This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Simplify
[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"
afde5508 35#include "uni_keywords.h"
a0ed51b3 36
806547a7 37static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 38static const char unees[] =
806547a7 39 "Malformed UTF-8 character (unexpected end of string)";
fb7e7255
KW
40
41/* Be sure to synchronize this message with the similar one in regcomp.c */
760c7c2f 42static const char cp_above_legal_max[] =
d22ec717
KW
43 "Use of code point 0x%" UVXf " is not allowed; the"
44 " permissible max is 0x%" UVXf;
760c7c2f 45
d22ec717 46#define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
901b21bf 47
48ef279e 48/*
ccfc67b7 49=head1 Unicode Support
7fefc6c1 50These are various utility functions for manipulating UTF8-encoded
72d33970 51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
9cbfb8ab
KW
59void
60Perl__force_out_malformed_utf8_message(pTHX_
61 const U8 *const p, /* First byte in UTF-8 sequence */
62 const U8 * const e, /* Final byte in sequence (may include
63 multiple chars */
64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
65 usually 0, or some DISALLOW flags */
66 const bool die_here) /* If TRUE, this function does not return */
67{
68 /* This core-only function is to be called when a malformed UTF-8 character
69 * is found, in order to output the detailed information about the
70 * malformation before dieing. The reason it exists is for the occasions
71 * when such a malformation is fatal, but warnings might be turned off, so
72 * that normally they would not be actually output. This ensures that they
73 * do get output. Because a sequence may be malformed in more than one
74 * way, multiple messages may be generated, so we can't make them fatal, as
75 * that would cause the first one to die.
76 *
77 * Instead we pretend -W was passed to perl, then die afterwards. The
78 * flexibility is here to return to the caller so they can finish up and
79 * die themselves */
80 U32 errors;
81
82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
83
84 ENTER;
c15a80f3 85 SAVEI8(PL_dowarn);
9cbfb8ab
KW
86 SAVESPTR(PL_curcop);
87
88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
89 if (PL_curcop) {
90 PL_curcop->cop_warnings = pWARN_ALL;
91 }
92
93 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
94
95 LEAVE;
96
97 if (! errors) {
98 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
99 " be called only when there are errors found");
100 }
101
102 if (die_here) {
103 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
104 }
105}
106
bb07812e
KW
107STATIC HV *
108S_new_msg_hv(pTHX_ const char * const message, /* The message text */
109 U32 categories, /* Packed warning categories */
110 U32 flag) /* Flag associated with this message */
111{
112 /* Creates, populates, and returns an HV* that describes an error message
113 * for the translators between UTF8 and code point */
114
115 SV* msg_sv = newSVpv(message, 0);
116 SV* category_sv = newSVuv(categories);
117 SV* flag_bit_sv = newSVuv(flag);
118
119 HV* msg_hv = newHV();
120
121 PERL_ARGS_ASSERT_NEW_MSG_HV;
122
2b672cf5
KW
123 (void) hv_stores(msg_hv, "text", msg_sv);
124 (void) hv_stores(msg_hv, "warn_categories", category_sv);
125 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
126
127 return msg_hv;
128}
129
eaf7a4d2 130/*
378516de 131=for apidoc uvoffuni_to_utf8_flags
eebe1485 132
a27992cc 133THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
134Instead, B<Almost all code should use L</uvchr_to_utf8> or
135L</uvchr_to_utf8_flags>>.
a27992cc 136
de69f3af
KW
137This function is like them, but the input is a strict Unicode
138(as opposed to native) code point. Only in very rare circumstances should code
139not be using the native code point.
949cf498 140
efa9cd84 141For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 142
eebe1485
SC
143=cut
144*/
145
33f38593
KW
146U8 *
147Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
148{
149 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
150
151 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
152}
153
c94c2f39
KW
154/* All these formats take a single UV code point argument */
155const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
156const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
157 " is not recommended for open interchange";
158const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
159 " may not be portable";
57ff5f59
KW
160const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
161 " Unicode, requires a Perl extension," \
162 " and so is not portable";
c94c2f39 163
33f38593 164#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
165 STMT_START { \
166 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
167 U32 category = packWARN(WARN_SURROGATE); \
168 const char * format = surrogate_cp_format; \
169 if (msgs) { \
170 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
171 category, \
172 UNICODE_GOT_SURROGATE); \
173 } \
174 else { \
175 Perl_ck_warner_d(aTHX_ category, format, uv); \
176 } \
8ee1cdcb
KW
177 } \
178 if (flags & UNICODE_DISALLOW_SURROGATE) { \
179 return NULL; \
180 } \
181 } STMT_END;
182
33f38593 183#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
184 STMT_START { \
185 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
186 U32 category = packWARN(WARN_NONCHAR); \
187 const char * format = nonchar_cp_format; \
188 if (msgs) { \
189 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
190 category, \
191 UNICODE_GOT_NONCHAR); \
192 } \
193 else { \
194 Perl_ck_warner_d(aTHX_ category, format, uv); \
195 } \
8ee1cdcb
KW
196 } \
197 if (flags & UNICODE_DISALLOW_NONCHAR) { \
198 return NULL; \
199 } \
200 } STMT_END;
201
ba6ed43c
KW
202/* Use shorter names internally in this file */
203#define SHIFT UTF_ACCUMULATION_SHIFT
204#undef MARK
205#define MARK UTF_CONTINUATION_MARK
206#define MASK UTF_CONTINUATION_MASK
207
33f38593
KW
208/*
209=for apidoc uvchr_to_utf8_flags_msgs
210
211THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
212
213Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
214
215This function is for code that wants any warning and/or error messages to be
216returned to the caller rather than be displayed. All messages that would have
884a31ee 217been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
218
219It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
220placed after all the others, C<msgs>. If this parameter is 0, this function
221behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
222be a pointer to an C<HV *> variable, in which this function creates a new HV to
223contain any appropriate messages. The hash has three key-value pairs, as
224follows:
225
226=over 4
227
228=item C<text>
229
230The text of the message as a C<SVpv>.
231
232=item C<warn_categories>
233
234The warning category (or categories) packed into a C<SVuv>.
235
236=item C<flag>
237
238A single flag bit associated with this message, in a C<SVuv>.
239The bit corresponds to some bit in the C<*errors> return value,
240such as C<UNICODE_GOT_SURROGATE>.
241
242=back
243
244It's important to note that specifying this parameter as non-null will cause
245any warnings this function would otherwise generate to be suppressed, and
246instead be placed in C<*msgs>. The caller can check the lexical warnings state
247(or not) when choosing what to do with the returned messages.
248
249The caller, of course, is responsible for freeing any returned HV.
250
251=cut
252*/
253
254/* Undocumented; we don't want people using this. Instead they should use
255 * uvchr_to_utf8_flags_msgs() */
dfe13c55 256U8 *
33f38593 257Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 258{
33f38593
KW
259 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
260
261 if (msgs) {
262 *msgs = NULL;
263 }
7918f24d 264
2d1545e5 265 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 266 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
267 return d;
268 }
facc1dc2 269
3ea68d71 270 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
271 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
272 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
273 return d;
274 }
d9432125 275
ba6ed43c
KW
276 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
277 * below, the 16 is for start bytes E0-EF (which are all the possible ones
278 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
279 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
280 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
281 * 0x800-0xFFFF on ASCII */
282 if (uv < (16 * (1U << (2 * SHIFT)))) {
283 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
284 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
285 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
286
287#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
288 aren't tested here */
289 /* The most likely code points in this range are below the surrogates.
290 * Do an extra test to quickly exclude those. */
291 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
292 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
293 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
294 {
33f38593 295 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
296 }
297 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 298 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 299 }
ba6ed43c
KW
300 }
301#endif
302 return d;
303 }
304
305 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
306 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
307 * happen starting with 4-byte characters on ASCII platforms. We unify the
308 * code for these with EBCDIC, even though some of them require 5-bytes on
309 * those, because khw believes the code saving is worth the very slight
310 * performance hit on these high EBCDIC code points. */
311
312 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
d22ec717
KW
313 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
314 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
a5bf80e0 315 }
33f38593
KW
316 if ( (flags & UNICODE_WARN_SUPER)
317 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 318 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 319 {
33f38593
KW
320 const char * format = super_cp_format;
321 U32 category = packWARN(WARN_NON_UNICODE);
322 U32 flag = UNICODE_GOT_SUPER;
323
324 /* Choose the more dire applicable warning */
325 if (UNICODE_IS_PERL_EXTENDED(uv)) {
326 format = perl_extended_cp_format;
327 if (flags & (UNICODE_WARN_PERL_EXTENDED
328 |UNICODE_DISALLOW_PERL_EXTENDED))
329 {
330 flag = UNICODE_GOT_PERL_EXTENDED;
331 }
332 }
a5bf80e0 333
33f38593
KW
334 if (msgs) {
335 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
336 category, flag);
337 }
338 else {
339 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
340 }
a5bf80e0 341 }
56576a04 342 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
343 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
344 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
345 {
346 return NULL;
347 }
348 }
ba6ed43c 349 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 350 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 351 }
d9432125 352
ba6ed43c
KW
353 /* Test for and handle 4-byte result. In the test immediately below, the
354 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
355 * characters). The 3 is for 3 continuation bytes; these each contribute
356 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
357 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
358 * 0x1_0000-0x1F_FFFF on ASCII */
359 if (uv < (8 * (1U << (3 * SHIFT)))) {
360 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
362 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
363 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
364
365#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
366 characters. The end-plane non-characters for EBCDIC were
367 handled just above */
368 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 369 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 370 }
ba6ed43c 371 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 372 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
373 }
374#endif
375
376 return d;
377 }
378
379 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
380 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
381 * format. The unrolled version above turns out to not save all that much
382 * time, and at these high code points (well above the legal Unicode range
383 * on ASCII platforms, and well above anything in common use in EBCDIC),
384 * khw believes that less code outweighs slight performance gains. */
385
d9432125 386 {
5aaebcb3 387 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
388 U8 *p = d+len-1;
389 while (p > d) {
4c8cd605 390 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
391 uv >>= UTF_ACCUMULATION_SHIFT;
392 }
4c8cd605 393 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
394 return d+len;
395 }
a0ed51b3 396}
a5bf80e0 397
646ca15d 398/*
07693fe6
KW
399=for apidoc uvchr_to_utf8
400
bcb1a2d4 401Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 402of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
403C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
404the byte after the end of the new character. In other words,
07693fe6
KW
405
406 d = uvchr_to_utf8(d, uv);
407
408is the recommended wide native character-aware way of saying
409
410 *(d++) = uv;
411
d22ec717
KW
412This function accepts any code point from 0..C<IV_MAX> as input.
413C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
414
415It is possible to forbid or warn on non-Unicode code points, or those that may
416be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 417
07693fe6
KW
418=cut
419*/
420
de69f3af
KW
421/* This is also a macro */
422PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
423
07693fe6
KW
424U8 *
425Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
426{
de69f3af 427 return uvchr_to_utf8(d, uv);
07693fe6
KW
428}
429
de69f3af
KW
430/*
431=for apidoc uvchr_to_utf8_flags
432
433Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 434of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
435C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
436the byte after the end of the new character. In other words,
de69f3af
KW
437
438 d = uvchr_to_utf8_flags(d, uv, flags);
439
440or, in most cases,
441
442 d = uvchr_to_utf8_flags(d, uv, 0);
443
444This is the Unicode-aware way of saying
445
446 *(d++) = uv;
447
d22ec717
KW
448If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
449input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
450
451Specifying C<flags> can further restrict what is allowed and not warned on, as
452follows:
de69f3af 453
796b6530 454If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
455the function will raise a warning, provided UTF8 warnings are enabled. If
456instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
457NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 458
760c7c2f
KW
459Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
460affect how the function handles a Unicode non-character.
93e6dbd6 461
760c7c2f
KW
462And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
463affect the handling of code points that are above the Unicode maximum of
4640x10FFFF. Languages other than Perl may not be able to accept files that
465contain these.
93e6dbd6
KW
466
467The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
468the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
469three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
470allowed inputs to the strict UTF-8 traditionally defined by Unicode.
471Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
472C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
473above-Unicode and surrogate flags, but not the non-character ones, as
474defined in
475L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
476See L<perlunicode/Noncharacter code points>.
93e6dbd6 477
57ff5f59
KW
478Extremely high code points were never specified in any standard, and require an
479extension to UTF-8 to express, which Perl does. It is likely that programs
480written in something other than Perl would not be able to read files that
481contain these; nor would Perl understand files written by something that uses a
482different extension. For these reasons, there is a separate set of flags that
483can warn and/or disallow these extremely high code points, even if other
484above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
485and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
486L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
487treat all above-Unicode code points, including these, as malformations. (Note
488that the Unicode standard considers anything above 0x10FFFF to be illegal, but
489there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
490
491A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
492retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
493C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
494C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
495platforms,these flags can apply to code points that actually do fit in 31 bits.
496The new names accurately describe the situation in all cases.
de69f3af 497
de69f3af
KW
498=cut
499*/
500
501/* This is also a macro */
502PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
503
07693fe6
KW
504U8 *
505Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
506{
de69f3af 507 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
508}
509
57ff5f59
KW
510#ifndef UV_IS_QUAD
511
e050c007
KW
512STATIC int
513S_is_utf8_cp_above_31_bits(const U8 * const s,
514 const U8 * const e,
515 const bool consider_overlongs)
83dc0f42
KW
516{
517 /* Returns TRUE if the first code point represented by the Perl-extended-
518 * UTF-8-encoded string starting at 's', and looking no further than 'e -
519 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
520 *
521 * The function handles the case where the input bytes do not include all
522 * the ones necessary to represent a full character. That is, they may be
523 * the intial bytes of the representation of a code point, but possibly
524 * the final ones necessary for the complete representation may be beyond
525 * 'e - 1'.
