This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Warn on too high a code point if portable warn enabled
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
fb7e7255
KW
39
40/* Be sure to synchronize this message with the similar one in regcomp.c */
760c7c2f 41static const char cp_above_legal_max[] =
d22ec717
KW
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
760c7c2f 44
48ef279e 45/*
ccfc67b7 46=head1 Unicode Support
7fefc6c1 47These are various utility functions for manipulating UTF8-encoded
72d33970 48strings. For the uninitiated, this is a method of representing arbitrary
61296642 49Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
50characters in the ASCII range are unmodified, and a zero byte never appears
51within non-zero characters.
166f8a29 52
eaf7a4d2
CS
53=cut
54*/
55
dd051059
DM
56/* helper for Perl__force_out_malformed_utf8_message(). Like
57 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
58 * PL_compiling */
59
60static void
61S_restore_cop_warnings(pTHX_ void *p)
62{
63 if (!specialWARN(PL_curcop->cop_warnings))
64 PerlMemShared_free(PL_curcop->cop_warnings);
65 PL_curcop->cop_warnings = (STRLEN*)p;
66}
67
68
9cbfb8ab
KW
69void
70Perl__force_out_malformed_utf8_message(pTHX_
71 const U8 *const p, /* First byte in UTF-8 sequence */
72 const U8 * const e, /* Final byte in sequence (may include
73 multiple chars */
74 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
75 usually 0, or some DISALLOW flags */
76 const bool die_here) /* If TRUE, this function does not return */
77{
78 /* This core-only function is to be called when a malformed UTF-8 character
79 * is found, in order to output the detailed information about the
80 * malformation before dieing. The reason it exists is for the occasions
81 * when such a malformation is fatal, but warnings might be turned off, so
82 * that normally they would not be actually output. This ensures that they
83 * do get output. Because a sequence may be malformed in more than one
84 * way, multiple messages may be generated, so we can't make them fatal, as
85 * that would cause the first one to die.
86 *
87 * Instead we pretend -W was passed to perl, then die afterwards. The
88 * flexibility is here to return to the caller so they can finish up and
89 * die themselves */
90 U32 errors;
91
92 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
93
94 ENTER;
c15a80f3 95 SAVEI8(PL_dowarn);
9cbfb8ab
KW
96 SAVESPTR(PL_curcop);
97
98 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
99 if (PL_curcop) {
dd051059
DM
100 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
101 * than PL_compiling */
102 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
103 (void*)PL_curcop->cop_warnings);
9cbfb8ab
KW
104 PL_curcop->cop_warnings = pWARN_ALL;
105 }
106
107 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
108
109 LEAVE;
110
111 if (! errors) {
112 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
113 " be called only when there are errors found");
114 }
115
116 if (die_here) {
117 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
118 }
119}
120
bb07812e
KW
121STATIC HV *
122S_new_msg_hv(pTHX_ const char * const message, /* The message text */
123 U32 categories, /* Packed warning categories */
124 U32 flag) /* Flag associated with this message */
125{
126 /* Creates, populates, and returns an HV* that describes an error message
127 * for the translators between UTF8 and code point */
128
129 SV* msg_sv = newSVpv(message, 0);
130 SV* category_sv = newSVuv(categories);
131 SV* flag_bit_sv = newSVuv(flag);
132
133 HV* msg_hv = newHV();
134
135 PERL_ARGS_ASSERT_NEW_MSG_HV;
136
2b672cf5
KW
137 (void) hv_stores(msg_hv, "text", msg_sv);
138 (void) hv_stores(msg_hv, "warn_categories", category_sv);
139 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
140
141 return msg_hv;
142}
143
eaf7a4d2 144/*
378516de 145=for apidoc uvoffuni_to_utf8_flags
eebe1485 146
a27992cc 147THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
148Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
149L<perlapi/uvchr_to_utf8_flags>>.
a27992cc 150
de69f3af
KW
151This function is like them, but the input is a strict Unicode
152(as opposed to native) code point. Only in very rare circumstances should code
153not be using the native code point.
949cf498 154
09232555 155For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
949cf498 156
eebe1485
SC
157=cut
158*/
159
33f38593
KW
160U8 *
161Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
162{
163 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
164
165 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
166}
167
c94c2f39
KW
168/* All these formats take a single UV code point argument */
169const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
170const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
171 " is not recommended for open interchange";
172const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
173 " may not be portable";
57ff5f59
KW
174const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
175 " Unicode, requires a Perl extension," \
176 " and so is not portable";
c94c2f39 177
33f38593 178#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
179 STMT_START { \
180 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
181 U32 category = packWARN(WARN_SURROGATE); \
182 const char * format = surrogate_cp_format; \
183 if (msgs) { \
184 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
185 category, \
186 UNICODE_GOT_SURROGATE); \
187 } \
188 else { \
189 Perl_ck_warner_d(aTHX_ category, format, uv); \
190 } \
8ee1cdcb
KW
191 } \
192 if (flags & UNICODE_DISALLOW_SURROGATE) { \
193 return NULL; \
194 } \
195 } STMT_END;
196
33f38593 197#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
198 STMT_START { \
199 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
200 U32 category = packWARN(WARN_NONCHAR); \
201 const char * format = nonchar_cp_format; \
202 if (msgs) { \
203 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
204 category, \
205 UNICODE_GOT_NONCHAR); \
206 } \
207 else { \
208 Perl_ck_warner_d(aTHX_ category, format, uv); \
209 } \
8ee1cdcb
KW
210 } \
211 if (flags & UNICODE_DISALLOW_NONCHAR) { \
212 return NULL; \
213 } \
214 } STMT_END;
215
ba6ed43c
KW
216/* Use shorter names internally in this file */
217#define SHIFT UTF_ACCUMULATION_SHIFT
218#undef MARK
219#define MARK UTF_CONTINUATION_MARK
220#define MASK UTF_CONTINUATION_MASK
221
33f38593
KW
222/*
223=for apidoc uvchr_to_utf8_flags_msgs
224
225THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
226
227Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
228
229This function is for code that wants any warning and/or error messages to be
230returned to the caller rather than be displayed. All messages that would have
884a31ee 231been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
232
233It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
234placed after all the others, C<msgs>. If this parameter is 0, this function
235behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
236be a pointer to an C<HV *> variable, in which this function creates a new HV to
237contain any appropriate messages. The hash has three key-value pairs, as
238follows:
239
240=over 4
241
242=item C<text>
243
244The text of the message as a C<SVpv>.
245
246=item C<warn_categories>
247
248The warning category (or categories) packed into a C<SVuv>.
249
250=item C<flag>
251
252A single flag bit associated with this message, in a C<SVuv>.
253The bit corresponds to some bit in the C<*errors> return value,
254such as C<UNICODE_GOT_SURROGATE>.
255
256=back
257
258It's important to note that specifying this parameter as non-null will cause
259any warnings this function would otherwise generate to be suppressed, and
260instead be placed in C<*msgs>. The caller can check the lexical warnings state
261(or not) when choosing what to do with the returned messages.
262
263The caller, of course, is responsible for freeing any returned HV.
264
265=cut
266*/
267
268/* Undocumented; we don't want people using this. Instead they should use
269 * uvchr_to_utf8_flags_msgs() */
dfe13c55 270U8 *
33f38593 271Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 272{
33f38593
KW
273 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
274
275 if (msgs) {
276 *msgs = NULL;
277 }
7918f24d 278
2d1545e5 279 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 280 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
281 return d;
282 }
facc1dc2 283
3ea68d71 284 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
285 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
286 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
287 return d;
288 }
d9432125 289
ba6ed43c
KW
290 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
291 * below, the 16 is for start bytes E0-EF (which are all the possible ones
292 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
293 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
294 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
295 * 0x800-0xFFFF on ASCII */
296 if (uv < (16 * (1U << (2 * SHIFT)))) {
297 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
298 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
299 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
300
301#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
302 aren't tested here */
303 /* The most likely code points in this range are below the surrogates.
304 * Do an extra test to quickly exclude those. */
305 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
306 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
307 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
308 {
33f38593 309 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
310 }
311 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 312 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 313 }
ba6ed43c
KW
314 }
315#endif
316 return d;
317 }
318
319 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
320 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
321 * happen starting with 4-byte characters on ASCII platforms. We unify the
322 * code for these with EBCDIC, even though some of them require 5-bytes on
323 * those, because khw believes the code saving is worth the very slight
324 * performance hit on these high EBCDIC code points. */
325
326 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
24b4c303
KW
327 if (UNLIKELY( uv > MAX_LEGAL_CP
328 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
329 {
40606899 330 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP);
a5bf80e0 331 }
33f38593
KW
332 if ( (flags & UNICODE_WARN_SUPER)
333 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 334 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 335 {
33f38593
KW
336 const char * format = super_cp_format;
337 U32 category = packWARN(WARN_NON_UNICODE);
338 U32 flag = UNICODE_GOT_SUPER;
339
340 /* Choose the more dire applicable warning */
341 if (UNICODE_IS_PERL_EXTENDED(uv)) {
342 format = perl_extended_cp_format;
dc4a6683 343 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
33f38593
KW
344 if (flags & (UNICODE_WARN_PERL_EXTENDED
345 |UNICODE_DISALLOW_PERL_EXTENDED))
346 {
347 flag = UNICODE_GOT_PERL_EXTENDED;
348 }
349 }
a5bf80e0 350
33f38593
KW
351 if (msgs) {
352 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
353 category, flag);
354 }
dc4a6683
KW
355 else if ( ckWARN_d(WARN_NON_UNICODE)
356 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
357 && ckWARN(WARN_PORTABLE)))
358 {
359 Perl_warner(aTHX_ category, format, uv);
33f38593 360 }
a5bf80e0 361 }
56576a04 362 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
363 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
364 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
365 {
366 return NULL;
367 }
368 }
ba6ed43c 369 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 370 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 371 }
d9432125 372
ba6ed43c
KW
373 /* Test for and handle 4-byte result. In the test immediately below, the
374 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
375 * characters). The 3 is for 3 continuation bytes; these each contribute
376 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
377 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
378 * 0x1_0000-0x1F_FFFF on ASCII */
379 if (uv < (8 * (1U << (3 * SHIFT)))) {
380 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
381 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
382 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
383 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
384
385#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
386 characters. The end-plane non-characters for EBCDIC were
387 handled just above */
388 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 389 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 390 }
ba6ed43c 391 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 392 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
393 }
394#endif
395
396 return d;
397 }
398
399 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
400 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
401 * format. The unrolled version above turns out to not save all that much
402 * time, and at these high code points (well above the legal Unicode range
403 * on ASCII platforms, and well above anything in common use in EBCDIC),
404 * khw believes that less code outweighs slight performance gains. */
405
d9432125 406 {
5aaebcb3 407 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
408 U8 *p = d+len-1;
409 while (p > d) {
957a9e81
KW
410 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
411 uv >>= SHIFT;
1d72bdf6 412 }
4c8cd605 413 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
414 return d+len;
415 }
a0ed51b3 416}
a5bf80e0 417
646ca15d 418/*
07693fe6
KW
419=for apidoc uvchr_to_utf8
420
bcb1a2d4 421Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 422of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
423C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
424the byte after the end of the new character. In other words,
07693fe6
KW
425
426 d = uvchr_to_utf8(d, uv);
427
428is the recommended wide native character-aware way of saying
429
430 *(d++) = uv;
431
d22ec717
KW
432This function accepts any code point from 0..C<IV_MAX> as input.
433C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
434
435It is possible to forbid or warn on non-Unicode code points, or those that may
436be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 437
07693fe6
KW
438=cut
439*/
440
de69f3af
KW
441/* This is also a macro */
442PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
443
07693fe6
KW
444U8 *
445Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
446{
de69f3af 447 return uvchr_to_utf8(d, uv);
07693fe6
KW
448}
449
de69f3af
KW
450/*
451=for apidoc uvchr_to_utf8_flags
452
453Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 454of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
455C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
456the byte after the end of the new character. In other words,
de69f3af
KW
457
458 d = uvchr_to_utf8_flags(d, uv, flags);
459
460or, in most cases,
461
462 d = uvchr_to_utf8_flags(d, uv, 0);
463
464This is the Unicode-aware way of saying
465
466 *(d++) = uv;
467
d22ec717
KW
468If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
469input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
470
471Specifying C<flags> can further restrict what is allowed and not warned on, as
472follows:
de69f3af 473
796b6530 474If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
475the function will raise a warning, provided UTF8 warnings are enabled. If
476instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
477NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 478
760c7c2f
KW
479Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
480affect how the function handles a Unicode non-character.
93e6dbd6 481
760c7c2f
KW
482And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
483affect the handling of code points that are above the Unicode maximum of
4840x10FFFF. Languages other than Perl may not be able to accept files that
485contain these.
93e6dbd6
KW
486
487The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
488the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
489three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
490allowed inputs to the strict UTF-8 traditionally defined by Unicode.
491Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
492C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
493above-Unicode and surrogate flags, but not the non-character ones, as
494defined in
e2176993 495L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f 496See L<perlunicode/Noncharacter code points>.
93e6dbd6 497
57ff5f59
KW
498Extremely high code points were never specified in any standard, and require an
499extension to UTF-8 to express, which Perl does. It is likely that programs
500written in something other than Perl would not be able to read files that
501contain these; nor would Perl understand files written by something that uses a
502different extension. For these reasons, there is a separate set of flags that
503can warn and/or disallow these extremely high code points, even if other
504above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
505and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
506L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
507treat all above-Unicode code points, including these, as malformations. (Note
508that the Unicode standard considers anything above 0x10FFFF to be illegal, but
509there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
510
511A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
512retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
513C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
514C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
515platforms,these flags can apply to code points that actually do fit in 31 bits.
