This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.h: Add assertions to 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"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
901b21bf 38
48ef279e 39/*
ccfc67b7 40=head1 Unicode Support
7fefc6c1 41These are various utility functions for manipulating UTF8-encoded
72d33970 42strings. For the uninitiated, this is a method of representing arbitrary
61296642 43Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
44characters in the ASCII range are unmodified, and a zero byte never appears
45within non-zero characters.
166f8a29 46
eaf7a4d2
CS
47=cut
48*/
49
50/*
7bbfa158 51=for apidoc is_invariant_string
eaf7a4d2 52
b8ed1ca9
KW
53Returns true iff the first C<len> bytes of the string C<s> are the same
54regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
55EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
56machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 57definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
58so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
59EBCDIC).
eaf7a4d2 60
75200dff
KW
61If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
62use this option, that C<s> can't have embedded C<NUL> characters and has to
63have a terminating C<NUL> byte).
9f7e3d64 64
a1433954 65See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
66
67=cut
68*/
69
70bool
7bbfa158 71Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
72{
73 const U8* const send = s + (len ? len : strlen((const char *)s));
74 const U8* x = s;
75
7bbfa158 76 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
77
78 for (; x < send; ++x) {
79 if (!UTF8_IS_INVARIANT(*x))
80 break;
81 }
82
83 return x == send;
84}
85
86/*
378516de 87=for apidoc uvoffuni_to_utf8_flags
eebe1485 88
a27992cc 89THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
90Instead, B<Almost all code should use L</uvchr_to_utf8> or
91L</uvchr_to_utf8_flags>>.
a27992cc 92
de69f3af
KW
93This function is like them, but the input is a strict Unicode
94(as opposed to native) code point. Only in very rare circumstances should code
95not be using the native code point.
949cf498 96
efa9cd84 97For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 98
eebe1485
SC
99=cut
100*/
101
dfe13c55 102U8 *
378516de 103Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 104{
378516de 105 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 106
d9432125
KW
107 if (UNI_IS_INVARIANT(uv)) {
108 *d++ = (U8) LATIN1_TO_NATIVE(uv);
109 return d;
110 }
111
9415f659
KW
112#ifdef EBCDIC
113 /* Not representable in UTF-EBCDIC */
114 flags |= UNICODE_DISALLOW_FE_FF;
115#endif
116
979f77b6
KW
117 /* The first problematic code point is the first surrogate */
118 if (uv >= UNICODE_SURROGATE_FIRST
0cfa64bf 119 && ckWARN3_d(WARN_SURROGATE, WARN_NON_UNICODE, WARN_NONCHAR))
979f77b6 120 {
949cf498
KW
121 if (UNICODE_IS_SURROGATE(uv)) {
122 if (flags & UNICODE_WARN_SURROGATE) {
8457b38f 123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE),
949cf498
KW
124 "UTF-16 surrogate U+%04"UVXf, uv);
125 }
126 if (flags & UNICODE_DISALLOW_SURROGATE) {
127 return NULL;
128 }
129 }
130 else if (UNICODE_IS_SUPER(uv)) {
131 if (flags & UNICODE_WARN_SUPER
132 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_WARN_FE_FF)))
133 {
8457b38f 134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
949cf498
KW
135 "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv);
136 }
137 if (flags & UNICODE_DISALLOW_SUPER
138 || (UNICODE_IS_FE_FF(uv) && (flags & UNICODE_DISALLOW_FE_FF)))
139 {
9415f659
KW
140#ifdef EBCDIC
141 Perl_die(aTHX_ "Can't represent character for Ox%"UVXf" on this platform", uv);
661d43c4 142 NOT_REACHED; /* NOTREACHED */
9415f659 143#endif
949cf498
KW
144 return NULL;
145 }
146 }
147 else if (UNICODE_IS_NONCHAR(uv)) {
148 if (flags & UNICODE_WARN_NONCHAR) {
8457b38f 149 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR),
ba707cdc 150 "Unicode non-character U+%04"UVXf" is not recommended for open interchange",
949cf498
KW
151 uv);
152 }
153 if (flags & UNICODE_DISALLOW_NONCHAR) {
154 return NULL;
155 }
156 }
507b9800 157 }
d9432125 158
2d331972 159#if defined(EBCDIC)
d9432125 160 {
5aaebcb3 161 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
162 U8 *p = d+len-1;
163 while (p > d) {
bc3632a8 164 *p-- = (U8) I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
165 uv >>= UTF_ACCUMULATION_SHIFT;
166 }
bc3632a8 167 *p = (U8) I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
168 return d+len;
169 }
170#else /* Non loop style */
a0ed51b3 171 if (uv < 0x800) {
eb160463
GS
172 *d++ = (U8)(( uv >> 6) | 0xc0);
173 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
174 return d;
175 }
176 if (uv < 0x10000) {
eb160463
GS
177 *d++ = (U8)(( uv >> 12) | 0xe0);
178 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
179 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
180 return d;
181 }
182 if (uv < 0x200000) {
eb160463
GS
183 *d++ = (U8)(( uv >> 18) | 0xf0);
184 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
185 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
186 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
187 return d;
188 }
189 if (uv < 0x4000000) {
eb160463
GS
190 *d++ = (U8)(( uv >> 24) | 0xf8);
191 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
192 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
193 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
194 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
195 return d;
196 }
197 if (uv < 0x80000000) {
eb160463
GS
198 *d++ = (U8)(( uv >> 30) | 0xfc);
199 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
200 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
201 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
202 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
203 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
204 return d;
205 }
6588300d 206#ifdef UTF8_QUAD_MAX
d7578b48 207 if (uv < UTF8_QUAD_MAX)
a0ed51b3
LW
208#endif
209 {
eb160463
GS
210 *d++ = 0xfe; /* Can't match U+FEFF! */
211 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
212 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
213 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
214 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
215 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
216 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
217 return d;
218 }
6588300d 219#ifdef UTF8_QUAD_MAX
a0ed51b3 220 {
eb160463
GS
221 *d++ = 0xff; /* Can't match U+FFFE! */
222 *d++ = 0x80; /* 6 Reserved bits */
223 *d++ = (U8)(((uv >> 60) & 0x0f) | 0x80); /* 2 Reserved bits */
224 *d++ = (U8)(((uv >> 54) & 0x3f) | 0x80);
225 *d++ = (U8)(((uv >> 48) & 0x3f) | 0x80);
226 *d++ = (U8)(((uv >> 42) & 0x3f) | 0x80);
227 *d++ = (U8)(((uv >> 36) & 0x3f) | 0x80);
228 *d++ = (U8)(((uv >> 30) & 0x3f) | 0x80);
229 *d++ = (U8)(((uv >> 24) & 0x3f) | 0x80);
230 *d++ = (U8)(((uv >> 18) & 0x3f) | 0x80);
231 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
232 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
233 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
234 return d;
235 }
236#endif
537124e4 237#endif /* Non loop style */
a0ed51b3 238}
646ca15d 239/*
07693fe6
KW
240=for apidoc uvchr_to_utf8
241
bcb1a2d4 242Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 243of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
244C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
245the byte after the end of the new character. In other words,
07693fe6
KW
246
247 d = uvchr_to_utf8(d, uv);
248
249is the recommended wide native character-aware way of saying
250
251 *(d++) = uv;
252
de69f3af
KW
253This function accepts any UV as input. To forbid or warn on non-Unicode code
254points, or those that may be problematic, see L</uvchr_to_utf8_flags>.
255
07693fe6
KW
256=cut
257*/
258
de69f3af
KW
259/* This is also a macro */
260PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
261
07693fe6
KW
262U8 *
263Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
264{
de69f3af 265 return uvchr_to_utf8(d, uv);
07693fe6
KW
266}
267
de69f3af
KW
268/*
269=for apidoc uvchr_to_utf8_flags
270
271Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 272of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
273C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
274the byte after the end of the new character. In other words,
de69f3af
KW
275
276 d = uvchr_to_utf8_flags(d, uv, flags);
277
278or, in most cases,
279
280 d = uvchr_to_utf8_flags(d, uv, 0);
281
282This is the Unicode-aware way of saying
283
284 *(d++) = uv;
285
286This function will convert to UTF-8 (and not warn) even code points that aren't
287legal Unicode or are problematic, unless C<flags> contains one or more of the
288following flags:
289
796b6530 290If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
de69f3af 291the function will raise a warning, provided UTF8 warnings are enabled. If instead
796b6530 292C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return NULL.
de69f3af
KW
293If both flags are set, the function will both warn and return NULL.
294
796b6530 295The C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
de69f3af 296affect how the function handles a Unicode non-character. And likewise, the
796b6530 297C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags affect the handling of
de69f3af
KW
298code points that are
299above the Unicode maximum of 0x10FFFF. Code points above 0x7FFF_FFFF (which are
300even less portable) can be warned and/or disallowed even if other above-Unicode
796b6530
KW
301code points are accepted, by the C<UNICODE_WARN_FE_FF> and
302C<UNICODE_DISALLOW_FE_FF> flags.
de69f3af 303
796b6530
KW
304And finally, the flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all four of
305the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
306four DISALLOW flags.
de69f3af
KW
307
308=cut
309*/
310
311/* This is also a macro */
312PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
313
07693fe6
KW
314U8 *
315Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
316{
de69f3af 317 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
318}
319
320/*
87cea99e 321=for apidoc is_utf8_string
6662521e 322
a1433954 323Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 324UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
325using C<strlen(s)> (which means if you use this option, that C<s> can't have
326embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
327that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 328
7bbfa158 329See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 330
6662521e
GS
331=cut
332*/
333
8e84507e 334bool
668b6d8d 335Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 336{
35da51f7 337 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 338 const U8* x = s;
067a85ef 339
7918f24d 340 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 341
6662521e 342 while (x < send) {
6302f837
KW
343 STRLEN len = isUTF8_CHAR(x, send);
344 if (UNLIKELY(! len)) {
345 return FALSE;
346 }
347 x += len;
6662521e 348 }
768c67ee 349
067a85ef 350 return TRUE;
6662521e
GS
351}
352
67e989fb 353/*
814fafa7
NC
354Implemented as a macro in utf8.h
355
87cea99e 356=for apidoc is_utf8_string_loc
814fafa7 357
a1433954
KW
358Like L</is_utf8_string> but stores the location of the failure (in the
359case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
360"utf8ness success") in the C<ep>.
361
a1433954 362See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 363
87cea99e 364=for apidoc is_utf8_string_loclen
81cd54e3 365
a1433954
KW
366Like L</is_utf8_string>() but stores the location of the failure (in the
367case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
368"utf8ness success") in the C<ep>, and the number of UTF-8
369encoded characters in the C<el>.
370
a1433954 371See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
372
373=cut
374*/
375
376bool
668b6d8d 377Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 378{
35da51f7 379 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 380 const U8* x = s;
3ebfea28 381 STRLEN outlen = 0;
7918f24d
NC
382
383 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 384
81cd54e3 385 while (x < send) {
6302f837
KW
386 STRLEN len = isUTF8_CHAR(x, send);
387 if (UNLIKELY(! len)) {
388 goto out;
389 }
390 x += len;
391 outlen++;
81cd54e3 392 }
768c67ee
JH
393
394 out:
3ebfea28
AL
395 if (el)
396 *el = outlen;
397
768c67ee
JH
398 if (ep)
399 *ep = x;
3ebfea28 400 return (x == send);
81cd54e3
JH
401}
402
403/*
768c67ee 404
de69f3af 405=for apidoc utf8n_to_uvchr
378516de
KW
406
407THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 408Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 409
9041c2e3 410Bottom level UTF-8 decode routine.
de69f3af 411Returns the native code point value of the first character in the string C<s>,
746afd53
KW
412which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
413C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
414the length, in bytes, of that character.
949cf498
KW
415
416The value of C<flags> determines the behavior when C<s> does not point to a
417well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
418zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
419next possible position in C<s> that could begin a non-malformed character.
420Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
421
422Various ALLOW flags can be set in C<flags> to allow (and not warn on)
423individual types of malformations, such as the sequence being overlong (that
424is, when there is a shorter sequence that can express the same code point;
425overlong sequences are expressly forbidden in the UTF-8 standard due to
426potential security issues). Another malformation example is the first byte of
427a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
428flags. For allowed 0 length strings, this function returns 0; for allowed
429overlong sequences, the computed code point is returned; for all other allowed
430malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
431determinable reasonable value.
949cf498 432
796b6530 433The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
434flags) malformation is found. If this flag is set, the routine assumes that
435the caller will raise a warning, and this function will silently just set
d088425d
KW
436C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
437
75200dff 438Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 439character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 440in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
441first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
442input had an error.
949cf498
KW
443
444Certain code points are considered problematic. These are Unicode surrogates,
746afd53 445Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 446By default these are considered regular code points, but certain situations
5eafe189 447warrant special handling for them. If C<flags> contains
796b6530
KW
448C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
449malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
450C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
451Unicode maximum) can be set to disallow these categories individually.
452
453The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
454C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
455raised for their respective categories, but otherwise the code points are
456considered valid (not malformations). To get a category to both be treated as
457a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 458(But note that warnings are not raised if lexically disabled nor if
796b6530 459C<UTF8_CHECK_ONLY> is also specified.)
949cf498
KW
460
461Very large code points (above 0x7FFF_FFFF) are considered more problematic than
462the others that are above the Unicode legal maximum. There are several
eb83ed87
KW
463reasons: they requre at least 32 bits to represent them on ASCII platforms, are
464not representable at all on EBCDIC platforms, and the original UTF-8
465specification never went above this number (the current 0x10FFFF limit was
466imposed later). (The smaller ones, those that fit into 32 bits, are
467representable by a UV on ASCII platforms, but not by an IV, which means that
468the number of operations that can be performed on them is quite restricted.)
469The UTF-8 encoding on ASCII platforms for these large code points begins with a
796b6530 470byte containing 0xFE or 0xFF. The C<UTF8_DISALLOW_FE_FF> flag will cause them to
eb83ed87 471be treated as malformations, while allowing smaller above-Unicode code points.
796b6530 472(Of course C<UTF8_DISALLOW_SUPER> will treat all above-Unicode code points,
72d33970 473including these, as malformations.)
796b6530 474Similarly, C<UTF8_WARN_FE_FF> acts just like
eb83ed87 475the other WARN flags, but applies just to these code points.
