This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Suppress overflow warning in bop.t.
[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)";
760c7c2f
KW
38static const char cp_above_legal_max[] =
39 "It is deprecated to use code point 0x%"UVXf"; the permissible max is 0x%"UVXf"";
40
41#define MAX_NON_DEPRECATED_CP (IV_MAX)
901b21bf 42
48ef279e 43/*
ccfc67b7 44=head1 Unicode Support
7fefc6c1 45These are various utility functions for manipulating UTF8-encoded
72d33970 46strings. For the uninitiated, this is a method of representing arbitrary
61296642 47Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
48characters in the ASCII range are unmodified, and a zero byte never appears
49within non-zero characters.
166f8a29 50
eaf7a4d2
CS
51=cut
52*/
53
54/*
7bbfa158 55=for apidoc is_invariant_string
eaf7a4d2 56
b8ed1ca9
KW
57Returns true iff the first C<len> bytes of the string C<s> are the same
58regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
59EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
60machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 61definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
62so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
63EBCDIC).
eaf7a4d2 64
75200dff
KW
65If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
66use this option, that C<s> can't have embedded C<NUL> characters and has to
67have a terminating C<NUL> byte).
9f7e3d64 68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
7bbfa158 75Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
7bbfa158 80 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
378516de 91=for apidoc uvoffuni_to_utf8_flags
eebe1485 92
a27992cc 93THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
94Instead, B<Almost all code should use L</uvchr_to_utf8> or
95L</uvchr_to_utf8_flags>>.
a27992cc 96
de69f3af
KW
97This function is like them, but the input is a strict Unicode
98(as opposed to native) code point. Only in very rare circumstances should code
99not be using the native code point.
949cf498 100
efa9cd84 101For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 102
eebe1485
SC
103=cut
104*/
105
8ee1cdcb
KW
106#define HANDLE_UNICODE_SURROGATE(uv, flags) \
107 STMT_START { \
108 if (flags & UNICODE_WARN_SURROGATE) { \
109 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
110 "UTF-16 surrogate U+%04"UVXf, uv); \
111 } \
112 if (flags & UNICODE_DISALLOW_SURROGATE) { \
113 return NULL; \
114 } \
115 } STMT_END;
116
117#define HANDLE_UNICODE_NONCHAR(uv, flags) \
118 STMT_START { \
119 if (flags & UNICODE_WARN_NONCHAR) { \
120 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
121 "Unicode non-character U+%04"UVXf" is not " \
122 "recommended for open interchange", uv); \
123 } \
124 if (flags & UNICODE_DISALLOW_NONCHAR) { \
125 return NULL; \
126 } \
127 } STMT_END;
128
ba6ed43c
KW
129/* Use shorter names internally in this file */
130#define SHIFT UTF_ACCUMULATION_SHIFT
131#undef MARK
132#define MARK UTF_CONTINUATION_MARK
133#define MASK UTF_CONTINUATION_MASK
134
dfe13c55 135U8 *
378516de 136Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 137{
378516de 138 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 139
2d1545e5 140 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 141 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
142 return d;
143 }
3ea68d71
KW
144 if (uv <= MAX_UTF8_TWO_BYTE) {
145 *d++ = UTF8_TWO_BYTE_HI(uv);
146 *d++ = UTF8_TWO_BYTE_LO(uv);
147 return d;
148 }
d9432125 149
ba6ed43c
KW
150 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
151 * below, the 16 is for start bytes E0-EF (which are all the possible ones
152 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
153 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
154 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
155 * 0x800-0xFFFF on ASCII */
156 if (uv < (16 * (1U << (2 * SHIFT)))) {
157 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
158 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
159 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
160
161#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
162 aren't tested here */
163 /* The most likely code points in this range are below the surrogates.
164 * Do an extra test to quickly exclude those. */
165 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
166 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
167 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
168 {
8ee1cdcb
KW
169 HANDLE_UNICODE_NONCHAR(uv, flags);
170 }
171 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
172 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 173 }
ba6ed43c
KW
174 }
175#endif
176 return d;
177 }
178
179 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
180 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
181 * happen starting with 4-byte characters on ASCII platforms. We unify the
182 * code for these with EBCDIC, even though some of them require 5-bytes on
183 * those, because khw believes the code saving is worth the very slight
184 * performance hit on these high EBCDIC code points. */
185
186 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
187 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
188 && ckWARN_d(WARN_DEPRECATED))
189 {
190 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
191 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
192 }
193 if ( (flags & UNICODE_WARN_SUPER)
194 || ( UNICODE_IS_ABOVE_31_BIT(uv)
195 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
196 {
197 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
198
199 /* Choose the more dire applicable warning */
200 (UNICODE_IS_ABOVE_31_BIT(uv))
201 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
202 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
203 uv);
204 }
205 if (flags & UNICODE_DISALLOW_SUPER
206 || ( UNICODE_IS_ABOVE_31_BIT(uv)
207 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
208 {
209 return NULL;
210 }
211 }
ba6ed43c
KW
212 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
213 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 214 }
d9432125 215
ba6ed43c
KW
216 /* Test for and handle 4-byte result. In the test immediately below, the
217 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
218 * characters). The 3 is for 3 continuation bytes; these each contribute
219 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
220 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
221 * 0x1_0000-0x1F_FFFF on ASCII */
222 if (uv < (8 * (1U << (3 * SHIFT)))) {
223 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
224 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
225 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
226 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
227
228#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
229 characters. The end-plane non-characters for EBCDIC were
230 handled just above */
231 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
232 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 233 }
ba6ed43c
KW
234 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
235 HANDLE_UNICODE_SURROGATE(uv, flags);
236 }
237#endif
238
239 return d;
240 }
241
242 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
243 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
244 * format. The unrolled version above turns out to not save all that much
245 * time, and at these high code points (well above the legal Unicode range
246 * on ASCII platforms, and well above anything in common use in EBCDIC),
247 * khw believes that less code outweighs slight performance gains. */
248
d9432125 249 {
5aaebcb3 250 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
251 U8 *p = d+len-1;
252 while (p > d) {
4c8cd605 253 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
254 uv >>= UTF_ACCUMULATION_SHIFT;
255 }
4c8cd605 256 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
257 return d+len;
258 }
a0ed51b3 259}
a5bf80e0 260
646ca15d 261/*
07693fe6
KW
262=for apidoc uvchr_to_utf8
263
bcb1a2d4 264Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 265of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
266C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
267the byte after the end of the new character. In other words,
07693fe6
KW
268
269 d = uvchr_to_utf8(d, uv);
270
271is the recommended wide native character-aware way of saying
272
273 *(d++) = uv;
274
760c7c2f
KW
275This function accepts any UV as input, but very high code points (above
276C<IV_MAX> on the platform) will raise a deprecation warning. This is
277typically 0x7FFF_FFFF in a 32-bit word.
278
279It is possible to forbid or warn on non-Unicode code points, or those that may
280be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 281
07693fe6
KW
282=cut
283*/
284
de69f3af
KW
285/* This is also a macro */
286PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
287
07693fe6
KW
288U8 *
289Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
290{
de69f3af 291 return uvchr_to_utf8(d, uv);
07693fe6
KW
292}
293
de69f3af
KW
294/*
295=for apidoc uvchr_to_utf8_flags
296
297Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 298of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
299C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
300the byte after the end of the new character. In other words,
de69f3af
KW
301
302 d = uvchr_to_utf8_flags(d, uv, flags);
303
304or, in most cases,
305
306 d = uvchr_to_utf8_flags(d, uv, 0);
307
308This is the Unicode-aware way of saying
309
310 *(d++) = uv;
311
760c7c2f
KW
312If C<flags> is 0, this function accepts any UV as input, but very high code
313points (above C<IV_MAX> for the platform) will raise a deprecation warning.
314This is typically 0x7FFF_FFFF in a 32-bit word.
315
316Specifying C<flags> can further restrict what is allowed and not warned on, as
317follows:
de69f3af 318
796b6530 319If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
320the function will raise a warning, provided UTF8 warnings are enabled. If
321instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
322NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 323
760c7c2f
KW
324Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
325affect how the function handles a Unicode non-character.
93e6dbd6 326
760c7c2f
KW
327And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
328affect the handling of code points that are above the Unicode maximum of
3290x10FFFF. Languages other than Perl may not be able to accept files that
330contain these.
93e6dbd6
KW
331
332The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
333the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
334three DISALLOW flags.
335
ab8e6d41
KW
336Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
337so using them is more problematic than other above-Unicode code points. Perl
338invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
339likely that non-Perl languages will not be able to read files that contain
340these that written by the perl interpreter; nor would Perl understand files
341written by something that uses a different extension. For these reasons, there
342is a separate set of flags that can warn and/or disallow these extremely high
343code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
344C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
345are entirely independent from the deprecation warning for code points above
346C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
347code point that needs more than 31 bits to represent. When that happens,
348effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
34932-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
350above-Unicode code points, including these, as malformations; and
351C<UNICODE_WARN_SUPER> warns on these.)
352
353On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
354extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
355than on ASCII. Prior to that, code points 2**31 and higher were simply
356unrepresentable, and a different, incompatible method was used to represent
357code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
358and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
359platforms, warning and disallowing 2**31 and higher.
de69f3af 360
de69f3af
KW
361=cut
362*/
363
364/* This is also a macro */
365PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
366
07693fe6
KW
367U8 *
368Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
369{
de69f3af 370 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
371}
372
373/*
87cea99e 374=for apidoc is_utf8_string
6662521e 375
a1433954 376Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 377UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
378using C<strlen(s)> (which means if you use this option, that C<s> can't have
379embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
380that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 381
7bbfa158 382See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 383
6662521e
GS
384=cut
385*/
386
8e84507e 387bool
668b6d8d 388Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 389{
35da51f7 390 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 391 const U8* x = s;
067a85ef 392
7918f24d 393 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 394
6662521e 395 while (x < send) {
6302f837
KW
396 STRLEN len = isUTF8_CHAR(x, send);
397 if (UNLIKELY(! len)) {
398 return FALSE;
399 }
400 x += len;
6662521e 401 }
768c67ee 402
067a85ef 403 return TRUE;
6662521e
GS
404}
405
67e989fb 406/*
814fafa7
NC
407Implemented as a macro in utf8.h
408
87cea99e 409=for apidoc is_utf8_string_loc
814fafa7 410
a1433954
KW
411Like L</is_utf8_string> but stores the location of the failure (in the
412case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
413"utf8ness success") in the C<ep>.
414
a1433954 415See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 416
87cea99e 417=for apidoc is_utf8_string_loclen
81cd54e3 418
a1433954
KW
419Like L</is_utf8_string>() but stores the location of the failure (in the
420case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
421"utf8ness success") in the C<ep>, and the number of UTF-8
422encoded characters in the C<el>.
423
a1433954 424See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
425
426=cut
427*/
428
429bool
668b6d8d 430Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 431{
35da51f7 432 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 433 const U8* x = s;
3ebfea28 434 STRLEN outlen = 0;
7918f24d
NC
435
436 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 437
81cd54e3 438 while (x < send) {
6302f837
KW
439 STRLEN len = isUTF8_CHAR(x, send);
440 if (UNLIKELY(! len)) {
441 goto out;
442 }
443 x += len;
444 outlen++;
81cd54e3 445 }
768c67ee
JH
446
447 out:
3ebfea28
AL
448 if (el)
449 *el = outlen;
450
768c67ee
JH
451 if (ep)
452 *ep = x;
3ebfea28 453 return (x == send);
81cd54e3
JH
454}
455
456/*
768c67ee 457
de69f3af 458=for apidoc utf8n_to_uvchr
378516de
KW
459
460THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 461Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 462
9041c2e3 463Bottom level UTF-8 decode routine.
de69f3af 464Returns the native code point value of the first character in the string C<s>,
746afd53
KW
465which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
466C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
467the length, in bytes, of that character.
949cf498
KW
468
469The value of C<flags> determines the behavior when C<s> does not point to a
470well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
471zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
472next possible position in C<s> that could begin a non-malformed character.
473Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
474
475Various ALLOW flags can be set in C<flags> to allow (and not warn on)
476individual types of malformations, such as the sequence being overlong (that
477is, when there is a shorter sequence that can express the same code point;
478overlong sequences are expressly forbidden in the UTF-8 standard due to
479potential security issues). Another malformation example is the first byte of
480a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
481flags. For allowed 0 length strings, this function returns 0; for allowed
482overlong sequences, the computed code point is returned; for all other allowed
483malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
484determinable reasonable value.
949cf498 485
796b6530 486The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
487flags) malformation is found. If this flag is set, the routine assumes that
488the caller will raise a warning, and this function will silently just set
d088425d
KW
489C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
490
75200dff 491Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 492character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 493in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
494first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
495input had an error.
949cf498
KW
496
497Certain code points are considered problematic. These are Unicode surrogates,
746afd53 498Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 499By default these are considered regular code points, but certain situations
5eafe189 500warrant special handling for them. If C<flags> contains
796b6530
KW
501C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
502malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
503C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
504Unicode maximum) can be set to disallow these categories individually.
505
506The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
507C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
508raised for their respective categories, but otherwise the code points are
509considered valid (not malformations). To get a category to both be treated as
510a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 511(But note that warnings are not raised if lexically disabled nor if
796b6530 512C<UTF8_CHECK_ONLY> is also specified.)
949cf498 513
760c7c2f
KW
514It is now deprecated to have very high code points (above C<IV_MAX> on the
515platforms) and this function will raise a deprecation warning for these (unless
516such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
517in a 32-bit word.
ab8e6d41
KW
518
519Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
520so using them is more problematic than other above-Unicode code points. Perl
521invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
522likely that non-Perl languages will not be able to read files that contain
523these that written by the perl interpreter; nor would Perl understand files
524written by something that uses a different extension. For these reasons, there
525is a separate set of flags that can warn and/or disallow these extremely high
526code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
527C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
528are entirely independent from the deprecation warning for code points above
529C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
530code point that needs more than 31 bits to represent. When that happens,
531effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
53232-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
533above-Unicode code points, including these, as malformations; and
534C<UTF8_WARN_SUPER> warns on these.)