516The new names accurately describe the situation in all cases.
de69f3af 517
de69f3af
KW
518=cut
519*/
520
521/* This is also a macro */
522PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
523
07693fe6
KW
524U8 *
525Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
526{
de69f3af 527 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
528}
529
57ff5f59
KW
530#ifndef UV_IS_QUAD
531
e050c007
KW
532STATIC int
533S_is_utf8_cp_above_31_bits(const U8 * const s,
534 const U8 * const e,
535 const bool consider_overlongs)
83dc0f42
KW
536{
537 /* Returns TRUE if the first code point represented by the Perl-extended-
538 * UTF-8-encoded string starting at 's', and looking no further than 'e -
539 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
540 *
541 * The function handles the case where the input bytes do not include all
542 * the ones necessary to represent a full character. That is, they may be
543 * the intial bytes of the representation of a code point, but possibly
544 * the final ones necessary for the complete representation may be beyond
545 * 'e - 1'.
546 *
e050c007
KW
547 * The function also can handle the case where the input is an overlong
548 * sequence. If 'consider_overlongs' is 0, the function assumes the
549 * input is not overlong, without checking, and will return based on that
550 * assumption. If this parameter is 1, the function will go to the trouble
551 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 552 *
e050c007 553 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
554 */
555
e050c007
KW
556 const STRLEN len = e - s;
557 int is_overlong;
558
559 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
560
561 assert(! UTF8_IS_INVARIANT(*s) && e > s);
562
83dc0f42
KW
563#ifdef EBCDIC
564
e050c007 565 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 566
e050c007
KW
567 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
568 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
569 * also be the start byte for a 31-bit code point; we need at least 2
570 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
571 * the start byte for an overlong sequence, but for 30-bit or smaller code
572 * points, so we don't have to worry about overlongs on EBCDIC.) */
573 if (*s != 0xFE) {
574 return 0;
575 }
83dc0f42 576
e050c007
KW
577 if (len == 1) {
578 return -1;
579 }
83dc0f42 580
e050c007 581#else
83dc0f42 582
e050c007
KW
583 /* On ASCII, FE and FF are the only start bytes that can evaluate to
584 * needing more than 31 bits. */
585 if (LIKELY(*s < 0xFE)) {
586 return 0;
587 }
83dc0f42 588
e050c007
KW
589 /* What we have left are FE and FF. Both of these require more than 31
590 * bits unless they are for overlongs. */
591 if (! consider_overlongs) {
592 return 1;
593 }
83dc0f42 594
e050c007
KW
595 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
596 * above 31 bits. But we need more than one byte to discern this, so if
597 * passed just the start byte, it could be an overlong evaluating to
598 * smaller */
599 if (len == 1) {
600 return -1;
601 }
83dc0f42 602
e050c007
KW
603 /* Having excluded len==1, and knowing that FE and FF are both valid start
604 * bytes, we can call the function below to see if the sequence is
605 * overlong. (We don't need the full generality of the called function,
606 * but for these huge code points, speed shouldn't be a consideration, and
607 * the compiler does have enough information, since it's static to this
608 * file, to optimize to just the needed parts.) */
609 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 610
e050c007
KW
611 /* If it isn't overlong, more than 31 bits are required. */
612 if (is_overlong == 0) {
613 return 1;
614 }
83dc0f42 615
e050c007
KW
616 /* If it is indeterminate if it is overlong, return that */
617 if (is_overlong < 0) {
618 return -1;
619 }
620
621 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
622 * the max it can be is 2**31 - 1 */
623 if (*s == 0xFE) {
624 return 0;
83dc0f42
KW
625 }
626
e050c007
KW
627#endif
628
629 /* Here, ASCII and EBCDIC rejoin:
630 * On ASCII: We have an overlong sequence starting with FF
631 * On EBCDIC: We have a sequence starting with FE. */
632
633 { /* For C89, use a block so the declaration can be close to its use */
634
635#ifdef EBCDIC
636
5f995336
KW
637 /* U+7FFFFFFF (2 ** 31 - 1)
638 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
639 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
640 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
641 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
642 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
643 * U+80000000 (2 ** 31):
644 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
645 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
646 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
647 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
648 *
649 * and since we know that *s = \xfe, any continuation sequcence
650 * following it that is gt the below is above 31 bits
651 [0] [1] [2] [3] [4] [5] [6] */
652 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
653
654#else
655
656 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
657 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
658 * FF overlong for U+80000000 (2 ** 31):
659 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
660 * and since we know that *s = \xff, any continuation sequcence
661 * following it that is gt the below is above 30 bits
662 [0] [1] [2] [3] [4] [5] [6] */
663 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 664
83dc0f42
KW
665
666#endif
e050c007
KW
667 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
668 const STRLEN cmp_len = MIN(conts_len, len - 1);
669
670 /* Now compare the continuation bytes in s with the ones we have
671 * compiled in that are for the largest 30 bit code point. If we have
672 * enough bytes available to determine the answer, or the bytes we do
673 * have differ from them, we can compare the two to get a definitive
674 * answer (Note that in UTF-EBCDIC, the two lowest possible
675 * continuation bytes are \x41 and \x42.) */
676 if (cmp_len >= conts_len || memNE(s + 1,
677 conts_for_highest_30_bit,
678 cmp_len))
679 {
680 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
681 }
83dc0f42 682
e050c007
KW
683 /* Here, all the bytes we have are the same as the highest 30-bit code
684 * point, but we are missing so many bytes that we can't make the
685 * determination */
686 return -1;
687 }
83dc0f42
KW
688}
689
57ff5f59
KW
690#endif
691
d6be65ae 692PERL_STATIC_INLINE int
12a4bed3
KW
693S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
694{
d6be65ae
KW
695 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
696 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
697 * it isn't, and -1 if there isn't enough information to tell. This last
698 * return value can happen if the sequence is incomplete, missing some
699 * trailing bytes that would form a complete character. If there are
700 * enough bytes to make a definitive decision, this function does so.
701 * Usually 2 bytes sufficient.
702 *
703 * Overlongs can occur whenever the number of continuation bytes changes.
704 * That means whenever the number of leading 1 bits in a start byte
705 * increases from the next lower start byte. That happens for start bytes
706 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
707 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
708 * ASCII platforms: C0, C1
709 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 710 */
12a4bed3
KW
711
712 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
713 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
714
715 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
716 assert(len > 1 && UTF8_IS_START(*s));
717
718 /* Each platform has overlongs after the start bytes given above (expressed
719 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
720 * the logic is the same, except the E0 overlong has already been excluded
721 * on EBCDIC platforms. The values below were found by manually
722 * inspecting the UTF-8 patterns. See the tables in utf8.h and
723 * utfebcdic.h. */
724
725# ifdef EBCDIC
726# define F0_ABOVE_OVERLONG 0xB0
727# define F8_ABOVE_OVERLONG 0xA8
728# define FC_ABOVE_OVERLONG 0xA4
729# define FE_ABOVE_OVERLONG 0xA2
730# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
731 /* I8(0xfe) is FF */
732# else
733
734 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 735 return 1;
12a4bed3
KW
736 }
737
738# define F0_ABOVE_OVERLONG 0x90
739# define F8_ABOVE_OVERLONG 0x88
740# define FC_ABOVE_OVERLONG 0x84
741# define FE_ABOVE_OVERLONG 0x82
742# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
743# endif
744
745
746 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
747 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
748 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
749 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
750 {
d6be65ae 751 return 1;
12a4bed3
KW
752 }
753
b0b342d4 754 /* Check for the FF overlong */
d6be65ae 755 return isFF_OVERLONG(s, len);
b0b342d4
KW
756}
757
8d6204cc 758PERL_STATIC_INLINE int
b0b342d4
KW
759S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
760{
8d6204cc
KW
761 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
762 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
763 * it isn't, and -1 if there isn't enough information to tell. This last
764 * return value can happen if the sequence is incomplete, missing some
765 * trailing bytes that would form a complete character. If there are
766 * enough bytes to make a definitive decision, this function does so. */
767
b0b342d4 768 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 769
8d6204cc
KW
770 /* To be an FF overlong, all the available bytes must match */
771 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
772 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
773 {
774 return 0;
775 }
776
777 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
778 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 779 * utfebcdic.h. */
8d6204cc
KW
780 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
781 return 1;
782 }
12a4bed3 783
8d6204cc
KW
784 /* The missing bytes could cause the result to go one way or the other, so
785 * the result is indeterminate */
786 return -1;
12a4bed3
KW
787}
788
d22ec717 789#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
790# ifdef EBCDIC /* Actually is I8 */
791# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 792 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
793# else
794# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 795 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
796# endif
797#endif
798
c285bbc4 799PERL_STATIC_INLINE int
e050c007
KW
800S_does_utf8_overflow(const U8 * const s,
801 const U8 * e,
802 const bool consider_overlongs)
a77c906e 803{
c285bbc4 804 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
805 * 'e' - 1 would overflow an IV on this platform; that is if it represents
806 * a code point larger than the highest representable code point. It
807 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
808 * enough information to tell. This last return value can happen if the
809 * sequence is incomplete, missing some trailing bytes that would form a
810 * complete character. If there are enough bytes to make a definitive
811 * decision, this function does so.
c285bbc4 812 *
e050c007
KW
813 * If 'consider_overlongs' is TRUE, the function checks for the possibility
814 * that the sequence is an overlong that doesn't overflow. Otherwise, it
815 * assumes the sequence is not an overlong. This can give different
816 * results only on ASCII 32-bit platforms.
817 *
c285bbc4
KW
818 * (For ASCII platforms, we could use memcmp() because we don't have to
819 * convert each byte to I8, but it's very rare input indeed that would
820 * approach overflow, so the loop below will likely only get executed once.)
821 *
822 * 'e' - 1 must not be beyond a full character. */
a77c906e 823
a77c906e
KW
824
825 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
826 assert(s <= e && s + UTF8SKIP(s) >= e);
827
d22ec717
KW
828#if ! defined(UV_IS_QUAD)
829
830 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
831
832#else
833
834 PERL_UNUSED_ARG(consider_overlongs);
835
836 {
837 const STRLEN len = e - s;
838 const U8 *x;
839 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
840
841 for (x = s; x < e; x++, y++) {
842
843 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
844 continue;
845 }
846
847 /* If this byte is larger than the corresponding highest UTF-8
848 * byte, the sequence overflow; otherwise the byte is less than,
849 * and so the sequence doesn't overflow */
850 return NATIVE_UTF8_TO_I8(*x) > *y;
851
852 }
853
854 /* Got to the end and all bytes are the same. If the input is a whole
855 * character, it doesn't overflow. And if it is a partial character,
856 * there's not enough information to tell */
857 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
858 return -1;
859 }
860
861 return 0;
862 }
863
864#endif
865
866}
867
868#if 0
869
870/* This is the portions of the above function that deal with UV_MAX instead of
871 * IV_MAX. They are left here in case we want to combine them so that internal
872 * uses can have larger code points. The only logic difference is that the
873 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
874 * different logic.
875 */
876
877/* Anything larger than this will overflow the word if it were converted into a UV */
878#if defined(UV_IS_QUAD)
879# ifdef EBCDIC /* Actually is I8 */
880# define HIGHEST_REPRESENTABLE_UTF8 \
881 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
882# else
883# define HIGHEST_REPRESENTABLE_UTF8 \
884 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
885# endif
886#else /* 32-bit */
887# ifdef EBCDIC
888# define HIGHEST_REPRESENTABLE_UTF8 \
889 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
890# else
891# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
892# endif
893#endif
894
a77c906e
KW
895#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
896
897 /* On 32 bit ASCII machines, many overlongs that start with FF don't
898 * overflow */
e050c007 899 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
900
901 /* To be such an overlong, the first bytes of 's' must match
902 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
903 * don't have any additional bytes available, the sequence, when
904 * completed might or might not fit in 32 bits. But if we have that
905 * next byte, we can tell for sure. If it is <= 0x83, then it does
906 * fit. */
907 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
908 return -1;
909 }
910
911 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
912 }
913
d22ec717
KW
914/* Starting with the #else, the rest of the function is identical except
915 * 1. we need to move the 'len' declaration to be global to the function
916 * 2. the endif move to just after the UNUSED_ARG.
917 * An empty endif is given just below to satisfy the preprocessor
918 */
a77c906e
KW
919#endif
920
d22ec717 921#endif
a77c906e 922
12a4bed3
KW
923#undef F0_ABOVE_OVERLONG
924#undef F8_ABOVE_OVERLONG
925#undef FC_ABOVE_OVERLONG
926#undef FE_ABOVE_OVERLONG
927#undef FF_OVERLONG_PREFIX
928
35f8c9bd 929STRLEN
1376b35c 930Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 931{
2b479609 932 STRLEN len;
12a4bed3 933 const U8 *x;
35f8c9bd 934
2b479609
KW
935 /* A helper function that should not be called directly.
936 *
937 * This function returns non-zero if the string beginning at 's' and
938 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
939 * code point; otherwise it returns 0. The examination stops after the
940 * first code point in 's' is validated, not looking at the rest of the
941 * input. If 'e' is such that there are not enough bytes to represent a
942 * complete code point, this function will return non-zero anyway, if the
943 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
944 * excluded by 'flags'.
945 *
946 * A non-zero return gives the number of bytes required to represent the
947 * code point. Be aware that if the input is for a partial character, the
948 * return will be larger than 'e - s'.
949 *
950 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
951 * The caller should have excluded the possibility of it being invariant
952 * before calling this function.
2b479609
KW
953 *
954 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
955 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
956 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
957 * disallowed by the flags. If the input is only for a partial character,
958 * the function will return non-zero if there is any sequence of
959 * well-formed UTF-8 that, when appended to the input sequence, could
960 * result in an allowed code point; otherwise it returns 0. Non characters
961 * cannot be determined based on partial character input. But many of the
962 * other excluded types can be determined with just the first one or two
963 * bytes.