526 *
e050c007
KW
527 * The function also can handle the case where the input is an overlong
528 * sequence. If 'consider_overlongs' is 0, the function assumes the
529 * input is not overlong, without checking, and will return based on that
530 * assumption. If this parameter is 1, the function will go to the trouble
531 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 532 *
e050c007 533 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
534 */
535
e050c007
KW
536 const STRLEN len = e - s;
537 int is_overlong;
538
539 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
540
541 assert(! UTF8_IS_INVARIANT(*s) && e > s);
542
83dc0f42
KW
543#ifdef EBCDIC
544
e050c007 545 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 546
e050c007
KW
547 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
548 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
549 * also be the start byte for a 31-bit code point; we need at least 2
550 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
551 * the start byte for an overlong sequence, but for 30-bit or smaller code
552 * points, so we don't have to worry about overlongs on EBCDIC.) */
553 if (*s != 0xFE) {
554 return 0;
555 }
83dc0f42 556
e050c007
KW
557 if (len == 1) {
558 return -1;
559 }
83dc0f42 560
e050c007 561#else
83dc0f42 562
e050c007
KW
563 /* On ASCII, FE and FF are the only start bytes that can evaluate to
564 * needing more than 31 bits. */
565 if (LIKELY(*s < 0xFE)) {
566 return 0;
567 }
83dc0f42 568
e050c007
KW
569 /* What we have left are FE and FF. Both of these require more than 31
570 * bits unless they are for overlongs. */
571 if (! consider_overlongs) {
572 return 1;
573 }
83dc0f42 574
e050c007
KW
575 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
576 * above 31 bits. But we need more than one byte to discern this, so if
577 * passed just the start byte, it could be an overlong evaluating to
578 * smaller */
579 if (len == 1) {
580 return -1;
581 }
83dc0f42 582
e050c007
KW
583 /* Having excluded len==1, and knowing that FE and FF are both valid start
584 * bytes, we can call the function below to see if the sequence is
585 * overlong. (We don't need the full generality of the called function,
586 * but for these huge code points, speed shouldn't be a consideration, and
587 * the compiler does have enough information, since it's static to this
588 * file, to optimize to just the needed parts.) */
589 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 590
e050c007
KW
591 /* If it isn't overlong, more than 31 bits are required. */
592 if (is_overlong == 0) {
593 return 1;
594 }
83dc0f42 595
e050c007
KW
596 /* If it is indeterminate if it is overlong, return that */
597 if (is_overlong < 0) {
598 return -1;
599 }
600
601 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
602 * the max it can be is 2**31 - 1 */
603 if (*s == 0xFE) {
604 return 0;
83dc0f42
KW
605 }
606
e050c007
KW
607#endif
608
609 /* Here, ASCII and EBCDIC rejoin:
610 * On ASCII: We have an overlong sequence starting with FF
611 * On EBCDIC: We have a sequence starting with FE. */
612
613 { /* For C89, use a block so the declaration can be close to its use */
614
615#ifdef EBCDIC
616
5f995336
KW
617 /* U+7FFFFFFF (2 ** 31 - 1)
618 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
619 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
620 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
621 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
622 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
623 * U+80000000 (2 ** 31):
624 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
625 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
626 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
627 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
628 *
629 * and since we know that *s = \xfe, any continuation sequcence
630 * following it that is gt the below is above 31 bits
631 [0] [1] [2] [3] [4] [5] [6] */
632 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
633
634#else
635
636 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
637 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
638 * FF overlong for U+80000000 (2 ** 31):
639 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
640 * and since we know that *s = \xff, any continuation sequcence
641 * following it that is gt the below is above 30 bits
642 [0] [1] [2] [3] [4] [5] [6] */
643 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 644
83dc0f42
KW
645
646#endif
e050c007
KW
647 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
648 const STRLEN cmp_len = MIN(conts_len, len - 1);
649
650 /* Now compare the continuation bytes in s with the ones we have
651 * compiled in that are for the largest 30 bit code point. If we have
652 * enough bytes available to determine the answer, or the bytes we do
653 * have differ from them, we can compare the two to get a definitive
654 * answer (Note that in UTF-EBCDIC, the two lowest possible
655 * continuation bytes are \x41 and \x42.) */
656 if (cmp_len >= conts_len || memNE(s + 1,
657 conts_for_highest_30_bit,
658 cmp_len))
659 {
660 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
661 }
83dc0f42 662
e050c007
KW
663 /* Here, all the bytes we have are the same as the highest 30-bit code
664 * point, but we are missing so many bytes that we can't make the
665 * determination */
666 return -1;
667 }
83dc0f42
KW
668}
669
57ff5f59
KW
670#endif
671
d6be65ae 672PERL_STATIC_INLINE int
12a4bed3
KW
673S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
674{
d6be65ae
KW
675 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
676 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
677 * it isn't, and -1 if there isn't enough information to tell. This last
678 * return value can happen if the sequence is incomplete, missing some
679 * trailing bytes that would form a complete character. If there are
680 * enough bytes to make a definitive decision, this function does so.
681 * Usually 2 bytes sufficient.
682 *
683 * Overlongs can occur whenever the number of continuation bytes changes.
684 * That means whenever the number of leading 1 bits in a start byte
685 * increases from the next lower start byte. That happens for start bytes
686 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
687 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
688 * ASCII platforms: C0, C1
689 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 690 */
12a4bed3
KW
691
692 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
693 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
694
695 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
696 assert(len > 1 && UTF8_IS_START(*s));
697
698 /* Each platform has overlongs after the start bytes given above (expressed
699 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
700 * the logic is the same, except the E0 overlong has already been excluded
701 * on EBCDIC platforms. The values below were found by manually
702 * inspecting the UTF-8 patterns. See the tables in utf8.h and
703 * utfebcdic.h. */
704
705# ifdef EBCDIC
706# define F0_ABOVE_OVERLONG 0xB0
707# define F8_ABOVE_OVERLONG 0xA8
708# define FC_ABOVE_OVERLONG 0xA4
709# define FE_ABOVE_OVERLONG 0xA2
710# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
711 /* I8(0xfe) is FF */
712# else
713
714 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 715 return 1;
12a4bed3
KW
716 }
717
718# define F0_ABOVE_OVERLONG 0x90
719# define F8_ABOVE_OVERLONG 0x88
720# define FC_ABOVE_OVERLONG 0x84
721# define FE_ABOVE_OVERLONG 0x82
722# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
723# endif
724
725
726 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
727 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
728 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
729 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
730 {
d6be65ae 731 return 1;
12a4bed3
KW
732 }
733
b0b342d4 734 /* Check for the FF overlong */
d6be65ae 735 return isFF_OVERLONG(s, len);
b0b342d4
KW
736}
737
8d6204cc 738PERL_STATIC_INLINE int
b0b342d4
KW
739S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
740{
8d6204cc
KW
741 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
742 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
743 * it isn't, and -1 if there isn't enough information to tell. This last
744 * return value can happen if the sequence is incomplete, missing some
745 * trailing bytes that would form a complete character. If there are
746 * enough bytes to make a definitive decision, this function does so. */
747
b0b342d4 748 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 749
8d6204cc
KW
750 /* To be an FF overlong, all the available bytes must match */
751 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
752 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
753 {
754 return 0;
755 }
756
757 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
758 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 759 * utfebcdic.h. */
8d6204cc
KW
760 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
761 return 1;
762 }
12a4bed3 763
8d6204cc
KW
764 /* The missing bytes could cause the result to go one way or the other, so
765 * the result is indeterminate */
766 return -1;
12a4bed3
KW
767}
768
d22ec717 769#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
770# ifdef EBCDIC /* Actually is I8 */
771# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 772 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
773# else
774# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 775 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
776# endif
777#endif
778
c285bbc4 779PERL_STATIC_INLINE int
e050c007
KW
780S_does_utf8_overflow(const U8 * const s,
781 const U8 * e,
782 const bool consider_overlongs)
a77c906e 783{
c285bbc4 784 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
785 * 'e' - 1 would overflow an IV on this platform; that is if it represents
786 * a code point larger than the highest representable code point. It
787 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
788 * enough information to tell. This last return value can happen if the
789 * sequence is incomplete, missing some trailing bytes that would form a
790 * complete character. If there are enough bytes to make a definitive
791 * decision, this function does so.
c285bbc4 792 *
e050c007
KW
793 * If 'consider_overlongs' is TRUE, the function checks for the possibility
794 * that the sequence is an overlong that doesn't overflow. Otherwise, it
795 * assumes the sequence is not an overlong. This can give different
796 * results only on ASCII 32-bit platforms.
797 *
c285bbc4
KW
798 * (For ASCII platforms, we could use memcmp() because we don't have to
799 * convert each byte to I8, but it's very rare input indeed that would
800 * approach overflow, so the loop below will likely only get executed once.)
801 *
802 * 'e' - 1 must not be beyond a full character. */
a77c906e 803
a77c906e
KW
804
805 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
806 assert(s <= e && s + UTF8SKIP(s) >= e);
807
d22ec717
KW
808#if ! defined(UV_IS_QUAD)
809
810 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
811
812#else
813
814 PERL_UNUSED_ARG(consider_overlongs);
815
816 {
817 const STRLEN len = e - s;
818 const U8 *x;
819 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
820
821 for (x = s; x < e; x++, y++) {
822
823 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
824 continue;
825 }
826
827 /* If this byte is larger than the corresponding highest UTF-8
828 * byte, the sequence overflow; otherwise the byte is less than,
829 * and so the sequence doesn't overflow */
830 return NATIVE_UTF8_TO_I8(*x) > *y;
831
832 }
833
834 /* Got to the end and all bytes are the same. If the input is a whole
835 * character, it doesn't overflow. And if it is a partial character,
836 * there's not enough information to tell */
837 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
838 return -1;
839 }
840
841 return 0;
842 }
843
844#endif
845
846}
847
848#if 0
849
850/* This is the portions of the above function that deal with UV_MAX instead of
851 * IV_MAX. They are left here in case we want to combine them so that internal
852 * uses can have larger code points. The only logic difference is that the
853 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
854 * different logic.
855 */
856
857/* Anything larger than this will overflow the word if it were converted into a UV */
858#if defined(UV_IS_QUAD)
859# ifdef EBCDIC /* Actually is I8 */
860# define HIGHEST_REPRESENTABLE_UTF8 \
861 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
862# else
863# define HIGHEST_REPRESENTABLE_UTF8 \
864 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
865# endif
866#else /* 32-bit */
867# ifdef EBCDIC
868# define HIGHEST_REPRESENTABLE_UTF8 \
869 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
870# else
871# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
872# endif
873#endif
874
a77c906e
KW
875#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
876
877 /* On 32 bit ASCII machines, many overlongs that start with FF don't
878 * overflow */
e050c007 879 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
880
881 /* To be such an overlong, the first bytes of 's' must match
882 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
883 * don't have any additional bytes available, the sequence, when
884 * completed might or might not fit in 32 bits. But if we have that
885 * next byte, we can tell for sure. If it is <= 0x83, then it does
886 * fit. */
887 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
888 return -1;
889 }
890
891 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
892 }
893
d22ec717
KW
894/* Starting with the #else, the rest of the function is identical except
895 * 1. we need to move the 'len' declaration to be global to the function
896 * 2. the endif move to just after the UNUSED_ARG.
897 * An empty endif is given just below to satisfy the preprocessor
898 */
a77c906e
KW
899#endif
900
d22ec717 901#endif
a77c906e 902
12a4bed3
KW
903#undef F0_ABOVE_OVERLONG
904#undef F8_ABOVE_OVERLONG
905#undef FC_ABOVE_OVERLONG
906#undef FE_ABOVE_OVERLONG
907#undef FF_OVERLONG_PREFIX
908
35f8c9bd 909STRLEN
edc2c47a 910Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 911{
2b479609 912 STRLEN len;
12a4bed3 913 const U8 *x;
35f8c9bd 914
2b479609
KW
915 /* A helper function that should not be called directly.
916 *
917 * This function returns non-zero if the string beginning at 's' and
918 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
919 * code point; otherwise it returns 0. The examination stops after the
920 * first code point in 's' is validated, not looking at the rest of the
921 * input. If 'e' is such that there are not enough bytes to represent a
922 * complete code point, this function will return non-zero anyway, if the
923 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
924 * excluded by 'flags'.
925 *
926 * A non-zero return gives the number of bytes required to represent the
927 * code point. Be aware that if the input is for a partial character, the
928 * return will be larger than 'e - s'.
929 *
930 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
931 * The caller should have excluded the possibility of it being invariant
932 * before calling this function.
2b479609
KW
933 *
934 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
935 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
936 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
937 * disallowed by the flags. If the input is only for a partial character,
938 * the function will return non-zero if there is any sequence of
939 * well-formed UTF-8 that, when appended to the input sequence, could
940 * result in an allowed code point; otherwise it returns 0. Non characters
941 * cannot be determined based on partial character input. But many of the
942 * other excluded types can be determined with just the first one or two
943 * bytes.
944 *
945 */
946
947 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
948
949 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 950 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 951 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 952
2b479609 953 /* A variant char must begin with a start byte */
35f8c9bd
KW
954 if (UNLIKELY(! UTF8_IS_START(*s))) {
955 return 0;
956 }
957
edc2c47a
KW
958 /* Examine a maximum of a single whole code point */
959 if (e - s > UTF8SKIP(s)) {
960 e = s + UTF8SKIP(s);
961 }
962
2b479609
KW
963 len = e - s;
964
965 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
966 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 967
56576a04
KW
968 /* Here, we are disallowing some set of largish code points, and the
969 * first byte indicates the sequence is for a code point that could be
970 * in the excluded set. We generally don't have to look beyond this or
971 * the second byte to see if the sequence is actually for one of the
972 * excluded classes. The code below is derived from this table:
973 *
2b479609
KW
974 * UTF-8 UTF-EBCDIC I8
975 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
976 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
977 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
978 *
56576a04
KW
979 * Keep in mind that legal continuation bytes range between \x80..\xBF
980 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
981 * continuation bytes. Hence, we don't have to test the upper edge
982 * because if any of those is encountered, the sequence is malformed,
983 * and would fail elsewhere in this function.
984 *
985 * The code here likewise assumes that there aren't other
986 * malformations; again the function should fail elsewhere because of
987 * these. For example, an overlong beginning with FC doesn't actually
988 * have to be a super; it could actually represent a small code point,
989 * even U+0000. But, since overlongs (and other malformations) are
990 * illegal, the function should return FALSE in either case.
