This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Inline is_utf8_invariant_string()
[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 38static const char cp_above_legal_max[] =
2d212e86 39 "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf"";
760c7c2f 40
114d9c4d 41#define MAX_NON_DEPRECATED_CP ((UV) (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/*
378516de 55=for apidoc uvoffuni_to_utf8_flags
eebe1485 56
a27992cc 57THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
58Instead, B<Almost all code should use L</uvchr_to_utf8> or
59L</uvchr_to_utf8_flags>>.
a27992cc 60
de69f3af
KW
61This function is like them, but the input is a strict Unicode
62(as opposed to native) code point. Only in very rare circumstances should code
63not be using the native code point.
949cf498 64
efa9cd84 65For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 66
eebe1485
SC
67=cut
68*/
69
8ee1cdcb
KW
70#define HANDLE_UNICODE_SURROGATE(uv, flags) \
71 STMT_START { \
72 if (flags & UNICODE_WARN_SURROGATE) { \
73 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
74 "UTF-16 surrogate U+%04"UVXf, uv); \
75 } \
76 if (flags & UNICODE_DISALLOW_SURROGATE) { \
77 return NULL; \
78 } \
79 } STMT_END;
80
81#define HANDLE_UNICODE_NONCHAR(uv, flags) \
82 STMT_START { \
83 if (flags & UNICODE_WARN_NONCHAR) { \
84 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
85 "Unicode non-character U+%04"UVXf" is not " \
86 "recommended for open interchange", uv); \
87 } \
88 if (flags & UNICODE_DISALLOW_NONCHAR) { \
89 return NULL; \
90 } \
91 } STMT_END;
92
ba6ed43c
KW
93/* Use shorter names internally in this file */
94#define SHIFT UTF_ACCUMULATION_SHIFT
95#undef MARK
96#define MARK UTF_CONTINUATION_MARK
97#define MASK UTF_CONTINUATION_MASK
98
dfe13c55 99U8 *
378516de 100Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 101{
378516de 102 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 103
2d1545e5 104 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 105 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
106 return d;
107 }
facc1dc2 108
3ea68d71 109 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
110 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
111 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
112 return d;
113 }
d9432125 114
ba6ed43c
KW
115 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
116 * below, the 16 is for start bytes E0-EF (which are all the possible ones
117 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
118 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
119 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
120 * 0x800-0xFFFF on ASCII */
121 if (uv < (16 * (1U << (2 * SHIFT)))) {
122 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
123 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
124 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
125
126#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
127 aren't tested here */
128 /* The most likely code points in this range are below the surrogates.
129 * Do an extra test to quickly exclude those. */
130 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
131 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
132 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
133 {
8ee1cdcb
KW
134 HANDLE_UNICODE_NONCHAR(uv, flags);
135 }
136 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
137 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 138 }
ba6ed43c
KW
139 }
140#endif
141 return d;
142 }
143
144 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
145 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
146 * happen starting with 4-byte characters on ASCII platforms. We unify the
147 * code for these with EBCDIC, even though some of them require 5-bytes on
148 * those, because khw believes the code saving is worth the very slight
149 * performance hit on these high EBCDIC code points. */
150
151 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
152 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
153 && ckWARN_d(WARN_DEPRECATED))
154 {
155 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
156 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
157 }
158 if ( (flags & UNICODE_WARN_SUPER)
159 || ( UNICODE_IS_ABOVE_31_BIT(uv)
160 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
161 {
162 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
163
164 /* Choose the more dire applicable warning */
165 (UNICODE_IS_ABOVE_31_BIT(uv))
166 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
167 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
168 uv);
169 }
170 if (flags & UNICODE_DISALLOW_SUPER
171 || ( UNICODE_IS_ABOVE_31_BIT(uv)
172 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
173 {
174 return NULL;
175 }
176 }
ba6ed43c
KW
177 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
178 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 179 }
d9432125 180
ba6ed43c
KW
181 /* Test for and handle 4-byte result. In the test immediately below, the
182 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
183 * characters). The 3 is for 3 continuation bytes; these each contribute
184 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
185 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
186 * 0x1_0000-0x1F_FFFF on ASCII */
187 if (uv < (8 * (1U << (3 * SHIFT)))) {
188 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
189 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
190 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
191 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
192
193#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
194 characters. The end-plane non-characters for EBCDIC were
195 handled just above */
196 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
197 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 198 }
ba6ed43c
KW
199 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
200 HANDLE_UNICODE_SURROGATE(uv, flags);
201 }
202#endif
203
204 return d;
205 }
206
207 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
208 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
209 * format. The unrolled version above turns out to not save all that much
210 * time, and at these high code points (well above the legal Unicode range
211 * on ASCII platforms, and well above anything in common use in EBCDIC),
212 * khw believes that less code outweighs slight performance gains. */
213
d9432125 214 {
5aaebcb3 215 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
216 U8 *p = d+len-1;
217 while (p > d) {
4c8cd605 218 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
219 uv >>= UTF_ACCUMULATION_SHIFT;
220 }
4c8cd605 221 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
222 return d+len;
223 }
a0ed51b3 224}
a5bf80e0 225
646ca15d 226/*
07693fe6
KW
227=for apidoc uvchr_to_utf8
228
bcb1a2d4 229Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 230of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
231C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
232the byte after the end of the new character. In other words,
07693fe6
KW
233
234 d = uvchr_to_utf8(d, uv);
235
236is the recommended wide native character-aware way of saying
237
238 *(d++) = uv;
239
760c7c2f
KW
240This function accepts any UV as input, but very high code points (above
241C<IV_MAX> on the platform) will raise a deprecation warning. This is
242typically 0x7FFF_FFFF in a 32-bit word.
243
244It is possible to forbid or warn on non-Unicode code points, or those that may
245be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 246
07693fe6
KW
247=cut
248*/
249
de69f3af
KW
250/* This is also a macro */
251PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
252
07693fe6
KW
253U8 *
254Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
255{
de69f3af 256 return uvchr_to_utf8(d, uv);
07693fe6
KW
257}
258
de69f3af
KW
259/*
260=for apidoc uvchr_to_utf8_flags
261
262Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 263of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
264C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
265the byte after the end of the new character. In other words,
de69f3af
KW
266
267 d = uvchr_to_utf8_flags(d, uv, flags);
268
269or, in most cases,
270
271 d = uvchr_to_utf8_flags(d, uv, 0);
272
273This is the Unicode-aware way of saying
274
275 *(d++) = uv;
276
760c7c2f
KW
277If C<flags> is 0, this function accepts any UV as input, but very high code
278points (above C<IV_MAX> for the platform) will raise a deprecation warning.
279This is typically 0x7FFF_FFFF in a 32-bit word.
280
281Specifying C<flags> can further restrict what is allowed and not warned on, as
282follows:
de69f3af 283
796b6530 284If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
285the function will raise a warning, provided UTF8 warnings are enabled. If
286instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
287NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 288
760c7c2f
KW
289Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
290affect how the function handles a Unicode non-character.
93e6dbd6 291
760c7c2f
KW
292And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
293affect the handling of code points that are above the Unicode maximum of
2940x10FFFF. Languages other than Perl may not be able to accept files that
295contain these.
93e6dbd6
KW
296
297The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
298the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
299three DISALLOW flags.
300
ab8e6d41
KW
301Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
302so using them is more problematic than other above-Unicode code points. Perl
303invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
304likely that non-Perl languages will not be able to read files that contain
305these that written by the perl interpreter; nor would Perl understand files
306written by something that uses a different extension. For these reasons, there
307is a separate set of flags that can warn and/or disallow these extremely high
308code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
309C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
310are entirely independent from the deprecation warning for code points above
311C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
312code point that needs more than 31 bits to represent. When that happens,
313effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
31432-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
315above-Unicode code points, including these, as malformations; and
316C<UNICODE_WARN_SUPER> warns on these.)
317
318On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
319extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
320than on ASCII. Prior to that, code points 2**31 and higher were simply
321unrepresentable, and a different, incompatible method was used to represent
322code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
323and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
324platforms, warning and disallowing 2**31 and higher.
de69f3af 325
de69f3af
KW
326=cut
327*/
328
329/* This is also a macro */
330PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
331
07693fe6
KW
332U8 *
333Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
334{
de69f3af 335 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
336}
337
338/*
87cea99e 339=for apidoc is_utf8_string
6662521e 340
a1433954 341Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 342UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
343using C<strlen(s)> (which means if you use this option, that C<s> can't have
344embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
345that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 346
8871a094
KW
347See also L</is_utf8_invariant_string>(), L</is_utf8_string_loclen>(), and
348L</is_utf8_string_loc>().
768c67ee 349
6662521e
GS
350=cut
351*/
352
8e84507e 353bool
668b6d8d 354Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 355{
35da51f7 356 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 357 const U8* x = s;
067a85ef 358
7918f24d 359 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 360
6662521e 361 while (x < send) {
6302f837
KW
362 STRLEN len = isUTF8_CHAR(x, send);
363 if (UNLIKELY(! len)) {
364 return FALSE;
365 }
366 x += len;
6662521e 367 }
768c67ee 368
067a85ef 369 return TRUE;
6662521e
GS
370}
371
67e989fb 372/*
814fafa7
NC
373Implemented as a macro in utf8.h
374
87cea99e 375=for apidoc is_utf8_string_loc
814fafa7 376
a1433954
KW
377Like L</is_utf8_string> but stores the location of the failure (in the
378case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
379"utf8ness success") in the C<ep>.
380
a1433954 381See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 382
87cea99e 383=for apidoc is_utf8_string_loclen
81cd54e3 384
a1433954
KW
385Like L</is_utf8_string>() but stores the location of the failure (in the
386case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
387"utf8ness success") in the C<ep>, and the number of UTF-8
388encoded characters in the C<el>.
389
a1433954 390See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
391
392=cut
393*/
394
395bool
668b6d8d 396Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 397{
35da51f7 398 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 399 const U8* x = s;
3ebfea28 400 STRLEN outlen = 0;
7918f24d
NC
401
402 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 403
81cd54e3 404 while (x < send) {
6302f837
KW
405 STRLEN len = isUTF8_CHAR(x, send);
406 if (UNLIKELY(! len)) {
9f590ad4 407 break;
6302f837
KW
408 }
409 x += len;
410 outlen++;
81cd54e3 411 }
768c67ee 412
3ebfea28
AL
413 if (el)
414 *el = outlen;
415
768c67ee
JH
416 if (ep)
417 *ep = x;
3ebfea28 418 return (x == send);
81cd54e3
JH
419}
420
421/*
768c67ee 422
de69f3af 423=for apidoc utf8n_to_uvchr
378516de
KW
424
425THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 426Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 427
9041c2e3 428Bottom level UTF-8 decode routine.
de69f3af 429Returns the native code point value of the first character in the string C<s>,
746afd53
KW
430which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
431C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
432the length, in bytes, of that character.
949cf498
KW
433
434The value of C<flags> determines the behavior when C<s> does not point to a
435well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
436zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
437next possible position in C<s> that could begin a non-malformed character.
438Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
439
440Various ALLOW flags can be set in C<flags> to allow (and not warn on)
441individual types of malformations, such as the sequence being overlong (that
442is, when there is a shorter sequence that can express the same code point;
443overlong sequences are expressly forbidden in the UTF-8 standard due to
444potential security issues). Another malformation example is the first byte of
445a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
446flags. For allowed 0 length strings, this function returns 0; for allowed
447overlong sequences, the computed code point is returned; for all other allowed
448malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
449determinable reasonable value.
949cf498 450
796b6530 451The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
452flags) malformation is found. If this flag is set, the routine assumes that
453the caller will raise a warning, and this function will silently just set
d088425d
KW
454C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
455
75200dff 456Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 457character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 458in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
459first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
460input had an error.
949cf498
KW
461
462Certain code points are considered problematic. These are Unicode surrogates,
746afd53 463Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 464By default these are considered regular code points, but certain situations
5eafe189 465warrant special handling for them. If C<flags> contains
796b6530
KW
466C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
467malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
468C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
469Unicode maximum) can be set to disallow these categories individually.
470
471The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
472C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
473raised for their respective categories, but otherwise the code points are
474considered valid (not malformations). To get a category to both be treated as
475a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 476(But note that warnings are not raised if lexically disabled nor if
796b6530 477C<UTF8_CHECK_ONLY> is also specified.)
949cf498 478
760c7c2f
KW
479It is now deprecated to have very high code points (above C<IV_MAX> on the
480platforms) and this function will raise a deprecation warning for these (unless
481such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
482in a 32-bit word.
ab8e6d41
KW
483
484Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
485so using them is more problematic than other above-Unicode code points. Perl
486invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
487likely that non-Perl languages will not be able to read files that contain
488these that written by the perl interpreter; nor would Perl understand files
489written by something that uses a different extension. For these reasons, there
490is a separate set of flags that can warn and/or disallow these extremely high
491code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
492C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
493are entirely independent from the deprecation warning for code points above
494C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
495code point that needs more than 31 bits to represent. When that happens,
496effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
49732-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
498above-Unicode code points, including these, as malformations; and
499C<UTF8_WARN_SUPER> warns on these.)
500
501On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
502extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
503than on ASCII. Prior to that, code points 2**31 and higher were simply
504unrepresentable, and a different, incompatible method was used to represent
505code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
506and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
507platforms, warning and disallowing 2**31 and higher.
949cf498
KW
508
509All other code points corresponding to Unicode characters, including private
510use and those yet to be assigned, are never considered malformed and never
511warn.
