This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
msgrcv: properly downgrade the receive buffer
[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 39
48ef279e 40/*
7fefc6c1 41These are various utility functions for manipulating UTF8-encoded
72d33970 42strings. For the uninitiated, this is a method of representing arbitrary
61296642 43Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
44characters in the ASCII range are unmodified, and a zero byte never appears
45within non-zero characters.
eaf7a4d2
CS
46*/
47
dd051059
DM
48/* helper for Perl__force_out_malformed_utf8_message(). Like
49 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
50 * PL_compiling */
51
52static void
53S_restore_cop_warnings(pTHX_ void *p)
54{
1943af61 55 free_and_set_cop_warnings(PL_curcop, (STRLEN*) p);
dd051059
DM
56}
57
58
9cbfb8ab
KW
59void
60Perl__force_out_malformed_utf8_message(pTHX_
61 const U8 *const p, /* First byte in UTF-8 sequence */
62 const U8 * const e, /* Final byte in sequence (may include
63 multiple chars */
64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
65 usually 0, or some DISALLOW flags */
66 const bool die_here) /* If TRUE, this function does not return */
67{
68 /* This core-only function is to be called when a malformed UTF-8 character
69 * is found, in order to output the detailed information about the
70 * malformation before dieing. The reason it exists is for the occasions
71 * when such a malformation is fatal, but warnings might be turned off, so
72 * that normally they would not be actually output. This ensures that they
73 * do get output. Because a sequence may be malformed in more than one
74 * way, multiple messages may be generated, so we can't make them fatal, as
75 * that would cause the first one to die.
76 *
77 * Instead we pretend -W was passed to perl, then die afterwards. The
78 * flexibility is here to return to the caller so they can finish up and
79 * die themselves */
80 U32 errors;
81
82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
83
84 ENTER;
c15a80f3 85 SAVEI8(PL_dowarn);
9cbfb8ab
KW
86 SAVESPTR(PL_curcop);
87
88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
89 if (PL_curcop) {
dd051059
DM
90 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
91 * than PL_compiling */
92 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
93 (void*)PL_curcop->cop_warnings);
9cbfb8ab
KW
94 PL_curcop->cop_warnings = pWARN_ALL;
95 }
96
97 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
98
99 LEAVE;
100
101 if (! errors) {
102 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
103 " be called only when there are errors found");
104 }
105
106 if (die_here) {
107 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
108 }
109}
110
bb07812e
KW
111STATIC HV *
112S_new_msg_hv(pTHX_ const char * const message, /* The message text */
113 U32 categories, /* Packed warning categories */
114 U32 flag) /* Flag associated with this message */
115{
116 /* Creates, populates, and returns an HV* that describes an error message
117 * for the translators between UTF8 and code point */
118
119 SV* msg_sv = newSVpv(message, 0);
120 SV* category_sv = newSVuv(categories);
121 SV* flag_bit_sv = newSVuv(flag);
122
123 HV* msg_hv = newHV();
124
125 PERL_ARGS_ASSERT_NEW_MSG_HV;
126
2b672cf5
KW
127 (void) hv_stores(msg_hv, "text", msg_sv);
128 (void) hv_stores(msg_hv, "warn_categories", category_sv);
129 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
130
131 return msg_hv;
132}
133
eaf7a4d2 134/*
378516de 135=for apidoc uvoffuni_to_utf8_flags
eebe1485 136
a27992cc 137THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
138Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
139L<perlapi/uvchr_to_utf8_flags>>.
a27992cc 140
de69f3af
KW
141This function is like them, but the input is a strict Unicode
142(as opposed to native) code point. Only in very rare circumstances should code
143not be using the native code point.
949cf498 144
09232555 145For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
949cf498 146
eebe1485
SC
147=cut
148*/
149
33f38593
KW
150U8 *
151Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
152{
153 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
154
155 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
156}
157
c94c2f39
KW
158/* All these formats take a single UV code point argument */
159const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
160const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
161 " is not recommended for open interchange";
162const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
163 " may not be portable";
c94c2f39 164
33f38593 165#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
166 STMT_START { \
167 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
168 U32 category = packWARN(WARN_SURROGATE); \
169 const char * format = surrogate_cp_format; \
170 if (msgs) { \
171 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
172 category, \
173 UNICODE_GOT_SURROGATE); \
174 } \
175 else { \
176 Perl_ck_warner_d(aTHX_ category, format, uv); \
177 } \
8ee1cdcb
KW
178 } \
179 if (flags & UNICODE_DISALLOW_SURROGATE) { \
180 return NULL; \
181 } \
182 } STMT_END;
183
33f38593 184#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
185 STMT_START { \
186 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
187 U32 category = packWARN(WARN_NONCHAR); \
188 const char * format = nonchar_cp_format; \
189 if (msgs) { \
190 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
191 category, \
192 UNICODE_GOT_NONCHAR); \
193 } \
194 else { \
195 Perl_ck_warner_d(aTHX_ category, format, uv); \
196 } \
8ee1cdcb
KW
197 } \
198 if (flags & UNICODE_DISALLOW_NONCHAR) { \
199 return NULL; \
200 } \
201 } STMT_END;
202
ba6ed43c
KW
203/* Use shorter names internally in this file */
204#define SHIFT UTF_ACCUMULATION_SHIFT
205#undef MARK
206#define MARK UTF_CONTINUATION_MARK
207#define MASK UTF_CONTINUATION_MASK
208
33f38593
KW
209/*
210=for apidoc uvchr_to_utf8_flags_msgs
211
212THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
213
214Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
215
216This function is for code that wants any warning and/or error messages to be
217returned to the caller rather than be displayed. All messages that would have
884a31ee 218been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
219
220It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
221placed after all the others, C<msgs>. If this parameter is 0, this function
222behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
223be a pointer to an C<HV *> variable, in which this function creates a new HV to
224contain any appropriate messages. The hash has three key-value pairs, as
225follows:
226
227=over 4
228
229=item C<text>
230
231The text of the message as a C<SVpv>.
232
233=item C<warn_categories>
234
235The warning category (or categories) packed into a C<SVuv>.
236
237=item C<flag>
238
239A single flag bit associated with this message, in a C<SVuv>.
240The bit corresponds to some bit in the C<*errors> return value,
241such as C<UNICODE_GOT_SURROGATE>.
242
243=back
244
245It's important to note that specifying this parameter as non-null will cause
246any warnings this function would otherwise generate to be suppressed, and
247instead be placed in C<*msgs>. The caller can check the lexical warnings state
248(or not) when choosing what to do with the returned messages.
249
250The caller, of course, is responsible for freeing any returned HV.
251
252=cut
253*/
254
255/* Undocumented; we don't want people using this. Instead they should use
256 * uvchr_to_utf8_flags_msgs() */
dfe13c55 257U8 *
33f38593 258Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 259{
33f38593
KW
260 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
261
262 if (msgs) {
263 *msgs = NULL;
264 }
7918f24d 265
2d1545e5 266 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 267 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
268 return d;
269 }
facc1dc2 270
3ea68d71 271 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
272 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
273 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
274 return d;
275 }
d9432125 276
ba6ed43c
KW
277 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
278 * below, the 16 is for start bytes E0-EF (which are all the possible ones
279 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
280 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
281 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
282 * 0x800-0xFFFF on ASCII */
283 if (uv < (16 * (1U << (2 * SHIFT)))) {
284 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
285 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
286 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
287
288#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
289 aren't tested here */
290 /* The most likely code points in this range are below the surrogates.
291 * Do an extra test to quickly exclude those. */
292 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
293 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
294 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
295 {
33f38593 296 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
297 }
298 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 299 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 300 }
ba6ed43c
KW
301 }
302#endif
303 return d;
304 }
305
306 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
307 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
308 * happen starting with 4-byte characters on ASCII platforms. We unify the
309 * code for these with EBCDIC, even though some of them require 5-bytes on
310 * those, because khw believes the code saving is worth the very slight
311 * performance hit on these high EBCDIC code points. */
312
313 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
24b4c303
KW
314 if (UNLIKELY( uv > MAX_LEGAL_CP
315 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
316 {
fb2f0a6a 317 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv));
a5bf80e0 318 }
33f38593
KW
319 if ( (flags & UNICODE_WARN_SUPER)
320 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 321 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 322 {
33f38593
KW
323 const char * format = super_cp_format;
324 U32 category = packWARN(WARN_NON_UNICODE);
325 U32 flag = UNICODE_GOT_SUPER;
326
327 /* Choose the more dire applicable warning */
328 if (UNICODE_IS_PERL_EXTENDED(uv)) {
8911f9b0 329 format = PL_extended_cp_format;
dc4a6683 330 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
33f38593
KW
331 if (flags & (UNICODE_WARN_PERL_EXTENDED
332 |UNICODE_DISALLOW_PERL_EXTENDED))
333 {
334 flag = UNICODE_GOT_PERL_EXTENDED;
335 }
336 }
a5bf80e0 337
33f38593
KW
338 if (msgs) {
339 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
340 category, flag);
341 }
dc4a6683
KW
342 else if ( ckWARN_d(WARN_NON_UNICODE)
343 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
344 && ckWARN(WARN_PORTABLE)))
345 {
346 Perl_warner(aTHX_ category, format, uv);
33f38593 347 }
a5bf80e0 348 }
56576a04 349 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
350 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
351 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
352 {
353 return NULL;
354 }
355 }
ba6ed43c 356 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 357 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 358 }
d9432125 359
ba6ed43c
KW
360 /* Test for and handle 4-byte result. In the test immediately below, the
361 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
362 * characters). The 3 is for 3 continuation bytes; these each contribute
363 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
364 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
365 * 0x1_0000-0x1F_FFFF on ASCII */
366 if (uv < (8 * (1U << (3 * SHIFT)))) {
367 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
368 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
369 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
370 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
371
372#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
373 characters. The end-plane non-characters for EBCDIC were
374 handled just above */
375 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 376 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 377 }
ba6ed43c 378 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 379 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
380 }
381#endif
382
383 return d;
384 }
385
386 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
387 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
388 * format. The unrolled version above turns out to not save all that much
389 * time, and at these high code points (well above the legal Unicode range
390 * on ASCII platforms, and well above anything in common use in EBCDIC),
391 * khw believes that less code outweighs slight performance gains. */
392
d9432125 393 {
5aaebcb3 394 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
395 U8 *p = d+len-1;
396 while (p > d) {
957a9e81
KW
397 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
398 uv >>= SHIFT;
1d72bdf6 399 }
4c8cd605 400 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
401 return d+len;
402 }
a0ed51b3 403}
a5bf80e0 404
646ca15d 405/*
07693fe6
KW
406=for apidoc uvchr_to_utf8
407
bcb1a2d4 408Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 409of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
410C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
411the byte after the end of the new character. In other words,
07693fe6
KW
412
413 d = uvchr_to_utf8(d, uv);
414
415is the recommended wide native character-aware way of saying
416
417 *(d++) = uv;
418
d22ec717
KW
419This function accepts any code point from 0..C<IV_MAX> as input.
420C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
421
422It is possible to forbid or warn on non-Unicode code points, or those that may
423be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 424
07693fe6
KW
425=cut
426*/
427
de69f3af
KW
428/* This is also a macro */
429PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
430
07693fe6
KW
431U8 *
432Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
433{
de69f3af 434 return uvchr_to_utf8(d, uv);
07693fe6
KW
435}
436
de69f3af
KW
437/*
438=for apidoc uvchr_to_utf8_flags
439
440Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 441of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
442C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
443the byte after the end of the new character. In other words,
de69f3af
KW
444
445 d = uvchr_to_utf8_flags(d, uv, flags);
446
447or, in most cases,
448
449 d = uvchr_to_utf8_flags(d, uv, 0);
450
451This is the Unicode-aware way of saying
452
453 *(d++) = uv;
454
d22ec717
KW
455If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
456input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
457
458Specifying C<flags> can further restrict what is allowed and not warned on, as
459follows:
de69f3af 460
796b6530 461If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
462the function will raise a warning, provided UTF8 warnings are enabled. If
463instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
464NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 465
760c7c2f
KW
466Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
467affect how the function handles a Unicode non-character.
93e6dbd6 468
760c7c2f
KW
469And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
470affect the handling of code points that are above the Unicode maximum of
4710x10FFFF. Languages other than Perl may not be able to accept files that
472contain these.
93e6dbd6
KW
473
474The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
475the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
476three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
477allowed inputs to the strict UTF-8 traditionally defined by Unicode.
478Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
479C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
480above-Unicode and surrogate flags, but not the non-character ones, as
481defined in
e2176993 482L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f 483See L<perlunicode/Noncharacter code points>.
93e6dbd6 484
57ff5f59
KW
485Extremely high code points were never specified in any standard, and require an
486extension to UTF-8 to express, which Perl does. It is likely that programs
487written in something other than Perl would not be able to read files that
488contain these; nor would Perl understand files written by something that uses a
489different extension. For these reasons, there is a separate set of flags that
490can warn and/or disallow these extremely high code points, even if other
491above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
492and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
eb992c6f 493C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
57ff5f59
KW
494treat all above-Unicode code points, including these, as malformations. (Note
495that the Unicode standard considers anything above 0x10FFFF to be illegal, but
496there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
497
498A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
499retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
500C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
501C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
502platforms,these flags can apply to code points that actually do fit in 31 bits.