964 *
965 */
966
1376b35c 967 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
2b479609
KW
968
969 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 970 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 971 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 972
2b479609 973 /* A variant char must begin with a start byte */
35f8c9bd
KW
974 if (UNLIKELY(! UTF8_IS_START(*s))) {
975 return 0;
976 }
977
edc2c47a
KW
978 /* Examine a maximum of a single whole code point */
979 if (e - s > UTF8SKIP(s)) {
980 e = s + UTF8SKIP(s);
981 }
982
2b479609
KW
983 len = e - s;
984
985 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
986 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 987
56576a04
KW
988 /* Here, we are disallowing some set of largish code points, and the
989 * first byte indicates the sequence is for a code point that could be
990 * in the excluded set. We generally don't have to look beyond this or
991 * the second byte to see if the sequence is actually for one of the
992 * excluded classes. The code below is derived from this table:
993 *
2b479609
KW
994 * UTF-8 UTF-EBCDIC I8
995 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
996 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
997 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
998 *
56576a04
KW
999 * Keep in mind that legal continuation bytes range between \x80..\xBF
1000 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
1001 * continuation bytes. Hence, we don't have to test the upper edge
1002 * because if any of those is encountered, the sequence is malformed,
1003 * and would fail elsewhere in this function.
1004 *
1005 * The code here likewise assumes that there aren't other
1006 * malformations; again the function should fail elsewhere because of
1007 * these. For example, an overlong beginning with FC doesn't actually
1008 * have to be a super; it could actually represent a small code point,
1009 * even U+0000. But, since overlongs (and other malformations) are
1010 * illegal, the function should return FALSE in either case.
2b479609
KW
1011 */
1012
1013#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1014# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 1015# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 1016
19794540
KW
1017# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1018 /* B6 and B7 */ \
1019 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1020# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1021#else
1022# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1023# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1024# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1025# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1026#endif
1027
1028 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1029 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1030 {
2b479609
KW
1031 return 0; /* Above Unicode */
1032 }
1033
d044b7a7 1034 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1035 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1036 {
57ff5f59 1037 return 0;
2b479609
KW
1038 }
1039
1040 if (len > 1) {
1041 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1042
1043 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1044 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1045 {
1046 return 0; /* Above Unicode */
1047 }
1048
1049 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1050 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1051 {
1052 return 0; /* Surrogate */
1053 }
1054
1055 if ( (flags & UTF8_DISALLOW_NONCHAR)
1056 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1057 {
1058 return 0; /* Noncharacter code point */
1059 }
1060 }
1061 }
1062
1063 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1064 for (x = s + 1; x < e; x++) {
1065 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1066 return 0;
1067 }
1068 }
1069
af13dd8a 1070 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1071 * overlong. */
d6be65ae 1072 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1073 return 0;
af13dd8a
KW
1074 }
1075
12a4bed3
KW
1076 /* And finally, that the code point represented fits in a word on this
1077 * platform */
e050c007
KW
1078 if (0 < does_utf8_overflow(s, e,
1079 0 /* Don't consider overlongs */
1080 ))
1081 {
12a4bed3 1082 return 0;
35f8c9bd
KW
1083 }
1084
2b479609 1085 return UTF8SKIP(s);
35f8c9bd
KW
1086}
1087
7e2f38b2 1088char *
63ab03b3 1089Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1090{
1091 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1092 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1093 * Currently, there are only two formats, so it is currently a bool:
1094 * 0 \xab
1095 * 1 ab (that is a space between two hex digit bytes)
1096 */
7cf8d05d
KW
1097
1098 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1099 trailing NUL */
63ab03b3
KW
1100 const U8 * s = start;
1101 const U8 * const e = start + len;
7cf8d05d
KW
1102 char * output;
1103 char * d;
1104
1105 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1106
1107 Newx(output, output_len, char);
1108 SAVEFREEPV(output);
1109
1110 d = output;
63ab03b3 1111 for (s = start; s < e; s++) {
7cf8d05d
KW
1112 const unsigned high_nibble = (*s & 0xF0) >> 4;
1113 const unsigned low_nibble = (*s & 0x0F);
1114
7e2f38b2 1115 if (format) {
63ab03b3
KW
1116 if (s > start) {
1117 *d++ = ' ';
1118 }
7e2f38b2
KW
1119 }
1120 else {
1121 *d++ = '\\';
1122 *d++ = 'x';
1123 }
7cf8d05d
KW
1124
1125 if (high_nibble < 10) {
1126 *d++ = high_nibble + '0';
1127 }
1128 else {
1129 *d++ = high_nibble - 10 + 'a';
1130 }
1131
1132 if (low_nibble < 10) {
1133 *d++ = low_nibble + '0';
1134 }
1135 else {
1136 *d++ = low_nibble - 10 + 'a';
1137 }
1138 }
1139
1140 *d = '\0';
1141 return output;
1142}
1143
806547a7 1144PERL_STATIC_INLINE char *
7cf8d05d
KW
1145S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1146
421da25c 1147 /* Max number of bytes to print */
3cc6a05e 1148 STRLEN print_len,
7cf8d05d
KW
1149
1150 /* Which one is the non-continuation */
1151 const STRLEN non_cont_byte_pos,
1152
1153 /* How many bytes should there be? */
1154 const STRLEN expect_len)
806547a7
KW
1155{
1156 /* Return the malformation warning text for an unexpected continuation
1157 * byte. */
1158
7cf8d05d 1159 const char * const where = (non_cont_byte_pos == 1)
806547a7 1160 ? "immediately"
7cf8d05d
KW
1161 : Perl_form(aTHX_ "%d bytes",
1162 (int) non_cont_byte_pos);
421da25c
KW
1163 const U8 * x = s + non_cont_byte_pos;
1164 const U8 * e = s + print_len;
806547a7
KW
1165
1166 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1167
7cf8d05d
KW
1168 /* We don't need to pass this parameter, but since it has already been
1169 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1170 assert(expect_len == UTF8SKIP(s));
1171
421da25c
KW
1172 /* As a defensive coding measure, don't output anything past a NUL. Such
1173 * bytes shouldn't be in the middle of a malformation, and could mark the
1174 * end of the allocated string, and what comes after is undefined */
1175 for (; x < e; x++) {
1176 if (*x == '\0') {
1177 x++; /* Output this particular NUL */
1178 break;
1179 }
1180 }
1181
7cf8d05d
KW
1182 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1183 " %s after start byte 0x%02x; need %d bytes, got %d)",
1184 malformed_text,
421da25c 1185 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1186 *(s + non_cont_byte_pos),
1187 where,
1188 *s,
1189 (int) expect_len,
1190 (int) non_cont_byte_pos);
806547a7
KW
1191}
1192
35f8c9bd
KW
1193/*
1194
de69f3af 1195=for apidoc utf8n_to_uvchr
378516de
KW
1196
1197THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1198Most code should use L</utf8_to_uvchr_buf>() rather than call this
1199directly.
67e989fb 1200
9041c2e3 1201Bottom level UTF-8 decode routine.
de69f3af 1202Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1203which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1204C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1205the length, in bytes, of that character.
949cf498
KW
1206
1207The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1208well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1209causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1210is the next possible position in C<s> that could begin a non-malformed
1211character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1212is raised. Some UTF-8 input sequences may contain multiple malformations.
1213This function tries to find every possible one in each call, so multiple
56576a04 1214warnings can be raised for the same sequence.
949cf498
KW
1215
1216Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1217individual types of malformations, such as the sequence being overlong (that
1218is, when there is a shorter sequence that can express the same code point;
1219overlong sequences are expressly forbidden in the UTF-8 standard due to
1220potential security issues). Another malformation example is the first byte of
1221a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1222flags. Even if allowed, this function generally returns the Unicode
1223REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1224F<utf8.h> to override this behavior for the overlong malformations, but don't
1225do that except for very specialized purposes.
949cf498 1226
796b6530 1227The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1228flags) malformation is found. If this flag is set, the routine assumes that
1229the caller will raise a warning, and this function will silently just set
d088425d
KW
1230C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1231
75200dff 1232Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1233character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1234in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1235be set to 1. To disambiguate, upon a zero return, see if the first byte of
1236C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1237error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1238
1239Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1240Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1241By default these are considered regular code points, but certain situations
ecc1615f
KW
1242warrant special handling for them, which can be specified using the C<flags>
1243parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1244three classes are treated as malformations and handled as such. The flags
1245C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1246C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1247disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1248restricts the allowed inputs to the strict UTF-8 traditionally defined by
1249Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1250definition given by
e2176993 1251L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f
KW
1252The difference between traditional strictness and C9 strictness is that the
1253latter does not forbid non-character code points. (They are still discouraged,
1254however.) For more discussion see L<perlunicode/Noncharacter code points>.
1255
1256The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1257C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1258C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1259raised for their respective categories, but otherwise the code points are
1260considered valid (not malformations). To get a category to both be treated as
1261a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1262(But note that warnings are not raised if lexically disabled nor if
796b6530 1263C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1264
57ff5f59
KW
1265Extremely high code points were never specified in any standard, and require an
1266extension to UTF-8 to express, which Perl does. It is likely that programs
1267written in something other than Perl would not be able to read files that
1268contain these; nor would Perl understand files written by something that uses a
1269different extension. For these reasons, there is a separate set of flags that
1270can warn and/or disallow these extremely high code points, even if other
1271above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1272C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1273L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1274above-Unicode code points, including these, as malformations.
1275(Note that the Unicode standard considers anything above 0x10FFFF to be
1276illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1277(2**31 -1))
1278
1279A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1280retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1281C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1282C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1283can apply to code points that actually do fit in 31 bits. This happens on
1284EBCDIC platforms, and sometimes when the L<overlong
1285malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1286describe the situation in all cases.
1287
ab8e6d41 1288
949cf498
KW
1289All other code points corresponding to Unicode characters, including private
1290use and those yet to be assigned, are never considered malformed and never
1291warn.
67e989fb 1292
5af38e47
KW
1293=for apidoc Amnh||UTF8_CHECK_ONLY
1294=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1295=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1296=for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1297=for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1298=for apidoc Amnh||UTF8_DISALLOW_SUPER
1299=for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1300=for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1301=for apidoc Amnh||UTF8_WARN_SURROGATE
1302=for apidoc Amnh||UTF8_WARN_NONCHAR
1303=for apidoc Amnh||UTF8_WARN_SUPER
1304=for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1305=for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1306
37607a96 1307=cut
f9380377
KW
1308
1309Also implemented as a macro in utf8.h
1310*/
1311
1312UV
e6a4ffc3
KW
1313Perl_utf8n_to_uvchr(const U8 *s,
1314 STRLEN curlen,
1315 STRLEN *retlen,
1316 const U32 flags)
f9380377
KW
1317{
1318 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1319
1320 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1321}
1322
1323/*
1324
1325=for apidoc utf8n_to_uvchr_error
1326
1327THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1328Most code should use L</utf8_to_uvchr_buf>() rather than call this
1329directly.
f9380377
KW
1330
1331This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1332are when an error is found. If you also need to know the generated warning
1333messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1334
1335It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1336all the others, C<errors>. If this parameter is 0, this function behaves
1337identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1338to a C<U32> variable, which this function sets to indicate any errors found.
1339Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1340C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1341of these bits will be set if a malformation is found, even if the input
7a65503b 1342C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1343exceptions are noted:
1344
1345=over 4
1346
57ff5f59 1347=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1348
57ff5f59
KW
1349The input sequence is not standard UTF-8, but a Perl extension. This bit is
1350set only if the input C<flags> parameter contains either the
1351C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1352
1353Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1354and so some extension must be used to express them. Perl uses a natural
1355extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1356extension to represent even higher ones, so that any code point that fits in a
135764-bit word can be represented. Text using these extensions is not likely to
1358be portable to non-Perl code. We lump both of these extensions together and
1359refer to them as Perl extended UTF-8. There exist other extensions that people
1360have invented, incompatible with Perl's.
1361
1362On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1363extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1364than on ASCII. Prior to that, code points 2**31 and higher were simply
1365unrepresentable, and a different, incompatible method was used to represent
1366code points between 2**30 and 2**31 - 1.
1367
1368On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1369Perl extended UTF-8 is used.
1370
1371In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1372may use for backward compatibility. That name is misleading, as this flag may
1373be set when the code point actually does fit in 31 bits. This happens on
1374EBCDIC platforms, and sometimes when the L<overlong
1375malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1376describes the situation in all cases.
f9380377
KW
1377
1378=item C<UTF8_GOT_CONTINUATION>
1379
1380The input sequence was malformed in that the first byte was a a UTF-8
1381continuation byte.
1382
1383=item C<UTF8_GOT_EMPTY>
1384
1385The input C<curlen> parameter was 0.
1386
1387=item C<UTF8_GOT_LONG>
1388
1389The input sequence was malformed in that there is some other sequence that
1390evaluates to the same code point, but that sequence is shorter than this one.
1391
fecaf136
KW
1392Until Unicode 3.1, it was legal for programs to accept this malformation, but
1393it was discovered that this created security issues.
1394
f9380377
KW
1395=item C<UTF8_GOT_NONCHAR>
1396
1397The code point represented by the input UTF-8 sequence is for a Unicode
1398non-character code point.
1399This bit is set only if the input C<flags> parameter contains either the
1400C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1401
1402=item C<UTF8_GOT_NON_CONTINUATION>
1403
1404The input sequence was malformed in that a non-continuation type byte was found
00d976bb
KW
1405in a position where only a continuation type one should be. See also
1406L</C<UTF8_GOT_SHORT>>.
f9380377
KW
1407
1408=item C<UTF8_GOT_OVERFLOW>
1409
1410The input sequence was malformed in that it is for a code point that is not
d22ec717 1411representable in the number of bits available in an IV on the current platform.
f9380377
KW
1412
1413=item C<UTF8_GOT_SHORT>
1414
1415The input sequence was malformed in that C<curlen> is smaller than required for
1416a complete sequence. In other words, the input is for a partial character
1417sequence.
1418
00d976bb
KW
1419
1420C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1421sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1422that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1423sequence was looked at. If no other flags are present, it means that the
1424sequence was valid as far as it went. Depending on the application, this could
1425mean one of three things:
1426
1427=over
1428
1429=item *
1430
1431The C<curlen> length parameter passed in was too small, and the function was
1432prevented from examining all the necessary bytes.