67e989fb 512
37607a96
PK
513=cut
514*/
67e989fb 515
a0ed51b3 516UV
de69f3af 517Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 518{
d4c19fe8 519 const U8 * const s0 = s;
eb83ed87 520 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 521 U8 * send;
eb83ed87
KW
522 UV uv = *s;
523 STRLEN expectlen;
949cf498 524 SV* sv = NULL;
eb83ed87
KW
525 UV outlier_ret = 0; /* return value when input is in error or problematic
526 */
527 UV pack_warn = 0; /* Save result of packWARN() for later */
528 bool unexpected_non_continuation = FALSE;
529 bool overflowed = FALSE;
2f8f112e 530 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 531
eb83ed87 532 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 533
de69f3af 534 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 535
eb83ed87
KW
536 /* The order of malformation tests here is important. We should consume as
537 * few bytes as possible in order to not skip any valid character. This is
538 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
539 * http://unicode.org/reports/tr36 for more discussion as to why. For
540 * example, once we've done a UTF8SKIP, we can tell the expected number of
541 * bytes, and could fail right off the bat if the input parameters indicate
542 * that there are too few available. But it could be that just that first
543 * byte is garbled, and the intended character occupies fewer bytes. If we
544 * blindly assumed that the first byte is correct, and skipped based on
545 * that number, we could skip over a valid input character. So instead, we
546 * always examine the sequence byte-by-byte.
547 *
548 * We also should not consume too few bytes, otherwise someone could inject
549 * things. For example, an input could be deliberately designed to
550 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 551 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
552 * which is actually part of of the overflowing sequence), that could look
553 * legitimate to the caller, which could discard the initial partial
554 * sequence and process the rest, inappropriately */
555
556 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 557 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
558 if (retlen) {
559 *retlen = 0;
560 }
a0dbb045 561
eb83ed87
KW
562 if (flags & UTF8_ALLOW_EMPTY) {
563 return 0;
564 }
565 if (! (flags & UTF8_CHECK_ONLY)) {
566 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
567 }
0c443dc2
JH
568 goto malformed;
569 }
570
eb83ed87
KW
571 expectlen = UTF8SKIP(s);
572
573 /* A well-formed UTF-8 character, as the vast majority of calls to this
574 * function will be for, has this expected length. For efficiency, set
575 * things up here to return it. It will be overriden only in those rare
576 * cases where a malformation is found */
577 if (retlen) {
578 *retlen = expectlen;
579 }
580
581 /* An invariant is trivially well-formed */
1d72bdf6 582 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 583 return uv;
a0ed51b3 584 }
67e989fb 585
eb83ed87 586 /* A continuation character can't start a valid sequence */
b5b9af04 587 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
588 if (flags & UTF8_ALLOW_CONTINUATION) {
589 if (retlen) {
590 *retlen = 1;
591 }
592 return UNICODE_REPLACEMENT;
593 }
ba210ebe 594
eb83ed87
KW
595 if (! (flags & UTF8_CHECK_ONLY)) {
596 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
597 }
598 curlen = 1;
ba210ebe
JH
599 goto malformed;
600 }
9041c2e3 601
dcd27b3c
KW
602 /* Here is not a continuation byte, nor an invariant. The only thing left
603 * is a start byte (possibly for an overlong) */
604
534752c1
KW
605 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
606 * that indicate the number of bytes in the character's whole UTF-8
607 * sequence, leaving just the bits that are part of the value. */
608 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 609
eb83ed87
KW
610 /* Now, loop through the remaining bytes in the character's sequence,
611 * accumulating each into the working value as we go. Be sure to not look
612 * past the end of the input string */
0b8d30e8
KW
613 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
614
eb83ed87 615 for (s = s0 + 1; s < send; s++) {
b5b9af04 616 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
617 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
618
619 /* The original implementors viewed this malformation as more
620 * serious than the others (though I, khw, don't understand
621 * why, since other malformations also give very very wrong
622 * results), so there is no way to turn off checking for it.
623 * Set a flag, but keep going in the loop, so that we absorb
624 * the rest of the bytes that comprise the character. */
625 overflowed = TRUE;
626 overflow_byte = *s; /* Save for warning message's use */
627 }
8850bf83 628 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
629 }
630 else {
631 /* Here, found a non-continuation before processing all expected
632 * bytes. This byte begins a new character, so quit, even if
633 * allowing this malformation. */
634 unexpected_non_continuation = TRUE;
635 break;
636 }
637 } /* End of loop through the character's bytes */
638
639 /* Save how many bytes were actually in the character */
640 curlen = s - s0;
641
642 /* The loop above finds two types of malformations: non-continuation and/or
643 * overflow. The non-continuation malformation is really a too-short
644 * malformation, as it means that the current character ended before it was
645 * expected to (being terminated prematurely by the beginning of the next
646 * character, whereas in the too-short malformation there just are too few
647 * bytes available to hold the character. In both cases, the check below
648 * that we have found the expected number of bytes would fail if executed.)
649 * Thus the non-continuation malformation is really unnecessary, being a
650 * subset of the too-short malformation. But there may be existing
651 * applications that are expecting the non-continuation type, so we retain
652 * it, and return it in preference to the too-short malformation. (If this
653 * code were being written from scratch, the two types might be collapsed
654 * into one.) I, khw, am also giving priority to returning the
655 * non-continuation and too-short malformations over overflow when multiple
656 * ones are present. I don't know of any real reason to prefer one over
657 * the other, except that it seems to me that multiple-byte errors trumps
658 * errors from a single byte */
b5b9af04 659 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
660 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
661 if (! (flags & UTF8_CHECK_ONLY)) {
662 if (curlen == 1) {
663 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
664 }
665 else {
666 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
667 }
668 }
eb83ed87
KW
669 goto malformed;
670 }
671 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
672
673 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
674 * as what the original expectations were. */
675 do_overlong_test = FALSE;
eb83ed87
KW
676 if (retlen) {
677 *retlen = curlen;
678 }
679 }
b5b9af04 680 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
681 if (! (flags & UTF8_ALLOW_SHORT)) {
682 if (! (flags & UTF8_CHECK_ONLY)) {
683 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 684 }
eb83ed87
KW
685 goto malformed;
686 }
687 uv = UNICODE_REPLACEMENT;
2f8f112e 688 do_overlong_test = FALSE;
eb83ed87
KW
689 if (retlen) {
690 *retlen = curlen;
691 }
692 }
693
b5b9af04 694 if (UNLIKELY(overflowed)) {
eb83ed87 695 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 696 goto malformed;
eb83ed87 697 }
eb83ed87 698
2f8f112e 699 if (do_overlong_test
5aaebcb3 700 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
701 && ! (flags & UTF8_ALLOW_LONG))
702 {
eb83ed87
KW
703 /* The overlong malformation has lower precedence than the others.
704 * Note that if this malformation is allowed, we return the actual
705 * value, instead of the replacement character. This is because this
706 * value is actually well-defined. */
707 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 708 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
709 }
710 goto malformed;
711 }
712
1a89bb6c 713 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
714 * problematic code point that is not allowed by the input parameters. */
715 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
716 && ((flags & ( UTF8_DISALLOW_NONCHAR
717 |UTF8_DISALLOW_SURROGATE
718 |UTF8_DISALLOW_SUPER
719 |UTF8_DISALLOW_ABOVE_31_BIT
720 |UTF8_WARN_NONCHAR
721 |UTF8_WARN_SURROGATE
722 |UTF8_WARN_SUPER
723 |UTF8_WARN_ABOVE_31_BIT))
724 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
725 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 726 {
949cf498 727 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
728
729 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
730 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 731 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 732 && ckWARN_d(WARN_SURROGATE))
eb83ed87 733 {
111d382d 734 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 735 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
736 }
737 if (flags & UTF8_DISALLOW_SURROGATE) {
738 goto disallowed;
739 }
740 }
949cf498 741 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 742 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 743 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 744 {
0bcdd8f6
KW
745 sv = sv_2mortal(Perl_newSVpvf(aTHX_
746 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
747 uv));
54f4afef 748 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 749 }
c0236afe
KW
750
751 /* The maximum code point ever specified by a standard was
752 * 2**31 - 1. Anything larger than that is a Perl extension that
753 * very well may not be understood by other applications (including
754 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
755 * these code points are indicated by the first UTF-8 byte being
d35f2ca5
KW
756 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
757 * and before possibly bailing out, so that the slightly more dire
c0236afe
KW
758 * warning will override the regular one. */
759 if (
760#ifndef EBCDIC
761 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
762#else
763 /* The I8 for 2**31 (U+80000000) is
764 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
765 * and it turns out that on all EBCDIC pages recognized that
766 * the UTF-EBCDIC for that code point is
767 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
768 * For the next lower code point, the 1047 UTF-EBCDIC is
769 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
770 * The other code pages differ only in the bytes following
771 * \x42. Thus the following works (the minimum continuation
772 * byte is \x41). */
773 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
774 || s0[2] > 0x41
775 || s0[3] > 0x41
776 || s0[4] > 0x41
777 || s0[5] > 0x41
778 || s0[6] > 0x41
779 || s0[7] > 0x42)
780#endif
d35f2ca5
KW
781 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
782 |UTF8_DISALLOW_ABOVE_31_BIT)))
ea5ced44 783 {
0bcdd8f6 784 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 785 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 786 && ckWARN_d(WARN_UTF8))
ea5ced44 787 {
0bcdd8f6
KW
788 sv = sv_2mortal(Perl_newSVpvf(aTHX_
789 "Code point 0x%"UVXf" is not Unicode, and not portable",
790 uv));
ea5ced44
KW
791 pack_warn = packWARN(WARN_UTF8);
792 }
d35f2ca5 793 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
794 goto disallowed;
795 }
796 }
c0236afe 797
949cf498
KW
798 if (flags & UTF8_DISALLOW_SUPER) {
799 goto disallowed;
800 }
760c7c2f
KW
801
802 /* The deprecated warning overrides any non-deprecated one */
803 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
804 {
805 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
806 uv, MAX_NON_DEPRECATED_CP));
807 pack_warn = packWARN(WARN_DEPRECATED);
808 }
949cf498 809 }
4190d317
KW
810 else if (UNICODE_IS_NONCHAR(uv)) {
811 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 812 && ckWARN_d(WARN_NONCHAR))
4190d317 813 {
ba707cdc 814 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 815 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
816 }
817 if (flags & UTF8_DISALLOW_NONCHAR) {
818 goto disallowed;
819 }
820 }
949cf498 821
eb83ed87 822 if (sv) {
de69f3af
KW
823 outlier_ret = uv; /* Note we don't bother to convert to native,
824 as all the outlier code points are the same
825 in both ASCII and EBCDIC */
eb83ed87
KW
826 goto do_warn;
827 }
828
949cf498
KW
829 /* Here, this is not considered a malformed character, so drop through
830 * to return it */
a0ed51b3 831 }
ba210ebe 832
de69f3af 833 return UNI_TO_NATIVE(uv);
ba210ebe 834
eb83ed87
KW
835 /* There are three cases which get to beyond this point. In all 3 cases:
836 * <sv> if not null points to a string to print as a warning.
837 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
838 * set.
839 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
840 * This is done by initializing it to 0, and changing it only
841 * for case 1).
842 * The 3 cases are:
843 * 1) The input is valid but problematic, and to be warned about. The
844 * return value is the resultant code point; <*retlen> is set to
845 * <curlen>, the number of bytes that comprise the code point.
846 * <pack_warn> contains the result of packWARN() for the warning
847 * types. The entry point for this case is the label <do_warn>;
848 * 2) The input is a valid code point but disallowed by the parameters to
849 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
850 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
851 * comprise the code point. <pack_warn> contains the result of
852 * packWARN() for the warning types. The entry point for this case is
853 * the label <disallowed>.
854 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
855 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
856 * bytes that comprise the malformation. All such malformations are
857 * assumed to be warning type <utf8>. The entry point for this case
858 * is the label <malformed>.
859 */
949cf498 860
7b52d656 861 malformed:
ba210ebe 862
eb83ed87
KW
863 if (sv && ckWARN_d(WARN_UTF8)) {
864 pack_warn = packWARN(WARN_UTF8);
865 }
866
7b52d656 867 disallowed:
eb83ed87 868
fcc8fcf6 869 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 870 if (retlen)
10edeb5d 871 *retlen = ((STRLEN) -1);
ba210ebe
JH
872 return 0;
873 }
874
7b52d656 875 do_warn:
5b311467 876
eb83ed87
KW
877 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
878 if warnings are to be raised. */
f555bc63 879 const char * const string = SvPVX_const(sv);
a0dbb045 880
f555bc63
KW
881 if (PL_op)
882 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
883 else
884 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
885 }
886
eb83ed87
KW
887 if (retlen) {
888 *retlen = curlen;
889 }
ba210ebe 890
eb83ed87 891 return outlier_ret;
a0ed51b3
LW
892}
893
8e84507e 894/*
ec5f19d0
KW
895=for apidoc utf8_to_uvchr_buf
896
897Returns the native code point of the first character in the string C<s> which
898is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 899C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 900
524080c4
KW
901If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
902enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 903C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 904(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 905C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 906the next possible position in C<s> that could begin a non-malformed character.
de69f3af 907See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 908returned.
ec5f19d0 909
760c7c2f
KW
910Code points above the platform's C<IV_MAX> will raise a deprecation warning,
911unless those are turned off.
912
ec5f19d0
KW
913=cut
914*/
915
916
917UV
918Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
919{
ec5f19d0
KW
920 assert(s < send);
921
922 return utf8n_to_uvchr(s, send - s, retlen,
923 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
924}
925
926/*
ec5f19d0
KW
927=for apidoc utf8_to_uvuni_buf
928
de69f3af
KW
929Only in very rare circumstances should code need to be dealing in Unicode
930(as opposed to native) code points. In those few cases, use
931C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
932
933Returns the Unicode (not-native) code point of the first character in the
934string C<s> which
ec5f19d0
KW
935is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
936C<retlen> will be set to the length, in bytes, of that character.