503The new names accurately describe the situation in all cases.
de69f3af 504
d145625f
KW
505=for apidoc Amnh||UNICODE_DISALLOW_ABOVE_31_BIT
506=for apidoc Amnh||UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE
507=for apidoc Amnh||UNICODE_DISALLOW_ILLEGAL_INTERCHANGE
508=for apidoc Amnh||UNICODE_DISALLOW_NONCHAR
509=for apidoc Amnh||UNICODE_DISALLOW_PERL_EXTENDED
510=for apidoc Amnh||UNICODE_DISALLOW_SUPER
511=for apidoc Amnh||UNICODE_DISALLOW_SURROGATE
512=for apidoc Amnh||UNICODE_WARN_ABOVE_31_BIT
513=for apidoc Amnh||UNICODE_WARN_ILLEGAL_C9_INTERCHANGE
514=for apidoc Amnh||UNICODE_WARN_ILLEGAL_INTERCHANGE
515=for apidoc Amnh||UNICODE_WARN_NONCHAR
516=for apidoc Amnh||UNICODE_WARN_PERL_EXTENDED
517=for apidoc Amnh||UNICODE_WARN_SUPER
518=for apidoc Amnh||UNICODE_WARN_SURROGATE
519
de69f3af
KW
520=cut
521*/
522
523/* This is also a macro */
524PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
525
07693fe6
KW
526U8 *
527Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
528{
de69f3af 529 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
530}
531
57ff5f59
KW
532#ifndef UV_IS_QUAD
533
e050c007
KW
534STATIC int
535S_is_utf8_cp_above_31_bits(const U8 * const s,
536 const U8 * const e,
537 const bool consider_overlongs)
83dc0f42
KW
538{
539 /* Returns TRUE if the first code point represented by the Perl-extended-
540 * UTF-8-encoded string starting at 's', and looking no further than 'e -
541 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
542 *
543 * The function handles the case where the input bytes do not include all
544 * the ones necessary to represent a full character. That is, they may be
545 * the intial bytes of the representation of a code point, but possibly
546 * the final ones necessary for the complete representation may be beyond
547 * 'e - 1'.
548 *
e050c007
KW
549 * The function also can handle the case where the input is an overlong
550 * sequence. If 'consider_overlongs' is 0, the function assumes the
551 * input is not overlong, without checking, and will return based on that
552 * assumption. If this parameter is 1, the function will go to the trouble
553 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 554 *
e050c007 555 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
556 */
557
e050c007
KW
558 const STRLEN len = e - s;
559 int is_overlong;
560
561 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
562
563 assert(! UTF8_IS_INVARIANT(*s) && e > s);
564
83dc0f42
KW
565#ifdef EBCDIC
566
e050c007 567 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 568
e050c007
KW
569 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
570 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
571 * also be the start byte for a 31-bit code point; we need at least 2
572 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
573 * the start byte for an overlong sequence, but for 30-bit or smaller code
574 * points, so we don't have to worry about overlongs on EBCDIC.) */
575 if (*s != 0xFE) {
576 return 0;
577 }
83dc0f42 578
e050c007
KW
579 if (len == 1) {
580 return -1;
581 }
83dc0f42 582
e050c007 583#else
83dc0f42 584
e050c007
KW
585 /* On ASCII, FE and FF are the only start bytes that can evaluate to
586 * needing more than 31 bits. */
587 if (LIKELY(*s < 0xFE)) {
588 return 0;
589 }
83dc0f42 590
e050c007
KW
591 /* What we have left are FE and FF. Both of these require more than 31
592 * bits unless they are for overlongs. */
593 if (! consider_overlongs) {
594 return 1;
595 }
83dc0f42 596
e050c007
KW
597 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
598 * above 31 bits. But we need more than one byte to discern this, so if
599 * passed just the start byte, it could be an overlong evaluating to
600 * smaller */
601 if (len == 1) {
602 return -1;
603 }
83dc0f42 604
e050c007
KW
605 /* Having excluded len==1, and knowing that FE and FF are both valid start
606 * bytes, we can call the function below to see if the sequence is
607 * overlong. (We don't need the full generality of the called function,
608 * but for these huge code points, speed shouldn't be a consideration, and
609 * the compiler does have enough information, since it's static to this
610 * file, to optimize to just the needed parts.) */
611 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 612
e050c007
KW
613 /* If it isn't overlong, more than 31 bits are required. */
614 if (is_overlong == 0) {
615 return 1;
616 }
83dc0f42 617
e050c007
KW
618 /* If it is indeterminate if it is overlong, return that */
619 if (is_overlong < 0) {
620 return -1;
621 }
622
623 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
624 * the max it can be is 2**31 - 1 */
625 if (*s == 0xFE) {
626 return 0;
83dc0f42
KW
627 }
628
e050c007
KW
629#endif
630
631 /* Here, ASCII and EBCDIC rejoin:
632 * On ASCII: We have an overlong sequence starting with FF
633 * On EBCDIC: We have a sequence starting with FE. */
634
635 { /* For C89, use a block so the declaration can be close to its use */
636
637#ifdef EBCDIC
638
5f995336
KW
639 /* U+7FFFFFFF (2 ** 31 - 1)
640 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
641 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
642 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
643 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
644 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
645 * U+80000000 (2 ** 31):
646 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
647 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
648 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
649 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
650 *
651 * and since we know that *s = \xfe, any continuation sequcence
652 * following it that is gt the below is above 31 bits
653 [0] [1] [2] [3] [4] [5] [6] */
654 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
655
656#else
657
658 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
659 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
660 * FF overlong for U+80000000 (2 ** 31):
661 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
662 * and since we know that *s = \xff, any continuation sequcence
663 * following it that is gt the below is above 30 bits
664 [0] [1] [2] [3] [4] [5] [6] */
665 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 666
83dc0f42
KW
667
668#endif
e050c007
KW
669 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
670 const STRLEN cmp_len = MIN(conts_len, len - 1);
671
672 /* Now compare the continuation bytes in s with the ones we have
673 * compiled in that are for the largest 30 bit code point. If we have
674 * enough bytes available to determine the answer, or the bytes we do
675 * have differ from them, we can compare the two to get a definitive
676 * answer (Note that in UTF-EBCDIC, the two lowest possible
677 * continuation bytes are \x41 and \x42.) */
678 if (cmp_len >= conts_len || memNE(s + 1,
679 conts_for_highest_30_bit,
680 cmp_len))
681 {
682 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
683 }
83dc0f42 684
e050c007
KW
685 /* Here, all the bytes we have are the same as the highest 30-bit code
686 * point, but we are missing so many bytes that we can't make the
687 * determination */
688 return -1;
689 }
83dc0f42
KW
690}
691
57ff5f59
KW
692#endif
693
d6be65ae 694PERL_STATIC_INLINE int
12a4bed3
KW
695S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
696{
d6be65ae
KW
697 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
698 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
699 * it isn't, and -1 if there isn't enough information to tell. This last
700 * return value can happen if the sequence is incomplete, missing some
701 * trailing bytes that would form a complete character. If there are
702 * enough bytes to make a definitive decision, this function does so.
703 * Usually 2 bytes sufficient.
704 *
705 * Overlongs can occur whenever the number of continuation bytes changes.
706 * That means whenever the number of leading 1 bits in a start byte
707 * increases from the next lower start byte. That happens for start bytes
708 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
709 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
710 * ASCII platforms: C0, C1
711 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 712 */
12a4bed3
KW
713
714 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
715 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
716
717 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
718 assert(len > 1 && UTF8_IS_START(*s));
719
720 /* Each platform has overlongs after the start bytes given above (expressed
721 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
722 * the logic is the same, except the E0 overlong has already been excluded
723 * on EBCDIC platforms. The values below were found by manually
724 * inspecting the UTF-8 patterns. See the tables in utf8.h and
725 * utfebcdic.h. */
726
727# ifdef EBCDIC
728# define F0_ABOVE_OVERLONG 0xB0
729# define F8_ABOVE_OVERLONG 0xA8
730# define FC_ABOVE_OVERLONG 0xA4
731# define FE_ABOVE_OVERLONG 0xA2
732# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
733 /* I8(0xfe) is FF */
734# else
735
736 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 737 return 1;
12a4bed3
KW
738 }
739
740# define F0_ABOVE_OVERLONG 0x90
741# define F8_ABOVE_OVERLONG 0x88
742# define FC_ABOVE_OVERLONG 0x84
743# define FE_ABOVE_OVERLONG 0x82
744# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
745# endif
746
747
748 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
749 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
750 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
751 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
752 {
d6be65ae 753 return 1;
12a4bed3
KW
754 }
755
b0b342d4 756 /* Check for the FF overlong */
d6be65ae 757 return isFF_OVERLONG(s, len);
b0b342d4
KW
758}
759
8d6204cc 760PERL_STATIC_INLINE int
b0b342d4
KW
761S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
762{
8d6204cc
KW
763 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
764 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
765 * it isn't, and -1 if there isn't enough information to tell. This last
766 * return value can happen if the sequence is incomplete, missing some
767 * trailing bytes that would form a complete character. If there are
768 * enough bytes to make a definitive decision, this function does so. */
769
b0b342d4 770 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 771
8d6204cc
KW
772 /* To be an FF overlong, all the available bytes must match */
773 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
774 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
775 {
776 return 0;
777 }
778
779 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
780 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 781 * utfebcdic.h. */
8d6204cc
KW
782 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
783 return 1;
784 }
12a4bed3 785
8d6204cc
KW
786 /* The missing bytes could cause the result to go one way or the other, so
787 * the result is indeterminate */
788 return -1;
12a4bed3
KW
789}
790
d22ec717 791#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
792# ifdef EBCDIC /* Actually is I8 */
793# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 794 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
795# else
796# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 797 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
798# endif
799#endif
800
c285bbc4 801PERL_STATIC_INLINE int
e050c007
KW
802S_does_utf8_overflow(const U8 * const s,
803 const U8 * e,
804 const bool consider_overlongs)
a77c906e 805{
c285bbc4 806 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
807 * 'e' - 1 would overflow an IV on this platform; that is if it represents
808 * a code point larger than the highest representable code point. It
809 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
810 * enough information to tell. This last return value can happen if the
811 * sequence is incomplete, missing some trailing bytes that would form a
812 * complete character. If there are enough bytes to make a definitive
813 * decision, this function does so.
c285bbc4 814 *
e050c007
KW
815 * If 'consider_overlongs' is TRUE, the function checks for the possibility
816 * that the sequence is an overlong that doesn't overflow. Otherwise, it
817 * assumes the sequence is not an overlong. This can give different
818 * results only on ASCII 32-bit platforms.
819 *
c285bbc4
KW
820 * (For ASCII platforms, we could use memcmp() because we don't have to
821 * convert each byte to I8, but it's very rare input indeed that would
822 * approach overflow, so the loop below will likely only get executed once.)
823 *
824 * 'e' - 1 must not be beyond a full character. */
a77c906e 825
a77c906e
KW
826
827 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
828 assert(s <= e && s + UTF8SKIP(s) >= e);
829
d22ec717
KW
830#if ! defined(UV_IS_QUAD)
831
832 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
833
834#else
835
836 PERL_UNUSED_ARG(consider_overlongs);
837
838 {
839 const STRLEN len = e - s;
840 const U8 *x;
841 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
842
843 for (x = s; x < e; x++, y++) {
844
845 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
846 continue;
847 }
848
849 /* If this byte is larger than the corresponding highest UTF-8
850 * byte, the sequence overflow; otherwise the byte is less than,
851 * and so the sequence doesn't overflow */
852 return NATIVE_UTF8_TO_I8(*x) > *y;
853
854 }
855
856 /* Got to the end and all bytes are the same. If the input is a whole
857 * character, it doesn't overflow. And if it is a partial character,
858 * there's not enough information to tell */
859 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
860 return -1;
861 }
862
863 return 0;
864 }
865
866#endif
867
868}
869
870#if 0
871
872/* This is the portions of the above function that deal with UV_MAX instead of
873 * IV_MAX. They are left here in case we want to combine them so that internal
874 * uses can have larger code points. The only logic difference is that the
875 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
876 * different logic.
877 */
878
879/* Anything larger than this will overflow the word if it were converted into a UV */
880#if defined(UV_IS_QUAD)
881# ifdef EBCDIC /* Actually is I8 */
882# define HIGHEST_REPRESENTABLE_UTF8 \
883 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
884# else
885# define HIGHEST_REPRESENTABLE_UTF8 \
886 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
887# endif
888#else /* 32-bit */
889# ifdef EBCDIC
890# define HIGHEST_REPRESENTABLE_UTF8 \
891 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
892# else
893# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
894# endif
895#endif
896
a77c906e
KW
897#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
898
899 /* On 32 bit ASCII machines, many overlongs that start with FF don't
900 * overflow */
e050c007 901 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
902
903 /* To be such an overlong, the first bytes of 's' must match
904 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
905 * don't have any additional bytes available, the sequence, when
906 * completed might or might not fit in 32 bits. But if we have that
907 * next byte, we can tell for sure. If it is <= 0x83, then it does
908 * fit. */
909 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
910 return -1;
911 }
912
913 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
914 }
915
d22ec717
KW
916/* Starting with the #else, the rest of the function is identical except
917 * 1. we need to move the 'len' declaration to be global to the function
918 * 2. the endif move to just after the UNUSED_ARG.
919 * An empty endif is given just below to satisfy the preprocessor
920 */
a77c906e
KW
921#endif
922
d22ec717 923#endif
a77c906e 924
12a4bed3
KW
925#undef F0_ABOVE_OVERLONG
926#undef F8_ABOVE_OVERLONG
927#undef FC_ABOVE_OVERLONG
928#undef FE_ABOVE_OVERLONG
929#undef FF_OVERLONG_PREFIX
930
35f8c9bd 931STRLEN
1376b35c 932Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 933{
2b479609 934 STRLEN len;
12a4bed3 935 const U8 *x;
35f8c9bd 936
2b479609
KW
937 /* A helper function that should not be called directly.
938 *
939 * This function returns non-zero if the string beginning at 's' and
940 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
941 * code point; otherwise it returns 0. The examination stops after the
942 * first code point in 's' is validated, not looking at the rest of the
943 * input. If 'e' is such that there are not enough bytes to represent a
944 * complete code point, this function will return non-zero anyway, if the
945 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
946 * excluded by 'flags'.
947 *
948 * A non-zero return gives the number of bytes required to represent the
949 * code point. Be aware that if the input is for a partial character, the
950 * return will be larger than 'e - s'.
951 *
952 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
953 * The caller should have excluded the possibility of it being invariant
954 * before calling this function.
2b479609
KW
955 *
956 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
957 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
958 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
959 * disallowed by the flags. If the input is only for a partial character,
960 * the function will return non-zero if there is any sequence of
961 * well-formed UTF-8 that, when appended to the input sequence, could
962 * result in an allowed code point; otherwise it returns 0. Non characters
963 * cannot be determined based on partial character input. But many of the
964 * other excluded types can be determined with just the first one or two
965 * bytes.
