This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regen/reentr.pl: Use Configure'd types
[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;
343 if (flags & (UNICODE_WARN_PERL_EXTENDED
344 |UNICODE_DISALLOW_PERL_EXTENDED))
345 {
346 flag = UNICODE_GOT_PERL_EXTENDED;
347 }
348 }
a5bf80e0 349
33f38593
KW
350 if (msgs) {
351 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
352 category, flag);
353 }
354 else {
355 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
356 }
a5bf80e0 357 }
56576a04 358 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
359 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
360 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
361 {
362 return NULL;
363 }
364 }
ba6ed43c 365 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 366 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 367 }
d9432125 368
ba6ed43c
KW
369 /* Test for and handle 4-byte result. In the test immediately below, the
370 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
371 * characters). The 3 is for 3 continuation bytes; these each contribute
372 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
373 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
374 * 0x1_0000-0x1F_FFFF on ASCII */
375 if (uv < (8 * (1U << (3 * SHIFT)))) {
376 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
377 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
378 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
379 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
380
381#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
382 characters. The end-plane non-characters for EBCDIC were
383 handled just above */
384 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 385 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 386 }
ba6ed43c 387 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 388 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
389 }
390#endif
391
392 return d;
393 }
394
395 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
396 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
397 * format. The unrolled version above turns out to not save all that much
398 * time, and at these high code points (well above the legal Unicode range
399 * on ASCII platforms, and well above anything in common use in EBCDIC),
400 * khw believes that less code outweighs slight performance gains. */
401
d9432125 402 {
5aaebcb3 403 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
404 U8 *p = d+len-1;
405 while (p > d) {
957a9e81
KW
406 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
407 uv >>= SHIFT;
1d72bdf6 408 }
4c8cd605 409 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
410 return d+len;
411 }
a0ed51b3 412}
a5bf80e0 413
646ca15d 414/*
07693fe6
KW
415=for apidoc uvchr_to_utf8
416
bcb1a2d4 417Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 418of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
419C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
420the byte after the end of the new character. In other words,
07693fe6
KW
421
422 d = uvchr_to_utf8(d, uv);
423
424is the recommended wide native character-aware way of saying
425
426 *(d++) = uv;
427
d22ec717
KW
428This function accepts any code point from 0..C<IV_MAX> as input.
429C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
430
431It is possible to forbid or warn on non-Unicode code points, or those that may
432be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 433
07693fe6
KW
434=cut
435*/
436
de69f3af
KW
437/* This is also a macro */
438PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
439
07693fe6
KW
440U8 *
441Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
442{
de69f3af 443 return uvchr_to_utf8(d, uv);
07693fe6
KW
444}
445
de69f3af
KW
446/*
447=for apidoc uvchr_to_utf8_flags
448
449Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 450of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
451C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
452the byte after the end of the new character. In other words,
de69f3af
KW
453
454 d = uvchr_to_utf8_flags(d, uv, flags);
455
456or, in most cases,
457
458 d = uvchr_to_utf8_flags(d, uv, 0);
459
460This is the Unicode-aware way of saying
461
462 *(d++) = uv;
463
d22ec717
KW
464If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
465input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
466
467Specifying C<flags> can further restrict what is allowed and not warned on, as
468follows:
de69f3af 469
796b6530 470If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
471the function will raise a warning, provided UTF8 warnings are enabled. If
472instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
473NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 474
760c7c2f
KW
475Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
476affect how the function handles a Unicode non-character.
93e6dbd6 477
760c7c2f
KW
478And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
479affect the handling of code points that are above the Unicode maximum of
4800x10FFFF. Languages other than Perl may not be able to accept files that
481contain these.
93e6dbd6
KW
482
483The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
484the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
485three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
486allowed inputs to the strict UTF-8 traditionally defined by Unicode.
487Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
488C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
489above-Unicode and surrogate flags, but not the non-character ones, as
490defined in
e2176993 491L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f 492See L<perlunicode/Noncharacter code points>.
93e6dbd6 493
57ff5f59
KW
494Extremely high code points were never specified in any standard, and require an
495extension to UTF-8 to express, which Perl does. It is likely that programs
496written in something other than Perl would not be able to read files that
497contain these; nor would Perl understand files written by something that uses a
498different extension. For these reasons, there is a separate set of flags that
499can warn and/or disallow these extremely high code points, even if other
500above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
501and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
502L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
503treat all above-Unicode code points, including these, as malformations. (Note
504that the Unicode standard considers anything above 0x10FFFF to be illegal, but
505there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
506
507A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
508retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
509C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
510C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
511platforms,these flags can apply to code points that actually do fit in 31 bits.
512The new names accurately describe the situation in all cases.
de69f3af 513
de69f3af
KW
514=cut
515*/
516
517/* This is also a macro */
518PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
519
07693fe6
KW
520U8 *
521Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
522{
de69f3af 523 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
524}
525
57ff5f59
KW
526#ifndef UV_IS_QUAD
527
e050c007
KW
528STATIC int
529S_is_utf8_cp_above_31_bits(const U8 * const s,
530 const U8 * const e,
531 const bool consider_overlongs)
83dc0f42
KW
532{
533 /* Returns TRUE if the first code point represented by the Perl-extended-
534 * UTF-8-encoded string starting at 's', and looking no further than 'e -
535 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
536 *
537 * The function handles the case where the input bytes do not include all
538 * the ones necessary to represent a full character. That is, they may be
539 * the intial bytes of the representation of a code point, but possibly
540 * the final ones necessary for the complete representation may be beyond
541 * 'e - 1'.
542 *
e050c007
KW
543 * The function also can handle the case where the input is an overlong
544 * sequence. If 'consider_overlongs' is 0, the function assumes the
545 * input is not overlong, without checking, and will return based on that
546 * assumption. If this parameter is 1, the function will go to the trouble
547 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 548 *
e050c007 549 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
550 */
551
e050c007
KW
552 const STRLEN len = e - s;
553 int is_overlong;
554
555 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
556
557 assert(! UTF8_IS_INVARIANT(*s) && e > s);
558
83dc0f42
KW
559#ifdef EBCDIC
560
e050c007 561 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 562
e050c007
KW
563 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
564 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
565 * also be the start byte for a 31-bit code point; we need at least 2
566 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
567 * the start byte for an overlong sequence, but for 30-bit or smaller code
568 * points, so we don't have to worry about overlongs on EBCDIC.) */
569 if (*s != 0xFE) {
570 return 0;
571 }
83dc0f42 572
e050c007
KW
573 if (len == 1) {
574 return -1;
575 }
83dc0f42 576
e050c007 577#else
83dc0f42 578
e050c007
KW
579 /* On ASCII, FE and FF are the only start bytes that can evaluate to
580 * needing more than 31 bits. */
581 if (LIKELY(*s < 0xFE)) {
582 return 0;
583 }
83dc0f42 584
e050c007
KW
585 /* What we have left are FE and FF. Both of these require more than 31
586 * bits unless they are for overlongs. */
587 if (! consider_overlongs) {
588 return 1;
589 }
83dc0f42 590
e050c007
KW
591 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
592 * above 31 bits. But we need more than one byte to discern this, so if
593 * passed just the start byte, it could be an overlong evaluating to
594 * smaller */
595 if (len == 1) {
596 return -1;
597 }
83dc0f42 598
e050c007
KW
599 /* Having excluded len==1, and knowing that FE and FF are both valid start
600 * bytes, we can call the function below to see if the sequence is
601 * overlong. (We don't need the full generality of the called function,
602 * but for these huge code points, speed shouldn't be a consideration, and
603 * the compiler does have enough information, since it's static to this
604 * file, to optimize to just the needed parts.) */
605 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 606
e050c007
KW
607 /* If it isn't overlong, more than 31 bits are required. */
608 if (is_overlong == 0) {
609 return 1;
610 }
83dc0f42 611
e050c007
KW
612 /* If it is indeterminate if it is overlong, return that */
613 if (is_overlong < 0) {
614 return -1;
615 }
616
617 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
618 * the max it can be is 2**31 - 1 */
619 if (*s == 0xFE) {
620 return 0;
83dc0f42
KW
621 }
622
e050c007
KW
623#endif
624
625 /* Here, ASCII and EBCDIC rejoin:
626 * On ASCII: We have an overlong sequence starting with FF
627 * On EBCDIC: We have a sequence starting with FE. */
628
629 { /* For C89, use a block so the declaration can be close to its use */
630
631#ifdef EBCDIC
632
5f995336
KW
633 /* U+7FFFFFFF (2 ** 31 - 1)
634 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
635 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
636 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
637 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
638 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
639 * U+80000000 (2 ** 31):
640 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
641 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
642 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
643 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
644 *
645 * and since we know that *s = \xfe, any continuation sequcence
646 * following it that is gt the below is above 31 bits
647 [0] [1] [2] [3] [4] [5] [6] */
648 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
649
650#else
651
652 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
653 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
654 * FF overlong for U+80000000 (2 ** 31):
655 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
656 * and since we know that *s = \xff, any continuation sequcence
657 * following it that is gt the below is above 30 bits
658 [0] [1] [2] [3] [4] [5] [6] */
659 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 660
83dc0f42
KW
661
662#endif
e050c007
KW
663 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
664 const STRLEN cmp_len = MIN(conts_len, len - 1);
665
666 /* Now compare the continuation bytes in s with the ones we have
667 * compiled in that are for the largest 30 bit code point. If we have
668 * enough bytes available to determine the answer, or the bytes we do
669 * have differ from them, we can compare the two to get a definitive
670 * answer (Note that in UTF-EBCDIC, the two lowest possible
671 * continuation bytes are \x41 and \x42.) */
672 if (cmp_len >= conts_len || memNE(s + 1,
673 conts_for_highest_30_bit,
674 cmp_len))
675 {
676 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
677 }
83dc0f42 678
e050c007
KW
679 /* Here, all the bytes we have are the same as the highest 30-bit code
680 * point, but we are missing so many bytes that we can't make the
681 * determination */
682 return -1;
683 }
83dc0f42
KW
684}
685
57ff5f59
KW
686#endif
687
d6be65ae 688PERL_STATIC_INLINE int
12a4bed3
KW
689S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
690{
d6be65ae
KW
691 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
692 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
693 * it isn't, and -1 if there isn't enough information to tell. This last
694 * return value can happen if the sequence is incomplete, missing some
695 * trailing bytes that would form a complete character. If there are
696 * enough bytes to make a definitive decision, this function does so.
697 * Usually 2 bytes sufficient.
698 *
699 * Overlongs can occur whenever the number of continuation bytes changes.