1433
1434=item *
1435
1436The buffer being looked at is based on reading data, and the data received so
1437far stopped in the middle of a character, so that the next read will
1438read the remainder of this character. (It is up to the caller to deal with the
1439split bytes somehow.)
1440
1441=item *
1442
1443This is a real error, and the partial sequence is all we're going to get.
1444
1445=back
1446
f9380377
KW
1447=item C<UTF8_GOT_SUPER>
1448
1449The input sequence was malformed in that it is for a non-Unicode code point;
1450that is, one above the legal Unicode maximum.
1451This bit is set only if the input C<flags> parameter contains either the
1452C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1453
1454=item C<UTF8_GOT_SURROGATE>
1455
1456The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1457code point.
1458This bit is set only if the input C<flags> parameter contains either the
1459C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1460
1461=back
1462
133551d8
KW
1463To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1464flag to suppress any warnings, and then examine the C<*errors> return.
1465
f9380377 1466=cut
37657a5b
KW
1467
1468Also implemented as a macro in utf8.h
37607a96 1469*/
67e989fb 1470
a0ed51b3 1471UV
e6a4ffc3 1472Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1473 STRLEN curlen,
1474 STRLEN *retlen,
1475 const U32 flags,
1476 U32 * errors)
1477{
1478 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1479
1480 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1481}
1482
1483/*
1484
1485=for apidoc utf8n_to_uvchr_msgs
1486
1487THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1488Most code should use L</utf8_to_uvchr_buf>() rather than call this
1489directly.
37657a5b
KW
1490
1491This function is for code that needs to know what the precise malformation(s)
1492are when an error is found, and wants the corresponding warning and/or error
1493messages to be returned to the caller rather than be displayed. All messages
1494that would have been displayed if all lexcial warnings are enabled will be
1495returned.
1496
1497It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1498placed after all the others, C<msgs>. If this parameter is 0, this function
1499behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1500be a pointer to an C<AV *> variable, in which this function creates a new AV to
1501contain any appropriate messages. The elements of the array are ordered so
1502that the first message that would have been displayed is in the 0th element,
1503and so on. Each element is a hash with three key-value pairs, as follows:
1504
1505=over 4
1506
1507=item C<text>
1508
1509The text of the message as a C<SVpv>.
1510
1511=item C<warn_categories>
1512
1513The warning category (or categories) packed into a C<SVuv>.
1514
1515=item C<flag>
1516
1517A single flag bit associated with this message, in a C<SVuv>.
1518The bit corresponds to some bit in the C<*errors> return value,
1519such as C<UTF8_GOT_LONG>.
1520
1521=back
1522
1523It's important to note that specifying this parameter as non-null will cause
1524any warnings this function would otherwise generate to be suppressed, and
1525instead be placed in C<*msgs>. The caller can check the lexical warnings state
1526(or not) when choosing what to do with the returned messages.
1527
1528If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1529no AV is created.
1530
1531The caller, of course, is responsible for freeing any returned AV.
1532
1533=cut
1534*/
1535
1536UV
e6a4ffc3 1537Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1538 STRLEN curlen,
1539 STRLEN *retlen,
1540 const U32 flags,
1541 U32 * errors,
1542 AV ** msgs)
a0ed51b3 1543{
d4c19fe8 1544 const U8 * const s0 = s;
2b9519f0 1545 const U8 * send = s0 + curlen;
5af9f822
KW
1546 U32 possible_problems; /* A bit is set here for each potential problem
1547 found as we go along */
1548 UV uv;
1549 STRLEN expectlen; /* How long should this sequence be? */
1550 STRLEN avail_len; /* When input is too short, gives what that is */
1551 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1552 gets set and discarded */
a0dbb045 1553
2b5e7bc2
KW
1554 /* The below are used only if there is both an overlong malformation and a
1555 * too short one. Otherwise the first two are set to 's0' and 'send', and
1556 * the third not used at all */
5af9f822 1557 U8 * adjusted_s0;
e9f2c446
KW
1558 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1559 routine; see [perl #130921] */
5af9f822 1560 UV uv_so_far;
e6a4ffc3 1561 dTHX;
5af9f822 1562
e6a4ffc3 1563 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1564
1565 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1566 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1567 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1568 * final case. */
1569
1570#ifndef EBCDIC
1571
1572 /* Each of the affected Hanguls starts with \xED */
1573
1574 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1575 if (retlen) {
1576 *retlen = 3;
1577 }
1578 if (errors) {
1579 *errors = 0;
1580 }
1581 if (msgs) {
1582 *msgs = NULL;
1583 }
1584
1585 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1586 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1587 | (s0[2] & UTF_CONTINUATION_MASK);
1588 }
1589
1590#endif
1591
1592 /* In conjunction with the exhaustive tests that can be enabled in
1593 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1594 * what it is intended to do, and that no flaws in it are masked by
1595 * dropping down and executing the code below
1596 assert(! isUTF8_CHAR(s0, send)
1597 || UTF8_IS_SURROGATE(s0, send)
1598 || UTF8_IS_SUPER(s0, send)
1599 || UTF8_IS_NONCHAR(s0,send));
1600 */
1601
1602 s = s0;
1603 uv = *s0;
1604 possible_problems = 0;
1605 expectlen = 0;
1606 avail_len = 0;
1607 discard_errors = 0;
1608 adjusted_s0 = (U8 *) s0;
1609 uv_so_far = 0;
1610
f9380377
KW
1611 if (errors) {
1612 *errors = 0;
1613 }
1614 else {
1615 errors = &discard_errors;
1616 }
a0dbb045 1617
eb83ed87
KW
1618 /* The order of malformation tests here is important. We should consume as
1619 * few bytes as possible in order to not skip any valid character. This is
1620 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
e2176993 1621 * https://unicode.org/reports/tr36 for more discussion as to why. For
eb83ed87
KW
1622 * example, once we've done a UTF8SKIP, we can tell the expected number of
1623 * bytes, and could fail right off the bat if the input parameters indicate
1624 * that there are too few available. But it could be that just that first
1625 * byte is garbled, and the intended character occupies fewer bytes. If we
1626 * blindly assumed that the first byte is correct, and skipped based on
1627 * that number, we could skip over a valid input character. So instead, we
1628 * always examine the sequence byte-by-byte.
1629 *
1630 * We also should not consume too few bytes, otherwise someone could inject
1631 * things. For example, an input could be deliberately designed to
1632 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1633 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1634 * which is actually part of of the overflowing sequence), that could look
1635 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1636 * sequence and process the rest, inappropriately.
1637 *
1638 * Some possible input sequences are malformed in more than one way. This
1639 * function goes to lengths to try to find all of them. This is necessary
1640 * for correctness, as the inputs may allow one malformation but not
1641 * another, and if we abandon searching for others after finding the
1642 * allowed one, we could allow in something that shouldn't have been.
1643 */
eb83ed87 1644
b5b9af04 1645 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1646 possible_problems |= UTF8_GOT_EMPTY;
1647 curlen = 0;
5a48568d 1648 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1649 goto ready_to_handle_errors;
0c443dc2
JH
1650 }
1651
eb83ed87
KW
1652 expectlen = UTF8SKIP(s);
1653
1654 /* A well-formed UTF-8 character, as the vast majority of calls to this
1655 * function will be for, has this expected length. For efficiency, set
1656 * things up here to return it. It will be overriden only in those rare
1657 * cases where a malformation is found */
1658 if (retlen) {
1659 *retlen = expectlen;
1660 }
1661
eb83ed87 1662 /* A continuation character can't start a valid sequence */
b5b9af04 1663 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1664 possible_problems |= UTF8_GOT_CONTINUATION;
1665 curlen = 1;
1666 uv = UNICODE_REPLACEMENT;
1667 goto ready_to_handle_errors;
ba210ebe 1668 }
9041c2e3 1669
dcd27b3c 1670 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1671 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1672 * because it excludes start bytes like \xC0 that always lead to
1673 * overlongs.) */
dcd27b3c 1674
534752c1
KW
1675 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1676 * that indicate the number of bytes in the character's whole UTF-8
1677 * sequence, leaving just the bits that are part of the value. */
1678 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1679
e308b348
KW
1680 /* Setup the loop end point, making sure to not look past the end of the
1681 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1682 if (UNLIKELY(curlen < expectlen)) {
1683 possible_problems |= UTF8_GOT_SHORT;
1684 avail_len = curlen;
e308b348
KW
1685 }
1686 else {
2b9519f0 1687 send = (U8*) s0 + expectlen;
e308b348 1688 }
e308b348 1689
eb83ed87 1690 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1691 * accumulating each into the working value as we go. */
eb83ed87 1692 for (s = s0 + 1; s < send; s++) {
b5b9af04 1693 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1694 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1695 continue;
1696 }
1697
1698 /* Here, found a non-continuation before processing all expected bytes.
1699 * This byte indicates the beginning of a new character, so quit, even
1700 * if allowing this malformation. */
2b5e7bc2 1701 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1702 break;
eb83ed87
KW
1703 } /* End of loop through the character's bytes */
1704
1705 /* Save how many bytes were actually in the character */
1706 curlen = s - s0;
1707
2b5e7bc2
KW
1708 /* Note that there are two types of too-short malformation. One is when
1709 * there is actual wrong data before the normal termination of the
1710 * sequence. The other is that the sequence wasn't complete before the end
1711 * of the data we are allowed to look at, based on the input 'curlen'.
1712 * This means that we were passed data for a partial character, but it is
1713 * valid as far as we saw. The other is definitely invalid. This
1714 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1715 * separate.
1716 *
1717 * A convenience macro that matches either of the too-short conditions. */
1718# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1719
1720 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1721 uv_so_far = uv;
1722 uv = UNICODE_REPLACEMENT;
1723 }
2b5e7bc2 1724
08e73697
KW
1725 /* Check for overflow. The algorithm requires us to not look past the end
1726 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1727 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1728 1 /* Do consider overlongs */
1729 )))
1730 {
2b5e7bc2
KW
1731 possible_problems |= UTF8_GOT_OVERFLOW;
1732 uv = UNICODE_REPLACEMENT;
eb83ed87 1733 }
eb83ed87 1734
2b5e7bc2
KW
1735 /* Check for overlong. If no problems so far, 'uv' is the correct code
1736 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1737 * we must look at the UTF-8 byte sequence itself to see if it is for an
1738 * overlong */
1739 if ( ( LIKELY(! possible_problems)
1740 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1741 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1742 && ( UNLIKELY(! UTF8_IS_START(*s0))
1743 || ( curlen > 1
d6be65ae 1744 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1745 s - s0))))))
2f8f112e 1746 {
2b5e7bc2
KW
1747 possible_problems |= UTF8_GOT_LONG;
1748
abc28b54 1749 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1750
abc28b54
KW
1751 /* The calculation in the 'true' branch of this 'if'
1752 * below won't work if overflows, and isn't needed
1753 * anyway. Further below we handle all overflow
1754 * cases */
1755 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1756 {
2b5e7bc2
KW
1757 UV min_uv = uv_so_far;
1758 STRLEN i;
1759
1760 /* Here, the input is both overlong and is missing some trailing
1761 * bytes. There is no single code point it could be for, but there
1762 * may be enough information present to determine if what we have
1763 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1764 * The code further below has the intelligence to determine this,
1765 * but just for non-overlong UTF-8 sequences. What we do here is
1766 * calculate the smallest code point the input could represent if
1767 * there were no too short malformation. Then we compute and save
1768 * the UTF-8 for that, which is what the code below looks at
1769 * instead of the raw input. It turns out that the smallest such
1770 * code point is all we need. */
2b5e7bc2
KW
1771 for (i = curlen; i < expectlen; i++) {
1772 min_uv = UTF8_ACCUMULATE(min_uv,
1773 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1774 }
1775
e9f2c446 1776 adjusted_s0 = temp_char_buf;
57ff5f59 1777 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1778 }
eb83ed87
KW
1779 }
1780
56576a04
KW
1781 /* Here, we have found all the possible problems, except for when the input
1782 * is for a problematic code point not allowed by the input parameters. */
1783
06188866
KW
1784 /* uv is valid for overlongs */
1785 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1786
1787 /* isn't problematic if < this */
1788 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1789 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1790
1791 /* if overflow, we know without looking further
1792 * precisely which of the problematic types it is,
1793 * and we deal with those in the overflow handling
1794 * code */
1795 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1796 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1797 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1798 && ((flags & ( UTF8_DISALLOW_NONCHAR
1799 |UTF8_DISALLOW_SURROGATE
1800 |UTF8_DISALLOW_SUPER
d044b7a7 1801 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1802 |UTF8_WARN_NONCHAR
1803 |UTF8_WARN_SURROGATE
1804 |UTF8_WARN_SUPER
d22ec717 1805 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1806 {
2b5e7bc2
KW
1807 /* If there were no malformations, or the only malformation is an
1808 * overlong, 'uv' is valid */
1809 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1810 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1811 possible_problems |= UTF8_GOT_SURROGATE;
1812 }
1813 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1814 possible_problems |= UTF8_GOT_SUPER;
1815 }
1816 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1817 possible_problems |= UTF8_GOT_NONCHAR;
1818 }
1819 }
1820 else { /* Otherwise, need to look at the source UTF-8, possibly
1821 adjusted to be non-overlong */
1822
1823 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1824 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1825 {
2b5e7bc2
KW
1826 possible_problems |= UTF8_GOT_SUPER;
1827 }
1828 else if (curlen > 1) {
1829 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1830 NATIVE_UTF8_TO_I8(*adjusted_s0),
1831 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1832 {
2b5e7bc2 1833 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1834 }
2b5e7bc2
KW
1835 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1836 NATIVE_UTF8_TO_I8(*adjusted_s0),
1837 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1838 {
1839 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1840 }
1841 }
c0236afe 1842
2b5e7bc2
KW
1843 /* We need a complete well-formed UTF-8 character to discern
1844 * non-characters, so can't look for them here */
1845 }
1846 }
949cf498 1847
2b5e7bc2
KW
1848 ready_to_handle_errors:
1849
1850 /* At this point:
1851 * curlen contains the number of bytes in the sequence that
1852 * this call should advance the input by.
e308b348
KW
1853 * avail_len gives the available number of bytes passed in, but
1854 * only if this is less than the expected number of
1855 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1856 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1857 * is set in it for each potential problem found.
