This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Move some static fcns from regcomp.c to invlist_inline.h
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
fb7e7255
KW
39
40/* Be sure to synchronize this message with the similar one in regcomp.c */
760c7c2f 41static const char cp_above_legal_max[] =
d22ec717
KW
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
760c7c2f 44
48ef279e 45/*
ccfc67b7 46=head1 Unicode Support
7fefc6c1 47These are various utility functions for manipulating UTF8-encoded
72d33970 48strings. For the uninitiated, this is a method of representing arbitrary
61296642 49Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
50characters in the ASCII range are unmodified, and a zero byte never appears
51within non-zero characters.
166f8a29 52
eaf7a4d2
CS
53=cut
54*/
55
dd051059
DM
56/* helper for Perl__force_out_malformed_utf8_message(). Like
57 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
58 * PL_compiling */
59
60static void
61S_restore_cop_warnings(pTHX_ void *p)
62{
63 if (!specialWARN(PL_curcop->cop_warnings))
64 PerlMemShared_free(PL_curcop->cop_warnings);
65 PL_curcop->cop_warnings = (STRLEN*)p;
66}
67
68
9cbfb8ab
KW
69void
70Perl__force_out_malformed_utf8_message(pTHX_
71 const U8 *const p, /* First byte in UTF-8 sequence */
72 const U8 * const e, /* Final byte in sequence (may include
73 multiple chars */
74 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
75 usually 0, or some DISALLOW flags */
76 const bool die_here) /* If TRUE, this function does not return */
77{
78 /* This core-only function is to be called when a malformed UTF-8 character
79 * is found, in order to output the detailed information about the
80 * malformation before dieing. The reason it exists is for the occasions
81 * when such a malformation is fatal, but warnings might be turned off, so
82 * that normally they would not be actually output. This ensures that they
83 * do get output. Because a sequence may be malformed in more than one
84 * way, multiple messages may be generated, so we can't make them fatal, as
85 * that would cause the first one to die.
86 *
87 * Instead we pretend -W was passed to perl, then die afterwards. The
88 * flexibility is here to return to the caller so they can finish up and
89 * die themselves */
90 U32 errors;
91
92 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
93
94 ENTER;
c15a80f3 95 SAVEI8(PL_dowarn);
9cbfb8ab
KW
96 SAVESPTR(PL_curcop);
97
98 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
99 if (PL_curcop) {
dd051059
DM
100 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
101 * than PL_compiling */
102 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
103 (void*)PL_curcop->cop_warnings);
9cbfb8ab
KW
104 PL_curcop->cop_warnings = pWARN_ALL;
105 }
106
107 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
108
109 LEAVE;
110
111 if (! errors) {
112 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
113 " be called only when there are errors found");
114 }
115
116 if (die_here) {
117 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
118 }
119}
120
bb07812e
KW
121STATIC HV *
122S_new_msg_hv(pTHX_ const char * const message, /* The message text */
123 U32 categories, /* Packed warning categories */
124 U32 flag) /* Flag associated with this message */
125{
126 /* Creates, populates, and returns an HV* that describes an error message
127 * for the translators between UTF8 and code point */
128
129 SV* msg_sv = newSVpv(message, 0);
130 SV* category_sv = newSVuv(categories);
131 SV* flag_bit_sv = newSVuv(flag);
132
133 HV* msg_hv = newHV();
134
135 PERL_ARGS_ASSERT_NEW_MSG_HV;
136
2b672cf5
KW
137 (void) hv_stores(msg_hv, "text", msg_sv);
138 (void) hv_stores(msg_hv, "warn_categories", category_sv);
139 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
140
141 return msg_hv;
142}
143
eaf7a4d2 144/*
378516de 145=for apidoc uvoffuni_to_utf8_flags
eebe1485 146
a27992cc 147THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
148Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
149L<perlapi/uvchr_to_utf8_flags>>.
a27992cc 150
de69f3af
KW
151This function is like them, but the input is a strict Unicode
152(as opposed to native) code point. Only in very rare circumstances should code
153not be using the native code point.
949cf498 154
09232555 155For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
949cf498 156
eebe1485
SC
157=cut
158*/
159
33f38593
KW
160U8 *
161Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
162{
163 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
164
165 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
166}
167
c94c2f39
KW
168/* All these formats take a single UV code point argument */
169const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
170const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
171 " is not recommended for open interchange";
172const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
173 " may not be portable";
57ff5f59
KW
174const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
175 " Unicode, requires a Perl extension," \
176 " and so is not portable";
c94c2f39 177
33f38593 178#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
179 STMT_START { \
180 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
181 U32 category = packWARN(WARN_SURROGATE); \
182 const char * format = surrogate_cp_format; \
183 if (msgs) { \
184 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
185 category, \
186 UNICODE_GOT_SURROGATE); \
187 } \
188 else { \
189 Perl_ck_warner_d(aTHX_ category, format, uv); \
190 } \
8ee1cdcb
KW
191 } \
192 if (flags & UNICODE_DISALLOW_SURROGATE) { \
193 return NULL; \
194 } \
195 } STMT_END;
196
33f38593 197#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
198 STMT_START { \
199 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
200 U32 category = packWARN(WARN_NONCHAR); \
201 const char * format = nonchar_cp_format; \
202 if (msgs) { \
203 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
204 category, \
205 UNICODE_GOT_NONCHAR); \
206 } \
207 else { \
208 Perl_ck_warner_d(aTHX_ category, format, uv); \
209 } \
8ee1cdcb
KW
210 } \
211 if (flags & UNICODE_DISALLOW_NONCHAR) { \
212 return NULL; \
213 } \
214 } STMT_END;
215
ba6ed43c
KW
216/* Use shorter names internally in this file */
217#define SHIFT UTF_ACCUMULATION_SHIFT
218#undef MARK
219#define MARK UTF_CONTINUATION_MARK
220#define MASK UTF_CONTINUATION_MASK
221
33f38593
KW
222/*
223=for apidoc uvchr_to_utf8_flags_msgs
224
225THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
226
227Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
228
229This function is for code that wants any warning and/or error messages to be
230returned to the caller rather than be displayed. All messages that would have
884a31ee 231been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
232
233It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
234placed after all the others, C<msgs>. If this parameter is 0, this function
235behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
236be a pointer to an C<HV *> variable, in which this function creates a new HV to
237contain any appropriate messages. The hash has three key-value pairs, as
238follows:
239
240=over 4
241
242=item C<text>
243
244The text of the message as a C<SVpv>.
245
246=item C<warn_categories>
247
248The warning category (or categories) packed into a C<SVuv>.
249
250=item C<flag>
251
252A single flag bit associated with this message, in a C<SVuv>.
253The bit corresponds to some bit in the C<*errors> return value,
254such as C<UNICODE_GOT_SURROGATE>.
255
256=back
257
258It's important to note that specifying this parameter as non-null will cause
259any warnings this function would otherwise generate to be suppressed, and
260instead be placed in C<*msgs>. The caller can check the lexical warnings state
261(or not) when choosing what to do with the returned messages.
262
263The caller, of course, is responsible for freeing any returned HV.
264
265=cut
266*/
267
268/* Undocumented; we don't want people using this. Instead they should use
269 * uvchr_to_utf8_flags_msgs() */
dfe13c55 270U8 *
33f38593 271Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 272{
33f38593
KW
273 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
274
275 if (msgs) {
276 *msgs = NULL;
277 }
7918f24d 278
2d1545e5 279 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 280 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
281 return d;
282 }
facc1dc2 283
3ea68d71 284 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
285 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
286 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
287 return d;
288 }
d9432125 289
ba6ed43c
KW
290 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
291 * below, the 16 is for start bytes E0-EF (which are all the possible ones
292 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
293 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
294 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
295 * 0x800-0xFFFF on ASCII */
296 if (uv < (16 * (1U << (2 * SHIFT)))) {
297 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
298 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
299 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
300
301#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
302 aren't tested here */
303 /* The most likely code points in this range are below the surrogates.
304 * Do an extra test to quickly exclude those. */
305 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
306 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
307 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
308 {
33f38593 309 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
310 }
311 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 312 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 313 }
ba6ed43c
KW
314 }
315#endif
316 return d;
317 }
318
319 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
320 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
321 * happen starting with 4-byte characters on ASCII platforms. We unify the
322 * code for these with EBCDIC, even though some of them require 5-bytes on
323 * those, because khw believes the code saving is worth the very slight
324 * performance hit on these high EBCDIC code points. */
325
326 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
40606899
KW
327 if (UNLIKELY(uv > MAX_LEGAL_CP)) {
328 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP);
a5bf80e0 329 }
33f38593
KW
330 if ( (flags & UNICODE_WARN_SUPER)
331 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 332 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 333 {
33f38593
KW
334 const char * format = super_cp_format;
335 U32 category = packWARN(WARN_NON_UNICODE);
336 U32 flag = UNICODE_GOT_SUPER;
337
338 /* Choose the more dire applicable warning */
339 if (UNICODE_IS_PERL_EXTENDED(uv)) {
340 format = perl_extended_cp_format;
341 if (flags & (UNICODE_WARN_PERL_EXTENDED
342 |UNICODE_DISALLOW_PERL_EXTENDED))
343 {
344 flag = UNICODE_GOT_PERL_EXTENDED;
345 }
346 }
a5bf80e0 347
33f38593
KW
348 if (msgs) {
349 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
350 category, flag);
351 }
352 else {
353 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
354 }
a5bf80e0 355 }
56576a04 356 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
357 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
358 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
359 {
360 return NULL;
361 }
362 }
ba6ed43c 363 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 364 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 365 }
d9432125 366
ba6ed43c
KW
367 /* Test for and handle 4-byte result. In the test immediately below, the
368 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
369 * characters). The 3 is for 3 continuation bytes; these each contribute
370 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
371 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
372 * 0x1_0000-0x1F_FFFF on ASCII */
373 if (uv < (8 * (1U << (3 * SHIFT)))) {
374 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
375 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
376 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
377 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
378
379#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
380 characters. The end-plane non-characters for EBCDIC were
381 handled just above */
382 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 383 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 384 }
ba6ed43c 385 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 386 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
387 }
388#endif
389
390 return d;
391 }
392
393 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
394 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
395 * format. The unrolled version above turns out to not save all that much
396 * time, and at these high code points (well above the legal Unicode range
397 * on ASCII platforms, and well above anything in common use in EBCDIC),
398 * khw believes that less code outweighs slight performance gains. */
399
d9432125 400 {
5aaebcb3 401 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
402 U8 *p = d+len-1;
403 while (p > d) {
957a9e81
KW
404 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
405 uv >>= SHIFT;
1d72bdf6 406 }
4c8cd605 407 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
408 return d+len;
409 }
a0ed51b3 410}
a5bf80e0 411
646ca15d 412/*
07693fe6
KW
413=for apidoc uvchr_to_utf8
414
bcb1a2d4 415Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 416of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
417C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
418the byte after the end of the new character. In other words,
07693fe6
KW
419
420 d = uvchr_to_utf8(d, uv);
421
422is the recommended wide native character-aware way of saying
423
424 *(d++) = uv;
425
d22ec717
KW
426This function accepts any code point from 0..C<IV_MAX> as input.
427C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
428
429It is possible to forbid or warn on non-Unicode code points, or those that may
430be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 431
07693fe6
KW
432=cut
433*/
434
de69f3af
KW
435/* This is also a macro */
436PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
437
07693fe6
KW
438U8 *
439Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
440{
de69f3af 441 return uvchr_to_utf8(d, uv);
07693fe6
KW
442}
443
de69f3af
KW
444/*
445=for apidoc uvchr_to_utf8_flags
446
447Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 448of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
449C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
450the byte after the end of the new character. In other words,
de69f3af
KW
451
452 d = uvchr_to_utf8_flags(d, uv, flags);
453
454or, in most cases,
455
456 d = uvchr_to_utf8_flags(d, uv, 0);
457
458This is the Unicode-aware way of saying
459
460 *(d++) = uv;
461
d22ec717
KW
462If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
463input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
464
465Specifying C<flags> can further restrict what is allowed and not warned on, as
466follows:
de69f3af 467
796b6530 468If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
469the function will raise a warning, provided UTF8 warnings are enabled. If
470instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
471NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 472
760c7c2f
KW
473Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
474affect how the function handles a Unicode non-character.
93e6dbd6 475
760c7c2f
KW
476And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
477affect the handling of code points that are above the Unicode maximum of
4780x10FFFF. Languages other than Perl may not be able to accept files that
479contain these.
93e6dbd6
KW
480
481The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
482the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
483three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
484allowed inputs to the strict UTF-8 traditionally defined by Unicode.
485Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
486C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
487above-Unicode and surrogate flags, but not the non-character ones, as
488defined in
e2176993 489L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f 490See L<perlunicode/Noncharacter code points>.
93e6dbd6 491
57ff5f59
KW
492Extremely high code points were never specified in any standard, and require an
493extension to UTF-8 to express, which Perl does. It is likely that programs
494written in something other than Perl would not be able to read files that
495contain these; nor would Perl understand files written by something that uses a
496different extension. For these reasons, there is a separate set of flags that
497can warn and/or disallow these extremely high code points, even if other
498above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
499and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
500L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
501treat all above-Unicode code points, including these, as malformations. (Note
502that the Unicode standard considers anything above 0x10FFFF to be illegal, but
503there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
504
505A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
506retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
507C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
508C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
509platforms,these flags can apply to code points that actually do fit in 31 bits.
510The new names accurately describe the situation in all cases.
de69f3af 511
de69f3af
KW
512=cut
513*/
514
515/* This is also a macro */
516PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
517
07693fe6
KW
518U8 *
519Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
520{
de69f3af 521 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
522}
523
57ff5f59
KW
524#ifndef UV_IS_QUAD
525
e050c007
KW
526STATIC int
527S_is_utf8_cp_above_31_bits(const U8 * const s,
528 const U8 * const e,
529 const bool consider_overlongs)
83dc0f42
KW
530{
531 /* Returns TRUE if the first code point represented by the Perl-extended-
532 * UTF-8-encoded string starting at 's', and looking no further than 'e -
533 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
534 *
535 * The function handles the case where the input bytes do not include all
536 * the ones necessary to represent a full character. That is, they may be
537 * the intial bytes of the representation of a code point, but possibly
538 * the final ones necessary for the complete representation may be beyond
539 * 'e - 1'.
