This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Fix typo in comment, add some comments
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 38static const char cp_above_legal_max[] =
2d212e86 39 "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf"";
760c7c2f 40
114d9c4d 41#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 42
48ef279e 43/*
ccfc67b7 44=head1 Unicode Support
7fefc6c1 45These are various utility functions for manipulating UTF8-encoded
72d33970 46strings. For the uninitiated, this is a method of representing arbitrary
61296642 47Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
48characters in the ASCII range are unmodified, and a zero byte never appears
49within non-zero characters.
166f8a29 50
eaf7a4d2
CS
51=cut
52*/
53
54/*
378516de 55=for apidoc uvoffuni_to_utf8_flags
eebe1485 56
a27992cc 57THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
58Instead, B<Almost all code should use L</uvchr_to_utf8> or
59L</uvchr_to_utf8_flags>>.
a27992cc 60
de69f3af
KW
61This function is like them, but the input is a strict Unicode
62(as opposed to native) code point. Only in very rare circumstances should code
63not be using the native code point.
949cf498 64
efa9cd84 65For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 66
eebe1485
SC
67=cut
68*/
69
8ee1cdcb
KW
70#define HANDLE_UNICODE_SURROGATE(uv, flags) \
71 STMT_START { \
72 if (flags & UNICODE_WARN_SURROGATE) { \
73 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
74 "UTF-16 surrogate U+%04"UVXf, uv); \
75 } \
76 if (flags & UNICODE_DISALLOW_SURROGATE) { \
77 return NULL; \
78 } \
79 } STMT_END;
80
81#define HANDLE_UNICODE_NONCHAR(uv, flags) \
82 STMT_START { \
83 if (flags & UNICODE_WARN_NONCHAR) { \
84 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
85 "Unicode non-character U+%04"UVXf" is not " \
86 "recommended for open interchange", uv); \
87 } \
88 if (flags & UNICODE_DISALLOW_NONCHAR) { \
89 return NULL; \
90 } \
91 } STMT_END;
92
ba6ed43c
KW
93/* Use shorter names internally in this file */
94#define SHIFT UTF_ACCUMULATION_SHIFT
95#undef MARK
96#define MARK UTF_CONTINUATION_MARK
97#define MASK UTF_CONTINUATION_MASK
98
dfe13c55 99U8 *
378516de 100Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 101{
378516de 102 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 103
2d1545e5 104 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 105 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
106 return d;
107 }
facc1dc2 108
3ea68d71 109 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
110 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
111 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
112 return d;
113 }
d9432125 114
ba6ed43c
KW
115 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
116 * below, the 16 is for start bytes E0-EF (which are all the possible ones
117 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
118 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
119 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
120 * 0x800-0xFFFF on ASCII */
121 if (uv < (16 * (1U << (2 * SHIFT)))) {
122 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
123 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
124 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
125
126#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
127 aren't tested here */
128 /* The most likely code points in this range are below the surrogates.
129 * Do an extra test to quickly exclude those. */
130 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
131 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
132 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
133 {
8ee1cdcb
KW
134 HANDLE_UNICODE_NONCHAR(uv, flags);
135 }
136 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
137 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 138 }
ba6ed43c
KW
139 }
140#endif
141 return d;
142 }
143
144 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
145 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
146 * happen starting with 4-byte characters on ASCII platforms. We unify the
147 * code for these with EBCDIC, even though some of them require 5-bytes on
148 * those, because khw believes the code saving is worth the very slight
149 * performance hit on these high EBCDIC code points. */
150
151 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
152 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
153 && ckWARN_d(WARN_DEPRECATED))
154 {
155 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
156 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
157 }
158 if ( (flags & UNICODE_WARN_SUPER)
159 || ( UNICODE_IS_ABOVE_31_BIT(uv)
160 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
161 {
162 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
163
164 /* Choose the more dire applicable warning */
165 (UNICODE_IS_ABOVE_31_BIT(uv))
166 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
167 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
168 uv);
169 }
170 if (flags & UNICODE_DISALLOW_SUPER
171 || ( UNICODE_IS_ABOVE_31_BIT(uv)
172 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
173 {
174 return NULL;
175 }
176 }
ba6ed43c
KW
177 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
178 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 179 }
d9432125 180
ba6ed43c
KW
181 /* Test for and handle 4-byte result. In the test immediately below, the
182 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
183 * characters). The 3 is for 3 continuation bytes; these each contribute
184 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
185 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
186 * 0x1_0000-0x1F_FFFF on ASCII */
187 if (uv < (8 * (1U << (3 * SHIFT)))) {
188 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
189 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
190 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
191 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
192
193#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
194 characters. The end-plane non-characters for EBCDIC were
195 handled just above */
196 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
197 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 198 }
ba6ed43c
KW
199 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
200 HANDLE_UNICODE_SURROGATE(uv, flags);
201 }
202#endif
203
204 return d;
205 }
206
207 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
208 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
209 * format. The unrolled version above turns out to not save all that much
210 * time, and at these high code points (well above the legal Unicode range
211 * on ASCII platforms, and well above anything in common use in EBCDIC),
212 * khw believes that less code outweighs slight performance gains. */
213
d9432125 214 {
5aaebcb3 215 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
216 U8 *p = d+len-1;
217 while (p > d) {
4c8cd605 218 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
219 uv >>= UTF_ACCUMULATION_SHIFT;
220 }
4c8cd605 221 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
222 return d+len;
223 }
a0ed51b3 224}
a5bf80e0 225
646ca15d 226/*
07693fe6
KW
227=for apidoc uvchr_to_utf8
228
bcb1a2d4 229Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 230of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
231C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
232the byte after the end of the new character. In other words,
07693fe6
KW
233
234 d = uvchr_to_utf8(d, uv);
235
236is the recommended wide native character-aware way of saying
237
238 *(d++) = uv;
239
760c7c2f
KW
240This function accepts any UV as input, but very high code points (above
241C<IV_MAX> on the platform) will raise a deprecation warning. This is
242typically 0x7FFF_FFFF in a 32-bit word.
243
244It is possible to forbid or warn on non-Unicode code points, or those that may
245be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 246
07693fe6
KW
247=cut
248*/
249
de69f3af
KW
250/* This is also a macro */
251PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
252
07693fe6
KW
253U8 *
254Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
255{
de69f3af 256 return uvchr_to_utf8(d, uv);
07693fe6
KW
257}
258
de69f3af
KW
259/*
260=for apidoc uvchr_to_utf8_flags
261
262Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 263of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
264C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
265the byte after the end of the new character. In other words,
de69f3af
KW
266
267 d = uvchr_to_utf8_flags(d, uv, flags);
268
269or, in most cases,
270
271 d = uvchr_to_utf8_flags(d, uv, 0);
272
273This is the Unicode-aware way of saying
274
275 *(d++) = uv;
276
760c7c2f
KW
277If C<flags> is 0, this function accepts any UV as input, but very high code
278points (above C<IV_MAX> for the platform) will raise a deprecation warning.
279This is typically 0x7FFF_FFFF in a 32-bit word.
280
281Specifying C<flags> can further restrict what is allowed and not warned on, as
282follows:
de69f3af 283
796b6530 284If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
285the function will raise a warning, provided UTF8 warnings are enabled. If
286instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
287NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 288
760c7c2f
KW
289Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
290affect how the function handles a Unicode non-character.
93e6dbd6 291
760c7c2f
KW
292And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
293affect the handling of code points that are above the Unicode maximum of
2940x10FFFF. Languages other than Perl may not be able to accept files that
295contain these.
93e6dbd6
KW
296
297The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
298the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
299three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
300allowed inputs to the strict UTF-8 traditionally defined by Unicode.
301Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
302C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
303above-Unicode and surrogate flags, but not the non-character ones, as
304defined in
305L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
306See L<perlunicode/Noncharacter code points>.
93e6dbd6 307
ab8e6d41
KW
308Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
309so using them is more problematic than other above-Unicode code points. Perl
310invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
311likely that non-Perl languages will not be able to read files that contain
312these that written by the perl interpreter; nor would Perl understand files
313written by something that uses a different extension. For these reasons, there
314is a separate set of flags that can warn and/or disallow these extremely high
315code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
316C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
317are entirely independent from the deprecation warning for code points above
318C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
319code point that needs more than 31 bits to represent. When that happens,
320effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
32132-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
322above-Unicode code points, including these, as malformations; and
323C<UNICODE_WARN_SUPER> warns on these.)
324
325On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
326extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
327than on ASCII. Prior to that, code points 2**31 and higher were simply
328unrepresentable, and a different, incompatible method was used to represent
329code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
330and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
331platforms, warning and disallowing 2**31 and higher.
de69f3af 332
de69f3af
KW
333=cut
334*/
335
336/* This is also a macro */
337PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
338
07693fe6
KW
339U8 *
340Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
341{
de69f3af 342 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
343}
344
83dc0f42
KW
345PERL_STATIC_INLINE bool
346S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
347{
348 /* Returns TRUE if the first code point represented by the Perl-extended-
349 * UTF-8-encoded string starting at 's', and looking no further than 'e -
350 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
351 *
352 * The function handles the case where the input bytes do not include all
353 * the ones necessary to represent a full character. That is, they may be
354 * the intial bytes of the representation of a code point, but possibly
355 * the final ones necessary for the complete representation may be beyond
356 * 'e - 1'.
357 *
358 * The function assumes that the sequence is well-formed UTF-8 as far as it
359 * goes, and is for a UTF-8 variant code point. If the sequence is
360 * incomplete, the function returns FALSE if there is any well-formed
361 * UTF-8 byte sequence that can complete it in such a way that a code point
362 * < 2**31 is produced; otherwise it returns TRUE.
363 *
364 * Getting this exactly right is slightly tricky, and has to be done in
365 * several places in this file, so is centralized here. It is based on the
366 * following table:
367 *
368 * U+7FFFFFFF (2 ** 31 - 1)
369 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
370 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
371 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
372 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
373 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
374 * U+80000000 (2 ** 31):
375 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
376 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
377 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
378 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
379 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
380 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
381 */
382
383#ifdef EBCDIC
384
385 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
386 const U8 * const prefix = "\x41\x41\x41\x41\x41\x41\x42";
387 const STRLEN prefix_len = sizeof(prefix) - 1;
388 const STRLEN len = e - s;
389 const cmp_len = MIN(prefix_len, len - 1);
390
391#else
392
393 PERL_UNUSED_ARG(e);
394
395#endif
396
397 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
398
399 assert(! UTF8_IS_INVARIANT(*s));
400
401#ifndef EBCDIC
402
403 /* Technically, a start byte of FE can be for a code point that fits into
404 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
405 * malformation. */
406 return (*s >= 0xFE);
407
408#else
409
410 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
411 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
412 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
413 * bits. */
414 if (*s != 0xFE || len == 1) {
415 return FALSE;
416 }
417
418 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
419 * \x41 and \x42. */
420 return cBOOL(memGT(s + 1, prefix, cmp_len));
421
422#endif
423
424}
425
07693fe6 426/*
768c67ee 427
35f8c9bd
KW
428A helper function for the macro isUTF8_CHAR(), which should be used instead of
429this function. The macro will handle smaller code points directly saving time,
430using this function as a fall-back for higher code points. This function
431assumes that it is not called with an invariant character, and that
432's + len - 1' is within bounds of the string 's'.
433
434Tests if the string C<s> of at least length 'len' is a valid variant UTF-8
435character. 0 is returned if not, otherwise, 'len' is returned.
436
4dab108f
KW
437It is written in such a way that if 'len' is set to less than a full
438character's length, it will test if the bytes ending there form the legal
439beginning of partial character.
440
35f8c9bd
KW
441*/
442
443STRLEN
444Perl__is_utf8_char_slow(const U8 * const s, const STRLEN len)
445{
446 const U8 *e;
447 const U8 *x, *y;
448
449 PERL_ARGS_ASSERT__IS_UTF8_CHAR_SLOW;
450
451 if (UNLIKELY(! UTF8_IS_START(*s))) {
452 return 0;
453 }
454
455 e = s + len;
456
457 for (x = s + 1; x < e; x++) {
458 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
459 return 0;
460 }
461 }
462
463#ifndef EBCDIC
464
465 /* Here is syntactically valid. Make sure this isn't the start of an
466 * overlong. These values were found by manually inspecting the UTF-8
467 * patterns. See the tables in utf8.h and utfebcdic.h */
468
469 /* This is not needed on modern perls where C0 and C1 are not considered
470 * start bytes. */
471#if 0
472 if (UNLIKELY(*s < 0xC2)) {
473 return 0;
474 }
475#endif
476
477 if (len > 1) {
478 if ( (*s == 0xE0 && UNLIKELY(s[1] < 0xA0))
479 || (*s == 0xF0 && UNLIKELY(s[1] < 0x90))
480 || (*s == 0xF8 && UNLIKELY(s[1] < 0x88))
481 || (*s == 0xFC && UNLIKELY(s[1] < 0x84))
482 || (*s == 0xFE && UNLIKELY(s[1] < 0x82)))
483 {
484 return 0;
485 }
486 if ((len > 6 && UNLIKELY(*s == 0xFF) && UNLIKELY(s[6] < 0x81))) {
487 return 0;
488 }
489 }
490
491#else /* For EBCDIC, we use I8, which is the same on all code pages */
492 {
493 const U8 s0 = NATIVE_UTF8_TO_I8(*s);
494
495 /* On modern perls C0-C4 aren't considered start bytes */
496 if ( /* s0 < 0xC5 || */ s0 == 0xE0) {
497 return 0;
498 }
499
500 if (len >= 1) {
501 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
502
503 if ( (s0 == 0xF0 && UNLIKELY(s1 < 0xB0))
504 || (s0 == 0xF8 && UNLIKELY(s1 < 0xA8))
505 || (s0 == 0xFC && UNLIKELY(s1 < 0xA4))
506 || (s0 == 0xFE && UNLIKELY(s1 < 0x82)))
507 {
508 return 0;
509 }
510 if ((len > 7 && UNLIKELY(s0 == 0xFF) && UNLIKELY(s[7] < 0xA1))) {
511 return 0;
512 }
513 }
514 }
515
516#endif
517
3a498dae
KW
518 /* Finally, see if this would overflow a UV on this platform. See if the
519 * UTF8 for this code point is larger than that for the highest
520 * representable code point. (For ASCII platforms, we could use memcmp()
521 * because we don't have to convert each byte to I8, but it's very rare
522 * input indeed that would approach overflow, so the loop below will likely
523 * only get executed once */
35f8c9bd
KW
524 y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
525
526 for (x = s; x < e; x++, y++) {
527
3a498dae 528 /* If the same as this byte, go on to the next */
35f8c9bd
KW
529 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
530 continue;
531 }
532
533 /* If this is larger, it overflows */
534 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
535 return 0;
536 }
537
538 /* But if smaller, it won't */
539 break;
540 }
541
542 return len;
543}
544
545/*
546
de69f3af 547=for apidoc utf8n_to_uvchr
378516de
KW
548
549THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 550Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 551
9041c2e3 552Bottom level UTF-8 decode routine.
de69f3af 553Returns the native code point value of the first character in the string C<s>,
746afd53
KW
554which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
555C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
556the length, in bytes, of that character.
