This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: White-space only
[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/*
7bbfa158 55=for apidoc is_invariant_string
eaf7a4d2 56
b8ed1ca9
KW
57Returns true iff the first C<len> bytes of the string C<s> are the same
58regardless of the UTF-8 encoding of the string (or UTF-EBCDIC encoding on
59EBCDIC machines). That is, if they are UTF-8 invariant. On ASCII-ish
60machines, all the ASCII characters and only the ASCII characters fit this
7bbfa158 61definition. On EBCDIC machines, the ASCII-range characters are invariant, but
b8ed1ca9
KW
62so also are the C1 controls and C<\c?> (which isn't in the ASCII range on
63EBCDIC).
eaf7a4d2 64
75200dff
KW
65If C<len> is 0, it will be calculated using C<strlen(s)>, (which means if you
66use this option, that C<s> can't have embedded C<NUL> characters and has to
67have a terminating C<NUL> byte).
9f7e3d64 68
a1433954 69See also L</is_utf8_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
eaf7a4d2
CS
70
71=cut
72*/
73
74bool
7bbfa158 75Perl_is_invariant_string(const U8 *s, STRLEN len)
eaf7a4d2
CS
76{
77 const U8* const send = s + (len ? len : strlen((const char *)s));
78 const U8* x = s;
79
7bbfa158 80 PERL_ARGS_ASSERT_IS_INVARIANT_STRING;
eaf7a4d2
CS
81
82 for (; x < send; ++x) {
83 if (!UTF8_IS_INVARIANT(*x))
84 break;
85 }
86
87 return x == send;
88}
89
90/*
378516de 91=for apidoc uvoffuni_to_utf8_flags
eebe1485 92
a27992cc 93THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
94Instead, B<Almost all code should use L</uvchr_to_utf8> or
95L</uvchr_to_utf8_flags>>.
a27992cc 96
de69f3af
KW
97This function is like them, but the input is a strict Unicode
98(as opposed to native) code point. Only in very rare circumstances should code
99not be using the native code point.
949cf498 100
efa9cd84 101For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 102
eebe1485
SC
103=cut
104*/
105
8ee1cdcb
KW
106#define HANDLE_UNICODE_SURROGATE(uv, flags) \
107 STMT_START { \
108 if (flags & UNICODE_WARN_SURROGATE) { \
109 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
110 "UTF-16 surrogate U+%04"UVXf, uv); \
111 } \
112 if (flags & UNICODE_DISALLOW_SURROGATE) { \
113 return NULL; \
114 } \
115 } STMT_END;
116
117#define HANDLE_UNICODE_NONCHAR(uv, flags) \
118 STMT_START { \
119 if (flags & UNICODE_WARN_NONCHAR) { \
120 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
121 "Unicode non-character U+%04"UVXf" is not " \
122 "recommended for open interchange", uv); \
123 } \
124 if (flags & UNICODE_DISALLOW_NONCHAR) { \
125 return NULL; \
126 } \
127 } STMT_END;
128
ba6ed43c
KW
129/* Use shorter names internally in this file */
130#define SHIFT UTF_ACCUMULATION_SHIFT
131#undef MARK
132#define MARK UTF_CONTINUATION_MARK
133#define MASK UTF_CONTINUATION_MASK
134
dfe13c55 135U8 *
378516de 136Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 137{
378516de 138 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 139
2d1545e5 140 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 141 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
142 return d;
143 }
facc1dc2 144
3ea68d71 145 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
146 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
147 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
148 return d;
149 }
d9432125 150
ba6ed43c
KW
151 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
152 * below, the 16 is for start bytes E0-EF (which are all the possible ones
153 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
154 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
155 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
156 * 0x800-0xFFFF on ASCII */
157 if (uv < (16 * (1U << (2 * SHIFT)))) {
158 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
159 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
160 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
161
162#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
163 aren't tested here */
164 /* The most likely code points in this range are below the surrogates.
165 * Do an extra test to quickly exclude those. */
166 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
167 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
168 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
169 {
8ee1cdcb
KW
170 HANDLE_UNICODE_NONCHAR(uv, flags);
171 }
172 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
173 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 174 }
ba6ed43c
KW
175 }
176#endif
177 return d;
178 }
179
180 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
181 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
182 * happen starting with 4-byte characters on ASCII platforms. We unify the
183 * code for these with EBCDIC, even though some of them require 5-bytes on
184 * those, because khw believes the code saving is worth the very slight
185 * performance hit on these high EBCDIC code points. */
186
187 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
188 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
189 && ckWARN_d(WARN_DEPRECATED))
190 {
191 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
192 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
193 }
194 if ( (flags & UNICODE_WARN_SUPER)
195 || ( UNICODE_IS_ABOVE_31_BIT(uv)
196 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
197 {
198 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
199
200 /* Choose the more dire applicable warning */
201 (UNICODE_IS_ABOVE_31_BIT(uv))
202 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
203 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
204 uv);
205 }
206 if (flags & UNICODE_DISALLOW_SUPER
207 || ( UNICODE_IS_ABOVE_31_BIT(uv)
208 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
209 {
210 return NULL;
211 }
212 }
ba6ed43c
KW
213 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
214 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 215 }
d9432125 216
ba6ed43c
KW
217 /* Test for and handle 4-byte result. In the test immediately below, the
218 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
219 * characters). The 3 is for 3 continuation bytes; these each contribute
220 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
221 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
222 * 0x1_0000-0x1F_FFFF on ASCII */
223 if (uv < (8 * (1U << (3 * SHIFT)))) {
224 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
225 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
226 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
227 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
228
229#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
230 characters. The end-plane non-characters for EBCDIC were
231 handled just above */
232 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
233 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 234 }
ba6ed43c
KW
235 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
236 HANDLE_UNICODE_SURROGATE(uv, flags);
237 }
238#endif
239
240 return d;
241 }
242
243 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
244 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
245 * format. The unrolled version above turns out to not save all that much
246 * time, and at these high code points (well above the legal Unicode range
247 * on ASCII platforms, and well above anything in common use in EBCDIC),
248 * khw believes that less code outweighs slight performance gains. */
249
d9432125 250 {
5aaebcb3 251 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
252 U8 *p = d+len-1;
253 while (p > d) {
4c8cd605 254 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
255 uv >>= UTF_ACCUMULATION_SHIFT;
256 }
4c8cd605 257 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
258 return d+len;
259 }
a0ed51b3 260}
a5bf80e0 261
646ca15d 262/*
07693fe6
KW
263=for apidoc uvchr_to_utf8
264
bcb1a2d4 265Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 266of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
267C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
268the byte after the end of the new character. In other words,
07693fe6
KW
269
270 d = uvchr_to_utf8(d, uv);
271
272is the recommended wide native character-aware way of saying
273
274 *(d++) = uv;
275
760c7c2f
KW
276This function accepts any UV as input, but very high code points (above
277C<IV_MAX> on the platform) will raise a deprecation warning. This is
278typically 0x7FFF_FFFF in a 32-bit word.
279
280It is possible to forbid or warn on non-Unicode code points, or those that may
281be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 282
07693fe6
KW
283=cut
284*/
285
de69f3af
KW
286/* This is also a macro */
287PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
288
07693fe6
KW
289U8 *
290Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
291{
de69f3af 292 return uvchr_to_utf8(d, uv);
07693fe6
KW
293}
294
de69f3af
KW
295/*
296=for apidoc uvchr_to_utf8_flags
297
298Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 299of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
300C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
301the byte after the end of the new character. In other words,
de69f3af
KW
302
303 d = uvchr_to_utf8_flags(d, uv, flags);
304
305or, in most cases,
306
307 d = uvchr_to_utf8_flags(d, uv, 0);
308
309This is the Unicode-aware way of saying
310
311 *(d++) = uv;
312
760c7c2f
KW
313If C<flags> is 0, this function accepts any UV as input, but very high code
314points (above C<IV_MAX> for the platform) will raise a deprecation warning.
315This is typically 0x7FFF_FFFF in a 32-bit word.
316
317Specifying C<flags> can further restrict what is allowed and not warned on, as
318follows:
de69f3af 319
796b6530 320If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
321the function will raise a warning, provided UTF8 warnings are enabled. If
322instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
323NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 324
760c7c2f
KW
325Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
326affect how the function handles a Unicode non-character.
93e6dbd6 327
760c7c2f
KW
328And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
329affect the handling of code points that are above the Unicode maximum of
3300x10FFFF. Languages other than Perl may not be able to accept files that
331contain these.
93e6dbd6
KW
332
333The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
334the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
335three DISALLOW flags.
336
ab8e6d41
KW
337Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
338so using them is more problematic than other above-Unicode code points. Perl
339invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
340likely that non-Perl languages will not be able to read files that contain
341these that written by the perl interpreter; nor would Perl understand files
342written by something that uses a different extension. For these reasons, there
343is a separate set of flags that can warn and/or disallow these extremely high
344code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
345C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
346are entirely independent from the deprecation warning for code points above
347C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
348code point that needs more than 31 bits to represent. When that happens,
349effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
35032-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
351above-Unicode code points, including these, as malformations; and
352C<UNICODE_WARN_SUPER> warns on these.)
353
354On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
355extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
356than on ASCII. Prior to that, code points 2**31 and higher were simply
357unrepresentable, and a different, incompatible method was used to represent
358code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
359and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
360platforms, warning and disallowing 2**31 and higher.
de69f3af 361
de69f3af
KW
362=cut
363*/
364
365/* This is also a macro */
366PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
367
07693fe6
KW
368U8 *
369Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
370{
de69f3af 371 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
372}
373
374/*
87cea99e 375=for apidoc is_utf8_string
6662521e 376
a1433954 377Returns true if the first C<len> bytes of string C<s> form a valid
9f7e3d64 378UTF-8 string, false otherwise. If C<len> is 0, it will be calculated
75200dff
KW
379using C<strlen(s)> (which means if you use this option, that C<s> can't have
380embedded C<NUL> characters and has to have a terminating C<NUL> byte). Note
381that all characters being ASCII constitute 'a valid UTF-8 string'.
6662521e 382
7bbfa158 383See also L</is_invariant_string>(), L</is_utf8_string_loclen>(), and L</is_utf8_string_loc>().
768c67ee 384
6662521e
GS
385=cut
386*/
387
8e84507e 388bool
668b6d8d 389Perl_is_utf8_string(const U8 *s, STRLEN len)
6662521e 390{
35da51f7 391 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 392 const U8* x = s;
067a85ef 393
7918f24d 394 PERL_ARGS_ASSERT_IS_UTF8_STRING;
1aa99e6b 395
6662521e 396 while (x < send) {
6302f837
KW
397 STRLEN len = isUTF8_CHAR(x, send);
398 if (UNLIKELY(! len)) {
399 return FALSE;
400 }
401 x += len;
6662521e 402 }
768c67ee 403
067a85ef 404 return TRUE;
6662521e
GS
405}
406
67e989fb 407/*
814fafa7
NC
408Implemented as a macro in utf8.h
409
87cea99e 410=for apidoc is_utf8_string_loc
814fafa7 411
a1433954
KW
412Like L</is_utf8_string> but stores the location of the failure (in the
413case of "utf8ness failure") or the location C<s>+C<len> (in the case of
814fafa7
NC
414"utf8ness success") in the C<ep>.
415
a1433954 416See also L</is_utf8_string_loclen>() and L</is_utf8_string>().
814fafa7 417
87cea99e 418=for apidoc is_utf8_string_loclen
81cd54e3 419
a1433954
KW
420Like L</is_utf8_string>() but stores the location of the failure (in the
421case of "utf8ness failure") or the location C<s>+C<len> (in the case of
768c67ee
JH
422"utf8ness success") in the C<ep>, and the number of UTF-8
423encoded characters in the C<el>.
424
a1433954 425See also L</is_utf8_string_loc>() and L</is_utf8_string>().
81cd54e3
JH
426
427=cut
428*/
429
430bool
668b6d8d 431Perl_is_utf8_string_loclen(const U8 *s, STRLEN len, const U8 **ep, STRLEN *el)
81cd54e3 432{
35da51f7 433 const U8* const send = s + (len ? len : strlen((const char *)s));
7fc63493 434 const U8* x = s;
3ebfea28 435 STRLEN outlen = 0;
7918f24d
NC
436
437 PERL_ARGS_ASSERT_IS_UTF8_STRING_LOCLEN;
81cd54e3 438
81cd54e3 439 while (x < send) {
6302f837
KW
440 STRLEN len = isUTF8_CHAR(x, send);
441 if (UNLIKELY(! len)) {
442 goto out;
443 }
444 x += len;
445 outlen++;
81cd54e3 446 }
768c67ee
JH
447
448 out:
3ebfea28
AL
449 if (el)
450 *el = outlen;
451
768c67ee
JH
452 if (ep)
453 *ep = x;
3ebfea28 454 return (x == send);
81cd54e3
JH
455}
456
457/*
768c67ee 458
de69f3af 459=for apidoc utf8n_to_uvchr
378516de
KW
460
461THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 462Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 463
9041c2e3 464Bottom level UTF-8 decode routine.
de69f3af 465Returns the native code point value of the first character in the string C<s>,
746afd53
KW
466which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
467C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
468the length, in bytes, of that character.
949cf498
KW
469
470The value of C<flags> determines the behavior when C<s> does not point to a
471well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
472zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
473next possible position in C<s> that could begin a non-malformed character.
474Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
475
476Various ALLOW flags can be set in C<flags> to allow (and not warn on)
477individual types of malformations, such as the sequence being overlong (that
478is, when there is a shorter sequence that can express the same code point;
479overlong sequences are expressly forbidden in the UTF-8 standard due to
480potential security issues). Another malformation example is the first byte of
481a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
482flags. For allowed 0 length strings, this function returns 0; for allowed
483overlong sequences, the computed code point is returned; for all other allowed
484malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
485determinable reasonable value.
949cf498 486
796b6530 487The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
488flags) malformation is found. If this flag is set, the routine assumes that
489the caller will raise a warning, and this function will silently just set
d088425d
KW
490C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
491
75200dff 492Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 493character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
d088425d 494in both cases, 0 is returned. To disambiguate, upon a zero return, see if the
75200dff
KW
495first byte of C<s> is 0 as well. If so, the input was a C<NUL>; if not, the
496input had an error.