2b479609
KW
991 */
992
993#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
994# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 995# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 996
19794540
KW
997# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
998 /* B6 and B7 */ \
999 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1000# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1001#else
1002# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1003# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1004# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1005# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1006#endif
1007
1008 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1009 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1010 {
2b479609
KW
1011 return 0; /* Above Unicode */
1012 }
1013
d044b7a7 1014 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1015 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1016 {
57ff5f59 1017 return 0;
2b479609
KW
1018 }
1019
1020 if (len > 1) {
1021 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1022
1023 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1024 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1025 {
1026 return 0; /* Above Unicode */
1027 }
1028
1029 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1030 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1031 {
1032 return 0; /* Surrogate */
1033 }
1034
1035 if ( (flags & UTF8_DISALLOW_NONCHAR)
1036 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1037 {
1038 return 0; /* Noncharacter code point */
1039 }
1040 }
1041 }
1042
1043 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1044 for (x = s + 1; x < e; x++) {
1045 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1046 return 0;
1047 }
1048 }
1049
af13dd8a 1050 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1051 * overlong. */
d6be65ae 1052 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1053 return 0;
af13dd8a
KW
1054 }
1055
12a4bed3
KW
1056 /* And finally, that the code point represented fits in a word on this
1057 * platform */
e050c007
KW
1058 if (0 < does_utf8_overflow(s, e,
1059 0 /* Don't consider overlongs */
1060 ))
1061 {
12a4bed3 1062 return 0;
35f8c9bd
KW
1063 }
1064
2b479609 1065 return UTF8SKIP(s);
35f8c9bd
KW
1066}
1067
7e2f38b2 1068char *
63ab03b3 1069Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1070{
1071 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1072 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1073 * Currently, there are only two formats, so it is currently a bool:
1074 * 0 \xab
1075 * 1 ab (that is a space between two hex digit bytes)
1076 */
7cf8d05d
KW
1077
1078 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1079 trailing NUL */
63ab03b3
KW
1080 const U8 * s = start;
1081 const U8 * const e = start + len;
7cf8d05d
KW
1082 char * output;
1083 char * d;
1084
1085 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1086
1087 Newx(output, output_len, char);
1088 SAVEFREEPV(output);
1089
1090 d = output;
63ab03b3 1091 for (s = start; s < e; s++) {
7cf8d05d
KW
1092 const unsigned high_nibble = (*s & 0xF0) >> 4;
1093 const unsigned low_nibble = (*s & 0x0F);
1094
7e2f38b2 1095 if (format) {
63ab03b3
KW
1096 if (s > start) {
1097 *d++ = ' ';
1098 }
7e2f38b2
KW
1099 }
1100 else {
1101 *d++ = '\\';
1102 *d++ = 'x';
1103 }
7cf8d05d
KW
1104
1105 if (high_nibble < 10) {
1106 *d++ = high_nibble + '0';
1107 }
1108 else {
1109 *d++ = high_nibble - 10 + 'a';
1110 }
1111
1112 if (low_nibble < 10) {
1113 *d++ = low_nibble + '0';
1114 }
1115 else {
1116 *d++ = low_nibble - 10 + 'a';
1117 }
1118 }
1119
1120 *d = '\0';
1121 return output;
1122}
1123
806547a7 1124PERL_STATIC_INLINE char *
7cf8d05d
KW
1125S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1126
421da25c 1127 /* Max number of bytes to print */
3cc6a05e 1128 STRLEN print_len,
7cf8d05d
KW
1129
1130 /* Which one is the non-continuation */
1131 const STRLEN non_cont_byte_pos,
1132
1133 /* How many bytes should there be? */
1134 const STRLEN expect_len)
806547a7
KW
1135{
1136 /* Return the malformation warning text for an unexpected continuation
1137 * byte. */
1138
7cf8d05d 1139 const char * const where = (non_cont_byte_pos == 1)
806547a7 1140 ? "immediately"
7cf8d05d
KW
1141 : Perl_form(aTHX_ "%d bytes",
1142 (int) non_cont_byte_pos);
421da25c
KW
1143 const U8 * x = s + non_cont_byte_pos;
1144 const U8 * e = s + print_len;
806547a7
KW
1145
1146 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1147
7cf8d05d
KW
1148 /* We don't need to pass this parameter, but since it has already been
1149 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1150 assert(expect_len == UTF8SKIP(s));
1151
421da25c
KW
1152 /* As a defensive coding measure, don't output anything past a NUL. Such
1153 * bytes shouldn't be in the middle of a malformation, and could mark the
1154 * end of the allocated string, and what comes after is undefined */
1155 for (; x < e; x++) {
1156 if (*x == '\0') {
1157 x++; /* Output this particular NUL */
1158 break;
1159 }
1160 }
1161
7cf8d05d
KW
1162 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1163 " %s after start byte 0x%02x; need %d bytes, got %d)",
1164 malformed_text,
421da25c 1165 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1166 *(s + non_cont_byte_pos),
1167 where,
1168 *s,
1169 (int) expect_len,
1170 (int) non_cont_byte_pos);
806547a7
KW
1171}
1172
35f8c9bd
KW
1173/*
1174
de69f3af 1175=for apidoc utf8n_to_uvchr
378516de
KW
1176
1177THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1178Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1179
9041c2e3 1180Bottom level UTF-8 decode routine.
de69f3af 1181Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1182which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1183C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1184the length, in bytes, of that character.
949cf498
KW
1185
1186The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1187well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1188causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1189is the next possible position in C<s> that could begin a non-malformed
1190character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1191is raised. Some UTF-8 input sequences may contain multiple malformations.
1192This function tries to find every possible one in each call, so multiple
56576a04 1193warnings can be raised for the same sequence.
949cf498
KW
1194
1195Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1196individual types of malformations, such as the sequence being overlong (that
1197is, when there is a shorter sequence that can express the same code point;
1198overlong sequences are expressly forbidden in the UTF-8 standard due to
1199potential security issues). Another malformation example is the first byte of
1200a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1201flags. Even if allowed, this function generally returns the Unicode
1202REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1203F<utf8.h> to override this behavior for the overlong malformations, but don't
1204do that except for very specialized purposes.
949cf498 1205
796b6530 1206The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1207flags) malformation is found. If this flag is set, the routine assumes that
1208the caller will raise a warning, and this function will silently just set
d088425d
KW
1209C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1210
75200dff 1211Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1212character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1213in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1214be set to 1. To disambiguate, upon a zero return, see if the first byte of
1215C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1216error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1217
1218Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1219Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1220By default these are considered regular code points, but certain situations
ecc1615f
KW
1221warrant special handling for them, which can be specified using the C<flags>
1222parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1223three classes are treated as malformations and handled as such. The flags
1224C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1225C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1226disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1227restricts the allowed inputs to the strict UTF-8 traditionally defined by
1228Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1229definition given by
1230L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1231The difference between traditional strictness and C9 strictness is that the
1232latter does not forbid non-character code points. (They are still discouraged,
1233however.) For more discussion see L<perlunicode/Noncharacter code points>.
1234
1235The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1236C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1237C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1238raised for their respective categories, but otherwise the code points are
1239considered valid (not malformations). To get a category to both be treated as
1240a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1241(But note that warnings are not raised if lexically disabled nor if
796b6530 1242C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1243
57ff5f59
KW
1244Extremely high code points were never specified in any standard, and require an
1245extension to UTF-8 to express, which Perl does. It is likely that programs
1246written in something other than Perl would not be able to read files that
1247contain these; nor would Perl understand files written by something that uses a
1248different extension. For these reasons, there is a separate set of flags that
1249can warn and/or disallow these extremely high code points, even if other
1250above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1251C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1252L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1253above-Unicode code points, including these, as malformations.
1254(Note that the Unicode standard considers anything above 0x10FFFF to be
1255illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1256(2**31 -1))
1257
1258A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1259retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1260C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1261C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1262can apply to code points that actually do fit in 31 bits. This happens on
1263EBCDIC platforms, and sometimes when the L<overlong
1264malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1265describe the situation in all cases.
1266
ab8e6d41 1267
949cf498
KW
1268All other code points corresponding to Unicode characters, including private
1269use and those yet to be assigned, are never considered malformed and never
1270warn.
67e989fb 1271
37607a96 1272=cut
f9380377
KW
1273
1274Also implemented as a macro in utf8.h
1275*/
1276
1277UV
1278Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1279 STRLEN curlen,
1280 STRLEN *retlen,
1281 const U32 flags)
1282{
1283 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1284
1285 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1286}
1287
2b9519f0
KW
1288/* The tables below come from http://bjoern.hoehrmann.de/utf-8/decoder/dfa/,
1289 * which requires this copyright notice */
1290
1291/* Copyright (c) 2008-2009 Bjoern Hoehrmann <bjoern@hoehrmann.de>
1292
1293Permission is hereby granted, free of charge, to any person obtaining a copy of
1294this software and associated documentation files (the "Software"), to deal in
1295the Software without restriction, including without limitation the rights to
1296use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
1297of the Software, and to permit persons to whom the Software is furnished to do
1298so, subject to the following conditions:
1299
1300The above copyright notice and this permission notice shall be included in all
1301copies or substantial portions of the Software.
1302
1303THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
1304IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
1305FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
1306AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
1307LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
1308OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
1309SOFTWARE.
1310
1311*/
1312
1313#if 0
1314static U8 utf8d_C9[] = {
1315 /* The first part of the table maps bytes to character classes that
1316 * to reduce the size of the transition table and create bitmasks. */
1317 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1318 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1319 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1320 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1321 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1322 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1323 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1324 10,3,3,3,3,3,3,3,3,3,3,3,3,4,3,3, 11,6,6,6,5,8,8,8,8,8,8,8,8,8,8,8, /*-FF*/
1325
1326 /* The second part is a transition table that maps a combination
1327 * of a state of the automaton and a character class to a state. */
1328 0,12,24,36,60,96,84,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,
1329 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,
1330 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,
1331 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,
1332 12,36,12,12,12,12,12,12,12,12,12,12
1333};
1334
1335#endif
1336
1337#ifndef EBCDIC
1338
1339/* This is a version of the above table customized for Perl that doesn't
1340 * exclude surrogates and accepts start bytes up through F7 (representing
1341 * 2**21 - 1). */
1342static U8 dfa_tab_for_perl[] = {
1343 /* The first part of the table maps bytes to character classes to reduce
1344 * the size of the transition table and create bitmasks. */
1345 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1346 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1347 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1348 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1349 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1350 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1351 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1352 10,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, 11,4,4,4,4,4,4,4,8,8,8,8,8,8,8,8, /*-FF*/
1353
1354 /* The second part is a transition table that maps a combination
1355 * of a state of the automaton and a character class to a state. */
1356 0,12,24,36,96,12,12,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,/*23*/
1357 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,/*47*/
1358 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,/*71*/
1359 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,/*95*/
1360 12,36,12,12,12,12,12,36,12,36,12,12 /* 96- 107 */
1361
1362 /* The customization was to repurpose the surrogates type '4' to instead be
1363 * for start bytes F1-F7. Types 5 and 6 are now unused, and their entries in
1364 * the transition part of the table are set to 12, so are illegal.
1365 *
1366 * To do higher code points would require expansion and some rearrangement of
1367 * the table. The type '1' entries for continuation bytes 80-8f would have to
1368 * be split into several types, because they aren't treated uniformly for
1369 * higher start bytes, since overlongs for F8 are 80-87; FC: 80-83; and FE:
1370 * 80-81. We start needing to worry about overflow if FE is included.
1371 * Ignoring, FE and FF, we could use type 5 for F9-FB, and 6 for FD (remember
1372 * from the web site that these are used to right shift). FE would
1373 * necessarily be type 7; and FF, type 8. And new states would have to be
1374 * created for F8 and FC (and FE and FF if used), so quite a bit of work would
1375 * be involved.
1376 *
1377 * XXX Better would be to customize the table so that the noncharacters are
1378 * excluded. This again is non trivial, but doing so would simplify the code
1379 * that uses this, and might make it small enough to make it inlinable */
1380};
1381
1382#endif
1383
f9380377
KW
1384/*
1385
1386=for apidoc utf8n_to_uvchr_error
1387
1388THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1389Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1390
1391This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1392are when an error is found. If you also need to know the generated warning
1393messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1394
1395It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1396all the others, C<errors>. If this parameter is 0, this function behaves
1397identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1398to a C<U32> variable, which this function sets to indicate any errors found.
1399Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1400C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1401of these bits will be set if a malformation is found, even if the input
7a65503b 1402C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1403exceptions are noted:
1404
1405=over 4
1406
57ff5f59 1407=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1408
57ff5f59
KW
1409The input sequence is not standard UTF-8, but a Perl extension. This bit is
1410set only if the input C<flags> parameter contains either the
1411C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1412
1413Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1414and so some extension must be used to express them. Perl uses a natural
1415extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1416extension to represent even higher ones, so that any code point that fits in a
141764-bit word can be represented. Text using these extensions is not likely to
1418be portable to non-Perl code. We lump both of these extensions together and
1419refer to them as Perl extended UTF-8. There exist other extensions that people
1420have invented, incompatible with Perl's.
1421
1422On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1423extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1424than on ASCII. Prior to that, code points 2**31 and higher were simply
1425unrepresentable, and a different, incompatible method was used to represent
1426code points between 2**30 and 2**31 - 1.
1427
1428On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1429Perl extended UTF-8 is used.
1430
1431In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1432may use for backward compatibility. That name is misleading, as this flag may
1433be set when the code point actually does fit in 31 bits. This happens on
1434EBCDIC platforms, and sometimes when the L<overlong
1435malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1436describes the situation in all cases.
f9380377
KW
1437
1438=item C<UTF8_GOT_CONTINUATION>
1439
1440The input sequence was malformed in that the first byte was a a UTF-8
1441continuation byte.
1442
1443=item C<UTF8_GOT_EMPTY>
1444
1445The input C<curlen> parameter was 0.
