This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlapi: Clarify utf8n_to_uvchr entry
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
27da23d5
JH
36static const char unees[] =
37 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 38static const char cp_above_legal_max[] =
2d212e86 39 "Use of code point 0x%"UVXf" is deprecated; the permissible max is 0x%"UVXf"";
760c7c2f 40
114d9c4d 41#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 42
48ef279e 43/*
ccfc67b7 44=head1 Unicode Support
7fefc6c1 45These are various utility functions for manipulating UTF8-encoded
72d33970 46strings. For the uninitiated, this is a method of representing arbitrary
61296642 47Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
48characters in the ASCII range are unmodified, and a zero byte never appears
49within non-zero characters.
166f8a29 50
eaf7a4d2
CS
51=cut
52*/
53
54/*
378516de 55=for apidoc uvoffuni_to_utf8_flags
eebe1485 56
a27992cc 57THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
58Instead, B<Almost all code should use L</uvchr_to_utf8> or
59L</uvchr_to_utf8_flags>>.
a27992cc 60
de69f3af
KW
61This function is like them, but the input is a strict Unicode
62(as opposed to native) code point. Only in very rare circumstances should code
63not be using the native code point.
949cf498 64
efa9cd84 65For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 66
eebe1485
SC
67=cut
68*/
69
8ee1cdcb
KW
70#define HANDLE_UNICODE_SURROGATE(uv, flags) \
71 STMT_START { \
72 if (flags & UNICODE_WARN_SURROGATE) { \
73 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
74 "UTF-16 surrogate U+%04"UVXf, uv); \
75 } \
76 if (flags & UNICODE_DISALLOW_SURROGATE) { \
77 return NULL; \
78 } \
79 } STMT_END;
80
81#define HANDLE_UNICODE_NONCHAR(uv, flags) \
82 STMT_START { \
83 if (flags & UNICODE_WARN_NONCHAR) { \
84 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
85 "Unicode non-character U+%04"UVXf" is not " \
86 "recommended for open interchange", uv); \
87 } \
88 if (flags & UNICODE_DISALLOW_NONCHAR) { \
89 return NULL; \
90 } \
91 } STMT_END;
92
ba6ed43c
KW
93/* Use shorter names internally in this file */
94#define SHIFT UTF_ACCUMULATION_SHIFT
95#undef MARK
96#define MARK UTF_CONTINUATION_MARK
97#define MASK UTF_CONTINUATION_MASK
98
dfe13c55 99U8 *
378516de 100Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 101{
378516de 102 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 103
2d1545e5 104 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 105 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
106 return d;
107 }
facc1dc2 108
3ea68d71 109 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
110 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
111 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
112 return d;
113 }
d9432125 114
ba6ed43c
KW
115 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
116 * below, the 16 is for start bytes E0-EF (which are all the possible ones
117 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
118 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
119 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
120 * 0x800-0xFFFF on ASCII */
121 if (uv < (16 * (1U << (2 * SHIFT)))) {
122 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
123 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
124 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
125
126#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
127 aren't tested here */
128 /* The most likely code points in this range are below the surrogates.
129 * Do an extra test to quickly exclude those. */
130 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
131 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
132 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
133 {
8ee1cdcb
KW
134 HANDLE_UNICODE_NONCHAR(uv, flags);
135 }
136 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
137 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 138 }
ba6ed43c
KW
139 }
140#endif
141 return d;
142 }
143
144 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
145 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
146 * happen starting with 4-byte characters on ASCII platforms. We unify the
147 * code for these with EBCDIC, even though some of them require 5-bytes on
148 * those, because khw believes the code saving is worth the very slight
149 * performance hit on these high EBCDIC code points. */
150
151 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
152 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
153 && ckWARN_d(WARN_DEPRECATED))
154 {
155 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
156 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
157 }
158 if ( (flags & UNICODE_WARN_SUPER)
159 || ( UNICODE_IS_ABOVE_31_BIT(uv)
160 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
161 {
162 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
163
164 /* Choose the more dire applicable warning */
165 (UNICODE_IS_ABOVE_31_BIT(uv))
166 ? "Code point 0x%"UVXf" is not Unicode, and not portable"
167 : "Code point 0x%"UVXf" is not Unicode, may not be portable",
168 uv);
169 }
170 if (flags & UNICODE_DISALLOW_SUPER
171 || ( UNICODE_IS_ABOVE_31_BIT(uv)
172 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
173 {
174 return NULL;
175 }
176 }
ba6ed43c
KW
177 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
178 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 179 }
d9432125 180
ba6ed43c
KW
181 /* Test for and handle 4-byte result. In the test immediately below, the
182 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
183 * characters). The 3 is for 3 continuation bytes; these each contribute
184 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
185 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
186 * 0x1_0000-0x1F_FFFF on ASCII */
187 if (uv < (8 * (1U << (3 * SHIFT)))) {
188 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
189 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
190 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
191 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
192
193#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
194 characters. The end-plane non-characters for EBCDIC were
195 handled just above */
196 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
197 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 198 }
ba6ed43c
KW
199 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
200 HANDLE_UNICODE_SURROGATE(uv, flags);
201 }
202#endif
203
204 return d;
205 }
206
207 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
208 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
209 * format. The unrolled version above turns out to not save all that much
210 * time, and at these high code points (well above the legal Unicode range
211 * on ASCII platforms, and well above anything in common use in EBCDIC),
212 * khw believes that less code outweighs slight performance gains. */
213
d9432125 214 {
5aaebcb3 215 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
216 U8 *p = d+len-1;
217 while (p > d) {
4c8cd605 218 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
219 uv >>= UTF_ACCUMULATION_SHIFT;
220 }
4c8cd605 221 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
222 return d+len;
223 }
a0ed51b3 224}
a5bf80e0 225
646ca15d 226/*
07693fe6
KW
227=for apidoc uvchr_to_utf8
228
bcb1a2d4 229Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 230of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
231C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
232the byte after the end of the new character. In other words,
07693fe6
KW
233
234 d = uvchr_to_utf8(d, uv);
235
236is the recommended wide native character-aware way of saying
237
238 *(d++) = uv;
239
760c7c2f
KW
240This function accepts any UV as input, but very high code points (above
241C<IV_MAX> on the platform) will raise a deprecation warning. This is
242typically 0x7FFF_FFFF in a 32-bit word.
243
244It is possible to forbid or warn on non-Unicode code points, or those that may
245be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 246
07693fe6
KW
247=cut
248*/
249
de69f3af
KW
250/* This is also a macro */
251PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
252
07693fe6
KW
253U8 *
254Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
255{
de69f3af 256 return uvchr_to_utf8(d, uv);
07693fe6
KW
257}
258
de69f3af
KW
259/*
260=for apidoc uvchr_to_utf8_flags
261
262Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 263of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
264C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
265the byte after the end of the new character. In other words,
de69f3af
KW
266
267 d = uvchr_to_utf8_flags(d, uv, flags);
268
269or, in most cases,
270
271 d = uvchr_to_utf8_flags(d, uv, 0);
272
273This is the Unicode-aware way of saying
274
275 *(d++) = uv;
276
760c7c2f
KW
277If C<flags> is 0, this function accepts any UV as input, but very high code
278points (above C<IV_MAX> for the platform) will raise a deprecation warning.
279This is typically 0x7FFF_FFFF in a 32-bit word.
280
281Specifying C<flags> can further restrict what is allowed and not warned on, as
282follows:
de69f3af 283
796b6530 284If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
285the function will raise a warning, provided UTF8 warnings are enabled. If
286instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
287NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 288
760c7c2f
KW
289Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
290affect how the function handles a Unicode non-character.
93e6dbd6 291
760c7c2f
KW
292And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
293affect the handling of code points that are above the Unicode maximum of
2940x10FFFF. Languages other than Perl may not be able to accept files that
295contain these.
93e6dbd6
KW
296
297The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
298the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
299three DISALLOW flags.
300
ab8e6d41
KW
301Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
302so using them is more problematic than other above-Unicode code points. Perl
303invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
304likely that non-Perl languages will not be able to read files that contain
305these that written by the perl interpreter; nor would Perl understand files
306written by something that uses a different extension. For these reasons, there
307is a separate set of flags that can warn and/or disallow these extremely high
308code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
309C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
310are entirely independent from the deprecation warning for code points above
311C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
312code point that needs more than 31 bits to represent. When that happens,
313effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
31432-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
315above-Unicode code points, including these, as malformations; and
316C<UNICODE_WARN_SUPER> warns on these.)
317
318On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
319extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
320than on ASCII. Prior to that, code points 2**31 and higher were simply
321unrepresentable, and a different, incompatible method was used to represent
322code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
323and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
324platforms, warning and disallowing 2**31 and higher.
de69f3af 325
de69f3af
KW
326=cut
327*/
328
329/* This is also a macro */
330PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
331
07693fe6
KW
332U8 *
333Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
334{
de69f3af 335 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
336}
337
338/*
768c67ee 339
35f8c9bd
KW
340A helper function for the macro isUTF8_CHAR(), which should be used instead of
341this function. The macro will handle smaller code points directly saving time,
342using this function as a fall-back for higher code points. This function
343assumes that it is not called with an invariant character, and that
344's + len - 1' is within bounds of the string 's'.
345
346Tests if the string C<s> of at least length 'len' is a valid variant UTF-8
347character. 0 is returned if not, otherwise, 'len' is returned.
348
4dab108f
KW
349It is written in such a way that if 'len' is set to less than a full
350character's length, it will test if the bytes ending there form the legal
351beginning of partial character.
352
35f8c9bd
KW
353*/
354
355STRLEN
356Perl__is_utf8_char_slow(const U8 * const s, const STRLEN len)
357{
358 const U8 *e;
359 const U8 *x, *y;
360
361 PERL_ARGS_ASSERT__IS_UTF8_CHAR_SLOW;
362
363 if (UNLIKELY(! UTF8_IS_START(*s))) {
364 return 0;
365 }
366
367 e = s + len;
368
369 for (x = s + 1; x < e; x++) {
370 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
371 return 0;
372 }
373 }
374
375#ifndef EBCDIC
376
377 /* Here is syntactically valid. Make sure this isn't the start of an
378 * overlong. These values were found by manually inspecting the UTF-8
379 * patterns. See the tables in utf8.h and utfebcdic.h */
380
381 /* This is not needed on modern perls where C0 and C1 are not considered
382 * start bytes. */
383#if 0
384 if (UNLIKELY(*s < 0xC2)) {
385 return 0;
386 }
387#endif
388
389 if (len > 1) {
390 if ( (*s == 0xE0 && UNLIKELY(s[1] < 0xA0))
391 || (*s == 0xF0 && UNLIKELY(s[1] < 0x90))
392 || (*s == 0xF8 && UNLIKELY(s[1] < 0x88))
393 || (*s == 0xFC && UNLIKELY(s[1] < 0x84))
394 || (*s == 0xFE && UNLIKELY(s[1] < 0x82)))
395 {
396 return 0;
397 }
398 if ((len > 6 && UNLIKELY(*s == 0xFF) && UNLIKELY(s[6] < 0x81))) {
399 return 0;
400 }
401 }
402
403#else /* For EBCDIC, we use I8, which is the same on all code pages */
404 {
405 const U8 s0 = NATIVE_UTF8_TO_I8(*s);
406
407 /* On modern perls C0-C4 aren't considered start bytes */
408 if ( /* s0 < 0xC5 || */ s0 == 0xE0) {
409 return 0;
410 }
411
412 if (len >= 1) {
413 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
414
415 if ( (s0 == 0xF0 && UNLIKELY(s1 < 0xB0))
416 || (s0 == 0xF8 && UNLIKELY(s1 < 0xA8))
417 || (s0 == 0xFC && UNLIKELY(s1 < 0xA4))
418 || (s0 == 0xFE && UNLIKELY(s1 < 0x82)))
419 {
420 return 0;
421 }
422 if ((len > 7 && UNLIKELY(s0 == 0xFF) && UNLIKELY(s[7] < 0xA1))) {
423 return 0;
424 }
425 }
426 }
427
428#endif
429
430 /* Now see if this would overflow a UV on this platform. See if the UTF8
431 * for this code point is larger than that for the highest representable
432 * code point */
433 y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
434
435 for (x = s; x < e; x++, y++) {
436
437 /* If the same at this byte, go on to the next */
438 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
439 continue;
440 }
441
442 /* If this is larger, it overflows */
443 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
444 return 0;
445 }
446
447 /* But if smaller, it won't */
448 break;
449 }
450
451 return len;
452}
453
454/*
455
de69f3af 456=for apidoc utf8n_to_uvchr
378516de
KW
457
458THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 459Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 460
9041c2e3 461Bottom level UTF-8 decode routine.
de69f3af 462Returns the native code point value of the first character in the string C<s>,
746afd53
KW
463which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
464C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
465the length, in bytes, of that character.
949cf498
KW
466
467The value of C<flags> determines the behavior when C<s> does not point to a
468well-formed UTF-8 character. If C<flags> is 0, when a malformation is found,
524080c4
KW
469zero is returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>) is the
470next possible position in C<s> that could begin a non-malformed character.
471Also, if UTF-8 warnings haven't been lexically disabled, a warning is raised.
949cf498
KW
472
473Various ALLOW flags can be set in C<flags> to allow (and not warn on)
474individual types of malformations, such as the sequence being overlong (that
475is, when there is a shorter sequence that can express the same code point;
476overlong sequences are expressly forbidden in the UTF-8 standard due to
477potential security issues). Another malformation example is the first byte of
478a character not being a legal first byte. See F<utf8.h> for the list of such
524080c4
KW
479flags. For allowed 0 length strings, this function returns 0; for allowed
480overlong sequences, the computed code point is returned; for all other allowed
481malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
482determinable reasonable value.
949cf498 483
796b6530 484The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
485flags) malformation is found. If this flag is set, the routine assumes that
486the caller will raise a warning, and this function will silently just set
d088425d
KW
487C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
488
75200dff 489Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 490character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
491in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
492be set to 1. To disambiguate, upon a zero return, see if the first byte of
493C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
494error.
949cf498
KW
495
496Certain code points are considered problematic. These are Unicode surrogates,
746afd53 497Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 498By default these are considered regular code points, but certain situations
5eafe189 499warrant special handling for them. If C<flags> contains
796b6530
KW
500C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all three classes are treated as
501malformations and handled as such. The flags C<UTF8_DISALLOW_SURROGATE>,
502C<UTF8_DISALLOW_NONCHAR>, and C<UTF8_DISALLOW_SUPER> (meaning above the legal
503Unicode maximum) can be set to disallow these categories individually.
504
505The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
506C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
507raised for their respective categories, but otherwise the code points are
508considered valid (not malformations). To get a category to both be treated as
509a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 510(But note that warnings are not raised if lexically disabled nor if
796b6530 511C<UTF8_CHECK_ONLY> is also specified.)
949cf498 512
760c7c2f
KW
513It is now deprecated to have very high code points (above C<IV_MAX> on the
514platforms) and this function will raise a deprecation warning for these (unless
515such warnings are turned off). This value, is typically 0x7FFF_FFFF (2**31 -1)
516in a 32-bit word.
ab8e6d41
KW
517
518Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
519so using them is more problematic than other above-Unicode code points. Perl
520invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
521likely that non-Perl languages will not be able to read files that contain
522these that written by the perl interpreter; nor would Perl understand files
523written by something that uses a different extension. For these reasons, there
524is a separate set of flags that can warn and/or disallow these extremely high
525code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
526C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
527are entirely independent from the deprecation warning for code points above
528C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
529code point that needs more than 31 bits to represent. When that happens,
530effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
53132-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
532above-Unicode code points, including these, as malformations; and
533C<UTF8_WARN_SUPER> warns on these.)