966 *
967 */
968
1376b35c 969 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
2b479609
KW
970
971 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 972 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 973 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 974
2b479609 975 /* A variant char must begin with a start byte */
35f8c9bd
KW
976 if (UNLIKELY(! UTF8_IS_START(*s))) {
977 return 0;
978 }
979
edc2c47a
KW
980 /* Examine a maximum of a single whole code point */
981 if (e - s > UTF8SKIP(s)) {
982 e = s + UTF8SKIP(s);
983 }
984
2b479609
KW
985 len = e - s;
986
987 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
988 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 989
56576a04
KW
990 /* Here, we are disallowing some set of largish code points, and the
991 * first byte indicates the sequence is for a code point that could be
992 * in the excluded set. We generally don't have to look beyond this or
993 * the second byte to see if the sequence is actually for one of the
994 * excluded classes. The code below is derived from this table:
995 *
2b479609
KW
996 * UTF-8 UTF-EBCDIC I8
997 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
998 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
999 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
1000 *
56576a04
KW
1001 * Keep in mind that legal continuation bytes range between \x80..\xBF
1002 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
1003 * continuation bytes. Hence, we don't have to test the upper edge
1004 * because if any of those is encountered, the sequence is malformed,
1005 * and would fail elsewhere in this function.
1006 *
1007 * The code here likewise assumes that there aren't other
1008 * malformations; again the function should fail elsewhere because of
1009 * these. For example, an overlong beginning with FC doesn't actually
1010 * have to be a super; it could actually represent a small code point,
1011 * even U+0000. But, since overlongs (and other malformations) are
1012 * illegal, the function should return FALSE in either case.
2b479609
KW
1013 */
1014
1015#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1016# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 1017# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 1018
19794540
KW
1019# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1020 /* B6 and B7 */ \
1021 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1022# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1023#else
1024# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1025# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1026# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1027# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1028#endif
1029
1030 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1031 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1032 {
2b479609
KW
1033 return 0; /* Above Unicode */
1034 }
1035
d044b7a7 1036 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1037 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1038 {
57ff5f59 1039 return 0;
2b479609
KW
1040 }
1041
1042 if (len > 1) {
1043 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1044
1045 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1046 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1047 {
1048 return 0; /* Above Unicode */
1049 }
1050
1051 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1052 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1053 {
1054 return 0; /* Surrogate */
1055 }
1056
1057 if ( (flags & UTF8_DISALLOW_NONCHAR)
1058 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1059 {
1060 return 0; /* Noncharacter code point */
1061 }
1062 }
1063 }
1064
1065 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1066 for (x = s + 1; x < e; x++) {
1067 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1068 return 0;
1069 }
1070 }
1071
af13dd8a 1072 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1073 * overlong. */
d6be65ae 1074 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1075 return 0;
af13dd8a
KW
1076 }
1077
12a4bed3
KW
1078 /* And finally, that the code point represented fits in a word on this
1079 * platform */
e050c007
KW
1080 if (0 < does_utf8_overflow(s, e,
1081 0 /* Don't consider overlongs */
1082 ))
1083 {
12a4bed3 1084 return 0;
35f8c9bd
KW
1085 }
1086
2b479609 1087 return UTF8SKIP(s);
35f8c9bd
KW
1088}
1089
7e2f38b2 1090char *
63ab03b3 1091Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1092{
1093 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1094 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1095 * Currently, there are only two formats, so it is currently a bool:
1096 * 0 \xab
1097 * 1 ab (that is a space between two hex digit bytes)
1098 */
7cf8d05d
KW
1099
1100 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1101 trailing NUL */
63ab03b3
KW
1102 const U8 * s = start;
1103 const U8 * const e = start + len;
7cf8d05d
KW
1104 char * output;
1105 char * d;
1106
1107 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1108
1109 Newx(output, output_len, char);
1110 SAVEFREEPV(output);
1111
1112 d = output;
63ab03b3 1113 for (s = start; s < e; s++) {
7cf8d05d
KW
1114 const unsigned high_nibble = (*s & 0xF0) >> 4;
1115 const unsigned low_nibble = (*s & 0x0F);
1116
7e2f38b2 1117 if (format) {
63ab03b3
KW
1118 if (s > start) {
1119 *d++ = ' ';
1120 }
7e2f38b2
KW
1121 }
1122 else {
1123 *d++ = '\\';
1124 *d++ = 'x';
1125 }
7cf8d05d
KW
1126
1127 if (high_nibble < 10) {
1128 *d++ = high_nibble + '0';
1129 }
1130 else {
1131 *d++ = high_nibble - 10 + 'a';
1132 }
1133
1134 if (low_nibble < 10) {
1135 *d++ = low_nibble + '0';
1136 }
1137 else {
1138 *d++ = low_nibble - 10 + 'a';
1139 }
1140 }
1141
1142 *d = '\0';
1143 return output;
1144}
1145
806547a7 1146PERL_STATIC_INLINE char *
7cf8d05d
KW
1147S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1148
421da25c 1149 /* Max number of bytes to print */
3cc6a05e 1150 STRLEN print_len,
7cf8d05d
KW
1151
1152 /* Which one is the non-continuation */
1153 const STRLEN non_cont_byte_pos,
1154
1155 /* How many bytes should there be? */
1156 const STRLEN expect_len)
806547a7
KW
1157{
1158 /* Return the malformation warning text for an unexpected continuation
1159 * byte. */
1160
7cf8d05d 1161 const char * const where = (non_cont_byte_pos == 1)
806547a7 1162 ? "immediately"
7cf8d05d
KW
1163 : Perl_form(aTHX_ "%d bytes",
1164 (int) non_cont_byte_pos);
421da25c
KW
1165 const U8 * x = s + non_cont_byte_pos;
1166 const U8 * e = s + print_len;
806547a7
KW
1167
1168 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1169
7cf8d05d
KW
1170 /* We don't need to pass this parameter, but since it has already been
1171 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1172 assert(expect_len == UTF8SKIP(s));
1173
421da25c
KW
1174 /* As a defensive coding measure, don't output anything past a NUL. Such
1175 * bytes shouldn't be in the middle of a malformation, and could mark the
1176 * end of the allocated string, and what comes after is undefined */
1177 for (; x < e; x++) {
1178 if (*x == '\0') {
1179 x++; /* Output this particular NUL */
1180 break;
1181 }
1182 }
1183
7cf8d05d
KW
1184 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1185 " %s after start byte 0x%02x; need %d bytes, got %d)",
1186 malformed_text,
421da25c 1187 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1188 *(s + non_cont_byte_pos),
1189 where,
1190 *s,
1191 (int) expect_len,
1192 (int) non_cont_byte_pos);
806547a7
KW
1193}
1194
35f8c9bd
KW
1195/*
1196
de69f3af 1197=for apidoc utf8n_to_uvchr
378516de
KW
1198
1199THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1200Most code should use L</utf8_to_uvchr_buf>() rather than call this
1201directly.
67e989fb 1202
9041c2e3 1203Bottom level UTF-8 decode routine.
de69f3af 1204Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1205which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1206C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1207the length, in bytes, of that character.
949cf498
KW
1208
1209The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1210well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1211causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1212is the next possible position in C<s> that could begin a non-malformed
1213character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1214is raised. Some UTF-8 input sequences may contain multiple malformations.
1215This function tries to find every possible one in each call, so multiple
56576a04 1216warnings can be raised for the same sequence.
949cf498
KW
1217
1218Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1219individual types of malformations, such as the sequence being overlong (that
1220is, when there is a shorter sequence that can express the same code point;
1221overlong sequences are expressly forbidden in the UTF-8 standard due to
1222potential security issues). Another malformation example is the first byte of
1223a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1224flags. Even if allowed, this function generally returns the Unicode
1225REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1226F<utf8.h> to override this behavior for the overlong malformations, but don't
1227do that except for very specialized purposes.
949cf498 1228
796b6530 1229The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1230flags) malformation is found. If this flag is set, the routine assumes that
1231the caller will raise a warning, and this function will silently just set
d088425d
KW
1232C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1233
75200dff 1234Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1235character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1236in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1237be set to 1. To disambiguate, upon a zero return, see if the first byte of
1238C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1239error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1240
1241Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1242Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1243By default these are considered regular code points, but certain situations
ecc1615f
KW
1244warrant special handling for them, which can be specified using the C<flags>
1245parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1246three classes are treated as malformations and handled as such. The flags
1247C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1248C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1249disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1250restricts the allowed inputs to the strict UTF-8 traditionally defined by
1251Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1252definition given by
e2176993 1253L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f
KW
1254The difference between traditional strictness and C9 strictness is that the
1255latter does not forbid non-character code points. (They are still discouraged,
1256however.) For more discussion see L<perlunicode/Noncharacter code points>.
1257
1258The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1259C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1260C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1261raised for their respective categories, but otherwise the code points are
1262considered valid (not malformations). To get a category to both be treated as
1263a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1264(But note that warnings are not raised if lexically disabled nor if
796b6530 1265C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1266
57ff5f59
KW
1267Extremely high code points were never specified in any standard, and require an
1268extension to UTF-8 to express, which Perl does. It is likely that programs
1269written in something other than Perl would not be able to read files that
1270contain these; nor would Perl understand files written by something that uses a
1271different extension. For these reasons, there is a separate set of flags that
1272can warn and/or disallow these extremely high code points, even if other
1273above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1274C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
eb992c6f 1275C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
57ff5f59
KW
1276above-Unicode code points, including these, as malformations.
1277(Note that the Unicode standard considers anything above 0x10FFFF to be
1278illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1279(2**31 -1))
1280
1281A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1282retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1283C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1284C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1285can apply to code points that actually do fit in 31 bits. This happens on
1286EBCDIC platforms, and sometimes when the L<overlong
1287malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1288describe the situation in all cases.
1289
ab8e6d41 1290
949cf498
KW
1291All other code points corresponding to Unicode characters, including private
1292use and those yet to be assigned, are never considered malformed and never
1293warn.
67e989fb 1294
5af38e47
KW
1295=for apidoc Amnh||UTF8_CHECK_ONLY
1296=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1297=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1298=for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1299=for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1300=for apidoc Amnh||UTF8_DISALLOW_SUPER
1301=for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1302=for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1303=for apidoc Amnh||UTF8_WARN_SURROGATE
1304=for apidoc Amnh||UTF8_WARN_NONCHAR
1305=for apidoc Amnh||UTF8_WARN_SUPER
1306=for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1307=for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1308
37607a96 1309=cut
f9380377
KW
1310
1311Also implemented as a macro in utf8.h
1312*/
1313
1314UV
e6a4ffc3
KW
1315Perl_utf8n_to_uvchr(const U8 *s,
1316 STRLEN curlen,
1317 STRLEN *retlen,
1318 const U32 flags)
f9380377
KW
1319{
1320 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1321
1322 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1323}
1324
1325/*
1326
1327=for apidoc utf8n_to_uvchr_error
1328
1329THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1330Most code should use L</utf8_to_uvchr_buf>() rather than call this
1331directly.
f9380377
KW
1332
1333This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1334are when an error is found. If you also need to know the generated warning
1335messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1336
1337It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1338all the others, C<errors>. If this parameter is 0, this function behaves
1339identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1340to a C<U32> variable, which this function sets to indicate any errors found.
1341Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1342C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1343of these bits will be set if a malformation is found, even if the input
7a65503b 1344C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1345exceptions are noted:
1346
1347=over 4
1348
57ff5f59 1349=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1350
57ff5f59
KW
1351The input sequence is not standard UTF-8, but a Perl extension. This bit is
1352set only if the input C<flags> parameter contains either the
1353C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1354
1355Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1356and so some extension must be used to express them. Perl uses a natural
1357extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1358extension to represent even higher ones, so that any code point that fits in a
135964-bit word can be represented. Text using these extensions is not likely to
1360be portable to non-Perl code. We lump both of these extensions together and
1361refer to them as Perl extended UTF-8. There exist other extensions that people
1362have invented, incompatible with Perl's.
1363
1364On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1365extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1366than on ASCII. Prior to that, code points 2**31 and higher were simply
1367unrepresentable, and a different, incompatible method was used to represent
1368code points between 2**30 and 2**31 - 1.
1369
1370On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1371Perl extended UTF-8 is used.
1372
1373In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1374may use for backward compatibility. That name is misleading, as this flag may
1375be set when the code point actually does fit in 31 bits. This happens on
1376EBCDIC platforms, and sometimes when the L<overlong
1377malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1378describes the situation in all cases.
f9380377
KW
1379
1380=item C<UTF8_GOT_CONTINUATION>
1381
a3815e44 1382The input sequence was malformed in that the first byte was a UTF-8
f9380377
KW
1383continuation byte.
1384
1385=item C<UTF8_GOT_EMPTY>
1386
1387The input C<curlen> parameter was 0.
1388
1389=item C<UTF8_GOT_LONG>
1390
1391The input sequence was malformed in that there is some other sequence that
1392evaluates to the same code point, but that sequence is shorter than this one.
1393
fecaf136
KW
1394Until Unicode 3.1, it was legal for programs to accept this malformation, but
1395it was discovered that this created security issues.
1396
f9380377
KW
1397=item C<UTF8_GOT_NONCHAR>
1398
1399The code point represented by the input UTF-8 sequence is for a Unicode
1400non-character code point.
1401This bit is set only if the input C<flags> parameter contains either the
1402C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1403
1404=item C<UTF8_GOT_NON_CONTINUATION>
1405
1406The input sequence was malformed in that a non-continuation type byte was found
00d976bb 1407in a position where only a continuation type one should be. See also
eb992c6f 1408C<L</UTF8_GOT_SHORT>>.
f9380377
KW
1409
1410=item C<UTF8_GOT_OVERFLOW>
1411
1412The input sequence was malformed in that it is for a code point that is not
d22ec717 1413representable in the number of bits available in an IV on the current platform.
f9380377
KW
1414
1415=item C<UTF8_GOT_SHORT>
1416
1417The input sequence was malformed in that C<curlen> is smaller than required for
1418a complete sequence. In other words, the input is for a partial character
1419sequence.
1420
00d976bb
KW
1421
1422C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1423sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1424that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1425sequence was looked at. If no other flags are present, it means that the
1426sequence was valid as far as it went. Depending on the application, this could
1427mean one of three things:
1428
1429=over
1430
1431=item *
1432
1433The C<curlen> length parameter passed in was too small, and the function was
1434prevented from examining all the necessary bytes.