535
536On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
537extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
538than on ASCII. Prior to that, code points 2**31 and higher were simply
539unrepresentable, and a different, incompatible method was used to represent
540code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
541and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
542platforms, warning and disallowing 2**31 and higher.
949cf498
KW
543
544All other code points corresponding to Unicode characters, including private
545use and those yet to be assigned, are never considered malformed and never
546warn.
67e989fb 547
37607a96
PK
548=cut
549*/
67e989fb 550
a0ed51b3 551UV
de69f3af 552Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 553{
d4c19fe8 554 const U8 * const s0 = s;
eb83ed87 555 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 556 U8 * send;
eb83ed87
KW
557 UV uv = *s;
558 STRLEN expectlen;
949cf498 559 SV* sv = NULL;
eb83ed87
KW
560 UV outlier_ret = 0; /* return value when input is in error or problematic
561 */
562 UV pack_warn = 0; /* Save result of packWARN() for later */
563 bool unexpected_non_continuation = FALSE;
564 bool overflowed = FALSE;
2f8f112e 565 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 566
eb83ed87 567 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 568
de69f3af 569 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 570
eb83ed87
KW
571 /* The order of malformation tests here is important. We should consume as
572 * few bytes as possible in order to not skip any valid character. This is
573 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
574 * http://unicode.org/reports/tr36 for more discussion as to why. For
575 * example, once we've done a UTF8SKIP, we can tell the expected number of
576 * bytes, and could fail right off the bat if the input parameters indicate
577 * that there are too few available. But it could be that just that first
578 * byte is garbled, and the intended character occupies fewer bytes. If we
579 * blindly assumed that the first byte is correct, and skipped based on
580 * that number, we could skip over a valid input character. So instead, we
581 * always examine the sequence byte-by-byte.
582 *
583 * We also should not consume too few bytes, otherwise someone could inject
584 * things. For example, an input could be deliberately designed to
585 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 586 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
587 * which is actually part of of the overflowing sequence), that could look
588 * legitimate to the caller, which could discard the initial partial
589 * sequence and process the rest, inappropriately */
590
591 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 592 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
593 if (retlen) {
594 *retlen = 0;
595 }
a0dbb045 596
eb83ed87
KW
597 if (flags & UTF8_ALLOW_EMPTY) {
598 return 0;
599 }
600 if (! (flags & UTF8_CHECK_ONLY)) {
601 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
602 }
0c443dc2
JH
603 goto malformed;
604 }
605
eb83ed87
KW
606 expectlen = UTF8SKIP(s);
607
608 /* A well-formed UTF-8 character, as the vast majority of calls to this
609 * function will be for, has this expected length. For efficiency, set
610 * things up here to return it. It will be overriden only in those rare
611 * cases where a malformation is found */
612 if (retlen) {
613 *retlen = expectlen;
614 }
615
616 /* An invariant is trivially well-formed */
1d72bdf6 617 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 618 return uv;
a0ed51b3 619 }
67e989fb 620
eb83ed87 621 /* A continuation character can't start a valid sequence */
b5b9af04 622 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
623 if (flags & UTF8_ALLOW_CONTINUATION) {
624 if (retlen) {
625 *retlen = 1;
626 }
627 return UNICODE_REPLACEMENT;
628 }
ba210ebe 629
eb83ed87
KW
630 if (! (flags & UTF8_CHECK_ONLY)) {
631 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
632 }
633 curlen = 1;
ba210ebe
JH
634 goto malformed;
635 }
9041c2e3 636
dcd27b3c
KW
637 /* Here is not a continuation byte, nor an invariant. The only thing left
638 * is a start byte (possibly for an overlong) */
639
1d72bdf6 640#ifdef EBCDIC
bc3632a8 641 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
642#endif
643
eb83ed87
KW
644 /* Remove the leading bits that indicate the number of bytes in the
645 * character's whole UTF-8 sequence, leaving just the bits that are part of
646 * the value */
647 uv &= UTF_START_MASK(expectlen);
ba210ebe 648
eb83ed87
KW
649 /* Now, loop through the remaining bytes in the character's sequence,
650 * accumulating each into the working value as we go. Be sure to not look
651 * past the end of the input string */
0b8d30e8
KW
652 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
653
eb83ed87 654 for (s = s0 + 1; s < send; s++) {
b5b9af04 655 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
656 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
657
658 /* The original implementors viewed this malformation as more
659 * serious than the others (though I, khw, don't understand
660 * why, since other malformations also give very very wrong
661 * results), so there is no way to turn off checking for it.
662 * Set a flag, but keep going in the loop, so that we absorb
663 * the rest of the bytes that comprise the character. */
664 overflowed = TRUE;
665 overflow_byte = *s; /* Save for warning message's use */
666 }
8850bf83 667 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
668 }
669 else {
670 /* Here, found a non-continuation before processing all expected
671 * bytes. This byte begins a new character, so quit, even if
672 * allowing this malformation. */
673 unexpected_non_continuation = TRUE;
674 break;
675 }
676 } /* End of loop through the character's bytes */
677
678 /* Save how many bytes were actually in the character */
679 curlen = s - s0;
680
681 /* The loop above finds two types of malformations: non-continuation and/or
682 * overflow. The non-continuation malformation is really a too-short
683 * malformation, as it means that the current character ended before it was
684 * expected to (being terminated prematurely by the beginning of the next
685 * character, whereas in the too-short malformation there just are too few
686 * bytes available to hold the character. In both cases, the check below
687 * that we have found the expected number of bytes would fail if executed.)
688 * Thus the non-continuation malformation is really unnecessary, being a
689 * subset of the too-short malformation. But there may be existing
690 * applications that are expecting the non-continuation type, so we retain
691 * it, and return it in preference to the too-short malformation. (If this
692 * code were being written from scratch, the two types might be collapsed
693 * into one.) I, khw, am also giving priority to returning the
694 * non-continuation and too-short malformations over overflow when multiple
695 * ones are present. I don't know of any real reason to prefer one over
696 * the other, except that it seems to me that multiple-byte errors trumps
697 * errors from a single byte */
b5b9af04 698 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
699 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
700 if (! (flags & UTF8_CHECK_ONLY)) {
701 if (curlen == 1) {
702 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
703 }
704 else {
705 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
706 }
707 }
eb83ed87
KW
708 goto malformed;
709 }
710 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
711
712 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
713 * as what the original expectations were. */
714 do_overlong_test = FALSE;
eb83ed87
KW
715 if (retlen) {
716 *retlen = curlen;
717 }
718 }
b5b9af04 719 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
720 if (! (flags & UTF8_ALLOW_SHORT)) {
721 if (! (flags & UTF8_CHECK_ONLY)) {
722 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 723 }
eb83ed87
KW
724 goto malformed;
725 }
726 uv = UNICODE_REPLACEMENT;
2f8f112e 727 do_overlong_test = FALSE;
eb83ed87
KW
728 if (retlen) {
729 *retlen = curlen;
730 }
731 }
732
b5b9af04 733 if (UNLIKELY(overflowed)) {
eb83ed87 734 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 735 goto malformed;
eb83ed87 736 }
eb83ed87 737
2f8f112e 738 if (do_overlong_test
5aaebcb3 739 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
740 && ! (flags & UTF8_ALLOW_LONG))
741 {
eb83ed87
KW
742 /* The overlong malformation has lower precedence than the others.
743 * Note that if this malformation is allowed, we return the actual
744 * value, instead of the replacement character. This is because this
745 * value is actually well-defined. */
746 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 747 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
748 }
749 goto malformed;
750 }
751
1a89bb6c 752 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
753 * problematic code point that is not allowed by the input parameters. */
754 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
755 && ((flags & ( UTF8_DISALLOW_NONCHAR
756 |UTF8_DISALLOW_SURROGATE
757 |UTF8_DISALLOW_SUPER
758 |UTF8_DISALLOW_ABOVE_31_BIT
759 |UTF8_WARN_NONCHAR
760 |UTF8_WARN_SURROGATE
761 |UTF8_WARN_SUPER
762 |UTF8_WARN_ABOVE_31_BIT))
763 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
764 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 765 {
949cf498 766 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
767
768 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
769 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 770 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 771 && ckWARN_d(WARN_SURROGATE))
eb83ed87 772 {
111d382d 773 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 774 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
775 }
776 if (flags & UTF8_DISALLOW_SURROGATE) {
777 goto disallowed;
778 }
779 }
949cf498 780 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 781 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 782 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 783 {
0bcdd8f6
KW
784 sv = sv_2mortal(Perl_newSVpvf(aTHX_
785 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
786 uv));
54f4afef 787 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 788 }
c0236afe
KW
789
790 /* The maximum code point ever specified by a standard was
791 * 2**31 - 1. Anything larger than that is a Perl extension that
792 * very well may not be understood by other applications (including
793 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
794 * these code points are indicated by the first UTF-8 byte being
d35f2ca5
KW
795 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
796 * and before possibly bailing out, so that the slightly more dire
c0236afe
KW
797 * warning will override the regular one. */
798 if (
799#ifndef EBCDIC
800 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
801#else
802 /* The I8 for 2**31 (U+80000000) is
803 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
804 * and it turns out that on all EBCDIC pages recognized that
805 * the UTF-EBCDIC for that code point is
806 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
807 * For the next lower code point, the 1047 UTF-EBCDIC is
808 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
809 * The other code pages differ only in the bytes following
810 * \x42. Thus the following works (the minimum continuation
811 * byte is \x41). */
812 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
813 || s0[2] > 0x41
814 || s0[3] > 0x41
815 || s0[4] > 0x41
816 || s0[5] > 0x41
817 || s0[6] > 0x41
818 || s0[7] > 0x42)
819#endif
d35f2ca5
KW
820 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
821 |UTF8_DISALLOW_ABOVE_31_BIT)))
ea5ced44 822 {
0bcdd8f6 823 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 824 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 825 && ckWARN_d(WARN_UTF8))
ea5ced44 826 {
0bcdd8f6
KW
827 sv = sv_2mortal(Perl_newSVpvf(aTHX_
828 "Code point 0x%"UVXf" is not Unicode, and not portable",
829 uv));
ea5ced44
KW
830 pack_warn = packWARN(WARN_UTF8);
831 }
d35f2ca5 832 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
833 goto disallowed;
834 }
835 }
c0236afe 836
949cf498
KW
837 if (flags & UTF8_DISALLOW_SUPER) {
838 goto disallowed;
839 }
760c7c2f
KW
840
841 /* The deprecated warning overrides any non-deprecated one */
842 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
843 {
844 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
845 uv, MAX_NON_DEPRECATED_CP));
846 pack_warn = packWARN(WARN_DEPRECATED);
847 }
949cf498 848 }
4190d317
KW
849 else if (UNICODE_IS_NONCHAR(uv)) {
850 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 851 && ckWARN_d(WARN_NONCHAR))
4190d317 852 {
ba707cdc 853 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 854 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
855 }
856 if (flags & UTF8_DISALLOW_NONCHAR) {
857 goto disallowed;
858 }
859 }
949cf498 860
eb83ed87 861 if (sv) {
de69f3af
KW
862 outlier_ret = uv; /* Note we don't bother to convert to native,
863 as all the outlier code points are the same
864 in both ASCII and EBCDIC */
eb83ed87
KW
865 goto do_warn;
866 }
867
949cf498
KW
868 /* Here, this is not considered a malformed character, so drop through
869 * to return it */
a0ed51b3 870 }
ba210ebe 871
de69f3af 872 return UNI_TO_NATIVE(uv);
ba210ebe 873
eb83ed87
KW
874 /* There are three cases which get to beyond this point. In all 3 cases:
875 * <sv> if not null points to a string to print as a warning.
876 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
877 * set.
878 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
879 * This is done by initializing it to 0, and changing it only
880 * for case 1).
881 * The 3 cases are:
882 * 1) The input is valid but problematic, and to be warned about. The
883 * return value is the resultant code point; <*retlen> is set to
884 * <curlen>, the number of bytes that comprise the code point.
885 * <pack_warn> contains the result of packWARN() for the warning
886 * types. The entry point for this case is the label <do_warn>;
887 * 2) The input is a valid code point but disallowed by the parameters to
888 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
889 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
890 * comprise the code point. <pack_warn> contains the result of
891 * packWARN() for the warning types. The entry point for this case is
892 * the label <disallowed>.
893 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
894 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
895 * bytes that comprise the malformation. All such malformations are
896 * assumed to be warning type <utf8>. The entry point for this case
897 * is the label <malformed>.
898 */
949cf498 899
7b52d656 900 malformed:
ba210ebe 901
eb83ed87
KW
902 if (sv && ckWARN_d(WARN_UTF8)) {
903 pack_warn = packWARN(WARN_UTF8);
904 }
905
7b52d656 906 disallowed:
eb83ed87 907
fcc8fcf6 908 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 909 if (retlen)
10edeb5d 910 *retlen = ((STRLEN) -1);
ba210ebe
JH
911 return 0;
912 }
913
7b52d656 914 do_warn:
5b311467 915
eb83ed87
KW
916 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
917 if warnings are to be raised. */
f555bc63 918 const char * const string = SvPVX_const(sv);
a0dbb045 919
f555bc63
KW
920 if (PL_op)
921 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
922 else
923 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
924 }
925
eb83ed87
KW
926 if (retlen) {
927 *retlen = curlen;
928 }
ba210ebe 929
eb83ed87 930 return outlier_ret;
a0ed51b3
LW
931}
932
8e84507e 933/*
ec5f19d0
KW
934=for apidoc utf8_to_uvchr_buf
935
936Returns the native code point of the first character in the string C<s> which
937is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 938C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 939
524080c4
KW
940If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
941enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 942C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 943(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 944C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 945the next possible position in C<s> that could begin a non-malformed character.
de69f3af 946See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 947returned.
ec5f19d0 948
760c7c2f
KW
949Code points above the platform's C<IV_MAX> will raise a deprecation warning,
950unless those are turned off.