534
535On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
536extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
537than on ASCII. Prior to that, code points 2**31 and higher were simply
538unrepresentable, and a different, incompatible method was used to represent
539code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
540and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
541platforms, warning and disallowing 2**31 and higher.
949cf498
KW
542
543All other code points corresponding to Unicode characters, including private
544use and those yet to be assigned, are never considered malformed and never
545warn.
67e989fb 546
37607a96
PK
547=cut
548*/
67e989fb 549
a0ed51b3 550UV
de69f3af 551Perl_utf8n_to_uvchr(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
a0ed51b3 552{
d4c19fe8 553 const U8 * const s0 = s;
eb83ed87 554 U8 overflow_byte = '\0'; /* Save byte in case of overflow */
0b8d30e8 555 U8 * send;
eb83ed87
KW
556 UV uv = *s;
557 STRLEN expectlen;
949cf498 558 SV* sv = NULL;
eb83ed87
KW
559 UV outlier_ret = 0; /* return value when input is in error or problematic
560 */
561 UV pack_warn = 0; /* Save result of packWARN() for later */
562 bool unexpected_non_continuation = FALSE;
563 bool overflowed = FALSE;
2f8f112e 564 bool do_overlong_test = TRUE; /* May have to skip this test */
a0dbb045 565
eb83ed87 566 const char* const malformed_text = "Malformed UTF-8 character";
7918f24d 567
de69f3af 568 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
a0dbb045 569
eb83ed87
KW
570 /* The order of malformation tests here is important. We should consume as
571 * few bytes as possible in order to not skip any valid character. This is
572 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
573 * http://unicode.org/reports/tr36 for more discussion as to why. For
574 * example, once we've done a UTF8SKIP, we can tell the expected number of
575 * bytes, and could fail right off the bat if the input parameters indicate
576 * that there are too few available. But it could be that just that first
577 * byte is garbled, and the intended character occupies fewer bytes. If we
578 * blindly assumed that the first byte is correct, and skipped based on
579 * that number, we could skip over a valid input character. So instead, we
580 * always examine the sequence byte-by-byte.
581 *
582 * We also should not consume too few bytes, otherwise someone could inject
583 * things. For example, an input could be deliberately designed to
584 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 585 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
586 * which is actually part of of the overflowing sequence), that could look
587 * legitimate to the caller, which could discard the initial partial
588 * sequence and process the rest, inappropriately */
589
590 /* Zero length strings, if allowed, of necessity are zero */
b5b9af04 591 if (UNLIKELY(curlen == 0)) {
eb83ed87
KW
592 if (retlen) {
593 *retlen = 0;
594 }
a0dbb045 595
eb83ed87
KW
596 if (flags & UTF8_ALLOW_EMPTY) {
597 return 0;
598 }
599 if (! (flags & UTF8_CHECK_ONLY)) {
600 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (empty string)", malformed_text));
601 }
0c443dc2
JH
602 goto malformed;
603 }
604
eb83ed87
KW
605 expectlen = UTF8SKIP(s);
606
607 /* A well-formed UTF-8 character, as the vast majority of calls to this
608 * function will be for, has this expected length. For efficiency, set
609 * things up here to return it. It will be overriden only in those rare
610 * cases where a malformation is found */
611 if (retlen) {
612 *retlen = expectlen;
613 }
614
615 /* An invariant is trivially well-formed */
1d72bdf6 616 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 617 return uv;
a0ed51b3 618 }
67e989fb 619
eb83ed87 620 /* A continuation character can't start a valid sequence */
b5b9af04 621 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
eb83ed87
KW
622 if (flags & UTF8_ALLOW_CONTINUATION) {
623 if (retlen) {
624 *retlen = 1;
625 }
626 return UNICODE_REPLACEMENT;
627 }
ba210ebe 628
eb83ed87
KW
629 if (! (flags & UTF8_CHECK_ONLY)) {
630 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected continuation byte 0x%02x, with no preceding start byte)", malformed_text, *s0));
631 }
632 curlen = 1;
ba210ebe
JH
633 goto malformed;
634 }
9041c2e3 635
dcd27b3c
KW
636 /* Here is not a continuation byte, nor an invariant. The only thing left
637 * is a start byte (possibly for an overlong) */
638
534752c1
KW
639 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
640 * that indicate the number of bytes in the character's whole UTF-8
641 * sequence, leaving just the bits that are part of the value. */
642 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 643
eb83ed87
KW
644 /* Now, loop through the remaining bytes in the character's sequence,
645 * accumulating each into the working value as we go. Be sure to not look
646 * past the end of the input string */
0b8d30e8
KW
647 send = (U8*) s0 + ((expectlen <= curlen) ? expectlen : curlen);
648
eb83ed87 649 for (s = s0 + 1; s < send; s++) {
b5b9af04 650 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
eb83ed87
KW
651 if (uv & UTF_ACCUMULATION_OVERFLOW_MASK) {
652
653 /* The original implementors viewed this malformation as more
654 * serious than the others (though I, khw, don't understand
655 * why, since other malformations also give very very wrong
656 * results), so there is no way to turn off checking for it.
657 * Set a flag, but keep going in the loop, so that we absorb
658 * the rest of the bytes that comprise the character. */
659 overflowed = TRUE;
660 overflow_byte = *s; /* Save for warning message's use */
661 }
8850bf83 662 uv = UTF8_ACCUMULATE(uv, *s);
eb83ed87
KW
663 }
664 else {
665 /* Here, found a non-continuation before processing all expected
666 * bytes. This byte begins a new character, so quit, even if
667 * allowing this malformation. */
668 unexpected_non_continuation = TRUE;
669 break;
670 }
671 } /* End of loop through the character's bytes */
672
673 /* Save how many bytes were actually in the character */
674 curlen = s - s0;
675
676 /* The loop above finds two types of malformations: non-continuation and/or
677 * overflow. The non-continuation malformation is really a too-short
678 * malformation, as it means that the current character ended before it was
679 * expected to (being terminated prematurely by the beginning of the next
680 * character, whereas in the too-short malformation there just are too few
681 * bytes available to hold the character. In both cases, the check below
682 * that we have found the expected number of bytes would fail if executed.)
683 * Thus the non-continuation malformation is really unnecessary, being a
684 * subset of the too-short malformation. But there may be existing
685 * applications that are expecting the non-continuation type, so we retain
686 * it, and return it in preference to the too-short malformation. (If this
687 * code were being written from scratch, the two types might be collapsed
688 * into one.) I, khw, am also giving priority to returning the
689 * non-continuation and too-short malformations over overflow when multiple
690 * ones are present. I don't know of any real reason to prefer one over
691 * the other, except that it seems to me that multiple-byte errors trumps
692 * errors from a single byte */
b5b9af04 693 if (UNLIKELY(unexpected_non_continuation)) {
eb83ed87
KW
694 if (!(flags & UTF8_ALLOW_NON_CONTINUATION)) {
695 if (! (flags & UTF8_CHECK_ONLY)) {
696 if (curlen == 1) {
697 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (unexpected non-continuation byte 0x%02x, immediately after start byte 0x%02x)", malformed_text, *s, *s0));
698 }
699 else {
700 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
701 }
702 }
eb83ed87
KW
703 goto malformed;
704 }
705 uv = UNICODE_REPLACEMENT;
2f8f112e
KW
706
707 /* Skip testing for overlongs, as the REPLACEMENT may not be the same
708 * as what the original expectations were. */
709 do_overlong_test = FALSE;
eb83ed87
KW
710 if (retlen) {
711 *retlen = curlen;
712 }
713 }
b5b9af04 714 else if (UNLIKELY(curlen < expectlen)) {
eb83ed87
KW
715 if (! (flags & UTF8_ALLOW_SHORT)) {
716 if (! (flags & UTF8_CHECK_ONLY)) {
717 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 718 }
eb83ed87
KW
719 goto malformed;
720 }
721 uv = UNICODE_REPLACEMENT;
2f8f112e 722 do_overlong_test = FALSE;
eb83ed87
KW
723 if (retlen) {
724 *retlen = curlen;
725 }
726 }
727
b5b9af04 728 if (UNLIKELY(overflowed)) {
eb83ed87 729 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "%s (overflow at byte 0x%02x, after start byte 0x%02x)", malformed_text, overflow_byte, *s0));
ba210ebe 730 goto malformed;
eb83ed87 731 }
eb83ed87 732
2f8f112e 733 if (do_overlong_test
5aaebcb3 734 && expectlen > (STRLEN) OFFUNISKIP(uv)
2f8f112e
KW
735 && ! (flags & UTF8_ALLOW_LONG))
736 {
eb83ed87
KW
737 /* The overlong malformation has lower precedence than the others.
738 * Note that if this malformation is allowed, we return the actual
739 * value, instead of the replacement character. This is because this
740 * value is actually well-defined. */
741 if (! (flags & UTF8_CHECK_ONLY)) {
5aaebcb3 742 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
743 }
744 goto malformed;
745 }
746
1a89bb6c 747 /* Here, the input is considered to be well-formed, but it still could be a
eb83ed87
KW
748 * problematic code point that is not allowed by the input parameters. */
749 if (uv >= UNICODE_SURROGATE_FIRST /* isn't problematic if < this */
760c7c2f
KW
750 && ((flags & ( UTF8_DISALLOW_NONCHAR
751 |UTF8_DISALLOW_SURROGATE
752 |UTF8_DISALLOW_SUPER
753 |UTF8_DISALLOW_ABOVE_31_BIT
754 |UTF8_WARN_NONCHAR
755 |UTF8_WARN_SURROGATE
756 |UTF8_WARN_SUPER
757 |UTF8_WARN_ABOVE_31_BIT))
758 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
759 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 760 {
949cf498 761 if (UNICODE_IS_SURROGATE(uv)) {
ea5ced44
KW
762
763 /* By adding UTF8_CHECK_ONLY to the test, we avoid unnecessary
764 * generation of the sv, since no warnings are raised under CHECK */
eb83ed87 765 if ((flags & (UTF8_WARN_SURROGATE|UTF8_CHECK_ONLY)) == UTF8_WARN_SURROGATE
54f4afef 766 && ckWARN_d(WARN_SURROGATE))
eb83ed87 767 {
111d382d 768 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "UTF-16 surrogate U+%04"UVXf"", uv));
54f4afef 769 pack_warn = packWARN(WARN_SURROGATE);
949cf498
KW
770 }
771 if (flags & UTF8_DISALLOW_SURROGATE) {
772 goto disallowed;
773 }
774 }
949cf498 775 else if ((uv > PERL_UNICODE_MAX)) {
eb83ed87 776 if ((flags & (UTF8_WARN_SUPER|UTF8_CHECK_ONLY)) == UTF8_WARN_SUPER
ea5ced44 777 && ckWARN_d(WARN_NON_UNICODE))
eb83ed87 778 {
0bcdd8f6
KW
779 sv = sv_2mortal(Perl_newSVpvf(aTHX_
780 "Code point 0x%04"UVXf" is not Unicode, may not be portable",
781 uv));
54f4afef 782 pack_warn = packWARN(WARN_NON_UNICODE);
949cf498 783 }
c0236afe
KW
784
785 /* The maximum code point ever specified by a standard was
786 * 2**31 - 1. Anything larger than that is a Perl extension that
787 * very well may not be understood by other applications (including
788 * earlier perl versions on EBCDIC platforms). On ASCII platforms,
789 * these code points are indicated by the first UTF-8 byte being
d35f2ca5
KW
790 * 0xFE or 0xFF. We test for these after the regular SUPER ones,
791 * and before possibly bailing out, so that the slightly more dire
c0236afe
KW
792 * warning will override the regular one. */
793 if (
794#ifndef EBCDIC
795 (*s0 & 0xFE) == 0xFE /* matches both FE, FF */
796#else
797 /* The I8 for 2**31 (U+80000000) is
798 * \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
799 * and it turns out that on all EBCDIC pages recognized that
800 * the UTF-EBCDIC for that code point is
801 * \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
802 * For the next lower code point, the 1047 UTF-EBCDIC is
803 * \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
804 * The other code pages differ only in the bytes following
805 * \x42. Thus the following works (the minimum continuation
806 * byte is \x41). */
807 *s0 == 0xFE && send - s0 > 7 && ( s0[1] > 0x41
808 || s0[2] > 0x41
809 || s0[3] > 0x41
810 || s0[4] > 0x41
811 || s0[5] > 0x41
812 || s0[6] > 0x41
813 || s0[7] > 0x42)
814#endif
d35f2ca5
KW
815 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER
816 |UTF8_DISALLOW_ABOVE_31_BIT)))
ea5ced44 817 {
0bcdd8f6 818 if ( ! (flags & UTF8_CHECK_ONLY)
d35f2ca5 819 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
0bcdd8f6 820 && ckWARN_d(WARN_UTF8))
ea5ced44 821 {
0bcdd8f6
KW
822 sv = sv_2mortal(Perl_newSVpvf(aTHX_
823 "Code point 0x%"UVXf" is not Unicode, and not portable",
824 uv));
ea5ced44
KW
825 pack_warn = packWARN(WARN_UTF8);
826 }
d35f2ca5 827 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
ea5ced44
KW
828 goto disallowed;
829 }
830 }
c0236afe 831
949cf498
KW
832 if (flags & UTF8_DISALLOW_SUPER) {
833 goto disallowed;
834 }
760c7c2f
KW
835
836 /* The deprecated warning overrides any non-deprecated one */
837 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP) && ckWARN_d(WARN_DEPRECATED))
838 {
839 sv = sv_2mortal(Perl_newSVpvf(aTHX_ cp_above_legal_max,
840 uv, MAX_NON_DEPRECATED_CP));
841 pack_warn = packWARN(WARN_DEPRECATED);
842 }
949cf498 843 }
4190d317
KW
844 else if (UNICODE_IS_NONCHAR(uv)) {
845 if ((flags & (UTF8_WARN_NONCHAR|UTF8_CHECK_ONLY)) == UTF8_WARN_NONCHAR
54f4afef 846 && ckWARN_d(WARN_NONCHAR))
4190d317 847 {
ba707cdc 848 sv = sv_2mortal(Perl_newSVpvf(aTHX_ "Unicode non-character U+%04"UVXf" is not recommended for open interchange", uv));
54f4afef 849 pack_warn = packWARN(WARN_NONCHAR);
4190d317
KW
850 }
851 if (flags & UTF8_DISALLOW_NONCHAR) {
852 goto disallowed;
853 }
854 }
949cf498 855
eb83ed87 856 if (sv) {
de69f3af
KW
857 outlier_ret = uv; /* Note we don't bother to convert to native,
858 as all the outlier code points are the same
859 in both ASCII and EBCDIC */
eb83ed87
KW
860 goto do_warn;
861 }
862
949cf498
KW
863 /* Here, this is not considered a malformed character, so drop through
864 * to return it */
a0ed51b3 865 }
ba210ebe 866
de69f3af 867 return UNI_TO_NATIVE(uv);
ba210ebe 868
eb83ed87
KW
869 /* There are three cases which get to beyond this point. In all 3 cases:
870 * <sv> if not null points to a string to print as a warning.
871 * <curlen> is what <*retlen> should be set to if UTF8_CHECK_ONLY isn't
872 * set.