937
524080c4
KW
938If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
939enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
940NULL) to -1. If those warnings are off, the computed value if well-defined (or
941the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
942is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
943next possible position in C<s> that could begin a non-malformed character.
de69f3af 944See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 945
760c7c2f
KW
946Code points above the platform's C<IV_MAX> will raise a deprecation warning,
947unless those are turned off.
948
ec5f19d0
KW
949=cut
950*/
951
952UV
953Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
954{
955 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
956
957 assert(send > s);
958
959 /* Call the low level routine asking for checks */
de69f3af
KW
960 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
961 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
962}
963
b76347f2 964/*
87cea99e 965=for apidoc utf8_length
b76347f2
JH
966
967Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
968Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
969up past C<e>, croaks.
b76347f2
JH
970
971=cut
972*/
973
974STRLEN
35a4481c 975Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
976{
977 STRLEN len = 0;
978
7918f24d
NC
979 PERL_ARGS_ASSERT_UTF8_LENGTH;
980
8850bf83
JH
981 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
982 * the bitops (especially ~) can create illegal UTF-8.
983 * In other words: in Perl UTF-8 is not just for Unicode. */
984
a3b680e6
AL
985 if (e < s)
986 goto warn_and_return;
b76347f2 987 while (s < e) {
4cbf4130 988 s += UTF8SKIP(s);
8e91ec7f
AV
989 len++;
990 }
991
992 if (e != s) {
993 len--;
994 warn_and_return:
9b387841
NC
995 if (PL_op)
996 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
997 "%s in %s", unees, OP_DESC(PL_op));
998 else
61a12c31 999 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1000 }
1001
1002 return len;
1003}
1004
b06226ff 1005/*
87cea99e 1006=for apidoc utf8_distance
b06226ff 1007
1e54db1a 1008Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1009and C<b>.
1010
1011WARNING: use only if you *know* that the pointers point inside the
1012same UTF-8 buffer.
1013
37607a96
PK
1014=cut
1015*/
a0ed51b3 1016
02eb7b47 1017IV
35a4481c 1018Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1019{
7918f24d
NC
1020 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1021
bf1665bc 1022 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1023}
1024
b06226ff 1025/*
87cea99e 1026=for apidoc utf8_hop
b06226ff 1027
8850bf83
JH
1028Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1029forward or backward.
b06226ff
JH
1030
1031WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1032the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1033on the first byte of character or just after the last byte of a character.
b06226ff 1034
37607a96
PK
1035=cut
1036*/
a0ed51b3
LW
1037
1038U8 *
4caf7d8c 1039Perl_utf8_hop(const U8 *s, SSize_t off)
a0ed51b3 1040{
7918f24d
NC
1041 PERL_ARGS_ASSERT_UTF8_HOP;
1042
8850bf83
JH
1043 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1044 * the bitops (especially ~) can create illegal UTF-8.
1045 * In other words: in Perl UTF-8 is not just for Unicode. */
1046
a0ed51b3
LW
1047 if (off >= 0) {
1048 while (off--)
1049 s += UTF8SKIP(s);
1050 }
1051 else {
1052 while (off++) {
1053 s--;
8850bf83
JH
1054 while (UTF8_IS_CONTINUATION(*s))
1055 s--;
a0ed51b3
LW
1056 }
1057 }
4373e329 1058 return (U8 *)s;
a0ed51b3
LW
1059}
1060
6940069f 1061/*
fed3ba5d
NC
1062=for apidoc bytes_cmp_utf8
1063
a1433954 1064Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1065sequence of characters (stored as UTF-8)
1066in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1067equal, -1 or -2 if the first string is less than the second string, +1 or +2
1068if the first string is greater than the second string.
1069
1070-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1071longer string. -2 or +2 is returned if
1072there was a difference between characters
fed3ba5d
NC
1073within the strings.
1074
1075=cut
1076*/
1077
1078int
1079Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1080{
1081 const U8 *const bend = b + blen;
1082 const U8 *const uend = u + ulen;
1083
1084 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1085
1086 while (b < bend && u < uend) {
1087 U8 c = *u++;
1088 if (!UTF8_IS_INVARIANT(c)) {
1089 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1090 if (u < uend) {
1091 U8 c1 = *u++;
1092 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1093 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1094 } else {
1095 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1096 "Malformed UTF-8 character "
1097 "(unexpected non-continuation byte 0x%02x"
1098 ", immediately after start byte 0x%02x)"
1099 /* Dear diag.t, it's in the pod. */
1100 "%s%s", c1, c,
1101 PL_op ? " in " : "",
1102 PL_op ? OP_DESC(PL_op) : "");
1103 return -2;
1104 }
1105 } else {
1106 if (PL_op)
1107 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1108 "%s in %s", unees, OP_DESC(PL_op));
1109 else
61a12c31 1110 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1111 return -2; /* Really want to return undef :-) */
1112 }
1113 } else {
1114 return -2;
1115 }
1116 }
1117 if (*b != c) {
1118 return *b < c ? -2 : +2;
1119 }
1120 ++b;
1121 }
1122
1123 if (b == bend && u == uend)
1124 return 0;
1125
1126 return b < bend ? +1 : -1;
1127}
1128
1129/*
87cea99e 1130=for apidoc utf8_to_bytes
6940069f 1131
2bbc8d55 1132Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1133Unlike L</bytes_to_utf8>, this over-writes the original string, and
1134updates C<len> to contain the new length.
67e989fb 1135Returns zero on failure, setting C<len> to -1.
6940069f 1136
a1433954 1137If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1138
6940069f
GS
1139=cut
1140*/
1141
1142U8 *
37607a96 1143Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1144{
d4c19fe8
AL
1145 U8 * const save = s;
1146 U8 * const send = s + *len;
6940069f 1147 U8 *d;
246fae53 1148
7918f24d 1149 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1150 PERL_UNUSED_CONTEXT;
7918f24d 1151
1e54db1a 1152 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1153 while (s < send) {
d59937ca
KW
1154 if (! UTF8_IS_INVARIANT(*s)) {
1155 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1156 *len = ((STRLEN) -1);
1157 return 0;
1158 }
1159 s++;
dcad2880 1160 }
d59937ca 1161 s++;
246fae53 1162 }
dcad2880
JH
1163
1164 d = s = save;
6940069f 1165 while (s < send) {
80e0b38f
KW
1166 U8 c = *s++;
1167 if (! UTF8_IS_INVARIANT(c)) {
1168 /* Then it is two-byte encoded */
a62b247b 1169 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1170 s++;
1171 }
1172 *d++ = c;
6940069f
GS
1173 }
1174 *d = '\0';
246fae53 1175 *len = d - save;
6940069f
GS
1176 return save;
1177}
1178
1179/*
87cea99e 1180=for apidoc bytes_from_utf8
f9a63242 1181
2bbc8d55 1182Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1183Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1184the newly-created string, and updates C<len> to contain the new
1185length. Returns the original string if no conversion occurs, C<len>
72d33970 1186is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 11870 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1188in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1189
37607a96
PK
1190=cut
1191*/
f9a63242
JH
1192
1193U8 *
e1ec3a88 1194Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1195{
f9a63242 1196 U8 *d;
e1ec3a88
AL
1197 const U8 *start = s;
1198 const U8 *send;
f9a63242
JH
1199 I32 count = 0;
1200
7918f24d 1201 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1202 PERL_UNUSED_CONTEXT;
f9a63242 1203 if (!*is_utf8)
73d840c0 1204 return (U8 *)start;
f9a63242 1205
1e54db1a 1206 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1207 for (send = s + *len; s < send;) {
d59937ca
KW
1208 if (! UTF8_IS_INVARIANT(*s)) {
1209 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1210 return (U8 *)start;
d59937ca
KW
1211 }
1212 count++;
1213 s++;
db42d148 1214 }
d59937ca 1215 s++;
f9a63242
JH
1216 }
1217
35da51f7 1218 *is_utf8 = FALSE;
f9a63242 1219
212542aa 1220 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1221 s = start; start = d;
f9a63242
JH
1222 while (s < send) {
1223 U8 c = *s++;
1a91c45d 1224 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1225 /* Then it is two-byte encoded */
a62b247b 1226 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1227 s++;
c4d5f83a
NIS
1228 }
1229 *d++ = c;
f9a63242
JH
1230 }
1231 *d = '\0';
1232 *len = d - start;
73d840c0 1233 return (U8 *)start;
f9a63242
JH
1234}
1235
1236/*
87cea99e 1237=for apidoc bytes_to_utf8
6940069f 1238
ff97e5cf
KW
1239Converts a string C<s> of length C<len> bytes from the native encoding into
1240UTF-8.
6662521e 1241Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1242reflect the new length in bytes.
6940069f 1243
75200dff 1244A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1245
1246If you want to convert to UTF-8 from encodings other than
1247the native (Latin1 or EBCDIC),
a1433954 1248see L</sv_recode_to_utf8>().
c9ada85f 1249
497711e7 1250=cut
6940069f
GS
1251*/
1252
c682ebef
FC
1253/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1254 likewise need duplication. */
1255
6940069f 1256U8*
35a4481c 1257Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1258{
35a4481c 1259 const U8 * const send = s + (*len);
6940069f
GS
1260 U8 *d;
1261 U8 *dst;
7918f24d
NC
1262
1263 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1264 PERL_UNUSED_CONTEXT;
6940069f 1265
212542aa 1266 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1267 dst = d;
1268
1269 while (s < send) {
55d09dc8
KW
1270 append_utf8_from_native_byte(*s, &d);
1271 s++;
6940069f
GS
1272 }
1273 *d = '\0';
6662521e 1274 *len = d-dst;
6940069f
GS
1275 return dst;
1276}
1277
a0ed51b3 1278/*
dea0fc0b 1279 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1280 *
1281 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1282 * We optimize for native, for obvious reasons. */
1283
1284U8*
dea0fc0b 1285Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1286{
dea0fc0b
JH
1287 U8* pend;
1288 U8* dstart = d;
1289
7918f24d
NC
1290 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1291
dea0fc0b 1292 if (bytelen & 1)
f5992bc4 1293 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1294
1295 pend = p + bytelen;
1296
a0ed51b3 1297 while (p < pend) {
dea0fc0b
JH
1298 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1299 p += 2;
2d1545e5 1300 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1301 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1302 continue;
1303 }
56d37426
KW
1304 if (uv <= MAX_UTF8_TWO_BYTE) {
1305 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1306 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1307 continue;
1308 }
46956fad
KW
1309#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1310#define LAST_HIGH_SURROGATE 0xDBFF
1311#define FIRST_LOW_SURROGATE 0xDC00
1312#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1313
1314 /* This assumes that most uses will be in the first Unicode plane, not
1315 * needing surrogates */
1316 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1317 && uv <= UNICODE_SURROGATE_LAST))
1318 {
1319 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1320 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1321 }
1322 else {
01ea242b 1323 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1324 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1325 || UNLIKELY(low > LAST_LOW_SURROGATE))
1326 {
01ea242b 1327 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1328 }
1329 p += 2;
46956fad
KW
1330 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1331 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1332 }
a0ed51b3 1333 }
56d37426
KW
1334#ifdef EBCDIC
1335 d = uvoffuni_to_utf8_flags(d, uv, 0);
1336#else
a0ed51b3 1337 if (uv < 0x10000) {
eb160463
GS
1338 *d++ = (U8)(( uv >> 12) | 0xe0);
1339 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1340 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1341 continue;
1342 }
1343 else {
eb160463
GS
1344 *d++ = (U8)(( uv >> 18) | 0xf0);
1345 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1346 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1347 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1348 continue;
1349 }
56d37426 1350#endif
a0ed51b3 1351 }
dea0fc0b 1352 *newlen = d - dstart;
a0ed51b3
LW
1353 return d;
1354}
1355
1356/* Note: this one is slightly destructive of the source. */
1357
1358U8*
dea0fc0b 1359Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1360{
1361 U8* s = (U8*)p;
d4c19fe8 1362 U8* const send = s + bytelen;
7918f24d
NC
1363
1364 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1365
e0ea5e2d
NC
1366 if (bytelen & 1)
1367 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1368 (UV)bytelen);
1369
a0ed51b3 1370 while (s < send) {
d4c19fe8 1371 const U8 tmp = s[0];
a0ed51b3
LW
1372 s[0] = s[1];
1373 s[1] = tmp;
1374 s += 2;
1375 }
dea0fc0b 1376 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1377}
1378
922e8cb4
KW
1379bool
1380Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1381{
1382 U8 tmpbuf[UTF8_MAXBYTES+1];
1383 uvchr_to_utf8(tmpbuf, c);
1384 return _is_utf8_FOO(classnum, tmpbuf);
1385}
1386
f9ae8fb6
JD
1387/* Internal function so we can deprecate the external one, and call
1388 this one from other deprecated functions in this file */
1389
f2645549
KW
1390bool
1391Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1392{
f2645549 1393 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1394
1395 if (*p == '_')
1396 return TRUE;
f25ce844 1397 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1398}
1399
5092f92a 1400bool
eba68aa0
KW
1401Perl__is_uni_perl_idcont(pTHX_ UV c)
1402{
1403 U8 tmpbuf[UTF8_MAXBYTES+1];
1404 uvchr_to_utf8(tmpbuf, c);
1405 return _is_utf8_perl_idcont(tmpbuf);
1406}
1407
1408bool
f91dcd13
KW
1409Perl__is_uni_perl_idstart(pTHX_ UV c)
1410{
1411 U8 tmpbuf[UTF8_MAXBYTES+1];
1412 uvchr_to_utf8(tmpbuf, c);
1413 return _is_utf8_perl_idstart(tmpbuf);
1414}
1415
3a4c58c9
KW
1416UV
1417Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1418{
1419 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1420 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1421 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1422 * either "SS" or "Ss". Which one to use is passed into the routine in
1423 * 'S_or_s' to avoid a test */
1424
1425 UV converted = toUPPER_LATIN1_MOD(c);
1426
1427 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1428
1429 assert(S_or_s == 'S' || S_or_s == 's');
1430
6f2d5cbc 1431 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1432 characters in this range */
3a4c58c9
KW
1433 *p = (U8) converted;
1434 *lenp = 1;
1435 return converted;
1436 }
1437
1438 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1439 * which it maps to one of them, so as to only have to have one check for
1440 * it in the main case */
1441 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1442 switch (c) {
1443 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1444 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1445 break;
1446 case MICRO_SIGN:
1447 converted = GREEK_CAPITAL_LETTER_MU;
1448 break;
79e064b9
KW
1449#if UNICODE_MAJOR_VERSION > 2 \
1450 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1451 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1452 case LATIN_SMALL_LETTER_SHARP_S:
1453 *(p)++ = 'S';
1454 *p = S_or_s;
1455 *lenp = 2;
1456 return 'S';
79e064b9 1457#endif
3a4c58c9
KW
1458 default:
1459 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1460 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1461 }
1462 }
1463
1464 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1465 *p = UTF8_TWO_BYTE_LO(converted);
1466 *lenp = 2;
1467
1468 return converted;
1469}
1470
50bda2c3
KW
1471/* Call the function to convert a UTF-8 encoded character to the specified case.