700 * That means whenever the number of leading 1 bits in a start byte
701 * increases from the next lower start byte. That happens for start bytes
702 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
703 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
704 * ASCII platforms: C0, C1
705 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 706 */
12a4bed3
KW
707
708 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
709 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
710
711 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
712 assert(len > 1 && UTF8_IS_START(*s));
713
714 /* Each platform has overlongs after the start bytes given above (expressed
715 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
716 * the logic is the same, except the E0 overlong has already been excluded
717 * on EBCDIC platforms. The values below were found by manually
718 * inspecting the UTF-8 patterns. See the tables in utf8.h and
719 * utfebcdic.h. */
720
721# ifdef EBCDIC
722# define F0_ABOVE_OVERLONG 0xB0
723# define F8_ABOVE_OVERLONG 0xA8
724# define FC_ABOVE_OVERLONG 0xA4
725# define FE_ABOVE_OVERLONG 0xA2
726# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
727 /* I8(0xfe) is FF */
728# else
729
730 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 731 return 1;
12a4bed3
KW
732 }
733
734# define F0_ABOVE_OVERLONG 0x90
735# define F8_ABOVE_OVERLONG 0x88
736# define FC_ABOVE_OVERLONG 0x84
737# define FE_ABOVE_OVERLONG 0x82
738# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
739# endif
740
741
742 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
743 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
744 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
745 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
746 {
d6be65ae 747 return 1;
12a4bed3
KW
748 }
749
b0b342d4 750 /* Check for the FF overlong */
d6be65ae 751 return isFF_OVERLONG(s, len);
b0b342d4
KW
752}
753
8d6204cc 754PERL_STATIC_INLINE int
b0b342d4
KW
755S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
756{
8d6204cc
KW
757 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
758 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
759 * it isn't, and -1 if there isn't enough information to tell. This last
760 * return value can happen if the sequence is incomplete, missing some
761 * trailing bytes that would form a complete character. If there are
762 * enough bytes to make a definitive decision, this function does so. */
763
b0b342d4 764 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 765
8d6204cc
KW
766 /* To be an FF overlong, all the available bytes must match */
767 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
768 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
769 {
770 return 0;
771 }
772
773 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
774 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 775 * utfebcdic.h. */
8d6204cc
KW
776 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
777 return 1;
778 }
12a4bed3 779
8d6204cc
KW
780 /* The missing bytes could cause the result to go one way or the other, so
781 * the result is indeterminate */
782 return -1;
12a4bed3
KW
783}
784
d22ec717 785#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
786# ifdef EBCDIC /* Actually is I8 */
787# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 788 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
789# else
790# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 791 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
792# endif
793#endif
794
c285bbc4 795PERL_STATIC_INLINE int
e050c007
KW
796S_does_utf8_overflow(const U8 * const s,
797 const U8 * e,
798 const bool consider_overlongs)
a77c906e 799{
c285bbc4 800 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
801 * 'e' - 1 would overflow an IV on this platform; that is if it represents
802 * a code point larger than the highest representable code point. It
803 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
804 * enough information to tell. This last return value can happen if the
805 * sequence is incomplete, missing some trailing bytes that would form a
806 * complete character. If there are enough bytes to make a definitive
807 * decision, this function does so.
c285bbc4 808 *
e050c007
KW
809 * If 'consider_overlongs' is TRUE, the function checks for the possibility
810 * that the sequence is an overlong that doesn't overflow. Otherwise, it
811 * assumes the sequence is not an overlong. This can give different
812 * results only on ASCII 32-bit platforms.
813 *
c285bbc4
KW
814 * (For ASCII platforms, we could use memcmp() because we don't have to
815 * convert each byte to I8, but it's very rare input indeed that would
816 * approach overflow, so the loop below will likely only get executed once.)
817 *
818 * 'e' - 1 must not be beyond a full character. */
a77c906e 819
a77c906e
KW
820
821 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
822 assert(s <= e && s + UTF8SKIP(s) >= e);
823
d22ec717
KW
824#if ! defined(UV_IS_QUAD)
825
826 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
827
828#else
829
830 PERL_UNUSED_ARG(consider_overlongs);
831
832 {
833 const STRLEN len = e - s;
834 const U8 *x;
835 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
836
837 for (x = s; x < e; x++, y++) {
838
839 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
840 continue;
841 }
842
843 /* If this byte is larger than the corresponding highest UTF-8
844 * byte, the sequence overflow; otherwise the byte is less than,
845 * and so the sequence doesn't overflow */
846 return NATIVE_UTF8_TO_I8(*x) > *y;
847
848 }
849
850 /* Got to the end and all bytes are the same. If the input is a whole
851 * character, it doesn't overflow. And if it is a partial character,
852 * there's not enough information to tell */
853 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
854 return -1;
855 }
856
857 return 0;
858 }
859
860#endif
861
862}
863
864#if 0
865
866/* This is the portions of the above function that deal with UV_MAX instead of
867 * IV_MAX. They are left here in case we want to combine them so that internal
868 * uses can have larger code points. The only logic difference is that the
869 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
870 * different logic.
871 */
872
873/* Anything larger than this will overflow the word if it were converted into a UV */
874#if defined(UV_IS_QUAD)
875# ifdef EBCDIC /* Actually is I8 */
876# define HIGHEST_REPRESENTABLE_UTF8 \
877 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
878# else
879# define HIGHEST_REPRESENTABLE_UTF8 \
880 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
881# endif
882#else /* 32-bit */
883# ifdef EBCDIC
884# define HIGHEST_REPRESENTABLE_UTF8 \
885 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
886# else
887# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
888# endif
889#endif
890
a77c906e
KW
891#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
892
893 /* On 32 bit ASCII machines, many overlongs that start with FF don't
894 * overflow */
e050c007 895 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
896
897 /* To be such an overlong, the first bytes of 's' must match
898 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
899 * don't have any additional bytes available, the sequence, when
900 * completed might or might not fit in 32 bits. But if we have that
901 * next byte, we can tell for sure. If it is <= 0x83, then it does
902 * fit. */
903 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
904 return -1;
905 }
906
907 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
908 }
909
d22ec717
KW
910/* Starting with the #else, the rest of the function is identical except
911 * 1. we need to move the 'len' declaration to be global to the function
912 * 2. the endif move to just after the UNUSED_ARG.
913 * An empty endif is given just below to satisfy the preprocessor
914 */
a77c906e
KW
915#endif
916
d22ec717 917#endif
a77c906e 918
12a4bed3
KW
919#undef F0_ABOVE_OVERLONG
920#undef F8_ABOVE_OVERLONG
921#undef FC_ABOVE_OVERLONG
922#undef FE_ABOVE_OVERLONG
923#undef FF_OVERLONG_PREFIX
924
35f8c9bd 925STRLEN
1376b35c 926Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 927{
2b479609 928 STRLEN len;
12a4bed3 929 const U8 *x;
35f8c9bd 930
2b479609
KW
931 /* A helper function that should not be called directly.
932 *
933 * This function returns non-zero if the string beginning at 's' and
934 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
935 * code point; otherwise it returns 0. The examination stops after the
936 * first code point in 's' is validated, not looking at the rest of the
937 * input. If 'e' is such that there are not enough bytes to represent a
938 * complete code point, this function will return non-zero anyway, if the
939 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
940 * excluded by 'flags'.
941 *
942 * A non-zero return gives the number of bytes required to represent the
943 * code point. Be aware that if the input is for a partial character, the
944 * return will be larger than 'e - s'.
945 *
946 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
947 * The caller should have excluded the possibility of it being invariant
948 * before calling this function.
2b479609
KW
949 *
950 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
951 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
952 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
953 * disallowed by the flags. If the input is only for a partial character,
954 * the function will return non-zero if there is any sequence of
955 * well-formed UTF-8 that, when appended to the input sequence, could
956 * result in an allowed code point; otherwise it returns 0. Non characters
957 * cannot be determined based on partial character input. But many of the
958 * other excluded types can be determined with just the first one or two
959 * bytes.
960 *
961 */
962
1376b35c 963 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
2b479609
KW
964
965 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 966 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 967 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 968
2b479609 969 /* A variant char must begin with a start byte */
35f8c9bd
KW
970 if (UNLIKELY(! UTF8_IS_START(*s))) {
971 return 0;
972 }
973
edc2c47a
KW
974 /* Examine a maximum of a single whole code point */
975 if (e - s > UTF8SKIP(s)) {
976 e = s + UTF8SKIP(s);
977 }
978
2b479609
KW
979 len = e - s;
980
981 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
982 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 983
56576a04
KW
984 /* Here, we are disallowing some set of largish code points, and the
985 * first byte indicates the sequence is for a code point that could be
986 * in the excluded set. We generally don't have to look beyond this or
987 * the second byte to see if the sequence is actually for one of the
988 * excluded classes. The code below is derived from this table:
989 *
2b479609
KW
990 * UTF-8 UTF-EBCDIC I8
991 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
992 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
993 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
994 *
56576a04
KW
995 * Keep in mind that legal continuation bytes range between \x80..\xBF
996 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
997 * continuation bytes. Hence, we don't have to test the upper edge
998 * because if any of those is encountered, the sequence is malformed,
999 * and would fail elsewhere in this function.
1000 *
1001 * The code here likewise assumes that there aren't other
1002 * malformations; again the function should fail elsewhere because of
1003 * these. For example, an overlong beginning with FC doesn't actually
1004 * have to be a super; it could actually represent a small code point,
1005 * even U+0000. But, since overlongs (and other malformations) are
1006 * illegal, the function should return FALSE in either case.
2b479609
KW
1007 */
1008
1009#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1010# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 1011# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 1012
19794540
KW
1013# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1014 /* B6 and B7 */ \
1015 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1016# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1017#else
1018# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1019# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1020# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1021# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1022#endif
1023
1024 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1025 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1026 {
2b479609
KW
1027 return 0; /* Above Unicode */
1028 }
1029
d044b7a7 1030 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1031 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1032 {
57ff5f59 1033 return 0;
2b479609
KW
1034 }
1035
1036 if (len > 1) {
1037 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1038
1039 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1040 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1041 {
1042 return 0; /* Above Unicode */
1043 }
1044
1045 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1046 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1047 {
1048 return 0; /* Surrogate */
1049 }
1050
1051 if ( (flags & UTF8_DISALLOW_NONCHAR)
1052 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1053 {
1054 return 0; /* Noncharacter code point */
1055 }
1056 }
1057 }
1058
1059 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1060 for (x = s + 1; x < e; x++) {
1061 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1062 return 0;
1063 }
1064 }
1065
af13dd8a 1066 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1067 * overlong. */
d6be65ae 1068 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1069 return 0;
af13dd8a
KW
1070 }
1071
12a4bed3
KW
1072 /* And finally, that the code point represented fits in a word on this
1073 * platform */
e050c007
KW
1074 if (0 < does_utf8_overflow(s, e,
1075 0 /* Don't consider overlongs */
1076 ))
1077 {
12a4bed3 1078 return 0;
35f8c9bd
KW
1079 }
1080
2b479609 1081 return UTF8SKIP(s);
35f8c9bd
KW
1082}
1083
7e2f38b2 1084char *
63ab03b3 1085Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1086{
1087 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1088 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1089 * Currently, there are only two formats, so it is currently a bool:
1090 * 0 \xab
1091 * 1 ab (that is a space between two hex digit bytes)
1092 */
7cf8d05d
KW
1093
1094 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1095 trailing NUL */
63ab03b3
KW
1096 const U8 * s = start;
1097 const U8 * const e = start + len;
7cf8d05d
KW
1098 char * output;
1099 char * d;
1100
1101 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1102
1103 Newx(output, output_len, char);
1104 SAVEFREEPV(output);
1105
1106 d = output;
63ab03b3 1107 for (s = start; s < e; s++) {
7cf8d05d
KW
1108 const unsigned high_nibble = (*s & 0xF0) >> 4;
1109 const unsigned low_nibble = (*s & 0x0F);
1110
7e2f38b2 1111 if (format) {
63ab03b3
KW
1112 if (s > start) {
1113 *d++ = ' ';
1114 }
7e2f38b2
KW
1115 }
1116 else {
1117 *d++ = '\\';
1118 *d++ = 'x';
1119 }
7cf8d05d
KW
1120
1121 if (high_nibble < 10) {
1122 *d++ = high_nibble + '0';
1123 }
1124 else {
1125 *d++ = high_nibble - 10 + 'a';
1126 }
1127
1128 if (low_nibble < 10) {
1129 *d++ = low_nibble + '0';
1130 }
1131 else {
1132 *d++ = low_nibble - 10 + 'a';
1133 }
1134 }
1135
1136 *d = '\0';
1137 return output;
1138}
1139
806547a7 1140PERL_STATIC_INLINE char *
7cf8d05d
KW
1141S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1142
421da25c 1143 /* Max number of bytes to print */
3cc6a05e 1144 STRLEN print_len,
7cf8d05d
KW
1145
1146 /* Which one is the non-continuation */
1147 const STRLEN non_cont_byte_pos,
1148
1149 /* How many bytes should there be? */
1150 const STRLEN expect_len)
806547a7
KW
1151{
1152 /* Return the malformation warning text for an unexpected continuation
1153 * byte. */
1154
7cf8d05d 1155 const char * const where = (non_cont_byte_pos == 1)
806547a7 1156 ? "immediately"
7cf8d05d
KW
1157 : Perl_form(aTHX_ "%d bytes",
1158 (int) non_cont_byte_pos);
421da25c
KW
1159 const U8 * x = s + non_cont_byte_pos;
1160 const U8 * e = s + print_len;
806547a7
KW
1161
1162 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1163
7cf8d05d
KW
1164 /* We don't need to pass this parameter, but since it has already been
1165 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1166 assert(expect_len == UTF8SKIP(s));
1167
421da25c
KW
1168 /* As a defensive coding measure, don't output anything past a NUL. Such
1169 * bytes shouldn't be in the middle of a malformation, and could mark the
1170 * end of the allocated string, and what comes after is undefined */
1171 for (; x < e; x++) {
1172 if (*x == '\0') {
1173 x++; /* Output this particular NUL */
1174 break;
1175 }
1176 }
1177
7cf8d05d
KW
1178 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1179 " %s after start byte 0x%02x; need %d bytes, got %d)",
1180 malformed_text,
421da25c 1181 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1182 *(s + non_cont_byte_pos),
1183 where,
1184 *s,
1185 (int) expect_len,
1186 (int) non_cont_byte_pos);
806547a7
KW
1187}
1188
35f8c9bd
KW
1189/*
1190
de69f3af 1191=for apidoc utf8n_to_uvchr
378516de
KW
1192
1193THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1194Most code should use L</utf8_to_uvchr_buf>() rather than call this
1195directly.
