This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Make is_utf8_char_buf() a macro
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
81e983c1 34#include "inline_invlist.c"
b24b43f7 35#include "charclass_invlists.h"
a0ed51b3 36
27da23d5
JH
37static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
a0ed51b3 42
166f8a29 43This file contains various utility functions for manipulating UTF8-encoded
72d33970 44strings. For the uninitiated, this is a method of representing arbitrary
61296642 45Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
46characters in the ASCII range are unmodified, and a zero byte never appears
47within non-zero characters.
166f8a29 48
eaf7a4d2
CS
49=cut
50*/
51
52/*
53=for apidoc is_ascii_string
54
a1433954 55Returns true if the first C<len> bytes of the string C<s> are the same whether
970ea3cb
KW
56or not the string is encoded in UTF-8 (or UTF-EBCDIC on EBCDIC machines). That
57is, if they are invariant. On ASCII-ish machines, only ASCII characters
58fit this definition, hence the function's name.
eaf7a4d2 59
75200dff
KW
60If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
61use this option, that C<s> can't have embedded C<NUL> characters and has to
62have a terminating C<NUL> byte).
9f7e3d64 63
a1433954 64See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
65
66=cut
67*/
68
69bool
668b6d8d 70Perl_is_ascii_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
71{
72 const U8* const send = s + (len ? len : strlen((const char *)s));
73 const U8* x = s;
74
75 PERL_ARGS_ASSERT_IS_ASCII_STRING;
eaf7a4d2
CS
76
77 for (; x < send; ++x) {
78 if (!UTF8_IS_INVARIANT(*x))
79 break;
80 }
81
82 return x == send;
83}
84
85/*
378516de 86=for apidoc uvoffuni_to_utf8_flags
eebe1485 87
a27992cc 88THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
89Instead, B<Almost all code should use L</uvchr_to_utf8> or
90L</uvchr_to_utf8_flags>>.
a27992cc 91
de69f3af
KW
92This function is like them, but the input is a strict Unicode
93(as opposed to native) code point. Only in very rare circumstances should code
94not be using the native code point.
949cf498 95
de69f3af 96For details, see the description for L</uvchr_to_utf8_flags>>.
949cf498 97
eebe1485
SC
98=cut
99*/
100
dfe13c55 101U8 *
378516de 102Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 103{
378516de 104 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 105
d9432125
KW
106 if (UNI_IS_INVARIANT(uv)) {
107 *d++ = (U8) LATIN1_TO_NATIVE(uv);
108 return d;
109 }
110
979f77b6
KW
111 /* The first problematic code point is the first surrogate */
112 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 113 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 114 {
949cf498
KW
115 if (UNICODE_IS_SURROGATE(uv)) {
116 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 117 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
118 "UTF-16 surrogate U+%04"UVXf, uv);
119 }
120 if (flags & UNICODE_DISALLOW_SURROGATE) {
121 return NULL;
122 }
123 }
124 else if (UNICODE_IS_SUPER(uv)) {
125 if (flags & UNICODE_WARN_SUPER
126 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
127 {
8457b38f 128 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
129 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
130 }
131 if (flags & UNICODE_DISALLOW_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
133 {
134 return NULL;
135 }
136 }
137 else if (UNICODE_IS_NONCHAR(uv)) {
138 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 139 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
949cf498
KW
140 "Unicode non-character U+%04"UVXf" is illegal for open interchange",
141 uv);
142 }
143 if (flags & UNICODE_DISALLOW_NONCHAR) {
144 return NULL;
145 }
146 }
507b9800 147 }
d9432125 148
2d331972 149#if defined(EBCDIC)
d9432125 150 {
5aaebcb3 151 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
152 U8 *p = d+len-1;
153 while (p > d) {
bc3632a8 154 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
155 uv >>= UTF_ACCUMULATION_SHIFT;
156 }
bc3632a8 157 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
158 return d+len;
159 }
160#else /* Non loop style */
a0ed51b3 161 if (uv < 0x800) {
eb160463
GS
162 *d++ = (U8)(( uv >> 6) | 0xc0);
163 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
164 return d;
165 }
166 if (uv < 0x10000) {
eb160463
GS
167 *d++ = (U8)(( uv >> 12) | 0xe0);
168 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
169 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
170 return d;
171 }
172 if (uv < 0x200000) {
eb160463
GS
173 *d++ = (U8)(( uv >> 18) | 0xf0);
174 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
175 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
176 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
177 return d;
178 }
179 if (uv < 0x4000000) {
eb160463
GS
180 *d++ = (U8)(( uv >> 24) | 0xf8);
181 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
182 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
183 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
184 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
185 return d;
186 }
187 if (uv < 0x80000000) {
eb160463
GS
188 *d++ = (U8)(( uv >> 30) | 0xfc);
189 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
190 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
191 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
193 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
194 return d;
195 }
6588300d 196#ifdef UTF8_QUAD_MAX
d7578b48 197 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
198#endif
199 {
eb160463
GS
200 *d++ = 0xfe; /* Can't match U+FEFF! */
201 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
203 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
204 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
205 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
206 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
207 return d;
208 }
6588300d 209#ifdef UTF8_QUAD_MAX
a0ed51b3 210 {
eb160463
GS
211 *d++ = 0xff; /* Can't match U+FFFE! */
212 *d++ = 0x80; /* 6 Reserved bits */
213 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
214 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
216 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
217 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
218 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
219 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
220 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
221 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
222 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
223 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
224 return d;
225 }
226#endif
537124e4 227#endif /* Non loop style */
a0ed51b3 228}
646ca15d 229/*
07693fe6
KW
230=for apidoc uvchr_to_utf8
231
bcb1a2d4 232Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
233of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
234C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
235the byte after the end of the new character. In other words,
07693fe6
KW
236
237 d = uvchr_to_utf8(d, uv);
238
239is the recommended wide native character-aware way of saying
240
241 *(d++) = uv;
242
de69f3af
KW
243This function accepts any UV as input. To forbid or warn on non-Unicode code
244points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
245
07693fe6
KW
246=cut
247*/
248
de69f3af
KW
249/* This is also a macro */
250PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
251
07693fe6
KW
252U8 *
253Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
254{
de69f3af 255 return uvchr_to_utf8(d, uv);
07693fe6
KW
256}
257
de69f3af
KW
258/*
259=for apidoc uvchr_to_utf8_flags
260
261Adds the UTF-8 representation of the native code point C<uv> to the end
c749c9fd
KW
262of the string C<d>; C<d> should have at least C<UNISKIP(uv)+1> (up to
263C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
264the byte after the end of the new character. In other words,
de69f3af
KW
265
266 d = uvchr_to_utf8_flags(d, uv, flags);
267
268or, in most cases,
269
270 d = uvchr_to_utf8_flags(d, uv, 0);
271
272This is the Unicode-aware way of saying
273
274 *(d++) = uv;
275
276This function will convert to UTF-8 (and not warn) even code points that aren't
277legal Unicode or are problematic, unless C<flags> contains one or more of the
278following flags:
279
280If C<uv> is a Unicode surrogate code point and UNICODE_WARN_SURROGATE is set,
281the function will raise a warning, provided UTF8 warnings are enabled. If instead
282UNICODE_DISALLOW_SURROGATE is set, the function will fail and return NULL.
283If both flags are set, the function will both warn and return NULL.
284
4c3cfd5d 285The UNICODE_WARN_NONCHAR and UNICODE_DISALLOW_NONCHAR flags
de69f3af 286affect how the function handles a Unicode non-character. And likewise, the
4c3cfd5d 287UNICODE_WARN_SUPER and UNICODE_DISALLOW_SUPER flags affect the handling of
de69f3af
KW
288code points that are
289above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
290even less portable) can be warned and/or disallowed even if other above-Unicode
291code points are accepted, by the UNICODE_WARN_FE_FF and UNICODE_DISALLOW_FE_FF
292flags.
293
294And finally, the flag UNICODE_WARN_ILLEGAL_INTERCHANGE selects all four of the
295above WARN flags; and UNICODE_DISALLOW_ILLEGAL_INTERCHANGE selects all four
296DISALLOW flags.
297
298=cut
299*/
300
301/* This is also a macro */
302PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
303
07693fe6
KW
304U8 *
305Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
306{
de69f3af 307 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
308}
309
310/*
492a624f
KW
311=for apidoc is_utf8_char_buf
312
3cedd9d9 313This is identical to the macro L</isUTF8_CHAR>.
492a624f
KW
314
315=cut */
316
317STRLEN
318Perl_is_utf8_char_buf(const U8 *buf, const U8* buf_end)
319{
320
492a624f
KW
321 PERL_ARGS_ASSERT_IS_UTF8_CHAR_BUF;
322
6302f837 323 return isUTF8_CHAR(buf, buf_end);
492a624f
KW
324}
325
326/*
87cea99e 327=for apidoc is_utf8_char
eebe1485 328
5da9da9e 329Tests if some arbitrary number of bytes begins in a valid UTF-8
2bbc8d55
SP
330character. Note that an INVARIANT (i.e. ASCII on non-EBCDIC machines)
331character is a valid UTF-8 character. The actual number of bytes in the UTF-8
332character will be returned if it is valid, otherwise 0.
9041c2e3 333
76848387 334This function is deprecated due to the possibility that malformed input could
a1433954 335cause reading beyond the end of the input buffer. Use L</is_utf8_char_buf>
76848387 336instead.
e0328548 337
82686b01 338=cut */
76848387 339
067a85ef 340STRLEN
668b6d8d 341Perl_is_utf8_char(const U8 *s)
386d01d6 342{
7918f24d 343 PERL_ARGS_ASSERT_IS_UTF8_CHAR;
492a624f 344
76848387 345 /* Assumes we have enough space, which is why this is deprecated */
6302f837 346 return isUTF8_CHAR(s, s + UTF8SKIP(s));
386d01d6
GS
347}
348
eaf7a4d2 349
6662521e 350/*
87cea99e 351=for apidoc is_utf8_string
6662521e 352
a1433954 353Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 354UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
355using C<strlen(s)> (which means if you use this option, that C<s> can't have
356embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
357that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 358
a1433954 359See also L</is_ascii_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 360
6662521e
GS
361=cut
362*/
363
8e84507e 364bool
668b6d8d 365Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 366{
35da51f7 367 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 368 const U8* x = s;
067a85ef 369
7918f24d 370 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 371
6662521e 372 while (x < send) {
6302f837
KW
373 STRLEN len = isUTF8_CHAR(x, send);
374 if (UNLIKELY(! len)) {
375 return FALSE;
376 }
377 x += len;
6662521e 378 }
768c67ee 379
067a85ef 380 return TRUE;
6662521e
GS
381}
382
67e989fb 383/*
814fafa7
NC
384Implemented as a macro in utf8.h
385
87cea99e 386=for apidoc is_utf8_string_loc
814fafa7 387
a1433954
KW
388Like L</is_utf8_string> but stores the location of the failure (in the
389case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
390"utf8ness success") in the C<ep>.
391
a1433954 392See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 393
87cea99e 394=for apidoc is_utf8_string_loclen
81cd54e3 395
a1433954
KW
396Like L</is_utf8_string>() but stores the location of the failure (in the
397case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
398"utf8ness success") in the C<ep>, and the number of UTF-8
399encoded characters in the C<el>.
400
a1433954 401See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
402
403=cut
404*/
405
406bool
668b6d8d 407Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 408{
35da51f7 409 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 410 const U8* x = s;
3ebfea28 411 STRLEN outlen = 0;
7918f24d
NC
412
413 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 414
81cd54e3 415 while (x < send) {
6302f837
KW
416 STRLEN len = isUTF8_CHAR(x, send);
417 if (UNLIKELY(! len)) {
418 goto out;
419 }
420 x += len;
421 outlen++;
81cd54e3 422 }
768c67ee
JH
423
424 out:
3ebfea28
AL
425 if (el)
426 *el = outlen;
427
768c67ee
JH
428 if (ep)
429 *ep = x;
3ebfea28 430 return (x == send);
81cd54e3
JH
431}
432
433/*
768c67ee 434
de69f3af 435=for apidoc utf8n_to_uvchr
378516de
KW
436
437THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 438Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 439
9041c2e3 440Bottom level UTF-8 decode routine.
de69f3af 441Returns the native code point value of the first character in the string C<s>,
746afd53
KW
442which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
443C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
444the length, in bytes, of that character.
949cf498
KW
445
446The value of C<flags> determines the behavior when C<s> does not point to a
447well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
448zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
449next possible position in C<s> that could begin a non-malformed character.
450Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
451
452Various ALLOW flags can be set in C<flags> to allow (and not warn on)
453individual types of malformations, such as the sequence being overlong (that
454is, when there is a shorter sequence that can express the same code point;
455overlong sequences are expressly forbidden in the UTF-8 standard due to
456potential security issues). Another malformation example is the first byte of
457a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
458flags. For allowed 0 length strings, this function returns 0; for allowed
459overlong sequences, the computed code point is returned; for all other allowed
460malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
461determinable reasonable value.
949cf498
KW
462
463The UTF8_CHECK_ONLY flag overrides the behavior when a non-allowed (by other
464flags) malformation is found. If this flag is set, the routine assumes that
465the caller will raise a warning, and this function will silently just set
d088425d
KW
466C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
467
75200dff 468Note that this API requires disambiguation between successful decoding a C<NUL>
d088425d
KW
469character, and an error return (unless the UTF8_CHECK_ONLY flag is set), as
470in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
471first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
472input had an error.
949cf498
KW
473
474Certain code points are considered problematic. These are Unicode surrogates,
746afd53 475Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 476By default these are considered regular code points, but certain situations
5eafe189 477warrant special handling for them. If C<flags> contains
949cf498
KW
478UTF8_DISALLOW_ILLEGAL_INTERCHANGE, all three classes are treated as
479malformations and handled as such. The flags UTF8_DISALLOW_SURROGATE,
480UTF8_DISALLOW_NONCHAR, and UTF8_DISALLOW_SUPER (meaning above the legal Unicode
481maximum) can be set to disallow these categories individually.
482
483The flags UTF8_WARN_ILLEGAL_INTERCHANGE, UTF8_WARN_SURROGATE,
484UTF8_WARN_NONCHAR, and UTF8_WARN_SUPER will cause warning messages to be raised
485for their respective categories, but otherwise the code points are considered
486valid (not malformations). To get a category to both be treated as a
487malformation and raise a warning, specify both the WARN and DISALLOW flags.
488(But note that warnings are not raised if lexically disabled nor if
489UTF8_CHECK_ONLY is also specified.)