949cf498
KW
557
558The value of C<flags> determines the behavior when C<s> does not point to a
559well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
560zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
561next possible position in C<s> that could begin a non-malformed character.
562Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
563
564Various ALLOW flags can be set in C<flags> to allow (and not warn on)
565individual types of malformations, such as the sequence being overlong (that
566is, when there is a shorter sequence that can express the same code point;
567overlong sequences are expressly forbidden in the UTF-8 standard due to
568potential security issues). Another malformation example is the first byte of
569a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
570flags. For allowed 0 length strings, this function returns 0; for allowed
571overlong sequences, the computed code point is returned; for all other allowed
572malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
573determinable reasonable value.
949cf498 574
796b6530 575The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
576flags) malformation is found. If this flag is set, the routine assumes that
577the caller will raise a warning, and this function will silently just set
d088425d
KW
578C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
579
75200dff 580Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 581character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
582in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
583be set to 1. To disambiguate, upon a zero return, see if the first byte of
584C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
585error.
949cf498
KW
586
587Certain code points are considered problematic. These are Unicode surrogates,
746afd53 588Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 589By default these are considered regular code points, but certain situations
ecc1615f
KW
590warrant special handling for them, which can be specified using the C<flags>
591parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
592three classes are treated as malformations and handled as such. The flags
593C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
594C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
595disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
596restricts the allowed inputs to the strict UTF-8 traditionally defined by
597Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
598definition given by
599L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
600The difference between traditional strictness and C9 strictness is that the
601latter does not forbid non-character code points. (They are still discouraged,
602however.) For more discussion see L<perlunicode/Noncharacter code points>.
603
604The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
605C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
606C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
607raised for their respective categories, but otherwise the code points are
608considered valid (not malformations). To get a category to both be treated as
609a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 610(But note that warnings are not raised if lexically disabled nor if
796b6530 611C<UTF8_CHECK_ONLY> is also specified.)
949cf498 612
760c7c2f
KW
613It is now deprecated to have very high code points (above C<IV_MAX> on the
614platforms) and this function will raise a deprecation warning for these (unless
615such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
616in a 32-bit word.
ab8e6d41
KW
617
618Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
619so using them is more problematic than other above-Unicode code points. Perl
620invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
621likely that non-Perl languages will not be able to read files that contain
622these that written by the perl interpreter; nor would Perl understand files
623written by something that uses a different extension. For these reasons, there
624is a separate set of flags that can warn and/or disallow these extremely high
625code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
626C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
627are entirely independent from the deprecation warning for code points above
628C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
629code point that needs more than 31 bits to represent. When that happens,
630effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
63132-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
632above-Unicode code points, including these, as malformations; and
633C<UTF8_WARN_SUPER> warns on these.)
634
635On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
636extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
637than on ASCII. Prior to that, code points 2**31 and higher were simply
638unrepresentable, and a different, incompatible method was used to represent
639code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
640and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
641platforms, warning and disallowing 2**31 and higher.
949cf498
KW
642
643All other code points corresponding to Unicode characters, including private
644use and those yet to be assigned, are never considered malformed and never
645warn.
67e989fb 646
37607a96
PK
647=cut
648*/
67e989fb 649
a0ed51b3 650UV
de69f3af 651Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 652{
d4c19fe8 653 const U8 * const s0 = s;
eb83ed87 654 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 655 U8 * send;
eb83ed87
KW
656 UV uv = *s;
657 STRLEN expectlen;
949cf498 658 SV* sv = NULL;
eb83ed87
KW
659 UV outlier_ret = 0; /* return value when input is in error or problematic
660 */
661 UV pack_warn = 0; /* Save result of packWARN() for later */
662 bool unexpected_non_continuation = FALSE;
663 bool overflowed = FALSE;
2f8f112e 664 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 665
eb83ed87 666 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 667
de69f3af 668 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 669
eb83ed87
KW
670 /* The order of malformation tests here is important. We should consume as
671 * few bytes as possible in order to not skip any valid character. This is
672 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
673 * http://unicode.org/reports/tr36 for more discussion as to why. For
674 * example, once we've done a UTF8SKIP, we can tell the expected number of
675 * bytes, and could fail right off the bat if the input parameters indicate
676 * that there are too few available. But it could be that just that first
677 * byte is garbled, and the intended character occupies fewer bytes. If we
678 * blindly assumed that the first byte is correct, and skipped based on
679 * that number, we could skip over a valid input character. So instead, we
680 * always examine the sequence byte-by-byte.
681 *
682 * We also should not consume too few bytes, otherwise someone could inject
683 * things. For example, an input could be deliberately designed to
684 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 685 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
686 * which is actually part of of the overflowing sequence), that could look
687 * legitimate to the caller, which could discard the initial partial
688 * sequence and process the rest, inappropriately */
689
690 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 691 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
692 if (retlen) {
693 *retlen = 0;
694 }
a0dbb045 695
eb83ed87
KW
696 if (flags & UTF8_ALLOW_EMPTY) {
697 return 0;
698 }
699 if (! (flags & UTF8_CHECK_ONLY)) {
700 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
701 }
0c443dc2
JH
702 goto malformed;
703 }
704
eb83ed87
KW
705 expectlen = UTF8SKIP(s);
706
707 /* A well-formed UTF-8 character, as the vast majority of calls to this
708 * function will be for, has this expected length. For efficiency, set
709 * things up here to return it. It will be overriden only in those rare
710 * cases where a malformation is found */
711 if (retlen) {
712 *retlen = expectlen;
713 }
714
715 /* An invariant is trivially well-formed */
1d72bdf6 716 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 717 return uv;
a0ed51b3 718 }
67e989fb 719
eb83ed87 720 /* A continuation character can't start a valid sequence */
b5b9af04 721 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
722 if (flags & UTF8_ALLOW_CONTINUATION) {
723 if (retlen) {
724 *retlen = 1;
725 }
726 return UNICODE_REPLACEMENT;
727 }
ba210ebe 728
eb83ed87
KW
729 if (! (flags & UTF8_CHECK_ONLY)) {
730 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
731 }
732 curlen = 1;
ba210ebe
JH
733 goto malformed;
734 }
9041c2e3 735
dcd27b3c
KW
736 /* Here is not a continuation byte, nor an invariant. The only thing left
737 * is a start byte (possibly for an overlong) */
738
534752c1
KW
739 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
740 * that indicate the number of bytes in the character's whole UTF-8
741 * sequence, leaving just the bits that are part of the value. */
742 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 743
eb83ed87
KW
744 /* Now, loop through the remaining bytes in the character's sequence,
745 * accumulating each into the working value as we go. Be sure to not look
746 * past the end of the input string */
0b8d30e8
KW
747 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
748
eb83ed87 749 for (s = s0 + 1; s < send; s++) {
b5b9af04 750 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
751 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
752
753 /* The original implementors viewed this malformation as more
754 * serious than the others (though I, khw, don't understand
755 * why, since other malformations also give very very wrong
756 * results), so there is no way to turn off checking for it.
757 * Set a flag, but keep going in the loop, so that we absorb
758 * the rest of the bytes that comprise the character. */
759 overflowed = TRUE;
760 overflow_byte = *s; /* Save for warning message's use */
761 }
8850bf83 762 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
763 }
764 else {
765 /* Here, found a non-continuation before processing all expected
766 * bytes. This byte begins a new character, so quit, even if
767 * allowing this malformation. */
768 unexpected_non_continuation = TRUE;
769 break;
770 }
771 } /* End of loop through the character's bytes */
772
773 /* Save how many bytes were actually in the character */
774 curlen = s - s0;
775
776 /* The loop above finds two types of malformations: non-continuation and/or
777 * overflow. The non-continuation malformation is really a too-short
778 * malformation, as it means that the current character ended before it was
779 * expected to (being terminated prematurely by the beginning of the next
780 * character, whereas in the too-short malformation there just are too few
781 * bytes available to hold the character. In both cases, the check below
782 * that we have found the expected number of bytes would fail if executed.)
783 * Thus the non-continuation malformation is really unnecessary, being a
784 * subset of the too-short malformation. But there may be existing
785 * applications that are expecting the non-continuation type, so we retain
786 * it, and return it in preference to the too-short malformation. (If this
787 * code were being written from scratch, the two types might be collapsed
788 * into one.) I, khw, am also giving priority to returning the
789 * non-continuation and too-short malformations over overflow when multiple
790 * ones are present. I don't know of any real reason to prefer one over
791 * the other, except that it seems to me that multiple-byte errors trumps
792 * errors from a single byte */
b5b9af04 793 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
794 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
795 if (! (flags & UTF8_CHECK_ONLY)) {
796 if (curlen == 1) {
797 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
798 }
799 else {
800 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
801 }
802 }
eb83ed87
KW
803 goto malformed;
804 }
805 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
806
807 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
808 * as what the original expectations were. */
809 do_overlong_test = FALSE;
eb83ed87
KW
810 if (retlen) {
811 *retlen = curlen;
812 }
813 }
b5b9af04 814 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
815 if (! (flags & UTF8_ALLOW_SHORT)) {
816 if (! (flags & UTF8_CHECK_ONLY)) {
817 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 818 }
eb83ed87
KW
819 goto malformed;
820 }
821 uv = UNICODE_REPLACEMENT;
2f8f112e 822 do_overlong_test = FALSE;
eb83ed87
KW
823 if (retlen) {
824 *retlen = curlen;
825 }
826 }
827
b5b9af04 828 if (UNLIKELY(overflowed)) {
eb83ed87 829 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 830 goto malformed;
eb83ed87 831 }
eb83ed87 832
2f8f112e 833 if (do_overlong_test
5aaebcb3 834 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
835 && ! (flags & UTF8_ALLOW_LONG))
836 {
eb83ed87
KW
837 /* The overlong malformation has lower precedence than the others.
838 * Note that if this malformation is allowed, we return the actual
839 * value, instead of the replacement character. This is because this
840 * value is actually well-defined. */
841 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 842 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
843 }
844 goto malformed;
845 }
846
1a89bb6c 847 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
848 * problematic code point that is not allowed by the input parameters. */
849 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
850 && ((flags & ( UTF8_DISALLOW_NONCHAR
851 |UTF8_DISALLOW_SURROGATE
852 |UTF8_DISALLOW_SUPER
853 |UTF8_DISALLOW_ABOVE_31_BIT
854 |UTF8_WARN_NONCHAR
855 |UTF8_WARN_SURROGATE
856 |UTF8_WARN_SUPER
857 |UTF8_WARN_ABOVE_31_BIT))
858 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
859 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 860 {
949cf498 861 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
862
863 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
864 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 865 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 866 && ckWARN_d(WARN_SURROGATE))
eb83ed87 867 {
111d382d 868 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 869 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
870 }
871 if (flags & UTF8_DISALLOW_SURROGATE) {
872 goto disallowed;
873 }
874 }
949cf498 875 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 876 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 877 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 878 {
0bcdd8f6
KW
879 sv = sv_2mortal(Perl_newSVpvf(aTHX_
880 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
881 uv));
54f4afef 882 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 883 }
c0236afe
KW
884
885 /* The maximum code point ever specified by a standard was
886 * 2**31 - 1. Anything larger than that is a Perl extension that
887 * very well may not be understood by other applications (including
83dc0f42
KW
888 * earlier perl versions on EBCDIC platforms). We test for these
889 * after the regular SUPER ones, and before possibly bailing out,
890 * so that the slightly more dire warning will override the regular
891 * one. */
892 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
893 |UTF8_WARN_SUPER
894 |UTF8_DISALLOW_ABOVE_31_BIT))
895 && UNLIKELY(is_utf8_cp_above_31_bits(s0, send)))
ea5ced44 896 {
0bcdd8f6 897 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 898 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 899 && ckWARN_d(WARN_UTF8))
ea5ced44 900 {
0bcdd8f6
KW
901 sv = sv_2mortal(Perl_newSVpvf(aTHX_
902 "Code point 0x%"UVXf" is not Unicode, and not portable",
903 uv));
ea5ced44
KW
904 pack_warn = packWARN(WARN_UTF8);
905 }
d35f2ca5 906 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
907 goto disallowed;
908 }
909 }
c0236afe 910
949cf498
KW
911 if (flags & UTF8_DISALLOW_SUPER) {
912 goto disallowed;
913 }
760c7c2f
KW
914
915 /* The deprecated warning overrides any non-deprecated one */
916 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
917 {
918 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
919 uv, MAX_NON_DEPRECATED_CP));
920 pack_warn = packWARN(WARN_DEPRECATED);
921 }
949cf498 922 }
4190d317
KW
923 else if (UNICODE_IS_NONCHAR(uv)) {
924 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 925 && ckWARN_d(WARN_NONCHAR))
4190d317 926 {
ba707cdc 927 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 928 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
929 }
930 if (flags & UTF8_DISALLOW_NONCHAR) {
931 goto disallowed;
932 }
933 }
949cf498 934
eb83ed87 935 if (sv) {
de69f3af
KW
936 outlier_ret = uv; /* Note we don't bother to convert to native,
937 as all the outlier code points are the same
938 in both ASCII and EBCDIC */
eb83ed87
KW
939 goto do_warn;
940 }
941
949cf498
KW
942 /* Here, this is not considered a malformed character, so drop through
943 * to return it */
a0ed51b3 944 }
ba210ebe 945
de69f3af 946 return UNI_TO_NATIVE(uv);
ba210ebe 947
eb83ed87
KW
948 /* There are three cases which get to beyond this point. In all 3 cases:
949 * <sv> if not null points to a string to print as a warning.