873 * <outlier_ret> is what return value to use if UTF8_CHECK_ONLY isn't set.
874 * This is done by initializing it to 0, and changing it only
875 * for case 1).
876 * The 3 cases are:
877 * 1) The input is valid but problematic, and to be warned about. The
878 * return value is the resultant code point; <*retlen> is set to
879 * <curlen>, the number of bytes that comprise the code point.
880 * <pack_warn> contains the result of packWARN() for the warning
881 * types. The entry point for this case is the label <do_warn>;
882 * 2) The input is a valid code point but disallowed by the parameters to
883 * this function. The return value is 0. If UTF8_CHECK_ONLY is set,
884 * <*relen> is -1; otherwise it is <curlen>, the number of bytes that
885 * comprise the code point. <pack_warn> contains the result of
886 * packWARN() for the warning types. The entry point for this case is
887 * the label <disallowed>.
888 * 3) The input is malformed. The return value is 0. If UTF8_CHECK_ONLY
889 * is set, <*relen> is -1; otherwise it is <curlen>, the number of
890 * bytes that comprise the malformation. All such malformations are
891 * assumed to be warning type <utf8>. The entry point for this case
892 * is the label <malformed>.
893 */
949cf498 894
7b52d656 895 malformed:
ba210ebe 896
eb83ed87
KW
897 if (sv && ckWARN_d(WARN_UTF8)) {
898 pack_warn = packWARN(WARN_UTF8);
899 }
900
7b52d656 901 disallowed:
eb83ed87 902
fcc8fcf6 903 if (flags & UTF8_CHECK_ONLY) {
ba210ebe 904 if (retlen)
10edeb5d 905 *retlen = ((STRLEN) -1);
ba210ebe
JH
906 return 0;
907 }
908
7b52d656 909 do_warn:
5b311467 910
eb83ed87
KW
911 if (pack_warn) { /* <pack_warn> was initialized to 0, and changed only
912 if warnings are to be raised. */
f555bc63 913 const char * const string = SvPVX_const(sv);
a0dbb045 914
f555bc63
KW
915 if (PL_op)
916 Perl_warner(aTHX_ pack_warn, "%s in %s", string, OP_DESC(PL_op));
917 else
918 Perl_warner(aTHX_ pack_warn, "%s", string);
a0dbb045
JH
919 }
920
eb83ed87
KW
921 if (retlen) {
922 *retlen = curlen;
923 }
ba210ebe 924
eb83ed87 925 return outlier_ret;
a0ed51b3
LW
926}
927
8e84507e 928/*
ec5f19d0
KW
929=for apidoc utf8_to_uvchr_buf
930
931Returns the native code point of the first character in the string C<s> which
932is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 933C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 934
524080c4
KW
935If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
936enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 937C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 938(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 939C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 940the next possible position in C<s> that could begin a non-malformed character.
de69f3af 941See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 942returned.
ec5f19d0 943
760c7c2f
KW
944Code points above the platform's C<IV_MAX> will raise a deprecation warning,
945unless those are turned off.
946
ec5f19d0 947=cut
52be2536
KW
948
949Also implemented as a macro in utf8.h
950
ec5f19d0
KW
951*/
952
953
954UV
955Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
956{
ec5f19d0
KW
957 assert(s < send);
958
959 return utf8n_to_uvchr(s, send - s, retlen,
960 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
961}
962
52be2536
KW
963/* This is marked as deprecated
964 *
ec5f19d0
KW
965=for apidoc utf8_to_uvuni_buf
966
de69f3af
KW
967Only in very rare circumstances should code need to be dealing in Unicode
968(as opposed to native) code points. In those few cases, use
969C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
970
971Returns the Unicode (not-native) code point of the first character in the
972string C<s> which
ec5f19d0
KW
973is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
974C<retlen> will be set to the length, in bytes, of that character.
975
524080c4
KW
976If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
977enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
978NULL) to -1. If those warnings are off, the computed value if well-defined (or
979the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
980is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
981next possible position in C<s> that could begin a non-malformed character.
de69f3af 982See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 983
760c7c2f
KW
984Code points above the platform's C<IV_MAX> will raise a deprecation warning,
985unless those are turned off.
986
ec5f19d0
KW
987=cut
988*/
989
990UV
991Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
992{
993 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
994
995 assert(send > s);
996
5962d97e
KW
997 /* Call the low level routine, asking for checks */
998 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
999}
1000
b76347f2 1001/*
87cea99e 1002=for apidoc utf8_length
b76347f2
JH
1003
1004Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1005Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1006up past C<e>, croaks.
b76347f2
JH
1007
1008=cut
1009*/
1010
1011STRLEN
35a4481c 1012Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1013{
1014 STRLEN len = 0;
1015
7918f24d
NC
1016 PERL_ARGS_ASSERT_UTF8_LENGTH;
1017
8850bf83
JH
1018 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1019 * the bitops (especially ~) can create illegal UTF-8.
1020 * In other words: in Perl UTF-8 is not just for Unicode. */
1021
a3b680e6
AL
1022 if (e < s)
1023 goto warn_and_return;
b76347f2 1024 while (s < e) {
4cbf4130 1025 s += UTF8SKIP(s);
8e91ec7f
AV
1026 len++;
1027 }
1028
1029 if (e != s) {
1030 len--;
1031 warn_and_return:
9b387841
NC
1032 if (PL_op)
1033 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1034 "%s in %s", unees, OP_DESC(PL_op));
1035 else
61a12c31 1036 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1037 }
1038
1039 return len;
1040}
1041
b06226ff 1042/*
fed3ba5d
NC
1043=for apidoc bytes_cmp_utf8
1044
a1433954 1045Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1046sequence of characters (stored as UTF-8)
1047in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1048equal, -1 or -2 if the first string is less than the second string, +1 or +2
1049if the first string is greater than the second string.
1050
1051-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1052longer string. -2 or +2 is returned if
1053there was a difference between characters
fed3ba5d
NC
1054within the strings.
1055
1056=cut
1057*/
1058
1059int
1060Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1061{
1062 const U8 *const bend = b + blen;
1063 const U8 *const uend = u + ulen;
1064
1065 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1066
1067 while (b < bend && u < uend) {
1068 U8 c = *u++;
1069 if (!UTF8_IS_INVARIANT(c)) {
1070 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1071 if (u < uend) {
1072 U8 c1 = *u++;
1073 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1074 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d
NC
1075 } else {
1076 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1077 "Malformed UTF-8 character "
1078 "(unexpected non-continuation byte 0x%02x"
1079 ", immediately after start byte 0x%02x)"
1080 /* Dear diag.t, it's in the pod. */
1081 "%s%s", c1, c,
1082 PL_op ? " in " : "",
1083 PL_op ? OP_DESC(PL_op) : "");
1084 return -2;
1085 }
1086 } else {
1087 if (PL_op)
1088 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1089 "%s in %s", unees, OP_DESC(PL_op));
1090 else
61a12c31 1091 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1092 return -2; /* Really want to return undef :-) */
1093 }
1094 } else {
1095 return -2;
1096 }
1097 }
1098 if (*b != c) {
1099 return *b < c ? -2 : +2;
1100 }
1101 ++b;
1102 }
1103
1104 if (b == bend && u == uend)
1105 return 0;
1106
1107 return b < bend ? +1 : -1;
1108}
1109
1110/*
87cea99e 1111=for apidoc utf8_to_bytes
6940069f 1112
2bbc8d55 1113Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1114Unlike L</bytes_to_utf8>, this over-writes the original string, and
1115updates C<len> to contain the new length.
67e989fb 1116Returns zero on failure, setting C<len> to -1.
6940069f 1117
a1433954 1118If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1119
6940069f
GS
1120=cut
1121*/
1122
1123U8 *
37607a96 1124Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1125{
d4c19fe8
AL
1126 U8 * const save = s;
1127 U8 * const send = s + *len;
6940069f 1128 U8 *d;
246fae53 1129
7918f24d 1130 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1131 PERL_UNUSED_CONTEXT;
7918f24d 1132
1e54db1a 1133 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1134 while (s < send) {
d59937ca
KW
1135 if (! UTF8_IS_INVARIANT(*s)) {
1136 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1137 *len = ((STRLEN) -1);
1138 return 0;
1139 }
1140 s++;
dcad2880 1141 }
d59937ca 1142 s++;
246fae53 1143 }
dcad2880
JH
1144
1145 d = s = save;
6940069f 1146 while (s < send) {
80e0b38f
KW
1147 U8 c = *s++;
1148 if (! UTF8_IS_INVARIANT(c)) {
1149 /* Then it is two-byte encoded */
a62b247b 1150 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1151 s++;
1152 }
1153 *d++ = c;
6940069f
GS
1154 }
1155 *d = '\0';
246fae53 1156 *len = d - save;
6940069f
GS
1157 return save;
1158}
1159
1160/*
87cea99e 1161=for apidoc bytes_from_utf8
f9a63242 1162
2bbc8d55 1163Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1164Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1165the newly-created string, and updates C<len> to contain the new
1166length. Returns the original string if no conversion occurs, C<len>
72d33970 1167is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 11680 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1169in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1170
37607a96
PK
1171=cut
1172*/
f9a63242
JH
1173
1174U8 *
e1ec3a88 1175Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1176{
f9a63242 1177 U8 *d;
e1ec3a88
AL
1178 const U8 *start = s;
1179 const U8 *send;
f9a63242
JH
1180 I32 count = 0;
1181
7918f24d 1182 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1183 PERL_UNUSED_CONTEXT;
f9a63242 1184 if (!*is_utf8)
73d840c0 1185 return (U8 *)start;
f9a63242 1186
1e54db1a 1187 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1188 for (send = s + *len; s < send;) {
d59937ca
KW
1189 if (! UTF8_IS_INVARIANT(*s)) {
1190 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1191 return (U8 *)start;
d59937ca
KW
1192 }
1193 count++;
1194 s++;
db42d148 1195 }
d59937ca 1196 s++;
f9a63242
JH
1197 }
1198
35da51f7 1199 *is_utf8 = FALSE;
f9a63242 1200
212542aa 1201 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1202 s = start; start = d;
f9a63242
JH
1203 while (s < send) {
1204 U8 c = *s++;
1a91c45d 1205 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1206 /* Then it is two-byte encoded */
a62b247b 1207 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1208 s++;
c4d5f83a
NIS
1209 }
1210 *d++ = c;
f9a63242
JH
1211 }
1212 *d = '\0';
1213 *len = d - start;
73d840c0 1214 return (U8 *)start;
f9a63242
JH
1215}
1216
1217/*
87cea99e 1218=for apidoc bytes_to_utf8
6940069f 1219
ff97e5cf
KW
1220Converts a string C<s> of length C<len> bytes from the native encoding into
1221UTF-8.
6662521e 1222Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1223reflect the new length in bytes.
6940069f 1224
75200dff 1225A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1226
1227If you want to convert to UTF-8 from encodings other than
1228the native (Latin1 or EBCDIC),
a1433954 1229see L</sv_recode_to_utf8>().
c9ada85f 1230
497711e7 1231=cut
6940069f
GS
1232*/
1233
c682ebef
FC
1234/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1235 likewise need duplication. */
1236
6940069f 1237U8*
35a4481c 1238Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1239{
35a4481c 1240 const U8 * const send = s + (*len);
6940069f
GS
1241 U8 *d;
1242 U8 *dst;
7918f24d
NC
1243
1244 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 1245 PERL_UNUSED_CONTEXT;
6940069f 1246
212542aa 1247 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
1248 dst = d;
1249
1250 while (s < send) {
55d09dc8
KW
1251 append_utf8_from_native_byte(*s, &d);
1252 s++;
6940069f
GS
1253 }
1254 *d = '\0';
6662521e 1255 *len = d-dst;
6940069f
GS
1256 return dst;
1257}
1258
a0ed51b3 1259/*
dea0fc0b 1260 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
1261 *
1262 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1263 * We optimize for native, for obvious reasons. */
1264
1265U8*
dea0fc0b 1266Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 1267{
dea0fc0b
JH
1268 U8* pend;
1269 U8* dstart = d;
1270
7918f24d
NC
1271 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1272
dea0fc0b 1273 if (bytelen & 1)
f5992bc4 1274 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %"UVuf, (UV)bytelen);
dea0fc0b
JH
1275
1276 pend = p + bytelen;
1277
a0ed51b3 1278 while (p < pend) {
dea0fc0b
JH
1279 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1280 p += 2;
2d1545e5 1281 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 1282 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
1283 continue;
1284 }
56d37426
KW
1285 if (uv <= MAX_UTF8_TWO_BYTE) {
1286 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1287 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
1288 continue;
1289 }
46956fad
KW
1290#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1291#define LAST_HIGH_SURROGATE 0xDBFF
1292#define FIRST_LOW_SURROGATE 0xDC00
1293#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
1294
1295 /* This assumes that most uses will be in the first Unicode plane, not
1296 * needing surrogates */
1297 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
1298 && uv <= UNICODE_SURROGATE_LAST))
1299 {
1300 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
1301 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
1302 }
1303 else {
01ea242b 1304 UV low = (p[0] << 8) + p[1];
e23c50db
KW
1305 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
1306 || UNLIKELY(low > LAST_LOW_SURROGATE))
1307 {
01ea242b 1308 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
1309 }
1310 p += 2;
46956fad
KW
1311 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
1312 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 1313 }
a0ed51b3 1314 }
56d37426
KW
1315#ifdef EBCDIC
1316 d = uvoffuni_to_utf8_flags(d, uv, 0);
1317#else
a0ed51b3 1318 if (uv < 0x10000) {
eb160463
GS
1319 *d++ = (U8)(( uv >> 12) | 0xe0);
1320 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1321 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1322 continue;
1323 }
1324 else {
eb160463
GS
1325 *d++ = (U8)(( uv >> 18) | 0xf0);
1326 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
1327 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
1328 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
1329 continue;
1330 }
56d37426 1331#endif
a0ed51b3 1332 }
dea0fc0b 1333 *newlen = d - dstart;
a0ed51b3
LW
1334 return d;
1335}
1336
1337/* Note: this one is slightly destructive of the source. */
1338
1339U8*
dea0fc0b 1340Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
1341{
1342 U8* s = (U8*)p;
d4c19fe8 1343 U8* const send = s + bytelen;
7918f24d
NC
1344
1345 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
1346
e0ea5e2d
NC
1347 if (bytelen & 1)
1348 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %"UVuf,
1349 (UV)bytelen);
1350
a0ed51b3 1351 while (s < send) {
d4c19fe8 1352 const U8 tmp = s[0];
a0ed51b3
LW
1353 s[0] = s[1];
1354 s[1] = tmp;
1355 s += 2;
1356 }
dea0fc0b 1357 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
1358}
1359
922e8cb4
KW
1360bool
1361Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
1362{
1363 U8 tmpbuf[UTF8_MAXBYTES+1];
1364 uvchr_to_utf8(tmpbuf, c);
1365 return _is_utf8_FOO(classnum, tmpbuf);
1366}
1367
f9ae8fb6
JD
1368/* Internal function so we can deprecate the external one, and call
1369 this one from other deprecated functions in this file */
1370
f2645549
KW
1371bool
1372Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 1373{
f2645549 1374 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
1375
1376 if (*p == '_')
1377 return TRUE;
f25ce844 1378 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
1379}
1380
5092f92a 1381bool
eba68aa0
KW
1382Perl__is_uni_perl_idcont(pTHX_ UV c)
1383{
1384 U8 tmpbuf[UTF8_MAXBYTES+1];
1385 uvchr_to_utf8(tmpbuf, c);
1386 return _is_utf8_perl_idcont(tmpbuf);
1387}
1388
1389bool
f91dcd13
KW
1390Perl__is_uni_perl_idstart(pTHX_ UV c)
1391{
1392 U8 tmpbuf[UTF8_MAXBYTES+1];
1393 uvchr_to_utf8(tmpbuf, c);
1394 return _is_utf8_perl_idstart(tmpbuf);
1395}
1396
3a4c58c9
KW
1397UV
1398Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
1399{
1400 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1401 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
1402 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
1403 * either "SS" or "Ss". Which one to use is passed into the routine in
1404 * 'S_or_s' to avoid a test */
1405
1406 UV converted = toUPPER_LATIN1_MOD(c);
1407
1408 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
1409
1410 assert(S_or_s == 'S' || S_or_s == 's');
1411
6f2d5cbc 1412 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 1413 characters in this range */
3a4c58c9
KW
1414 *p = (U8) converted;
1415 *lenp = 1;
1416 return converted;
1417 }
1418
1419 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
1420 * which it maps to one of them, so as to only have to have one check for
1421 * it in the main case */
1422 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
1423 switch (c) {
1424 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
1425 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
1426 break;
1427 case MICRO_SIGN:
1428 converted = GREEK_CAPITAL_LETTER_MU;
1429 break;
79e064b9
KW
1430#if UNICODE_MAJOR_VERSION > 2 \
1431 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
1432 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
1433 case LATIN_SMALL_LETTER_SHARP_S:
1434 *(p)++ = 'S';
1435 *p = S_or_s;
1436 *lenp = 2;
1437 return 'S';
79e064b9 1438#endif
3a4c58c9
KW
1439 default:
1440 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 1441 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
1442 }
1443 }
1444
1445 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1446 *p = UTF8_TWO_BYTE_LO(converted);
1447 *lenp = 2;
1448
1449 return converted;
1450}
1451
50bda2c3
KW
1452/* Call the function to convert a UTF-8 encoded character to the specified case.