949cf498
KW
497
498Certain code points are considered problematic. These are Unicode surrogates,
746afd53 499Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 500By default these are considered regular code points, but certain situations
5eafe189 501warrant special handling for them. If C<flags> contains
796b6530
KW
502C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
503malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
504C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
505Unicode maximum) can be set to disallow these categories individually.
506
507The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
508C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
509raised for their respective categories, but otherwise the code points are
510considered valid (not malformations). To get a category to both be treated as
511a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 512(But note that warnings are not raised if lexically disabled nor if
796b6530 513C<UTF8_CHECK_ONLY> is also specified.)
949cf498 514
760c7c2f
KW
515It is now deprecated to have very high code points (above C<IV_MAX> on the
516platforms) and this function will raise a deprecation warning for these (unless
517such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
518in a 32-bit word.
ab8e6d41
KW
519
520Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
521so using them is more problematic than other above-Unicode code points. Perl
522invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
523likely that non-Perl languages will not be able to read files that contain
524these that written by the perl interpreter; nor would Perl understand files
525written by something that uses a different extension. For these reasons, there
526is a separate set of flags that can warn and/or disallow these extremely high
527code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
528C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
529are entirely independent from the deprecation warning for code points above
530C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
531code point that needs more than 31 bits to represent. When that happens,
532effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
53332-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
534above-Unicode code points, including these, as malformations; and
535C<UTF8_WARN_SUPER> warns on these.)
536
537On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
538extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
539than on ASCII. Prior to that, code points 2**31 and higher were simply
540unrepresentable, and a different, incompatible method was used to represent
541code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
542and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
543platforms, warning and disallowing 2**31 and higher.
949cf498
KW
544
545All other code points corresponding to Unicode characters, including private
546use and those yet to be assigned, are never considered malformed and never
547warn.
67e989fb 548
37607a96
PK
549=cut
550*/
67e989fb 551
a0ed51b3 552UV
de69f3af 553Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 554{
d4c19fe8 555 const U8 * const s0 = s;
eb83ed87 556 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 557 U8 * send;
eb83ed87
KW
558 UV uv = *s;
559 STRLEN expectlen;
949cf498 560 SV* sv = NULL;
eb83ed87
KW
561 UV outlier_ret = 0; /* return value when input is in error or problematic
562 */
563 UV pack_warn = 0; /* Save result of packWARN() for later */
564 bool unexpected_non_continuation = FALSE;
565 bool overflowed = FALSE;
2f8f112e 566 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 567
eb83ed87 568 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 569
de69f3af 570 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 571
eb83ed87
KW
572 /* The order of malformation tests here is important. We should consume as
573 * few bytes as possible in order to not skip any valid character. This is
574 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
575 * http://unicode.org/reports/tr36 for more discussion as to why. For
576 * example, once we've done a UTF8SKIP, we can tell the expected number of
577 * bytes, and could fail right off the bat if the input parameters indicate
578 * that there are too few available. But it could be that just that first
579 * byte is garbled, and the intended character occupies fewer bytes. If we
580 * blindly assumed that the first byte is correct, and skipped based on
581 * that number, we could skip over a valid input character. So instead, we
582 * always examine the sequence byte-by-byte.
583 *
584 * We also should not consume too few bytes, otherwise someone could inject
585 * things. For example, an input could be deliberately designed to
586 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 587 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
588 * which is actually part of of the overflowing sequence), that could look
589 * legitimate to the caller, which could discard the initial partial
590 * sequence and process the rest, inappropriately */
591
592 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 593 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
594 if (retlen) {
595 *retlen = 0;
596 }
a0dbb045 597
eb83ed87
KW
598 if (flags & UTF8_ALLOW_EMPTY) {
599 return 0;
600 }
601 if (! (flags & UTF8_CHECK_ONLY)) {
602 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
603 }
0c443dc2
JH
604 goto malformed;
605 }
606
eb83ed87
KW
607 expectlen = UTF8SKIP(s);
608
609 /* A well-formed UTF-8 character, as the vast majority of calls to this
610 * function will be for, has this expected length. For efficiency, set
611 * things up here to return it. It will be overriden only in those rare
612 * cases where a malformation is found */
613 if (retlen) {
614 *retlen = expectlen;
615 }
616
617 /* An invariant is trivially well-formed */
1d72bdf6 618 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 619 return uv;
a0ed51b3 620 }
67e989fb 621
eb83ed87 622 /* A continuation character can't start a valid sequence */
b5b9af04 623 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
624 if (flags & UTF8_ALLOW_CONTINUATION) {
625 if (retlen) {
626 *retlen = 1;
627 }
628 return UNICODE_REPLACEMENT;
629 }
ba210ebe 630
eb83ed87
KW
631 if (! (flags & UTF8_CHECK_ONLY)) {
632 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
633 }
634 curlen = 1;
ba210ebe
JH
635 goto malformed;
636 }
9041c2e3 637
dcd27b3c
KW
638 /* Here is not a continuation byte, nor an invariant. The only thing left
639 * is a start byte (possibly for an overlong) */
640
1d72bdf6 641#ifdef EBCDIC
bc3632a8 642 uv = NATIVE_UTF8_TO_I8(uv);
1d72bdf6
NIS
643#endif
644
eb83ed87
KW
645 /* Remove the leading bits that indicate the number of bytes in the
646 * character's whole UTF-8 sequence, leaving just the bits that are part of
647 * the value */
648 uv &= UTF_START_MASK(expectlen);
ba210ebe 649
eb83ed87
KW
650 /* Now, loop through the remaining bytes in the character's sequence,
651 * accumulating each into the working value as we go. Be sure to not look
652 * past the end of the input string */
0b8d30e8
KW
653 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
654
eb83ed87 655 for (s = s0 + 1; s < send; s++) {
b5b9af04 656 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
657 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
658
659 /* The original implementors viewed this malformation as more
660 * serious than the others (though I, khw, don't understand
661 * why, since other malformations also give very very wrong
662 * results), so there is no way to turn off checking for it.
663 * Set a flag, but keep going in the loop, so that we absorb
664 * the rest of the bytes that comprise the character. */
665 overflowed = TRUE;
666 overflow_byte = *s; /* Save for warning message's use */
667 }
8850bf83 668 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
669 }
670 else {
671 /* Here, found a non-continuation before processing all expected
672 * bytes. This byte begins a new character, so quit, even if
673 * allowing this malformation. */
674 unexpected_non_continuation = TRUE;
675 break;
676 }
677 } /* End of loop through the character's bytes */
678
679 /* Save how many bytes were actually in the character */
680 curlen = s - s0;
681
682 /* The loop above finds two types of malformations: non-continuation and/or
683 * overflow. The non-continuation malformation is really a too-short
684 * malformation, as it means that the current character ended before it was
685 * expected to (being terminated prematurely by the beginning of the next
686 * character, whereas in the too-short malformation there just are too few
687 * bytes available to hold the character. In both cases, the check below
688 * that we have found the expected number of bytes would fail if executed.)
689 * Thus the non-continuation malformation is really unnecessary, being a
690 * subset of the too-short malformation. But there may be existing
691 * applications that are expecting the non-continuation type, so we retain
692 * it, and return it in preference to the too-short malformation. (If this
693 * code were being written from scratch, the two types might be collapsed
694 * into one.) I, khw, am also giving priority to returning the
695 * non-continuation and too-short malformations over overflow when multiple
696 * ones are present. I don't know of any real reason to prefer one over
697 * the other, except that it seems to me that multiple-byte errors trumps
698 * errors from a single byte */
b5b9af04 699 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
700 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
701 if (! (flags & UTF8_CHECK_ONLY)) {
702 if (curlen == 1) {
703 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
704 }
705 else {
706 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
707 }
708 }
eb83ed87
KW
709 goto malformed;
710 }
711 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
712
713 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
714 * as what the original expectations were. */
715 do_overlong_test = FALSE;
eb83ed87
KW
716 if (retlen) {
717 *retlen = curlen;
718 }
719 }
b5b9af04 720 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
721 if (! (flags & UTF8_ALLOW_SHORT)) {
722 if (! (flags & UTF8_CHECK_ONLY)) {
723 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 724 }
eb83ed87
KW
725 goto malformed;
726 }
727 uv = UNICODE_REPLACEMENT;
2f8f112e 728 do_overlong_test = FALSE;
eb83ed87
KW
729 if (retlen) {
730 *retlen = curlen;
731 }
732 }
733
b5b9af04 734 if (UNLIKELY(overflowed)) {
eb83ed87 735 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 736 goto malformed;
eb83ed87 737 }
eb83ed87 738
2f8f112e 739 if (do_overlong_test
5aaebcb3 740 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
741 && ! (flags & UTF8_ALLOW_LONG))
742 {
eb83ed87
KW
743 /* The overlong malformation has lower precedence than the others.
744 * Note that if this malformation is allowed, we return the actual
745 * value, instead of the replacement character. This is because this
746 * value is actually well-defined. */
747 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 748 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
749 }
750 goto malformed;
751 }
752
1a89bb6c 753 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
754 * problematic code point that is not allowed by the input parameters. */
755 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
756 && ((flags & ( UTF8_DISALLOW_NONCHAR
757 |UTF8_DISALLOW_SURROGATE
758 |UTF8_DISALLOW_SUPER
759 |UTF8_DISALLOW_ABOVE_31_BIT
760 |UTF8_WARN_NONCHAR
761 |UTF8_WARN_SURROGATE
762 |UTF8_WARN_SUPER
763 |UTF8_WARN_ABOVE_31_BIT))
764 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
765 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 766 {
949cf498 767 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
768
769 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
770 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 771 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 772 && ckWARN_d(WARN_SURROGATE))
eb83ed87 773 {
111d382d 774 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 775 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
776 }
777 if (flags & UTF8_DISALLOW_SURROGATE) {
778 goto disallowed;
779 }
780 }
949cf498 781 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 782 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 783 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 784 {
0bcdd8f6
KW
785 sv = sv_2mortal(Perl_newSVpvf(aTHX_
786 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
787 uv));
54f4afef 788 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 789 }
c0236afe
KW
790
791 /* The maximum code point ever specified by a standard was
792 * 2**31 - 1. Anything larger than that is a Perl extension that
793 * very well may not be understood by other applications (including
794 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
795 * these code points are indicated by the first UTF-8 byte being
d35f2ca5
KW
796 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
797 * and before possibly bailing out, so that the slightly more dire
c0236afe
KW
798 * warning will override the regular one. */
799 if (
800#ifndef EBCDIC
801 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
802#else
803 /* The I8 for 2**31 (U+80000000) is
804 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
805 * and it turns out that on all EBCDIC pages recognized that
806 * the UTF-EBCDIC for that code point is
807 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
808 * For the next lower code point, the 1047 UTF-EBCDIC is
809 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
810 * The other code pages differ only in the bytes following
811 * \x42. Thus the following works (the minimum continuation
812 * byte is \x41). */
813 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
814 || s0[2] > 0x41
815 || s0[3] > 0x41
816 || s0[4] > 0x41
817 || s0[5] > 0x41
818 || s0[6] > 0x41
819 || s0[7] > 0x42)
820#endif
d35f2ca5
KW
821 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
822 |UTF8_DISALLOW_ABOVE_31_BIT)))
ea5ced44 823 {
0bcdd8f6 824 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 825 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 826 && ckWARN_d(WARN_UTF8))
ea5ced44 827 {
0bcdd8f6
KW
828 sv = sv_2mortal(Perl_newSVpvf(aTHX_
829 "Code point 0x%"UVXf" is not Unicode, and not portable",
830 uv));
ea5ced44
KW
831 pack_warn = packWARN(WARN_UTF8);
832 }
d35f2ca5 833 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
834 goto disallowed;
835 }
836 }
c0236afe 837
949cf498
KW
838 if (flags & UTF8_DISALLOW_SUPER) {
839 goto disallowed;
840 }
760c7c2f
KW
841
842 /* The deprecated warning overrides any non-deprecated one */
843 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
844 {
845 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
846 uv, MAX_NON_DEPRECATED_CP));
847 pack_warn = packWARN(WARN_DEPRECATED);
848 }
949cf498 849 }
4190d317
KW
850 else if (UNICODE_IS_NONCHAR(uv)) {
851 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 852 && ckWARN_d(WARN_NONCHAR))
4190d317 853 {
ba707cdc 854 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 855 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
856 }
857 if (flags & UTF8_DISALLOW_NONCHAR) {
858 goto disallowed;
859 }
860 }
949cf498 861
eb83ed87 862 if (sv) {
de69f3af
KW
863 outlier_ret = uv; /* Note we don't bother to convert to native,
864 as all the outlier code points are the same
865 in both ASCII and EBCDIC */
eb83ed87
KW
866 goto do_warn;
867 }
868
949cf498
KW
869 /* Here, this is not considered a malformed character, so drop through
870 * to return it */
a0ed51b3 871 }
ba210ebe 872
de69f3af 873 return UNI_TO_NATIVE(uv);
ba210ebe 874
eb83ed87
KW
875 /* There are three cases which get to beyond this point. In all 3 cases:
876 * <sv> if not null points to a string to print as a warning.
877 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
878 * set.
879 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
880 * This is done by initializing it to 0, and changing it only
881 * for case 1).
882 * The 3 cases are:
883 * 1) The input is valid but problematic, and to be warned about. The
884 * return value is the resultant code point; <*retlen> is set to
885 * <curlen>, the number of bytes that comprise the code point.
886 * <pack_warn> contains the result of packWARN() for the warning
887 * types. The entry point for this case is the label <do_warn>;
888 * 2) The input is a valid code point but disallowed by the parameters to
889 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
890 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
891 * comprise the code point. <pack_warn> contains the result of
892 * packWARN() for the warning types. The entry point for this case is
893 * the label <disallowed>.
894 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
895 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
896 * bytes that comprise the malformation. All such malformations are
897 * assumed to be warning type <utf8>. The entry point for this case
898 * is the label <malformed>.