1446
1447=item C<UTF8_GOT_LONG>
1448
1449The input sequence was malformed in that there is some other sequence that
1450evaluates to the same code point, but that sequence is shorter than this one.
1451
fecaf136
KW
1452Until Unicode 3.1, it was legal for programs to accept this malformation, but
1453it was discovered that this created security issues.
1454
f9380377
KW
1455=item C<UTF8_GOT_NONCHAR>
1456
1457The code point represented by the input UTF-8 sequence is for a Unicode
1458non-character code point.
1459This bit is set only if the input C<flags> parameter contains either the
1460C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1461
1462=item C<UTF8_GOT_NON_CONTINUATION>
1463
1464The input sequence was malformed in that a non-continuation type byte was found
1465in a position where only a continuation type one should be.
1466
1467=item C<UTF8_GOT_OVERFLOW>
1468
1469The input sequence was malformed in that it is for a code point that is not
d22ec717 1470representable in the number of bits available in an IV on the current platform.
f9380377
KW
1471
1472=item C<UTF8_GOT_SHORT>
1473
1474The input sequence was malformed in that C<curlen> is smaller than required for
1475a complete sequence. In other words, the input is for a partial character
1476sequence.
1477
1478=item C<UTF8_GOT_SUPER>
1479
1480The input sequence was malformed in that it is for a non-Unicode code point;
1481that is, one above the legal Unicode maximum.
1482This bit is set only if the input C<flags> parameter contains either the
1483C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1484
1485=item C<UTF8_GOT_SURROGATE>
1486
1487The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1488code point.
1489This bit is set only if the input C<flags> parameter contains either the
1490C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1491
1492=back
1493
133551d8
KW
1494To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1495flag to suppress any warnings, and then examine the C<*errors> return.
1496
f9380377 1497=cut
37657a5b
KW
1498
1499Also implemented as a macro in utf8.h
37607a96 1500*/
67e989fb 1501
a0ed51b3 1502UV
f9380377 1503Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
37657a5b
KW
1504 STRLEN curlen,
1505 STRLEN *retlen,
1506 const U32 flags,
1507 U32 * errors)
1508{
1509 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1510
1511 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1512}
1513
1514/*
1515
1516=for apidoc utf8n_to_uvchr_msgs
1517
1518THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1519Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1520
1521This function is for code that needs to know what the precise malformation(s)
1522are when an error is found, and wants the corresponding warning and/or error
1523messages to be returned to the caller rather than be displayed. All messages
1524that would have been displayed if all lexcial warnings are enabled will be
1525returned.
1526
1527It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1528placed after all the others, C<msgs>. If this parameter is 0, this function
1529behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1530be a pointer to an C<AV *> variable, in which this function creates a new AV to
1531contain any appropriate messages. The elements of the array are ordered so
1532that the first message that would have been displayed is in the 0th element,
1533and so on. Each element is a hash with three key-value pairs, as follows:
1534
1535=over 4
1536
1537=item C<text>
1538
1539The text of the message as a C<SVpv>.
1540
1541=item C<warn_categories>
1542
1543The warning category (or categories) packed into a C<SVuv>.
1544
1545=item C<flag>
1546
1547A single flag bit associated with this message, in a C<SVuv>.
1548The bit corresponds to some bit in the C<*errors> return value,
1549such as C<UTF8_GOT_LONG>.
1550
1551=back
1552
1553It's important to note that specifying this parameter as non-null will cause
1554any warnings this function would otherwise generate to be suppressed, and
1555instead be placed in C<*msgs>. The caller can check the lexical warnings state
1556(or not) when choosing what to do with the returned messages.
1557
1558If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1559no AV is created.
1560
1561The caller, of course, is responsible for freeing any returned AV.
1562
1563=cut
1564*/
1565
1566UV
1567Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1568 STRLEN curlen,
1569 STRLEN *retlen,
1570 const U32 flags,
1571 U32 * errors,
1572 AV ** msgs)
a0ed51b3 1573{
d4c19fe8 1574 const U8 * const s0 = s;
2b9519f0 1575 const U8 * send = s0 + curlen;
2b5e7bc2
KW
1576 U32 possible_problems = 0; /* A bit is set here for each potential problem
1577 found as we go along */
2b672cf5 1578 UV uv = (UV) -1;
2b5e7bc2
KW
1579 STRLEN expectlen = 0; /* How long should this sequence be?
1580 (initialized to silence compilers' wrong
1581 warning) */
e308b348 1582 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1583 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1584 this gets set and discarded */
a0dbb045 1585
2b5e7bc2
KW
1586 /* The below are used only if there is both an overlong malformation and a
1587 * too short one. Otherwise the first two are set to 's0' and 'send', and
1588 * the third not used at all */
1589 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1590 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1591 routine; see [perl #130921] */
2b5e7bc2 1592 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1593
2b9519f0
KW
1594 UV state = 0;
1595
37657a5b 1596 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
f9380377
KW
1597
1598 if (errors) {
1599 *errors = 0;
1600 }
1601 else {
1602 errors = &discard_errors;
1603 }
a0dbb045 1604
eb83ed87
KW
1605 /* The order of malformation tests here is important. We should consume as
1606 * few bytes as possible in order to not skip any valid character. This is
1607 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1608 * http://unicode.org/reports/tr36 for more discussion as to why. For
1609 * example, once we've done a UTF8SKIP, we can tell the expected number of
1610 * bytes, and could fail right off the bat if the input parameters indicate
1611 * that there are too few available. But it could be that just that first
1612 * byte is garbled, and the intended character occupies fewer bytes. If we
1613 * blindly assumed that the first byte is correct, and skipped based on
1614 * that number, we could skip over a valid input character. So instead, we
1615 * always examine the sequence byte-by-byte.
1616 *
1617 * We also should not consume too few bytes, otherwise someone could inject
1618 * things. For example, an input could be deliberately designed to
1619 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1620 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1621 * which is actually part of of the overflowing sequence), that could look
1622 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1623 * sequence and process the rest, inappropriately.
1624 *
1625 * Some possible input sequences are malformed in more than one way. This
1626 * function goes to lengths to try to find all of them. This is necessary
1627 * for correctness, as the inputs may allow one malformation but not
1628 * another, and if we abandon searching for others after finding the
1629 * allowed one, we could allow in something that shouldn't have been.
1630 */
eb83ed87 1631
b5b9af04 1632 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1633 possible_problems |= UTF8_GOT_EMPTY;
1634 curlen = 0;
5a48568d 1635 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1636 goto ready_to_handle_errors;
0c443dc2
JH
1637 }
1638
eb83ed87
KW
1639 expectlen = UTF8SKIP(s);
1640
1641 /* A well-formed UTF-8 character, as the vast majority of calls to this
1642 * function will be for, has this expected length. For efficiency, set
1643 * things up here to return it. It will be overriden only in those rare
1644 * cases where a malformation is found */
1645 if (retlen) {
1646 *retlen = expectlen;
1647 }
1648
1649 /* An invariant is trivially well-formed */
2b9519f0
KW
1650 if (UTF8_IS_INVARIANT(*s0)) {
1651 return *s0;
a0ed51b3 1652 }
67e989fb 1653
2b9519f0
KW
1654#ifndef EBCDIC
1655
1656 /* Measurements show that this dfa is somewhat faster than the regular code
1657 * below, so use it first, dropping down for the non-normal cases. */
1658
1659# define PERL_UTF8_DECODE_REJECT 12
1660
1661 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
1662 UV type = dfa_tab_for_perl[*s];
1663
1664 if (state != 0) {
1665 uv = (*s & 0x3fu) | (uv << UTF_ACCUMULATION_SHIFT);
1666 state = dfa_tab_for_perl[256 + state + type];
1667 }
1668 else {
1669 uv = (0xff >> type) & (*s);
1670 state = dfa_tab_for_perl[256 + type];
1671 }
1672
1673 if (state == 0) {
1674
1675 /* If this could be a code point that the flags don't allow (the first
1676 * surrogate is the first such possible one), delve further, but we already
1677 * have calculated 'uv' */
1678 if ( (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1679 |UTF8_WARN_ILLEGAL_INTERCHANGE))
1680 && uv >= UNICODE_SURROGATE_FIRST)
1681 {
1682 curlen = s + 1 - s0;
1683 goto got_uv;
1684 }
1685
1686 return uv;
1687 }
1688
1689 s++;
1690 }
1691
1692 /* Here, is some sort of failure. Use the full mechanism */
1693
1694 uv = *s0;
1695
1696#endif
1697
eb83ed87 1698 /* A continuation character can't start a valid sequence */
b5b9af04 1699 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1700 possible_problems |= UTF8_GOT_CONTINUATION;
1701 curlen = 1;
1702 uv = UNICODE_REPLACEMENT;
1703 goto ready_to_handle_errors;
ba210ebe 1704 }
9041c2e3 1705
dcd27b3c 1706 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1707 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1708 * because it excludes start bytes like \xC0 that always lead to
1709 * overlongs.) */
dcd27b3c 1710
534752c1
KW
1711 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1712 * that indicate the number of bytes in the character's whole UTF-8
1713 * sequence, leaving just the bits that are part of the value. */
1714 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1715
e308b348
KW
1716 /* Setup the loop end point, making sure to not look past the end of the
1717 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1718 if (UNLIKELY(curlen < expectlen)) {
1719 possible_problems |= UTF8_GOT_SHORT;
1720 avail_len = curlen;
e308b348
KW
1721 }
1722 else {
2b9519f0 1723 send = (U8*) s0 + expectlen;
e308b348 1724 }
e308b348 1725
eb83ed87 1726 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1727 * accumulating each into the working value as we go. */
eb83ed87 1728 for (s = s0 + 1; s < send; s++) {
b5b9af04 1729 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1730 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1731 continue;
1732 }
1733
1734 /* Here, found a non-continuation before processing all expected bytes.
1735 * This byte indicates the beginning of a new character, so quit, even
1736 * if allowing this malformation. */
2b5e7bc2 1737 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1738 break;
eb83ed87
KW
1739 } /* End of loop through the character's bytes */
1740
1741 /* Save how many bytes were actually in the character */
1742 curlen = s - s0;
1743
2b5e7bc2
KW
1744 /* Note that there are two types of too-short malformation. One is when
1745 * there is actual wrong data before the normal termination of the
1746 * sequence. The other is that the sequence wasn't complete before the end
1747 * of the data we are allowed to look at, based on the input 'curlen'.
1748 * This means that we were passed data for a partial character, but it is
1749 * valid as far as we saw. The other is definitely invalid. This
1750 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1751 * separate.
1752 *
1753 * A convenience macro that matches either of the too-short conditions. */
1754# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1755
1756 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1757 uv_so_far = uv;
1758 uv = UNICODE_REPLACEMENT;
1759 }
2b5e7bc2 1760
08e73697
KW
1761 /* Check for overflow. The algorithm requires us to not look past the end
1762 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1763 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1764 1 /* Do consider overlongs */
1765 )))
1766 {
2b5e7bc2
KW
1767 possible_problems |= UTF8_GOT_OVERFLOW;
1768 uv = UNICODE_REPLACEMENT;
eb83ed87 1769 }
eb83ed87 1770
2b5e7bc2
KW
1771 /* Check for overlong. If no problems so far, 'uv' is the correct code
1772 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1773 * we must look at the UTF-8 byte sequence itself to see if it is for an
1774 * overlong */
1775 if ( ( LIKELY(! possible_problems)
1776 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1777 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1778 && ( UNLIKELY(! UTF8_IS_START(*s0))
1779 || ( curlen > 1
d6be65ae 1780 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1781 s - s0))))))
2f8f112e 1782 {
2b5e7bc2
KW
1783 possible_problems |= UTF8_GOT_LONG;
1784
abc28b54 1785 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1786
abc28b54
KW
1787 /* The calculation in the 'true' branch of this 'if'
1788 * below won't work if overflows, and isn't needed
1789 * anyway. Further below we handle all overflow
1790 * cases */
1791 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1792 {
2b5e7bc2
KW
1793 UV min_uv = uv_so_far;
1794 STRLEN i;
1795
1796 /* Here, the input is both overlong and is missing some trailing
1797 * bytes. There is no single code point it could be for, but there
1798 * may be enough information present to determine if what we have
1799 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1800 * The code further below has the intelligence to determine this,
1801 * but just for non-overlong UTF-8 sequences. What we do here is
1802 * calculate the smallest code point the input could represent if
1803 * there were no too short malformation. Then we compute and save
1804 * the UTF-8 for that, which is what the code below looks at
1805 * instead of the raw input. It turns out that the smallest such
1806 * code point is all we need. */
2b5e7bc2
KW
1807 for (i = curlen; i < expectlen; i++) {
1808 min_uv = UTF8_ACCUMULATE(min_uv,
1809 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1810 }
1811
e9f2c446 1812 adjusted_s0 = temp_char_buf;
57ff5f59 1813 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1814 }
eb83ed87
KW
1815 }
1816
2b9519f0
KW
1817 got_uv:
1818
56576a04
KW
1819 /* Here, we have found all the possible problems, except for when the input
1820 * is for a problematic code point not allowed by the input parameters. */
1821
06188866
KW
1822 /* uv is valid for overlongs */
1823 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1824
1825 /* isn't problematic if < this */
1826 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1827 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1828
1829 /* if overflow, we know without looking further
1830 * precisely which of the problematic types it is,
1831 * and we deal with those in the overflow handling
1832 * code */
1833 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1834 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1835 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1836 && ((flags & ( UTF8_DISALLOW_NONCHAR
1837 |UTF8_DISALLOW_SURROGATE
1838 |UTF8_DISALLOW_SUPER
d044b7a7 1839 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1840 |UTF8_WARN_NONCHAR
1841 |UTF8_WARN_SURROGATE
1842 |UTF8_WARN_SUPER
d22ec717 1843 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1844 {
2b5e7bc2
KW
1845 /* If there were no malformations, or the only malformation is an
1846 * overlong, 'uv' is valid */
1847 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1848 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1849 possible_problems |= UTF8_GOT_SURROGATE;
1850 }
1851 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1852 possible_problems |= UTF8_GOT_SUPER;
1853 }
1854 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1855 possible_problems |= UTF8_GOT_NONCHAR;
1856 }
1857 }
1858 else { /* Otherwise, need to look at the source UTF-8, possibly
1859 adjusted to be non-overlong */
1860
1861 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1862 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1863 {
2b5e7bc2
KW
1864 possible_problems |= UTF8_GOT_SUPER;
1865 }
1866 else if (curlen > 1) {
1867 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1868 NATIVE_UTF8_TO_I8(*adjusted_s0),
1869 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1870 {
2b5e7bc2 1871 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1872 }
2b5e7bc2
KW
1873 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1874 NATIVE_UTF8_TO_I8(*adjusted_s0),
1875 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1876 {
1877 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1878 }
1879 }
c0236afe 1880
2b5e7bc2
KW
1881 /* We need a complete well-formed UTF-8 character to discern
1882 * non-characters, so can't look for them here */
1883 }
1884 }
949cf498 1885
2b5e7bc2
KW
1886 ready_to_handle_errors:
1887
1888 /* At this point:
1889 * curlen contains the number of bytes in the sequence that
1890 * this call should advance the input by.
e308b348
KW
1891 * avail_len gives the available number of bytes passed in, but
1892 * only if this is less than the expected number of
1893 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1894 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1895 * is set in it for each potential problem found.