1858 * uv contains the code point the input sequence
1859 * represents; or if there is a problem that prevents
1860 * a well-defined value from being computed, it is
1861 * some subsitute value, typically the REPLACEMENT
1862 * CHARACTER.
1863 * s0 points to the first byte of the character
56576a04
KW
1864 * s points to just after were we left off processing
1865 * the character
1866 * send points to just after where that character should
1867 * end, based on how many bytes the start byte tells
1868 * us should be in it, but no further than s0 +
1869 * avail_len
2b5e7bc2 1870 */
eb83ed87 1871
2b5e7bc2
KW
1872 if (UNLIKELY(possible_problems)) {
1873 bool disallowed = FALSE;
1874 const U32 orig_problems = possible_problems;
1875
37657a5b
KW
1876 if (msgs) {
1877 *msgs = NULL;
1878 }
1879
2b5e7bc2 1880 while (possible_problems) { /* Handle each possible problem */
9fde5914 1881 U32 pack_warn = 0;
2b5e7bc2 1882 char * message = NULL;
37657a5b 1883 U32 this_flag_bit = 0;
2b5e7bc2
KW
1884
1885 /* Each 'if' clause handles one problem. They are ordered so that
1886 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1887 * ones; this is kinda in decreasing severity order. But the
1888 * overlong must come last, as it changes 'uv' looked at by the
1889 * others */
2b5e7bc2
KW
1890 if (possible_problems & UTF8_GOT_OVERFLOW) {
1891
56576a04
KW
1892 /* Overflow means also got a super and are using Perl's
1893 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1894 possible_problems
d044b7a7 1895 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1896 *errors |= UTF8_GOT_OVERFLOW;
1897
1898 /* But the API says we flag all errors found */
1899 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1900 *errors |= UTF8_GOT_SUPER;
1901 }
ddb65933 1902 if (flags
d044b7a7 1903 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1904 {
d044b7a7 1905 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1906 }
2b5e7bc2 1907
d60baaa7 1908 /* Disallow if any of the three categories say to */
56576a04 1909 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1910 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1911 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1912 {
1913 disallowed = TRUE;
1914 }
1915
d22ec717
KW
1916 /* Likewise, warn if any say to */
1917 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1918 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1919 {
2b5e7bc2 1920
ddb65933
KW
1921 /* The warnings code explicitly says it doesn't handle the
1922 * case of packWARN2 and two categories which have
1923 * parent-child relationship. Even if it works now to
1924 * raise the warning if either is enabled, it wouldn't
1925 * necessarily do so in the future. We output (only) the
56576a04 1926 * most dire warning */
ddb65933 1927 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1928 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1929 pack_warn = packWARN(WARN_UTF8);
1930 }
37657a5b 1931 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1932 pack_warn = packWARN(WARN_NON_UNICODE);
1933 }
1934 if (pack_warn) {
1935 message = Perl_form(aTHX_ "%s: %s (overflows)",
1936 malformed_text,
05b9033b 1937 _byte_dump_string(s0, curlen, 0));
37657a5b 1938 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1939 }
2b5e7bc2
KW
1940 }
1941 }
1942 }
1943 else if (possible_problems & UTF8_GOT_EMPTY) {
1944 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1945 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1946
1947 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1948
1949 /* This so-called malformation is now treated as a bug in
1950 * the caller. If you have nothing to decode, skip calling
1951 * this function */
1952 assert(0);
1953
2b5e7bc2 1954 disallowed = TRUE;
37657a5b
KW
1955 if ( (msgs
1956 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1957 {
2b5e7bc2
KW
1958 pack_warn = packWARN(WARN_UTF8);
1959 message = Perl_form(aTHX_ "%s (empty string)",
1960 malformed_text);
37657a5b 1961 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1962 }
1963 }
1964 }
1965 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1966 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1967 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1968
1969 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1970 disallowed = TRUE;
37657a5b
KW
1971 if (( msgs
1972 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1973 {
2b5e7bc2
KW
1974 pack_warn = packWARN(WARN_UTF8);
1975 message = Perl_form(aTHX_
1976 "%s: %s (unexpected continuation byte 0x%02x,"
1977 " with no preceding start byte)",
1978 malformed_text,
7e2f38b2 1979 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1980 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1981 }
1982 }
1983 }
2b5e7bc2
KW
1984 else if (possible_problems & UTF8_GOT_SHORT) {
1985 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1986 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1987
1988 if (! (flags & UTF8_ALLOW_SHORT)) {
1989 disallowed = TRUE;
37657a5b
KW
1990 if (( msgs
1991 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1992 {
2b5e7bc2
KW
1993 pack_warn = packWARN(WARN_UTF8);
1994 message = Perl_form(aTHX_
56576a04
KW
1995 "%s: %s (too short; %d byte%s available, need %d)",
1996 malformed_text,
1997 _byte_dump_string(s0, send - s0, 0),
1998 (int)avail_len,
1999 avail_len == 1 ? "" : "s",
2000 (int)expectlen);
37657a5b 2001 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
2002 }
2003 }
ba210ebe 2004
2b5e7bc2 2005 }
e308b348
KW
2006 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2007 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2008 *errors |= UTF8_GOT_NON_CONTINUATION;
2009
2010 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2011 disallowed = TRUE;
37657a5b
KW
2012 if (( msgs
2013 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2014 {
99a765e9
KW
2015
2016 /* If we don't know for sure that the input length is
2017 * valid, avoid as much as possible reading past the
2018 * end of the buffer */
2019 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
100de20c
KW
2020 ? (int) (s - s0)
2021 : (int) (send - s0);
e308b348
KW
2022 pack_warn = packWARN(WARN_UTF8);
2023 message = Perl_form(aTHX_ "%s",
2024 unexpected_non_continuation_text(s0,
99a765e9 2025 printlen,
e308b348
KW
2026 s - s0,
2027 (int) expectlen));
37657a5b 2028 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2029 }
2030 }
2031 }
2b5e7bc2
KW
2032 else if (possible_problems & UTF8_GOT_SURROGATE) {
2033 possible_problems &= ~UTF8_GOT_SURROGATE;
2034
f9380377
KW
2035 if (flags & UTF8_WARN_SURROGATE) {
2036 *errors |= UTF8_GOT_SURROGATE;
2037
2038 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2039 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2040 {
2b5e7bc2
KW
2041 pack_warn = packWARN(WARN_SURROGATE);
2042
2043 /* These are the only errors that can occur with a
2044 * surrogate when the 'uv' isn't valid */
2045 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2046 message = Perl_form(aTHX_
2047 "UTF-16 surrogate (any UTF-8 sequence that"
2048 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2049 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2050 }
2051 else {
c94c2f39 2052 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2053 }
37657a5b 2054 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2055 }
2b5e7bc2 2056 }
ba210ebe 2057
2b5e7bc2
KW
2058 if (flags & UTF8_DISALLOW_SURROGATE) {
2059 disallowed = TRUE;
f9380377 2060 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2061 }
2062 }
2063 else if (possible_problems & UTF8_GOT_SUPER) {
2064 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2065
f9380377
KW
2066 if (flags & UTF8_WARN_SUPER) {
2067 *errors |= UTF8_GOT_SUPER;
2068
2069 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2070 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2071 {
2b5e7bc2
KW
2072 pack_warn = packWARN(WARN_NON_UNICODE);
2073
2074 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2075 message = Perl_form(aTHX_
2076 "Any UTF-8 sequence that starts with"
2077 " \"%s\" is for a non-Unicode code point,"
2078 " may not be portable",
7e2f38b2 2079 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2080 }
2081 else {
c94c2f39 2082 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2083 }
37657a5b 2084 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2085 }
2b5e7bc2 2086 }
ba210ebe 2087
57ff5f59
KW
2088 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2089 * and before possibly bailing out, so that the more dire
2090 * warning will override the regular one. */
2091 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2092 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2093 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
dc4a6683
KW
2094 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2095 || ckWARN(WARN_PORTABLE))))
2b5e7bc2 2096 {
dc4a6683 2097 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2b5e7bc2 2098
57ff5f59
KW
2099 /* If it is an overlong that evaluates to a code point
2100 * that doesn't have to use the Perl extended UTF-8, it
2101 * still used it, and so we output a message that
2102 * doesn't refer to the code point. The same is true
2103 * if there was a SHORT malformation where the code
2104 * point is not valid. In that case, 'uv' will have
2105 * been set to the REPLACEMENT CHAR, and the message
2106 * below without the code point in it will be selected
2107 * */
2108 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2109 message = Perl_form(aTHX_
57ff5f59 2110 perl_extended_cp_format, uv);
2b5e7bc2
KW
2111 }
2112 else {
2113 message = Perl_form(aTHX_
57ff5f59
KW
2114 "Any UTF-8 sequence that starts with"
2115 " \"%s\" is a Perl extension, and"
2116 " so is not portable",
2117 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2118 }
37657a5b 2119 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2120 }
2121
d044b7a7
KW
2122 if (flags & ( UTF8_WARN_PERL_EXTENDED
2123 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2124 {
d044b7a7 2125 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2126
d044b7a7 2127 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2128 disallowed = TRUE;
2129 }
2b5e7bc2
KW
2130 }
2131 }
eb83ed87 2132
2b5e7bc2 2133 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2134 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2135 disallowed = TRUE;
2136 }
2b5e7bc2
KW
2137 }
2138 else if (possible_problems & UTF8_GOT_NONCHAR) {
2139 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2140
f9380377
KW
2141 if (flags & UTF8_WARN_NONCHAR) {
2142 *errors |= UTF8_GOT_NONCHAR;
2143
2144 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2145 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2146 {
2b5e7bc2
KW
2147 /* The code above should have guaranteed that we don't
2148 * get here with errors other than overlong */
2149 assert (! (orig_problems
2150 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2151
2152 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2153 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2154 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2155 }
2b5e7bc2 2156 }
5b311467 2157
2b5e7bc2
KW
2158 if (flags & UTF8_DISALLOW_NONCHAR) {
2159 disallowed = TRUE;
f9380377 2160 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2161 }
6c64cd9d
KW
2162 }
2163 else if (possible_problems & UTF8_GOT_LONG) {
2164 possible_problems &= ~UTF8_GOT_LONG;
2165 *errors |= UTF8_GOT_LONG;
2166
2167 if (flags & UTF8_ALLOW_LONG) {
2168
2169 /* We don't allow the actual overlong value, unless the
2170 * special extra bit is also set */
2171 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2172 & ~UTF8_ALLOW_LONG)))
2173 {
2174 uv = UNICODE_REPLACEMENT;
2175 }
2176 }
2177 else {
2178 disallowed = TRUE;
2179
37657a5b
KW
2180 if (( msgs
2181 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2182 {
6c64cd9d
KW
2183 pack_warn = packWARN(WARN_UTF8);
2184
2185 /* These error types cause 'uv' to be something that
2186 * isn't what was intended, so can't use it in the
2187 * message. The other error types either can't
2188 * generate an overlong, or else the 'uv' is valid */
2189 if (orig_problems &
2190 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2191 {
2192 message = Perl_form(aTHX_
2193 "%s: %s (any UTF-8 sequence that starts"
2194 " with \"%s\" is overlong which can and"
2195 " should be represented with a"
2196 " different, shorter sequence)",
2197 malformed_text,
2198 _byte_dump_string(s0, send - s0, 0),
2199 _byte_dump_string(s0, curlen, 0));
2200 }
2201 else {
2202 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2203 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2204 uv, 0);
d819dc50
KW
2205 /* Don't use U+ for non-Unicode code points, which
2206 * includes those in the Latin1 range */
2207 const char * preface = ( uv > PERL_UNICODE_MAX
2208#ifdef EBCDIC
2209 || uv <= 0xFF
2210#endif
2211 )
2212 ? "0x"
2213 : "U+";
6c64cd9d
KW
2214 message = Perl_form(aTHX_
2215 "%s: %s (overlong; instead use %s to represent"
2216 " %s%0*" UVXf ")",
2217 malformed_text,
2218 _byte_dump_string(s0, send - s0, 0),
2219 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2220 preface,
2221 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2222 small code points */
1be62ab9 2223 UNI_TO_NATIVE(uv));
6c64cd9d 2224 }
37657a5b 2225 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2226 }
2227 }
2b5e7bc2
KW
2228 } /* End of looking through the possible flags */
2229
2230 /* Display the message (if any) for the problem being handled in
2231 * this iteration of the loop */
2232 if (message) {
37657a5b 2233 if (msgs) {
37657a5b
KW
2234 assert(this_flag_bit);
2235
2236 if (*msgs == NULL) {
2237 *msgs = newAV();
2238 }
2239
bb07812e
KW
2240 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2241 pack_warn,
2242 this_flag_bit)));
37657a5b
KW
2243 }
2244 else if (PL_op)
2b5e7bc2
KW
2245 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2246 OP_DESC(PL_op));
2247 else
2248 Perl_warner(aTHX_ pack_warn, "%s", message);
2249 }
ddb65933 2250 } /* End of 'while (possible_problems)' */
a0dbb045 2251
2b5e7bc2
KW
2252 /* Since there was a possible problem, the returned length may need to
2253 * be changed from the one stored at the beginning of this function.