1435
1436=item *
1437
1438The buffer being looked at is based on reading data, and the data received so
1439far stopped in the middle of a character, so that the next read will
1440read the remainder of this character. (It is up to the caller to deal with the
1441split bytes somehow.)
1442
1443=item *
1444
1445This is a real error, and the partial sequence is all we're going to get.
1446
1447=back
1448
f9380377
KW
1449=item C<UTF8_GOT_SUPER>
1450
1451The input sequence was malformed in that it is for a non-Unicode code point;
1452that is, one above the legal Unicode maximum.
1453This bit is set only if the input C<flags> parameter contains either the
1454C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1455
1456=item C<UTF8_GOT_SURROGATE>
1457
1458The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1459code point.
1460This bit is set only if the input C<flags> parameter contains either the
1461C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1462
1463=back
1464
133551d8
KW
1465To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1466flag to suppress any warnings, and then examine the C<*errors> return.
1467
d145625f
KW
1468=for apidoc Amnh||UTF8_GOT_PERL_EXTENDED
1469=for apidoc Amnh||UTF8_GOT_CONTINUATION
1470=for apidoc Amnh||UTF8_GOT_EMPTY
1471=for apidoc Amnh||UTF8_GOT_LONG
1472=for apidoc Amnh||UTF8_GOT_NONCHAR
1473=for apidoc Amnh||UTF8_GOT_NON_CONTINUATION
1474=for apidoc Amnh||UTF8_GOT_OVERFLOW
1475=for apidoc Amnh||UTF8_GOT_SHORT
1476=for apidoc Amnh||UTF8_GOT_SUPER
1477=for apidoc Amnh||UTF8_GOT_SURROGATE
1478
f9380377 1479=cut
37657a5b
KW
1480
1481Also implemented as a macro in utf8.h
37607a96 1482*/
67e989fb 1483
a0ed51b3 1484UV
e6a4ffc3 1485Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1486 STRLEN curlen,
1487 STRLEN *retlen,
1488 const U32 flags,
1489 U32 * errors)
1490{
1491 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1492
1493 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1494}
1495
1496/*
1497
1498=for apidoc utf8n_to_uvchr_msgs
1499
1500THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1501Most code should use L</utf8_to_uvchr_buf>() rather than call this
1502directly.
37657a5b
KW
1503
1504This function is for code that needs to know what the precise malformation(s)
1505are when an error is found, and wants the corresponding warning and/or error
1506messages to be returned to the caller rather than be displayed. All messages
f1460a66 1507that would have been displayed if all lexical warnings are enabled will be
37657a5b
KW
1508returned.
1509
1510It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1511placed after all the others, C<msgs>. If this parameter is 0, this function
1512behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1513be a pointer to an C<AV *> variable, in which this function creates a new AV to
1514contain any appropriate messages. The elements of the array are ordered so
1515that the first message that would have been displayed is in the 0th element,
1516and so on. Each element is a hash with three key-value pairs, as follows:
1517
1518=over 4
1519
1520=item C<text>
1521
1522The text of the message as a C<SVpv>.
1523
1524=item C<warn_categories>
1525
1526The warning category (or categories) packed into a C<SVuv>.
1527
1528=item C<flag>
1529
1530A single flag bit associated with this message, in a C<SVuv>.
1531The bit corresponds to some bit in the C<*errors> return value,
1532such as C<UTF8_GOT_LONG>.
1533
1534=back
1535
1536It's important to note that specifying this parameter as non-null will cause
1537any warnings this function would otherwise generate to be suppressed, and
1538instead be placed in C<*msgs>. The caller can check the lexical warnings state
1539(or not) when choosing what to do with the returned messages.
1540
1541If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1542no AV is created.
1543
1544The caller, of course, is responsible for freeing any returned AV.
1545
1546=cut
1547*/
1548
1549UV
e6a4ffc3 1550Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1551 STRLEN curlen,
1552 STRLEN *retlen,
1553 const U32 flags,
1554 U32 * errors,
1555 AV ** msgs)
a0ed51b3 1556{
d4c19fe8 1557 const U8 * const s0 = s;
2b9519f0 1558 const U8 * send = s0 + curlen;
5af9f822
KW
1559 U32 possible_problems; /* A bit is set here for each potential problem
1560 found as we go along */
1561 UV uv;
1562 STRLEN expectlen; /* How long should this sequence be? */
1563 STRLEN avail_len; /* When input is too short, gives what that is */
1564 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1565 gets set and discarded */
a0dbb045 1566
2b5e7bc2
KW
1567 /* The below are used only if there is both an overlong malformation and a
1568 * too short one. Otherwise the first two are set to 's0' and 'send', and
1569 * the third not used at all */
5af9f822 1570 U8 * adjusted_s0;
e9f2c446
KW
1571 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1572 routine; see [perl #130921] */
5af9f822 1573 UV uv_so_far;
e6a4ffc3 1574 dTHX;
5af9f822 1575
e6a4ffc3 1576 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1577
1578 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1579 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1580 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1581 * final case. */
1582
1583#ifndef EBCDIC
1584
1585 /* Each of the affected Hanguls starts with \xED */
1586
1587 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1588 if (retlen) {
1589 *retlen = 3;
1590 }
1591 if (errors) {
1592 *errors = 0;
1593 }
1594 if (msgs) {
1595 *msgs = NULL;
1596 }
1597
1598 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1599 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1600 | (s0[2] & UTF_CONTINUATION_MASK);
1601 }
1602
1603#endif
1604
1605 /* In conjunction with the exhaustive tests that can be enabled in
1606 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1607 * what it is intended to do, and that no flaws in it are masked by
1608 * dropping down and executing the code below
1609 assert(! isUTF8_CHAR(s0, send)
1610 || UTF8_IS_SURROGATE(s0, send)
1611 || UTF8_IS_SUPER(s0, send)
1612 || UTF8_IS_NONCHAR(s0,send));
1613 */
1614
1615 s = s0;
1616 uv = *s0;
1617 possible_problems = 0;
1618 expectlen = 0;
1619 avail_len = 0;
1620 discard_errors = 0;
1621 adjusted_s0 = (U8 *) s0;
1622 uv_so_far = 0;
1623
f9380377
KW
1624 if (errors) {
1625 *errors = 0;
1626 }
1627 else {
1628 errors = &discard_errors;
1629 }
a0dbb045 1630
eb83ed87
KW
1631 /* The order of malformation tests here is important. We should consume as
1632 * few bytes as possible in order to not skip any valid character. This is
1633 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
e2176993 1634 * https://unicode.org/reports/tr36 for more discussion as to why. For
eb83ed87
KW
1635 * example, once we've done a UTF8SKIP, we can tell the expected number of
1636 * bytes, and could fail right off the bat if the input parameters indicate
1637 * that there are too few available. But it could be that just that first
1638 * byte is garbled, and the intended character occupies fewer bytes. If we
1639 * blindly assumed that the first byte is correct, and skipped based on
1640 * that number, we could skip over a valid input character. So instead, we
1641 * always examine the sequence byte-by-byte.
1642 *
1643 * We also should not consume too few bytes, otherwise someone could inject
1644 * things. For example, an input could be deliberately designed to
1645 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1646 * returning to the caller C<*retlen> pointing to the very next byte (one
a3815e44 1647 * which is actually part of the overflowing sequence), that could look
eb83ed87 1648 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1649 * sequence and process the rest, inappropriately.
1650 *
1651 * Some possible input sequences are malformed in more than one way. This
1652 * function goes to lengths to try to find all of them. This is necessary
1653 * for correctness, as the inputs may allow one malformation but not
1654 * another, and if we abandon searching for others after finding the
1655 * allowed one, we could allow in something that shouldn't have been.
1656 */
eb83ed87 1657
b5b9af04 1658 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1659 possible_problems |= UTF8_GOT_EMPTY;
1660 curlen = 0;
5a48568d 1661 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1662 goto ready_to_handle_errors;
0c443dc2
JH
1663 }
1664
eb83ed87
KW
1665 expectlen = UTF8SKIP(s);
1666
1667 /* A well-formed UTF-8 character, as the vast majority of calls to this
1668 * function will be for, has this expected length. For efficiency, set
1669 * things up here to return it. It will be overriden only in those rare
1670 * cases where a malformation is found */
1671 if (retlen) {
1672 *retlen = expectlen;
1673 }
1674
eb83ed87 1675 /* A continuation character can't start a valid sequence */
b5b9af04 1676 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1677 possible_problems |= UTF8_GOT_CONTINUATION;
1678 curlen = 1;
1679 uv = UNICODE_REPLACEMENT;
1680 goto ready_to_handle_errors;
ba210ebe 1681 }
9041c2e3 1682
dcd27b3c 1683 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1684 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1685 * because it excludes start bytes like \xC0 that always lead to
1686 * overlongs.) */
dcd27b3c 1687
534752c1
KW
1688 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1689 * that indicate the number of bytes in the character's whole UTF-8
1690 * sequence, leaving just the bits that are part of the value. */
1691 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1692
e308b348
KW
1693 /* Setup the loop end point, making sure to not look past the end of the
1694 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1695 if (UNLIKELY(curlen < expectlen)) {
1696 possible_problems |= UTF8_GOT_SHORT;
1697 avail_len = curlen;
e308b348
KW
1698 }
1699 else {
2b9519f0 1700 send = (U8*) s0 + expectlen;
e308b348 1701 }
e308b348 1702
eb83ed87 1703 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1704 * accumulating each into the working value as we go. */
eb83ed87 1705 for (s = s0 + 1; s < send; s++) {
b5b9af04 1706 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1707 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1708 continue;
1709 }
1710
1711 /* Here, found a non-continuation before processing all expected bytes.
1712 * This byte indicates the beginning of a new character, so quit, even
1713 * if allowing this malformation. */
2b5e7bc2 1714 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1715 break;
eb83ed87
KW
1716 } /* End of loop through the character's bytes */
1717
1718 /* Save how many bytes were actually in the character */
1719 curlen = s - s0;
1720
2b5e7bc2
KW
1721 /* Note that there are two types of too-short malformation. One is when
1722 * there is actual wrong data before the normal termination of the
1723 * sequence. The other is that the sequence wasn't complete before the end
1724 * of the data we are allowed to look at, based on the input 'curlen'.
1725 * This means that we were passed data for a partial character, but it is
1726 * valid as far as we saw. The other is definitely invalid. This
1727 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1728 * separate.
1729 *
1730 * A convenience macro that matches either of the too-short conditions. */
1731# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1732
1733 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1734 uv_so_far = uv;
1735 uv = UNICODE_REPLACEMENT;
1736 }
2b5e7bc2 1737
08e73697
KW
1738 /* Check for overflow. The algorithm requires us to not look past the end
1739 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1740 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1741 1 /* Do consider overlongs */
1742 )))
1743 {
2b5e7bc2
KW
1744 possible_problems |= UTF8_GOT_OVERFLOW;
1745 uv = UNICODE_REPLACEMENT;
eb83ed87 1746 }
eb83ed87 1747
2b5e7bc2
KW
1748 /* Check for overlong. If no problems so far, 'uv' is the correct code
1749 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1750 * we must look at the UTF-8 byte sequence itself to see if it is for an
1751 * overlong */
1752 if ( ( LIKELY(! possible_problems)
1753 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1754 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1755 && ( UNLIKELY(! UTF8_IS_START(*s0))
1756 || ( curlen > 1
d6be65ae 1757 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1758 s - s0))))))
2f8f112e 1759 {
2b5e7bc2
KW
1760 possible_problems |= UTF8_GOT_LONG;
1761
abc28b54 1762 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1763
abc28b54
KW
1764 /* The calculation in the 'true' branch of this 'if'
1765 * below won't work if overflows, and isn't needed
1766 * anyway. Further below we handle all overflow
1767 * cases */
1768 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1769 {
2b5e7bc2
KW
1770 UV min_uv = uv_so_far;
1771 STRLEN i;
1772
1773 /* Here, the input is both overlong and is missing some trailing
1774 * bytes. There is no single code point it could be for, but there
1775 * may be enough information present to determine if what we have
1776 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1777 * The code further below has the intelligence to determine this,
1778 * but just for non-overlong UTF-8 sequences. What we do here is
1779 * calculate the smallest code point the input could represent if
1780 * there were no too short malformation. Then we compute and save
1781 * the UTF-8 for that, which is what the code below looks at
1782 * instead of the raw input. It turns out that the smallest such
1783 * code point is all we need. */
2b5e7bc2
KW
1784 for (i = curlen; i < expectlen; i++) {
1785 min_uv = UTF8_ACCUMULATE(min_uv,
1786 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1787 }
1788
e9f2c446 1789 adjusted_s0 = temp_char_buf;
57ff5f59 1790 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1791 }
eb83ed87
KW
1792 }
1793
56576a04
KW
1794 /* Here, we have found all the possible problems, except for when the input
1795 * is for a problematic code point not allowed by the input parameters. */
1796
06188866
KW
1797 /* uv is valid for overlongs */
1798 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1799
1800 /* isn't problematic if < this */
1801 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1802 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1803
1804 /* if overflow, we know without looking further
1805 * precisely which of the problematic types it is,
1806 * and we deal with those in the overflow handling
1807 * code */
1808 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1809 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1810 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1811 && ((flags & ( UTF8_DISALLOW_NONCHAR
1812 |UTF8_DISALLOW_SURROGATE
1813 |UTF8_DISALLOW_SUPER
d044b7a7 1814 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1815 |UTF8_WARN_NONCHAR
1816 |UTF8_WARN_SURROGATE
1817 |UTF8_WARN_SUPER
d22ec717 1818 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1819 {
2b5e7bc2
KW
1820 /* If there were no malformations, or the only malformation is an
1821 * overlong, 'uv' is valid */
1822 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1823 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1824 possible_problems |= UTF8_GOT_SURROGATE;
1825 }
1826 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1827 possible_problems |= UTF8_GOT_SUPER;
1828 }
1829 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1830 possible_problems |= UTF8_GOT_NONCHAR;
1831 }
1832 }
1833 else { /* Otherwise, need to look at the source UTF-8, possibly
1834 adjusted to be non-overlong */
1835
1836 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1837 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1838 {
2b5e7bc2
KW
1839 possible_problems |= UTF8_GOT_SUPER;
1840 }
1841 else if (curlen > 1) {
1842 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1843 NATIVE_UTF8_TO_I8(*adjusted_s0),
1844 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1845 {
2b5e7bc2 1846 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1847 }
2b5e7bc2
KW
1848 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1849 NATIVE_UTF8_TO_I8(*adjusted_s0),
1850 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1851 {
1852 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1853 }
1854 }
c0236afe 1855
2b5e7bc2
KW
1856 /* We need a complete well-formed UTF-8 character to discern
1857 * non-characters, so can't look for them here */
1858 }
1859 }
949cf498 1860
2b5e7bc2
KW
1861 ready_to_handle_errors:
1862
1863 /* At this point:
1864 * curlen contains the number of bytes in the sequence that
1865 * this call should advance the input by.
e308b348
KW
1866 * avail_len gives the available number of bytes passed in, but
1867 * only if this is less than the expected number of
1868 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1869 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1870 * is set in it for each potential problem found.