949cf498
KW
476
477All other code points corresponding to Unicode characters, including private
478use and those yet to be assigned, are never considered malformed and never
479warn.
67e989fb 480
37607a96
PK
481=cut
482*/
67e989fb 483
a0ed51b3 484UV
de69f3af 485Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 486{
d4c19fe8 487 const U8 * const s0 = s;
eb83ed87 488 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 489 U8 * send;
eb83ed87
KW
490 UV uv = *s;
491 STRLEN expectlen;
949cf498 492 SV* sv = NULL;
eb83ed87
KW
493 UV outlier_ret = 0; /* return value when input is in error or problematic
494 */
495 UV pack_warn = 0; /* Save result of packWARN() for later */
496 bool unexpected_non_continuation = FALSE;
497 bool overflowed = FALSE;
2f8f112e 498 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 499
eb83ed87 500 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 501
de69f3af 502 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 503
eb83ed87
KW
504 /* The order of malformation tests here is important. We should consume as
505 * few bytes as possible in order to not skip any valid character. This is
506 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
507 * http://unicode.org/reports/tr36 for more discussion as to why. For
508 * example, once we've done a UTF8SKIP, we can tell the expected number of
509 * bytes, and could fail right off the bat if the input parameters indicate
510 * that there are too few available. But it could be that just that first
511 * byte is garbled, and the intended character occupies fewer bytes. If we
512 * blindly assumed that the first byte is correct, and skipped based on
513 * that number, we could skip over a valid input character. So instead, we
514 * always examine the sequence byte-by-byte.
515 *
516 * We also should not consume too few bytes, otherwise someone could inject
517 * things. For example, an input could be deliberately designed to
518 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 519 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
520 * which is actually part of of the overflowing sequence), that could look
521 * legitimate to the caller, which could discard the initial partial
522 * sequence and process the rest, inappropriately */
523
524 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 525 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
526 if (retlen) {
527 *retlen = 0;
528 }
a0dbb045 529
eb83ed87
KW
530 if (flags & UTF8_ALLOW_EMPTY) {
531 return 0;
532 }
533 if (! (flags & UTF8_CHECK_ONLY)) {
534 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
535 }
0c443dc2
JH
536 goto malformed;
537 }
538
eb83ed87
KW
539 expectlen = UTF8SKIP(s);
540
541 /* A well-formed UTF-8 character, as the vast majority of calls to this
542 * function will be for, has this expected length. For efficiency, set
543 * things up here to return it. It will be overriden only in those rare
544 * cases where a malformation is found */
545 if (retlen) {
546 *retlen = expectlen;
547 }
548
549 /* An invariant is trivially well-formed */
1d72bdf6 550 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 551 return uv;
a0ed51b3 552 }
67e989fb 553
eb83ed87 554 /* A continuation character can't start a valid sequence */
b5b9af04 555 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
556 if (flags & UTF8_ALLOW_CONTINUATION) {
557 if (retlen) {
558 *retlen = 1;
559 }
560 return UNICODE_REPLACEMENT;
561 }
ba210ebe 562
eb83ed87
KW
563 if (! (flags & UTF8_CHECK_ONLY)) {
564 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
565 }
566 curlen = 1;
ba210ebe
JH
567 goto malformed;
568 }
9041c2e3 569
dcd27b3c
KW
570 /* Here is not a continuation byte, nor an invariant. The only thing left
571 * is a start byte (possibly for an overlong) */
572
1d72bdf6 573#ifdef EBCDIC
bc3632a8 574 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
575#endif
576
eb83ed87
KW
577 /* Remove the leading bits that indicate the number of bytes in the
578 * character's whole UTF-8 sequence, leaving just the bits that are part of
579 * the value */
580 uv &= UTF_START_MASK(expectlen);
ba210ebe 581
eb83ed87
KW
582 /* Now, loop through the remaining bytes in the character's sequence,
583 * accumulating each into the working value as we go. Be sure to not look
584 * past the end of the input string */
0b8d30e8
KW
585 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
586
eb83ed87 587 for (s = s0 + 1; s < send; s++) {
b5b9af04 588 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
589#ifndef EBCDIC /* Can't overflow in EBCDIC */
590 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
591
592 /* The original implementors viewed this malformation as more
593 * serious than the others (though I, khw, don't understand
594 * why, since other malformations also give very very wrong
595 * results), so there is no way to turn off checking for it.
596 * Set a flag, but keep going in the loop, so that we absorb
597 * the rest of the bytes that comprise the character. */
598 overflowed = TRUE;
599 overflow_byte = *s; /* Save for warning message's use */
600 }
601#endif
8850bf83 602 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
603 }
604 else {
605 /* Here, found a non-continuation before processing all expected
606 * bytes. This byte begins a new character, so quit, even if
607 * allowing this malformation. */
608 unexpected_non_continuation = TRUE;
609 break;
610 }
611 } /* End of loop through the character's bytes */
612
613 /* Save how many bytes were actually in the character */
614 curlen = s - s0;
615
616 /* The loop above finds two types of malformations: non-continuation and/or
617 * overflow. The non-continuation malformation is really a too-short
618 * malformation, as it means that the current character ended before it was
619 * expected to (being terminated prematurely by the beginning of the next
620 * character, whereas in the too-short malformation there just are too few
621 * bytes available to hold the character. In both cases, the check below
622 * that we have found the expected number of bytes would fail if executed.)
623 * Thus the non-continuation malformation is really unnecessary, being a
624 * subset of the too-short malformation. But there may be existing
625 * applications that are expecting the non-continuation type, so we retain
626 * it, and return it in preference to the too-short malformation. (If this
627 * code were being written from scratch, the two types might be collapsed
628 * into one.) I, khw, am also giving priority to returning the
629 * non-continuation and too-short malformations over overflow when multiple
630 * ones are present. I don't know of any real reason to prefer one over
631 * the other, except that it seems to me that multiple-byte errors trumps
632 * errors from a single byte */
b5b9af04 633 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
634 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
635 if (! (flags & UTF8_CHECK_ONLY)) {
636 if (curlen == 1) {
637 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
638 }
639 else {
640 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
641 }
642 }
eb83ed87
KW
643 goto malformed;
644 }
645 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
646
647 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
648 * as what the original expectations were. */
649 do_overlong_test = FALSE;
eb83ed87
KW
650 if (retlen) {
651 *retlen = curlen;
652 }
653 }
b5b9af04 654 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
655 if (! (flags & UTF8_ALLOW_SHORT)) {
656 if (! (flags & UTF8_CHECK_ONLY)) {
657 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 658 }
eb83ed87
KW
659 goto malformed;
660 }
661 uv = UNICODE_REPLACEMENT;
2f8f112e 662 do_overlong_test = FALSE;
eb83ed87
KW
663 if (retlen) {
664 *retlen = curlen;
665 }
666 }
667
ea5ced44 668#ifndef EBCDIC /* EBCDIC can't overflow */
b5b9af04 669 if (UNLIKELY(overflowed)) {
eb83ed87 670 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 671 goto malformed;
eb83ed87
KW
672 }
673#endif
674
2f8f112e 675 if (do_overlong_test
5aaebcb3 676 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
677 && ! (flags & UTF8_ALLOW_LONG))
678 {
eb83ed87
KW
679 /* The overlong malformation has lower precedence than the others.
680 * Note that if this malformation is allowed, we return the actual
681 * value, instead of the replacement character. This is because this
682 * value is actually well-defined. */
683 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 684 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
685 }
686 goto malformed;
687 }
688
1a89bb6c 689 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
690 * problematic code point that is not allowed by the input parameters. */
691 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
692 && (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
693 |UTF8_WARN_ILLEGAL_INTERCHANGE)))
694 {
949cf498 695 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
696
697 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
698 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 699 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 700 && ckWARN_d(WARN_SURROGATE))
eb83ed87 701 {
111d382d 702 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 703 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
704 }
705 if (flags & UTF8_DISALLOW_SURROGATE) {
706 goto disallowed;
707 }
708 }
949cf498 709 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 710 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 711 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 712 {
111d382d 713 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%04"UVXf" is not Unicode, may not be portable", uv));
54f4afef 714 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 715 }
ea5ced44
KW
716#ifndef EBCDIC /* EBCDIC always allows FE, FF */
717
718 /* The first byte being 0xFE or 0xFF is a subset of the SUPER code
719 * points. We test for these after the regular SUPER ones, and
720 * before possibly bailing out, so that the more dire warning
721 * overrides the regular one, if applicable */
722 if ((*s0 & 0xFE) == 0xFE /* matches both FE, FF */
723 && (flags & (UTF8_WARN_FE_FF|UTF8_DISALLOW_FE_FF)))
724 {
725 if ((flags & (UTF8_WARN_FE_FF|UTF8_CHECK_ONLY))
726 == UTF8_WARN_FE_FF
727 && ckWARN_d(WARN_UTF8))
728 {
729 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Code point 0x%"UVXf" is not Unicode, and not portable", uv));
730 pack_warn = packWARN(WARN_UTF8);
731 }
732 if (flags & UTF8_DISALLOW_FE_FF) {
733 goto disallowed;
734 }
735 }
736#endif
949cf498
KW
737 if (flags & UTF8_DISALLOW_SUPER) {
738 goto disallowed;
739 }
740 }
4190d317
KW
741 else if (UNICODE_IS_NONCHAR(uv)) {
742 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 743 && ckWARN_d(WARN_NONCHAR))
4190d317 744 {
ba707cdc 745 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 746 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
747 }
748 if (flags & UTF8_DISALLOW_NONCHAR) {
749 goto disallowed;
750 }
751 }
949cf498 752
eb83ed87 753 if (sv) {
de69f3af
KW
754 outlier_ret = uv; /* Note we don't bother to convert to native,
755 as all the outlier code points are the same
756 in both ASCII and EBCDIC */
eb83ed87
KW
757 goto do_warn;
758 }
759
949cf498
KW
760 /* Here, this is not considered a malformed character, so drop through
761 * to return it */
a0ed51b3 762 }
ba210ebe 763
de69f3af 764 return UNI_TO_NATIVE(uv);
ba210ebe 765
eb83ed87
KW
766 /* There are three cases which get to beyond this point. In all 3 cases:
767 * <sv> if not null points to a string to print as a warning.
768 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
769 * set.
770 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
771 * This is done by initializing it to 0, and changing it only
772 * for case 1).
773 * The 3 cases are:
774 * 1) The input is valid but problematic, and to be warned about. The
775 * return value is the resultant code point; <*retlen> is set to
776 * <curlen>, the number of bytes that comprise the code point.
777 * <pack_warn> contains the result of packWARN() for the warning
778 * types. The entry point for this case is the label <do_warn>;
779 * 2) The input is a valid code point but disallowed by the parameters to
780 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
781 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
782 * comprise the code point. <pack_warn> contains the result of
783 * packWARN() for the warning types. The entry point for this case is
784 * the label <disallowed>.
785 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
786 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
787 * bytes that comprise the malformation. All such malformations are
788 * assumed to be warning type <utf8>. The entry point for this case
789 * is the label <malformed>.
790 */
949cf498 791
7b52d656 792 malformed:
ba210ebe 793
eb83ed87
KW
794 if (sv && ckWARN_d(WARN_UTF8)) {
795 pack_warn = packWARN(WARN_UTF8);
796 }
797
7b52d656 798 disallowed:
eb83ed87 799
fcc8fcf6 800 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 801 if (retlen)
10edeb5d 802 *retlen = ((STRLEN) -1);
ba210ebe
JH
803 return 0;
804 }
805
7b52d656 806 do_warn:
5b311467 807
eb83ed87
KW
808 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
809 if warnings are to be raised. */
f555bc63 810 const char * const string = SvPVX_const(sv);
a0dbb045 811
f555bc63
KW
812 if (PL_op)
813 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
814 else
815 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
816 }
817
eb83ed87
KW
818 if (retlen) {
819 *retlen = curlen;
820 }
ba210ebe 821
eb83ed87 822 return outlier_ret;
a0ed51b3
LW
823}
824
8e84507e 825/*
ec5f19d0
KW
826=for apidoc utf8_to_uvchr_buf
827
828Returns the native code point of the first character in the string C<s> which
829is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 830C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 831
524080c4
KW
832If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
833enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 834C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 835(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 836C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 837the next possible position in C<s> that could begin a non-malformed character.
de69f3af 838See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 839returned.
ec5f19d0
KW
840
841=cut
842*/
843
844
845UV
846Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
847{
ec5f19d0
KW
848 assert(s < send);
849
850 return utf8n_to_uvchr(s, send - s, retlen,
851 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
852}
853
27d6c58a 854/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 855 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 856 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
857
858UV
859Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
860{
010ab96b
KW
861 UV expectlen = UTF8SKIP(s);
862 const U8* send = s + expectlen;
9ff2f0f7 863 UV uv = *s;
3986bb7c 864
27d6c58a 865 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 866 PERL_UNUSED_CONTEXT;
27d6c58a 867
010ab96b
KW
868 if (retlen) {
869 *retlen = expectlen;
870 }
871
872 /* An invariant is trivially returned */
873 if (expectlen == 1) {
9ff2f0f7 874 return uv;
010ab96b
KW
875 }
876
9ff2f0f7
KW
877#ifdef EBCDIC
878 uv = NATIVE_UTF8_TO_I8(uv);
879#endif
880
010ab96b
KW
881 /* Remove the leading bits that indicate the number of bytes, leaving just
882 * the bits that are part of the value */
883 uv &= UTF_START_MASK(expectlen);
884
885 /* Now, loop through the remaining bytes, accumulating each into the
886 * working total as we go. (I khw tried unrolling the loop for up to 4
887 * bytes, but there was no performance improvement) */
888 for (++s; s < send; s++) {
889 uv = UTF8_ACCUMULATE(uv, *s);
890 }
891
3986bb7c 892 return UNI_TO_NATIVE(uv);
010ab96b 893
27d6c58a
KW
894}
895
ec5f19d0 896/*
ec5f19d0
KW
897=for apidoc utf8_to_uvuni_buf
898
de69f3af
KW
899Only in very rare circumstances should code need to be dealing in Unicode
900(as opposed to native) code points. In those few cases, use
901C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
902
903Returns the Unicode (not-native) code point of the first character in the
904string C<s> which
ec5f19d0
KW
905is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
906C<retlen> will be set to the length, in bytes, of that character.
