RETURN(x) \
2306 PL_curpm= old_PL_curpm; \
2307 return x
2308
c4a5db0c 2309 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2310 * by calling utf8_heavy.pl in the general case. The returned value may be
2311 * the swash's inversion list instead if the input parameters allow it.
2312 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2313 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2314 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2315 *
2316 * This interface should only be used by functions that won't destroy or
2317 * adversely change the swash, as doing so affects all other uses of the
2318 * swash in the program; the general public should use 'Perl_swash_init'
2319 * instead.
2320 *
2321 * pkg is the name of the package that <name> should be in.
2322 * name is the name of the swash to find. Typically it is a Unicode
2323 * property name, including user-defined ones
2324 * listsv is a string to initialize the swash with. It must be of the form
2325 * documented as the subroutine return value in
2326 * L<perlunicode/User-Defined Character Properties>
2327 * minbits is the number of bits required to represent each data element.
2328 * It is '1' for binary properties.
2329 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2330 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2331 * flags_p if non-NULL is the address of various input and output flag bits
2332 * to the routine, as follows: ('I' means is input to the routine;
2333 * 'O' means output from the routine. Only flags marked O are
2334 * meaningful on return.)
2335 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2336 * came from a user-defined property. (I O)
5d3d13d1
KW
2337 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2338 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2339 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2340 * return of an inversion list instead of a swash hash if this routine
2341 * thinks that would result in faster execution of swash_fetch() later
2342 * on. (I)
9a53f6cf
KW
2343 *
2344 * Thus there are three possible inputs to find the swash: <name>,
2345 * <listsv>, and <invlist>. At least one must be specified. The result
2346 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2347 * actions can intersect, etc. what <name> gives. To avoid going out to
2348 * disk at all, <invlist> should specify completely what the swash should
2349 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2350 *
2351 * <invlist> is only valid for binary properties */
c4a5db0c 2352
2c1f00b9
YO
2353 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2354
c4a5db0c 2355 SV* retval = &PL_sv_undef;
83199d38 2356 HV* swash_hv = NULL;
87367d5f
KW
2357 const int invlist_swash_boundary =
2358 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2359 ? 512 /* Based on some benchmarking, but not extensive, see commit
2360 message */
2361 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2362
2363 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2364 assert(! invlist || minbits == 1);
2365
2c1f00b9
YO
2366 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2367 that triggered the swash init and the swash init perl logic itself.
2368 See perl #122747 */
2369
9a53f6cf
KW
2370 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2371 * so */
2372 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2373 dSP;
2374 const size_t pkg_len = strlen(pkg);
2375 const size_t name_len = strlen(name);
2376 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2377 SV* errsv_save;
2378 GV *method;
2379
2380 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2381
2382 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2383 ENTER;
69794297 2384 SAVEHINTS();
650f067c
JL
2385 /* We might get here via a subroutine signature which uses a utf8
2386 * parameter name, at which point PL_subname will have been set
2387 * but not yet used. */
2388 save_item(PL_subname);
69794297
KW
2389 if (PL_parser && PL_parser->error_count)
2390 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2391 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2392 if (!method) { /* demand load utf8 */
2393 ENTER;
db2c6cb3
FC
2394 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2395 GvSV(PL_errgv) = NULL;
1a419e6b 2396#ifndef NO_TAINT_SUPPORT
69794297
KW
2397 /* It is assumed that callers of this routine are not passing in
2398 * any user derived data. */
284167a5
SM
2399 SAVEBOOL(TAINT_get);
2400 TAINT_NOT;
2401#endif
69794297
KW
2402 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2403 NULL);
eed484f9 2404 {
db2c6cb3
FC
2405 /* Not ERRSV, as there is no need to vivify a scalar we are
2406 about to discard. */
2407 SV * const errsv = GvSV(PL_errgv);
2408 if (!SvTRUE(errsv)) {
2409 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2410 SvREFCNT_dec(errsv);
2411 }
eed484f9 2412 }
69794297
KW
2413 LEAVE;
2414 }
2415 SPAGAIN;
2416 PUSHMARK(SP);
2417 EXTEND(SP,5);
2418 mPUSHp(pkg, pkg_len);
2419 mPUSHp(name, name_len);
2420 PUSHs(listsv);
2421 mPUSHi(minbits);
2422 mPUSHi(none);
2423 PUTBACK;
db2c6cb3
FC
2424 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2425 GvSV(PL_errgv) = NULL;
69794297
KW
2426 /* If we already have a pointer to the method, no need to use
2427 * call_method() to repeat the lookup. */
c41800a8
KW
2428 if (method
2429 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2430 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2431 {
2432 retval = *PL_stack_sp--;
2433 SvREFCNT_inc(retval);
2434 }
eed484f9 2435 {
db2c6cb3
FC
2436 /* Not ERRSV. See above. */
2437 SV * const errsv = GvSV(PL_errgv);
2438 if (!SvTRUE(errsv)) {
2439 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2440 SvREFCNT_dec(errsv);
2441 }
eed484f9 2442 }
ce3b816e 2443 LEAVE;
69794297
KW
2444 POPSTACK;
2445 if (IN_PERL_COMPILETIME) {
2446 CopHINTS_set(PL_curcop, PL_hints);
2447 }
2448 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2449 if (SvPOK(retval))
2450
2451 /* If caller wants to handle missing properties, let them */
5d3d13d1 2452 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2453 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2454 }
2455 Perl_croak(aTHX_
2456 "Can't find Unicode property definition \"%"SVf"\"",
2457 SVfARG(retval));
a25b5927 2458 NOT_REACHED; /* NOTREACHED */
69794297 2459 }