490
491Very large code points (above 0x7FFF_FFFF) are considered more problematic than
492the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
493reasons: they requre at least 32 bits to represent them on ASCII platforms, are
494not representable at all on EBCDIC platforms, and the original UTF-8
495specification never went above this number (the current 0x10FFFF limit was
496imposed later). (The smaller ones, those that fit into 32 bits, are
497representable by a UV on ASCII platforms, but not by an IV, which means that
498the number of operations that can be performed on them is quite restricted.)
499The UTF-8 encoding on ASCII platforms for these large code points begins with a
500byte containing 0xFE or 0xFF. The UTF8_DISALLOW_FE_FF flag will cause them to
501be treated as malformations, while allowing smaller above-Unicode code points.
502(Of course UTF8_DISALLOW_SUPER will treat all above-Unicode code points,
72d33970
FC
503including these, as malformations.)
504Similarly, UTF8_WARN_FE_FF acts just like
eb83ed87 505the other WARN flags, but applies just to these code points.
949cf498
KW
506
507All other code points corresponding to Unicode characters, including private
508use and those yet to be assigned, are never considered malformed and never
509warn.
67e989fb 510
37607a96
PK
511=cut
512*/
67e989fb 513
a0ed51b3 514UV
de69f3af 515Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 516{
97aff369 517 dVAR;
d4c19fe8 518 const U8 * const s0 = s;
eb83ed87 519 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 520 U8 * send;
eb83ed87
KW
521 UV uv = *s;
522 STRLEN expectlen;
949cf498 523 SV* sv = NULL;
eb83ed87
KW
524 UV outlier_ret = 0; /* return value when input is in error or problematic
525 */
526 UV pack_warn = 0; /* Save result of packWARN() for later */
527 bool unexpected_non_continuation = FALSE;
528 bool overflowed = FALSE;
2f8f112e 529 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 530
eb83ed87 531 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 532
de69f3af 533 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 534
eb83ed87
KW
535 /* The order of malformation tests here is important. We should consume as
536 * few bytes as possible in order to not skip any valid character. This is
537 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
538 * http://unicode.org/reports/tr36 for more discussion as to why. For
539 * example, once we've done a UTF8SKIP, we can tell the expected number of
540 * bytes, and could fail right off the bat if the input parameters indicate
541 * that there are too few available. But it could be that just that first
542 * byte is garbled, and the intended character occupies fewer bytes. If we
543 * blindly assumed that the first byte is correct, and skipped based on
544 * that number, we could skip over a valid input character. So instead, we
545 * always examine the sequence byte-by-byte.
546 *
547 * We also should not consume too few bytes, otherwise someone could inject
548 * things. For example, an input could be deliberately designed to
549 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 550 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
551 * which is actually part of of the overflowing sequence), that could look
552 * legitimate to the caller, which could discard the initial partial
553 * sequence and process the rest, inappropriately */
554
555 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 556 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
557 if (retlen) {
558 *retlen = 0;
559 }
a0dbb045 560
eb83ed87
KW
561 if (flags & UTF8_ALLOW_EMPTY) {
562 return 0;
563 }
564 if (! (flags & UTF8_CHECK_ONLY)) {
565 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
566 }
0c443dc2
JH
567 goto malformed;
568 }
569
eb83ed87
KW
570 expectlen = UTF8SKIP(s);
571
572 /* A well-formed UTF-8 character, as the vast majority of calls to this
573 * function will be for, has this expected length. For efficiency, set
574 * things up here to return it. It will be overriden only in those rare
575 * cases where a malformation is found */
576 if (retlen) {
577 *retlen = expectlen;
578 }
579
580 /* An invariant is trivially well-formed */
1d72bdf6 581 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 582 return uv;
a0ed51b3 583 }
67e989fb 584
eb83ed87 585 /* A continuation character can't start a valid sequence */
b5b9af04 586 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
587 if (flags & UTF8_ALLOW_CONTINUATION) {
588 if (retlen) {
589 *retlen = 1;
590 }
591 return UNICODE_REPLACEMENT;
592 }
ba210ebe 593
eb83ed87
KW
594 if (! (flags & UTF8_CHECK_ONLY)) {
595 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
596 }
597 curlen = 1;
ba210ebe
JH
598 goto malformed;
599 }
9041c2e3 600
dcd27b3c
KW
601 /* Here is not a continuation byte, nor an invariant. The only thing left
602 * is a start byte (possibly for an overlong) */
603
1d72bdf6 604#ifdef EBCDIC
bc3632a8 605 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
606#endif
607
eb83ed87
KW
608 /* Remove the leading bits that indicate the number of bytes in the
609 * character's whole UTF-8 sequence, leaving just the bits that are part of
610 * the value */
611 uv &= UTF_START_MASK(expectlen);
ba210ebe 612
eb83ed87
KW
613 /* Now, loop through the remaining bytes in the character's sequence,
614 * accumulating each into the working value as we go. Be sure to not look
615 * past the end of the input string */
0b8d30e8
KW
616 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
617
eb83ed87 618 for (s = s0 + 1; s < send; s++) {
b5b9af04 619 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
620#ifndef EBCDIC /* Can't overflow in EBCDIC */
621 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
622
623 /* The original implementors viewed this malformation as more
624 * serious than the others (though I, khw, don't understand
625 * why, since other malformations also give very very wrong
626 * results), so there is no way to turn off checking for it.
627 * Set a flag, but keep going in the loop, so that we absorb
628 * the rest of the bytes that comprise the character. */
629 overflowed = TRUE;
630 overflow_byte = *s; /* Save for warning message's use */
631 }
632#endif
8850bf83 633 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
634 }
635 else {
636 /* Here, found a non-continuation before processing all expected
637 * bytes. This byte begins a new character, so quit, even if
638 * allowing this malformation. */
639 unexpected_non_continuation = TRUE;
640 break;
641 }
642 } /* End of loop through the character's bytes */
643
644 /* Save how many bytes were actually in the character */
645 curlen = s - s0;
646
647 /* The loop above finds two types of malformations: non-continuation and/or
648 * overflow. The non-continuation malformation is really a too-short
649 * malformation, as it means that the current character ended before it was
650 * expected to (being terminated prematurely by the beginning of the next
651 * character, whereas in the too-short malformation there just are too few
652 * bytes available to hold the character. In both cases, the check below
653 * that we have found the expected number of bytes would fail if executed.)
654 * Thus the non-continuation malformation is really unnecessary, being a
655 * subset of the too-short malformation. But there may be existing
656 * applications that are expecting the non-continuation type, so we retain
657 * it, and return it in preference to the too-short malformation. (If this
658 * code were being written from scratch, the two types might be collapsed
659 * into one.) I, khw, am also giving priority to returning the
660 * non-continuation and too-short malformations over overflow when multiple
661 * ones are present. I don't know of any real reason to prefer one over
662 * the other, except that it seems to me that multiple-byte errors trumps
663 * errors from a single byte */
b5b9af04 664 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
665 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
666 if (! (flags & UTF8_CHECK_ONLY)) {
667 if (curlen == 1) {
668 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
669 }
670 else {
671 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
672 }
673 }
eb83ed87
KW
674 goto malformed;
675 }
676 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
677
678 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
679 * as what the original expectations were. */
680 do_overlong_test = FALSE;
eb83ed87
KW
681 if (retlen) {
682 *retlen = curlen;
683 }
684 }
b5b9af04 685 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
686 if (! (flags & UTF8_ALLOW_SHORT)) {
687 if (! (flags & UTF8_CHECK_ONLY)) {
688 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 689 }
eb83ed87
KW
690 goto malformed;
691 }
692 uv = UNICODE_REPLACEMENT;
2f8f112e 693 do_overlong_test = FALSE;
eb83ed87
KW
694 if (retlen) {
695 *retlen = curlen;
696 }
697 }
698
ea5ced44 699#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 700 if (UNLIKELY(overflowed)) {
eb83ed87 701 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 702 goto malformed;
eb83ed87
KW
703 }
704#endif
705
2f8f112e 706 if (do_overlong_test
5aaebcb3 707 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
708 && ! (flags & UTF8_ALLOW_LONG))
709 {
eb83ed87
KW
710 /* The overlong malformation has lower precedence than the others.
711 * Note that if this malformation is allowed, we return the actual
712 * value, instead of the replacement character. This is because this
713 * value is actually well-defined. */
714 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 715 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
716 }
717 goto malformed;
718 }
719
1a89bb6c 720 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
721 * problematic code point that is not allowed by the input parameters. */
722 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
723 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
724 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
725 {
949cf498 726 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
727
728 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
729 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 730 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 731 && ckWARN_d(WARN_SURROGATE))
eb83ed87 732 {
111d382d 733 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 734 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
735 }
736 if (flags & UTF8_DISALLOW_SURROGATE) {
737 goto disallowed;
738 }
739 }
949cf498 740 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 741 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 742 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 743 {
111d382d 744 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 745 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 746 }
ea5ced44
KW
747#ifndef EBCDIC /* EBCDIC always allows FE, FF */
748
749 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
750 * points. We test for these after the regular SUPER ones, and
751 * before possibly bailing out, so that the more dire warning
752 * overrides the regular one, if applicable */
753 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
754 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
755 {
756 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
757 == UTF8_WARN_FE_FF
758 && ckWARN_d(WARN_UTF8))
759 {
760 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
761 pack_warn = packWARN(WARN_UTF8);
762 }
763 if (flags & UTF8_DISALLOW_FE_FF) {
764 goto disallowed;
765 }
766 }
767#endif
949cf498
KW
768 if (flags & UTF8_DISALLOW_SUPER) {
769 goto disallowed;
770 }
771 }
4190d317
KW
772 else if (UNICODE_IS_NONCHAR(uv)) {
773 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 774 && ckWARN_d(WARN_NONCHAR))
4190d317
KW
775 {
776 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is illegal for open interchange", uv));
54f4afef 777 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
778 }
779 if (flags & UTF8_DISALLOW_NONCHAR) {
780 goto disallowed;
781 }
782 }
949cf498 783
eb83ed87 784 if (sv) {
de69f3af
KW
785 outlier_ret = uv; /* Note we don't bother to convert to native,
786 as all the outlier code points are the same
787 in both ASCII and EBCDIC */
eb83ed87
KW
788 goto do_warn;
789 }
790
949cf498
KW
791 /* Here, this is not considered a malformed character, so drop through
792 * to return it */
a0ed51b3 793 }
ba210ebe 794
de69f3af 795 return UNI_TO_NATIVE(uv);
ba210ebe 796
eb83ed87
KW
797 /* There are three cases which get to beyond this point. In all 3 cases:
798 * <sv> if not null points to a string to print as a warning.
799 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
800 * set.
801 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
802 * This is done by initializing it to 0, and changing it only
803 * for case 1).
804 * The 3 cases are:
805 * 1) The input is valid but problematic, and to be warned about. The
806 * return value is the resultant code point; <*retlen> is set to
807 * <curlen>, the number of bytes that comprise the code point.
808 * <pack_warn> contains the result of packWARN() for the warning
809 * types. The entry point for this case is the label <do_warn>;
810 * 2) The input is a valid code point but disallowed by the parameters to
811 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
812 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
813 * comprise the code point. <pack_warn> contains the result of
814 * packWARN() for the warning types. The entry point for this case is
815 * the label <disallowed>.
816 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
817 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
818 * bytes that comprise the malformation. All such malformations are
819 * assumed to be warning type <utf8>. The entry point for this case
820 * is the label <malformed>.
821 */
949cf498 822
ba210ebe
JH
823malformed:
824
eb83ed87
KW
825 if (sv && ckWARN_d(WARN_UTF8)) {
826 pack_warn = packWARN(WARN_UTF8);
827 }
828
829disallowed:
830
fcc8fcf6 831 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 832 if (retlen)
10edeb5d 833 *retlen = ((STRLEN) -1);
ba210ebe
JH
834 return 0;
835 }
836
eb83ed87 837do_warn:
5b311467 838
eb83ed87
KW
839 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
840 if warnings are to be raised. */
f555bc63 841 const char * const string = SvPVX_const(sv);
a0dbb045 842
f555bc63
KW
843 if (PL_op)
844 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
845 else
846 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
847 }
848
eb83ed87
KW
849 if (retlen) {
850 *retlen = curlen;
851 }
ba210ebe 852
eb83ed87 853 return outlier_ret;
a0ed51b3
LW
854}
855
8e84507e 856/*
ec5f19d0
KW
857=for apidoc utf8_to_uvchr_buf
858
859Returns the native code point of the first character in the string C<s> which
860is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 861C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 862
524080c4
KW
863If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
864enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
173db420
KW
865NULL) to -1. If those warnings are off, the computed value, if well-defined
866(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
867C<*retlen> is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is
868the next possible position in C<s> that could begin a non-malformed character.
de69f3af 869See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 870returned.
ec5f19d0
KW
871
872=cut
873*/
874
875
876UV
877Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
878{
ec5f19d0
KW
879 assert(s < send);
880
881 return utf8n_to_uvchr(s, send - s, retlen,
882 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
883}
884
27d6c58a 885/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 886 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 887 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
888
889UV
890Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
891{
010ab96b
KW
892 UV expectlen = UTF8SKIP(s);
893 const U8* send = s + expectlen;
9ff2f0f7 894 UV uv = *s;
3986bb7c 895
27d6c58a
KW
896 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
897
010ab96b
KW
898 if (retlen) {
899 *retlen = expectlen;
900 }
901
902 /* An invariant is trivially returned */
903 if (expectlen == 1) {
9ff2f0f7 904 return uv;
010ab96b
KW
905 }
906
9ff2f0f7
KW
907#ifdef EBCDIC
908 uv = NATIVE_UTF8_TO_I8(uv);
909#endif
910
010ab96b
KW
911 /* Remove the leading bits that indicate the number of bytes, leaving just
912 * the bits that are part of the value */
913 uv &= UTF_START_MASK(expectlen);
914
915 /* Now, loop through the remaining bytes, accumulating each into the
916 * working total as we go. (I khw tried unrolling the loop for up to 4
917 * bytes, but there was no performance improvement) */
918 for (++s; s < send; s++) {
919 uv = UTF8_ACCUMULATE(uv, *s);
920 }
921
3986bb7c 922 return UNI_TO_NATIVE(uv);
010ab96b 923
27d6c58a
KW
924}
925
ec5f19d0 926/*
87cea99e 927=for apidoc utf8_to_uvchr
9041c2e3 928
6ee84de2 929Returns the native code point of the first character in the string C<s>
1e54db1a 930which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
931length, in bytes, of that character.