950 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
951 * set.
952 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
953 * This is done by initializing it to 0, and changing it only
954 * for case 1).
955 * The 3 cases are:
956 * 1) The input is valid but problematic, and to be warned about. The
957 * return value is the resultant code point; <*retlen> is set to
958 * <curlen>, the number of bytes that comprise the code point.
959 * <pack_warn> contains the result of packWARN() for the warning
960 * types. The entry point for this case is the label <do_warn>;
961 * 2) The input is a valid code point but disallowed by the parameters to
962 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
963 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
964 * comprise the code point. <pack_warn> contains the result of
965 * packWARN() for the warning types. The entry point for this case is
966 * the label <disallowed>.
967 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
968 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
969 * bytes that comprise the malformation. All such malformations are
970 * assumed to be warning type <utf8>. The entry point for this case
971 * is the label <malformed>.
972 */
949cf498 973
7b52d656 974 malformed:
ba210ebe 975
eb83ed87
KW
976 if (sv && ckWARN_d(WARN_UTF8)) {
977 pack_warn = packWARN(WARN_UTF8);
978 }
979
7b52d656 980 disallowed:
eb83ed87 981
fcc8fcf6 982 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 983 if (retlen)
10edeb5d 984 *retlen = ((STRLEN) -1);
ba210ebe
JH
985 return 0;
986 }
987
7b52d656 988 do_warn:
5b311467 989
eb83ed87
KW
990 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
991 if warnings are to be raised. */
f555bc63 992 const char * const string = SvPVX_const(sv);
a0dbb045 993
f555bc63
KW
994 if (PL_op)
995 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
996 else
997 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
998 }
999
eb83ed87
KW
1000 if (retlen) {
1001 *retlen = curlen;
1002 }
ba210ebe 1003
eb83ed87 1004 return outlier_ret;
a0ed51b3
LW
1005}
1006
8e84507e 1007/*
ec5f19d0
KW
1008=for apidoc utf8_to_uvchr_buf
1009
1010Returns the native code point of the first character in the string C<s> which
1011is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1012C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1013
524080c4
KW
1014If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1015enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1016C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1017(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1018C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1019the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1020See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1021returned.
ec5f19d0 1022
760c7c2f
KW
1023Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1024unless those are turned off.
1025
ec5f19d0 1026=cut
52be2536
KW
1027
1028Also implemented as a macro in utf8.h
1029
ec5f19d0
KW
1030*/
1031
1032
1033UV
1034Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1035{
ec5f19d0
KW
1036 assert(s < send);
1037
1038 return utf8n_to_uvchr(s, send - s, retlen,
1039 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1040}
1041
52be2536
KW
1042/* This is marked as deprecated
1043 *
ec5f19d0
KW
1044=for apidoc utf8_to_uvuni_buf
1045
de69f3af
KW
1046Only in very rare circumstances should code need to be dealing in Unicode
1047(as opposed to native) code points. In those few cases, use
1048C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1049
1050Returns the Unicode (not-native) code point of the first character in the
1051string C<s> which
ec5f19d0
KW
1052is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1053C<retlen> will be set to the length, in bytes, of that character.
1054
524080c4
KW
1055If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1056enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1057NULL) to -1. If those warnings are off, the computed value if well-defined (or
1058the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1059is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1060next possible position in C<s> that could begin a non-malformed character.
de69f3af 1061See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1062
760c7c2f
KW
1063Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1064unless those are turned off.
1065
ec5f19d0
KW
1066=cut
1067*/
1068
1069UV
1070Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1071{
1072 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1073
1074 assert(send > s);
1075
5962d97e
KW
1076 /* Call the low level routine, asking for checks */
1077 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1078}
1079
b76347f2 1080/*
87cea99e 1081=for apidoc utf8_length
b76347f2
JH
1082
1083Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1084Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1085up past C<e>, croaks.
b76347f2
JH
1086
1087=cut
1088*/
1089
1090STRLEN
35a4481c 1091Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1092{
1093 STRLEN len = 0;
1094
7918f24d
NC
1095 PERL_ARGS_ASSERT_UTF8_LENGTH;
1096
8850bf83
JH
1097 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1098 * the bitops (especially ~) can create illegal UTF-8.
1099 * In other words: in Perl UTF-8 is not just for Unicode. */
1100
a3b680e6
AL
1101 if (e < s)
1102 goto warn_and_return;
b76347f2 1103 while (s < e) {
4cbf4130 1104 s += UTF8SKIP(s);
8e91ec7f
AV
1105 len++;
1106 }
1107
1108 if (e != s) {
1109 len--;
1110 warn_and_return:
9b387841
NC
1111 if (PL_op)
1112 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1113 "%s in %s", unees, OP_DESC(PL_op));
1114 else
61a12c31 1115 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1116 }
1117
1118 return len;
1119}
1120
b06226ff 1121/*
fed3ba5d
NC
1122=for apidoc bytes_cmp_utf8
1123
a1433954 1124Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1125sequence of characters (stored as UTF-8)
1126in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1127equal, -1 or -2 if the first string is less than the second string, +1 or +2
1128if the first string is greater than the second string.
1129
1130-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1131longer string. -2 or +2 is returned if
1132there was a difference between characters
fed3ba5d
NC
1133within the strings.
1134
1135=cut
1136*/
1137
1138int
1139Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1140{
1141 const U8 *const bend = b + blen;
1142 const U8 *const uend = u + ulen;
1143
1144 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1145
1146 while (b < bend && u < uend) {
1147 U8 c = *u++;
1148 if (!UTF8_IS_INVARIANT(c)) {
1149 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1150 if (u < uend) {
1151 U8 c1 = *u++;
1152 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1153 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1154 } else {
1155 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1156 "Malformed UTF-8 character "
1157 "(unexpected non-continuation byte 0x%02x"
1158 ", immediately after start byte 0x%02x)"
1159 /* Dear diag.t, it's in the pod. */
1160 "%s%s", c1, c,
1161 PL_op ? " in " : "",
1162 PL_op ? OP_DESC(PL_op) : "");
1163 return -2;
1164 }
1165 } else {
1166 if (PL_op)
1167 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1168 "%s in %s", unees, OP_DESC(PL_op));
1169 else
61a12c31 1170 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1171 return -2; /* Really want to return undef :-) */
1172 }
1173 } else {
1174 return -2;
1175 }
1176 }
1177 if (*b != c) {
1178 return *b < c ? -2 : +2;
1179 }
1180 ++b;
1181 }
1182
1183 if (b == bend && u == uend)
1184 return 0;
1185
1186 return b < bend ? +1 : -1;
1187}
1188
1189/*
87cea99e 1190=for apidoc utf8_to_bytes
6940069f 1191
2bbc8d55 1192Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1193Unlike L</bytes_to_utf8>, this over-writes the original string, and
1194updates C<len> to contain the new length.
67e989fb 1195Returns zero on failure, setting C<len> to -1.
6940069f 1196
a1433954 1197If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1198
6940069f
GS
1199=cut
1200*/
1201
1202U8 *
37607a96 1203Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1204{
d4c19fe8
AL
1205 U8 * const save = s;
1206 U8 * const send = s + *len;
6940069f 1207 U8 *d;
246fae53 1208
7918f24d 1209 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1210 PERL_UNUSED_CONTEXT;
7918f24d 1211
1e54db1a 1212 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1213 while (s < send) {
d59937ca
KW
1214 if (! UTF8_IS_INVARIANT(*s)) {
1215 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1216 *len = ((STRLEN) -1);
1217 return 0;
1218 }
1219 s++;
dcad2880 1220 }
d59937ca 1221 s++;
246fae53 1222 }
dcad2880
JH
1223
1224 d = s = save;
6940069f 1225 while (s < send) {
80e0b38f
KW
1226 U8 c = *s++;
1227 if (! UTF8_IS_INVARIANT(c)) {
1228 /* Then it is two-byte encoded */
a62b247b 1229 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1230 s++;
1231 }
1232 *d++ = c;
6940069f
GS
1233 }
1234 *d = '\0';
246fae53 1235 *len = d - save;
6940069f
GS
1236 return save;
1237}
1238
1239/*
87cea99e 1240=for apidoc bytes_from_utf8
f9a63242 1241
2bbc8d55 1242Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1243Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1244the newly-created string, and updates C<len> to contain the new
1245length. Returns the original string if no conversion occurs, C<len>
72d33970 1246is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 12470 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1248in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1249
37607a96
PK
1250=cut
1251*/
f9a63242
JH
1252
1253U8 *
e1ec3a88 1254Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1255{
f9a63242 1256 U8 *d;
e1ec3a88
AL
1257 const U8 *start = s;
1258 const U8 *send;
f9a63242
JH
1259 I32 count = 0;
1260
7918f24d 1261 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1262 PERL_UNUSED_CONTEXT;
f9a63242 1263 if (!*is_utf8)
73d840c0 1264 return (U8 *)start;
f9a63242 1265
1e54db1a 1266 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1267 for (send = s + *len; s < send;) {
d59937ca
KW
1268 if (! UTF8_IS_INVARIANT(*s)) {
1269 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1270 return (U8 *)start;
d59937ca
KW
1271 }
1272 count++;
1273 s++;
db42d148 1274 }
d59937ca 1275 s++;
f9a63242
JH
1276 }
1277
35da51f7 1278 *is_utf8 = FALSE;
f9a63242 1279
212542aa 1280 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1281 s = start; start = d;
f9a63242
JH
1282 while (s < send) {
1283 U8 c = *s++;
1a91c45d 1284 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1285 /* Then it is two-byte encoded */
a62b247b 1286 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1287 s++;
c4d5f83a
NIS
1288 }
1289 *d++ = c;
f9a63242
JH
1290 }
1291 *d = '\0';
1292 *len = d - start;
73d840c0 1293 return (U8 *)start;
f9a63242
JH
1294}
1295
1296/*
87cea99e 1297=for apidoc bytes_to_utf8
6940069f 1298
ff97e5cf
KW
1299Converts a string C<s> of length C<len> bytes from the native encoding into
1300UTF-8.
6662521e 1301Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1302reflect the new length in bytes.
6940069f 1303
75200dff 1304A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1305
1306If you want to convert to UTF-8 from encodings other than
1307the native (Latin1 or EBCDIC),
a1433954 1308see L</sv_recode_to_utf8>().
c9ada85f 1309
497711e7 1310=cut
6940069f
GS
1311*/
1312
c682ebef
FC
1313/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1314 likewise need duplication. */
1315
6940069f 1316U8*
35a4481c 1317Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1318{
35a4481c 1319 const U8 * const send = s + (*len);
6940069f
GS
1320 U8 *d;
1321 U8 *dst;
7918f24d
NC
1322
1323 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1324 PERL_UNUSED_CONTEXT;
6940069f 1325
212542aa 1326 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1327 dst = d;
1328
1329 while (s < send) {
55d09dc8
KW
1330 append_utf8_from_native_byte(*s, &d);
1331 s++;
6940069f
GS
1332 }
1333 *d = '\0';
6662521e 1334 *len = d-dst;
6940069f
GS
1335 return dst;
1336}
1337
a0ed51b3 1338/*
dea0fc0b 1339 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1340 *
1341 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1342 * We optimize for native, for obvious reasons. */
1343
1344U8*
dea0fc0b 1345Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1346{
dea0fc0b
JH
1347 U8* pend;
1348 U8* dstart = d;
1349
7918f24d
NC
1350 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1351
dea0fc0b 1352 if (bytelen & 1)
f5992bc4 1353 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1354
1355 pend = p + bytelen;
1356
a0ed51b3 1357 while (p < pend) {
dea0fc0b
JH
1358 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1359 p += 2;
2d1545e5 1360 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1361 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1362 continue;
1363 }
56d37426
KW
1364 if (uv <= MAX_UTF8_TWO_BYTE) {
1365 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1366 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1367 continue;
1368 }
46956fad
KW
1369#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1370#define LAST_HIGH_SURROGATE 0xDBFF
1371#define FIRST_LOW_SURROGATE 0xDC00
1372#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1373
1374 /* This assumes that most uses will be in the first Unicode plane, not
1375 * needing surrogates */
1376 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1377 && uv <= UNICODE_SURROGATE_LAST))
1378 {
1379 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1380 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1381 }
1382 else {
01ea242b 1383 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1384 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1385 || UNLIKELY(low > LAST_LOW_SURROGATE))
1386 {
01ea242b 1387 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1388 }
1389 p += 2;
46956fad
KW
1390 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1391 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1392 }
a0ed51b3 1393 }
56d37426
KW
1394#ifdef EBCDIC
1395 d = uvoffuni_to_utf8_flags(d, uv, 0);
1396#else
a0ed51b3 1397 if (uv < 0x10000) {
eb160463
GS
1398 *d++ = (U8)(( uv >> 12) | 0xe0);
1399 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1400 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1401 continue;
1402 }
1403 else {
eb160463
GS
1404 *d++ = (U8)(( uv >> 18) | 0xf0);
1405 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1406 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1407 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1408 continue;
1409 }
56d37426 1410#endif
a0ed51b3 1411 }
dea0fc0b 1412 *newlen = d - dstart;
a0ed51b3
LW
1413 return d;
1414}
1415
1416/* Note: this one is slightly destructive of the source. */
1417
1418U8*
dea0fc0b 1419Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1420{
1421 U8* s = (U8*)p;
d4c19fe8 1422 U8* const send = s + bytelen;
7918f24d
NC
1423
1424 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1425
e0ea5e2d
NC
1426 if (bytelen & 1)
1427 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1428 (UV)bytelen);
1429
a0ed51b3 1430 while (s < send) {
d4c19fe8 1431 const U8 tmp = s[0];
a0ed51b3
LW
1432 s[0] = s[1];
1433 s[1] = tmp;
1434 s += 2;
1435 }
dea0fc0b 1436 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1437}
1438
922e8cb4
KW
1439bool
1440Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1441{
1442 U8 tmpbuf[UTF8_MAXBYTES+1];
1443 uvchr_to_utf8(tmpbuf, c);
1444 return _is_utf8_FOO(classnum, tmpbuf);
1445}
1446
f9ae8fb6
JD
1447/* Internal function so we can deprecate the external one, and call
1448 this one from other deprecated functions in this file */
1449
f2645549
KW
1450bool
1451Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1452{
f2645549 1453 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1454
1455 if (*p == '_')
1456 return TRUE;
f25ce844 1457 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1458}
1459
5092f92a 1460bool
eba68aa0
KW
1461Perl__is_uni_perl_idcont(pTHX_ UV c)
1462{
1463 U8 tmpbuf[UTF8_MAXBYTES+1];
1464 uvchr_to_utf8(tmpbuf, c);
1465 return _is_utf8_perl_idcont(tmpbuf);
1466}
1467
1468bool
f91dcd13
KW
1469Perl__is_uni_perl_idstart(pTHX_ UV c)
1470{
1471 U8 tmpbuf[UTF8_MAXBYTES+1];
1472 uvchr_to_utf8(tmpbuf, c);
1473 return _is_utf8_perl_idstart(tmpbuf);
1474}
1475
3a4c58c9
KW
1476UV
1477Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1478{
1479 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1480 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1481 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1482 * either "SS" or "Ss". Which one to use is passed into the routine in
1483 * 'S_or_s' to avoid a test */
1484
1485 UV converted = toUPPER_LATIN1_MOD(c);
1486
1487 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1488
1489 assert(S_or_s == 'S' || S_or_s == 's');
1490
6f2d5cbc 1491 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1492 characters in this range */
3a4c58c9
KW
1493 *p = (U8) converted;
1494 *lenp = 1;
1495 return converted;
1496 }
1497
1498 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1499 * which it maps to one of them, so as to only have to have one check for
1500 * it in the main case */
1501 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1502 switch (c) {
1503 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1504 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1505 break;
1506 case MICRO_SIGN:
1507 converted = GREEK_CAPITAL_LETTER_MU;
1508 break;
79e064b9
KW
1509#if UNICODE_MAJOR_VERSION > 2 \
1510 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1511 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1512 case LATIN_SMALL_LETTER_SHARP_S:
1513 *(p)++ = 'S';
1514 *p = S_or_s;
1515 *lenp = 2;
1516 return 'S';
79e064b9 1517#endif
3a4c58c9
KW
1518 default:
1519 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1520 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1521 }
1522 }
1523
1524 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1525 *p = UTF8_TWO_BYTE_LO(converted);
1526 *lenp = 2;
1527
1528 return converted;
1529}
1530
50bda2c3
KW
1531/* Call the function to convert a UTF-8 encoded character to the specified case.