9a53f6cf 2460 } /* End of calling the module to find the swash */
36eb48b4 2461
83199d38
KW
2462 /* If this operation fetched a swash, and we will need it later, get it */
2463 if (retval != &PL_sv_undef
2464 && (minbits == 1 || (flags_p
2465 && ! (*flags_p
2466 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2467 {
2468 swash_hv = MUTABLE_HV(SvRV(retval));
2469
2470 /* If we don't already know that there is a user-defined component to
2471 * this swash, and the user has indicated they wish to know if there is
2472 * one (by passing <flags_p>), find out */
2473 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2474 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2475 if (user_defined && SvUV(*user_defined)) {
2476 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2477 }
2478 }
2479 }
2480
36eb48b4
KW
2481 /* Make sure there is an inversion list for binary properties */
2482 if (minbits == 1) {
2483 SV** swash_invlistsvp = NULL;
2484 SV* swash_invlist = NULL;
9a53f6cf 2485 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2486 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2487 an unclaimed reference count */
36eb48b4 2488
9a53f6cf 2489 /* If this operation fetched a swash, get its already existing
83199d38 2490 * inversion list, or create one for it */
36eb48b4 2491
83199d38 2492 if (swash_hv) {
5c9f4bd2 2493 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2494 if (swash_invlistsvp) {
2495 swash_invlist = *swash_invlistsvp;
2496 invlist_in_swash_is_valid = TRUE;
2497 }
2498 else {
36eb48b4 2499 swash_invlist = _swash_to_invlist(retval);
02c85471 2500 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2501 }
2502 }
2503
2504 /* If an inversion list was passed in, have to include it */
2505 if (invlist) {
2506
2507 /* Any fetched swash will by now have an inversion list in it;
2508 * otherwise <swash_invlist> will be NULL, indicating that we
2509 * didn't fetch a swash */
2510 if (swash_invlist) {
2511
2512 /* Add the passed-in inversion list, which invalidates the one
2513 * already stored in the swash */
2514 invlist_in_swash_is_valid = FALSE;
2515 _invlist_union(invlist, swash_invlist, &swash_invlist);
2516 }
2517 else {
2518
87367d5f
KW
2519 /* Here, there is no swash already. Set up a minimal one, if
2520 * we are going to return a swash */
2521 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2522 swash_hv = newHV();
4aca0fe6 2523 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2524 }
9a53f6cf
KW
2525 swash_invlist = invlist;
2526 }
9a53f6cf
KW
2527 }
2528
2529 /* Here, we have computed the union of all the passed-in data. It may
2530 * be that there was an inversion list in the swash which didn't get
538e84ed 2531 * touched; otherwise save the computed one */
87367d5f
KW
2532 if (! invlist_in_swash_is_valid
2533 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2534 {
5c9f4bd2 2535 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2536 {
2537 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2538 }
cc34d8c5
FC
2539 /* We just stole a reference count. */
2540 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2541 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2542 }
87367d5f 2543
dbfdbd26
KW
2544 SvREADONLY_on(swash_invlist);
2545
c41800a8 2546 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2547 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2548 SvREFCNT_dec(retval);
02c85471
FC
2549 if (!swash_invlist_unclaimed)
2550 SvREFCNT_inc_simple_void_NN(swash_invlist);
2551 retval = newRV_noinc(swash_invlist);
87367d5f 2552 }
36eb48b4
KW
2553 }
2554
2c1f00b9
YO
2555 CORE_SWASH_INIT_RETURN(retval);
2556#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2557}
2558
035d37be
JH
2559
2560/* This API is wrong for special case conversions since we may need to
2561 * return several Unicode characters for a single Unicode character
2562 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2563 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2564 * multiple values. --jhi
2565 * For those, you should use to_utf8_case() instead */
b0e3252e 2566/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2567
2568/* Note:
2569 * Returns the value of property/mapping C<swash> for the first character
2570 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2571 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2572 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2573 *
2574 * A "swash" is a hash which contains initially the keys/values set up by
2575 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2576 * property for all possible code points. Things are stored in a compact form
2577 * (see utf8_heavy.pl) so that calculation is required to find the actual
2578 * property value for a given code point. As code points are looked up, new
2579 * key/value pairs are added to the hash, so that the calculation doesn't have
2580 * to ever be re-done. Further, each calculation is done, not just for the
2581 * desired one, but for a whole block of code points adjacent to that one.
2582 * For binary properties on ASCII machines, the block is usually for 64 code
2583 * points, starting with a code point evenly divisible by 64. Thus if the
2584 * property value for code point 257 is requested, the code goes out and
2585 * calculates the property values for all 64 code points between 256 and 319,
2586 * and stores these as a single 64-bit long bit vector, called a "swatch",
2587 * under the key for code point 256. The key is the UTF-8 encoding for code
2588 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2589 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2590 * for code point 258 is then requested, this code realizes that it would be
2591 * stored under the key for 256, and would find that value and extract the
2592 * relevant bit, offset from 256.
2593 *
2594 * Non-binary properties are stored in as many bits as necessary to represent
2595 * their values (32 currently, though the code is more general than that), not
2596 * as single bits, but the principal is the same: the value for each key is a
2597 * vector that encompasses the property values for all code points whose UTF-8
2598 * representations are represented by the key. That is, for all code points
2599 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2600 * bytes of that.