1453 * Note that there may be more than one character in the result.
1454 * INP is a pointer to the first byte of the input character
1455 * OUTP will be set to the first byte of the string of changed characters. It
1456 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
1457 * LENP will be set to the length in bytes of the string of changed characters
1458 *
1459 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
1460#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
1461#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
1462#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 1463
b9992569
KW
1464/* This additionally has the input parameter 'specials', which if non-zero will
1465 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 1466 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 1467#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 1468
84afefe6
JH
1469UV
1470Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1471{
a1433954
KW
1472 /* Convert the Unicode character whose ordinal is <c> to its uppercase
1473 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
1474 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
1475 * the changed version may be longer than the original character.
1476 *
1477 * The ordinal of the first character of the changed version is returned
1478 * (but note, as explained above, that there may be more.) */
1479
7918f24d
NC
1480 PERL_ARGS_ASSERT_TO_UNI_UPPER;
1481
3a4c58c9
KW
1482 if (c < 256) {
1483 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
1484 }
1485
0ebc6274 1486 uvchr_to_utf8(p, c);
b9992569 1487 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
1488}
1489
84afefe6
JH
1490UV
1491Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1492{
7918f24d
NC
1493 PERL_ARGS_ASSERT_TO_UNI_TITLE;
1494
3a4c58c9
KW
1495 if (c < 256) {
1496 return _to_upper_title_latin1((U8) c, p, lenp, 's');
1497 }
1498
0ebc6274 1499 uvchr_to_utf8(p, c);
b9992569 1500 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
1501}
1502
afc16117 1503STATIC U8
81611534 1504S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
afc16117
KW
1505{
1506 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 1507 * those, converting the result to UTF-8. Since the result is always just
a1433954 1508 * one character, we allow <p> to be NULL */
afc16117
KW
1509
1510 U8 converted = toLOWER_LATIN1(c);
1511
1512 if (p != NULL) {
6f2d5cbc 1513 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
1514 *p = converted;
1515 *lenp = 1;
1516 }
1517 else {
430c9760
KW
1518 /* Result is known to always be < 256, so can use the EIGHT_BIT
1519 * macros */
1520 *p = UTF8_EIGHT_BIT_HI(converted);
1521 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
1522 *lenp = 2;
1523 }
1524 }
1525 return converted;
1526}
1527
84afefe6
JH
1528UV
1529Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 1530{
7918f24d
NC
1531 PERL_ARGS_ASSERT_TO_UNI_LOWER;
1532
afc16117
KW
1533 if (c < 256) {
1534 return to_lower_latin1((U8) c, p, lenp);
bca00c02
KW
1535 }
1536
afc16117 1537 uvchr_to_utf8(p, c);
b9992569 1538 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
1539}
1540
84afefe6 1541UV
51910141 1542Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 1543{
51910141 1544 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 1545 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 1546 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
1547 *
1548 * Not to be used for locale folds
51910141 1549 */
f673fad4 1550
a1dde8de
KW
1551 UV converted;
1552
1553 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 1554 PERL_UNUSED_CONTEXT;
a1dde8de 1555
1ca267a5
KW
1556 assert (! (flags & FOLD_FLAGS_LOCALE));
1557
659a7c2d 1558 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
1559 converted = GREEK_SMALL_LETTER_MU;
1560 }
9b63e895
KW
1561#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
1562 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
1563 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
1564 else if ( (flags & FOLD_FLAGS_FULL)
1565 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
1566 {
1ca267a5
KW
1567 /* If can't cross 127/128 boundary, can't return "ss"; instead return
1568 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
1569 * under those circumstances. */
1570 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
1571 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
1572 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
1573 p, *lenp, U8);
1574 return LATIN_SMALL_LETTER_LONG_S;
1575 }
1576 else {
4f489194
KW
1577 *(p)++ = 's';
1578 *p = 's';
1579 *lenp = 2;
1580 return 's';
1ca267a5 1581 }
a1dde8de 1582 }
9b63e895 1583#endif
a1dde8de
KW
1584 else { /* In this range the fold of all other characters is their lower
1585 case */
1586 converted = toLOWER_LATIN1(c);
1587 }
1588
6f2d5cbc 1589 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
1590 *p = (U8) converted;
1591 *lenp = 1;
1592 }
1593 else {
1594 *(p)++ = UTF8_TWO_BYTE_HI(converted);
1595 *p = UTF8_TWO_BYTE_LO(converted);
1596 *lenp = 2;
1597 }
1598
1599 return converted;
1600}
1601
1602UV
31f05a37 1603Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 1604{
4b593389 1605
a0270393
KW
1606 /* Not currently externally documented, and subject to change
1607 * <flags> bits meanings:
1608 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
1609 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
1610 * locale are to be used.
a0270393
KW
1611 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
1612 */
4b593389 1613
36bb2ab6 1614 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 1615
780fcc9f
KW
1616 if (flags & FOLD_FLAGS_LOCALE) {
1617 /* Treat a UTF-8 locale as not being in locale at all */
1618 if (IN_UTF8_CTYPE_LOCALE) {
1619 flags &= ~FOLD_FLAGS_LOCALE;
1620 }
1621 else {
1622 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 1623 goto needs_full_generality;
780fcc9f 1624 }
31f05a37
KW
1625 }
1626
a1dde8de 1627 if (c < 256) {
e7b7ac46 1628 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 1629 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
1630 }
1631
2f306ab9 1632 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
1633 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
1634 uvchr_to_utf8(p, c);
b9992569 1635 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
1636 }
1637 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
1638 the special flags. */
1639 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
1640
1641 needs_full_generality:
a0270393 1642 uvchr_to_utf8(utf8_c, c);
445bf929 1643 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 1644 }
84afefe6
JH
1645}
1646
26483009 1647PERL_STATIC_INLINE bool
5141f98e 1648S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 1649 const char *const swashname, SV* const invlist)
bde6a22d 1650{
ea317ccb
KW
1651 /* returns a boolean giving whether or not the UTF8-encoded character that
1652 * starts at <p> is in the swash indicated by <swashname>. <swash>
1653 * contains a pointer to where the swash indicated by <swashname>
1654 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
1655 * look at <*swash> and only generate a swash if it is not null. <invlist>
1656 * is NULL or an inversion list that defines the swash. If not null, it
1657 * saves time during initialization of the swash.
ea317ccb
KW
1658 *
1659 * Note that it is assumed that the buffer length of <p> is enough to
1660 * contain all the bytes that comprise the character. Thus, <*p> should
1661 * have been checked before this call for mal-formedness enough to assure
1662 * that. */
1663
7918f24d
NC
1664 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
1665
492a624f 1666 /* The API should have included a length for the UTF-8 character in <p>,
28123549 1667 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
1668 * as far as there being enough bytes available in it to accommodate the
1669 * character without reading beyond the end, and pass that number on to the
1670 * validating routine */
6302f837 1671 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
28123549
KW
1672 if (ckWARN_d(WARN_UTF8)) {
1673 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
9816f121 1674 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
28123549
KW
1675 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
1676 what the malformation is */
1677 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
1678 }
1679 }
1680 return FALSE;
1681 }
87367d5f
KW
1682 if (!*swash) {
1683 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
1684 *swash = _core_swash_init("utf8",
1685
1686 /* Only use the name if there is no inversion
1687 * list; otherwise will go out to disk */
1688 (invlist) ? "" : swashname,
1689
1690 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 1691 }
28123549 1692
bde6a22d
NC
1693 return swash_fetch(*swash, p, TRUE) != 0;
1694}
1695
1696bool
922e8cb4
KW
1697Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
1698{
922e8cb4
KW
1699 PERL_ARGS_ASSERT__IS_UTF8_FOO;
1700
1701 assert(classnum < _FIRST_NON_SWASH_CC);
1702
f25ce844
KW
1703 return is_utf8_common(p,
1704 &PL_utf8_swash_ptrs[classnum],
1705 swash_property_names[classnum],
1706 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
1707}
1708
1709bool
f2645549 1710Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 1711{
f2645549 1712 SV* invlist = NULL;
7918f24d 1713
f2645549 1714 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 1715
f2645549
KW
1716 if (! PL_utf8_perl_idstart) {
1717 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
1718 }
60071a22 1719 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
1720}
1721
1722bool
f2645549 1723Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 1724{
f2645549 1725 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
1726
1727 if (*p == '_')
1728 return TRUE;
f25ce844 1729 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
1730}
1731
1732bool
eba68aa0
KW
1733Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
1734{
b24b43f7 1735 SV* invlist = NULL;
eba68aa0
KW
1736
1737 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
1738
b24b43f7
KW
1739 if (! PL_utf8_perl_idcont) {
1740 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
1741 }
60071a22 1742 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
1743}
1744
eba68aa0 1745bool
f2645549 1746Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 1747{
f2645549 1748 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 1749
f25ce844 1750 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
1751}
1752
1753bool
f2645549 1754Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 1755{
f2645549 1756 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 1757
f25ce844 1758 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
1759}
1760
1761bool
7dbf68d2
KW
1762Perl__is_utf8_mark(pTHX_ const U8 *p)
1763{
7dbf68d2
KW
1764 PERL_ARGS_ASSERT__IS_UTF8_MARK;
1765
f25ce844 1766 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
1767}
1768
6b5c0936 1769/*
87cea99e 1770=for apidoc to_utf8_case
6b5c0936 1771
9da1e7cb
KW
1772Instead use the appropriate one of L</toUPPER_utf8>,
1773L</toTITLE_utf8>,
1774L</toLOWER_utf8>,
1775or L</toFOLD_utf8>.
1776
6fae5207 1777C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
1778the character that is being converted. This routine assumes that the character
1779at C<p> is well-formed.
6b5c0936 1780
6fae5207
KW
1781C<ustrp> is a pointer to the character buffer to put the
1782conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
1783of the result.
1784
6fae5207 1785C<swashp> is a pointer to the swash to use.
6b5c0936 1786
a1433954 1787Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 1788and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 1789but not always, a multicharacter mapping), is tried first.
6b5c0936 1790
4a8240a3
KW
1791C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
1792any special mappings; C<""> means to use the special mappings. Values other
1793than these two are treated as the name of the hash containing the special
1794mappings, like C<"utf8::ToSpecLower">.
6b5c0936 1795
796b6530
KW
1796C<normal> is a string like C<"ToLower"> which means the swash
1797C<%utf8::ToLower>.
0134edef 1798
760c7c2f
KW
1799Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1800unless those are turned off.
1801
0134edef 1802=cut */
6b5c0936 1803
2104c8d9 1804UV
9a957fbc
AL
1805Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
1806 SV **swashp, const char *normal, const char *special)
a0ed51b3 1807{
b9992569
KW
1808 PERL_ARGS_ASSERT_TO_UTF8_CASE;
1809
1810 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
1811}
1812
1813 /* change namve uv1 to 'from' */
6a4a25f4 1814STATIC UV
b9992569
KW
1815S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
1816 SV **swashp, const char *normal, const char *special)
1817{
0134edef 1818 STRLEN len = 0;
7918f24d 1819
b9992569 1820 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 1821
36eaa811
KW
1822 /* For code points that don't change case, we already know that the output
1823 * of this function is the unchanged input, so we can skip doing look-ups
1824 * for them. Unfortunately the case-changing code points are scattered
1825 * around. But there are some long consecutive ranges where there are no
1826 * case changing code points. By adding tests, we can eliminate the lookup
1827 * for all the ones in such ranges. This is currently done here only for
1828 * just a few cases where the scripts are in common use in modern commerce
1829 * (and scripts adjacent to those which can be included without additional
1830 * tests). */
1831
1832 if (uv1 >= 0x0590) {
1833 /* This keeps from needing further processing the code points most
1834 * likely to be used in the following non-cased scripts: Hebrew,
1835 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
1836 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
1837 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
1838 if (uv1 < 0x10A0) {
1839 goto cases_to_self;
1840 }
1841
1842 /* The following largish code point ranges also don't have case
1843 * changes, but khw didn't think they warranted extra tests to speed
1844 * them up (which would slightly slow down everything else above them):
1845 * 1100..139F Hangul Jamo, Ethiopic
1846 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
1847 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
1848 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
1849 * Combining Diacritical Marks Extended, Balinese,
1850 * Sundanese, Batak, Lepcha, Ol Chiki
1851 * 2000..206F General Punctuation
1852 */
1853
1854 if (uv1 >= 0x2D30) {
1855
1856 /* This keeps the from needing further processing the code points
1857 * most likely to be used in the following non-cased major scripts:
1858 * CJK, Katakana, Hiragana, plus some less-likely scripts.