907
524080c4
KW
908If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
909enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
910NULL) to -1. If those warnings are off, the computed value if well-defined (or
911the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
912is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
913next possible position in C<s> that could begin a non-malformed character.
de69f3af 914See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
915
916=cut
917*/
918
919UV
920Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
921{
922 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
923
924 assert(send > s);
925
926 /* Call the low level routine asking for checks */
de69f3af
KW
927 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
928 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
929}
930
b76347f2 931/*
87cea99e 932=for apidoc utf8_length
b76347f2
JH
933
934Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
935Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
936up past C<e>, croaks.
b76347f2
JH
937
938=cut
939*/
940
941STRLEN
35a4481c 942Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
943{
944 STRLEN len = 0;
945
7918f24d
NC
946 PERL_ARGS_ASSERT_UTF8_LENGTH;
947
8850bf83
JH
948 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
949 * the bitops (especially ~) can create illegal UTF-8.
950 * In other words: in Perl UTF-8 is not just for Unicode. */
951
a3b680e6
AL
952 if (e < s)
953 goto warn_and_return;
b76347f2 954 while (s < e) {
4cbf4130 955 s += UTF8SKIP(s);
8e91ec7f
AV
956 len++;
957 }
958
959 if (e != s) {
960 len--;
961 warn_and_return:
9b387841
NC
962 if (PL_op)
963 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
964 "%s in %s", unees, OP_DESC(PL_op));
965 else
61a12c31 966 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
967 }
968
969 return len;
970}
971
b06226ff 972/*
87cea99e 973=for apidoc utf8_distance
b06226ff 974
1e54db1a 975Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
976and C<b>.
977
978WARNING: use only if you *know* that the pointers point inside the
979same UTF-8 buffer.
980
37607a96
PK
981=cut
982*/
a0ed51b3 983
02eb7b47 984IV
35a4481c 985Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 986{
7918f24d
NC
987 PERL_ARGS_ASSERT_UTF8_DISTANCE;
988
bf1665bc 989 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
990}
991
b06226ff 992/*
87cea99e 993=for apidoc utf8_hop
b06226ff 994
8850bf83
JH
995Return the UTF-8 pointer C<s> displaced by C<off> characters, either
996forward or backward.
b06226ff
JH
997
998WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
999the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1000on the first byte of character or just after the last byte of a character.
b06226ff 1001
37607a96
PK
1002=cut
1003*/
a0ed51b3
LW
1004
1005U8 *
ddeaf645 1006Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1007{
7918f24d
NC
1008 PERL_ARGS_ASSERT_UTF8_HOP;
1009
8850bf83
JH
1010 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1011 * the bitops (especially ~) can create illegal UTF-8.
1012 * In other words: in Perl UTF-8 is not just for Unicode. */
1013
a0ed51b3
LW
1014 if (off >= 0) {
1015 while (off--)
1016 s += UTF8SKIP(s);
1017 }
1018 else {
1019 while (off++) {
1020 s--;
8850bf83
JH
1021 while (UTF8_IS_CONTINUATION(*s))
1022 s--;
a0ed51b3
LW
1023 }
1024 }
4373e329 1025 return (U8 *)s;
a0ed51b3
LW
1026}
1027
6940069f 1028/*
fed3ba5d
NC
1029=for apidoc bytes_cmp_utf8
1030
a1433954 1031Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1032sequence of characters (stored as UTF-8)
1033in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1034equal, -1 or -2 if the first string is less than the second string, +1 or +2
1035if the first string is greater than the second string.
1036
1037-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1038longer string. -2 or +2 is returned if
1039there was a difference between characters
fed3ba5d
NC
1040within the strings.
1041
1042=cut
1043*/
1044
1045int
1046Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1047{
1048 const U8 *const bend = b + blen;
1049 const U8 *const uend = u + ulen;
1050
1051 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1052
1053 while (b < bend && u < uend) {
1054 U8 c = *u++;
1055 if (!UTF8_IS_INVARIANT(c)) {
1056 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1057 if (u < uend) {
1058 U8 c1 = *u++;
1059 if (UTF8_IS_CONTINUATION(c1)) {
94bb8c36 1060 c = TWO_BYTE_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1061 } else {
1062 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1063 "Malformed UTF-8 character "
1064 "(unexpected non-continuation byte 0x%02x"
1065 ", immediately after start byte 0x%02x)"
1066 /* Dear diag.t, it's in the pod. */
1067 "%s%s", c1, c,
1068 PL_op ? " in " : "",
1069 PL_op ? OP_DESC(PL_op) : "");
1070 return -2;
1071 }
1072 } else {
1073 if (PL_op)
1074 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1075 "%s in %s", unees, OP_DESC(PL_op));
1076 else
61a12c31 1077 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1078 return -2; /* Really want to return undef :-) */
1079 }
1080 } else {
1081 return -2;
1082 }
1083 }
1084 if (*b != c) {
1085 return *b < c ? -2 : +2;
1086 }
1087 ++b;
1088 }
1089
1090 if (b == bend && u == uend)
1091 return 0;
1092
1093 return b < bend ? +1 : -1;
1094}
1095
1096/*
87cea99e 1097=for apidoc utf8_to_bytes
6940069f 1098
2bbc8d55 1099Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1100Unlike L</bytes_to_utf8>, this over-writes the original string, and
1101updates C<len> to contain the new length.
67e989fb 1102Returns zero on failure, setting C<len> to -1.
6940069f 1103
a1433954 1104If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1105
6940069f
GS
1106=cut
1107*/
1108
1109U8 *
37607a96 1110Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1111{
d4c19fe8
AL
1112 U8 * const save = s;
1113 U8 * const send = s + *len;
6940069f 1114 U8 *d;
246fae53 1115
7918f24d 1116 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1117 PERL_UNUSED_CONTEXT;
7918f24d 1118
1e54db1a 1119 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1120 while (s < send) {
d59937ca
KW
1121 if (! UTF8_IS_INVARIANT(*s)) {
1122 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1123 *len = ((STRLEN) -1);
1124 return 0;
1125 }
1126 s++;
dcad2880 1127 }
d59937ca 1128 s++;
246fae53 1129 }
dcad2880
JH
1130
1131 d = s = save;
6940069f 1132 while (s < send) {
80e0b38f
KW
1133 U8 c = *s++;
1134 if (! UTF8_IS_INVARIANT(c)) {
1135 /* Then it is two-byte encoded */
1136 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1137 s++;
1138 }
1139 *d++ = c;
6940069f
GS
1140 }
1141 *d = '\0';
246fae53 1142 *len = d - save;
6940069f
GS
1143 return save;
1144}
1145
1146/*
87cea99e 1147=for apidoc bytes_from_utf8
f9a63242 1148
2bbc8d55 1149Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1150Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1151the newly-created string, and updates C<len> to contain the new
1152length. Returns the original string if no conversion occurs, C<len>
72d33970 1153is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 11540 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1155in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1156
37607a96
PK
1157=cut
1158*/
f9a63242
JH
1159
1160U8 *
e1ec3a88 1161Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1162{
f9a63242 1163 U8 *d;
e1ec3a88
AL
1164 const U8 *start = s;
1165 const U8 *send;
f9a63242
JH
1166 I32 count = 0;
1167
7918f24d 1168 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1169 PERL_UNUSED_CONTEXT;
f9a63242 1170 if (!*is_utf8)
73d840c0 1171 return (U8 *)start;
f9a63242 1172
1e54db1a 1173 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1174 for (send = s + *len; s < send;) {
d59937ca
KW
1175 if (! UTF8_IS_INVARIANT(*s)) {
1176 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1177 return (U8 *)start;
d59937ca
KW
1178 }
1179 count++;
1180 s++;
db42d148 1181 }
d59937ca 1182 s++;
f9a63242
JH
1183 }
1184
35da51f7 1185 *is_utf8 = FALSE;
f9a63242 1186
212542aa 1187 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1188 s = start; start = d;
f9a63242
JH
1189 while (s < send) {
1190 U8 c = *s++;
1a91c45d 1191 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1192 /* Then it is two-byte encoded */
1a91c45d
KW
1193 c = TWO_BYTE_UTF8_TO_NATIVE(c, *s);
1194 s++;
c4d5f83a
NIS
1195 }
1196 *d++ = c;
f9a63242
JH
1197 }
1198 *d = '\0';
1199 *len = d - start;
73d840c0 1200 return (U8 *)start;
f9a63242
JH
1201}
1202
1203/*
87cea99e 1204=for apidoc bytes_to_utf8
6940069f 1205
ff97e5cf
KW
1206Converts a string C<s> of length C<len> bytes from the native encoding into
1207UTF-8.
6662521e 1208Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1209reflect the new length in bytes.
6940069f 1210
75200dff 1211A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1212
1213If you want to convert to UTF-8 from encodings other than
1214the native (Latin1 or EBCDIC),
a1433954 1215see L</sv_recode_to_utf8>().
c9ada85f 1216
497711e7 1217=cut
6940069f
GS
1218*/
1219
c682ebef
FC
1220/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1221 likewise need duplication. */
1222
6940069f 1223U8*
35a4481c 1224Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1225{
35a4481c 1226 const U8 * const send = s + (*len);
6940069f
GS
1227 U8 *d;
1228 U8 *dst;
7918f24d
NC
1229
1230 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1231 PERL_UNUSED_CONTEXT;
6940069f 1232
212542aa 1233 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1234 dst = d;
1235
1236 while (s < send) {
55d09dc8
KW
1237 append_utf8_from_native_byte(*s, &d);
1238 s++;
6940069f
GS
1239 }
1240 *d = '\0';
6662521e 1241 *len = d-dst;
6940069f
GS
1242 return dst;
1243}
1244
a0ed51b3 1245/*
dea0fc0b 1246 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1247 *
1248 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1249 * We optimize for native, for obvious reasons. */
1250
1251U8*
dea0fc0b 1252Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1253{
dea0fc0b
JH
1254 U8* pend;
1255 U8* dstart = d;
1256
7918f24d
NC
1257 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1258
dea0fc0b 1259 if (bytelen & 1)
f5992bc4 1260 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1261
1262 pend = p + bytelen;
1263
a0ed51b3 1264 while (p < pend) {
dea0fc0b
JH
1265 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1266 p += 2;
56d37426
KW
1267 if (UNI_IS_INVARIANT(uv)) {
1268 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1269 continue;
1270 }
56d37426
KW
1271 if (uv <= MAX_UTF8_TWO_BYTE) {
1272 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1273 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1274 continue;
1275 }
46956fad
KW
1276#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1277#define LAST_HIGH_SURROGATE 0xDBFF
1278#define FIRST_LOW_SURROGATE 0xDC00
1279#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1280
1281 /* This assumes that most uses will be in the first Unicode plane, not
1282 * needing surrogates */
1283 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1284 && uv <= UNICODE_SURROGATE_LAST))
1285 {
1286 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1287 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1288 }
1289 else {
01ea242b 1290 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1291 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1292 || UNLIKELY(low > LAST_LOW_SURROGATE))
1293 {
01ea242b 1294 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1295 }
1296 p += 2;
46956fad
KW
1297 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1298 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1299 }
a0ed51b3 1300 }
56d37426
KW
1301#ifdef EBCDIC
1302 d = uvoffuni_to_utf8_flags(d, uv, 0);
1303#else
a0ed51b3 1304 if (uv < 0x10000) {
eb160463
GS
1305 *d++ = (U8)(( uv >> 12) | 0xe0);
1306 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1307 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1308 continue;
1309 }
1310 else {
eb160463
GS
1311 *d++ = (U8)(( uv >> 18) | 0xf0);
1312 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1313 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1314 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1315 continue;
1316 }
56d37426 1317#endif
a0ed51b3 1318 }
dea0fc0b 1319 *newlen = d - dstart;
a0ed51b3
LW
1320 return d;
1321}
1322
1323/* Note: this one is slightly destructive of the source. */
1324
1325U8*
dea0fc0b 1326Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1327{
1328 U8* s = (U8*)p;
d4c19fe8 1329 U8* const send = s + bytelen;
7918f24d
NC
1330
1331 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1332
e0ea5e2d
NC
1333 if (bytelen & 1)
1334 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1335 (UV)bytelen);
1336
a0ed51b3 1337 while (s < send) {
d4c19fe8 1338 const U8 tmp = s[0];
a0ed51b3
LW
1339 s[0] = s[1];
1340 s[1] = tmp;
1341 s += 2;
1342 }
dea0fc0b 1343 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1344}
1345
922e8cb4
KW
1346bool
1347Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1348{
1349 U8 tmpbuf[UTF8_MAXBYTES+1];
1350 uvchr_to_utf8(tmpbuf, c);
1351 return _is_utf8_FOO(classnum, tmpbuf);
1352}
1353
f9ae8fb6
JD
1354/* Internal function so we can deprecate the external one, and call
1355 this one from other deprecated functions in this file */
1356
f2645549
KW
1357bool
1358Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1359{
f2645549 1360 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1361
1362 if (*p == '_')
1363 return TRUE;
f25ce844 1364 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1365}
1366
5092f92a 1367bool
eba68aa0
KW
1368Perl__is_uni_perl_idcont(pTHX_ UV c)
1369{
1370 U8 tmpbuf[UTF8_MAXBYTES+1];
1371 uvchr_to_utf8(tmpbuf, c);
1372 return _is_utf8_perl_idcont(tmpbuf);
1373}
1374
1375bool
f91dcd13
KW
1376Perl__is_uni_perl_idstart(pTHX_ UV c)
1377{
1378 U8 tmpbuf[UTF8_MAXBYTES+1];
1379 uvchr_to_utf8(tmpbuf, c);
1380 return _is_utf8_perl_idstart(tmpbuf);
1381}
1382
3a4c58c9
KW
1383UV
1384Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1385{
1386 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1387 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1388 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1389 * either "SS" or "Ss". Which one to use is passed into the routine in
1390 * 'S_or_s' to avoid a test */
1391
1392 UV converted = toUPPER_LATIN1_MOD(c);
1393
1394 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1395
1396 assert(S_or_s == 'S' || S_or_s == 's');
1397
6f2d5cbc 1398 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1399 characters in this range */
3a4c58c9
KW
1400 *p = (U8) converted;
1401 *lenp = 1;
1402 return converted;
1403 }
1404
1405 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1406 * which it maps to one of them, so as to only have to have one check for
1407 * it in the main case */
1408 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1409 switch (c) {
1410 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1411 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1412 break;
1413 case MICRO_SIGN:
1414 converted = GREEK_CAPITAL_LETTER_MU;
1415 break;
79e064b9
KW
1416#if UNICODE_MAJOR_VERSION > 2 \
1417 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1418 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1419 case LATIN_SMALL_LETTER_SHARP_S:
1420 *(p)++ = 'S';
1421 *p = S_or_s;
1422 *lenp = 2;
1423 return 'S';
79e064b9 1424#endif
3a4c58c9
KW
1425 default:
1426 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1427 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1428 }
1429 }
1430
1431 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1432 *p = UTF8_TWO_BYTE_LO(converted);
1433 *lenp = 2;
1434
1435 return converted;
1436}
1437
50bda2c3
KW
1438/* Call the function to convert a UTF-8 encoded character to the specified case.