1871 * uv contains the code point the input sequence
1872 * represents; or if there is a problem that prevents
1873 * a well-defined value from being computed, it is
1874 * some subsitute value, typically the REPLACEMENT
1875 * CHARACTER.
1876 * s0 points to the first byte of the character
56576a04
KW
1877 * s points to just after were we left off processing
1878 * the character
1879 * send points to just after where that character should
1880 * end, based on how many bytes the start byte tells
1881 * us should be in it, but no further than s0 +
1882 * avail_len
2b5e7bc2 1883 */
eb83ed87 1884
2b5e7bc2
KW
1885 if (UNLIKELY(possible_problems)) {
1886 bool disallowed = FALSE;
1887 const U32 orig_problems = possible_problems;
1888
37657a5b
KW
1889 if (msgs) {
1890 *msgs = NULL;
1891 }
1892
2b5e7bc2 1893 while (possible_problems) { /* Handle each possible problem */
9fde5914 1894 U32 pack_warn = 0;
2b5e7bc2 1895 char * message = NULL;
37657a5b 1896 U32 this_flag_bit = 0;
2b5e7bc2
KW
1897
1898 /* Each 'if' clause handles one problem. They are ordered so that
1899 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1900 * ones; this is kinda in decreasing severity order. But the
1901 * overlong must come last, as it changes 'uv' looked at by the
1902 * others */
2b5e7bc2
KW
1903 if (possible_problems & UTF8_GOT_OVERFLOW) {
1904
56576a04
KW
1905 /* Overflow means also got a super and are using Perl's
1906 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1907 possible_problems
d044b7a7 1908 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1909 *errors |= UTF8_GOT_OVERFLOW;
1910
1911 /* But the API says we flag all errors found */
1912 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1913 *errors |= UTF8_GOT_SUPER;
1914 }
ddb65933 1915 if (flags
d044b7a7 1916 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1917 {
d044b7a7 1918 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1919 }
2b5e7bc2 1920
d60baaa7 1921 /* Disallow if any of the three categories say to */
56576a04 1922 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1923 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1924 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1925 {
1926 disallowed = TRUE;
1927 }
1928
d22ec717
KW
1929 /* Likewise, warn if any say to */
1930 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1931 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1932 {
2b5e7bc2 1933
ddb65933
KW
1934 /* The warnings code explicitly says it doesn't handle the
1935 * case of packWARN2 and two categories which have
1936 * parent-child relationship. Even if it works now to
1937 * raise the warning if either is enabled, it wouldn't
1938 * necessarily do so in the future. We output (only) the
56576a04 1939 * most dire warning */
ddb65933 1940 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1941 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1942 pack_warn = packWARN(WARN_UTF8);
1943 }
37657a5b 1944 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1945 pack_warn = packWARN(WARN_NON_UNICODE);
1946 }
1947 if (pack_warn) {
1948 message = Perl_form(aTHX_ "%s: %s (overflows)",
1949 malformed_text,
05b9033b 1950 _byte_dump_string(s0, curlen, 0));
37657a5b 1951 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1952 }
2b5e7bc2
KW
1953 }
1954 }
1955 }
1956 else if (possible_problems & UTF8_GOT_EMPTY) {
1957 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1958 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1959
1960 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1961
1962 /* This so-called malformation is now treated as a bug in
1963 * the caller. If you have nothing to decode, skip calling
1964 * this function */
1965 assert(0);
1966
2b5e7bc2 1967 disallowed = TRUE;
37657a5b
KW
1968 if ( (msgs
1969 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1970 {
2b5e7bc2
KW
1971 pack_warn = packWARN(WARN_UTF8);
1972 message = Perl_form(aTHX_ "%s (empty string)",
1973 malformed_text);
37657a5b 1974 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1975 }
1976 }
1977 }
1978 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1979 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1980 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1981
1982 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1983 disallowed = TRUE;
37657a5b
KW
1984 if (( msgs
1985 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1986 {
2b5e7bc2
KW
1987 pack_warn = packWARN(WARN_UTF8);
1988 message = Perl_form(aTHX_
1989 "%s: %s (unexpected continuation byte 0x%02x,"
1990 " with no preceding start byte)",
1991 malformed_text,
7e2f38b2 1992 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1993 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1994 }
1995 }
1996 }
2b5e7bc2
KW
1997 else if (possible_problems & UTF8_GOT_SHORT) {
1998 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1999 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
2000
2001 if (! (flags & UTF8_ALLOW_SHORT)) {
2002 disallowed = TRUE;
37657a5b
KW
2003 if (( msgs
2004 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2005 {
2b5e7bc2
KW
2006 pack_warn = packWARN(WARN_UTF8);
2007 message = Perl_form(aTHX_
56576a04
KW
2008 "%s: %s (too short; %d byte%s available, need %d)",
2009 malformed_text,
2010 _byte_dump_string(s0, send - s0, 0),
2011 (int)avail_len,
2012 avail_len == 1 ? "" : "s",
2013 (int)expectlen);
37657a5b 2014 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
2015 }
2016 }
ba210ebe 2017
2b5e7bc2 2018 }
e308b348
KW
2019 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2020 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2021 *errors |= UTF8_GOT_NON_CONTINUATION;
2022
2023 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2024 disallowed = TRUE;
37657a5b
KW
2025 if (( msgs
2026 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2027 {
99a765e9
KW
2028
2029 /* If we don't know for sure that the input length is
2030 * valid, avoid as much as possible reading past the
2031 * end of the buffer */
2032 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
100de20c
KW
2033 ? (int) (s - s0)
2034 : (int) (send - s0);
e308b348
KW
2035 pack_warn = packWARN(WARN_UTF8);
2036 message = Perl_form(aTHX_ "%s",
2037 unexpected_non_continuation_text(s0,
99a765e9 2038 printlen,
e308b348
KW
2039 s - s0,
2040 (int) expectlen));
37657a5b 2041 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2042 }
2043 }
2044 }
2b5e7bc2
KW
2045 else if (possible_problems & UTF8_GOT_SURROGATE) {
2046 possible_problems &= ~UTF8_GOT_SURROGATE;
2047
f9380377
KW
2048 if (flags & UTF8_WARN_SURROGATE) {
2049 *errors |= UTF8_GOT_SURROGATE;
2050
2051 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2052 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2053 {
2b5e7bc2
KW
2054 pack_warn = packWARN(WARN_SURROGATE);
2055
2056 /* These are the only errors that can occur with a
2057 * surrogate when the 'uv' isn't valid */
2058 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2059 message = Perl_form(aTHX_
2060 "UTF-16 surrogate (any UTF-8 sequence that"
2061 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2062 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2063 }
2064 else {
c94c2f39 2065 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2066 }
37657a5b 2067 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2068 }
2b5e7bc2 2069 }
ba210ebe 2070
2b5e7bc2
KW
2071 if (flags & UTF8_DISALLOW_SURROGATE) {
2072 disallowed = TRUE;
f9380377 2073 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2074 }
2075 }
2076 else if (possible_problems & UTF8_GOT_SUPER) {
2077 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2078
f9380377
KW
2079 if (flags & UTF8_WARN_SUPER) {
2080 *errors |= UTF8_GOT_SUPER;
2081
2082 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2083 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2084 {
2b5e7bc2
KW
2085 pack_warn = packWARN(WARN_NON_UNICODE);
2086
2087 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2088 message = Perl_form(aTHX_
2089 "Any UTF-8 sequence that starts with"
2090 " \"%s\" is for a non-Unicode code point,"
2091 " may not be portable",
7e2f38b2 2092 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2093 }
2094 else {
c94c2f39 2095 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2096 }
37657a5b 2097 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2098 }
2b5e7bc2 2099 }
ba210ebe 2100
57ff5f59
KW
2101 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2102 * and before possibly bailing out, so that the more dire
2103 * warning will override the regular one. */
2104 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2105 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2106 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
dc4a6683
KW
2107 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2108 || ckWARN(WARN_PORTABLE))))
2b5e7bc2 2109 {
dc4a6683 2110 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2b5e7bc2 2111
57ff5f59
KW
2112 /* If it is an overlong that evaluates to a code point
2113 * that doesn't have to use the Perl extended UTF-8, it
2114 * still used it, and so we output a message that
2115 * doesn't refer to the code point. The same is true
2116 * if there was a SHORT malformation where the code
2117 * point is not valid. In that case, 'uv' will have
2118 * been set to the REPLACEMENT CHAR, and the message
2119 * below without the code point in it will be selected
2120 * */
2121 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2122 message = Perl_form(aTHX_
8911f9b0 2123 PL_extended_cp_format, uv);
2b5e7bc2
KW
2124 }
2125 else {
2126 message = Perl_form(aTHX_
57ff5f59
KW
2127 "Any UTF-8 sequence that starts with"
2128 " \"%s\" is a Perl extension, and"
2129 " so is not portable",
2130 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2131 }
37657a5b 2132 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2133 }
2134
d044b7a7
KW
2135 if (flags & ( UTF8_WARN_PERL_EXTENDED
2136 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2137 {
d044b7a7 2138 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2139
d044b7a7 2140 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2141 disallowed = TRUE;
2142 }
2b5e7bc2
KW
2143 }
2144 }
eb83ed87 2145
2b5e7bc2 2146 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2147 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2148 disallowed = TRUE;
2149 }
2b5e7bc2
KW
2150 }
2151 else if (possible_problems & UTF8_GOT_NONCHAR) {
2152 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2153
f9380377
KW
2154 if (flags & UTF8_WARN_NONCHAR) {
2155 *errors |= UTF8_GOT_NONCHAR;
2156
2157 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2158 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2159 {
2b5e7bc2
KW
2160 /* The code above should have guaranteed that we don't
2161 * get here with errors other than overlong */
2162 assert (! (orig_problems
2163 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2164
2165 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2166 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2167 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2168 }
2b5e7bc2 2169 }
5b311467 2170
2b5e7bc2
KW
2171 if (flags & UTF8_DISALLOW_NONCHAR) {
2172 disallowed = TRUE;
f9380377 2173 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2174 }
6c64cd9d
KW
2175 }
2176 else if (possible_problems & UTF8_GOT_LONG) {
2177 possible_problems &= ~UTF8_GOT_LONG;
2178 *errors |= UTF8_GOT_LONG;
2179
2180 if (flags & UTF8_ALLOW_LONG) {
2181
2182 /* We don't allow the actual overlong value, unless the
2183 * special extra bit is also set */
2184 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2185 & ~UTF8_ALLOW_LONG)))
2186 {
2187 uv = UNICODE_REPLACEMENT;
2188 }
2189 }
2190 else {
2191 disallowed = TRUE;
2192
37657a5b
KW
2193 if (( msgs
2194 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2195 {
6c64cd9d
KW
2196 pack_warn = packWARN(WARN_UTF8);
2197
2198 /* These error types cause 'uv' to be something that
2199 * isn't what was intended, so can't use it in the
2200 * message. The other error types either can't
2201 * generate an overlong, or else the 'uv' is valid */
2202 if (orig_problems &
2203 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2204 {
2205 message = Perl_form(aTHX_
2206 "%s: %s (any UTF-8 sequence that starts"
2207 " with \"%s\" is overlong which can and"
2208 " should be represented with a"
2209 " different, shorter sequence)",
2210 malformed_text,
2211 _byte_dump_string(s0, send - s0, 0),
2212 _byte_dump_string(s0, curlen, 0));
2213 }
2214 else {
2215 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2216 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2217 uv, 0);
d819dc50
KW
2218 /* Don't use U+ for non-Unicode code points, which
2219 * includes those in the Latin1 range */
2220 const char * preface = ( uv > PERL_UNICODE_MAX
2221#ifdef EBCDIC
2222 || uv <= 0xFF
2223#endif
2224 )
2225 ? "0x"
2226 : "U+";
6c64cd9d
KW
2227 message = Perl_form(aTHX_
2228 "%s: %s (overlong; instead use %s to represent"
2229 " %s%0*" UVXf ")",
2230 malformed_text,
2231 _byte_dump_string(s0, send - s0, 0),
2232 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2233 preface,
2234 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2235 small code points */
1be62ab9 2236 UNI_TO_NATIVE(uv));
6c64cd9d 2237 }
37657a5b 2238 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2239 }
2240 }
2b5e7bc2
KW
2241 } /* End of looking through the possible flags */
2242
2243 /* Display the message (if any) for the problem being handled in
2244 * this iteration of the loop */
2245 if (message) {
37657a5b 2246 if (msgs) {
37657a5b
KW
2247 assert(this_flag_bit);
2248
2249 if (*msgs == NULL) {
2250 *msgs = newAV();
2251 }
2252
bb07812e
KW
2253 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2254 pack_warn,
2255 this_flag_bit)));
37657a5b
KW
2256 }
2257 else if (PL_op)
2b5e7bc2
KW
2258 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2259 OP_DESC(PL_op));
2260 else
2261 Perl_warner(aTHX_ pack_warn, "%s", message);
2262 }
ddb65933 2263 } /* End of 'while (possible_problems)' */
a0dbb045 2264
2b5e7bc2
KW
2265 /* Since there was a possible problem, the returned length may need to
2266 * be changed from the one stored at the beginning of this function.