1896 * uv contains the code point the input sequence
1897 * represents; or if there is a problem that prevents
1898 * a well-defined value from being computed, it is
1899 * some subsitute value, typically the REPLACEMENT
1900 * CHARACTER.
1901 * s0 points to the first byte of the character
56576a04
KW
1902 * s points to just after were we left off processing
1903 * the character
1904 * send points to just after where that character should
1905 * end, based on how many bytes the start byte tells
1906 * us should be in it, but no further than s0 +
1907 * avail_len
2b5e7bc2 1908 */
eb83ed87 1909
2b5e7bc2
KW
1910 if (UNLIKELY(possible_problems)) {
1911 bool disallowed = FALSE;
1912 const U32 orig_problems = possible_problems;
1913
37657a5b
KW
1914 if (msgs) {
1915 *msgs = NULL;
1916 }
1917
2b5e7bc2
KW
1918 while (possible_problems) { /* Handle each possible problem */
1919 UV pack_warn = 0;
1920 char * message = NULL;
37657a5b 1921 U32 this_flag_bit = 0;
2b5e7bc2
KW
1922
1923 /* Each 'if' clause handles one problem. They are ordered so that
1924 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1925 * ones; this is kinda in decreasing severity order. But the
1926 * overlong must come last, as it changes 'uv' looked at by the
1927 * others */
2b5e7bc2
KW
1928 if (possible_problems & UTF8_GOT_OVERFLOW) {
1929
56576a04
KW
1930 /* Overflow means also got a super and are using Perl's
1931 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1932 possible_problems
d044b7a7 1933 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1934 *errors |= UTF8_GOT_OVERFLOW;
1935
1936 /* But the API says we flag all errors found */
1937 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1938 *errors |= UTF8_GOT_SUPER;
1939 }
ddb65933 1940 if (flags
d044b7a7 1941 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1942 {
d044b7a7 1943 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1944 }
2b5e7bc2 1945
d60baaa7 1946 /* Disallow if any of the three categories say to */
56576a04 1947 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1948 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1949 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1950 {
1951 disallowed = TRUE;
1952 }
1953
d22ec717
KW
1954 /* Likewise, warn if any say to */
1955 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1956 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1957 {
2b5e7bc2 1958
ddb65933
KW
1959 /* The warnings code explicitly says it doesn't handle the
1960 * case of packWARN2 and two categories which have
1961 * parent-child relationship. Even if it works now to
1962 * raise the warning if either is enabled, it wouldn't
1963 * necessarily do so in the future. We output (only) the
56576a04 1964 * most dire warning */
ddb65933 1965 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1966 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1967 pack_warn = packWARN(WARN_UTF8);
1968 }
37657a5b 1969 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1970 pack_warn = packWARN(WARN_NON_UNICODE);
1971 }
1972 if (pack_warn) {
1973 message = Perl_form(aTHX_ "%s: %s (overflows)",
1974 malformed_text,
05b9033b 1975 _byte_dump_string(s0, curlen, 0));
37657a5b 1976 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1977 }
2b5e7bc2
KW
1978 }
1979 }
1980 }
1981 else if (possible_problems & UTF8_GOT_EMPTY) {
1982 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1983 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1984
1985 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1986
1987 /* This so-called malformation is now treated as a bug in
1988 * the caller. If you have nothing to decode, skip calling
1989 * this function */
1990 assert(0);
1991
2b5e7bc2 1992 disallowed = TRUE;
37657a5b
KW
1993 if ( (msgs
1994 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1995 {
2b5e7bc2
KW
1996 pack_warn = packWARN(WARN_UTF8);
1997 message = Perl_form(aTHX_ "%s (empty string)",
1998 malformed_text);
37657a5b 1999 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
2000 }
2001 }
2002 }
2003 else if (possible_problems & UTF8_GOT_CONTINUATION) {
2004 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 2005 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
2006
2007 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
2008 disallowed = TRUE;
37657a5b
KW
2009 if (( msgs
2010 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2011 {
2b5e7bc2
KW
2012 pack_warn = packWARN(WARN_UTF8);
2013 message = Perl_form(aTHX_
2014 "%s: %s (unexpected continuation byte 0x%02x,"
2015 " with no preceding start byte)",
2016 malformed_text,
7e2f38b2 2017 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 2018 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
2019 }
2020 }
2021 }
2b5e7bc2
KW
2022 else if (possible_problems & UTF8_GOT_SHORT) {
2023 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 2024 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
2025
2026 if (! (flags & UTF8_ALLOW_SHORT)) {
2027 disallowed = TRUE;
37657a5b
KW
2028 if (( msgs
2029 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2030 {
2b5e7bc2
KW
2031 pack_warn = packWARN(WARN_UTF8);
2032 message = Perl_form(aTHX_
56576a04
KW
2033 "%s: %s (too short; %d byte%s available, need %d)",
2034 malformed_text,
2035 _byte_dump_string(s0, send - s0, 0),
2036 (int)avail_len,
2037 avail_len == 1 ? "" : "s",
2038 (int)expectlen);
37657a5b 2039 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
2040 }
2041 }
ba210ebe 2042
2b5e7bc2 2043 }
e308b348
KW
2044 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2045 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2046 *errors |= UTF8_GOT_NON_CONTINUATION;
2047
2048 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2049 disallowed = TRUE;
37657a5b
KW
2050 if (( msgs
2051 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2052 {
99a765e9
KW
2053
2054 /* If we don't know for sure that the input length is
2055 * valid, avoid as much as possible reading past the
2056 * end of the buffer */
2057 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2058 ? s - s0
2059 : send - s0;
e308b348
KW
2060 pack_warn = packWARN(WARN_UTF8);
2061 message = Perl_form(aTHX_ "%s",
2062 unexpected_non_continuation_text(s0,
99a765e9 2063 printlen,
e308b348
KW
2064 s - s0,
2065 (int) expectlen));
37657a5b 2066 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2067 }
2068 }
2069 }
2b5e7bc2
KW
2070 else if (possible_problems & UTF8_GOT_SURROGATE) {
2071 possible_problems &= ~UTF8_GOT_SURROGATE;
2072
f9380377
KW
2073 if (flags & UTF8_WARN_SURROGATE) {
2074 *errors |= UTF8_GOT_SURROGATE;
2075
2076 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2077 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2078 {
2b5e7bc2
KW
2079 pack_warn = packWARN(WARN_SURROGATE);
2080
2081 /* These are the only errors that can occur with a
2082 * surrogate when the 'uv' isn't valid */
2083 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2084 message = Perl_form(aTHX_
2085 "UTF-16 surrogate (any UTF-8 sequence that"
2086 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2087 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2088 }
2089 else {
c94c2f39 2090 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2091 }
37657a5b 2092 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2093 }
2b5e7bc2 2094 }
ba210ebe 2095
2b5e7bc2
KW
2096 if (flags & UTF8_DISALLOW_SURROGATE) {
2097 disallowed = TRUE;
f9380377 2098 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2099 }
2100 }
2101 else if (possible_problems & UTF8_GOT_SUPER) {
2102 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2103
f9380377
KW
2104 if (flags & UTF8_WARN_SUPER) {
2105 *errors |= UTF8_GOT_SUPER;
2106
2107 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2108 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2109 {
2b5e7bc2
KW
2110 pack_warn = packWARN(WARN_NON_UNICODE);
2111
2112 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2113 message = Perl_form(aTHX_
2114 "Any UTF-8 sequence that starts with"
2115 " \"%s\" is for a non-Unicode code point,"
2116 " may not be portable",
7e2f38b2 2117 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2118 }
2119 else {
c94c2f39 2120 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2121 }
37657a5b 2122 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2123 }
2b5e7bc2 2124 }
ba210ebe 2125
57ff5f59
KW
2126 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2127 * and before possibly bailing out, so that the more dire
2128 * warning will override the regular one. */
2129 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2130 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2131 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2132 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2133 {
db0f09e6 2134 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2135
57ff5f59
KW
2136 /* If it is an overlong that evaluates to a code point
2137 * that doesn't have to use the Perl extended UTF-8, it
2138 * still used it, and so we output a message that
2139 * doesn't refer to the code point. The same is true
2140 * if there was a SHORT malformation where the code
2141 * point is not valid. In that case, 'uv' will have
2142 * been set to the REPLACEMENT CHAR, and the message
2143 * below without the code point in it will be selected
2144 * */
2145 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2146 message = Perl_form(aTHX_
57ff5f59 2147 perl_extended_cp_format, uv);
2b5e7bc2
KW
2148 }
2149 else {
2150 message = Perl_form(aTHX_
57ff5f59
KW
2151 "Any UTF-8 sequence that starts with"
2152 " \"%s\" is a Perl extension, and"
2153 " so is not portable",
2154 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2155 }
37657a5b 2156 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2157 }
2158
d044b7a7
KW
2159 if (flags & ( UTF8_WARN_PERL_EXTENDED
2160 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2161 {
d044b7a7 2162 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2163
d044b7a7 2164 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2165 disallowed = TRUE;
2166 }
2b5e7bc2
KW
2167 }
2168 }
eb83ed87 2169
2b5e7bc2 2170 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2171 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2172 disallowed = TRUE;
2173 }
2b5e7bc2
KW
2174 }
2175 else if (possible_problems & UTF8_GOT_NONCHAR) {
2176 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2177
f9380377
KW
2178 if (flags & UTF8_WARN_NONCHAR) {
2179 *errors |= UTF8_GOT_NONCHAR;
2180
2181 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2182 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2183 {
2b5e7bc2
KW
2184 /* The code above should have guaranteed that we don't
2185 * get here with errors other than overlong */
2186 assert (! (orig_problems
2187 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2188
2189 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2190 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2191 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2192 }
2b5e7bc2 2193 }
5b311467 2194
2b5e7bc2
KW
2195 if (flags & UTF8_DISALLOW_NONCHAR) {
2196 disallowed = TRUE;
f9380377 2197 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2198 }
6c64cd9d
KW
2199 }
2200 else if (possible_problems & UTF8_GOT_LONG) {
2201 possible_problems &= ~UTF8_GOT_LONG;
2202 *errors |= UTF8_GOT_LONG;
2203
2204 if (flags & UTF8_ALLOW_LONG) {
2205
2206 /* We don't allow the actual overlong value, unless the
2207 * special extra bit is also set */
2208 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2209 & ~UTF8_ALLOW_LONG)))
2210 {
2211 uv = UNICODE_REPLACEMENT;
2212 }
2213 }
2214 else {
2215 disallowed = TRUE;
2216
37657a5b
KW
2217 if (( msgs
2218 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2219 {
6c64cd9d
KW
2220 pack_warn = packWARN(WARN_UTF8);
2221
2222 /* These error types cause 'uv' to be something that
2223 * isn't what was intended, so can't use it in the
2224 * message. The other error types either can't
2225 * generate an overlong, or else the 'uv' is valid */
2226 if (orig_problems &
2227 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2228 {
2229 message = Perl_form(aTHX_
2230 "%s: %s (any UTF-8 sequence that starts"
2231 " with \"%s\" is overlong which can and"
2232 " should be represented with a"
2233 " different, shorter sequence)",
2234 malformed_text,
2235 _byte_dump_string(s0, send - s0, 0),
2236 _byte_dump_string(s0, curlen, 0));
2237 }
2238 else {
2239 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2240 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2241 uv, 0);
d819dc50
KW
2242 /* Don't use U+ for non-Unicode code points, which
2243 * includes those in the Latin1 range */
2244 const char * preface = ( uv > PERL_UNICODE_MAX
2245#ifdef EBCDIC
2246 || uv <= 0xFF
2247#endif
2248 )
2249 ? "0x"
2250 : "U+";
6c64cd9d
KW
2251 message = Perl_form(aTHX_
2252 "%s: %s (overlong; instead use %s to represent"
2253 " %s%0*" UVXf ")",
2254 malformed_text,
2255 _byte_dump_string(s0, send - s0, 0),
2256 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2257 preface,
2258 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2259 small code points */
1be62ab9 2260 UNI_TO_NATIVE(uv));
6c64cd9d 2261 }
37657a5b 2262 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2263 }
2264 }
2b5e7bc2
KW
2265 } /* End of looking through the possible flags */
2266
2267 /* Display the message (if any) for the problem being handled in
2268 * this iteration of the loop */
2269 if (message) {
37657a5b 2270 if (msgs) {
37657a5b
KW
2271 assert(this_flag_bit);
2272
2273 if (*msgs == NULL) {
2274 *msgs = newAV();
2275 }
2276
bb07812e
KW
2277 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2278 pack_warn,
2279 this_flag_bit)));
37657a5b
KW
2280 }
2281 else if (PL_op)
2b5e7bc2
KW
2282 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2283 OP_DESC(PL_op));
2284 else
2285 Perl_warner(aTHX_ pack_warn, "%s", message);
2286 }
ddb65933 2287 } /* End of 'while (possible_problems)' */
a0dbb045 2288
2b5e7bc2
KW
2289 /* Since there was a possible problem, the returned length may need to
2290 * be changed from the one stored at the beginning of this function.