2254 * Instead of trying to figure out if that's needed, just do it. */
2255 if (retlen) {
2256 *retlen = curlen;
2257 }
a0dbb045 2258
2b5e7bc2
KW
2259 if (disallowed) {
2260 if (flags & UTF8_CHECK_ONLY && retlen) {
2261 *retlen = ((STRLEN) -1);
2262 }
2263 return 0;
2264 }
eb83ed87 2265 }
ba210ebe 2266
2b5e7bc2 2267 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2268}
2269
8e84507e 2270/*
ec5f19d0
KW
2271=for apidoc utf8_to_uvchr_buf
2272
2273Returns the native code point of the first character in the string C<s> which
2274is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2275C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2276
524080c4
KW
2277If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2278enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2279C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2280(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2281C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2282the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2283See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2284returned.
ec5f19d0
KW
2285
2286=cut
52be2536
KW
2287
2288Also implemented as a macro in utf8.h
2289
ec5f19d0
KW
2290*/
2291
2292
2293UV
2294Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2295{
7f974d7e
KW
2296 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2297
9a9a6c98 2298 return utf8_to_uvchr_buf_helper(s, send, retlen);
ec5f19d0
KW
2299}
2300
52be2536
KW
2301/* This is marked as deprecated
2302 *
ec5f19d0
KW
2303=for apidoc utf8_to_uvuni_buf
2304
de69f3af
KW
2305Only in very rare circumstances should code need to be dealing in Unicode
2306(as opposed to native) code points. In those few cases, use
09232555
KW
2307C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2308If you are not absolutely sure this is one of those cases, then assume it isn't
2309and use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2310
2311Returns the Unicode (not-native) code point of the first character in the
2312string C<s> which
ec5f19d0
KW
2313is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2314C<retlen> will be set to the length, in bytes, of that character.
2315
524080c4
KW
2316If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2317enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2318NULL) to -1. If those warnings are off, the computed value if well-defined (or
2319the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2320is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2321next possible position in C<s> that could begin a non-malformed character.
09232555
KW
2322See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2323returned.
ec5f19d0
KW
2324
2325=cut
2326*/
2327
2328UV
2329Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2330{
2331 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2332
2333 assert(send > s);
2334
5962d97e 2335 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2336}
2337
b76347f2 2338/*
87cea99e 2339=for apidoc utf8_length
b76347f2 2340
b2e7ed74
KW
2341Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2342at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2343same place, it returns 0 with no warning raised.
2344
2345If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2346and returns the number of valid characters.
b76347f2
JH
2347
2348=cut
2349*/
2350
2351STRLEN
35a4481c 2352Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2353{
2354 STRLEN len = 0;
2355
7918f24d
NC
2356 PERL_ARGS_ASSERT_UTF8_LENGTH;
2357
8850bf83
JH
2358 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2359 * the bitops (especially ~) can create illegal UTF-8.
2360 * In other words: in Perl UTF-8 is not just for Unicode. */
2361
12c43b0a 2362 if (UNLIKELY(e < s))
a3b680e6 2363 goto warn_and_return;
b76347f2 2364 while (s < e) {
4cbf4130 2365 s += UTF8SKIP(s);
8e91ec7f
AV
2366 len++;
2367 }
2368
12c43b0a 2369 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2370 len--;
2371 warn_and_return:
9b387841
NC
2372 if (PL_op)
2373 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2374 "%s in %s", unees, OP_DESC(PL_op));
2375 else
61a12c31 2376 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2377 }
2378
2379 return len;
2380}
2381
b06226ff 2382/*
fed3ba5d
NC
2383=for apidoc bytes_cmp_utf8
2384
a1433954 2385Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2386sequence of characters (stored as UTF-8)
2387in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2388equal, -1 or -2 if the first string is less than the second string, +1 or +2
2389if the first string is greater than the second string.
2390
2391-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2392longer string. -2 or +2 is returned if
2393there was a difference between characters
fed3ba5d
NC
2394within the strings.
2395
2396=cut
2397*/
2398
2399int
2400Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2401{
2402 const U8 *const bend = b + blen;
2403 const U8 *const uend = u + ulen;
2404
2405 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2406
2407 while (b < bend && u < uend) {
2408 U8 c = *u++;
2409 if (!UTF8_IS_INVARIANT(c)) {
2410 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2411 if (u < uend) {
2412 U8 c1 = *u++;
2413 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2414 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2415 } else {
2b5e7bc2 2416 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2417 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2418 "%s %s%s",
2419 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2420 PL_op ? " in " : "",
2421 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2422 return -2;
2423 }
2424 } else {
2425 if (PL_op)
2426 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2427 "%s in %s", unees, OP_DESC(PL_op));
2428 else
61a12c31 2429 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2430 return -2; /* Really want to return undef :-) */
2431 }
2432 } else {
2433 return -2;
2434 }
2435 }
2436 if (*b != c) {
2437 return *b < c ? -2 : +2;
2438 }
2439 ++b;
2440 }
2441
2442 if (b == bend && u == uend)
2443 return 0;
2444
2445 return b < bend ? +1 : -1;
2446}
2447
2448/*
87cea99e 2449=for apidoc utf8_to_bytes
6940069f 2450
3bc0c78c 2451Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2452Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2453updates C<*lenp> to contain the new length.
3bc0c78c
KW
2454Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2455
2456Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2457having saved the value of C<*lenp> before the call, and subtracting the
2458after-call value of C<*lenp> from it.
6940069f 2459
a1433954 2460If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2461
6940069f
GS
2462=cut
2463*/
2464
2465U8 *
09af0336 2466Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2467{
9fe0d3c2 2468 U8 * first_variant;
246fae53 2469
7918f24d 2470 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2471 PERL_UNUSED_CONTEXT;
7918f24d 2472
9fe0d3c2 2473 /* This is a no-op if no variants at all in the input */
09af0336 2474 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2475 return s;
2476 }
2477
2478 {
3c5aa262 2479 U8 * const save = s;
09af0336 2480 U8 * const send = s + *lenp;
3c5aa262
KW
2481 U8 * d;
2482
2483 /* Nothing before the first variant needs to be changed, so start the real
2484 * work there */
2485 s = first_variant;
2486 while (s < send) {
2487 if (! UTF8_IS_INVARIANT(*s)) {
2488 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2489 *lenp = ((STRLEN) -1);
3c5aa262
KW
2490 return 0;
2491 }
2492 s++;
d59937ca
KW
2493 }
2494 s++;
dcad2880 2495 }
dcad2880 2496
3c5aa262
KW
2497 /* Is downgradable, so do it */
2498 d = s = first_variant;
2499 while (s < send) {
2500 U8 c = *s++;
2501 if (! UVCHR_IS_INVARIANT(c)) {
2502 /* Then it is two-byte encoded */
2503 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2504 s++;
2505 }
2506 *d++ = c;
2507 }
2508 *d = '\0';
09af0336 2509 *lenp = d - save;
3c5aa262
KW
2510
2511 return save;
9fe0d3c2 2512 }
6940069f
GS
2513}
2514
2515/*
87cea99e 2516=for apidoc bytes_from_utf8
f9a63242 2517
09af0336 2518Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2519byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2520actually encoded in UTF-8.
2521
2522Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2523the input string.
2524
41ae6089
KW
2525Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2526not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2527C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2528
41ae6089 2529Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2530newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2531returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2532caller is responsible for arranging for the memory used by this string to get
2533freed.
f9a63242 2534
3bc0c78c 2535Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2536having saved the value of C<*lenp> before the call, and subtracting the
2537after-call value of C<*lenp> from it.
3bc0c78c 2538
37607a96 2539=cut
976c1b08
KW
2540
2541There is a macro that avoids this function call, but this is retained for
2542anyone who calls it with the Perl_ prefix */
f9a63242
JH
2543
2544U8 *
41ae6089 2545Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2546{
7918f24d 2547 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2548 PERL_UNUSED_CONTEXT;
f9a63242 2549
976c1b08
KW
2550 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2551}
2552
2553/*
df6bd76f
KW
2554=for comment
2555skip apidoc
2556This is not currently externally documented because we don't want people to use
2557it for now. XXX Perhaps that is too paranoid, and it should be documented?
2558
2559=for apidoc bytes_from_utf8_loc
976c1b08
KW
2560
2561Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2562to store the location of the first character in C<"s"> that cannot be
2563converted to non-UTF8.
2564
2565If that parameter is C<NULL>, this function behaves identically to
2566C<bytes_from_utf8>.
2567
2568Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2569C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2570
2571Otherwise, the function returns a newly created C<NUL>-terminated string
2572containing the non-UTF8 equivalent of the convertible first portion of
2573C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2574If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2575and C<*first_non_downgradable> is set to C<NULL>.
2576
2577Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2578first character in the original string that wasn't converted. C<*is_utf8p> is
2579unchanged. Note that the new string may have length 0.
2580
2581Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2582C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2583converts as many characters in it as possible stopping at the first one it
385b74be 2584finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2585set to point to that. The function returns the portion that could be converted
2586in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2587not including the terminating C<NUL>. If the very first character in the
2588original could not be converted, C<*lenp> will be 0, and the new string will
2589contain just a single C<NUL>. If the entire input string was converted,
2590C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2591
2592Upon successful return, the number of variants in the converted portion of the
2593string can be computed by having saved the value of C<*lenp> before the call,
2594and subtracting the after-call value of C<*lenp> from it.
2595
2596=cut
2597
2598
2599*/
2600
2601U8 *
2602Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2603{
2604 U8 *d;
2605 const U8 *original = s;
2606 U8 *converted_start;
2607 const U8 *send = s + *lenp;
f9a63242 2608
976c1b08 2609 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2610
976c1b08
KW
2611 if (! *is_utf8p) {
2612 if (first_unconverted) {
2613 *first_unconverted = NULL;
2614 }
2615
2616 return (U8 *) original;
2617 }
2618
2619 Newx(d, (*lenp) + 1, U8);
2620
2621 converted_start = d;
7299a045
KW
2622 while (s < send) {
2623 U8 c = *s++;
2624 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2625
2626 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2627 * have to stop now */
2628 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2629 if (first_unconverted) {
2630 *first_unconverted = s - 1;
2631 goto finish_and_return;
2632 }
2633 else {
2634 Safefree(converted_start);
2635 return (U8 *) original;
2636 }
2637 }
2638
7299a045
KW
2639 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2640 s++;
38af28cf 2641 }
7299a045
KW
2642 *d++ = c;
2643 }
170a1c22 2644
976c1b08
KW
2645 /* Here, converted the whole of the input */
2646 *is_utf8p = FALSE;
2647 if (first_unconverted) {
2648 *first_unconverted = NULL;
170a1c22 2649 }
976c1b08
KW
2650
2651 finish_and_return:
46a08a6f
KW
2652 *d = '\0';
2653 *lenp = d - converted_start;
976c1b08
KW
2654
2655 /* Trim unused space */
2656 Renew(converted_start, *lenp + 1, U8);
2657
2658 return converted_start;
f9a63242
JH
2659}
2660
2661/*
87cea99e 2662=for apidoc bytes_to_utf8
6940069f 2663
09af0336 2664Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2665UTF-8.
09af0336 2666Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2667reflect the new length in bytes. The caller is responsible for arranging for
2668the memory used by this string to get freed.
6940069f 2669
3bc0c78c 2670Upon successful return, the number of variants in the string can be computed by
23b37b12 2671having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2672after-call value of C<*lenp>.
2673
75200dff 2674A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2675
2676If you want to convert to UTF-8 from encodings other than
2677the native (Latin1 or EBCDIC),
a1433954 2678see L</sv_recode_to_utf8>().
c9ada85f 2679
497711e7 2680=cut
6940069f
GS
2681*/
2682
2683U8*
09af0336 2684Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2685{
09af0336 2686 const U8 * const send = s + (*lenp);
6940069f
GS
2687 U8 *d;
2688 U8 *dst;
7918f24d
NC
2689
2690 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2691 PERL_UNUSED_CONTEXT;
6940069f 2692
d4662719
KW
2693 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2694 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2695 dst = d;
2696
2697 while (s < send) {
55d09dc8
KW
2698 append_utf8_from_native_byte(*s, &d);
2699 s++;
6940069f 2700 }
2e11cf67 2701
6940069f 2702 *d = '\0';
09af0336 2703 *lenp = d-dst;
2e11cf67 2704
6940069f
GS
2705 return dst;
2706}
2707
a0ed51b3 2708/*
624504c5
KW
2709 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2710 * use utf16_to_utf8_reversed().
a0ed51b3 2711 *
624504c5
KW
2712 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2713 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2714 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2715 *
2716 * These functions don't check for overflow. The worst case is every code
2717 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2718 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2719 * destination must be pre-extended to 2 times the source length.