1439 * Note that there may be more than one character in the result.
1440 * INP is a pointer to the first byte of the input character
1441 * OUTP will be set to the first byte of the string of changed characters. It
1442 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1443 * LENP will be set to the length in bytes of the string of changed characters
1444 *
1445 * The functions return the ordinal of the first character in the string of OUTP */
4a8240a3
KW
1446#define CALL_UPPER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_toupper, "ToUc", "")
1447#define CALL_TITLE_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_totitle, "ToTc", "")
1448#define CALL_LOWER_CASE(INP, OUTP, LENP) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tolower, "ToLc", "")
50bda2c3
KW
1449
1450/* This additionally has the input parameter SPECIALS, which if non-zero will
1451 * cause this to use the SPECIALS hash for folding (meaning get full case
1452 * folding); otherwise, when zero, this implies a simple case fold */
4a8240a3 1453#define CALL_FOLD_CASE(INP, OUTP, LENP, SPECIALS) Perl_to_utf8_case(aTHX_ INP, OUTP, LENP, &PL_utf8_tofold, "ToCf", (SPECIALS) ? "" : NULL)
c3fd2246 1454
84afefe6
JH
1455UV
1456Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1457{
a1433954
KW
1458 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1459 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1460 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1461 * the changed version may be longer than the original character.
1462 *
1463 * The ordinal of the first character of the changed version is returned
1464 * (but note, as explained above, that there may be more.) */
1465
7918f24d
NC
1466 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1467
3a4c58c9
KW
1468 if (c < 256) {
1469 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1470 }
1471
0ebc6274 1472 uvchr_to_utf8(p, c);
3a4c58c9 1473 return CALL_UPPER_CASE(p, p, lenp);
a0ed51b3
LW
1474}
1475
84afefe6
JH
1476UV
1477Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1478{
7918f24d
NC
1479 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1480
3a4c58c9
KW
1481 if (c < 256) {
1482 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1483 }
1484
0ebc6274 1485 uvchr_to_utf8(p, c);
3a4c58c9 1486 return CALL_TITLE_CASE(p, p, lenp);
a0ed51b3
LW
1487}
1488
afc16117 1489STATIC U8
81611534 1490S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1491{
1492 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1493 * those, converting the result to UTF-8. Since the result is always just
a1433954 1494 * one character, we allow <p> to be NULL */
afc16117
KW
1495
1496 U8 converted = toLOWER_LATIN1(c);
1497
1498 if (p != NULL) {
6f2d5cbc 1499 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1500 *p = converted;
1501 *lenp = 1;
1502 }
1503 else {
430c9760
KW
1504 /* Result is known to always be < 256, so can use the EIGHT_BIT
1505 * macros */
1506 *p = UTF8_EIGHT_BIT_HI(converted);
1507 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1508 *lenp = 2;
1509 }
1510 }
1511 return converted;
1512}
1513
84afefe6
JH
1514UV
1515Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1516{
7918f24d
NC
1517 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1518
afc16117
KW
1519 if (c < 256) {
1520 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1521 }
1522
afc16117 1523 uvchr_to_utf8(p, c);
968c5e6a 1524 return CALL_LOWER_CASE(p, p, lenp);
a0ed51b3
LW
1525}
1526
84afefe6 1527UV
51910141 1528Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1529{
51910141 1530 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1531 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1532 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1533 *
1534 * Not to be used for locale folds
51910141 1535 */
f673fad4 1536
a1dde8de
KW
1537 UV converted;
1538
1539 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1540 PERL_UNUSED_CONTEXT;
a1dde8de 1541
1ca267a5
KW
1542 assert (! (flags & FOLD_FLAGS_LOCALE));
1543
a1dde8de
KW
1544 if (c == MICRO_SIGN) {
1545 converted = GREEK_SMALL_LETTER_MU;
1546 }
9b63e895
KW
1547#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1548 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1549 || UNICODE_DOT_DOT_VERSION > 0)
51910141 1550 else if ((flags & FOLD_FLAGS_FULL) && c == LATIN_SMALL_LETTER_SHARP_S) {
1ca267a5
KW
1551
1552 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1553 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1554 * under those circumstances. */
1555 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1556 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1557 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1558 p, *lenp, U8);
1559 return LATIN_SMALL_LETTER_LONG_S;
1560 }
1561 else {
4f489194
KW
1562 *(p)++ = 's';
1563 *p = 's';
1564 *lenp = 2;
1565 return 's';
1ca267a5 1566 }
a1dde8de 1567 }
9b63e895 1568#endif
a1dde8de
KW
1569 else { /* In this range the fold of all other characters is their lower
1570 case */
1571 converted = toLOWER_LATIN1(c);
1572 }
1573
6f2d5cbc 1574 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1575 *p = (U8) converted;
1576 *lenp = 1;
1577 }
1578 else {
1579 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1580 *p = UTF8_TWO_BYTE_LO(converted);
1581 *lenp = 2;
1582 }
1583
1584 return converted;
1585}
1586
1587UV
31f05a37 1588Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1589{
4b593389 1590
a0270393
KW
1591 /* Not currently externally documented, and subject to change
1592 * <flags> bits meanings:
1593 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1594 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1595 * locale are to be used.
a0270393
KW
1596 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1597 */
4b593389 1598
36bb2ab6 1599 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1600
780fcc9f
KW
1601 if (flags & FOLD_FLAGS_LOCALE) {
1602 /* Treat a UTF-8 locale as not being in locale at all */
1603 if (IN_UTF8_CTYPE_LOCALE) {
1604 flags &= ~FOLD_FLAGS_LOCALE;
1605 }
1606 else {
1607 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1608 goto needs_full_generality;
780fcc9f 1609 }
31f05a37
KW
1610 }
1611
a1dde8de 1612 if (c < 256) {
e7b7ac46 1613 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1614 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1615 }
1616
2f306ab9 1617 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1618 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1619 uvchr_to_utf8(p, c);
1620 return CALL_FOLD_CASE(p, p, lenp, flags & FOLD_FLAGS_FULL);
1621 }
1622 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1623 the special flags. */
1624 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1625
1626 needs_full_generality:
a0270393 1627 uvchr_to_utf8(utf8_c, c);
445bf929 1628 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1629 }
84afefe6
JH
1630}
1631
26483009 1632PERL_STATIC_INLINE bool
5141f98e 1633S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1634 const char *const swashname, SV* const invlist)
bde6a22d 1635{
ea317ccb
KW
1636 /* returns a boolean giving whether or not the UTF8-encoded character that
1637 * starts at <p> is in the swash indicated by <swashname>. <swash>
1638 * contains a pointer to where the swash indicated by <swashname>
1639 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1640 * look at <*swash> and only generate a swash if it is not null. <invlist>
1641 * is NULL or an inversion list that defines the swash. If not null, it
1642 * saves time during initialization of the swash.
ea317ccb
KW
1643 *
1644 * Note that it is assumed that the buffer length of <p> is enough to
1645 * contain all the bytes that comprise the character. Thus, <*p> should
1646 * have been checked before this call for mal-formedness enough to assure
1647 * that. */
1648
7918f24d
NC
1649 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1650
492a624f 1651 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1652 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1653 * as far as there being enough bytes available in it to accommodate the
1654 * character without reading beyond the end, and pass that number on to the
1655 * validating routine */
6302f837 1656 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1657 if (ckWARN_d(WARN_UTF8)) {
1658 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1659 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1660 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1661 what the malformation is */
1662 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1663 }
1664 }
1665 return FALSE;
1666 }
87367d5f
KW
1667 if (!*swash) {
1668 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1669 *swash = _core_swash_init("utf8",
1670
1671 /* Only use the name if there is no inversion
1672 * list; otherwise will go out to disk */
1673 (invlist) ? "" : swashname,
1674
1675 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1676 }
28123549 1677
bde6a22d
NC
1678 return swash_fetch(*swash, p, TRUE) != 0;
1679}
1680
1681bool
922e8cb4
KW
1682Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1683{
922e8cb4
KW
1684 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1685
1686 assert(classnum < _FIRST_NON_SWASH_CC);
1687
f25ce844
KW
1688 return is_utf8_common(p,
1689 &PL_utf8_swash_ptrs[classnum],
1690 swash_property_names[classnum],
1691 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1692}
1693
1694bool
f2645549 1695Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1696{
f2645549 1697 SV* invlist = NULL;
7918f24d 1698
f2645549 1699 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1700
f2645549
KW
1701 if (! PL_utf8_perl_idstart) {
1702 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1703 }
60071a22 1704 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1705}
1706
1707bool
f2645549 1708Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1709{
f2645549 1710 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1711
1712 if (*p == '_')
1713 return TRUE;
f25ce844 1714 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1715}
1716
1717bool
eba68aa0
KW
1718Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1719{
b24b43f7 1720 SV* invlist = NULL;
eba68aa0
KW
1721
1722 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1723
b24b43f7
KW
1724 if (! PL_utf8_perl_idcont) {
1725 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1726 }
60071a22 1727 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1728}
1729
eba68aa0 1730bool
f2645549 1731Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1732{
f2645549 1733 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1734
f25ce844 1735 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1736}
1737
1738bool
f2645549 1739Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1740{
f2645549 1741 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1742
f25ce844 1743 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1744}
1745
1746bool
7dbf68d2
KW
1747Perl__is_utf8_mark(pTHX_ const U8 *p)
1748{
7dbf68d2
KW
1749 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1750
f25ce844 1751 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1752}
1753
6b5c0936 1754/*
87cea99e 1755=for apidoc to_utf8_case
6b5c0936 1756
6fae5207 1757C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1758the character that is being converted. This routine assumes that the character
1759at C<p> is well-formed.
6b5c0936 1760
6fae5207
KW
1761C<ustrp> is a pointer to the character buffer to put the
1762conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1763of the result.
1764
6fae5207 1765C<swashp> is a pointer to the swash to use.
6b5c0936 1766
a1433954 1767Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1768and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1769but not always, a multicharacter mapping), is tried first.
6b5c0936 1770
4a8240a3
KW
1771C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1772any special mappings; C<""> means to use the special mappings. Values other
1773than these two are treated as the name of the hash containing the special
1774mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1775
796b6530
KW
1776C<normal> is a string like C<"ToLower"> which means the swash
1777C<%utf8::ToLower>.
0134edef
JH
1778
1779=cut */
6b5c0936 1780
2104c8d9 1781UV
9a957fbc
AL
1782Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1783 SV **swashp, const char *normal, const char *special)
a0ed51b3 1784{
0134edef 1785 STRLEN len = 0;
f4cd282c 1786 const UV uv1 = valid_utf8_to_uvchr(p, NULL);
7918f24d
NC
1787
1788 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1789
9ae3ac1a
KW
1790 /* Note that swash_fetch() doesn't output warnings for these because it
1791 * assumes we will */
8457b38f 1792 if (uv1 >= UNICODE_SURROGATE_FIRST) {
9ae3ac1a 1793 if (uv1 <= UNICODE_SURROGATE_LAST) {
8457b38f
KW
1794 if (ckWARN_d(WARN_SURROGATE)) {
1795 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1796 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1797 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1798 }
9ae3ac1a
KW
1799 }
1800 else if (UNICODE_IS_SUPER(uv1)) {
8457b38f
KW
1801 if (ckWARN_d(WARN_NON_UNICODE)) {
1802 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1803 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1804 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1805 }
9ae3ac1a
KW
1806 }
1807
1808 /* Note that non-characters are perfectly legal, so no warning should
1809 * be given */
1810 }
1811
0134edef 1812 if (!*swashp) /* load on-demand */
5ab9d2ef 1813 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1814
a6f87d8c 1815 if (special) {
0134edef 1816 /* It might be "special" (sometimes, but not always,
2a37f04d 1817 * a multicharacter mapping) */
4a8240a3 1818 HV *hv = NULL;
b08cf34e
JH
1819 SV **svp;
1820
4a8240a3
KW
1821 /* If passed in the specials name, use that; otherwise use any
1822 * given in the swash */
1823 if (*special != '\0') {
1824 hv = get_hv(special, 0);
1825 }
1826 else {
1827 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1828 if (svp) {
1829 hv = MUTABLE_HV(SvRV(*svp));
1830 }
1831 }
1832
176fe009 1833 if (hv
5f560d8a 1834 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
1835 && (*svp))
1836 {
cfd0369c 1837 const char *s;
47654450 1838
cfd0369c 1839 s = SvPV_const(*svp, len);
47654450 1840 if (len == 1)
f4cd282c 1841 /* EIGHTBIT */
c80e42f3 1842 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1843 else {
d2dcd0fb 1844 Copy(s, ustrp, len, U8);
29e98929 1845 }
983ffd37 1846 }
0134edef
JH
1847 }
1848
1849 if (!len && *swashp) {
4a4088c4 1850 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 1851
0134edef
JH
1852 if (uv2) {
1853 /* It was "normal" (a single character mapping). */
f4cd282c 1854 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1855 }
1856 }
1feea2c7 1857
cbe07460
KW
1858 if (len) {
1859 if (lenp) {
1860 *lenp = len;
1861 }
1862 return valid_utf8_to_uvchr(ustrp, 0);
1863 }
1864
1865 /* Here, there was no mapping defined, which means that the code point maps
1866 * to itself. Return the inputs */
bfdf22ec 1867 len = UTF8SKIP(p);
ca9fab46
KW
1868 if (p != ustrp) { /* Don't copy onto itself */
1869 Copy(p, ustrp, len, U8);
1870 }
0134edef 1871
2a37f04d
JH
1872 if (lenp)
1873 *lenp = len;
1874
f4cd282c 1875 return uv1;
cbe07460 1876
a0ed51b3
LW
1877}
1878
051a06d4 1879STATIC UV
357aadde 1880S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 1881{
4a4088c4 1882 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
1883 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1884 * result contains a character that crosses the 255/256 boundary, disallow
1885 * the change, and return the original code point. See L<perlfunc/lc> for
1886 * why;
051a06d4 1887 *
a1433954
KW
1888 * p points to the original string whose case was changed; assumed
1889 * by this routine to be well-formed
051a06d4
KW
1890 * result the code point of the first character in the changed-case string
1891 * ustrp points to the changed-case string (<result> represents its first char)
1892 * lenp points to the length of <ustrp> */
1893
1894 UV original; /* To store the first code point of <p> */
1895
1896 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
1897
a4f12ed7 1898 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
1899
1900 /* We know immediately if the first character in the string crosses the
1901 * boundary, so can skip */
1902 if (result > 255) {
1903
1904 /* Look at every character in the result; if any cross the
1905 * boundary, the whole thing is disallowed */
1906 U8* s = ustrp + UTF8SKIP(ustrp);
1907 U8* e = ustrp + *lenp;
1908 while (s < e) {
a4f12ed7 1909 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
1910 goto bad_crossing;
1911 }
1912 s += UTF8SKIP(s);
1913 }
1914
613abc6d
KW
1915 /* Here, no characters crossed, result is ok as-is, but we warn. */
1916 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
1917 return result;
1918 }
1919
7b52d656 1920 bad_crossing:
051a06d4
KW
1921
1922 /* Failed, have to return the original */
4b88fb76 1923 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
1924
1925 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
1926 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
1927 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
1928 "resolved to \"\\x{%"UVXf"}\".",
357aadde 1929 OP_DESC(PL_op),
ab0b796c
KW
1930 original,
1931 original);
051a06d4
KW
1932 Copy(p, ustrp, *lenp, char);
1933 return original;
1934}
1935
d3e79532 1936/*
87cea99e 1937=for apidoc to_utf8_upper
d3e79532 1938
1f607577 1939Instead use L</toUPPER_utf8>.