2267 * Instead of trying to figure out if that's needed, just do it. */
2268 if (retlen) {
2269 *retlen = curlen;
2270 }
a0dbb045 2271
2b5e7bc2
KW
2272 if (disallowed) {
2273 if (flags & UTF8_CHECK_ONLY && retlen) {
2274 *retlen = ((STRLEN) -1);
2275 }
2276 return 0;
2277 }
eb83ed87 2278 }
ba210ebe 2279
2b5e7bc2 2280 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2281}
2282
8e84507e 2283/*
ec5f19d0
KW
2284=for apidoc utf8_to_uvchr_buf
2285
2286Returns the native code point of the first character in the string C<s> which
2287is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2288C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2289
524080c4
KW
2290If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2291enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2292C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2293(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2294C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2295the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2296See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2297returned.
ec5f19d0
KW
2298
2299=cut
52be2536
KW
2300
2301Also implemented as a macro in utf8.h
2302
ec5f19d0
KW
2303*/
2304
2305
2306UV
2307Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2308{
7f974d7e
KW
2309 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2310
9a9a6c98 2311 return utf8_to_uvchr_buf_helper(s, send, retlen);
ec5f19d0
KW
2312}
2313
52be2536
KW
2314/* This is marked as deprecated
2315 *
ec5f19d0
KW
2316=for apidoc utf8_to_uvuni_buf
2317
de69f3af
KW
2318Only in very rare circumstances should code need to be dealing in Unicode
2319(as opposed to native) code points. In those few cases, use
09232555
KW
2320C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2321If you are not absolutely sure this is one of those cases, then assume it isn't
2322and use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2323
2324Returns the Unicode (not-native) code point of the first character in the
2325string C<s> which
ec5f19d0
KW
2326is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2327C<retlen> will be set to the length, in bytes, of that character.
2328
524080c4
KW
2329If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2330enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2331NULL) to -1. If those warnings are off, the computed value if well-defined (or
2332the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2333is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2334next possible position in C<s> that could begin a non-malformed character.
09232555
KW
2335See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2336returned.
ec5f19d0
KW
2337
2338=cut
2339*/
2340
2341UV
2342Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2343{
2344 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2345
2346 assert(send > s);
2347
5962d97e 2348 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2349}
2350
b76347f2 2351/*
87cea99e 2352=for apidoc utf8_length
b76347f2 2353
b2e7ed74
KW
2354Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2355at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2356same place, it returns 0 with no warning raised.
2357
2358If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2359and returns the number of valid characters.
b76347f2
JH
2360
2361=cut
2362*/
2363
2364STRLEN
35a4481c 2365Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2366{
2367 STRLEN len = 0;
2368
7918f24d
NC
2369 PERL_ARGS_ASSERT_UTF8_LENGTH;
2370
8850bf83
JH
2371 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2372 * the bitops (especially ~) can create illegal UTF-8.
2373 * In other words: in Perl UTF-8 is not just for Unicode. */
2374
12c43b0a 2375 if (UNLIKELY(e < s))
a3b680e6 2376 goto warn_and_return;
b76347f2 2377 while (s < e) {
4cbf4130 2378 s += UTF8SKIP(s);
8e91ec7f
AV
2379 len++;
2380 }
2381
12c43b0a 2382 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2383 len--;
2384 warn_and_return:
9b387841
NC
2385 if (PL_op)
2386 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2387 "%s in %s", unees, OP_DESC(PL_op));
2388 else
61a12c31 2389 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2390 }
2391
2392 return len;
2393}
2394
b06226ff 2395/*
fed3ba5d
NC
2396=for apidoc bytes_cmp_utf8
2397
a1433954 2398Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2399sequence of characters (stored as UTF-8)
2400in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2401equal, -1 or -2 if the first string is less than the second string, +1 or +2
2402if the first string is greater than the second string.
2403
2404-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2405longer string. -2 or +2 is returned if
2406there was a difference between characters
fed3ba5d
NC
2407within the strings.
2408
2409=cut
2410*/
2411
2412int
2413Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2414{
2415 const U8 *const bend = b + blen;
2416 const U8 *const uend = u + ulen;
2417
2418 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2419
2420 while (b < bend && u < uend) {
2421 U8 c = *u++;
2422 if (!UTF8_IS_INVARIANT(c)) {
2423 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2424 if (u < uend) {
2425 U8 c1 = *u++;
2426 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2427 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2428 } else {
2b5e7bc2 2429 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2430 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2431 "%s %s%s",
2432 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2433 PL_op ? " in " : "",
2434 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2435 return -2;
2436 }
2437 } else {
2438 if (PL_op)
2439 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2440 "%s in %s", unees, OP_DESC(PL_op));
2441 else
61a12c31 2442 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2443 return -2; /* Really want to return undef :-) */
2444 }
2445 } else {
2446 return -2;
2447 }
2448 }
2449 if (*b != c) {
2450 return *b < c ? -2 : +2;
2451 }
2452 ++b;
2453 }
2454
2455 if (b == bend && u == uend)
2456 return 0;
2457
2458 return b < bend ? +1 : -1;
2459}
2460
2461/*
87cea99e 2462=for apidoc utf8_to_bytes
6940069f 2463
3bc0c78c 2464Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2465Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2466updates C<*lenp> to contain the new length.
3bc0c78c
KW
2467Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2468
2469Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2470having saved the value of C<*lenp> before the call, and subtracting the
2471after-call value of C<*lenp> from it.
6940069f 2472
a1433954 2473If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2474
6940069f
GS
2475=cut
2476*/
2477
2478U8 *
09af0336 2479Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2480{
9fe0d3c2 2481 U8 * first_variant;
246fae53 2482
7918f24d 2483 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2484 PERL_UNUSED_CONTEXT;
7918f24d 2485
9fe0d3c2 2486 /* This is a no-op if no variants at all in the input */
09af0336 2487 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2488 return s;
2489 }
2490
2491 {
3c5aa262 2492 U8 * const save = s;
09af0336 2493 U8 * const send = s + *lenp;
3c5aa262
KW
2494 U8 * d;
2495
2496 /* Nothing before the first variant needs to be changed, so start the real
2497 * work there */
2498 s = first_variant;
2499 while (s < send) {
2500 if (! UTF8_IS_INVARIANT(*s)) {
2501 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2502 *lenp = ((STRLEN) -1);
3c5aa262
KW
2503 return 0;
2504 }
2505 s++;
d59937ca
KW
2506 }
2507 s++;
dcad2880 2508 }
dcad2880 2509
3c5aa262
KW
2510 /* Is downgradable, so do it */
2511 d = s = first_variant;
2512 while (s < send) {
2513 U8 c = *s++;
2514 if (! UVCHR_IS_INVARIANT(c)) {
2515 /* Then it is two-byte encoded */
2516 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2517 s++;
2518 }
2519 *d++ = c;
2520 }
2521 *d = '\0';
09af0336 2522 *lenp = d - save;
3c5aa262
KW
2523
2524 return save;
9fe0d3c2 2525 }
6940069f
GS
2526}
2527
2528/*
87cea99e 2529=for apidoc bytes_from_utf8
f9a63242 2530
09af0336 2531Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2532byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2533actually encoded in UTF-8.
2534
2535Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2536the input string.
2537
41ae6089
KW
2538Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2539not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2540C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2541
41ae6089 2542Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2543newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2544returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2545caller is responsible for arranging for the memory used by this string to get
2546freed.
f9a63242 2547
3bc0c78c 2548Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2549having saved the value of C<*lenp> before the call, and subtracting the
2550after-call value of C<*lenp> from it.
3bc0c78c 2551
37607a96 2552=cut
976c1b08
KW
2553
2554There is a macro that avoids this function call, but this is retained for
2555anyone who calls it with the Perl_ prefix */
f9a63242
JH
2556
2557U8 *
41ae6089 2558Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2559{
7918f24d 2560 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2561 PERL_UNUSED_CONTEXT;
f9a63242 2562
976c1b08
KW
2563 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2564}
2565
2566/*
df6bd76f 2567=for apidoc bytes_from_utf8_loc
976c1b08 2568
eda578be
KW
2569Like C<L<perlapi/bytes_from_utf8>()>, but takes an extra parameter, a pointer
2570to where to store the location of the first character in C<"s"> that cannot be
976c1b08
KW
2571converted to non-UTF8.
2572
2573If that parameter is C<NULL>, this function behaves identically to
2574C<bytes_from_utf8>.
2575
2576Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2577C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2578
2579Otherwise, the function returns a newly created C<NUL>-terminated string
2580containing the non-UTF8 equivalent of the convertible first portion of
2581C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2582If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2583and C<*first_non_downgradable> is set to C<NULL>.
2584
8505db87 2585Otherwise, C<*first_non_downgradable> is set to point to the first byte of the
976c1b08
KW
2586first character in the original string that wasn't converted. C<*is_utf8p> is
2587unchanged. Note that the new string may have length 0.
2588
2589Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2590C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2591converts as many characters in it as possible stopping at the first one it
385b74be 2592finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2593set to point to that. The function returns the portion that could be converted
2594in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2595not including the terminating C<NUL>. If the very first character in the
2596original could not be converted, C<*lenp> will be 0, and the new string will
2597contain just a single C<NUL>. If the entire input string was converted,
2598C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2599
2600Upon successful return, the number of variants in the converted portion of the
2601string can be computed by having saved the value of C<*lenp> before the call,
2602and subtracting the after-call value of C<*lenp> from it.
2603
2604=cut
2605
2606
2607*/
2608
2609U8 *
2610Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2611{
2612 U8 *d;
2613 const U8 *original = s;
2614 U8 *converted_start;
2615 const U8 *send = s + *lenp;
f9a63242 2616
976c1b08 2617 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2618
976c1b08
KW
2619 if (! *is_utf8p) {
2620 if (first_unconverted) {
2621 *first_unconverted = NULL;
2622 }
2623
2624 return (U8 *) original;
2625 }
2626
2627 Newx(d, (*lenp) + 1, U8);
2628
2629 converted_start = d;
7299a045
KW
2630 while (s < send) {
2631 U8 c = *s++;
2632 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2633
2634 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2635 * have to stop now */
2636 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2637 if (first_unconverted) {
2638 *first_unconverted = s - 1;
2639 goto finish_and_return;
2640 }
2641 else {
2642 Safefree(converted_start);
2643 return (U8 *) original;
2644 }
2645 }
2646
7299a045
KW
2647 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2648 s++;
38af28cf 2649 }
7299a045
KW
2650 *d++ = c;
2651 }
170a1c22 2652
976c1b08
KW
2653 /* Here, converted the whole of the input */
2654 *is_utf8p = FALSE;
2655 if (first_unconverted) {
2656 *first_unconverted = NULL;
170a1c22 2657 }
976c1b08
KW
2658
2659 finish_and_return:
46a08a6f
KW
2660 *d = '\0';
2661 *lenp = d - converted_start;
976c1b08
KW
2662
2663 /* Trim unused space */
2664 Renew(converted_start, *lenp + 1, U8);
2665
2666 return converted_start;
f9a63242
JH
2667}
2668
2669/*
87cea99e 2670=for apidoc bytes_to_utf8
6940069f 2671
09af0336 2672Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2673UTF-8.
09af0336 2674Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2675reflect the new length in bytes. The caller is responsible for arranging for
2676the memory used by this string to get freed.
6940069f 2677
3bc0c78c 2678Upon successful return, the number of variants in the string can be computed by
23b37b12 2679having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2680after-call value of C<*lenp>.
2681
75200dff 2682A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2683
2684If you want to convert to UTF-8 from encodings other than
2685the native (Latin1 or EBCDIC),
a1433954 2686see L</sv_recode_to_utf8>().
c9ada85f 2687
497711e7 2688=cut
6940069f
GS
2689*/
2690
2691U8*
09af0336 2692Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2693{
09af0336 2694 const U8 * const send = s + (*lenp);
6940069f
GS
2695 U8 *d;
2696 U8 *dst;
7918f24d
NC
2697
2698 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2699 PERL_UNUSED_CONTEXT;
6940069f 2700
d4662719
KW
2701 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2702 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2703 dst = d;
2704
2705 while (s < send) {
55d09dc8
KW
2706 append_utf8_from_native_byte(*s, &d);
2707 s++;
6940069f 2708 }
2e11cf67 2709
6940069f 2710 *d = '\0';
09af0336 2711 *lenp = d-dst;
2e11cf67 2712
6940069f
GS
2713 return dst;
2714}
2715
a0ed51b3 2716/*
624504c5
KW
2717 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2718 * use utf16_to_utf8_reversed().
a0ed51b3 2719 *
624504c5
KW
2720 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2721 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2722 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2723 *
2724 * These functions don't check for overflow. The worst case is every code
2725 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2726 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2727 * destination must be pre-extended to 2 times the source length.