67e989fb 1196
9041c2e3 1197Bottom level UTF-8 decode routine.
de69f3af 1198Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1199which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1200C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1201the length, in bytes, of that character.
949cf498
KW
1202
1203The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1204well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1205causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1206is the next possible position in C<s> that could begin a non-malformed
1207character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1208is raised. Some UTF-8 input sequences may contain multiple malformations.
1209This function tries to find every possible one in each call, so multiple
56576a04 1210warnings can be raised for the same sequence.
949cf498
KW
1211
1212Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1213individual types of malformations, such as the sequence being overlong (that
1214is, when there is a shorter sequence that can express the same code point;
1215overlong sequences are expressly forbidden in the UTF-8 standard due to
1216potential security issues). Another malformation example is the first byte of
1217a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1218flags. Even if allowed, this function generally returns the Unicode
1219REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1220F<utf8.h> to override this behavior for the overlong malformations, but don't
1221do that except for very specialized purposes.
949cf498 1222
796b6530 1223The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1224flags) malformation is found. If this flag is set, the routine assumes that
1225the caller will raise a warning, and this function will silently just set
d088425d
KW
1226C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1227
75200dff 1228Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1229character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1230in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1231be set to 1. To disambiguate, upon a zero return, see if the first byte of
1232C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1233error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1234
1235Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1236Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1237By default these are considered regular code points, but certain situations
ecc1615f
KW
1238warrant special handling for them, which can be specified using the C<flags>
1239parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1240three classes are treated as malformations and handled as such. The flags
1241C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1242C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1243disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1244restricts the allowed inputs to the strict UTF-8 traditionally defined by
1245Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1246definition given by
e2176993 1247L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f
KW
1248The difference between traditional strictness and C9 strictness is that the
1249latter does not forbid non-character code points. (They are still discouraged,
1250however.) For more discussion see L<perlunicode/Noncharacter code points>.
1251
1252The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1253C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1254C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1255raised for their respective categories, but otherwise the code points are
1256considered valid (not malformations). To get a category to both be treated as
1257a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1258(But note that warnings are not raised if lexically disabled nor if
796b6530 1259C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1260
57ff5f59
KW
1261Extremely high code points were never specified in any standard, and require an
1262extension to UTF-8 to express, which Perl does. It is likely that programs
1263written in something other than Perl would not be able to read files that
1264contain these; nor would Perl understand files written by something that uses a
1265different extension. For these reasons, there is a separate set of flags that
1266can warn and/or disallow these extremely high code points, even if other
1267above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1268C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1269L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1270above-Unicode code points, including these, as malformations.
1271(Note that the Unicode standard considers anything above 0x10FFFF to be
1272illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1273(2**31 -1))
1274
1275A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1276retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1277C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1278C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1279can apply to code points that actually do fit in 31 bits. This happens on
1280EBCDIC platforms, and sometimes when the L<overlong
1281malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1282describe the situation in all cases.
1283
ab8e6d41 1284
949cf498
KW
1285All other code points corresponding to Unicode characters, including private
1286use and those yet to be assigned, are never considered malformed and never
1287warn.
67e989fb 1288
5af38e47
KW
1289=for apidoc Amnh||UTF8_CHECK_ONLY
1290=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1291=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1292=for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1293=for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1294=for apidoc Amnh||UTF8_DISALLOW_SUPER
1295=for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1296=for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1297=for apidoc Amnh||UTF8_WARN_SURROGATE
1298=for apidoc Amnh||UTF8_WARN_NONCHAR
1299=for apidoc Amnh||UTF8_WARN_SUPER
1300=for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1301=for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1302
37607a96 1303=cut
f9380377
KW
1304
1305Also implemented as a macro in utf8.h
1306*/
1307
1308UV
e6a4ffc3
KW
1309Perl_utf8n_to_uvchr(const U8 *s,
1310 STRLEN curlen,
1311 STRLEN *retlen,
1312 const U32 flags)
f9380377
KW
1313{
1314 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1315
1316 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1317}
1318
1319/*
1320
1321=for apidoc utf8n_to_uvchr_error
1322
1323THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1324Most code should use L</utf8_to_uvchr_buf>() rather than call this
1325directly.
f9380377
KW
1326
1327This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1328are when an error is found. If you also need to know the generated warning
1329messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1330
1331It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1332all the others, C<errors>. If this parameter is 0, this function behaves
1333identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1334to a C<U32> variable, which this function sets to indicate any errors found.
1335Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1336C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1337of these bits will be set if a malformation is found, even if the input
7a65503b 1338C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1339exceptions are noted:
1340
1341=over 4
1342
57ff5f59 1343=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1344
57ff5f59
KW
1345The input sequence is not standard UTF-8, but a Perl extension. This bit is
1346set only if the input C<flags> parameter contains either the
1347C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1348
1349Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1350and so some extension must be used to express them. Perl uses a natural
1351extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1352extension to represent even higher ones, so that any code point that fits in a
135364-bit word can be represented. Text using these extensions is not likely to
1354be portable to non-Perl code. We lump both of these extensions together and
1355refer to them as Perl extended UTF-8. There exist other extensions that people
1356have invented, incompatible with Perl's.
1357
1358On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1359extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1360than on ASCII. Prior to that, code points 2**31 and higher were simply
1361unrepresentable, and a different, incompatible method was used to represent
1362code points between 2**30 and 2**31 - 1.
1363
1364On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1365Perl extended UTF-8 is used.
1366
1367In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1368may use for backward compatibility. That name is misleading, as this flag may
1369be set when the code point actually does fit in 31 bits. This happens on
1370EBCDIC platforms, and sometimes when the L<overlong
1371malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1372describes the situation in all cases.
f9380377
KW
1373
1374=item C<UTF8_GOT_CONTINUATION>
1375
1376The input sequence was malformed in that the first byte was a a UTF-8
1377continuation byte.
1378
1379=item C<UTF8_GOT_EMPTY>
1380
1381The input C<curlen> parameter was 0.
1382
1383=item C<UTF8_GOT_LONG>
1384
1385The input sequence was malformed in that there is some other sequence that
1386evaluates to the same code point, but that sequence is shorter than this one.
1387
fecaf136
KW
1388Until Unicode 3.1, it was legal for programs to accept this malformation, but
1389it was discovered that this created security issues.
1390
f9380377
KW
1391=item C<UTF8_GOT_NONCHAR>
1392
1393The code point represented by the input UTF-8 sequence is for a Unicode
1394non-character code point.
1395This bit is set only if the input C<flags> parameter contains either the
1396C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1397
1398=item C<UTF8_GOT_NON_CONTINUATION>
1399
1400The input sequence was malformed in that a non-continuation type byte was found
00d976bb
KW
1401in a position where only a continuation type one should be. See also
1402L</C<UTF8_GOT_SHORT>>.
f9380377
KW
1403
1404=item C<UTF8_GOT_OVERFLOW>
1405
1406The input sequence was malformed in that it is for a code point that is not
d22ec717 1407representable in the number of bits available in an IV on the current platform.
f9380377
KW
1408
1409=item C<UTF8_GOT_SHORT>
1410
1411The input sequence was malformed in that C<curlen> is smaller than required for
1412a complete sequence. In other words, the input is for a partial character
1413sequence.
1414
00d976bb
KW
1415
1416C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1417sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1418that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1419sequence was looked at. If no other flags are present, it means that the
1420sequence was valid as far as it went. Depending on the application, this could
1421mean one of three things:
1422
1423=over
1424
1425=item *
1426
1427The C<curlen> length parameter passed in was too small, and the function was
1428prevented from examining all the necessary bytes.
1429
1430=item *
1431
1432The buffer being looked at is based on reading data, and the data received so
1433far stopped in the middle of a character, so that the next read will
1434read the remainder of this character. (It is up to the caller to deal with the
1435split bytes somehow.)
1436
1437=item *
1438
1439This is a real error, and the partial sequence is all we're going to get.
1440
1441=back
1442
f9380377
KW
1443=item C<UTF8_GOT_SUPER>
1444
1445The input sequence was malformed in that it is for a non-Unicode code point;
1446that is, one above the legal Unicode maximum.
1447This bit is set only if the input C<flags> parameter contains either the
1448C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1449
1450=item C<UTF8_GOT_SURROGATE>
1451
1452The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1453code point.
1454This bit is set only if the input C<flags> parameter contains either the
1455C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1456
1457=back
1458
133551d8
KW
1459To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1460flag to suppress any warnings, and then examine the C<*errors> return.
1461
f9380377 1462=cut
37657a5b
KW
1463
1464Also implemented as a macro in utf8.h
37607a96 1465*/
67e989fb 1466
a0ed51b3 1467UV
e6a4ffc3 1468Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1469 STRLEN curlen,
1470 STRLEN *retlen,
1471 const U32 flags,
1472 U32 * errors)
1473{
1474 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1475
1476 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1477}
1478
1479/*
1480
1481=for apidoc utf8n_to_uvchr_msgs
1482
1483THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1484Most code should use L</utf8_to_uvchr_buf>() rather than call this
1485directly.
37657a5b
KW
1486
1487This function is for code that needs to know what the precise malformation(s)
1488are when an error is found, and wants the corresponding warning and/or error
1489messages to be returned to the caller rather than be displayed. All messages
1490that would have been displayed if all lexcial warnings are enabled will be
1491returned.
1492
1493It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1494placed after all the others, C<msgs>. If this parameter is 0, this function
1495behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1496be a pointer to an C<AV *> variable, in which this function creates a new AV to
1497contain any appropriate messages. The elements of the array are ordered so
1498that the first message that would have been displayed is in the 0th element,
1499and so on. Each element is a hash with three key-value pairs, as follows:
1500
1501=over 4
1502
1503=item C<text>
1504
1505The text of the message as a C<SVpv>.
1506
1507=item C<warn_categories>
1508
1509The warning category (or categories) packed into a C<SVuv>.
1510
1511=item C<flag>
1512
1513A single flag bit associated with this message, in a C<SVuv>.
1514The bit corresponds to some bit in the C<*errors> return value,
1515such as C<UTF8_GOT_LONG>.
1516
1517=back
1518
1519It's important to note that specifying this parameter as non-null will cause
1520any warnings this function would otherwise generate to be suppressed, and
1521instead be placed in C<*msgs>. The caller can check the lexical warnings state
1522(or not) when choosing what to do with the returned messages.