1532 * Note that there may be more than one character in the result.
1533 * INP is a pointer to the first byte of the input character
1534 * OUTP will be set to the first byte of the string of changed characters. It
1535 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1536 * LENP will be set to the length in bytes of the string of changed characters
1537 *
1538 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
1539#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
1540#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
1541#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 1542
b9992569
KW
1543/* This additionally has the input parameter 'specials', which if non-zero will
1544 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 1545 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 1546#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 1547
84afefe6
JH
1548UV
1549Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1550{
a1433954
KW
1551 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1552 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1553 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1554 * the changed version may be longer than the original character.
1555 *
1556 * The ordinal of the first character of the changed version is returned
1557 * (but note, as explained above, that there may be more.) */
1558
7918f24d
NC
1559 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1560
3a4c58c9
KW
1561 if (c < 256) {
1562 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1563 }
1564
0ebc6274 1565 uvchr_to_utf8(p, c);
b9992569 1566 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
1567}
1568
84afefe6
JH
1569UV
1570Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1571{
7918f24d
NC
1572 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1573
3a4c58c9
KW
1574 if (c < 256) {
1575 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1576 }
1577
0ebc6274 1578 uvchr_to_utf8(p, c);
b9992569 1579 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
1580}
1581
afc16117 1582STATIC U8
81611534 1583S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1584{
1585 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1586 * those, converting the result to UTF-8. Since the result is always just
a1433954 1587 * one character, we allow <p> to be NULL */
afc16117
KW
1588
1589 U8 converted = toLOWER_LATIN1(c);
1590
1591 if (p != NULL) {
6f2d5cbc 1592 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1593 *p = converted;
1594 *lenp = 1;
1595 }
1596 else {
430c9760
KW
1597 /* Result is known to always be < 256, so can use the EIGHT_BIT
1598 * macros */
1599 *p = UTF8_EIGHT_BIT_HI(converted);
1600 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1601 *lenp = 2;
1602 }
1603 }
1604 return converted;
1605}
1606
84afefe6
JH
1607UV
1608Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1609{
7918f24d
NC
1610 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1611
afc16117
KW
1612 if (c < 256) {
1613 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1614 }
1615
afc16117 1616 uvchr_to_utf8(p, c);
b9992569 1617 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
1618}
1619
84afefe6 1620UV
51910141 1621Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1622{
51910141 1623 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1624 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1625 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1626 *
1627 * Not to be used for locale folds
51910141 1628 */
f673fad4 1629
a1dde8de
KW
1630 UV converted;
1631
1632 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1633 PERL_UNUSED_CONTEXT;
a1dde8de 1634
1ca267a5
KW
1635 assert (! (flags & FOLD_FLAGS_LOCALE));
1636
659a7c2d 1637 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
1638 converted = GREEK_SMALL_LETTER_MU;
1639 }
9b63e895
KW
1640#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1641 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1642 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
1643 else if ( (flags & FOLD_FLAGS_FULL)
1644 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
1645 {
1ca267a5
KW
1646 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1647 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1648 * under those circumstances. */
1649 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1650 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1651 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1652 p, *lenp, U8);
1653 return LATIN_SMALL_LETTER_LONG_S;
1654 }
1655 else {
4f489194
KW
1656 *(p)++ = 's';
1657 *p = 's';
1658 *lenp = 2;
1659 return 's';
1ca267a5 1660 }
a1dde8de 1661 }
9b63e895 1662#endif
a1dde8de
KW
1663 else { /* In this range the fold of all other characters is their lower
1664 case */
1665 converted = toLOWER_LATIN1(c);
1666 }
1667
6f2d5cbc 1668 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1669 *p = (U8) converted;
1670 *lenp = 1;
1671 }
1672 else {
1673 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1674 *p = UTF8_TWO_BYTE_LO(converted);
1675 *lenp = 2;
1676 }
1677
1678 return converted;
1679}
1680
1681UV
31f05a37 1682Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1683{
4b593389 1684
a0270393
KW
1685 /* Not currently externally documented, and subject to change
1686 * <flags> bits meanings:
1687 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1688 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1689 * locale are to be used.
a0270393
KW
1690 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1691 */
4b593389 1692
36bb2ab6 1693 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1694
780fcc9f
KW
1695 if (flags & FOLD_FLAGS_LOCALE) {
1696 /* Treat a UTF-8 locale as not being in locale at all */
1697 if (IN_UTF8_CTYPE_LOCALE) {
1698 flags &= ~FOLD_FLAGS_LOCALE;
1699 }
1700 else {
1701 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1702 goto needs_full_generality;
780fcc9f 1703 }
31f05a37
KW
1704 }
1705
a1dde8de 1706 if (c < 256) {
e7b7ac46 1707 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1708 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1709 }
1710
2f306ab9 1711 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1712 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1713 uvchr_to_utf8(p, c);
b9992569 1714 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
1715 }
1716 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1717 the special flags. */
1718 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1719
1720 needs_full_generality:
a0270393 1721 uvchr_to_utf8(utf8_c, c);
445bf929 1722 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1723 }
84afefe6
JH
1724}
1725
26483009 1726PERL_STATIC_INLINE bool
5141f98e 1727S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1728 const char *const swashname, SV* const invlist)
bde6a22d 1729{
ea317ccb
KW
1730 /* returns a boolean giving whether or not the UTF8-encoded character that
1731 * starts at <p> is in the swash indicated by <swashname>. <swash>
1732 * contains a pointer to where the swash indicated by <swashname>
1733 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1734 * look at <*swash> and only generate a swash if it is not null. <invlist>
1735 * is NULL or an inversion list that defines the swash. If not null, it
1736 * saves time during initialization of the swash.
ea317ccb
KW
1737 *
1738 * Note that it is assumed that the buffer length of <p> is enough to
1739 * contain all the bytes that comprise the character. Thus, <*p> should
1740 * have been checked before this call for mal-formedness enough to assure
1741 * that. */
1742
7918f24d
NC
1743 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1744
492a624f 1745 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1746 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1747 * as far as there being enough bytes available in it to accommodate the
1748 * character without reading beyond the end, and pass that number on to the
1749 * validating routine */
6302f837 1750 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1751 if (ckWARN_d(WARN_UTF8)) {
1752 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1753 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1754 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1755 what the malformation is */
1756 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1757 }
1758 }
1759 return FALSE;
1760 }
87367d5f
KW
1761 if (!*swash) {
1762 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1763 *swash = _core_swash_init("utf8",
1764
1765 /* Only use the name if there is no inversion
1766 * list; otherwise will go out to disk */
1767 (invlist) ? "" : swashname,
1768
1769 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1770 }
28123549 1771
bde6a22d
NC
1772 return swash_fetch(*swash, p, TRUE) != 0;
1773}
1774
1775bool
922e8cb4
KW
1776Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1777{
922e8cb4
KW
1778 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1779
1780 assert(classnum < _FIRST_NON_SWASH_CC);
1781
f25ce844
KW
1782 return is_utf8_common(p,
1783 &PL_utf8_swash_ptrs[classnum],
1784 swash_property_names[classnum],
1785 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1786}
1787
1788bool
f2645549 1789Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1790{
f2645549 1791 SV* invlist = NULL;
7918f24d 1792
f2645549 1793 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1794
f2645549
KW
1795 if (! PL_utf8_perl_idstart) {
1796 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1797 }
60071a22 1798 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1799}
1800
1801bool
f2645549 1802Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1803{
f2645549 1804 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1805
1806 if (*p == '_')
1807 return TRUE;
f25ce844 1808 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1809}
1810
1811bool
eba68aa0
KW
1812Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1813{
b24b43f7 1814 SV* invlist = NULL;
eba68aa0
KW
1815
1816 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1817
b24b43f7
KW
1818 if (! PL_utf8_perl_idcont) {
1819 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1820 }
60071a22 1821 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1822}
1823
eba68aa0 1824bool
f2645549 1825Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1826{
f2645549 1827 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1828
f25ce844 1829 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1830}
1831
1832bool
f2645549 1833Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1834{
f2645549 1835 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1836
f25ce844 1837 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1838}
1839
1840bool
7dbf68d2
KW
1841Perl__is_utf8_mark(pTHX_ const U8 *p)
1842{
7dbf68d2
KW
1843 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1844
f25ce844 1845 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1846}
1847
6b5c0936 1848/*
87cea99e 1849=for apidoc to_utf8_case
6b5c0936 1850
9da1e7cb
KW
1851Instead use the appropriate one of L</toUPPER_utf8>,
1852L</toTITLE_utf8>,
1853L</toLOWER_utf8>,
1854or L</toFOLD_utf8>.
1855
6fae5207 1856C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1857the character that is being converted. This routine assumes that the character
1858at C<p> is well-formed.
6b5c0936 1859
6fae5207
KW
1860C<ustrp> is a pointer to the character buffer to put the
1861conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1862of the result.
1863
6fae5207 1864C<swashp> is a pointer to the swash to use.
6b5c0936 1865
a1433954 1866Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1867and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1868but not always, a multicharacter mapping), is tried first.
6b5c0936 1869
4a8240a3
KW
1870C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1871any special mappings; C<""> means to use the special mappings. Values other
1872than these two are treated as the name of the hash containing the special
1873mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1874
796b6530
KW
1875C<normal> is a string like C<"ToLower"> which means the swash
1876C<%utf8::ToLower>.
0134edef 1877
760c7c2f
KW
1878Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1879unless those are turned off.
1880
0134edef 1881=cut */
6b5c0936 1882
2104c8d9 1883UV
9a957fbc
AL
1884Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1885 SV **swashp, const char *normal, const char *special)
a0ed51b3 1886{
b9992569
KW
1887 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1888
1889 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
1890}
1891
1892 /* change namve uv1 to 'from' */
6a4a25f4 1893STATIC UV
b9992569
KW
1894S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
1895 SV **swashp, const char *normal, const char *special)
1896{
0134edef 1897 STRLEN len = 0;
7918f24d 1898
b9992569 1899 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 1900
36eaa811
KW
1901 /* For code points that don't change case, we already know that the output
1902 * of this function is the unchanged input, so we can skip doing look-ups
1903 * for them. Unfortunately the case-changing code points are scattered
1904 * around. But there are some long consecutive ranges where there are no
1905 * case changing code points. By adding tests, we can eliminate the lookup
1906 * for all the ones in such ranges. This is currently done here only for
1907 * just a few cases where the scripts are in common use in modern commerce
1908 * (and scripts adjacent to those which can be included without additional
1909 * tests). */
1910
1911 if (uv1 >= 0x0590) {
1912 /* This keeps from needing further processing the code points most
1913 * likely to be used in the following non-cased scripts: Hebrew,
1914 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
1915 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
1916 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
1917 if (uv1 < 0x10A0) {
1918 goto cases_to_self;
1919 }
1920
1921 /* The following largish code point ranges also don't have case
1922 * changes, but khw didn't think they warranted extra tests to speed
1923 * them up (which would slightly slow down everything else above them):
1924 * 1100..139F Hangul Jamo, Ethiopic
1925 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
1926 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
1927 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
1928 * Combining Diacritical Marks Extended, Balinese,
1929 * Sundanese, Batak, Lepcha, Ol Chiki
1930 * 2000..206F General Punctuation
1931 */
1932
1933 if (uv1 >= 0x2D30) {
1934
1935 /* This keeps the from needing further processing the code points
1936 * most likely to be used in the following non-cased major scripts:
1937 * CJK, Katakana, Hiragana, plus some less-likely scripts.