1472 * Note that there may be more than one character in the result.
1473 * INP is a pointer to the first byte of the input character
1474 * OUTP will be set to the first byte of the string of changed characters. It
1475 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1476 * LENP will be set to the length in bytes of the string of changed characters
1477 *
1478 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
1479#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
1480#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
1481#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 1482
b9992569
KW
1483/* This additionally has the input parameter 'specials', which if non-zero will
1484 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 1485 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 1486#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 1487
84afefe6
JH
1488UV
1489Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1490{
a1433954
KW
1491 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1492 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1493 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1494 * the changed version may be longer than the original character.
1495 *
1496 * The ordinal of the first character of the changed version is returned
1497 * (but note, as explained above, that there may be more.) */
1498
7918f24d
NC
1499 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1500
3a4c58c9
KW
1501 if (c < 256) {
1502 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1503 }
1504
0ebc6274 1505 uvchr_to_utf8(p, c);
b9992569 1506 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
1507}
1508
84afefe6
JH
1509UV
1510Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1511{
7918f24d
NC
1512 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1513
3a4c58c9
KW
1514 if (c < 256) {
1515 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1516 }
1517
0ebc6274 1518 uvchr_to_utf8(p, c);
b9992569 1519 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
1520}
1521
afc16117 1522STATIC U8
81611534 1523S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1524{
1525 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1526 * those, converting the result to UTF-8. Since the result is always just
a1433954 1527 * one character, we allow <p> to be NULL */
afc16117
KW
1528
1529 U8 converted = toLOWER_LATIN1(c);
1530
1531 if (p != NULL) {
6f2d5cbc 1532 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1533 *p = converted;
1534 *lenp = 1;
1535 }
1536 else {
430c9760
KW
1537 /* Result is known to always be < 256, so can use the EIGHT_BIT
1538 * macros */
1539 *p = UTF8_EIGHT_BIT_HI(converted);
1540 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1541 *lenp = 2;
1542 }
1543 }
1544 return converted;
1545}
1546
84afefe6
JH
1547UV
1548Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1549{
7918f24d
NC
1550 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1551
afc16117
KW
1552 if (c < 256) {
1553 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1554 }
1555
afc16117 1556 uvchr_to_utf8(p, c);
b9992569 1557 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
1558}
1559
84afefe6 1560UV
51910141 1561Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1562{
51910141 1563 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1564 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1565 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1566 *
1567 * Not to be used for locale folds
51910141 1568 */
f673fad4 1569
a1dde8de
KW
1570 UV converted;
1571
1572 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1573 PERL_UNUSED_CONTEXT;
a1dde8de 1574
1ca267a5
KW
1575 assert (! (flags & FOLD_FLAGS_LOCALE));
1576
659a7c2d 1577 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
1578 converted = GREEK_SMALL_LETTER_MU;
1579 }
9b63e895
KW
1580#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1581 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1582 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
1583 else if ( (flags & FOLD_FLAGS_FULL)
1584 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
1585 {
1ca267a5
KW
1586 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1587 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1588 * under those circumstances. */
1589 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1590 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1591 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1592 p, *lenp, U8);
1593 return LATIN_SMALL_LETTER_LONG_S;
1594 }
1595 else {
4f489194
KW
1596 *(p)++ = 's';
1597 *p = 's';
1598 *lenp = 2;
1599 return 's';
1ca267a5 1600 }
a1dde8de 1601 }
9b63e895 1602#endif
a1dde8de
KW
1603 else { /* In this range the fold of all other characters is their lower
1604 case */
1605 converted = toLOWER_LATIN1(c);
1606 }
1607
6f2d5cbc 1608 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1609 *p = (U8) converted;
1610 *lenp = 1;
1611 }
1612 else {
1613 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1614 *p = UTF8_TWO_BYTE_LO(converted);
1615 *lenp = 2;
1616 }
1617
1618 return converted;
1619}
1620
1621UV
31f05a37 1622Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1623{
4b593389 1624
a0270393
KW
1625 /* Not currently externally documented, and subject to change
1626 * <flags> bits meanings:
1627 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1628 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1629 * locale are to be used.
a0270393
KW
1630 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1631 */
4b593389 1632
36bb2ab6 1633 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1634
780fcc9f
KW
1635 if (flags & FOLD_FLAGS_LOCALE) {
1636 /* Treat a UTF-8 locale as not being in locale at all */
1637 if (IN_UTF8_CTYPE_LOCALE) {
1638 flags &= ~FOLD_FLAGS_LOCALE;
1639 }
1640 else {
1641 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1642 goto needs_full_generality;
780fcc9f 1643 }
31f05a37
KW
1644 }
1645
a1dde8de 1646 if (c < 256) {
e7b7ac46 1647 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1648 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1649 }
1650
2f306ab9 1651 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1652 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1653 uvchr_to_utf8(p, c);
b9992569 1654 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
1655 }
1656 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1657 the special flags. */
1658 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1659
1660 needs_full_generality:
a0270393 1661 uvchr_to_utf8(utf8_c, c);
445bf929 1662 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1663 }
84afefe6
JH
1664}
1665
26483009 1666PERL_STATIC_INLINE bool
5141f98e 1667S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1668 const char *const swashname, SV* const invlist)
bde6a22d 1669{
ea317ccb
KW
1670 /* returns a boolean giving whether or not the UTF8-encoded character that
1671 * starts at <p> is in the swash indicated by <swashname>. <swash>
1672 * contains a pointer to where the swash indicated by <swashname>
1673 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1674 * look at <*swash> and only generate a swash if it is not null. <invlist>
1675 * is NULL or an inversion list that defines the swash. If not null, it
1676 * saves time during initialization of the swash.
ea317ccb
KW
1677 *
1678 * Note that it is assumed that the buffer length of <p> is enough to
1679 * contain all the bytes that comprise the character. Thus, <*p> should
1680 * have been checked before this call for mal-formedness enough to assure
1681 * that. */
1682
7918f24d
NC
1683 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1684
492a624f 1685 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1686 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1687 * as far as there being enough bytes available in it to accommodate the
1688 * character without reading beyond the end, and pass that number on to the
1689 * validating routine */
6302f837 1690 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1691 if (ckWARN_d(WARN_UTF8)) {
1692 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1693 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1694 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1695 what the malformation is */
1696 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1697 }
1698 }
1699 return FALSE;
1700 }
87367d5f
KW
1701 if (!*swash) {
1702 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1703 *swash = _core_swash_init("utf8",
1704
1705 /* Only use the name if there is no inversion
1706 * list; otherwise will go out to disk */
1707 (invlist) ? "" : swashname,
1708
1709 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1710 }
28123549 1711
bde6a22d
NC
1712 return swash_fetch(*swash, p, TRUE) != 0;
1713}
1714
1715bool
922e8cb4
KW
1716Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1717{
922e8cb4
KW
1718 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1719
1720 assert(classnum < _FIRST_NON_SWASH_CC);
1721
f25ce844
KW
1722 return is_utf8_common(p,
1723 &PL_utf8_swash_ptrs[classnum],
1724 swash_property_names[classnum],
1725 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1726}
1727
1728bool
f2645549 1729Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1730{
f2645549 1731 SV* invlist = NULL;
7918f24d 1732
f2645549 1733 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1734
f2645549
KW
1735 if (! PL_utf8_perl_idstart) {
1736 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1737 }
60071a22 1738 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1739}
1740
1741bool
f2645549 1742Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1743{
f2645549 1744 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1745
1746 if (*p == '_')
1747 return TRUE;
f25ce844 1748 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1749}
1750
1751bool
eba68aa0
KW
1752Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1753{
b24b43f7 1754 SV* invlist = NULL;
eba68aa0
KW
1755
1756 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1757
b24b43f7
KW
1758 if (! PL_utf8_perl_idcont) {
1759 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1760 }
60071a22 1761 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1762}
1763
eba68aa0 1764bool
f2645549 1765Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1766{
f2645549 1767 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1768
f25ce844 1769 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1770}
1771
1772bool
f2645549 1773Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1774{
f2645549 1775 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1776
f25ce844 1777 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1778}
1779
1780bool
7dbf68d2
KW
1781Perl__is_utf8_mark(pTHX_ const U8 *p)
1782{
7dbf68d2
KW
1783 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1784
f25ce844 1785 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1786}
1787
6b5c0936 1788/*
87cea99e 1789=for apidoc to_utf8_case
6b5c0936 1790
9da1e7cb
KW
1791Instead use the appropriate one of L</toUPPER_utf8>,
1792L</toTITLE_utf8>,
1793L</toLOWER_utf8>,
1794or L</toFOLD_utf8>.
1795
6fae5207 1796C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1797the character that is being converted. This routine assumes that the character
1798at C<p> is well-formed.
6b5c0936 1799
6fae5207
KW
1800C<ustrp> is a pointer to the character buffer to put the
1801conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1802of the result.
1803
6fae5207 1804C<swashp> is a pointer to the swash to use.
6b5c0936 1805
a1433954 1806Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1807and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1808but not always, a multicharacter mapping), is tried first.
6b5c0936 1809
4a8240a3
KW
1810C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1811any special mappings; C<""> means to use the special mappings. Values other
1812than these two are treated as the name of the hash containing the special
1813mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1814
796b6530
KW
1815C<normal> is a string like C<"ToLower"> which means the swash
1816C<%utf8::ToLower>.
0134edef 1817
760c7c2f
KW
1818Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1819unless those are turned off.
1820
0134edef 1821=cut */
6b5c0936 1822
2104c8d9 1823UV
9a957fbc
AL
1824Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1825 SV **swashp, const char *normal, const char *special)
a0ed51b3 1826{
b9992569
KW
1827 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1828
1829 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
1830}
1831
1832 /* change namve uv1 to 'from' */
6a4a25f4 1833STATIC UV
b9992569
KW
1834S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
1835 SV **swashp, const char *normal, const char *special)
1836{
0134edef 1837 STRLEN len = 0;
7918f24d 1838
b9992569 1839 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 1840
36eaa811
KW
1841 /* For code points that don't change case, we already know that the output
1842 * of this function is the unchanged input, so we can skip doing look-ups
1843 * for them. Unfortunately the case-changing code points are scattered
1844 * around. But there are some long consecutive ranges where there are no
1845 * case changing code points. By adding tests, we can eliminate the lookup
1846 * for all the ones in such ranges. This is currently done here only for
1847 * just a few cases where the scripts are in common use in modern commerce
1848 * (and scripts adjacent to those which can be included without additional
1849 * tests). */
1850
1851 if (uv1 >= 0x0590) {
1852 /* This keeps from needing further processing the code points most
1853 * likely to be used in the following non-cased scripts: Hebrew,
1854 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
1855 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
1856 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
1857 if (uv1 < 0x10A0) {
1858 goto cases_to_self;
1859 }
1860
1861 /* The following largish code point ranges also don't have case
1862 * changes, but khw didn't think they warranted extra tests to speed
1863 * them up (which would slightly slow down everything else above them):
1864 * 1100..139F Hangul Jamo, Ethiopic
1865 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
1866 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
1867 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
1868 * Combining Diacritical Marks Extended, Balinese,
1869 * Sundanese, Batak, Lepcha, Ol Chiki
1870 * 2000..206F General Punctuation
1871 */
1872
1873 if (uv1 >= 0x2D30) {
1874
1875 /* This keeps the from needing further processing the code points
1876 * most likely to be used in the following non-cased major scripts:
1877 * CJK, Katakana, Hiragana, plus some less-likely scripts.