1523
1524If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1525no AV is created.
1526
1527The caller, of course, is responsible for freeing any returned AV.
1528
1529=cut
1530*/
1531
1532UV
e6a4ffc3 1533Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1534 STRLEN curlen,
1535 STRLEN *retlen,
1536 const U32 flags,
1537 U32 * errors,
1538 AV ** msgs)
a0ed51b3 1539{
d4c19fe8 1540 const U8 * const s0 = s;
2b9519f0 1541 const U8 * send = s0 + curlen;
5af9f822
KW
1542 U32 possible_problems; /* A bit is set here for each potential problem
1543 found as we go along */
1544 UV uv;
1545 STRLEN expectlen; /* How long should this sequence be? */
1546 STRLEN avail_len; /* When input is too short, gives what that is */
1547 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1548 gets set and discarded */
a0dbb045 1549
2b5e7bc2
KW
1550 /* The below are used only if there is both an overlong malformation and a
1551 * too short one. Otherwise the first two are set to 's0' and 'send', and
1552 * the third not used at all */
5af9f822 1553 U8 * adjusted_s0;
e9f2c446
KW
1554 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1555 routine; see [perl #130921] */
5af9f822 1556 UV uv_so_far;
e6a4ffc3 1557 dTHX;
5af9f822 1558
e6a4ffc3 1559 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1560
1561 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1562 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1563 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1564 * final case. */
1565
1566#ifndef EBCDIC
1567
1568 /* Each of the affected Hanguls starts with \xED */
1569
1570 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1571 if (retlen) {
1572 *retlen = 3;
1573 }
1574 if (errors) {
1575 *errors = 0;
1576 }
1577 if (msgs) {
1578 *msgs = NULL;
1579 }
1580
1581 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1582 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1583 | (s0[2] & UTF_CONTINUATION_MASK);
1584 }
1585
1586#endif
1587
1588 /* In conjunction with the exhaustive tests that can be enabled in
1589 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1590 * what it is intended to do, and that no flaws in it are masked by
1591 * dropping down and executing the code below
1592 assert(! isUTF8_CHAR(s0, send)
1593 || UTF8_IS_SURROGATE(s0, send)
1594 || UTF8_IS_SUPER(s0, send)
1595 || UTF8_IS_NONCHAR(s0,send));
1596 */
1597
1598 s = s0;
1599 uv = *s0;
1600 possible_problems = 0;
1601 expectlen = 0;
1602 avail_len = 0;
1603 discard_errors = 0;
1604 adjusted_s0 = (U8 *) s0;
1605 uv_so_far = 0;
1606
f9380377
KW
1607 if (errors) {
1608 *errors = 0;
1609 }
1610 else {
1611 errors = &discard_errors;
1612 }
a0dbb045 1613
eb83ed87
KW
1614 /* The order of malformation tests here is important. We should consume as
1615 * few bytes as possible in order to not skip any valid character. This is
1616 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
e2176993 1617 * https://unicode.org/reports/tr36 for more discussion as to why. For
eb83ed87
KW
1618 * example, once we've done a UTF8SKIP, we can tell the expected number of
1619 * bytes, and could fail right off the bat if the input parameters indicate
1620 * that there are too few available. But it could be that just that first
1621 * byte is garbled, and the intended character occupies fewer bytes. If we
1622 * blindly assumed that the first byte is correct, and skipped based on
1623 * that number, we could skip over a valid input character. So instead, we
1624 * always examine the sequence byte-by-byte.
1625 *
1626 * We also should not consume too few bytes, otherwise someone could inject
1627 * things. For example, an input could be deliberately designed to
1628 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1629 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1630 * which is actually part of of the overflowing sequence), that could look
1631 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1632 * sequence and process the rest, inappropriately.
1633 *
1634 * Some possible input sequences are malformed in more than one way. This
1635 * function goes to lengths to try to find all of them. This is necessary
1636 * for correctness, as the inputs may allow one malformation but not
1637 * another, and if we abandon searching for others after finding the
1638 * allowed one, we could allow in something that shouldn't have been.
1639 */
eb83ed87 1640
b5b9af04 1641 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1642 possible_problems |= UTF8_GOT_EMPTY;
1643 curlen = 0;
5a48568d 1644 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1645 goto ready_to_handle_errors;
0c443dc2
JH
1646 }
1647
eb83ed87
KW
1648 expectlen = UTF8SKIP(s);
1649
1650 /* A well-formed UTF-8 character, as the vast majority of calls to this
1651 * function will be for, has this expected length. For efficiency, set
1652 * things up here to return it. It will be overriden only in those rare
1653 * cases where a malformation is found */
1654 if (retlen) {
1655 *retlen = expectlen;
1656 }
1657
eb83ed87 1658 /* A continuation character can't start a valid sequence */
b5b9af04 1659 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1660 possible_problems |= UTF8_GOT_CONTINUATION;
1661 curlen = 1;
1662 uv = UNICODE_REPLACEMENT;
1663 goto ready_to_handle_errors;
ba210ebe 1664 }
9041c2e3 1665
dcd27b3c 1666 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1667 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1668 * because it excludes start bytes like \xC0 that always lead to
1669 * overlongs.) */
dcd27b3c 1670
534752c1
KW
1671 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1672 * that indicate the number of bytes in the character's whole UTF-8
1673 * sequence, leaving just the bits that are part of the value. */
1674 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1675
e308b348
KW
1676 /* Setup the loop end point, making sure to not look past the end of the
1677 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1678 if (UNLIKELY(curlen < expectlen)) {
1679 possible_problems |= UTF8_GOT_SHORT;
1680 avail_len = curlen;
e308b348
KW
1681 }
1682 else {
2b9519f0 1683 send = (U8*) s0 + expectlen;
e308b348 1684 }
e308b348 1685
eb83ed87 1686 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1687 * accumulating each into the working value as we go. */
eb83ed87 1688 for (s = s0 + 1; s < send; s++) {
b5b9af04 1689 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1690 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1691 continue;
1692 }
1693
1694 /* Here, found a non-continuation before processing all expected bytes.
1695 * This byte indicates the beginning of a new character, so quit, even
1696 * if allowing this malformation. */
2b5e7bc2 1697 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1698 break;
eb83ed87
KW
1699 } /* End of loop through the character's bytes */
1700
1701 /* Save how many bytes were actually in the character */
1702 curlen = s - s0;
1703
2b5e7bc2
KW
1704 /* Note that there are two types of too-short malformation. One is when
1705 * there is actual wrong data before the normal termination of the
1706 * sequence. The other is that the sequence wasn't complete before the end
1707 * of the data we are allowed to look at, based on the input 'curlen'.
1708 * This means that we were passed data for a partial character, but it is
1709 * valid as far as we saw. The other is definitely invalid. This
1710 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1711 * separate.
1712 *
1713 * A convenience macro that matches either of the too-short conditions. */
1714# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1715
1716 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1717 uv_so_far = uv;
1718 uv = UNICODE_REPLACEMENT;
1719 }
2b5e7bc2 1720
08e73697
KW
1721 /* Check for overflow. The algorithm requires us to not look past the end
1722 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1723 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1724 1 /* Do consider overlongs */
1725 )))
1726 {
2b5e7bc2
KW
1727 possible_problems |= UTF8_GOT_OVERFLOW;
1728 uv = UNICODE_REPLACEMENT;
eb83ed87 1729 }
eb83ed87 1730
2b5e7bc2
KW
1731 /* Check for overlong. If no problems so far, 'uv' is the correct code
1732 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1733 * we must look at the UTF-8 byte sequence itself to see if it is for an
1734 * overlong */
1735 if ( ( LIKELY(! possible_problems)
1736 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1737 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1738 && ( UNLIKELY(! UTF8_IS_START(*s0))
1739 || ( curlen > 1
d6be65ae 1740 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1741 s - s0))))))
2f8f112e 1742 {
2b5e7bc2
KW
1743 possible_problems |= UTF8_GOT_LONG;
1744
abc28b54 1745 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1746
abc28b54
KW
1747 /* The calculation in the 'true' branch of this 'if'
1748 * below won't work if overflows, and isn't needed
1749 * anyway. Further below we handle all overflow
1750 * cases */
1751 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1752 {
2b5e7bc2
KW
1753 UV min_uv = uv_so_far;
1754 STRLEN i;
1755
1756 /* Here, the input is both overlong and is missing some trailing
1757 * bytes. There is no single code point it could be for, but there
1758 * may be enough information present to determine if what we have
1759 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1760 * The code further below has the intelligence to determine this,
1761 * but just for non-overlong UTF-8 sequences. What we do here is
1762 * calculate the smallest code point the input could represent if
1763 * there were no too short malformation. Then we compute and save
1764 * the UTF-8 for that, which is what the code below looks at
1765 * instead of the raw input. It turns out that the smallest such
1766 * code point is all we need. */
2b5e7bc2
KW
1767 for (i = curlen; i < expectlen; i++) {
1768 min_uv = UTF8_ACCUMULATE(min_uv,
1769 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1770 }
1771
e9f2c446 1772 adjusted_s0 = temp_char_buf;
57ff5f59 1773 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1774 }
eb83ed87
KW
1775 }
1776
56576a04
KW
1777 /* Here, we have found all the possible problems, except for when the input
1778 * is for a problematic code point not allowed by the input parameters. */
1779
06188866
KW
1780 /* uv is valid for overlongs */
1781 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1782
1783 /* isn't problematic if < this */
1784 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1785 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1786
1787 /* if overflow, we know without looking further
1788 * precisely which of the problematic types it is,
1789 * and we deal with those in the overflow handling
1790 * code */
1791 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1792 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1793 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1794 && ((flags & ( UTF8_DISALLOW_NONCHAR
1795 |UTF8_DISALLOW_SURROGATE
1796 |UTF8_DISALLOW_SUPER
d044b7a7 1797 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1798 |UTF8_WARN_NONCHAR
1799 |UTF8_WARN_SURROGATE
1800 |UTF8_WARN_SUPER
d22ec717 1801 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1802 {
2b5e7bc2
KW
1803 /* If there were no malformations, or the only malformation is an
1804 * overlong, 'uv' is valid */
1805 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1806 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1807 possible_problems |= UTF8_GOT_SURROGATE;
1808 }
1809 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1810 possible_problems |= UTF8_GOT_SUPER;
1811 }
1812 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1813 possible_problems |= UTF8_GOT_NONCHAR;
1814 }
1815 }
1816 else { /* Otherwise, need to look at the source UTF-8, possibly
1817 adjusted to be non-overlong */
1818
1819 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1820 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1821 {
2b5e7bc2
KW
1822 possible_problems |= UTF8_GOT_SUPER;
1823 }
1824 else if (curlen > 1) {
1825 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1826 NATIVE_UTF8_TO_I8(*adjusted_s0),
1827 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1828 {
2b5e7bc2 1829 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1830 }
2b5e7bc2
KW
1831 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1832 NATIVE_UTF8_TO_I8(*adjusted_s0),
1833 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1834 {
1835 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1836 }
1837 }
c0236afe 1838
2b5e7bc2
KW
1839 /* We need a complete well-formed UTF-8 character to discern
1840 * non-characters, so can't look for them here */
1841 }
1842 }
949cf498 1843
2b5e7bc2
KW
1844 ready_to_handle_errors:
1845
1846 /* At this point:
1847 * curlen contains the number of bytes in the sequence that
1848 * this call should advance the input by.
e308b348
KW
1849 * avail_len gives the available number of bytes passed in, but
1850 * only if this is less than the expected number of
1851 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1852 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1853 * is set in it for each potential problem found.
1854 * uv contains the code point the input sequence
1855 * represents; or if there is a problem that prevents
1856 * a well-defined value from being computed, it is
1857 * some subsitute value, typically the REPLACEMENT
1858 * CHARACTER.