1859 *
1860 * (0x2D30 above might have to be changed to 2F00 in the unlikely
1861 * event that Unicode eventually allocates the unused block as of
1862 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
1863 * that the test suite will start having failures to alert you
1864 * should that happen) */
1865 if (uv1 < 0xA640) {
1866 goto cases_to_self;
1867 }
1868
1869 if (uv1 >= 0xAC00) {
1870 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
1871 if (ckWARN_d(WARN_SURROGATE)) {
1872 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1873 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
1874 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04"UVXf"", desc, uv1);
1875 }
1876 goto cases_to_self;
1877 }
36eaa811
KW
1878
1879 /* AC00..FAFF Catches Hangul syllables and private use, plus
1880 * some others */
1881 if (uv1 < 0xFB00) {
1882 goto cases_to_self;
1883
1884 }
1885
5af9bc97
KW
1886 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
1887 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
1888 && ckWARN_d(WARN_DEPRECATED))
1889 {
1890 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
1891 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
1892 }
1893 if (ckWARN_d(WARN_NON_UNICODE)) {
1894 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
1895 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
1896 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04"UVXf"", desc, uv1);
1897 }
1898 goto cases_to_self;
1899 }
3bfc1e70
KW
1900#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
1901 if (UNLIKELY(uv1
1902 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
1903 {
1904
1905 /* As of this writing, this means we avoid swash creation
1906 * for anything beyond low Plane 1 */
1907 goto cases_to_self;
1908 }
1909#endif
36eaa811
KW
1910 }
1911 }
9ae3ac1a 1912
36eaa811
KW
1913 /* Note that non-characters are perfectly legal, so no warning should
1914 * be given. There are so few of them, that it isn't worth the extra
1915 * tests to avoid swash creation */
9ae3ac1a
KW
1916 }
1917
0134edef 1918 if (!*swashp) /* load on-demand */
5ab9d2ef 1919 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 1920
a6f87d8c 1921 if (special) {
0134edef 1922 /* It might be "special" (sometimes, but not always,
2a37f04d 1923 * a multicharacter mapping) */
4a8240a3 1924 HV *hv = NULL;
b08cf34e
JH
1925 SV **svp;
1926
4a8240a3
KW
1927 /* If passed in the specials name, use that; otherwise use any
1928 * given in the swash */
1929 if (*special != '\0') {
1930 hv = get_hv(special, 0);
1931 }
1932 else {
1933 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
1934 if (svp) {
1935 hv = MUTABLE_HV(SvRV(*svp));
1936 }
1937 }
1938
176fe009 1939 if (hv
5f560d8a 1940 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
1941 && (*svp))
1942 {
cfd0369c 1943 const char *s;
47654450 1944
cfd0369c 1945 s = SvPV_const(*svp, len);
47654450 1946 if (len == 1)
f4cd282c 1947 /* EIGHTBIT */
c80e42f3 1948 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 1949 else {
d2dcd0fb 1950 Copy(s, ustrp, len, U8);
29e98929 1951 }
983ffd37 1952 }
0134edef
JH
1953 }
1954
1955 if (!len && *swashp) {
4a4088c4 1956 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 1957
0134edef
JH
1958 if (uv2) {
1959 /* It was "normal" (a single character mapping). */
f4cd282c 1960 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
1961 }
1962 }
1feea2c7 1963
cbe07460
KW
1964 if (len) {
1965 if (lenp) {
1966 *lenp = len;
1967 }
1968 return valid_utf8_to_uvchr(ustrp, 0);
1969 }
1970
1971 /* Here, there was no mapping defined, which means that the code point maps
1972 * to itself. Return the inputs */
e24dfe9c 1973 cases_to_self:
bfdf22ec 1974 len = UTF8SKIP(p);
ca9fab46
KW
1975 if (p != ustrp) { /* Don't copy onto itself */
1976 Copy(p, ustrp, len, U8);
1977 }
0134edef 1978
2a37f04d
JH
1979 if (lenp)
1980 *lenp = len;
1981
f4cd282c 1982 return uv1;
cbe07460 1983
a0ed51b3
LW
1984}
1985
051a06d4 1986STATIC UV
357aadde 1987S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 1988{
4a4088c4 1989 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
1990 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
1991 * result contains a character that crosses the 255/256 boundary, disallow
1992 * the change, and return the original code point. See L<perlfunc/lc> for
1993 * why;
051a06d4 1994 *
a1433954
KW
1995 * p points to the original string whose case was changed; assumed
1996 * by this routine to be well-formed
051a06d4
KW
1997 * result the code point of the first character in the changed-case string
1998 * ustrp points to the changed-case string (<result> represents its first char)
1999 * lenp points to the length of <ustrp> */
2000
2001 UV original; /* To store the first code point of <p> */
2002
2003 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2004
a4f12ed7 2005 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2006
2007 /* We know immediately if the first character in the string crosses the
2008 * boundary, so can skip */
2009 if (result > 255) {
2010
2011 /* Look at every character in the result; if any cross the
2012 * boundary, the whole thing is disallowed */
2013 U8* s = ustrp + UTF8SKIP(ustrp);
2014 U8* e = ustrp + *lenp;
2015 while (s < e) {
a4f12ed7 2016 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2017 goto bad_crossing;
2018 }
2019 s += UTF8SKIP(s);
2020 }
2021
613abc6d
KW
2022 /* Here, no characters crossed, result is ok as-is, but we warn. */
2023 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2024 return result;
2025 }
2026
7b52d656 2027 bad_crossing:
051a06d4
KW
2028
2029 /* Failed, have to return the original */
4b88fb76 2030 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2031
2032 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2033 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2034 "Can't do %s(\"\\x{%"UVXf"}\") on non-UTF-8 locale; "
2035 "resolved to \"\\x{%"UVXf"}\".",
357aadde 2036 OP_DESC(PL_op),
ab0b796c
KW
2037 original,
2038 original);
051a06d4
KW
2039 Copy(p, ustrp, *lenp, char);
2040 return original;
2041}
2042
d3e79532 2043/*
87cea99e 2044=for apidoc to_utf8_upper
d3e79532 2045
1f607577 2046Instead use L</toUPPER_utf8>.
a1433954 2047
d3e79532
JH
2048=cut */
2049
051a06d4 2050/* Not currently externally documented, and subject to change:
31f05a37
KW
2051 * <flags> is set iff iff the rules from the current underlying locale are to
2052 * be used. */
051a06d4 2053
2104c8d9 2054UV
31f05a37 2055Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2056{
051a06d4
KW
2057 UV result;
2058
2059 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2060
780fcc9f
KW
2061 if (flags) {
2062 /* Treat a UTF-8 locale as not being in locale at all */
2063 if (IN_UTF8_CTYPE_LOCALE) {
2064 flags = FALSE;
2065 }
2066 else {
2067 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2068 }
31f05a37
KW
2069 }
2070
3a4c58c9 2071 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2072 if (flags) {
2073 result = toUPPER_LC(*p);
2074 }
2075 else {
81c6c7ce 2076 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
051a06d4 2077 }
3a4c58c9
KW
2078 }
2079 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2080 if (flags) {
a62b247b 2081 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2082 result = toUPPER_LC(c);
051a06d4
KW
2083 }
2084 else {
a62b247b 2085 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2086 ustrp, lenp, 'S');
051a06d4
KW
2087 }
2088 }
4a4088c4 2089 else { /* UTF-8, ord above 255 */
b9992569 2090 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2091
2092 if (flags) {
357aadde 2093 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2094 }
2095 return result;
2096 }
2097
4a4088c4 2098 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2099 if (UTF8_IS_INVARIANT(result)) {
2100 *ustrp = (U8) result;
2101 *lenp = 1;
2102 }
2103 else {
62cb07ea
KW
2104 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2105 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2106 *lenp = 2;
3a4c58c9 2107 }
baa60164 2108
051a06d4 2109 return result;
983ffd37 2110}
a0ed51b3 2111
d3e79532 2112/*
87cea99e 2113=for apidoc to_utf8_title
d3e79532 2114
1f607577 2115Instead use L</toTITLE_utf8>.
a1433954 2116
d3e79532
JH
2117=cut */
2118
051a06d4 2119/* Not currently externally documented, and subject to change:
31f05a37
KW
2120 * <flags> is set iff the rules from the current underlying locale are to be
2121 * used. Since titlecase is not defined in POSIX, for other than a
2122 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2123 */
051a06d4 2124
983ffd37 2125UV
31f05a37 2126Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2127{
051a06d4
KW
2128 UV result;
2129
2130 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 2131
780fcc9f
KW
2132 if (flags) {
2133 /* Treat a UTF-8 locale as not being in locale at all */
2134 if (IN_UTF8_CTYPE_LOCALE) {
2135 flags = FALSE;
2136 }
2137 else {
2138 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2139 }
31f05a37
KW
2140 }
2141
3a4c58c9 2142 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2143 if (flags) {
2144 result = toUPPER_LC(*p);
2145 }
2146 else {
81c6c7ce 2147 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
051a06d4 2148 }
3a4c58c9
KW
2149 }
2150 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2151 if (flags) {
a62b247b 2152 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2153 result = toUPPER_LC(c);
051a06d4
KW
2154 }
2155 else {
a62b247b 2156 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2157 ustrp, lenp, 's');
051a06d4
KW
2158 }
2159 }
4a4088c4 2160 else { /* UTF-8, ord above 255 */
b9992569 2161 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2162
2163 if (flags) {
357aadde 2164 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4
KW
2165 }
2166 return result;
2167 }
2168
4a4088c4 2169 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2170 if (UTF8_IS_INVARIANT(result)) {
2171 *ustrp = (U8) result;
2172 *lenp = 1;
2173 }
2174 else {
62cb07ea
KW
2175 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2176 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4 2177 *lenp = 2;
3a4c58c9
KW
2178 }
2179
051a06d4 2180 return result;
a0ed51b3
LW
2181}
2182
d3e79532 2183/*
87cea99e 2184=for apidoc to_utf8_lower
d3e79532 2185
1f607577 2186Instead use L</toLOWER_utf8>.
a1433954 2187
d3e79532
JH
2188=cut */
2189
051a06d4 2190/* Not currently externally documented, and subject to change:
31f05a37
KW
2191 * <flags> is set iff iff the rules from the current underlying locale are to
2192 * be used.
2193 */
051a06d4 2194
2104c8d9 2195UV
31f05a37 2196Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2197{
051a06d4
KW
2198 UV result;
2199
051a06d4 2200 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 2201
780fcc9f
KW
2202 if (flags) {
2203 /* Treat a UTF-8 locale as not being in locale at all */
2204 if (IN_UTF8_CTYPE_LOCALE) {
2205 flags = FALSE;
2206 }
2207 else {
2208 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2209 }
31f05a37
KW
2210 }
2211
968c5e6a 2212 if (UTF8_IS_INVARIANT(*p)) {
051a06d4
KW
2213 if (flags) {
2214 result = toLOWER_LC(*p);
2215 }
2216 else {
81c6c7ce 2217 return to_lower_latin1(*p, ustrp, lenp);
051a06d4 2218 }
968c5e6a
KW
2219 }
2220 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2221 if (flags) {
a62b247b 2222 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2223 result = toLOWER_LC(c);
051a06d4
KW
2224 }
2225 else {
a62b247b 2226 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
81c6c7ce 2227 ustrp, lenp);
051a06d4 2228 }
968c5e6a 2229 }
4a4088c4 2230 else { /* UTF-8, ord above 255 */
b9992569 2231 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
051a06d4
KW
2232
2233 if (flags) {
357aadde 2234 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2235 }
968c5e6a 2236
051a06d4
KW
2237 return result;
2238 }
2239
4a4088c4 2240 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2241 if (UTF8_IS_INVARIANT(result)) {
2242 *ustrp = (U8) result;
2243 *lenp = 1;
2244 }
2245 else {
62cb07ea
KW
2246 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2247 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2248 *lenp = 2;
2249 }
2250
051a06d4 2251 return result;
b4e400f9
JH
2252}
2253
d3e79532 2254/*
87cea99e 2255=for apidoc to_utf8_fold
d3e79532 2256
1f607577 2257Instead use L</toFOLD_utf8>.
a1433954 2258
d3e79532
JH
2259=cut */
2260
051a06d4
KW
2261/* Not currently externally documented, and subject to change,
2262 * in <flags>
31f05a37
KW
2263 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2264 * locale are to be used.
051a06d4
KW
2265 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2266 * otherwise simple folds
a0270393
KW
2267 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2268 * prohibited
445bf929 2269 */
36bb2ab6 2270
b4e400f9 2271UV
445bf929 2272Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 2273{
051a06d4
KW
2274 UV result;
2275
36bb2ab6 2276 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 2277
a0270393
KW
2278 /* These are mutually exclusive */
2279 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2280
50ba90ff
KW
2281 assert(p != ustrp); /* Otherwise overwrites */
2282
780fcc9f
KW
2283 if (flags & FOLD_FLAGS_LOCALE) {
2284 /* Treat a UTF-8 locale as not being in locale at all */
2285 if (IN_UTF8_CTYPE_LOCALE) {
2286 flags &= ~FOLD_FLAGS_LOCALE;
2287 }
2288 else {
2289 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2290 }
31f05a37
KW
2291 }
2292
a1dde8de 2293 if (UTF8_IS_INVARIANT(*p)) {
051a06d4 2294 if (flags & FOLD_FLAGS_LOCALE) {
d22b930b 2295 result = toFOLD_LC(*p);
051a06d4
KW
2296 }
2297 else {
81c6c7ce 2298 return _to_fold_latin1(*p, ustrp, lenp,
1ca267a5 2299 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2300 }
a1dde8de
KW
2301 }
2302 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
051a06d4 2303 if (flags & FOLD_FLAGS_LOCALE) {
a62b247b 2304 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
68067e4e 2305 result = toFOLD_LC(c);
051a06d4
KW
2306 }
2307 else {
a62b247b 2308 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
51910141 2309 ustrp, lenp,
1ca267a5 2310 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
051a06d4 2311 }
a1dde8de 2312 }
4a4088c4 2313 else { /* UTF-8, ord above 255 */
b9992569 2314 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 2315
1ca267a5
KW
2316 if (flags & FOLD_FLAGS_LOCALE) {
2317
76f2ffcd 2318# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
2319 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
2320
2321# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
2322# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
2323
2324 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 2325
538e84ed
KW
2326 /* Special case these two characters, as what normally gets
2327 * returned under locale doesn't work */
76f2ffcd
KW
2328 if (UTF8SKIP(p) == cap_sharp_s_len
2329 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 2330 {
ab0b796c
KW
2331 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2332 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2333 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
2334 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
2335 goto return_long_s;
2336 }
0766489e
KW
2337 else
2338#endif
2339 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 2340 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 2341 {
ab0b796c
KW
2342 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2343 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2344 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
2345 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
2346 goto return_ligature_st;
2347 }
74894415
KW
2348
2349#if UNICODE_MAJOR_VERSION == 3 \
2350 && UNICODE_DOT_VERSION == 0 \
2351 && UNICODE_DOT_DOT_VERSION == 1
2352# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
2353
2354 /* And special case this on this Unicode version only, for the same
2355 * reaons the other two are special cased. They would cross the
2356 * 255/256 boundary which is forbidden under /l, and so the code
2357 * wouldn't catch that they are equivalent (which they are only in
2358 * this release) */
2359 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
2360 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
2361 {
2362 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2363 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2364 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
2365 "resolved to \"\\x{0131}\".");
2366 goto return_dotless_i;
2367 }
2368#endif
2369
357aadde 2370 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 2371 }
a0270393
KW
2372 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
2373 return result;
2374 }
2375 else {
4a4088c4 2376 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
2377 * character above the ASCII range, and the result should not
2378 * contain an ASCII character. */
a0270393
KW
2379
2380 UV original; /* To store the first code point of <p> */
2381
2382 /* Look at every character in the result; if any cross the
2383 * boundary, the whole thing is disallowed */
2384 U8* s = ustrp;
2385 U8* e = ustrp + *lenp;
2386 while (s < e) {
2387 if (isASCII(*s)) {
2388 /* Crossed, have to return the original */
2389 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 2390
9fc2026f 2391 /* But in these instances, there is an alternative we can
1ca267a5 2392 * return that is valid */
0766489e
KW
2393 if (original == LATIN_SMALL_LETTER_SHARP_S
2394#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
2395 || original == LATIN_CAPITAL_LETTER_SHARP_S
2396#endif
2397 ) {
1ca267a5
KW
2398 goto return_long_s;
2399 }
9fc2026f
KW
2400 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
2401 goto return_ligature_st;
2402 }
74894415
KW
2403#if UNICODE_MAJOR_VERSION == 3 \
2404 && UNICODE_DOT_VERSION == 0 \
2405 && UNICODE_DOT_DOT_VERSION == 1
2406
2407 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
2408 goto return_dotless_i;
2409 }
2410#endif
a0270393
KW
2411 Copy(p, ustrp, *lenp, char);
2412 return original;
2413 }
2414 s += UTF8SKIP(s);
2415 }
051a06d4 2416
a0270393
KW
2417 /* Here, no characters crossed, result is ok as-is */
2418 return result;
2419 }
051a06d4
KW
2420 }
2421
4a4088c4 2422 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
2423 if (UTF8_IS_INVARIANT(result)) {
2424 *ustrp = (U8) result;
2425 *lenp = 1;
2426 }
2427 else {
62cb07ea
KW
2428 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2429 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
2430 *lenp = 2;
2431 }
2432
051a06d4 2433 return result;
1ca267a5
KW
2434
2435 return_long_s:
2436 /* Certain folds to 'ss' are prohibited by the options, but they do allow
2437 * folds to a string of two of these characters. By returning this
2438 * instead, then, e.g.,
2439 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
2440 * works. */
2441
2442 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2443 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2444 ustrp, *lenp, U8);
2445 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
2446
2447 return_ligature_st:
2448 /* Two folds to 'st' are prohibited by the options; instead we pick one and
2449 * have the other one fold to it */
2450
2451 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
2452 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
2453 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
2454
2455#if UNICODE_MAJOR_VERSION == 3 \
2456 && UNICODE_DOT_VERSION == 0 \
2457 && UNICODE_DOT_DOT_VERSION == 1
2458
2459 return_dotless_i:
2460 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
2461 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
2462 return LATIN_SMALL_LETTER_DOTLESS_I;
2463
2464#endif
2465
a0ed51b3
LW
2466}
2467
711a919c 2468/* Note:
f90a9a02 2469 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
2470 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
2471 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
2472 */
c4a5db0c 2473
a0ed51b3 2474SV*
7fc63493 2475Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 2476{
c4a5db0c
KW
2477 PERL_ARGS_ASSERT_SWASH_INIT;
2478
2479 /* Returns a copy of a swash initiated by the called function. This is the
2480 * public interface, and returning a copy prevents others from doing
2481 * mischief on the original */
2482
5d3d13d1 2483 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
2484}
2485
2486SV*
5d3d13d1 2487Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 2488{
2c1f00b9
YO
2489
2490 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
2491 * use the following define */
2492
2493#define CORE_SWASH_INIT_RETURN(x) \
2494 PL_curpm= old_PL_curpm; \
2495 return x
2496
c4a5db0c 2497 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
2498 * by calling utf8_heavy.pl in the general case. The returned value may be
2499 * the swash's inversion list instead if the input parameters allow it.