1878 *
1879 * (0x2D30 above might have to be changed to 2F00 in the unlikely
1880 * event that Unicode eventually allocates the unused block as of
1881 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
1882 * that the test suite will start having failures to alert you
1883 * should that happen) */
1884 if (uv1 < 0xA640) {
1885 goto cases_to_self;
1886 }
1887
1888 if (uv1 >= 0xAC00) {
1889 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
1890 if (ckWARN_d(WARN_SURROGATE)) {
1891 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1892 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1893 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1894 }
1895 goto cases_to_self;
1896 }
36eaa811
KW
1897
1898 /* AC00..FAFF Catches Hangul syllables and private use, plus
1899 * some others */
1900 if (uv1 < 0xFB00) {
1901 goto cases_to_self;
1902
1903 }
1904
5af9bc97
KW
1905 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
1906 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1907 && ckWARN_d(WARN_DEPRECATED))
1908 {
1909 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1910 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1911 }
1912 if (ckWARN_d(WARN_NON_UNICODE)) {
1913 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1914 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1915 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1916 }
1917 goto cases_to_self;
1918 }
3bfc1e70
KW
1919#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
1920 if (UNLIKELY(uv1
1921 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
1922 {
1923
1924 /* As of this writing, this means we avoid swash creation
1925 * for anything beyond low Plane 1 */
1926 goto cases_to_self;
1927 }
1928#endif
36eaa811
KW
1929 }
1930 }
9ae3ac1a 1931
36eaa811
KW
1932 /* Note that non-characters are perfectly legal, so no warning should
1933 * be given. There are so few of them, that it isn't worth the extra
1934 * tests to avoid swash creation */
9ae3ac1a
KW
1935 }
1936
0134edef 1937 if (!*swashp) /* load on-demand */
5ab9d2ef 1938 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1939
a6f87d8c 1940 if (special) {
0134edef 1941 /* It might be "special" (sometimes, but not always,
2a37f04d 1942 * a multicharacter mapping) */
4a8240a3 1943 HV *hv = NULL;
b08cf34e
JH
1944 SV **svp;
1945
4a8240a3
KW
1946 /* If passed in the specials name, use that; otherwise use any
1947 * given in the swash */
1948 if (*special != '\0') {
1949 hv = get_hv(special, 0);
1950 }
1951 else {
1952 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1953 if (svp) {
1954 hv = MUTABLE_HV(SvRV(*svp));
1955 }
1956 }
1957
176fe009 1958 if (hv
5f560d8a 1959 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
1960 && (*svp))
1961 {
cfd0369c 1962 const char *s;
47654450 1963
cfd0369c 1964 s = SvPV_const(*svp, len);
47654450 1965 if (len == 1)
f4cd282c 1966 /* EIGHTBIT */
c80e42f3 1967 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1968 else {
d2dcd0fb 1969 Copy(s, ustrp, len, U8);
29e98929 1970 }
983ffd37 1971 }
0134edef
JH
1972 }
1973
1974 if (!len && *swashp) {
4a4088c4 1975 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 1976
0134edef
JH
1977 if (uv2) {
1978 /* It was "normal" (a single character mapping). */
f4cd282c 1979 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1980 }
1981 }
1feea2c7 1982
cbe07460
KW
1983 if (len) {
1984 if (lenp) {
1985 *lenp = len;
1986 }
1987 return valid_utf8_to_uvchr(ustrp, 0);
1988 }
1989
1990 /* Here, there was no mapping defined, which means that the code point maps
1991 * to itself. Return the inputs */
e24dfe9c 1992 cases_to_self:
bfdf22ec 1993 len = UTF8SKIP(p);
ca9fab46
KW
1994 if (p != ustrp) { /* Don't copy onto itself */
1995 Copy(p, ustrp, len, U8);
1996 }
0134edef 1997
2a37f04d
JH
1998 if (lenp)
1999 *lenp = len;
2000
f4cd282c 2001 return uv1;
cbe07460 2002
a0ed51b3
LW
2003}
2004
051a06d4 2005STATIC UV
357aadde 2006S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2007{
4a4088c4 2008 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2009 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2010 * result contains a character that crosses the 255/256 boundary, disallow
2011 * the change, and return the original code point. See L<perlfunc/lc> for
2012 * why;
051a06d4 2013 *
a1433954
KW
2014 * p points to the original string whose case was changed; assumed
2015 * by this routine to be well-formed
051a06d4
KW
2016 * result the code point of the first character in the changed-case string
2017 * ustrp points to the changed-case string (<result> represents its first char)
2018 * lenp points to the length of <ustrp> */
2019
2020 UV original; /* To store the first code point of <p> */
2021
2022 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2023
a4f12ed7 2024 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2025
2026 /* We know immediately if the first character in the string crosses the
2027 * boundary, so can skip */
2028 if (result > 255) {
2029
2030 /* Look at every character in the result; if any cross the
2031 * boundary, the whole thing is disallowed */
2032 U8* s = ustrp + UTF8SKIP(ustrp);
2033 U8* e = ustrp + *lenp;
2034 while (s < e) {
a4f12ed7 2035 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2036 goto bad_crossing;
2037 }
2038 s += UTF8SKIP(s);
2039 }
2040
613abc6d
KW
2041 /* Here, no characters crossed, result is ok as-is, but we warn. */
2042 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2043 return result;
2044 }
2045
7b52d656 2046 bad_crossing:
051a06d4
KW
2047
2048 /* Failed, have to return the original */
4b88fb76 2049 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2050
2051 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2052 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2053 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2054 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2055 OP_DESC(PL_op),
ab0b796c
KW
2056 original,
2057 original);
051a06d4
KW
2058 Copy(p, ustrp, *lenp, char);
2059 return original;
2060}
2061
d3e79532 2062/*
87cea99e 2063=for apidoc to_utf8_upper
d3e79532 2064
1f607577 2065Instead use L</toUPPER_utf8>.
a1433954 2066
d3e79532
JH
2067=cut */
2068
051a06d4 2069/* Not currently externally documented, and subject to change:
31f05a37
KW
2070 * <flags> is set iff iff the rules from the current underlying locale are to
2071 * be used. */
051a06d4 2072
2104c8d9 2073UV
31f05a37 2074Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2075{
051a06d4
KW
2076 UV result;
2077
2078 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2079
780fcc9f
KW
2080 if (flags) {
2081 /* Treat a UTF-8 locale as not being in locale at all */
2082 if (IN_UTF8_CTYPE_LOCALE) {
2083 flags = FALSE;
2084 }
2085 else {
2086 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2087 }
31f05a37
KW
2088 }
2089
3a4c58c9 2090 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2091 if (flags) {
2092 result = toUPPER_LC(*p);
2093 }
2094 else {
81c6c7ce 2095 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2096 }
3a4c58c9
KW
2097 }
2098 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2099 if (flags) {
a62b247b 2100 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2101 result = toUPPER_LC(c);
051a06d4
KW
2102 }
2103 else {
a62b247b 2104 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2105 ustrp, lenp, 'S');
051a06d4
KW
2106 }
2107 }
4a4088c4 2108 else { /* UTF-8, ord above 255 */
b9992569 2109 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2110
2111 if (flags) {
357aadde 2112 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2113 }
2114 return result;
2115 }
2116
4a4088c4 2117 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2118 if (UTF8_IS_INVARIANT(result)) {
2119 *ustrp = (U8) result;
2120 *lenp = 1;
2121 }
2122 else {
62cb07ea
KW
2123 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2124 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2125 *lenp = 2;
3a4c58c9 2126 }
baa60164 2127
051a06d4 2128 return result;
983ffd37 2129}
a0ed51b3 2130
d3e79532 2131/*
87cea99e 2132=for apidoc to_utf8_title
d3e79532 2133
1f607577 2134Instead use L</toTITLE_utf8>.
a1433954 2135
d3e79532
JH
2136=cut */
2137
051a06d4 2138/* Not currently externally documented, and subject to change:
31f05a37
KW
2139 * <flags> is set iff the rules from the current underlying locale are to be
2140 * used. Since titlecase is not defined in POSIX, for other than a
2141 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2142 */
051a06d4 2143
983ffd37 2144UV
31f05a37 2145Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2146{
051a06d4
KW
2147 UV result;
2148
2149 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2150
780fcc9f
KW
2151 if (flags) {
2152 /* Treat a UTF-8 locale as not being in locale at all */
2153 if (IN_UTF8_CTYPE_LOCALE) {
2154 flags = FALSE;
2155 }
2156 else {
2157 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2158 }
31f05a37
KW
2159 }
2160
3a4c58c9 2161 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2162 if (flags) {
2163 result = toUPPER_LC(*p);
2164 }
2165 else {
81c6c7ce 2166 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2167 }
3a4c58c9
KW
2168 }
2169 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2170 if (flags) {
a62b247b 2171 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2172 result = toUPPER_LC(c);
051a06d4
KW
2173 }
2174 else {
a62b247b 2175 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2176 ustrp, lenp, 's');
051a06d4
KW
2177 }
2178 }
4a4088c4 2179 else { /* UTF-8, ord above 255 */
b9992569 2180 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2181
2182 if (flags) {
357aadde 2183 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2184 }
2185 return result;
2186 }
2187
4a4088c4 2188 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2189 if (UTF8_IS_INVARIANT(result)) {
2190 *ustrp = (U8) result;
2191 *lenp = 1;
2192 }
2193 else {
62cb07ea
KW
2194 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2195 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2196 *lenp = 2;
3a4c58c9
KW
2197 }
2198
051a06d4 2199 return result;
a0ed51b3
LW
2200}
2201
d3e79532 2202/*
87cea99e 2203=for apidoc to_utf8_lower
d3e79532 2204
1f607577 2205Instead use L</toLOWER_utf8>.
a1433954 2206
d3e79532
JH
2207=cut */
2208
051a06d4 2209/* Not currently externally documented, and subject to change:
31f05a37
KW
2210 * <flags> is set iff iff the rules from the current underlying locale are to
2211 * be used.
2212 */
051a06d4 2213
2104c8d9 2214UV
31f05a37 2215Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2216{
051a06d4
KW
2217 UV result;
2218
051a06d4 2219 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2220
780fcc9f
KW
2221 if (flags) {
2222 /* Treat a UTF-8 locale as not being in locale at all */
2223 if (IN_UTF8_CTYPE_LOCALE) {
2224 flags = FALSE;
2225 }
2226 else {
2227 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2228 }
31f05a37
KW
2229 }
2230
968c5e6a 2231 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2232 if (flags) {
2233 result = toLOWER_LC(*p);
2234 }
2235 else {
81c6c7ce 2236 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2237 }
968c5e6a
KW
2238 }
2239 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2240 if (flags) {
a62b247b 2241 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2242 result = toLOWER_LC(c);
051a06d4
KW
2243 }
2244 else {
a62b247b 2245 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2246 ustrp, lenp);
051a06d4 2247 }
968c5e6a 2248 }
4a4088c4 2249 else { /* UTF-8, ord above 255 */
b9992569 2250 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2251
2252 if (flags) {
357aadde 2253 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2254 }
968c5e6a 2255
051a06d4
KW
2256 return result;
2257 }
2258
4a4088c4 2259 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2260 if (UTF8_IS_INVARIANT(result)) {
2261 *ustrp = (U8) result;
2262 *lenp = 1;
2263 }
2264 else {
62cb07ea
KW
2265 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2266 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2267 *lenp = 2;
2268 }
2269
051a06d4 2270 return result;
b4e400f9
JH
2271}
2272
d3e79532 2273/*
87cea99e 2274=for apidoc to_utf8_fold
d3e79532 2275
1f607577 2276Instead use L</toFOLD_utf8>.
a1433954 2277
d3e79532
JH
2278=cut */
2279
051a06d4
KW
2280/* Not currently externally documented, and subject to change,
2281 * in <flags>
31f05a37
KW
2282 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2283 * locale are to be used.