1859 * s0 points to the first byte of the character
56576a04
KW
1860 * s points to just after were we left off processing
1861 * the character
1862 * send points to just after where that character should
1863 * end, based on how many bytes the start byte tells
1864 * us should be in it, but no further than s0 +
1865 * avail_len
2b5e7bc2 1866 */
eb83ed87 1867
2b5e7bc2
KW
1868 if (UNLIKELY(possible_problems)) {
1869 bool disallowed = FALSE;
1870 const U32 orig_problems = possible_problems;
1871
37657a5b
KW
1872 if (msgs) {
1873 *msgs = NULL;
1874 }
1875
2b5e7bc2 1876 while (possible_problems) { /* Handle each possible problem */
9fde5914 1877 U32 pack_warn = 0;
2b5e7bc2 1878 char * message = NULL;
37657a5b 1879 U32 this_flag_bit = 0;
2b5e7bc2
KW
1880
1881 /* Each 'if' clause handles one problem. They are ordered so that
1882 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1883 * ones; this is kinda in decreasing severity order. But the
1884 * overlong must come last, as it changes 'uv' looked at by the
1885 * others */
2b5e7bc2
KW
1886 if (possible_problems & UTF8_GOT_OVERFLOW) {
1887
56576a04
KW
1888 /* Overflow means also got a super and are using Perl's
1889 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1890 possible_problems
d044b7a7 1891 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1892 *errors |= UTF8_GOT_OVERFLOW;
1893
1894 /* But the API says we flag all errors found */
1895 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1896 *errors |= UTF8_GOT_SUPER;
1897 }
ddb65933 1898 if (flags
d044b7a7 1899 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1900 {
d044b7a7 1901 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1902 }
2b5e7bc2 1903
d60baaa7 1904 /* Disallow if any of the three categories say to */
56576a04 1905 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1906 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1907 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1908 {
1909 disallowed = TRUE;
1910 }
1911
d22ec717
KW
1912 /* Likewise, warn if any say to */
1913 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1914 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1915 {
2b5e7bc2 1916
ddb65933
KW
1917 /* The warnings code explicitly says it doesn't handle the
1918 * case of packWARN2 and two categories which have
1919 * parent-child relationship. Even if it works now to
1920 * raise the warning if either is enabled, it wouldn't
1921 * necessarily do so in the future. We output (only) the
56576a04 1922 * most dire warning */
ddb65933 1923 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1924 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1925 pack_warn = packWARN(WARN_UTF8);
1926 }
37657a5b 1927 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1928 pack_warn = packWARN(WARN_NON_UNICODE);
1929 }
1930 if (pack_warn) {
1931 message = Perl_form(aTHX_ "%s: %s (overflows)",
1932 malformed_text,
05b9033b 1933 _byte_dump_string(s0, curlen, 0));
37657a5b 1934 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1935 }
2b5e7bc2
KW
1936 }
1937 }
1938 }
1939 else if (possible_problems & UTF8_GOT_EMPTY) {
1940 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1941 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1942
1943 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1944
1945 /* This so-called malformation is now treated as a bug in
1946 * the caller. If you have nothing to decode, skip calling
1947 * this function */
1948 assert(0);
1949
2b5e7bc2 1950 disallowed = TRUE;
37657a5b
KW
1951 if ( (msgs
1952 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1953 {
2b5e7bc2
KW
1954 pack_warn = packWARN(WARN_UTF8);
1955 message = Perl_form(aTHX_ "%s (empty string)",
1956 malformed_text);
37657a5b 1957 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1958 }
1959 }
1960 }
1961 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1962 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1963 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1964
1965 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1966 disallowed = TRUE;
37657a5b
KW
1967 if (( msgs
1968 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1969 {
2b5e7bc2
KW
1970 pack_warn = packWARN(WARN_UTF8);
1971 message = Perl_form(aTHX_
1972 "%s: %s (unexpected continuation byte 0x%02x,"
1973 " with no preceding start byte)",
1974 malformed_text,
7e2f38b2 1975 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1976 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1977 }
1978 }
1979 }
2b5e7bc2
KW
1980 else if (possible_problems & UTF8_GOT_SHORT) {
1981 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1982 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1983
1984 if (! (flags & UTF8_ALLOW_SHORT)) {
1985 disallowed = TRUE;
37657a5b
KW
1986 if (( msgs
1987 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1988 {
2b5e7bc2
KW
1989 pack_warn = packWARN(WARN_UTF8);
1990 message = Perl_form(aTHX_
56576a04
KW
1991 "%s: %s (too short; %d byte%s available, need %d)",
1992 malformed_text,
1993 _byte_dump_string(s0, send - s0, 0),
1994 (int)avail_len,
1995 avail_len == 1 ? "" : "s",
1996 (int)expectlen);
37657a5b 1997 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1998 }
1999 }
ba210ebe 2000
2b5e7bc2 2001 }
e308b348
KW
2002 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2003 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2004 *errors |= UTF8_GOT_NON_CONTINUATION;
2005
2006 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2007 disallowed = TRUE;
37657a5b
KW
2008 if (( msgs
2009 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2010 {
99a765e9
KW
2011
2012 /* If we don't know for sure that the input length is
2013 * valid, avoid as much as possible reading past the
2014 * end of the buffer */
2015 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
100de20c
KW
2016 ? (int) (s - s0)
2017 : (int) (send - s0);
e308b348
KW
2018 pack_warn = packWARN(WARN_UTF8);
2019 message = Perl_form(aTHX_ "%s",
2020 unexpected_non_continuation_text(s0,
99a765e9 2021 printlen,
e308b348
KW
2022 s - s0,
2023 (int) expectlen));
37657a5b 2024 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2025 }
2026 }
2027 }
2b5e7bc2
KW
2028 else if (possible_problems & UTF8_GOT_SURROGATE) {
2029 possible_problems &= ~UTF8_GOT_SURROGATE;
2030
f9380377
KW
2031 if (flags & UTF8_WARN_SURROGATE) {
2032 *errors |= UTF8_GOT_SURROGATE;
2033
2034 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2035 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2036 {
2b5e7bc2
KW
2037 pack_warn = packWARN(WARN_SURROGATE);
2038
2039 /* These are the only errors that can occur with a
2040 * surrogate when the 'uv' isn't valid */
2041 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2042 message = Perl_form(aTHX_
2043 "UTF-16 surrogate (any UTF-8 sequence that"
2044 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2045 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2046 }
2047 else {
c94c2f39 2048 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2049 }
37657a5b 2050 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2051 }
2b5e7bc2 2052 }
ba210ebe 2053
2b5e7bc2
KW
2054 if (flags & UTF8_DISALLOW_SURROGATE) {
2055 disallowed = TRUE;
f9380377 2056 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2057 }
2058 }
2059 else if (possible_problems & UTF8_GOT_SUPER) {
2060 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2061
f9380377
KW
2062 if (flags & UTF8_WARN_SUPER) {
2063 *errors |= UTF8_GOT_SUPER;
2064
2065 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2066 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2067 {
2b5e7bc2
KW
2068 pack_warn = packWARN(WARN_NON_UNICODE);
2069
2070 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2071 message = Perl_form(aTHX_
2072 "Any UTF-8 sequence that starts with"
2073 " \"%s\" is for a non-Unicode code point,"
2074 " may not be portable",
7e2f38b2 2075 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2076 }
2077 else {
c94c2f39 2078 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2079 }
37657a5b 2080 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2081 }
2b5e7bc2 2082 }
ba210ebe 2083
57ff5f59
KW
2084 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2085 * and before possibly bailing out, so that the more dire
2086 * warning will override the regular one. */
2087 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2088 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2089 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2090 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2091 {
db0f09e6 2092 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2093
57ff5f59
KW
2094 /* If it is an overlong that evaluates to a code point
2095 * that doesn't have to use the Perl extended UTF-8, it
2096 * still used it, and so we output a message that
2097 * doesn't refer to the code point. The same is true
2098 * if there was a SHORT malformation where the code
2099 * point is not valid. In that case, 'uv' will have
2100 * been set to the REPLACEMENT CHAR, and the message
2101 * below without the code point in it will be selected
2102 * */
2103 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2104 message = Perl_form(aTHX_
57ff5f59 2105 perl_extended_cp_format, uv);
2b5e7bc2
KW
2106 }
2107 else {
2108 message = Perl_form(aTHX_
57ff5f59
KW
2109 "Any UTF-8 sequence that starts with"
2110 " \"%s\" is a Perl extension, and"
2111 " so is not portable",
2112 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2113 }
37657a5b 2114 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2115 }
2116
d044b7a7
KW
2117 if (flags & ( UTF8_WARN_PERL_EXTENDED
2118 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2119 {
d044b7a7 2120 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2121
d044b7a7 2122 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2123 disallowed = TRUE;
2124 }
2b5e7bc2
KW
2125 }
2126 }
eb83ed87 2127
2b5e7bc2 2128 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2129 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2130 disallowed = TRUE;
2131 }
2b5e7bc2
KW
2132 }
2133 else if (possible_problems & UTF8_GOT_NONCHAR) {
2134 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2135
f9380377
KW
2136 if (flags & UTF8_WARN_NONCHAR) {
2137 *errors |= UTF8_GOT_NONCHAR;
2138
2139 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2140 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2141 {
2b5e7bc2
KW
2142 /* The code above should have guaranteed that we don't
2143 * get here with errors other than overlong */
2144 assert (! (orig_problems
2145 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2146
2147 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2148 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2149 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2150 }
2b5e7bc2 2151 }
5b311467 2152
2b5e7bc2
KW
2153 if (flags & UTF8_DISALLOW_NONCHAR) {
2154 disallowed = TRUE;
f9380377 2155 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2156 }
6c64cd9d
KW
2157 }
2158 else if (possible_problems & UTF8_GOT_LONG) {
2159 possible_problems &= ~UTF8_GOT_LONG;
2160 *errors |= UTF8_GOT_LONG;
2161
2162 if (flags & UTF8_ALLOW_LONG) {
2163
2164 /* We don't allow the actual overlong value, unless the
2165 * special extra bit is also set */
2166 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2167 & ~UTF8_ALLOW_LONG)))
2168 {
2169 uv = UNICODE_REPLACEMENT;
2170 }
2171 }
2172 else {
2173 disallowed = TRUE;
2174
37657a5b
KW
2175 if (( msgs
2176 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2177 {
6c64cd9d
KW
2178 pack_warn = packWARN(WARN_UTF8);
2179
2180 /* These error types cause 'uv' to be something that
2181 * isn't what was intended, so can't use it in the
2182 * message. The other error types either can't
2183 * generate an overlong, or else the 'uv' is valid */
2184 if (orig_problems &
2185 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2186 {
2187 message = Perl_form(aTHX_
2188 "%s: %s (any UTF-8 sequence that starts"
2189 " with \"%s\" is overlong which can and"
2190 " should be represented with a"
2191 " different, shorter sequence)",
2192 malformed_text,
2193 _byte_dump_string(s0, send - s0, 0),
2194 _byte_dump_string(s0, curlen, 0));
2195 }
2196 else {
2197 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2198 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2199 uv, 0);
d819dc50
KW
2200 /* Don't use U+ for non-Unicode code points, which
2201 * includes those in the Latin1 range */
2202 const char * preface = ( uv > PERL_UNICODE_MAX
2203#ifdef EBCDIC
2204 || uv <= 0xFF
2205#endif
2206 )
2207 ? "0x"
2208 : "U+";
6c64cd9d
KW
2209 message = Perl_form(aTHX_
2210 "%s: %s (overlong; instead use %s to represent"
2211 " %s%0*" UVXf ")",
2212 malformed_text,
2213 _byte_dump_string(s0, send - s0, 0),
2214 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2215 preface,
2216 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2217 small code points */
1be62ab9 2218 UNI_TO_NATIVE(uv));
6c64cd9d 2219 }
37657a5b 2220 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2221 }
2222 }
2b5e7bc2
KW
2223 } /* End of looking through the possible flags */
2224
2225 /* Display the message (if any) for the problem being handled in
2226 * this iteration of the loop */
2227 if (message) {
37657a5b 2228 if (msgs) {
37657a5b
KW
2229 assert(this_flag_bit);
2230
2231 if (*msgs == NULL) {
2232 *msgs = newAV();
2233 }
2234
bb07812e
KW
2235 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2236 pack_warn,
2237 this_flag_bit)));
37657a5b
KW
2238 }
2239 else if (PL_op)
2b5e7bc2
KW
2240 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2241 OP_DESC(PL_op));
2242 else
2243 Perl_warner(aTHX_ pack_warn, "%s", message);
2244 }
ddb65933 2245 } /* End of 'while (possible_problems)' */
a0dbb045 2246
2b5e7bc2
KW
2247 /* Since there was a possible problem, the returned length may need to
2248 * be changed from the one stored at the beginning of this function.