899 */
949cf498 900
7b52d656 901 malformed:
ba210ebe 902
eb83ed87
KW
903 if (sv && ckWARN_d(WARN_UTF8)) {
904 pack_warn = packWARN(WARN_UTF8);
905 }
906
7b52d656 907 disallowed:
eb83ed87 908
fcc8fcf6 909 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 910 if (retlen)
10edeb5d 911 *retlen = ((STRLEN) -1);
ba210ebe
JH
912 return 0;
913 }
914
7b52d656 915 do_warn:
5b311467 916
eb83ed87
KW
917 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
918 if warnings are to be raised. */
f555bc63 919 const char * const string = SvPVX_const(sv);
a0dbb045 920
f555bc63
KW
921 if (PL_op)
922 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
923 else
924 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
925 }
926
eb83ed87
KW
927 if (retlen) {
928 *retlen = curlen;
929 }
ba210ebe 930
eb83ed87 931 return outlier_ret;
a0ed51b3
LW
932}
933
8e84507e 934/*
ec5f19d0
KW
935=for apidoc utf8_to_uvchr_buf
936
937Returns the native code point of the first character in the string C<s> which
938is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 939C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 940
524080c4
KW
941If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
942enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 943C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 944(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 945C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 946the next possible position in C<s> that could begin a non-malformed character.
de69f3af 947See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 948returned.
ec5f19d0 949
760c7c2f
KW
950Code points above the platform's C<IV_MAX> will raise a deprecation warning,
951unless those are turned off.
952
ec5f19d0
KW
953=cut
954*/
955
956
957UV
958Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
959{
ec5f19d0
KW
960 assert(s < send);
961
962 return utf8n_to_uvchr(s, send - s, retlen,
963 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
964}
965
27d6c58a 966/* Like L</utf8_to_uvchr_buf>(), but should only be called when it is known that
3986bb7c 967 * there are no malformations in the input UTF-8 string C<s>. surrogates,
57b0056d 968 * non-character code points, and non-Unicode code points are allowed. */
27d6c58a
KW
969
970UV
971Perl_valid_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
972{
010ab96b
KW
973 UV expectlen = UTF8SKIP(s);
974 const U8* send = s + expectlen;
9ff2f0f7 975 UV uv = *s;
3986bb7c 976
27d6c58a 977 PERL_ARGS_ASSERT_VALID_UTF8_TO_UVCHR;
81611534 978 PERL_UNUSED_CONTEXT;
27d6c58a 979
010ab96b
KW
980 if (retlen) {
981 *retlen = expectlen;
982 }
983
984 /* An invariant is trivially returned */
985 if (expectlen == 1) {
9ff2f0f7 986 return uv;
010ab96b
KW
987 }
988
9ff2f0f7
KW
989#ifdef EBCDIC
990 uv = NATIVE_UTF8_TO_I8(uv);
991#endif
992
010ab96b
KW
993 /* Remove the leading bits that indicate the number of bytes, leaving just
994 * the bits that are part of the value */
995 uv &= UTF_START_MASK(expectlen);
996
997 /* Now, loop through the remaining bytes, accumulating each into the
998 * working total as we go. (I khw tried unrolling the loop for up to 4
999 * bytes, but there was no performance improvement) */
1000 for (++s; s < send; s++) {
1001 uv = UTF8_ACCUMULATE(uv, *s);
1002 }
1003
3986bb7c 1004 return UNI_TO_NATIVE(uv);
010ab96b 1005
27d6c58a
KW
1006}
1007
ec5f19d0 1008/*
ec5f19d0
KW
1009=for apidoc utf8_to_uvuni_buf
1010
de69f3af
KW
1011Only in very rare circumstances should code need to be dealing in Unicode
1012(as opposed to native) code points. In those few cases, use
1013C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1014
1015Returns the Unicode (not-native) code point of the first character in the
1016string C<s> which
ec5f19d0
KW
1017is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1018C<retlen> will be set to the length, in bytes, of that character.
1019
524080c4
KW
1020If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1021enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1022NULL) to -1. If those warnings are off, the computed value if well-defined (or
1023the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1024is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1025next possible position in C<s> that could begin a non-malformed character.
de69f3af 1026See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1027
760c7c2f
KW
1028Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1029unless those are turned off.
1030
ec5f19d0
KW
1031=cut
1032*/
1033
1034UV
1035Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1036{
1037 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1038
1039 assert(send > s);
1040
1041 /* Call the low level routine asking for checks */
de69f3af
KW
1042 return NATIVE_TO_UNI(Perl_utf8n_to_uvchr(aTHX_ s, send -s, retlen,
1043 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY));
ec5f19d0
KW
1044}
1045
b76347f2 1046/*
87cea99e 1047=for apidoc utf8_length
b76347f2
JH
1048
1049Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1050Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1051up past C<e>, croaks.
b76347f2
JH
1052
1053=cut
1054*/
1055
1056STRLEN
35a4481c 1057Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1058{
1059 STRLEN len = 0;
1060
7918f24d
NC
1061 PERL_ARGS_ASSERT_UTF8_LENGTH;
1062
8850bf83
JH
1063 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1064 * the bitops (especially ~) can create illegal UTF-8.
1065 * In other words: in Perl UTF-8 is not just for Unicode. */
1066
a3b680e6
AL
1067 if (e < s)
1068 goto warn_and_return;
b76347f2 1069 while (s < e) {
4cbf4130 1070 s += UTF8SKIP(s);
8e91ec7f
AV
1071 len++;
1072 }
1073
1074 if (e != s) {
1075 len--;
1076 warn_and_return:
9b387841
NC
1077 if (PL_op)
1078 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1079 "%s in %s", unees, OP_DESC(PL_op));
1080 else
61a12c31 1081 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1082 }
1083
1084 return len;
1085}
1086
b06226ff 1087/*
87cea99e 1088=for apidoc utf8_distance
b06226ff 1089
1e54db1a 1090Returns the number of UTF-8 characters between the UTF-8 pointers C<a>
b06226ff
JH
1091and C<b>.
1092
1093WARNING: use only if you *know* that the pointers point inside the
1094same UTF-8 buffer.
1095
37607a96
PK
1096=cut
1097*/
a0ed51b3 1098
02eb7b47 1099IV
35a4481c 1100Perl_utf8_distance(pTHX_ const U8 *a, const U8 *b)
a0ed51b3 1101{
7918f24d
NC
1102 PERL_ARGS_ASSERT_UTF8_DISTANCE;
1103
bf1665bc 1104 return (a < b) ? -1 * (IV) utf8_length(a, b) : (IV) utf8_length(b, a);
a0ed51b3
LW
1105}
1106
b06226ff 1107/*
87cea99e 1108=for apidoc utf8_hop
b06226ff 1109
8850bf83
JH
1110Return the UTF-8 pointer C<s> displaced by C<off> characters, either
1111forward or backward.
b06226ff
JH
1112
1113WARNING: do not use the following unless you *know* C<off> is within
8850bf83
JH
1114the UTF-8 data pointed to by C<s> *and* that on entry C<s> is aligned
1115on the first byte of character or just after the last byte of a character.
b06226ff 1116
37607a96
PK
1117=cut
1118*/
a0ed51b3
LW
1119
1120U8 *
4caf7d8c 1121Perl_utf8_hop(const U8 *s, SSize_t off)
a0ed51b3 1122{
7918f24d
NC
1123 PERL_ARGS_ASSERT_UTF8_HOP;
1124
8850bf83
JH
1125 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g
1126 * the bitops (especially ~) can create illegal UTF-8.
1127 * In other words: in Perl UTF-8 is not just for Unicode. */
1128
a0ed51b3
LW
1129 if (off >= 0) {
1130 while (off--)
1131 s += UTF8SKIP(s);
1132 }
1133 else {
1134 while (off++) {
1135 s--;
8850bf83
JH
1136 while (UTF8_IS_CONTINUATION(*s))
1137 s--;
a0ed51b3
LW
1138 }
1139 }
4373e329 1140 return (U8 *)s;
a0ed51b3
LW
1141}
1142
6940069f 1143/*
fed3ba5d
NC
1144=for apidoc bytes_cmp_utf8
1145
a1433954 1146Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1147sequence of characters (stored as UTF-8)
1148in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1149equal, -1 or -2 if the first string is less than the second string, +1 or +2
1150if the first string is greater than the second string.
1151
1152-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1153longer string. -2 or +2 is returned if
1154there was a difference between characters
fed3ba5d
NC
1155within the strings.
1156
1157=cut
1158*/
1159
1160int
1161Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1162{
1163 const U8 *const bend = b + blen;
1164 const U8 *const uend = u + ulen;
1165
1166 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1167
1168 while (b < bend && u < uend) {
1169 U8 c = *u++;
1170 if (!UTF8_IS_INVARIANT(c)) {
1171 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1172 if (u < uend) {
1173 U8 c1 = *u++;
1174 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1175 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1176 } else {
1177 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1178 "Malformed UTF-8 character "
1179 "(unexpected non-continuation byte 0x%02x"
1180 ", immediately after start byte 0x%02x)"
1181 /* Dear diag.t, it's in the pod. */
1182 "%s%s", c1, c,
1183 PL_op ? " in " : "",
1184 PL_op ? OP_DESC(PL_op) : "");
1185 return -2;
1186 }
1187 } else {
1188 if (PL_op)
1189 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1190 "%s in %s", unees, OP_DESC(PL_op));
1191 else
61a12c31 1192 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1193 return -2; /* Really want to return undef :-) */
1194 }
1195 } else {
1196 return -2;
1197 }
1198 }
1199 if (*b != c) {
1200 return *b < c ? -2 : +2;
1201 }
1202 ++b;
1203 }
1204
1205 if (b == bend && u == uend)
1206 return 0;
1207
1208 return b < bend ? +1 : -1;
1209}
1210
1211/*
87cea99e 1212=for apidoc utf8_to_bytes
6940069f 1213
2bbc8d55 1214Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1215Unlike L</bytes_to_utf8>, this over-writes the original string, and
1216updates C<len> to contain the new length.
67e989fb 1217Returns zero on failure, setting C<len> to -1.
6940069f 1218
a1433954 1219If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1220
6940069f
GS
1221=cut
1222*/
1223
1224U8 *
37607a96 1225Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1226{
d4c19fe8
AL
1227 U8 * const save = s;
1228 U8 * const send = s + *len;
6940069f 1229 U8 *d;
246fae53 1230
7918f24d 1231 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1232 PERL_UNUSED_CONTEXT;
7918f24d 1233
1e54db1a 1234 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1235 while (s < send) {
d59937ca
KW
1236 if (! UTF8_IS_INVARIANT(*s)) {
1237 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1238 *len = ((STRLEN) -1);
1239 return 0;
1240 }
1241 s++;
dcad2880 1242 }
d59937ca 1243 s++;
246fae53 1244 }
dcad2880
JH
1245
1246 d = s = save;
6940069f 1247 while (s < send) {
80e0b38f
KW
1248 U8 c = *s++;
1249 if (! UTF8_IS_INVARIANT(c)) {
1250 /* Then it is two-byte encoded */
a62b247b 1251 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1252 s++;
1253 }
1254 *d++ = c;
6940069f
GS
1255 }
1256 *d = '\0';
246fae53 1257 *len = d - save;
6940069f
GS
1258 return save;
1259}
1260
1261/*
87cea99e 1262=for apidoc bytes_from_utf8
f9a63242 1263
2bbc8d55 1264Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1265Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1266the newly-created string, and updates C<len> to contain the new
1267length. Returns the original string if no conversion occurs, C<len>
72d33970 1268is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 12690 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1270in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1271
37607a96
PK
1272=cut
1273*/
f9a63242
JH
1274
1275U8 *
e1ec3a88 1276Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1277{
f9a63242 1278 U8 *d;
e1ec3a88
AL
1279 const U8 *start = s;
1280 const U8 *send;
f9a63242
JH
1281 I32 count = 0;
1282
7918f24d 1283 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1284 PERL_UNUSED_CONTEXT;
f9a63242 1285 if (!*is_utf8)
73d840c0 1286 return (U8 *)start;
f9a63242 1287
1e54db1a 1288 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1289 for (send = s + *len; s < send;) {
d59937ca
KW
1290 if (! UTF8_IS_INVARIANT(*s)) {
1291 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1292 return (U8 *)start;
d59937ca
KW
1293 }
1294 count++;
1295 s++;
db42d148 1296 }
d59937ca 1297 s++;
f9a63242
JH
1298 }
1299
35da51f7 1300 *is_utf8 = FALSE;
f9a63242 1301
212542aa 1302 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1303 s = start; start = d;
f9a63242
JH
1304 while (s < send) {
1305 U8 c = *s++;
1a91c45d 1306 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1307 /* Then it is two-byte encoded */
a62b247b 1308 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1309 s++;
c4d5f83a
NIS
1310 }
1311 *d++ = c;
f9a63242
JH
1312 }
1313 *d = '\0';
1314 *len = d - start;
73d840c0 1315 return (U8 *)start;
f9a63242
JH
1316}
1317
1318/*
87cea99e 1319=for apidoc bytes_to_utf8
6940069f 1320
ff97e5cf
KW
1321Converts a string C<s> of length C<len> bytes from the native encoding into
1322UTF-8.
6662521e 1323Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1324reflect the new length in bytes.