051a06d4
KW
2284 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2285 * otherwise simple folds
a0270393
KW
2286 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2287 * prohibited
445bf929 2288 */
36bb2ab6 2289
b4e400f9 2290UV
445bf929 2291Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2292{
051a06d4
KW
2293 UV result;
2294
36bb2ab6 2295 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2296
a0270393
KW
2297 /* These are mutually exclusive */
2298 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2299
50ba90ff
KW
2300 assert(p != ustrp); /* Otherwise overwrites */
2301
780fcc9f
KW
2302 if (flags & FOLD_FLAGS_LOCALE) {
2303 /* Treat a UTF-8 locale as not being in locale at all */
2304 if (IN_UTF8_CTYPE_LOCALE) {
2305 flags &= ~FOLD_FLAGS_LOCALE;
2306 }
2307 else {
2308 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2309 }
31f05a37
KW
2310 }
2311
a1dde8de 2312 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2313 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2314 result = toFOLD_LC(*p);
051a06d4
KW
2315 }
2316 else {
81c6c7ce 2317 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2318 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2319 }
a1dde8de
KW
2320 }
2321 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2322 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2323 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2324 result = toFOLD_LC(c);
051a06d4
KW
2325 }
2326 else {
a62b247b 2327 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2328 ustrp, lenp,
1ca267a5 2329 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2330 }
a1dde8de 2331 }
4a4088c4 2332 else { /* UTF-8, ord above 255 */
b9992569 2333 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2334
1ca267a5
KW
2335 if (flags & FOLD_FLAGS_LOCALE) {
2336
76f2ffcd 2337# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2338 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2339
2340# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2341# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2342
2343 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2344
538e84ed
KW
2345 /* Special case these two characters, as what normally gets
2346 * returned under locale doesn't work */
76f2ffcd
KW
2347 if (UTF8SKIP(p) == cap_sharp_s_len
2348 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2349 {
ab0b796c
KW
2350 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2351 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2352 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2353 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2354 goto return_long_s;
2355 }
0766489e
KW
2356 else
2357#endif
2358 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2359 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2360 {
ab0b796c
KW
2361 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2362 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2363 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2364 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2365 goto return_ligature_st;
2366 }
74894415
KW
2367
2368#if UNICODE_MAJOR_VERSION == 3 \
2369 && UNICODE_DOT_VERSION == 0 \
2370 && UNICODE_DOT_DOT_VERSION == 1
2371# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2372
2373 /* And special case this on this Unicode version only, for the same
2374 * reaons the other two are special cased. They would cross the
2375 * 255/256 boundary which is forbidden under /l, and so the code
2376 * wouldn't catch that they are equivalent (which they are only in
2377 * this release) */
2378 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2379 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2380 {
2381 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2382 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2383 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2384 "resolved to \"\\x{0131}\".");
2385 goto return_dotless_i;
2386 }
2387#endif
2388
357aadde 2389 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2390 }
a0270393
KW
2391 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2392 return result;
2393 }
2394 else {
4a4088c4 2395 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2396 * character above the ASCII range, and the result should not
2397 * contain an ASCII character. */
a0270393
KW
2398
2399 UV original; /* To store the first code point of <p> */
2400
2401 /* Look at every character in the result; if any cross the
2402 * boundary, the whole thing is disallowed */
2403 U8* s = ustrp;
2404 U8* e = ustrp + *lenp;
2405 while (s < e) {
2406 if (isASCII(*s)) {
2407 /* Crossed, have to return the original */
2408 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2409
9fc2026f 2410 /* But in these instances, there is an alternative we can
1ca267a5 2411 * return that is valid */
0766489e
KW
2412 if (original == LATIN_SMALL_LETTER_SHARP_S
2413#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2414 || original == LATIN_CAPITAL_LETTER_SHARP_S
2415#endif
2416 ) {
1ca267a5
KW
2417 goto return_long_s;
2418 }
9fc2026f
KW
2419 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2420 goto return_ligature_st;
2421 }
74894415
KW
2422#if UNICODE_MAJOR_VERSION == 3 \
2423 && UNICODE_DOT_VERSION == 0 \
2424 && UNICODE_DOT_DOT_VERSION == 1
2425
2426 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2427 goto return_dotless_i;
2428 }
2429#endif
a0270393
KW
2430 Copy(p, ustrp, *lenp, char);
2431 return original;
2432 }
2433 s += UTF8SKIP(s);
2434 }
051a06d4 2435
a0270393
KW
2436 /* Here, no characters crossed, result is ok as-is */
2437 return result;
2438 }
051a06d4
KW
2439 }
2440
4a4088c4 2441 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2442 if (UTF8_IS_INVARIANT(result)) {
2443 *ustrp = (U8) result;
2444 *lenp = 1;
2445 }
2446 else {
62cb07ea
KW
2447 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2448 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2449 *lenp = 2;
2450 }
2451
051a06d4 2452 return result;
1ca267a5
KW
2453
2454 return_long_s:
2455 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2456 * folds to a string of two of these characters. By returning this
2457 * instead, then, e.g.,
2458 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2459 * works. */
2460
2461 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2462 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2463 ustrp, *lenp, U8);
2464 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2465
2466 return_ligature_st:
2467 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2468 * have the other one fold to it */
2469
2470 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2471 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2472 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2473
2474#if UNICODE_MAJOR_VERSION == 3 \
2475 && UNICODE_DOT_VERSION == 0 \
2476 && UNICODE_DOT_DOT_VERSION == 1
2477
2478 return_dotless_i:
2479 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2480 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2481 return LATIN_SMALL_LETTER_DOTLESS_I;
2482
2483#endif
2484
a0ed51b3
LW
2485}
2486
711a919c 2487/* Note:
f90a9a02 2488 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2489 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2490 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2491 */
c4a5db0c 2492
a0ed51b3 2493SV*
7fc63493 2494Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2495{
c4a5db0c
KW
2496 PERL_ARGS_ASSERT_SWASH_INIT;
2497
2498 /* Returns a copy of a swash initiated by the called function. This is the
2499 * public interface, and returning a copy prevents others from doing
2500 * mischief on the original */
2501
5d3d13d1 2502 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2503}
2504
2505SV*
5d3d13d1 2506Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2507{
2c1f00b9
YO
2508
2509 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2510 * use the following define */
2511
2512#define CORE_SWASH_INIT_RETURN(x) \
2513 PL_curpm= old_PL_curpm; \
2514 return x
2515
c4a5db0c 2516 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2517 * by calling utf8_heavy.pl in the general case. The returned value may be
2518 * the swash's inversion list instead if the input parameters allow it.
2519 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2520 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2521 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2522 *
2523 * This interface should only be used by functions that won't destroy or
2524 * adversely change the swash, as doing so affects all other uses of the
2525 * swash in the program; the general public should use 'Perl_swash_init'
2526 * instead.
2527 *
2528 * pkg is the name of the package that <name> should be in.
2529 * name is the name of the swash to find. Typically it is a Unicode
2530 * property name, including user-defined ones
2531 * listsv is a string to initialize the swash with. It must be of the form
2532 * documented as the subroutine return value in
2533 * L<perlunicode/User-Defined Character Properties>
2534 * minbits is the number of bits required to represent each data element.
2535 * It is '1' for binary properties.
2536 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2537 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2538 * flags_p if non-NULL is the address of various input and output flag bits
2539 * to the routine, as follows: ('I' means is input to the routine;
2540 * 'O' means output from the routine. Only flags marked O are
2541 * meaningful on return.)
2542 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2543 * came from a user-defined property. (I O)
5d3d13d1
KW
2544 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2545 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2546 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2547 * return of an inversion list instead of a swash hash if this routine
2548 * thinks that would result in faster execution of swash_fetch() later
2549 * on. (I)
9a53f6cf
KW
2550 *
2551 * Thus there are three possible inputs to find the swash: <name>,
2552 * <listsv>, and <invlist>. At least one must be specified. The result
2553 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2554 * actions can intersect, etc. what <name> gives. To avoid going out to
2555 * disk at all, <invlist> should specify completely what the swash should
2556 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2557 *
2558 * <invlist> is only valid for binary properties */
c4a5db0c 2559
2c1f00b9
YO
2560 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2561
c4a5db0c 2562 SV* retval = &PL_sv_undef;
83199d38 2563 HV* swash_hv = NULL;
87367d5f
KW
2564 const int invlist_swash_boundary =
2565 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2566 ? 512 /* Based on some benchmarking, but not extensive, see commit
2567 message */
2568 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2569
2570 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2571 assert(! invlist || minbits == 1);
2572
2c1f00b9
YO
2573 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2574 that triggered the swash init and the swash init perl logic itself.
2575 See perl #122747 */
2576
9a53f6cf
KW
2577 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2578 * so */
2579 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2580 dSP;
2581 const size_t pkg_len = strlen(pkg);
2582 const size_t name_len = strlen(name);
2583 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2584 SV* errsv_save;
2585 GV *method;
2586
2587 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2588
2589 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2590 ENTER;
69794297 2591 SAVEHINTS();
2782061f 2592 save_re_context();
650f067c
JL
2593 /* We might get here via a subroutine signature which uses a utf8
2594 * parameter name, at which point PL_subname will have been set
2595 * but not yet used. */
2596 save_item(PL_subname);
69794297
KW
2597 if (PL_parser && PL_parser->error_count)
2598 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2599 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2600 if (!method) { /* demand load UTF-8 */
69794297 2601 ENTER;
db2c6cb3
FC
2602 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2603 GvSV(PL_errgv) = NULL;
1a419e6b 2604#ifndef NO_TAINT_SUPPORT
69794297
KW
2605 /* It is assumed that callers of this routine are not passing in
2606 * any user derived data. */
2782061f
DM
2607 /* Need to do this after save_re_context() as it will set
2608 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2609 * in Perl_magic_get). Even line to create errsv_save can turn on
2610 * PL_tainted. */
284167a5
SM
2611 SAVEBOOL(TAINT_get);
2612 TAINT_NOT;
2613#endif
69794297
KW
2614 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2615 NULL);
eed484f9 2616 {
db2c6cb3
FC
2617 /* Not ERRSV, as there is no need to vivify a scalar we are
2618 about to discard. */
2619 SV * const errsv = GvSV(PL_errgv);
2620 if (!SvTRUE(errsv)) {
2621 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2622 SvREFCNT_dec(errsv);
2623 }
eed484f9 2624 }
69794297
KW
2625 LEAVE;
2626 }
2627 SPAGAIN;
2628 PUSHMARK(SP);
2629 EXTEND(SP,5);
2630 mPUSHp(pkg, pkg_len);
2631 mPUSHp(name, name_len);
2632 PUSHs(listsv);
2633 mPUSHi(minbits);
2634 mPUSHi(none);
2635 PUTBACK;
db2c6cb3
FC
2636 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2637 GvSV(PL_errgv) = NULL;
69794297
KW
2638 /* If we already have a pointer to the method, no need to use
2639 * call_method() to repeat the lookup. */
c41800a8
KW
2640 if (method
2641 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2642 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2643 {
2644 retval = *PL_stack_sp--;
2645 SvREFCNT_inc(retval);
2646 }
eed484f9 2647 {
db2c6cb3
FC
2648 /* Not ERRSV. See above. */
2649 SV * const errsv = GvSV(PL_errgv);
2650 if (!SvTRUE(errsv)) {
2651 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2652 SvREFCNT_dec(errsv);
2653 }
eed484f9 2654 }
ce3b816e 2655 LEAVE;
69794297
KW
2656 POPSTACK;
2657 if (IN_PERL_COMPILETIME) {
2658 CopHINTS_set(PL_curcop, PL_hints);
2659 }
2660 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 2661 if (SvPOK(retval)) {
69794297
KW
2662
2663 /* If caller wants to handle missing properties, let them */
5d3d13d1 2664 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2665 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2666 }
2667 Perl_croak(aTHX_
2668 "Can't find Unicode property definition \"%"SVf"\"",
2669 SVfARG(retval));
a25b5927 2670 NOT_REACHED; /* NOTREACHED */
d95e4a00 2671 }
69794297 2672 }
9a53f6cf 2673 } /* End of calling the module to find the swash */
36eb48b4 2674
83199d38
KW
2675 /* If this operation fetched a swash, and we will need it later, get it */
2676 if (retval != &PL_sv_undef
2677 && (minbits == 1 || (flags_p
2678 && ! (*flags_p
2679 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2680 {
2681 swash_hv = MUTABLE_HV(SvRV(retval));
2682
2683 /* If we don't already know that there is a user-defined component to
2684 * this swash, and the user has indicated they wish to know if there is
2685 * one (by passing <flags_p>), find out */
2686 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2687 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2688 if (user_defined && SvUV(*user_defined)) {
2689 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2690 }
2691 }
2692 }
2693
36eb48b4
KW
2694 /* Make sure there is an inversion list for binary properties */
2695 if (minbits == 1) {
2696 SV** swash_invlistsvp = NULL;
2697 SV* swash_invlist = NULL;
9a53f6cf 2698 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2699 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2700 an unclaimed reference count */
36eb48b4 2701
9a53f6cf 2702 /* If this operation fetched a swash, get its already existing
83199d38 2703 * inversion list, or create one for it */
36eb48b4 2704
83199d38 2705 if (swash_hv) {
5c9f4bd2 2706 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2707 if (swash_invlistsvp) {
2708 swash_invlist = *swash_invlistsvp;
2709 invlist_in_swash_is_valid = TRUE;
2710 }
2711 else {
36eb48b4 2712 swash_invlist = _swash_to_invlist(retval);
02c85471 2713 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2714 }
2715 }
2716
2717 /* If an inversion list was passed in, have to include it */
2718 if (invlist) {
2719
2720 /* Any fetched swash will by now have an inversion list in it;
2721 * otherwise <swash_invlist> will be NULL, indicating that we
2722 * didn't fetch a swash */
2723 if (swash_invlist) {
2724
2725 /* Add the passed-in inversion list, which invalidates the one
2726 * already stored in the swash */
2727 invlist_in_swash_is_valid = FALSE;
2728 _invlist_union(invlist, swash_invlist, &swash_invlist);
2729 }
2730 else {
2731
87367d5f
KW
2732 /* Here, there is no swash already. Set up a minimal one, if
2733 * we are going to return a swash */
2734 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2735 swash_hv = newHV();
4aca0fe6 2736 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2737 }
9a53f6cf
KW
2738 swash_invlist = invlist;
2739 }
9a53f6cf
KW
2740 }
2741
2742 /* Here, we have computed the union of all the passed-in data. It may
2743 * be that there was an inversion list in the swash which didn't get
538e84ed 2744 * touched; otherwise save the computed one */
87367d5f
KW
2745 if (! invlist_in_swash_is_valid
2746 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2747 {
5c9f4bd2 2748 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2749 {
2750 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2751 }
cc34d8c5
FC
2752 /* We just stole a reference count. */
2753 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2754 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2755 }
87367d5f 2756
dbfdbd26
KW
2757 SvREADONLY_on(swash_invlist);
2758
c41800a8 2759 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2760 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2761 SvREFCNT_dec(retval);
02c85471
FC
2762 if (!swash_invlist_unclaimed)
2763 SvREFCNT_inc_simple_void_NN(swash_invlist);
2764 retval = newRV_noinc(swash_invlist);
87367d5f 2765 }
36eb48b4
KW
2766 }
2767
2c1f00b9
YO
2768 CORE_SWASH_INIT_RETURN(retval);
2769#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2770}
2771
035d37be
JH
2772
2773/* This API is wrong for special case conversions since we may need to
2774 * return several Unicode characters for a single Unicode character
2775 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2776 * the lower-level routine, and it is similarly broken for returning
38684baa 2777 * multiple values. --jhi
b9992569 2778 * For those, you should use S__to_utf8_case() instead */
b0e3252e 2779/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2780
2781/* Note:
2782 * Returns the value of property/mapping C<swash> for the first character
2783 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2784 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2785 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2786 *
2787 * A "swash" is a hash which contains initially the keys/values set up by
2788 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2789 * property for all possible code points. Things are stored in a compact form
2790 * (see utf8_heavy.pl) so that calculation is required to find the actual
2791 * property value for a given code point. As code points are looked up, new
2792 * key/value pairs are added to the hash, so that the calculation doesn't have
2793 * to ever be re-done. Further, each calculation is done, not just for the
2794 * desired one, but for a whole block of code points adjacent to that one.
2795 * For binary properties on ASCII machines, the block is usually for 64 code
2796 * points, starting with a code point evenly divisible by 64. Thus if the
2797 * property value for code point 257 is requested, the code goes out and
2798 * calculates the property values for all 64 code points between 256 and 319,
2799 * and stores these as a single 64-bit long bit vector, called a "swatch",
2800 * under the key for code point 256. The key is the UTF-8 encoding for code
2801 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2802 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2803 * for code point 258 is then requested, this code realizes that it would be
2804 * stored under the key for 256, and would find that value and extract the
2805 * relevant bit, offset from 256.