1938 *
1939 * (0x2D30 above might have to be changed to 2F00 in the unlikely
1940 * event that Unicode eventually allocates the unused block as of
1941 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
1942 * that the test suite will start having failures to alert you
1943 * should that happen) */
1944 if (uv1 < 0xA640) {
1945 goto cases_to_self;
1946 }
1947
1948 if (uv1 >= 0xAC00) {
1949 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
1950 if (ckWARN_d(WARN_SURROGATE)) {
1951 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1952 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1953 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1954 }
1955 goto cases_to_self;
1956 }
36eaa811
KW
1957
1958 /* AC00..FAFF Catches Hangul syllables and private use, plus
1959 * some others */
1960 if (uv1 < 0xFB00) {
1961 goto cases_to_self;
1962
1963 }
1964
5af9bc97
KW
1965 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
1966 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1967 && ckWARN_d(WARN_DEPRECATED))
1968 {
1969 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1970 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1971 }
1972 if (ckWARN_d(WARN_NON_UNICODE)) {
1973 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1974 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1975 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1976 }
1977 goto cases_to_self;
1978 }
3bfc1e70
KW
1979#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
1980 if (UNLIKELY(uv1
1981 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
1982 {
1983
1984 /* As of this writing, this means we avoid swash creation
1985 * for anything beyond low Plane 1 */
1986 goto cases_to_self;
1987 }
1988#endif
36eaa811
KW
1989 }
1990 }
9ae3ac1a 1991
36eaa811
KW
1992 /* Note that non-characters are perfectly legal, so no warning should
1993 * be given. There are so few of them, that it isn't worth the extra
1994 * tests to avoid swash creation */
9ae3ac1a
KW
1995 }
1996
0134edef 1997 if (!*swashp) /* load on-demand */
5ab9d2ef 1998 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1999
a6f87d8c 2000 if (special) {
0134edef 2001 /* It might be "special" (sometimes, but not always,
2a37f04d 2002 * a multicharacter mapping) */
4a8240a3 2003 HV *hv = NULL;
b08cf34e
JH
2004 SV **svp;
2005
4a8240a3
KW
2006 /* If passed in the specials name, use that; otherwise use any
2007 * given in the swash */
2008 if (*special != '\0') {
2009 hv = get_hv(special, 0);
2010 }
2011 else {
2012 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2013 if (svp) {
2014 hv = MUTABLE_HV(SvRV(*svp));
2015 }
2016 }
2017
176fe009 2018 if (hv
5f560d8a 2019 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2020 && (*svp))
2021 {
cfd0369c 2022 const char *s;
47654450 2023
cfd0369c 2024 s = SvPV_const(*svp, len);
47654450 2025 if (len == 1)
f4cd282c 2026 /* EIGHTBIT */
c80e42f3 2027 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2028 else {
d2dcd0fb 2029 Copy(s, ustrp, len, U8);
29e98929 2030 }
983ffd37 2031 }
0134edef
JH
2032 }
2033
2034 if (!len && *swashp) {
4a4088c4 2035 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2036
0134edef
JH
2037 if (uv2) {
2038 /* It was "normal" (a single character mapping). */
f4cd282c 2039 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2040 }
2041 }
1feea2c7 2042
cbe07460
KW
2043 if (len) {
2044 if (lenp) {
2045 *lenp = len;
2046 }
2047 return valid_utf8_to_uvchr(ustrp, 0);
2048 }
2049
2050 /* Here, there was no mapping defined, which means that the code point maps
2051 * to itself. Return the inputs */
e24dfe9c 2052 cases_to_self:
bfdf22ec 2053 len = UTF8SKIP(p);
ca9fab46
KW
2054 if (p != ustrp) { /* Don't copy onto itself */
2055 Copy(p, ustrp, len, U8);
2056 }
0134edef 2057
2a37f04d
JH
2058 if (lenp)
2059 *lenp = len;
2060
f4cd282c 2061 return uv1;
cbe07460 2062
a0ed51b3
LW
2063}
2064
051a06d4 2065STATIC UV
357aadde 2066S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2067{
4a4088c4 2068 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2069 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2070 * result contains a character that crosses the 255/256 boundary, disallow
2071 * the change, and return the original code point. See L<perlfunc/lc> for
2072 * why;
051a06d4 2073 *
a1433954
KW
2074 * p points to the original string whose case was changed; assumed
2075 * by this routine to be well-formed
051a06d4
KW
2076 * result the code point of the first character in the changed-case string
2077 * ustrp points to the changed-case string (<result> represents its first char)
2078 * lenp points to the length of <ustrp> */
2079
2080 UV original; /* To store the first code point of <p> */
2081
2082 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2083
a4f12ed7 2084 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2085
2086 /* We know immediately if the first character in the string crosses the
2087 * boundary, so can skip */
2088 if (result > 255) {
2089
2090 /* Look at every character in the result; if any cross the
2091 * boundary, the whole thing is disallowed */
2092 U8* s = ustrp + UTF8SKIP(ustrp);
2093 U8* e = ustrp + *lenp;
2094 while (s < e) {
a4f12ed7 2095 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2096 goto bad_crossing;
2097 }
2098 s += UTF8SKIP(s);
2099 }
2100
613abc6d
KW
2101 /* Here, no characters crossed, result is ok as-is, but we warn. */
2102 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2103 return result;
2104 }
2105
7b52d656 2106 bad_crossing:
051a06d4
KW
2107
2108 /* Failed, have to return the original */
4b88fb76 2109 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2110
2111 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2112 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2113 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2114 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2115 OP_DESC(PL_op),
ab0b796c
KW
2116 original,
2117 original);
051a06d4
KW
2118 Copy(p, ustrp, *lenp, char);
2119 return original;
2120}
2121
d3e79532 2122/*
87cea99e 2123=for apidoc to_utf8_upper
d3e79532 2124
1f607577 2125Instead use L</toUPPER_utf8>.
a1433954 2126
d3e79532
JH
2127=cut */
2128
051a06d4 2129/* Not currently externally documented, and subject to change:
31f05a37
KW
2130 * <flags> is set iff iff the rules from the current underlying locale are to
2131 * be used. */
051a06d4 2132
2104c8d9 2133UV
31f05a37 2134Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2135{
051a06d4
KW
2136 UV result;
2137
2138 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2139
780fcc9f
KW
2140 if (flags) {
2141 /* Treat a UTF-8 locale as not being in locale at all */
2142 if (IN_UTF8_CTYPE_LOCALE) {
2143 flags = FALSE;
2144 }
2145 else {
2146 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2147 }
31f05a37
KW
2148 }
2149
3a4c58c9 2150 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2151 if (flags) {
2152 result = toUPPER_LC(*p);
2153 }
2154 else {
81c6c7ce 2155 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2156 }
3a4c58c9
KW
2157 }
2158 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2159 if (flags) {
a62b247b 2160 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2161 result = toUPPER_LC(c);
051a06d4
KW
2162 }
2163 else {
a62b247b 2164 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2165 ustrp, lenp, 'S');
051a06d4
KW
2166 }
2167 }
4a4088c4 2168 else { /* UTF-8, ord above 255 */
b9992569 2169 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2170
2171 if (flags) {
357aadde 2172 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2173 }
2174 return result;
2175 }
2176
4a4088c4 2177 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2178 if (UTF8_IS_INVARIANT(result)) {
2179 *ustrp = (U8) result;
2180 *lenp = 1;
2181 }
2182 else {
62cb07ea
KW
2183 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2184 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2185 *lenp = 2;
3a4c58c9 2186 }
baa60164 2187
051a06d4 2188 return result;
983ffd37 2189}
a0ed51b3 2190
d3e79532 2191/*
87cea99e 2192=for apidoc to_utf8_title
d3e79532 2193
1f607577 2194Instead use L</toTITLE_utf8>.
a1433954 2195
d3e79532
JH
2196=cut */
2197
051a06d4 2198/* Not currently externally documented, and subject to change:
31f05a37
KW
2199 * <flags> is set iff the rules from the current underlying locale are to be
2200 * used. Since titlecase is not defined in POSIX, for other than a
2201 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2202 */
051a06d4 2203
983ffd37 2204UV
31f05a37 2205Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2206{
051a06d4
KW
2207 UV result;
2208
2209 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2210
780fcc9f
KW
2211 if (flags) {
2212 /* Treat a UTF-8 locale as not being in locale at all */
2213 if (IN_UTF8_CTYPE_LOCALE) {
2214 flags = FALSE;
2215 }
2216 else {
2217 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2218 }
31f05a37
KW
2219 }
2220
3a4c58c9 2221 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2222 if (flags) {
2223 result = toUPPER_LC(*p);
2224 }
2225 else {
81c6c7ce 2226 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2227 }
3a4c58c9
KW
2228 }
2229 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2230 if (flags) {
a62b247b 2231 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2232 result = toUPPER_LC(c);
051a06d4
KW
2233 }
2234 else {
a62b247b 2235 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2236 ustrp, lenp, 's');
051a06d4
KW
2237 }
2238 }
4a4088c4 2239 else { /* UTF-8, ord above 255 */
b9992569 2240 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2241
2242 if (flags) {
357aadde 2243 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2244 }
2245 return result;
2246 }
2247
4a4088c4 2248 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2249 if (UTF8_IS_INVARIANT(result)) {
2250 *ustrp = (U8) result;
2251 *lenp = 1;
2252 }
2253 else {
62cb07ea
KW
2254 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2255 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2256 *lenp = 2;
3a4c58c9
KW
2257 }
2258
051a06d4 2259 return result;
a0ed51b3
LW
2260}
2261
d3e79532 2262/*
87cea99e 2263=for apidoc to_utf8_lower
d3e79532 2264
1f607577 2265Instead use L</toLOWER_utf8>.
a1433954 2266
d3e79532
JH
2267=cut */
2268
051a06d4 2269/* Not currently externally documented, and subject to change:
31f05a37
KW
2270 * <flags> is set iff iff the rules from the current underlying locale are to
2271 * be used.
2272 */
051a06d4 2273
2104c8d9 2274UV
31f05a37 2275Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2276{
051a06d4
KW
2277 UV result;
2278
051a06d4 2279 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2280
780fcc9f
KW
2281 if (flags) {
2282 /* Treat a UTF-8 locale as not being in locale at all */
2283 if (IN_UTF8_CTYPE_LOCALE) {
2284 flags = FALSE;
2285 }
2286 else {
2287 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2288 }
31f05a37
KW
2289 }
2290
968c5e6a 2291 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2292 if (flags) {
2293 result = toLOWER_LC(*p);
2294 }
2295 else {
81c6c7ce 2296 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2297 }
968c5e6a
KW
2298 }
2299 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2300 if (flags) {
a62b247b 2301 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2302 result = toLOWER_LC(c);
051a06d4
KW
2303 }
2304 else {
a62b247b 2305 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2306 ustrp, lenp);
051a06d4 2307 }
968c5e6a 2308 }
4a4088c4 2309 else { /* UTF-8, ord above 255 */
b9992569 2310 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2311
2312 if (flags) {
357aadde 2313 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2314 }
968c5e6a 2315
051a06d4
KW
2316 return result;
2317 }
2318
4a4088c4 2319 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2320 if (UTF8_IS_INVARIANT(result)) {
2321 *ustrp = (U8) result;
2322 *lenp = 1;
2323 }
2324 else {
62cb07ea
KW
2325 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2326 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2327 *lenp = 2;
2328 }
2329
051a06d4 2330 return result;
b4e400f9
JH
2331}
2332
d3e79532 2333/*
87cea99e 2334=for apidoc to_utf8_fold
d3e79532 2335
1f607577 2336Instead use L</toFOLD_utf8>.
a1433954 2337
d3e79532
JH
2338=cut */
2339
051a06d4
KW
2340/* Not currently externally documented, and subject to change,
2341 * in <flags>
31f05a37
KW
2342 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2343 * locale are to be used.