2720 *
2721 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2722
2723U8*
f46dcac2 2724Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3 2725{
dea0fc0b
JH
2726 U8* pend;
2727 U8* dstart = d;
2728
7918f24d
NC
2729 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2730
dea0fc0b 2731 if (bytelen & 1)
56576a04
KW
2732 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2733 (UV)bytelen);
dea0fc0b
JH
2734
2735 pend = p + bytelen;
2736
a0ed51b3 2737 while (p < pend) {
dea0fc0b
JH
2738 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2739 p += 2;
2d1545e5 2740 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2741 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2742 continue;
2743 }
56d37426
KW
2744 if (uv <= MAX_UTF8_TWO_BYTE) {
2745 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2746 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2747 continue;
2748 }
ffd0a9d3 2749
46956fad
KW
2750#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2751#define LAST_HIGH_SURROGATE 0xDBFF
2752#define FIRST_LOW_SURROGATE 0xDC00
2753#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2754#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2755
2756 /* This assumes that most uses will be in the first Unicode plane, not
2757 * needing surrogates */
b497502c
KW
2758 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2759 UNICODE_SURROGATE_LAST)))
e23c50db
KW
2760 {
2761 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2762 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2763 }
2764 else {
01ea242b 2765 UV low = (p[0] << 8) + p[1];
b497502c
KW
2766 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2767 LAST_LOW_SURROGATE)))
e23c50db 2768 {
01ea242b 2769 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2770 }
2771 p += 2;
46956fad 2772 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2773 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2774 }
a0ed51b3 2775 }
56d37426
KW
2776#ifdef EBCDIC
2777 d = uvoffuni_to_utf8_flags(d, uv, 0);
2778#else
ffd0a9d3 2779 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2780 *d++ = (U8)(( uv >> 12) | 0xe0);
2781 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2782 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2783 continue;
2784 }
2785 else {
eb160463
GS
2786 *d++ = (U8)(( uv >> 18) | 0xf0);
2787 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2788 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2789 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2790 continue;
2791 }
56d37426 2792#endif
a0ed51b3 2793 }
dea0fc0b 2794 *newlen = d - dstart;
a0ed51b3
LW
2795 return d;
2796}
2797
2798/* Note: this one is slightly destructive of the source. */
2799
2800U8*
f46dcac2 2801Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3
LW
2802{
2803 U8* s = (U8*)p;
d4c19fe8 2804 U8* const send = s + bytelen;
7918f24d
NC
2805
2806 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2807
e0ea5e2d 2808 if (bytelen & 1)
147e3846 2809 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2810 (UV)bytelen);
2811
a0ed51b3 2812 while (s < send) {
d4c19fe8 2813 const U8 tmp = s[0];
a0ed51b3
LW
2814 s[0] = s[1];
2815 s[1] = tmp;
2816 s += 2;
2817 }
dea0fc0b 2818 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2819}
2820
922e8cb4
KW
2821bool
2822Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2823{
1565c085 2824 dVAR;
dc31b55c 2825 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2826}
2827
5092f92a 2828bool
eba68aa0
KW
2829Perl__is_uni_perl_idcont(pTHX_ UV c)
2830{
1565c085 2831 dVAR;
c12658c9 2832 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2833}
2834
2835bool
f91dcd13
KW
2836Perl__is_uni_perl_idstart(pTHX_ UV c)
2837{
1565c085 2838 dVAR;
c12658c9 2839 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2840}
2841
3a4c58c9 2842UV
56576a04
KW
2843Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2844 const char S_or_s)
3a4c58c9
KW
2845{
2846 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2847 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2848 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2849 * either "SS" or "Ss". Which one to use is passed into the routine in
2850 * 'S_or_s' to avoid a test */
2851
2852 UV converted = toUPPER_LATIN1_MOD(c);
2853
2854 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2855
2856 assert(S_or_s == 'S' || S_or_s == 's');
2857
6f2d5cbc 2858 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2859 characters in this range */
3a4c58c9
KW
2860 *p = (U8) converted;
2861 *lenp = 1;
2862 return converted;
2863 }
2864
2865 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2866 * which it maps to one of them, so as to only have to have one check for
2867 * it in the main case */
2868 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2869 switch (c) {
2870 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2871 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2872 break;
2873 case MICRO_SIGN:
2874 converted = GREEK_CAPITAL_LETTER_MU;
2875 break;
79e064b9
KW
2876#if UNICODE_MAJOR_VERSION > 2 \
2877 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2878 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2879 case LATIN_SMALL_LETTER_SHARP_S:
2880 *(p)++ = 'S';
2881 *p = S_or_s;
2882 *lenp = 2;
2883 return 'S';
79e064b9 2884#endif
3a4c58c9 2885 default:
56576a04
KW
2886 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2887 " '%c' to map to '%c'",
2888 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2889 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2890 }
2891 }
2892
2893 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2894 *p = UTF8_TWO_BYTE_LO(converted);
2895 *lenp = 2;
2896
2897 return converted;
2898}
2899
fe63c520
KW
2900/* If compiled on an early Unicode version, there may not be auxiliary tables
2901 * */
2902#ifndef HAS_UC_AUX_TABLES
2903# define UC_AUX_TABLE_ptrs NULL
2904# define UC_AUX_TABLE_lengths NULL
2905#endif
2906#ifndef HAS_TC_AUX_TABLES
2907# define TC_AUX_TABLE_ptrs NULL
2908# define TC_AUX_TABLE_lengths NULL
2909#endif
2910#ifndef HAS_LC_AUX_TABLES
2911# define LC_AUX_TABLE_ptrs NULL
2912# define LC_AUX_TABLE_lengths NULL
2913#endif
2914#ifndef HAS_CF_AUX_TABLES
2915# define CF_AUX_TABLE_ptrs NULL
2916# define CF_AUX_TABLE_lengths NULL
2917#endif
2918#ifndef HAS_UC_AUX_TABLES
2919# define UC_AUX_TABLE_ptrs NULL
2920# define UC_AUX_TABLE_lengths NULL
2921#endif
2922
50bda2c3
KW
2923/* Call the function to convert a UTF-8 encoded character to the specified case.
2924 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2925 * 's' is a pointer to the first byte of the input character
2926 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2927 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2928 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2929 *
56576a04 2930 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2931 * 'd' */
56576a04 2932#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2933 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2934 Uppercase_Mapping_invmap, \
2935 UC_AUX_TABLE_ptrs, \
2936 UC_AUX_TABLE_lengths, \
2937 "uppercase")
56576a04 2938#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2939 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2940 Titlecase_Mapping_invmap, \
2941 TC_AUX_TABLE_ptrs, \
2942 TC_AUX_TABLE_lengths, \
2943 "titlecase")
56576a04 2944#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2945 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2946 Lowercase_Mapping_invmap, \
2947 LC_AUX_TABLE_ptrs, \
2948 LC_AUX_TABLE_lengths, \
2949 "lowercase")
2950
50bda2c3 2951
b9992569
KW
2952/* This additionally has the input parameter 'specials', which if non-zero will
2953 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2954 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2955#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2956 (specials) \
2957 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2958 Case_Folding_invmap, \
2959 CF_AUX_TABLE_ptrs, \
2960 CF_AUX_TABLE_lengths, \
2961 "foldcase") \
2962 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2963 Simple_Case_Folding_invmap, \
2964 NULL, NULL, \
2965 "foldcase")
c3fd2246 2966
84afefe6
JH
2967UV
2968Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2969{
a1433954
KW
2970 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2971 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2972 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2973 * the changed version may be longer than the original character.
2974 *
2975 * The ordinal of the first character of the changed version is returned
2976 * (but note, as explained above, that there may be more.) */
2977
1565c085 2978 dVAR;
7918f24d
NC
2979 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2980
3a4c58c9
KW
2981 if (c < 256) {
2982 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2983 }
2984
a13f1de4 2985 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2986}
2987
84afefe6
JH
2988UV
2989Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2990{
1565c085 2991 dVAR;
7918f24d
NC
2992 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2993
3a4c58c9
KW
2994 if (c < 256) {
2995 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2996 }
2997
a13f1de4 2998 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2999}
3000
afc16117 3001STATIC U8
eaf412bf 3002S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
3003{
3004 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 3005 * those, converting the result to UTF-8. Since the result is always just
a1433954 3006 * one character, we allow <p> to be NULL */
afc16117
KW
3007
3008 U8 converted = toLOWER_LATIN1(c);
3009
eaf412bf
KW
3010 PERL_UNUSED_ARG(dummy);
3011
afc16117 3012 if (p != NULL) {
6f2d5cbc 3013 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3014 *p = converted;
3015 *lenp = 1;
3016 }
3017 else {
430c9760
KW
3018 /* Result is known to always be < 256, so can use the EIGHT_BIT
3019 * macros */
3020 *p = UTF8_EIGHT_BIT_HI(converted);
3021 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3022 *lenp = 2;
3023 }
3024 }
3025 return converted;
3026}
3027
84afefe6
JH
3028UV
3029Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3030{
1565c085 3031 dVAR;
7918f24d
NC
3032 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3033
afc16117 3034 if (c < 256) {
eaf412bf 3035 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3036 }
3037
a13f1de4 3038 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3039}
3040
84afefe6 3041UV
7c0ab950 3042Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3043{
51910141 3044 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3045 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3046 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3047 *
3048 * Not to be used for locale folds
51910141 3049 */
f673fad4 3050
a1dde8de
KW
3051 UV converted;
3052
3053 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3054
1ca267a5
KW
3055 assert (! (flags & FOLD_FLAGS_LOCALE));
3056
659a7c2d 3057 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3058 converted = GREEK_SMALL_LETTER_MU;
3059 }
9b63e895
KW
3060#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3061 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3062 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3063 else if ( (flags & FOLD_FLAGS_FULL)
3064 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3065 {
1ca267a5
KW
3066 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3067 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3068 * under those circumstances. */
3069 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3070 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3071 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3072 p, *lenp, U8);
3073 return LATIN_SMALL_LETTER_LONG_S;
3074 }
3075 else {
4f489194
KW
3076 *(p)++ = 's';
3077 *p = 's';
3078 *lenp = 2;
3079 return 's';
1ca267a5 3080 }
a1dde8de 3081 }
9b63e895 3082#endif
a1dde8de
KW
3083 else { /* In this range the fold of all other characters is their lower
3084 case */
3085 converted = toLOWER_LATIN1(c);
3086 }
3087
6f2d5cbc 3088 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3089 *p = (U8) converted;
3090 *lenp = 1;
3091 }
3092 else {
3093 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3094 *p = UTF8_TWO_BYTE_LO(converted);
3095 *lenp = 2;
3096 }
3097
3098 return converted;
3099}
3100
3101UV
31f05a37 3102Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3103{
4b593389 3104
a0270393
KW
3105 /* Not currently externally documented, and subject to change
3106 * <flags> bits meanings:
3107 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3108 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3109 * locale are to be used.
a0270393
KW
3110 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3111 */
4b593389 3112
1565c085 3113 dVAR;
36bb2ab6 3114 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3115
780fcc9f 3116 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3117 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3118 * except for potentially warning */
8b7358b9 3119 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3120 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3121 flags &= ~FOLD_FLAGS_LOCALE;
3122 }
3123 else {
e7b7ac46 3124 goto needs_full_generality;
780fcc9f 3125 }
31f05a37
KW
3126 }
3127
a1dde8de 3128 if (c < 256) {
e7b7ac46 3129 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3130 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3131 }
3132
2f306ab9 3133 /* Here, above 255. If no special needs, just use the macro */
a0270393 3134 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3135 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3136 }
567b353c 3137 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3138 the special flags. */
3139 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3140
3141 needs_full_generality:
a0270393 3142 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3143 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3144 p, lenp, flags);
a0270393 3145 }
84afefe6
JH
3146}
3147
26483009 3148PERL_STATIC_INLINE bool
dd1a3ba7
KW
3149S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3150 SV* const invlist)
da8c1a98
KW
3151{
3152 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3153 * starts at <p>, and extending no further than <e - 1> is in the inversion
3154 * list <invlist>. */
da8c1a98 3155
b68ffe0c
KW
3156 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3157
dd1a3ba7 3158 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
da8c1a98 3159
b68ffe0c 3160 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3161 _force_out_malformed_utf8_message(p, e, 0, 1);
3162 NOT_REACHED; /* NOTREACHED */
3163 }
3164
eb1f4bb4 3165 assert(invlist);
b68ffe0c 3166 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3167}
3168
059703b0 3169#if 0 /* Not currently used, but may be needed in the future */
dd1a3ba7
KW
3170PERLVAR(I, seen_deprecated_macro, HV *)
3171
34aeb2e9
KW
3172STATIC void
3173S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3174 const char * const alternative,
3175 const bool use_locale,
3176 const char * const file,
3177 const unsigned line)
3178{
3179 const char * key;
3180
3181 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3182
3183 if (ckWARN_d(WARN_DEPRECATED)) {
3184
3185 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3186 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3187 if (! PL_seen_deprecated_macro) {
3188 PL_seen_deprecated_macro = newHV();
3189 }
3190 if (! hv_store(PL_seen_deprecated_macro, key,
3191 strlen(key), &PL_sv_undef, 0))
3192 {
3193 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3194 }
3195
c44e9413 3196 if (instr(file, "mathoms.c")) {
607313a1 3197 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3198 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3199 " will be removed. Avoid this message by"
3200 " converting to use %s().\n",
3201 file, line, name, alternative);
3202 }
3203 else {
34aeb2e9 3204 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3205 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3206 " require an additional parameter. Avoid this"
3207 " message by converting to use %s().\n",
3208 file, line, name, alternative);
607313a1 3209 }
34aeb2e9
KW
3210 }
3211 }
3212}
059703b0 3213#endif
922e8cb4
KW
3214
3215bool
dd1a3ba7 3216Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
da8c1a98 3217{
5fd49982 3218 dVAR;
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{
5fd49982 3227 dVAR;
dd1a3ba7 3228 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
da8c1a98 3229
dd1a3ba7 3230 return is_utf8_common(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3231}
3232
3233bool
dd1a3ba7 3234Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
c11ff943 3235{
5fd49982 3236 dVAR;
dd1a3ba7 3237 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
7dbf68d2 3238
dd1a3ba7 3239 return is_utf8_common(p, e, PL_utf8_perl_idcont);
7dbf68d2
KW
3240}
3241
6a4a25f4 3242STATIC UV
30613bdc
KW
3243S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3244 U8* ustrp, STRLEN *lenp,
40d2776f
KW
3245 SV *invlist, const I32 * const invmap,
3246 const U32 * const * const aux_tables,
30613bdc
KW
3247 const U8 * const aux_table_lengths,
3248 const char * const normal)
b9992569 3249{
0134edef 3250 STRLEN len = 0;
7918f24d 3251
30613bdc
KW
3252 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3253 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3254 * to name the new case in any generated messages, as a fallback if the
3255 * operation being used is not available. The new case is given by the
3256 * data structures in the remaining arguments.