540 *
e050c007
KW
541 * The function also can handle the case where the input is an overlong
542 * sequence. If 'consider_overlongs' is 0, the function assumes the
543 * input is not overlong, without checking, and will return based on that
544 * assumption. If this parameter is 1, the function will go to the trouble
545 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 546 *
e050c007 547 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
548 */
549
e050c007
KW
550 const STRLEN len = e - s;
551 int is_overlong;
552
553 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
554
555 assert(! UTF8_IS_INVARIANT(*s) && e > s);
556
83dc0f42
KW
557#ifdef EBCDIC
558
e050c007 559 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 560
e050c007
KW
561 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
562 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
563 * also be the start byte for a 31-bit code point; we need at least 2
564 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
565 * the start byte for an overlong sequence, but for 30-bit or smaller code
566 * points, so we don't have to worry about overlongs on EBCDIC.) */
567 if (*s != 0xFE) {
568 return 0;
569 }
83dc0f42 570
e050c007
KW
571 if (len == 1) {
572 return -1;
573 }
83dc0f42 574
e050c007 575#else
83dc0f42 576
e050c007
KW
577 /* On ASCII, FE and FF are the only start bytes that can evaluate to
578 * needing more than 31 bits. */
579 if (LIKELY(*s < 0xFE)) {
580 return 0;
581 }
83dc0f42 582
e050c007
KW
583 /* What we have left are FE and FF. Both of these require more than 31
584 * bits unless they are for overlongs. */
585 if (! consider_overlongs) {
586 return 1;
587 }
83dc0f42 588
e050c007
KW
589 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
590 * above 31 bits. But we need more than one byte to discern this, so if
591 * passed just the start byte, it could be an overlong evaluating to
592 * smaller */
593 if (len == 1) {
594 return -1;
595 }
83dc0f42 596
e050c007
KW
597 /* Having excluded len==1, and knowing that FE and FF are both valid start
598 * bytes, we can call the function below to see if the sequence is
599 * overlong. (We don't need the full generality of the called function,
600 * but for these huge code points, speed shouldn't be a consideration, and
601 * the compiler does have enough information, since it's static to this
602 * file, to optimize to just the needed parts.) */
603 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 604
e050c007
KW
605 /* If it isn't overlong, more than 31 bits are required. */
606 if (is_overlong == 0) {
607 return 1;
608 }
83dc0f42 609
e050c007
KW
610 /* If it is indeterminate if it is overlong, return that */
611 if (is_overlong < 0) {
612 return -1;
613 }
614
615 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
616 * the max it can be is 2**31 - 1 */
617 if (*s == 0xFE) {
618 return 0;
83dc0f42
KW
619 }
620
e050c007
KW
621#endif
622
623 /* Here, ASCII and EBCDIC rejoin:
624 * On ASCII: We have an overlong sequence starting with FF
625 * On EBCDIC: We have a sequence starting with FE. */
626
627 { /* For C89, use a block so the declaration can be close to its use */
628
629#ifdef EBCDIC
630
5f995336
KW
631 /* U+7FFFFFFF (2 ** 31 - 1)
632 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
633 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
634 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
635 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
636 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
637 * U+80000000 (2 ** 31):
638 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
639 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
640 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
641 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
642 *
643 * and since we know that *s = \xfe, any continuation sequcence
644 * following it that is gt the below is above 31 bits
645 [0] [1] [2] [3] [4] [5] [6] */
646 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
647
648#else
649
650 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
651 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
652 * FF overlong for U+80000000 (2 ** 31):
653 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
654 * and since we know that *s = \xff, any continuation sequcence
655 * following it that is gt the below is above 30 bits
656 [0] [1] [2] [3] [4] [5] [6] */
657 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 658
83dc0f42
KW
659
660#endif
e050c007
KW
661 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
662 const STRLEN cmp_len = MIN(conts_len, len - 1);
663
664 /* Now compare the continuation bytes in s with the ones we have
665 * compiled in that are for the largest 30 bit code point. If we have
666 * enough bytes available to determine the answer, or the bytes we do
667 * have differ from them, we can compare the two to get a definitive
668 * answer (Note that in UTF-EBCDIC, the two lowest possible
669 * continuation bytes are \x41 and \x42.) */
670 if (cmp_len >= conts_len || memNE(s + 1,
671 conts_for_highest_30_bit,
672 cmp_len))
673 {
674 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
675 }
83dc0f42 676
e050c007
KW
677 /* Here, all the bytes we have are the same as the highest 30-bit code
678 * point, but we are missing so many bytes that we can't make the
679 * determination */
680 return -1;
681 }
83dc0f42
KW
682}
683
57ff5f59
KW
684#endif
685
d6be65ae 686PERL_STATIC_INLINE int
12a4bed3
KW
687S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
688{
d6be65ae
KW
689 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
690 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
691 * it isn't, and -1 if there isn't enough information to tell. This last
692 * return value can happen if the sequence is incomplete, missing some
693 * trailing bytes that would form a complete character. If there are
694 * enough bytes to make a definitive decision, this function does so.
695 * Usually 2 bytes sufficient.
696 *
697 * Overlongs can occur whenever the number of continuation bytes changes.
698 * That means whenever the number of leading 1 bits in a start byte
699 * increases from the next lower start byte. That happens for start bytes
700 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
701 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
702 * ASCII platforms: C0, C1
703 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 704 */
12a4bed3
KW
705
706 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
707 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
708
709 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
710 assert(len > 1 && UTF8_IS_START(*s));
711
712 /* Each platform has overlongs after the start bytes given above (expressed
713 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
714 * the logic is the same, except the E0 overlong has already been excluded
715 * on EBCDIC platforms. The values below were found by manually
716 * inspecting the UTF-8 patterns. See the tables in utf8.h and
717 * utfebcdic.h. */
718
719# ifdef EBCDIC
720# define F0_ABOVE_OVERLONG 0xB0
721# define F8_ABOVE_OVERLONG 0xA8
722# define FC_ABOVE_OVERLONG 0xA4
723# define FE_ABOVE_OVERLONG 0xA2
724# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
725 /* I8(0xfe) is FF */
726# else
727
728 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 729 return 1;
12a4bed3
KW
730 }
731
732# define F0_ABOVE_OVERLONG 0x90
733# define F8_ABOVE_OVERLONG 0x88
734# define FC_ABOVE_OVERLONG 0x84
735# define FE_ABOVE_OVERLONG 0x82
736# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
737# endif
738
739
740 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
741 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
742 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
743 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
744 {
d6be65ae 745 return 1;
12a4bed3
KW
746 }
747
b0b342d4 748 /* Check for the FF overlong */
d6be65ae 749 return isFF_OVERLONG(s, len);
b0b342d4
KW
750}
751
8d6204cc 752PERL_STATIC_INLINE int
b0b342d4
KW
753S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
754{
8d6204cc
KW
755 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
756 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
757 * it isn't, and -1 if there isn't enough information to tell. This last
758 * return value can happen if the sequence is incomplete, missing some
759 * trailing bytes that would form a complete character. If there are
760 * enough bytes to make a definitive decision, this function does so. */
761
b0b342d4 762 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 763
8d6204cc
KW
764 /* To be an FF overlong, all the available bytes must match */
765 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
766 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
767 {
768 return 0;
769 }
770
771 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
772 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 773 * utfebcdic.h. */
8d6204cc
KW
774 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
775 return 1;
776 }
12a4bed3 777
8d6204cc
KW
778 /* The missing bytes could cause the result to go one way or the other, so
779 * the result is indeterminate */
780 return -1;
12a4bed3
KW
781}
782
d22ec717 783#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
784# ifdef EBCDIC /* Actually is I8 */
785# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 786 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
787# else
788# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 789 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
790# endif
791#endif
792
c285bbc4 793PERL_STATIC_INLINE int
e050c007
KW
794S_does_utf8_overflow(const U8 * const s,
795 const U8 * e,
796 const bool consider_overlongs)
a77c906e 797{
c285bbc4 798 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
799 * 'e' - 1 would overflow an IV on this platform; that is if it represents
800 * a code point larger than the highest representable code point. It
801 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
802 * enough information to tell. This last return value can happen if the
803 * sequence is incomplete, missing some trailing bytes that would form a
804 * complete character. If there are enough bytes to make a definitive
805 * decision, this function does so.
c285bbc4 806 *
e050c007
KW
807 * If 'consider_overlongs' is TRUE, the function checks for the possibility
808 * that the sequence is an overlong that doesn't overflow. Otherwise, it
809 * assumes the sequence is not an overlong. This can give different
810 * results only on ASCII 32-bit platforms.
811 *
c285bbc4
KW
812 * (For ASCII platforms, we could use memcmp() because we don't have to
813 * convert each byte to I8, but it's very rare input indeed that would
814 * approach overflow, so the loop below will likely only get executed once.)
815 *
816 * 'e' - 1 must not be beyond a full character. */
a77c906e 817
a77c906e
KW
818
819 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
820 assert(s <= e && s + UTF8SKIP(s) >= e);
821
d22ec717
KW
822#if ! defined(UV_IS_QUAD)
823
824 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
825
826#else
827
828 PERL_UNUSED_ARG(consider_overlongs);
829
830 {
831 const STRLEN len = e - s;
832 const U8 *x;
833 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
834
835 for (x = s; x < e; x++, y++) {
836
837 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
838 continue;
839 }
840
841 /* If this byte is larger than the corresponding highest UTF-8
842 * byte, the sequence overflow; otherwise the byte is less than,
843 * and so the sequence doesn't overflow */
844 return NATIVE_UTF8_TO_I8(*x) > *y;
845
846 }
847
848 /* Got to the end and all bytes are the same. If the input is a whole
849 * character, it doesn't overflow. And if it is a partial character,
850 * there's not enough information to tell */
851 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
852 return -1;
853 }
854
855 return 0;
856 }
857
858#endif
859
860}
861
862#if 0
863
864/* This is the portions of the above function that deal with UV_MAX instead of
865 * IV_MAX. They are left here in case we want to combine them so that internal
866 * uses can have larger code points. The only logic difference is that the
867 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
868 * different logic.
869 */
870
871/* Anything larger than this will overflow the word if it were converted into a UV */
872#if defined(UV_IS_QUAD)
873# ifdef EBCDIC /* Actually is I8 */
874# define HIGHEST_REPRESENTABLE_UTF8 \
875 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
876# else
877# define HIGHEST_REPRESENTABLE_UTF8 \
878 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
879# endif
880#else /* 32-bit */
881# ifdef EBCDIC
882# define HIGHEST_REPRESENTABLE_UTF8 \
883 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
884# else
885# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
886# endif
887#endif
888
a77c906e
KW
889#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
890
891 /* On 32 bit ASCII machines, many overlongs that start with FF don't
892 * overflow */
e050c007 893 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
894
895 /* To be such an overlong, the first bytes of 's' must match
896 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
897 * don't have any additional bytes available, the sequence, when
898 * completed might or might not fit in 32 bits. But if we have that
899 * next byte, we can tell for sure. If it is <= 0x83, then it does
900 * fit. */
901 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
902 return -1;
903 }
904
905 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
906 }
907
d22ec717
KW
908/* Starting with the #else, the rest of the function is identical except
909 * 1. we need to move the 'len' declaration to be global to the function
910 * 2. the endif move to just after the UNUSED_ARG.
911 * An empty endif is given just below to satisfy the preprocessor
912 */
a77c906e
KW
913#endif
914
d22ec717 915#endif
a77c906e 916
12a4bed3
KW
917#undef F0_ABOVE_OVERLONG
918#undef F8_ABOVE_OVERLONG
919#undef FC_ABOVE_OVERLONG
920#undef FE_ABOVE_OVERLONG
921#undef FF_OVERLONG_PREFIX
922
35f8c9bd 923STRLEN
1376b35c 924Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 925{
2b479609 926 STRLEN len;
12a4bed3 927 const U8 *x;
35f8c9bd 928
2b479609
KW
929 /* A helper function that should not be called directly.
930 *
931 * This function returns non-zero if the string beginning at 's' and
932 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
933 * code point; otherwise it returns 0. The examination stops after the
934 * first code point in 's' is validated, not looking at the rest of the
935 * input. If 'e' is such that there are not enough bytes to represent a
936 * complete code point, this function will return non-zero anyway, if the
937 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
938 * excluded by 'flags'.
939 *
940 * A non-zero return gives the number of bytes required to represent the
941 * code point. Be aware that if the input is for a partial character, the
942 * return will be larger than 'e - s'.
943 *
944 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
945 * The caller should have excluded the possibility of it being invariant
946 * before calling this function.
2b479609
KW
947 *
948 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
949 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
950 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
951 * disallowed by the flags. If the input is only for a partial character,
952 * the function will return non-zero if there is any sequence of
953 * well-formed UTF-8 that, when appended to the input sequence, could
954 * result in an allowed code point; otherwise it returns 0. Non characters
955 * cannot be determined based on partial character input. But many of the
956 * other excluded types can be determined with just the first one or two
957 * bytes.
958 *
959 */
960
1376b35c 961 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
2b479609
KW
962
963 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 964 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 965 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 966
2b479609 967 /* A variant char must begin with a start byte */
35f8c9bd
KW
968 if (UNLIKELY(! UTF8_IS_START(*s))) {
969 return 0;
970 }
971
edc2c47a
KW
972 /* Examine a maximum of a single whole code point */
973 if (e - s > UTF8SKIP(s)) {
974 e = s + UTF8SKIP(s);
975 }
976
2b479609
KW
977 len = e - s;
978
979 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
980 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 981
56576a04
KW
982 /* Here, we are disallowing some set of largish code points, and the
983 * first byte indicates the sequence is for a code point that could be
984 * in the excluded set. We generally don't have to look beyond this or
985 * the second byte to see if the sequence is actually for one of the
986 * excluded classes. The code below is derived from this table:
987 *
2b479609
KW
988 * UTF-8 UTF-EBCDIC I8
989 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
990 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
991 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
992 *
56576a04
KW
993 * Keep in mind that legal continuation bytes range between \x80..\xBF
994 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
995 * continuation bytes. Hence, we don't have to test the upper edge
996 * because if any of those is encountered, the sequence is malformed,
997 * and would fail elsewhere in this function.
998 *
999 * The code here likewise assumes that there aren't other
1000 * malformations; again the function should fail elsewhere because of
1001 * these. For example, an overlong beginning with FC doesn't actually
1002 * have to be a super; it could actually represent a small code point,
1003 * even U+0000. But, since overlongs (and other malformations) are
1004 * illegal, the function should return FALSE in either case.
2b479609
KW
1005 */
1006
1007#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1008# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 1009# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 1010
19794540
KW
1011# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1012 /* B6 and B7 */ \
1013 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1014# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1015#else
1016# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1017# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1018# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1019# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1020#endif
1021
1022 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1023 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1024 {
2b479609
KW
1025 return 0; /* Above Unicode */
1026 }
1027
d044b7a7 1028 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1029 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1030 {
57ff5f59 1031 return 0;
2b479609
KW
1032 }
1033
1034 if (len > 1) {
1035 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1036
1037 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1038 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1039 {
1040 return 0; /* Above Unicode */
1041 }
1042
1043 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1044 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1045 {
1046 return 0; /* Surrogate */
1047 }
1048
1049 if ( (flags & UTF8_DISALLOW_NONCHAR)
1050 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1051 {
1052 return 0; /* Noncharacter code point */
1053 }
1054 }
1055 }
1056
1057 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1058 for (x = s + 1; x < e; x++) {
1059 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1060 return 0;
1061 }
1062 }
1063
af13dd8a 1064 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1065 * overlong. */
d6be65ae 1066 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1067 return 0;
af13dd8a
KW
1068 }
1069
12a4bed3
KW
1070 /* And finally, that the code point represented fits in a word on this
1071 * platform */
e050c007
KW
1072 if (0 < does_utf8_overflow(s, e,
1073 0 /* Don't consider overlongs */
1074 ))
1075 {
12a4bed3 1076 return 0;
35f8c9bd
KW
1077 }
1078
2b479609 1079 return UTF8SKIP(s);
35f8c9bd
KW
1080}
1081
7e2f38b2 1082char *
63ab03b3 1083Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1084{
1085 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1086 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1087 * Currently, there are only two formats, so it is currently a bool:
1088 * 0 \xab
1089 * 1 ab (that is a space between two hex digit bytes)
1090 */
7cf8d05d
KW
1091
1092 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1093 trailing NUL */
63ab03b3
KW
1094 const U8 * s = start;
1095 const U8 * const e = start + len;
7cf8d05d
KW
1096 char * output;
1097 char * d;
1098
1099 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1100
1101 Newx(output, output_len, char);
1102 SAVEFREEPV(output);
1103
1104 d = output;
63ab03b3 1105 for (s = start; s < e; s++) {
7cf8d05d
KW
1106 const unsigned high_nibble = (*s & 0xF0) >> 4;
1107 const unsigned low_nibble = (*s & 0x0F);
1108
7e2f38b2 1109 if (format) {
63ab03b3
KW
1110 if (s > start) {
1111 *d++ = ' ';
1112 }
7e2f38b2
KW
1113 }
1114 else {
1115 *d++ = '\\';
1116 *d++ = 'x';
1117 }
7cf8d05d
KW
1118
1119 if (high_nibble < 10) {
1120 *d++ = high_nibble + '0';
1121 }
1122 else {
1123 *d++ = high_nibble - 10 + 'a';
1124 }
1125
1126 if (low_nibble < 10) {
1127 *d++ = low_nibble + '0';
1128 }
1129 else {
1130 *d++ = low_nibble - 10 + 'a';
1131 }
1132 }
1133
1134 *d = '\0';
1135 return output;
1136}
1137
806547a7 1138PERL_STATIC_INLINE char *
7cf8d05d
KW
1139S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1140
421da25c 1141 /* Max number of bytes to print */
3cc6a05e 1142 STRLEN print_len,
7cf8d05d
KW
1143
1144 /* Which one is the non-continuation */
1145 const STRLEN non_cont_byte_pos,
1146
1147 /* How many bytes should there be? */
1148 const STRLEN expect_len)
806547a7
KW
1149{
1150 /* Return the malformation warning text for an unexpected continuation
1151 * byte. */
1152
7cf8d05d 1153 const char * const where = (non_cont_byte_pos == 1)
806547a7 1154 ? "immediately"
7cf8d05d
KW
1155 : Perl_form(aTHX_ "%d bytes",
1156 (int) non_cont_byte_pos);
421da25c
KW
1157 const U8 * x = s + non_cont_byte_pos;
1158 const U8 * e = s + print_len;
806547a7
KW
1159
1160 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1161
7cf8d05d
KW
1162 /* We don't need to pass this parameter, but since it has already been
1163 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1164 assert(expect_len == UTF8SKIP(s));
1165
421da25c
KW
1166 /* As a defensive coding measure, don't output anything past a NUL. Such
1167 * bytes shouldn't be in the middle of a malformation, and could mark the
1168 * end of the allocated string, and what comes after is undefined */
1169 for (; x < e; x++) {
1170 if (*x == '\0') {
1171 x++; /* Output this particular NUL */
1172 break;
1173 }
1174 }
1175
7cf8d05d
KW
1176 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1177 " %s after start byte 0x%02x; need %d bytes, got %d)",
1178 malformed_text,
421da25c 1179 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1180 *(s + non_cont_byte_pos),
1181 where,
1182 *s,
1183 (int) expect_len,
1184 (int) non_cont_byte_pos);
806547a7
KW
1185}
1186
35f8c9bd
KW
1187/*
1188
de69f3af 1189=for apidoc utf8n_to_uvchr
378516de
KW
1190
1191THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1192Most code should use L</utf8_to_uvchr_buf>() rather than call this
1193directly.
67e989fb 1194
9041c2e3 1195Bottom level UTF-8 decode routine.
de69f3af 1196Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1197which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1198C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1199the length, in bytes, of that character.
949cf498
KW
1200
1201The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1202well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1203causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1204is the next possible position in C<s> that could begin a non-malformed
1205character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1206is raised. Some UTF-8 input sequences may contain multiple malformations.
1207This function tries to find every possible one in each call, so multiple
56576a04 1208warnings can be raised for the same sequence.
949cf498
KW
1209
1210Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1211individual types of malformations, such as the sequence being overlong (that
1212is, when there is a shorter sequence that can express the same code point;
1213overlong sequences are expressly forbidden in the UTF-8 standard due to
1214potential security issues). Another malformation example is the first byte of
1215a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1216flags. Even if allowed, this function generally returns the Unicode
1217REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1218F<utf8.h> to override this behavior for the overlong malformations, but don't
1219do that except for very specialized purposes.
949cf498 1220
796b6530 1221The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1222flags) malformation is found. If this flag is set, the routine assumes that
1223the caller will raise a warning, and this function will silently just set
d088425d
KW
1224C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1225
75200dff 1226Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1227character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1228in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1229be set to 1. To disambiguate, upon a zero return, see if the first byte of
1230C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1231error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1232
1233Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1234Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1235By default these are considered regular code points, but certain situations
ecc1615f
KW
1236warrant special handling for them, which can be specified using the C<flags>
1237parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1238three classes are treated as malformations and handled as such. The flags
1239C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1240C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1241disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1242restricts the allowed inputs to the strict UTF-8 traditionally defined by
1243Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1244definition given by
e2176993 1245L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f
KW
1246The difference between traditional strictness and C9 strictness is that the
1247latter does not forbid non-character code points. (They are still discouraged,
1248however.) For more discussion see L<perlunicode/Noncharacter code points>.
1249
1250The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1251C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1252C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1253raised for their respective categories, but otherwise the code points are
1254considered valid (not malformations). To get a category to both be treated as
1255a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1256(But note that warnings are not raised if lexically disabled nor if
796b6530 1257C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1258
57ff5f59
KW
1259Extremely high code points were never specified in any standard, and require an
1260extension to UTF-8 to express, which Perl does. It is likely that programs
1261written in something other than Perl would not be able to read files that
1262contain these; nor would Perl understand files written by something that uses a
1263different extension. For these reasons, there is a separate set of flags that
1264can warn and/or disallow these extremely high code points, even if other
1265above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1266C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1267L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1268above-Unicode code points, including these, as malformations.
1269(Note that the Unicode standard considers anything above 0x10FFFF to be
1270illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1271(2**31 -1))
1272
1273A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1274retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1275C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1276C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1277can apply to code points that actually do fit in 31 bits. This happens on
1278EBCDIC platforms, and sometimes when the L<overlong
1279malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1280describe the situation in all cases.
1281
ab8e6d41 1282
949cf498
KW
1283All other code points corresponding to Unicode characters, including private
1284use and those yet to be assigned, are never considered malformed and never
1285warn.
67e989fb 1286
37607a96 1287=cut
f9380377
KW
1288
1289Also implemented as a macro in utf8.h
1290*/
1291
1292UV
e6a4ffc3
KW
1293Perl_utf8n_to_uvchr(const U8 *s,
1294 STRLEN curlen,
1295 STRLEN *retlen,
1296 const U32 flags)
f9380377
KW
1297{
1298 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1299
1300 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1301}
1302
1303/*
1304
1305=for apidoc utf8n_to_uvchr_error
1306
1307THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1308Most code should use L</utf8_to_uvchr_buf>() rather than call this
1309directly.
f9380377
KW
1310
1311This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1312are when an error is found. If you also need to know the generated warning
1313messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1314
1315It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1316all the others, C<errors>. If this parameter is 0, this function behaves
1317identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1318to a C<U32> variable, which this function sets to indicate any errors found.
1319Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1320C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1321of these bits will be set if a malformation is found, even if the input
7a65503b 1322C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1323exceptions are noted:
1324
1325=over 4
1326
57ff5f59 1327=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1328
57ff5f59
KW
1329The input sequence is not standard UTF-8, but a Perl extension. This bit is
1330set only if the input C<flags> parameter contains either the
1331C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1332
1333Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1334and so some extension must be used to express them. Perl uses a natural
1335extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1336extension to represent even higher ones, so that any code point that fits in a
133764-bit word can be represented. Text using these extensions is not likely to
1338be portable to non-Perl code. We lump both of these extensions together and
1339refer to them as Perl extended UTF-8. There exist other extensions that people
1340have invented, incompatible with Perl's.
1341
1342On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1343extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1344than on ASCII. Prior to that, code points 2**31 and higher were simply
1345unrepresentable, and a different, incompatible method was used to represent
1346code points between 2**30 and 2**31 - 1.
1347
1348On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1349Perl extended UTF-8 is used.
1350
1351In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1352may use for backward compatibility. That name is misleading, as this flag may
1353be set when the code point actually does fit in 31 bits. This happens on
1354EBCDIC platforms, and sometimes when the L<overlong
1355malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1356describes the situation in all cases.
f9380377
KW
1357
1358=item C<UTF8_GOT_CONTINUATION>
1359
1360The input sequence was malformed in that the first byte was a a UTF-8
1361continuation byte.
1362
1363=item C<UTF8_GOT_EMPTY>
1364
1365The input C<curlen> parameter was 0.
1366
1367=item C<UTF8_GOT_LONG>
1368
1369The input sequence was malformed in that there is some other sequence that
1370evaluates to the same code point, but that sequence is shorter than this one.
1371
fecaf136
KW
1372Until Unicode 3.1, it was legal for programs to accept this malformation, but
1373it was discovered that this created security issues.
1374
f9380377
KW
1375=item C<UTF8_GOT_NONCHAR>
1376
1377The code point represented by the input UTF-8 sequence is for a Unicode
1378non-character code point.
1379This bit is set only if the input C<flags> parameter contains either the
1380C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1381
1382=item C<UTF8_GOT_NON_CONTINUATION>
1383
1384The input sequence was malformed in that a non-continuation type byte was found
00d976bb
KW
1385in a position where only a continuation type one should be. See also
1386L</C<UTF8_GOT_SHORT>>.
f9380377
KW
1387
1388=item C<UTF8_GOT_OVERFLOW>
1389
1390The input sequence was malformed in that it is for a code point that is not
d22ec717 1391representable in the number of bits available in an IV on the current platform.
f9380377
KW
1392
1393=item C<UTF8_GOT_SHORT>
1394
1395The input sequence was malformed in that C<curlen> is smaller than required for
1396a complete sequence. In other words, the input is for a partial character
1397sequence.
1398
00d976bb
KW
1399
1400C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1401sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1402that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1403sequence was looked at. If no other flags are present, it means that the
1404sequence was valid as far as it went. Depending on the application, this could
1405mean one of three things:
1406
1407=over
1408
1409=item *
1410
1411The C<curlen> length parameter passed in was too small, and the function was
1412prevented from examining all the necessary bytes.
1413
1414=item *
1415
1416The buffer being looked at is based on reading data, and the data received so
1417far stopped in the middle of a character, so that the next read will
1418read the remainder of this character. (It is up to the caller to deal with the
1419split bytes somehow.)
1420
1421=item *
1422
1423This is a real error, and the partial sequence is all we're going to get.
1424
1425=back
1426
f9380377
KW
1427=item C<UTF8_GOT_SUPER>
1428
1429The input sequence was malformed in that it is for a non-Unicode code point;
1430that is, one above the legal Unicode maximum.
1431This bit is set only if the input C<flags> parameter contains either the
1432C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1433
1434=item C<UTF8_GOT_SURROGATE>
1435
1436The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1437code point.
1438This bit is set only if the input C<flags> parameter contains either the
1439C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1440
1441=back
1442
133551d8
KW
1443To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1444flag to suppress any warnings, and then examine the C<*errors> return.
1445
f9380377 1446=cut
37657a5b
KW
1447
1448Also implemented as a macro in utf8.h
37607a96 1449*/
67e989fb 1450
a0ed51b3 1451UV
e6a4ffc3 1452Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1453 STRLEN curlen,
1454 STRLEN *retlen,
1455 const U32 flags,
1456 U32 * errors)
1457{
1458 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1459
1460 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1461}
1462
1463/*
1464
1465=for apidoc utf8n_to_uvchr_msgs
1466
1467THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1468Most code should use L</utf8_to_uvchr_buf>() rather than call this
1469directly.
37657a5b
KW
1470
1471This function is for code that needs to know what the precise malformation(s)
1472are when an error is found, and wants the corresponding warning and/or error
1473messages to be returned to the caller rather than be displayed. All messages
1474that would have been displayed if all lexcial warnings are enabled will be
1475returned.
1476
1477It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1478placed after all the others, C<msgs>. If this parameter is 0, this function
1479behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1480be a pointer to an C<AV *> variable, in which this function creates a new AV to
1481contain any appropriate messages. The elements of the array are ordered so
1482that the first message that would have been displayed is in the 0th element,
1483and so on. Each element is a hash with three key-value pairs, as follows:
1484
1485=over 4
1486
1487=item C<text>
1488
1489The text of the message as a C<SVpv>.
1490
1491=item C<warn_categories>
1492
1493The warning category (or categories) packed into a C<SVuv>.
1494
1495=item C<flag>
1496
1497A single flag bit associated with this message, in a C<SVuv>.
1498The bit corresponds to some bit in the C<*errors> return value,
1499such as C<UTF8_GOT_LONG>.
1500
1501=back
1502
1503It's important to note that specifying this parameter as non-null will cause
1504any warnings this function would otherwise generate to be suppressed, and
1505instead be placed in C<*msgs>. The caller can check the lexical warnings state
1506(or not) when choosing what to do with the returned messages.
1507
1508If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1509no AV is created.
1510
1511The caller, of course, is responsible for freeing any returned AV.
1512
1513=cut
1514*/
1515
1516UV
e6a4ffc3 1517Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1518 STRLEN curlen,
1519 STRLEN *retlen,
1520 const U32 flags,
1521 U32 * errors,
1522 AV ** msgs)
a0ed51b3 1523{
d4c19fe8 1524 const U8 * const s0 = s;
2b9519f0 1525 const U8 * send = s0 + curlen;
5af9f822
KW
1526 U32 possible_problems; /* A bit is set here for each potential problem
1527 found as we go along */
1528 UV uv;
1529 STRLEN expectlen; /* How long should this sequence be? */
1530 STRLEN avail_len; /* When input is too short, gives what that is */
1531 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1532 gets set and discarded */
a0dbb045 1533
2b5e7bc2
KW
1534 /* The below are used only if there is both an overlong malformation and a
1535 * too short one. Otherwise the first two are set to 's0' and 'send', and
1536 * the third not used at all */
5af9f822 1537 U8 * adjusted_s0;
e9f2c446
KW
1538 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1539 routine; see [perl #130921] */
5af9f822 1540 UV uv_so_far;
e6a4ffc3 1541 dTHX;
5af9f822 1542
e6a4ffc3 1543 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1544
1545 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1546 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1547 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1548 * final case. */
1549
1550#ifndef EBCDIC
1551
1552 /* Each of the affected Hanguls starts with \xED */
1553
1554 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1555 if (retlen) {
1556 *retlen = 3;
1557 }
1558 if (errors) {
1559 *errors = 0;
1560 }
1561 if (msgs) {
1562 *msgs = NULL;
1563 }
1564
1565 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1566 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1567 | (s0[2] & UTF_CONTINUATION_MASK);
1568 }
1569
1570#endif
1571
1572 /* In conjunction with the exhaustive tests that can be enabled in
1573 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1574 * what it is intended to do, and that no flaws in it are masked by
1575 * dropping down and executing the code below
1576 assert(! isUTF8_CHAR(s0, send)
1577 || UTF8_IS_SURROGATE(s0, send)
1578 || UTF8_IS_SUPER(s0, send)
1579 || UTF8_IS_NONCHAR(s0,send));
1580 */
1581
1582 s = s0;
1583 uv = *s0;
1584 possible_problems = 0;
1585 expectlen = 0;
1586 avail_len = 0;
1587 discard_errors = 0;
1588 adjusted_s0 = (U8 *) s0;
1589 uv_so_far = 0;
1590
f9380377
KW
1591 if (errors) {
1592 *errors = 0;
1593 }
1594 else {
1595 errors = &discard_errors;
1596 }
a0dbb045 1597
eb83ed87
KW
1598 /* The order of malformation tests here is important. We should consume as
1599 * few bytes as possible in order to not skip any valid character. This is
1600 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
e2176993 1601 * https://unicode.org/reports/tr36 for more discussion as to why. For
eb83ed87
KW
1602 * example, once we've done a UTF8SKIP, we can tell the expected number of
1603 * bytes, and could fail right off the bat if the input parameters indicate
1604 * that there are too few available. But it could be that just that first
1605 * byte is garbled, and the intended character occupies fewer bytes. If we
1606 * blindly assumed that the first byte is correct, and skipped based on
1607 * that number, we could skip over a valid input character. So instead, we
1608 * always examine the sequence byte-by-byte.
1609 *
1610 * We also should not consume too few bytes, otherwise someone could inject
1611 * things. For example, an input could be deliberately designed to
1612 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1613 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1614 * which is actually part of of the overflowing sequence), that could look
1615 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1616 * sequence and process the rest, inappropriately.
1617 *
1618 * Some possible input sequences are malformed in more than one way. This
1619 * function goes to lengths to try to find all of them. This is necessary
1620 * for correctness, as the inputs may allow one malformation but not
1621 * another, and if we abandon searching for others after finding the
1622 * allowed one, we could allow in something that shouldn't have been.
1623 */
eb83ed87 1624
b5b9af04 1625 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1626 possible_problems |= UTF8_GOT_EMPTY;
1627 curlen = 0;
5a48568d 1628 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1629 goto ready_to_handle_errors;
0c443dc2
JH
1630 }
1631
eb83ed87
KW
1632 expectlen = UTF8SKIP(s);
1633
1634 /* A well-formed UTF-8 character, as the vast majority of calls to this
1635 * function will be for, has this expected length. For efficiency, set
1636 * things up here to return it. It will be overriden only in those rare
1637 * cases where a malformation is found */
1638 if (retlen) {
1639 *retlen = expectlen;
1640 }
1641
eb83ed87 1642 /* A continuation character can't start a valid sequence */
b5b9af04 1643 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1644 possible_problems |= UTF8_GOT_CONTINUATION;
1645 curlen = 1;
1646 uv = UNICODE_REPLACEMENT;
1647 goto ready_to_handle_errors;
ba210ebe 1648 }
9041c2e3 1649
dcd27b3c 1650 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1651 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1652 * because it excludes start bytes like \xC0 that always lead to
1653 * overlongs.) */
dcd27b3c 1654
534752c1
KW
1655 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1656 * that indicate the number of bytes in the character's whole UTF-8
1657 * sequence, leaving just the bits that are part of the value. */
1658 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1659
e308b348
KW
1660 /* Setup the loop end point, making sure to not look past the end of the
1661 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1662 if (UNLIKELY(curlen < expectlen)) {
1663 possible_problems |= UTF8_GOT_SHORT;
1664 avail_len = curlen;
e308b348
KW
1665 }
1666 else {
2b9519f0 1667 send = (U8*) s0 + expectlen;
e308b348 1668 }
e308b348 1669
eb83ed87 1670 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1671 * accumulating each into the working value as we go. */
eb83ed87 1672 for (s = s0 + 1; s < send; s++) {
b5b9af04 1673 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1674 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1675 continue;
1676 }
1677
1678 /* Here, found a non-continuation before processing all expected bytes.
1679 * This byte indicates the beginning of a new character, so quit, even
1680 * if allowing this malformation. */
2b5e7bc2 1681 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1682 break;
eb83ed87
KW
1683 } /* End of loop through the character's bytes */
1684
1685 /* Save how many bytes were actually in the character */
1686 curlen = s - s0;
1687
2b5e7bc2
KW
1688 /* Note that there are two types of too-short malformation. One is when
1689 * there is actual wrong data before the normal termination of the
1690 * sequence. The other is that the sequence wasn't complete before the end
1691 * of the data we are allowed to look at, based on the input 'curlen'.
1692 * This means that we were passed data for a partial character, but it is
1693 * valid as far as we saw. The other is definitely invalid. This
1694 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1695 * separate.
1696 *
1697 * A convenience macro that matches either of the too-short conditions. */
1698# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1699
1700 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1701 uv_so_far = uv;
1702 uv = UNICODE_REPLACEMENT;
1703 }
2b5e7bc2 1704
08e73697
KW
1705 /* Check for overflow. The algorithm requires us to not look past the end
1706 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1707 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1708 1 /* Do consider overlongs */
1709 )))
1710 {
2b5e7bc2
KW
1711 possible_problems |= UTF8_GOT_OVERFLOW;
1712 uv = UNICODE_REPLACEMENT;
eb83ed87 1713 }
eb83ed87 1714
2b5e7bc2
KW
1715 /* Check for overlong. If no problems so far, 'uv' is the correct code
1716 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1717 * we must look at the UTF-8 byte sequence itself to see if it is for an
1718 * overlong */
1719 if ( ( LIKELY(! possible_problems)
1720 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1721 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1722 && ( UNLIKELY(! UTF8_IS_START(*s0))
1723 || ( curlen > 1
d6be65ae 1724 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1725 s - s0))))))
2f8f112e 1726 {
2b5e7bc2
KW
1727 possible_problems |= UTF8_GOT_LONG;
1728
abc28b54 1729 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1730
abc28b54
KW
1731 /* The calculation in the 'true' branch of this 'if'
1732 * below won't work if overflows, and isn't needed
1733 * anyway. Further below we handle all overflow
1734 * cases */
1735 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1736 {
2b5e7bc2
KW
1737 UV min_uv = uv_so_far;
1738 STRLEN i;
1739
1740 /* Here, the input is both overlong and is missing some trailing
1741 * bytes. There is no single code point it could be for, but there
1742 * may be enough information present to determine if what we have
1743 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1744 * The code further below has the intelligence to determine this,
1745 * but just for non-overlong UTF-8 sequences. What we do here is
1746 * calculate the smallest code point the input could represent if
1747 * there were no too short malformation. Then we compute and save
1748 * the UTF-8 for that, which is what the code below looks at
1749 * instead of the raw input. It turns out that the smallest such
1750 * code point is all we need. */
2b5e7bc2
KW
1751 for (i = curlen; i < expectlen; i++) {
1752 min_uv = UTF8_ACCUMULATE(min_uv,
1753 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1754 }
1755
e9f2c446 1756 adjusted_s0 = temp_char_buf;
57ff5f59 1757 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1758 }
eb83ed87
KW
1759 }
1760
56576a04
KW
1761 /* Here, we have found all the possible problems, except for when the input
1762 * is for a problematic code point not allowed by the input parameters. */
1763
06188866
KW
1764 /* uv is valid for overlongs */
1765 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1766
1767 /* isn't problematic if < this */
1768 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1769 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1770
1771 /* if overflow, we know without looking further
1772 * precisely which of the problematic types it is,
1773 * and we deal with those in the overflow handling
1774 * code */
1775 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1776 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1777 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1778 && ((flags & ( UTF8_DISALLOW_NONCHAR
1779 |UTF8_DISALLOW_SURROGATE
1780 |UTF8_DISALLOW_SUPER
d044b7a7 1781 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1782 |UTF8_WARN_NONCHAR
1783 |UTF8_WARN_SURROGATE
1784 |UTF8_WARN_SUPER
d22ec717 1785 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1786 {
2b5e7bc2
KW
1787 /* If there were no malformations, or the only malformation is an
1788 * overlong, 'uv' is valid */
1789 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1790 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1791 possible_problems |= UTF8_GOT_SURROGATE;
1792 }
1793 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1794 possible_problems |= UTF8_GOT_SUPER;
1795 }
1796 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1797 possible_problems |= UTF8_GOT_NONCHAR;
1798 }
1799 }
1800 else { /* Otherwise, need to look at the source UTF-8, possibly
1801 adjusted to be non-overlong */
1802
1803 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1804 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1805 {
2b5e7bc2
KW
1806 possible_problems |= UTF8_GOT_SUPER;
1807 }
1808 else if (curlen > 1) {
1809 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1810 NATIVE_UTF8_TO_I8(*adjusted_s0),
1811 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1812 {
2b5e7bc2 1813 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1814 }
2b5e7bc2
KW
1815 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1816 NATIVE_UTF8_TO_I8(*adjusted_s0),
1817 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1818 {
1819 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1820 }
1821 }
c0236afe 1822
2b5e7bc2
KW
1823 /* We need a complete well-formed UTF-8 character to discern
1824 * non-characters, so can't look for them here */
1825 }
1826 }
949cf498 1827
2b5e7bc2
KW
1828 ready_to_handle_errors:
1829
1830 /* At this point:
1831 * curlen contains the number of bytes in the sequence that
1832 * this call should advance the input by.
e308b348
KW
1833 * avail_len gives the available number of bytes passed in, but
1834 * only if this is less than the expected number of
1835 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1836 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1837 * is set in it for each potential problem found.
1838 * uv contains the code point the input sequence
1839 * represents; or if there is a problem that prevents
1840 * a well-defined value from being computed, it is
1841 * some subsitute value, typically the REPLACEMENT
1842 * CHARACTER.
1843 * s0 points to the first byte of the character
56576a04
KW
1844 * s points to just after were we left off processing
1845 * the character
1846 * send points to just after where that character should
1847 * end, based on how many bytes the start byte tells
1848 * us should be in it, but no further than s0 +
1849 * avail_len
2b5e7bc2 1850 */
eb83ed87 1851
2b5e7bc2
KW
1852 if (UNLIKELY(possible_problems)) {
1853 bool disallowed = FALSE;
1854 const U32 orig_problems = possible_problems;
1855
37657a5b
KW
1856 if (msgs) {
1857 *msgs = NULL;
1858 }
1859
2b5e7bc2
KW
1860 while (possible_problems) { /* Handle each possible problem */
1861 UV pack_warn = 0;
1862 char * message = NULL;
37657a5b 1863 U32 this_flag_bit = 0;
2b5e7bc2
KW
1864
1865 /* Each 'if' clause handles one problem. They are ordered so that
1866 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1867 * ones; this is kinda in decreasing severity order. But the
1868 * overlong must come last, as it changes 'uv' looked at by the
1869 * others */
2b5e7bc2
KW
1870 if (possible_problems & UTF8_GOT_OVERFLOW) {
1871
56576a04
KW
1872 /* Overflow means also got a super and are using Perl's
1873 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1874 possible_problems
d044b7a7 1875 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1876 *errors |= UTF8_GOT_OVERFLOW;
1877
1878 /* But the API says we flag all errors found */
1879 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1880 *errors |= UTF8_GOT_SUPER;
1881 }
ddb65933 1882 if (flags
d044b7a7 1883 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1884 {
d044b7a7 1885 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1886 }
2b5e7bc2 1887
d60baaa7 1888 /* Disallow if any of the three categories say to */
56576a04 1889 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1890 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1891 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1892 {
1893 disallowed = TRUE;
1894 }
1895
d22ec717
KW
1896 /* Likewise, warn if any say to */
1897 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1898 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1899 {
2b5e7bc2 1900
ddb65933
KW
1901 /* The warnings code explicitly says it doesn't handle the
1902 * case of packWARN2 and two categories which have
1903 * parent-child relationship. Even if it works now to
1904 * raise the warning if either is enabled, it wouldn't
1905 * necessarily do so in the future. We output (only) the
56576a04 1906 * most dire warning */
ddb65933 1907 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1908 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1909 pack_warn = packWARN(WARN_UTF8);
1910 }
37657a5b 1911 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1912 pack_warn = packWARN(WARN_NON_UNICODE);
1913 }
1914 if (pack_warn) {
1915 message = Perl_form(aTHX_ "%s: %s (overflows)",
1916 malformed_text,
05b9033b 1917 _byte_dump_string(s0, curlen, 0));
37657a5b 1918 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1919 }
2b5e7bc2
KW
1920 }
1921 }
1922 }
1923 else if (possible_problems & UTF8_GOT_EMPTY) {
1924 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1925 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1926
1927 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1928
1929 /* This so-called malformation is now treated as a bug in
1930 * the caller. If you have nothing to decode, skip calling
1931 * this function */
1932 assert(0);
1933
2b5e7bc2 1934 disallowed = TRUE;
37657a5b
KW
1935 if ( (msgs
1936 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1937 {
2b5e7bc2
KW
1938 pack_warn = packWARN(WARN_UTF8);
1939 message = Perl_form(aTHX_ "%s (empty string)",
1940 malformed_text);
37657a5b 1941 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1942 }
1943 }
1944 }
1945 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1946 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1947 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1948
1949 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1950 disallowed = TRUE;
37657a5b
KW
1951 if (( msgs
1952 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1953 {
2b5e7bc2
KW
1954 pack_warn = packWARN(WARN_UTF8);
1955 message = Perl_form(aTHX_
1956 "%s: %s (unexpected continuation byte 0x%02x,"
1957 " with no preceding start byte)",
1958 malformed_text,
7e2f38b2 1959 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1960 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1961 }
1962 }
1963 }
2b5e7bc2
KW
1964 else if (possible_problems & UTF8_GOT_SHORT) {
1965 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1966 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1967
1968 if (! (flags & UTF8_ALLOW_SHORT)) {
1969 disallowed = TRUE;
37657a5b
KW
1970 if (( msgs
1971 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1972 {
2b5e7bc2
KW
1973 pack_warn = packWARN(WARN_UTF8);
1974 message = Perl_form(aTHX_
56576a04
KW
1975 "%s: %s (too short; %d byte%s available, need %d)",
1976 malformed_text,
1977 _byte_dump_string(s0, send - s0, 0),
1978 (int)avail_len,
1979 avail_len == 1 ? "" : "s",
1980 (int)expectlen);
37657a5b 1981 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1982 }
1983 }
ba210ebe 1984
2b5e7bc2 1985 }
e308b348
KW
1986 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1987 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1988 *errors |= UTF8_GOT_NON_CONTINUATION;
1989
1990 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1991 disallowed = TRUE;
37657a5b
KW
1992 if (( msgs
1993 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1994 {
99a765e9
KW
1995
1996 /* If we don't know for sure that the input length is
1997 * valid, avoid as much as possible reading past the
1998 * end of the buffer */
1999 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2000 ? s - s0
2001 : send - s0;
e308b348
KW
2002 pack_warn = packWARN(WARN_UTF8);
2003 message = Perl_form(aTHX_ "%s",
2004 unexpected_non_continuation_text(s0,
99a765e9 2005 printlen,
e308b348
KW
2006 s - s0,
2007 (int) expectlen));
37657a5b 2008 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2009 }
2010 }
2011 }
2b5e7bc2
KW
2012 else if (possible_problems & UTF8_GOT_SURROGATE) {
2013 possible_problems &= ~UTF8_GOT_SURROGATE;
2014
f9380377
KW
2015 if (flags & UTF8_WARN_SURROGATE) {
2016 *errors |= UTF8_GOT_SURROGATE;
2017
2018 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2019 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2020 {
2b5e7bc2
KW
2021 pack_warn = packWARN(WARN_SURROGATE);
2022
2023 /* These are the only errors that can occur with a
2024 * surrogate when the 'uv' isn't valid */
2025 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2026 message = Perl_form(aTHX_
2027 "UTF-16 surrogate (any UTF-8 sequence that"
2028 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2029 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2030 }
2031 else {
c94c2f39 2032 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2033 }
37657a5b 2034 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2035 }
2b5e7bc2 2036 }
ba210ebe 2037
2b5e7bc2
KW
2038 if (flags & UTF8_DISALLOW_SURROGATE) {
2039 disallowed = TRUE;
f9380377 2040 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2041 }
2042 }
2043 else if (possible_problems & UTF8_GOT_SUPER) {
2044 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2045
f9380377
KW
2046 if (flags & UTF8_WARN_SUPER) {
2047 *errors |= UTF8_GOT_SUPER;
2048
2049 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2050 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2051 {
2b5e7bc2
KW
2052 pack_warn = packWARN(WARN_NON_UNICODE);
2053
2054 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2055 message = Perl_form(aTHX_
2056 "Any UTF-8 sequence that starts with"
2057 " \"%s\" is for a non-Unicode code point,"
2058 " may not be portable",
7e2f38b2 2059 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2060 }
2061 else {
c94c2f39 2062 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2063 }
37657a5b 2064 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2065 }
2b5e7bc2 2066 }
ba210ebe 2067
57ff5f59
KW
2068 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2069 * and before possibly bailing out, so that the more dire
2070 * warning will override the regular one. */
2071 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2072 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2073 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2074 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2075 {
db0f09e6 2076 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2077
57ff5f59
KW
2078 /* If it is an overlong that evaluates to a code point
2079 * that doesn't have to use the Perl extended UTF-8, it
2080 * still used it, and so we output a message that
2081 * doesn't refer to the code point. The same is true
2082 * if there was a SHORT malformation where the code
2083 * point is not valid. In that case, 'uv' will have
2084 * been set to the REPLACEMENT CHAR, and the message
2085 * below without the code point in it will be selected
2086 * */
2087 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2088 message = Perl_form(aTHX_
57ff5f59 2089 perl_extended_cp_format, uv);
2b5e7bc2
KW
2090 }
2091 else {
2092 message = Perl_form(aTHX_
57ff5f59
KW
2093 "Any UTF-8 sequence that starts with"
2094 " \"%s\" is a Perl extension, and"
2095 " so is not portable",
2096 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2097 }
37657a5b 2098 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2099 }
2100
d044b7a7
KW
2101 if (flags & ( UTF8_WARN_PERL_EXTENDED
2102 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2103 {
d044b7a7 2104 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2105
d044b7a7 2106 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2107 disallowed = TRUE;
2108 }
2b5e7bc2
KW
2109 }
2110 }
eb83ed87 2111
2b5e7bc2 2112 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2113 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2114 disallowed = TRUE;
2115 }
2b5e7bc2
KW
2116 }
2117 else if (possible_problems & UTF8_GOT_NONCHAR) {
2118 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2119
f9380377
KW
2120 if (flags & UTF8_WARN_NONCHAR) {
2121 *errors |= UTF8_GOT_NONCHAR;
2122
2123 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2124 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2125 {
2b5e7bc2
KW
2126 /* The code above should have guaranteed that we don't
2127 * get here with errors other than overlong */
2128 assert (! (orig_problems
2129 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2130
2131 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2132 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2133 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2134 }
2b5e7bc2 2135 }
5b311467 2136
2b5e7bc2
KW
2137 if (flags & UTF8_DISALLOW_NONCHAR) {
2138 disallowed = TRUE;
f9380377 2139 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2140 }
6c64cd9d
KW
2141 }
2142 else if (possible_problems & UTF8_GOT_LONG) {
2143 possible_problems &= ~UTF8_GOT_LONG;
2144 *errors |= UTF8_GOT_LONG;
2145
2146 if (flags & UTF8_ALLOW_LONG) {
2147
2148 /* We don't allow the actual overlong value, unless the
2149 * special extra bit is also set */
2150 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2151 & ~UTF8_ALLOW_LONG)))
2152 {
2153 uv = UNICODE_REPLACEMENT;
2154 }
2155 }
2156 else {
2157 disallowed = TRUE;
2158
37657a5b
KW
2159 if (( msgs
2160 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2161 {
6c64cd9d
KW
2162 pack_warn = packWARN(WARN_UTF8);
2163
2164 /* These error types cause 'uv' to be something that
2165 * isn't what was intended, so can't use it in the
2166 * message. The other error types either can't
2167 * generate an overlong, or else the 'uv' is valid */
2168 if (orig_problems &
2169 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2170 {
2171 message = Perl_form(aTHX_
2172 "%s: %s (any UTF-8 sequence that starts"
2173 " with \"%s\" is overlong which can and"
2174 " should be represented with a"
2175 " different, shorter sequence)",
2176 malformed_text,
2177 _byte_dump_string(s0, send - s0, 0),
2178 _byte_dump_string(s0, curlen, 0));
2179 }
2180 else {
2181 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2182 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2183 uv, 0);
d819dc50
KW
2184 /* Don't use U+ for non-Unicode code points, which
2185 * includes those in the Latin1 range */
2186 const char * preface = ( uv > PERL_UNICODE_MAX
2187#ifdef EBCDIC
2188 || uv <= 0xFF
2189#endif
2190 )
2191 ? "0x"
2192 : "U+";
6c64cd9d
KW
2193 message = Perl_form(aTHX_
2194 "%s: %s (overlong; instead use %s to represent"
2195 " %s%0*" UVXf ")",
2196 malformed_text,
2197 _byte_dump_string(s0, send - s0, 0),
2198 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2199 preface,
2200 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2201 small code points */
1be62ab9 2202 UNI_TO_NATIVE(uv));
6c64cd9d 2203 }
37657a5b 2204 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2205 }
2206 }
2b5e7bc2
KW
2207 } /* End of looking through the possible flags */
2208
2209 /* Display the message (if any) for the problem being handled in
2210 * this iteration of the loop */
2211 if (message) {
37657a5b 2212 if (msgs) {
37657a5b
KW
2213 assert(this_flag_bit);
2214
2215 if (*msgs == NULL) {
2216 *msgs = newAV();
2217 }
2218
bb07812e
KW
2219 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2220 pack_warn,
2221 this_flag_bit)));
37657a5b
KW
2222 }
2223 else if (PL_op)
2b5e7bc2
KW
2224 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2225 OP_DESC(PL_op));
2226 else
2227 Perl_warner(aTHX_ pack_warn, "%s", message);
2228 }
ddb65933 2229 } /* End of 'while (possible_problems)' */
a0dbb045 2230
2b5e7bc2
KW
2231 /* Since there was a possible problem, the returned length may need to
2232 * be changed from the one stored at the beginning of this function.
2233 * Instead of trying to figure out if that's needed, just do it. */
2234 if (retlen) {
2235 *retlen = curlen;
2236 }
a0dbb045 2237
2b5e7bc2
KW
2238 if (disallowed) {
2239 if (flags & UTF8_CHECK_ONLY && retlen) {
2240 *retlen = ((STRLEN) -1);
2241 }
2242 return 0;
2243 }
eb83ed87 2244 }
ba210ebe 2245
2b5e7bc2 2246 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2247}
2248
8e84507e 2249/*
ec5f19d0
KW
2250=for apidoc utf8_to_uvchr_buf
2251
2252Returns the native code point of the first character in the string C<s> which
2253is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2254C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2255
524080c4
KW
2256If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2257enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2258C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2259(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2260C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2261the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2262See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2263returned.
ec5f19d0
KW
2264
2265=cut
52be2536
KW
2266
2267Also implemented as a macro in utf8.h
2268
ec5f19d0
KW
2269*/
2270
2271
2272UV
2273Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2274{
7f974d7e
KW
2275 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2276
9a9a6c98 2277 return utf8_to_uvchr_buf_helper(s, send, retlen);
ec5f19d0
KW
2278}
2279
52be2536
KW
2280/* This is marked as deprecated
2281 *
ec5f19d0
KW
2282=for apidoc utf8_to_uvuni_buf
2283
de69f3af
KW
2284Only in very rare circumstances should code need to be dealing in Unicode
2285(as opposed to native) code points. In those few cases, use
09232555
KW
2286C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2287If you are not absolutely sure this is one of those cases, then assume it isn't
2288and use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2289
2290Returns the Unicode (not-native) code point of the first character in the
2291string C<s> which
ec5f19d0
KW
2292is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2293C<retlen> will be set to the length, in bytes, of that character.
2294
524080c4
KW
2295If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2296enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2297NULL) to -1. If those warnings are off, the computed value if well-defined (or
2298the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2299is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2300next possible position in C<s> that could begin a non-malformed character.
09232555
KW
2301See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2302returned.
ec5f19d0
KW
2303
2304=cut
2305*/
2306
2307UV
2308Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2309{
2310 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2311
2312 assert(send > s);
2313
5962d97e 2314 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2315}
2316
b76347f2 2317/*
87cea99e 2318=for apidoc utf8_length
b76347f2 2319
b2e7ed74
KW
2320Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2321at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2322same place, it returns 0 with no warning raised.
2323
2324If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2325and returns the number of valid characters.
b76347f2
JH
2326
2327=cut
2328*/
2329
2330STRLEN
35a4481c 2331Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2332{
2333 STRLEN len = 0;
2334
7918f24d
NC
2335 PERL_ARGS_ASSERT_UTF8_LENGTH;
2336
8850bf83
JH
2337 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2338 * the bitops (especially ~) can create illegal UTF-8.
2339 * In other words: in Perl UTF-8 is not just for Unicode. */
2340
12c43b0a 2341 if (UNLIKELY(e < s))
a3b680e6 2342 goto warn_and_return;
b76347f2 2343 while (s < e) {
4cbf4130 2344 s += UTF8SKIP(s);
8e91ec7f
AV
2345 len++;
2346 }
2347
12c43b0a 2348 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2349 len--;
2350 warn_and_return:
9b387841
NC
2351 if (PL_op)
2352 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2353 "%s in %s", unees, OP_DESC(PL_op));
2354 else
61a12c31 2355 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2356 }
2357
2358 return len;
2359}
2360
b06226ff 2361/*
fed3ba5d
NC
2362=for apidoc bytes_cmp_utf8
2363
a1433954 2364Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2365sequence of characters (stored as UTF-8)
2366in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2367equal, -1 or -2 if the first string is less than the second string, +1 or +2
2368if the first string is greater than the second string.
2369
2370-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2371longer string. -2 or +2 is returned if
2372there was a difference between characters
fed3ba5d
NC
2373within the strings.
2374
2375=cut
2376*/
2377
2378int
2379Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2380{
2381 const U8 *const bend = b + blen;
2382 const U8 *const uend = u + ulen;
2383
2384 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2385
2386 while (b < bend && u < uend) {
2387 U8 c = *u++;
2388 if (!UTF8_IS_INVARIANT(c)) {
2389 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2390 if (u < uend) {
2391 U8 c1 = *u++;
2392 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2393 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2394 } else {
2b5e7bc2 2395 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2396 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2397 "%s %s%s",
2398 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2399 PL_op ? " in " : "",
2400 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2401 return -2;
2402 }
2403 } else {
2404 if (PL_op)
2405 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2406 "%s in %s", unees, OP_DESC(PL_op));
2407 else
61a12c31 2408 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2409 return -2; /* Really want to return undef :-) */
2410 }
2411 } else {
2412 return -2;
2413 }
2414 }
2415 if (*b != c) {
2416 return *b < c ? -2 : +2;
2417 }
2418 ++b;
2419 }
2420
2421 if (b == bend && u == uend)
2422 return 0;
2423
2424 return b < bend ? +1 : -1;
2425}
2426
2427/*
87cea99e 2428=for apidoc utf8_to_bytes
6940069f 2429
3bc0c78c 2430Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2431Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2432updates C<*lenp> to contain the new length.
3bc0c78c
KW
2433Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2434
2435Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2436having saved the value of C<*lenp> before the call, and subtracting the
2437after-call value of C<*lenp> from it.
6940069f 2438
a1433954 2439If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2440
6940069f
GS
2441=cut
2442*/
2443
2444U8 *
09af0336 2445Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2446{
9fe0d3c2 2447 U8 * first_variant;
246fae53 2448
7918f24d 2449 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2450 PERL_UNUSED_CONTEXT;
7918f24d 2451
9fe0d3c2 2452 /* This is a no-op if no variants at all in the input */
09af0336 2453 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2454 return s;
2455 }
2456
2457 {
3c5aa262 2458 U8 * const save = s;
09af0336 2459 U8 * const send = s + *lenp;
3c5aa262
KW
2460 U8 * d;
2461
2462 /* Nothing before the first variant needs to be changed, so start the real
2463 * work there */
2464 s = first_variant;
2465 while (s < send) {
2466 if (! UTF8_IS_INVARIANT(*s)) {
2467 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2468 *lenp = ((STRLEN) -1);
3c5aa262
KW
2469 return 0;
2470 }
2471 s++;
d59937ca
KW
2472 }
2473 s++;
dcad2880 2474 }
dcad2880 2475
3c5aa262
KW
2476 /* Is downgradable, so do it */
2477 d = s = first_variant;
2478 while (s < send) {
2479 U8 c = *s++;
2480 if (! UVCHR_IS_INVARIANT(c)) {
2481 /* Then it is two-byte encoded */
2482 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2483 s++;
2484 }
2485 *d++ = c;
2486 }
2487 *d = '\0';
09af0336 2488 *lenp = d - save;
3c5aa262
KW
2489
2490 return save;
9fe0d3c2 2491 }
6940069f
GS
2492}
2493
2494/*
87cea99e 2495=for apidoc bytes_from_utf8
f9a63242 2496
09af0336 2497Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2498byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2499actually encoded in UTF-8.
2500
2501Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2502the input string.
2503
41ae6089
KW
2504Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2505not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2506C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2507
41ae6089 2508Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2509newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2510returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2511caller is responsible for arranging for the memory used by this string to get
2512freed.
f9a63242 2513
3bc0c78c 2514Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2515having saved the value of C<*lenp> before the call, and subtracting the
2516after-call value of C<*lenp> from it.
3bc0c78c 2517
37607a96 2518=cut
976c1b08
KW
2519
2520There is a macro that avoids this function call, but this is retained for
2521anyone who calls it with the Perl_ prefix */
f9a63242
JH
2522
2523U8 *
41ae6089 2524Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2525{
7918f24d 2526 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2527 PERL_UNUSED_CONTEXT;
f9a63242 2528
976c1b08
KW
2529 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2530}
2531
2532/*
2533No = here because currently externally undocumented
2534for apidoc bytes_from_utf8_loc
2535
2536Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2537to store the location of the first character in C<"s"> that cannot be
2538converted to non-UTF8.
2539
2540If that parameter is C<NULL>, this function behaves identically to
2541C<bytes_from_utf8>.
2542
2543Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2544C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2545
2546Otherwise, the function returns a newly created C<NUL>-terminated string
2547containing the non-UTF8 equivalent of the convertible first portion of
2548C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2549If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2550and C<*first_non_downgradable> is set to C<NULL>.
2551
2552Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2553first character in the original string that wasn't converted. C<*is_utf8p> is
2554unchanged. Note that the new string may have length 0.
2555
2556Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2557C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2558converts as many characters in it as possible stopping at the first one it
385b74be 2559finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2560set to point to that. The function returns the portion that could be converted
2561in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2562not including the terminating C<NUL>. If the very first character in the
2563original could not be converted, C<*lenp> will be 0, and the new string will
2564contain just a single C<NUL>. If the entire input string was converted,
2565C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2566
2567Upon successful return, the number of variants in the converted portion of the
2568string can be computed by having saved the value of C<*lenp> before the call,
2569and subtracting the after-call value of C<*lenp> from it.
2570
2571=cut
2572
2573
2574*/
2575
2576U8 *
2577Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2578{
2579 U8 *d;
2580 const U8 *original = s;
2581 U8 *converted_start;
2582 const U8 *send = s + *lenp;
f9a63242 2583
976c1b08 2584 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2585
976c1b08
KW
2586 if (! *is_utf8p) {
2587 if (first_unconverted) {
2588 *first_unconverted = NULL;
2589 }
2590
2591 return (U8 *) original;
2592 }
2593
2594 Newx(d, (*lenp) + 1, U8);
2595
2596 converted_start = d;
7299a045
KW
2597 while (s < send) {
2598 U8 c = *s++;
2599 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2600
2601 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2602 * have to stop now */
2603 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2604 if (first_unconverted) {
2605 *first_unconverted = s - 1;
2606 goto finish_and_return;
2607 }
2608 else {
2609 Safefree(converted_start);
2610 return (U8 *) original;
2611 }
2612 }
2613
7299a045
KW
2614 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2615 s++;
38af28cf 2616 }
7299a045
KW
2617 *d++ = c;
2618 }
170a1c22 2619
976c1b08
KW
2620 /* Here, converted the whole of the input */
2621 *is_utf8p = FALSE;
2622 if (first_unconverted) {
2623 *first_unconverted = NULL;
170a1c22 2624 }
976c1b08
KW
2625
2626 finish_and_return:
46a08a6f
KW
2627 *d = '\0';
2628 *lenp = d - converted_start;
976c1b08
KW
2629
2630 /* Trim unused space */
2631 Renew(converted_start, *lenp + 1, U8);
2632
2633 return converted_start;
f9a63242
JH
2634}
2635
2636/*
87cea99e 2637=for apidoc bytes_to_utf8
6940069f 2638
09af0336 2639Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2640UTF-8.
09af0336 2641Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2642reflect the new length in bytes. The caller is responsible for arranging for
2643the memory used by this string to get freed.
6940069f 2644
3bc0c78c 2645Upon successful return, the number of variants in the string can be computed by
23b37b12 2646having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2647after-call value of C<*lenp>.
2648
75200dff 2649A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2650
2651If you want to convert to UTF-8 from encodings other than
2652the native (Latin1 or EBCDIC),
a1433954 2653see L</sv_recode_to_utf8>().
c9ada85f 2654
497711e7 2655=cut
6940069f
GS
2656*/
2657
2658U8*
09af0336 2659Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2660{
09af0336 2661 const U8 * const send = s + (*lenp);
6940069f
GS
2662 U8 *d;
2663 U8 *dst;
7918f24d
NC
2664
2665 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2666 PERL_UNUSED_CONTEXT;
6940069f 2667
d4662719
KW
2668 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2669 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2670 dst = d;
2671
2672 while (s < send) {
55d09dc8
KW
2673 append_utf8_from_native_byte(*s, &d);
2674 s++;
6940069f 2675 }
2e11cf67 2676
6940069f 2677 *d = '\0';
09af0336 2678 *lenp = d-dst;
2e11cf67 2679
6940069f
GS
2680 return dst;
2681}
2682
a0ed51b3 2683/*
624504c5
KW
2684 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2685 * use utf16_to_utf8_reversed().
a0ed51b3 2686 *
624504c5
KW
2687 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2688 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2689 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2690 *
2691 * These functions don't check for overflow. The worst case is every code
2692 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2693 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2694 * destination must be pre-extended to 2 times the source length.
2695 *
2696 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2697
2698U8*
dea0fc0b 2699Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2700{
dea0fc0b
JH
2701 U8* pend;
2702 U8* dstart = d;
2703
7918f24d
NC
2704 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2705
dea0fc0b 2706 if (bytelen & 1)
56576a04
KW
2707 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2708 (UV)bytelen);
dea0fc0b
JH
2709
2710 pend = p + bytelen;
2711
a0ed51b3 2712 while (p < pend) {
dea0fc0b
JH
2713 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2714 p += 2;
2d1545e5 2715 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2716 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2717 continue;
2718 }
56d37426
KW
2719 if (uv <= MAX_UTF8_TWO_BYTE) {
2720 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2721 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2722 continue;
2723 }
ffd0a9d3 2724
46956fad
KW
2725#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2726#define LAST_HIGH_SURROGATE 0xDBFF
2727#define FIRST_LOW_SURROGATE 0xDC00
2728#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2729#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2730
2731 /* This assumes that most uses will be in the first Unicode plane, not
2732 * needing surrogates */
2733 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2734 && uv <= UNICODE_SURROGATE_LAST))
2735 {
2736 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2737 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2738 }
2739 else {
01ea242b 2740 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2741 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2742 || UNLIKELY(low > LAST_LOW_SURROGATE))
2743 {
01ea242b 2744 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2745 }
2746 p += 2;
46956fad 2747 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2748 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2749 }
a0ed51b3 2750 }
56d37426
KW
2751#ifdef EBCDIC
2752 d = uvoffuni_to_utf8_flags(d, uv, 0);
2753#else
ffd0a9d3 2754 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2755 *d++ = (U8)(( uv >> 12) | 0xe0);
2756 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2757 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2758 continue;
2759 }
2760 else {
eb160463
GS
2761 *d++ = (U8)(( uv >> 18) | 0xf0);
2762 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2763 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2764 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2765 continue;
2766 }
56d37426 2767#endif
a0ed51b3 2768 }
dea0fc0b 2769 *newlen = d - dstart;
a0ed51b3
LW
2770 return d;
2771}
2772
2773/* Note: this one is slightly destructive of the source. */
2774
2775U8*
dea0fc0b 2776Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2777{
2778 U8* s = (U8*)p;
d4c19fe8 2779 U8* const send = s + bytelen;
7918f24d
NC
2780
2781 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2782
e0ea5e2d 2783 if (bytelen & 1)
147e3846 2784 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2785 (UV)bytelen);
2786
a0ed51b3 2787 while (s < send) {
d4c19fe8 2788 const U8 tmp = s[0];
a0ed51b3
LW
2789 s[0] = s[1];
2790 s[1] = tmp;
2791 s += 2;
2792 }
dea0fc0b 2793 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2794}
2795
922e8cb4
KW
2796bool
2797Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2798{
1565c085 2799 dVAR;
dc31b55c 2800 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2801}
2802
5092f92a 2803bool
eba68aa0
KW
2804Perl__is_uni_perl_idcont(pTHX_ UV c)
2805{
1565c085 2806 dVAR;
c12658c9 2807 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2808}
2809
2810bool
f91dcd13
KW
2811Perl__is_uni_perl_idstart(pTHX_ UV c)
2812{
1565c085 2813 dVAR;
c12658c9 2814 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2815}
2816
3a4c58c9 2817UV
56576a04
KW
2818Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2819 const char S_or_s)
3a4c58c9
KW
2820{
2821 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2822 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2823 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2824 * either "SS" or "Ss". Which one to use is passed into the routine in
2825 * 'S_or_s' to avoid a test */
2826
2827 UV converted = toUPPER_LATIN1_MOD(c);
2828
2829 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2830
2831 assert(S_or_s == 'S' || S_or_s == 's');
2832
6f2d5cbc 2833 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2834 characters in this range */
3a4c58c9
KW
2835 *p = (U8) converted;
2836 *lenp = 1;
2837 return converted;
2838 }
2839
2840 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2841 * which it maps to one of them, so as to only have to have one check for
2842 * it in the main case */
2843 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2844 switch (c) {
2845 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2846 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2847 break;
2848 case MICRO_SIGN:
2849 converted = GREEK_CAPITAL_LETTER_MU;
2850 break;
79e064b9
KW
2851#if UNICODE_MAJOR_VERSION > 2 \
2852 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2853 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2854 case LATIN_SMALL_LETTER_SHARP_S:
2855 *(p)++ = 'S';
2856 *p = S_or_s;
2857 *lenp = 2;
2858 return 'S';
79e064b9 2859#endif
3a4c58c9 2860 default:
56576a04
KW
2861 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2862 " '%c' to map to '%c'",
2863 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2864 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2865 }
2866 }
2867
2868 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2869 *p = UTF8_TWO_BYTE_LO(converted);
2870 *lenp = 2;
2871
2872 return converted;
2873}
2874
fe63c520
KW
2875/* If compiled on an early Unicode version, there may not be auxiliary tables
2876 * */
2877#ifndef HAS_UC_AUX_TABLES
2878# define UC_AUX_TABLE_ptrs NULL
2879# define UC_AUX_TABLE_lengths NULL
2880#endif
2881#ifndef HAS_TC_AUX_TABLES
2882# define TC_AUX_TABLE_ptrs NULL
2883# define TC_AUX_TABLE_lengths NULL
2884#endif
2885#ifndef HAS_LC_AUX_TABLES
2886# define LC_AUX_TABLE_ptrs NULL
2887# define LC_AUX_TABLE_lengths NULL
2888#endif
2889#ifndef HAS_CF_AUX_TABLES
2890# define CF_AUX_TABLE_ptrs NULL
2891# define CF_AUX_TABLE_lengths NULL
2892#endif
2893#ifndef HAS_UC_AUX_TABLES
2894# define UC_AUX_TABLE_ptrs NULL
2895# define UC_AUX_TABLE_lengths NULL
2896#endif
2897
50bda2c3
KW
2898/* Call the function to convert a UTF-8 encoded character to the specified case.
2899 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2900 * 's' is a pointer to the first byte of the input character
2901 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2902 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2903 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2904 *
56576a04 2905 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2906 * 'd' */
56576a04 2907#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2908 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2909 Uppercase_Mapping_invmap, \
2910 UC_AUX_TABLE_ptrs, \
2911 UC_AUX_TABLE_lengths, \
2912 "uppercase")
56576a04 2913#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2914 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2915 Titlecase_Mapping_invmap, \
2916 TC_AUX_TABLE_ptrs, \
2917 TC_AUX_TABLE_lengths, \
2918 "titlecase")
56576a04 2919#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2920 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2921 Lowercase_Mapping_invmap, \
2922 LC_AUX_TABLE_ptrs, \
2923 LC_AUX_TABLE_lengths, \
2924 "lowercase")
2925
50bda2c3 2926
b9992569
KW
2927/* This additionally has the input parameter 'specials', which if non-zero will
2928 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2929 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2930#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2931 (specials) \
2932 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2933 Case_Folding_invmap, \
2934 CF_AUX_TABLE_ptrs, \
2935 CF_AUX_TABLE_lengths, \
2936 "foldcase") \
2937 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2938 Simple_Case_Folding_invmap, \
2939 NULL, NULL, \
2940 "foldcase")
c3fd2246 2941
84afefe6
JH
2942UV
2943Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2944{
a1433954
KW
2945 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2946 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2947 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2948 * the changed version may be longer than the original character.
2949 *
2950 * The ordinal of the first character of the changed version is returned
2951 * (but note, as explained above, that there may be more.) */
2952
1565c085 2953 dVAR;
7918f24d
NC
2954 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2955
3a4c58c9
KW
2956 if (c < 256) {
2957 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2958 }
2959
a13f1de4 2960 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2961}
2962
84afefe6
JH
2963UV
2964Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2965{
1565c085 2966 dVAR;
7918f24d
NC
2967 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2968
3a4c58c9
KW
2969 if (c < 256) {
2970 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2971 }
2972
a13f1de4 2973 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2974}
2975
afc16117 2976STATIC U8
eaf412bf 2977S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2978{
2979 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2980 * those, converting the result to UTF-8. Since the result is always just
a1433954 2981 * one character, we allow <p> to be NULL */
afc16117
KW
2982
2983 U8 converted = toLOWER_LATIN1(c);
2984
eaf412bf
KW
2985 PERL_UNUSED_ARG(dummy);
2986
afc16117 2987 if (p != NULL) {
6f2d5cbc 2988 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2989 *p = converted;
2990 *lenp = 1;
2991 }
2992 else {
430c9760
KW
2993 /* Result is known to always be < 256, so can use the EIGHT_BIT
2994 * macros */
2995 *p = UTF8_EIGHT_BIT_HI(converted);
2996 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2997 *lenp = 2;
2998 }
2999 }
3000 return converted;
3001}
3002
84afefe6
JH
3003UV
3004Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3005{
1565c085 3006 dVAR;
7918f24d
NC
3007 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3008
afc16117 3009 if (c < 256) {
eaf412bf 3010 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3011 }
3012
a13f1de4 3013 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3014}
3015
84afefe6 3016UV
7c0ab950 3017Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3018{
51910141 3019 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3020 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3021 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3022 *
3023 * Not to be used for locale folds
51910141 3024 */
f673fad4 3025
a1dde8de
KW
3026 UV converted;
3027
3028 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3029
1ca267a5
KW
3030 assert (! (flags & FOLD_FLAGS_LOCALE));
3031
659a7c2d 3032 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3033 converted = GREEK_SMALL_LETTER_MU;
3034 }
9b63e895
KW
3035#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3036 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3037 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3038 else if ( (flags & FOLD_FLAGS_FULL)
3039 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3040 {
1ca267a5
KW
3041 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3042 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3043 * under those circumstances. */
3044 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3045 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3046 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3047 p, *lenp, U8);
3048 return LATIN_SMALL_LETTER_LONG_S;
3049 }
3050 else {
4f489194
KW
3051 *(p)++ = 's';
3052 *p = 's';
3053 *lenp = 2;
3054 return 's';
1ca267a5 3055 }
a1dde8de 3056 }
9b63e895 3057#endif
a1dde8de
KW
3058 else { /* In this range the fold of all other characters is their lower
3059 case */
3060 converted = toLOWER_LATIN1(c);
3061 }
3062
6f2d5cbc 3063 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3064 *p = (U8) converted;
3065 *lenp = 1;
3066 }
3067 else {
3068 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3069 *p = UTF8_TWO_BYTE_LO(converted);
3070 *lenp = 2;
3071 }
3072
3073 return converted;
3074}
3075
3076UV
31f05a37 3077Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3078{
4b593389 3079
a0270393
KW
3080 /* Not currently externally documented, and subject to change
3081 * <flags> bits meanings:
3082 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3083 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3084 * locale are to be used.
a0270393
KW
3085 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3086 */
4b593389 3087
1565c085 3088 dVAR;
36bb2ab6 3089 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3090
780fcc9f 3091 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3092 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3093 * except for potentially warning */
8b7358b9 3094 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3095 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3096 flags &= ~FOLD_FLAGS_LOCALE;
3097 }
3098 else {
e7b7ac46 3099 goto needs_full_generality;
780fcc9f 3100 }
31f05a37
KW
3101 }
3102
a1dde8de 3103 if (c < 256) {
e7b7ac46 3104 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3105 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3106 }
3107
2f306ab9 3108 /* Here, above 255. If no special needs, just use the macro */
a0270393 3109 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3110 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3111 }
567b353c 3112 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3113 the special flags. */
3114 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3115
3116 needs_full_generality:
a0270393 3117 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3118 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3119 p, lenp, flags);
a0270393 3120 }
84afefe6
JH
3121}
3122
26483009 3123PERL_STATIC_INLINE bool
dd1a3ba7
KW
3124S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3125 SV* const invlist)
da8c1a98
KW
3126{
3127 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3128 * starts at <p>, and extending no further than <e - 1> is in the inversion
3129 * list <invlist>. */
da8c1a98 3130
b68ffe0c
KW
3131 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3132
dd1a3ba7 3133 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
da8c1a98 3134
b68ffe0c 3135 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3136 _force_out_malformed_utf8_message(p, e, 0, 1);
3137 NOT_REACHED; /* NOTREACHED */
3138 }
3139
eb1f4bb4 3140 assert(invlist);
b68ffe0c 3141 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3142}
3143
059703b0 3144#if 0 /* Not currently used, but may be needed in the future */
dd1a3ba7
KW
3145PERLVAR(I, seen_deprecated_macro, HV *)
3146
34aeb2e9
KW
3147STATIC void
3148S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3149 const char * const alternative,
3150 const bool use_locale,
3151 const char * const file,
3152 const unsigned line)
3153{
3154 const char * key;
3155
3156 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3157
3158 if (ckWARN_d(WARN_DEPRECATED)) {
3159
3160 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3161 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3162 if (! PL_seen_deprecated_macro) {
3163 PL_seen_deprecated_macro = newHV();
3164 }
3165 if (! hv_store(PL_seen_deprecated_macro, key,
3166 strlen(key), &PL_sv_undef, 0))
3167 {
3168 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3169 }
3170
c44e9413 3171 if (instr(file, "mathoms.c")) {
607313a1 3172 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3173 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3174 " will be removed. Avoid this message by"
3175 " converting to use %s().\n",
3176 file, line, name, alternative);
3177 }
3178 else {
34aeb2e9 3179 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3180 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3181 " require an additional parameter. Avoid this"
3182 " message by converting to use %s().\n",
3183 file, line, name, alternative);
607313a1 3184 }
34aeb2e9
KW
3185 }
3186 }
3187}
059703b0 3188#endif
922e8cb4
KW
3189
3190bool
dd1a3ba7 3191Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
da8c1a98 3192{
dd1a3ba7 3193 PERL_ARGS_ASSERT__IS_UTF8_FOO;
da8c1a98 3194
dd1a3ba7 3195 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
da8c1a98
KW
3196}
3197
3198bool
dd1a3ba7 3199Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
da8c1a98 3200{
dd1a3ba7 3201 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
da8c1a98 3202
dd1a3ba7 3203 return is_utf8_common(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3204}
3205
3206bool
dd1a3ba7 3207Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
c11ff943 3208{
dd1a3ba7 3209 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
7dbf68d2 3210
dd1a3ba7 3211 return is_utf8_common(p, e, PL_utf8_perl_idcont);
7dbf68d2
KW
3212}
3213
6a4a25f4 3214STATIC UV
30613bdc
KW
3215S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3216 U8* ustrp, STRLEN *lenp,
341bb5b7 3217 SV *invlist, const int * const invmap,
e39a4130 3218 const unsigned int * const * const aux_tables,
30613bdc
KW
3219 const U8 * const aux_table_lengths,
3220 const char * const normal)
b9992569 3221{
0134edef 3222 STRLEN len = 0;
7918f24d 3223
30613bdc
KW
3224 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3225 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3226 * to name the new case in any generated messages, as a fallback if the
3227 * operation being used is not available. The new case is given by the
3228 * data structures in the remaining arguments.
3229 *
3230 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3231 * entire changed case string, and the return value is the first code point
3232 * in that string */
3233
b9992569 3234 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3235
36eaa811
KW
3236 /* For code points that don't change case, we already know that the output
3237 * of this function is the unchanged input, so we can skip doing look-ups
3238 * for them. Unfortunately the case-changing code points are scattered
3239 * around. But there are some long consecutive ranges where there are no
3240 * case changing code points. By adding tests, we can eliminate the lookup
3241 * for all the ones in such ranges. This is currently done here only for
3242 * just a few cases where the scripts are in common use in modern commerce
3243 * (and scripts adjacent to those which can be included without additional
3244 * tests). */
3245
3246 if (uv1 >= 0x0590) {
3247 /* This keeps from needing further processing the code points most
3248 * likely to be used in the following non-cased scripts: Hebrew,
3249 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3250 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3251 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3252 if (uv1 < 0x10A0) {
3253 goto cases_to_self;
3254 }
3255
3256 /* The following largish code point ranges also don't have case
3257 * changes, but khw didn't think they warranted extra tests to speed
3258 * them up (which would slightly slow down everything else above them):
3259 * 1100..139F Hangul Jamo, Ethiopic
3260 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3261 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3262 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3263 * Combining Diacritical Marks Extended, Balinese,
3264 * Sundanese, Batak, Lepcha, Ol Chiki
3265 * 2000..206F General Punctuation
3266 */
3267
3268 if (uv1 >= 0x2D30) {
3269
3270 /* This keeps the from needing further processing the code points
3271 * most likely to be used in the following non-cased major scripts:
3272 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3273 *
3274 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3275 * event that Unicode eventually allocates the unused block as of
3276 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3277 * that the test suite will start having failures to alert you
3278 * should that happen) */
3279 if (uv1 < 0xA640) {
3280 goto cases_to_self;
3281 }
3282
3283 if (uv1 >= 0xAC00) {
3284 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3285 if (ckWARN_d(WARN_SURROGATE)) {
3286 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3287 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3288 "Operation \"%s\" returns its argument for"
3289 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3290 }
3291 goto cases_to_self;
3292 }
36eaa811
KW
3293
3294 /* AC00..FAFF Catches Hangul syllables and private use, plus
3295 * some others */
3296 if (uv1 < 0xFB00) {
3297 goto cases_to_self;
36eaa811
KW
3298 }
3299
5af9bc97 3300 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3301 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
d22ec717 3302 Perl_croak(aTHX_ cp_above_legal_max, uv1,
40606899 3303 MAX_LEGAL_CP);
5af9bc97
KW
3304 }
3305 if (ckWARN_d(WARN_NON_UNICODE)) {
3306 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3307 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3308 "Operation \"%s\" returns its argument for"
3309 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3310 }
3311 goto cases_to_self;
3312 }
3bfc1e70
KW
3313#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3314 if (UNLIKELY(uv1
3315 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3316 {
3317
56576a04
KW
3318 /* As of Unicode 10.0, this means we avoid swash creation
3319 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3320 goto cases_to_self;
3321 }
3322#endif
36eaa811
KW
3323 }
3324 }
9ae3ac1a 3325
36eaa811 3326 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3327 * be given. */
9ae3ac1a
KW
3328 }
3329
8946fcd9
KW
3330 {
3331 unsigned int i;
e39a4130 3332 const unsigned int * cp_list;
8946fcd9 3333 U8 * d;
69352d88
KW
3334
3335 /* 'index' is guaranteed to be non-negative, as this is an inversion
3336 * map that covers all possible inputs. See [perl #133365] */
8946fcd9
KW
3337 SSize_t index = _invlist_search(invlist, uv1);
3338 IV base = invmap[index];
0134edef 3339
30613bdc
KW
3340 /* The data structures are set up so that if 'base' is non-negative,
3341 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3342 if (base >= 0) {
3343 IV lc;
b08cf34e 3344
8946fcd9
KW
3345 if (base == 0) {
3346 goto cases_to_self;
4a8240a3 3347 }
4a8240a3 3348
30613bdc 3349 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3350 lc = base + uv1 - invlist_array(invlist)[index];
3351 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3352 return lc;
3353 }
1feea2c7 3354
30613bdc
KW
3355 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3356 * requires an auxiliary table look up. abs(base) gives the index into
3357 * a list of such tables which points to the proper aux table. And a
3358 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3359 cp_list = aux_tables[-base];
30613bdc
KW
3360
3361 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3362 d = ustrp;
3363 for (i = 0; i < aux_table_lengths[-base]; i++) {
3364 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3365 }
8946fcd9
KW
3366 *d = '\0';
3367 *lenp = d - ustrp;
3368
3369 return cp_list[0];
cbe07460
KW
3370 }
3371
3372 /* Here, there was no mapping defined, which means that the code point maps
3373 * to itself. Return the inputs */
e24dfe9c 3374 cases_to_self:
a13f1de4
KW
3375 if (p) {
3376 len = UTF8SKIP(p);
3377 if (p != ustrp) { /* Don't copy onto itself */
3378 Copy(p, ustrp, len, U8);
3379 }
3380 *lenp = len;
3381 }
3382 else {
3383 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3384 }
2a37f04d 3385
f4cd282c 3386 return uv1;
cbe07460 3387
a0ed51b3
LW
3388}
3389
b74fe592 3390Size_t
e39a4130
KW
3391Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3392 const unsigned int ** remaining_folds_to)
b74fe592
KW
3393{
3394 /* Returns the count of the number of code points that fold to the input
3395 * 'cp' (besides itself).
3396 *
3397 * If the return is 0, there is nothing else that folds to it, and
3398 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3399 *
3400 * If the return is 1, '*first_folds_to' is set to the single code point,
3401 * and '*remaining_folds_to' is set to NULL.
3402 *
3403 * Otherwise, '*first_folds_to' is set to a code point, and
3404 * '*remaining_fold_to' is set to an array that contains the others. The
3405 * length of this array is the returned count minus 1.
3406 *
3407 * The reason for this convolution is to avoid having to deal with
3408 * allocating and freeing memory. The lists are already constructed, so
3409 * the return can point to them, but single code points aren't, so would
3410 * need to be constructed if we didn't employ something like this API */
3411
1565c085 3412 dVAR;
69352d88
KW
3413 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3414 * that covers all possible inputs. See [perl #133365] */
b74fe592
KW
3415 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3416 int base = _Perl_IVCF_invmap[index];
3417
3418 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3419
3420 if (base == 0) { /* No fold */
3421 *first_folds_to = 0;
3422 *remaining_folds_to = NULL;
3423 return 0;
3424 }
3425
3426#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3427
3428 assert(base > 0);
3429
3430#else
3431
3432 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3433
3434 /* The data structure is set up so that the absolute value of 'base' is
3435 * an index into a table of pointers to arrays, with the array
3436 * corresponding to the index being the list of code points that fold
3437 * to 'cp', and the parallel array containing the length of the list
3438 * array */
3439 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3440 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3441 *first_folds_to
3442 */
3443 return IVCF_AUX_TABLE_lengths[-base];
3444 }
3445
3446#endif
3447
3448 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3449 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3450 *remaining_folds_to = NULL;
3451 return 1;
3452}
3453
051a06d4 3454STATIC UV
56576a04
KW
3455S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3456 U8* const ustrp, STRLEN *lenp)
051a06d4 3457{
4a4088c4 3458 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3459 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3460 * result contains a character that crosses the 255/256 boundary, disallow
3461 * the change, and return the original code point. See L<perlfunc/lc> for
3462 * why;
051a06d4 3463 *
a1433954
KW
3464 * p points to the original string whose case was changed; assumed
3465 * by this routine to be well-formed
051a06d4 3466 * result the code point of the first character in the changed-case string
56576a04
KW
3467 * ustrp points to the changed-case string (<result> represents its
3468 * first char)
051a06d4
KW
3469 * lenp points to the length of <ustrp> */
3470
3471 UV original; /* To store the first code point of <p> */
3472
3473 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3474
a4f12ed7 3475 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3476
3477 /* We know immediately if the first character in the string crosses the
5e45c680 3478 * boundary, so can skip testing */
051a06d4
KW
3479 if (result > 255) {
3480
3481 /* Look at every character in the result; if any cross the
3482 * boundary, the whole thing is disallowed */
3483 U8* s = ustrp + UTF8SKIP(ustrp);
3484 U8* e = ustrp + *lenp;
3485 while (s < e) {
a4f12ed7 3486 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3487 goto bad_crossing;
3488 }
3489 s += UTF8SKIP(s);
3490 }
3491
613abc6d
KW
3492 /* Here, no characters crossed, result is ok as-is, but we warn. */
3493 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3494 return result;
3495 }
3496
7b52d656 3497 bad_crossing:
051a06d4
KW
3498
3499 /* Failed, have to return the original */
4b88fb76 3500 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3501
3502 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3503 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3504 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3505 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3506 OP_DESC(PL_op),
ab0b796c
KW
3507 original,
3508 original);
051a06d4
KW
3509 Copy(p, ustrp, *lenp, char);
3510 return original;
3511}
3512
b257a28c
KW
3513STATIC UV
3514S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3515 U8 * ustrp, STRLEN *lenp)
3516{
3517 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3518 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3519 * Otherwise, it returns the first code point of the Turkic foldcased
3520 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3521 * contain *lenp bytes
3522 *
3523 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3524 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3525 * DOTLESS I */
3526
3527 PERL_ARGS_ASSERT_TURKIC_FC;
3528 assert(e > p);
3529
3530 if (UNLIKELY(*p == 'I')) {
3531 *lenp = 2;
3532 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3533 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3534 return LATIN_SMALL_LETTER_DOTLESS_I;
3535 }
3536
3537 if (UNLIKELY(memBEGINs(p, e - p,
3538 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3539 {
3540 *lenp = 1;
3541 *ustrp = 'i';
3542 return 'i';
3543 }
3544
3545 return 0;
3546}
3547
3548STATIC UV
3549S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3550 U8 * ustrp, STRLEN *lenp)
3551{
3552 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3553 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3554 * Otherwise, it returns the first code point of the Turkic lowercased
3555 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3556 * contain *lenp bytes */
3557
1565c085 3558 dVAR;
b257a28c
KW
3559 PERL_ARGS_ASSERT_TURKIC_LC;
3560 assert(e > p0);
3561
3562 /* A 'I' requires context as to what to do */
3563 if (UNLIKELY(*p0 == 'I')) {
3564 const U8 * p = p0 + 1;
3565
3566 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3567 * modified by a dot above lowercases to 'i' even in turkic locales. */
3568 while (p < e) {
3569 UV cp;
3570
3571 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3572 ustrp[0] = 'i';
3573 *lenp = 1;
3574 return 'i';
3575 }
3576
3577 /* For the dot above to modify the 'I', it must be part of a
3578 * combining sequence immediately following the 'I', and no other
3579 * modifier with a ccc of 230 may intervene */
3580 cp = utf8_to_uvchr_buf(p, e, NULL);
3581 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3582 break;
3583 }
3584
3585 /* Here the combining sequence continues */
3586 p += UTF8SKIP(p);
3587 }
3588 }
3589
3590 /* In all other cases the lc is the same as the fold */
3591 return turkic_fc(p0, e, ustrp, lenp);
3592}
3593
3594STATIC UV
3595S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3596 U8 * ustrp, STRLEN *lenp)
3597{
3598 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3599 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3600 * Otherwise, it returns the first code point of the Turkic upper or
3601 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3602 * ustrp will contain *lenp bytes
3603 *
3604 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3605 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3606 * DOTLESS I */
3607
3608 PERL_ARGS_ASSERT_TURKIC_UC;
3609 assert(e > p);
3610
3611 if (*p == 'i') {
3612 *lenp = 2;
3613 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3614 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3615 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3616 }
3617
3618 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3619 *lenp = 1;
3620 *ustrp = 'I';
3621 return 'I';
3622 }
3623
3624 return 0;
3625}
3626
eaf412bf
KW
3627/* The process for changing the case is essentially the same for the four case
3628 * change types, except there are complications for folding. Otherwise the
3629 * difference is only which case to change to. To make sure that they all do
3630 * the same thing, the bodies of the functions are extracted out into the
3631 * following two macros. The functions are written with the same variable
3632 * names, and these are known and used inside these macros. It would be
3633 * better, of course, to have inline functions to do it, but since different
3634 * macros are called, depending on which case is being changed to, this is not
3635 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3636 * function can start with the common start macro, then finish with its special
3637 * handling; while the other three cases can just use the common end macro.
3638 *
3639 * The algorithm is to use the proper (passed in) macro or function to change
3640 * the case for code points that are below 256. The macro is used if using
3641 * locale rules for the case change; the function if not. If the code point is
3642 * above 255, it is computed from the input UTF-8, and another macro is called
3643 * to do the conversion. If necessary, the output is converted to UTF-8. If
3644 * using a locale, we have to check that the change did not cross the 255/256
3645 * boundary, see check_locale_boundary_crossing() for further details.
3646 *
3647 * The macros are split with the correct case change for the below-256 case
3648 * stored into 'result', and in the middle of an else clause for the above-255
3649 * case. At that point in the 'else', 'result' is not the final result, but is
3650 * the input code point calculated from the UTF-8. The fold code needs to
3651 * realize all this and take it from there.
3652 *
b257a28c
KW
3653 * To deal with Turkic locales, the function specified by the parameter
3654 * 'turkic' is called when appropriate.
3655 *
eaf412bf
KW
3656 * If you read the two macros as sequential, it's easier to understand what's
3657 * going on. */
3658#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
b257a28c 3659 L1_func_extra_param, turkic) \
a239b1e2 3660 \
eaf412bf 3661 if (flags & (locale_flags)) { \
8b7358b9 3662 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
eaf412bf 3663 if (IN_UTF8_CTYPE_LOCALE) { \
b257a28c
KW
3664 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3665 UV ret = turkic(p, e, ustrp, lenp); \
3666 if (ret) return ret; \
3667 } \
3668 \
3669 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3670 * all */ \
eaf412bf
KW
3671 flags &= ~(locale_flags); \
3672 } \
eaf412bf
KW
3673 } \
3674 \
3675 if (UTF8_IS_INVARIANT(*p)) { \
3676 if (flags & (locale_flags)) { \
3677 result = LC_L1_change_macro(*p); \
3678 } \
3679 else { \
3680 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3681 } \
3682 } \
a239b1e2 3683 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
1a751160 3684 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
eaf412bf 3685 if (flags & (locale_flags)) { \
1a751160 3686 result = LC_L1_change_macro(c); \
eaf412bf
KW
3687 } \
3688 else { \
1a751160 3689 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
eaf412bf
KW
3690 } \
3691 } \
fa8ab374
KW
3692 else { /* malformed UTF-8 or ord above 255 */ \
3693 STRLEN len_result; \
fa8ab374
KW
3694 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3695 if (len_result == (STRLEN) -1) { \
059703b0 3696 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
fa8ab374 3697 }
eaf412bf
KW
3698
3699#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3700 result = change_macro(result, p, ustrp, lenp); \
3701 \
3702 if (flags & (locale_flags)) { \
3703 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3704 } \
3705 return result; \
3706 } \
3707 \
3708 /* Here, used locale rules. Convert back to UTF-8 */ \
3709 if (UTF8_IS_INVARIANT(result)) { \
3710 *ustrp = (U8) result; \
3711 *lenp = 1; \
3712 } \
3713 else { \
3714 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3715 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3716 *lenp = 2; \
3717 } \
3718 \
3719 return result;
3720