6940069f 1325
75200dff 1326A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1327
1328If you want to convert to UTF-8 from encodings other than
1329the native (Latin1 or EBCDIC),
a1433954 1330see L</sv_recode_to_utf8>().
c9ada85f 1331
497711e7 1332=cut
6940069f
GS
1333*/
1334
c682ebef
FC
1335/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1336 likewise need duplication. */
1337
6940069f 1338U8*
35a4481c 1339Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1340{
35a4481c 1341 const U8 * const send = s + (*len);
6940069f
GS
1342 U8 *d;
1343 U8 *dst;
7918f24d
NC
1344
1345 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1346 PERL_UNUSED_CONTEXT;
6940069f 1347
212542aa 1348 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1349 dst = d;
1350
1351 while (s < send) {
55d09dc8
KW
1352 append_utf8_from_native_byte(*s, &d);
1353 s++;
6940069f
GS
1354 }
1355 *d = '\0';
6662521e 1356 *len = d-dst;
6940069f
GS
1357 return dst;
1358}
1359
a0ed51b3 1360/*
dea0fc0b 1361 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1362 *
1363 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1364 * We optimize for native, for obvious reasons. */
1365
1366U8*
dea0fc0b 1367Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1368{
dea0fc0b
JH
1369 U8* pend;
1370 U8* dstart = d;
1371
7918f24d
NC
1372 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1373
dea0fc0b 1374 if (bytelen & 1)
f5992bc4 1375 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1376
1377 pend = p + bytelen;
1378
a0ed51b3 1379 while (p < pend) {
dea0fc0b
JH
1380 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1381 p += 2;
2d1545e5 1382 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1383 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1384 continue;
1385 }
56d37426
KW
1386 if (uv <= MAX_UTF8_TWO_BYTE) {
1387 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1388 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1389 continue;
1390 }
46956fad
KW
1391#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1392#define LAST_HIGH_SURROGATE 0xDBFF
1393#define FIRST_LOW_SURROGATE 0xDC00
1394#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1395
1396 /* This assumes that most uses will be in the first Unicode plane, not
1397 * needing surrogates */
1398 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1399 && uv <= UNICODE_SURROGATE_LAST))
1400 {
1401 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1402 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1403 }
1404 else {
01ea242b 1405 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1406 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1407 || UNLIKELY(low > LAST_LOW_SURROGATE))
1408 {
01ea242b 1409 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1410 }
1411 p += 2;
46956fad
KW
1412 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1413 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1414 }
a0ed51b3 1415 }
56d37426
KW
1416#ifdef EBCDIC
1417 d = uvoffuni_to_utf8_flags(d, uv, 0);
1418#else
a0ed51b3 1419 if (uv < 0x10000) {
eb160463
GS
1420 *d++ = (U8)(( uv >> 12) | 0xe0);
1421 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1422 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1423 continue;
1424 }
1425 else {
eb160463
GS
1426 *d++ = (U8)(( uv >> 18) | 0xf0);
1427 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1428 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1429 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1430 continue;
1431 }
56d37426 1432#endif
a0ed51b3 1433 }
dea0fc0b 1434 *newlen = d - dstart;
a0ed51b3
LW
1435 return d;
1436}
1437
1438/* Note: this one is slightly destructive of the source. */
1439
1440U8*
dea0fc0b 1441Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1442{
1443 U8* s = (U8*)p;
d4c19fe8 1444 U8* const send = s + bytelen;
7918f24d
NC
1445
1446 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1447
e0ea5e2d
NC
1448 if (bytelen & 1)
1449 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1450 (UV)bytelen);
1451
a0ed51b3 1452 while (s < send) {
d4c19fe8 1453 const U8 tmp = s[0];
a0ed51b3
LW
1454 s[0] = s[1];
1455 s[1] = tmp;
1456 s += 2;
1457 }
dea0fc0b 1458 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1459}
1460
922e8cb4
KW
1461bool
1462Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1463{
1464 U8 tmpbuf[UTF8_MAXBYTES+1];
1465 uvchr_to_utf8(tmpbuf, c);
1466 return _is_utf8_FOO(classnum, tmpbuf);
1467}
1468
f9ae8fb6
JD
1469/* Internal function so we can deprecate the external one, and call
1470 this one from other deprecated functions in this file */
1471
f2645549
KW
1472bool
1473Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1474{
f2645549 1475 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1476
1477 if (*p == '_')
1478 return TRUE;
f25ce844 1479 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1480}
1481
5092f92a 1482bool
eba68aa0
KW
1483Perl__is_uni_perl_idcont(pTHX_ UV c)
1484{
1485 U8 tmpbuf[UTF8_MAXBYTES+1];
1486 uvchr_to_utf8(tmpbuf, c);
1487 return _is_utf8_perl_idcont(tmpbuf);
1488}
1489
1490bool
f91dcd13
KW
1491Perl__is_uni_perl_idstart(pTHX_ UV c)
1492{
1493 U8 tmpbuf[UTF8_MAXBYTES+1];
1494 uvchr_to_utf8(tmpbuf, c);
1495 return _is_utf8_perl_idstart(tmpbuf);
1496}
1497
3a4c58c9
KW
1498UV
1499Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1500{
1501 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1502 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1503 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1504 * either "SS" or "Ss". Which one to use is passed into the routine in
1505 * 'S_or_s' to avoid a test */
1506
1507 UV converted = toUPPER_LATIN1_MOD(c);
1508
1509 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1510
1511 assert(S_or_s == 'S' || S_or_s == 's');
1512
6f2d5cbc 1513 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1514 characters in this range */
3a4c58c9
KW
1515 *p = (U8) converted;
1516 *lenp = 1;
1517 return converted;
1518 }
1519
1520 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1521 * which it maps to one of them, so as to only have to have one check for
1522 * it in the main case */
1523 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1524 switch (c) {
1525 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1526 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1527 break;
1528 case MICRO_SIGN:
1529 converted = GREEK_CAPITAL_LETTER_MU;
1530 break;
79e064b9
KW
1531#if UNICODE_MAJOR_VERSION > 2 \
1532 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1533 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1534 case LATIN_SMALL_LETTER_SHARP_S:
1535 *(p)++ = 'S';
1536 *p = S_or_s;
1537 *lenp = 2;
1538 return 'S';
79e064b9 1539#endif
3a4c58c9
KW
1540 default:
1541 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1542 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1543 }
1544 }
1545
1546 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1547 *p = UTF8_TWO_BYTE_LO(converted);
1548 *lenp = 2;
1549
1550 return converted;
1551}
1552
50bda2c3
KW
1553/* Call the function to convert a UTF-8 encoded character to the specified case.
1554 * Note that there may be more than one character in the result.
1555 * INP is a pointer to the first byte of the input character
1556 * OUTP will be set to the first byte of the string of changed characters. It
1557 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1558 * LENP will be set to the length in bytes of the string of changed characters
1559 *
1560 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
1561#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
1562#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
1563#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 1564
b9992569
KW
1565/* This additionally has the input parameter 'specials', which if non-zero will
1566 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 1567 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 1568#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 1569
84afefe6
JH
1570UV
1571Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1572{
a1433954
KW
1573 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1574 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1575 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1576 * the changed version may be longer than the original character.
1577 *
1578 * The ordinal of the first character of the changed version is returned
1579 * (but note, as explained above, that there may be more.) */
1580
7918f24d
NC
1581 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1582
3a4c58c9
KW
1583 if (c < 256) {
1584 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1585 }
1586
0ebc6274 1587 uvchr_to_utf8(p, c);
b9992569 1588 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
1589}
1590
84afefe6
JH
1591UV
1592Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1593{
7918f24d
NC
1594 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1595
3a4c58c9
KW
1596 if (c < 256) {
1597 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1598 }
1599
0ebc6274 1600 uvchr_to_utf8(p, c);
b9992569 1601 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
1602}
1603
afc16117 1604STATIC U8
81611534 1605S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1606{
1607 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1608 * those, converting the result to UTF-8. Since the result is always just
a1433954 1609 * one character, we allow <p> to be NULL */
afc16117
KW
1610
1611 U8 converted = toLOWER_LATIN1(c);
1612
1613 if (p != NULL) {
6f2d5cbc 1614 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1615 *p = converted;
1616 *lenp = 1;
1617 }
1618 else {
430c9760
KW
1619 /* Result is known to always be < 256, so can use the EIGHT_BIT
1620 * macros */
1621 *p = UTF8_EIGHT_BIT_HI(converted);
1622 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1623 *lenp = 2;
1624 }
1625 }
1626 return converted;
1627}
1628
84afefe6
JH
1629UV
1630Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1631{
7918f24d
NC
1632 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1633
afc16117
KW
1634 if (c < 256) {
1635 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1636 }
1637
afc16117 1638 uvchr_to_utf8(p, c);
b9992569 1639 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
1640}
1641
84afefe6 1642UV
51910141 1643Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1644{
51910141 1645 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1646 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1647 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1648 *
1649 * Not to be used for locale folds
51910141 1650 */
f673fad4 1651
a1dde8de
KW
1652 UV converted;
1653
1654 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1655 PERL_UNUSED_CONTEXT;
a1dde8de 1656
1ca267a5
KW
1657 assert (! (flags & FOLD_FLAGS_LOCALE));
1658
659a7c2d 1659 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
1660 converted = GREEK_SMALL_LETTER_MU;
1661 }
9b63e895
KW
1662#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1663 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1664 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
1665 else if ( (flags & FOLD_FLAGS_FULL)
1666 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
1667 {
1ca267a5
KW
1668 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1669 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1670 * under those circumstances. */
1671 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1672 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1673 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1674 p, *lenp, U8);
1675 return LATIN_SMALL_LETTER_LONG_S;
1676 }
1677 else {
4f489194
KW
1678 *(p)++ = 's';
1679 *p = 's';
1680 *lenp = 2;
1681 return 's';
1ca267a5 1682 }
a1dde8de 1683 }
9b63e895 1684#endif
a1dde8de
KW
1685 else { /* In this range the fold of all other characters is their lower
1686 case */
1687 converted = toLOWER_LATIN1(c);
1688 }
1689
6f2d5cbc 1690 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1691 *p = (U8) converted;
1692 *lenp = 1;
1693 }
1694 else {
1695 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1696 *p = UTF8_TWO_BYTE_LO(converted);
1697 *lenp = 2;
1698 }
1699
1700 return converted;
1701}
1702
1703UV
31f05a37 1704Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1705{
4b593389 1706
a0270393
KW
1707 /* Not currently externally documented, and subject to change
1708 * <flags> bits meanings:
1709 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1710 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1711 * locale are to be used.
a0270393
KW
1712 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1713 */
4b593389 1714
36bb2ab6 1715 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1716
780fcc9f
KW
1717 if (flags & FOLD_FLAGS_LOCALE) {
1718 /* Treat a UTF-8 locale as not being in locale at all */
1719 if (IN_UTF8_CTYPE_LOCALE) {
1720 flags &= ~FOLD_FLAGS_LOCALE;
1721 }
1722 else {
1723 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1724 goto needs_full_generality;
780fcc9f 1725 }
31f05a37
KW
1726 }
1727
a1dde8de 1728 if (c < 256) {
e7b7ac46 1729 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1730 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1731 }
1732
2f306ab9 1733 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1734 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1735 uvchr_to_utf8(p, c);
b9992569 1736 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
1737 }
1738 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1739 the special flags. */
1740 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1741
1742 needs_full_generality:
a0270393 1743 uvchr_to_utf8(utf8_c, c);
445bf929 1744 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1745 }
84afefe6
JH
1746}
1747
26483009 1748PERL_STATIC_INLINE bool
5141f98e 1749S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1750 const char *const swashname, SV* const invlist)
bde6a22d 1751{
ea317ccb
KW
1752 /* returns a boolean giving whether or not the UTF8-encoded character that
1753 * starts at <p> is in the swash indicated by <swashname>. <swash>
1754 * contains a pointer to where the swash indicated by <swashname>
1755 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1756 * look at <*swash> and only generate a swash if it is not null. <invlist>
1757 * is NULL or an inversion list that defines the swash. If not null, it
1758 * saves time during initialization of the swash.
ea317ccb
KW
1759 *
1760 * Note that it is assumed that the buffer length of <p> is enough to
1761 * contain all the bytes that comprise the character. Thus, <*p> should
1762 * have been checked before this call for mal-formedness enough to assure
1763 * that. */
1764
7918f24d
NC
1765 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1766
492a624f 1767 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1768 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1769 * as far as there being enough bytes available in it to accommodate the
1770 * character without reading beyond the end, and pass that number on to the
1771 * validating routine */
6302f837 1772 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1773 if (ckWARN_d(WARN_UTF8)) {
1774 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1775 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1776 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1777 what the malformation is */
1778 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1779 }
1780 }
1781 return FALSE;
1782 }
87367d5f
KW
1783 if (!*swash) {
1784 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1785 *swash = _core_swash_init("utf8",
1786
1787 /* Only use the name if there is no inversion
1788 * list; otherwise will go out to disk */
1789 (invlist) ? "" : swashname,
1790
1791 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1792 }
28123549 1793
bde6a22d
NC
1794 return swash_fetch(*swash, p, TRUE) != 0;
1795}
1796
1797bool
922e8cb4
KW
1798Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1799{
922e8cb4
KW
1800 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1801
1802 assert(classnum < _FIRST_NON_SWASH_CC);
1803
f25ce844
KW
1804 return is_utf8_common(p,
1805 &PL_utf8_swash_ptrs[classnum],
1806 swash_property_names[classnum],
1807 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1808}
1809
1810bool
f2645549 1811Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1812{
f2645549 1813 SV* invlist = NULL;
7918f24d 1814
f2645549 1815 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1816
f2645549
KW
1817 if (! PL_utf8_perl_idstart) {
1818 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1819 }
60071a22 1820 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1821}
1822
1823bool
f2645549 1824Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1825{
f2645549 1826 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1827
1828 if (*p == '_')
1829 return TRUE;
f25ce844 1830 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1831}
1832
1833bool
eba68aa0
KW
1834Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1835{
b24b43f7 1836 SV* invlist = NULL;
eba68aa0
KW
1837
1838 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1839
b24b43f7
KW
1840 if (! PL_utf8_perl_idcont) {
1841 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1842 }
60071a22 1843 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1844}
1845
eba68aa0 1846bool
f2645549 1847Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1848{
f2645549 1849 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1850
f25ce844 1851 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1852}
1853
1854bool
f2645549 1855Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1856{
f2645549 1857 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1858
f25ce844 1859 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1860}
1861
1862bool
7dbf68d2
KW
1863Perl__is_utf8_mark(pTHX_ const U8 *p)
1864{
7dbf68d2
KW
1865 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1866
f25ce844 1867 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1868}
1869
6b5c0936 1870/*
87cea99e 1871=for apidoc to_utf8_case
6b5c0936 1872
9da1e7cb
KW
1873Instead use the appropriate one of L</toUPPER_utf8>,
1874L</toTITLE_utf8>,
1875L</toLOWER_utf8>,
1876or L</toFOLD_utf8>.