2249 * Instead of trying to figure out if that's needed, just do it. */
2250 if (retlen) {
2251 *retlen = curlen;
2252 }
a0dbb045 2253
2b5e7bc2
KW
2254 if (disallowed) {
2255 if (flags & UTF8_CHECK_ONLY && retlen) {
2256 *retlen = ((STRLEN) -1);
2257 }
2258 return 0;
2259 }
eb83ed87 2260 }
ba210ebe 2261
2b5e7bc2 2262 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2263}
2264
8e84507e 2265/*
ec5f19d0
KW
2266=for apidoc utf8_to_uvchr_buf
2267
2268Returns the native code point of the first character in the string C<s> which
2269is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2270C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2271
524080c4
KW
2272If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2273enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2274C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2275(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2276C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2277the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2278See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2279returned.
ec5f19d0
KW
2280
2281=cut
52be2536
KW
2282
2283Also implemented as a macro in utf8.h
2284
ec5f19d0
KW
2285*/
2286
2287
2288UV
2289Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2290{
7f974d7e
KW
2291 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2292
9a9a6c98 2293 return utf8_to_uvchr_buf_helper(s, send, retlen);
ec5f19d0
KW
2294}
2295
52be2536
KW
2296/* This is marked as deprecated
2297 *
ec5f19d0
KW
2298=for apidoc utf8_to_uvuni_buf
2299
de69f3af
KW
2300Only in very rare circumstances should code need to be dealing in Unicode
2301(as opposed to native) code points. In those few cases, use
09232555
KW
2302C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2303If you are not absolutely sure this is one of those cases, then assume it isn't
2304and use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2305
2306Returns the Unicode (not-native) code point of the first character in the
2307string C<s> which
ec5f19d0
KW
2308is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2309C<retlen> will be set to the length, in bytes, of that character.
2310
524080c4
KW
2311If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2312enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2313NULL) to -1. If those warnings are off, the computed value if well-defined (or
2314the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2315is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2316next possible position in C<s> that could begin a non-malformed character.
09232555
KW
2317See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2318returned.
ec5f19d0
KW
2319
2320=cut
2321*/
2322
2323UV
2324Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2325{
2326 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2327
2328 assert(send > s);
2329
5962d97e 2330 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2331}
2332
b76347f2 2333/*
87cea99e 2334=for apidoc utf8_length
b76347f2 2335
b2e7ed74
KW
2336Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2337at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2338same place, it returns 0 with no warning raised.
2339
2340If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2341and returns the number of valid characters.
b76347f2
JH
2342
2343=cut
2344*/
2345
2346STRLEN
35a4481c 2347Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2348{
2349 STRLEN len = 0;
2350
7918f24d
NC
2351 PERL_ARGS_ASSERT_UTF8_LENGTH;
2352
8850bf83
JH
2353 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2354 * the bitops (especially ~) can create illegal UTF-8.
2355 * In other words: in Perl UTF-8 is not just for Unicode. */
2356
12c43b0a 2357 if (UNLIKELY(e < s))
a3b680e6 2358 goto warn_and_return;
b76347f2 2359 while (s < e) {
4cbf4130 2360 s += UTF8SKIP(s);
8e91ec7f
AV
2361 len++;
2362 }
2363
12c43b0a 2364 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2365 len--;
2366 warn_and_return:
9b387841
NC
2367 if (PL_op)
2368 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2369 "%s in %s", unees, OP_DESC(PL_op));
2370 else
61a12c31 2371 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2372 }
2373
2374 return len;
2375}
2376
b06226ff 2377/*
fed3ba5d
NC
2378=for apidoc bytes_cmp_utf8
2379
a1433954 2380Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2381sequence of characters (stored as UTF-8)
2382in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2383equal, -1 or -2 if the first string is less than the second string, +1 or +2
2384if the first string is greater than the second string.
2385
2386-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2387longer string. -2 or +2 is returned if
2388there was a difference between characters
fed3ba5d
NC
2389within the strings.
2390
2391=cut
2392*/
2393
2394int
2395Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2396{
2397 const U8 *const bend = b + blen;
2398 const U8 *const uend = u + ulen;
2399
2400 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2401
2402 while (b < bend && u < uend) {
2403 U8 c = *u++;
2404 if (!UTF8_IS_INVARIANT(c)) {
2405 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2406 if (u < uend) {
2407 U8 c1 = *u++;
2408 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2409 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2410 } else {
2b5e7bc2 2411 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2412 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2413 "%s %s%s",
2414 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2415 PL_op ? " in " : "",
2416 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2417 return -2;
2418 }
2419 } else {
2420 if (PL_op)
2421 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2422 "%s in %s", unees, OP_DESC(PL_op));
2423 else
61a12c31 2424 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2425 return -2; /* Really want to return undef :-) */
2426 }
2427 } else {
2428 return -2;
2429 }
2430 }
2431 if (*b != c) {
2432 return *b < c ? -2 : +2;
2433 }
2434 ++b;
2435 }
2436
2437 if (b == bend && u == uend)
2438 return 0;
2439
2440 return b < bend ? +1 : -1;
2441}
2442
2443/*
87cea99e 2444=for apidoc utf8_to_bytes
6940069f 2445
3bc0c78c 2446Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2447Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2448updates C<*lenp> to contain the new length.
3bc0c78c
KW
2449Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2450
2451Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2452having saved the value of C<*lenp> before the call, and subtracting the
2453after-call value of C<*lenp> from it.
6940069f 2454
a1433954 2455If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2456
6940069f
GS
2457=cut
2458*/
2459
2460U8 *
09af0336 2461Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2462{
9fe0d3c2 2463 U8 * first_variant;
246fae53 2464
7918f24d 2465 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2466 PERL_UNUSED_CONTEXT;
7918f24d 2467
9fe0d3c2 2468 /* This is a no-op if no variants at all in the input */
09af0336 2469 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2470 return s;
2471 }
2472
2473 {
3c5aa262 2474 U8 * const save = s;
09af0336 2475 U8 * const send = s + *lenp;
3c5aa262
KW
2476 U8 * d;
2477
2478 /* Nothing before the first variant needs to be changed, so start the real
2479 * work there */
2480 s = first_variant;
2481 while (s < send) {
2482 if (! UTF8_IS_INVARIANT(*s)) {
2483 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2484 *lenp = ((STRLEN) -1);
3c5aa262
KW
2485 return 0;
2486 }
2487 s++;
d59937ca
KW
2488 }
2489 s++;
dcad2880 2490 }
dcad2880 2491
3c5aa262
KW
2492 /* Is downgradable, so do it */
2493 d = s = first_variant;
2494 while (s < send) {
2495 U8 c = *s++;
2496 if (! UVCHR_IS_INVARIANT(c)) {
2497 /* Then it is two-byte encoded */
2498 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2499 s++;
2500 }
2501 *d++ = c;
2502 }
2503 *d = '\0';
09af0336 2504 *lenp = d - save;
3c5aa262
KW
2505
2506 return save;
9fe0d3c2 2507 }
6940069f
GS
2508}
2509
2510/*
87cea99e 2511=for apidoc bytes_from_utf8
f9a63242 2512
09af0336 2513Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2514byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2515actually encoded in UTF-8.
2516
2517Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2518the input string.
2519
41ae6089
KW
2520Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2521not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2522C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2523
41ae6089 2524Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2525newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2526returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2527caller is responsible for arranging for the memory used by this string to get
2528freed.
f9a63242 2529
3bc0c78c 2530Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2531having saved the value of C<*lenp> before the call, and subtracting the
2532after-call value of C<*lenp> from it.
3bc0c78c 2533
37607a96 2534=cut
976c1b08
KW
2535
2536There is a macro that avoids this function call, but this is retained for
2537anyone who calls it with the Perl_ prefix */
f9a63242
JH
2538
2539U8 *
41ae6089 2540Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2541{
7918f24d 2542 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2543 PERL_UNUSED_CONTEXT;
f9a63242 2544
976c1b08
KW
2545 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2546}
2547
2548/*
df6bd76f
KW
2549=for comment
2550skip apidoc
2551This is not currently externally documented because we don't want people to use
2552it for now. XXX Perhaps that is too paranoid, and it should be documented?
2553
2554=for apidoc bytes_from_utf8_loc
976c1b08
KW
2555
2556Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2557to store the location of the first character in C<"s"> that cannot be
2558converted to non-UTF8.
2559
2560If that parameter is C<NULL>, this function behaves identically to
2561C<bytes_from_utf8>.
2562
2563Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2564C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2565
2566Otherwise, the function returns a newly created C<NUL>-terminated string
2567containing the non-UTF8 equivalent of the convertible first portion of
2568C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2569If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2570and C<*first_non_downgradable> is set to C<NULL>.
2571
2572Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2573first character in the original string that wasn't converted. C<*is_utf8p> is
2574unchanged. Note that the new string may have length 0.
2575
2576Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2577C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2578converts as many characters in it as possible stopping at the first one it
385b74be 2579finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2580set to point to that. The function returns the portion that could be converted
2581in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2582not including the terminating C<NUL>. If the very first character in the
2583original could not be converted, C<*lenp> will be 0, and the new string will
2584contain just a single C<NUL>. If the entire input string was converted,
2585C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2586
2587Upon successful return, the number of variants in the converted portion of the
2588string can be computed by having saved the value of C<*lenp> before the call,
2589and subtracting the after-call value of C<*lenp> from it.
2590
2591=cut
2592
2593
2594*/
2595
2596U8 *
2597Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2598{
2599 U8 *d;
2600 const U8 *original = s;
2601 U8 *converted_start;
2602 const U8 *send = s + *lenp;
f9a63242 2603
976c1b08 2604 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2605
976c1b08
KW
2606 if (! *is_utf8p) {
2607 if (first_unconverted) {
2608 *first_unconverted = NULL;
2609 }
2610
2611 return (U8 *) original;
2612 }
2613
2614 Newx(d, (*lenp) + 1, U8);
2615
2616 converted_start = d;
7299a045
KW
2617 while (s < send) {
2618 U8 c = *s++;
2619 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2620
2621 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2622 * have to stop now */
2623 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2624 if (first_unconverted) {
2625 *first_unconverted = s - 1;
2626 goto finish_and_return;
2627 }
2628 else {
2629 Safefree(converted_start);
2630 return (U8 *) original;
2631 }
2632 }
2633
7299a045
KW
2634 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2635 s++;
38af28cf 2636 }
7299a045
KW
2637 *d++ = c;
2638 }
170a1c22 2639
976c1b08
KW
2640 /* Here, converted the whole of the input */
2641 *is_utf8p = FALSE;
2642 if (first_unconverted) {
2643 *first_unconverted = NULL;
170a1c22 2644 }
976c1b08
KW
2645
2646 finish_and_return:
46a08a6f
KW
2647 *d = '\0';
2648 *lenp = d - converted_start;
976c1b08
KW
2649
2650 /* Trim unused space */
2651 Renew(converted_start, *lenp + 1, U8);
2652
2653 return converted_start;
f9a63242
JH
2654}
2655
2656/*
87cea99e 2657=for apidoc bytes_to_utf8
6940069f 2658
09af0336 2659Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2660UTF-8.