2806 *
2807 * Non-binary properties are stored in as many bits as necessary to represent
2808 * their values (32 currently, though the code is more general than that), not
fc273927 2809 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
2810 * vector that encompasses the property values for all code points whose UTF-8
2811 * representations are represented by the key. That is, for all code points
2812 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2813 * bytes of that.
680c470c 2814 */
a0ed51b3 2815UV
680c470c 2816Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2817{
ef8f7699 2818 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2819 U32 klen;
2820 U32 off;
9b56a019 2821 STRLEN slen = 0;
7d85a32c 2822 STRLEN needents;
cfd0369c 2823 const U8 *tmps = NULL;
979f2922 2824 SV *swatch;
08fb1ac5 2825 const U8 c = *ptr;
3568d838 2826
7918f24d
NC
2827 PERL_ARGS_ASSERT_SWASH_FETCH;
2828
87367d5f
KW
2829 /* If it really isn't a hash, it isn't really swash; must be an inversion
2830 * list */
2831 if (SvTYPE(hv) != SVt_PVHV) {
2832 return _invlist_contains_cp((SV*)hv,
2833 (do_utf8)
2834 ? valid_utf8_to_uvchr(ptr, NULL)
2835 : c);
2836 }
2837
08fb1ac5
KW
2838 /* We store the values in a "swatch" which is a vec() value in a swash
2839 * hash. Code points 0-255 are a single vec() stored with key length
2840 * (klen) 0. All other code points have a UTF-8 representation
2841 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2842 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2843 * length for them is the length of the encoded char - 1. ptr[klen] is the
2844 * final byte in the sequence representing the character */
2845 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2846 klen = 0;
2847 needents = 256;
2848 off = c;
3568d838 2849 }
08fb1ac5
KW
2850 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2851 klen = 0;
2852 needents = 256;
a62b247b 2853 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2854 }
2855 else {
08fb1ac5
KW
2856 klen = UTF8SKIP(ptr) - 1;
2857
2858 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2859 * the vec is the final byte in the sequence. (In EBCDIC this is
2860 * converted to I8 to get consecutive values.) To help you visualize
2861 * all this:
2862 * Straight 1047 After final byte
2863 * UTF-8 UTF-EBCDIC I8 transform
2864 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2865 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2866 * ...
2867 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2868 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2869 * ...
2870 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2871 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2872 * ...
2873 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2874 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2875 * ...
2876 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2877 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2878 *
2879 * (There are no discontinuities in the elided (...) entries.)
2880 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2881 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2882 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2883 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2884 * index into the vec() swatch (after subtracting 0x80, which we
2885 * actually do with an '&').
2886 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2887 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2888 * dicontinuities which go away by transforming it into I8, and we
2889 * effectively subtract 0xA0 to get the index. */
979f2922 2890 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2891 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2892 }
7d85a32c 2893
a0ed51b3 2894 /*
4a4088c4 2895 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2896 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2897 * it's nothing to sniff at.) Pity we usually come through at least
2898 * two function calls to get here...
2899 *
2900 * NB: this code assumes that swatches are never modified, once generated!
2901 */
2902
3568d838 2903 if (hv == PL_last_swash_hv &&
a0ed51b3 2904 klen == PL_last_swash_klen &&
27da23d5 2905 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2906 {
2907 tmps = PL_last_swash_tmps;
2908 slen = PL_last_swash_slen;
2909 }
2910 else {
2911 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2912 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2913
b0e3252e 2914 /* If not cached, generate it via swatch_get */
979f2922 2915 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2916 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2917 {
2918 if (klen) {
2919 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2920 swatch = swatch_get(swash,
2921 code_point & ~((UV)needents - 1),
2922 needents);
2923 }
2924 else { /* For the first 256 code points, the swatch has a key of
2925 length 0 */
2926 swatch = swatch_get(swash, 0, needents);
2927 }
979f2922 2928
923e4eb5 2929 if (IN_PERL_COMPILETIME)
623e6609 2930 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2931
979f2922 2932 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2933
979f2922
ST
2934 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2935 || (slen << 3) < needents)
5637ef5b
NC
2936 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2937 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2938 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2939 }
2940
2941 PL_last_swash_hv = hv;
16d8f38a 2942 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2943 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2944 /* FIXME change interpvar.h? */
2945 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2946 PL_last_swash_slen = slen;
2947 if (klen)
2948 Copy(ptr, PL_last_swash_key, klen, U8);
2949 }
2950
9faf8d75 2951 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2952 case 1:
e7aca353 2953 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2954 case 8:
e7aca353 2955 return ((UV) tmps[off]);
a0ed51b3
LW
2956 case 16:
2957 off <<= 1;
e7aca353
JH
2958 return
2959 ((UV) tmps[off ] << 8) +
2960 ((UV) tmps[off + 1]);
a0ed51b3
LW
2961 case 32:
2962 off <<= 2;
e7aca353
JH
2963 return
2964 ((UV) tmps[off ] << 24) +
2965 ((UV) tmps[off + 1] << 16) +
2966 ((UV) tmps[off + 2] << 8) +
2967 ((UV) tmps[off + 3]);
a0ed51b3 2968 }
5637ef5b
NC
2969 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2970 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2971 NORETURN_FUNCTION_END;
a0ed51b3 2972}
2b9d42f0 2973
319009ee
KW
2974/* Read a single line of the main body of the swash input text. These are of
2975 * the form:
2976 * 0053 0056 0073
2977 * where each number is hex. The first two numbers form the minimum and
2978 * maximum of a range, and the third is the value associated with the range.
2979 * Not all swashes should have a third number
2980 *
2981 * On input: l points to the beginning of the line to be examined; it points
2982 * to somewhere in the string of the whole input text, and is
2983 * terminated by a \n or the null string terminator.
2984 * lend points to the null terminator of that string
2985 * wants_value is non-zero if the swash expects a third number
2986 * typestr is the name of the swash's mapping, like 'ToLower'
2987 * On output: *min, *max, and *val are set to the values read from the line.
2988 * returns a pointer just beyond the line examined. If there was no
2989 * valid min number on the line, returns lend+1
2990 */
2991
2992STATIC U8*
2993S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2994 const bool wants_value, const U8* const typestr)
2995{
2996 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2997 STRLEN numlen; /* Length of the number */
02470786
KW
2998 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2999 | PERL_SCAN_DISALLOW_PREFIX
3000 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3001
3002 /* nl points to the next \n in the scan */
3003 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3004
95543e92
KW
3005 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3006
319009ee
KW
3007 /* Get the first number on the line: the range minimum */
3008 numlen = lend - l;
3009 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 3010 *max = *min; /* So can never return without setting max */
319009ee
KW
3011 if (numlen) /* If found a hex number, position past it */
3012 l += numlen;
3013 else if (nl) { /* Else, go handle next line, if any */
3014 return nl + 1; /* 1 is length of "\n" */
3015 }
3016 else { /* Else, no next line */
3017 return lend + 1; /* to LIST's end at which \n is not found */
3018 }
3019
3020 /* The max range value follows, separated by a BLANK */
3021 if (isBLANK(*l)) {
3022 ++l;
02470786
KW
3023 flags = PERL_SCAN_SILENT_ILLDIGIT
3024 | PERL_SCAN_DISALLOW_PREFIX
3025 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3026 numlen = lend - l;
3027 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3028 if (numlen)
3029 l += numlen;
3030 else /* If no value here, it is a single element range */
3031 *max = *min;
3032
3033 /* Non-binary tables have a third entry: what the first element of the
24303724 3034 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3035 if (wants_value) {
3036 if (isBLANK(*l)) {
3037 ++l;
f2a7d0fc
KW
3038 flags = PERL_SCAN_SILENT_ILLDIGIT
3039 | PERL_SCAN_DISALLOW_PREFIX
3040 | PERL_SCAN_SILENT_NON_PORTABLE;
3041 numlen = lend - l;
3042 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3043 if (numlen)
3044 l += numlen;
3045 else
3046 *val = 0;
319009ee
KW
3047 }
3048 else {
3049 *val = 0;
3050 if (typeto) {
dcbac5bb 3051 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3052 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3053 typestr, l);
3054 }
3055 }
3056 }
3057 else
3058 *val = 0; /* bits == 1, then any val should be ignored */
3059 }
3060 else { /* Nothing following range min, should be single element with no
3061 mapping expected */
319009ee
KW
3062 if (wants_value) {
3063 *val = 0;
3064 if (typeto) {
dcbac5bb 3065 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3066 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3067 }
3068 }
3069 else
3070 *val = 0; /* bits == 1, then val should be ignored */
3071 }
3072
3073 /* Position to next line if any, or EOF */
3074 if (nl)
3075 l = nl + 1;
3076 else
3077 l = lend;
3078
3079 return l;
3080}
3081
979f2922
ST
3082/* Note:
3083 * Returns a swatch (a bit vector string) for a code point sequence
3084 * that starts from the value C<start> and comprises the number C<span>.
3085 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3086 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3087 */
3088STATIC SV*
b0e3252e 3089S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3090{
3091 SV *swatch;
77f9f126 3092 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3093 STRLEN lcur, xcur, scur;
ef8f7699 3094 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3095 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3096
88d45d28
KW
3097 SV** listsvp = NULL; /* The string containing the main body of the table */
3098 SV** extssvp = NULL;
3099 SV** invert_it_svp = NULL;
3100 U8* typestr = NULL;
786861f5
KW
3101 STRLEN bits;
3102 STRLEN octets; /* if bits == 1, then octets == 0 */
3103 UV none;
3104 UV end = start + span;
972dd592 3105
36eb48b4 3106 if (invlistsvp == NULL) {
786861f5
KW
3107 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3108 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3109 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3110 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3111 listsvp = hv_fetchs(hv, "LIST", FALSE);
3112 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3113
3114 bits = SvUV(*bitssvp);
3115 none = SvUV(*nonesvp);
3116 typestr = (U8*)SvPV_nolen(*typesvp);
3117 }
36eb48b4
KW
3118 else {
3119 bits = 1;
3120 none = 0;
3121 }
786861f5 3122 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3123
b0e3252e 3124 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3125
979f2922 3126 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3127 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3128 (UV)bits);
979f2922
ST
3129 }
3130
84ea5ef6
KW
3131 /* If overflowed, use the max possible */
3132 if (end < start) {
3133 end = UV_MAX;
3134 span = end - start;
3135 }
3136
979f2922 3137 /* create and initialize $swatch */
979f2922 3138 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3139 swatch = newSV(scur);
3140 SvPOK_on(swatch);
979f2922
ST
3141 s = (U8*)SvPVX(swatch);
3142 if (octets && none) {
0bd48802 3143 const U8* const e = s + scur;
979f2922
ST
3144 while (s < e) {
3145 if (bits == 8)
3146 *s++ = (U8)(none & 0xff);
3147 else if (bits == 16) {
3148 *s++ = (U8)((none >> 8) & 0xff);
3149 *s++ = (U8)( none & 0xff);
3150 }
3151 else if (bits == 32) {
3152 *s++ = (U8)((none >> 24) & 0xff);
3153 *s++ = (U8)((none >> 16) & 0xff);
3154 *s++ = (U8)((none >> 8) & 0xff);
3155 *s++ = (U8)( none & 0xff);
3156 }
3157 }
3158 *s = '\0';
3159 }
3160 else {
3161 (void)memzero((U8*)s, scur + 1);
3162 }
3163 SvCUR_set(swatch, scur);
3164 s = (U8*)SvPVX(swatch);
3165
36eb48b4
KW
3166 if (invlistsvp) { /* If has an inversion list set up use that */
3167 _invlist_populate_swatch(*invlistsvp, start, end, s);
3168 return swatch;
3169 }
3170
3171 /* read $swash->{LIST} */
979f2922
ST
3172 l = (U8*)SvPV(*listsvp, lcur);
3173 lend = l + lcur;
3174 while (l < lend) {
8ed25d53 3175 UV min, max, val, upper;
95543e92
KW
3176 l = swash_scan_list_line(l, lend, &min, &max, &val,
3177 cBOOL(octets), typestr);
319009ee 3178 if (l > lend) {
979f2922
ST
3179 break;
3180 }
3181
972dd592 3182 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3183 if (max < start)
3184 continue;
3185
8ed25d53
KW
3186 /* <end> is generally 1 beyond where we want to set things, but at the
3187 * platform's infinity, where we can't go any higher, we want to
3188 * include the code point at <end> */
3189 upper = (max < end)
3190 ? max
3191 : (max != UV_MAX || end != UV_MAX)
3192 ? end - 1
3193 : end;
3194
979f2922 3195 if (octets) {
35da51f7 3196 UV key;
979f2922
ST
3197 if (min < start) {
3198 if (!none || val < none) {
3199 val += start - min;
3200 }
3201 min = start;
3202 }
8ed25d53 3203 for (key = min; key <= upper; key++) {
979f2922 3204 STRLEN offset;
979f2922
ST
3205 /* offset must be non-negative (start <= min <= key < end) */
3206 offset = octets * (key - start);
3207 if (bits == 8)
3208 s[offset] = (U8)(val & 0xff);
3209 else if (bits == 16) {
3210 s[offset ] = (U8)((val >> 8) & 0xff);
3211 s[offset + 1] = (U8)( val & 0xff);
3212 }
3213 else if (bits == 32) {
3214 s[offset ] = (U8)((val >> 24) & 0xff);
3215 s[offset + 1] = (U8)((val >> 16) & 0xff);
3216 s[offset + 2] = (U8)((val >> 8) & 0xff);