932
4b88fb76 933Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31
KW
934malformed input could cause reading beyond the end of the input buffer, which
935is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4b88fb76 936
524080c4
KW
937If C<s> points to one of the detected malformations, and UTF8 warnings are
938enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
939NULL) to -1. If those warnings are off, the computed value if well-defined (or
940the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
941is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
942next possible position in C<s> that could begin a non-malformed character.
de69f3af 943See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
9041c2e3
NIS
944
945=cut
946*/
947
948UV
7fc63493 949Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
9041c2e3 950{
7918f24d
NC
951 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
952
2ff6c191 953 return utf8_to_uvchr_buf(s, s + UTF8_MAXBYTES, retlen);
9041c2e3
NIS
954}
955
956/*
ec5f19d0
KW
957=for apidoc utf8_to_uvuni_buf
958
de69f3af
KW
959Only in very rare circumstances should code need to be dealing in Unicode
960(as opposed to native) code points. In those few cases, use
961C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
962
963Returns the Unicode (not-native) code point of the first character in the
964string C<s> which
ec5f19d0
KW
965is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
966C<retlen> will be set to the length, in bytes, of that character.
967
524080c4
KW
968If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
969enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
970NULL) to -1. If those warnings are off, the computed value if well-defined (or
971the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
972is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
973next possible position in C<s> that could begin a non-malformed character.
de69f3af 974See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
975
976=cut
977*/
978
979UV
980Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
981{
982 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
983
984 assert(send > s);
985
986 /* Call the low level routine asking for checks */
de69f3af
KW
987 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
988 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
989}
990
5495102a
KW
991/* DEPRECATED!
992 * Like L</utf8_to_uvuni_buf>(), but should only be called when it is known that
2114036c 993 * there are no malformations in the input UTF-8 string C<s>. Surrogates,
3986bb7c 994 * non-character code points, and non-Unicode code points are allowed */
27d6c58a
KW
995
996UV
997Perl_valid_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
998{
999 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVUNI;
1000
010ab96b 1001 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
27d6c58a
KW
1002}
1003
ec5f19d0 1004/*
87cea99e 1005=for apidoc utf8_to_uvuni
9041c2e3
NIS
1006
1007Returns the Unicode code point of the first character in the string C<s>
1e54db1a 1008which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
9041c2e3
NIS
1009length, in bytes, of that character.
1010
4b88fb76 1011Some, but not all, UTF-8 malformations are detected, and in fact, some
977c1d31 1012malformed input could cause reading beyond the end of the input buffer, which
4f83cdcd
KW
1013is one reason why this function is deprecated. The other is that only in
1014extremely limited circumstances should the Unicode versus native code point be
de69f3af 1015of any interest to you. See L</utf8_to_uvuni_buf> for alternatives.
9041c2e3 1016
524080c4
KW
1017If C<s> points to one of the detected malformations, and UTF8 warnings are
1018enabled, zero is returned and C<*retlen> is set (if C<retlen> doesn't point to
1019NULL) to -1. If those warnings are off, the computed value if well-defined (or
1020the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1021is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1022next possible position in C<s> that could begin a non-malformed character.
de69f3af 1023See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
8e84507e
NIS
1024
1025=cut
1026*/
1027
1028UV
7fc63493 1029Perl_utf8_to_uvuni(pTHX_ const U8 *s, STRLEN *retlen)
8e84507e 1030{
7918f24d
NC
1031 PERL_ARGS_ASSERT_UTF8_TO_UVUNI;
1032
5495102a 1033 return NATIVE_TO_UNI(valid_utf8_to_uvchr(s, retlen));
8e84507e
NIS
1034}
1035
b76347f2 1036/*
87cea99e 1037=for apidoc utf8_length
b76347f2
JH
1038
1039Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1040Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1041up past C<e>, croaks.
b76347f2
JH
1042
1043=cut
1044*/
1045
1046STRLEN
35a4481c 1047Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2 1048{
97aff369 1049 dVAR;
b76347f2
JH
1050 STRLEN len = 0;
1051
7918f24d
NC
1052 PERL_ARGS_ASSERT_UTF8_LENGTH;
1053
8850bf83
JH
1054 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1055 * the bitops (especially ~) can create illegal UTF-8.
1056 * In other words: in Perl UTF-8 is not just for Unicode. */
1057
a3b680e6
AL
1058 if (e < s)
1059 goto warn_and_return;
b76347f2 1060 while (s < e) {
4cbf4130 1061 s += UTF8SKIP(s);
8e91ec7f
AV
1062 len++;
1063 }
1064
1065 if (e != s) {
1066 len--;
1067 warn_and_return:
9b387841
NC
1068 if (PL_op)
1069 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1070 "%s in %s", unees, OP_DESC(PL_op));
1071 else
61a12c31 1072 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1073 }
1074
1075 return len;
1076}
1077
b06226ff 1078/*
87cea99e 1079=for apidoc utf8_distance
b06226ff 1080
1e54db1a 1081Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1082and C<b>.
1083
1084WARNING: use only if you *know* that the pointers point inside the
1085same UTF-8 buffer.
1086
37607a96
PK
1087=cut
1088*/
a0ed51b3 1089
02eb7b47 1090IV
35a4481c 1091Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1092{
7918f24d
NC
1093 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1094
bf1665bc 1095 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1096}
1097
b06226ff 1098/*
87cea99e 1099=for apidoc utf8_hop
b06226ff 1100
8850bf83
JH
1101Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1102forward or backward.
b06226ff
JH
1103
1104WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1105the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1106on the first byte of character or just after the last byte of a character.
b06226ff 1107
37607a96
PK
1108=cut
1109*/
a0ed51b3
LW
1110
1111U8 *
4373e329 1112Perl_utf8_hop(pTHX_ const U8 *s, I32 off)
a0ed51b3 1113{
7918f24d
NC
1114 PERL_ARGS_ASSERT_UTF8_HOP;
1115
96a5add6 1116 PERL_UNUSED_CONTEXT;
8850bf83
JH
1117 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1118 * the bitops (especially ~) can create illegal UTF-8.
1119 * In other words: in Perl UTF-8 is not just for Unicode. */
1120
a0ed51b3
LW
1121 if (off >= 0) {
1122 while (off--)
1123 s += UTF8SKIP(s);
1124 }
1125 else {
1126 while (off++) {
1127 s--;
8850bf83
JH
1128 while (UTF8_IS_CONTINUATION(*s))
1129 s--;
a0ed51b3
LW
1130 }
1131 }
4373e329 1132 return (U8 *)s;
a0ed51b3
LW
1133}
1134
6940069f 1135/*
fed3ba5d
NC
1136=for apidoc bytes_cmp_utf8
1137
a1433954 1138Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1139sequence of characters (stored as UTF-8)
1140in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1141equal, -1 or -2 if the first string is less than the second string, +1 or +2
1142if the first string is greater than the second string.
1143
1144-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1145longer string. -2 or +2 is returned if
1146there was a difference between characters
fed3ba5d
NC
1147within the strings.
1148
1149=cut
1150*/
1151
1152int
1153Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1154{
1155 const U8 *const bend = b + blen;
1156 const U8 *const uend = u + ulen;
1157
1158 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1159
1160 PERL_UNUSED_CONTEXT;
1161
1162 while (b < bend && u < uend) {
1163 U8 c = *u++;
1164 if (!UTF8_IS_INVARIANT(c)) {
1165 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1166 if (u < uend) {
1167 U8 c1 = *u++;
1168 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1169 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1170 } else {
1171 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1172 "Malformed UTF-8 character "
1173 "(unexpected non-continuation byte 0x%02x"
1174 ", immediately after start byte 0x%02x)"
1175 /* Dear diag.t, it's in the pod. */
1176 "%s%s", c1, c,
1177 PL_op ? " in " : "",
1178 PL_op ? OP_DESC(PL_op) : "");
1179 return -2;
1180 }
1181 } else {
1182 if (PL_op)
1183 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1184 "%s in %s", unees, OP_DESC(PL_op));
1185 else
61a12c31 1186 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1187 return -2; /* Really want to return undef :-) */
1188 }
1189 } else {
1190 return -2;
1191 }
1192 }
1193 if (*b != c) {
1194 return *b < c ? -2 : +2;
1195 }
1196 ++b;
1197 }
1198
1199 if (b == bend && u == uend)
1200 return 0;
1201
1202 return b < bend ? +1 : -1;
1203}
1204
1205/*
87cea99e 1206=for apidoc utf8_to_bytes
6940069f 1207
2bbc8d55 1208Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1209Unlike L</bytes_to_utf8>, this over-writes the original string, and
1210updates C<len> to contain the new length.
67e989fb 1211Returns zero on failure, setting C<len> to -1.
6940069f 1212
a1433954 1213If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1214
6940069f
GS
1215=cut
1216*/
1217
1218U8 *
37607a96 1219Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1220{
d4c19fe8
AL
1221 U8 * const save = s;
1222 U8 * const send = s + *len;
6940069f 1223 U8 *d;
246fae53 1224
7918f24d
NC
1225 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1226
1e54db1a 1227 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1228 while (s < send) {
d59937ca
KW
1229 if (! UTF8_IS_INVARIANT(*s)) {
1230 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1231 *len = ((STRLEN) -1);
1232 return 0;
1233 }
1234 s++;
dcad2880 1235 }
d59937ca 1236 s++;
246fae53 1237 }
dcad2880
JH
1238
1239 d = s = save;
6940069f 1240 while (s < send) {
80e0b38f
KW
1241 U8 c = *s++;
1242 if (! UTF8_IS_INVARIANT(c)) {
1243 /* Then it is two-byte encoded */
1244 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1245 s++;
1246 }
1247 *d++ = c;
6940069f
GS
1248 }
1249 *d = '\0';
246fae53 1250 *len = d - save;
6940069f
GS
1251 return save;
1252}
1253
1254/*
87cea99e 1255=for apidoc bytes_from_utf8
f9a63242 1256
2bbc8d55 1257Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1258Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1259the newly-created string, and updates C<len> to contain the new
1260length. Returns the original string if no conversion occurs, C<len>
72d33970 1261is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55
SP
12620 if C<s> is converted or consisted entirely of characters that are invariant
1263in utf8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1264
37607a96
PK
1265=cut
1266*/
f9a63242
JH
1267
1268U8 *
e1ec3a88 1269Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1270{
f9a63242 1271 U8 *d;
e1ec3a88
AL
1272 const U8 *start = s;
1273 const U8 *send;
f9a63242
JH
1274 I32 count = 0;
1275
7918f24d
NC
1276 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1277
96a5add6 1278 PERL_UNUSED_CONTEXT;
f9a63242 1279 if (!*is_utf8)
73d840c0 1280 return (U8 *)start;
f9a63242 1281
1e54db1a 1282 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1283 for (send = s + *len; s < send;) {
d59937ca
KW
1284 if (! UTF8_IS_INVARIANT(*s)) {
1285 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1286 return (U8 *)start;
d59937ca
KW
1287 }
1288 count++;
1289 s++;
db42d148 1290 }
d59937ca 1291 s++;
f9a63242
JH
1292 }
1293
35da51f7 1294 *is_utf8 = FALSE;
f9a63242 1295
212542aa 1296 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1297 s = start; start = d;
f9a63242
JH
1298 while (s < send) {
1299 U8 c = *s++;
1a91c45d 1300 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1301 /* Then it is two-byte encoded */
1a91c45d
KW
1302 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1303 s++;
c4d5f83a
NIS
1304 }
1305 *d++ = c;
f9a63242
JH
1306 }
1307 *d = '\0';
1308 *len = d - start;
73d840c0 1309 return (U8 *)start;
f9a63242
JH
1310}
1311
1312/*
87cea99e 1313=for apidoc bytes_to_utf8
6940069f 1314
ff97e5cf
KW
1315Converts a string C<s> of length C<len> bytes from the native encoding into
1316UTF-8.
6662521e 1317Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1318reflect the new length in bytes.
6940069f 1319
75200dff 1320A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1321
1322If you want to convert to UTF-8 from encodings other than
1323the native (Latin1 or EBCDIC),
a1433954 1324see L</sv_recode_to_utf8>().