09af0336 2661Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2662reflect the new length in bytes. The caller is responsible for arranging for
2663the memory used by this string to get freed.
6940069f 2664
3bc0c78c 2665Upon successful return, the number of variants in the string can be computed by
23b37b12 2666having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2667after-call value of C<*lenp>.
2668
75200dff 2669A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2670
2671If you want to convert to UTF-8 from encodings other than
2672the native (Latin1 or EBCDIC),
a1433954 2673see L</sv_recode_to_utf8>().
c9ada85f 2674
497711e7 2675=cut
6940069f
GS
2676*/
2677
2678U8*
09af0336 2679Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2680{
09af0336 2681 const U8 * const send = s + (*lenp);
6940069f
GS
2682 U8 *d;
2683 U8 *dst;
7918f24d
NC
2684
2685 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2686 PERL_UNUSED_CONTEXT;
6940069f 2687
d4662719
KW
2688 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2689 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2690 dst = d;
2691
2692 while (s < send) {
55d09dc8
KW
2693 append_utf8_from_native_byte(*s, &d);
2694 s++;
6940069f 2695 }
2e11cf67 2696
6940069f 2697 *d = '\0';
09af0336 2698 *lenp = d-dst;
2e11cf67 2699
6940069f
GS
2700 return dst;
2701}
2702
a0ed51b3 2703/*
624504c5
KW
2704 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2705 * use utf16_to_utf8_reversed().
a0ed51b3 2706 *
624504c5
KW
2707 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2708 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2709 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2710 *
2711 * These functions don't check for overflow. The worst case is every code
2712 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2713 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2714 * destination must be pre-extended to 2 times the source length.
2715 *
2716 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2717
2718U8*
f46dcac2 2719Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3 2720{
dea0fc0b
JH
2721 U8* pend;
2722 U8* dstart = d;
2723
7918f24d
NC
2724 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2725
dea0fc0b 2726 if (bytelen & 1)
56576a04
KW
2727 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2728 (UV)bytelen);
dea0fc0b
JH
2729
2730 pend = p + bytelen;
2731
a0ed51b3 2732 while (p < pend) {
dea0fc0b
JH
2733 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2734 p += 2;
2d1545e5 2735 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2736 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2737 continue;
2738 }
56d37426
KW
2739 if (uv <= MAX_UTF8_TWO_BYTE) {
2740 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2741 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2742 continue;
2743 }
ffd0a9d3 2744
46956fad
KW
2745#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2746#define LAST_HIGH_SURROGATE 0xDBFF
2747#define FIRST_LOW_SURROGATE 0xDC00
2748#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2749#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2750
2751 /* This assumes that most uses will be in the first Unicode plane, not
2752 * needing surrogates */
b497502c
KW
2753 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2754 UNICODE_SURROGATE_LAST)))
e23c50db
KW
2755 {
2756 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2757 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2758 }
2759 else {
01ea242b 2760 UV low = (p[0] << 8) + p[1];
b497502c
KW
2761 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2762 LAST_LOW_SURROGATE)))
e23c50db 2763 {
01ea242b 2764 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2765 }
2766 p += 2;
46956fad 2767 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2768 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2769 }
a0ed51b3 2770 }
56d37426
KW
2771#ifdef EBCDIC
2772 d = uvoffuni_to_utf8_flags(d, uv, 0);
2773#else
ffd0a9d3 2774 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2775 *d++ = (U8)(( uv >> 12) | 0xe0);
2776 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2777 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2778 continue;
2779 }
2780 else {
eb160463
GS
2781 *d++ = (U8)(( uv >> 18) | 0xf0);
2782 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2783 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2784 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2785 continue;
2786 }
56d37426 2787#endif
a0ed51b3 2788 }
dea0fc0b 2789 *newlen = d - dstart;
a0ed51b3
LW
2790 return d;
2791}
2792
2793/* Note: this one is slightly destructive of the source. */
2794
2795U8*
f46dcac2 2796Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3
LW
2797{
2798 U8* s = (U8*)p;
d4c19fe8 2799 U8* const send = s + bytelen;
7918f24d
NC
2800
2801 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2802
e0ea5e2d 2803 if (bytelen & 1)
147e3846 2804 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2805 (UV)bytelen);
2806
a0ed51b3 2807 while (s < send) {
d4c19fe8 2808 const U8 tmp = s[0];
a0ed51b3
LW
2809 s[0] = s[1];
2810 s[1] = tmp;
2811 s += 2;
2812 }
dea0fc0b 2813 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2814}
2815
922e8cb4
KW
2816bool
2817Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2818{
1565c085 2819 dVAR;
dc31b55c 2820 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2821}
2822
5092f92a 2823bool
eba68aa0
KW
2824Perl__is_uni_perl_idcont(pTHX_ UV c)
2825{
1565c085 2826 dVAR;
c12658c9 2827 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2828}
2829
2830bool
f91dcd13
KW
2831Perl__is_uni_perl_idstart(pTHX_ UV c)
2832{
1565c085 2833 dVAR;
c12658c9 2834 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2835}
2836
3a4c58c9 2837UV
56576a04
KW
2838Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2839 const char S_or_s)
3a4c58c9
KW
2840{
2841 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2842 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2843 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2844 * either "SS" or "Ss". Which one to use is passed into the routine in
2845 * 'S_or_s' to avoid a test */
2846
2847 UV converted = toUPPER_LATIN1_MOD(c);
2848
2849 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2850
2851 assert(S_or_s == 'S' || S_or_s == 's');
2852
6f2d5cbc 2853 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2854 characters in this range */
3a4c58c9
KW
2855 *p = (U8) converted;
2856 *lenp = 1;
2857 return converted;
2858 }
2859
2860 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2861 * which it maps to one of them, so as to only have to have one check for
2862 * it in the main case */
2863 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2864 switch (c) {
2865 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2866 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2867 break;
2868 case MICRO_SIGN:
2869 converted = GREEK_CAPITAL_LETTER_MU;
2870 break;
79e064b9
KW
2871#if UNICODE_MAJOR_VERSION > 2 \
2872 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2873 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2874 case LATIN_SMALL_LETTER_SHARP_S:
2875 *(p)++ = 'S';
2876 *p = S_or_s;
2877 *lenp = 2;
2878 return 'S';
79e064b9 2879#endif
3a4c58c9 2880 default:
56576a04
KW
2881 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2882 " '%c' to map to '%c'",
2883 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2884 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2885 }
2886 }
2887
2888 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2889 *p = UTF8_TWO_BYTE_LO(converted);
2890 *lenp = 2;
2891
2892 return converted;
2893}
2894
fe63c520
KW
2895/* If compiled on an early Unicode version, there may not be auxiliary tables
2896 * */
2897#ifndef HAS_UC_AUX_TABLES
2898# define UC_AUX_TABLE_ptrs NULL
2899# define UC_AUX_TABLE_lengths NULL
2900#endif
2901#ifndef HAS_TC_AUX_TABLES
2902# define TC_AUX_TABLE_ptrs NULL
2903# define TC_AUX_TABLE_lengths NULL
2904#endif
2905#ifndef HAS_LC_AUX_TABLES
2906# define LC_AUX_TABLE_ptrs NULL
2907# define LC_AUX_TABLE_lengths NULL
2908#endif
2909#ifndef HAS_CF_AUX_TABLES
2910# define CF_AUX_TABLE_ptrs NULL
2911# define CF_AUX_TABLE_lengths NULL
2912#endif
2913#ifndef HAS_UC_AUX_TABLES
2914# define UC_AUX_TABLE_ptrs NULL
2915# define UC_AUX_TABLE_lengths NULL
2916#endif
2917
50bda2c3
KW
2918/* Call the function to convert a UTF-8 encoded character to the specified case.
2919 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2920 * 's' is a pointer to the first byte of the input character
2921 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2922 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2923 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2924 *
56576a04 2925 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2926 * 'd' */
56576a04 2927#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2928 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2929 Uppercase_Mapping_invmap, \
2930 UC_AUX_TABLE_ptrs, \
2931 UC_AUX_TABLE_lengths, \
2932 "uppercase")
56576a04 2933#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2934 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2935 Titlecase_Mapping_invmap, \
2936 TC_AUX_TABLE_ptrs, \
2937 TC_AUX_TABLE_lengths, \
2938 "titlecase")
56576a04 2939#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2940 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2941 Lowercase_Mapping_invmap, \
2942 LC_AUX_TABLE_ptrs, \
2943 LC_AUX_TABLE_lengths, \
2944 "lowercase")
2945
50bda2c3 2946
b9992569
KW
2947/* This additionally has the input parameter 'specials', which if non-zero will
2948 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2949 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2950#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2951 (specials) \
2952 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2953 Case_Folding_invmap, \
2954 CF_AUX_TABLE_ptrs, \
2955 CF_AUX_TABLE_lengths, \
2956 "foldcase") \
2957 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2958 Simple_Case_Folding_invmap, \
2959 NULL, NULL, \
2960 "foldcase")
c3fd2246 2961
84afefe6
JH
2962UV
2963Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2964{
a1433954
KW
2965 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2966 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2967 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2968 * the changed version may be longer than the original character.
2969 *
2970 * The ordinal of the first character of the changed version is returned
2971 * (but note, as explained above, that there may be more.) */
2972
1565c085 2973 dVAR;
7918f24d
NC
2974 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2975
3a4c58c9
KW
2976 if (c < 256) {
2977 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2978 }
2979
a13f1de4 2980 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2981}
2982
84afefe6
JH
2983UV
2984Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2985{
1565c085 2986 dVAR;
7918f24d
NC
2987 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2988
3a4c58c9
KW
2989 if (c < 256) {
2990 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2991 }
2992
a13f1de4 2993 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2994}
2995
afc16117 2996STATIC U8
eaf412bf 2997S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2998{
2999 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 3000 * those, converting the result to UTF-8. Since the result is always just
a1433954 3001 * one character, we allow <p> to be NULL */
afc16117
KW
3002
3003 U8 converted = toLOWER_LATIN1(c);
3004
eaf412bf
KW
3005 PERL_UNUSED_ARG(dummy);
3006
afc16117 3007 if (p != NULL) {
6f2d5cbc 3008 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3009 *p = converted;
3010 *lenp = 1;
3011 }
3012 else {
430c9760
KW
3013 /* Result is known to always be < 256, so can use the EIGHT_BIT
3014 * macros */
3015 *p = UTF8_EIGHT_BIT_HI(converted);
3016 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3017 *lenp = 2;
3018 }
3019 }
3020 return converted;
3021}
3022
84afefe6
JH
3023UV
3024Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3025{
1565c085 3026 dVAR;
7918f24d
NC
3027 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3028
afc16117 3029 if (c < 256) {
eaf412bf 3030 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3031 }
3032
a13f1de4 3033 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3034}
3035
84afefe6 3036UV
7c0ab950 3037Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3038{
51910141 3039 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3040 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3041 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3042 *
3043 * Not to be used for locale folds
51910141 3044 */
f673fad4 3045
a1dde8de
KW
3046 UV converted;
3047
3048 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3049
1ca267a5
KW
3050 assert (! (flags & FOLD_FLAGS_LOCALE));
3051
659a7c2d 3052 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3053 converted = GREEK_SMALL_LETTER_MU;
3054 }
9b63e895
KW
3055#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3056 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3057 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3058 else if ( (flags & FOLD_FLAGS_FULL)
3059 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3060 {
1ca267a5
KW
3061 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3062 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3063 * under those circumstances. */
3064 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3065 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3066 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3067 p, *lenp, U8);
3068 return LATIN_SMALL_LETTER_LONG_S;
3069 }
3070 else {
4f489194
KW
3071 *(p)++ = 's';
3072 *p = 's';
3073 *lenp = 2;
3074 return 's';
1ca267a5 3075 }
a1dde8de 3076 }
9b63e895 3077#endif
a1dde8de
KW
3078 else { /* In this range the fold of all other characters is their lower
3079 case */
3080 converted = toLOWER_LATIN1(c);
3081 }
3082
6f2d5cbc 3083 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3084 *p = (U8) converted;
3085 *lenp = 1;
3086 }
3087 else {
3088 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3089 *p = UTF8_TWO_BYTE_LO(converted);
3090 *lenp = 2;
3091 }
3092
3093 return converted;
3094}
3095
3096UV
31f05a37 3097Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3098{
4b593389 3099
a0270393
KW
3100 /* Not currently externally documented, and subject to change
3101 * <flags> bits meanings:
3102 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3103 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3104 * locale are to be used.