051a06d4
KW
2344 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2345 * otherwise simple folds
a0270393
KW
2346 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2347 * prohibited
445bf929 2348 */
36bb2ab6 2349
b4e400f9 2350UV
445bf929 2351Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2352{
051a06d4
KW
2353 UV result;
2354
36bb2ab6 2355 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2356
a0270393
KW
2357 /* These are mutually exclusive */
2358 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2359
50ba90ff
KW
2360 assert(p != ustrp); /* Otherwise overwrites */
2361
780fcc9f
KW
2362 if (flags & FOLD_FLAGS_LOCALE) {
2363 /* Treat a UTF-8 locale as not being in locale at all */
2364 if (IN_UTF8_CTYPE_LOCALE) {
2365 flags &= ~FOLD_FLAGS_LOCALE;
2366 }
2367 else {
2368 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2369 }
31f05a37
KW
2370 }
2371
a1dde8de 2372 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2373 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2374 result = toFOLD_LC(*p);
051a06d4
KW
2375 }
2376 else {
81c6c7ce 2377 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2378 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2379 }
a1dde8de
KW
2380 }
2381 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2382 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2383 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2384 result = toFOLD_LC(c);
051a06d4
KW
2385 }
2386 else {
a62b247b 2387 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2388 ustrp, lenp,
1ca267a5 2389 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2390 }
a1dde8de 2391 }
4a4088c4 2392 else { /* UTF-8, ord above 255 */
b9992569 2393 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2394
1ca267a5
KW
2395 if (flags & FOLD_FLAGS_LOCALE) {
2396
76f2ffcd 2397# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2398 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2399
2400# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2401# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2402
2403 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2404
538e84ed
KW
2405 /* Special case these two characters, as what normally gets
2406 * returned under locale doesn't work */
76f2ffcd
KW
2407 if (UTF8SKIP(p) == cap_sharp_s_len
2408 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2409 {
ab0b796c
KW
2410 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2411 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2412 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2413 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2414 goto return_long_s;
2415 }
0766489e
KW
2416 else
2417#endif
2418 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2419 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2420 {
ab0b796c
KW
2421 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2422 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2423 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2424 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2425 goto return_ligature_st;
2426 }
74894415
KW
2427
2428#if UNICODE_MAJOR_VERSION == 3 \
2429 && UNICODE_DOT_VERSION == 0 \
2430 && UNICODE_DOT_DOT_VERSION == 1
2431# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2432
2433 /* And special case this on this Unicode version only, for the same
2434 * reaons the other two are special cased. They would cross the
2435 * 255/256 boundary which is forbidden under /l, and so the code
2436 * wouldn't catch that they are equivalent (which they are only in
2437 * this release) */
2438 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2439 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2440 {
2441 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2442 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2443 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2444 "resolved to \"\\x{0131}\".");
2445 goto return_dotless_i;
2446 }
2447#endif
2448
357aadde 2449 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2450 }
a0270393
KW
2451 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2452 return result;
2453 }
2454 else {
4a4088c4 2455 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2456 * character above the ASCII range, and the result should not
2457 * contain an ASCII character. */
a0270393
KW
2458
2459 UV original; /* To store the first code point of <p> */
2460
2461 /* Look at every character in the result; if any cross the
2462 * boundary, the whole thing is disallowed */
2463 U8* s = ustrp;
2464 U8* e = ustrp + *lenp;
2465 while (s < e) {
2466 if (isASCII(*s)) {
2467 /* Crossed, have to return the original */
2468 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2469
9fc2026f 2470 /* But in these instances, there is an alternative we can
1ca267a5 2471 * return that is valid */
0766489e
KW
2472 if (original == LATIN_SMALL_LETTER_SHARP_S
2473#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2474 || original == LATIN_CAPITAL_LETTER_SHARP_S
2475#endif
2476 ) {
1ca267a5
KW
2477 goto return_long_s;
2478 }
9fc2026f
KW
2479 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2480 goto return_ligature_st;
2481 }
74894415
KW
2482#if UNICODE_MAJOR_VERSION == 3 \
2483 && UNICODE_DOT_VERSION == 0 \
2484 && UNICODE_DOT_DOT_VERSION == 1
2485
2486 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2487 goto return_dotless_i;
2488 }
2489#endif
a0270393
KW
2490 Copy(p, ustrp, *lenp, char);
2491 return original;
2492 }
2493 s += UTF8SKIP(s);
2494 }
051a06d4 2495
a0270393
KW
2496 /* Here, no characters crossed, result is ok as-is */
2497 return result;
2498 }
051a06d4
KW
2499 }
2500
4a4088c4 2501 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2502 if (UTF8_IS_INVARIANT(result)) {
2503 *ustrp = (U8) result;
2504 *lenp = 1;
2505 }
2506 else {
62cb07ea
KW
2507 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2508 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2509 *lenp = 2;
2510 }
2511
051a06d4 2512 return result;
1ca267a5
KW
2513
2514 return_long_s:
2515 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2516 * folds to a string of two of these characters. By returning this
2517 * instead, then, e.g.,
2518 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2519 * works. */
2520
2521 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2522 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2523 ustrp, *lenp, U8);
2524 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2525
2526 return_ligature_st:
2527 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2528 * have the other one fold to it */
2529
2530 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2531 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2532 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2533
2534#if UNICODE_MAJOR_VERSION == 3 \
2535 && UNICODE_DOT_VERSION == 0 \
2536 && UNICODE_DOT_DOT_VERSION == 1
2537
2538 return_dotless_i:
2539 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2540 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2541 return LATIN_SMALL_LETTER_DOTLESS_I;
2542
2543#endif
2544
a0ed51b3
LW
2545}
2546
711a919c 2547/* Note:
f90a9a02 2548 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2549 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2550 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2551 */
c4a5db0c 2552
a0ed51b3 2553SV*
7fc63493 2554Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2555{
c4a5db0c
KW
2556 PERL_ARGS_ASSERT_SWASH_INIT;
2557
2558 /* Returns a copy of a swash initiated by the called function. This is the
2559 * public interface, and returning a copy prevents others from doing
2560 * mischief on the original */
2561
5d3d13d1 2562 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2563}
2564
2565SV*
5d3d13d1 2566Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2567{
2c1f00b9
YO
2568
2569 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2570 * use the following define */
2571
2572#define CORE_SWASH_INIT_RETURN(x) \
2573 PL_curpm= old_PL_curpm; \
2574 return x
2575
c4a5db0c 2576 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2577 * by calling utf8_heavy.pl in the general case. The returned value may be
2578 * the swash's inversion list instead if the input parameters allow it.
2579 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2580 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2581 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2582 *
2583 * This interface should only be used by functions that won't destroy or
2584 * adversely change the swash, as doing so affects all other uses of the
2585 * swash in the program; the general public should use 'Perl_swash_init'
2586 * instead.
2587 *
2588 * pkg is the name of the package that <name> should be in.
2589 * name is the name of the swash to find. Typically it is a Unicode
2590 * property name, including user-defined ones
2591 * listsv is a string to initialize the swash with. It must be of the form
2592 * documented as the subroutine return value in
2593 * L<perlunicode/User-Defined Character Properties>
2594 * minbits is the number of bits required to represent each data element.
2595 * It is '1' for binary properties.
2596 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2597 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2598 * flags_p if non-NULL is the address of various input and output flag bits
2599 * to the routine, as follows: ('I' means is input to the routine;
2600 * 'O' means output from the routine. Only flags marked O are
2601 * meaningful on return.)
2602 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2603 * came from a user-defined property. (I O)
5d3d13d1
KW
2604 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2605 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2606 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2607 * return of an inversion list instead of a swash hash if this routine
2608 * thinks that would result in faster execution of swash_fetch() later
2609 * on. (I)
9a53f6cf
KW
2610 *
2611 * Thus there are three possible inputs to find the swash: <name>,
2612 * <listsv>, and <invlist>. At least one must be specified. The result
2613 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2614 * actions can intersect, etc. what <name> gives. To avoid going out to
2615 * disk at all, <invlist> should specify completely what the swash should
2616 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2617 *
2618 * <invlist> is only valid for binary properties */
c4a5db0c 2619
2c1f00b9
YO
2620 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2621
c4a5db0c 2622 SV* retval = &PL_sv_undef;
83199d38 2623 HV* swash_hv = NULL;
87367d5f
KW
2624 const int invlist_swash_boundary =
2625 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2626 ? 512 /* Based on some benchmarking, but not extensive, see commit
2627 message */
2628 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2629
2630 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2631 assert(! invlist || minbits == 1);
2632
2c1f00b9
YO
2633 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2634 that triggered the swash init and the swash init perl logic itself.
2635 See perl #122747 */
2636
9a53f6cf
KW
2637 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2638 * so */
2639 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2640 dSP;
2641 const size_t pkg_len = strlen(pkg);
2642 const size_t name_len = strlen(name);
2643 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2644 SV* errsv_save;
2645 GV *method;
2646
2647 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2648
2649 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2650 ENTER;
69794297 2651 SAVEHINTS();
2782061f 2652 save_re_context();
650f067c
JL
2653 /* We might get here via a subroutine signature which uses a utf8
2654 * parameter name, at which point PL_subname will have been set
2655 * but not yet used. */
2656 save_item(PL_subname);
69794297
KW
2657 if (PL_parser && PL_parser->error_count)
2658 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2659 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2660 if (!method) { /* demand load UTF-8 */
69794297 2661 ENTER;
db2c6cb3
FC
2662 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2663 GvSV(PL_errgv) = NULL;
1a419e6b 2664#ifndef NO_TAINT_SUPPORT
69794297
KW
2665 /* It is assumed that callers of this routine are not passing in
2666 * any user derived data. */
2782061f
DM
2667 /* Need to do this after save_re_context() as it will set
2668 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2669 * in Perl_magic_get). Even line to create errsv_save can turn on
2670 * PL_tainted. */
284167a5
SM
2671 SAVEBOOL(TAINT_get);
2672 TAINT_NOT;
2673#endif
69794297
KW
2674 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2675 NULL);
eed484f9 2676 {
db2c6cb3
FC
2677 /* Not ERRSV, as there is no need to vivify a scalar we are
2678 about to discard. */
2679 SV * const errsv = GvSV(PL_errgv);
2680 if (!SvTRUE(errsv)) {
2681 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2682 SvREFCNT_dec(errsv);
2683 }
eed484f9 2684 }
69794297
KW
2685 LEAVE;
2686 }
2687 SPAGAIN;
2688 PUSHMARK(SP);
2689 EXTEND(SP,5);
2690 mPUSHp(pkg, pkg_len);
2691 mPUSHp(name, name_len);
2692 PUSHs(listsv);
2693 mPUSHi(minbits);
2694 mPUSHi(none);
2695 PUTBACK;
db2c6cb3
FC
2696 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2697 GvSV(PL_errgv) = NULL;
69794297
KW
2698 /* If we already have a pointer to the method, no need to use
2699 * call_method() to repeat the lookup. */
c41800a8
KW
2700 if (method
2701 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2702 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2703 {
2704 retval = *PL_stack_sp--;
2705 SvREFCNT_inc(retval);
2706 }
eed484f9 2707 {
db2c6cb3
FC
2708 /* Not ERRSV. See above. */
2709 SV * const errsv = GvSV(PL_errgv);
2710 if (!SvTRUE(errsv)) {
2711 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2712 SvREFCNT_dec(errsv);
2713 }
eed484f9 2714 }
ce3b816e 2715 LEAVE;
69794297
KW
2716 POPSTACK;
2717 if (IN_PERL_COMPILETIME) {
2718 CopHINTS_set(PL_curcop, PL_hints);
2719 }
2720 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 2721 if (SvPOK(retval)) {
69794297
KW
2722
2723 /* If caller wants to handle missing properties, let them */
5d3d13d1 2724 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2725 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2726 }
2727 Perl_croak(aTHX_
2728 "Can't find Unicode property definition \"%"SVf"\"",
2729 SVfARG(retval));
a25b5927 2730 NOT_REACHED; /* NOTREACHED */
d95e4a00 2731 }
69794297 2732 }
9a53f6cf 2733 } /* End of calling the module to find the swash */
36eb48b4 2734
83199d38
KW
2735 /* If this operation fetched a swash, and we will need it later, get it */
2736 if (retval != &PL_sv_undef
2737 && (minbits == 1 || (flags_p
2738 && ! (*flags_p
2739 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2740 {
2741 swash_hv = MUTABLE_HV(SvRV(retval));
2742
2743 /* If we don't already know that there is a user-defined component to
2744 * this swash, and the user has indicated they wish to know if there is
2745 * one (by passing <flags_p>), find out */
2746 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2747 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2748 if (user_defined && SvUV(*user_defined)) {
2749 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2750 }
2751 }
2752 }
2753
36eb48b4
KW
2754 /* Make sure there is an inversion list for binary properties */
2755 if (minbits == 1) {
2756 SV** swash_invlistsvp = NULL;
2757 SV* swash_invlist = NULL;
9a53f6cf 2758 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2759 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2760 an unclaimed reference count */
36eb48b4 2761
9a53f6cf 2762 /* If this operation fetched a swash, get its already existing
83199d38 2763 * inversion list, or create one for it */
36eb48b4 2764
83199d38 2765 if (swash_hv) {
5c9f4bd2 2766 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2767 if (swash_invlistsvp) {
2768 swash_invlist = *swash_invlistsvp;
2769 invlist_in_swash_is_valid = TRUE;
2770 }
2771 else {
36eb48b4 2772 swash_invlist = _swash_to_invlist(retval);
02c85471 2773 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2774 }
2775 }
2776
2777 /* If an inversion list was passed in, have to include it */
2778 if (invlist) {
2779
2780 /* Any fetched swash will by now have an inversion list in it;
2781 * otherwise <swash_invlist> will be NULL, indicating that we
2782 * didn't fetch a swash */
2783 if (swash_invlist) {
2784
2785 /* Add the passed-in inversion list, which invalidates the one
2786 * already stored in the swash */
2787 invlist_in_swash_is_valid = FALSE;
2788 _invlist_union(invlist, swash_invlist, &swash_invlist);
2789 }
2790 else {
2791
87367d5f
KW
2792 /* Here, there is no swash already. Set up a minimal one, if
2793 * we are going to return a swash */
2794 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2795 swash_hv = newHV();
4aca0fe6 2796 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2797 }
9a53f6cf
KW
2798 swash_invlist = invlist;
2799 }
9a53f6cf
KW
2800 }
2801
2802 /* Here, we have computed the union of all the passed-in data. It may
2803 * be that there was an inversion list in the swash which didn't get
538e84ed 2804 * touched; otherwise save the computed one */
87367d5f
KW
2805 if (! invlist_in_swash_is_valid
2806 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2807 {
5c9f4bd2 2808 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2809 {
2810 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2811 }
cc34d8c5
FC
2812 /* We just stole a reference count. */
2813 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2814 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2815 }
87367d5f 2816
dbfdbd26
KW
2817 SvREADONLY_on(swash_invlist);
2818
c41800a8 2819 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2820 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2821 SvREFCNT_dec(retval);
02c85471
FC
2822 if (!swash_invlist_unclaimed)
2823 SvREFCNT_inc_simple_void_NN(swash_invlist);
2824 retval = newRV_noinc(swash_invlist);
87367d5f 2825 }
36eb48b4
KW
2826 }
2827
2c1f00b9
YO
2828 CORE_SWASH_INIT_RETURN(retval);
2829#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2830}
2831
035d37be
JH
2832
2833/* This API is wrong for special case conversions since we may need to
2834 * return several Unicode characters for a single Unicode character
2835 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2836 * the lower-level routine, and it is similarly broken for returning
38684baa 2837 * multiple values. --jhi
b9992569 2838 * For those, you should use S__to_utf8_case() instead */
b0e3252e 2839/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2840
2841/* Note:
2842 * Returns the value of property/mapping C<swash> for the first character
2843 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2844 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2845 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2846 *
2847 * A "swash" is a hash which contains initially the keys/values set up by
2848 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2849 * property for all possible code points. Things are stored in a compact form
2850 * (see utf8_heavy.pl) so that calculation is required to find the actual
2851 * property value for a given code point. As code points are looked up, new
2852 * key/value pairs are added to the hash, so that the calculation doesn't have
2853 * to ever be re-done. Further, each calculation is done, not just for the
2854 * desired one, but for a whole block of code points adjacent to that one.