c9ada85f 1325
497711e7 1326=cut
6940069f
GS
1327*/
1328
c682ebef
FC
1329/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1330 likewise need duplication. */
1331
6940069f 1332U8*
35a4481c 1333Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1334{
35a4481c 1335 const U8 * const send = s + (*len);
6940069f
GS
1336 U8 *d;
1337 U8 *dst;
7918f24d
NC
1338
1339 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1340 PERL_UNUSED_CONTEXT;
6940069f 1341
212542aa 1342 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1343 dst = d;
1344
1345 while (s < send) {
55d09dc8
KW
1346 append_utf8_from_native_byte(*s, &d);
1347 s++;
6940069f
GS
1348 }
1349 *d = '\0';
6662521e 1350 *len = d-dst;
6940069f
GS
1351 return dst;
1352}
1353
a0ed51b3 1354/*
dea0fc0b 1355 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1356 *
1357 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1358 * We optimize for native, for obvious reasons. */
1359
1360U8*
dea0fc0b 1361Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1362{
dea0fc0b
JH
1363 U8* pend;
1364 U8* dstart = d;
1365
7918f24d
NC
1366 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1367
dea0fc0b 1368 if (bytelen & 1)
f5992bc4 1369 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1370
1371 pend = p + bytelen;
1372
a0ed51b3 1373 while (p < pend) {
dea0fc0b
JH
1374 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1375 p += 2;
56d37426
KW
1376 if (UNI_IS_INVARIANT(uv)) {
1377 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1378 continue;
1379 }
56d37426
KW
1380 if (uv <= MAX_UTF8_TWO_BYTE) {
1381 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1382 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1383 continue;
1384 }
46956fad
KW
1385#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1386#define LAST_HIGH_SURROGATE 0xDBFF
1387#define FIRST_LOW_SURROGATE 0xDC00
1388#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
1389 if (uv >= FIRST_HIGH_SURROGATE && uv <= LAST_HIGH_SURROGATE) {
01ea242b 1390 if (p >= pend) {
dea0fc0b 1391 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
01ea242b
NC
1392 } else {
1393 UV low = (p[0] << 8) + p[1];
1394 p += 2;
46956fad 1395 if (low < FIRST_LOW_SURROGATE || low > LAST_LOW_SURROGATE)
01ea242b 1396 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
46956fad
KW
1397 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1398 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1399 }
46956fad 1400 } else if (uv >= FIRST_LOW_SURROGATE && uv <= LAST_LOW_SURROGATE) {
dbde1951 1401 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
a0ed51b3 1402 }
56d37426
KW
1403#ifdef EBCDIC
1404 d = uvoffuni_to_utf8_flags(d, uv, 0);
1405#else
a0ed51b3 1406 if (uv < 0x10000) {
eb160463
GS
1407 *d++ = (U8)(( uv >> 12) | 0xe0);
1408 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1409 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1410 continue;
1411 }
1412 else {
eb160463
GS
1413 *d++ = (U8)(( uv >> 18) | 0xf0);
1414 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1415 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1416 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1417 continue;
1418 }
56d37426 1419#endif
a0ed51b3 1420 }
dea0fc0b 1421 *newlen = d - dstart;
a0ed51b3
LW
1422 return d;
1423}
1424
1425/* Note: this one is slightly destructive of the source. */
1426
1427U8*
dea0fc0b 1428Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1429{
1430 U8* s = (U8*)p;
d4c19fe8 1431 U8* const send = s + bytelen;
7918f24d
NC
1432
1433 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1434
e0ea5e2d
NC
1435 if (bytelen & 1)
1436 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1437 (UV)bytelen);
1438
a0ed51b3 1439 while (s < send) {
d4c19fe8 1440 const U8 tmp = s[0];
a0ed51b3
LW
1441 s[0] = s[1];
1442 s[1] = tmp;
1443 s += 2;
1444 }
dea0fc0b 1445 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1446}
1447
922e8cb4
KW
1448bool
1449Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1450{
1451 U8 tmpbuf[UTF8_MAXBYTES+1];
1452 uvchr_to_utf8(tmpbuf, c);
1453 return _is_utf8_FOO(classnum, tmpbuf);
1454}
1455
f9ae8fb6
JD
1456/* Internal function so we can deprecate the external one, and call
1457 this one from other deprecated functions in this file */
1458
1459PERL_STATIC_INLINE bool
61b19385
KW
1460S_is_utf8_idfirst(pTHX_ const U8 *p)
1461{
1462 dVAR;
1463
1464 if (*p == '_')
1465 return TRUE;
1466 /* is_utf8_idstart would be more logical. */
f25ce844 1467 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1468}
1469
5092f92a 1470bool
84afefe6 1471Perl_is_uni_idfirst(pTHX_ UV c)
a0ed51b3 1472{
89ebb4a3 1473 U8 tmpbuf[UTF8_MAXBYTES+1];
230880c1 1474 uvchr_to_utf8(tmpbuf, c);
61b19385 1475 return S_is_utf8_idfirst(aTHX_ tmpbuf);
a0ed51b3
LW
1476}
1477
1478bool
eba68aa0
KW
1479Perl__is_uni_perl_idcont(pTHX_ UV c)
1480{
1481 U8 tmpbuf[UTF8_MAXBYTES+1];
1482 uvchr_to_utf8(tmpbuf, c);
1483 return _is_utf8_perl_idcont(tmpbuf);
1484}
1485
1486bool
f91dcd13
KW
1487Perl__is_uni_perl_idstart(pTHX_ UV c)
1488{
1489 U8 tmpbuf[UTF8_MAXBYTES+1];
1490 uvchr_to_utf8(tmpbuf, c);
1491 return _is_utf8_perl_idstart(tmpbuf);
1492}
1493
3a4c58c9
KW
1494UV
1495Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1496{
1497 /* We have the latin1-range values compiled into the core, so just use
1498 * those, converting the result to utf8. The only difference between upper
1499 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1500 * either "SS" or "Ss". Which one to use is passed into the routine in
1501 * 'S_or_s' to avoid a test */
1502
1503 UV converted = toUPPER_LATIN1_MOD(c);
1504
1505 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1506
1507 assert(S_or_s == 'S' || S_or_s == 's');
1508
6f2d5cbc 1509 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1510 characters in this range */
3a4c58c9
KW
1511 *p = (U8) converted;
1512 *lenp = 1;
1513 return converted;
1514 }
1515
1516 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1517 * which it maps to one of them, so as to only have to have one check for
1518 * it in the main case */
1519 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1520 switch (c) {
1521 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1522 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1523 break;
1524 case MICRO_SIGN:
1525 converted = GREEK_CAPITAL_LETTER_MU;
1526 break;
1527 case LATIN_SMALL_LETTER_SHARP_S:
1528 *(p)++ = 'S';
1529 *p = S_or_s;
1530 *lenp = 2;
1531 return 'S';
1532 default:
1533 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
118e2215 1534 assert(0); /* NOTREACHED */
3a4c58c9
KW
1535 }
1536 }
1537
1538 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1539 *p = UTF8_TWO_BYTE_LO(converted);
1540 *lenp = 2;
1541
1542 return converted;
1543}
1544
50bda2c3
KW
1545/* Call the function to convert a UTF-8 encoded character to the specified case.
1546 * Note that there may be more than one character in the result.
1547 * INP is a pointer to the first byte of the input character
1548 * OUTP will be set to the first byte of the string of changed characters. It
1549 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1550 * LENP will be set to the length in bytes of the string of changed characters
1551 *
1552 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1553#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1554#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1555#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1556
1557/* This additionally has the input parameter SPECIALS, which if non-zero will
1558 * cause this to use the SPECIALS hash for folding (meaning get full case
1559 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1560#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1561
84afefe6
JH
1562UV
1563Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1564{
3a4c58c9
KW
1565 dVAR;
1566
a1433954
KW
1567 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1568 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1569 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1570 * the changed version may be longer than the original character.
1571 *
1572 * The ordinal of the first character of the changed version is returned
1573 * (but note, as explained above, that there may be more.) */
1574
7918f24d
NC
1575 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1576
3a4c58c9
KW
1577 if (c < 256) {
1578 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1579 }
1580
0ebc6274 1581 uvchr_to_utf8(p, c);
3a4c58c9 1582 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1583}
1584
84afefe6
JH
1585UV
1586Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1587{
3a4c58c9
KW
1588 dVAR;
1589
7918f24d
NC
1590 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1591
3a4c58c9
KW
1592 if (c < 256) {
1593 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1594 }
1595
0ebc6274 1596 uvchr_to_utf8(p, c);
3a4c58c9 1597 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1598}
1599
afc16117
KW
1600STATIC U8
1601S_to_lower_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp)
1602{
1603 /* We have the latin1-range values compiled into the core, so just use
1604 * those, converting the result to utf8. Since the result is always just
a1433954 1605 * one character, we allow <p> to be NULL */
afc16117
KW
1606
1607 U8 converted = toLOWER_LATIN1(c);
1608
1609 if (p != NULL) {
6f2d5cbc 1610 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1611 *p = converted;
1612 *lenp = 1;
1613 }
1614 else {
1615 *p = UTF8_TWO_BYTE_HI(converted);
1616 *(p+1) = UTF8_TWO_BYTE_LO(converted);
1617 *lenp = 2;
1618 }
1619 }
1620 return converted;
1621}
1622
84afefe6
JH
1623UV
1624Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1625{
968c5e6a
KW
1626 dVAR;
1627
7918f24d
NC
1628 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1629
afc16117
KW
1630 if (c < 256) {
1631 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1632 }
1633
afc16117 1634 uvchr_to_utf8(p, c);
968c5e6a 1635 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1636}
1637
84afefe6 1638UV
51910141 1639Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1640{
51910141 1641 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1642 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1643 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1644 *
1645 * Not to be used for locale folds
51910141 1646 */
f673fad4 1647
a1dde8de
KW
1648 UV converted;
1649
1650 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
1651
1ca267a5
KW
1652 assert (! (flags & FOLD_FLAGS_LOCALE));
1653
a1dde8de
KW
1654 if (c == MICRO_SIGN) {
1655 converted = GREEK_SMALL_LETTER_MU;
1656 }
51910141 1657 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1658
1659 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1660 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1661 * under those circumstances. */
1662 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1663 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1664 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1665 p, *lenp, U8);
1666 return LATIN_SMALL_LETTER_LONG_S;
1667 }
1668 else {
4f489194
KW
1669 *(p)++ = 's';
1670 *p = 's';
1671 *lenp = 2;
1672 return 's';
1ca267a5 1673 }
a1dde8de
KW
1674 }
1675 else { /* In this range the fold of all other characters is their lower
1676 case */
1677 converted = toLOWER_LATIN1(c);
1678 }
1679
6f2d5cbc 1680 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1681 *p = (U8) converted;
1682 *lenp = 1;
1683 }
1684 else {
1685 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1686 *p = UTF8_TWO_BYTE_LO(converted);
1687 *lenp = 2;
1688 }
1689
1690 return converted;
1691}
1692
1693UV
31f05a37 1694Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1695{
4b593389 1696
a0270393
KW
1697 /* Not currently externally documented, and subject to change
1698 * <flags> bits meanings:
1699 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1700 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1701 * locale are to be used.
a0270393
KW
1702 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1703 */
4b593389 1704
36bb2ab6 1705 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1706
31f05a37
KW
1707 /* Tread a UTF-8 locale as not being in locale at all */
1708 if (IN_UTF8_CTYPE_LOCALE) {
1709 flags &= ~FOLD_FLAGS_LOCALE;
1710 }
1711
a1dde8de 1712 if (c < 256) {
a0270393 1713 UV result = _to_fold_latin1((U8) c, p, lenp,
31f05a37 1714 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a0270393
KW
1715 /* It is illegal for the fold to cross the 255/256 boundary under
1716 * locale; in this case return the original */
1717 return (result > 256 && flags & FOLD_FLAGS_LOCALE)
1718 ? c
1719 : result;
a1dde8de
KW
1720 }
1721
a0270393
KW
1722 /* If no special needs, just use the macro */
1723 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1724 uvchr_to_utf8(p, c);
1725 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1726 }
1727 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1728 the special flags. */
1729 U8 utf8_c[UTF8_MAXBYTES + 1];
1730 uvchr_to_utf8(utf8_c, c);
445bf929 1731 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1732 }
84afefe6
JH
1733}
1734
26483009 1735PERL_STATIC_INLINE bool
5141f98e 1736S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1737 const char *const swashname, SV* const invlist)
bde6a22d 1738{
ea317ccb
KW
1739 /* returns a boolean giving whether or not the UTF8-encoded character that
1740 * starts at <p> is in the swash indicated by <swashname>. <swash>
1741 * contains a pointer to where the swash indicated by <swashname>
1742 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1743 * look at <*swash> and only generate a swash if it is not null. <invlist>
1744 * is NULL or an inversion list that defines the swash. If not null, it
1745 * saves time during initialization of the swash.
ea317ccb
KW
1746 *
1747 * Note that it is assumed that the buffer length of <p> is enough to
1748 * contain all the bytes that comprise the character. Thus, <*p> should
1749 * have been checked before this call for mal-formedness enough to assure
1750 * that. */
1751
97aff369 1752 dVAR;
7918f24d
NC
1753
1754 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1755
492a624f 1756 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1757 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1758 * as far as there being enough bytes available in it to accommodate the
1759 * character without reading beyond the end, and pass that number on to the
1760 * validating routine */
6302f837 1761 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1762 if (ckWARN_d(WARN_UTF8)) {
1763 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1764 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1765 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1766 what the malformation is */
1767 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1768 }
1769 }
1770 return FALSE;
1771 }
87367d5f
KW
1772 if (!*swash) {
1773 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1774 *swash = _core_swash_init("utf8",
1775
1776 /* Only use the name if there is no inversion
1777 * list; otherwise will go out to disk */
1778 (invlist) ? "" : swashname,
1779
1780 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1781 }
28123549 1782
bde6a22d
NC
1783 return swash_fetch(*swash, p, TRUE) != 0;
1784}
1785
1786bool
922e8cb4
KW
1787Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1788{
1789 dVAR;
1790
1791 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1792
1793 assert(classnum < _FIRST_NON_SWASH_CC);
1794
f25ce844
KW
1795 return is_utf8_common(p,
1796 &PL_utf8_swash_ptrs[classnum],
1797 swash_property_names[classnum],
1798 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1799}
1800
1801bool
7fc63493 1802Perl_is_utf8_idfirst(pTHX_ const U8 *p) /* The naming is historical. */
a0ed51b3 1803{
97aff369 1804 dVAR;
7918f24d
NC
1805
1806 PERL_ARGS_ASSERT_IS_UTF8_IDFIRST;
1807
61b19385 1808 return S_is_utf8_idfirst(aTHX_ p);
82686b01
JH
1809}
1810
1811bool
c11ff943
KW
1812Perl_is_utf8_xidfirst(pTHX_ const U8 *p) /* The naming is historical. */
1813{
1814 dVAR;
1815
1816 PERL_ARGS_ASSERT_IS_UTF8_XIDFIRST;
1817
1818 if (*p == '_')
1819 return TRUE;
1820 /* is_utf8_idstart would be more logical. */
f25ce844 1821 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1822}
1823
1824bool
d65654cb 1825Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
b6912c02
KW
1826{
1827 dVAR;
b24b43f7 1828 SV* invlist = NULL;
b6912c02 1829
d65654cb 1830 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
b6912c02 1831
b24b43f7
KW
1832 if (! PL_utf8_perl_idstart) {
1833 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1834 }
1835 return is_utf8_common(p, &PL_utf8_perl_idstart, "", invlist);
b6912c02
KW
1836}
1837
1838bool
eba68aa0
KW
1839Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1840{
1841 dVAR;
b24b43f7 1842 SV* invlist = NULL;
eba68aa0
KW
1843
1844 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1845
b24b43f7
KW
1846 if (! PL_utf8_perl_idcont) {
1847 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1848 }
1849 return is_utf8_common(p, &PL_utf8_perl_idcont, "", invlist);
eba68aa0
KW
1850}
1851
1852
1853bool
7fc63493 1854Perl_is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1855{
97aff369 1856 dVAR;
7918f24d
NC
1857
1858 PERL_ARGS_ASSERT_IS_UTF8_IDCONT;
1859
f25ce844 1860 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1861}
1862
1863bool
c11ff943
KW
1864Perl_is_utf8_xidcont(pTHX_ const U8 *p)
1865{
1866 dVAR;
1867
1868 PERL_ARGS_ASSERT_IS_UTF8_XIDCONT;
1869
f25ce844 1870 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1871}
1872
1873bool
7dbf68d2
KW
1874Perl__is_utf8_mark(pTHX_ const U8 *p)
1875{
1876 dVAR;
1877
1878 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1879
f25ce844 1880 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1881}
1882
6b5c0936 1883/*
87cea99e 1884=for apidoc to_utf8_case
6b5c0936 1885
6fae5207 1886C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1887the character that is being converted. This routine assumes that the character
1888at C<p> is well-formed.
6b5c0936 1889
6fae5207
KW
1890C<ustrp> is a pointer to the character buffer to put the
1891conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1892of the result.
1893
6fae5207 1894C<swashp> is a pointer to the swash to use.
6b5c0936 1895
a1433954 1896Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
6fae5207 1897and loaded by SWASHNEW, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1898but not always, a multicharacter mapping), is tried first.
6b5c0936 1899
4a8240a3
KW
1900C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1901any special mappings; C<""> means to use the special mappings. Values other
1902than these two are treated as the name of the hash containing the special
1903mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1904
6fae5207 1905C<normal> is a string like "ToLower" which means the swash
0134edef
JH
1906%utf8::ToLower.