951
ec5f19d0
KW
952=cut
953*/
954
955
956UV
957Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
958{
ec5f19d0
KW
959 assert(s < send);
960
961 return utf8n_to_uvchr(s, send - s, retlen,
962 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
963}
964
27d6c58a 965/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 966 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 967 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
968
969UV
970Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
971{
010ab96b
KW
972 UV expectlen = UTF8SKIP(s);
973 const U8* send = s + expectlen;
9ff2f0f7 974 UV uv = *s;
3986bb7c 975
27d6c58a 976 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 977 PERL_UNUSED_CONTEXT;
27d6c58a 978
010ab96b
KW
979 if (retlen) {
980 *retlen = expectlen;
981 }
982
983 /* An invariant is trivially returned */
984 if (expectlen == 1) {
9ff2f0f7 985 return uv;
010ab96b
KW
986 }
987
9ff2f0f7
KW
988#ifdef EBCDIC
989 uv = NATIVE_UTF8_TO_I8(uv);
990#endif
991
010ab96b
KW
992 /* Remove the leading bits that indicate the number of bytes, leaving just
993 * the bits that are part of the value */
994 uv &= UTF_START_MASK(expectlen);
995
996 /* Now, loop through the remaining bytes, accumulating each into the
997 * working total as we go. (I khw tried unrolling the loop for up to 4
998 * bytes, but there was no performance improvement) */
999 for (++s; s < send; s++) {
1000 uv = UTF8_ACCUMULATE(uv, *s);
1001 }
1002
3986bb7c 1003 return UNI_TO_NATIVE(uv);
010ab96b 1004
27d6c58a
KW
1005}
1006
ec5f19d0 1007/*
ec5f19d0
KW
1008=for apidoc utf8_to_uvuni_buf
1009
de69f3af
KW
1010Only in very rare circumstances should code need to be dealing in Unicode
1011(as opposed to native) code points. In those few cases, use
1012C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1013
1014Returns the Unicode (not-native) code point of the first character in the
1015string C<s> which
ec5f19d0
KW
1016is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1017C<retlen> will be set to the length, in bytes, of that character.
1018
524080c4
KW
1019If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1020enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1021NULL) to -1. If those warnings are off, the computed value if well-defined (or
1022the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1023is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1024next possible position in C<s> that could begin a non-malformed character.
de69f3af 1025See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1026
760c7c2f
KW
1027Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1028unless those are turned off.
1029
ec5f19d0
KW
1030=cut
1031*/
1032
1033UV
1034Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1035{
1036 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1037
1038 assert(send > s);
1039
1040 /* Call the low level routine asking for checks */
de69f3af
KW
1041 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1042 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1043}
1044
b76347f2 1045/*
87cea99e 1046=for apidoc utf8_length
b76347f2
JH
1047
1048Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1049Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1050up past C<e>, croaks.
b76347f2
JH
1051
1052=cut
1053*/
1054
1055STRLEN
35a4481c 1056Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1057{
1058 STRLEN len = 0;
1059
7918f24d
NC
1060 PERL_ARGS_ASSERT_UTF8_LENGTH;
1061
8850bf83
JH
1062 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1063 * the bitops (especially ~) can create illegal UTF-8.
1064 * In other words: in Perl UTF-8 is not just for Unicode. */
1065
a3b680e6
AL
1066 if (e < s)
1067 goto warn_and_return;
b76347f2 1068 while (s < e) {
4cbf4130 1069 s += UTF8SKIP(s);
8e91ec7f
AV
1070 len++;
1071 }
1072
1073 if (e != s) {
1074 len--;
1075 warn_and_return:
9b387841
NC
1076 if (PL_op)
1077 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1078 "%s in %s", unees, OP_DESC(PL_op));
1079 else
61a12c31 1080 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1081 }
1082
1083 return len;
1084}
1085
b06226ff 1086/*
87cea99e 1087=for apidoc utf8_distance
b06226ff 1088
1e54db1a 1089Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1090and C<b>.
1091
1092WARNING: use only if you *know* that the pointers point inside the
1093same UTF-8 buffer.
1094
37607a96
PK
1095=cut
1096*/
a0ed51b3 1097
02eb7b47 1098IV
35a4481c 1099Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1100{
7918f24d
NC
1101 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1102
bf1665bc 1103 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1104}
1105
b06226ff 1106/*
87cea99e 1107=for apidoc utf8_hop
b06226ff 1108
8850bf83
JH
1109Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1110forward or backward.
b06226ff
JH
1111
1112WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1113the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1114on the first byte of character or just after the last byte of a character.
b06226ff 1115
37607a96
PK
1116=cut
1117*/
a0ed51b3
LW
1118
1119U8 *
ddeaf645 1120Perl_utf8_hop(const U8 *s, I32 off)
a0ed51b3 1121{
7918f24d
NC
1122 PERL_ARGS_ASSERT_UTF8_HOP;
1123
8850bf83
JH
1124 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1125 * the bitops (especially ~) can create illegal UTF-8.
1126 * In other words: in Perl UTF-8 is not just for Unicode. */
1127
a0ed51b3
LW
1128 if (off >= 0) {
1129 while (off--)
1130 s += UTF8SKIP(s);
1131 }
1132 else {
1133 while (off++) {
1134 s--;
8850bf83
JH
1135 while (UTF8_IS_CONTINUATION(*s))
1136 s--;
a0ed51b3
LW
1137 }
1138 }
4373e329 1139 return (U8 *)s;
a0ed51b3
LW
1140}
1141
6940069f 1142/*
fed3ba5d
NC
1143=for apidoc bytes_cmp_utf8
1144
a1433954 1145Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1146sequence of characters (stored as UTF-8)
1147in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1148equal, -1 or -2 if the first string is less than the second string, +1 or +2
1149if the first string is greater than the second string.
1150
1151-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1152longer string. -2 or +2 is returned if
1153there was a difference between characters
fed3ba5d
NC
1154within the strings.
1155
1156=cut
1157*/
1158
1159int
1160Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1161{
1162 const U8 *const bend = b + blen;
1163 const U8 *const uend = u + ulen;
1164
1165 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1166
1167 while (b < bend && u < uend) {
1168 U8 c = *u++;
1169 if (!UTF8_IS_INVARIANT(c)) {
1170 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1171 if (u < uend) {
1172 U8 c1 = *u++;
1173 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1174 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1175 } else {
1176 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1177 "Malformed UTF-8 character "
1178 "(unexpected non-continuation byte 0x%02x"
1179 ", immediately after start byte 0x%02x)"
1180 /* Dear diag.t, it's in the pod. */
1181 "%s%s", c1, c,
1182 PL_op ? " in " : "",
1183 PL_op ? OP_DESC(PL_op) : "");
1184 return -2;
1185 }
1186 } else {
1187 if (PL_op)
1188 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1189 "%s in %s", unees, OP_DESC(PL_op));
1190 else
61a12c31 1191 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1192 return -2; /* Really want to return undef :-) */
1193 }
1194 } else {
1195 return -2;
1196 }
1197 }
1198 if (*b != c) {
1199 return *b < c ? -2 : +2;
1200 }
1201 ++b;
1202 }
1203
1204 if (b == bend && u == uend)
1205 return 0;
1206
1207 return b < bend ? +1 : -1;
1208}
1209
1210/*
87cea99e 1211=for apidoc utf8_to_bytes
6940069f 1212
2bbc8d55 1213Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1214Unlike L</bytes_to_utf8>, this over-writes the original string, and
1215updates C<len> to contain the new length.
67e989fb 1216Returns zero on failure, setting C<len> to -1.
6940069f 1217
a1433954 1218If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1219
6940069f
GS
1220=cut
1221*/
1222
1223U8 *
37607a96 1224Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1225{
d4c19fe8
AL
1226 U8 * const save = s;
1227 U8 * const send = s + *len;
6940069f 1228 U8 *d;
246fae53 1229
7918f24d 1230 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1231 PERL_UNUSED_CONTEXT;
7918f24d 1232
1e54db1a 1233 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1234 while (s < send) {
d59937ca
KW
1235 if (! UTF8_IS_INVARIANT(*s)) {
1236 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1237 *len = ((STRLEN) -1);
1238 return 0;
1239 }
1240 s++;
dcad2880 1241 }
d59937ca 1242 s++;
246fae53 1243 }
dcad2880
JH
1244
1245 d = s = save;
6940069f 1246 while (s < send) {
80e0b38f
KW
1247 U8 c = *s++;
1248 if (! UTF8_IS_INVARIANT(c)) {
1249 /* Then it is two-byte encoded */
a62b247b 1250 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1251 s++;
1252 }
1253 *d++ = c;
6940069f
GS
1254 }
1255 *d = '\0';
246fae53 1256 *len = d - save;
6940069f
GS
1257 return save;
1258}
1259
1260/*
87cea99e 1261=for apidoc bytes_from_utf8
f9a63242 1262
2bbc8d55 1263Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1264Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1265the newly-created string, and updates C<len> to contain the new
1266length. Returns the original string if no conversion occurs, C<len>
72d33970 1267is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 12680 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1269in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1270
37607a96
PK
1271=cut
1272*/
f9a63242
JH
1273
1274U8 *
e1ec3a88 1275Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1276{
f9a63242 1277 U8 *d;
e1ec3a88
AL
1278 const U8 *start = s;
1279 const U8 *send;
f9a63242
JH
1280 I32 count = 0;
1281
7918f24d 1282 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1283 PERL_UNUSED_CONTEXT;
f9a63242 1284 if (!*is_utf8)
73d840c0 1285 return (U8 *)start;
f9a63242 1286
1e54db1a 1287 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1288 for (send = s + *len; s < send;) {
d59937ca
KW
1289 if (! UTF8_IS_INVARIANT(*s)) {
1290 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1291 return (U8 *)start;
d59937ca
KW
1292 }
1293 count++;
1294 s++;
db42d148 1295 }
d59937ca 1296 s++;
f9a63242
JH
1297 }
1298
35da51f7 1299 *is_utf8 = FALSE;
f9a63242 1300
212542aa 1301 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1302 s = start; start = d;
f9a63242
JH
1303 while (s < send) {
1304 U8 c = *s++;
1a91c45d 1305 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1306 /* Then it is two-byte encoded */
a62b247b 1307 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1308 s++;
c4d5f83a
NIS
1309 }
1310 *d++ = c;
f9a63242
JH
1311 }
1312 *d = '\0';
1313 *len = d - start;
73d840c0 1314 return (U8 *)start;
f9a63242
JH
1315}
1316
1317/*
87cea99e 1318=for apidoc bytes_to_utf8
6940069f 1319
ff97e5cf
KW
1320Converts a string C<s> of length C<len> bytes from the native encoding into
1321UTF-8.
6662521e 1322Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1323reflect the new length in bytes.
6940069f 1324
75200dff 1325A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1326
1327If you want to convert to UTF-8 from encodings other than
1328the native (Latin1 or EBCDIC),
a1433954 1329see L</sv_recode_to_utf8>().