2855 * For binary properties on ASCII machines, the block is usually for 64 code
2856 * points, starting with a code point evenly divisible by 64. Thus if the
2857 * property value for code point 257 is requested, the code goes out and
2858 * calculates the property values for all 64 code points between 256 and 319,
2859 * and stores these as a single 64-bit long bit vector, called a "swatch",
2860 * under the key for code point 256. The key is the UTF-8 encoding for code
2861 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2862 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2863 * for code point 258 is then requested, this code realizes that it would be
2864 * stored under the key for 256, and would find that value and extract the
2865 * relevant bit, offset from 256.
2866 *
2867 * Non-binary properties are stored in as many bits as necessary to represent
2868 * their values (32 currently, though the code is more general than that), not
fc273927 2869 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
2870 * vector that encompasses the property values for all code points whose UTF-8
2871 * representations are represented by the key. That is, for all code points
2872 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2873 * bytes of that.
680c470c 2874 */
a0ed51b3 2875UV
680c470c 2876Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2877{
ef8f7699 2878 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2879 U32 klen;
2880 U32 off;
9b56a019 2881 STRLEN slen = 0;
7d85a32c 2882 STRLEN needents;
cfd0369c 2883 const U8 *tmps = NULL;
979f2922 2884 SV *swatch;
08fb1ac5 2885 const U8 c = *ptr;
3568d838 2886
7918f24d
NC
2887 PERL_ARGS_ASSERT_SWASH_FETCH;
2888
87367d5f
KW
2889 /* If it really isn't a hash, it isn't really swash; must be an inversion
2890 * list */
2891 if (SvTYPE(hv) != SVt_PVHV) {
2892 return _invlist_contains_cp((SV*)hv,
2893 (do_utf8)
2894 ? valid_utf8_to_uvchr(ptr, NULL)
2895 : c);
2896 }
2897
08fb1ac5
KW
2898 /* We store the values in a "swatch" which is a vec() value in a swash
2899 * hash. Code points 0-255 are a single vec() stored with key length
2900 * (klen) 0. All other code points have a UTF-8 representation
2901 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2902 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2903 * length for them is the length of the encoded char - 1. ptr[klen] is the
2904 * final byte in the sequence representing the character */
2905 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2906 klen = 0;
2907 needents = 256;
2908 off = c;
3568d838 2909 }
08fb1ac5
KW
2910 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2911 klen = 0;
2912 needents = 256;
a62b247b 2913 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2914 }
2915 else {
08fb1ac5
KW
2916 klen = UTF8SKIP(ptr) - 1;
2917
2918 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2919 * the vec is the final byte in the sequence. (In EBCDIC this is
2920 * converted to I8 to get consecutive values.) To help you visualize
2921 * all this:
2922 * Straight 1047 After final byte
2923 * UTF-8 UTF-EBCDIC I8 transform
2924 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2925 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2926 * ...
2927 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2928 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2929 * ...
2930 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2931 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2932 * ...
2933 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2934 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2935 * ...
2936 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2937 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2938 *
2939 * (There are no discontinuities in the elided (...) entries.)
2940 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2941 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2942 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2943 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2944 * index into the vec() swatch (after subtracting 0x80, which we
2945 * actually do with an '&').
2946 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2947 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2948 * dicontinuities which go away by transforming it into I8, and we
2949 * effectively subtract 0xA0 to get the index. */
979f2922 2950 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2951 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2952 }
7d85a32c 2953
a0ed51b3 2954 /*
4a4088c4 2955 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2956 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2957 * it's nothing to sniff at.) Pity we usually come through at least
2958 * two function calls to get here...
2959 *
2960 * NB: this code assumes that swatches are never modified, once generated!
2961 */
2962
3568d838 2963 if (hv == PL_last_swash_hv &&
a0ed51b3 2964 klen == PL_last_swash_klen &&
27da23d5 2965 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2966 {
2967 tmps = PL_last_swash_tmps;
2968 slen = PL_last_swash_slen;
2969 }
2970 else {
2971 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2972 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2973
b0e3252e 2974 /* If not cached, generate it via swatch_get */
979f2922 2975 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2976 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2977 {
2978 if (klen) {
2979 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2980 swatch = swatch_get(swash,
2981 code_point & ~((UV)needents - 1),
2982 needents);
2983 }
2984 else { /* For the first 256 code points, the swatch has a key of
2985 length 0 */
2986 swatch = swatch_get(swash, 0, needents);
2987 }
979f2922 2988
923e4eb5 2989 if (IN_PERL_COMPILETIME)
623e6609 2990 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2991
979f2922 2992 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2993
979f2922
ST
2994 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2995 || (slen << 3) < needents)
5637ef5b
NC
2996 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2997 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2998 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2999 }
3000
3001 PL_last_swash_hv = hv;
16d8f38a 3002 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3003 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3004 /* FIXME change interpvar.h? */
3005 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3006 PL_last_swash_slen = slen;
3007 if (klen)
3008 Copy(ptr, PL_last_swash_key, klen, U8);
3009 }
3010
9faf8d75 3011 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3012 case 1:
e7aca353 3013 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3014 case 8:
e7aca353 3015 return ((UV) tmps[off]);
a0ed51b3
LW
3016 case 16:
3017 off <<= 1;
e7aca353
JH
3018 return
3019 ((UV) tmps[off ] << 8) +
3020 ((UV) tmps[off + 1]);
a0ed51b3
LW
3021 case 32:
3022 off <<= 2;
e7aca353
JH
3023 return
3024 ((UV) tmps[off ] << 24) +
3025 ((UV) tmps[off + 1] << 16) +
3026 ((UV) tmps[off + 2] << 8) +
3027 ((UV) tmps[off + 3]);
a0ed51b3 3028 }
5637ef5b
NC
3029 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3030 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 3031 NORETURN_FUNCTION_END;
a0ed51b3 3032}
2b9d42f0 3033
319009ee
KW
3034/* Read a single line of the main body of the swash input text. These are of
3035 * the form:
3036 * 0053 0056 0073
3037 * where each number is hex. The first two numbers form the minimum and
3038 * maximum of a range, and the third is the value associated with the range.
3039 * Not all swashes should have a third number
3040 *
3041 * On input: l points to the beginning of the line to be examined; it points
3042 * to somewhere in the string of the whole input text, and is
3043 * terminated by a \n or the null string terminator.
3044 * lend points to the null terminator of that string
3045 * wants_value is non-zero if the swash expects a third number
3046 * typestr is the name of the swash's mapping, like 'ToLower'
3047 * On output: *min, *max, and *val are set to the values read from the line.
3048 * returns a pointer just beyond the line examined. If there was no
3049 * valid min number on the line, returns lend+1
3050 */
3051
3052STATIC U8*
3053S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3054 const bool wants_value, const U8* const typestr)
3055{
3056 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3057 STRLEN numlen; /* Length of the number */
02470786
KW
3058 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3059 | PERL_SCAN_DISALLOW_PREFIX
3060 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3061
3062 /* nl points to the next \n in the scan */
3063 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3064
95543e92
KW
3065 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3066
319009ee
KW
3067 /* Get the first number on the line: the range minimum */
3068 numlen = lend - l;
3069 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 3070 *max = *min; /* So can never return without setting max */
319009ee
KW
3071 if (numlen) /* If found a hex number, position past it */
3072 l += numlen;
3073 else if (nl) { /* Else, go handle next line, if any */
3074 return nl + 1; /* 1 is length of "\n" */
3075 }
3076 else { /* Else, no next line */
3077 return lend + 1; /* to LIST's end at which \n is not found */
3078 }
3079
3080 /* The max range value follows, separated by a BLANK */
3081 if (isBLANK(*l)) {
3082 ++l;
02470786
KW
3083 flags = PERL_SCAN_SILENT_ILLDIGIT
3084 | PERL_SCAN_DISALLOW_PREFIX
3085 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3086 numlen = lend - l;
3087 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3088 if (numlen)
3089 l += numlen;
3090 else /* If no value here, it is a single element range */
3091 *max = *min;
3092
3093 /* Non-binary tables have a third entry: what the first element of the
24303724 3094 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3095 if (wants_value) {
3096 if (isBLANK(*l)) {
3097 ++l;
f2a7d0fc
KW
3098 flags = PERL_SCAN_SILENT_ILLDIGIT
3099 | PERL_SCAN_DISALLOW_PREFIX
3100 | PERL_SCAN_SILENT_NON_PORTABLE;
3101 numlen = lend - l;
3102 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3103 if (numlen)
3104 l += numlen;
3105 else
3106 *val = 0;
319009ee
KW
3107 }
3108 else {
3109 *val = 0;
3110 if (typeto) {
dcbac5bb 3111 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3112 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3113 typestr, l);
3114 }
3115 }
3116 }
3117 else
3118 *val = 0; /* bits == 1, then any val should be ignored */
3119 }
3120 else { /* Nothing following range min, should be single element with no
3121 mapping expected */
319009ee
KW
3122 if (wants_value) {
3123 *val = 0;
3124 if (typeto) {
dcbac5bb 3125 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3126 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3127 }
3128 }
3129 else
3130 *val = 0; /* bits == 1, then val should be ignored */
3131 }
3132
3133 /* Position to next line if any, or EOF */
3134 if (nl)
3135 l = nl + 1;
3136 else
3137 l = lend;
3138
3139 return l;
3140}
3141
979f2922
ST
3142/* Note:
3143 * Returns a swatch (a bit vector string) for a code point sequence
3144 * that starts from the value C<start> and comprises the number C<span>.