1907
1908=cut */
6b5c0936 1909
2104c8d9 1910UV
9a957fbc
AL
1911Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1912 SV **swashp, const char *normal, const char *special)
a0ed51b3 1913{
97aff369 1914 dVAR;
0134edef 1915 STRLEN len = 0;
f4cd282c 1916 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1917
1918 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1919
9ae3ac1a
KW
1920 /* Note that swash_fetch() doesn't output warnings for these because it
1921 * assumes we will */
8457b38f 1922 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1923 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1924 if (ckWARN_d(WARN_SURROGATE)) {
1925 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1926 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1927 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1928 }
9ae3ac1a
KW
1929 }
1930 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1931 if (ckWARN_d(WARN_NON_UNICODE)) {
1932 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1933 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1934 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1935 }
9ae3ac1a
KW
1936 }
1937
1938 /* Note that non-characters are perfectly legal, so no warning should
1939 * be given */
1940 }
1941
0134edef 1942 if (!*swashp) /* load on-demand */
5ab9d2ef 1943 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1944
a6f87d8c 1945 if (special) {
0134edef 1946 /* It might be "special" (sometimes, but not always,
2a37f04d 1947 * a multicharacter mapping) */
4a8240a3 1948 HV *hv = NULL;
b08cf34e
JH
1949 SV **svp;
1950
4a8240a3
KW
1951 /* If passed in the specials name, use that; otherwise use any
1952 * given in the swash */
1953 if (*special != '\0') {
1954 hv = get_hv(special, 0);
1955 }
1956 else {
1957 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1958 if (svp) {
1959 hv = MUTABLE_HV(SvRV(*svp));
1960 }
1961 }
1962
176fe009
KW
1963 if (hv
1964 && (svp = hv_fetch(hv, (const char*)p, UNISKIP(uv1), FALSE))
1965 && (*svp))
1966 {
cfd0369c 1967 const char *s;
47654450 1968
cfd0369c 1969 s = SvPV_const(*svp, len);
47654450 1970 if (len == 1)
f4cd282c 1971 /* EIGHTBIT */
c80e42f3 1972 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1973 else {
d2dcd0fb 1974 Copy(s, ustrp, len, U8);
29e98929 1975 }
983ffd37 1976 }
0134edef
JH
1977 }
1978
1979 if (!len && *swashp) {
f4cd282c 1980 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is utf8 */);
d4c19fe8 1981
0134edef
JH
1982 if (uv2) {
1983 /* It was "normal" (a single character mapping). */
f4cd282c 1984 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1985 }
1986 }
1feea2c7 1987
cbe07460
KW
1988 if (len) {
1989 if (lenp) {
1990 *lenp = len;
1991 }
1992 return valid_utf8_to_uvchr(ustrp, 0);
1993 }
1994
1995 /* Here, there was no mapping defined, which means that the code point maps
1996 * to itself. Return the inputs */
bfdf22ec 1997 len = UTF8SKIP(p);
ca9fab46
KW
1998 if (p != ustrp) { /* Don't copy onto itself */
1999 Copy(p, ustrp, len, U8);
2000 }
0134edef 2001
2a37f04d
JH
2002 if (lenp)
2003 *lenp = len;
2004
f4cd282c 2005 return uv1;
cbe07460 2006
a0ed51b3
LW
2007}
2008
051a06d4
KW
2009STATIC UV
2010S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2011{
2012 /* This is called when changing the case of a utf8-encoded character above
31f05a37
KW
2013 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2014 * result contains a character that crosses the 255/256 boundary, disallow
2015 * the change, and return the original code point. See L<perlfunc/lc> for
2016 * why;
051a06d4 2017 *
a1433954
KW
2018 * p points to the original string whose case was changed; assumed
2019 * by this routine to be well-formed
051a06d4
KW
2020 * result the code point of the first character in the changed-case string
2021 * ustrp points to the changed-case string (<result> represents its first char)
2022 * lenp points to the length of <ustrp> */
2023
2024 UV original; /* To store the first code point of <p> */
2025
2026 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2027
a4f12ed7 2028 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2029
2030 /* We know immediately if the first character in the string crosses the
2031 * boundary, so can skip */
2032 if (result > 255) {
2033
2034 /* Look at every character in the result; if any cross the
2035 * boundary, the whole thing is disallowed */
2036 U8* s = ustrp + UTF8SKIP(ustrp);
2037 U8* e = ustrp + *lenp;
2038 while (s < e) {
a4f12ed7 2039 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2040 goto bad_crossing;
2041 }
2042 s += UTF8SKIP(s);
2043 }
2044
2045 /* Here, no characters crossed, result is ok as-is */
2046 return result;
2047 }
2048
2049bad_crossing:
2050
2051 /* Failed, have to return the original */
4b88fb76 2052 original = valid_utf8_to_uvchr(p, lenp);
051a06d4
KW
2053 Copy(p, ustrp, *lenp, char);
2054 return original;
2055}
2056
d3e79532 2057/*
87cea99e 2058=for apidoc to_utf8_upper
d3e79532 2059
1f607577 2060Instead use L</toUPPER_utf8>.
a1433954 2061
d3e79532
JH
2062=cut */
2063
051a06d4 2064/* Not currently externally documented, and subject to change:
31f05a37
KW
2065 * <flags> is set iff iff the rules from the current underlying locale are to
2066 * be used. */
051a06d4 2067
2104c8d9 2068UV
31f05a37 2069Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2070{
97aff369 2071 dVAR;
7918f24d 2072
051a06d4
KW
2073 UV result;
2074
2075 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2076
31f05a37
KW
2077 if (flags && IN_UTF8_CTYPE_LOCALE) {
2078 flags = FALSE;
2079 }
2080
3a4c58c9 2081 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2082 if (flags) {
2083 result = toUPPER_LC(*p);
2084 }
2085 else {
81c6c7ce 2086 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2087 }
3a4c58c9
KW
2088 }
2089 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2090 if (flags) {
a6d8b88b 2091 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2092 result = toUPPER_LC(c);
051a06d4
KW
2093 }
2094 else {
94bb8c36 2095 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2096 ustrp, lenp, 'S');
051a06d4
KW
2097 }
2098 }
2099 else { /* utf8, ord above 255 */
2100 result = CALL_UPPER_CASE(p, ustrp, lenp);
2101
2102 if (flags) {
2103 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2104 }
2105 return result;
2106 }
2107
2108 /* Here, used locale rules. Convert back to utf8 */
2109 if (UTF8_IS_INVARIANT(result)) {
2110 *ustrp = (U8) result;
2111 *lenp = 1;
2112 }
2113 else {
62cb07ea
KW
2114 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2115 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2116 *lenp = 2;
3a4c58c9 2117 }
baa60164 2118
051a06d4 2119 return result;
983ffd37 2120}
a0ed51b3 2121
d3e79532 2122/*
87cea99e 2123=for apidoc to_utf8_title
d3e79532 2124
1f607577 2125Instead use L</toTITLE_utf8>.
a1433954 2126
d3e79532
JH
2127=cut */
2128
051a06d4 2129/* Not currently externally documented, and subject to change:
31f05a37
KW
2130 * <flags> is set iff the rules from the current underlying locale are to be
2131 * used. Since titlecase is not defined in POSIX, for other than a
2132 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2133 */
051a06d4 2134
983ffd37 2135UV
31f05a37 2136Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2137{
97aff369 2138 dVAR;
7918f24d 2139
051a06d4
KW
2140 UV result;
2141
2142 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2143
31f05a37
KW
2144 if (flags && IN_UTF8_CTYPE_LOCALE) {
2145 flags = FALSE;
2146 }
2147
3a4c58c9 2148 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2149 if (flags) {
2150 result = toUPPER_LC(*p);
2151 }
2152 else {
81c6c7ce 2153 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2154 }
3a4c58c9
KW
2155 }
2156 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2157 if (flags) {
a6d8b88b 2158 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2159 result = toUPPER_LC(c);
051a06d4
KW
2160 }
2161 else {
94bb8c36 2162 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2163 ustrp, lenp, 's');
051a06d4
KW
2164 }
2165 }
2166 else { /* utf8, ord above 255 */
2167 result = CALL_TITLE_CASE(p, ustrp, lenp);
2168
2169 if (flags) {
2170 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2171 }
2172 return result;
2173 }
2174
2175 /* Here, used locale rules. Convert back to utf8 */
2176 if (UTF8_IS_INVARIANT(result)) {
2177 *ustrp = (U8) result;
2178 *lenp = 1;
2179 }
2180 else {
62cb07ea
KW
2181 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2182 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2183 *lenp = 2;
3a4c58c9
KW
2184 }
2185
051a06d4 2186 return result;
a0ed51b3
LW
2187}
2188
d3e79532 2189/*
87cea99e 2190=for apidoc to_utf8_lower
d3e79532 2191
1f607577 2192Instead use L</toLOWER_utf8>.
a1433954 2193
d3e79532
JH
2194=cut */
2195
051a06d4 2196/* Not currently externally documented, and subject to change:
31f05a37
KW
2197 * <flags> is set iff iff the rules from the current underlying locale are to
2198 * be used.
2199 */
051a06d4 2200
2104c8d9 2201UV
31f05a37 2202Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2203{
051a06d4
KW
2204 UV result;
2205
97aff369 2206 dVAR;
7918f24d 2207
051a06d4 2208 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2209
31f05a37
KW
2210 if (flags && IN_UTF8_CTYPE_LOCALE) {
2211 flags = FALSE;
2212 }
2213
968c5e6a 2214 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2215 if (flags) {
2216 result = toLOWER_LC(*p);
2217 }
2218 else {
81c6c7ce 2219 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2220 }
968c5e6a
KW
2221 }
2222 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2223 if (flags) {
a6d8b88b 2224 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2225 result = toLOWER_LC(c);
051a06d4
KW
2226 }
2227 else {
94bb8c36 2228 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2229 ustrp, lenp);
051a06d4 2230 }
968c5e6a 2231 }
051a06d4
KW
2232 else { /* utf8, ord above 255 */
2233 result = CALL_LOWER_CASE(p, ustrp, lenp);
2234
2235 if (flags) {
2236 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2237 }
968c5e6a 2238
051a06d4
KW
2239 return result;
2240 }
2241
2242 /* Here, used locale rules. Convert back to utf8 */
2243 if (UTF8_IS_INVARIANT(result)) {
2244 *ustrp = (U8) result;
2245 *lenp = 1;
2246 }
2247 else {
62cb07ea
KW
2248 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2249 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2250 *lenp = 2;
2251 }
2252
051a06d4 2253 return result;
b4e400f9
JH
2254}
2255
d3e79532 2256/*
87cea99e 2257=for apidoc to_utf8_fold
d3e79532 2258
1f607577 2259Instead use L</toFOLD_utf8>.
a1433954 2260
d3e79532
JH
2261=cut */
2262
051a06d4
KW
2263/* Not currently externally documented, and subject to change,
2264 * in <flags>
31f05a37
KW
2265 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2266 * locale are to be used.
051a06d4
KW
2267 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2268 * otherwise simple folds
a0270393
KW
2269 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2270 * prohibited
445bf929 2271 */
36bb2ab6 2272
b4e400f9 2273UV
445bf929 2274Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2275{
97aff369 2276 dVAR;
7918f24d 2277
051a06d4
KW
2278 UV result;
2279
36bb2ab6 2280 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2281
a0270393
KW
2282 /* These are mutually exclusive */
2283 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2284
50ba90ff
KW
2285 assert(p != ustrp); /* Otherwise overwrites */
2286
31f05a37
KW
2287 if (flags & FOLD_FLAGS_LOCALE && IN_UTF8_CTYPE_LOCALE) {
2288 flags &= ~FOLD_FLAGS_LOCALE;
2289 }
2290
a1dde8de 2291 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2292 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2293 result = toFOLD_LC(*p);
051a06d4
KW
2294 }
2295 else {
81c6c7ce 2296 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2297 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2298 }
a1dde8de
KW
2299 }
2300 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2301 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2302 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2303 result = toFOLD_LC(c);
051a06d4
KW
2304 }
2305 else {
94bb8c36 2306 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2307 ustrp, lenp,
1ca267a5 2308 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2309 }
a1dde8de 2310 }
051a06d4 2311 else { /* utf8, ord above 255 */
a0270393 2312 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2313
1ca267a5
KW
2314 if (flags & FOLD_FLAGS_LOCALE) {
2315
538e84ed
KW
2316 /* Special case these two characters, as what normally gets
2317 * returned under locale doesn't work */
1ca267a5
KW
2318 if (UTF8SKIP(p) == sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1
2319 && memEQ((char *) p, LATIN_CAPITAL_LETTER_SHARP_S_UTF8,
2320 sizeof(LATIN_CAPITAL_LETTER_SHARP_S_UTF8) - 1))
2321 {
2322 goto return_long_s;
2323 }
9fc2026f
KW
2324 else if (UTF8SKIP(p) == sizeof(LATIN_SMALL_LIGATURE_LONG_S_T) - 1
2325 && memEQ((char *) p, LATIN_SMALL_LIGATURE_LONG_S_T_UTF8,
2326 sizeof(LATIN_SMALL_LIGATURE_LONG_S_T_UTF8) - 1))
2327 {
2328 goto return_ligature_st;
2329 }
a0270393 2330 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2331 }
a0270393
KW
2332 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2333 return result;
2334 }
2335 else {
2336 /* This is called when changing the case of a utf8-encoded
9fc2026f
KW
2337 * character above the ASCII range, and the result should not
2338 * contain an ASCII character. */
a0270393
KW
2339
2340 UV original; /* To store the first code point of <p> */
2341
2342 /* Look at every character in the result; if any cross the
2343 * boundary, the whole thing is disallowed */
2344 U8* s = ustrp;
2345 U8* e = ustrp + *lenp;
2346 while (s < e) {
2347 if (isASCII(*s)) {
2348 /* Crossed, have to return the original */
2349 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2350
9fc2026f 2351 /* But in these instances, there is an alternative we can
1ca267a5 2352 * return that is valid */
9fc2026f
KW
2353 if (original == LATIN_CAPITAL_LETTER_SHARP_S
2354 || original == LATIN_SMALL_LETTER_SHARP_S)
2355 {
1ca267a5
KW
2356 goto return_long_s;
2357 }
9fc2026f
KW
2358 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2359 goto return_ligature_st;
2360 }
a0270393
KW
2361 Copy(p, ustrp, *lenp, char);
2362 return original;
2363 }
2364 s += UTF8SKIP(s);
2365 }
051a06d4 2366
a0270393
KW
2367 /* Here, no characters crossed, result is ok as-is */
2368 return result;
2369 }
051a06d4
KW
2370 }
2371
2372 /* Here, used locale rules. Convert back to utf8 */
2373 if (UTF8_IS_INVARIANT(result)) {
2374 *ustrp = (U8) result;
2375 *lenp = 1;
2376 }
2377 else {
62cb07ea
KW
2378 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2379 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2380 *lenp = 2;
2381 }
2382
051a06d4 2383 return result;
1ca267a5
KW
2384
2385 return_long_s:
2386 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2387 * folds to a string of two of these characters. By returning this
2388 * instead, then, e.g.,
2389 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2390 * works. */
2391
2392 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2393 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2394 ustrp, *lenp, U8);
2395 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2396
2397 return_ligature_st:
2398 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2399 * have the other one fold to it */
2400
2401 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2402 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2403 return LATIN_SMALL_LIGATURE_ST;
a0ed51b3
LW
2404}
2405
711a919c 2406/* Note:
f90a9a02 2407 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2408 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2409 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2410 */
c4a5db0c 2411
a0ed51b3 2412SV*
7fc63493 2413Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2414{
c4a5db0c
KW
2415 PERL_ARGS_ASSERT_SWASH_INIT;
2416
2417 /* Returns a copy of a swash initiated by the called function. This is the
2418 * public interface, and returning a copy prevents others from doing
2419 * mischief on the original */
2420
5d3d13d1 2421 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2422}
2423
2424SV*
5d3d13d1 2425Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c
KW
2426{
2427 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2428 * by calling utf8_heavy.pl in the general case. The returned value may be
2429 * the swash's inversion list instead if the input parameters allow it.