2728 *
2729 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2730
2731U8*
f46dcac2 2732Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3 2733{
dea0fc0b
JH
2734 U8* pend;
2735 U8* dstart = d;
2736
7918f24d
NC
2737 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2738
dea0fc0b 2739 if (bytelen & 1)
56576a04
KW
2740 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2741 (UV)bytelen);
dea0fc0b
JH
2742
2743 pend = p + bytelen;
2744
a0ed51b3 2745 while (p < pend) {
dea0fc0b
JH
2746 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2747 p += 2;
2d1545e5 2748 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2749 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2750 continue;
2751 }
56d37426
KW
2752 if (uv <= MAX_UTF8_TWO_BYTE) {
2753 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2754 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2755 continue;
2756 }
ffd0a9d3 2757
46956fad
KW
2758#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2759#define LAST_HIGH_SURROGATE 0xDBFF
2760#define FIRST_LOW_SURROGATE 0xDC00
2761#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2762#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2763
2764 /* This assumes that most uses will be in the first Unicode plane, not
2765 * needing surrogates */
b497502c
KW
2766 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2767 UNICODE_SURROGATE_LAST)))
e23c50db
KW
2768 {
2769 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2770 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2771 }
2772 else {
01ea242b 2773 UV low = (p[0] << 8) + p[1];
b497502c
KW
2774 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2775 LAST_LOW_SURROGATE)))
e23c50db 2776 {
01ea242b 2777 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2778 }
2779 p += 2;
46956fad 2780 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2781 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2782 }
a0ed51b3 2783 }
56d37426
KW
2784#ifdef EBCDIC
2785 d = uvoffuni_to_utf8_flags(d, uv, 0);
2786#else
ffd0a9d3 2787 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2788 *d++ = (U8)(( uv >> 12) | 0xe0);
2789 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2790 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2791 continue;
2792 }
2793 else {
eb160463
GS
2794 *d++ = (U8)(( uv >> 18) | 0xf0);
2795 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2796 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2797 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2798 continue;
2799 }
56d37426 2800#endif
a0ed51b3 2801 }
dea0fc0b 2802 *newlen = d - dstart;
a0ed51b3
LW
2803 return d;
2804}
2805
2806/* Note: this one is slightly destructive of the source. */
2807
2808U8*
f46dcac2 2809Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3
LW
2810{
2811 U8* s = (U8*)p;
d4c19fe8 2812 U8* const send = s + bytelen;
7918f24d
NC
2813
2814 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2815
e0ea5e2d 2816 if (bytelen & 1)
147e3846 2817 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2818 (UV)bytelen);
2819
a0ed51b3 2820 while (s < send) {
d4c19fe8 2821 const U8 tmp = s[0];
a0ed51b3
LW
2822 s[0] = s[1];
2823 s[1] = tmp;
2824 s += 2;
2825 }
dea0fc0b 2826 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2827}
2828
922e8cb4
KW
2829bool
2830Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2831{
dc31b55c 2832 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2833}
2834
5092f92a 2835bool
eba68aa0
KW
2836Perl__is_uni_perl_idcont(pTHX_ UV c)
2837{
c12658c9 2838 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2839}
2840
2841bool
f91dcd13
KW
2842Perl__is_uni_perl_idstart(pTHX_ UV c)
2843{
c12658c9 2844 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2845}
2846
3a4c58c9 2847UV
56576a04
KW
2848Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2849 const char S_or_s)
3a4c58c9
KW
2850{
2851 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2852 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2853 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2854 * either "SS" or "Ss". Which one to use is passed into the routine in
2855 * 'S_or_s' to avoid a test */
2856
2857 UV converted = toUPPER_LATIN1_MOD(c);
2858
2859 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2860
2861 assert(S_or_s == 'S' || S_or_s == 's');
2862
6f2d5cbc 2863 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2864 characters in this range */
3a4c58c9
KW
2865 *p = (U8) converted;
2866 *lenp = 1;
2867 return converted;
2868 }
2869
2870 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2871 * which it maps to one of them, so as to only have to have one check for
2872 * it in the main case */
2873 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2874 switch (c) {
2875 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2876 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2877 break;
2878 case MICRO_SIGN:
2879 converted = GREEK_CAPITAL_LETTER_MU;
2880 break;
79e064b9
KW
2881#if UNICODE_MAJOR_VERSION > 2 \
2882 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2883 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2884 case LATIN_SMALL_LETTER_SHARP_S:
2885 *(p)++ = 'S';
2886 *p = S_or_s;
2887 *lenp = 2;
2888 return 'S';
79e064b9 2889#endif
3a4c58c9 2890 default:
56576a04
KW
2891 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2892 " '%c' to map to '%c'",
2893 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2894 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2895 }
2896 }
2897
2898 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2899 *p = UTF8_TWO_BYTE_LO(converted);
2900 *lenp = 2;
2901
2902 return converted;
2903}
2904
fe63c520
KW
2905/* If compiled on an early Unicode version, there may not be auxiliary tables
2906 * */
2907#ifndef HAS_UC_AUX_TABLES
2908# define UC_AUX_TABLE_ptrs NULL
2909# define UC_AUX_TABLE_lengths NULL
2910#endif
2911#ifndef HAS_TC_AUX_TABLES
2912# define TC_AUX_TABLE_ptrs NULL
2913# define TC_AUX_TABLE_lengths NULL
2914#endif
2915#ifndef HAS_LC_AUX_TABLES
2916# define LC_AUX_TABLE_ptrs NULL
2917# define LC_AUX_TABLE_lengths NULL
2918#endif
2919#ifndef HAS_CF_AUX_TABLES
2920# define CF_AUX_TABLE_ptrs NULL
2921# define CF_AUX_TABLE_lengths NULL
2922#endif
2923#ifndef HAS_UC_AUX_TABLES
2924# define UC_AUX_TABLE_ptrs NULL
2925# define UC_AUX_TABLE_lengths NULL
2926#endif
2927
50bda2c3
KW
2928/* Call the function to convert a UTF-8 encoded character to the specified case.
2929 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2930 * 's' is a pointer to the first byte of the input character
2931 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2932 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2933 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2934 *
56576a04 2935 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2936 * 'd' */
56576a04 2937#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2938 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2939 Uppercase_Mapping_invmap, \
2940 UC_AUX_TABLE_ptrs, \
2941 UC_AUX_TABLE_lengths, \
2942 "uppercase")
56576a04 2943#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2944 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2945 Titlecase_Mapping_invmap, \
2946 TC_AUX_TABLE_ptrs, \
2947 TC_AUX_TABLE_lengths, \
2948 "titlecase")
56576a04 2949#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2950 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2951 Lowercase_Mapping_invmap, \
2952 LC_AUX_TABLE_ptrs, \
2953 LC_AUX_TABLE_lengths, \
2954 "lowercase")
2955
50bda2c3 2956
b9992569
KW
2957/* This additionally has the input parameter 'specials', which if non-zero will
2958 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2959 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2960#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2961 (specials) \
2962 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2963 Case_Folding_invmap, \
2964 CF_AUX_TABLE_ptrs, \
2965 CF_AUX_TABLE_lengths, \
2966 "foldcase") \
2967 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2968 Simple_Case_Folding_invmap, \
2969 NULL, NULL, \
2970 "foldcase")
c3fd2246 2971
84afefe6
JH
2972UV
2973Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2974{
a1433954
KW
2975 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2976 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2977 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2978 * the changed version may be longer than the original character.
2979 *
2980 * The ordinal of the first character of the changed version is returned
2981 * (but note, as explained above, that there may be more.) */
2982
7918f24d
NC
2983 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2984
3a4c58c9
KW
2985 if (c < 256) {
2986 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2987 }
2988
a13f1de4 2989 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2990}
2991
84afefe6
JH
2992UV
2993Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2994{
7918f24d
NC
2995 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2996
3a4c58c9
KW
2997 if (c < 256) {
2998 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2999 }
3000
a13f1de4 3001 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3002}
3003
afc16117 3004STATIC U8
eaf412bf 3005S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
3006{
3007 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 3008 * those, converting the result to UTF-8. Since the result is always just
a1433954 3009 * one character, we allow <p> to be NULL */
afc16117
KW
3010
3011 U8 converted = toLOWER_LATIN1(c);
3012
eaf412bf
KW
3013 PERL_UNUSED_ARG(dummy);
3014
afc16117 3015 if (p != NULL) {
6f2d5cbc 3016 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3017 *p = converted;
3018 *lenp = 1;
3019 }
3020 else {
430c9760
KW
3021 /* Result is known to always be < 256, so can use the EIGHT_BIT
3022 * macros */
3023 *p = UTF8_EIGHT_BIT_HI(converted);
3024 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3025 *lenp = 2;
3026 }
3027 }
3028 return converted;
3029}
3030
84afefe6
JH
3031UV
3032Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3033{
7918f24d
NC
3034 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3035
afc16117 3036 if (c < 256) {
eaf412bf 3037 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3038 }
3039
a13f1de4 3040 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3041}
3042
84afefe6 3043UV
7c0ab950 3044Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3045{
51910141 3046 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3047 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3048 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3049 *
3050 * Not to be used for locale folds
51910141 3051 */
f673fad4 3052
a1dde8de
KW
3053 UV converted;
3054
3055 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3056
1ca267a5
KW
3057 assert (! (flags & FOLD_FLAGS_LOCALE));
3058
659a7c2d 3059 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3060 converted = GREEK_SMALL_LETTER_MU;
3061 }
9b63e895
KW
3062#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3063 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3064 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3065 else if ( (flags & FOLD_FLAGS_FULL)
3066 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3067 {
1ca267a5
KW
3068 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3069 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3070 * under those circumstances. */
3071 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3072 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3073 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3074 p, *lenp, U8);
3075 return LATIN_SMALL_LETTER_LONG_S;
3076 }
3077 else {
4f489194
KW
3078 *(p)++ = 's';
3079 *p = 's';
3080 *lenp = 2;
3081 return 's';
1ca267a5 3082 }
a1dde8de 3083 }
9b63e895 3084#endif
a1dde8de
KW
3085 else { /* In this range the fold of all other characters is their lower
3086 case */
3087 converted = toLOWER_LATIN1(c);
3088 }
3089
6f2d5cbc 3090 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3091 *p = (U8) converted;
3092 *lenp = 1;
3093 }
3094 else {
3095 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3096 *p = UTF8_TWO_BYTE_LO(converted);
3097 *lenp = 2;
3098 }
3099
3100 return converted;
3101}
3102
3103UV
31f05a37 3104Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3105{
4b593389 3106
a0270393
KW
3107 /* Not currently externally documented, and subject to change
3108 * <flags> bits meanings:
3109 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3110 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3111 * locale are to be used.
a0270393
KW
3112 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3113 */
4b593389 3114
36bb2ab6 3115 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3116
780fcc9f 3117 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3118 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3119 * except for potentially warning */
8b7358b9 3120 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3121 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3122 flags &= ~FOLD_FLAGS_LOCALE;
3123 }
3124 else {
e7b7ac46 3125 goto needs_full_generality;
780fcc9f 3126 }
31f05a37
KW
3127 }
3128
a1dde8de 3129 if (c < 256) {
e7b7ac46 3130 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3131 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3132 }
3133
2f306ab9 3134 /* Here, above 255. If no special needs, just use the macro */
a0270393 3135 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3136 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3137 }
567b353c 3138 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3139 the special flags. */
3140 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3141
3142 needs_full_generality:
a0270393 3143 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3144 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3145 p, lenp, flags);
a0270393 3146 }
84afefe6
JH
3147}
3148
26483009 3149PERL_STATIC_INLINE bool
dd1a3ba7
KW
3150S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3151 SV* const invlist)
da8c1a98
KW
3152{
3153 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3154 * starts at <p>, and extending no further than <e - 1> is in the inversion
3155 * list <invlist>. */
da8c1a98 3156
b68ffe0c
KW
3157 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3158
dd1a3ba7 3159 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
da8c1a98 3160
b68ffe0c 3161 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3162 _force_out_malformed_utf8_message(p, e, 0, 1);
3163 NOT_REACHED; /* NOTREACHED */
3164 }
3165
eb1f4bb4 3166 assert(invlist);
b68ffe0c 3167 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3168}
3169
059703b0 3170#if 0 /* Not currently used, but may be needed in the future */
dd1a3ba7
KW
3171PERLVAR(I, seen_deprecated_macro, HV *)
3172
34aeb2e9
KW
3173STATIC void
3174S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3175 const char * const alternative,
3176 const bool use_locale,
3177 const char * const file,
3178 const unsigned line)
3179{
3180 const char * key;
3181
3182 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3183
3184 if (ckWARN_d(WARN_DEPRECATED)) {
3185
3186 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3187 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3188 if (! PL_seen_deprecated_macro) {
3189 PL_seen_deprecated_macro = newHV();
3190 }
3191 if (! hv_store(PL_seen_deprecated_macro, key,
3192 strlen(key), &PL_sv_undef, 0))
3193 {
3194 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3195 }
3196
c44e9413 3197 if (instr(file, "mathoms.c")) {
607313a1 3198 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3199 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3200 " will be removed. Avoid this message by"
3201 " converting to use %s().\n",
3202 file, line, name, alternative);
3203 }
3204 else {
34aeb2e9 3205 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3206 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3207 " require an additional parameter. Avoid this"
3208 " message by converting to use %s().\n",
3209 file, line, name, alternative);
607313a1 3210 }
34aeb2e9
KW
3211 }
3212 }
3213}
059703b0 3214#endif
922e8cb4
KW
3215
3216bool
dd1a3ba7 3217Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
da8c1a98 3218{
dd1a3ba7 3219 PERL_ARGS_ASSERT__IS_UTF8_FOO;
da8c1a98 3220
dd1a3ba7 3221 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
da8c1a98
KW
3222}
3223
3224bool
dd1a3ba7 3225Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
da8c1a98 3226{
dd1a3ba7 3227 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
da8c1a98 3228
dd1a3ba7 3229 return is_utf8_common(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3230}
3231
3232bool
dd1a3ba7 3233Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
c11ff943 3234{
dd1a3ba7 3235 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
7dbf68d2 3236
dd1a3ba7 3237 return is_utf8_common(p, e, PL_utf8_perl_idcont);
7dbf68d2
KW
3238}
3239
6a4a25f4 3240STATIC UV
30613bdc
KW
3241S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3242 U8* ustrp, STRLEN *lenp,
40d2776f
KW
3243 SV *invlist, const I32 * const invmap,
3244 const U32 * const * const aux_tables,
30613bdc
KW
3245 const U8 * const aux_table_lengths,
3246 const char * const normal)
b9992569 3247{
0134edef 3248 STRLEN len = 0;
7918f24d 3249
30613bdc
KW
3250 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3251 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3252 * to name the new case in any generated messages, as a fallback if the
3253 * operation being used is not available. The new case is given by the
3254 * data structures in the remaining arguments.