680c470c 2601 */
a0ed51b3 2602UV
680c470c 2603Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2604{
ef8f7699 2605 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2606 U32 klen;
2607 U32 off;
9b56a019 2608 STRLEN slen = 0;
7d85a32c 2609 STRLEN needents;
cfd0369c 2610 const U8 *tmps = NULL;
979f2922 2611 SV *swatch;
08fb1ac5 2612 const U8 c = *ptr;
3568d838 2613
7918f24d
NC
2614 PERL_ARGS_ASSERT_SWASH_FETCH;
2615
87367d5f
KW
2616 /* If it really isn't a hash, it isn't really swash; must be an inversion
2617 * list */
2618 if (SvTYPE(hv) != SVt_PVHV) {
2619 return _invlist_contains_cp((SV*)hv,
2620 (do_utf8)
2621 ? valid_utf8_to_uvchr(ptr, NULL)
2622 : c);
2623 }
2624
08fb1ac5
KW
2625 /* We store the values in a "swatch" which is a vec() value in a swash
2626 * hash. Code points 0-255 are a single vec() stored with key length
2627 * (klen) 0. All other code points have a UTF-8 representation
2628 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2629 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2630 * length for them is the length of the encoded char - 1. ptr[klen] is the
2631 * final byte in the sequence representing the character */
2632 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2633 klen = 0;
2634 needents = 256;
2635 off = c;
3568d838 2636 }
08fb1ac5
KW
2637 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2638 klen = 0;
2639 needents = 256;
2640 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2641 }
2642 else {
08fb1ac5
KW
2643 klen = UTF8SKIP(ptr) - 1;
2644
2645 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2646 * the vec is the final byte in the sequence. (In EBCDIC this is
2647 * converted to I8 to get consecutive values.) To help you visualize
2648 * all this:
2649 * Straight 1047 After final byte
2650 * UTF-8 UTF-EBCDIC I8 transform
2651 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2652 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2653 * ...
2654 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2655 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2656 * ...
2657 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2658 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2659 * ...
2660 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2661 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2662 * ...
2663 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2664 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2665 *
2666 * (There are no discontinuities in the elided (...) entries.)
2667 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2668 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2669 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2670 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2671 * index into the vec() swatch (after subtracting 0x80, which we
2672 * actually do with an '&').
2673 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2674 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2675 * dicontinuities which go away by transforming it into I8, and we
2676 * effectively subtract 0xA0 to get the index. */
979f2922 2677 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2678 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2679 }
7d85a32c 2680
a0ed51b3
LW
2681 /*
2682 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2683 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2684 * it's nothing to sniff at.) Pity we usually come through at least
2685 * two function calls to get here...
2686 *
2687 * NB: this code assumes that swatches are never modified, once generated!
2688 */
2689
3568d838 2690 if (hv == PL_last_swash_hv &&
a0ed51b3 2691 klen == PL_last_swash_klen &&
27da23d5 2692 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2693 {
2694 tmps = PL_last_swash_tmps;
2695 slen = PL_last_swash_slen;
2696 }
2697 else {
2698 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2699 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2700
b0e3252e 2701 /* If not cached, generate it via swatch_get */
979f2922 2702 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2703 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2704 {
2705 if (klen) {
2706 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2707 swatch = swatch_get(swash,
2708 code_point & ~((UV)needents - 1),
2709 needents);
2710 }
2711 else { /* For the first 256 code points, the swatch has a key of
2712 length 0 */
2713 swatch = swatch_get(swash, 0, needents);
2714 }
979f2922 2715
923e4eb5 2716 if (IN_PERL_COMPILETIME)
623e6609 2717 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2718
979f2922 2719 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2720
979f2922
ST
2721 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2722 || (slen << 3) < needents)
5637ef5b
NC
2723 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2724 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2725 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2726 }
2727
2728 PL_last_swash_hv = hv;
16d8f38a 2729 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2730 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2731 /* FIXME change interpvar.h? */
2732 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2733 PL_last_swash_slen = slen;
2734 if (klen)
2735 Copy(ptr, PL_last_swash_key, klen, U8);
2736 }
2737
9faf8d75 2738 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2739 case 1:
e7aca353 2740 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2741 case 8:
e7aca353 2742 return ((UV) tmps[off]);
a0ed51b3
LW
2743 case 16:
2744 off <<= 1;
e7aca353
JH
2745 return
2746 ((UV) tmps[off ] << 8) +
2747 ((UV) tmps[off + 1]);
a0ed51b3
LW
2748 case 32:
2749 off <<= 2;
e7aca353
JH
2750 return
2751 ((UV) tmps[off ] << 24) +
2752 ((UV) tmps[off + 1] << 16) +
2753 ((UV) tmps[off + 2] << 8) +
2754 ((UV) tmps[off + 3]);
a0ed51b3 2755 }
5637ef5b
NC
2756 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2757 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2758 NORETURN_FUNCTION_END;
a0ed51b3 2759}
2b9d42f0 2760
319009ee
KW
2761/* Read a single line of the main body of the swash input text. These are of
2762 * the form:
2763 * 0053 0056 0073
2764 * where each number is hex. The first two numbers form the minimum and
2765 * maximum of a range, and the third is the value associated with the range.
2766 * Not all swashes should have a third number
2767 *
2768 * On input: l points to the beginning of the line to be examined; it points
2769 * to somewhere in the string of the whole input text, and is
2770 * terminated by a \n or the null string terminator.
2771 * lend points to the null terminator of that string
2772 * wants_value is non-zero if the swash expects a third number
2773 * typestr is the name of the swash's mapping, like 'ToLower'
2774 * On output: *min, *max, and *val are set to the values read from the line.