c9ada85f 1330
497711e7 1331=cut
6940069f
GS
1332*/
1333
c682ebef
FC
1334/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1335 likewise need duplication. */
1336
6940069f 1337U8*
35a4481c 1338Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1339{
35a4481c 1340 const U8 * const send = s + (*len);
6940069f
GS
1341 U8 *d;
1342 U8 *dst;
7918f24d
NC
1343
1344 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1345 PERL_UNUSED_CONTEXT;
6940069f 1346
212542aa 1347 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1348 dst = d;
1349
1350 while (s < send) {
55d09dc8
KW
1351 append_utf8_from_native_byte(*s, &d);
1352 s++;
6940069f
GS
1353 }
1354 *d = '\0';
6662521e 1355 *len = d-dst;
6940069f
GS
1356 return dst;
1357}
1358
a0ed51b3 1359/*
dea0fc0b 1360 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1361 *
1362 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1363 * We optimize for native, for obvious reasons. */
1364
1365U8*
dea0fc0b 1366Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1367{
dea0fc0b
JH
1368 U8* pend;
1369 U8* dstart = d;
1370
7918f24d
NC
1371 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1372
dea0fc0b 1373 if (bytelen & 1)
f5992bc4 1374 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1375
1376 pend = p + bytelen;
1377
a0ed51b3 1378 while (p < pend) {
dea0fc0b
JH
1379 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1380 p += 2;
2d1545e5 1381 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1382 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1383 continue;
1384 }
56d37426
KW
1385 if (uv <= MAX_UTF8_TWO_BYTE) {
1386 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1387 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1388 continue;
1389 }
46956fad
KW
1390#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1391#define LAST_HIGH_SURROGATE 0xDBFF
1392#define FIRST_LOW_SURROGATE 0xDC00
1393#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1394
1395 /* This assumes that most uses will be in the first Unicode plane, not
1396 * needing surrogates */
1397 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1398 && uv <= UNICODE_SURROGATE_LAST))
1399 {
1400 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1401 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1402 }
1403 else {
01ea242b 1404 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1405 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1406 || UNLIKELY(low > LAST_LOW_SURROGATE))
1407 {
01ea242b 1408 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1409 }
1410 p += 2;
46956fad
KW
1411 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1412 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1413 }
a0ed51b3 1414 }
56d37426
KW
1415#ifdef EBCDIC
1416 d = uvoffuni_to_utf8_flags(d, uv, 0);
1417#else
a0ed51b3 1418 if (uv < 0x10000) {
eb160463
GS
1419 *d++ = (U8)(( uv >> 12) | 0xe0);
1420 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1421 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1422 continue;
1423 }
1424 else {
eb160463
GS
1425 *d++ = (U8)(( uv >> 18) | 0xf0);
1426 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1427 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1428 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1429 continue;
1430 }
56d37426 1431#endif
a0ed51b3 1432 }
dea0fc0b 1433 *newlen = d - dstart;
a0ed51b3
LW
1434 return d;
1435}
1436
1437/* Note: this one is slightly destructive of the source. */
1438
1439U8*
dea0fc0b 1440Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1441{
1442 U8* s = (U8*)p;
d4c19fe8 1443 U8* const send = s + bytelen;
7918f24d
NC
1444
1445 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1446
e0ea5e2d
NC
1447 if (bytelen & 1)
1448 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1449 (UV)bytelen);
1450
a0ed51b3 1451 while (s < send) {
d4c19fe8 1452 const U8 tmp = s[0];
a0ed51b3
LW
1453 s[0] = s[1];
1454 s[1] = tmp;
1455 s += 2;
1456 }
dea0fc0b 1457 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1458}
1459
922e8cb4
KW
1460bool
1461Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1462{
1463 U8 tmpbuf[UTF8_MAXBYTES+1];
1464 uvchr_to_utf8(tmpbuf, c);
1465 return _is_utf8_FOO(classnum, tmpbuf);
1466}
1467
f9ae8fb6
JD
1468/* Internal function so we can deprecate the external one, and call
1469 this one from other deprecated functions in this file */
1470
f2645549
KW
1471bool
1472Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1473{
f2645549 1474 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1475
1476 if (*p == '_')
1477 return TRUE;
f25ce844 1478 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1479}
1480
5092f92a 1481bool
eba68aa0
KW
1482Perl__is_uni_perl_idcont(pTHX_ UV c)
1483{
1484 U8 tmpbuf[UTF8_MAXBYTES+1];
1485 uvchr_to_utf8(tmpbuf, c);
1486 return _is_utf8_perl_idcont(tmpbuf);
1487}
1488
1489bool
f91dcd13
KW
1490Perl__is_uni_perl_idstart(pTHX_ UV c)
1491{
1492 U8 tmpbuf[UTF8_MAXBYTES+1];
1493 uvchr_to_utf8(tmpbuf, c);
1494 return _is_utf8_perl_idstart(tmpbuf);
1495}
1496
3a4c58c9
KW
1497UV
1498Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1499{
1500 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1501 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1502 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1503 * either "SS" or "Ss". Which one to use is passed into the routine in
1504 * 'S_or_s' to avoid a test */
1505
1506 UV converted = toUPPER_LATIN1_MOD(c);
1507
1508 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1509
1510 assert(S_or_s == 'S' || S_or_s == 's');
1511
6f2d5cbc 1512 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1513 characters in this range */
3a4c58c9
KW
1514 *p = (U8) converted;
1515 *lenp = 1;
1516 return converted;
1517 }
1518
1519 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1520 * which it maps to one of them, so as to only have to have one check for
1521 * it in the main case */
1522 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1523 switch (c) {
1524 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1525 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1526 break;
1527 case MICRO_SIGN:
1528 converted = GREEK_CAPITAL_LETTER_MU;
1529 break;
79e064b9
KW
1530#if UNICODE_MAJOR_VERSION > 2 \
1531 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1532 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1533 case LATIN_SMALL_LETTER_SHARP_S:
1534 *(p)++ = 'S';
1535 *p = S_or_s;
1536 *lenp = 2;
1537 return 'S';
79e064b9 1538#endif
3a4c58c9
KW
1539 default:
1540 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1541 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1542 }
1543 }
1544
1545 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1546 *p = UTF8_TWO_BYTE_LO(converted);
1547 *lenp = 2;
1548
1549 return converted;
1550}
1551
50bda2c3
KW
1552/* Call the function to convert a UTF-8 encoded character to the specified case.
1553 * Note that there may be more than one character in the result.
1554 * INP is a pointer to the first byte of the input character
1555 * OUTP will be set to the first byte of the string of changed characters. It
1556 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1557 * LENP will be set to the length in bytes of the string of changed characters
1558 *
1559 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
1560#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
1561#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
1562#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 1563
b9992569
KW
1564/* This additionally has the input parameter 'specials', which if non-zero will
1565 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 1566 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 1567#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 1568
84afefe6
JH
1569UV
1570Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1571{
a1433954
KW
1572 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1573 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1574 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1575 * the changed version may be longer than the original character.
1576 *
1577 * The ordinal of the first character of the changed version is returned
1578 * (but note, as explained above, that there may be more.) */
1579
7918f24d
NC
1580 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1581
3a4c58c9
KW
1582 if (c < 256) {
1583 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1584 }
1585
0ebc6274 1586 uvchr_to_utf8(p, c);
b9992569 1587 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
1588}
1589
84afefe6
JH
1590UV
1591Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1592{
7918f24d
NC
1593 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1594
3a4c58c9
KW
1595 if (c < 256) {
1596 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1597 }
1598
0ebc6274 1599 uvchr_to_utf8(p, c);
b9992569 1600 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
1601}
1602
afc16117 1603STATIC U8
81611534 1604S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1605{
1606 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1607 * those, converting the result to UTF-8. Since the result is always just
a1433954 1608 * one character, we allow <p> to be NULL */
afc16117
KW
1609
1610 U8 converted = toLOWER_LATIN1(c);
1611
1612 if (p != NULL) {
6f2d5cbc 1613 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1614 *p = converted;
1615 *lenp = 1;
1616 }
1617 else {
430c9760
KW
1618 /* Result is known to always be < 256, so can use the EIGHT_BIT
1619 * macros */
1620 *p = UTF8_EIGHT_BIT_HI(converted);
1621 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1622 *lenp = 2;
1623 }
1624 }
1625 return converted;
1626}
1627
84afefe6
JH
1628UV
1629Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1630{
7918f24d
NC
1631 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1632
afc16117
KW
1633 if (c < 256) {
1634 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1635 }
1636
afc16117 1637 uvchr_to_utf8(p, c);
b9992569 1638 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
1639}
1640
84afefe6 1641UV
51910141 1642Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1643{
51910141 1644 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1645 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1646 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1647 *
1648 * Not to be used for locale folds
51910141 1649 */
f673fad4 1650
a1dde8de
KW
1651 UV converted;
1652
1653 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1654 PERL_UNUSED_CONTEXT;
a1dde8de 1655
1ca267a5
KW
1656 assert (! (flags & FOLD_FLAGS_LOCALE));
1657
659a7c2d 1658 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
1659 converted = GREEK_SMALL_LETTER_MU;
1660 }
9b63e895
KW
1661#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1662 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1663 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
1664 else if ( (flags & FOLD_FLAGS_FULL)
1665 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
1666 {
1ca267a5
KW
1667 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1668 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1669 * under those circumstances. */
1670 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1671 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1672 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1673 p, *lenp, U8);
1674 return LATIN_SMALL_LETTER_LONG_S;
1675 }
1676 else {
4f489194
KW
1677 *(p)++ = 's';
1678 *p = 's';
1679 *lenp = 2;
1680 return 's';
1ca267a5 1681 }
a1dde8de 1682 }
9b63e895 1683#endif
a1dde8de
KW
1684 else { /* In this range the fold of all other characters is their lower
1685 case */
1686 converted = toLOWER_LATIN1(c);
1687 }
1688
6f2d5cbc 1689 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1690 *p = (U8) converted;
1691 *lenp = 1;
1692 }
1693 else {
1694 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1695 *p = UTF8_TWO_BYTE_LO(converted);
1696 *lenp = 2;
1697 }
1698
1699 return converted;
1700}
1701
1702UV
31f05a37 1703Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1704{
4b593389 1705
a0270393
KW
1706 /* Not currently externally documented, and subject to change
1707 * <flags> bits meanings:
1708 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1709 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1710 * locale are to be used.
a0270393
KW
1711 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1712 */
4b593389 1713
36bb2ab6 1714 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1715
780fcc9f
KW
1716 if (flags & FOLD_FLAGS_LOCALE) {
1717 /* Treat a UTF-8 locale as not being in locale at all */
1718 if (IN_UTF8_CTYPE_LOCALE) {
1719 flags &= ~FOLD_FLAGS_LOCALE;
1720 }
1721 else {
1722 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1723 goto needs_full_generality;
780fcc9f 1724 }
31f05a37
KW
1725 }
1726
a1dde8de 1727 if (c < 256) {
e7b7ac46 1728 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1729 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1730 }
1731
2f306ab9 1732 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1733 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1734 uvchr_to_utf8(p, c);
b9992569 1735 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
1736 }
1737 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1738 the special flags. */
1739 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1740
1741 needs_full_generality:
a0270393 1742 uvchr_to_utf8(utf8_c, c);
445bf929 1743 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1744 }
84afefe6
JH
1745}
1746
26483009 1747PERL_STATIC_INLINE bool
5141f98e 1748S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1749 const char *const swashname, SV* const invlist)
bde6a22d 1750{
ea317ccb
KW
1751 /* returns a boolean giving whether or not the UTF8-encoded character that
1752 * starts at <p> is in the swash indicated by <swashname>. <swash>
1753 * contains a pointer to where the swash indicated by <swashname>
1754 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1755 * look at <*swash> and only generate a swash if it is not null. <invlist>
1756 * is NULL or an inversion list that defines the swash. If not null, it
1757 * saves time during initialization of the swash.
ea317ccb
KW
1758 *
1759 * Note that it is assumed that the buffer length of <p> is enough to
1760 * contain all the bytes that comprise the character. Thus, <*p> should
1761 * have been checked before this call for mal-formedness enough to assure
1762 * that. */
1763
7918f24d
NC
1764 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1765
492a624f 1766 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1767 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1768 * as far as there being enough bytes available in it to accommodate the
1769 * character without reading beyond the end, and pass that number on to the
1770 * validating routine */
6302f837 1771 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1772 if (ckWARN_d(WARN_UTF8)) {
1773 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1774 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1775 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1776 what the malformation is */
1777 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1778 }
1779 }
1780 return FALSE;
1781 }
87367d5f
KW
1782 if (!*swash) {
1783 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1784 *swash = _core_swash_init("utf8",
1785
1786 /* Only use the name if there is no inversion
1787 * list; otherwise will go out to disk */
1788 (invlist) ? "" : swashname,
1789
1790 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1791 }
28123549 1792
bde6a22d
NC
1793 return swash_fetch(*swash, p, TRUE) != 0;
1794}
1795
1796bool
922e8cb4
KW
1797Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1798{
922e8cb4
KW
1799 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1800
1801 assert(classnum < _FIRST_NON_SWASH_CC);
1802
f25ce844
KW
1803 return is_utf8_common(p,
1804 &PL_utf8_swash_ptrs[classnum],
1805 swash_property_names[classnum],
1806 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1807}
1808
1809bool
f2645549 1810Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1811{
f2645549 1812 SV* invlist = NULL;
7918f24d 1813
f2645549 1814 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1815
f2645549
KW
1816 if (! PL_utf8_perl_idstart) {
1817 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1818 }
60071a22 1819 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1820}
1821
1822bool
f2645549 1823Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1824{
f2645549 1825 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1826
1827 if (*p == '_')
1828 return TRUE;
f25ce844 1829 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1830}
1831
1832bool
eba68aa0
KW
1833Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1834{
b24b43f7 1835 SV* invlist = NULL;
eba68aa0
KW
1836
1837 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1838
b24b43f7
KW
1839 if (! PL_utf8_perl_idcont) {
1840 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1841 }
60071a22 1842 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1843}
1844
eba68aa0 1845bool
f2645549 1846Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1847{
f2645549 1848 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1849
f25ce844 1850 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1851}
1852
1853bool
f2645549 1854Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1855{
f2645549 1856 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1857
f25ce844 1858 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1859}
1860
1861bool
7dbf68d2
KW
1862Perl__is_utf8_mark(pTHX_ const U8 *p)
1863{
7dbf68d2
KW
1864 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1865
f25ce844 1866 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1867}
1868
6b5c0936 1869/*
87cea99e 1870=for apidoc to_utf8_case
6b5c0936 1871
9da1e7cb
KW
1872Instead use the appropriate one of L</toUPPER_utf8>,
1873L</toTITLE_utf8>,
1874L</toLOWER_utf8>,
1875or L</toFOLD_utf8>.
1876
6fae5207 1877C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1878the character that is being converted. This routine assumes that the character
1879at C<p> is well-formed.
6b5c0936 1880
6fae5207
KW
1881C<ustrp> is a pointer to the character buffer to put the
1882conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1883of the result.
1884
6fae5207 1885C<swashp> is a pointer to the swash to use.
6b5c0936 1886
a1433954 1887Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1888and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1889but not always, a multicharacter mapping), is tried first.
6b5c0936 1890
4a8240a3
KW
1891C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1892any special mappings; C<""> means to use the special mappings. Values other
1893than these two are treated as the name of the hash containing the special
1894mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1895
796b6530
KW
1896C<normal> is a string like C<"ToLower"> which means the swash
1897C<%utf8::ToLower>.
0134edef 1898
760c7c2f
KW
1899Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1900unless those are turned off.