3255 *
3256 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3257 * entire changed case string, and the return value is the first code point
3258 * in that string */
3259
b9992569 3260 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3261
36eaa811
KW
3262 /* For code points that don't change case, we already know that the output
3263 * of this function is the unchanged input, so we can skip doing look-ups
3264 * for them. Unfortunately the case-changing code points are scattered
3265 * around. But there are some long consecutive ranges where there are no
3266 * case changing code points. By adding tests, we can eliminate the lookup
3267 * for all the ones in such ranges. This is currently done here only for
3268 * just a few cases where the scripts are in common use in modern commerce
3269 * (and scripts adjacent to those which can be included without additional
3270 * tests). */
3271
3272 if (uv1 >= 0x0590) {
3273 /* This keeps from needing further processing the code points most
3274 * likely to be used in the following non-cased scripts: Hebrew,
3275 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3276 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3277 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3278 if (uv1 < 0x10A0) {
3279 goto cases_to_self;
3280 }
3281
3282 /* The following largish code point ranges also don't have case
3283 * changes, but khw didn't think they warranted extra tests to speed
3284 * them up (which would slightly slow down everything else above them):
3285 * 1100..139F Hangul Jamo, Ethiopic
3286 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3287 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3288 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3289 * Combining Diacritical Marks Extended, Balinese,
3290 * Sundanese, Batak, Lepcha, Ol Chiki
3291 * 2000..206F General Punctuation
3292 */
3293
3294 if (uv1 >= 0x2D30) {
3295
3296 /* This keeps the from needing further processing the code points
3297 * most likely to be used in the following non-cased major scripts:
3298 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3299 *
3300 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3301 * event that Unicode eventually allocates the unused block as of
3302 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3303 * that the test suite will start having failures to alert you
3304 * should that happen) */
3305 if (uv1 < 0xA640) {
3306 goto cases_to_self;
3307 }
3308
3309 if (uv1 >= 0xAC00) {
3310 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3311 if (ckWARN_d(WARN_SURROGATE)) {
3312 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3313 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3314 "Operation \"%s\" returns its argument for"
3315 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3316 }
3317 goto cases_to_self;
3318 }
36eaa811
KW
3319
3320 /* AC00..FAFF Catches Hangul syllables and private use, plus
3321 * some others */
3322 if (uv1 < 0xFB00) {
3323 goto cases_to_self;
36eaa811
KW
3324 }
3325
5af9bc97 3326 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3327 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
fb2f0a6a 3328 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1));
5af9bc97
KW
3329 }
3330 if (ckWARN_d(WARN_NON_UNICODE)) {
3331 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3332 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3333 "Operation \"%s\" returns its argument for"
3334 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3335 }
3336 goto cases_to_self;
3337 }
c62fdeb7
KW
3338#ifdef HIGHEST_CASE_CHANGING_CP
3339 if (UNLIKELY(uv1 > HIGHEST_CASE_CHANGING_CP)) {
3bfc1e70 3340
3bfc1e70
KW
3341 goto cases_to_self;
3342 }
3343#endif
36eaa811
KW
3344 }
3345 }
9ae3ac1a 3346
36eaa811 3347 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3348 * be given. */
9ae3ac1a
KW
3349 }
3350
8946fcd9
KW
3351 {
3352 unsigned int i;
22197324 3353 const U32 * cp_list;
8946fcd9 3354 U8 * d;
69352d88
KW
3355
3356 /* 'index' is guaranteed to be non-negative, as this is an inversion
3357 * map that covers all possible inputs. See [perl #133365] */
8946fcd9 3358 SSize_t index = _invlist_search(invlist, uv1);
22197324 3359 I32 base = invmap[index];
0134edef 3360
30613bdc
KW
3361 /* The data structures are set up so that if 'base' is non-negative,
3362 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3363 if (base >= 0) {
3364 IV lc;
b08cf34e 3365
8946fcd9
KW
3366 if (base == 0) {
3367 goto cases_to_self;
4a8240a3 3368 }
4a8240a3 3369
30613bdc 3370 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3371 lc = base + uv1 - invlist_array(invlist)[index];
3372 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3373 return lc;
3374 }
1feea2c7 3375
30613bdc
KW
3376 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3377 * requires an auxiliary table look up. abs(base) gives the index into
3378 * a list of such tables which points to the proper aux table. And a
3379 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3380 cp_list = aux_tables[-base];
30613bdc
KW
3381
3382 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3383 d = ustrp;
3384 for (i = 0; i < aux_table_lengths[-base]; i++) {
3385 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3386 }
8946fcd9
KW
3387 *d = '\0';
3388 *lenp = d - ustrp;
3389
3390 return cp_list[0];
cbe07460
KW
3391 }
3392
3393 /* Here, there was no mapping defined, which means that the code point maps
3394 * to itself. Return the inputs */
e24dfe9c 3395 cases_to_self:
a13f1de4
KW
3396 if (p) {
3397 len = UTF8SKIP(p);
3398 if (p != ustrp) { /* Don't copy onto itself */
3399 Copy(p, ustrp, len, U8);
3400 }
3401 *lenp = len;
3402 }
3403 else {
3404 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3405 }
2a37f04d 3406
f4cd282c 3407 return uv1;
cbe07460 3408
a0ed51b3
LW
3409}
3410
b74fe592 3411Size_t
1b292063 3412Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
40d2776f 3413 const U32 ** remaining_folds_to)
b74fe592
KW
3414{
3415 /* Returns the count of the number of code points that fold to the input
3416 * 'cp' (besides itself).
3417 *
3418 * If the return is 0, there is nothing else that folds to it, and
3419 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3420 *
3421 * If the return is 1, '*first_folds_to' is set to the single code point,
3422 * and '*remaining_folds_to' is set to NULL.
3423 *
3424 * Otherwise, '*first_folds_to' is set to a code point, and
3425 * '*remaining_fold_to' is set to an array that contains the others. The
3426 * length of this array is the returned count minus 1.
3427 *
3428 * The reason for this convolution is to avoid having to deal with
3429 * allocating and freeing memory. The lists are already constructed, so
3430 * the return can point to them, but single code points aren't, so would
1b292063
KW
3431 * need to be constructed if we didn't employ something like this API
3432 *
3433 * The code points returned by this function are all legal Unicode, which
3434 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3435 * constructed with this size (to save space and memory), and we return
3436 * pointers, so they must be this size */
b74fe592 3437
69352d88
KW
3438 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3439 * that covers all possible inputs. See [perl #133365] */
b74fe592 3440 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
40d2776f 3441 I32 base = _Perl_IVCF_invmap[index];
b74fe592
KW
3442
3443 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3444
3445 if (base == 0) { /* No fold */
3446 *first_folds_to = 0;
3447 *remaining_folds_to = NULL;
3448 return 0;
3449 }
3450
3451#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3452
3453 assert(base > 0);
3454
3455#else
3456
3457 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3458
3459 /* The data structure is set up so that the absolute value of 'base' is
3460 * an index into a table of pointers to arrays, with the array
3461 * corresponding to the index being the list of code points that fold
3462 * to 'cp', and the parallel array containing the length of the list
3463 * array */
3464 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
99f30495
KW
3465 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3466 /* +1 excludes first_folds_to */
b74fe592
KW
3467 return IVCF_AUX_TABLE_lengths[-base];
3468 }
3469
3470#endif
3471
3472 /* Only the single code point. This works like 'fc(G) = G - A + a' */
40d2776f
KW
3473 *first_folds_to = (U32) (base + cp
3474 - invlist_array(PL_utf8_foldclosures)[index]);
b74fe592
KW
3475 *remaining_folds_to = NULL;
3476 return 1;
3477}
3478
051a06d4 3479STATIC UV
56576a04
KW
3480S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3481 U8* const ustrp, STRLEN *lenp)
051a06d4 3482{
4a4088c4 3483 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3484 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3485 * result contains a character that crosses the 255/256 boundary, disallow
3486 * the change, and return the original code point. See L<perlfunc/lc> for
3487 * why;
051a06d4 3488 *
a1433954
KW
3489 * p points to the original string whose case was changed; assumed
3490 * by this routine to be well-formed
051a06d4 3491 * result the code point of the first character in the changed-case string
56576a04
KW
3492 * ustrp points to the changed-case string (<result> represents its
3493 * first char)
051a06d4
KW
3494 * lenp points to the length of <ustrp> */
3495
3496 UV original; /* To store the first code point of <p> */
3497
3498 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3499
a4f12ed7 3500 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3501
3502 /* We know immediately if the first character in the string crosses the
5e45c680 3503 * boundary, so can skip testing */
051a06d4
KW
3504 if (result > 255) {
3505
3506 /* Look at every character in the result; if any cross the
3507 * boundary, the whole thing is disallowed */
3508 U8* s = ustrp + UTF8SKIP(ustrp);
3509 U8* e = ustrp + *lenp;
3510 while (s < e) {
a4f12ed7 3511 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3512 goto bad_crossing;
3513 }
3514 s += UTF8SKIP(s);
3515 }
3516
613abc6d
KW
3517 /* Here, no characters crossed, result is ok as-is, but we warn. */
3518 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3519 return result;
3520 }
3521
7b52d656 3522 bad_crossing:
051a06d4
KW
3523
3524 /* Failed, have to return the original */
4b88fb76 3525 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3526
3527 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3528 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3529 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3530 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3531 OP_DESC(PL_op),
ab0b796c
KW
3532 original,
3533 original);
051a06d4
KW
3534 Copy(p, ustrp, *lenp, char);
3535 return original;
3536}
3537
b257a28c
KW
3538STATIC UV
3539S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3540 U8 * ustrp, STRLEN *lenp)
3541{
3542 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3543 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3544 * Otherwise, it returns the first code point of the Turkic foldcased
3545 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3546 * contain *lenp bytes
3547 *
3548 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3549 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3550 * DOTLESS I */
3551
3552 PERL_ARGS_ASSERT_TURKIC_FC;
3553 assert(e > p);
3554
3555 if (UNLIKELY(*p == 'I')) {
3556 *lenp = 2;
3557 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3558 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3559 return LATIN_SMALL_LETTER_DOTLESS_I;
3560 }
3561
3562 if (UNLIKELY(memBEGINs(p, e - p,
3563 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3564 {
3565 *lenp = 1;
3566 *ustrp = 'i';
3567 return 'i';
3568 }
3569
3570 return 0;
3571}
3572
3573STATIC UV
3574S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3575 U8 * ustrp, STRLEN *lenp)
3576{
3577 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3578 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3579 * Otherwise, it returns the first code point of the Turkic lowercased
3580 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3581 * contain *lenp bytes */
3582
3583 PERL_ARGS_ASSERT_TURKIC_LC;
3584 assert(e > p0);
3585
3586 /* A 'I' requires context as to what to do */
3587 if (UNLIKELY(*p0 == 'I')) {
3588 const U8 * p = p0 + 1;
3589
3590 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3591 * modified by a dot above lowercases to 'i' even in turkic locales. */
3592 while (p < e) {
3593 UV cp;
3594
3595 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3596 ustrp[0] = 'i';
3597 *lenp = 1;
3598 return 'i';
3599 }
3600
3601 /* For the dot above to modify the 'I', it must be part of a
3602 * combining sequence immediately following the 'I', and no other
3603 * modifier with a ccc of 230 may intervene */
3604 cp = utf8_to_uvchr_buf(p, e, NULL);
3605 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3606 break;
3607 }
3608
3609 /* Here the combining sequence continues */
3610 p += UTF8SKIP(p);
3611 }
3612 }
3613
3614 /* In all other cases the lc is the same as the fold */
3615 return turkic_fc(p0, e, ustrp, lenp);
3616}
3617
3618STATIC UV
3619S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3620 U8 * ustrp, STRLEN *lenp)
3621{
3622 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3623 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3624 * Otherwise, it returns the first code point of the Turkic upper or
3625 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3626 * ustrp will contain *lenp bytes
3627 *
3628 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
a3815e44 3629 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
b257a28c
KW
3630 * DOTLESS I */
3631
3632 PERL_ARGS_ASSERT_TURKIC_UC;
3633 assert(e > p);
3634
3635 if (*p == 'i') {
3636 *lenp = 2;
3637 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3638 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3639 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3640 }
3641
3642 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3643 *lenp = 1;
3644 *ustrp = 'I';
3645 return 'I';
3646 }
3647
3648 return 0;
3649}
3650
eaf412bf
KW
3651/* The process for changing the case is essentially the same for the four case
3652 * change types, except there are complications for folding. Otherwise the
3653 * difference is only which case to change to. To make sure that they all do
3654 * the same thing, the bodies of the functions are extracted out into the
3655 * following two macros. The functions are written with the same variable
3656 * names, and these are known and used inside these macros. It would be
3657 * better, of course, to have inline functions to do it, but since different
3658 * macros are called, depending on which case is being changed to, this is not
3659 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3660 * function can start with the common start macro, then finish with its special
3661 * handling; while the other three cases can just use the common end macro.
3662 *
3663 * The algorithm is to use the proper (passed in) macro or function to change
3664 * the case for code points that are below 256. The macro is used if using
3665 * locale rules for the case change; the function if not. If the code point is
3666 * above 255, it is computed from the input UTF-8, and another macro is called
3667 * to do the conversion. If necessary, the output is converted to UTF-8. If
3668 * using a locale, we have to check that the change did not cross the 255/256
3669 * boundary, see check_locale_boundary_crossing() for further details.
3670 *
3671 * The macros are split with the correct case change for the below-256 case
3672 * stored into 'result', and in the middle of an else clause for the above-255
3673 * case. At that point in the 'else', 'result' is not the final result, but is
3674 * the input code point calculated from the UTF-8. The fold code needs to
3675 * realize all this and take it from there.
3676 *
b257a28c
KW
3677 * To deal with Turkic locales, the function specified by the parameter
3678 * 'turkic' is called when appropriate.
3679 *
eaf412bf
KW
3680 * If you read the two macros as sequential, it's easier to understand what's
3681 * going on. */
3682#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
b257a28c 3683 L1_func_extra_param, turkic) \
a239b1e2 3684 \
eaf412bf 3685 if (flags & (locale_flags)) { \
8b7358b9 3686 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
eaf412bf 3687 if (IN_UTF8_CTYPE_LOCALE) { \
b257a28c
KW
3688 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3689 UV ret = turkic(p, e, ustrp, lenp); \
3690 if (ret) return ret; \
3691 } \
3692 \
3693 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3694 * all */ \
eaf412bf
KW
3695 flags &= ~(locale_flags); \
3696 } \
eaf412bf
KW
3697 } \
3698 \
3699 if (UTF8_IS_INVARIANT(*p)) { \
3700 if (flags & (locale_flags)) { \
3701 result = LC_L1_change_macro(*p); \
3702 } \
3703 else { \
3704 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3705 } \
3706 } \
a239b1e2 3707 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
1a751160 3708 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
eaf412bf 3709 if (flags & (locale_flags)) { \
1a751160