1877
6fae5207 1878C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1879the character that is being converted. This routine assumes that the character
1880at C<p> is well-formed.
6b5c0936 1881
6fae5207
KW
1882C<ustrp> is a pointer to the character buffer to put the
1883conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1884of the result.
1885
6fae5207 1886C<swashp> is a pointer to the swash to use.
6b5c0936 1887
a1433954 1888Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1889and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1890but not always, a multicharacter mapping), is tried first.
6b5c0936 1891
4a8240a3
KW
1892C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1893any special mappings; C<""> means to use the special mappings. Values other
1894than these two are treated as the name of the hash containing the special
1895mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1896
796b6530
KW
1897C<normal> is a string like C<"ToLower"> which means the swash
1898C<%utf8::ToLower>.
0134edef 1899
760c7c2f
KW
1900Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1901unless those are turned off.
1902
0134edef 1903=cut */
6b5c0936 1904
2104c8d9 1905UV
9a957fbc
AL
1906Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1907 SV **swashp, const char *normal, const char *special)
a0ed51b3 1908{
b9992569
KW
1909 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1910
1911 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
1912}
1913
1914 /* change namve uv1 to 'from' */
6a4a25f4 1915STATIC UV
b9992569
KW
1916S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
1917 SV **swashp, const char *normal, const char *special)
1918{
0134edef 1919 STRLEN len = 0;
7918f24d 1920
b9992569 1921 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 1922
36eaa811
KW
1923 /* For code points that don't change case, we already know that the output
1924 * of this function is the unchanged input, so we can skip doing look-ups
1925 * for them. Unfortunately the case-changing code points are scattered
1926 * around. But there are some long consecutive ranges where there are no
1927 * case changing code points. By adding tests, we can eliminate the lookup
1928 * for all the ones in such ranges. This is currently done here only for
1929 * just a few cases where the scripts are in common use in modern commerce
1930 * (and scripts adjacent to those which can be included without additional
1931 * tests). */
1932
1933 if (uv1 >= 0x0590) {
1934 /* This keeps from needing further processing the code points most
1935 * likely to be used in the following non-cased scripts: Hebrew,
1936 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
1937 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
1938 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
1939 if (uv1 < 0x10A0) {
1940 goto cases_to_self;
1941 }
1942
1943 /* The following largish code point ranges also don't have case
1944 * changes, but khw didn't think they warranted extra tests to speed
1945 * them up (which would slightly slow down everything else above them):
1946 * 1100..139F Hangul Jamo, Ethiopic
1947 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
1948 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
1949 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
1950 * Combining Diacritical Marks Extended, Balinese,
1951 * Sundanese, Batak, Lepcha, Ol Chiki
1952 * 2000..206F General Punctuation
1953 */
1954
1955 if (uv1 >= 0x2D30) {
1956
1957 /* This keeps the from needing further processing the code points
1958 * most likely to be used in the following non-cased major scripts:
1959 * CJK, Katakana, Hiragana, plus some less-likely scripts.
1960 *
1961 * (0x2D30 above might have to be changed to 2F00 in the unlikely
1962 * event that Unicode eventually allocates the unused block as of
1963 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
1964 * that the test suite will start having failures to alert you
1965 * should that happen) */
1966 if (uv1 < 0xA640) {
1967 goto cases_to_self;
1968 }
1969
1970 if (uv1 >= 0xAC00) {
1971 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
1972 if (ckWARN_d(WARN_SURROGATE)) {
1973 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1974 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1975 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1976 }
1977 goto cases_to_self;
1978 }
36eaa811
KW
1979
1980 /* AC00..FAFF Catches Hangul syllables and private use, plus
1981 * some others */
1982 if (uv1 < 0xFB00) {
1983 goto cases_to_self;
1984
1985 }
1986
5af9bc97
KW
1987 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
1988 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1989 && ckWARN_d(WARN_DEPRECATED))
1990 {
1991 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1992 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1993 }
1994 if (ckWARN_d(WARN_NON_UNICODE)) {
1995 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1996 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1997 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1998 }
1999 goto cases_to_self;
2000 }
3bfc1e70
KW
2001#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2002 if (UNLIKELY(uv1
2003 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2004 {
2005
2006 /* As of this writing, this means we avoid swash creation
2007 * for anything beyond low Plane 1 */
2008 goto cases_to_self;
2009 }
2010#endif
36eaa811
KW
2011 }
2012 }
9ae3ac1a 2013
36eaa811
KW
2014 /* Note that non-characters are perfectly legal, so no warning should
2015 * be given. There are so few of them, that it isn't worth the extra
2016 * tests to avoid swash creation */
9ae3ac1a
KW
2017 }
2018
0134edef 2019 if (!*swashp) /* load on-demand */
5ab9d2ef 2020 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2021
a6f87d8c 2022 if (special) {
0134edef 2023 /* It might be "special" (sometimes, but not always,
2a37f04d 2024 * a multicharacter mapping) */
4a8240a3 2025 HV *hv = NULL;
b08cf34e
JH
2026 SV **svp;
2027
4a8240a3
KW
2028 /* If passed in the specials name, use that; otherwise use any
2029 * given in the swash */
2030 if (*special != '\0') {
2031 hv = get_hv(special, 0);
2032 }
2033 else {
2034 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2035 if (svp) {
2036 hv = MUTABLE_HV(SvRV(*svp));
2037 }
2038 }
2039
176fe009 2040 if (hv
5f560d8a 2041 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2042 && (*svp))
2043 {
cfd0369c 2044 const char *s;
47654450 2045
cfd0369c 2046 s = SvPV_const(*svp, len);
47654450 2047 if (len == 1)
f4cd282c 2048 /* EIGHTBIT */
c80e42f3 2049 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2050 else {
d2dcd0fb 2051 Copy(s, ustrp, len, U8);
29e98929 2052 }
983ffd37 2053 }
0134edef
JH
2054 }
2055
2056 if (!len && *swashp) {
4a4088c4 2057 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2058
0134edef
JH
2059 if (uv2) {
2060 /* It was "normal" (a single character mapping). */
f4cd282c 2061 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2062 }
2063 }
1feea2c7 2064
cbe07460
KW
2065 if (len) {
2066 if (lenp) {
2067 *lenp = len;
2068 }
2069 return valid_utf8_to_uvchr(ustrp, 0);
2070 }
2071
2072 /* Here, there was no mapping defined, which means that the code point maps
2073 * to itself. Return the inputs */
e24dfe9c 2074 cases_to_self:
bfdf22ec 2075 len = UTF8SKIP(p);
ca9fab46
KW
2076 if (p != ustrp) { /* Don't copy onto itself */
2077 Copy(p, ustrp, len, U8);
2078 }
0134edef 2079
2a37f04d
JH
2080 if (lenp)
2081 *lenp = len;
2082
f4cd282c 2083 return uv1;
cbe07460 2084
a0ed51b3
LW
2085}
2086
051a06d4 2087STATIC UV
357aadde 2088S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2089{
4a4088c4 2090 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2091 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2092 * result contains a character that crosses the 255/256 boundary, disallow
2093 * the change, and return the original code point. See L<perlfunc/lc> for
2094 * why;
051a06d4 2095 *
a1433954
KW
2096 * p points to the original string whose case was changed; assumed
2097 * by this routine to be well-formed
051a06d4
KW
2098 * result the code point of the first character in the changed-case string
2099 * ustrp points to the changed-case string (<result> represents its first char)
2100 * lenp points to the length of <ustrp> */
2101
2102 UV original; /* To store the first code point of <p> */
2103
2104 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2105
a4f12ed7 2106 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2107
2108 /* We know immediately if the first character in the string crosses the
2109 * boundary, so can skip */
2110 if (result > 255) {
2111
2112 /* Look at every character in the result; if any cross the
2113 * boundary, the whole thing is disallowed */
2114 U8* s = ustrp + UTF8SKIP(ustrp);
2115 U8* e = ustrp + *lenp;
2116 while (s < e) {
a4f12ed7 2117 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2118 goto bad_crossing;
2119 }
2120 s += UTF8SKIP(s);
2121 }
2122
613abc6d
KW
2123 /* Here, no characters crossed, result is ok as-is, but we warn. */
2124 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2125 return result;
2126 }
2127
7b52d656 2128 bad_crossing:
051a06d4
KW
2129
2130 /* Failed, have to return the original */
4b88fb76 2131 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2132
2133 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2134 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2135 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2136 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2137 OP_DESC(PL_op),
ab0b796c
KW
2138 original,
2139 original);
051a06d4
KW
2140 Copy(p, ustrp, *lenp, char);
2141 return original;
2142}
2143
d3e79532 2144/*
87cea99e 2145=for apidoc to_utf8_upper
d3e79532 2146
1f607577 2147Instead use L</toUPPER_utf8>.
a1433954 2148
d3e79532
JH
2149=cut */
2150
051a06d4 2151/* Not currently externally documented, and subject to change:
31f05a37
KW
2152 * <flags> is set iff iff the rules from the current underlying locale are to
2153 * be used. */
051a06d4 2154
2104c8d9 2155UV
31f05a37 2156Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2157{
051a06d4
KW
2158 UV result;
2159
2160 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2161
780fcc9f
KW
2162 if (flags) {
2163 /* Treat a UTF-8 locale as not being in locale at all */
2164 if (IN_UTF8_CTYPE_LOCALE) {
2165 flags = FALSE;
2166 }
2167 else {
2168 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2169 }
31f05a37
KW
2170 }
2171
3a4c58c9 2172 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2173 if (flags) {
2174 result = toUPPER_LC(*p);
2175 }
2176 else {
81c6c7ce 2177 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2178 }
3a4c58c9
KW
2179 }
2180 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2181 if (flags) {
a62b247b 2182 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2183 result = toUPPER_LC(c);
051a06d4
KW
2184 }
2185 else {
a62b247b 2186 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2187 ustrp, lenp, 'S');
051a06d4
KW
2188 }
2189 }
4a4088c4 2190 else { /* UTF-8, ord above 255 */
b9992569 2191 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2192
2193 if (flags) {
357aadde 2194 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2195 }
2196 return result;
2197 }
2198
4a4088c4 2199 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2200 if (UTF8_IS_INVARIANT(result)) {
2201 *ustrp = (U8) result;
2202 *lenp = 1;
2203 }
2204 else {
62cb07ea
KW
2205 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2206 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2207 *lenp = 2;
3a4c58c9 2208 }
baa60164 2209
051a06d4 2210 return result;
983ffd37 2211}
a0ed51b3 2212
d3e79532 2213/*
87cea99e 2214=for apidoc to_utf8_title
d3e79532 2215
1f607577 2216Instead use L</toTITLE_utf8>.
a1433954 2217
d3e79532
JH
2218=cut */
2219
051a06d4 2220/* Not currently externally documented, and subject to change:
31f05a37
KW
2221 * <flags> is set iff the rules from the current underlying locale are to be
2222 * used. Since titlecase is not defined in POSIX, for other than a
2223 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2224 */
051a06d4 2225
983ffd37 2226UV
31f05a37 2227Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2228{
051a06d4
KW
2229 UV result;
2230
2231 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2232
780fcc9f
KW
2233 if (flags) {
2234 /* Treat a UTF-8 locale as not being in locale at all */
2235 if (IN_UTF8_CTYPE_LOCALE) {
2236 flags = FALSE;
2237 }
2238 else {
2239 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2240 }
31f05a37
KW
2241 }
2242
3a4c58c9 2243 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2244 if (flags) {
2245 result = toUPPER_LC(*p);
2246 }
2247 else {
81c6c7ce 2248 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2249 }
3a4c58c9
KW
2250 }
2251 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2252 if (flags) {
a62b247b 2253 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2254 result = toUPPER_LC(c);
051a06d4
KW
2255 }
2256 else {
a62b247b 2257 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2258 ustrp, lenp, 's');
051a06d4
KW
2259 }
2260 }
4a4088c4 2261 else { /* UTF-8, ord above 255 */
b9992569 2262 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2263
2264 if (flags) {
357aadde 2265 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2266 }
2267 return result;
2268 }
2269
4a4088c4 2270 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2271 if (UTF8_IS_INVARIANT(result)) {
2272 *ustrp = (U8) result;
2273 *lenp = 1;
2274 }
2275 else {
62cb07ea
KW
2276 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2277 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2278 *lenp = 2;
3a4c58c9
KW
2279 }
2280
051a06d4 2281 return result;
a0ed51b3
LW
2282}
2283
d3e79532 2284/*
87cea99e 2285=for apidoc to_utf8_lower
d3e79532 2286
1f607577 2287Instead use L</toLOWER_utf8>.
a1433954 2288
d3e79532
JH
2289=cut */
2290
051a06d4 2291/* Not currently externally documented, and subject to change:
31f05a37
KW
2292 * <flags> is set iff iff the rules from the current underlying locale are to
2293 * be used.
2294 */
051a06d4 2295
2104c8d9 2296UV
31f05a37 2297Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2298{
051a06d4
KW
2299 UV result;
2300
051a06d4 2301 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2302
780fcc9f
KW
2303 if (flags) {
2304 /* Treat a UTF-8 locale as not being in locale at all */
2305 if (IN_UTF8_CTYPE_LOCALE) {
2306 flags = FALSE;
2307 }
2308 else {
2309 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2310 }
31f05a37
KW
2311 }
2312
968c5e6a 2313 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2314 if (flags) {
2315 result = toLOWER_LC(*p);
2316 }
2317 else {
81c6c7ce 2318 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2319 }
968c5e6a
KW
2320 }
2321 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2322 if (flags) {
a62b247b 2323 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2324 result = toLOWER_LC(c);
051a06d4
KW
2325 }
2326 else {
a62b247b 2327 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2328 ustrp, lenp);
051a06d4 2329 }
968c5e6a 2330 }
4a4088c4 2331 else { /* UTF-8, ord above 255 */
b9992569 2332 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2333
2334 if (flags) {
357aadde 2335 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2336 }
968c5e6a 2337
051a06d4
KW
2338 return result;
2339 }
2340
4a4088c4 2341 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2342 if (UTF8_IS_INVARIANT(result)) {
2343 *ustrp = (U8) result;
2344 *lenp = 1;
2345 }
2346 else {
62cb07ea
KW
2347 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2348 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2349 *lenp = 2;
2350 }
2351
051a06d4 2352 return result;
b4e400f9
JH
2353}
2354
d3e79532 2355/*
87cea99e 2356=for apidoc to_utf8_fold
d3e79532 2357
1f607577 2358Instead use L</toFOLD_utf8>.
a1433954 2359
d3e79532
JH
2360=cut */
2361
051a06d4
KW
2362/* Not currently externally documented, and subject to change,
2363 * in <flags>
31f05a37
KW
2364 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2365 * locale are to be used.