2430 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2431 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2432 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2433 *
2434 * This interface should only be used by functions that won't destroy or
2435 * adversely change the swash, as doing so affects all other uses of the
2436 * swash in the program; the general public should use 'Perl_swash_init'
2437 * instead.
2438 *
2439 * pkg is the name of the package that <name> should be in.
2440 * name is the name of the swash to find. Typically it is a Unicode
2441 * property name, including user-defined ones
2442 * listsv is a string to initialize the swash with. It must be of the form
2443 * documented as the subroutine return value in
2444 * L<perlunicode/User-Defined Character Properties>
2445 * minbits is the number of bits required to represent each data element.
2446 * It is '1' for binary properties.
2447 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2448 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2449 * flags_p if non-NULL is the address of various input and output flag bits
2450 * to the routine, as follows: ('I' means is input to the routine;
2451 * 'O' means output from the routine. Only flags marked O are
2452 * meaningful on return.)
2453 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2454 * came from a user-defined property. (I O)
5d3d13d1
KW
2455 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2456 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2457 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2458 * return of an inversion list instead of a swash hash if this routine
2459 * thinks that would result in faster execution of swash_fetch() later
2460 * on. (I)
9a53f6cf
KW
2461 *
2462 * Thus there are three possible inputs to find the swash: <name>,
2463 * <listsv>, and <invlist>. At least one must be specified. The result
2464 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2465 * actions can intersect, etc. what <name> gives. To avoid going out to
2466 * disk at all, <invlist> should specify completely what the swash should
2467 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2468 *
2469 * <invlist> is only valid for binary properties */
c4a5db0c 2470
27da23d5 2471 dVAR;
c4a5db0c 2472 SV* retval = &PL_sv_undef;
83199d38 2473 HV* swash_hv = NULL;
87367d5f
KW
2474 const int invlist_swash_boundary =
2475 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2476 ? 512 /* Based on some benchmarking, but not extensive, see commit
2477 message */
2478 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2479
2480 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2481 assert(! invlist || minbits == 1);
2482
2483 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2484 * so */
2485 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2486 dSP;
2487 const size_t pkg_len = strlen(pkg);
2488 const size_t name_len = strlen(name);
2489 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2490 SV* errsv_save;
2491 GV *method;
2492
2493 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2494
2495 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2496 ENTER;
69794297
KW
2497 SAVEHINTS();
2498 save_re_context();
650f067c
JL
2499 /* We might get here via a subroutine signature which uses a utf8
2500 * parameter name, at which point PL_subname will have been set
2501 * but not yet used. */
2502 save_item(PL_subname);
69794297
KW
2503 if (PL_parser && PL_parser->error_count)
2504 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2505 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
2506 if (!method) { /* demand load utf8 */
2507 ENTER;
db2c6cb3
FC
2508 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2509 GvSV(PL_errgv) = NULL;
69794297
KW
2510 /* It is assumed that callers of this routine are not passing in
2511 * any user derived data. */
2512 /* Need to do this after save_re_context() as it will set
2513 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2514 * in Perl_magic_get). Even line to create errsv_save can turn on
2515 * PL_tainted. */
284167a5
SM
2516#ifndef NO_TAINT_SUPPORT
2517 SAVEBOOL(TAINT_get);
2518 TAINT_NOT;
2519#endif
69794297
KW
2520 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2521 NULL);
eed484f9 2522 {
db2c6cb3
FC
2523 /* Not ERRSV, as there is no need to vivify a scalar we are
2524 about to discard. */
2525 SV * const errsv = GvSV(PL_errgv);
2526 if (!SvTRUE(errsv)) {
2527 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2528 SvREFCNT_dec(errsv);
2529 }
eed484f9 2530 }
69794297
KW
2531 LEAVE;
2532 }
2533 SPAGAIN;
2534 PUSHMARK(SP);
2535 EXTEND(SP,5);
2536 mPUSHp(pkg, pkg_len);
2537 mPUSHp(name, name_len);
2538 PUSHs(listsv);
2539 mPUSHi(minbits);
2540 mPUSHi(none);
2541 PUTBACK;
db2c6cb3
FC
2542 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2543 GvSV(PL_errgv) = NULL;
69794297
KW
2544 /* If we already have a pointer to the method, no need to use
2545 * call_method() to repeat the lookup. */
c41800a8
KW
2546 if (method
2547 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2548 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2549 {
2550 retval = *PL_stack_sp--;
2551 SvREFCNT_inc(retval);
2552 }
eed484f9 2553 {
db2c6cb3
FC
2554 /* Not ERRSV. See above. */
2555 SV * const errsv = GvSV(PL_errgv);
2556 if (!SvTRUE(errsv)) {
2557 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2558 SvREFCNT_dec(errsv);
2559 }
eed484f9 2560 }
ce3b816e 2561 LEAVE;
69794297
KW
2562 POPSTACK;
2563 if (IN_PERL_COMPILETIME) {
2564 CopHINTS_set(PL_curcop, PL_hints);
2565 }
2566 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2567 if (SvPOK(retval))
2568
2569 /* If caller wants to handle missing properties, let them */
5d3d13d1 2570 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
69794297
KW
2571 return NULL;
2572 }
2573 Perl_croak(aTHX_
2574 "Can't find Unicode property definition \"%"SVf"\"",
2575 SVfARG(retval));
2576 Perl_croak(aTHX_ "SWASHNEW didn't return an HV ref");
2577 }
9a53f6cf 2578 } /* End of calling the module to find the swash */
36eb48b4 2579
83199d38
KW
2580 /* If this operation fetched a swash, and we will need it later, get it */
2581 if (retval != &PL_sv_undef
2582 && (minbits == 1 || (flags_p
2583 && ! (*flags_p
2584 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2585 {
2586 swash_hv = MUTABLE_HV(SvRV(retval));
2587
2588 /* If we don't already know that there is a user-defined component to
2589 * this swash, and the user has indicated they wish to know if there is
2590 * one (by passing <flags_p>), find out */
2591 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2592 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2593 if (user_defined && SvUV(*user_defined)) {
2594 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2595 }
2596 }
2597 }
2598
36eb48b4
KW
2599 /* Make sure there is an inversion list for binary properties */
2600 if (minbits == 1) {
2601 SV** swash_invlistsvp = NULL;
2602 SV* swash_invlist = NULL;
9a53f6cf 2603 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2604 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2605 an unclaimed reference count */
36eb48b4 2606
9a53f6cf 2607 /* If this operation fetched a swash, get its already existing
83199d38 2608 * inversion list, or create one for it */
36eb48b4 2609
83199d38 2610 if (swash_hv) {
5c9f4bd2 2611 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2612 if (swash_invlistsvp) {
2613 swash_invlist = *swash_invlistsvp;
2614 invlist_in_swash_is_valid = TRUE;
2615 }
2616 else {
36eb48b4 2617 swash_invlist = _swash_to_invlist(retval);
02c85471 2618 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2619 }
2620 }
2621
2622 /* If an inversion list was passed in, have to include it */
2623 if (invlist) {
2624
2625 /* Any fetched swash will by now have an inversion list in it;
2626 * otherwise <swash_invlist> will be NULL, indicating that we
2627 * didn't fetch a swash */
2628 if (swash_invlist) {
2629
2630 /* Add the passed-in inversion list, which invalidates the one
2631 * already stored in the swash */
2632 invlist_in_swash_is_valid = FALSE;
2633 _invlist_union(invlist, swash_invlist, &swash_invlist);
2634 }
2635 else {
2636
87367d5f
KW
2637 /* Here, there is no swash already. Set up a minimal one, if
2638 * we are going to return a swash */
2639 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2640 swash_hv = newHV();
4aca0fe6 2641 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2642 }
9a53f6cf
KW
2643 swash_invlist = invlist;
2644 }
9a53f6cf
KW
2645 }
2646
2647 /* Here, we have computed the union of all the passed-in data. It may
2648 * be that there was an inversion list in the swash which didn't get
538e84ed 2649 * touched; otherwise save the computed one */
87367d5f
KW
2650 if (! invlist_in_swash_is_valid
2651 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2652 {
5c9f4bd2 2653 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2654 {
2655 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2656 }
cc34d8c5
FC
2657 /* We just stole a reference count. */
2658 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2659 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2660 }
87367d5f 2661
dbfdbd26
KW
2662 SvREADONLY_on(swash_invlist);
2663
c41800a8 2664 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2665 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2666 SvREFCNT_dec(retval);
02c85471
FC
2667 if (!swash_invlist_unclaimed)
2668 SvREFCNT_inc_simple_void_NN(swash_invlist);
2669 retval = newRV_noinc(swash_invlist);
87367d5f 2670 }
36eb48b4
KW
2671 }
2672
a0ed51b3
LW
2673 return retval;
2674}
2675
035d37be
JH
2676
2677/* This API is wrong for special case conversions since we may need to
2678 * return several Unicode characters for a single Unicode character
2679 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2680 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2681 * multiple values. --jhi
2682 * For those, you should use to_utf8_case() instead */
b0e3252e 2683/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2684
2685/* Note:
2686 * Returns the value of property/mapping C<swash> for the first character
2687 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3d0f8846
KW
2688 * assumed to be in well-formed utf8. If C<do_utf8> is false, the string C<ptr>
2689 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2690 *
2691 * A "swash" is a hash which contains initially the keys/values set up by
2692 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2693 * property for all possible code points. Things are stored in a compact form
2694 * (see utf8_heavy.pl) so that calculation is required to find the actual
2695 * property value for a given code point. As code points are looked up, new
2696 * key/value pairs are added to the hash, so that the calculation doesn't have
2697 * to ever be re-done. Further, each calculation is done, not just for the
2698 * desired one, but for a whole block of code points adjacent to that one.
2699 * For binary properties on ASCII machines, the block is usually for 64 code
2700 * points, starting with a code point evenly divisible by 64. Thus if the
2701 * property value for code point 257 is requested, the code goes out and
2702 * calculates the property values for all 64 code points between 256 and 319,
2703 * and stores these as a single 64-bit long bit vector, called a "swatch",
2704 * under the key for code point 256. The key is the UTF-8 encoding for code
2705 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2706 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2707 * for code point 258 is then requested, this code realizes that it would be
2708 * stored under the key for 256, and would find that value and extract the
2709 * relevant bit, offset from 256.
2710 *
2711 * Non-binary properties are stored in as many bits as necessary to represent
2712 * their values (32 currently, though the code is more general than that), not
2713 * as single bits, but the principal is the same: the value for each key is a
2714 * vector that encompasses the property values for all code points whose UTF-8
2715 * representations are represented by the key. That is, for all code points
2716 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2717 * bytes of that.
680c470c 2718 */
a0ed51b3 2719UV
680c470c 2720Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2721{
27da23d5 2722 dVAR;
ef8f7699 2723 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2724 U32 klen;
2725 U32 off;
9b56a019 2726 STRLEN slen = 0;
7d85a32c 2727 STRLEN needents;
cfd0369c 2728 const U8 *tmps = NULL;
979f2922 2729 SV *swatch;
08fb1ac5 2730 const U8 c = *ptr;
3568d838 2731
7918f24d
NC
2732 PERL_ARGS_ASSERT_SWASH_FETCH;
2733
87367d5f
KW
2734 /* If it really isn't a hash, it isn't really swash; must be an inversion
2735 * list */
2736 if (SvTYPE(hv) != SVt_PVHV) {
2737 return _invlist_contains_cp((SV*)hv,
2738 (do_utf8)
2739 ? valid_utf8_to_uvchr(ptr, NULL)
2740 : c);
2741 }
2742
08fb1ac5
KW
2743 /* We store the values in a "swatch" which is a vec() value in a swash
2744 * hash. Code points 0-255 are a single vec() stored with key length
2745 * (klen) 0. All other code points have a UTF-8 representation
2746 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2747 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2748 * length for them is the length of the encoded char - 1. ptr[klen] is the
2749 * final byte in the sequence representing the character */
2750 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2751 klen = 0;
2752 needents = 256;
2753 off = c;
3568d838 2754 }
08fb1ac5
KW
2755 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2756 klen = 0;
2757 needents = 256;
2758 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2759 }
2760 else {
08fb1ac5
KW
2761 klen = UTF8SKIP(ptr) - 1;
2762
2763 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2764 * the vec is the final byte in the sequence. (In EBCDIC this is
2765 * converted to I8 to get consecutive values.) To help you visualize
2766 * all this:
2767 * Straight 1047 After final byte
2768 * UTF-8 UTF-EBCDIC I8 transform
2769 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2770 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2771 * ...
2772 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2773 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2774 * ...
2775 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2776 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2777 * ...
2778 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2779 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2780 * ...
2781 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2782 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2783 *
2784 * (There are no discontinuities in the elided (...) entries.)
2785 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2786 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2787 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2788 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2789 * index into the vec() swatch (after subtracting 0x80, which we
2790 * actually do with an '&').
2791 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2792 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2793 * dicontinuities which go away by transforming it into I8, and we
2794 * effectively subtract 0xA0 to get the index. */
979f2922 2795 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2796 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2797 }
7d85a32c 2798
a0ed51b3
LW
2799 /*
2800 * This single-entry cache saves about 1/3 of the utf8 overhead in test
2801 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2802 * it's nothing to sniff at.) Pity we usually come through at least
2803 * two function calls to get here...