2775 * returns a pointer just beyond the line examined. If there was no
2776 * valid min number on the line, returns lend+1
2777 */
2778
2779STATIC U8*
2780S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2781 const bool wants_value, const U8* const typestr)
2782{
2783 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2784 STRLEN numlen; /* Length of the number */
02470786
KW
2785 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2786 | PERL_SCAN_DISALLOW_PREFIX
2787 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2788
2789 /* nl points to the next \n in the scan */
2790 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2791
95543e92
KW
2792 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2793
319009ee
KW
2794 /* Get the first number on the line: the range minimum */
2795 numlen = lend - l;
2796 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2797 *max = *min; /* So can never return without setting max */
319009ee
KW
2798 if (numlen) /* If found a hex number, position past it */
2799 l += numlen;
2800 else if (nl) { /* Else, go handle next line, if any */
2801 return nl + 1; /* 1 is length of "\n" */
2802 }
2803 else { /* Else, no next line */
2804 return lend + 1; /* to LIST's end at which \n is not found */
2805 }
2806
2807 /* The max range value follows, separated by a BLANK */
2808 if (isBLANK(*l)) {
2809 ++l;
02470786
KW
2810 flags = PERL_SCAN_SILENT_ILLDIGIT
2811 | PERL_SCAN_DISALLOW_PREFIX
2812 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2813 numlen = lend - l;
2814 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2815 if (numlen)
2816 l += numlen;
2817 else /* If no value here, it is a single element range */
2818 *max = *min;
2819
2820 /* Non-binary tables have a third entry: what the first element of the
24303724 2821 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2822 if (wants_value) {
2823 if (isBLANK(*l)) {
2824 ++l;
f2a7d0fc
KW
2825 flags = PERL_SCAN_SILENT_ILLDIGIT
2826 | PERL_SCAN_DISALLOW_PREFIX
2827 | PERL_SCAN_SILENT_NON_PORTABLE;
2828 numlen = lend - l;
2829 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2830 if (numlen)
2831 l += numlen;
2832 else
2833 *val = 0;
319009ee
KW
2834 }
2835 else {
2836 *val = 0;
2837 if (typeto) {
dcbac5bb 2838 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2839 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2840 typestr, l);
2841 }
2842 }
2843 }
2844 else
2845 *val = 0; /* bits == 1, then any val should be ignored */
2846 }
2847 else { /* Nothing following range min, should be single element with no
2848 mapping expected */
319009ee
KW
2849 if (wants_value) {
2850 *val = 0;
2851 if (typeto) {
dcbac5bb 2852 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2853 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2854 }
2855 }
2856 else
2857 *val = 0; /* bits == 1, then val should be ignored */
2858 }
2859
2860 /* Position to next line if any, or EOF */
2861 if (nl)
2862 l = nl + 1;
2863 else
2864 l = lend;
2865
2866 return l;
2867}
2868
979f2922
ST
2869/* Note:
2870 * Returns a swatch (a bit vector string) for a code point sequence
2871 * that starts from the value C<start> and comprises the number C<span>.
2872 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2873 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2874 */
2875STATIC SV*
b0e3252e 2876S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2877{
2878 SV *swatch;
77f9f126 2879 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2880 STRLEN lcur, xcur, scur;
ef8f7699 2881 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2882 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2883
88d45d28
KW
2884 SV** listsvp = NULL; /* The string containing the main body of the table */
2885 SV** extssvp = NULL;
2886 SV** invert_it_svp = NULL;
2887 U8* typestr = NULL;
786861f5
KW
2888 STRLEN bits;
2889 STRLEN octets; /* if bits == 1, then octets == 0 */
2890 UV none;
2891 UV end = start + span;
972dd592 2892
36eb48b4 2893 if (invlistsvp == NULL) {
786861f5
KW
2894 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2895 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2896 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2897 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2898 listsvp = hv_fetchs(hv, "LIST", FALSE);
2899 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2900
2901 bits = SvUV(*bitssvp);
2902 none = SvUV(*nonesvp);
2903 typestr = (U8*)SvPV_nolen(*typesvp);
2904 }
36eb48b4
KW
2905 else {
2906 bits = 1;
2907 none = 0;
2908 }
786861f5 2909 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2910
b0e3252e 2911 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2912
979f2922 2913 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 2914 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 2915 (UV)bits);
979f2922
ST
2916 }
2917
84ea5ef6
KW
2918 /* If overflowed, use the max possible */
2919 if (end < start) {
2920 end = UV_MAX;
2921 span = end - start;
2922 }
2923
979f2922 2924 /* create and initialize $swatch */
979f2922 2925 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
2926 swatch = newSV(scur);
2927 SvPOK_on(swatch);
979f2922
ST
2928 s = (U8*)SvPVX(swatch);
2929 if (octets && none) {
0bd48802 2930 const U8* const e = s + scur;
979f2922
ST
2931 while (s < e) {
2932 if (bits == 8)
2933 *s++ = (U8)(none & 0xff);
2934 else if (bits == 16) {
2935 *s++ = (U8)((none >> 8) & 0xff);
2936 *s++ = (U8)( none & 0xff);
2937 }
2938 else if (bits == 32) {
2939 *s++ = (U8)((none >> 24) & 0xff);
2940 *s++ = (U8)((none >> 16) & 0xff);
2941 *s++ = (U8)((none >> 8) & 0xff);
2942 *s++ = (U8)( none & 0xff);
2943 }
2944 }
2945 *s = '\0';
2946 }
2947 else {
2948 (void)memzero((U8*)s, scur + 1);
2949 }
2950 SvCUR_set(swatch, scur);
2951 s = (U8*)SvPVX(swatch);
2952
36eb48b4
KW
2953 if (invlistsvp) { /* If has an inversion list set up use that */
2954 _invlist_populate_swatch(*invlistsvp, start, end, s);
2955 return swatch;
2956 }
2957
2958 /* read $swash->{LIST} */
979f2922
ST
2959 l = (U8*)SvPV(*listsvp, lcur);
2960 lend = l + lcur;
2961 while (l < lend) {
8ed25d53 2962 UV min, max, val, upper;
95543e92
KW
2963 l = swash_scan_list_line(l, lend, &min, &max, &val,
2964 cBOOL(octets), typestr);
319009ee 2965 if (l > lend) {
979f2922
ST
2966 break;
2967 }
2968
972dd592 2969 /* If looking for something beyond this range, go try the next one */
979f2922
ST