1901
0134edef 1902=cut */
6b5c0936 1903
2104c8d9 1904UV
9a957fbc
AL
1905Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1906 SV **swashp, const char *normal, const char *special)
a0ed51b3 1907{
b9992569
KW
1908 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1909
1910 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
1911}
1912
1913 /* change namve uv1 to 'from' */
1914UV
1915S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
1916 SV **swashp, const char *normal, const char *special)
1917{
0134edef 1918 STRLEN len = 0;
7918f24d 1919
b9992569 1920 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 1921
36eaa811
KW
1922 /* For code points that don't change case, we already know that the output
1923 * of this function is the unchanged input, so we can skip doing look-ups
1924 * for them. Unfortunately the case-changing code points are scattered
1925 * around. But there are some long consecutive ranges where there are no
1926 * case changing code points. By adding tests, we can eliminate the lookup
1927 * for all the ones in such ranges. This is currently done here only for
1928 * just a few cases where the scripts are in common use in modern commerce
1929 * (and scripts adjacent to those which can be included without additional
1930 * tests). */
1931
1932 if (uv1 >= 0x0590) {
1933 /* This keeps from needing further processing the code points most
1934 * likely to be used in the following non-cased scripts: Hebrew,
1935 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
1936 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
1937 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
1938 if (uv1 < 0x10A0) {
1939 goto cases_to_self;
1940 }
1941
1942 /* The following largish code point ranges also don't have case
1943 * changes, but khw didn't think they warranted extra tests to speed
1944 * them up (which would slightly slow down everything else above them):
1945 * 1100..139F Hangul Jamo, Ethiopic
1946 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
1947 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
1948 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
1949 * Combining Diacritical Marks Extended, Balinese,
1950 * Sundanese, Batak, Lepcha, Ol Chiki
1951 * 2000..206F General Punctuation
1952 */
1953
1954 if (uv1 >= 0x2D30) {
1955
1956 /* This keeps the from needing further processing the code points
1957 * most likely to be used in the following non-cased major scripts:
1958 * CJK, Katakana, Hiragana, plus some less-likely scripts.
1959 *
1960 * (0x2D30 above might have to be changed to 2F00 in the unlikely
1961 * event that Unicode eventually allocates the unused block as of
1962 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
1963 * that the test suite will start having failures to alert you
1964 * should that happen) */
1965 if (uv1 < 0xA640) {
1966 goto cases_to_self;
1967 }
1968
1969 if (uv1 >= 0xAC00) {
1970 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
1971 if (ckWARN_d(WARN_SURROGATE)) {
1972 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1973 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1974 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1975 }
1976 goto cases_to_self;
1977 }
36eaa811
KW
1978
1979 /* AC00..FAFF Catches Hangul syllables and private use, plus
1980 * some others */
1981 if (uv1 < 0xFB00) {
1982 goto cases_to_self;
1983
1984 }
1985
5af9bc97
KW
1986 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
1987 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1988 && ckWARN_d(WARN_DEPRECATED))
1989 {
1990 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1991 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1992 }
1993 if (ckWARN_d(WARN_NON_UNICODE)) {
1994 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1995 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1996 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1997 }
1998 goto cases_to_self;
1999 }
3bfc1e70
KW
2000#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2001 if (UNLIKELY(uv1
2002 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2003 {
2004
2005 /* As of this writing, this means we avoid swash creation
2006 * for anything beyond low Plane 1 */
2007 goto cases_to_self;
2008 }
2009#endif
36eaa811
KW
2010 }
2011 }
9ae3ac1a 2012
36eaa811
KW
2013 /* Note that non-characters are perfectly legal, so no warning should
2014 * be given. There are so few of them, that it isn't worth the extra
2015 * tests to avoid swash creation */
9ae3ac1a
KW
2016 }
2017
0134edef 2018 if (!*swashp) /* load on-demand */
5ab9d2ef 2019 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2020
a6f87d8c 2021 if (special) {
0134edef 2022 /* It might be "special" (sometimes, but not always,
2a37f04d 2023 * a multicharacter mapping) */
4a8240a3 2024 HV *hv = NULL;
b08cf34e
JH
2025 SV **svp;
2026
4a8240a3
KW
2027 /* If passed in the specials name, use that; otherwise use any
2028 * given in the swash */
2029 if (*special != '\0') {
2030 hv = get_hv(special, 0);
2031 }
2032 else {
2033 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2034 if (svp) {
2035 hv = MUTABLE_HV(SvRV(*svp));
2036 }
2037 }
2038
176fe009 2039 if (hv
5f560d8a 2040 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2041 && (*svp))
2042 {
cfd0369c 2043 const char *s;
47654450 2044
cfd0369c 2045 s = SvPV_const(*svp, len);
47654450 2046 if (len == 1)
f4cd282c 2047 /* EIGHTBIT */
c80e42f3 2048 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2049 else {
d2dcd0fb 2050 Copy(s, ustrp, len, U8);
29e98929 2051 }
983ffd37 2052 }
0134edef
JH
2053 }
2054
2055 if (!len && *swashp) {
4a4088c4 2056 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2057
0134edef
JH
2058 if (uv2) {
2059 /* It was "normal" (a single character mapping). */
f4cd282c 2060 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2061 }
2062 }
1feea2c7 2063
cbe07460
KW
2064 if (len) {
2065 if (lenp) {
2066 *lenp = len;
2067 }
2068 return valid_utf8_to_uvchr(ustrp, 0);
2069 }
2070
2071 /* Here, there was no mapping defined, which means that the code point maps
2072 * to itself. Return the inputs */
e24dfe9c 2073 cases_to_self:
bfdf22ec 2074 len = UTF8SKIP(p);
ca9fab46
KW
2075 if (p != ustrp) { /* Don't copy onto itself */
2076 Copy(p, ustrp, len, U8);
2077 }
0134edef 2078
2a37f04d
JH
2079 if (lenp)
2080 *lenp = len;
2081
f4cd282c 2082 return uv1;
cbe07460 2083
a0ed51b3
LW
2084}
2085
051a06d4 2086STATIC UV
357aadde 2087S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2088{
4a4088c4 2089 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2090 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2091 * result contains a character that crosses the 255/256 boundary, disallow
2092 * the change, and return the original code point. See L<perlfunc/lc> for
2093 * why;
051a06d4 2094 *
a1433954
KW
2095 * p points to the original string whose case was changed; assumed
2096 * by this routine to be well-formed
051a06d4
KW
2097 * result the code point of the first character in the changed-case string
2098 * ustrp points to the changed-case string (<result> represents its first char)
2099 * lenp points to the length of <ustrp> */
2100
2101 UV original; /* To store the first code point of <p> */
2102
2103 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2104
a4f12ed7 2105 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2106
2107 /* We know immediately if the first character in the string crosses the
2108 * boundary, so can skip */
2109 if (result > 255) {
2110
2111 /* Look at every character in the result; if any cross the
2112 * boundary, the whole thing is disallowed */
2113 U8* s = ustrp + UTF8SKIP(ustrp);
2114 U8* e = ustrp + *lenp;
2115 while (s < e) {
a4f12ed7 2116 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2117 goto bad_crossing;
2118 }
2119 s += UTF8SKIP(s);
2120 }
2121
613abc6d
KW
2122 /* Here, no characters crossed, result is ok as-is, but we warn. */
2123 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2124 return result;
2125 }
2126
7b52d656 2127 bad_crossing:
051a06d4
KW
2128
2129 /* Failed, have to return the original */
4b88fb76 2130 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2131
2132 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2133 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2134 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2135 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2136 OP_DESC(PL_op),
ab0b796c
KW
2137 original,
2138 original);
051a06d4
KW
2139 Copy(p, ustrp, *lenp, char);
2140 return original;
2141}
2142
d3e79532 2143/*
87cea99e 2144=for apidoc to_utf8_upper
d3e79532 2145
1f607577 2146Instead use L</toUPPER_utf8>.
a1433954 2147
d3e79532
JH
2148=cut */
2149
051a06d4 2150/* Not currently externally documented, and subject to change:
31f05a37
KW
2151 * <flags> is set iff iff the rules from the current underlying locale are to
2152 * be used. */
051a06d4 2153
2104c8d9 2154UV
31f05a37 2155Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2156{
051a06d4
KW
2157 UV result;
2158
2159 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2160
780fcc9f
KW
2161 if (flags) {
2162 /* Treat a UTF-8 locale as not being in locale at all */
2163 if (IN_UTF8_CTYPE_LOCALE) {
2164 flags = FALSE;
2165 }
2166 else {
2167 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2168 }
31f05a37
KW
2169 }
2170
3a4c58c9 2171 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2172 if (flags) {
2173 result = toUPPER_LC(*p);
2174 }
2175 else {
81c6c7ce 2176 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2177 }
3a4c58c9
KW
2178 }
2179 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2180 if (flags) {
a62b247b 2181 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2182 result = toUPPER_LC(c);
051a06d4
KW
2183 }
2184 else {
a62b247b 2185 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2186 ustrp, lenp, 'S');
051a06d4
KW
2187 }
2188 }
4a4088c4 2189 else { /* UTF-8, ord above 255 */
b9992569 2190 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2191
2192 if (flags) {
357aadde 2193 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2194 }
2195 return result;
2196 }
2197
4a4088c4 2198 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2199 if (UTF8_IS_INVARIANT(result)) {
2200 *ustrp = (U8) result;
2201 *lenp = 1;
2202 }
2203 else {
62cb07ea
KW
2204 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2205 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2206 *lenp = 2;
3a4c58c9 2207 }
baa60164 2208
051a06d4 2209 return result;
983ffd37 2210}
a0ed51b3 2211
d3e79532 2212/*
87cea99e 2213=for apidoc to_utf8_title
d3e79532 2214
1f607577 2215Instead use L</toTITLE_utf8>.
a1433954 2216
d3e79532
JH
2217=cut */
2218
051a06d4 2219/* Not currently externally documented, and subject to change:
31f05a37
KW
2220 * <flags> is set iff the rules from the current underlying locale are to be
2221 * used. Since titlecase is not defined in POSIX, for other than a
2222 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2223 */
051a06d4 2224
983ffd37 2225UV
31f05a37 2226Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2227{
051a06d4
KW
2228 UV result;
2229
2230 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2231
780fcc9f
KW
2232 if (flags) {
2233 /* Treat a UTF-8 locale as not being in locale at all */
2234 if (IN_UTF8_CTYPE_LOCALE) {
2235 flags = FALSE;
2236 }
2237 else {
2238 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2239 }
31f05a37
KW
2240 }
2241
3a4c58c9 2242 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2243 if (flags) {
2244 result = toUPPER_LC(*p);
2245 }
2246 else {
81c6c7ce 2247 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2248 }
3a4c58c9
KW
2249 }
2250 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2251 if (flags) {
a62b247b 2252 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2253 result = toUPPER_LC(c);
051a06d4
KW
2254 }
2255 else {
a62b247b 2256 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2257 ustrp, lenp, 's');
051a06d4
KW
2258 }
2259 }
4a4088c4 2260 else { /* UTF-8, ord above 255 */
b9992569 2261 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2262
2263 if (flags) {
357aadde 2264 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2265 }
2266 return result;
2267 }
2268
4a4088c4 2269 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2270 if (UTF8_IS_INVARIANT(result)) {
2271 *ustrp = (U8) result;
2272 *lenp = 1;
2273 }
2274 else {
62cb07ea
KW
2275 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2276 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2277 *lenp = 2;
3a4c58c9
KW
2278 }
2279
051a06d4 2280 return result;
a0ed51b3
LW
2281}
2282
d3e79532 2283/*
87cea99e 2284=for apidoc to_utf8_lower
d3e79532 2285
1f607577 2286Instead use L</toLOWER_utf8>.
a1433954 2287
d3e79532
JH
2288=cut */
2289
051a06d4 2290/* Not currently externally documented, and subject to change:
31f05a37
KW
2291 * <flags> is set iff iff the rules from the current underlying locale are to
2292 * be used.
2293 */
051a06d4 2294
2104c8d9 2295UV
31f05a37 2296Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2297{
051a06d4
KW
2298 UV result;
2299
051a06d4 2300 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2301
780fcc9f
KW
2302 if (flags) {
2303 /* Treat a UTF-8 locale as not being in locale at all */
2304 if (IN_UTF8_CTYPE_LOCALE) {
2305 flags = FALSE;
2306 }
2307 else {
2308 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2309 }
31f05a37
KW
2310 }
2311
968c5e6a 2312 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2313 if (flags) {
2314 result = toLOWER_LC(*p);
2315 }
2316 else {
81c6c7ce 2317 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2318 }
968c5e6a
KW
2319 }
2320 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2321 if (flags) {
a62b247b 2322 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2323 result = toLOWER_LC(c);
051a06d4
KW
2324 }
2325 else {
a62b247b 2326 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2327 ustrp, lenp);
051a06d4 2328 }
968c5e6a 2329 }
4a4088c4 2330 else { /* UTF-8, ord above 255 */
b9992569 2331 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2332
2333 if (flags) {
357aadde 2334 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2335 }
968c5e6a 2336
051a06d4
KW
2337 return result;
2338 }
2339
4a4088c4 2340 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2341 if (UTF8_IS_INVARIANT(result)) {
2342 *ustrp = (U8) result;
2343 *lenp = 1;
2344 }
2345 else {
62cb07ea
KW
2346 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2347 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2348 *lenp = 2;
2349 }
2350
051a06d4 2351 return result;
b4e400f9
JH
2352}
2353
d3e79532 2354/*
87cea99e 2355=for apidoc to_utf8_fold
d3e79532 2356
1f607577 2357Instead use L</toFOLD_utf8>.
a1433954 2358
d3e79532
JH
2359=cut */
2360
051a06d4
KW
2361/* Not currently externally documented, and subject to change,
2362 * in <flags>
31f05a37
KW
2363 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2364 * locale are to be used.