051a06d4
KW
2366 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2367 * otherwise simple folds
a0270393
KW
2368 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2369 * prohibited
445bf929 2370 */
36bb2ab6 2371
b4e400f9 2372UV
445bf929 2373Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2374{
051a06d4
KW
2375 UV result;
2376
36bb2ab6 2377 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2378
a0270393
KW
2379 /* These are mutually exclusive */
2380 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2381
50ba90ff
KW
2382 assert(p != ustrp); /* Otherwise overwrites */
2383
780fcc9f
KW
2384 if (flags & FOLD_FLAGS_LOCALE) {
2385 /* Treat a UTF-8 locale as not being in locale at all */
2386 if (IN_UTF8_CTYPE_LOCALE) {
2387 flags &= ~FOLD_FLAGS_LOCALE;
2388 }
2389 else {
2390 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2391 }
31f05a37
KW
2392 }
2393
a1dde8de 2394 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2395 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2396 result = toFOLD_LC(*p);
051a06d4
KW
2397 }
2398 else {
81c6c7ce 2399 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2400 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2401 }
a1dde8de
KW
2402 }
2403 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2404 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2405 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2406 result = toFOLD_LC(c);
051a06d4
KW
2407 }
2408 else {
a62b247b 2409 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2410 ustrp, lenp,
1ca267a5 2411 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2412 }
a1dde8de 2413 }
4a4088c4 2414 else { /* UTF-8, ord above 255 */
b9992569 2415 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2416
1ca267a5
KW
2417 if (flags & FOLD_FLAGS_LOCALE) {
2418
76f2ffcd 2419# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2420 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2421
2422# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2423# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2424
2425 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2426
538e84ed
KW
2427 /* Special case these two characters, as what normally gets
2428 * returned under locale doesn't work */
76f2ffcd
KW
2429 if (UTF8SKIP(p) == cap_sharp_s_len
2430 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2431 {
ab0b796c
KW
2432 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2433 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2434 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2435 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2436 goto return_long_s;
2437 }
0766489e
KW
2438 else
2439#endif
2440 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2441 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2442 {
ab0b796c
KW
2443 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2444 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2445 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2446 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2447 goto return_ligature_st;
2448 }
74894415
KW
2449
2450#if UNICODE_MAJOR_VERSION == 3 \
2451 && UNICODE_DOT_VERSION == 0 \
2452 && UNICODE_DOT_DOT_VERSION == 1
2453# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2454
2455 /* And special case this on this Unicode version only, for the same
2456 * reaons the other two are special cased. They would cross the
2457 * 255/256 boundary which is forbidden under /l, and so the code
2458 * wouldn't catch that they are equivalent (which they are only in
2459 * this release) */
2460 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2461 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2462 {
2463 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2464 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2465 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2466 "resolved to \"\\x{0131}\".");
2467 goto return_dotless_i;
2468 }
2469#endif
2470
357aadde 2471 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2472 }
a0270393
KW
2473 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2474 return result;
2475 }
2476 else {
4a4088c4 2477 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2478 * character above the ASCII range, and the result should not
2479 * contain an ASCII character. */
a0270393
KW
2480
2481 UV original; /* To store the first code point of <p> */
2482
2483 /* Look at every character in the result; if any cross the
2484 * boundary, the whole thing is disallowed */
2485 U8* s = ustrp;
2486 U8* e = ustrp + *lenp;
2487 while (s < e) {
2488 if (isASCII(*s)) {
2489 /* Crossed, have to return the original */
2490 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2491
9fc2026f 2492 /* But in these instances, there is an alternative we can
1ca267a5 2493 * return that is valid */
0766489e
KW
2494 if (original == LATIN_SMALL_LETTER_SHARP_S
2495#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2496 || original == LATIN_CAPITAL_LETTER_SHARP_S
2497#endif
2498 ) {
1ca267a5
KW
2499 goto return_long_s;
2500 }
9fc2026f
KW
2501 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2502 goto return_ligature_st;
2503 }
74894415
KW
2504#if UNICODE_MAJOR_VERSION == 3 \
2505 && UNICODE_DOT_VERSION == 0 \
2506 && UNICODE_DOT_DOT_VERSION == 1
2507
2508 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2509 goto return_dotless_i;
2510 }
2511#endif
a0270393
KW
2512 Copy(p, ustrp, *lenp, char);
2513 return original;
2514 }
2515 s += UTF8SKIP(s);
2516 }
051a06d4 2517
a0270393
KW
2518 /* Here, no characters crossed, result is ok as-is */
2519 return result;
2520 }
051a06d4
KW
2521 }
2522
4a4088c4 2523 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2524 if (UTF8_IS_INVARIANT(result)) {
2525 *ustrp = (U8) result;
2526 *lenp = 1;
2527 }
2528 else {
62cb07ea
KW
2529 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2530 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2531 *lenp = 2;
2532 }
2533
051a06d4 2534 return result;
1ca267a5
KW
2535
2536 return_long_s:
2537 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2538 * folds to a string of two of these characters. By returning this
2539 * instead, then, e.g.,
2540 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2541 * works. */
2542
2543 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2544 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2545 ustrp, *lenp, U8);
2546 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2547
2548 return_ligature_st:
2549 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2550 * have the other one fold to it */
2551
2552 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2553 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2554 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2555
2556#if UNICODE_MAJOR_VERSION == 3 \
2557 && UNICODE_DOT_VERSION == 0 \
2558 && UNICODE_DOT_DOT_VERSION == 1
2559
2560 return_dotless_i:
2561 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2562 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2563 return LATIN_SMALL_LETTER_DOTLESS_I;
2564
2565#endif
2566
a0ed51b3
LW
2567}
2568
711a919c 2569/* Note:
f90a9a02 2570 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2571 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2572 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2573 */
c4a5db0c 2574
a0ed51b3 2575SV*
7fc63493 2576Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2577{
c4a5db0c
KW
2578 PERL_ARGS_ASSERT_SWASH_INIT;
2579
2580 /* Returns a copy of a swash initiated by the called function. This is the
2581 * public interface, and returning a copy prevents others from doing
2582 * mischief on the original */
2583
5d3d13d1 2584 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2585}
2586
2587SV*
5d3d13d1 2588Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2589{
2c1f00b9
YO
2590
2591 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2592 * use the following define */
2593
2594#define CORE_SWASH_INIT_RETURN(x) \
2595 PL_curpm= old_PL_curpm; \
2596 return x
2597
c4a5db0c 2598 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2599 * by calling utf8_heavy.pl in the general case. The returned value may be
2600 * the swash's inversion list instead if the input parameters allow it.
2601 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2602 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2603 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2604 *
2605 * This interface should only be used by functions that won't destroy or
2606 * adversely change the swash, as doing so affects all other uses of the
2607 * swash in the program; the general public should use 'Perl_swash_init'
2608 * instead.
2609 *
2610 * pkg is the name of the package that <name> should be in.
2611 * name is the name of the swash to find. Typically it is a Unicode
2612 * property name, including user-defined ones
2613 * listsv is a string to initialize the swash with. It must be of the form
2614 * documented as the subroutine return value in
2615 * L<perlunicode/User-Defined Character Properties>
2616 * minbits is the number of bits required to represent each data element.
2617 * It is '1' for binary properties.
2618 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2619 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2620 * flags_p if non-NULL is the address of various input and output flag bits
2621 * to the routine, as follows: ('I' means is input to the routine;
2622 * 'O' means output from the routine. Only flags marked O are
2623 * meaningful on return.)
2624 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2625 * came from a user-defined property. (I O)
5d3d13d1
KW
2626 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2627 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2628 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2629 * return of an inversion list instead of a swash hash if this routine
2630 * thinks that would result in faster execution of swash_fetch() later
2631 * on. (I)
9a53f6cf
KW
2632 *
2633 * Thus there are three possible inputs to find the swash: <name>,
2634 * <listsv>, and <invlist>. At least one must be specified. The result
2635 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2636 * actions can intersect, etc. what <name> gives. To avoid going out to
2637 * disk at all, <invlist> should specify completely what the swash should
2638 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2639 *
2640 * <invlist> is only valid for binary properties */
c4a5db0c 2641
2c1f00b9
YO
2642 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2643
c4a5db0c 2644 SV* retval = &PL_sv_undef;
83199d38 2645 HV* swash_hv = NULL;
87367d5f
KW
2646 const int invlist_swash_boundary =
2647 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2648 ? 512 /* Based on some benchmarking, but not extensive, see commit
2649 message */
2650 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2651
2652 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2653 assert(! invlist || minbits == 1);
2654
2c1f00b9
YO
2655 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2656 that triggered the swash init and the swash init perl logic itself.
2657 See perl #122747 */
2658
9a53f6cf
KW
2659 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2660 * so */
2661 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2662 dSP;
2663 const size_t pkg_len = strlen(pkg);
2664 const size_t name_len = strlen(name);
2665 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2666 SV* errsv_save;
2667 GV *method;
2668
2669 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2670
2671 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2672 ENTER;
69794297 2673 SAVEHINTS();
2782061f 2674 save_re_context();
650f067c
JL
2675 /* We might get here via a subroutine signature which uses a utf8
2676 * parameter name, at which point PL_subname will have been set
2677 * but not yet used. */
2678 save_item(PL_subname);
69794297
KW
2679 if (PL_parser && PL_parser->error_count)
2680 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2681 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2682 if (!method) { /* demand load UTF-8 */
69794297 2683 ENTER;
db2c6cb3
FC
2684 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2685 GvSV(PL_errgv) = NULL;
1a419e6b 2686#ifndef NO_TAINT_SUPPORT
69794297
KW
2687 /* It is assumed that callers of this routine are not passing in
2688 * any user derived data. */
2782061f
DM
2689 /* Need to do this after save_re_context() as it will set
2690 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2691 * in Perl_magic_get). Even line to create errsv_save can turn on
2692 * PL_tainted. */
284167a5
SM
2693 SAVEBOOL(TAINT_get);
2694 TAINT_NOT;
2695#endif
69794297
KW
2696 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2697 NULL);
eed484f9 2698 {
db2c6cb3
FC
2699 /* Not ERRSV, as there is no need to vivify a scalar we are
2700 about to discard. */
2701 SV * const errsv = GvSV(PL_errgv);
2702 if (!SvTRUE(errsv)) {
2703 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2704 SvREFCNT_dec(errsv);
2705 }
eed484f9 2706 }
69794297
KW
2707 LEAVE;
2708 }
2709 SPAGAIN;
2710 PUSHMARK(SP);
2711 EXTEND(SP,5);
2712 mPUSHp(pkg, pkg_len);
2713 mPUSHp(name, name_len);
2714 PUSHs(listsv);
2715 mPUSHi(minbits);
2716 mPUSHi(none);
2717 PUTBACK;
db2c6cb3
FC
2718 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2719 GvSV(PL_errgv) = NULL;
69794297
KW
2720 /* If we already have a pointer to the method, no need to use
2721 * call_method() to repeat the lookup. */
c41800a8
KW
2722 if (method
2723 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2724 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2725 {
2726 retval = *PL_stack_sp--;
2727 SvREFCNT_inc(retval);
2728 }
eed484f9 2729 {
db2c6cb3
FC
2730 /* Not ERRSV. See above. */
2731 SV * const errsv = GvSV(PL_errgv);
2732 if (!SvTRUE(errsv)) {
2733 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2734 SvREFCNT_dec(errsv);
2735 }
eed484f9 2736 }
ce3b816e 2737 LEAVE;
69794297
KW
2738 POPSTACK;
2739 if (IN_PERL_COMPILETIME) {
2740 CopHINTS_set(PL_curcop, PL_hints);
2741 }
2742 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 2743 if (SvPOK(retval)) {
69794297
KW
2744
2745 /* If caller wants to handle missing properties, let them */
5d3d13d1 2746 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2747 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2748 }
2749 Perl_croak(aTHX_
2750 "Can't find Unicode property definition \"%"SVf"\"",
2751 SVfARG(retval));
a25b5927 2752 NOT_REACHED; /* NOTREACHED */
d95e4a00 2753 }
69794297 2754 }
9a53f6cf 2755 } /* End of calling the module to find the swash */
36eb48b4 2756
83199d38
KW
2757 /* If this operation fetched a swash, and we will need it later, get it */
2758 if (retval != &PL_sv_undef
2759 && (minbits == 1 || (flags_p
2760 && ! (*flags_p
2761 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2762 {
2763 swash_hv = MUTABLE_HV(SvRV(retval));
2764
2765 /* If we don't already know that there is a user-defined component to
2766 * this swash, and the user has indicated they wish to know if there is
2767 * one (by passing <flags_p>), find out */
2768 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2769 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2770 if (user_defined && SvUV(*user_defined)) {
2771 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2772 }
2773 }
2774 }
2775
36eb48b4
KW
2776 /* Make sure there is an inversion list for binary properties */
2777 if (minbits == 1) {
2778 SV** swash_invlistsvp = NULL;
2779 SV* swash_invlist = NULL;
9a53f6cf 2780 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2781 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2782 an unclaimed reference count */
36eb48b4 2783
9a53f6cf 2784 /* If this operation fetched a swash, get its already existing
83199d38 2785 * inversion list, or create one for it */
36eb48b4 2786
83199d38 2787 if (swash_hv) {
5c9f4bd2 2788 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2789 if (swash_invlistsvp) {
2790 swash_invlist = *swash_invlistsvp;
2791 invlist_in_swash_is_valid = TRUE;
2792 }
2793 else {
36eb48b4 2794 swash_invlist = _swash_to_invlist(retval);
02c85471 2795 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2796 }
2797 }
2798
2799 /* If an inversion list was passed in, have to include it */
2800 if (invlist) {
2801
2802 /* Any fetched swash will by now have an inversion list in it;
2803 * otherwise <swash_invlist> will be NULL, indicating that we
2804 * didn't fetch a swash */
2805 if (swash_invlist) {
2806
2807 /* Add the passed-in inversion list, which invalidates the one
2808 * already stored in the swash */
2809 invlist_in_swash_is_valid = FALSE;
2810 _invlist_union(invlist, swash_invlist, &swash_invlist);
2811 }
2812 else {
2813
87367d5f
KW
2814 /* Here, there is no swash already. Set up a minimal one, if
2815 * we are going to return a swash */
2816 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2817 swash_hv = newHV();
4aca0fe6 2818 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2819 }
9a53f6cf
KW
2820 swash_invlist = invlist;
2821 }
9a53f6cf
KW
2822 }
2823
2824 /* Here, we have computed the union of all the passed-in data. It may
2825 * be that there was an inversion list in the swash which didn't get
538e84ed 2826 * touched; otherwise save the computed one */
87367d5f
KW
2827 if (! invlist_in_swash_is_valid
2828 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2829 {
5c9f4bd2 2830 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2831 {
2832 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2833 }
cc34d8c5
FC
2834 /* We just stole a reference count. */
2835 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2836 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2837 }
87367d5f 2838
dbfdbd26
KW
2839 SvREADONLY_on(swash_invlist);
2840
c41800a8 2841 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2842 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2843 SvREFCNT_dec(retval);
02c85471
FC
2844 if (!swash_invlist_unclaimed)
2845 SvREFCNT_inc_simple_void_NN(swash_invlist);
2846 retval = newRV_noinc(swash_invlist);
87367d5f 2847 }
36eb48b4
KW
2848 }
2849
2c1f00b9
YO
2850 CORE_SWASH_INIT_RETURN(retval);
2851#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2852}
2853
035d37be
JH
2854
2855/* This API is wrong for special case conversions since we may need to
2856 * return several Unicode characters for a single Unicode character
2857 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2858 * the lower-level routine, and it is similarly broken for returning
38684baa 2859 * multiple values. --jhi
b9992569 2860 * For those, you should use S__to_utf8_case() instead */
b0e3252e 2861/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2862
2863/* Note:
2864 * Returns the value of property/mapping C<swash> for the first character
2865 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2866 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2867 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2868 *
2869 * A "swash" is a hash which contains initially the keys/values set up by
2870 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2871 * property for all possible code points. Things are stored in a compact form
2872 * (see utf8_heavy.pl) so that calculation is required to find the actual
2873 * property value for a given code point. As code points are looked up, new
2874 * key/value pairs are added to the hash, so that the calculation doesn't have
2875 * to ever be re-done. Further, each calculation is done, not just for the
2876 * desired one, but for a whole block of code points adjacent to that one.