2804 *
2805 * NB: this code assumes that swatches are never modified, once generated!
2806 */
2807
3568d838 2808 if (hv == PL_last_swash_hv &&
a0ed51b3 2809 klen == PL_last_swash_klen &&
27da23d5 2810 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2811 {
2812 tmps = PL_last_swash_tmps;
2813 slen = PL_last_swash_slen;
2814 }
2815 else {
2816 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2817 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2818
b0e3252e 2819 /* If not cached, generate it via swatch_get */
979f2922 2820 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2821 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2822 {
2823 if (klen) {
2824 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2825 swatch = swatch_get(swash,
2826 code_point & ~((UV)needents - 1),
2827 needents);
2828 }
2829 else { /* For the first 256 code points, the swatch has a key of
2830 length 0 */
2831 swatch = swatch_get(swash, 0, needents);
2832 }
979f2922 2833
923e4eb5 2834 if (IN_PERL_COMPILETIME)
623e6609 2835 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2836
979f2922 2837 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2838
979f2922
ST
2839 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2840 || (slen << 3) < needents)
5637ef5b
NC
2841 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2842 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2843 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2844 }
2845
2846 PL_last_swash_hv = hv;
16d8f38a 2847 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2848 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2849 /* FIXME change interpvar.h? */
2850 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2851 PL_last_swash_slen = slen;
2852 if (klen)
2853 Copy(ptr, PL_last_swash_key, klen, U8);
2854 }
2855
9faf8d75 2856 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2857 case 1:
e7aca353 2858 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2859 case 8:
e7aca353 2860 return ((UV) tmps[off]);
a0ed51b3
LW
2861 case 16:
2862 off <<= 1;
e7aca353
JH
2863 return
2864 ((UV) tmps[off ] << 8) +
2865 ((UV) tmps[off + 1]);
a0ed51b3
LW
2866 case 32:
2867 off <<= 2;
e7aca353
JH
2868 return
2869 ((UV) tmps[off ] << 24) +
2870 ((UV) tmps[off + 1] << 16) +
2871 ((UV) tmps[off + 2] << 8) +
2872 ((UV) tmps[off + 3]);
a0ed51b3 2873 }
5637ef5b
NC
2874 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2875 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2876 NORETURN_FUNCTION_END;
a0ed51b3 2877}
2b9d42f0 2878
319009ee
KW
2879/* Read a single line of the main body of the swash input text. These are of
2880 * the form:
2881 * 0053 0056 0073
2882 * where each number is hex. The first two numbers form the minimum and
2883 * maximum of a range, and the third is the value associated with the range.
2884 * Not all swashes should have a third number
2885 *
2886 * On input: l points to the beginning of the line to be examined; it points
2887 * to somewhere in the string of the whole input text, and is
2888 * terminated by a \n or the null string terminator.
2889 * lend points to the null terminator of that string
2890 * wants_value is non-zero if the swash expects a third number
2891 * typestr is the name of the swash's mapping, like 'ToLower'
2892 * On output: *min, *max, and *val are set to the values read from the line.
2893 * returns a pointer just beyond the line examined. If there was no
2894 * valid min number on the line, returns lend+1
2895 */
2896
2897STATIC U8*
2898S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2899 const bool wants_value, const U8* const typestr)
2900{
2901 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2902 STRLEN numlen; /* Length of the number */
02470786
KW
2903 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2904 | PERL_SCAN_DISALLOW_PREFIX
2905 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2906
2907 /* nl points to the next \n in the scan */
2908 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2909
95543e92
KW
2910 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2911
319009ee
KW
2912 /* Get the first number on the line: the range minimum */
2913 numlen = lend - l;
2914 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2915 *max = *min; /* So can never return without setting max */
319009ee
KW
2916 if (numlen) /* If found a hex number, position past it */
2917 l += numlen;
2918 else if (nl) { /* Else, go handle next line, if any */
2919 return nl + 1; /* 1 is length of "\n" */
2920 }
2921 else { /* Else, no next line */
2922 return lend + 1; /* to LIST's end at which \n is not found */
2923 }
2924
2925 /* The max range value follows, separated by a BLANK */
2926 if (isBLANK(*l)) {
2927 ++l;
02470786
KW
2928 flags = PERL_SCAN_SILENT_ILLDIGIT
2929 | PERL_SCAN_DISALLOW_PREFIX
2930 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2931 numlen = lend - l;
2932 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2933 if (numlen)
2934 l += numlen;
2935 else /* If no value here, it is a single element range */
2936 *max = *min;
2937
2938 /* Non-binary tables have a third entry: what the first element of the
24303724 2939 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2940 if (wants_value) {
2941 if (isBLANK(*l)) {
2942 ++l;
f2a7d0fc
KW
2943 flags = PERL_SCAN_SILENT_ILLDIGIT
2944 | PERL_SCAN_DISALLOW_PREFIX
2945 | PERL_SCAN_SILENT_NON_PORTABLE;
2946 numlen = lend - l;
2947 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2948 if (numlen)
2949 l += numlen;
2950 else
2951 *val = 0;
319009ee
KW
2952 }
2953 else {
2954 *val = 0;
2955 if (typeto) {
dcbac5bb 2956 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2957 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2958 typestr, l);
2959 }
2960 }
2961 }
2962 else
2963 *val = 0; /* bits == 1, then any val should be ignored */
2964 }
2965 else { /* Nothing following range min, should be single element with no
2966 mapping expected */
319009ee
KW
2967 if (wants_value) {
2968 *val = 0;
2969 if (typeto) {
dcbac5bb 2970 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2971 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2972 }
2973 }
2974 else
2975 *val = 0; /* bits == 1, then val should be ignored */
2976 }
2977
2978 /* Position to next line if any, or EOF */
2979 if (nl)
2980 l = nl + 1;
2981 else
2982 l = lend;
2983
2984 return l;
2985}
2986
979f2922
ST
2987/* Note:
2988 * Returns a swatch (a bit vector string) for a code point sequence
2989 * that starts from the value C<start> and comprises the number C<span>.
2990 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2991 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2992 */
2993STATIC SV*
b0e3252e 2994S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2995{
2996 SV *swatch;
77f9f126 2997 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2998 STRLEN lcur, xcur, scur;
ef8f7699 2999 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3000 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3001
88d45d28
KW
3002 SV** listsvp = NULL; /* The string containing the main body of the table */
3003 SV** extssvp = NULL;
3004 SV** invert_it_svp = NULL;
3005 U8* typestr = NULL;
786861f5
KW
3006 STRLEN bits;
3007 STRLEN octets; /* if bits == 1, then octets == 0 */
3008 UV none;
3009 UV end = start + span;
972dd592 3010
36eb48b4 3011 if (invlistsvp == NULL) {
786861f5
KW
3012 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3013 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3014 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3015 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3016 listsvp = hv_fetchs(hv, "LIST", FALSE);
3017 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3018
3019 bits = SvUV(*bitssvp);
3020 none = SvUV(*nonesvp);
3021 typestr = (U8*)SvPV_nolen(*typesvp);
3022 }
36eb48b4
KW
3023 else {
3024 bits = 1;
3025 none = 0;
3026 }
786861f5 3027 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3028
b0e3252e 3029 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3030
979f2922 3031 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3032 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3033 (UV)bits);
979f2922
ST
3034 }
3035
84ea5ef6
KW
3036 /* If overflowed, use the max possible */
3037 if (end < start) {
3038 end = UV_MAX;
3039 span = end - start;
3040 }
3041
979f2922 3042 /* create and initialize $swatch */
979f2922 3043 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3044 swatch = newSV(scur);
3045 SvPOK_on(swatch);
979f2922
ST
3046 s = (U8*)SvPVX(swatch);
3047 if (octets && none) {
0bd48802 3048 const U8* const e = s + scur;
979f2922
ST
3049 while (s < e) {
3050 if (bits == 8)
3051 *s++ = (U8)(none & 0xff);
3052 else if (bits == 16) {
3053 *s++ = (U8)((none >> 8) & 0xff);
3054 *s++ = (U8)( none & 0xff);
3055 }
3056 else if (bits == 32) {
3057 *s++ = (U8)((none >> 24) & 0xff);
3058 *s++ = (U8)((none >> 16) & 0xff);
3059 *s++ = (U8)((none >> 8) & 0xff);
3060 *s++ = (U8)( none & 0xff);
3061 }
3062 }
3063 *s = '\0';
3064 }
3065 else {
3066 (void)memzero((U8*)s, scur + 1);
3067 }
3068 SvCUR_set(swatch, scur);
3069 s = (U8*)SvPVX(swatch);
3070
36eb48b4
KW
3071 if (invlistsvp) { /* If has an inversion list set up use that */
3072 _invlist_populate_swatch(*invlistsvp, start, end, s);
3073 return swatch;
3074 }
3075
3076 /* read $swash->{LIST} */
979f2922
ST
3077 l = (U8*)SvPV(*listsvp, lcur);
3078 lend = l + lcur;
3079 while (l < lend) {
8ed25d53 3080 UV min, max, val, upper;
95543e92
KW
3081 l = swash_scan_list_line(l, lend, &min, &max, &val,
3082 cBOOL(octets), typestr);
319009ee 3083 if (l > lend) {
979f2922
ST
3084 break;
3085 }
3086
972dd592 3087 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3088 if (max < start)
3089 continue;
3090
8ed25d53
KW
3091 /* <end> is generally 1 beyond where we want to set things, but at the
3092 * platform's infinity, where we can't go any higher, we want to
3093 * include the code point at <end> */
3094 upper = (max < end)
3095 ? max
3096 : (max != UV_MAX || end != UV_MAX)
3097 ? end - 1
3098 : end;
3099
979f2922 3100 if (octets) {
35da51f7 3101 UV key;
979f2922
ST
3102 if (min < start) {
3103 if (!none || val < none) {
3104 val += start - min;
3105 }
3106 min = start;
3107 }
8ed25d53 3108 for (key = min; key <= upper; key++) {
979f2922 3109 STRLEN offset;
979f2922
ST
3110 /* offset must be non-negative (start <= min <= key < end) */
3111 offset = octets * (key - start);
3112 if (bits == 8)
3113 s[offset] = (U8)(val & 0xff);
3114 else if (bits == 16) {
3115 s[offset ] = (U8)((val >> 8) & 0xff);
3116 s[offset + 1] = (U8)( val & 0xff);
3117 }
3118 else if (bits == 32) {
3119 s[offset ] = (U8)((val >> 24) & 0xff);
3120 s[offset + 1] = (U8)((val >> 16) & 0xff);
3121 s[offset + 2] = (U8)((val >> 8) & 0xff);
3122 s[offset + 3] = (U8)( val & 0xff);
3123 }
3124
3125 if (!none || val < none)
3126 ++val;
3127 }
3128 }
711a919c 3129 else { /* bits == 1, then val should be ignored */
35da51f7 3130 UV key;
979f2922
ST
3131 if (min < start)
3132 min = start;
6cb05c12 3133
8ed25d53 3134 for (key = min; key <= upper; key++) {
0bd48802 3135 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3136 s[offset >> 3] |= 1 << (offset & 7);
3137 }
3138 }
3139 } /* while */
979f2922 3140
9479a769 3141 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3142 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3143
3144 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3145 * be 0, and their inversion should also be 0, as we don't succeed any
3146 * Unicode property matches for non-Unicode code points */
3147 if (start <= PERL_UNICODE_MAX) {
3148
3149 /* The code below assumes that we never cross the
3150 * Unicode/above-Unicode boundary in a range, as otherwise we would
3151 * have to figure out where to stop flipping the bits. Since this
3152 * boundary is divisible by a large power of 2, and swatches comes
3153 * in small powers of 2, this should be a valid assumption */
3154 assert(start + span - 1 <= PERL_UNICODE_MAX);
3155
507a8485
KW
3156 send = s + scur;
3157 while (s < send) {
3158 *s = ~(*s);
3159 s++;
3160 }
0bda3001 3161 }
77f9f126
KW
3162 }
3163
d73c39c5
KW
3164 /* read $swash->{EXTRAS}
3165 * This code also copied to swash_to_invlist() below */
979f2922
ST
3166 x = (U8*)SvPV(*extssvp, xcur);
3167 xend = x + xcur;
3168 while (x < xend) {
3169 STRLEN namelen;
3170 U8 *namestr;
3171 SV** othersvp;
3172 HV* otherhv;
3173 STRLEN otherbits;
3174 SV **otherbitssvp, *other;
711a919c 3175 U8 *s, *o, *nl;
979f2922
ST
3176 STRLEN slen, olen;
3177
35da51f7 3178 const U8 opc = *x++;
979f2922
ST
3179 if (opc == '\n')
3180 continue;
3181
3182 nl = (U8*)memchr(x, '\n', xend - x);
3183
3184 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3185 if (nl) {
3186 x = nl + 1; /* 1 is length of "\n" */
3187 continue;
3188 }
3189 else {
3190 x = xend; /* to EXTRAS' end at which \n is not found */
3191 break;
3192 }
3193 }
3194
3195 namestr = x;
3196 if (nl) {
3197 namelen = nl - namestr;
3198 x = nl + 1;
3199 }
3200 else {
3201 namelen = xend - namestr;
3202 x = xend;
3203 }
3204
3205 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3206 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3207 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3208 otherbits = (STRLEN)SvUV(*otherbitssvp);
3209 if (bits < otherbits)
5637ef5b
NC
3210 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3211 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3212
3213 /* The "other" swatch must be destroyed after. */
b0e3252e 3214 other = swatch_get(*othersvp, start, span);
979f2922
ST
3215 o = (U8*)SvPV(other, olen);
3216
3217 if (!olen)
b0e3252e 3218 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3219
3220 s = (U8*)SvPV(swatch, slen);
3221 if (bits == 1 && otherbits == 1) {
3222 if (slen != olen)
5637ef5b
NC
3223 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3224 "mismatch, slen=%"UVuf", olen=%"UVuf,
3225 (UV)slen, (UV)olen);
979f2922
ST
3226
3227 switch (opc) {
3228 case '+':
3229 while (slen--)
3230 *s++ |= *o++;
3231 break;
3232 case '!':
3233 while (slen--)
3234 *s++ |= ~*o++;
3235 break;
3236 case '-':
3237 while (slen--)
3238 *s++ &= ~*o++;
3239 break;
3240 case '&':
3241 while (slen--)
3242 *s++ &= *o++;
3243 break;
3244 default:
3245 break;
3246 }
3247 }
711a919c 3248 else {
979f2922
ST
3249 STRLEN otheroctets = otherbits >> 3;
3250 STRLEN offset = 0;
35da51f7 3251 U8* const send = s + slen;
979f2922
ST
3252
3253 while (s < send) {
3254 UV otherval = 0;
3255
3256 if (otherbits == 1) {
3257 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3258 ++offset;
3259 }
3260 else {
3261 STRLEN vlen = otheroctets;
3262 otherval = *o++;
3263 while (--vlen) {
3264 otherval <<= 8;
3265 otherval |= *o++;
3266 }
3267 }
3268
711a919c 3269 if (opc == '+' && otherval)
6f207bd3 3270 NOOP; /* replace with otherval */
979f2922
ST
3271 else if (opc == '!' && !otherval)
3272 otherval = 1;
3273 else if (opc == '-' && otherval)
3274 otherval = 0;
3275 else if (opc == '&' && !otherval)
3276 otherval = 0;
3277 else {
711a919c 3278 s += octets; /* no replacement */
979f2922
ST
3279 continue;
3280 }
3281
3282 if (bits == 8)
3283 *s++ = (U8)( otherval & 0xff);
3284 else if (bits == 16) {
3285 *s++ = (U8)((otherval >> 8) & 0xff);
3286 *s++ = (U8)( otherval & 0xff);
3287 }
3288 else if (bits == 32) {
3289 *s++ = (U8)((otherval >> 24) & 0xff);
3290 *s++ = (U8)((otherval >> 16) & 0xff);
3291 *s++ = (U8)((otherval >> 8) & 0xff);
3292 *s++ = (U8)( otherval & 0xff);
3293 }
3294 }
3295 }
3296 sv_free(other); /* through with it! */
3297 } /* while */
3298 return swatch;
3299}
3300
064c021d 3301HV*
4c2e1131 3302Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3303{
3304
79a2a0e8 3305 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3306 * Can't be used on a property that is subject to user override, as it
3307 * relies on the value of SPECIALS in the swash which would be set by
3308 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3309 * for overridden properties
064c021d
KW
3310 *
3311 * Returns a hash which is the inversion and closure of a swash mapping.