2970 if (max < start)
2971 continue;
2972
8ed25d53
KW
2973 /* <end> is generally 1 beyond where we want to set things, but at the
2974 * platform's infinity, where we can't go any higher, we want to
2975 * include the code point at <end> */
2976 upper = (max < end)
2977 ? max
2978 : (max != UV_MAX || end != UV_MAX)
2979 ? end - 1
2980 : end;
2981
979f2922 2982 if (octets) {
35da51f7 2983 UV key;
979f2922
ST
2984 if (min < start) {
2985 if (!none || val < none) {
2986 val += start - min;
2987 }
2988 min = start;
2989 }
8ed25d53 2990 for (key = min; key <= upper; key++) {
979f2922 2991 STRLEN offset;
979f2922
ST
2992 /* offset must be non-negative (start <= min <= key < end) */
2993 offset = octets * (key - start);
2994 if (bits == 8)
2995 s[offset] = (U8)(val & 0xff);
2996 else if (bits == 16) {
2997 s[offset ] = (U8)((val >> 8) & 0xff);
2998 s[offset + 1] = (U8)( val & 0xff);
2999 }
3000 else if (bits == 32) {
3001 s[offset ] = (U8)((val >> 24) & 0xff);
3002 s[offset + 1] = (U8)((val >> 16) & 0xff);
3003 s[offset + 2] = (U8)((val >> 8) & 0xff);
3004 s[offset + 3] = (U8)( val & 0xff);
3005 }
3006
3007 if (!none || val < none)
3008 ++val;
3009 }
3010 }
711a919c 3011 else { /* bits == 1, then val should be ignored */
35da51f7 3012 UV key;
979f2922
ST
3013 if (min < start)
3014 min = start;
6cb05c12 3015
8ed25d53 3016 for (key = min; key <= upper; key++) {
0bd48802 3017 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3018 s[offset >> 3] |= 1 << (offset & 7);
3019 }
3020 }
3021 } /* while */
979f2922 3022
9479a769 3023 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3024 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3025
3026 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3027 * be 0, and their inversion should also be 0, as we don't succeed any
3028 * Unicode property matches for non-Unicode code points */
3029 if (start <= PERL_UNICODE_MAX) {
3030
3031 /* The code below assumes that we never cross the
3032 * Unicode/above-Unicode boundary in a range, as otherwise we would
3033 * have to figure out where to stop flipping the bits. Since this
3034 * boundary is divisible by a large power of 2, and swatches comes
3035 * in small powers of 2, this should be a valid assumption */
3036 assert(start + span - 1 <= PERL_UNICODE_MAX);
3037
507a8485
KW
3038 send = s + scur;
3039 while (s < send) {
3040 *s = ~(*s);
3041 s++;
3042 }
0bda3001 3043 }
77f9f126
KW
3044 }
3045
d73c39c5
KW
3046 /* read $swash->{EXTRAS}
3047 * This code also copied to swash_to_invlist() below */
979f2922
ST
3048 x = (U8*)SvPV(*extssvp, xcur);
3049 xend = x + xcur;
3050 while (x < xend) {
3051 STRLEN namelen;
3052 U8 *namestr;
3053 SV** othersvp;
3054 HV* otherhv;
3055 STRLEN otherbits;
3056 SV **otherbitssvp, *other;
711a919c 3057 U8 *s, *o, *nl;
979f2922
ST
3058 STRLEN slen, olen;
3059
35da51f7 3060 const U8 opc = *x++;
979f2922
ST
3061 if (opc == '\n')
3062 continue;
3063
3064 nl = (U8*)memchr(x, '\n', xend - x);
3065
3066 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3067 if (nl) {
3068 x = nl + 1; /* 1 is length of "\n" */
3069 continue;
3070 }
3071 else {
3072 x = xend; /* to EXTRAS' end at which \n is not found */
3073 break;
3074 }
3075 }
3076
3077 namestr = x;
3078 if (nl) {
3079 namelen = nl - namestr;
3080 x = nl + 1;
3081 }
3082 else {
3083 namelen = xend - namestr;
3084 x = xend;
3085 }
3086
3087 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3088 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3089 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3090 otherbits = (STRLEN)SvUV(*otherbitssvp);
3091 if (bits < otherbits)
5637ef5b
NC
3092 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3093 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3094
3095 /* The "other" swatch must be destroyed after. */
b0e3252e 3096 other = swatch_get(*othersvp, start, span);
979f2922
ST
3097 o = (U8*)SvPV(other, olen);
3098
3099 if (!olen)
b0e3252e 3100 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3101
3102 s = (U8*)SvPV(swatch, slen);
3103 if (bits == 1 && otherbits == 1) {
3104 if (slen != olen)
5637ef5b
NC
3105 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3106 "mismatch, slen=%"UVuf", olen=%"UVuf,
3107 (UV)slen, (UV)olen);
979f2922
ST
3108
3109 switch (opc) {
3110 case '+':
3111 while (slen--)
3112 *s++ |= *o++;
3113 break;
3114 case '!':
3115 while (slen--)
3116 *s++ |= ~*o++;
3117 break;
3118 case '-':
3119 while (slen--)
3120 *s++ &= ~*o++;
3121 break;
3122 case '&':
3123 while (slen--)
3124 *s++ &= *o++;
3125 break;
3126 default:
3127 break;
3128 }
3129 }
711a919c 3130 else {
979f2922
ST
3131 STRLEN otheroctets = otherbits >> 3;
3132 STRLEN offset = 0;
35da51f7 3133 U8* const send = s + slen;
979f2922
ST
3134
3135 while (s < send) {
3136 UV otherval = 0;
3137
3138 if (otherbits == 1) {
3139 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3140 ++offset;
3141 }
3142 else {
3143 STRLEN vlen = otheroctets;
3144 otherval = *o++;
3145 while (--vlen) {
3146 otherval <<= 8;
3147 otherval |= *o++;
3148 }
3149 }
3150
711a919c 3151 if (opc == '+' && otherval)
6f207bd3 3152 NOOP; /* replace with otherval */
979f2922
ST
3153 else if (opc == '!' && !otherval)
3154 otherval = 1;
3155 else if (opc == '-' && otherval)
3156 otherval = 0;
3157 else if (opc == '&' && !otherval)
3158 otherval = 0;
3159 else {
711a919c 3160 s += octets; /* no replacement */
979f2922
ST
3161 continue;
3162 }
3163
3164 if (bits == 8)
3165 *s++ = (U8)( otherval & 0xff);
3166 else if (bits == 16) {
3167 *s++ = (U8)((otherval >> 8) & 0xff);
3168 *s++ = (U8)( otherval & 0xff);
3169 }
3170 else if (bits == 32) {
3171 *s++ = (U8)((otherval >> 24) & 0xff);
3172 *s++ = (U8)((otherval >> 16) & 0xff);
3173 *s++ = (U8)((otherval >> 8) & 0xff);
3174 *s++ = (U8)( otherval & 0xff);
3175 }
3176 }
3177 }
3178 sv_free(other); /* through with it! */
3179 } /* while */
3180 return swatch;
3181}
3182
064c021d 3183HV*
4c2e1131 3184Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3185{
3186
79a2a0e8 3187 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3188 * Can't be used on a property that is subject to user override, as it
3189 * relies on the value of SPECIALS in the swash which would be set by
3190 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3191 * for overridden properties
064c021d
KW
3192 *
3193 * Returns a hash which is the inversion and closure of a swash mapping.