2877 * For binary properties on ASCII machines, the block is usually for 64 code
2878 * points, starting with a code point evenly divisible by 64. Thus if the
2879 * property value for code point 257 is requested, the code goes out and
2880 * calculates the property values for all 64 code points between 256 and 319,
2881 * and stores these as a single 64-bit long bit vector, called a "swatch",
2882 * under the key for code point 256. The key is the UTF-8 encoding for code
2883 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2884 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2885 * for code point 258 is then requested, this code realizes that it would be
2886 * stored under the key for 256, and would find that value and extract the
2887 * relevant bit, offset from 256.
2888 *
2889 * Non-binary properties are stored in as many bits as necessary to represent
2890 * their values (32 currently, though the code is more general than that), not
fc273927 2891 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
2892 * vector that encompasses the property values for all code points whose UTF-8
2893 * representations are represented by the key. That is, for all code points
2894 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2895 * bytes of that.
680c470c 2896 */
a0ed51b3 2897UV
680c470c 2898Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2899{
ef8f7699 2900 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2901 U32 klen;
2902 U32 off;
9b56a019 2903 STRLEN slen = 0;
7d85a32c 2904 STRLEN needents;
cfd0369c 2905 const U8 *tmps = NULL;
979f2922 2906 SV *swatch;
08fb1ac5 2907 const U8 c = *ptr;
3568d838 2908
7918f24d
NC
2909 PERL_ARGS_ASSERT_SWASH_FETCH;
2910
87367d5f
KW
2911 /* If it really isn't a hash, it isn't really swash; must be an inversion
2912 * list */
2913 if (SvTYPE(hv) != SVt_PVHV) {
2914 return _invlist_contains_cp((SV*)hv,
2915 (do_utf8)
2916 ? valid_utf8_to_uvchr(ptr, NULL)
2917 : c);
2918 }
2919
08fb1ac5
KW
2920 /* We store the values in a "swatch" which is a vec() value in a swash
2921 * hash. Code points 0-255 are a single vec() stored with key length
2922 * (klen) 0. All other code points have a UTF-8 representation
2923 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2924 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2925 * length for them is the length of the encoded char - 1. ptr[klen] is the
2926 * final byte in the sequence representing the character */
2927 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2928 klen = 0;
2929 needents = 256;
2930 off = c;
3568d838 2931 }
08fb1ac5
KW
2932 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2933 klen = 0;
2934 needents = 256;
a62b247b 2935 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2936 }
2937 else {
08fb1ac5
KW
2938 klen = UTF8SKIP(ptr) - 1;
2939
2940 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2941 * the vec is the final byte in the sequence. (In EBCDIC this is
2942 * converted to I8 to get consecutive values.) To help you visualize
2943 * all this:
2944 * Straight 1047 After final byte
2945 * UTF-8 UTF-EBCDIC I8 transform
2946 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2947 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2948 * ...
2949 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2950 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2951 * ...
2952 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2953 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2954 * ...
2955 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2956 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2957 * ...
2958 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2959 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2960 *
2961 * (There are no discontinuities in the elided (...) entries.)
2962 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2963 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2964 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2965 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2966 * index into the vec() swatch (after subtracting 0x80, which we
2967 * actually do with an '&').
2968 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2969 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2970 * dicontinuities which go away by transforming it into I8, and we
2971 * effectively subtract 0xA0 to get the index. */
979f2922 2972 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2973 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2974 }
7d85a32c 2975
a0ed51b3 2976 /*
4a4088c4 2977 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2978 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2979 * it's nothing to sniff at.) Pity we usually come through at least
2980 * two function calls to get here...
2981 *
2982 * NB: this code assumes that swatches are never modified, once generated!
2983 */
2984
3568d838 2985 if (hv == PL_last_swash_hv &&
a0ed51b3 2986 klen == PL_last_swash_klen &&
27da23d5 2987 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2988 {
2989 tmps = PL_last_swash_tmps;
2990 slen = PL_last_swash_slen;
2991 }
2992 else {
2993 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2994 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2995
b0e3252e 2996 /* If not cached, generate it via swatch_get */
979f2922 2997 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2998 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2999 {
3000 if (klen) {
3001 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3002 swatch = swatch_get(swash,
3003 code_point & ~((UV)needents - 1),
3004 needents);
3005 }
3006 else { /* For the first 256 code points, the swatch has a key of
3007 length 0 */
3008 swatch = swatch_get(swash, 0, needents);
3009 }
979f2922 3010
923e4eb5 3011 if (IN_PERL_COMPILETIME)
623e6609 3012 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3013
979f2922 3014 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3015
979f2922
ST
3016 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3017 || (slen << 3) < needents)
5637ef5b
NC
3018 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3019 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
3020 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3021 }
3022
3023 PL_last_swash_hv = hv;
16d8f38a 3024 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3025 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3026 /* FIXME change interpvar.h? */
3027 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3028 PL_last_swash_slen = slen;
3029 if (klen)
3030 Copy(ptr, PL_last_swash_key, klen, U8);
3031 }
3032
9faf8d75 3033 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3034 case 1:
e7aca353 3035 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3036 case 8:
e7aca353 3037 return ((UV) tmps[off]);
a0ed51b3
LW
3038 case 16:
3039 off <<= 1;
e7aca353
JH
3040 return
3041 ((UV) tmps[off ] << 8) +
3042 ((UV) tmps[off + 1]);
a0ed51b3
LW
3043 case 32:
3044 off <<= 2;
e7aca353
JH
3045 return
3046 ((UV) tmps[off ] << 24) +
3047 ((UV) tmps[off + 1] << 16) +
3048 ((UV) tmps[off + 2] << 8) +
3049 ((UV) tmps[off + 3]);
a0ed51b3 3050 }
5637ef5b
NC
3051 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3052 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3053 NORETURN_FUNCTION_END;
a0ed51b3 3054}
2b9d42f0 3055
319009ee
KW
3056/* Read a single line of the main body of the swash input text. These are of
3057 * the form:
3058 * 0053 0056 0073
3059 * where each number is hex. The first two numbers form the minimum and
3060 * maximum of a range, and the third is the value associated with the range.
3061 * Not all swashes should have a third number
3062 *
3063 * On input: l points to the beginning of the line to be examined; it points
3064 * to somewhere in the string of the whole input text, and is
3065 * terminated by a \n or the null string terminator.
3066 * lend points to the null terminator of that string
3067 * wants_value is non-zero if the swash expects a third number
3068 * typestr is the name of the swash's mapping, like 'ToLower'
3069 * On output: *min, *max, and *val are set to the values read from the line.
3070 * returns a pointer just beyond the line examined. If there was no
3071 * valid min number on the line, returns lend+1
3072 */
3073
3074STATIC U8*
3075S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3076 const bool wants_value, const U8* const typestr)
3077{
3078 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3079 STRLEN numlen; /* Length of the number */
02470786
KW
3080 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3081 | PERL_SCAN_DISALLOW_PREFIX
3082 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3083
3084 /* nl points to the next \n in the scan */
3085 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3086
95543e92
KW
3087 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3088
319009ee
KW
3089 /* Get the first number on the line: the range minimum */
3090 numlen = lend - l;
3091 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 3092 *max = *min; /* So can never return without setting max */
319009ee
KW
3093 if (numlen) /* If found a hex number, position past it */
3094 l += numlen;
3095 else if (nl) { /* Else, go handle next line, if any */
3096 return nl + 1; /* 1 is length of "\n" */
3097 }
3098 else { /* Else, no next line */
3099 return lend + 1; /* to LIST's end at which \n is not found */
3100 }
3101
3102 /* The max range value follows, separated by a BLANK */
3103 if (isBLANK(*l)) {
3104 ++l;
02470786
KW
3105 flags = PERL_SCAN_SILENT_ILLDIGIT
3106 | PERL_SCAN_DISALLOW_PREFIX
3107 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3108 numlen = lend - l;
3109 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3110 if (numlen)
3111 l += numlen;
3112 else /* If no value here, it is a single element range */
3113 *max = *min;
3114
3115 /* Non-binary tables have a third entry: what the first element of the
24303724 3116 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3117 if (wants_value) {
3118 if (isBLANK(*l)) {
3119 ++l;
f2a7d0fc
KW
3120 flags = PERL_SCAN_SILENT_ILLDIGIT
3121 | PERL_SCAN_DISALLOW_PREFIX
3122 | PERL_SCAN_SILENT_NON_PORTABLE;
3123 numlen = lend - l;
3124 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3125 if (numlen)
3126 l += numlen;
3127 else
3128 *val = 0;
319009ee
KW
3129 }
3130 else {
3131 *val = 0;
3132 if (typeto) {
dcbac5bb 3133 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3134 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3135 typestr, l);
3136 }
3137 }
3138 }
3139 else
3140 *val = 0; /* bits == 1, then any val should be ignored */
3141 }
3142 else { /* Nothing following range min, should be single element with no
3143 mapping expected */
319009ee
KW
3144 if (wants_value) {
3145 *val = 0;
3146 if (typeto) {
dcbac5bb 3147 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3148 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3149 }
3150 }
3151 else
3152 *val = 0; /* bits == 1, then val should be ignored */
3153 }
3154
3155 /* Position to next line if any, or EOF */
3156 if (nl)
3157 l = nl + 1;
3158 else
3159 l = lend;
3160
3161 return l;
3162}
3163
979f2922
ST
3164/* Note:
3165 * Returns a swatch (a bit vector string) for a code point sequence
3166 * that starts from the value C<start> and comprises the number C<span>.