a1433954 1940
d3e79532
JH
1941=cut */
1942
051a06d4 1943/* Not currently externally documented, and subject to change:
31f05a37
KW
1944 * <flags> is set iff iff the rules from the current underlying locale are to
1945 * be used. */
051a06d4 1946
2104c8d9 1947UV
31f05a37 1948Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 1949{
051a06d4
KW
1950 UV result;
1951
1952 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 1953
780fcc9f
KW
1954 if (flags) {
1955 /* Treat a UTF-8 locale as not being in locale at all */
1956 if (IN_UTF8_CTYPE_LOCALE) {
1957 flags = FALSE;
1958 }
1959 else {
1960 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
1961 }
31f05a37
KW
1962 }
1963
3a4c58c9 1964 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
1965 if (flags) {
1966 result = toUPPER_LC(*p);
1967 }
1968 else {
81c6c7ce 1969 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 1970 }
3a4c58c9
KW
1971 }
1972 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 1973 if (flags) {
a6d8b88b 1974 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 1975 result = toUPPER_LC(c);
051a06d4
KW
1976 }
1977 else {
94bb8c36 1978 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 1979 ustrp, lenp, 'S');
051a06d4
KW
1980 }
1981 }
4a4088c4 1982 else { /* UTF-8, ord above 255 */
051a06d4
KW
1983 result = CALL_UPPER_CASE(p, ustrp, lenp);
1984
1985 if (flags) {
357aadde 1986 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
1987 }
1988 return result;
1989 }
1990
4a4088c4 1991 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
1992 if (UTF8_IS_INVARIANT(result)) {
1993 *ustrp = (U8) result;
1994 *lenp = 1;
1995 }
1996 else {
62cb07ea
KW
1997 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
1998 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 1999 *lenp = 2;
3a4c58c9 2000 }
baa60164 2001
051a06d4 2002 return result;
983ffd37 2003}
a0ed51b3 2004
d3e79532 2005/*
87cea99e 2006=for apidoc to_utf8_title
d3e79532 2007
1f607577 2008Instead use L</toTITLE_utf8>.
a1433954 2009
d3e79532
JH
2010=cut */
2011
051a06d4 2012/* Not currently externally documented, and subject to change:
31f05a37
KW
2013 * <flags> is set iff the rules from the current underlying locale are to be
2014 * used. Since titlecase is not defined in POSIX, for other than a
2015 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2016 */
051a06d4 2017
983ffd37 2018UV
31f05a37 2019Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2020{
051a06d4
KW
2021 UV result;
2022
2023 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2024
780fcc9f
KW
2025 if (flags) {
2026 /* Treat a UTF-8 locale as not being in locale at all */
2027 if (IN_UTF8_CTYPE_LOCALE) {
2028 flags = FALSE;
2029 }
2030 else {
2031 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2032 }
31f05a37
KW
2033 }
2034
3a4c58c9 2035 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2036 if (flags) {
2037 result = toUPPER_LC(*p);
2038 }
2039 else {
81c6c7ce 2040 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2041 }
3a4c58c9
KW
2042 }
2043 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2044 if (flags) {
a6d8b88b 2045 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2046 result = toUPPER_LC(c);
051a06d4
KW
2047 }
2048 else {
94bb8c36 2049 return _to_upper_title_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2050 ustrp, lenp, 's');
051a06d4
KW
2051 }
2052 }
4a4088c4 2053 else { /* UTF-8, ord above 255 */
051a06d4
KW
2054 result = CALL_TITLE_CASE(p, ustrp, lenp);
2055
2056 if (flags) {
357aadde 2057 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2058 }
2059 return result;
2060 }
2061
4a4088c4 2062 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2063 if (UTF8_IS_INVARIANT(result)) {
2064 *ustrp = (U8) result;
2065 *lenp = 1;
2066 }
2067 else {
62cb07ea
KW
2068 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2069 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2070 *lenp = 2;
3a4c58c9
KW
2071 }
2072
051a06d4 2073 return result;
a0ed51b3
LW
2074}
2075
d3e79532 2076/*
87cea99e 2077=for apidoc to_utf8_lower
d3e79532 2078
1f607577 2079Instead use L</toLOWER_utf8>.
a1433954 2080
d3e79532
JH
2081=cut */
2082
051a06d4 2083/* Not currently externally documented, and subject to change:
31f05a37
KW
2084 * <flags> is set iff iff the rules from the current underlying locale are to
2085 * be used.
2086 */
051a06d4 2087
2104c8d9 2088UV
31f05a37 2089Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2090{
051a06d4
KW
2091 UV result;
2092
051a06d4 2093 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2094
780fcc9f
KW
2095 if (flags) {
2096 /* Treat a UTF-8 locale as not being in locale at all */
2097 if (IN_UTF8_CTYPE_LOCALE) {
2098 flags = FALSE;
2099 }
2100 else {
2101 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2102 }
31f05a37
KW
2103 }
2104
968c5e6a 2105 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2106 if (flags) {
2107 result = toLOWER_LC(*p);
2108 }
2109 else {
81c6c7ce 2110 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2111 }
968c5e6a
KW
2112 }
2113 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2114 if (flags) {
a6d8b88b 2115 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2116 result = toLOWER_LC(c);
051a06d4
KW
2117 }
2118 else {
94bb8c36 2119 return to_lower_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2120 ustrp, lenp);
051a06d4 2121 }
968c5e6a 2122 }
4a4088c4 2123 else { /* UTF-8, ord above 255 */
051a06d4
KW
2124 result = CALL_LOWER_CASE(p, ustrp, lenp);
2125
2126 if (flags) {
357aadde 2127 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2128 }
968c5e6a 2129
051a06d4
KW
2130 return result;
2131 }
2132
4a4088c4 2133 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2134 if (UTF8_IS_INVARIANT(result)) {
2135 *ustrp = (U8) result;
2136 *lenp = 1;
2137 }
2138 else {
62cb07ea
KW
2139 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2140 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2141 *lenp = 2;
2142 }
2143
051a06d4 2144 return result;
b4e400f9
JH
2145}
2146
d3e79532 2147/*
87cea99e 2148=for apidoc to_utf8_fold
d3e79532 2149
1f607577 2150Instead use L</toFOLD_utf8>.
a1433954 2151
d3e79532
JH
2152=cut */
2153
051a06d4
KW
2154/* Not currently externally documented, and subject to change,
2155 * in <flags>
31f05a37
KW
2156 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2157 * locale are to be used.
051a06d4
KW
2158 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2159 * otherwise simple folds
a0270393
KW
2160 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2161 * prohibited
445bf929 2162 */
36bb2ab6 2163
b4e400f9 2164UV
445bf929 2165Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2166{
051a06d4
KW
2167 UV result;
2168
36bb2ab6 2169 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2170
a0270393
KW
2171 /* These are mutually exclusive */
2172 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2173
50ba90ff
KW
2174 assert(p != ustrp); /* Otherwise overwrites */
2175
780fcc9f
KW
2176 if (flags & FOLD_FLAGS_LOCALE) {
2177 /* Treat a UTF-8 locale as not being in locale at all */
2178 if (IN_UTF8_CTYPE_LOCALE) {
2179 flags &= ~FOLD_FLAGS_LOCALE;
2180 }
2181 else {
2182 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2183 }
31f05a37
KW
2184 }
2185
a1dde8de 2186 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2187 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2188 result = toFOLD_LC(*p);
051a06d4
KW
2189 }
2190 else {
81c6c7ce 2191 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2192 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2193 }
a1dde8de
KW
2194 }
2195 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2196 if (flags & FOLD_FLAGS_LOCALE) {
a6d8b88b 2197 U8 c = TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2198 result = toFOLD_LC(c);
051a06d4
KW
2199 }
2200 else {
94bb8c36 2201 return _to_fold_latin1(TWO_BYTE_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2202 ustrp, lenp,
1ca267a5 2203 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2204 }
a1dde8de 2205 }
4a4088c4 2206 else { /* UTF-8, ord above 255 */
a0270393 2207 result = CALL_FOLD_CASE(p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2208
1ca267a5
KW
2209 if (flags & FOLD_FLAGS_LOCALE) {
2210
76f2ffcd 2211# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2212 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2213
2214# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2215# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2216
2217 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2218
538e84ed
KW
2219 /* Special case these two characters, as what normally gets
2220 * returned under locale doesn't work */
76f2ffcd
KW
2221 if (UTF8SKIP(p) == cap_sharp_s_len
2222 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2223 {
ab0b796c
KW
2224 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2225 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2226 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2227 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2228 goto return_long_s;
2229 }
0766489e
KW
2230 else
2231#endif
2232 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2233 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2234 {
ab0b796c
KW
2235 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2236 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2237 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2238 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2239 goto return_ligature_st;
2240 }
74894415
KW
2241
2242#if UNICODE_MAJOR_VERSION == 3 \
2243 && UNICODE_DOT_VERSION == 0 \
2244 && UNICODE_DOT_DOT_VERSION == 1
2245# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2246
2247 /* And special case this on this Unicode version only, for the same
2248 * reaons the other two are special cased. They would cross the
2249 * 255/256 boundary which is forbidden under /l, and so the code
2250 * wouldn't catch that they are equivalent (which they are only in
2251 * this release) */
2252 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2253 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2254 {
2255 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2256 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2257 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2258 "resolved to \"\\x{0131}\".");
2259 goto return_dotless_i;
2260 }
2261#endif
2262
357aadde 2263 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2264 }
a0270393
KW
2265 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2266 return result;
2267 }
2268 else {
4a4088c4 2269 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2270 * character above the ASCII range, and the result should not
2271 * contain an ASCII character. */
a0270393
KW
2272
2273 UV original; /* To store the first code point of <p> */
2274
2275 /* Look at every character in the result; if any cross the
2276 * boundary, the whole thing is disallowed */
2277 U8* s = ustrp;
2278 U8* e = ustrp + *lenp;
2279 while (s < e) {
2280 if (isASCII(*s)) {
2281 /* Crossed, have to return the original */
2282 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2283
9fc2026f 2284 /* But in these instances, there is an alternative we can
1ca267a5 2285 * return that is valid */
0766489e
KW
2286 if (original == LATIN_SMALL_LETTER_SHARP_S
2287#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2288 || original == LATIN_CAPITAL_LETTER_SHARP_S
2289#endif
2290 ) {
1ca267a5
KW
2291 goto return_long_s;
2292 }
9fc2026f
KW
2293 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2294 goto return_ligature_st;
2295 }
74894415
KW
2296#if UNICODE_MAJOR_VERSION == 3 \
2297 && UNICODE_DOT_VERSION == 0 \
2298 && UNICODE_DOT_DOT_VERSION == 1
2299
2300 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2301 goto return_dotless_i;
2302 }
2303#endif
a0270393
KW
2304 Copy(p, ustrp, *lenp, char);
2305 return original;
2306 }
2307 s += UTF8SKIP(s);
2308 }
051a06d4 2309
a0270393
KW
2310 /* Here, no characters crossed, result is ok as-is */
2311 return result;
2312 }
051a06d4
KW
2313 }
2314
4a4088c4 2315 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2316 if (UTF8_IS_INVARIANT(result)) {
2317 *ustrp = (U8) result;
2318 *lenp = 1;
2319 }
2320 else {
62cb07ea
KW
2321 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2322 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2323 *lenp = 2;
2324 }
2325
051a06d4 2326 return result;
1ca267a5
KW
2327
2328 return_long_s:
2329 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2330 * folds to a string of two of these characters. By returning this
2331 * instead, then, e.g.,
2332 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2333 * works. */
2334
2335 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2336 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2337 ustrp, *lenp, U8);
2338 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2339
2340 return_ligature_st:
2341 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2342 * have the other one fold to it */
2343
2344 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2345 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2346 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2347
2348#if UNICODE_MAJOR_VERSION == 3 \
2349 && UNICODE_DOT_VERSION == 0 \
2350 && UNICODE_DOT_DOT_VERSION == 1
2351
2352 return_dotless_i:
2353 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2354 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2355 return LATIN_SMALL_LETTER_DOTLESS_I;
2356
2357#endif
2358
a0ed51b3
LW
2359}
2360
711a919c 2361/* Note:
f90a9a02 2362 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2363 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2364 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2365 */
c4a5db0c 2366
a0ed51b3 2367SV*
7fc63493 2368Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2369{
c4a5db0c
KW
2370 PERL_ARGS_ASSERT_SWASH_INIT;
2371
2372 /* Returns a copy of a swash initiated by the called function. This is the
2373 * public interface, and returning a copy prevents others from doing
2374 * mischief on the original */
2375
5d3d13d1 2376 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2377}
2378
2379SV*
5d3d13d1 2380Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2381{
2c1f00b9
YO
2382
2383 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2384 * use the following define */
2385
2386#define CORE_SWASH_INIT_RETURN(x) \
2387 PL_curpm= old_PL_curpm; \
2388 return x
2389
c4a5db0c 2390 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2391 * by calling utf8_heavy.pl in the general case. The returned value may be
2392 * the swash's inversion list instead if the input parameters allow it.