051a06d4
KW
2365 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2366 * otherwise simple folds
a0270393
KW
2367 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2368 * prohibited
445bf929 2369 */
36bb2ab6 2370
b4e400f9 2371UV
445bf929 2372Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2373{
051a06d4
KW
2374 UV result;
2375
36bb2ab6 2376 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2377
a0270393
KW
2378 /* These are mutually exclusive */
2379 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2380
50ba90ff
KW
2381 assert(p != ustrp); /* Otherwise overwrites */
2382
780fcc9f
KW
2383 if (flags & FOLD_FLAGS_LOCALE) {
2384 /* Treat a UTF-8 locale as not being in locale at all */
2385 if (IN_UTF8_CTYPE_LOCALE) {
2386 flags &= ~FOLD_FLAGS_LOCALE;
2387 }
2388 else {
2389 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2390 }
31f05a37
KW
2391 }
2392
a1dde8de 2393 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2394 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2395 result = toFOLD_LC(*p);
051a06d4
KW
2396 }
2397 else {
81c6c7ce 2398 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2399 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2400 }
a1dde8de
KW
2401 }
2402 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2403 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2404 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2405 result = toFOLD_LC(c);
051a06d4
KW
2406 }
2407 else {
a62b247b 2408 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2409 ustrp, lenp,
1ca267a5 2410 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2411 }
a1dde8de 2412 }
4a4088c4 2413 else { /* UTF-8, ord above 255 */
b9992569 2414 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2415
1ca267a5
KW
2416 if (flags & FOLD_FLAGS_LOCALE) {
2417
76f2ffcd 2418# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2419 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2420
2421# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2422# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2423
2424 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2425
538e84ed
KW
2426 /* Special case these two characters, as what normally gets
2427 * returned under locale doesn't work */
76f2ffcd
KW
2428 if (UTF8SKIP(p) == cap_sharp_s_len
2429 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2430 {
ab0b796c
KW
2431 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2432 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2433 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2434 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2435 goto return_long_s;
2436 }
0766489e
KW
2437 else
2438#endif
2439 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2440 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2441 {
ab0b796c
KW
2442 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2443 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2444 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2445 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2446 goto return_ligature_st;
2447 }
74894415
KW
2448
2449#if UNICODE_MAJOR_VERSION == 3 \
2450 && UNICODE_DOT_VERSION == 0 \
2451 && UNICODE_DOT_DOT_VERSION == 1
2452# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2453
2454 /* And special case this on this Unicode version only, for the same
2455 * reaons the other two are special cased. They would cross the
2456 * 255/256 boundary which is forbidden under /l, and so the code
2457 * wouldn't catch that they are equivalent (which they are only in
2458 * this release) */
2459 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2460 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2461 {
2462 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2463 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2464 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2465 "resolved to \"\\x{0131}\".");
2466 goto return_dotless_i;
2467 }
2468#endif
2469
357aadde 2470 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2471 }
a0270393
KW
2472 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2473 return result;
2474 }
2475 else {
4a4088c4 2476 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2477 * character above the ASCII range, and the result should not
2478 * contain an ASCII character. */
a0270393
KW
2479
2480 UV original; /* To store the first code point of <p> */
2481
2482 /* Look at every character in the result; if any cross the
2483 * boundary, the whole thing is disallowed */
2484 U8* s = ustrp;
2485 U8* e = ustrp + *lenp;
2486 while (s < e) {
2487 if (isASCII(*s)) {
2488 /* Crossed, have to return the original */
2489 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2490
9fc2026f 2491 /* But in these instances, there is an alternative we can
1ca267a5 2492 * return that is valid */
0766489e
KW
2493 if (original == LATIN_SMALL_LETTER_SHARP_S
2494#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2495 || original == LATIN_CAPITAL_LETTER_SHARP_S
2496#endif
2497 ) {
1ca267a5
KW
2498 goto return_long_s;
2499 }
9fc2026f
KW
2500 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2501 goto return_ligature_st;
2502 }
74894415
KW
2503#if UNICODE_MAJOR_VERSION == 3 \
2504 && UNICODE_DOT_VERSION == 0 \
2505 && UNICODE_DOT_DOT_VERSION == 1
2506
2507 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2508 goto return_dotless_i;
2509 }
2510#endif
a0270393
KW
2511 Copy(p, ustrp, *lenp, char);
2512 return original;
2513 }
2514 s += UTF8SKIP(s);
2515 }
051a06d4 2516
a0270393
KW
2517 /* Here, no characters crossed, result is ok as-is */
2518 return result;
2519 }
051a06d4
KW
2520 }
2521
4a4088c4 2522 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2523 if (UTF8_IS_INVARIANT(result)) {
2524 *ustrp = (U8) result;
2525 *lenp = 1;
2526 }
2527 else {
62cb07ea
KW
2528 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2529 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2530 *lenp = 2;
2531 }
2532
051a06d4 2533 return result;
1ca267a5
KW
2534
2535 return_long_s:
2536 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2537 * folds to a string of two of these characters. By returning this
2538 * instead, then, e.g.,
2539 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2540 * works. */
2541
2542 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2543 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2544 ustrp, *lenp, U8);
2545 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2546
2547 return_ligature_st:
2548 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2549 * have the other one fold to it */
2550
2551 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2552 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2553 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2554
2555#if UNICODE_MAJOR_VERSION == 3 \
2556 && UNICODE_DOT_VERSION == 0 \
2557 && UNICODE_DOT_DOT_VERSION == 1
2558
2559 return_dotless_i:
2560 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2561 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2562 return LATIN_SMALL_LETTER_DOTLESS_I;
2563
2564#endif
2565
a0ed51b3
LW
2566}
2567
711a919c 2568/* Note:
f90a9a02 2569 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2570 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2571 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2572 */
c4a5db0c 2573
a0ed51b3 2574SV*
7fc63493 2575Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2576{
c4a5db0c
KW
2577 PERL_ARGS_ASSERT_SWASH_INIT;
2578
2579 /* Returns a copy of a swash initiated by the called function. This is the
2580 * public interface, and returning a copy prevents others from doing
2581 * mischief on the original */
2582
5d3d13d1 2583 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2584}
2585
2586SV*
5d3d13d1 2587Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2588{
2c1f00b9
YO
2589
2590 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2591 * use the following define */
2592
2593#define CORE_SWASH_INIT_RETURN(x) \
2594 PL_curpm= old_PL_curpm; \
2595 return x
2596
c4a5db0c 2597 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2598 * by calling utf8_heavy.pl in the general case. The returned value may be
2599 * the swash's inversion list instead if the input parameters allow it.
2600 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2601 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2602 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2603 *
2604 * This interface should only be used by functions that won't destroy or
2605 * adversely change the swash, as doing so affects all other uses of the
2606 * swash in the program; the general public should use 'Perl_swash_init'
2607 * instead.
2608 *
2609 * pkg is the name of the package that <name> should be in.
2610 * name is the name of the swash to find. Typically it is a Unicode
2611 * property name, including user-defined ones
2612 * listsv is a string to initialize the swash with. It must be of the form
2613 * documented as the subroutine return value in
2614 * L<perlunicode/User-Defined Character Properties>
2615 * minbits is the number of bits required to represent each data element.
2616 * It is '1' for binary properties.
2617 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2618 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2619 * flags_p if non-NULL is the address of various input and output flag bits
2620 * to the routine, as follows: ('I' means is input to the routine;
2621 * 'O' means output from the routine. Only flags marked O are
2622 * meaningful on return.)
2623 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2624 * came from a user-defined property. (I O)
5d3d13d1
KW
2625 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2626 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2627 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2628 * return of an inversion list instead of a swash hash if this routine
2629 * thinks that would result in faster execution of swash_fetch() later
2630 * on. (I)
9a53f6cf
KW
2631 *
2632 * Thus there are three possible inputs to find the swash: <name>,
2633 * <listsv>, and <invlist>. At least one must be specified. The result
2634 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2635 * actions can intersect, etc. what <name> gives. To avoid going out to
2636 * disk at all, <invlist> should specify completely what the swash should
2637 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2638 *
2639 * <invlist> is only valid for binary properties */
c4a5db0c 2640
2c1f00b9
YO
2641 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2642
c4a5db0c 2643 SV* retval = &PL_sv_undef;
83199d38 2644 HV* swash_hv = NULL;
87367d5f
KW
2645 const int invlist_swash_boundary =
2646 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2647 ? 512 /* Based on some benchmarking, but not extensive, see commit
2648 message */
2649 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2650
2651 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2652 assert(! invlist || minbits == 1);
2653
2c1f00b9
YO
2654 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2655 that triggered the swash init and the swash init perl logic itself.
2656 See perl #122747 */
2657
9a53f6cf
KW
2658 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2659 * so */
2660 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2661 dSP;
2662 const size_t pkg_len = strlen(pkg);
2663 const size_t name_len = strlen(name);
2664 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2665 SV* errsv_save;
2666 GV *method;
2667
2668 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2669
2670 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2671 ENTER;
69794297 2672 SAVEHINTS();
2782061f 2673 save_re_context();
650f067c
JL
2674 /* We might get here via a subroutine signature which uses a utf8
2675 * parameter name, at which point PL_subname will have been set
2676 * but not yet used. */
2677 save_item(PL_subname);
69794297
KW
2678 if (PL_parser && PL_parser->error_count)
2679 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2680 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2681 if (!method) { /* demand load UTF-8 */
69794297 2682 ENTER;
db2c6cb3
FC
2683 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2684 GvSV(PL_errgv) = NULL;
1a419e6b 2685#ifndef NO_TAINT_SUPPORT
69794297
KW
2686 /* It is assumed that callers of this routine are not passing in
2687 * any user derived data. */
2782061f
DM
2688 /* Need to do this after save_re_context() as it will set
2689 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2690 * in Perl_magic_get). Even line to create errsv_save can turn on
2691 * PL_tainted. */
284167a5
SM
2692 SAVEBOOL(TAINT_get);
2693 TAINT_NOT;
2694#endif
69794297
KW
2695 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2696 NULL);
eed484f9 2697 {
db2c6cb3
FC
2698 /* Not ERRSV, as there is no need to vivify a scalar we are
2699 about to discard. */
2700 SV * const errsv = GvSV(PL_errgv);
2701 if (!SvTRUE(errsv)) {
2702 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2703 SvREFCNT_dec(errsv);
2704 }
eed484f9 2705 }
69794297
KW
2706 LEAVE;
2707 }
2708 SPAGAIN;
2709 PUSHMARK(SP);
2710 EXTEND(SP,5);
2711 mPUSHp(pkg, pkg_len);
2712 mPUSHp(name, name_len);
2713 PUSHs(listsv);
2714 mPUSHi(minbits);
2715 mPUSHi(none);
2716 PUTBACK;
db2c6cb3
FC
2717 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2718 GvSV(PL_errgv) = NULL;
69794297
KW
2719 /* If we already have a pointer to the method, no need to use
2720 * call_method() to repeat the lookup. */
c41800a8
KW
2721 if (method
2722 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2723 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2724 {
2725 retval = *PL_stack_sp--;
2726 SvREFCNT_inc(retval);
2727 }
eed484f9 2728 {
db2c6cb3
FC
2729 /* Not ERRSV. See above. */
2730 SV * const errsv = GvSV(PL_errgv);
2731 if (!SvTRUE(errsv)) {
2732 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2733 SvREFCNT_dec(errsv);
2734 }
eed484f9 2735 }
ce3b816e 2736 LEAVE;
69794297
KW
2737 POPSTACK;
2738 if (IN_PERL_COMPILETIME) {
2739 CopHINTS_set(PL_curcop, PL_hints);
2740 }
2741 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
2742 if (SvPOK(retval))
2743
2744 /* If caller wants to handle missing properties, let them */
5d3d13d1 2745 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2746 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2747 }
2748 Perl_croak(aTHX_
2749 "Can't find Unicode property definition \"%"SVf"\"",
2750 SVfARG(retval));
a25b5927 2751 NOT_REACHED; /* NOTREACHED */
69794297 2752 }
9a53f6cf 2753 } /* End of calling the module to find the swash */
36eb48b4 2754
83199d38
KW
2755 /* If this operation fetched a swash, and we will need it later, get it */
2756 if (retval != &PL_sv_undef
2757 && (minbits == 1 || (flags_p
2758 && ! (*flags_p
2759 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2760 {
2761 swash_hv = MUTABLE_HV(SvRV(retval));
2762
2763 /* If we don't already know that there is a user-defined component to
2764 * this swash, and the user has indicated they wish to know if there is
2765 * one (by passing <flags_p>), find out */
2766 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2767 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2768 if (user_defined && SvUV(*user_defined)) {
2769 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2770 }
2771 }
2772 }
2773
36eb48b4
KW
2774 /* Make sure there is an inversion list for binary properties */
2775 if (minbits == 1) {
2776 SV** swash_invlistsvp = NULL;
2777 SV* swash_invlist = NULL;
9a53f6cf 2778 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2779 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2780 an unclaimed reference count */
36eb48b4 2781
9a53f6cf 2782 /* If this operation fetched a swash, get its already existing
83199d38 2783 * inversion list, or create one for it */
36eb48b4 2784
83199d38 2785 if (swash_hv) {
5c9f4bd2 2786 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2787 if (swash_invlistsvp) {
2788 swash_invlist = *swash_invlistsvp;
2789 invlist_in_swash_is_valid = TRUE;
2790 }
2791 else {
36eb48b4 2792 swash_invlist = _swash_to_invlist(retval);
02c85471 2793 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2794 }
2795 }
2796
2797 /* If an inversion list was passed in, have to include it */
2798 if (invlist) {
2799
2800 /* Any fetched swash will by now have an inversion list in it;
2801 * otherwise <swash_invlist> will be NULL, indicating that we
2802 * didn't fetch a swash */
2803 if (swash_invlist) {
2804
2805 /* Add the passed-in inversion list, which invalidates the one
2806 * already stored in the swash */
2807 invlist_in_swash_is_valid = FALSE;
2808 _invlist_union(invlist, swash_invlist, &swash_invlist);
2809 }
2810 else {
2811
87367d5f
KW
2812 /* Here, there is no swash already. Set up a minimal one, if
2813 * we are going to return a swash */
2814 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2815 swash_hv = newHV();
4aca0fe6 2816 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2817 }
9a53f6cf
KW
2818 swash_invlist = invlist;
2819 }
9a53f6cf
KW
2820 }
2821
2822 /* Here, we have computed the union of all the passed-in data. It may
2823 * be that there was an inversion list in the swash which didn't get
538e84ed 2824 * touched; otherwise save the computed one */
87367d5f
KW
2825 if (! invlist_in_swash_is_valid
2826 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2827 {
5c9f4bd2 2828 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2829 {
2830 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2831 }
cc34d8c5
FC
2832 /* We just stole a reference count. */
2833 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2834 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2835 }
87367d5f 2836
dbfdbd26
KW
2837 SvREADONLY_on(swash_invlist);
2838
c41800a8 2839 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2840 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2841 SvREFCNT_dec(retval);
02c85471
FC
2842 if (!swash_invlist_unclaimed)
2843 SvREFCNT_inc_simple_void_NN(swash_invlist);
2844 retval = newRV_noinc(swash_invlist);
87367d5f 2845 }
36eb48b4
KW
2846 }
2847
2c1f00b9
YO
2848 CORE_SWASH_INIT_RETURN(retval);
2849#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2850}
2851
035d37be
JH
2852
2853/* This API is wrong for special case conversions since we may need to
2854 * return several Unicode characters for a single Unicode character
2855 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2856 * the lower-level routine, and it is similarly broken for returning
38684baa 2857 * multiple values. --jhi
b9992569 2858 * For those, you should use S__to_utf8_case() instead */
b0e3252e 2859/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2860
2861/* Note:
2862 * Returns the value of property/mapping C<swash> for the first character
2863 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2864 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2865 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2866 *
2867 * A "swash" is a hash which contains initially the keys/values set up by
2868 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2869 * property for all possible code points. Things are stored in a compact form
2870 * (see utf8_heavy.pl) so that calculation is required to find the actual
2871 * property value for a given code point. As code points are looked up, new
2872 * key/value pairs are added to the hash, so that the calculation doesn't have
2873 * to ever be re-done. Further, each calculation is done, not just for the
2874 * desired one, but for a whole block of code points adjacent to that one.