3312 * For example, consider the input lines:
3313 * 004B 006B
3314 * 004C 006C
3315 * 212A 006B
3316 *
3317 * The returned hash would have two keys, the utf8 for 006B and the utf8 for
3318 * 006C. The value for each key is an array. For 006C, the array would
17d5c697 3319 * have two elements, the utf8 for itself, and for 004C. For 006B, there
064c021d
KW
3320 * would be three elements in its array, the utf8 for 006B, 004B and 212A.
3321 *
538e84ed
KW
3322 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3323 * keys are only code points that are folded-to, so it isn't a full closure.
3324 *
064c021d
KW
3325 * Essentially, for any code point, it gives all the code points that map to
3326 * it, or the list of 'froms' for that point.
3327 *
5662e334
KW
3328 * Currently it ignores any additions or deletions from other swashes,
3329 * looking at just the main body of the swash, and if there are SPECIALS
3330 * in the swash, at that hash
3331 *
3332 * The specials hash can be extra code points, and most likely consists of
3333 * maps from single code points to multiple ones (each expressed as a string
3334 * of utf8 characters). This function currently returns only 1-1 mappings.
3335 * However consider this possible input in the specials hash:
3336 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3337 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3338 *
3339 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3340 * currently handle. But it also means that FB05 and FB06 are equivalent in
3341 * a 1-1 mapping which we should handle, and this relationship may not be in
3342 * the main table. Therefore this function examines all the multi-char
3343 * sequences and adds the 1-1 mappings that come out of that. */
064c021d
KW
3344
3345 U8 *l, *lend;
3346 STRLEN lcur;
3347 HV *const hv = MUTABLE_HV(SvRV(swash));
3348
923b6d4e
KW
3349 /* The string containing the main body of the table. This will have its
3350 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3351 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3352
3353 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3354 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3355 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3356 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3357 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3358 const STRLEN bits = SvUV(*bitssvp);
3359 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3360 const UV none = SvUV(*nonesvp);
5662e334 3361 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3362
3363 HV* ret = newHV();
3364
3365 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3366
3367 /* Must have at least 8 bits to get the mappings */
3368 if (bits != 8 && bits != 16 && bits != 32) {
3369 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3370 (UV)bits);
3371 }
3372
5662e334
KW
3373 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3374 mapping to more than one character */
3375
3376 /* Construct an inverse mapping hash for the specials */
3377 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3378 HV * specials_inverse = newHV();
3379 char *char_from; /* the lhs of the map */
3380 I32 from_len; /* its byte length */
3381 char *char_to; /* the rhs of the map */
3382 I32 to_len; /* its byte length */
3383 SV *sv_to; /* and in a sv */
3384 AV* from_list; /* list of things that map to each 'to' */
3385
3386 hv_iterinit(specials_hv);
3387
3388 /* The keys are the characters (in utf8) that map to the corresponding
3389 * utf8 string value. Iterate through the list creating the inverse
3390 * list. */
3391 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3392 SV** listp;
3393 if (! SvPOK(sv_to)) {
5637ef5b
NC
3394 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3395 "unexpectedly is not a string, flags=%lu",
3396 (unsigned long)SvFLAGS(sv_to));
5662e334 3397 }
4b88fb76 3398 /*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
3399
3400 /* Each key in the inverse list is a mapped-to value, and the key's
3401 * hash value is a list of the strings (each in utf8) that map to
3402 * it. Those strings are all one character long */
3403 if ((listp = hv_fetch(specials_inverse,
3404 SvPVX(sv_to),
3405 SvCUR(sv_to), 0)))
3406 {
3407 from_list = (AV*) *listp;
3408 }
3409 else { /* No entry yet for it: create one */
3410 from_list = newAV();
3411 if (! hv_store(specials_inverse,
3412 SvPVX(sv_to),
3413 SvCUR(sv_to),
3414 (SV*) from_list, 0))
3415 {
3416 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3417 }
3418 }
3419
3420 /* Here have the list associated with this 'to' (perhaps newly
3421 * created and empty). Just add to it. Note that we ASSUME that
3422 * the input is guaranteed to not have duplications, so we don't
3423 * check for that. Duplications just slow down execution time. */
3424 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3425 }
3426
3427 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3428 * it looking for cases like the FB05/FB06 examples above. There would
3429 * be an entry in the hash like
3430 * 'st' => [ FB05, FB06 ]
3431 * In this example we will create two lists that get stored in the
3432 * returned hash, 'ret':
3433 * FB05 => [ FB05, FB06 ]
3434 * FB06 => [ FB05, FB06 ]
3435 *
3436 * Note that there is nothing to do if the array only has one element.
3437 * (In the normal 1-1 case handled below, we don't have to worry about
3438 * two lists, as everything gets tied to the single list that is
3439 * generated for the single character 'to'. But here, we are omitting
3440 * that list, ('st' in the example), so must have multiple lists.) */
3441 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3442 &char_to, &to_len)))
3443 {
b9f2b683 3444 if (av_tindex(from_list) > 0) {
c70927a6 3445 SSize_t i;
5662e334
KW
3446
3447 /* We iterate over all combinations of i,j to place each code
3448 * point on each list */
b9f2b683 3449 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3450 SSize_t j;
5662e334
KW
3451 AV* i_list = newAV();
3452 SV** entryp = av_fetch(from_list, i, FALSE);
3453 if (entryp == NULL) {
3454 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3455 }
3456 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3457 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3458 }
3459 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3460 (SV*) i_list, FALSE))
3461 {
3462 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3463 }
3464
538e84ed 3465 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3466 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3467 entryp = av_fetch(from_list, j, FALSE);
3468 if (entryp == NULL) {
3469 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3470 }
3471
3472 /* When i==j this adds itself to the list */
4b88fb76
KW
3473 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3474 (U8*) SvPVX(*entryp),
3475 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3476 0)));
4637d003 3477 /*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
3478 }
3479 }
3480 }
3481 }
3482 SvREFCNT_dec(specials_inverse); /* done with it */
3483 } /* End of specials */
3484
064c021d
KW
3485 /* read $swash->{LIST} */
3486 l = (U8*)SvPV(*listsvp, lcur);
3487 lend = l + lcur;
3488
3489 /* Go through each input line */
3490 while (l < lend) {
3491 UV min, max, val;
3492 UV inverse;
95543e92
KW
3493 l = swash_scan_list_line(l, lend, &min, &max, &val,
3494 cBOOL(octets), typestr);
064c021d
KW
3495 if (l > lend) {
3496 break;
3497 }
3498
3499 /* Each element in the range is to be inverted */
3500 for (inverse = min; inverse <= max; inverse++) {
3501 AV* list;
064c021d
KW
3502 SV** listp;
3503 IV i;
3504 bool found_key = FALSE;
5662e334 3505 bool found_inverse = FALSE;
064c021d
KW
3506
3507 /* The key is the inverse mapping */
3508 char key[UTF8_MAXBYTES+1];
c80e42f3 3509 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3510 STRLEN key_len = key_end - key;
3511
064c021d
KW
3512 /* Get the list for the map */
3513 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3514 list = (AV*) *listp;
3515 }
3516 else { /* No entry yet for it: create one */
3517 list = newAV();
3518 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3519 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3520 }
3521 }
3522
5662e334
KW
3523 /* Look through list to see if this inverse mapping already is
3524 * listed, or if there is a mapping to itself already */
b9f2b683 3525 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3526 SV** entryp = av_fetch(list, i, FALSE);
3527 SV* entry;
3528 if (entryp == NULL) {
3529 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3530 }
3531 entry = *entryp;
5662e334 3532 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, SvUV(entry)));*/
56ca34ca 3533 if (SvUV(entry) == val) {
064c021d 3534 found_key = TRUE;
5662e334
KW
3535 }
3536 if (SvUV(entry) == inverse) {
3537 found_inverse = TRUE;
3538 }
3539
3540 /* No need to continue searching if found everything we are
3541 * looking for */
3542 if (found_key && found_inverse) {
064c021d
KW
3543 break;
3544 }
3545 }
56ca34ca
KW
3546
3547 /* Make sure there is a mapping to itself on the list */
064c021d 3548 if (! found_key) {
d397ff6a 3549 av_push(list, newSVuv(val));
4637d003 3550 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3551 }
3552
3553
3554 /* Simply add the value to the list */
5662e334
KW
3555 if (! found_inverse) {
3556 av_push(list, newSVuv(inverse));
4637d003 3557 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3558 }
064c021d 3559
b0e3252e 3560 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3561 * range. That makes more compact tables possible. You can
3562 * express the capitalization, for example, of all consecutive
3563 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3564 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3565 * and it's not documented; it appears to be used only in
b0e3252e 3566 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3567 * in case */
064c021d
KW
3568 if (!none || val < none) {
3569 ++val;
3570 }
3571 }
3572 }
3573
3574 return ret;
3575}
3576
a25abddc 3577SV*
d764b54e
KW
3578Perl__swash_to_invlist(pTHX_ SV* const swash)
3579{
3580
ed92f1b3
KW
3581 /* Subject to change or removal. For use only in one place in regcomp.c.
3582 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3583
3584 U8 *l, *lend;
3585 char *loc;
3586 STRLEN lcur;
3587 HV *const hv = MUTABLE_HV(SvRV(swash));
3588 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3589 U8 empty[] = "";
923b6d4e
KW
3590 SV** listsvp;
3591 SV** typesvp;
3592 SV** bitssvp;
3593 SV** extssvp;
3594 SV** invert_it_svp;
d764b54e 3595
923b6d4e
KW
3596 U8* typestr;
3597 STRLEN bits;
3598 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3599 U8 *x, *xend;
3600 STRLEN xcur;
d764b54e 3601
a25abddc 3602 SV* invlist;
d764b54e 3603
b81740c0
KW
3604 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3605
923b6d4e
KW
3606 /* If not a hash, it must be the swash's inversion list instead */
3607 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3608 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3609 }
3610
3611 /* The string containing the main body of the table */
3612 listsvp = hv_fetchs(hv, "LIST", FALSE);
3613 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3614 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3615 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3616 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3617
3618 typestr = (U8*)SvPV_nolen(*typesvp);
3619 bits = SvUV(*bitssvp);
3620 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3621
d764b54e 3622 /* read $swash->{LIST} */
b443038a
KW
3623 if (SvPOK(*listsvp)) {
3624 l = (U8*)SvPV(*listsvp, lcur);
3625 }
3626 else {
3627 /* LIST legitimately doesn't contain a string during compilation phases
3628 * of Perl itself, before the Unicode tables are generated. In this
3629 * case, just fake things up by creating an empty list */
3630 l = empty;
3631 lcur = 0;
3632 }
d764b54e
KW
3633 loc = (char *) l;
3634 lend = l + lcur;
3635
31aa6e0b
KW
3636 if (*l == 'V') { /* Inversion list format */
3637 char *after_strtol = (char *) lend;
3638 UV element0;
3639 UV* other_elements_ptr;
3640
3641 /* The first number is a count of the rest */
3642 l++;
3643 elements = Strtoul((char *)l, &after_strtol, 10);
eb092534
KW
3644 if (elements == 0) {
3645 invlist = _new_invlist(0);
3646 }
3647 else {
31aa6e0b 3648 l = (U8 *) after_strtol;
1f9f7d4c
KW
3649
3650 /* Get the 0th element, which is needed to setup the inversion list */
3651 element0 = (UV) Strtoul((char *)l, &after_strtol, 10);
3652 l = (U8 *) after_strtol;
3653 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
3654 elements--;
3655
3656 /* Then just populate the rest of the input */
3657 while (elements-- > 0) {
3658 if (l > lend) {
3659 Perl_croak(aTHX_ "panic: Expecting %"UVuf" more elements than available", elements);
3660 }
3661 *other_elements_ptr++ = (UV) Strtoul((char *)l, &after_strtol, 10);
3662 l = (U8 *) after_strtol;
3663 }
eb092534 3664 }
31aa6e0b
KW
3665 }
3666 else {
3667
1784d2f9
KW
3668 /* Scan the input to count the number of lines to preallocate array
3669 * size based on worst possible case, which is each line in the input
3670 * creates 2 elements in the inversion list: 1) the beginning of a
3671 * range in the list; 2) the beginning of a range not in the list. */
3672 while ((loc = (strchr(loc, '\n'))) != NULL) {
3673 elements += 2;
3674 loc++;
3675 }
d764b54e 3676
1784d2f9
KW
3677 /* If the ending is somehow corrupt and isn't a new line, add another
3678 * element for the final range that isn't in the inversion list */
3679 if (! (*lend == '\n'
3680 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
3681 {
3682 elements++;
3683 }
d764b54e 3684
1784d2f9 36