3145 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3146 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3147 */
3148STATIC SV*
b0e3252e 3149S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3150{
3151 SV *swatch;
77f9f126 3152 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3153 STRLEN lcur, xcur, scur;
ef8f7699 3154 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3155 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3156
88d45d28
KW
3157 SV** listsvp = NULL; /* The string containing the main body of the table */
3158 SV** extssvp = NULL;
3159 SV** invert_it_svp = NULL;
3160 U8* typestr = NULL;
786861f5
KW
3161 STRLEN bits;
3162 STRLEN octets; /* if bits == 1, then octets == 0 */
3163 UV none;
3164 UV end = start + span;
972dd592 3165
36eb48b4 3166 if (invlistsvp == NULL) {
786861f5
KW
3167 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3168 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3169 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3170 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3171 listsvp = hv_fetchs(hv, "LIST", FALSE);
3172 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3173
3174 bits = SvUV(*bitssvp);
3175 none = SvUV(*nonesvp);
3176 typestr = (U8*)SvPV_nolen(*typesvp);
3177 }
36eb48b4
KW
3178 else {
3179 bits = 1;
3180 none = 0;
3181 }
786861f5 3182 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3183
b0e3252e 3184 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3185
979f2922 3186 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3187 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3188 (UV)bits);
979f2922
ST
3189 }
3190
84ea5ef6
KW
3191 /* If overflowed, use the max possible */
3192 if (end < start) {
3193 end = UV_MAX;
3194 span = end - start;
3195 }
3196
979f2922 3197 /* create and initialize $swatch */
979f2922 3198 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3199 swatch = newSV(scur);
3200 SvPOK_on(swatch);
979f2922
ST
3201 s = (U8*)SvPVX(swatch);
3202 if (octets && none) {
0bd48802 3203 const U8* const e = s + scur;
979f2922
ST
3204 while (s < e) {
3205 if (bits == 8)
3206 *s++ = (U8)(none & 0xff);
3207 else if (bits == 16) {
3208 *s++ = (U8)((none >> 8) & 0xff);
3209 *s++ = (U8)( none & 0xff);
3210 }
3211 else if (bits == 32) {
3212 *s++ = (U8)((none >> 24) & 0xff);
3213 *s++ = (U8)((none >> 16) & 0xff);
3214 *s++ = (U8)((none >> 8) & 0xff);
3215 *s++ = (U8)( none & 0xff);
3216 }
3217 }
3218 *s = '\0';
3219 }
3220 else {
3221 (void)memzero((U8*)s, scur + 1);
3222 }
3223 SvCUR_set(swatch, scur);
3224 s = (U8*)SvPVX(swatch);
3225
36eb48b4
KW
3226 if (invlistsvp) { /* If has an inversion list set up use that */
3227 _invlist_populate_swatch(*invlistsvp, start, end, s);
3228 return swatch;
3229 }
3230
3231 /* read $swash->{LIST} */
979f2922
ST
3232 l = (U8*)SvPV(*listsvp, lcur);
3233 lend = l + lcur;
3234 while (l < lend) {
8ed25d53 3235 UV min, max, val, upper;
95543e92
KW
3236 l = swash_scan_list_line(l, lend, &min, &max, &val,
3237 cBOOL(octets), typestr);
319009ee 3238 if (l > lend) {
979f2922
ST
3239 break;
3240 }
3241
972dd592 3242 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3243 if (max < start)
3244 continue;
3245
8ed25d53
KW
3246 /* <end> is generally 1 beyond where we want to set things, but at the
3247 * platform's infinity, where we can't go any higher, we want to
3248 * include the code point at <end> */
3249 upper = (max < end)
3250 ? max
3251 : (max != UV_MAX || end != UV_MAX)
3252 ? end - 1
3253 : end;
3254
979f2922 3255 if (octets) {
35da51f7 3256 UV key;
979f2922
ST
3257 if (min < start) {
3258 if (!none || val < none) {
3259 val += start - min;
3260 }
3261 min = start;
3262 }
8ed25d53 3263 for (key = min; key <= upper; key++) {
979f2922 3264 STRLEN offset;
979f2922
ST
3265 /* offset must be non-negative (start <= min <= key < end) */
3266 offset = octets * (key - start);
3267 if (bits == 8)
3268 s[offset] = (U8)(val & 0xff);
3269 else if (bits == 16) {
3270 s[offset ] = (U8)((val >> 8) & 0xff);
3271 s[offset + 1] = (U8)( val & 0xff);
3272 }
3273 else if (bits == 32) {
3274 s[offset ] = (U8)((val >> 24) & 0xff);
3275 s[offset + 1] = (U8)((val >> 16) & 0xff);
3276 s[offset + 2] = (U8)((val >> 8) & 0xff);
3277 s[offset + 3] = (U8)( val & 0xff);
3278 }
3279
3280 if (!none || val < none)
3281 ++val;
3282 }
3283 }
711a919c 3284 else { /* bits == 1, then val should be ignored */
35da51f7 3285 UV key;
979f2922
ST
3286 if (min < start)
3287 min = start;
6cb05c12 3288
8ed25d53 3289 for (key = min; key <= upper; key++) {
0bd48802 3290 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3291 s[offset >> 3] |= 1 << (offset & 7);
3292 }
3293 }
3294 } /* while */
979f2922 3295
9479a769 3296 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3297 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3298
3299 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3300 * be 0, and their inversion should also be 0, as we don't succeed any
3301 * Unicode property matches for non-Unicode code points */
3302 if (start <= PERL_UNICODE_MAX) {
3303
3304 /* The code below assumes that we never cross the
3305 * Unicode/above-Unicode boundary in a range, as otherwise we would
3306 * have to figure out where to stop flipping the bits. Since this
3307 * boundary is divisible by a large power of 2, and swatches comes
3308 * in small powers of 2, this should be a valid assumption */
3309 assert(start + span - 1 <= PERL_UNICODE_MAX);
3310
507a8485
KW
3311 send = s + scur;
3312 while (s < send) {
3313 *s = ~(*s);
3314 s++;
3315 }
0bda3001 3316 }
77f9f126
KW
3317 }
3318
d73c39c5
KW
3319 /* read $swash->{EXTRAS}
3320 * This code also copied to swash_to_invlist() below */
979f2922
ST
3321 x = (U8*)SvPV(*extssvp, xcur);
3322 xend = x + xcur;
3323 while (x < xend) {
3324 STRLEN namelen;
3325 U8 *namestr;
3326 SV** othersvp;
3327 HV* otherhv;
3328 STRLEN otherbits;
3329 SV **otherbitssvp, *other;
711a919c 3330 U8 *s, *o, *nl;
979f2922
ST
3331 STRLEN slen, olen;
3332
35da51f7 3333 const U8 opc = *x++;
979f2922
ST
3334 if (opc == '\n')
3335 continue;
3336
3337 nl = (U8*)memchr(x, '\n', xend - x);
3338
3339 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3340 if (nl) {
3341 x = nl + 1; /* 1 is length of "\n" */
3342 continue;
3343 }
3344 else {
3345 x = xend; /* to EXTRAS' end at which \n is not found */
3346 break;
3347 }
3348 }
3349
3350 namestr = x;
3351 if (nl) {
3352 namelen = nl - namestr;
3353 x = nl + 1;
3354 }
3355 else {
3356 namelen = xend - namestr;
3357 x = xend;
3358 }
3359
3360 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3361 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3362 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3363 otherbits = (STRLEN)SvUV(*otherbitssvp);
3364 if (bits < otherbits)
5637ef5b
NC
3365 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3366 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3367
3368 /* The "other" swatch must be destroyed after. */
b0e3252e 3369 other = swatch_get(*othersvp, start, span);
979f2922
ST
3370 o = (U8*)SvPV(other, olen);
3371
3372 if (!olen)
b0e3252e 3373 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3374
3375 s = (U8*)SvPV(swatch, slen);
3376 if (bits == 1 && otherbits == 1) {
3377 if (slen != olen)
5637ef5b
NC
3378 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3379 "mismatch, slen=%"UVuf", olen=%"UVuf,
3380 (UV)slen, (UV)olen);
979f2922
ST
3381
3382 switch (opc) {
3383 case '+':
3384 while (slen--)
3385 *s++ |= *o++;
3386 break;
3387 case '!':
3388 while (slen--)
3389 *s++ |= ~*o++;
3390 break;
3391 case '-':
3392 while (slen--)
3393 *s++ &= ~*o++;
3394 break;
3395 case '&':
3396 while (slen--)
3397 *s++ &= *o++;
3398 break;
3399 default:
3400 break;
3401 }
3402 }
711a919c 3403 else {
979f2922
ST
3404 STRLEN otheroctets = otherbits >> 3;
3405 STRLEN offset = 0;
35da51f7 3406 U8* const send = s + slen;
979f2922
ST
3407
3408 while (s < send) {
3409 UV otherval = 0;
3410
3411 if (otherbits == 1) {
3412 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3413 ++offset;
3414 }
3415 else {
3416 STRLEN vlen = otheroctets;
3417 otherval = *o++;
3418 while (--vlen) {
3419 otherval <<= 8;
3420 otherval |= *o++;
3421 }
3422 }
3423
711a919c 3424 if (opc == '+' && otherval)
6f207bd3 3425 NOOP; /* replace with otherval */
979f2922
ST
3426 else if (opc == '!' && !otherval)
3427 otherval = 1;
3428 else if (opc == '-' && otherval)
3429 otherval = 0;
3430 else if (opc == '&' && !otherval)
3431 otherval = 0;
3432 else {
711a919c 3433 s += octets; /* no replacement */
979f2922
ST
3434 continue;
3435 }
3436
3437 if (bits == 8)
3438 *s++ = (U8)( otherval & 0xff);
3439 else if (bits == 16) {
3440 *s++ = (U8)((otherval >> 8) & 0xff);
3441 *s++ = (U8)( otherval & 0xff);
3442 }
3443 else if (bits == 32) {
3444 *s++ = (U8)((otherval >> 24) & 0xff);
3445 *s++ = (U8)((otherval >> 16) & 0xff);
3446 *s++ = (U8)((otherval >> 8) & 0xff);
3447 *s++ = (U8)( otherval & 0xff);
3448 }
3449 }
3450 }
3451 sv_free(other); /* through with it! */
3452 } /* while */
3453 return swatch;
3454}
3455
064c021d 3456HV*
4c2e1131 3457Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3458{
3459
79a2a0e8 3460 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3461 * Can't be used on a property that is subject to user override, as it
3462 * relies on the value of SPECIALS in the swash which would be set by
3463 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3464 * for overridden properties
064c021d
KW
3465 *
3466 * Returns a hash which is the inversion and closure of a swash mapping.
3467 * For example, consider the input lines:
3468 * 004B 006B
3469 * 004C 006C
3470 * 212A 006B
3471 *
4a4088c4 3472 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3473 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3474 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3475 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3476 *
538e84ed
KW
3477 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3478 * keys are only code points that are folded-to, so it isn't a full closure.
3479 *
064c021d
KW
3480 * Essentially, for any code point, it gives all the code points that map to
3481 * it, or the list of 'froms' for that point.
3482 *
5662e334
KW
3483 * Currently it ignores any additions or deletions from other swashes,
3484 * looking at just the main body of the swash, and if there are SPECIALS
3485 * in the swash, at that hash
3486 *
3487 * The specials hash can be extra code points, and most likely consists of
3488 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3489 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3490 * However consider this possible input in the specials hash:
3491 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3492 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3493 *
3494 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3495 * currently handle. But it also means that FB05 and FB06 are equivalent in
3496 * a 1-1 mapping which we should handle, and this relationship may not be in
3497 * the main table. Therefore this function examines all the multi-char
74894415
KW
3498 * sequences and adds the 1-1 mappings that come out of that.
3499 *
3500 * XXX This function was originally intended to be multipurpose, but its
3501 * only use is quite likely to remain for constructing the inversion of
3502 * the CaseFolding (//i) property. If it were more general purpose for
3503 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3504 * because certain folds are prohibited under /iaa and /il. As an example,
3505 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3506 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3507 * prohibited, so we would not figure out that they fold to each other.
3508 * Code could be written to automatically figure this out, similar to the
3509 * code that does this for multi-character folds, but this is the only case
3510 * where something like this is ever likely to happen, as all the single
7ee537e6 3511 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3512 * little special code that is compiled only for this Unicode version. This
3513 * is smaller and didn't require much coding time to do. But this makes
3514 * this routine strongly tied to being used just for CaseFolding. If ever
3515 * it should be generalized, this would have to be fixed */
064c021d
KW
3516
3517 U8 *l, *lend;
3518 STRLEN lcur;
3519 HV *const hv = MUTABLE_HV(SvRV(swash));
3520
923b6d4e
KW
3521 /* The string containing the main body of the table. This will have its
3522 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3523 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3524
3525 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3526 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3527 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3528 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3529 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3530 const STRLEN bits = SvUV(*bitssvp);
3531 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3532 const UV none = SvUV(*nonesvp);
5662e334 3533 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3534
3535 HV* ret = newHV();
3536
3537 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3538
3539 /* Must have at least 8 bits to get the mappings */
3540 if (bits != 8 && bits != 16 && bits != 32) {
3541 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3542 (UV)bits);
3543 }
3544
5662e334
KW
3545 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3546 mapping to more than one character */
3547
3548 /* Construct an inverse mapping hash for the specials */
3549 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3550 HV * specials_inverse = newHV();
3551 char *char_from; /* the lhs of the map */
3552 I32 from_len; /* its byte length */
3553 char *char_to; /* the rhs of the map */
3554 I32 to_len; /* its byte length */
3555 SV *sv_to; /* and in a sv */
3556 AV* from_list; /* list of things that map to each 'to' */
3557
3558 hv_iterinit(specials_hv);
3559
4a4088c4
KW
3560 /* The keys are the characters (in UTF-8) that map to the corresponding
3561 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3562 * list. */
3563 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3564 SV** listp;
3565 if (! SvPOK(sv_to)) {
5637ef5b
NC
3566 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3567 "unexpectedly is not a string, flags=%lu",
3568 (unsigned long)SvFLAGS(sv_to));
5662e334 3569 }
4b88fb76 3570 /*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
3571
3572 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3573 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3574 * it. Those strings are all one character long */
3575 if ((listp = hv_fetch(specials_inverse,
3576 SvPVX(sv_to),
3577 SvCUR(sv_to), 0)))
3578 {
3579 from_list = (AV*) *listp;
3580 }
3581 else { /* No entry yet for it: create one */
3582 from_list = newAV();
3583 if (! hv_store(specials_inverse,
3584 SvPVX(sv_to),
3585 SvCUR(sv_to),
3586 (SV*) from_list, 0))
3587 {
3588 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3589 }
3590 }
3591
3592 /* Here have the list associated with this 'to' (perhaps newly
3593 * created and empty). Just add to it. Note that we ASSUME that
3594 * the input is guaranteed to not have duplications, so we don't
3595 * check for that. Duplications just slow down execution time. */
3596 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3597 }
3598
3599 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3600 * it looking for cases like the FB05/FB06 examples above. There would
3601 * be an entry in the hash like
3602 * 'st' => [ FB05, FB06 ]
3603 * In this example we will create two lists that get stored in the
3604 * returned hash, 'ret':
3605 * FB05 => [ FB05, FB06 ]
3606 * FB06 => [ FB05, FB06 ]
3607 *
3608 * Note that there is nothing to do if the array only has one element.
3609 * (In the normal 1-1 case handled below, we don't have to worry about
3610 * two lists, as everything gets tied to the single list that is
3611 * generated for the single character 'to'. But here, we are omitting
3612 * that list, ('st' in the example), so must have multiple lists.) */
3613 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3614 &char_to, &to_len)))
3615 {
7e13d567 3616 if (av_tindex_nomg(from_list) > 0) {
c70927a6 3617 SSize_t i;
5662e334
KW
3618
3619 /* We iterate over all combinations of i,j to place each code
3620 * point on each list */
7e13d567 3621 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
c70927a6 3622 SSize_t j;
5662e334
KW
3623 AV* i_list = newAV();
3624 SV** entryp = av_fetch(from_list, i, FALSE);
3625 if (entryp == NULL) {
3626 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3627 }
3628 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3629 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3630 }
3631 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3632 (SV*) i_list, FALSE))
3633 {
3634 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3635 }
3636
538e84ed 3637 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
7e13d567 3638 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
5662e334
KW
3639 entryp = av_fetch(from_list, j, FALSE);
3640 if (entryp == NULL) {
3641 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3642 }
3643
3644 /* When i==j this adds itself to the list */
4b88fb76
KW
3645 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3646 (U8*) SvPVX(*entryp),
3647 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3648 0)));
4637d003 3649 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
5662e334
KW
3650 }
3651 }
3652 }
3653 }
3654 SvREFCNT_dec(specials_inverse); /* done with it */
3655 } /* End of specials */
3656
064c021d 3657 /* read $swash->{LIST} */
74894415
KW
3658
3659#if UNICODE_MAJOR_VERSION == 3 \
3660 && UNICODE_DOT_VERSION == 0 \
3661 && UNICODE_DOT_DOT_VERSION == 1
3662
3663 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3664 * rule so that things work under /iaa and /il */
3665
3666 SV * mod_listsv = sv_mortalcopy(*listsvp);
3667 sv_catpv(mod_listsv, "130\t130\t131\n");
3668 l = (U8*)SvPV(mod_listsv, lcur);
3669
3670#else
3671
064c021d 3672 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3673
3674#endif
3675
064c021d
KW
3676 lend = l + lcur;
3677
3678 /* Go through each input line */
3679 while (l < lend) {
3680 UV min, max, val;
3681 UV inverse;
95543e92
KW
3682 l = swash_scan_list_line(l, lend, &min, &max, &val,
3683 cBOOL(octets), typestr);
064c021d
KW
3684 if (l > lend) {
3685 break;
3686 }
3687
3688 /* Each element in the range is to be inverted */
3689 for (inverse = min; inverse <= max; inverse++) {
3690 AV* list;
064c021d
KW
3691 SV** listp;
3692 IV i;
3693 bool found_key = FALSE;
5662e334 3694 bool found_inverse = FALSE;
064c021d
KW
3695