2875 * For binary properties on ASCII machines, the block is usually for 64 code
2876 * points, starting with a code point evenly divisible by 64. Thus if the
2877 * property value for code point 257 is requested, the code goes out and
2878 * calculates the property values for all 64 code points between 256 and 319,
2879 * and stores these as a single 64-bit long bit vector, called a "swatch",
2880 * under the key for code point 256. The key is the UTF-8 encoding for code
2881 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2882 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2883 * for code point 258 is then requested, this code realizes that it would be
2884 * stored under the key for 256, and would find that value and extract the
2885 * relevant bit, offset from 256.
2886 *
2887 * Non-binary properties are stored in as many bits as necessary to represent
2888 * their values (32 currently, though the code is more general than that), not
2889 * as single bits, but the principal is the same: the value for each key is a
2890 * vector that encompasses the property values for all code points whose UTF-8
2891 * representations are represented by the key. That is, for all code points
2892 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2893 * bytes of that.
680c470c 2894 */
a0ed51b3 2895UV
680c470c 2896Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2897{
ef8f7699 2898 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2899 U32 klen;
2900 U32 off;
9b56a019 2901 STRLEN slen = 0;
7d85a32c 2902 STRLEN needents;
cfd0369c 2903 const U8 *tmps = NULL;
979f2922 2904 SV *swatch;
08fb1ac5 2905 const U8 c = *ptr;
3568d838 2906
7918f24d
NC
2907 PERL_ARGS_ASSERT_SWASH_FETCH;
2908
87367d5f
KW
2909 /* If it really isn't a hash, it isn't really swash; must be an inversion
2910 * list */
2911 if (SvTYPE(hv) != SVt_PVHV) {
2912 return _invlist_contains_cp((SV*)hv,
2913 (do_utf8)
2914 ? valid_utf8_to_uvchr(ptr, NULL)
2915 : c);
2916 }
2917
08fb1ac5
KW
2918 /* We store the values in a "swatch" which is a vec() value in a swash
2919 * hash. Code points 0-255 are a single vec() stored with key length
2920 * (klen) 0. All other code points have a UTF-8 representation
2921 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2922 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2923 * length for them is the length of the encoded char - 1. ptr[klen] is the
2924 * final byte in the sequence representing the character */
2925 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2926 klen = 0;
2927 needents = 256;
2928 off = c;
3568d838 2929 }
08fb1ac5
KW
2930 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2931 klen = 0;
2932 needents = 256;
a62b247b 2933 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2934 }
2935 else {
08fb1ac5
KW
2936 klen = UTF8SKIP(ptr) - 1;
2937
2938 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2939 * the vec is the final byte in the sequence. (In EBCDIC this is
2940 * converted to I8 to get consecutive values.) To help you visualize
2941 * all this:
2942 * Straight 1047 After final byte
2943 * UTF-8 UTF-EBCDIC I8 transform
2944 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2945 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2946 * ...
2947 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2948 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2949 * ...
2950 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2951 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2952 * ...
2953 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2954 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2955 * ...
2956 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2957 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2958 *
2959 * (There are no discontinuities in the elided (...) entries.)
2960 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2961 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2962 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2963 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2964 * index into the vec() swatch (after subtracting 0x80, which we
2965 * actually do with an '&').
2966 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2967 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2968 * dicontinuities which go away by transforming it into I8, and we
2969 * effectively subtract 0xA0 to get the index. */
979f2922 2970 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2971 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2972 }
7d85a32c 2973
a0ed51b3 2974 /*
4a4088c4 2975 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2976 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2977 * it's nothing to sniff at.) Pity we usually come through at least
2978 * two function calls to get here...
2979 *
2980 * NB: this code assumes that swatches are never modified, once generated!
2981 */
2982
3568d838 2983 if (hv == PL_last_swash_hv &&
a0ed51b3 2984 klen == PL_last_swash_klen &&
27da23d5 2985 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2986 {
2987 tmps = PL_last_swash_tmps;
2988 slen = PL_last_swash_slen;
2989 }
2990 else {
2991 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2992 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2993
b0e3252e 2994 /* If not cached, generate it via swatch_get */
979f2922 2995 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2996 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2997 {
2998 if (klen) {
2999 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3000 swatch = swatch_get(swash,
3001 code_point & ~((UV)needents - 1),
3002 needents);
3003 }
3004 else { /* For the first 256 code points, the swatch has a key of
3005 length 0 */
3006 swatch = swatch_get(swash, 0, needents);
3007 }
979f2922 3008
923e4eb5 3009 if (IN_PERL_COMPILETIME)
623e6609 3010 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3011
979f2922 3012 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3013
979f2922
ST
3014 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3015 || (slen << 3) < needents)
5637ef5b
NC
3016 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3017 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3018 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3019 }
3020
3021 PL_last_swash_hv = hv;
16d8f38a 3022 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3023 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3024 /* FIXME change interpvar.h? */
3025 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3026 PL_last_swash_slen = slen;
3027 if (klen)
3028 Copy(ptr, PL_last_swash_key, klen, U8);
3029 }
3030
9faf8d75 3031 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3032 case 1:
e7aca353 3033 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3034 case 8:
e7aca353 3035 return ((UV) tmps[off]);
a0ed51b3
LW
3036 case 16:
3037 off <<= 1;
e7aca353
JH
3038 return
3039 ((UV) tmps[off ] << 8) +
3040 ((UV) tmps[off + 1]);
a0ed51b3
LW
3041 case 32:
3042 off <<= 2;
e7aca353
JH
3043 return
3044 ((UV) tmps[off ] << 24) +
3045 ((UV) tmps[off + 1] << 16) +
3046 ((UV) tmps[off + 2] << 8) +
3047 ((UV) tmps[off + 3]);
a0ed51b3 3048 }
5637ef5b
NC
3049 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3050 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3051 NORETURN_FUNCTION_END;
a0ed51b3 3052}
2b9d42f0 3053
319009ee
KW
3054/* Read a single line of the main body of the swash input text. These are of
3055 * the form:
3056 * 0053 0056 0073
3057 * where each number is hex. The first two numbers form the minimum and
3058 * maximum of a range, and the third is the value associated with the range.
3059 * Not all swashes should have a third number
3060 *
3061 * On input: l points to the beginning of the line to be examined; it points
3062 * to somewhere in the string of the whole input text, and is
3063 * terminated by a \n or the null string terminator.
3064 * lend points to the null terminator of that string
3065 * wants_value is non-zero if the swash expects a third number
3066 * typestr is the name of the swash's mapping, like 'ToLower'
3067 * On output: *min, *max, and *val are set to the values read from the line.
3068 * returns a pointer just beyond the line examined. If there was no
3069 * valid min number on the line, returns lend+1
3070 */
3071
3072STATIC U8*
3073S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3074 const bool wants_value, const U8* const typestr)
3075{
3076 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3077 STRLEN numlen; /* Length of the number */
02470786
KW
3078 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3079 | PERL_SCAN_DISALLOW_PREFIX
3080 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3081
3082 /* nl points to the next \n in the scan */
3083 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3084
95543e92
KW
3085 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3086
319009ee
KW
3087 /* Get the first number on the line: the range minimum */
3088 numlen = lend - l;
3089 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 3090 *max = *min; /* So can never return without setting max */
319009ee
KW
3091 if (numlen) /* If found a hex number, position past it */
3092 l += numlen;
3093 else if (nl) { /* Else, go handle next line, if any */
3094 return nl + 1; /* 1 is length of "\n" */
3095 }
3096 else { /* Else, no next line */
3097 return lend + 1; /* to LIST's end at which \n is not found */
3098 }
3099
3100 /* The max range value follows, separated by a BLANK */
3101 if (isBLANK(*l)) {
3102 ++l;
02470786
KW
3103 flags = PERL_SCAN_SILENT_ILLDIGIT
3104 | PERL_SCAN_DISALLOW_PREFIX
3105 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3106 numlen = lend - l;
3107 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3108 if (numlen)
3109 l += numlen;
3110 else /* If no value here, it is a single element range */
3111 *max = *min;
3112
3113 /* Non-binary tables have a third entry: what the first element of the
24303724 3114 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3115 if (wants_value) {
3116 if (isBLANK(*l)) {
3117 ++l;
f2a7d0fc
KW
3118 flags = PERL_SCAN_SILENT_ILLDIGIT
3119 | PERL_SCAN_DISALLOW_PREFIX
3120 | PERL_SCAN_SILENT_NON_PORTABLE;
3121 numlen = lend - l;
3122 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3123 if (numlen)
3124 l += numlen;
3125 else
3126 *val = 0;
319009ee
KW
3127 }
3128 else {
3129 *val = 0;
3130 if (typeto) {
dcbac5bb 3131 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3132 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3133 typestr, l);
3134 }
3135 }
3136 }
3137 else
3138 *val = 0; /* bits == 1, then any val should be ignored */
3139 }
3140 else { /* Nothing following range min, should be single element with no
3141 mapping expected */
319009ee
KW
3142 if (wants_value) {
3143 *val = 0;
3144 if (typeto) {
dcbac5bb 3145 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3146 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3147 }
3148 }
3149 else
3150 *val = 0; /* bits == 1, then val should be ignored */
3151 }
3152
3153 /* Position to next line if any, or EOF */
3154 if (nl)
3155 l = nl + 1;
3156 else
3157 l = lend;
3158
3159 return l;
3160}
3161
979f2922
ST
3162/* Note:
3163 * Returns a swatch (a bit vector string) for a code point sequence
3164 * that starts from the value C<start> and comprises the number C<span>.