2393 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2394 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2395 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2396 *
2397 * This interface should only be used by functions that won't destroy or
2398 * adversely change the swash, as doing so affects all other uses of the
2399 * swash in the program; the general public should use 'Perl_swash_init'
2400 * instead.
2401 *
2402 * pkg is the name of the package that <name> should be in.
2403 * name is the name of the swash to find. Typically it is a Unicode
2404 * property name, including user-defined ones
2405 * listsv is a string to initialize the swash with. It must be of the form
2406 * documented as the subroutine return value in
2407 * L<perlunicode/User-Defined Character Properties>
2408 * minbits is the number of bits required to represent each data element.
2409 * It is '1' for binary properties.
2410 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2411 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2412 * flags_p if non-NULL is the address of various input and output flag bits
2413 * to the routine, as follows: ('I' means is input to the routine;
2414 * 'O' means output from the routine. Only flags marked O are
2415 * meaningful on return.)
2416 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2417 * came from a user-defined property. (I O)
5d3d13d1
KW
2418 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2419 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2420 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2421 * return of an inversion list instead of a swash hash if this routine
2422 * thinks that would result in faster execution of swash_fetch() later
2423 * on. (I)
9a53f6cf
KW
2424 *
2425 * Thus there are three possible inputs to find the swash: <name>,
2426 * <listsv>, and <invlist>. At least one must be specified. The result
2427 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2428 * actions can intersect, etc. what <name> gives. To avoid going out to
2429 * disk at all, <invlist> should specify completely what the swash should
2430 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2431 *
2432 * <invlist> is only valid for binary properties */
c4a5db0c 2433
2c1f00b9
YO
2434 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2435
c4a5db0c 2436 SV* retval = &PL_sv_undef;
83199d38 2437 HV* swash_hv = NULL;
87367d5f
KW
2438 const int invlist_swash_boundary =
2439 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2440 ? 512 /* Based on some benchmarking, but not extensive, see commit
2441 message */
2442 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2443
2444 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2445 assert(! invlist || minbits == 1);
2446
2c1f00b9
YO
2447 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2448 that triggered the swash init and the swash init perl logic itself.
2449 See perl #122747 */
2450
9a53f6cf
KW
2451 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2452 * so */
2453 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2454 dSP;
2455 const size_t pkg_len = strlen(pkg);
2456 const size_t name_len = strlen(name);
2457 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2458 SV* errsv_save;
2459 GV *method;
2460
2461 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2462
2463 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2464 ENTER;
69794297 2465 SAVEHINTS();
2782061f 2466 save_re_context();
650f067c
JL
2467 /* We might get here via a subroutine signature which uses a utf8
2468 * parameter name, at which point PL_subname will have been set
2469 * but not yet used. */
2470 save_item(PL_subname);
69794297
KW
2471 if (PL_parser && PL_parser->error_count)
2472 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2473 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2474 if (!method) { /* demand load UTF-8 */
69794297 2475 ENTER;
db2c6cb3
FC
2476 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2477 GvSV(PL_errgv) = NULL;
1a419e6b 2478#ifndef NO_TAINT_SUPPORT
69794297
KW
2479 /* It is assumed that callers of this routine are not passing in
2480 * any user derived data. */
2782061f
DM
2481 /* Need to do this after save_re_context() as it will set
2482 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2483 * in Perl_magic_get). Even line to create errsv_save can turn on
2484 * PL_tainted. */
284167a5
SM
2485 SAVEBOOL(TAINT_get);
2486 TAINT_NOT;
2487#endif
69794297
KW
2488 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2489 NULL);
eed484f9 2490 {
db2c6cb3
FC
2491 /* Not ERRSV, as there is no need to vivify a scalar we are
2492 about to discard. */
2493 SV * const errsv = GvSV(PL_errgv);
2494 if (!SvTRUE(errsv)) {
2495 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2496 SvREFCNT_dec(errsv);
2497 }
eed484f9 2498 }
69794297
KW
2499 LEAVE;
2500 }
2501 SPAGAIN;
2502 PUSHMARK(SP);
2503 EXTEND(SP,5);
2504 mPUSHp(pkg, pkg_len);
2505 mPUSHp(name, name_len);
2506 PUSHs(listsv);
2507 mPUSHi(minbits);
2508 mPUSHi(none);
2509 PUTBACK;
db2c6cb3
FC
2510 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2511 GvSV(PL_errgv) = NULL;
69794297
KW
2512 /* If we already have a pointer to the method, no need to use
2513 * call_method() to repeat the lookup. */
c41800a8
KW
2514 if (method
2515 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2516 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2517 {
2518 retval = *PL_stack_sp--;
2519 SvREFCNT_inc(retval);
2520 }
eed484f9 2521 {
db2c6cb3
FC
2522 /* Not ERRSV. See above. */
2523 SV * const errsv = GvSV(PL_errgv);
2524 if (!SvTRUE(errsv)) {
2525 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2526 SvREFCNT_dec(errsv);
2527 }
eed484f9 2528 }
ce3b816e 2529 LEAVE;
69794297
KW
2530 POPSTACK;
2531 if (IN_PERL_COMPILETIME) {
2532 CopHINTS_set(PL_curcop, PL_hints);
2533 }
2534 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2535 if (SvPOK(retval))
2536
2537 /* If caller wants to handle missing properties, let them */
5d3d13d1 2538 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2539 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2540 }
2541 Perl_croak(aTHX_
2542 "Can't find Unicode property definition \"%"SVf"\"",
2543 SVfARG(retval));
a25b5927 2544 NOT_REACHED; /* NOTREACHED */
69794297 2545 }
9a53f6cf 2546 } /* End of calling the module to find the swash */
36eb48b4 2547
83199d38
KW
2548 /* If this operation fetched a swash, and we will need it later, get it */
2549 if (retval != &PL_sv_undef
2550 && (minbits == 1 || (flags_p
2551 && ! (*flags_p
2552 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2553 {
2554 swash_hv = MUTABLE_HV(SvRV(retval));
2555
2556 /* If we don't already know that there is a user-defined component to
2557 * this swash, and the user has indicated they wish to know if there is
2558 * one (by passing <flags_p>), find out */
2559 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2560 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2561 if (user_defined && SvUV(*user_defined)) {
2562 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2563 }
2564 }
2565 }
2566
36eb48b4
KW
2567 /* Make sure there is an inversion list for binary properties */
2568 if (minbits == 1) {
2569 SV** swash_invlistsvp = NULL;
2570 SV* swash_invlist = NULL;
9a53f6cf 2571 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2572 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2573 an unclaimed reference count */
36eb48b4 2574
9a53f6cf 2575 /* If this operation fetched a swash, get its already existing
83199d38 2576 * inversion list, or create one for it */
36eb48b4 2577
83199d38 2578 if (swash_hv) {
5c9f4bd2 2579 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2580 if (swash_invlistsvp) {
2581 swash_invlist = *swash_invlistsvp;
2582 invlist_in_swash_is_valid = TRUE;
2583 }
2584 else {
36eb48b4 2585 swash_invlist = _swash_to_invlist(retval);
02c85471 2586 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2587 }
2588 }
2589
2590 /* If an inversion list was passed in, have to include it */
2591 if (invlist) {
2592
2593 /* Any fetched swash will by now have an inversion list in it;
2594 * otherwise <swash_invlist> will be NULL, indicating that we
2595 * didn't fetch a swash */
2596 if (swash_invlist) {
2597
2598 /* Add the passed-in inversion list, which invalidates the one
2599 * already stored in the swash */
2600 invlist_in_swash_is_valid = FALSE;
2601 _invlist_union(invlist, swash_invlist, &swash_invlist);
2602 }
2603 else {
2604
87367d5f
KW
2605 /* Here, there is no swash already. Set up a minimal one, if
2606 * we are going to return a swash */
2607 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2608 swash_hv = newHV();
4aca0fe6 2609 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2610 }
9a53f6cf
KW
2611 swash_invlist = invlist;
2612 }
9a53f6cf
KW
2613 }
2614
2615 /* Here, we have computed the union of all the passed-in data. It may
2616 * be that there was an inversion list in the swash which didn't get
538e84ed 2617 * touched; otherwise save the computed one */
87367d5f
KW
2618 if (! invlist_in_swash_is_valid
2619 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2620 {
5c9f4bd2 2621 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2622 {
2623 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2624 }
cc34d8c5
FC
2625 /* We just stole a reference count. */
2626 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2627 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2628 }
87367d5f 2629
dbfdbd26
KW
2630 SvREADONLY_on(swash_invlist);
2631
c41800a8 2632 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2633 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2634 SvREFCNT_dec(retval);
02c85471
FC
2635 if (!swash_invlist_unclaimed)
2636 SvREFCNT_inc_simple_void_NN(swash_invlist);
2637 retval = newRV_noinc(swash_invlist);
87367d5f 2638 }
36eb48b4
KW
2639 }
2640
2c1f00b9
YO
2641 CORE_SWASH_INIT_RETURN(retval);
2642#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2643}
2644
035d37be
JH
2645
2646/* This API is wrong for special case conversions since we may need to
2647 * return several Unicode characters for a single Unicode character
2648 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2649 * the lower-level routine, and it is similarly broken for returning
38684baa
KW
2650 * multiple values. --jhi
2651 * For those, you should use to_utf8_case() instead */
b0e3252e 2652/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2653
2654/* Note:
2655 * Returns the value of property/mapping C<swash> for the first character
2656 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2657 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2658 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2659 *
2660 * A "swash" is a hash which contains initially the keys/values set up by
2661 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2662 * property for all possible code points. Things are stored in a compact form
2663 * (see utf8_heavy.pl) so that calculation is required to find the actual
2664 * property value for a given code point. As code points are looked up, new
2665 * key/value pairs are added to the hash, so that the calculation doesn't have
2666 * to ever be re-done. Further, each calculation is done, not just for the
2667 * desired one, but for a whole block of code points adjacent to that one.
2668 * For binary properties on ASCII machines, the block is usually for 64 code
2669 * points, starting with a code point evenly divisible by 64. Thus if the
2670 * property value for code point 257 is requested, the code goes out and
2671 * calculates the property values for all 64 code points between 256 and 319,
2672 * and stores these as a single 64-bit long bit vector, called a "swatch",
2673 * under the key for code point 256. The key is the UTF-8 encoding for code
2674 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2675 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2676 * for code point 258 is then requested, this code realizes that it would be
2677 * stored under the key for 256, and would find that value and extract the
2678 * relevant bit, offset from 256.
2679 *
2680 * Non-binary properties are stored in as many bits as necessary to represent
2681 * their values (32 currently, though the code is more general than that), not
2682 * as single bits, but the principal is the same: the value for each key is a
2683 * vector that encompasses the property values for all code points whose UTF-8
2684 * representations are represented by the key. That is, for all code points
2685 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2686 * bytes of that.
680c470c 2687 */
a0ed51b3 2688UV
680c470c 2689Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2690{
ef8f7699 2691 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2692 U32 klen;
2693 U32 off;
9b56a019 2694 STRLEN slen = 0;
7d85a32c 2695 STRLEN needents;
cfd0369c 2696 const U8 *tmps = NULL;
979f2922 2697 SV *swatch;
08fb1ac5 2698 const U8 c = *ptr;
3568d838 2699
7918f24d
NC
2700 PERL_ARGS_ASSERT_SWASH_FETCH;
2701
87367d5f
KW
2702 /* If it really isn't a hash, it isn't really swash; must be an inversion
2703 * list */
2704 if (SvTYPE(hv) != SVt_PVHV) {
2705 return _invlist_contains_cp((SV*)hv,
2706 (do_utf8)
2707 ? valid_utf8_to_uvchr(ptr, NULL)
2708 : c);
2709 }
2710
08fb1ac5
KW
2711 /* We store the values in a "swatch" which is a vec() value in a swash
2712 * hash. Code points 0-255 are a single vec() stored with key length
2713 * (klen) 0. All other code points have a UTF-8 representation
2714 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2715 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2716 * length for them is the length of the encoded char - 1. ptr[klen] is the
2717 * final byte in the sequence representing the character */
2718 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2719 klen = 0;
2720 needents = 256;
2721 off = c;
3568d838 2722 }
08fb1ac5
KW
2723 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2724 klen = 0;
2725 needents = 256;
2726 off = TWO_BYTE_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2727 }
2728 else {
08fb1ac5
KW
2729 klen = UTF8SKIP(ptr) - 1;
2730
2731 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2732 * the vec is the final byte in the sequence. (In EBCDIC this is
2733 * converted to I8 to get consecutive values.) To help you visualize
2734 * all this:
2735 * Straight 1047 After final byte
2736 * UTF-8 UTF-EBCDIC I8 transform
2737 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2738 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2739 * ...
2740 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2741 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2742 * ...
2743 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2744 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2745 * ...
2746 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2747 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2748 * ...
2749 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2750 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2751 *
2752 * (There are no discontinuities in the elided (...) entries.)
2753 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2754 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2755 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2756 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2757 * index into the vec() swatch (after subtracting 0x80, which we
2758 * actually do with an '&').
2759 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2760 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2761 * dicontinuities which go away by transforming it into I8, and we
2762 * effectively subtract 0xA0 to get the index. */
979f2922 2763 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2764 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2765 }
7d85a32c 2766
a0ed51b3 2767 /*
4a4088c4 2768 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2769 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2770 * it's nothing to sniff at.) Pity we usually come through at least
2771 * two function calls to get here...
2772 *
2773 * NB: this code assumes that swatches are never modified, once generated!