3257 *
3258 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3259 * entire changed case string, and the return value is the first code point
3260 * in that string */
3261
b9992569 3262 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3263
36eaa811
KW
3264 /* For code points that don't change case, we already know that the output
3265 * of this function is the unchanged input, so we can skip doing look-ups
3266 * for them. Unfortunately the case-changing code points are scattered
3267 * around. But there are some long consecutive ranges where there are no
3268 * case changing code points. By adding tests, we can eliminate the lookup
3269 * for all the ones in such ranges. This is currently done here only for
3270 * just a few cases where the scripts are in common use in modern commerce
3271 * (and scripts adjacent to those which can be included without additional
3272 * tests). */
3273
3274 if (uv1 >= 0x0590) {
3275 /* This keeps from needing further processing the code points most
3276 * likely to be used in the following non-cased scripts: Hebrew,
3277 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3278 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3279 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3280 if (uv1 < 0x10A0) {
3281 goto cases_to_self;
3282 }
3283
3284 /* The following largish code point ranges also don't have case
3285 * changes, but khw didn't think they warranted extra tests to speed
3286 * them up (which would slightly slow down everything else above them):
3287 * 1100..139F Hangul Jamo, Ethiopic
3288 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3289 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3290 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3291 * Combining Diacritical Marks Extended, Balinese,
3292 * Sundanese, Batak, Lepcha, Ol Chiki
3293 * 2000..206F General Punctuation
3294 */
3295
3296 if (uv1 >= 0x2D30) {
3297
3298 /* This keeps the from needing further processing the code points
3299 * most likely to be used in the following non-cased major scripts:
3300 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3301 *
3302 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3303 * event that Unicode eventually allocates the unused block as of
3304 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3305 * that the test suite will start having failures to alert you
3306 * should that happen) */
3307 if (uv1 < 0xA640) {
3308 goto cases_to_self;
3309 }
3310
3311 if (uv1 >= 0xAC00) {
3312 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3313 if (ckWARN_d(WARN_SURROGATE)) {
3314 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3315 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3316 "Operation \"%s\" returns its argument for"
3317 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3318 }
3319 goto cases_to_self;
3320 }
36eaa811
KW
3321
3322 /* AC00..FAFF Catches Hangul syllables and private use, plus
3323 * some others */
3324 if (uv1 < 0xFB00) {
3325 goto cases_to_self;
36eaa811
KW
3326 }
3327
5af9bc97 3328 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3329 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
d22ec717 3330 Perl_croak(aTHX_ cp_above_legal_max, uv1,
40606899 3331 MAX_LEGAL_CP);
5af9bc97
KW
3332 }
3333 if (ckWARN_d(WARN_NON_UNICODE)) {
3334 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3335 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3336 "Operation \"%s\" returns its argument for"
3337 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3338 }
3339 goto cases_to_self;
3340 }
3bfc1e70
KW
3341#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3342 if (UNLIKELY(uv1
3343 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3344 {
3345
3bfc1e70
KW
3346 goto cases_to_self;
3347 }
3348#endif
36eaa811
KW
3349 }
3350 }
9ae3ac1a 3351
36eaa811 3352 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3353 * be given. */
9ae3ac1a
KW
3354 }
3355
8946fcd9
KW
3356 {
3357 unsigned int i;
22197324 3358 const U32 * cp_list;
8946fcd9 3359 U8 * d;
69352d88
KW
3360
3361 /* 'index' is guaranteed to be non-negative, as this is an inversion
3362 * map that covers all possible inputs. See [perl #133365] */
8946fcd9 3363 SSize_t index = _invlist_search(invlist, uv1);
22197324 3364 I32 base = invmap[index];
0134edef 3365
30613bdc
KW
3366 /* The data structures are set up so that if 'base' is non-negative,
3367 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3368 if (base >= 0) {
3369 IV lc;
b08cf34e 3370
8946fcd9
KW
3371 if (base == 0) {
3372 goto cases_to_self;
4a8240a3 3373 }
4a8240a3 3374
30613bdc 3375 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3376 lc = base + uv1 - invlist_array(invlist)[index];
3377 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3378 return lc;
3379 }
1feea2c7 3380
30613bdc
KW
3381 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3382 * requires an auxiliary table look up. abs(base) gives the index into
3383 * a list of such tables which points to the proper aux table. And a
3384 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3385 cp_list = aux_tables[-base];
30613bdc
KW
3386
3387 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3388 d = ustrp;
3389 for (i = 0; i < aux_table_lengths[-base]; i++) {
3390 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3391 }
8946fcd9
KW
3392 *d = '\0';
3393 *lenp = d - ustrp;
3394
3395 return cp_list[0];
cbe07460
KW
3396 }
3397
3398 /* Here, there was no mapping defined, which means that the code point maps
3399 * to itself. Return the inputs */
e24dfe9c 3400 cases_to_self:
a13f1de4
KW
3401 if (p) {
3402 len = UTF8SKIP(p);
3403 if (p != ustrp) { /* Don't copy onto itself */
3404 Copy(p, ustrp, len, U8);
3405 }
3406 *lenp = len;
3407 }
3408 else {
3409 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3410 }
2a37f04d 3411
f4cd282c 3412 return uv1;
cbe07460 3413
a0ed51b3
LW
3414}
3415
b74fe592 3416Size_t
1b292063 3417Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
40d2776f 3418 const U32 ** remaining_folds_to)
b74fe592
KW
3419{
3420 /* Returns the count of the number of code points that fold to the input
3421 * 'cp' (besides itself).
3422 *
3423 * If the return is 0, there is nothing else that folds to it, and
3424 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3425 *
3426 * If the return is 1, '*first_folds_to' is set to the single code point,
3427 * and '*remaining_folds_to' is set to NULL.
3428 *
3429 * Otherwise, '*first_folds_to' is set to a code point, and
3430 * '*remaining_fold_to' is set to an array that contains the others. The
3431 * length of this array is the returned count minus 1.
3432 *
3433 * The reason for this convolution is to avoid having to deal with
3434 * allocating and freeing memory. The lists are already constructed, so
3435 * the return can point to them, but single code points aren't, so would
1b292063
KW
3436 * need to be constructed if we didn't employ something like this API
3437 *
3438 * The code points returned by this function are all legal Unicode, which
3439 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3440 * constructed with this size (to save space and memory), and we return
3441 * pointers, so they must be this size */
b74fe592 3442
1565c085 3443 dVAR;
69352d88
KW
3444 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3445 * that covers all possible inputs. See [perl #133365] */
b74fe592 3446 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
40d2776f 3447 I32 base = _Perl_IVCF_invmap[index];
b74fe592
KW
3448
3449 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3450
3451 if (base == 0) { /* No fold */
3452 *first_folds_to = 0;
3453 *remaining_folds_to = NULL;
3454 return 0;
3455 }
3456
3457#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3458
3459 assert(base > 0);
3460
3461#else
3462
3463 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3464
3465 /* The data structure is set up so that the absolute value of 'base' is
3466 * an index into a table of pointers to arrays, with the array
3467 * corresponding to the index being the list of code points that fold
3468 * to 'cp', and the parallel array containing the length of the list
3469 * array */
3470 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
99f30495
KW
3471 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3472 /* +1 excludes first_folds_to */
b74fe592
KW
3473 return IVCF_AUX_TABLE_lengths[-base];
3474 }
3475
3476#endif
3477
3478 /* Only the single code point. This works like 'fc(G) = G - A + a' */
40d2776f
KW
3479 *first_folds_to = (U32) (base + cp
3480 - invlist_array(PL_utf8_foldclosures)[index]);
b74fe592
KW
3481 *remaining_folds_to = NULL;
3482 return 1;
3483}
3484
051a06d4 3485STATIC UV
56576a04
KW
3486S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3487 U8* const ustrp, STRLEN *lenp)
051a06d4 3488{
4a4088c4 3489 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3490 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3491 * result contains a character that crosses the 255/256 boundary, disallow
3492 * the change, and return the original code point. See L<perlfunc/lc> for
3493 * why;
051a06d4 3494 *
a1433954
KW
3495 * p points to the original string whose case was changed; assumed
3496 * by this routine to be well-formed
051a06d4 3497 * result the code point of the first character in the changed-case string
56576a04
KW
3498 * ustrp points to the changed-case string (<result> represents its
3499 * first char)
051a06d4
KW
3500 * lenp points to the length of <ustrp> */
3501
3502 UV original; /* To store the first code point of <p> */
3503
3504 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3505
a4f12ed7 3506 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3507
3508 /* We know immediately if the first character in the string crosses the
5e45c680 3509 * boundary, so can skip testing */
051a06d4
KW
3510 if (result > 255) {
3511
3512 /* Look at every character in the result; if any cross the
3513 * boundary, the whole thing is disallowed */
3514 U8* s = ustrp + UTF8SKIP(ustrp);
3515 U8* e = ustrp + *lenp;
3516 while (s < e) {
a4f12ed7 3517 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3518 goto bad_crossing;
3519 }
3520 s += UTF8SKIP(s);
3521 }
3522
613abc6d
KW
3523 /* Here, no characters crossed, result is ok as-is, but we warn. */
3524 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3525 return result;
3526 }
3527
7b52d656 3528 bad_crossing:
051a06d4
KW
3529
3530 /* Failed, have to return the original */
4b88fb76 3531 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3532
3533 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3534 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3535 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3536 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3537 OP_DESC(PL_op),
ab0b796c
KW
3538 original,
3539 original);
051a06d4
KW
3540 Copy(p, ustrp, *lenp, char);
3541 return original;
3542}
3543
b257a28c
KW
3544STATIC UV
3545S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3546 U8 * ustrp, STRLEN *lenp)
3547{
3548 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3549 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3550 * Otherwise, it returns the first code point of the Turkic foldcased
3551 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3552 * contain *lenp bytes
3553 *
3554 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3555 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3556 * DOTLESS I */
3557
3558 PERL_ARGS_ASSERT_TURKIC_FC;
3559 assert(e > p);
3560
3561 if (UNLIKELY(*p == 'I')) {
3562 *lenp = 2;
3563 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3564 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3565 return LATIN_SMALL_LETTER_DOTLESS_I;
3566 }
3567
3568 if (UNLIKELY(memBEGINs(p, e - p,
3569 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3570 {
3571 *lenp = 1;
3572 *ustrp = 'i';
3573 return 'i';
3574 }
3575
3576 return 0;
3577}
3578
3579STATIC UV
3580S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3581 U8 * ustrp, STRLEN *lenp)
3582{
3583 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3584 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3585 * Otherwise, it returns the first code point of the Turkic lowercased
3586 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3587 * contain *lenp bytes */
3588
1565c085 3589 dVAR;
b257a28c
KW
3590 PERL_ARGS_ASSERT_TURKIC_LC;
3591 assert(e > p0);
3592
3593 /* A 'I' requires context as to what to do */
3594 if (UNLIKELY(*p0 == 'I')) {
3595 const U8 * p = p0 + 1;
3596
3597 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3598 * modified by a dot above lowercases to 'i' even in turkic locales. */
3599 while (p < e) {
3600 UV cp;
3601
3602 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3603 ustrp[0] = 'i';
3604 *lenp = 1;
3605 return 'i';
3606 }
3607
3608 /* For the dot above to modify the 'I', it must be part of a
3609 * combining sequence immediately following the 'I', and no other
3610 * modifier with a ccc of 230 may intervene */
3611 cp = utf8_to_uvchr_buf(p, e, NULL);
3612 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3613 break;
3614 }
3615
3616 /* Here the combining sequence continues */
3617 p += UTF8SKIP(p);
3618 }
3619 }
3620
3621 /* In all other cases the lc is the same as the fold */
3622 return turkic_fc(p0, e, ustrp, lenp);
3623}
3624
3625STATIC UV
3626S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3627 U8 * ustrp, STRLEN *lenp)
3628{
3629 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3630 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3631 * Otherwise, it returns the first code point of the Turkic upper or
3632 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3633 * ustrp will contain *lenp bytes
3634 *
3635 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3636 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3637 * DOTLESS I */
3638
3639 PERL_ARGS_ASSERT_TURKIC_UC;
3640 assert(e > p);
3641
3642 if (*p == 'i') {
3643 *lenp = 2;
3644 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3645 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3646 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3647 }
3648
3649 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3650 *lenp = 1;
3651 *ustrp = 'I';
3652 return 'I';
3653 }
3654
3655 return 0;
3656}
3657
eaf412bf
KW
3658/* The process for changing the case is essentially the same for the four case
3659 * change types, except there are complications for folding. Otherwise the
3660 * difference is only which case to change to. To make sure that they all do
3661 * the same thing, the bodies of the functions are extracted out into the
3662 * following two macros. The functions are written with the same variable
3663 * names, and these are known and used inside these macros. It would be
3664 * better, of course, to have inline functions to do it, but since different
3665 * macros are called, depending on which case is being changed to, this is not
3666 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3667 * function can start with the common start macro, then finish with its special
3668 * handling; while the other three cases can just use the common end macro.
3669 *
3670 * The algorithm is to use the proper (passed in) macro or function to change
3671 * the case for code points that are below 256. The macro is used if using
3672 * locale rules for the case change; the function if not. If the code point is
3673 * above 255, it is computed from the input UTF-8, and another macro is called
3674 * to do the conversion. If necessary, the output is converted to UTF-8. If
3675 * using a locale, we have to check that the change did not cross the 255/256
3676 * boundary, see check_locale_boundary_crossing() for further details.
3677 *
3678 * The macros are split with the correct case change for the below-256 case
3679 * stored into 'result', and in the middle of an else clause for the above-255
3680 * case. At that point in the 'else', 'result' is not the final result, but is
3681 * the input code point calculated from the UTF-8. The fold code needs to
3682 * realize all this and take it from there.
3683 *
b257a28c
KW
3684 * To deal with Turkic locales, the function specified by the parameter
3685 * 'turkic' is called when appropriate.
3686 *
eaf412bf
KW
3687 * If you read the two macros as sequential, it's easier to understand what's
3688 * going on. */
3689#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
b257a28c 3690 L1_func_extra_param, turkic) \
a239b1e2 3691 \
eaf412bf 3692 if (flags & (locale_flags)) { \
8b7358b9 3693 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
eaf412bf 3694 if (IN_UTF8_CTYPE_LOCALE) { \
b257a28c
KW
3695 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3696 UV ret = turkic(p, e, ustrp, lenp); \
3697 if (ret) return ret; \
3698 } \
3699 \
3700 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3701 * all */ \
eaf412bf
KW
3702 flags &= ~(locale_flags); \
3703 } \
eaf412bf
KW
3704 } \
3705 \
3706 if (UTF8_IS_INVARIANT(*p)) { \
3707 if (flags & (locale_flags)) { \
3708 result = LC_L1_change_macro(*p); \
3709 } \
3710 else { \
3711 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3712 } \
3713 } \
a239b1e2 3714 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \