This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8_to_uvchr_buf() make behavior match docs
[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.
de69f3af
KW
148Instead, B<Almost all code should use L</uvchr_to_utf8> or
149L</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
efa9cd84 155For details, see the description for L</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
489L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
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
edc2c47a 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
961 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
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.
de69f3af 1192Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1193
9041c2e3 1194Bottom level UTF-8 decode routine.
de69f3af 1195Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1196which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1197C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1198the length, in bytes, of that character.
949cf498
KW
1199
1200The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1201well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1202causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1203is the next possible position in C<s> that could begin a non-malformed
1204character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1205is raised. Some UTF-8 input sequences may contain multiple malformations.
1206This function tries to find every possible one in each call, so multiple
56576a04 1207warnings can be raised for the same sequence.
949cf498
KW
1208
1209Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1210individual types of malformations, such as the sequence being overlong (that
1211is, when there is a shorter sequence that can express the same code point;
1212overlong sequences are expressly forbidden in the UTF-8 standard due to
1213potential security issues). Another malformation example is the first byte of
1214a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1215flags. Even if allowed, this function generally returns the Unicode
1216REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1217F<utf8.h> to override this behavior for the overlong malformations, but don't
1218do that except for very specialized purposes.
949cf498 1219
796b6530 1220The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1221flags) malformation is found. If this flag is set, the routine assumes that
1222the caller will raise a warning, and this function will silently just set
d088425d
KW
1223C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1224
75200dff 1225Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1226character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1227in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1228be set to 1. To disambiguate, upon a zero return, see if the first byte of
1229C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1230error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1231
1232Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1233Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1234By default these are considered regular code points, but certain situations
ecc1615f
KW
1235warrant special handling for them, which can be specified using the C<flags>
1236parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1237three classes are treated as malformations and handled as such. The flags
1238C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1239C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1240disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1241restricts the allowed inputs to the strict UTF-8 traditionally defined by
1242Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1243definition given by
1244L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1245The difference between traditional strictness and C9 strictness is that the
1246latter does not forbid non-character code points. (They are still discouraged,
1247however.) For more discussion see L<perlunicode/Noncharacter code points>.
1248
1249The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1250C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1251C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1252raised for their respective categories, but otherwise the code points are
1253considered valid (not malformations). To get a category to both be treated as
1254a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1255(But note that warnings are not raised if lexically disabled nor if
796b6530 1256C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1257
57ff5f59
KW
1258Extremely high code points were never specified in any standard, and require an
1259extension to UTF-8 to express, which Perl does. It is likely that programs
1260written in something other than Perl would not be able to read files that
1261contain these; nor would Perl understand files written by something that uses a
1262different extension. For these reasons, there is a separate set of flags that
1263can warn and/or disallow these extremely high code points, even if other
1264above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1265C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1266L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1267above-Unicode code points, including these, as malformations.
1268(Note that the Unicode standard considers anything above 0x10FFFF to be
1269illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1270(2**31 -1))
1271
1272A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1273retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1274C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1275C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1276can apply to code points that actually do fit in 31 bits. This happens on
1277EBCDIC platforms, and sometimes when the L<overlong
1278malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1279describe the situation in all cases.
1280
ab8e6d41 1281
949cf498
KW
1282All other code points corresponding to Unicode characters, including private
1283use and those yet to be assigned, are never considered malformed and never
1284warn.
67e989fb 1285
37607a96 1286=cut
f9380377
KW
1287
1288Also implemented as a macro in utf8.h
1289*/
1290
1291UV
e6a4ffc3
KW
1292Perl_utf8n_to_uvchr(const U8 *s,
1293 STRLEN curlen,
1294 STRLEN *retlen,
1295 const U32 flags)
f9380377
KW
1296{
1297 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1298
1299 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1300}
1301
1302/*
1303
1304=for apidoc utf8n_to_uvchr_error
1305
1306THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1307Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1308
1309This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1310are when an error is found. If you also need to know the generated warning
1311messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1312
1313It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1314all the others, C<errors>. If this parameter is 0, this function behaves
1315identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1316to a C<U32> variable, which this function sets to indicate any errors found.
1317Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1318C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1319of these bits will be set if a malformation is found, even if the input
7a65503b 1320C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1321exceptions are noted:
1322
1323=over 4
1324
57ff5f59 1325=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1326
57ff5f59
KW
1327The input sequence is not standard UTF-8, but a Perl extension. This bit is
1328set only if the input C<flags> parameter contains either the
1329C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1330
1331Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1332and so some extension must be used to express them. Perl uses a natural
1333extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1334extension to represent even higher ones, so that any code point that fits in a
133564-bit word can be represented. Text using these extensions is not likely to
1336be portable to non-Perl code. We lump both of these extensions together and
1337refer to them as Perl extended UTF-8. There exist other extensions that people
1338have invented, incompatible with Perl's.
1339
1340On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1341extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1342than on ASCII. Prior to that, code points 2**31 and higher were simply
1343unrepresentable, and a different, incompatible method was used to represent
1344code points between 2**30 and 2**31 - 1.
1345
1346On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1347Perl extended UTF-8 is used.
1348
1349In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1350may use for backward compatibility. That name is misleading, as this flag may
1351be set when the code point actually does fit in 31 bits. This happens on
1352EBCDIC platforms, and sometimes when the L<overlong
1353malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1354describes the situation in all cases.
f9380377
KW
1355
1356=item C<UTF8_GOT_CONTINUATION>
1357
1358The input sequence was malformed in that the first byte was a a UTF-8
1359continuation byte.
1360
1361=item C<UTF8_GOT_EMPTY>
1362
1363The input C<curlen> parameter was 0.
1364
1365=item C<UTF8_GOT_LONG>
1366
1367The input sequence was malformed in that there is some other sequence that
1368evaluates to the same code point, but that sequence is shorter than this one.
1369
fecaf136
KW
1370Until Unicode 3.1, it was legal for programs to accept this malformation, but
1371it was discovered that this created security issues.
1372
f9380377
KW
1373=item C<UTF8_GOT_NONCHAR>
1374
1375The code point represented by the input UTF-8 sequence is for a Unicode
1376non-character code point.
1377This bit is set only if the input C<flags> parameter contains either the
1378C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1379
1380=item C<UTF8_GOT_NON_CONTINUATION>
1381
1382The input sequence was malformed in that a non-continuation type byte was found
00d976bb
KW
1383in a position where only a continuation type one should be. See also
1384L</C<UTF8_GOT_SHORT>>.
f9380377
KW
1385
1386=item C<UTF8_GOT_OVERFLOW>
1387
1388The input sequence was malformed in that it is for a code point that is not
d22ec717 1389representable in the number of bits available in an IV on the current platform.
f9380377
KW
1390
1391=item C<UTF8_GOT_SHORT>
1392
1393The input sequence was malformed in that C<curlen> is smaller than required for
1394a complete sequence. In other words, the input is for a partial character
1395sequence.
1396
00d976bb
KW
1397
1398C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1399sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1400that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1401sequence was looked at. If no other flags are present, it means that the
1402sequence was valid as far as it went. Depending on the application, this could
1403mean one of three things:
1404
1405=over
1406
1407=item *
1408
1409The C<curlen> length parameter passed in was too small, and the function was
1410prevented from examining all the necessary bytes.
1411
1412=item *
1413
1414The buffer being looked at is based on reading data, and the data received so
1415far stopped in the middle of a character, so that the next read will
1416read the remainder of this character. (It is up to the caller to deal with the
1417split bytes somehow.)
1418
1419=item *
1420
1421This is a real error, and the partial sequence is all we're going to get.
1422
1423=back
1424
f9380377
KW
1425=item C<UTF8_GOT_SUPER>
1426
1427The input sequence was malformed in that it is for a non-Unicode code point;
1428that is, one above the legal Unicode maximum.
1429This bit is set only if the input C<flags> parameter contains either the
1430C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1431
1432=item C<UTF8_GOT_SURROGATE>
1433
1434The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1435code point.
1436This bit is set only if the input C<flags> parameter contains either the
1437C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1438
1439=back
1440
133551d8
KW
1441To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1442flag to suppress any warnings, and then examine the C<*errors> return.
1443
f9380377 1444=cut
37657a5b
KW
1445
1446Also implemented as a macro in utf8.h
37607a96 1447*/
67e989fb 1448
a0ed51b3 1449UV
e6a4ffc3 1450Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1451 STRLEN curlen,
1452 STRLEN *retlen,
1453 const U32 flags,
1454 U32 * errors)
1455{
1456 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1457
1458 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1459}
1460
1461/*
1462
1463=for apidoc utf8n_to_uvchr_msgs
1464
1465THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1466Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1467
1468This function is for code that needs to know what the precise malformation(s)
1469are when an error is found, and wants the corresponding warning and/or error
1470messages to be returned to the caller rather than be displayed. All messages
1471that would have been displayed if all lexcial warnings are enabled will be
1472returned.
1473
1474It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1475placed after all the others, C<msgs>. If this parameter is 0, this function
1476behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1477be a pointer to an C<AV *> variable, in which this function creates a new AV to
1478contain any appropriate messages. The elements of the array are ordered so
1479that the first message that would have been displayed is in the 0th element,
1480and so on. Each element is a hash with three key-value pairs, as follows:
1481
1482=over 4
1483
1484=item C<text>
1485
1486The text of the message as a C<SVpv>.
1487
1488=item C<warn_categories>
1489
1490The warning category (or categories) packed into a C<SVuv>.
1491
1492=item C<flag>
1493
1494A single flag bit associated with this message, in a C<SVuv>.
1495The bit corresponds to some bit in the C<*errors> return value,
1496such as C<UTF8_GOT_LONG>.
1497
1498=back
1499
1500It's important to note that specifying this parameter as non-null will cause
1501any warnings this function would otherwise generate to be suppressed, and
1502instead be placed in C<*msgs>. The caller can check the lexical warnings state
1503(or not) when choosing what to do with the returned messages.
1504
1505If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1506no AV is created.
1507
1508The caller, of course, is responsible for freeing any returned AV.
1509
1510=cut
1511*/
1512
1513UV
e6a4ffc3 1514Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1515 STRLEN curlen,
1516 STRLEN *retlen,
1517 const U32 flags,
1518 U32 * errors,
1519 AV ** msgs)
a0ed51b3 1520{
d4c19fe8 1521 const U8 * const s0 = s;
2b9519f0 1522 const U8 * send = s0 + curlen;
5af9f822
KW
1523 U32 possible_problems; /* A bit is set here for each potential problem
1524 found as we go along */
1525 UV uv;
1526 STRLEN expectlen; /* How long should this sequence be? */
1527 STRLEN avail_len; /* When input is too short, gives what that is */
1528 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1529 gets set and discarded */
a0dbb045 1530
2b5e7bc2
KW
1531 /* The below are used only if there is both an overlong malformation and a
1532 * too short one. Otherwise the first two are set to 's0' and 'send', and
1533 * the third not used at all */
5af9f822 1534 U8 * adjusted_s0;
e9f2c446
KW
1535 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1536 routine; see [perl #130921] */
5af9f822 1537 UV uv_so_far;
e6a4ffc3 1538 dTHX;
5af9f822 1539
e6a4ffc3 1540 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1541
1542 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1543 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1544 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1545 * final case. */
1546
1547#ifndef EBCDIC
1548
1549 /* Each of the affected Hanguls starts with \xED */
1550
1551 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1552 if (retlen) {
1553 *retlen = 3;
1554 }
1555 if (errors) {
1556 *errors = 0;
1557 }
1558 if (msgs) {
1559 *msgs = NULL;
1560 }
1561
1562 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1563 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1564 | (s0[2] & UTF_CONTINUATION_MASK);
1565 }
1566
1567#endif
1568
1569 /* In conjunction with the exhaustive tests that can be enabled in
1570 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1571 * what it is intended to do, and that no flaws in it are masked by
1572 * dropping down and executing the code below
1573 assert(! isUTF8_CHAR(s0, send)
1574 || UTF8_IS_SURROGATE(s0, send)
1575 || UTF8_IS_SUPER(s0, send)
1576 || UTF8_IS_NONCHAR(s0,send));
1577 */
1578
1579 s = s0;
1580 uv = *s0;
1581 possible_problems = 0;
1582 expectlen = 0;
1583 avail_len = 0;
1584 discard_errors = 0;
1585 adjusted_s0 = (U8 *) s0;
1586 uv_so_far = 0;
1587
f9380377
KW
1588 if (errors) {
1589 *errors = 0;
1590 }
1591 else {
1592 errors = &discard_errors;
1593 }
a0dbb045 1594
eb83ed87
KW
1595 /* The order of malformation tests here is important. We should consume as
1596 * few bytes as possible in order to not skip any valid character. This is
1597 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1598 * http://unicode.org/reports/tr36 for more discussion as to why. For
1599 * example, once we've done a UTF8SKIP, we can tell the expected number of
1600 * bytes, and could fail right off the bat if the input parameters indicate
1601 * that there are too few available. But it could be that just that first
1602 * byte is garbled, and the intended character occupies fewer bytes. If we
1603 * blindly assumed that the first byte is correct, and skipped based on
1604 * that number, we could skip over a valid input character. So instead, we
1605 * always examine the sequence byte-by-byte.
1606 *
1607 * We also should not consume too few bytes, otherwise someone could inject
1608 * things. For example, an input could be deliberately designed to
1609 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1610 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1611 * which is actually part of of the overflowing sequence), that could look
1612 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1613 * sequence and process the rest, inappropriately.
1614 *
1615 * Some possible input sequences are malformed in more than one way. This
1616 * function goes to lengths to try to find all of them. This is necessary
1617 * for correctness, as the inputs may allow one malformation but not
1618 * another, and if we abandon searching for others after finding the
1619 * allowed one, we could allow in something that shouldn't have been.
1620 */
eb83ed87 1621
b5b9af04 1622 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1623 possible_problems |= UTF8_GOT_EMPTY;
1624 curlen = 0;
5a48568d 1625 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1626 goto ready_to_handle_errors;
0c443dc2
JH
1627 }
1628
eb83ed87
KW
1629 expectlen = UTF8SKIP(s);
1630
1631 /* A well-formed UTF-8 character, as the vast majority of calls to this
1632 * function will be for, has this expected length. For efficiency, set
1633 * things up here to return it. It will be overriden only in those rare
1634 * cases where a malformation is found */
1635 if (retlen) {
1636 *retlen = expectlen;
1637 }
1638
eb83ed87 1639 /* A continuation character can't start a valid sequence */
b5b9af04 1640 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1641 possible_problems |= UTF8_GOT_CONTINUATION;
1642 curlen = 1;
1643 uv = UNICODE_REPLACEMENT;
1644 goto ready_to_handle_errors;
ba210ebe 1645 }
9041c2e3 1646
dcd27b3c 1647 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1648 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1649 * because it excludes start bytes like \xC0 that always lead to
1650 * overlongs.) */
dcd27b3c 1651
534752c1
KW
1652 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1653 * that indicate the number of bytes in the character's whole UTF-8
1654 * sequence, leaving just the bits that are part of the value. */
1655 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1656
e308b348
KW
1657 /* Setup the loop end point, making sure to not look past the end of the
1658 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1659 if (UNLIKELY(curlen < expectlen)) {
1660 possible_problems |= UTF8_GOT_SHORT;
1661 avail_len = curlen;
e308b348
KW
1662 }
1663 else {
2b9519f0 1664 send = (U8*) s0 + expectlen;
e308b348 1665 }
e308b348 1666
eb83ed87 1667 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1668 * accumulating each into the working value as we go. */
eb83ed87 1669 for (s = s0 + 1; s < send; s++) {
b5b9af04 1670 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1671 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1672 continue;
1673 }
1674
1675 /* Here, found a non-continuation before processing all expected bytes.
1676 * This byte indicates the beginning of a new character, so quit, even
1677 * if allowing this malformation. */
2b5e7bc2 1678 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1679 break;
eb83ed87
KW
1680 } /* End of loop through the character's bytes */
1681
1682 /* Save how many bytes were actually in the character */
1683 curlen = s - s0;
1684
2b5e7bc2
KW
1685 /* Note that there are two types of too-short malformation. One is when
1686 * there is actual wrong data before the normal termination of the
1687 * sequence. The other is that the sequence wasn't complete before the end
1688 * of the data we are allowed to look at, based on the input 'curlen'.
1689 * This means that we were passed data for a partial character, but it is
1690 * valid as far as we saw. The other is definitely invalid. This
1691 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1692 * separate.
1693 *
1694 * A convenience macro that matches either of the too-short conditions. */
1695# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1696
1697 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1698 uv_so_far = uv;
1699 uv = UNICODE_REPLACEMENT;
1700 }
2b5e7bc2 1701
08e73697
KW
1702 /* Check for overflow. The algorithm requires us to not look past the end
1703 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1704 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1705 1 /* Do consider overlongs */
1706 )))
1707 {
2b5e7bc2
KW
1708 possible_problems |= UTF8_GOT_OVERFLOW;
1709 uv = UNICODE_REPLACEMENT;
eb83ed87 1710 }
eb83ed87 1711
2b5e7bc2
KW
1712 /* Check for overlong. If no problems so far, 'uv' is the correct code
1713 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1714 * we must look at the UTF-8 byte sequence itself to see if it is for an
1715 * overlong */
1716 if ( ( LIKELY(! possible_problems)
1717 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1718 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1719 && ( UNLIKELY(! UTF8_IS_START(*s0))
1720 || ( curlen > 1
d6be65ae 1721 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1722 s - s0))))))
2f8f112e 1723 {
2b5e7bc2
KW
1724 possible_problems |= UTF8_GOT_LONG;
1725
abc28b54 1726 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1727
abc28b54
KW
1728 /* The calculation in the 'true' branch of this 'if'
1729 * below won't work if overflows, and isn't needed
1730 * anyway. Further below we handle all overflow
1731 * cases */
1732 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1733 {
2b5e7bc2
KW
1734 UV min_uv = uv_so_far;
1735 STRLEN i;
1736
1737 /* Here, the input is both overlong and is missing some trailing
1738 * bytes. There is no single code point it could be for, but there
1739 * may be enough information present to determine if what we have
1740 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1741 * The code further below has the intelligence to determine this,
1742 * but just for non-overlong UTF-8 sequences. What we do here is
1743 * calculate the smallest code point the input could represent if
1744 * there were no too short malformation. Then we compute and save
1745 * the UTF-8 for that, which is what the code below looks at
1746 * instead of the raw input. It turns out that the smallest such
1747 * code point is all we need. */
2b5e7bc2
KW
1748 for (i = curlen; i < expectlen; i++) {
1749 min_uv = UTF8_ACCUMULATE(min_uv,
1750 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1751 }
1752
e9f2c446 1753 adjusted_s0 = temp_char_buf;
57ff5f59 1754 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1755 }
eb83ed87
KW
1756 }
1757
56576a04
KW
1758 /* Here, we have found all the possible problems, except for when the input
1759 * is for a problematic code point not allowed by the input parameters. */
1760
06188866
KW
1761 /* uv is valid for overlongs */
1762 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1763
1764 /* isn't problematic if < this */
1765 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1766 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1767
1768 /* if overflow, we know without looking further
1769 * precisely which of the problematic types it is,
1770 * and we deal with those in the overflow handling
1771 * code */
1772 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1773 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1774 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1775 && ((flags & ( UTF8_DISALLOW_NONCHAR
1776 |UTF8_DISALLOW_SURROGATE
1777 |UTF8_DISALLOW_SUPER
d044b7a7 1778 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1779 |UTF8_WARN_NONCHAR
1780 |UTF8_WARN_SURROGATE
1781 |UTF8_WARN_SUPER
d22ec717 1782 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1783 {
2b5e7bc2
KW
1784 /* If there were no malformations, or the only malformation is an
1785 * overlong, 'uv' is valid */
1786 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1787 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1788 possible_problems |= UTF8_GOT_SURROGATE;
1789 }
1790 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1791 possible_problems |= UTF8_GOT_SUPER;
1792 }
1793 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1794 possible_problems |= UTF8_GOT_NONCHAR;
1795 }
1796 }
1797 else { /* Otherwise, need to look at the source UTF-8, possibly
1798 adjusted to be non-overlong */
1799
1800 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1801 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1802 {
2b5e7bc2
KW
1803 possible_problems |= UTF8_GOT_SUPER;
1804 }
1805 else if (curlen > 1) {
1806 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1807 NATIVE_UTF8_TO_I8(*adjusted_s0),
1808 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1809 {
2b5e7bc2 1810 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1811 }
2b5e7bc2
KW
1812 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1813 NATIVE_UTF8_TO_I8(*adjusted_s0),
1814 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1815 {
1816 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1817 }
1818 }
c0236afe 1819
2b5e7bc2
KW
1820 /* We need a complete well-formed UTF-8 character to discern
1821 * non-characters, so can't look for them here */
1822 }
1823 }
949cf498 1824
2b5e7bc2
KW
1825 ready_to_handle_errors:
1826
1827 /* At this point:
1828 * curlen contains the number of bytes in the sequence that
1829 * this call should advance the input by.
e308b348
KW
1830 * avail_len gives the available number of bytes passed in, but
1831 * only if this is less than the expected number of
1832 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1833 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1834 * is set in it for each potential problem found.
1835 * uv contains the code point the input sequence
1836 * represents; or if there is a problem that prevents
1837 * a well-defined value from being computed, it is
1838 * some subsitute value, typically the REPLACEMENT
1839 * CHARACTER.
1840 * s0 points to the first byte of the character
56576a04
KW
1841 * s points to just after were we left off processing
1842 * the character
1843 * send points to just after where that character should
1844 * end, based on how many bytes the start byte tells
1845 * us should be in it, but no further than s0 +
1846 * avail_len
2b5e7bc2 1847 */
eb83ed87 1848
2b5e7bc2
KW
1849 if (UNLIKELY(possible_problems)) {
1850 bool disallowed = FALSE;
1851 const U32 orig_problems = possible_problems;
1852
37657a5b
KW
1853 if (msgs) {
1854 *msgs = NULL;
1855 }
1856
2b5e7bc2
KW
1857 while (possible_problems) { /* Handle each possible problem */
1858 UV pack_warn = 0;
1859 char * message = NULL;
37657a5b 1860 U32 this_flag_bit = 0;
2b5e7bc2
KW
1861
1862 /* Each 'if' clause handles one problem. They are ordered so that
1863 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1864 * ones; this is kinda in decreasing severity order. But the
1865 * overlong must come last, as it changes 'uv' looked at by the
1866 * others */
2b5e7bc2
KW
1867 if (possible_problems & UTF8_GOT_OVERFLOW) {
1868
56576a04
KW
1869 /* Overflow means also got a super and are using Perl's
1870 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1871 possible_problems
d044b7a7 1872 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1873 *errors |= UTF8_GOT_OVERFLOW;
1874
1875 /* But the API says we flag all errors found */
1876 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1877 *errors |= UTF8_GOT_SUPER;
1878 }
ddb65933 1879 if (flags
d044b7a7 1880 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1881 {
d044b7a7 1882 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1883 }
2b5e7bc2 1884
d60baaa7 1885 /* Disallow if any of the three categories say to */
56576a04 1886 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1887 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1888 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1889 {
1890 disallowed = TRUE;
1891 }
1892
d22ec717
KW
1893 /* Likewise, warn if any say to */
1894 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1895 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1896 {
2b5e7bc2 1897
ddb65933
KW
1898 /* The warnings code explicitly says it doesn't handle the
1899 * case of packWARN2 and two categories which have
1900 * parent-child relationship. Even if it works now to
1901 * raise the warning if either is enabled, it wouldn't
1902 * necessarily do so in the future. We output (only) the
56576a04 1903 * most dire warning */
ddb65933 1904 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1905 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1906 pack_warn = packWARN(WARN_UTF8);
1907 }
37657a5b 1908 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1909 pack_warn = packWARN(WARN_NON_UNICODE);
1910 }
1911 if (pack_warn) {
1912 message = Perl_form(aTHX_ "%s: %s (overflows)",
1913 malformed_text,
05b9033b 1914 _byte_dump_string(s0, curlen, 0));
37657a5b 1915 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1916 }
2b5e7bc2
KW
1917 }
1918 }
1919 }
1920 else if (possible_problems & UTF8_GOT_EMPTY) {
1921 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1922 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1923
1924 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1925
1926 /* This so-called malformation is now treated as a bug in
1927 * the caller. If you have nothing to decode, skip calling
1928 * this function */
1929 assert(0);
1930
2b5e7bc2 1931 disallowed = TRUE;
37657a5b
KW
1932 if ( (msgs
1933 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1934 {
2b5e7bc2
KW
1935 pack_warn = packWARN(WARN_UTF8);
1936 message = Perl_form(aTHX_ "%s (empty string)",
1937 malformed_text);
37657a5b 1938 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1939 }
1940 }
1941 }
1942 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1943 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1944 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1945
1946 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1947 disallowed = TRUE;
37657a5b
KW
1948 if (( msgs
1949 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1950 {
2b5e7bc2
KW
1951 pack_warn = packWARN(WARN_UTF8);
1952 message = Perl_form(aTHX_
1953 "%s: %s (unexpected continuation byte 0x%02x,"
1954 " with no preceding start byte)",
1955 malformed_text,
7e2f38b2 1956 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1957 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1958 }
1959 }
1960 }
2b5e7bc2
KW
1961 else if (possible_problems & UTF8_GOT_SHORT) {
1962 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1963 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1964
1965 if (! (flags & UTF8_ALLOW_SHORT)) {
1966 disallowed = TRUE;
37657a5b
KW
1967 if (( msgs
1968 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1969 {
2b5e7bc2
KW
1970 pack_warn = packWARN(WARN_UTF8);
1971 message = Perl_form(aTHX_
56576a04
KW
1972 "%s: %s (too short; %d byte%s available, need %d)",
1973 malformed_text,
1974 _byte_dump_string(s0, send - s0, 0),
1975 (int)avail_len,
1976 avail_len == 1 ? "" : "s",
1977 (int)expectlen);
37657a5b 1978 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1979 }
1980 }
ba210ebe 1981
2b5e7bc2 1982 }
e308b348
KW
1983 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1984 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1985 *errors |= UTF8_GOT_NON_CONTINUATION;
1986
1987 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1988 disallowed = TRUE;
37657a5b
KW
1989 if (( msgs
1990 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1991 {
99a765e9
KW
1992
1993 /* If we don't know for sure that the input length is
1994 * valid, avoid as much as possible reading past the
1995 * end of the buffer */
1996 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1997 ? s - s0
1998 : send - s0;
e308b348
KW
1999 pack_warn = packWARN(WARN_UTF8);
2000 message = Perl_form(aTHX_ "%s",
2001 unexpected_non_continuation_text(s0,
99a765e9 2002 printlen,
e308b348
KW
2003 s - s0,
2004 (int) expectlen));
37657a5b 2005 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2006 }
2007 }
2008 }
2b5e7bc2
KW
2009 else if (possible_problems & UTF8_GOT_SURROGATE) {
2010 possible_problems &= ~UTF8_GOT_SURROGATE;
2011
f9380377
KW
2012 if (flags & UTF8_WARN_SURROGATE) {
2013 *errors |= UTF8_GOT_SURROGATE;
2014
2015 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2016 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2017 {
2b5e7bc2
KW
2018 pack_warn = packWARN(WARN_SURROGATE);
2019
2020 /* These are the only errors that can occur with a
2021 * surrogate when the 'uv' isn't valid */
2022 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2023 message = Perl_form(aTHX_
2024 "UTF-16 surrogate (any UTF-8 sequence that"
2025 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2026 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2027 }
2028 else {
c94c2f39 2029 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2030 }
37657a5b 2031 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2032 }
2b5e7bc2 2033 }
ba210ebe 2034
2b5e7bc2
KW
2035 if (flags & UTF8_DISALLOW_SURROGATE) {
2036 disallowed = TRUE;
f9380377 2037 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2038 }
2039 }
2040 else if (possible_problems & UTF8_GOT_SUPER) {
2041 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2042
f9380377
KW
2043 if (flags & UTF8_WARN_SUPER) {
2044 *errors |= UTF8_GOT_SUPER;
2045
2046 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2047 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2048 {
2b5e7bc2
KW
2049 pack_warn = packWARN(WARN_NON_UNICODE);
2050
2051 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2052 message = Perl_form(aTHX_
2053 "Any UTF-8 sequence that starts with"
2054 " \"%s\" is for a non-Unicode code point,"
2055 " may not be portable",
7e2f38b2 2056 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2057 }
2058 else {
c94c2f39 2059 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2060 }
37657a5b 2061 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2062 }
2b5e7bc2 2063 }
ba210ebe 2064
57ff5f59
KW
2065 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2066 * and before possibly bailing out, so that the more dire
2067 * warning will override the regular one. */
2068 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2069 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2070 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2071 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2072 {
db0f09e6 2073 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2074
57ff5f59
KW
2075 /* If it is an overlong that evaluates to a code point
2076 * that doesn't have to use the Perl extended UTF-8, it
2077 * still used it, and so we output a message that
2078 * doesn't refer to the code point. The same is true
2079 * if there was a SHORT malformation where the code
2080 * point is not valid. In that case, 'uv' will have
2081 * been set to the REPLACEMENT CHAR, and the message
2082 * below without the code point in it will be selected
2083 * */
2084 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2085 message = Perl_form(aTHX_
57ff5f59 2086 perl_extended_cp_format, uv);
2b5e7bc2
KW
2087 }
2088 else {
2089 message = Perl_form(aTHX_
57ff5f59
KW
2090 "Any UTF-8 sequence that starts with"
2091 " \"%s\" is a Perl extension, and"
2092 " so is not portable",
2093 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2094 }
37657a5b 2095 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2096 }
2097
d044b7a7
KW
2098 if (flags & ( UTF8_WARN_PERL_EXTENDED
2099 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2100 {
d044b7a7 2101 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2102
d044b7a7 2103 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2104 disallowed = TRUE;
2105 }
2b5e7bc2
KW
2106 }
2107 }
eb83ed87 2108
2b5e7bc2 2109 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2110 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2111 disallowed = TRUE;
2112 }
2b5e7bc2
KW
2113 }
2114 else if (possible_problems & UTF8_GOT_NONCHAR) {
2115 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2116
f9380377
KW
2117 if (flags & UTF8_WARN_NONCHAR) {
2118 *errors |= UTF8_GOT_NONCHAR;
2119
2120 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2121 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2122 {
2b5e7bc2
KW
2123 /* The code above should have guaranteed that we don't
2124 * get here with errors other than overlong */
2125 assert (! (orig_problems
2126 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2127
2128 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2129 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2130 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2131 }
2b5e7bc2 2132 }
5b311467 2133
2b5e7bc2
KW
2134 if (flags & UTF8_DISALLOW_NONCHAR) {
2135 disallowed = TRUE;
f9380377 2136 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2137 }
6c64cd9d
KW
2138 }
2139 else if (possible_problems & UTF8_GOT_LONG) {
2140 possible_problems &= ~UTF8_GOT_LONG;
2141 *errors |= UTF8_GOT_LONG;
2142
2143 if (flags & UTF8_ALLOW_LONG) {
2144
2145 /* We don't allow the actual overlong value, unless the
2146 * special extra bit is also set */
2147 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2148 & ~UTF8_ALLOW_LONG)))
2149 {
2150 uv = UNICODE_REPLACEMENT;
2151 }
2152 }
2153 else {
2154 disallowed = TRUE;
2155
37657a5b
KW
2156 if (( msgs
2157 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2158 {
6c64cd9d
KW
2159 pack_warn = packWARN(WARN_UTF8);
2160
2161 /* These error types cause 'uv' to be something that
2162 * isn't what was intended, so can't use it in the
2163 * message. The other error types either can't
2164 * generate an overlong, or else the 'uv' is valid */
2165 if (orig_problems &
2166 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2167 {
2168 message = Perl_form(aTHX_
2169 "%s: %s (any UTF-8 sequence that starts"
2170 " with \"%s\" is overlong which can and"
2171 " should be represented with a"
2172 " different, shorter sequence)",
2173 malformed_text,
2174 _byte_dump_string(s0, send - s0, 0),
2175 _byte_dump_string(s0, curlen, 0));
2176 }
2177 else {
2178 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2179 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2180 uv, 0);
d819dc50
KW
2181 /* Don't use U+ for non-Unicode code points, which
2182 * includes those in the Latin1 range */
2183 const char * preface = ( uv > PERL_UNICODE_MAX
2184#ifdef EBCDIC
2185 || uv <= 0xFF
2186#endif
2187 )
2188 ? "0x"
2189 : "U+";
6c64cd9d
KW
2190 message = Perl_form(aTHX_
2191 "%s: %s (overlong; instead use %s to represent"
2192 " %s%0*" UVXf ")",
2193 malformed_text,
2194 _byte_dump_string(s0, send - s0, 0),
2195 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2196 preface,
2197 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2198 small code points */
1be62ab9 2199 UNI_TO_NATIVE(uv));
6c64cd9d 2200 }
37657a5b 2201 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2202 }
2203 }
2b5e7bc2
KW
2204 } /* End of looking through the possible flags */
2205
2206 /* Display the message (if any) for the problem being handled in
2207 * this iteration of the loop */
2208 if (message) {
37657a5b 2209 if (msgs) {
37657a5b
KW
2210 assert(this_flag_bit);
2211
2212 if (*msgs == NULL) {
2213 *msgs = newAV();
2214 }
2215
bb07812e
KW
2216 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2217 pack_warn,
2218 this_flag_bit)));
37657a5b
KW
2219 }
2220 else if (PL_op)
2b5e7bc2
KW
2221 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2222 OP_DESC(PL_op));
2223 else
2224 Perl_warner(aTHX_ pack_warn, "%s", message);
2225 }
ddb65933 2226 } /* End of 'while (possible_problems)' */
a0dbb045 2227
2b5e7bc2
KW
2228 /* Since there was a possible problem, the returned length may need to
2229 * be changed from the one stored at the beginning of this function.
2230 * Instead of trying to figure out if that's needed, just do it. */
2231 if (retlen) {
2232 *retlen = curlen;
2233 }
a0dbb045 2234
2b5e7bc2
KW
2235 if (disallowed) {
2236 if (flags & UTF8_CHECK_ONLY && retlen) {
2237 *retlen = ((STRLEN) -1);
2238 }
2239 return 0;
2240 }
eb83ed87 2241 }
ba210ebe 2242
2b5e7bc2 2243 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2244}
2245
8e84507e 2246/*
ec5f19d0
KW
2247=for apidoc utf8_to_uvchr_buf
2248
2249Returns the native code point of the first character in the string C<s> which
2250is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2251C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2252
524080c4
KW
2253If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2254enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2255C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2256(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2257C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2258the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2259See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2260returned.
ec5f19d0
KW
2261
2262=cut
52be2536
KW
2263
2264Also implemented as a macro in utf8.h
2265
ec5f19d0
KW
2266*/
2267
2268
2269UV
2270Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2271{
7f974d7e
KW
2272 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2273
82651abe 2274 return _utf8_to_uvchr_buf(s, send, retlen);
ec5f19d0
KW
2275}
2276
52be2536
KW
2277/* This is marked as deprecated
2278 *
ec5f19d0
KW
2279=for apidoc utf8_to_uvuni_buf
2280
de69f3af
KW
2281Only in very rare circumstances should code need to be dealing in Unicode
2282(as opposed to native) code points. In those few cases, use
07dfe0a4
KW
2283C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2284are not absolutely sure this is one of those cases, then assume it isn't and
2285use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2286
2287Returns the Unicode (not-native) code point of the first character in the
2288string C<s> which
ec5f19d0
KW
2289is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2290C<retlen> will be set to the length, in bytes, of that character.
2291
524080c4
KW
2292If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2293enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2294NULL) to -1. If those warnings are off, the computed value if well-defined (or
2295the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2296is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2297next possible position in C<s> that could begin a non-malformed character.
de69f3af 2298See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2299
2300=cut
2301*/
2302
2303UV
2304Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2305{
2306 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2307
2308 assert(send > s);
2309
5962d97e 2310 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2311}
2312
b76347f2 2313/*
87cea99e 2314=for apidoc utf8_length
b76347f2 2315
b2e7ed74
KW
2316Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2317at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2318same place, it returns 0 with no warning raised.
2319
2320If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2321and returns the number of valid characters.
b76347f2
JH
2322
2323=cut
2324*/
2325
2326STRLEN
35a4481c 2327Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2328{
2329 STRLEN len = 0;
2330
7918f24d
NC
2331 PERL_ARGS_ASSERT_UTF8_LENGTH;
2332
8850bf83
JH
2333 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2334 * the bitops (especially ~) can create illegal UTF-8.
2335 * In other words: in Perl UTF-8 is not just for Unicode. */
2336
12c43b0a 2337 if (UNLIKELY(e < s))
a3b680e6 2338 goto warn_and_return;
b76347f2 2339 while (s < e) {
4cbf4130 2340 s += UTF8SKIP(s);
8e91ec7f
AV
2341 len++;
2342 }
2343
12c43b0a 2344 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2345 len--;
2346 warn_and_return:
9b387841
NC
2347 if (PL_op)
2348 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2349 "%s in %s", unees, OP_DESC(PL_op));
2350 else
61a12c31 2351 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2352 }
2353
2354 return len;
2355}
2356
b06226ff 2357/*
fed3ba5d
NC
2358=for apidoc bytes_cmp_utf8
2359
a1433954 2360Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2361sequence of characters (stored as UTF-8)
2362in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2363equal, -1 or -2 if the first string is less than the second string, +1 or +2
2364if the first string is greater than the second string.
2365
2366-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2367longer string. -2 or +2 is returned if
2368there was a difference between characters
fed3ba5d
NC
2369within the strings.
2370
2371=cut
2372*/
2373
2374int
2375Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2376{
2377 const U8 *const bend = b + blen;
2378 const U8 *const uend = u + ulen;
2379
2380 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2381
2382 while (b < bend && u < uend) {
2383 U8 c = *u++;
2384 if (!UTF8_IS_INVARIANT(c)) {
2385 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2386 if (u < uend) {
2387 U8 c1 = *u++;
2388 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2389 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2390 } else {
2b5e7bc2 2391 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2392 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2393 "%s %s%s",
2394 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2395 PL_op ? " in " : "",
2396 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2397 return -2;
2398 }
2399 } else {
2400 if (PL_op)
2401 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2402 "%s in %s", unees, OP_DESC(PL_op));
2403 else
61a12c31 2404 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2405 return -2; /* Really want to return undef :-) */
2406 }
2407 } else {
2408 return -2;
2409 }
2410 }
2411 if (*b != c) {
2412 return *b < c ? -2 : +2;
2413 }
2414 ++b;
2415 }
2416
2417 if (b == bend && u == uend)
2418 return 0;
2419
2420 return b < bend ? +1 : -1;
2421}
2422
2423/*
87cea99e 2424=for apidoc utf8_to_bytes
6940069f 2425
3bc0c78c 2426Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2427Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2428updates C<*lenp> to contain the new length.
3bc0c78c
KW
2429Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2430
2431Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2432having saved the value of C<*lenp> before the call, and subtracting the
2433after-call value of C<*lenp> from it.
6940069f 2434
a1433954 2435If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2436
6940069f
GS
2437=cut
2438*/
2439
2440U8 *
09af0336 2441Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2442{
9fe0d3c2 2443 U8 * first_variant;
246fae53 2444
7918f24d 2445 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2446 PERL_UNUSED_CONTEXT;
7918f24d 2447
9fe0d3c2 2448 /* This is a no-op if no variants at all in the input */
09af0336 2449 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2450 return s;
2451 }
2452
2453 {
3c5aa262 2454 U8 * const save = s;
09af0336 2455 U8 * const send = s + *lenp;
3c5aa262
KW
2456 U8 * d;
2457
2458 /* Nothing before the first variant needs to be changed, so start the real
2459 * work there */
2460 s = first_variant;
2461 while (s < send) {
2462 if (! UTF8_IS_INVARIANT(*s)) {
2463 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2464 *lenp = ((STRLEN) -1);
3c5aa262
KW
2465 return 0;
2466 }
2467 s++;
d59937ca
KW
2468 }
2469 s++;
dcad2880 2470 }
dcad2880 2471
3c5aa262
KW
2472 /* Is downgradable, so do it */
2473 d = s = first_variant;
2474 while (s < send) {
2475 U8 c = *s++;
2476 if (! UVCHR_IS_INVARIANT(c)) {
2477 /* Then it is two-byte encoded */
2478 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2479 s++;
2480 }
2481 *d++ = c;
2482 }
2483 *d = '\0';
09af0336 2484 *lenp = d - save;
3c5aa262
KW
2485
2486 return save;
9fe0d3c2 2487 }
6940069f
GS
2488}
2489
2490/*
87cea99e 2491=for apidoc bytes_from_utf8
f9a63242 2492
09af0336 2493Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2494byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2495actually encoded in UTF-8.
2496
2497Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2498the input string.
2499
41ae6089
KW
2500Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2501not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2502C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2503
41ae6089 2504Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2505newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2506returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2507caller is responsible for arranging for the memory used by this string to get
2508freed.
f9a63242 2509
3bc0c78c 2510Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2511having saved the value of C<*lenp> before the call, and subtracting the
2512after-call value of C<*lenp> from it.
3bc0c78c 2513
37607a96 2514=cut
976c1b08
KW
2515
2516There is a macro that avoids this function call, but this is retained for
2517anyone who calls it with the Perl_ prefix */
f9a63242
JH
2518
2519U8 *
41ae6089 2520Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2521{
7918f24d 2522 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2523 PERL_UNUSED_CONTEXT;
f9a63242 2524
976c1b08
KW
2525 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2526}
2527
2528/*
2529No = here because currently externally undocumented
2530for apidoc bytes_from_utf8_loc
2531
2532Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2533to store the location of the first character in C<"s"> that cannot be
2534converted to non-UTF8.
2535
2536If that parameter is C<NULL>, this function behaves identically to
2537C<bytes_from_utf8>.
2538
2539Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2540C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2541
2542Otherwise, the function returns a newly created C<NUL>-terminated string
2543containing the non-UTF8 equivalent of the convertible first portion of
2544C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2545If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2546and C<*first_non_downgradable> is set to C<NULL>.
2547
2548Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2549first character in the original string that wasn't converted. C<*is_utf8p> is
2550unchanged. Note that the new string may have length 0.
2551
2552Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2553C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2554converts as many characters in it as possible stopping at the first one it
385b74be 2555finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2556set to point to that. The function returns the portion that could be converted
2557in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2558not including the terminating C<NUL>. If the very first character in the
2559original could not be converted, C<*lenp> will be 0, and the new string will
2560contain just a single C<NUL>. If the entire input string was converted,
2561C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2562
2563Upon successful return, the number of variants in the converted portion of the
2564string can be computed by having saved the value of C<*lenp> before the call,
2565and subtracting the after-call value of C<*lenp> from it.
2566
2567=cut
2568
2569
2570*/
2571
2572U8 *
2573Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2574{
2575 U8 *d;
2576 const U8 *original = s;
2577 U8 *converted_start;
2578 const U8 *send = s + *lenp;
f9a63242 2579
976c1b08 2580 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2581
976c1b08
KW
2582 if (! *is_utf8p) {
2583 if (first_unconverted) {
2584 *first_unconverted = NULL;
2585 }
2586
2587 return (U8 *) original;
2588 }
2589
2590 Newx(d, (*lenp) + 1, U8);
2591
2592 converted_start = d;
7299a045
KW
2593 while (s < send) {
2594 U8 c = *s++;
2595 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2596
2597 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2598 * have to stop now */
2599 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2600 if (first_unconverted) {
2601 *first_unconverted = s - 1;
2602 goto finish_and_return;
2603 }
2604 else {
2605 Safefree(converted_start);
2606 return (U8 *) original;
2607 }
2608 }
2609
7299a045
KW
2610 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2611 s++;
38af28cf 2612 }
7299a045
KW
2613 *d++ = c;
2614 }
170a1c22 2615
976c1b08
KW
2616 /* Here, converted the whole of the input */
2617 *is_utf8p = FALSE;
2618 if (first_unconverted) {
2619 *first_unconverted = NULL;
170a1c22 2620 }
976c1b08
KW
2621
2622 finish_and_return:
46a08a6f
KW
2623 *d = '\0';
2624 *lenp = d - converted_start;
976c1b08
KW
2625
2626 /* Trim unused space */
2627 Renew(converted_start, *lenp + 1, U8);
2628
2629 return converted_start;
f9a63242
JH
2630}
2631
2632/*
87cea99e 2633=for apidoc bytes_to_utf8
6940069f 2634
09af0336 2635Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2636UTF-8.
09af0336 2637Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2638reflect the new length in bytes. The caller is responsible for arranging for
2639the memory used by this string to get freed.
6940069f 2640
3bc0c78c 2641Upon successful return, the number of variants in the string can be computed by
23b37b12 2642having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2643after-call value of C<*lenp>.
2644
75200dff 2645A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2646
2647If you want to convert to UTF-8 from encodings other than
2648the native (Latin1 or EBCDIC),
a1433954 2649see L</sv_recode_to_utf8>().
c9ada85f 2650
497711e7 2651=cut
6940069f
GS
2652*/
2653
2654U8*
09af0336 2655Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2656{
09af0336 2657 const U8 * const send = s + (*lenp);
6940069f
GS
2658 U8 *d;
2659 U8 *dst;
7918f24d
NC
2660
2661 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2662 PERL_UNUSED_CONTEXT;
6940069f 2663
d4662719
KW
2664 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2665 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2666 dst = d;
2667
2668 while (s < send) {
55d09dc8
KW
2669 append_utf8_from_native_byte(*s, &d);
2670 s++;
6940069f 2671 }
2e11cf67 2672
6940069f 2673 *d = '\0';
09af0336 2674 *lenp = d-dst;
2e11cf67 2675
6940069f
GS
2676 return dst;
2677}
2678
a0ed51b3 2679/*
624504c5
KW
2680 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2681 * use utf16_to_utf8_reversed().
a0ed51b3 2682 *
624504c5
KW
2683 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2684 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2685 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2686 *
2687 * These functions don't check for overflow. The worst case is every code
2688 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2689 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2690 * destination must be pre-extended to 2 times the source length.
2691 *
2692 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2693
2694U8*
dea0fc0b 2695Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2696{
dea0fc0b
JH
2697 U8* pend;
2698 U8* dstart = d;
2699
7918f24d
NC
2700 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2701
dea0fc0b 2702 if (bytelen & 1)
56576a04
KW
2703 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2704 (UV)bytelen);
dea0fc0b
JH
2705
2706 pend = p + bytelen;
2707
a0ed51b3 2708 while (p < pend) {
dea0fc0b
JH
2709 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2710 p += 2;
2d1545e5 2711 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2712 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2713 continue;
2714 }
56d37426
KW
2715 if (uv <= MAX_UTF8_TWO_BYTE) {
2716 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2717 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2718 continue;
2719 }
ffd0a9d3 2720
46956fad
KW
2721#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2722#define LAST_HIGH_SURROGATE 0xDBFF
2723#define FIRST_LOW_SURROGATE 0xDC00
2724#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2725#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2726
2727 /* This assumes that most uses will be in the first Unicode plane, not
2728 * needing surrogates */
2729 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2730 && uv <= UNICODE_SURROGATE_LAST))
2731 {
2732 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2733 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2734 }
2735 else {
01ea242b 2736 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2737 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2738 || UNLIKELY(low > LAST_LOW_SURROGATE))
2739 {
01ea242b 2740 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2741 }
2742 p += 2;
46956fad 2743 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2744 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2745 }
a0ed51b3 2746 }
56d37426
KW
2747#ifdef EBCDIC
2748 d = uvoffuni_to_utf8_flags(d, uv, 0);
2749#else
ffd0a9d3 2750 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2751 *d++ = (U8)(( uv >> 12) | 0xe0);
2752 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2753 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2754 continue;
2755 }
2756 else {
eb160463
GS
2757 *d++ = (U8)(( uv >> 18) | 0xf0);
2758 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2759 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2760 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2761 continue;
2762 }
56d37426 2763#endif
a0ed51b3 2764 }
dea0fc0b 2765 *newlen = d - dstart;
a0ed51b3
LW
2766 return d;
2767}
2768
2769/* Note: this one is slightly destructive of the source. */
2770
2771U8*
dea0fc0b 2772Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2773{
2774 U8* s = (U8*)p;
d4c19fe8 2775 U8* const send = s + bytelen;
7918f24d
NC
2776
2777 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2778
e0ea5e2d 2779 if (bytelen & 1)
147e3846 2780 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2781 (UV)bytelen);
2782
a0ed51b3 2783 while (s < send) {
d4c19fe8 2784 const U8 tmp = s[0];
a0ed51b3
LW
2785 s[0] = s[1];
2786 s[1] = tmp;
2787 s += 2;
2788 }
dea0fc0b 2789 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2790}
2791
922e8cb4
KW
2792bool
2793Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2794{
1565c085 2795 dVAR;
dc31b55c 2796 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2797}
2798
f9ae8fb6
JD
2799/* Internal function so we can deprecate the external one, and call
2800 this one from other deprecated functions in this file */
2801
f2645549
KW
2802bool
2803Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2804{
1565c085
DM
2805 dVAR;
2806
f2645549 2807 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2808
2809 if (*p == '_')
2810 return TRUE;
eb1f4bb4 2811 return is_utf8_common(p, PL_utf8_idstart);
61b19385
KW
2812}
2813
5092f92a 2814bool
eba68aa0
KW
2815Perl__is_uni_perl_idcont(pTHX_ UV c)
2816{
1565c085 2817 dVAR;
c12658c9 2818 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2819}
2820
2821bool
f91dcd13
KW
2822Perl__is_uni_perl_idstart(pTHX_ UV c)
2823{
1565c085 2824 dVAR;
c12658c9 2825 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2826}
2827
3a4c58c9 2828UV
56576a04
KW
2829Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2830 const char S_or_s)
3a4c58c9
KW
2831{
2832 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2833 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2834 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2835 * either "SS" or "Ss". Which one to use is passed into the routine in
2836 * 'S_or_s' to avoid a test */
2837
2838 UV converted = toUPPER_LATIN1_MOD(c);
2839
2840 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2841
2842 assert(S_or_s == 'S' || S_or_s == 's');
2843
6f2d5cbc 2844 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2845 characters in this range */
3a4c58c9
KW
2846 *p = (U8) converted;
2847 *lenp = 1;
2848 return converted;
2849 }
2850
2851 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2852 * which it maps to one of them, so as to only have to have one check for
2853 * it in the main case */
2854 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2855 switch (c) {
2856 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2857 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2858 break;
2859 case MICRO_SIGN:
2860 converted = GREEK_CAPITAL_LETTER_MU;
2861 break;
79e064b9
KW
2862#if UNICODE_MAJOR_VERSION > 2 \
2863 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2864 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2865 case LATIN_SMALL_LETTER_SHARP_S:
2866 *(p)++ = 'S';
2867 *p = S_or_s;
2868 *lenp = 2;
2869 return 'S';
79e064b9 2870#endif
3a4c58c9 2871 default:
56576a04
KW
2872 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2873 " '%c' to map to '%c'",
2874 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2875 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2876 }
2877 }
2878
2879 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2880 *p = UTF8_TWO_BYTE_LO(converted);
2881 *lenp = 2;
2882
2883 return converted;
2884}
2885
fe63c520
KW
2886/* If compiled on an early Unicode version, there may not be auxiliary tables
2887 * */
2888#ifndef HAS_UC_AUX_TABLES
2889# define UC_AUX_TABLE_ptrs NULL
2890# define UC_AUX_TABLE_lengths NULL
2891#endif
2892#ifndef HAS_TC_AUX_TABLES
2893# define TC_AUX_TABLE_ptrs NULL
2894# define TC_AUX_TABLE_lengths NULL
2895#endif
2896#ifndef HAS_LC_AUX_TABLES
2897# define LC_AUX_TABLE_ptrs NULL
2898# define LC_AUX_TABLE_lengths NULL
2899#endif
2900#ifndef HAS_CF_AUX_TABLES
2901# define CF_AUX_TABLE_ptrs NULL
2902# define CF_AUX_TABLE_lengths NULL
2903#endif
2904#ifndef HAS_UC_AUX_TABLES
2905# define UC_AUX_TABLE_ptrs NULL
2906# define UC_AUX_TABLE_lengths NULL
2907#endif
2908
50bda2c3
KW
2909/* Call the function to convert a UTF-8 encoded character to the specified case.
2910 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2911 * 's' is a pointer to the first byte of the input character
2912 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2913 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2914 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2915 *
56576a04 2916 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2917 * 'd' */
56576a04 2918#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2919 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2920 Uppercase_Mapping_invmap, \
2921 UC_AUX_TABLE_ptrs, \
2922 UC_AUX_TABLE_lengths, \
2923 "uppercase")
56576a04 2924#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2925 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2926 Titlecase_Mapping_invmap, \
2927 TC_AUX_TABLE_ptrs, \
2928 TC_AUX_TABLE_lengths, \
2929 "titlecase")
56576a04 2930#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2931 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2932 Lowercase_Mapping_invmap, \
2933 LC_AUX_TABLE_ptrs, \
2934 LC_AUX_TABLE_lengths, \
2935 "lowercase")
2936
50bda2c3 2937
b9992569
KW
2938/* This additionally has the input parameter 'specials', which if non-zero will
2939 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2940 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2941#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2942 (specials) \
2943 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2944 Case_Folding_invmap, \
2945 CF_AUX_TABLE_ptrs, \
2946 CF_AUX_TABLE_lengths, \
2947 "foldcase") \
2948 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2949 Simple_Case_Folding_invmap, \
2950 NULL, NULL, \
2951 "foldcase")
c3fd2246 2952
84afefe6
JH
2953UV
2954Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2955{
a1433954
KW
2956 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2957 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2958 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2959 * the changed version may be longer than the original character.
2960 *
2961 * The ordinal of the first character of the changed version is returned
2962 * (but note, as explained above, that there may be more.) */
2963
1565c085 2964 dVAR;
7918f24d
NC
2965 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2966
3a4c58c9
KW
2967 if (c < 256) {
2968 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2969 }
2970
a13f1de4 2971 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2972}
2973
84afefe6
JH
2974UV
2975Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2976{
1565c085 2977 dVAR;
7918f24d
NC
2978 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2979
3a4c58c9
KW
2980 if (c < 256) {
2981 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2982 }
2983
a13f1de4 2984 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2985}
2986
afc16117 2987STATIC U8
eaf412bf 2988S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2989{
2990 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2991 * those, converting the result to UTF-8. Since the result is always just
a1433954 2992 * one character, we allow <p> to be NULL */
afc16117
KW
2993
2994 U8 converted = toLOWER_LATIN1(c);
2995
eaf412bf
KW
2996 PERL_UNUSED_ARG(dummy);
2997
afc16117 2998 if (p != NULL) {
6f2d5cbc 2999 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3000 *p = converted;
3001 *lenp = 1;
3002 }
3003 else {
430c9760
KW
3004 /* Result is known to always be < 256, so can use the EIGHT_BIT
3005 * macros */
3006 *p = UTF8_EIGHT_BIT_HI(converted);
3007 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3008 *lenp = 2;
3009 }
3010 }
3011 return converted;
3012}
3013
84afefe6
JH
3014UV
3015Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3016{
1565c085 3017 dVAR;
7918f24d
NC
3018 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3019
afc16117 3020 if (c < 256) {
eaf412bf 3021 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3022 }
3023
a13f1de4 3024 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3025}
3026
84afefe6 3027UV
7c0ab950 3028Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3029{
51910141 3030 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3031 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3032 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3033 *
3034 * Not to be used for locale folds
51910141 3035 */
f673fad4 3036
a1dde8de
KW
3037 UV converted;
3038
3039 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3040
1ca267a5
KW
3041 assert (! (flags & FOLD_FLAGS_LOCALE));
3042
659a7c2d 3043 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3044 converted = GREEK_SMALL_LETTER_MU;
3045 }
9b63e895
KW
3046#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3047 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3048 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3049 else if ( (flags & FOLD_FLAGS_FULL)
3050 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3051 {
1ca267a5
KW
3052 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3053 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3054 * under those circumstances. */
3055 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3056 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3057 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3058 p, *lenp, U8);
3059 return LATIN_SMALL_LETTER_LONG_S;
3060 }
3061 else {
4f489194
KW
3062 *(p)++ = 's';
3063 *p = 's';
3064 *lenp = 2;
3065 return 's';
1ca267a5 3066 }
a1dde8de 3067 }
9b63e895 3068#endif
a1dde8de
KW
3069 else { /* In this range the fold of all other characters is their lower
3070 case */
3071 converted = toLOWER_LATIN1(c);
3072 }
3073
6f2d5cbc 3074 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3075 *p = (U8) converted;
3076 *lenp = 1;
3077 }
3078 else {
3079 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3080 *p = UTF8_TWO_BYTE_LO(converted);
3081 *lenp = 2;
3082 }
3083
3084 return converted;
3085}
3086
3087UV
31f05a37 3088Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3089{
4b593389 3090
a0270393
KW
3091 /* Not currently externally documented, and subject to change
3092 * <flags> bits meanings:
3093 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3094 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3095 * locale are to be used.
a0270393
KW
3096 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3097 */
4b593389 3098
1565c085 3099 dVAR;
36bb2ab6 3100 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3101
780fcc9f 3102 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3103 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3104 * except for potentially warning */
8b7358b9 3105 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3106 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3107 flags &= ~FOLD_FLAGS_LOCALE;
3108 }
3109 else {
e7b7ac46 3110 goto needs_full_generality;
780fcc9f 3111 }
31f05a37
KW
3112 }
3113
a1dde8de 3114 if (c < 256) {
e7b7ac46 3115 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3116 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3117 }
3118
2f306ab9 3119 /* Here, above 255. If no special needs, just use the macro */
a0270393 3120 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3121 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3122 }
567b353c 3123 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3124 the special flags. */
3125 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3126
3127 needs_full_generality:
a0270393 3128 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3129 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3130 p, lenp, flags);
a0270393 3131 }
84afefe6
JH
3132}
3133
26483009 3134PERL_STATIC_INLINE bool
eb1f4bb4 3135S_is_utf8_common(pTHX_ const U8 *const p, SV* const invlist)
bde6a22d 3136{
ea317ccb 3137 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4 3138 * starts at <p> is in the inversion list indicated by <invlist>.
ea317ccb
KW
3139 *
3140 * Note that it is assumed that the buffer length of <p> is enough to
3141 * contain all the bytes that comprise the character. Thus, <*p> should
3142 * have been checked before this call for mal-formedness enough to assure
016c8ffc
KW
3143 * that. This function, does make sure to not look past any NUL, so it is
3144 * safe to use on C, NUL-terminated, strings */
3145 STRLEN len = my_strnlen((char *) p, UTF8SKIP(p));
ea317ccb 3146
7918f24d
NC
3147 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3148
492a624f 3149 /* The API should have included a length for the UTF-8 character in <p>,
28123549 3150 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
3151 * as far as there being enough bytes available in it to accommodate the
3152 * character without reading beyond the end, and pass that number on to the
3153 * validating routine */
016c8ffc
KW
3154 if (! isUTF8_CHAR(p, p + len)) {
3155 _force_out_malformed_utf8_message(p, p + len, _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
3156 1 /* Die */ );
3157 NOT_REACHED; /* NOTREACHED */
28123549 3158 }
86ae6e94 3159
eb1f4bb4 3160 return is_utf8_common_with_len(p, p + len, invlist);
bde6a22d
NC
3161}
3162
da8c1a98 3163PERL_STATIC_INLINE bool
56576a04 3164S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
56576a04 3165 SV* const invlist)
da8c1a98
KW
3166{
3167 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3168 * starts at <p>, and extending no further than <e - 1> is in the inversion
3169 * list <invlist>. */
da8c1a98 3170
b68ffe0c
KW
3171 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3172
da8c1a98
KW
3173 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3174
b68ffe0c 3175 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3176 _force_out_malformed_utf8_message(p, e, 0, 1);
3177 NOT_REACHED; /* NOTREACHED */
3178 }
3179
eb1f4bb4 3180 assert(invlist);
b68ffe0c 3181 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3182}
3183
34aeb2e9
KW
3184STATIC void
3185S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3186 const char * const alternative,
3187 const bool use_locale,
3188 const char * const file,
3189 const unsigned line)
3190{
3191 const char * key;
3192
3193 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3194
3195 if (ckWARN_d(WARN_DEPRECATED)) {
3196
3197 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3198 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3199 if (! PL_seen_deprecated_macro) {
3200 PL_seen_deprecated_macro = newHV();
3201 }
3202 if (! hv_store(PL_seen_deprecated_macro, key,
3203 strlen(key), &PL_sv_undef, 0))
3204 {
3205 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3206 }
3207
c44e9413 3208 if (instr(file, "mathoms.c")) {
607313a1 3209 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3210 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3211 " will be removed. Avoid this message by"
3212 " converting to use %s().\n",
3213 file, line, name, alternative);
3214 }
3215 else {
34aeb2e9 3216 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3217 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3218 " require an additional parameter. Avoid this"
3219 " message by converting to use %s().\n",
3220 file, line, name, alternative);
607313a1 3221 }
34aeb2e9
KW
3222 }
3223 }
3224}
3225
bde6a22d 3226bool
34aeb2e9 3227Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3228 const U8 * const p,
34aeb2e9
KW
3229 const char * const name,
3230 const char * const alternative,
3231 const bool use_utf8,
3232 const bool use_locale,
3233 const char * const file,
3234 const unsigned line)
922e8cb4 3235{
1565c085 3236 dVAR;
922e8cb4
KW
3237 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3238
34aeb2e9
KW
3239 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3240
3241 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3242
3243 switch (classnum) {
3244 case _CC_WORDCHAR:
3245 case _CC_DIGIT:
3246 case _CC_ALPHA:
3247 case _CC_LOWER:
3248 case _CC_UPPER:
3249 case _CC_PUNCT:
3250 case _CC_PRINT:
3251 case _CC_ALPHANUMERIC:
3252 case _CC_GRAPH:
3253 case _CC_CASED:
3254
eb1f4bb4 3255 return is_utf8_common(p, PL_XPosix_ptrs[classnum]);
34aeb2e9
KW
3256
3257 case _CC_SPACE:
3258 return is_XPERLSPACE_high(p);
3259 case _CC_BLANK:
3260 return is_HORIZWS_high(p);
3261 case _CC_XDIGIT:
3262 return is_XDIGIT_high(p);
3263 case _CC_CNTRL:
3264 return 0;
3265 case _CC_ASCII:
3266 return 0;
3267 case _CC_VERTSPACE:
3268 return is_VERTWS_high(p);
3269 case _CC_IDFIRST:
eb1f4bb4 3270 return is_utf8_common(p, PL_utf8_perl_idstart);
34aeb2e9 3271 case _CC_IDCONT:
eb1f4bb4 3272 return is_utf8_common(p, PL_utf8_perl_idcont);
34aeb2e9
KW
3273 }
3274 }
3275
3276 /* idcont is the same as wordchar below 256 */
3277 if (classnum == _CC_IDCONT) {
3278 classnum = _CC_WORDCHAR;
3279 }
3280 else if (classnum == _CC_IDFIRST) {
3281 if (*p == '_') {
3282 return TRUE;
3283 }
3284 classnum = _CC_ALPHA;
3285 }
3286
3287 if (! use_locale) {
3288 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3289 return _generic_isCC(*p, classnum);
3290 }
922e8cb4 3291
34aeb2e9
KW
3292 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3293 }
3294 else {
3295 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3296 return isFOO_lc(classnum, *p);
3297 }
3298
3299 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3300 }
3301
3302 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3303}
3304
3305bool
da8c1a98
KW
3306Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3307 const U8 * const e)
3308{
1565c085 3309 dVAR;
da8c1a98
KW
3310 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3311
eb1f4bb4 3312 return is_utf8_common_with_len(p, e, PL_XPosix_ptrs[classnum]);
da8c1a98
KW
3313}
3314
3315bool
da8c1a98
KW
3316Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3317{
1565c085 3318 dVAR;
da8c1a98
KW
3319 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3320
eb1f4bb4 3321 return is_utf8_common_with_len(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3322}
3323
3324bool
f2645549 3325Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3326{
1565c085 3327 dVAR;
f2645549 3328 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3329
3330 if (*p == '_')
3331 return TRUE;
eb1f4bb4 3332 return is_utf8_common(p, PL_utf8_xidstart);
c11ff943
KW
3333}
3334
3335bool
da8c1a98
KW
3336Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3337{
1565c085 3338 dVAR;
da8c1a98
KW
3339 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3340
eb1f4bb4 3341 return is_utf8_common_with_len(p, e, PL_utf8_perl_idcont);
da8c1a98
KW
3342}
3343
3344bool
f2645549 3345Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3346{
1565c085 3347 dVAR;
f2645549 3348 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3349
eb1f4bb4 3350 return is_utf8_common(p, PL_utf8_idcont);
a0ed51b3
LW
3351}
3352
3353bool
f2645549 3354Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3355{
1565c085 3356 dVAR;
f2645549 3357 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3358
eb1f4bb4 3359 return is_utf8_common(p, PL_utf8_xidcont);
c11ff943
KW
3360}
3361
3362bool
7dbf68d2
KW
3363Perl__is_utf8_mark(pTHX_ const U8 *p)
3364{
1565c085 3365 dVAR;
7dbf68d2
KW
3366 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3367
eb1f4bb4 3368 return is_utf8_common(p, PL_utf8_mark);
7dbf68d2
KW
3369}
3370
6a4a25f4 3371STATIC UV
30613bdc
KW
3372S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3373 U8* ustrp, STRLEN *lenp,
341bb5b7 3374 SV *invlist, const int * const invmap,
e39a4130 3375 const unsigned int * const * const aux_tables,
30613bdc
KW
3376 const U8 * const aux_table_lengths,
3377 const char * const normal)
b9992569 3378{
0134edef 3379 STRLEN len = 0;
7918f24d 3380
30613bdc
KW
3381 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3382 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3383 * to name the new case in any generated messages, as a fallback if the
3384 * operation being used is not available. The new case is given by the
3385 * data structures in the remaining arguments.
3386 *
3387 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3388 * entire changed case string, and the return value is the first code point
3389 * in that string */
3390
b9992569 3391 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3392
36eaa811
KW
3393 /* For code points that don't change case, we already know that the output
3394 * of this function is the unchanged input, so we can skip doing look-ups
3395 * for them. Unfortunately the case-changing code points are scattered
3396 * around. But there are some long consecutive ranges where there are no
3397 * case changing code points. By adding tests, we can eliminate the lookup
3398 * for all the ones in such ranges. This is currently done here only for
3399 * just a few cases where the scripts are in common use in modern commerce
3400 * (and scripts adjacent to those which can be included without additional
3401 * tests). */
3402
3403 if (uv1 >= 0x0590) {
3404 /* This keeps from needing further processing the code points most
3405 * likely to be used in the following non-cased scripts: Hebrew,
3406 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3407 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3408 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3409 if (uv1 < 0x10A0) {
3410 goto cases_to_self;
3411 }
3412
3413 /* The following largish code point ranges also don't have case
3414 * changes, but khw didn't think they warranted extra tests to speed
3415 * them up (which would slightly slow down everything else above them):
3416 * 1100..139F Hangul Jamo, Ethiopic
3417 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3418 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3419 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3420 * Combining Diacritical Marks Extended, Balinese,
3421 * Sundanese, Batak, Lepcha, Ol Chiki
3422 * 2000..206F General Punctuation
3423 */
3424
3425 if (uv1 >= 0x2D30) {
3426
3427 /* This keeps the from needing further processing the code points
3428 * most likely to be used in the following non-cased major scripts:
3429 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3430 *
3431 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3432 * event that Unicode eventually allocates the unused block as of
3433 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3434 * that the test suite will start having failures to alert you
3435 * should that happen) */
3436 if (uv1 < 0xA640) {
3437 goto cases_to_self;
3438 }
3439
3440 if (uv1 >= 0xAC00) {
3441 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3442 if (ckWARN_d(WARN_SURROGATE)) {
3443 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3444 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3445 "Operation \"%s\" returns its argument for"
3446 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3447 }
3448 goto cases_to_self;
3449 }
36eaa811
KW
3450
3451 /* AC00..FAFF Catches Hangul syllables and private use, plus
3452 * some others */
3453 if (uv1 < 0xFB00) {
3454 goto cases_to_self;
36eaa811
KW
3455 }
3456
5af9bc97 3457 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3458 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
d22ec717 3459 Perl_croak(aTHX_ cp_above_legal_max, uv1,
40606899 3460 MAX_LEGAL_CP);
5af9bc97
KW
3461 }
3462 if (ckWARN_d(WARN_NON_UNICODE)) {
3463 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3464 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3465 "Operation \"%s\" returns its argument for"
3466 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3467 }
3468 goto cases_to_self;
3469 }
3bfc1e70
KW
3470#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3471 if (UNLIKELY(uv1
3472 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3473 {
3474
56576a04
KW
3475 /* As of Unicode 10.0, this means we avoid swash creation
3476 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3477 goto cases_to_self;
3478 }
3479#endif
36eaa811
KW
3480 }
3481 }
9ae3ac1a 3482
36eaa811 3483 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3484 * be given. */
9ae3ac1a
KW
3485 }
3486
8946fcd9
KW
3487 {
3488 unsigned int i;
e39a4130 3489 const unsigned int * cp_list;
8946fcd9 3490 U8 * d;
69352d88
KW
3491
3492 /* 'index' is guaranteed to be non-negative, as this is an inversion
3493 * map that covers all possible inputs. See [perl #133365] */
8946fcd9
KW
3494 SSize_t index = _invlist_search(invlist, uv1);
3495 IV base = invmap[index];
0134edef 3496
30613bdc
KW
3497 /* The data structures are set up so that if 'base' is non-negative,
3498 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3499 if (base >= 0) {
3500 IV lc;
b08cf34e 3501
8946fcd9
KW
3502 if (base == 0) {
3503 goto cases_to_self;
4a8240a3 3504 }
4a8240a3 3505
30613bdc 3506 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3507 lc = base + uv1 - invlist_array(invlist)[index];
3508 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3509 return lc;
3510 }
1feea2c7 3511
30613bdc
KW
3512 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3513 * requires an auxiliary table look up. abs(base) gives the index into
3514 * a list of such tables which points to the proper aux table. And a
3515 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3516 cp_list = aux_tables[-base];
30613bdc
KW
3517
3518 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3519 d = ustrp;
3520 for (i = 0; i < aux_table_lengths[-base]; i++) {
3521 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3522 }
8946fcd9
KW
3523 *d = '\0';
3524 *lenp = d - ustrp;
3525
3526 return cp_list[0];
cbe07460
KW
3527 }
3528
3529 /* Here, there was no mapping defined, which means that the code point maps
3530 * to itself. Return the inputs */
e24dfe9c 3531 cases_to_self:
a13f1de4
KW
3532 if (p) {
3533 len = UTF8SKIP(p);
3534 if (p != ustrp) { /* Don't copy onto itself */
3535 Copy(p, ustrp, len, U8);
3536 }
3537 *lenp = len;
3538 }
3539 else {
3540 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3541 }
2a37f04d 3542
f4cd282c 3543 return uv1;
cbe07460 3544
a0ed51b3
LW
3545}
3546
b74fe592 3547Size_t
e39a4130
KW
3548Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3549 const unsigned int ** remaining_folds_to)
b74fe592
KW
3550{
3551 /* Returns the count of the number of code points that fold to the input
3552 * 'cp' (besides itself).
3553 *
3554 * If the return is 0, there is nothing else that folds to it, and
3555 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3556 *
3557 * If the return is 1, '*first_folds_to' is set to the single code point,
3558 * and '*remaining_folds_to' is set to NULL.
3559 *
3560 * Otherwise, '*first_folds_to' is set to a code point, and
3561 * '*remaining_fold_to' is set to an array that contains the others. The
3562 * length of this array is the returned count minus 1.
3563 *
3564 * The reason for this convolution is to avoid having to deal with
3565 * allocating and freeing memory. The lists are already constructed, so
3566 * the return can point to them, but single code points aren't, so would
3567 * need to be constructed if we didn't employ something like this API */
3568
1565c085 3569 dVAR;
69352d88
KW
3570 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3571 * that covers all possible inputs. See [perl #133365] */
b74fe592
KW
3572 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3573 int base = _Perl_IVCF_invmap[index];
3574
3575 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3576
3577 if (base == 0) { /* No fold */
3578 *first_folds_to = 0;
3579 *remaining_folds_to = NULL;
3580 return 0;
3581 }
3582
3583#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3584
3585 assert(base > 0);
3586
3587#else
3588
3589 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3590
3591 /* The data structure is set up so that the absolute value of 'base' is
3592 * an index into a table of pointers to arrays, with the array
3593 * corresponding to the index being the list of code points that fold
3594 * to 'cp', and the parallel array containing the length of the list
3595 * array */
3596 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3597 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3598 *first_folds_to
3599 */
3600 return IVCF_AUX_TABLE_lengths[-base];
3601 }
3602
3603#endif
3604
3605 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3606 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3607 *remaining_folds_to = NULL;
3608 return 1;
3609}
3610
051a06d4 3611STATIC UV
56576a04
KW
3612S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3613 U8* const ustrp, STRLEN *lenp)
051a06d4 3614{
4a4088c4 3615 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3616 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3617 * result contains a character that crosses the 255/256 boundary, disallow
3618 * the change, and return the original code point. See L<perlfunc/lc> for
3619 * why;
051a06d4 3620 *
a1433954
KW
3621 * p points to the original string whose case was changed; assumed
3622 * by this routine to be well-formed
051a06d4 3623 * result the code point of the first character in the changed-case string
56576a04
KW
3624 * ustrp points to the changed-case string (<result> represents its
3625 * first char)
051a06d4
KW
3626 * lenp points to the length of <ustrp> */
3627
3628 UV original; /* To store the first code point of <p> */
3629
3630 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3631
a4f12ed7 3632 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3633
3634 /* We know immediately if the first character in the string crosses the
5e45c680 3635 * boundary, so can skip testing */
051a06d4
KW
3636 if (result > 255) {
3637
3638 /* Look at every character in the result; if any cross the
3639 * boundary, the whole thing is disallowed */
3640 U8* s = ustrp + UTF8SKIP(ustrp);
3641 U8* e = ustrp + *lenp;
3642 while (s < e) {
a4f12ed7 3643 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3644 goto bad_crossing;
3645 }
3646 s += UTF8SKIP(s);
3647 }
3648
613abc6d
KW
3649 /* Here, no characters crossed, result is ok as-is, but we warn. */
3650 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3651 return result;
3652 }
3653
7b52d656 3654 bad_crossing:
051a06d4
KW
3655
3656 /* Failed, have to return the original */
4b88fb76 3657 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3658
3659 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3660 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3661 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3662 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3663 OP_DESC(PL_op),
ab0b796c
KW
3664 original,
3665 original);
051a06d4
KW
3666 Copy(p, ustrp, *lenp, char);
3667 return original;
3668}
3669
607313a1
KW
3670STATIC U32
3671S_check_and_deprecate(pTHX_ const U8 *p,
3672 const U8 **e,
3673 const unsigned int type, /* See below */
3674 const bool use_locale, /* Is this a 'LC_'
3675 macro call? */
3676 const char * const file,
3677 const unsigned line)
3678{
3679 /* This is a temporary function to deprecate the unsafe calls to the case
3680 * changing macros and functions. It keeps all the special stuff in just
3681 * one place.
3682 *
3683 * It updates *e with the pointer to the end of the input string. If using
3684 * the old-style macros, *e is NULL on input, and so this function assumes
3685 * the input string is long enough to hold the entire UTF-8 sequence, and
3686 * sets *e accordingly, but it then returns a flag to pass the
3687 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3688 * using the full length if possible.
3689 *
3690 * It also does the assert that *e > p when *e is not NULL. This should be
3691 * migrated to the callers when this function gets deleted.
3692 *
3693 * The 'type' parameter is used for the caller to specify which case
3694 * changing function this is called from: */
3695
3696# define DEPRECATE_TO_UPPER 0
3697# define DEPRECATE_TO_TITLE 1
3698# define DEPRECATE_TO_LOWER 2
3699# define DEPRECATE_TO_FOLD 3
3700
3701 U32 utf8n_flags = 0;
3702 const char * name;
3703 const char * alternative;
3704
3705 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3706
3707 if (*e == NULL) {
3708 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
6f7ce6e8
KW
3709
3710 /* strnlen() makes this function safe for the common case of
3711 * NUL-terminated strings */
3712 *e = p + my_strnlen((char *) p, UTF8SKIP(p));
607313a1
KW
3713
3714 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3715 * there. It could be part of a larger path */