3194 * For example, consider the input lines:
3195 * 004B 006B
3196 * 004C 006C
3197 * 212A 006B
3198 *
3199 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3200 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3201 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3202 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3203 *
538e84ed
KW
3204 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3205 * keys are only code points that are folded-to, so it isn't a full closure.
3206 *
064c021d
KW
3207 * Essentially, for any code point, it gives all the code points that map to
3208 * it, or the list of 'froms' for that point.
3209 *
5662e334
KW
3210 * Currently it ignores any additions or deletions from other swashes,
3211 * looking at just the main body of the swash, and if there are SPECIALS
3212 * in the swash, at that hash
3213 *
3214 * The specials hash can be extra code points, and most likely consists of
3215 * maps from single code points to multiple ones (each expressed as a string
3216 * of utf8 characters). This function currently returns only 1-1 mappings.
3217 * However consider this possible input in the specials hash:
3218 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3219 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3220 *
3221 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3222 * currently handle. But it also means that FB05 and FB06 are equivalent in
3223 * a 1-1 mapping which we should handle, and this relationship may not be in
3224 * the main table. Therefore this function examines all the multi-char
3225 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3226
3227 U8 *l, *lend;
3228 STRLEN lcur;
3229 HV *const hv = MUTABLE_HV(SvRV(swash));
3230
923b6d4e
KW
3231 /* The string containing the main body of the table. This will have its
3232 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3233 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3234
3235 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3236 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3237 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3238 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3239 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3240 const STRLEN bits = SvUV(*bitssvp);
3241 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3242 const UV none = SvUV(*nonesvp);
5662e334 3243 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3244
3245 HV* ret = newHV();
3246
3247 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3248
3249 /* Must have at least 8 bits to get the mappings */
3250 if (bits != 8 && bits != 16 && bits != 32) {
3251 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3252 (UV)bits);
3253 }
3254
5662e334
KW
3255 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3256 mapping to more than one character */
3257
3258 /* Construct an inverse mapping hash for the specials */
3259 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3260 HV * specials_inverse = newHV();
3261 char *char_from; /* the lhs of the map */
3262 I32 from_len; /* its byte length */
3263 char *char_to; /* the rhs of the map */
3264 I32 to_len; /* its byte length */
3265 SV *sv_to; /* and in a sv */
3266 AV* from_list; /* list of things that map to each 'to' */
3267
3268 hv_iterinit(specials_hv);
3269
3270 /* The keys are the characters (in utf8) that map to the corresponding
3271 * utf8 string value. Iterate through the list creating the inverse
3272 * list. */
3273 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3274 SV** listp;
3275 if (! SvPOK(sv_to)) {
5637ef5b
NC
3276 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3277 "unexpectedly is not a string, flags=%lu",
3278 (unsigned long)SvFLAGS(sv_to));
5662e334 3279 }
4b88fb76 3280 /*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
3281
3282 /* Each key in the inverse list is a mapped-to value, and the key's
3283 * hash value is a list of the strings (each in utf8) that map to
3284 * it. Those strings are all one character long */
3285 if ((listp = hv_fetch(specials_inverse,
3286 SvPVX(sv_to),
3287 SvCUR(sv_to), 0)))
3288 {
3289 from_list = (AV*) *listp;
3290 }
3291 else { /* No entry yet for it: create one */
3292 from_list = newAV();
3293 if (! hv_store(specials_inverse,
3294 SvPVX(sv_to),
3295 SvCUR(sv_to),
3296 (SV*) from_list, 0))
3297 {
3298 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3299 }
3300 }
3301
3302 /* Here have the list associated with this 'to' (perhaps newly
3303 * created and empty). Just add to it. Note that we ASSUME that
3304 * the input is guaranteed to not have duplications, so we don't
3305 * check for that. Duplications just slow down execution time. */
3306 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3307 }
3308
3309 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3310 * it looking for cases like the FB05/FB06 examples above. There would
3311 * be an entry in the hash like
3312 * 'st' => [ FB05, FB06 ]
3313 * In this example we will create two lists that get stored in the
3314 * returned hash, 'ret':
3315 * FB05 => [ FB05, FB06 ]
3316 * FB06 => [ FB05, FB06 ]
3317 *
3318 * Note that there is nothing to do if the array only has one element.