2500 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
2501 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
2502 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
2503 *
2504 * This interface should only be used by functions that won't destroy or
2505 * adversely change the swash, as doing so affects all other uses of the
2506 * swash in the program; the general public should use 'Perl_swash_init'
2507 * instead.
2508 *
2509 * pkg is the name of the package that <name> should be in.
2510 * name is the name of the swash to find. Typically it is a Unicode
2511 * property name, including user-defined ones
2512 * listsv is a string to initialize the swash with. It must be of the form
2513 * documented as the subroutine return value in
2514 * L<perlunicode/User-Defined Character Properties>
2515 * minbits is the number of bits required to represent each data element.
2516 * It is '1' for binary properties.
2517 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 2518 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
2519 * flags_p if non-NULL is the address of various input and output flag bits
2520 * to the routine, as follows: ('I' means is input to the routine;
2521 * 'O' means output from the routine. Only flags marked O are
2522 * meaningful on return.)
2523 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
2524 * came from a user-defined property. (I O)
5d3d13d1
KW
2525 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
2526 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
2527 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
2528 * return of an inversion list instead of a swash hash if this routine
2529 * thinks that would result in faster execution of swash_fetch() later
2530 * on. (I)
9a53f6cf
KW
2531 *
2532 * Thus there are three possible inputs to find the swash: <name>,
2533 * <listsv>, and <invlist>. At least one must be specified. The result
2534 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
2535 * actions can intersect, etc. what <name> gives. To avoid going out to
2536 * disk at all, <invlist> should specify completely what the swash should
2537 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
2538 *
2539 * <invlist> is only valid for binary properties */
c4a5db0c 2540
2c1f00b9
YO
2541 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
2542
c4a5db0c 2543 SV* retval = &PL_sv_undef;
83199d38 2544 HV* swash_hv = NULL;
87367d5f
KW
2545 const int invlist_swash_boundary =
2546 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
2547 ? 512 /* Based on some benchmarking, but not extensive, see commit
2548 message */
2549 : -1; /* Never return just an inversion list */
9a53f6cf
KW
2550
2551 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
2552 assert(! invlist || minbits == 1);
2553
2c1f00b9
YO
2554 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
2555 that triggered the swash init and the swash init perl logic itself.
2556 See perl #122747 */
2557
9a53f6cf
KW
2558 /* If data was passed in to go out to utf8_heavy to find the swash of, do
2559 * so */
2560 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
2561 dSP;
2562 const size_t pkg_len = strlen(pkg);
2563 const size_t name_len = strlen(name);
2564 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
2565 SV* errsv_save;
2566 GV *method;
2567
2568 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
2569
2570 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 2571 ENTER;
69794297 2572 SAVEHINTS();
2782061f 2573 save_re_context();
650f067c
JL
2574 /* We might get here via a subroutine signature which uses a utf8
2575 * parameter name, at which point PL_subname will have been set
2576 * but not yet used. */
2577 save_item(PL_subname);
69794297
KW
2578 if (PL_parser && PL_parser->error_count)
2579 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
2580 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 2581 if (!method) { /* demand load UTF-8 */
69794297 2582 ENTER;
db2c6cb3
FC
2583 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2584 GvSV(PL_errgv) = NULL;
1a419e6b 2585#ifndef NO_TAINT_SUPPORT
69794297
KW
2586 /* It is assumed that callers of this routine are not passing in
2587 * any user derived data. */
2782061f
DM
2588 /* Need to do this after save_re_context() as it will set
2589 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
2590 * in Perl_magic_get). Even line to create errsv_save can turn on
2591 * PL_tainted. */
284167a5
SM
2592 SAVEBOOL(TAINT_get);
2593 TAINT_NOT;
2594#endif
69794297
KW
2595 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
2596 NULL);
eed484f9 2597 {
db2c6cb3
FC
2598 /* Not ERRSV, as there is no need to vivify a scalar we are
2599 about to discard. */
2600 SV * const errsv = GvSV(PL_errgv);
2601 if (!SvTRUE(errsv)) {
2602 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2603 SvREFCNT_dec(errsv);
2604 }
eed484f9 2605 }
69794297
KW
2606 LEAVE;
2607 }
2608 SPAGAIN;
2609 PUSHMARK(SP);
2610 EXTEND(SP,5);
2611 mPUSHp(pkg, pkg_len);
2612 mPUSHp(name, name_len);
2613 PUSHs(listsv);
2614 mPUSHi(minbits);
2615 mPUSHi(none);
2616 PUTBACK;
db2c6cb3
FC
2617 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
2618 GvSV(PL_errgv) = NULL;
69794297
KW
2619 /* If we already have a pointer to the method, no need to use
2620 * call_method() to repeat the lookup. */
c41800a8
KW
2621 if (method
2622 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
2623 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
2624 {
2625 retval = *PL_stack_sp--;
2626 SvREFCNT_inc(retval);
2627 }
eed484f9 2628 {
db2c6cb3
FC
2629 /* Not ERRSV. See above. */
2630 SV * const errsv = GvSV(PL_errgv);
2631 if (!SvTRUE(errsv)) {
2632 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
2633 SvREFCNT_dec(errsv);
2634 }
eed484f9 2635 }
ce3b816e 2636 LEAVE;
69794297
KW
2637 POPSTACK;
2638 if (IN_PERL_COMPILETIME) {
2639 CopHINTS_set(PL_curcop, PL_hints);
2640 }
2641 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 2642 if (SvPOK(retval)) {
69794297
KW
2643
2644 /* If caller wants to handle missing properties, let them */
5d3d13d1 2645 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 2646 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
2647 }
2648 Perl_croak(aTHX_
2649 "Can't find Unicode property definition \"%"SVf"\"",
2650 SVfARG(retval));
a25b5927 2651 NOT_REACHED; /* NOTREACHED */
d95e4a00 2652 }
69794297 2653 }
9a53f6cf 2654 } /* End of calling the module to find the swash */
36eb48b4 2655
83199d38
KW
2656 /* If this operation fetched a swash, and we will need it later, get it */
2657 if (retval != &PL_sv_undef
2658 && (minbits == 1 || (flags_p
2659 && ! (*flags_p
2660 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
2661 {
2662 swash_hv = MUTABLE_HV(SvRV(retval));
2663
2664 /* If we don't already know that there is a user-defined component to
2665 * this swash, and the user has indicated they wish to know if there is
2666 * one (by passing <flags_p>), find out */
2667 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
2668 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
2669 if (user_defined && SvUV(*user_defined)) {
2670 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
2671 }
2672 }
2673 }
2674
36eb48b4
KW
2675 /* Make sure there is an inversion list for binary properties */
2676 if (minbits == 1) {
2677 SV** swash_invlistsvp = NULL;
2678 SV* swash_invlist = NULL;
9a53f6cf 2679 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
2680 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
2681 an unclaimed reference count */
36eb48b4 2682
9a53f6cf 2683 /* If this operation fetched a swash, get its already existing
83199d38 2684 * inversion list, or create one for it */
36eb48b4 2685
83199d38 2686 if (swash_hv) {
5c9f4bd2 2687 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
2688 if (swash_invlistsvp) {
2689 swash_invlist = *swash_invlistsvp;
2690 invlist_in_swash_is_valid = TRUE;
2691 }
2692 else {
36eb48b4 2693 swash_invlist = _swash_to_invlist(retval);
02c85471 2694 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
2695 }
2696 }
2697
2698 /* If an inversion list was passed in, have to include it */
2699 if (invlist) {
2700
2701 /* Any fetched swash will by now have an inversion list in it;
2702 * otherwise <swash_invlist> will be NULL, indicating that we
2703 * didn't fetch a swash */
2704 if (swash_invlist) {
2705
2706 /* Add the passed-in inversion list, which invalidates the one
2707 * already stored in the swash */
2708 invlist_in_swash_is_valid = FALSE;
2709 _invlist_union(invlist, swash_invlist, &swash_invlist);
2710 }
2711 else {
2712
87367d5f
KW
2713 /* Here, there is no swash already. Set up a minimal one, if
2714 * we are going to return a swash */
2715 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 2716 swash_hv = newHV();
4aca0fe6 2717 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 2718 }
9a53f6cf
KW
2719 swash_invlist = invlist;
2720 }
9a53f6cf
KW
2721 }
2722
2723 /* Here, we have computed the union of all the passed-in data. It may
2724 * be that there was an inversion list in the swash which didn't get
538e84ed 2725 * touched; otherwise save the computed one */
87367d5f
KW
2726 if (! invlist_in_swash_is_valid
2727 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
2728 {
5c9f4bd2 2729 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
2730 {
2731 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2732 }
cc34d8c5
FC
2733 /* We just stole a reference count. */
2734 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
2735 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 2736 }
87367d5f 2737
dbfdbd26
KW
2738 SvREADONLY_on(swash_invlist);
2739
c41800a8 2740 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
2741 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
2742 SvREFCNT_dec(retval);
02c85471
FC
2743 if (!swash_invlist_unclaimed)
2744 SvREFCNT_inc_simple_void_NN(swash_invlist);
2745 retval = newRV_noinc(swash_invlist);
87367d5f 2746 }
36eb48b4
KW
2747 }
2748
2c1f00b9
YO
2749 CORE_SWASH_INIT_RETURN(retval);
2750#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
2751}
2752
035d37be
JH
2753
2754/* This API is wrong for special case conversions since we may need to
2755 * return several Unicode characters for a single Unicode character
2756 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
2757 * the lower-level routine, and it is similarly broken for returning
38684baa 2758 * multiple values. --jhi
b9992569 2759 * For those, you should use S__to_utf8_case() instead */
b0e3252e 2760/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
2761
2762/* Note:
2763 * Returns the value of property/mapping C<swash> for the first character
2764 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 2765 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 2766 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
2767 *
2768 * A "swash" is a hash which contains initially the keys/values set up by
2769 * SWASHNEW. The purpose is to be able to completely represent a Unicode
2770 * property for all possible code points. Things are stored in a compact form
2771 * (see utf8_heavy.pl) so that calculation is required to find the actual
2772 * property value for a given code point. As code points are looked up, new
2773 * key/value pairs are added to the hash, so that the calculation doesn't have
2774 * to ever be re-done. Further, each calculation is done, not just for the
2775 * desired one, but for a whole block of code points adjacent to that one.
2776 * For binary properties on ASCII machines, the block is usually for 64 code
2777 * points, starting with a code point evenly divisible by 64. Thus if the
2778 * property value for code point 257 is requested, the code goes out and
2779 * calculates the property values for all 64 code points between 256 and 319,
2780 * and stores these as a single 64-bit long bit vector, called a "swatch",
2781 * under the key for code point 256. The key is the UTF-8 encoding for code
2782 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
2783 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
2784 * for code point 258 is then requested, this code realizes that it would be
2785 * stored under the key for 256, and would find that value and extract the
2786 * relevant bit, offset from 256.
2787 *
2788 * Non-binary properties are stored in as many bits as necessary to represent
2789 * their values (32 currently, though the code is more general than that), not
fc273927 2790 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
2791 * vector that encompasses the property values for all code points whose UTF-8
2792 * representations are represented by the key. That is, for all code points
2793 * whose UTF-8 representations are length N bytes, and the key is the first N-1
2794 * bytes of that.
680c470c 2795 */
a0ed51b3 2796UV
680c470c 2797Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 2798{
ef8f7699 2799 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
2800 U32 klen;
2801 U32 off;
9b56a019 2802 STRLEN slen = 0;
7d85a32c 2803 STRLEN needents;
cfd0369c 2804 const U8 *tmps = NULL;
979f2922 2805 SV *swatch;
08fb1ac5 2806 const U8 c = *ptr;
3568d838 2807
7918f24d
NC
2808 PERL_ARGS_ASSERT_SWASH_FETCH;
2809
87367d5f
KW
2810 /* If it really isn't a hash, it isn't really swash; must be an inversion
2811 * list */
2812 if (SvTYPE(hv) != SVt_PVHV) {
2813 return _invlist_contains_cp((SV*)hv,
2814 (do_utf8)
2815 ? valid_utf8_to_uvchr(ptr, NULL)
2816 : c);
2817 }
2818
08fb1ac5
KW
2819 /* We store the values in a "swatch" which is a vec() value in a swash
2820 * hash. Code points 0-255 are a single vec() stored with key length
2821 * (klen) 0. All other code points have a UTF-8 representation
2822 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
2823 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
2824 * length for them is the length of the encoded char - 1. ptr[klen] is the
2825 * final byte in the sequence representing the character */
2826 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
2827 klen = 0;
2828 needents = 256;
2829 off = c;
3568d838 2830 }
08fb1ac5
KW
2831 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2832 klen = 0;
2833 needents = 256;
a62b247b 2834 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
2835 }
2836 else {
08fb1ac5
KW
2837 klen = UTF8SKIP(ptr) - 1;
2838
2839 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
2840 * the vec is the final byte in the sequence. (In EBCDIC this is
2841 * converted to I8 to get consecutive values.) To help you visualize
2842 * all this:
2843 * Straight 1047 After final byte
2844 * UTF-8 UTF-EBCDIC I8 transform
2845 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
2846 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
2847 * ...