3217 s[offset + 3] = (U8)( val & 0xff);
3218 }
3219
3220 if (!none || val < none)
3221 ++val;
3222 }
3223 }
711a919c 3224 else { /* bits == 1, then val should be ignored */
35da51f7 3225 UV key;
979f2922
ST
3226 if (min < start)
3227 min = start;
6cb05c12 3228
8ed25d53 3229 for (key = min; key <= upper; key++) {
0bd48802 3230 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3231 s[offset >> 3] |= 1 << (offset & 7);
3232 }
3233 }
3234 } /* while */
979f2922 3235
9479a769 3236 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3237 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3238
3239 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3240 * be 0, and their inversion should also be 0, as we don't succeed any
3241 * Unicode property matches for non-Unicode code points */
3242 if (start <= PERL_UNICODE_MAX) {
3243
3244 /* The code below assumes that we never cross the
3245 * Unicode/above-Unicode boundary in a range, as otherwise we would
3246 * have to figure out where to stop flipping the bits. Since this
3247 * boundary is divisible by a large power of 2, and swatches comes
3248 * in small powers of 2, this should be a valid assumption */
3249 assert(start + span - 1 <= PERL_UNICODE_MAX);
3250
507a8485
KW
3251 send = s + scur;
3252 while (s < send) {
3253 *s = ~(*s);
3254 s++;
3255 }
0bda3001 3256 }
77f9f126
KW
3257 }
3258
d73c39c5
KW
3259 /* read $swash->{EXTRAS}
3260 * This code also copied to swash_to_invlist() below */
979f2922
ST
3261 x = (U8*)SvPV(*extssvp, xcur);
3262 xend = x + xcur;
3263 while (x < xend) {
3264 STRLEN namelen;
3265 U8 *namestr;
3266 SV** othersvp;
3267 HV* otherhv;
3268 STRLEN otherbits;
3269 SV **otherbitssvp, *other;
711a919c 3270 U8 *s, *o, *nl;
979f2922
ST
3271 STRLEN slen, olen;
3272
35da51f7 3273 const U8 opc = *x++;
979f2922
ST
3274 if (opc == '\n')
3275 continue;
3276
3277 nl = (U8*)memchr(x, '\n', xend - x);
3278
3279 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3280 if (nl) {
3281 x = nl + 1; /* 1 is length of "\n" */
3282 continue;
3283 }
3284 else {
3285 x = xend; /* to EXTRAS' end at which \n is not found */
3286 break;
3287 }
3288 }
3289
3290 namestr = x;
3291 if (nl) {
3292 namelen = nl - namestr;
3293 x = nl + 1;
3294 }
3295 else {
3296 namelen = xend - namestr;
3297 x = xend;
3298 }
3299
3300 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3301 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3302 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3303 otherbits = (STRLEN)SvUV(*otherbitssvp);
3304 if (bits < otherbits)
5637ef5b
NC
3305 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3306 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3307
3308 /* The "other" swatch must be destroyed after. */
b0e3252e 3309 other = swatch_get(*othersvp, start, span);
979f2922
ST
3310 o = (U8*)SvPV(other, olen);
3311
3312 if (!olen)
b0e3252e 3313 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3314
3315 s = (U8*)SvPV(swatch, slen);
3316 if (bits == 1 && otherbits == 1) {
3317 if (slen != olen)
5637ef5b
NC
3318 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3319 "mismatch, slen=%"UVuf", olen=%"UVuf,
3320 (UV)slen, (UV)olen);
979f2922
ST
3321
3322 switch (opc) {
3323 case '+':
3324 while (slen--)
3325 *s++ |= *o++;
3326 break;
3327 case '!':
3328 while (slen--)
3329 *s++ |= ~*o++;
3330 break;
3331 case '-':
3332 while (slen--)
3333 *s++ &= ~*o++;
3334 break;
3335 case '&':
3336 while (slen--)
3337 *s++ &= *o++;
3338 break;
3339 default:
3340 break;
3341 }
3342 }
711a919c 3343 else {
979f2922
ST
3344 STRLEN otheroctets = otherbits >> 3;
3345 STRLEN offset = 0;
35da51f7 3346 U8* const send = s + slen;
979f2922
ST
3347
3348 while (s < send) {
3349 UV otherval = 0;
3350
3351 if (otherbits == 1) {
3352 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3353 ++offset;
3354 }
3355 else {
3356 STRLEN vlen = otheroctets;
3357 otherval = *o++;
3358 while (--vlen) {
3359 otherval <<= 8;
3360 otherval |= *o++;
3361 }
3362 }
3363
711a919c 3364 if (opc == '+' && otherval)
6f207bd3 3365 NOOP; /* replace with otherval */
979f2922
ST
3366 else if (opc == '!' && !otherval)
3367 otherval = 1;
3368 else if (opc == '-' && otherval)
3369 otherval = 0;
3370 else if (opc == '&' && !otherval)
3371 otherval = 0;
3372 else {
711a919c 3373 s += octets; /* no replacement */
979f2922
ST
3374 continue;
3375 }
3376
3377 if (bits == 8)
3378 *s++ = (U8)( otherval & 0xff);
3379 else if (bits == 16) {
3380 *s++ = (U8)((otherval >> 8) & 0xff);
3381 *s++ = (U8)( otherval & 0xff);
3382 }
3383 else if (bits == 32) {
3384 *s++ = (U8)((otherval >> 24) & 0xff);
3385 *s++ = (U8)((otherval >> 16) & 0xff);
3386 *s++ = (U8)((otherval >> 8) & 0xff);
3387 *s++ = (U8)( otherval & 0xff);
3388 }
3389 }
3390 }
3391 sv_free(other); /* through with it! */
3392 } /* while */
3393 return swatch;
3394}
3395
064c021d 3396HV*
4c2e1131 3397Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3398{
3399
79a2a0e8 3400 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3401 * Can't be used on a property that is subject to user override, as it
3402 * relies on the value of SPECIALS in the swash which would be set by
3403 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3404 * for overridden properties
064c021d
KW
3405 *
3406 * Returns a hash which is the inversion and closure of a swash mapping.
3407 * For example, consider the input lines:
3408 * 004B 006B
3409 * 004C 006C
3410 * 212A 006B
3411 *
4a4088c4 3412 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3413 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3414 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3415 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3416 *
538e84ed
KW
3417 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3418 * keys are only code points that are folded-to, so it isn't a full closure.
3419 *
064c021d
KW
3420 * Essentially, for any code point, it gives all the code points that map to
3421 * it, or the list of 'froms' for that point.
3422 *
5662e334
KW
3423 * Currently it ignores any additions or deletions from other swashes,
3424 * looking at just the main body of the swash, and if there are SPECIALS
3425 * in the swash, at that hash
3426 *
3427 * The specials hash can be extra code points, and most likely consists of
3428 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3429 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3430 * However consider this possible input in the specials hash:
3431 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3432 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3433 *
3434 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3435 * currently handle. But it also means that FB05 and FB06 are equivalent in
3436 * a 1-1 mapping which we should handle, and this relationship may not be in
3437 * the main table. Therefore this function examines all the multi-char
74894415
KW
3438 * sequences and adds the 1-1 mappings that come out of that.
3439 *
3440 * XXX This function was originally intended to be multipurpose, but its
3441 * only use is quite likely to remain for constructing the inversion of
3442 * the CaseFolding (//i) property. If it were more general purpose for
3443 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3444 * because certain folds are prohibited under /iaa and /il. As an example,
3445 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3446 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3447 * prohibited, so we would not figure out that they fold to each other.
3448 * Code could be written to automatically figure this out, similar to the
3449 * code that does this for multi-character folds, but this is the only case
3450 * where something like this is ever likely to happen, as all the single
7ee537e6 3451 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3452 * little special code that is compiled only for this Unicode version. This
3453 * is smaller and didn't require much coding time to do. But this makes
3454 * this routine strongly tied to being used just for CaseFolding. If ever
3455 * it should be generalized, this would have to be fixed */
064c021d
KW
3456
3457 U8 *l, *lend;
3458 STRLEN lcur;
3459 HV *const hv = MUTABLE_HV(SvRV(swash));
3460
923b6d4e
KW
3461 /* The string containing the main body of the table. This will have its
3462 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3463 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3464
3465 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3466 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3467 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3468 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3469 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3470 const STRLEN bits = SvUV(*bitssvp);
3471 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3472 const UV none = SvUV(*nonesvp);
5662e334 3473 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3474
3475 HV* ret = newHV();
3476
3477 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3478
3479 /* Must have at least 8 bits to get the mappings */
3480 if (bits != 8 && bits != 16 && bits != 32) {
3481 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3482 (UV)bits);
3483 }
3484
5662e334
KW
3485 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3486 mapping to more than one character */
3487
3488 /* Construct an inverse mapping hash for the specials */
3489 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3490 HV * specials_inverse = newHV();
3491 char *char_from; /* the lhs of the map */
3492 I32 from_len; /* its byte length */
3493 char *char_to; /* the rhs of the map */
3494 I32 to_len; /* its byte length */
3495 SV *sv_to; /* and in a sv */
3496 AV* from_list; /* list of things that map to each 'to' */
3497
3498 hv_iterinit(specials_hv);
3499
4a4088c4
KW
3500 /* The keys are the characters (in UTF-8) that map to the corresponding
3501 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3502 * list. */
3503 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3504 SV** listp;
3505 if (! SvPOK(sv_to)) {
5637ef5b
NC
3506 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3507 "unexpectedly is not a string, flags=%lu",
3508 (unsigned long)SvFLAGS(sv_to));
5662e334 3509 }
4b88fb76 3510 /*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
3511
3512 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3513 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3514 * it. Those strings are all one character long */
3515 if ((listp = hv_fetch(specials_inverse,
3516 SvPVX(sv_to),
3517 SvCUR(sv_to), 0)))
3518 {
3519 from_list = (AV*) *listp;
3520 }
3521 else { /* No entry yet for it: create one */
3522 from_list = newAV();
3523 if (! hv_store(specials_inverse,
3524 SvPVX(sv_to),
3525 SvCUR(sv_to),
3526 (SV*) from_list, 0))
3527 {
3528 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3529 }
3530 }
3531
3532 /* Here have the list associated with this 'to' (perhaps newly
3533 * created and empty). Just add to it. Note that we ASSUME that
3534 * the input is guaranteed to not have duplications, so we don't
3535 * check for that. Duplications just slow down execution time. */
3536 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3537 }
3538
3539 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3540 * it looking for cases like the FB05/FB06 examples above. There would
3541 * be an entry in the hash like
3542 * 'st' => [ FB05, FB06 ]
3543 * In this example we will create two lists that get stored in the
3544 * returned hash, 'ret':
3545 * FB05 => [ FB05, FB06 ]
3546 * FB06 => [ FB05, FB06 ]
3547 *
3548 * Note that there is nothing to do if the array only has one element.
3549 * (In the normal 1-1 case handled below, we don't have to worry about
3550 * two lists, as everything gets tied to the single list that is
3551 * generated for the single character 'to'. But here, we are omitting
3552 * that list, ('st' in the example), so must have multiple lists.) */
3553 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3554 &char_to, &to_len)))
3555 {
7e13d567 3556 if (av_tindex_nomg(from_list) > 0) {
c70927a6 3557 SSize_t i;
5662e334
KW
3558
3559 /* We iterate over all combinations of i,j to place each code
3560 * point on each list */
7e13d567 3561 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
c70927a6 3562 SSize_t j;
5662e334
KW
3563 AV* i_list = newAV();
3564 SV** entryp = av_fetch(from_list, i, FALSE);
3565 if (entryp == NULL) {
3566 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3567 }
3568 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3569 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3570 }
3571 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3572 (SV*) i_list, FALSE))
3573 {
3574 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3575 }
3576
538e84ed 3577 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
7e13d567 3578 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
5662e334
KW
3579 entryp = av_fetch(from_list, j, FALSE);
3580 if (entryp == NULL) {
3581 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3582 }
3583
3584 /* When i==j this adds itself to the list */
4b88fb76
KW
3585 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3586 (U8*) SvPVX(*entryp),
3587 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3588 0)));
4637d003 3589 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
5662e334
KW
3590 }
3591 }
3592 }
3593 }
3594 SvREFCNT_dec(specials_inverse); /* done with it */
3595 } /* End of specials */
3596
064c021d 3597 /* read $swash->{LIST} */
74894415
KW
3598
3599#if UNICODE_MAJOR_VERSION == 3 \
3600 && UNICODE_DOT_VERSION == 0 \
3601 && UNICODE_DOT_DOT_VERSION == 1
3602
3603 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3604 * rule so that things work under /iaa and /il */
3605
3606 SV * mod_listsv = sv_mortalcopy(*listsvp);
3607 sv_catpv(mod_listsv, "130\t130\t131\n");
3608 l = (U8*)SvPV(mod_listsv, lcur);
3609
3610#else
3611
064c021d 3612 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3613
3614#endif
3615
064c021d
KW
3616 lend = l + lcur;
3617
3618 /* Go through each input line */
3619 while (l < lend) {
3620 UV min, max, val;
3621 UV inverse;
95543e92
KW
3622 l = swash_scan_list_line(l, lend, &min, &max, &val,
3623 cBOOL(octets), typestr);
064c021d
KW
3624 if (l > lend) {
3625 break;
3626 }
3627
3628 /* Each element in the range is to be inverted */
3629 for (inverse = min; inverse <= max; inverse++) {
3630 AV* list;
064c021d
KW
3631 SV** listp;
3632 IV i;
3633 bool found_key = FALSE;
5662e334 3634 bool found_inverse = FALSE;
064c021d
KW
3635
3636 /* The key is the inverse mapping */
3637 char key[UTF8_MAXBYTES+1];
c80e42f3 3638 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3639 STRLEN key_len = key_end - key;
3640
064c021d
KW