3319 * (In the normal 1-1 case handled below, we don't have to worry about
3320 * two lists, as everything gets tied to the single list that is
3321 * generated for the single character 'to'. But here, we are omitting
3322 * that list, ('st' in the example), so must have multiple lists.) */
3323 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3324 &char_to, &to_len)))
3325 {
b9f2b683 3326 if (av_tindex(from_list) > 0) {
c70927a6 3327 SSize_t i;
5662e334
KW
3328
3329 /* We iterate over all combinations of i,j to place each code
3330 * point on each list */
b9f2b683 3331 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3332 SSize_t j;
5662e334
KW
3333 AV* i_list = newAV();
3334 SV** entryp = av_fetch(from_list, i, FALSE);
3335 if (entryp == NULL) {
3336 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3337 }
3338 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3339 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3340 }
3341 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3342 (SV*) i_list, FALSE))
3343 {
3344 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3345 }
3346
538e84ed 3347 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3348 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3349 entryp = av_fetch(from_list, j, FALSE);
3350 if (entryp == NULL) {
3351 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3352 }
3353
3354 /* When i==j this adds itself to the list */
4b88fb76
KW
3355 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3356 (U8*) SvPVX(*entryp),
3357 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3358 0)));
4637d003 3359 /*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
3360 }
3361 }
3362 }
3363 }
3364 SvREFCNT_dec(specials_inverse); /* done with it */
3365 } /* End of specials */
3366
064c021d
KW
3367 /* read $swash->{LIST} */
3368 l = (U8*)SvPV(*listsvp, lcur);
3369 lend = l + lcur;
3370
3371 /* Go through each input line */
3372 while (l < lend) {
3373 UV min, max, val;
3374 UV inverse;
95543e92
KW
3375 l = swash_scan_list_line(l, lend, &min, &max, &val,
3376 cBOOL(octets), typestr);
064c021d
KW
3377 if (l > lend) {
3378 break;
3379 }
3380
3381 /* Each element in the range is to be inverted */
3382 for (inverse = min; inverse <= max; inverse++) {
3383 AV* list;
064c021d
KW
3384 SV** listp;
3385 IV i;
3386 bool found_key = FALSE;
5662e334 3387 bool found_inverse = FALSE;
064c021d
KW
3388
3389 /* The key is the inverse mapping */
3390 char key[UTF8_MAXBYTES+1];
c80e42f3 3391 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3392 STRLEN key_len = key_end - key;
3393
064c021d
KW
3394 /* Get the list for the map */
3395 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3396 list = (AV*) *listp;
3397 }
3398 else { /* No entry yet for it: create one */
3399 list = newAV();
3400 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3401 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3402 }
3403 }
3404
5662e334
KW
3405 /* Look through list to see if this inverse mapping already is
3406 * listed, or if there is a mapping to itself already */
b9f2b683 3407 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3408 SV** entryp = av_fetch(list, i, FALSE);
3409 SV* entry;
414db8a1 3410 UV uv;
064c021d
KW
3411 if (entryp == NULL) {
3412 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3413 }
3414 entry = *entryp;
414db8a1
DM
3415 uv = SvUV(entry);
3416 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3417 if (uv == val) {
064c021d 3418 found_key = TRUE;
5662e334 3419 }
414db8a1 3420 if (uv == inverse) {
5662e334
KW
3421 found_inverse = TRUE;
3422 }
3423
3424 /* No need to continue searching if found everything we are
3425 * looking for */
3426 if (found_key && found_inverse) {
064c021d
KW
3427 break;
3428 }
3429 }
56ca34ca
KW
3430
3431 /* Make sure there is a mapping to itself on the list */
064c021d 3432 if (! found_key) {
d397ff6a 3433 av_push(list, newSVuv(val));
4637d003 3434 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3435 }
3436
3437
3438 /* Simply add the value to the list */
5662e334
KW
3439 if (! found_inverse) {
3440 av_push(list, newSVuv(inverse));
4637d003 3441 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3442 }
064c021d 3443
b0e3252e 3444 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3445 * range. That makes more compact tables possible. You can
3446 * express the capitalization, for example, of all consecutive
3447 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3448 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3449 * and it's not documented; it appears to be used only in
b0e3252e 3450 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3451 * in case */
064c021d
KW
3452 if (!none || val < none) {
3453 ++val;
3454 }
3455 }
3456 }
3457
3458 return ret;
3459}
3460
a25abddc 3461SV*
d764b54e
KW
3462Perl__swash_to_invlist(pTHX_ SV* const swash)
3463{
3464
ed92f1b3
KW
3465 /* Subject to change or removal. For use only in one place in regcomp.c.