2774 */
2775
3568d838 2776 if (hv == PL_last_swash_hv &&
a0ed51b3 2777 klen == PL_last_swash_klen &&
27da23d5 2778 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2779 {
2780 tmps = PL_last_swash_tmps;
2781 slen = PL_last_swash_slen;
2782 }
2783 else {
2784 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2785 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2786
b0e3252e 2787 /* If not cached, generate it via swatch_get */
979f2922 2788 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2789 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2790 {
2791 if (klen) {
2792 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2793 swatch = swatch_get(swash,
2794 code_point & ~((UV)needents - 1),
2795 needents);
2796 }
2797 else { /* For the first 256 code points, the swatch has a key of
2798 length 0 */
2799 swatch = swatch_get(swash, 0, needents);
2800 }
979f2922 2801
923e4eb5 2802 if (IN_PERL_COMPILETIME)
623e6609 2803 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2804
979f2922 2805 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2806
979f2922
ST
2807 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2808 || (slen << 3) < needents)
5637ef5b
NC
2809 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2810 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2811 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2812 }
2813
2814 PL_last_swash_hv = hv;
16d8f38a 2815 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2816 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2817 /* FIXME change interpvar.h? */
2818 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2819 PL_last_swash_slen = slen;
2820 if (klen)
2821 Copy(ptr, PL_last_swash_key, klen, U8);
2822 }
2823
9faf8d75 2824 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2825 case 1:
e7aca353 2826 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2827 case 8:
e7aca353 2828 return ((UV) tmps[off]);
a0ed51b3
LW
2829 case 16:
2830 off <<= 1;
e7aca353
JH
2831 return
2832 ((UV) tmps[off ] << 8) +
2833 ((UV) tmps[off + 1]);
a0ed51b3
LW
2834 case 32:
2835 off <<= 2;
e7aca353
JH
2836 return
2837 ((UV) tmps[off ] << 24) +
2838 ((UV) tmps[off + 1] << 16) +
2839 ((UV) tmps[off + 2] << 8) +
2840 ((UV) tmps[off + 3]);
a0ed51b3 2841 }
5637ef5b
NC
2842 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2843 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2844 NORETURN_FUNCTION_END;
a0ed51b3 2845}
2b9d42f0 2846
319009ee
KW
2847/* Read a single line of the main body of the swash input text. These are of
2848 * the form:
2849 * 0053 0056 0073
2850 * where each number is hex. The first two numbers form the minimum and
2851 * maximum of a range, and the third is the value associated with the range.
2852 * Not all swashes should have a third number
2853 *
2854 * On input: l points to the beginning of the line to be examined; it points
2855 * to somewhere in the string of the whole input text, and is
2856 * terminated by a \n or the null string terminator.
2857 * lend points to the null terminator of that string
2858 * wants_value is non-zero if the swash expects a third number
2859 * typestr is the name of the swash's mapping, like 'ToLower'
2860 * On output: *min, *max, and *val are set to the values read from the line.
2861 * returns a pointer just beyond the line examined. If there was no
2862 * valid min number on the line, returns lend+1
2863 */
2864
2865STATIC U8*
2866S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2867 const bool wants_value, const U8* const typestr)
2868{
2869 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2870 STRLEN numlen; /* Length of the number */
02470786
KW
2871 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2872 | PERL_SCAN_DISALLOW_PREFIX
2873 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2874
2875 /* nl points to the next \n in the scan */
2876 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2877
95543e92
KW
2878 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2879
319009ee
KW
2880 /* Get the first number on the line: the range minimum */
2881 numlen = lend - l;
2882 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2883 *max = *min; /* So can never return without setting max */
319009ee
KW
2884 if (numlen) /* If found a hex number, position past it */
2885 l += numlen;
2886 else if (nl) { /* Else, go handle next line, if any */
2887 return nl + 1; /* 1 is length of "\n" */
2888 }
2889 else { /* Else, no next line */
2890 return lend + 1; /* to LIST's end at which \n is not found */
2891 }
2892
2893 /* The max range value follows, separated by a BLANK */
2894 if (isBLANK(*l)) {
2895 ++l;
02470786
KW
2896 flags = PERL_SCAN_SILENT_ILLDIGIT
2897 | PERL_SCAN_DISALLOW_PREFIX
2898 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2899 numlen = lend - l;
2900 *max = grok_hex((char *)l, &numlen, &flags, NULL);
2901 if (numlen)
2902 l += numlen;
2903 else /* If no value here, it is a single element range */
2904 *max = *min;
2905
2906 /* Non-binary tables have a third entry: what the first element of the
24303724 2907 * range maps to. The map for those currently read here is in hex */
319009ee
KW
2908 if (wants_value) {
2909 if (isBLANK(*l)) {
2910 ++l;
f2a7d0fc
KW
2911 flags = PERL_SCAN_SILENT_ILLDIGIT
2912 | PERL_SCAN_DISALLOW_PREFIX
2913 | PERL_SCAN_SILENT_NON_PORTABLE;
2914 numlen = lend - l;
2915 *val = grok_hex((char *)l, &numlen, &flags, NULL);
2916 if (numlen)
2917 l += numlen;
2918 else
2919 *val = 0;
319009ee
KW
2920 }
2921 else {
2922 *val = 0;
2923 if (typeto) {
dcbac5bb 2924 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2925 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
2926 typestr, l);
2927 }
2928 }
2929 }
2930 else
2931 *val = 0; /* bits == 1, then any val should be ignored */
2932 }
2933 else { /* Nothing following range min, should be single element with no
2934 mapping expected */
319009ee
KW
2935 if (wants_value) {
2936 *val = 0;
2937 if (typeto) {
dcbac5bb 2938 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
2939 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
2940 }
2941 }
2942 else
2943 *val = 0; /* bits == 1, then val should be ignored */
2944 }
2945
2946 /* Position to next line if any, or EOF */
2947 if (nl)
2948 l = nl + 1;
2949 else
2950 l = lend;
2951
2952 return l;
2953}
2954
979f2922
ST
2955/* Note:
2956 * Returns a swatch (a bit vector string) for a code point sequence
2957 * that starts from the value C<start> and comprises the number C<span>.
2958 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
2959 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
2960 */
2961STATIC SV*
b0e3252e 2962S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
2963{
2964 SV *swatch;
77f9f126 2965 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 2966 STRLEN lcur, xcur, scur;
ef8f7699 2967 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 2968 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 2969
88d45d28
KW
2970 SV** listsvp = NULL; /* The string containing the main body of the table */
2971 SV** extssvp = NULL;
2972 SV** invert_it_svp = NULL;
2973 U8* typestr = NULL;
786861f5
KW
2974 STRLEN bits;
2975 STRLEN octets; /* if bits == 1, then octets == 0 */
2976 UV none;
2977 UV end = start + span;
972dd592 2978
36eb48b4 2979 if (invlistsvp == NULL) {
786861f5
KW
2980 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
2981 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
2982 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
2983 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
2984 listsvp = hv_fetchs(hv, "LIST", FALSE);
2985 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
2986
2987 bits = SvUV(*bitssvp);
2988 none = SvUV(*nonesvp);
2989 typestr = (U8*)SvPV_nolen(*typesvp);
2990 }
36eb48b4
KW
2991 else {
2992 bits = 1;
2993 none = 0;
2994 }
786861f5 2995 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 2996
b0e3252e 2997 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 2998
979f2922 2999 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3000 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3001 (UV)bits);
979f2922
ST
3002 }
3003
84ea5ef6
KW
3004 /* If overflowed, use the max possible */
3005 if (end < start) {
3006 end = UV_MAX;
3007 span = end - start;
3008 }
3009
979f2922 3010 /* create and initialize $swatch */
979f2922 3011 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3012 swatch = newSV(scur);
3013 SvPOK_on(swatch);
979f2922
ST
3014 s = (U8*)SvPVX(swatch);
3015 if (octets && none) {
0bd48802 3016 const U8* const e = s + scur;
979f2922
ST
3017 while (s < e) {
3018 if (bits == 8)
3019 *s++ = (U8)(none & 0xff);
3020 else if (bits == 16) {
3021 *s++ = (U8)((none >> 8) & 0xff);
3022 *s++ = (U8)( none & 0xff);
3023 }
3024 else if (bits == 32) {
3025 *s++ = (U8)((none >> 24) & 0xff);
3026 *s++ = (U8)((none >> 16) & 0xff);
3027 *s++ = (U8)((none >> 8) & 0xff);
3028 *s++ = (U8)( none & 0xff);
3029 }
3030 }
3031 *s = '\0';
3032 }
3033 else {
3034 (void)memzero((U8*)s, scur + 1);
3035 }
3036 SvCUR_set(swatch, scur);
3037 s = (U8*)SvPVX(swatch);
3038
36eb48b4
KW
3039 if (invlistsvp) { /* If has an inversion list set up use that */
3040 _invlist_populate_swatch(*invlistsvp, start, end, s);
3041 return swatch;
3042 }
3043
3044 /* read $swash->{LIST} */
979f2922
ST
3045 l = (U8*)SvPV(*listsvp, lcur);
3046 lend = l + lcur;
3047 while (l < lend) {
8ed25d53 3048 UV min, max, val, upper;
95543e92
KW
3049 l = swash_scan_list_line(l, lend, &min, &max, &val,
3050 cBOOL(octets), typestr);
319009ee 3051 if (l > lend) {
979f2922
ST
3052 break;
3053 }
3054
972dd592 3055 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3056 if (max < start)
3057 continue;
3058
8ed25d53
KW
3059 /* <end> is generally 1 beyond where we want to set things, but at the
3060 * platform's infinity, where we can't go any higher, we want to
3061 * include the code point at <end> */
3062 upper = (max < end)
3063 ? max
3064 : (max != UV_MAX || end != UV_MAX)
3065 ? end - 1
3066 : end;
3067
979f2922 3068 if (octets) {
35da51f7 3069 UV key;
979f2922
ST
3070 if (min < start) {
3071 if (!none || val < none) {
3072 val += start - min;
3073 }
3074 min = start;
3075 }
8ed25d53 3076 for (key = min; key <= upper; key++) {
979f2922 3077 STRLEN offset;
979f2922
ST
3078 /* offset must be non-negative (start <= min <= key < end) */
3079 offset = octets * (key - start);
3080 if (bits == 8)
3081 s[offset] = (U8)(val & 0xff);
3082 else if (bits == 16) {
3083 s[offset ] = (U8)((val >> 8) & 0xff);
3084 s[offset + 1] = (U8)( val & 0xff);
3085 }
3086 else if (bits == 32) {
3087 s[offset ] = (U8)((val >> 24) & 0xff);
3088 s[offset + 1] = (U8)((val >> 16) & 0xff);
3089 s[offset + 2] = (U8)((val >> 8) & 0xff);
3090 s[offset + 3] = (U8)( val & 0xff);
3091 }
3092
3093 if (!none || val < none)
3094 ++val;
3095 }
3096 }
711a919c 3097 else { /* bits == 1, then val should be ignored */
35da51f7 3098 UV key;
979f2922
ST
3099 if (min < start)
3100 min = start;
6cb05c12 3101
8ed25d53 3102 for (key = min; key <= upper; key++) {
0bd48802 3103 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3104 s[offset >> 3] |= 1 << (offset & 7);
3105 }
3106 }
3107 } /* while */
979f2922 3108
9479a769 3109 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3110 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3111
3112 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3113 * be 0, and their inversion should also be 0, as we don't succeed any
3114 * Unicode property matches for non-Unicode code points */
3115 if (start <= PERL_UNICODE_MAX) {
3116
3117 /* The code below assumes that we never cross the
3118 * Unicode/above-Unicode boundary in a range, as otherwise we would
3119 * have to figure out where to stop flipping the bits. Since this
3120 * boundary is divisible by a large power of 2, and swatches comes
3121 * in small powers of 2, this should be a valid assumption */
3122 assert(start + span - 1 <= PERL_UNICODE_MAX);
3123
507a8485
KW
3124 send = s + scur;
3125 while (s < send) {
3126 *s = ~(*s);
3127 s++;
3128 }
0bda3001 3129 }
77f9f126
KW
3130 }
3131
d73c39c5
KW
3132 /* read $swash->{EXTRAS}
3133 * This code also copied to swash_to_invlist() below */
979f2922
ST
3134 x = (U8*)SvPV(*extssvp, xcur);
3135 xend = x + xcur;
3136 while (x < xend) {
3137 STRLEN namelen;
3138 U8 *namestr;
3139 SV** othersvp;
3140 HV* otherhv;
3141 STRLEN otherbits;
3142 SV **otherbitssvp, *other;
711a919c 3143 U8 *s, *o, *nl;
979f2922
ST
3144 STRLEN slen, olen;
3145
35da51f7 3146 const U8 opc = *x++;
979f2922
ST
3147 if (opc == '\n')
3148 continue;
3149
3150 nl = (U8*)memchr(x, '\n', xend - x);
3151
3152 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3153 if (nl) {
3154 x = nl + 1; /* 1 is length of "\n" */
3155 continue;
3156 }
3157 else {
3158 x = xend; /* to EXTRAS' end at which \n is not found */
3159 break;
3160 }
3161 }
3162
3163 namestr = x;
3164 if (nl) {
3165 namelen = nl - namestr;
3166 x = nl + 1;
3167 }
3168 else {
3169 namelen = xend - namestr;
3170 x = xend;
3171 }
3172
3173 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3174 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3175 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3176 otherbits = (STRLEN)SvUV(*otherbitssvp);
3177 if (bits < otherbits)
5637ef5b
NC
3178 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3179 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3180
3181 /* The "other" swatch must be destroyed after. */
b0e3252e 3182 other = swatch_get(*othersvp, start, span);
979f2922
ST
3183 o = (U8*)SvPV(other, olen);
3184
3185 if (!olen)
b0e3252e 3186 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3187
3188 s = (U8*)SvPV(swatch, slen);
3189 if (bits == 1 && otherbits == 1) {
3190 if (slen != olen)
5637ef5b
NC
3191 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3192 "mismatch, slen=%"UVuf", olen=%"UVuf,
3193 (UV)slen, (UV)olen);
979f2922
ST
3194
3195 switch (opc) {
3196 case '+':
3197 while (slen--)
3198 *s++ |= *o++;
3199 break;
3200 case '!':
3201 while (slen--)
3202 *s++ |= ~*o++;
3203 break;
3204 case '-':
3205 while (slen--)
3206 *s++ &= ~*o++;
3207 break;
3208 case '&':
3209 while (slen--)
3210 *s++ &= *o++;
3211 break;
3212 default:
3213 break;
3214 }
3215 }
711a919c 3216 else {
979f2922
ST
3217 STRLEN otheroctets = otherbits >> 3;
3218 STRLEN offset = 0;
35da51f7 3219 U8* const send = s + slen;
979f2922
ST
3220
3221 while (s < send) {
3222 UV otherval = 0;
3223
3224 if (otherbits == 1) {
3225 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3226 ++offset;
3227 }
3228 else {
3229 STRLEN vlen = otheroctets;
3230 otherval = *o++;
3231 while (--vlen) {
3232 otherval <<= 8;
3233 otherval |= *o++;
3234 }
3235 }
3236
711a919c 3237 if (opc == '+' && otherval)
6f207bd3 3238 NOOP; /* replace with otherval */
979f2922
ST
3239 else if (opc == '!' && !otherval)
3240 otherval = 1;
3241 else if (opc == '-' && otherval)
3242 otherval = 0;
3243 else if (opc == '&' && !otherval)
3244 otherval = 0;
3245 else {
711a919c 3246 s += octets; /* no replacement */
979f2922
ST
3247 continue;
3248 }
3249
3250 if (bits == 8)
3251 *s++ = (U8)( otherval & 0xff);
3252 else if (bits == 16) {
3253 *s++ = (U8)((otherval >> 8) & 0xff);
3254 *s++ = (U8)( otherval & 0xff);
3255 }
3256 else if (bits == 32) {
3257 *s++ = (U8)((otherval >> 24) & 0xff);
3258 *s++ = (U8)((otherval >> 16) & 0xff);
3259 *s++ = (U8)((otherval >> 8) & 0xff);
3260 *s++ = (U8)( otherval & 0xff);
3261 }
3262 }
3263 }
3264 sv_free(other); /* through with it! */
3265 } /* while */
3266 return swatch;
3267}
3268
064c021d 3269HV*
4c2e1131 3270Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3271{
3272
79a2a0e8 3273 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3274 * Can't be used on a property that is subject to user override, as it
3275 * relies on the value of SPECIALS in the swash which would be set by
3276 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3277 * for overridden properties
064c021d
KW
3278 *
3279 * Returns a hash which is the inversion and closure of a swash mapping.