2848 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
2849 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
2850 * ...
2851 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
2852 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
2853 * ...
2854 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
2855 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
2856 * ...
2857 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
2858 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
2859 *
2860 * (There are no discontinuities in the elided (...) entries.)
2861 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
2862 * key for the next 31, up through U+043F, whose UTF-8 final byte is
2863 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
2864 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
2865 * index into the vec() swatch (after subtracting 0x80, which we
2866 * actually do with an '&').
2867 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
2868 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
2869 * dicontinuities which go away by transforming it into I8, and we
2870 * effectively subtract 0xA0 to get the index. */
979f2922 2871 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 2872 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 2873 }
7d85a32c 2874
a0ed51b3 2875 /*
4a4088c4 2876 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
2877 * suite. (That is, only 7-8% overall over just a hash cache. Still,
2878 * it's nothing to sniff at.) Pity we usually come through at least
2879 * two function calls to get here...
2880 *
2881 * NB: this code assumes that swatches are never modified, once generated!
2882 */
2883
3568d838 2884 if (hv == PL_last_swash_hv &&
a0ed51b3 2885 klen == PL_last_swash_klen &&
27da23d5 2886 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
2887 {
2888 tmps = PL_last_swash_tmps;
2889 slen = PL_last_swash_slen;
2890 }
2891 else {
2892 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 2893 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 2894
b0e3252e 2895 /* If not cached, generate it via swatch_get */
979f2922 2896 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
2897 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
2898 {
2899 if (klen) {
2900 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
2901 swatch = swatch_get(swash,
2902 code_point & ~((UV)needents - 1),
2903 needents);
2904 }
2905 else { /* For the first 256 code points, the swatch has a key of
2906 length 0 */
2907 swatch = swatch_get(swash, 0, needents);
2908 }
979f2922 2909
923e4eb5 2910 if (IN_PERL_COMPILETIME)
623e6609 2911 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 2912
979f2922 2913 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 2914
979f2922
ST
2915 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
2916 || (slen << 3) < needents)
5637ef5b
NC
2917 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
2918 "svp=%p, tmps=%p, slen=%"UVuf", needents=%"UVuf,
2919 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
2920 }
2921
2922 PL_last_swash_hv = hv;
16d8f38a 2923 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 2924 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
2925 /* FIXME change interpvar.h? */
2926 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
2927 PL_last_swash_slen = slen;
2928 if (klen)
2929 Copy(ptr, PL_last_swash_key, klen, U8);
2930 }
2931
9faf8d75 2932 switch ((int)((slen << 3) / needents)) {
a0ed51b3 2933 case 1:
e7aca353 2934 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 2935 case 8:
e7aca353 2936 return ((UV) tmps[off]);
a0ed51b3
LW
2937 case 16:
2938 off <<= 1;
e7aca353
JH
2939 return
2940 ((UV) tmps[off ] << 8) +
2941 ((UV) tmps[off + 1]);
a0ed51b3
LW
2942 case 32:
2943 off <<= 2;
e7aca353
JH
2944 return
2945 ((UV) tmps[off ] << 24) +
2946 ((UV) tmps[off + 1] << 16) +
2947 ((UV) tmps[off + 2] << 8) +
2948 ((UV) tmps[off + 3]);
a0ed51b3 2949 }
5637ef5b
NC
2950 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
2951 "slen=%"UVuf", needents=%"UVuf, (UV)slen, (UV)needents);
670f1322 2952 NORETURN_FUNCTION_END;
a0ed51b3 2953}
2b9d42f0 2954
319009ee
KW
2955/* Read a single line of the main body of the swash input text. These are of
2956 * the form:
2957 * 0053 0056 0073
2958 * where each number is hex. The first two numbers form the minimum and
2959 * maximum of a range, and the third is the value associated with the range.
2960 * Not all swashes should have a third number
2961 *
2962 * On input: l points to the beginning of the line to be examined; it points
2963 * to somewhere in the string of the whole input text, and is
2964 * terminated by a \n or the null string terminator.
2965 * lend points to the null terminator of that string
2966 * wants_value is non-zero if the swash expects a third number
2967 * typestr is the name of the swash's mapping, like 'ToLower'
2968 * On output: *min, *max, and *val are set to the values read from the line.
2969 * returns a pointer just beyond the line examined. If there was no
2970 * valid min number on the line, returns lend+1
2971 */
2972
2973STATIC U8*
2974S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
2975 const bool wants_value, const U8* const typestr)
2976{
2977 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
2978 STRLEN numlen; /* Length of the number */
02470786
KW
2979 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
2980 | PERL_SCAN_DISALLOW_PREFIX
2981 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
2982
2983 /* nl points to the next \n in the scan */
2984 U8* const nl = (U8*)memchr(l, '\n', lend - l);
2985
95543e92
KW
2986 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
2987
319009ee
KW
2988 /* Get the first number on the line: the range minimum */
2989 numlen = lend - l;
2990 *min = grok_hex((char *)l, &numlen, &flags, NULL);
c88850db 2991 *max = *min; /* So can never return without setting max */
319009ee
KW
2992 if (numlen) /* If found a hex number, position past it */
2993 l += numlen;
2994 else if (nl) { /* Else, go handle next line, if any */
2995 return nl + 1; /* 1 is length of "\n" */
2996 }
2997 else { /* Else, no next line */
2998 return lend + 1; /* to LIST's end at which \n is not found */
2999 }
3000
3001 /* The max range value follows, separated by a BLANK */
3002 if (isBLANK(*l)) {
3003 ++l;
02470786
KW
3004 flags = PERL_SCAN_SILENT_ILLDIGIT
3005 | PERL_SCAN_DISALLOW_PREFIX
3006 | PERL_SCAN_SILENT_NON_PORTABLE;
319009ee
KW
3007 numlen = lend - l;
3008 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3009 if (numlen)
3010 l += numlen;
3011 else /* If no value here, it is a single element range */
3012 *max = *min;
3013
3014 /* Non-binary tables have a third entry: what the first element of the
24303724 3015 * range maps to. The map for those currently read here is in hex */
319009ee
KW
3016 if (wants_value) {
3017 if (isBLANK(*l)) {
3018 ++l;
f2a7d0fc
KW
3019 flags = PERL_SCAN_SILENT_ILLDIGIT
3020 | PERL_SCAN_DISALLOW_PREFIX
3021 | PERL_SCAN_SILENT_NON_PORTABLE;
3022 numlen = lend - l;
3023 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3024 if (numlen)
3025 l += numlen;
3026 else
3027 *val = 0;
319009ee
KW
3028 }
3029 else {
3030 *val = 0;
3031 if (typeto) {
dcbac5bb 3032 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3033 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3034 typestr, l);
3035 }
3036 }
3037 }
3038 else
3039 *val = 0; /* bits == 1, then any val should be ignored */
3040 }
3041 else { /* Nothing following range min, should be single element with no
3042 mapping expected */
319009ee
KW
3043 if (wants_value) {
3044 *val = 0;
3045 if (typeto) {
dcbac5bb 3046 /* diag_listed_as: To%s: illegal mapping '%s' */
319009ee
KW
3047 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3048 }
3049 }
3050 else
3051 *val = 0; /* bits == 1, then val should be ignored */
3052 }
3053
3054 /* Position to next line if any, or EOF */
3055 if (nl)
3056 l = nl + 1;
3057 else
3058 l = lend;
3059
3060 return l;
3061}
3062
979f2922
ST
3063/* Note:
3064 * Returns a swatch (a bit vector string) for a code point sequence
3065 * that starts from the value C<start> and comprises the number C<span>.
3066 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3067 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3068 */
3069STATIC SV*
b0e3252e 3070S_swatch_get(pTHX_ SV* swash, UV start, UV span)
979f2922
ST
3071{
3072 SV *swatch;
77f9f126 3073 U8 *l, *lend, *x, *xend, *s, *send;
979f2922 3074 STRLEN lcur, xcur, scur;
ef8f7699 3075 HV *const hv = MUTABLE_HV(SvRV(swash));
5c9f4bd2 3076 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
36eb48b4 3077
88d45d28
KW
3078 SV** listsvp = NULL; /* The string containing the main body of the table */
3079 SV** extssvp = NULL;
3080 SV** invert_it_svp = NULL;
3081 U8* typestr = NULL;
786861f5
KW
3082 STRLEN bits;
3083 STRLEN octets; /* if bits == 1, then octets == 0 */
3084 UV none;
3085 UV end = start + span;
972dd592 3086
36eb48b4 3087 if (invlistsvp == NULL) {
786861f5
KW
3088 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3089 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3090 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3091 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3092 listsvp = hv_fetchs(hv, "LIST", FALSE);
3093 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3094
3095 bits = SvUV(*bitssvp);
3096 none = SvUV(*nonesvp);
3097 typestr = (U8*)SvPV_nolen(*typesvp);
3098 }
36eb48b4
KW
3099 else {
3100 bits = 1;
3101 none = 0;
3102 }
786861f5 3103 octets = bits >> 3; /* if bits == 1, then octets == 0 */
979f2922 3104
b0e3252e 3105 PERL_ARGS_ASSERT_SWATCH_GET;
7918f24d 3106
979f2922 3107 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
b0e3252e 3108 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %"UVuf,
660a4616 3109 (UV)bits);
979f2922
ST
3110 }
3111
84ea5ef6
KW
3112 /* If overflowed, use the max possible */
3113 if (end < start) {
3114 end = UV_MAX;
3115 span = end - start;
3116 }
3117
979f2922 3118 /* create and initialize $swatch */
979f2922 3119 scur = octets ? (span * octets) : (span + 7) / 8;
e524fe40
NC
3120 swatch = newSV(scur);
3121 SvPOK_on(swatch);
979f2922
ST
3122 s = (U8*)SvPVX(swatch);
3123 if (octets && none) {
0bd48802 3124 const U8* const e = s + scur;
979f2922
ST
3125 while (s < e) {
3126 if (bits == 8)
3127 *s++ = (U8)(none & 0xff);
3128 else if (bits == 16) {
3129 *s++ = (U8)((none >> 8) & 0xff);
3130 *s++ = (U8)( none & 0xff);
3131 }
3132 else if (bits == 32) {
3133 *s++ = (U8)((none >> 24) & 0xff);
3134 *s++ = (U8)((none >> 16) & 0xff);
3135 *s++ = (U8)((none >> 8) & 0xff);
3136 *s++ = (U8)( none & 0xff);
3137 }
3138 }
3139 *s = '\0';
3140 }
3141 else {
3142 (void)memzero((U8*)s, scur + 1);
3143 }
3144 SvCUR_set(swatch, scur);
3145 s = (U8*)SvPVX(swatch);
3146
36eb48b4
KW
3147 if (invlistsvp) { /* If has an inversion list set up use that */
3148 _invlist_populate_swatch(*invlistsvp, start, end, s);
3149 return swatch;
3150 }
3151
3152 /* read $swash->{LIST} */
979f2922
ST
3153 l = (U8*)SvPV(*listsvp, lcur);
3154 lend = l + lcur;
3155 while (l < lend) {
8ed25d53 3156 UV min, max, val, upper;
95543e92
KW
3157 l = swash_scan_list_line(l, lend, &min, &max, &val,
3158 cBOOL(octets), typestr);
319009ee 3159 if (l > lend) {
979f2922
ST
3160 break;
3161 }
3162
972dd592 3163 /* If looking for something beyond this range, go try the next one */
979f2922
ST
3164 if (max < start)
3165 continue;
3166
8ed25d53
KW
3167 /* <end> is generally 1 beyond where we want to set things, but at the
3168 * platform's infinity, where we can't go any higher, we want to
3169 * include the code point at <end> */
3170 upper = (max < end)
3171 ? max
3172 : (max != UV_MAX || end != UV_MAX)
3173 ? end - 1
3174 : end;
3175
979f2922 3176 if (octets) {
35da51f7 3177 UV key;
979f2922
ST
3178 if (min < start) {
3179 if (!none || val < none) {
3180 val += start - min;
3181 }
3182 min = start;
3183 }
8ed25d53 3184 for (key = min; key <= upper; key++) {
979f2922 3185 STRLEN offset;
979f2922
ST
3186 /* offset must be non-negative (start <= min <= key < end) */
3187 offset = octets * (key - start);
3188 if (bits == 8)
3189 s[offset] = (U8)(val & 0xff);
3190 else if (bits == 16) {
3191 s[offset ] = (U8)((val >> 8) & 0xff);
3192 s[offset + 1] = (U8)( val & 0xff);
3193 }
3194 else if (bits == 32) {
3195 s[offset ] = (U8)((val >> 24) & 0xff);
3196 s[offset + 1] = (U8)((val >> 16) & 0xff);
3197 s[offset + 2] = (U8)((val >> 8) & 0xff);
3198 s[offset + 3] = (U8)( val & 0xff);
3199 }
3200
3201 if (!none || val < none)
3202 ++val;
3203 }
3204 }
711a919c 3205 else { /* bits == 1, then val should be ignored */
35da51f7 3206 UV key;
979f2922
ST
3207 if (min < start)
3208 min = start;
6cb05c12 3209
8ed25d53 3210 for (key = min; key <= upper; key++) {
0bd48802 3211 const STRLEN offset = (STRLEN)(key - start);
979f2922
ST
3212 s[offset >> 3] |= 1 << (offset & 7);
3213 }
3214 }
3215 } /* while */
979f2922 3216
9479a769 3217 /* Invert if the data says it should be. Assumes that bits == 1 */
77f9f126 3218 if (invert_it_svp && SvUV(*invert_it_svp)) {
0bda3001
KW
3219
3220 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3221 * be 0, and their inversion should also be 0, as we don't succeed any
3222 * Unicode property matches for non-Unicode code points */
3223 if (start <= PERL_UNICODE_MAX) {
3224
3225 /* The code below assumes that we never cross the
3226 * Unicode/above-Unicode boundary in a range, as otherwise we would
3227 * have to figure out where to stop flipping the bits. Since this
3228 * boundary is divisible by a large power of 2, and swatches comes
3229 * in small powers of 2, this should be a valid assumption */
3230 assert(start + span - 1 <= PERL_UNICODE_MAX);
3231
507a8485
KW
3232 send = s + scur;
3233 while (s < send) {
3234 *s = ~(*s);
3235 s++;
3236 }
0bda3001 3237 }
77f9f126
KW
3238 }
3239
d73c39c5
KW
3240 /* read $swash->{EXTRAS}
3241 * This code also copied to swash_to_invlist() below */
979f2922
ST
3242 x = (U8*)SvPV(*extssvp, xcur);
3243 xend = x + xcur;
3244 while (x < xend) {
3245 STRLEN namelen;
3246 U8 *namestr;
3247 SV** othersvp;
3248 HV* otherhv;
3249 STRLEN otherbits;
3250 SV **otherbitssvp, *other;
711a919c 3251 U8 *s, *o, *nl;
979f2922
ST
3252 STRLEN slen, olen;
3253
35da51f7 3254 const U8 opc = *x++;
979f2922
ST
3255 if (opc == '\n')
3256 continue;
3257
3258 nl = (U8*)memchr(x, '\n', xend - x);
3259
3260 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3261 if (nl) {
3262 x = nl + 1; /* 1 is length of "\n" */
3263 continue;
3264 }
3265 else {
3266 x = xend; /* to EXTRAS' end at which \n is not found */
3267 break;
3268 }
3269 }
3270
3271 namestr = x;
3272 if (nl) {
3273 namelen = nl - namestr;
3274 x = nl + 1;
3275 }
3276 else {
3277 namelen = xend - namestr;
3278 x = xend;
3279 }
3280
3281 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
ef8f7699 3282 otherhv = MUTABLE_HV(SvRV(*othersvp));
017a3ce5 3283 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
979f2922
ST
3284 otherbits = (STRLEN)SvUV(*otherbitssvp);
3285 if (bits < otherbits)
5637ef5b
NC
3286 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3287 "bits=%"UVuf", otherbits=%"UVuf, (UV)bits, (UV)otherbits);
979f2922
ST
3288
3289 /* The "other" swatch must be destroyed after. */
b0e3252e 3290 other = swatch_get(*othersvp, start, span);
979f2922
ST
3291 o = (U8*)SvPV(other, olen);
3292
3293 if (!olen)
b0e3252e 3294 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
979f2922
ST
3295
3296 s = (U8*)SvPV(swatch, slen);
3297 if (bits == 1 && otherbits == 1) {
3298 if (slen != olen)
5637ef5b
NC
3299 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
3300 "mismatch, slen=%"UVuf", olen=%"UVuf,
3301 (UV)slen, (UV)olen);
979f2922
ST
3302
3303 switch (opc) {
3304 case '+':
3305 while (slen--)
3306 *s++ |= *o++;
3307 break;
3308 case '!':
3309 while (slen--)
3310 *s++ |= ~*o++;
3311 break;
3312 case '-':
3313 while (slen--)
3314 *s++ &= ~*o++;
3315 break;
3316 case '&':
3317 while (slen--)
3318 *s++ &= *o++;
3319 break;
3320 default:
3321 break;
3322 }
3323 }
711a919c 3324 else {
979f2922
ST
3325 STRLEN otheroctets = otherbits >> 3;
3326 STRLEN offset = 0;
35da51f7 3327 U8* const send = s + slen;
979f2922
ST
3328
3329 while (s < send) {
3330 UV otherval = 0;
3331
3332 if (otherbits == 1) {
3333 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
3334 ++offset;
3335 }
3336 else {
3337 STRLEN vlen = otheroctets;
3338 otherval = *o++;
3339 while (--vlen) {
3340 otherval <<= 8;
3341 otherval |= *o++;
3342 }
3343 }
3344
711a919c 3345 if (opc == '+' && otherval)
6f207bd3 3346 NOOP; /* replace with otherval */
979f2922
ST
3347 else if (opc == '!' && !otherval)
3348 otherval = 1;
3349 else if (opc == '-' && otherval)
3350 otherval = 0;
3351 else if (opc == '&' && !otherval)
3352 otherval = 0;
3353 else {
711a919c 3354 s += octets; /* no replacement */
979f2922
ST
3355 continue;
3356 }
3357
3358 if (bits == 8)
3359 *s++ = (U8)( otherval & 0xff);
3360 else if (bits == 16) {
3361 *s++ = (U8)((otherval >> 8) & 0xff);
3362 *s++ = (U8)( otherval & 0xff);
3363 }
3364 else if (bits == 32) {
3365 *s++ = (U8)((otherval >> 24) & 0xff);
3366 *s++ = (U8)((otherval >> 16) & 0xff);
3367 *s++ = (U8)((otherval >> 8) & 0xff);
3368 *s++ = (U8)( otherval & 0xff);
3369 }
3370 }
3371 }
3372 sv_free(other); /* through with it! */
3373 } /* while */
3374 return swatch;
3375}
3376
064c021d 3377HV*
4c2e1131 3378Perl__swash_inversion_hash(pTHX_ SV* const swash)
064c021d
KW
3379{
3380
79a2a0e8 3381 /* Subject to change or removal. For use only in regcomp.c and regexec.c
5662e334
KW
3382 * Can't be used on a property that is subject to user override, as it
3383 * relies on the value of SPECIALS in the swash which would be set by
3384 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
3385 * for overridden properties
064c021d
KW
3386 *
3387 * Returns a hash which is the inversion and closure of a swash mapping.