3167 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3168 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3169 */
3170STATIC SV*
b0e3252e 3171S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3172{
3173 SV *swatch;
77f9f126 3174 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3175 STRLEN lcur, xcur, scur;
ef8f7699 3176 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3177 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3178
88d45d28
KW
3179 SV** listsvp = NULL; /* The string containing the main body of the table */
3180 SV** extssvp = NULL;
3181 SV** invert_it_svp = NULL;
3182 U8* typestr = NULL;
786861f5
KW
3183 STRLEN bits;
3184 STRLEN octets; /* if bits == 1, then octets == 0 */
3185 UV none;
3186 UV end = start + span;
972dd592 3187
36eb48b4 3188 if (invlistsvp == NULL) {
786861f5
KW
3189 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3190 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3191 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3192 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3193 listsvp = hv_fetchs(hv, "LIST", FALSE);
3194 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3195
3196 bits = SvUV(*bitssvp);
3197 none = SvUV(*nonesvp);
3198 typestr = (U8*)SvPV_nolen(*typesvp);
3199 }
36eb48b4
KW
3200 else {
3201 bits = 1;
3202 none = 0;
3203 }
786861f5 3204 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3205
b0e3252e 3206 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3207
979f2922 3208 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3209 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3210 (UV)bits);
979f2922
ST
3211 }
3212
84ea5ef6
KW
3213 /* If overflowed, use the max possible */
3214 if (end < start) {
3215 end = UV_MAX;
3216 span = end - start;
3217 }
3218
979f2922 3219 /* create and initialize $swatch */
979f2922 3220 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3221 swatch = newSV(scur);
3222 SvPOK_on(swatch);
979f2922
ST
3223 s = (U8*)SvPVX(swatch);
3224 if (octets && none) {
0bd48802 3225 const U8* const e = s + scur;
979f2922
ST
3226 while (s < e) {
3227 if (bits == 8)
3228 *s++ = (U8)(none & 0xff);
3229 else if (bits == 16) {
3230 *s++ = (U8)((none >> 8) & 0xff);
3231 *s++ = (U8)( none & 0xff);
3232 }
3233 else if (bits == 32) {
3234 *s++ = (U8)((none >> 24) & 0xff);
3235 *s++ = (U8)((none >> 16) & 0xff);
3236 *s++ = (U8)((none >> 8) & 0xff);
3237 *s++ = (U8)( none & 0xff);
3238 }
3239 }
3240 *s = '\0';
3241 }
3242 else {
3243 (void)memzero((U8*)s, scur + 1);
3244 }
3245 SvCUR_set(swatch, scur);
3246 s = (U8*)SvPVX(swatch);
3247
36eb48b4
KW
3248 if (invlistsvp) { /* If has an inversion list set up use that */
3249 _invlist_populate_swatch(*invlistsvp, start, end, s);
3250 return swatch;
3251 }
3252
3253 /* read $swash->{LIST} */
979f2922
ST
3254 l = (U8*)SvPV(*listsvp, lcur);
3255 lend = l + lcur;
3256 while (l < lend) {
8ed25d53 3257 UV min, max, val, upper;
95543e92
KW
3258 l = swash_scan_list_line(l, lend, &min, &max, &val,
3259 cBOOL(octets), typestr);
319009ee 3260 if (l > lend) {
979f2922
ST
3261 break;
3262 }
3263
972dd592 3264 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3265 if (max < start)
3266 continue;
3267
8ed25d53
KW
3268 /* <end> is generally 1 beyond where we want to set things, but at the
3269 * platform's infinity, where we can't go any higher, we want to
3270 * include the code point at <end> */
3271 upper = (max < end)
3272 ? max
3273 : (max != UV_MAX || end != UV_MAX)
3274 ? end - 1
3275 : end;
3276
979f2922 3277 if (octets) {
35da51f7 3278 UV key;
979f2922
ST
3279 if (min < start) {
3280 if (!none || val < none) {
3281 val += start - min;
3282 }
3283 min = start;
3284 }
8ed25d53 3285 for (key = min; key <= upper; key++) {
979f2922 3286 STRLEN offset;
979f2922
ST
3287 /* offset must be non-negative (start <= min <= key < end) */
3288 offset = octets * (key - start);
3289 if (bits == 8)
3290 s[offset] = (U8)(val & 0xff);
3291 else if (bits == 16) {
3292 s[offset ] = (U8)((val >> 8) & 0xff);
3293 s[offset + 1] = (U8)( val & 0xff);
3294 }
3295 else if (bits == 32) {
3296 s[offset ] = (U8)((val >> 24) & 0xff);
3297 s[offset + 1] = (U8)((val >> 16) & 0xff);
3298 s[offset + 2] = (U8)((val >> 8) & 0xff);
3299 s[offset + 3] = (U8)( val & 0xff);
3300 }
3301
3302 if (!none || val < none)
3303 ++val;
3304 }
3305 }
711a919c 3306 else { /* bits == 1, then val should be ignored */
35da51f7 3307 UV key;
979f2922
ST
3308 if (min < start)
3309 min = start;
6cb05c12 3310
8ed25d53 3311 for (key = min; key <= upper; key++) {
0bd48802 3312 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3313 s[offset >> 3] |= 1 << (offset & 7);
3314 }
3315 }
3316 } /* while */
979f2922 3317
9479a769 3318 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3319 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3320
3321 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3322 * be 0, and their inversion should also be 0, as we don't succeed any
3323 * Unicode property matches for non-Unicode code points */
3324 if (start <= PERL_UNICODE_MAX) {
3325
3326 /* The code below assumes that we never cross the
3327 * Unicode/above-Unicode boundary in a range, as otherwise we would
3328 * have to figure out where to stop flipping the bits. Since this
3329 * boundary is divisible by a large power of 2, and swatches comes
3330 * in small powers of 2, this should be a valid assumption */
3331 assert(start + span - 1 <= PERL_UNICODE_MAX);
3332
507a8485
KW
3333 send = s + scur;
3334 while (s < send) {
3335 *s = ~(*s);
3336 s++;
3337 }
0bda3001 3338 }
77f9f126
KW
3339 }
3340
d73c39c5
KW
3341 /* read $swash->{EXTRAS}
3342 * This code also copied to swash_to_invlist() below */
979f2922
ST
3343 x = (U8*)SvPV(*extssvp, xcur);
3344 xend = x + xcur;
3345 while (x < xend) {
3346 STRLEN namelen;
3347 U8 *namestr;
3348 SV** othersvp;
3349 HV* otherhv;
3350 STRLEN otherbits;
3351 SV **otherbitssvp, *other;
711a919c 3352 U8 *s, *o, *nl;
979f2922
ST
3353 STRLEN slen, olen;
3354
35da51f7 3355 const U8 opc = *x++;
979f2922
ST
3356 if (opc == '\n')
3357 continue;
3358
3359 nl = (U8*)memchr(x, '\n', xend - x);
3360
3361 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3362 if (nl) {
3363 x = nl + 1; /* 1 is length of "\n" */
3364 continue;
3365 }
3366 else {
3367 x = xend; /* to EXTRAS' end at which \n is not found */
3368 break;
3369 }
3370 }
3371
3372 namestr = x;
3373 if (nl) {
3374 namelen = nl - namestr;
3375 x = nl + 1;
3376 }
3377 else {
3378 namelen = xend - namestr;
3379 x = xend;
3380 }
3381
3382 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3383 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3384 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3385 otherbits = (STRLEN)SvUV(*otherbitssvp);
3386 if (bits < otherbits)
5637ef5b
NC
3387 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3388 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3389
3390 /* The "other" swatch must be destroyed after. */
b0e3252e 3391 other = swatch_get(*othersvp, start, span);
979f2922
ST
3392 o = (U8*)SvPV(other, olen);
3393
3394 if (!olen)
b0e3252e 3395 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3396
3397 s = (U8*)SvPV(swatch, slen);
3398 if (bits == 1 && otherbits == 1) {
3399 if (slen != olen)
5637ef5b
NC
3400 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3401 "mismatch, slen=%"UVuf", olen=%"UVuf,
3402 (UV)slen, (UV)olen);
979f2922
ST
3403
3404 switch (opc) {
3405 case '+':
3406 while (slen--)
3407 *s++ |= *o++;
3408 break;
3409 case '!':
3410 while (slen--)
3411 *s++ |= ~*o++;
3412 break;
3413 case '-':
3414 while (slen--)
3415 *s++ &= ~*o++;
3416 break;
3417 case '&':
3418 while (slen--)
3419 *s++ &= *o++;
3420 break;
3421 default:
3422 break;
3423 }
3424 }
711a919c 3425 else {
979f2922
ST
3426 STRLEN otheroctets = otherbits >> 3;
3427 STRLEN offset = 0;
35da51f7 3428 U8* const send = s + slen;
979f2922
ST
3429
3430 while (s < send) {
3431 UV otherval = 0;
3432
3433 if (otherbits == 1) {
3434 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3435 ++offset;
3436 }
3437 else {
3438 STRLEN vlen = otheroctets;
3439 otherval = *o++;
3440 while (--vlen) {
3441 otherval <<= 8;
3442 otherval |= *o++;
3443 }
3444 }
3445
711a919c 3446 if (opc == '+' && otherval)
6f207bd3 3447 NOOP; /* replace with otherval */
979f2922
ST
3448 else if (opc == '!' && !otherval)
3449 otherval = 1;
3450 else if (opc == '-' && otherval)
3451 otherval = 0;
3452 else if (opc == '&' && !otherval)
3453 otherval = 0;
3454 else {
711a919c 3455 s += octets; /* no replacement */
979f2922
ST
3456 continue;
3457 }
3458
3459 if (bits == 8)
3460 *s++ = (U8)( otherval & 0xff);
3461 else if (bits == 16) {
3462 *s++ = (U8)((otherval >> 8) & 0xff);
3463 *s++ = (U8)( otherval & 0xff);
3464 }
3465 else if (bits == 32) {
3466 *s++ = (U8)((otherval >> 24) & 0xff);
3467 *s++ = (U8)((otherval >> 16) & 0xff);
3468 *s++ = (U8)((otherval >> 8) & 0xff);
3469 *s++ = (U8)( otherval & 0xff);
3470 }
3471 }
3472 }
3473 sv_free(other); /* through with it! */
3474 } /* while */
3475 return swatch;
3476}
3477
064c021d 3478HV*
4c2e1131 3479Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3480{
3481
79a2a0e8 3482 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3483 * Can't be used on a property that is subject to user override, as it
3484 * relies on the value of SPECIALS in the swash which would be set by
3485 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3486 * for overridden properties
064c021d
KW
3487 *
3488 * Returns a hash which is the inversion and closure of a swash mapping.
3489 * For example, consider the input lines:
3490 * 004B 006B
3491 * 004C 006C
3492 * 212A 006B
3493 *
4a4088c4 3494 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3495 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3496 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3497 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3498 *
538e84ed
KW
3499 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3500 * keys are only code points that are folded-to, so it isn't a full closure.
3501 *
064c021d
KW
3502 * Essentially, for any code point, it gives all the code points that map to
3503 * it, or the list of 'froms' for that point.
3504 *
5662e334
KW
3505 * Currently it ignores any additions or deletions from other swashes,
3506 * looking at just the main body of the swash, and if there are SPECIALS
3507 * in the swash, at that hash
3508 *
3509 * The specials hash can be extra code points, and most likely consists of
3510 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3511 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3512 * However consider this possible input in the specials hash:
3513 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3514 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3515 *
3516 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3517 * currently handle. But it also means that FB05 and FB06 are equivalent in
3518 * a 1-1 mapping which we should handle, and this relationship may not be in
3519 * the main table. Therefore this function examines all the multi-char
74894415
KW
3520 * sequences and adds the 1-1 mappings that come out of that.
3521 *
3522 * XXX This function was originally intended to be multipurpose, but its
3523 * only use is quite likely to remain for constructing the inversion of
3524 * the CaseFolding (//i) property. If it were more general purpose for
3525 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3526 * because certain folds are prohibited under /iaa and /il. As an example,
3527 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3528 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3529 * prohibited, so we would not figure out that they fold to each other.
3530 * Code could be written to automatically figure this out, similar to the
3531 * code that does this for multi-character folds, but this is the only case
3532 * where something like this is ever likely to happen, as all the single
7ee537e6 3533 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3534 * little special code that is compiled only for this Unicode version. This
3535 * is smaller and didn't require much coding time to do. But this makes
3536 * this routine strongly tied to being used just for CaseFolding. If ever
3537 * it should be generalized, this would have to be fixed */
064c021d
KW
3538
3539 U8 *l, *lend;
3540 STRLEN lcur;
3541 HV *const hv = MUTABLE_HV(SvRV(swash));
3542
923b6d4e
KW
3543 /* The string containing the main body of the table. This will have its
3544 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3545 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3546
3547 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3548 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3549 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3550 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3551 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3552 const STRLEN bits = SvUV(*bitssvp);
3553 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3554 const UV none = SvUV(*nonesvp);
5662e334 3555 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3556
3557 HV* ret = newHV();
3558
3559 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3560
3561 /* Must have at least 8 bits to get the mappings */
3562 if (bits != 8 && bits != 16 && bits != 32) {
3563 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3564 (UV)bits);
3565 }
3566
5662e334
KW
3567 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3568 mapping to more than one character */
3569
3570 /* Construct an inverse mapping hash for the specials */
3571 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3572 HV * specials_inverse = newHV();
3573 char *char_from; /* the lhs of the map */
3574 I32 from_len; /* its byte length */
3575 char *char_to; /* the rhs of the map */
3576 I32 to_len; /* its byte length */
3577 SV *sv_to; /* and in a sv */
3578 AV* from_list; /* list of things that map to each 'to' */
3579
3580 hv_iterinit(specials_hv);
3581
4a4088c4
KW
3582 /* The keys are the characters (in UTF-8) that map to the corresponding
3583 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3584 * list. */
3585 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3586 SV** listp;
3587 if (! SvPOK(sv_to)) {
5637ef5b
NC
3588 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3589 "unexpectedly is not a string, flags=%lu",
3590 (unsigned long)SvFLAGS(sv_to));
5662e334 3591 }
4b88fb76 3592 /*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
3593
3594 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3595 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3596 * it. Those strings are all one character long */
3597 if ((listp = hv_fetch(specials_inverse,
3598 SvPVX(sv_to),
3599 SvCUR(sv_to), 0)))
3600 {
3601 from_list = (AV*) *listp;
3602 }
3603 else { /* No entry yet for it: create one */
3604 from_list = newAV();
3605 if (! hv_store(specials_inverse,
3606 SvPVX(sv_to),
3607 SvCUR(sv_to),
3608 (SV*) from_list, 0))
3609 {
3610 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3611 }
3612 }
3613
3614 /* Here have the list associated with this 'to' (perhaps newly
3615 * created and empty). Just add to it. Note that we ASSUME that
3616 * the input is guaranteed to not have duplications, so we don't
3617 * check for that. Duplications just slow down execution time. */
3618 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3619 }
3620
3621 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3622 * it looking for cases like the FB05/FB06 examples above. There would
3623 * be an entry in the hash like
3624 * 'st' => [ FB05, FB06 ]
3625 * In this example we will create two lists that get stored in the
3626 * returned hash, 'ret':
3627 * FB05 => [ FB05, FB06 ]
3628 * FB06 => [ FB05, FB06 ]
3629 *
3630 * Note that there is nothing to do if the array only has one element.
3631 * (In the normal 1-1 case handled below, we don't have to worry about
3632 * two lists, as everything gets tied to the single list that is
3633 * generated for the single character 'to'. But here, we are omitting
3634 * that list, ('st' in the example), so must have multiple lists.) */
3635 while ((from_list = (AV *) hv_iternextsv(specials_inverse,