3165 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3166 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3167 */
3168STATIC SV*
b0e3252e 3169S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3170{
3171 SV *swatch;
77f9f126 3172 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3173 STRLEN lcur, xcur, scur;
ef8f7699 3174 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3175 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3176
88d45d28
KW
3177 SV** listsvp = NULL; /* The string containing the main body of the table */
3178 SV** extssvp = NULL;
3179 SV** invert_it_svp = NULL;
3180 U8* typestr = NULL;
786861f5
KW
3181 STRLEN bits;
3182 STRLEN octets; /* if bits == 1, then octets == 0 */
3183 UV none;
3184 UV end = start + span;
972dd592 3185
36eb48b4 3186 if (invlistsvp == NULL) {
786861f5
KW
3187 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3188 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3189 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3190 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3191 listsvp = hv_fetchs(hv, "LIST", FALSE);
3192 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3193
3194 bits = SvUV(*bitssvp);
3195 none = SvUV(*nonesvp);
3196 typestr = (U8*)SvPV_nolen(*typesvp);
3197 }
36eb48b4
KW
3198 else {
3199 bits = 1;
3200 none = 0;
3201 }
786861f5 3202 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3203
b0e3252e 3204 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3205
979f2922 3206 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3207 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3208 (UV)bits);
979f2922
ST
3209 }
3210
84ea5ef6
KW
3211 /* If overflowed, use the max possible */
3212 if (end < start) {
3213 end = UV_MAX;
3214 span = end - start;
3215 }
3216
979f2922 3217 /* create and initialize $swatch */
979f2922 3218 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3219 swatch = newSV(scur);
3220 SvPOK_on(swatch);
979f2922
ST
3221 s = (U8*)SvPVX(swatch);
3222 if (octets && none) {
0bd48802 3223 const U8* const e = s + scur;
979f2922
ST
3224 while (s < e) {
3225 if (bits == 8)
3226 *s++ = (U8)(none & 0xff);
3227 else if (bits == 16) {
3228 *s++ = (U8)((none >> 8) & 0xff);
3229 *s++ = (U8)( none & 0xff);
3230 }
3231 else if (bits == 32) {
3232 *s++ = (U8)((none >> 24) & 0xff);
3233 *s++ = (U8)((none >> 16) & 0xff);
3234 *s++ = (U8)((none >> 8) & 0xff);
3235 *s++ = (U8)( none & 0xff);
3236 }
3237 }
3238 *s = '\0';
3239 }
3240 else {
3241 (void)memzero((U8*)s, scur + 1);
3242 }
3243 SvCUR_set(swatch, scur);
3244 s = (U8*)SvPVX(swatch);
3245
36eb48b4
KW
3246 if (invlistsvp) { /* If has an inversion list set up use that */
3247 _invlist_populate_swatch(*invlistsvp, start, end, s);
3248 return swatch;
3249 }
3250
3251 /* read $swash->{LIST} */
979f2922
ST
3252 l = (U8*)SvPV(*listsvp, lcur);
3253 lend = l + lcur;
3254 while (l < lend) {
8ed25d53 3255 UV min, max, val, upper;
95543e92
KW
3256 l = swash_scan_list_line(l, lend, &min, &max, &val,
3257 cBOOL(octets), typestr);
319009ee 3258 if (l > lend) {
979f2922
ST
3259 break;
3260 }
3261
972dd592 3262 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3263 if (max < start)
3264 continue;
3265
8ed25d53
KW
3266 /* <end> is generally 1 beyond where we want to set things, but at the
3267 * platform's infinity, where we can't go any higher, we want to
3268 * include the code point at <end> */
3269 upper = (max < end)
3270 ? max
3271 : (max != UV_MAX || end != UV_MAX)
3272 ? end - 1
3273 : end;
3274
979f2922 3275 if (octets) {
35da51f7 3276 UV key;
979f2922
ST
3277 if (min < start) {
3278 if (!none || val < none) {
3279 val += start - min;
3280 }
3281 min = start;
3282 }
8ed25d53 3283 for (key = min; key <= upper; key++) {
979f2922 3284 STRLEN offset;
979f2922
ST
3285 /* offset must be non-negative (start <= min <= key < end) */
3286 offset = octets * (key - start);
3287 if (bits == 8)
3288 s[offset] = (U8)(val & 0xff);
3289 else if (bits == 16) {
3290 s[offset ] = (U8)((val >> 8) & 0xff);
3291 s[offset + 1] = (U8)( val & 0xff);
3292 }
3293 else if (bits == 32) {
3294 s[offset ] = (U8)((val >> 24) & 0xff);
3295 s[offset + 1] = (U8)((val >> 16) & 0xff);
3296 s[offset + 2] = (U8)((val >> 8) & 0xff);
3297 s[offset + 3] = (U8)( val & 0xff);
3298 }
3299
3300 if (!none || val < none)
3301 ++val;
3302 }
3303 }
711a919c 3304 else { /* bits == 1, then val should be ignored */
35da51f7 3305 UV key;
979f2922
ST
3306 if (min < start)
3307 min = start;
6cb05c12 3308
8ed25d53 3309 for (key = min; key <= upper; key++) {
0bd48802 3310 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3311 s[offset >> 3] |= 1 << (offset & 7);
3312 }
3313 }
3314 } /* while */
979f2922 3315
9479a769 3316 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3317 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3318
3319 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3320 * be 0, and their inversion should also be 0, as we don't succeed any
3321 * Unicode property matches for non-Unicode code points */
3322 if (start <= PERL_UNICODE_MAX) {
3323
3324 /* The code below assumes that we never cross the
3325 * Unicode/above-Unicode boundary in a range, as otherwise we would
3326 * have to figure out where to stop flipping the bits. Since this
3327 * boundary is divisible by a large power of 2, and swatches comes
3328 * in small powers of 2, this should be a valid assumption */
3329 assert(start + span - 1 <= PERL_UNICODE_MAX);
3330
507a8485
KW
3331 send = s + scur;
3332 while (s < send) {
3333 *s = ~(*s);
3334 s++;
3335 }
0bda3001 3336 }
77f9f126
KW
3337 }
3338
d73c39c5
KW
3339 /* read $swash->{EXTRAS}
3340 * This code also copied to swash_to_invlist() below */
979f2922
ST
3341 x = (U8*)SvPV(*extssvp, xcur);
3342 xend = x + xcur;
3343 while (x < xend) {
3344 STRLEN namelen;
3345 U8 *namestr;
3346 SV** othersvp;
3347 HV* otherhv;
3348 STRLEN otherbits;
3349 SV **otherbitssvp, *other;
711a919c 3350 U8 *s, *o, *nl;
979f2922
ST
3351 STRLEN slen, olen;
3352
35da51f7 3353 const U8 opc = *x++;
979f2922
ST
3354 if (opc == '\n')
3355 continue;
3356
3357 nl = (U8*)memchr(x, '\n', xend - x);
3358
3359 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3360 if (nl) {
3361 x = nl + 1; /* 1 is length of "\n" */
3362 continue;
3363 }
3364 else {
3365 x = xend; /* to EXTRAS' end at which \n is not found */
3366 break;
3367 }
3368 }
3369
3370 namestr = x;
3371 if (nl) {
3372 namelen = nl - namestr;
3373 x = nl + 1;
3374 }
3375 else {
3376 namelen = xend - namestr;
3377 x = xend;
3378 }
3379
3380 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3381 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3382 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3383 otherbits = (STRLEN)SvUV(*otherbitssvp);
3384 if (bits < otherbits)
5637ef5b
NC
3385 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3386 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3387
3388 /* The "other" swatch must be destroyed after. */
b0e3252e 3389 other = swatch_get(*othersvp, start, span);
979f2922
ST
3390 o = (U8*)SvPV(other, olen);
3391
3392 if (!olen)
b0e3252e 3393 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3394
3395 s = (U8*)SvPV(swatch, slen);
3396 if (bits == 1 && otherbits == 1) {
3397 if (slen != olen)
5637ef5b
NC
3398 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3399 "mismatch, slen=%"UVuf", olen=%"UVuf,
3400 (UV)slen, (UV)olen);
979f2922
ST
3401
3402 switch (opc) {
3403 case '+':
3404 while (slen--)
3405 *s++ |= *o++;
3406 break;
3407 case '!':
3408 while (slen--)
3409 *s++ |= ~*o++;
3410 break;
3411 case '-':
3412 while (slen--)
3413 *s++ &= ~*o++;
3414 break;
3415 case '&':
3416 while (slen--)
3417 *s++ &= *o++;
3418 break;
3419 default:
3420 break;
3421 }
3422 }
711a919c 3423 else {
979f2922
ST
3424 STRLEN otheroctets = otherbits >> 3;
3425 STRLEN offset = 0;
35da51f7 3426 U8* const send = s + slen;
979f2922
ST
3427
3428 while (s < send) {
3429 UV otherval = 0;
3430
3431 if (otherbits == 1) {
3432 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3433 ++offset;
3434 }
3435 else {
3436 STRLEN vlen = otheroctets;
3437 otherval = *o++;
3438 while (--vlen) {
3439 otherval <<= 8;
3440 otherval |= *o++;
3441 }
3442 }
3443
711a919c 3444 if (opc == '+' && otherval)
6f207bd3 3445 NOOP; /* replace with otherval */
979f2922
ST
3446 else if (opc == '!' && !otherval)
3447 otherval = 1;
3448 else if (opc == '-' && otherval)
3449 otherval = 0;
3450 else if (opc == '&' && !otherval)
3451 otherval = 0;
3452 else {
711a919c 3453 s += octets; /* no replacement */
979f2922
ST
3454 continue;
3455 }
3456
3457 if (bits == 8)
3458 *s++ = (U8)( otherval & 0xff);
3459 else if (bits == 16) {
3460 *s++ = (U8)((otherval >> 8) & 0xff);
3461 *s++ = (U8)( otherval & 0xff);
3462 }
3463 else if (bits == 32) {
3464 *s++ = (U8)((otherval >> 24) & 0xff);
3465 *s++ = (U8)((otherval >> 16) & 0xff);
3466 *s++ = (U8)((otherval >> 8) & 0xff);
3467 *s++ = (U8)( otherval & 0xff);
3468 }
3469 }
3470 }
3471 sv_free(other); /* through with it! */
3472 } /* while */
3473 return swatch;
3474}
3475
064c021d 3476HV*
4c2e1131 3477Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3478{
3479
79a2a0e8 3480 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3481 * Can't be used on a property that is subject to user override, as it
3482 * relies on the value of SPECIALS in the swash which would be set by
3483 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3484 * for overridden properties
064c021d
KW
3485 *
3486 * Returns a hash which is the inversion and closure of a swash mapping.
3487 * For example, consider the input lines:
3488 * 004B 006B
3489 * 004C 006C
3490 * 212A 006B
3491 *
4a4088c4 3492 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3493 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3494 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3495 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3496 *
538e84ed
KW
3497 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3498 * keys are only code points that are folded-to, so it isn't a full closure.
3499 *
064c021d
KW
3500 * Essentially, for any code point, it gives all the code points that map to
3501 * it, or the list of 'froms' for that point.
3502 *
5662e334
KW
3503 * Currently it ignores any additions or deletions from other swashes,
3504 * looking at just the main body of the swash, and if there are SPECIALS
3505 * in the swash, at that hash
3506 *
3507 * The specials hash can be extra code points, and most likely consists of
3508 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3509 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3510 * However consider this possible input in the specials hash:
3511 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3512 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3513 *
3514 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3515 * currently handle. But it also means that FB05 and FB06 are equivalent in
3516 * a 1-1 mapping which we should handle, and this relationship may not be in
3517 * the main table. Therefore this function examines all the multi-char
74894415
KW
3518 * sequences and adds the 1-1 mappings that come out of that.
3519 *
3520 * XXX This function was originally intended to be multipurpose, but its
3521 * only use is quite likely to remain for constructing the inversion of
3522 * the CaseFolding (//i) property. If it were more general purpose for
3523 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3524 * because certain folds are prohibited under /iaa and /il. As an example,
3525 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3526 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3527 * prohibited, so we would not figure out that they fold to each other.
3528 * Code could be written to automatically figure this out, similar to the
3529 * code that does this for multi-character folds, but this is the only case
3530 * where something like this is ever likely to happen, as all the single
7ee537e6 3531 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3532 * little special code that is compiled only for this Unicode version. This
3533 * is smaller and didn't require much coding time to do. But this makes
3534 * this routine strongly tied to being used just for CaseFolding. If ever
3535 * it should be generalized, this would have to be fixed */
064c021d
KW
3536
3537 U8 *l, *lend;
3538 STRLEN lcur;
3539 HV *const hv = MUTABLE_HV(SvRV(swash));
3540
923b6d4e
KW
3541 /* The string containing the main body of the table. This will have its
3542 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3543 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3544
3545 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3546 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3547 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3548 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3549 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3550 const STRLEN bits = SvUV(*bitssvp);
3551 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3552 const UV none = SvUV(*nonesvp);
5662e334 3553 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3554
3555 HV* ret = newHV();
3556
3557 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3558
3559 /* Must have at least 8 bits to get the mappings */
3560 if (bits != 8 && bits != 16 && bits != 32) {
3561 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3562 (UV)bits);
3563 }
3564
5662e334
KW
3565 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3566 mapping to more than one character */
3567
3568 /* Construct an inverse mapping hash for the specials */
3569 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3570 HV * specials_inverse = newHV();
3571 char *char_from; /* the lhs of the map */
3572 I32 from_len; /* its byte length */
3573 char *char_to; /* the rhs of the map */
3574 I32 to_len; /* its byte length */
3575 SV *sv_to; /* and in a sv */
3576 AV* from_list; /* list of things that map to each 'to' */
3577
3578 hv_iterinit(specials_hv);
3579
4a4088c4
KW
3580 /* The keys are the characters (in UTF-8) that map to the corresponding
3581 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3582 * list. */
3583 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3584 SV** listp;
3585 if (! SvPOK(sv_to)) {
5637ef5b
NC
3586 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3587 "unexpectedly is not a string, flags=%lu",
3588 (unsigned long)SvFLAGS(sv_to));
5662e334 3589 }
4b88fb76 3590 /*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
3591
3592 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3593 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3594 * it. Those strings are all one character long */
3595 if ((listp = hv_fetch(specials_inverse,
3596 SvPVX(sv_to),
3597 SvCUR(sv_to), 0)))
3598 {
3599 from_list = (AV*) *listp;
3600 }
3601 else { /* No entry yet for it: create one */
3602 from_list = newAV();
3603 if (! hv_store(specials_inverse,
3604 SvPVX(sv_to),
3605 SvCUR(sv_to),
3606 (SV*) from_list, 0))
3607 {
3608 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3609 }
3610 }
3611
3612 /* Here have the list associated with this 'to' (perhaps newly
3613 * created and empty). Just add to it. Note that we ASSUME that
3614 * the input is guaranteed to not have duplications, so we don't
3615 * check for that. Duplications just slow down execution time. */
3616 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3617 }
3618
3619 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3620 * it looking for cases like the FB05/FB06 examples above. There would
3621 * be an entry in the hash like
3622 * 'st' => [ FB05, FB06 ]
3623 * In this example we will create two lists that get stored in the
3624 * returned hash, 'ret':
3625 * FB05 => [ FB05, FB06 ]
3626 * FB06 => [ FB05, FB06 ]
3627 *
3628 * Note that there is nothing to do if the array only has one element.
3629 * (In the normal 1-1 case handled below, we don't have to worry about
3630 * two lists, as everything gets tied to the single list that is
3631 * generated for the single character 'to'. But here, we are omitting
3632 * that list, ('st' in the example), so must have multiple lists.) */
3633 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3634 &char_to, &to_len)))
3635 {
b9f2b683 3636 if (av_tindex(from_list) > 0) {
c70927a6 3637 SSize_t i;
5662e334
KW
3638
3639 /* We iterate over all combinations of i,j to place each code
3640 * point on each list */
b9f2b683 3641 for (i = 0; i <= av_tindex(from_list); i++) {
c70927a6 3642<