3466 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3467
3468 U8 *l, *lend;
3469 char *loc;
3470 STRLEN lcur;
3471 HV *const hv = MUTABLE_HV(SvRV(swash));
3472 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3473 U8 empty[] = "";
923b6d4e
KW
3474 SV** listsvp;
3475 SV** typesvp;
3476 SV** bitssvp;
3477 SV** extssvp;
3478 SV** invert_it_svp;
d764b54e 3479
923b6d4e
KW
3480 U8* typestr;
3481 STRLEN bits;
3482 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3483 U8 *x, *xend;
3484 STRLEN xcur;
d764b54e 3485
a25abddc 3486 SV* invlist;
d764b54e 3487
b81740c0
KW
3488 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3489
923b6d4e
KW
3490 /* If not a hash, it must be the swash's inversion list instead */
3491 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3492 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3493 }
3494
3495 /* The string containing the main body of the table */
3496 listsvp = hv_fetchs(hv, "LIST", FALSE);
3497 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3498 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3499 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3500 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3501
3502 typestr = (U8*)SvPV_nolen(*typesvp);
3503 bits = SvUV(*bitssvp);
3504 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3505
d764b54e 3506 /* read $swash->{LIST} */
b443038a
KW
3507 if (SvPOK(*listsvp)) {
3508 l = (U8*)SvPV(*listsvp, lcur);
3509 }
3510 else {
3511 /* LIST legitimately doesn't contain a string during compilation phases
3512 * of Perl itself, before the Unicode tables are generated. In this
3513 * case, just fake things up by creating an empty list */
3514 l = empty;
3515 lcur = 0;
3516 }
d764b54e
KW
3517 loc = (char *) l;
3518 lend = l + lcur;
3519
31aa6e0b 3520 if (*l == 'V') { /* Inversion list format */
99944878 3521 const char *after_atou = (char *) lend;
31aa6e0b
KW
3522 UV element0;
3523 UV* other_elements_ptr;
3524
3525 /* The first number is a count of the rest */
3526 l++;
99944878 3527 elements = grok_atou((const char *)l, &after_atou);
eb092534
KW
3528 if (elements == 0) {
3529 invlist = _new_invlist(0);
3530 }
3531 else {
99944878
JH
3532 while (isSPACE(*l)) l++;
3533 l = (U8 *) after_atou;
1f9f7d4c
KW
3534
3535 /* Get the 0th element, which is needed to setup the inversion list */
99944878
JH
3536 while (isSPACE(*l)) l++;
3537 element0 = (UV) grok_atou((const char *)l, &after_atou);
3538 l = (U8 *) after_atou;
1f9f7d4c
KW
3539 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3540 elements--;
3541
3542 /* Then just populate the rest of the input */
3543 while (elements-- > 0) {
3544 if (l > lend) {
3545 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3546 }
99944878
JH
3547 while (isSPACE(*l)) l++;
3548 *other_elements_ptr++ = (UV) grok_atou((const char *)l, &after_atou);
3549 l = (U8 *) after_atou;
1f9f7d4c 3550 }
eb092534 3551 }
31aa6e0b
KW
3552 }
3553 else {
3554
1784d2f9
KW
3555 /* Scan the input to count the number of lines to preallocate array
3556 * size based on worst possible case, which is each line in the input
3557 * creates 2 elements in the inversion list: 1) the beginning of a
3558 * range in the list; 2) the beginning of a range not in the list. */
3559 while ((loc = (strchr(loc, '\n'))) != NULL) {
3560 elements += 2;
3561 loc++;
3562 }
d764b54e 3563
1784d2f9
KW
3564 /* If the ending is somehow corrupt and isn't a new line, add another
3565 * element for the final range that isn't in the inversion list */
3566 if (! (*lend == '\n'
3567 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3568 {
3569 elements++;
3570 }
d764b54e 3571
1784d2f9 3572 invlist = _new_invlist(elements);
d764b54e 3573
1784d2f9
KW
3574 /* Now go through the input again, adding each range to the list */
3575 while (l < lend) {
3576 UV start, end;
3577 UV val; /* Not used by this function */
d764b54e 3578
95543e92
KW
3579 l = swash_scan_list_line(l, lend, &start, &end, &val,
3580 cBOOL(octets), typestr);
d764b54e 3581
1784d2f9
KW
3582 if (l > lend) {
3583 break;
3584 }
3585
3586 invlist = _add_range_to_invlist(invlist, start, end);
3587 }
31aa6e0b 3588 }
d764b54e 3589
77f9f126
KW
3590 /* Invert if the data says it should be */
3591 if (invert_it_svp && SvUV(*invert_it_svp)) {
25151030 3592 _invlist_invert(invlist);
77f9f126
KW
3593 }
3594
b0e3252e 3595 /* This code is copied from swatch_get()
d73c39c5
KW
3596 * read $swash->{EXTRAS} */
3597 x = (U8*)SvPV(*extssvp, xcur);
3598 xend = x + xcur;
3599 while (x < xend) {
3600 STRLEN namelen;
3601 U8 *namestr;
3602 SV** othersvp;
3603 HV* otherhv;
3604 STRLEN otherbits;
3605 SV **otherbitssvp, *other;
3606 U8 *nl;
3607
3608 const U8 opc = *x++;
3609 if (opc == '\n')
3610 continue;
3611
3612 nl = (U8*)memchr(x, '\n', xend - x);
3613
3614 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3615 if (nl) {
3616 x = nl + 1; /* 1 is length of "\n" */
3617 continue;
3618 }
3619 else {
3620 x = xend; /* to EXTRAS' end at which \n is not found */
3621 break;
3622 }
3623 }
3624
3625 namestr = x;
3626 if (nl) {
3627 namelen = nl - namestr;
3628 x = nl + 1;
3629 }
3630 else {
3631 namelen = xend - namestr;
3632 x = xend;
3633 }
3634
3635 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3636 otherhv = MUTABLE_HV(SvRV(*othersvp));
3637 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3638 otherbits = (STRLEN)SvUV(*otherbitssvp);
3639
3640 if (bits != otherbits || bits != 1) {
5637ef5b
NC
3641 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
3642 "properties, bits=%"UVuf", otherbits=%"UVuf,
3643 (UV)bits, (UV)otherbits);
d73c39c5
KW
3644 }
3645
3646 /* The "other" swatch must be destroyed after. */
3647 other = _swash_to_invlist((SV *)*othersvp);
3648
b0e3252e 3649 /* End of code copied from swatch_get() */
d73c39c5
KW
3650 switch (opc) {
3651 case '+':
3652 _invlist_union(invlist, other, &invlist);
3653 break;
3654 case '!':
6c46377d 3655 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
d73c39c5
KW
3656 break;
3657 case '-':
3658 _invlist_subtract(invlist, other, &invlist);
3659 break;
3660 case '&':
3661 _invlist_intersection(invlist, other, &invlist);
3662 break;
3663 default:
3664 break;
3665 }
3666 sv_free(other); /* through with it! */
3667 }
3668
dbfdbd26 3669 SvREADONLY_on(invlist);
d764b54e
KW
3670 return invlist;
3671}
3672
3fdfee00
KW
3673SV*
3674Perl__get_swash_invlist(pTHX_ SV* const swash)
3675{
872dd7e0 3676 SV** ptr;
3fdfee00
KW
3677
3678 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
3679
87367d5f 3680 if (! SvROK(swash)) {
872dd7e0
KW
3681 return NULL;
3682 }
3683
87367d5f
KW
3684 /* If it really isn't a hash, it isn't really swash;