3388 * For example, consider the input lines:
3389 * 004B 006B
3390 * 004C 006C
3391 * 212A 006B
3392 *
4a4088c4 3393 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
064c021d 3394 * 006C. The value for each key is an array. For 006C, the array would
4a4088c4
KW
3395 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
3396 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
064c021d 3397 *
538e84ed
KW
3398 * Note that there are no elements in the hash for 004B, 004C, 212A. The
3399 * keys are only code points that are folded-to, so it isn't a full closure.
3400 *
064c021d
KW
3401 * Essentially, for any code point, it gives all the code points that map to
3402 * it, or the list of 'froms' for that point.
3403 *
5662e334
KW
3404 * Currently it ignores any additions or deletions from other swashes,
3405 * looking at just the main body of the swash, and if there are SPECIALS
3406 * in the swash, at that hash
3407 *
3408 * The specials hash can be extra code points, and most likely consists of
3409 * maps from single code points to multiple ones (each expressed as a string
4a4088c4 3410 * of UTF-8 characters). This function currently returns only 1-1 mappings.
5662e334
KW
3411 * However consider this possible input in the specials hash:
3412 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
3413 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
3414 *
3415 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
3416 * currently handle. But it also means that FB05 and FB06 are equivalent in
3417 * a 1-1 mapping which we should handle, and this relationship may not be in
3418 * the main table. Therefore this function examines all the multi-char
74894415
KW
3419 * sequences and adds the 1-1 mappings that come out of that.
3420 *
3421 * XXX This function was originally intended to be multipurpose, but its
3422 * only use is quite likely to remain for constructing the inversion of
3423 * the CaseFolding (//i) property. If it were more general purpose for
3424 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
3425 * because certain folds are prohibited under /iaa and /il. As an example,
3426 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
3427 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
3428 * prohibited, so we would not figure out that they fold to each other.
3429 * Code could be written to automatically figure this out, similar to the
3430 * code that does this for multi-character folds, but this is the only case
3431 * where something like this is ever likely to happen, as all the single
7ee537e6 3432 * char folds to the 0-255 range are now quite settled. Instead there is a
74894415
KW
3433 * little special code that is compiled only for this Unicode version. This
3434 * is smaller and didn't require much coding time to do. But this makes
3435 * this routine strongly tied to being used just for CaseFolding. If ever
3436 * it should be generalized, this would have to be fixed */
064c021d
KW
3437
3438 U8 *l, *lend;
3439 STRLEN lcur;
3440 HV *const hv = MUTABLE_HV(SvRV(swash));
3441
923b6d4e
KW
3442 /* The string containing the main body of the table. This will have its
3443 * assertion fail if the swash has been converted to its inversion list */
064c021d
KW
3444 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
3445
3446 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3447 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3448 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3449 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
3450 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
3451 const STRLEN bits = SvUV(*bitssvp);
3452 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
3453 const UV none = SvUV(*nonesvp);
5662e334 3454 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
064c021d
KW
3455
3456 HV* ret = newHV();
3457
3458 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
3459
3460 /* Must have at least 8 bits to get the mappings */
3461 if (bits != 8 && bits != 16 && bits != 32) {
3462 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"UVuf,
3463 (UV)bits);
3464 }
3465
5662e334
KW
3466 if (specials_p) { /* It might be "special" (sometimes, but not always, a
3467 mapping to more than one character */
3468
3469 /* Construct an inverse mapping hash for the specials */
3470 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
3471 HV * specials_inverse = newHV();
3472 char *char_from; /* the lhs of the map */
3473 I32 from_len; /* its byte length */
3474 char *char_to; /* the rhs of the map */
3475 I32 to_len; /* its byte length */
3476 SV *sv_to; /* and in a sv */
3477 AV* from_list; /* list of things that map to each 'to' */
3478
3479 hv_iterinit(specials_hv);
3480
4a4088c4
KW
3481 /* The keys are the characters (in UTF-8) that map to the corresponding
3482 * UTF-8 string value. Iterate through the list creating the inverse
5662e334
KW
3483 * list. */
3484 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
3485 SV** listp;
3486 if (! SvPOK(sv_to)) {
5637ef5b
NC
3487 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
3488 "unexpectedly is not a string, flags=%lu",
3489 (unsigned long)SvFLAGS(sv_to));
5662e334 3490 }
4b88fb76 3491 /*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
3492
3493 /* Each key in the inverse list is a mapped-to value, and the key's
4a4088c4 3494 * hash value is a list of the strings (each in UTF-8) that map to
5662e334
KW
3495 * it. Those strings are all one character long */
3496 if ((listp = hv_fetch(specials_inverse,
3497 SvPVX(sv_to),
3498 SvCUR(sv_to), 0)))
3499 {
3500 from_list = (AV*) *listp;
3501 }
3502 else { /* No entry yet for it: create one */
3503 from_list = newAV();
3504 if (! hv_store(specials_inverse,
3505 SvPVX(sv_to),
3506 SvCUR(sv_to),
3507 (SV*) from_list, 0))
3508 {
3509 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3510 }
3511 }
3512
3513 /* Here have the list associated with this 'to' (perhaps newly
3514 * created and empty). Just add to it. Note that we ASSUME that
3515 * the input is guaranteed to not have duplications, so we don't
3516 * check for that. Duplications just slow down execution time. */
3517 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
3518 }
3519
3520 /* Here, 'specials_inverse' contains the inverse mapping. Go through
3521 * it looking for cases like the FB05/FB06 examples above. There would
3522 * be an entry in the hash like
3523 * 'st' => [ FB05, FB06 ]
3524 * In this example we will create two lists that get stored in the
3525 * returned hash, 'ret':
3526 * FB05 => [ FB05, FB06 ]
3527 * FB06 => [ FB05, FB06 ]
3528 *
3529 * Note that there is nothing to do if the array only has one element.
3530 * (In the normal 1-1 case handled below, we don't have to worry about
3531 * two lists, as everything gets tied to the single list that is
3532 * generated for the single character 'to'. But here, we are omitting
3533 * that list, ('st' in the example), so must have multiple lists.) */
3534 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
3535 &char_to, &to_len)))
3536 {
7e13d567 3537 if (av_tindex_nomg(from_list) > 0) {
c70927a6 3538 SSize_t i;
5662e334
KW
3539
3540 /* We iterate over all combinations of i,j to place each code
3541 * point on each list */
7e13d567 3542 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
c70927a6 3543 SSize_t j;
5662e334
KW
3544 AV* i_list = newAV();
3545 SV** entryp = av_fetch(from_list, i, FALSE);
3546 if (entryp == NULL) {
3547 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3548 }
3549 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
3550 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
3551 }
3552 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
3553 (SV*) i_list, FALSE))
3554 {
3555 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3556 }
3557
538e84ed 3558 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
7e13d567 3559 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
5662e334
KW
3560 entryp = av_fetch(from_list, j, FALSE);
3561 if (entryp == NULL) {
3562 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3563 }
3564
3565 /* When i==j this adds itself to the list */
4b88fb76
KW
3566 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
3567 (U8*) SvPVX(*entryp),
3568 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
3569 0)));
4637d003 3570 /*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
3571 }
3572 }
3573 }
3574 }
3575 SvREFCNT_dec(specials_inverse); /* done with it */
3576 } /* End of specials */
3577
064c021d 3578 /* read $swash->{LIST} */
74894415
KW
3579
3580#if UNICODE_MAJOR_VERSION == 3 \
3581 && UNICODE_DOT_VERSION == 0 \
3582 && UNICODE_DOT_DOT_VERSION == 1
3583
3584 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
3585 * rule so that things work under /iaa and /il */
3586
3587 SV * mod_listsv = sv_mortalcopy(*listsvp);
3588 sv_catpv(mod_listsv, "130\t130\t131\n");
3589 l = (U8*)SvPV(mod_listsv, lcur);
3590
3591#else
3592
064c021d 3593 l = (U8*)SvPV(*listsvp, lcur);
74894415
KW
3594
3595#endif
3596
064c021d
KW
3597 lend = l + lcur;
3598
3599 /* Go through each input line */
3600 while (l < lend) {
3601 UV min, max, val;
3602 UV inverse;
95543e92
KW
3603 l = swash_scan_list_line(l, lend, &min, &max, &val,
3604 cBOOL(octets), typestr);
064c021d
KW
3605 if (l > lend) {
3606 break;
3607 }
3608
3609 /* Each element in the range is to be inverted */
3610 for (inverse = min; inverse <= max; inverse++) {
3611 AV* list;
064c021d
KW
3612 SV** listp;
3613 IV i;
3614 bool found_key = FALSE;
5662e334 3615 bool found_inverse = FALSE;
064c021d
KW
3616
3617 /* The key is the inverse mapping */
3618 char key[UTF8_MAXBYTES+1];
c80e42f3 3619 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
064c021d
KW
3620 STRLEN key_len = key_end - key;
3621
064c021d
KW
3622 /* Get the list for the map */
3623 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
3624 list = (AV*) *listp;
3625 }
3626 else { /* No entry yet for it: create one */
3627 list = newAV();
3628 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
3629 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3630 }
3631 }
3632
5662e334
KW
3633 /* Look through list to see if this inverse mapping already is
3634 * listed, or if there is a mapping to itself already */
7e13d567 3635 for (i = 0; i <= av_tindex_nomg(list); i++) {
064c021d
KW
3636 SV** entryp = av_fetch(list, i, FALSE);
3637 SV* entry;
414db8a1 3638 UV uv;
064c021d
KW
3639 if (entryp == NULL) {
3640 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
3641 }
3642 entry = *entryp;
414db8a1
DM
3643 uv = SvUV(entry);
3644 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %"UVXf" contains %"UVXf"\n", val, uv));*/
3645 if (uv == val) {
064c021d 3646 found_key = TRUE;
5662e334 3647 }
414db8a1 3648 if (uv == inverse) {
5662e334
KW
3649 found_inverse = TRUE;
3650 }
3651
3652 /* No need to continue searching if found everything we are
3653 * looking for */
3654 if (found_key && found_inverse) {
064c021d
KW
3655 break;
3656 }
3657 }
56ca34ca
KW
3658
3659 /* Make sure there is a mapping to itself on the list */
064c021d 3660 if (! found_key) {
d397ff6a 3661 av_push(list, newSVuv(val));
4637d003 3662 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, val, val));*/
064c021d
KW
3663 }
3664
3665
3666 /* Simply add the value to the list */
5662e334
KW
3667 if (! found_inverse) {
3668 av_push(list, newSVuv(inverse));
4637d003 3669 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %"UVXf" to list for %"UVXf"\n", __FILE__, __LINE__, inverse, val));*/
5662e334 3670 }
064c021d 3671
b0e3252e 3672 /* swatch_get() increments the value of val for each element in the
064c021d
KW
3673 * range. That makes more compact tables possible. You can
3674 * express the capitalization, for example, of all consecutive
3675 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
3676 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
bd3f2f94 3677 * and it's not documented; it appears to be used only in
b0e3252e 3678 * implementing tr//; I copied the semantics from swatch_get(), just
bd3f2f94 3679 * in case */