2291 * Instead of trying to figure out if that's needed, just do it. */
2292 if (retlen) {
2293 *retlen = curlen;
2294 }
a0dbb045 2295
2b5e7bc2
KW
2296 if (disallowed) {
2297 if (flags & UTF8_CHECK_ONLY && retlen) {
2298 *retlen = ((STRLEN) -1);
2299 }
2300 return 0;
2301 }
eb83ed87 2302 }
ba210ebe 2303
2b5e7bc2 2304 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2305}
2306
8e84507e 2307/*
ec5f19d0
KW
2308=for apidoc utf8_to_uvchr_buf
2309
2310Returns the native code point of the first character in the string C<s> which
2311is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2312C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2313
524080c4
KW
2314If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2315enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2316C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2317(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2318C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2319the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2320See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2321returned.
ec5f19d0
KW
2322
2323=cut
52be2536
KW
2324
2325Also implemented as a macro in utf8.h
2326
ec5f19d0
KW
2327*/
2328
2329
2330UV
2331Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2332{
7f974d7e
KW
2333 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2334
ec5f19d0
KW
2335 assert(s < send);
2336
2337 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 2338 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
2339}
2340
52be2536
KW
2341/* This is marked as deprecated
2342 *
ec5f19d0
KW
2343=for apidoc utf8_to_uvuni_buf
2344
de69f3af
KW
2345Only in very rare circumstances should code need to be dealing in Unicode
2346(as opposed to native) code points. In those few cases, use
07dfe0a4
KW
2347C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2348are not absolutely sure this is one of those cases, then assume it isn't and
2349use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2350
2351Returns the Unicode (not-native) code point of the first character in the
2352string C<s> which
ec5f19d0
KW
2353is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2354C<retlen> will be set to the length, in bytes, of that character.
2355
524080c4
KW
2356If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2357enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2358NULL) to -1. If those warnings are off, the computed value if well-defined (or
2359the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2360is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2361next possible position in C<s> that could begin a non-malformed character.
de69f3af 2362See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2363
2364=cut
2365*/
2366
2367UV
2368Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2369{
2370 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2371
2372 assert(send > s);
2373
5962d97e 2374 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2375}
2376
b76347f2 2377/*
87cea99e 2378=for apidoc utf8_length
b76347f2 2379
b2e7ed74
KW
2380Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2381at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2382same place, it returns 0 with no warning raised.
2383
2384If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2385and returns the number of valid characters.
b76347f2
JH
2386
2387=cut
2388*/
2389
2390STRLEN
35a4481c 2391Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2392{
2393 STRLEN len = 0;
2394
7918f24d
NC
2395 PERL_ARGS_ASSERT_UTF8_LENGTH;
2396
8850bf83
JH
2397 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2398 * the bitops (especially ~) can create illegal UTF-8.
2399 * In other words: in Perl UTF-8 is not just for Unicode. */
2400
a3b680e6
AL
2401 if (e < s)
2402 goto warn_and_return;
b76347f2 2403 while (s < e) {
4cbf4130 2404 s += UTF8SKIP(s);
8e91ec7f
AV
2405 len++;
2406 }
2407
2408 if (e != s) {
2409 len--;
2410 warn_and_return:
9b387841
NC
2411 if (PL_op)
2412 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2413 "%s in %s", unees, OP_DESC(PL_op));
2414 else
61a12c31 2415 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2416 }
2417
2418 return len;
2419}
2420
b06226ff 2421/*
fed3ba5d
NC
2422=for apidoc bytes_cmp_utf8
2423
a1433954 2424Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2425sequence of characters (stored as UTF-8)
2426in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2427equal, -1 or -2 if the first string is less than the second string, +1 or +2
2428if the first string is greater than the second string.
2429
2430-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2431longer string. -2 or +2 is returned if
2432there was a difference between characters
fed3ba5d
NC
2433within the strings.
2434
2435=cut
2436*/
2437
2438int
2439Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2440{
2441 const U8 *const bend = b + blen;
2442 const U8 *const uend = u + ulen;
2443
2444 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2445
2446 while (b < bend && u < uend) {
2447 U8 c = *u++;
2448 if (!UTF8_IS_INVARIANT(c)) {
2449 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2450 if (u < uend) {
2451 U8 c1 = *u++;
2452 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2453 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2454 } else {
2b5e7bc2 2455 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2456 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2457 "%s %s%s",
2458 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2459 PL_op ? " in " : "",
2460 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2461 return -2;
2462 }
2463 } else {
2464 if (PL_op)
2465 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2466 "%s in %s", unees, OP_DESC(PL_op));
2467 else
61a12c31 2468 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2469 return -2; /* Really want to return undef :-) */
2470 }
2471 } else {
2472 return -2;
2473 }
2474 }
2475 if (*b != c) {
2476 return *b < c ? -2 : +2;
2477 }
2478 ++b;
2479 }
2480
2481 if (b == bend && u == uend)
2482 return 0;
2483
2484 return b < bend ? +1 : -1;
2485}
2486
2487/*
87cea99e 2488=for apidoc utf8_to_bytes
6940069f 2489
3bc0c78c 2490Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2491Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2492updates C<*lenp> to contain the new length.
3bc0c78c
KW
2493Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2494
2495Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2496having saved the value of C<*lenp> before the call, and subtracting the
2497after-call value of C<*lenp> from it.
6940069f 2498
a1433954 2499If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2500
6940069f
GS
2501=cut
2502*/
2503
2504U8 *
09af0336 2505Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2506{
9fe0d3c2 2507 U8 * first_variant;
246fae53 2508
7918f24d 2509 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2510 PERL_UNUSED_CONTEXT;
7918f24d 2511
9fe0d3c2 2512 /* This is a no-op if no variants at all in the input */
09af0336 2513 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2514 return s;
2515 }
2516
2517 {
3c5aa262 2518 U8 * const save = s;
09af0336 2519 U8 * const send = s + *lenp;
3c5aa262
KW
2520 U8 * d;
2521
2522 /* Nothing before the first variant needs to be changed, so start the real
2523 * work there */
2524 s = first_variant;
2525 while (s < send) {
2526 if (! UTF8_IS_INVARIANT(*s)) {
2527 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2528 *lenp = ((STRLEN) -1);
3c5aa262
KW
2529 return 0;
2530 }
2531 s++;
d59937ca
KW
2532 }
2533 s++;
dcad2880 2534 }
dcad2880 2535
3c5aa262
KW
2536 /* Is downgradable, so do it */
2537 d = s = first_variant;
2538 while (s < send) {
2539 U8 c = *s++;
2540 if (! UVCHR_IS_INVARIANT(c)) {
2541 /* Then it is two-byte encoded */
2542 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2543 s++;
2544 }
2545 *d++ = c;
2546 }
2547 *d = '\0';
09af0336 2548 *lenp = d - save;
3c5aa262
KW
2549
2550 return save;
9fe0d3c2 2551 }
6940069f
GS
2552}
2553
2554/*
87cea99e 2555=for apidoc bytes_from_utf8
f9a63242 2556
09af0336 2557Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2558byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2559actually encoded in UTF-8.
2560
2561Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2562the input string.
2563
41ae6089
KW
2564Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2565not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2566C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2567
41ae6089 2568Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2569newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2570returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2571caller is responsible for arranging for the memory used by this string to get
2572freed.
f9a63242 2573
3bc0c78c 2574Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2575having saved the value of C<*lenp> before the call, and subtracting the
2576after-call value of C<*lenp> from it.
3bc0c78c 2577
37607a96 2578=cut
976c1b08
KW
2579
2580There is a macro that avoids this function call, but this is retained for
2581anyone who calls it with the Perl_ prefix */
f9a63242
JH
2582
2583U8 *
41ae6089 2584Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2585{
7918f24d 2586 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2587 PERL_UNUSED_CONTEXT;
f9a63242 2588
976c1b08
KW
2589 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2590}
2591
2592/*
2593No = here because currently externally undocumented
2594for apidoc bytes_from_utf8_loc
2595
2596Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2597to store the location of the first character in C<"s"> that cannot be
2598converted to non-UTF8.
2599
2600If that parameter is C<NULL>, this function behaves identically to
2601C<bytes_from_utf8>.
2602
2603Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2604C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2605
2606Otherwise, the function returns a newly created C<NUL>-terminated string
2607containing the non-UTF8 equivalent of the convertible first portion of
2608C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2609If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2610and C<*first_non_downgradable> is set to C<NULL>.
2611
2612Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2613first character in the original string that wasn't converted. C<*is_utf8p> is
2614unchanged. Note that the new string may have length 0.
2615
2616Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2617C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2618converts as many characters in it as possible stopping at the first one it
385b74be 2619finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2620set to point to that. The function returns the portion that could be converted
2621in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2622not including the terminating C<NUL>. If the very first character in the
2623original could not be converted, C<*lenp> will be 0, and the new string will
2624contain just a single C<NUL>. If the entire input string was converted,
2625C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2626
2627Upon successful return, the number of variants in the converted portion of the
2628string can be computed by having saved the value of C<*lenp> before the call,
2629and subtracting the after-call value of C<*lenp> from it.
2630
2631=cut
2632
2633
2634*/
2635
2636U8 *
2637Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2638{
2639 U8 *d;
2640 const U8 *original = s;
2641 U8 *converted_start;
2642 const U8 *send = s + *lenp;
f9a63242 2643
976c1b08 2644 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2645
976c1b08
KW
2646 if (! *is_utf8p) {
2647 if (first_unconverted) {
2648 *first_unconverted = NULL;
2649 }
2650
2651 return (U8 *) original;
2652 }
2653
2654 Newx(d, (*lenp) + 1, U8);
2655
2656 converted_start = d;
7299a045
KW
2657 while (s < send) {
2658 U8 c = *s++;
2659 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2660
2661 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2662 * have to stop now */
2663 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2664 if (first_unconverted) {
2665 *first_unconverted = s - 1;
2666 goto finish_and_return;
2667 }
2668 else {
2669 Safefree(converted_start);
2670 return (U8 *) original;
2671 }
2672 }
2673
7299a045
KW
2674 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2675 s++;
38af28cf 2676 }
7299a045
KW
2677 *d++ = c;
2678 }
170a1c22 2679
976c1b08
KW
2680 /* Here, converted the whole of the input */
2681 *is_utf8p = FALSE;
2682 if (first_unconverted) {
2683 *first_unconverted = NULL;
170a1c22 2684 }
976c1b08
KW
2685
2686 finish_and_return:
46a08a6f
KW
2687 *d = '\0';
2688 *lenp = d - converted_start;
976c1b08
KW
2689
2690 /* Trim unused space */
2691 Renew(converted_start, *lenp + 1, U8);
2692
2693 return converted_start;
f9a63242
JH
2694}
2695
2696/*
87cea99e 2697=for apidoc bytes_to_utf8
6940069f 2698
09af0336 2699Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2700UTF-8.
09af0336 2701Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2702reflect the new length in bytes. The caller is responsible for arranging for
2703the memory used by this string to get freed.
6940069f 2704
3bc0c78c 2705Upon successful return, the number of variants in the string can be computed by
23b37b12 2706having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2707after-call value of C<*lenp>.
2708
75200dff 2709A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2710
2711If you want to convert to UTF-8 from encodings other than
2712the native (Latin1 or EBCDIC),
a1433954 2713see L</sv_recode_to_utf8>().
c9ada85f 2714
497711e7 2715=cut
6940069f
GS
2716*/
2717
2718U8*
09af0336 2719Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2720{
09af0336 2721 const U8 * const send = s + (*lenp);
6940069f
GS
2722 U8 *d;
2723 U8 *dst;
7918f24d
NC
2724
2725 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2726 PERL_UNUSED_CONTEXT;
6940069f 2727
09af0336 2728 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2729 dst = d;
2730
2731 while (s < send) {
55d09dc8
KW
2732 append_utf8_from_native_byte(*s, &d);
2733 s++;
6940069f 2734 }
2e11cf67 2735
6940069f 2736 *d = '\0';
09af0336 2737 *lenp = d-dst;
2e11cf67
KW
2738
2739 /* Trim unused space */
2740 Renew(dst, *lenp + 1, U8);
2741
6940069f
GS
2742 return dst;
2743}
2744
a0ed51b3 2745/*
624504c5
KW
2746 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2747 * use utf16_to_utf8_reversed().
a0ed51b3 2748 *
624504c5
KW
2749 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2750 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2751 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2752 *
2753 * These functions don't check for overflow. The worst case is every code
2754 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2755 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2756 * destination must be pre-extended to 2 times the source length.
2757 *
2758 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2759
2760U8*
dea0fc0b 2761Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2762{
dea0fc0b
JH
2763 U8* pend;
2764 U8* dstart = d;
2765
7918f24d
NC
2766 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2767
dea0fc0b 2768 if (bytelen & 1)
56576a04
KW
2769 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2770 (UV)bytelen);
dea0fc0b
JH
2771
2772 pend = p + bytelen;
2773
a0ed51b3 2774 while (p < pend) {
dea0fc0b
JH
2775 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2776 p += 2;
2d1545e5 2777 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2778 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2779 continue;
2780 }
56d37426
KW
2781 if (uv <= MAX_UTF8_TWO_BYTE) {
2782 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2783 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2784 continue;
2785 }
ffd0a9d3 2786
46956fad
KW
2787#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2788#define LAST_HIGH_SURROGATE 0xDBFF
2789#define FIRST_LOW_SURROGATE 0xDC00
2790#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2791#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2792
2793 /* This assumes that most uses will be in the first Unicode plane, not
2794 * needing surrogates */
2795 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2796 && uv <= UNICODE_SURROGATE_LAST))
2797 {
2798 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2799 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2800 }
2801 else {
01ea242b 2802 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2803 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2804 || UNLIKELY(low > LAST_LOW_SURROGATE))
2805 {
01ea242b 2806 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2807 }
2808 p += 2;
46956fad 2809 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2810 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2811 }
a0ed51b3 2812 }
56d37426
KW
2813#ifdef EBCDIC
2814 d = uvoffuni_to_utf8_flags(d, uv, 0);
2815#else
ffd0a9d3 2816 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2817 *d++ = (U8)(( uv >> 12) | 0xe0);
2818 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2819 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2820 continue;
2821 }
2822 else {
eb160463
GS
2823 *d++ = (U8)(( uv >> 18) | 0xf0);
2824 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2825 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2826 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2827 continue;
2828 }
56d37426 2829#endif
a0ed51b3 2830 }
dea0fc0b 2831 *newlen = d - dstart;
a0ed51b3
LW
2832 return d;
2833}
2834
2835/* Note: this one is slightly destructive of the source. */
2836
2837U8*
dea0fc0b 2838Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2839{
2840 U8* s = (U8*)p;
d4c19fe8 2841 U8* const send = s + bytelen;
7918f24d
NC
2842
2843 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2844
e0ea5e2d 2845 if (bytelen & 1)
147e3846 2846 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2847 (UV)bytelen);
2848
a0ed51b3 2849 while (s < send) {
d4c19fe8 2850 const U8 tmp = s[0];
a0ed51b3
LW
2851 s[0] = s[1];
2852 s[1] = tmp;
2853 s += 2;
2854 }
dea0fc0b 2855 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2856}
2857
922e8cb4
KW
2858bool
2859Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2860{
dc31b55c 2861 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2862}
2863
f9ae8fb6
JD
2864/* Internal function so we can deprecate the external one, and call
2865 this one from other deprecated functions in this file */
2866
f2645549
KW
2867bool
2868Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2869{
f2645549 2870 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2871
2872 if (*p == '_')
2873 return TRUE;
af0d7385
KW
2874 return is_utf8_common(p, NULL,
2875 "This is buggy if this gets used",
2876 PL_utf8_idstart);
61b19385
KW
2877}
2878
5092f92a 2879bool
eba68aa0
KW
2880Perl__is_uni_perl_idcont(pTHX_ UV c)
2881{
c12658c9 2882 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2883}
2884
2885bool
f91dcd13
KW
2886Perl__is_uni_perl_idstart(pTHX_ UV c)
2887{
c12658c9 2888 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2889}
2890
3a4c58c9 2891UV
56576a04
KW
2892Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2893 const char S_or_s)
3a4c58c9
KW
2894{
2895 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2896 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2897 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2898 * either "SS" or "Ss". Which one to use is passed into the routine in
2899 * 'S_or_s' to avoid a test */
2900
2901 UV converted = toUPPER_LATIN1_MOD(c);
2902
2903 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2904
2905 assert(S_or_s == 'S' || S_or_s == 's');
2906
6f2d5cbc 2907 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2908 characters in this range */
3a4c58c9
KW
2909 *p = (U8) converted;
2910 *lenp = 1;
2911 return converted;
2912 }
2913
2914 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2915 * which it maps to one of them, so as to only have to have one check for
2916 * it in the main case */
2917 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2918 switch (c) {
2919 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2920 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2921 break;
2922 case MICRO_SIGN:
2923 converted = GREEK_CAPITAL_LETTER_MU;
2924 break;
79e064b9
KW
2925#if UNICODE_MAJOR_VERSION > 2 \
2926 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2927 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2928 case LATIN_SMALL_LETTER_SHARP_S:
2929 *(p)++ = 'S';
2930 *p = S_or_s;
2931 *lenp = 2;
2932 return 'S';
79e064b9 2933#endif
3a4c58c9 2934 default:
56576a04
KW
2935 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2936 " '%c' to map to '%c'",
2937 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2938 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2939 }
2940 }
2941
2942 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2943 *p = UTF8_TWO_BYTE_LO(converted);
2944 *lenp = 2;
2945
2946 return converted;
2947}
2948
fe63c520
KW
2949/* If compiled on an early Unicode version, there may not be auxiliary tables
2950 * */
2951#ifndef HAS_UC_AUX_TABLES
2952# define UC_AUX_TABLE_ptrs NULL
2953# define UC_AUX_TABLE_lengths NULL
2954#endif
2955#ifndef HAS_TC_AUX_TABLES
2956# define TC_AUX_TABLE_ptrs NULL
2957# define TC_AUX_TABLE_lengths NULL
2958#endif
2959#ifndef HAS_LC_AUX_TABLES
2960# define LC_AUX_TABLE_ptrs NULL
2961# define LC_AUX_TABLE_lengths NULL
2962#endif
2963#ifndef HAS_CF_AUX_TABLES
2964# define CF_AUX_TABLE_ptrs NULL
2965# define CF_AUX_TABLE_lengths NULL
2966#endif
2967#ifndef HAS_UC_AUX_TABLES
2968# define UC_AUX_TABLE_ptrs NULL
2969# define UC_AUX_TABLE_lengths NULL
2970#endif
2971
50bda2c3
KW
2972/* Call the function to convert a UTF-8 encoded character to the specified case.
2973 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2974 * 's' is a pointer to the first byte of the input character
2975 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2976 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2977 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2978 *
56576a04 2979 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2980 * 'd' */
56576a04 2981#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2982 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2983 Uppercase_Mapping_invmap, \
2984 UC_AUX_TABLE_ptrs, \
2985 UC_AUX_TABLE_lengths, \
2986 "uppercase")
56576a04 2987#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2988 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2989 Titlecase_Mapping_invmap, \
2990 TC_AUX_TABLE_ptrs, \
2991 TC_AUX_TABLE_lengths, \
2992 "titlecase")
56576a04 2993#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2994 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2995 Lowercase_Mapping_invmap, \
2996 LC_AUX_TABLE_ptrs, \
2997 LC_AUX_TABLE_lengths, \
2998 "lowercase")
2999
50bda2c3 3000
b9992569
KW
3001/* This additionally has the input parameter 'specials', which if non-zero will
3002 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 3003 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 3004#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
3005 (specials) \
3006 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
3007 Case_Folding_invmap, \
3008 CF_AUX_TABLE_ptrs, \
3009 CF_AUX_TABLE_lengths, \
3010 "foldcase") \
3011 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
3012 Simple_Case_Folding_invmap, \
3013 NULL, NULL, \
3014 "foldcase")
c3fd2246 3015
84afefe6
JH
3016UV
3017Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3018{
a1433954
KW
3019 /* Convert the Unicode character whose ordinal is <c> to its uppercase
3020 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
3021 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
3022 * the changed version may be longer than the original character.
3023 *
3024 * The ordinal of the first character of the changed version is returned
3025 * (but note, as explained above, that there may be more.) */
3026
7918f24d
NC
3027 PERL_ARGS_ASSERT_TO_UNI_UPPER;
3028
3a4c58c9
KW
3029 if (c < 256) {
3030 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
3031 }
3032
0ebc6274 3033 uvchr_to_utf8(p, c);
b9992569 3034 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
3035}
3036
84afefe6
JH
3037UV
3038Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3039{
7918f24d
NC
3040 PERL_ARGS_ASSERT_TO_UNI_TITLE;
3041
3a4c58c9
KW
3042 if (c < 256) {
3043 return _to_upper_title_latin1((U8) c, p, lenp, 's');
3044 }
3045
0ebc6274 3046 uvchr_to_utf8(p, c);
b9992569 3047 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
3048}
3049
afc16117 3050STATIC U8
eaf412bf 3051S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
3052{
3053 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 3054 * those, converting the result to UTF-8. Since the result is always just
a1433954 3055 * one character, we allow <p> to be NULL */
afc16117
KW
3056
3057 U8 converted = toLOWER_LATIN1(c);
3058
eaf412bf
KW
3059 PERL_UNUSED_ARG(dummy);
3060
afc16117 3061 if (p != NULL) {
6f2d5cbc 3062 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3063 *p = converted;
3064 *lenp = 1;
3065 }
3066 else {
430c9760
KW
3067 /* Result is known to always be < 256, so can use the EIGHT_BIT
3068 * macros */
3069 *p = UTF8_EIGHT_BIT_HI(converted);
3070 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3071 *lenp = 2;
3072 }
3073 }
3074 return converted;
3075}
3076
84afefe6
JH
3077UV
3078Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3079{
7918f24d
NC
3080 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3081
afc16117 3082 if (c < 256) {
eaf412bf 3083 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3084 }
3085
afc16117 3086 uvchr_to_utf8(p, c);
b9992569 3087 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
3088}
3089
84afefe6 3090UV
7c0ab950 3091Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3092{
51910141 3093 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3094 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3095 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3096 *
3097 * Not to be used for locale folds
51910141 3098 */
f673fad4 3099
a1dde8de
KW
3100 UV converted;
3101
3102 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3103
1ca267a5
KW
3104 assert (! (flags & FOLD_FLAGS_LOCALE));
3105
659a7c2d 3106 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3107 converted = GREEK_SMALL_LETTER_MU;
3108 }
9b63e895
KW
3109#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3110 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3111 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3112 else if ( (flags & FOLD_FLAGS_FULL)
3113 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3114 {
1ca267a5
KW
3115 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3116 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3117 * under those circumstances. */
3118 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3119 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3120 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3121 p, *lenp, U8);
3122 return LATIN_SMALL_LETTER_LONG_S;
3123 }
3124 else {
4f489194
KW
3125 *(p)++ = 's';
3126 *p = 's';
3127 *lenp = 2;
3128 return 's';
1ca267a5 3129 }
a1dde8de 3130 }
9b63e895 3131#endif
a1dde8de
KW
3132 else { /* In this range the fold of all other characters is their lower
3133 case */
3134 converted = toLOWER_LATIN1(c);
3135 }
3136
6f2d5cbc 3137 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3138 *p = (U8) converted;
3139 *lenp = 1;
3140 }
3141 else {
3142 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3143 *p = UTF8_TWO_BYTE_LO(converted);
3144 *lenp = 2;
3145 }
3146
3147 return converted;
3148}
3149
3150UV
31f05a37 3151Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3152{
4b593389 3153
a0270393
KW
3154 /* Not currently externally documented, and subject to change
3155 * <flags> bits meanings:
3156 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3157 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3158 * locale are to be used.
a0270393
KW
3159 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3160 */
4b593389 3161
36bb2ab6 3162 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3163
780fcc9f 3164 if (flags & FOLD_FLAGS_LOCALE) {
8b7358b9
KW
3165 /* Treat a UTF-8 locale as not being in locale at all, except for
3166 * potentially warning */
3167 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
780fcc9f
KW
3168 if (IN_UTF8_CTYPE_LOCALE) {
3169 flags &= ~FOLD_FLAGS_LOCALE;
3170 }
3171 else {
e7b7ac46 3172 goto needs_full_generality;
780fcc9f 3173 }
31f05a37
KW
3174 }
3175
a1dde8de 3176 if (c < 256) {
e7b7ac46 3177 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3178 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3179 }
3180
2f306ab9 3181 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
3182 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3183 uvchr_to_utf8(p, c);
b9992569 3184 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3185 }
567b353c 3186 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3187 the special flags. */
3188 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3189
3190 needs_full_generality:
a0270393 3191 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3192 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3193 p, lenp, flags);
a0270393 3194 }
84afefe6
JH
3195}
3196
26483009 3197PERL_STATIC_INLINE bool
5141f98e 3198S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 3199 const char *const swashname, SV* const invlist)
bde6a22d 3200{
ea317ccb
KW
3201 /* returns a boolean giving whether or not the UTF8-encoded character that
3202 * starts at <p> is in the swash indicated by <swashname>. <swash>
3203 * contains a pointer to where the swash indicated by <swashname>
3204 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
3205 * look at <*swash> and only generate a swash if it is not null. <invlist>
3206 * is NULL or an inversion list that defines the swash. If not null, it
3207 * saves time during initialization of the swash.
ea317ccb
KW
3208 *
3209 * Note that it is assumed that the buffer length of <p> is enough to
3210 * contain all the bytes that comprise the character. Thus, <*p> should
3211 * have been checked before this call for mal-formedness enough to assure
3212 * that. */
3213
7918f24d
NC
3214 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3215
492a624f 3216 /* The API should have included a length for the UTF-8 character in <p>,
28123549 3217 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
3218 * as far as there being enough bytes available in it to accommodate the
3219 * character without reading beyond the end, and pass that number on to the
3220 * validating routine */
6302f837 3221 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 3222 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 3223 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
3224 1 /* Die */ );
3225 NOT_REACHED; /* NOTREACHED */
28123549 3226 }
86ae6e94 3227
dc31b55c
KW
3228 if (invlist) {
3229 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3230 }
3231
af0d7385
KW
3232 assert(swash);
3233
87367d5f
KW
3234 if (!*swash) {
3235 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
3236 *swash = _core_swash_init("utf8",
3237
3238 /* Only use the name if there is no inversion
3239 * list; otherwise will go out to disk */
3240 (invlist) ? "" : swashname,
3241
3242 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 3243 }
28123549 3244
bde6a22d
NC
3245 return swash_fetch(*swash, p, TRUE) != 0;
3246}
3247
da8c1a98 3248PERL_STATIC_INLINE bool
56576a04
KW
3249S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3250 SV **swash, const char *const swashname,
3251 SV* const invlist)
da8c1a98
KW
3252{
3253 /* returns a boolean giving whether or not the UTF8-encoded character that
3254 * starts at <p>, and extending no further than <e - 1> is in the swash
3255 * indicated by <swashname>. <swash> contains a pointer to where the swash
3256 * indicated by <swashname> is to be stored; which this routine will do, so
3257 * that future calls will look at <*swash> and only generate a swash if it
3258 * is not null. <invlist> is NULL or an inversion list that defines the
3259 * swash. If not null, it saves time during initialization of the swash.
3260 */
3261
3262 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3263
3264 if (! isUTF8_CHAR(p, e)) {
3265 _force_out_malformed_utf8_message(p, e, 0, 1);
3266 NOT_REACHED; /* NOTREACHED */
3267 }
3268
dc31b55c
KW
3269 if (invlist) {
3270 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3271 }
3272
af0d7385
KW
3273 assert(swash);
3274
da8c1a98
KW
3275 if (!*swash) {
3276 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3277 *swash = _core_swash_init("utf8",
3278
3279 /* Only use the name if there is no inversion
3280 * list; otherwise will go out to disk */
3281 (invlist) ? "" : swashname,
3282
3283 &PL_sv_undef, 1, 0, invlist, &flags);
3284 }
3285
3286 return swash_fetch(*swash, p, TRUE) != 0;
3287}
3288
34aeb2e9
KW
3289STATIC void
3290S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3291 const char * const alternative,
3292 const bool use_locale,
3293 const char * const file,
3294 const unsigned line)
3295{
3296 const char * key;
3297
3298 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3299
3300 if (ckWARN_d(WARN_DEPRECATED)) {
3301
3302 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3303 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3304 if (! PL_seen_deprecated_macro) {
3305 PL_seen_deprecated_macro = newHV();
3306 }
3307 if (! hv_store(PL_seen_deprecated_macro, key,
3308 strlen(key), &PL_sv_undef, 0))
3309 {
3310 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3311 }
3312
c44e9413 3313 if (instr(file, "mathoms.c")) {
607313a1
KW
3314 Perl_warner(aTHX_ WARN_DEPRECATED,
3315 "In %s, line %d, starting in Perl v5.30, %s()"
3316 " will be removed. Avoid this message by"
3317 " converting to use %s().\n",
3318 file, line, name, alternative);
3319 }
3320 else {
34aeb2e9
KW
3321 Perl_warner(aTHX_ WARN_DEPRECATED,
3322 "In %s, line %d, starting in Perl v5.30, %s() will"
3323 " require an additional parameter. Avoid this"
3324 " message by converting to use %s().\n",
3325 file, line, name, alternative);
607313a1 3326 }
34aeb2e9
KW
3327 }
3328 }
3329}
3330
bde6a22d 3331bool
34aeb2e9 3332Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3333 const U8 * const p,
34aeb2e9
KW
3334 const char * const name,
3335 const char * const alternative,
3336 const bool use_utf8,
3337 const bool use_locale,
3338 const char * const file,
3339 const unsigned line)
922e8cb4 3340{
922e8cb4
KW
3341 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3342
34aeb2e9
KW
3343 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3344
3345 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3346
3347 switch (classnum) {
3348 case _CC_WORDCHAR:
3349 case _CC_DIGIT:
3350 case _CC_ALPHA:
3351 case _CC_LOWER:
3352 case _CC_UPPER:
3353 case _CC_PUNCT:
3354 case _CC_PRINT:
3355 case _CC_ALPHANUMERIC:
3356 case _CC_GRAPH:
3357 case _CC_CASED:
3358
3359 return is_utf8_common(p,
af0d7385
KW
3360 NULL,
3361 "This is buggy if this gets used",
34aeb2e9
KW
3362 PL_XPosix_ptrs[classnum]);
3363
3364 case _CC_SPACE:
3365 return is_XPERLSPACE_high(p);
3366 case _CC_BLANK:
3367 return is_HORIZWS_high(p);
3368 case _CC_XDIGIT:
3369 return is_XDIGIT_high(p);
3370 case _CC_CNTRL:
3371 return 0;
3372 case _CC_ASCII:
3373 return 0;
3374 case _CC_VERTSPACE:
3375 return is_VERTWS_high(p);
3376 case _CC_IDFIRST:
af0d7385
KW
3377 return is_utf8_common(p, NULL,
3378 "This is buggy if this gets used",
3379 PL_utf8_perl_idstart);
34aeb2e9 3380 case _CC_IDCONT:
af0d7385
KW
3381 return is_utf8_common(p, NULL,
3382 "This is buggy if this gets used",
3383 PL_utf8_perl_idcont);
34aeb2e9
KW
3384 }
3385 }
3386
3387 /* idcont is the same as wordchar below 256 */
3388 if (classnum == _CC_IDCONT) {
3389 classnum = _CC_WORDCHAR;
3390 }
3391 else if (classnum == _CC_IDFIRST) {
3392 if (*p == '_') {
3393 return TRUE;
3394 }
3395 classnum = _CC_ALPHA;
3396 }
3397
3398 if (! use_locale) {
3399 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3400 return _generic_isCC(*p, classnum);
3401 }
922e8cb4 3402
34aeb2e9
KW
3403 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3404 }
3405 else {
3406 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3407 return isFOO_lc(classnum, *p);
3408 }
3409
3410 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3411 }
3412
3413 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3414}
3415
3416bool
da8c1a98
KW
3417Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3418 const U8 * const e)
3419{
3420 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3421
af0d7385
KW
3422 return is_utf8_common_with_len(p, e, NULL,
3423 "This is buggy if this gets used",
da8c1a98
KW
3424 PL_XPosix_ptrs[classnum]);
3425}
3426
3427bool
da8c1a98
KW
3428Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3429{
da8c1a98
KW
3430 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3431
af0d7385
KW
3432 return is_utf8_common_with_len(p, e, NULL,
3433 "This is buggy if this gets used",
3434 PL_utf8_perl_idstart);
da8c1a98
KW
3435}
3436
3437bool
f2645549 3438Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3439{
f2645549 3440 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3441
3442 if (*p == '_')
3443 return TRUE;
f25ce844 3444 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3445}
3446
3447bool
da8c1a98
KW
3448Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3449{
da8c1a98
KW
3450 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3451
af0d7385
KW
3452 return is_utf8_common_with_len(p, e, NULL,
3453 "This is buggy if this gets used",
3454 PL_utf8_perl_idcont);
da8c1a98
KW
3455}
3456
3457bool
f2645549 3458Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3459{
f2645549 3460 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3461
f25ce844 3462 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3463}
3464
3465bool
f2645549 3466Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3467{
f2645549 3468 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3469
5062866a 3470 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
c11ff943
KW
3471}
3472
3473bool
7dbf68d2
KW
3474Perl__is_utf8_mark(pTHX_ const U8 *p)
3475{
7dbf68d2
KW
3476 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3477
f25ce844 3478 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3479}
3480
6a4a25f4 3481STATIC UV
30613bdc
KW
3482S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3483 U8* ustrp, STRLEN *lenp,
341bb5b7 3484 SV *invlist, const int * const invmap,
e39a4130 3485 const unsigned int * const * const aux_tables,
30613bdc
KW
3486 const U8 * const aux_table_lengths,
3487 const char * const normal)
b9992569 3488{
0134edef 3489 STRLEN len = 0;
7918f24d 3490
30613bdc
KW
3491 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3492 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3493 * to name the new case in any generated messages, as a fallback if the
3494 * operation being used is not available. The new case is given by the
3495 * data structures in the remaining arguments.
3496 *
3497 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3498 * entire changed case string, and the return value is the first code point
3499 * in that string */
3500
b9992569 3501 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3502
36eaa811
KW
3503 /* For code points that don't change case, we already know that the output
3504 * of this function is the unchanged input, so we can skip doing look-ups
3505 * for them. Unfortunately the case-changing code points are scattered
3506 * around. But there are some long consecutive ranges where there are no
3507 * case changing code points. By adding tests, we can eliminate the lookup
3508 * for all the ones in such ranges. This is currently done here only for
3509 * just a few cases where the scripts are in common use in modern commerce
3510 * (and scripts adjacent to those which can be included without additional
3511 * tests). */
3512
3513 if (uv1 >= 0x0590) {
3514 /* This keeps from needing further processing the code points most
3515 * likely to be used in the following non-cased scripts: Hebrew,
3516 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3517 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3518 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3519 if (uv1 < 0x10A0) {
3520 goto cases_to_self;
3521 }
3522
3523 /* The following largish code point ranges also don't have case
3524 * changes, but khw didn't think they warranted extra tests to speed
3525 * them up (which would slightly slow down everything else above them):
3526 * 1100..139F Hangul Jamo, Ethiopic
3527 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3528 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3529 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3530 * Combining Diacritical Marks Extended, Balinese,
3531 * Sundanese, Batak, Lepcha, Ol Chiki
3532 * 2000..206F General Punctuation
3533 */
3534
3535 if (uv1 >= 0x2D30) {
3536
3537 /* This keeps the from needing further processing the code points
3538 * most likely to be used in the following non-cased major scripts:
3539 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3540 *
3541 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3542 * event that Unicode eventually allocates the unused block as of
3543 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3544 * that the test suite will start having failures to alert you
3545 * should that happen) */
3546 if (uv1 < 0xA640) {
3547 goto cases_to_self;
3548 }
3549
3550 if (uv1 >= 0xAC00) {
3551 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3552 if (ckWARN_d(WARN_SURROGATE)) {
3553 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3554 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3555 "Operation \"%s\" returns its argument for"
3556 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3557 }
3558 goto cases_to_self;
3559 }
36eaa811
KW
3560
3561 /* AC00..FAFF Catches Hangul syllables and private use, plus
3562 * some others */
3563 if (uv1 < 0xFB00) {
3564 goto cases_to_self;
36eaa811
KW
3565 }
3566
5af9bc97 3567 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3568 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3569 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3570 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3571 }
3572 if (ckWARN_d(WARN_NON_UNICODE)) {
3573 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3574 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3575 "Operation \"%s\" returns its argument for"
3576 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3577 }
3578 goto cases_to_self;
3579 }
3bfc1e70
KW
3580#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3581 if (UNLIKELY(uv1
3582 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3583 {
3584
56576a04
KW
3585 /* As of Unicode 10.0, this means we avoid swash creation
3586 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3587 goto cases_to_self;
3588 }
3589#endif
36eaa811
KW
3590 }
3591 }
9ae3ac1a 3592
36eaa811 3593 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3594 * be given. */
9ae3ac1a
KW
3595 }
3596
8946fcd9
KW
3597 {
3598 unsigned int i;
e39a4130 3599 const unsigned int * cp_list;
8946fcd9
KW
3600 U8 * d;
3601 SSize_t index = _invlist_search(invlist, uv1);
3602 IV base = invmap[index];
0134edef 3603
30613bdc
KW
3604 /* The data structures are set up so that if 'base' is non-negative,
3605 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3606 if (base >= 0) {
3607 IV lc;
b08cf34e 3608
8946fcd9
KW
3609 if (base == 0) {
3610 goto cases_to_self;
4a8240a3 3611 }
4a8240a3 3612
30613bdc 3613 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3614 lc = base + uv1 - invlist_array(invlist)[index];
3615 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3616 return lc;
3617 }
1feea2c7 3618
30613bdc
KW
3619 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3620 * requires an auxiliary table look up. abs(base) gives the index into
3621 * a list of such tables which points to the proper aux table. And a
3622 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3623 cp_list = aux_tables[-base];
30613bdc
KW
3624
3625 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3626 d = ustrp;
3627 for (i = 0; i < aux_table_lengths[-base]; i++) {
3628 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3629 }
8946fcd9
KW
3630 *d = '\0';
3631 *lenp = d - ustrp;
3632
3633 return cp_list[0];
cbe07460
KW
3634 }
3635
3636 /* Here, there was no mapping defined, which means that the code point maps
3637 * to itself. Return the inputs */
e24dfe9c 3638 cases_to_self:
bfdf22ec 3639 len = UTF8SKIP(p);
ca9fab46
KW
3640 if (p != ustrp) { /* Don't copy onto itself */
3641 Copy(p, ustrp, len, U8);
3642 }
0134edef 3643
2a37f04d
JH
3644 if (lenp)
3645 *lenp = len;
3646
f4cd282c 3647 return uv1;
cbe07460 3648
a0ed51b3
LW
3649}
3650
b74fe592 3651Size_t
e39a4130
KW
3652Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3653 const unsigned int ** remaining_folds_to)
b74fe592
KW
3654{
3655 /* Returns the count of the number of code points that fold to the input
3656 * 'cp' (besides itself).
3657 *
3658 * If the return is 0, there is nothing else that folds to it, and
3659 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3660 *
3661 * If the return is 1, '*first_folds_to' is set to the single code point,
3662 * and '*remaining_folds_to' is set to NULL.
3663 *
3664 * Otherwise, '*first_folds_to' is set to a code point, and
3665 * '*remaining_fold_to' is set to an array that contains the others. The
3666 * length of this array is the returned count minus 1.
3667 *
3668 * The reason for this convolution is to avoid having to deal with
3669 * allocating and freeing memory. The lists are already constructed, so
3670 * the return can point to them, but single code points aren't, so would
3671 * need to be constructed if we didn't employ something like this API */
3672
3673 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3674 int base = _Perl_IVCF_invmap[index];
3675
3676 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3677
3678 if (base == 0) { /* No fold */
3679 *first_folds_to = 0;
3680 *remaining_folds_to = NULL;
3681 return 0;
3682 }
3683
3684#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3685
3686 assert(base > 0);
3687
3688#else
3689
3690 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3691
3692 /* The data structure is set up so that the absolute value of 'base' is
3693 * an index into a table of pointers to arrays, with the array
3694 * corresponding to the index being the list of code points that fold
3695 * to 'cp', and the parallel array containing the length of the list
3696 * array */
3697 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3698 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3699 *first_folds_to
3700 */
3701 return IVCF_AUX_TABLE_lengths[-base];
3702 }
3703
3704#endif
3705
3706 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3707 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3708 *remaining_folds_to = NULL;
3709 return 1;
3710}
3711
051a06d4 3712STATIC UV
56576a04
KW
3713S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3714 U8* const ustrp, STRLEN *lenp)
051a06d4 3715{
4a4088c4 3716 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3717 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3718 * result contains a character that crosses the 255/256 boundary, disallow
3719 * the change, and return the original code point. See L<perlfunc/lc> for
3720 * why;
051a06d4 3721 *
a1433954
KW
3722 * p points to the original string whose case was changed; assumed
3723 * by this routine to be well-formed
051a06d4 3724 * result the code point of the first character in the changed-case string
56576a04
KW
3725 * ustrp points to the changed-case string (<result> represents its
3726 * first char)
051a06d4
KW
3727 * lenp points to the length of <ustrp> */
3728
3729 UV original; /* To store the first code point of <p> */
3730
3731 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3732
a4f12ed7 3733 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3734
3735 /* We know immediately if the first character in the string crosses the
5e45c680 3736 * boundary, so can skip testing */