3280 * For example, consider the input lines:
3281 * 004B 006B
3282 * 004C 006C
3283 * 212A 006B
3284 *
4a4088c4 3285 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3286 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3287 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3288 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3289 *
538e84ed
KW
3290 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3291 * keys are only code points that are folded-to, so it isn't a full closure.
3292 *
064c021d
KW
3293 * Essentially, for any code point, it gives all the code points that map to
3294 * it, or the list of 'froms' for that point.
3295 *
5662e334
KW
3296 * Currently it ignores any additions or deletions from other swashes,
3297 * looking at just the main body of the swash, and if there are SPECIALS
3298 * in the swash, at that hash
3299 *
3300 * The specials hash can be extra code points, and most likely consists of
3301 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3302 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3303 * However consider this possible input in the specials hash:
3304 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3305 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3306 *
3307 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3308 * currently handle. But it also means that FB05 and FB06 are equivalent in
3309 * a 1-1 mapping which we should handle, and this relationship may not be in
3310 * the main table. Therefore this function examines all the multi-char
74894415
KW
3311 * sequences and adds the 1-1 mappings that come out of that.
3312 *
3313 * XXX This function was originally intended to be multipurpose, but its
3314 * only use is quite likely to remain for constructing the inversion of
3315 * the CaseFolding (//i) property. If it were more general purpose for
3316 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3317 * because certain folds are prohibited under /iaa and /il. As an example,
3318 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3319 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3320 * prohibited, so we would not figure out that they fold to each other.
3321 * Code could be written to automatically figure this out, similar to the
3322 * code that does this for multi-character folds, but this is the only case
3323 * where something like this is ever likely to happen, as all the single
3324 * char folds to The 0-255 range are now quite settled. Instead there is a
3325 * little special code that is compiled only for this Unicode version. This
3326 * is smaller and didn't require much coding time to do. But this makes
3327 * this routine strongly tied to being used just for CaseFolding. If ever
3328 * it should be generalized, this would have to be fixed */
064c021d
KW
3329
3330 U8 *l, *lend;
3331 STRLEN lcur;
3332 HV *const hv = MUTABLE_HV(SvRV(swash));
3333
923b6d4e
KW
3334 /* The string containing the main body of the table. This will have its
3335 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3336 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3337
3338 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3339 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3340 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3341 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3342 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3343 const STRLEN bits = SvUV(*bitssvp);
3344 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3345 const UV none = SvUV(*nonesvp);
5662e334 3346 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3347
3348 HV* ret = newHV();
3349
3350 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3351
3352 /* Must have at least 8 bits to get the mappings */
3353 if (bits != 8 && bits != 16 && bits != 32) {
3354 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3355 (UV)bits);
3356 }
3357
5662e334
KW
3358 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3359 mapping to more than one character */
3360
3361 /* Construct an inverse mapping hash for the specials */
3362 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3363 HV * specials_inverse = newHV();
3364 char *char_from; /* the lhs of the map */
3365 I32 from_len; /* its byte length */
3366 char *char_to; /* the rhs of the map */
3367 I32 to_len; /* its byte length */
3368 SV *sv_to; /* and in a sv */
3369 AV* from_list; /* list of things that map to each 'to' */
3370
3371 hv_iterinit(specials_hv);
3372
4a4088c4
KW
3373 /* The keys are the characters (in UTF-8) that map to the corresponding
3374 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3375 * list. */
3376 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3377 SV** listp;
3378 if (! SvPOK(sv_to)) {
5637ef5b
NC
3379 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3380 "unexpectedly is not a string, flags=%lu",
3381 (unsigned long)SvFLAGS(sv_to));
5662e334 3382 }
4b88fb76 3383 /*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
3384
3385 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3386 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3387 * it. Those strings are all one character long */
3388 if ((listp = hv_fetch(specials_inverse,
3389 SvPVX(sv_to),
3390 SvCUR(sv_to), 0)))
3391 {
3392 from_list = (AV*) *listp;
3393 }
3394 else { /* No entry yet for it: create one */
3395 from_list = newAV();
3396 if (! hv_store(specials_inverse,
3397 SvPVX(sv_to),
3398 SvCUR(sv_to),
3399 (SV*) from_list, 0))
3400 {
3401 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3402 }
3403 }
3404
3405 /* Here have the list associated with this 'to' (perhaps newly
3406 * created and empty). Just add to it. Note that we ASSUME that
3407 * the input is guaranteed to not have duplications, so we don't
3408 * check for that. Duplications just slow down execution time. */
3409 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3410 }
3411
3412 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3413 * it looking for cases like the FB05/FB06 examples above. There would
3414 * be an entry in the hash like
3415 * 'st' => [ FB05, FB06 ]
3416 * In this example we will create two lists that get stored in the
3417 * returned hash, 'ret':
3418 * FB05 => [ FB05, FB06 ]
3419 * FB06 => [ FB05, FB06 ]
3420 *
3421 * Note that there is nothing to do if the array only has one element.
3422 * (In the normal 1-1 case handled below, we don't have to worry about
3423 * two lists, as everything gets tied to the single list that is
3424 * generated for the single character 'to'. But here, we are omitting
3425 * that list, ('st' in the example), so must have multiple lists.) */
3426 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3427 &char_to, &to_len)))
3428 {
b9f2b683 3429 if (av_tindex(from_list) > 0) {
c70927a6 3430 SSize_t i;
5662e334
KW
3431
3432 /* We iterate over all combinations of i,j to place each code
3433 * point on each list */
b9f2b683 3434 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3435 SSize_t j;
5662e334
KW
3436 AV* i_list = newAV();
3437 SV** entryp = av_fetch(from_list, i, FALSE);
3438 if (entryp == NULL) {
3439 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3440 }
3441 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3442 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3443 }
3444 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3445 (SV*) i_list, FALSE))
3446 {
3447 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3448 }
3449
538e84ed 3450 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
b9f2b683 3451 for (j = 0; j <= av_tindex(from_list); j++) {
5662e334
KW
3452 entryp = av_fetch(from_list, j, FALSE);
3453 if (entryp == NULL) {
3454 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3455 }
3456
3457 /* When i==j this adds itself to the list */
4b88fb76
KW
3458 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3459 (U8*) SvPVX(*entryp),
3460 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3461 0)));
4637d003 3462 /*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
3463 }
3464 }
3465 }
3466 }
3467 SvREFCNT_dec(specials_inverse); /* done with it */
3468 } /* End of specials */
3469
064c021d 3470 /* read $swash->{LIST} */
74894415
KW
3471
3472#if UNICODE_MAJOR_VERSION == 3 \
3473 && UNICODE_DOT_VERSION == 0 \
3474 && UNICODE_DOT_DOT_VERSION == 1
3475
3476 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3477 * rule so that things work under /iaa and /il */
3478
3479 SV * mod_listsv = sv_mortalcopy(*listsvp);
3480 sv_catpv(mod_listsv, "130\t130\t131\n");
3481 l = (U8*)SvPV(mod_listsv, lcur);
3482
3483#else
3484
064c021d 3485 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3486
3487#endif
3488
064c021d
KW
3489 lend = l + lcur;
3490
3491 /* Go through each input line */
3492 while (l < lend) {
3493 UV min, max, val;
3494 UV inverse;
95543e92
KW
3495 l = swash_scan_list_line(l, lend, &min, &max, &val,
3496 cBOOL(octets), typestr);
064c021d
KW
3497 if (l > lend) {
3498 break;
3499 }
3500
3501 /* Each element in the range is to be inverted */
3502 for (inverse = min; inverse <= max; inverse++) {
3503 AV* list;
064c021d
KW
3504 SV** listp;
3505 IV i;
3506 bool found_key = FALSE;
5662e334 3507 bool found_inverse = FALSE;
064c021d
KW
3508
3509 /* The key is the inverse mapping */
3510 char key[UTF8_MAXBYTES+1];
c80e42f3 3511 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3512 STRLEN key_len = key_end - key;
3513
064c021d
KW
3514 /* Get the list for the map */
3515 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3516 list = (AV*) *listp;
3517 }
3518 else { /* No entry yet for it: create one */
3519 list = newAV();
3520 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3521 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3522 }
3523 }
3524
5662e334
KW
3525 /* Look through list to see if this inverse mapping already is
3526 * listed, or if there is a mapping to itself already */
b9f2b683 3527 for (i = 0; i <= av_tindex(list); i++) {
064c021d
KW
3528 SV** entryp = av_fetch(list, i, FALSE);
3529 SV* entry;
414db8a1 3530 UV uv;
064c021d
KW
3531 if (entryp == NULL) {
3532 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3533 }
3534 entry = *entryp;
414db8a1
DM
3535 uv = SvUV(entry);
3536 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3537 if (uv == val) {
064c021d 3538 found_key = TRUE;
5662e334 3539 }
414db8a1 3540 if (uv == inverse) {
5662e334
KW
3541 found_inverse = TRUE;
3542 }
3543
3544 /* No need to continue searching if found everything we are
3545 * looking for */
3546 if (found_key && found_inverse) {
064c021d
KW
3547 break;
3548 }
3549 }
56ca34ca
KW
3550
3551 /* Make sure there is a mapping to itself on the list */
064c021d 3552 if (! found_key) {
d397ff6a 3553 av_push(list, newSVuv(val));
4637d003 3554 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3555 }
3556
3557
3558 /* Simply add the value to the list */
5662e334
KW
3559 if (! found_inverse) {
3560 av_push(list, newSVuv(inverse));
4637d003 3561 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3562 }
064c021d 3563
b0e3252e 3564 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3565 * range. That makes more compact tables possible. You can
3566 * express the capitalization, for example, of all consecutive
3567 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3568 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3569 * and it's not documented; it appears to be used only in
b0e3252e 3570 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3571 * in case */
064c021d
KW
3572 if (!none || val < none) {
3573 ++val;
3574 }
3575 }
3576 }
3577
3578 return ret;
3579}
3580
a25abddc 3581SV*
d764b54e
KW
3582Perl__swash_to_invlist(pTHX_ SV* const swash)
3583{
3584
ed92f1b3
KW
3585 /* Subject to change or removal. For use only in one place in regcomp.c.
3586 * Ownership is given to one reference count in the returned SV* */
d764b54e
KW
3587
3588 U8 *l, *lend;
3589 char *loc;
3590 STRLEN lcur;
3591 HV *const hv = MUTABLE_HV(SvRV(swash));
3592 UV elements = 0; /* Number of elements in the inversion list */
b443038a 3593 U8 empty[] = "";
923b6d4e
KW
3594 SV** listsvp;
3595 SV** typesvp;
3596 SV** bitssvp;
3597 SV** extssvp;
3598 SV** invert_it_svp;
d764b54e 3599
923b6d4e
KW
3600 U8* typestr;
3601 STRLEN bits;
3602 STRLEN octets; /* if bits == 1, then octets == 0 */
d73c39c5
KW
3603 U8 *x, *xend;
3604 STRLEN xcur;
d764b54e 3605
a25abddc 3606 SV* invlist;
d764b54e 3607
b81740c0
KW
3608 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
3609
923b6d4e
KW
3610 /* If not a hash, it must be the swash's inversion list instead */
3611 if (SvTYPE(hv) != SVt_PVHV) {
ed92f1b3 3612 return SvREFCNT_inc_simple_NN((SV*) hv);
923b6d4e
KW
3613 }
3614
3615 /* The string containing the main body of the table */
3616 listsvp = hv_fetchs(hv, "LIST", FALSE);
3617 typesvp = hv_fetchs(hv, "TYPE", FALSE);
3618 bitssvp = hv_fetchs(hv, "BITS", FALSE);
3619 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3620 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3621
3622 typestr = (U8*)SvPV_nolen(*typesvp);
3623 bits = SvUV(*bitssvp);
3624 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3625
d764b54e 3626 /* read $swash->{LIST} */
b443038a
KW
3627 if (SvPOK(*listsvp)) {
3628 l = (U8*)SvPV(*listsvp, lcur);
3629 }
3630 else {
3631 /* LIST legitimately doesn't contain a string during compilation phases
3632 * of Perl itself, before the Unicode tables are generated. In this
3633 * case, just fake things up by creating an empty list */
3634 l = empty;
3635 lcur = 0;