a0270393
KW
3105 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3106 */
4b593389 3107
1565c085 3108 dVAR;
36bb2ab6 3109 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3110
780fcc9f 3111 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3112 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3113 * except for potentially warning */
8b7358b9 3114 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3115 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3116 flags &= ~FOLD_FLAGS_LOCALE;
3117 }
3118 else {
e7b7ac46 3119 goto needs_full_generality;
780fcc9f 3120 }
31f05a37
KW
3121 }
3122
a1dde8de 3123 if (c < 256) {
e7b7ac46 3124 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3125 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3126 }
3127
2f306ab9 3128 /* Here, above 255. If no special needs, just use the macro */
a0270393 3129 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3130 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3131 }
567b353c 3132 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3133 the special flags. */
3134 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3135
3136 needs_full_generality:
a0270393 3137 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3138 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3139 p, lenp, flags);
a0270393 3140 }
84afefe6
JH
3141}
3142
26483009 3143PERL_STATIC_INLINE bool
dd1a3ba7
KW
3144S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3145 SV* const invlist)
da8c1a98
KW
3146{
3147 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3148 * starts at <p>, and extending no further than <e - 1> is in the inversion
3149 * list <invlist>. */
da8c1a98 3150
b68ffe0c
KW
3151 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3152
dd1a3ba7 3153 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
da8c1a98 3154
b68ffe0c 3155 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3156 _force_out_malformed_utf8_message(p, e, 0, 1);
3157 NOT_REACHED; /* NOTREACHED */
3158 }
3159
eb1f4bb4 3160 assert(invlist);
b68ffe0c 3161 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3162}
3163
059703b0 3164#if 0 /* Not currently used, but may be needed in the future */
dd1a3ba7
KW
3165PERLVAR(I, seen_deprecated_macro, HV *)
3166
34aeb2e9
KW
3167STATIC void
3168S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3169 const char * const alternative,
3170 const bool use_locale,
3171 const char * const file,
3172 const unsigned line)
3173{
3174 const char * key;
3175
3176 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3177
3178 if (ckWARN_d(WARN_DEPRECATED)) {
3179
3180 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3181 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3182 if (! PL_seen_deprecated_macro) {
3183 PL_seen_deprecated_macro = newHV();
3184 }
3185 if (! hv_store(PL_seen_deprecated_macro, key,
3186 strlen(key), &PL_sv_undef, 0))
3187 {
3188 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3189 }
3190
c44e9413 3191 if (instr(file, "mathoms.c")) {
607313a1 3192 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3193 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3194 " will be removed. Avoid this message by"
3195 " converting to use %s().\n",
3196 file, line, name, alternative);
3197 }
3198 else {
34aeb2e9 3199 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3200 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3201 " require an additional parameter. Avoid this"
3202 " message by converting to use %s().\n",
3203 file, line, name, alternative);
607313a1 3204 }
34aeb2e9
KW
3205 }
3206 }
3207}
059703b0 3208#endif
922e8cb4
KW
3209
3210bool
dd1a3ba7 3211Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
da8c1a98 3212{
5fd49982 3213 dVAR;
dd1a3ba7 3214 PERL_ARGS_ASSERT__IS_UTF8_FOO;
da8c1a98 3215
dd1a3ba7 3216 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
da8c1a98
KW
3217}
3218
3219bool
dd1a3ba7 3220Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
da8c1a98 3221{
5fd49982 3222 dVAR;
dd1a3ba7 3223 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
da8c1a98 3224
dd1a3ba7 3225 return is_utf8_common(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3226}
3227
3228bool
dd1a3ba7 3229Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
c11ff943 3230{
5fd49982 3231 dVAR;
dd1a3ba7 3232 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
7dbf68d2 3233
dd1a3ba7 3234 return is_utf8_common(p, e, PL_utf8_perl_idcont);
7dbf68d2
KW
3235}
3236
6a4a25f4 3237STATIC UV
30613bdc
KW
3238S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3239 U8* ustrp, STRLEN *lenp,
40d2776f
KW
3240 SV *invlist, const I32 * const invmap,
3241 const U32 * const * const aux_tables,
30613bdc
KW
3242 const U8 * const aux_table_lengths,
3243 const char * const normal)
b9992569 3244{
0134edef 3245 STRLEN len = 0;
7918f24d 3246
30613bdc
KW
3247 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3248 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3249 * to name the new case in any generated messages, as a fallback if the
3250 * operation being used is not available. The new case is given by the
3251 * data structures in the remaining arguments.
3252 *
3253 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3254 * entire changed case string, and the return value is the first code point
3255 * in that string */
3256
b9992569 3257 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3258
36eaa811
KW
3259 /* For code points that don't change case, we already know that the output
3260 * of this function is the unchanged input, so we can skip doing look-ups
3261 * for them. Unfortunately the case-changing code points are scattered
3262 * around. But there are some long consecutive ranges where there are no
3263 * case changing code points. By adding tests, we can eliminate the lookup
3264 * for all the ones in such ranges. This is currently done here only for
3265 * just a few cases where the scripts are in common use in modern commerce
3266 * (and scripts adjacent to those which can be included without additional
3267 * tests). */
3268
3269 if (uv1 >= 0x0590) {
3270 /* This keeps from needing further processing the code points most
3271 * likely to be used in the following non-cased scripts: Hebrew,
3272 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3273 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3274 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3275 if (uv1 < 0x10A0) {
3276 goto cases_to_self;
3277 }
3278
3279 /* The following largish code point ranges also don't have case
3280 * changes, but khw didn't think they warranted extra tests to speed
3281 * them up (which would slightly slow down everything else above them):
3282 * 1100..139F Hangul Jamo, Ethiopic
3283 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3284 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3285 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3286 * Combining Diacritical Marks Extended, Balinese,
3287 * Sundanese, Batak, Lepcha, Ol Chiki
3288 * 2000..206F General Punctuation
3289 */
3290
3291 if (uv1 >= 0x2D30) {
3292
3293 /* This keeps the from needing further processing the code points
3294 * most likely to be used in the following non-cased major scripts:
3295 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3296 *
3297 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3298 * event that Unicode eventually allocates the unused block as of
3299 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3300 * that the test suite will start having failures to alert you
3301 * should that happen) */
3302 if (uv1 < 0xA640) {
3303 goto cases_to_self;
3304 }
3305
3306 if (uv1 >= 0xAC00) {
3307 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3308 if (ckWARN_d(WARN_SURROGATE)) {
3309 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3310 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3311 "Operation \"%s\" returns its argument for"
3312 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3313 }
3314 goto cases_to_self;
3315 }
36eaa811
KW
3316
3317 /* AC00..FAFF Catches Hangul syllables and private use, plus
3318 * some others */
3319 if (uv1 < 0xFB00) {
3320 goto cases_to_self;
36eaa811
KW
3321 }
3322
5af9bc97 3323 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3324 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
d22ec717 3325 Perl_croak(aTHX_ cp_above_legal_max, uv1,
40606899 3326 MAX_LEGAL_CP);
5af9bc97
KW
3327 }
3328 if (ckWARN_d(WARN_NON_UNICODE)) {
3329 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3330 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3331 "Operation \"%s\" returns its argument for"
3332 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3333 }
3334 goto cases_to_self;
3335 }
3bfc1e70
KW
3336#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3337 if (UNLIKELY(uv1
3338 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3339 {
3340
3bfc1e70
KW
3341 goto cases_to_self;
3342 }
3343#endif
36eaa811
KW
3344 }
3345 }
9ae3ac1a 3346
36eaa811 3347 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3348 * be given. */
9ae3ac1a
KW
3349 }
3350
8946fcd9
KW
3351 {
3352 unsigned int i;
e39a4130 3353 const unsigned int * cp_list;
8946fcd9 3354 U8 * d;
69352d88
KW
3355
3356 /* 'index' is guaranteed to be non-negative, as this is an inversion
3357 * map that covers all possible inputs. See [perl #133365] */
8946fcd9
KW
3358 SSize_t index = _invlist_search(invlist, uv1);
3359 IV base = invmap[index];
0134edef 3360
30613bdc
KW
3361 /* The data structures are set up so that if 'base' is non-negative,
3362 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3363 if (base >= 0) {
3364 IV lc;
b08cf34e 3365
8946fcd9
KW
3366 if (base == 0) {
3367 goto cases_to_self;
4a8240a3 3368 }
4a8240a3 3369
30613bdc 3370 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3371 lc = base + uv1 - invlist_array(invlist)[index];
3372 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3373 return lc;
3374 }
1feea2c7 3375
30613bdc
KW
3376 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3377 * requires an auxiliary table look up. abs(base) gives the index into
3378 * a list of such tables which points to the proper aux table. And a
3379 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3380 cp_list = aux_tables[-base];
30613bdc
KW
3381
3382 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3383 d = ustrp;
3384 for (i = 0; i < aux_table_lengths[-base]; i++) {
3385 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3386 }
8946fcd9
KW
3387 *d = '\0';
3388 *lenp = d - ustrp;
3389
3390 return cp_list[0];
cbe07460
KW
3391 }
3392
3393 /* Here, there was no mapping defined, which means that the code point maps
3394 * to itself. Return the inputs */
e24dfe9c 3395 cases_to_self:
a13f1de4
KW
3396 if (p) {
3397 len = UTF8SKIP(p);
3398 if (p != ustrp) { /* Don't copy onto itself */
3399 Copy(p, ustrp, len, U8);
3400 }
3401 *lenp = len;
3402 }
3403 else {
3404 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3405 }
2a37f04d 3406
f4cd282c 3407 return uv1;
cbe07460 3408
a0ed51b3
LW
3409}
3410
b74fe592 3411Size_t
1b292063 3412Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
40d2776f 3413 const U32 ** remaining_folds_to)
b74fe592
KW
3414{
3415 /* Returns the count of the number of code points that fold to the input
3416 * 'cp' (besides itself).
3417 *
3418 * If the return is 0, there is nothing else that folds to it, and
3419 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3420 *
3421 * If the return is 1, '*first_folds_to' is set to the single code point,
3422 * and '*remaining_folds_to' is set to NULL.
3423 *
3424 * Otherwise, '*first_folds_to' is set to a code point, and
3425 * '*remaining_fold_to' is set to an array that contains the others. The
3426 * length of this array is the returned count minus 1.
3427 *
3428 * The reason for this convolution is to avoid having to deal with
3429 * allocating and freeing memory. The lists are already constructed, so
3430 * the return can point to them, but single code points aren't, so would
1b292063
KW
3431 * need to be constructed if we didn't employ something like this API
3432 *
3433 * The code points returned by this function are all legal Unicode, which
3434 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3435 * constructed with this size (to save space and memory), and we return
3436 * pointers, so they must be this size */
b74fe592 3437
1565c085 3438 dVAR;
69352d88
KW
3439 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3440 * that covers all possible inputs. See [perl #133365] */
b74fe592 3441 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
40d2776f 3442 I32 base = _Perl_IVCF_invmap[index];
b74fe592
KW
3443
3444 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3445
3446 if (base == 0) { /* No fold */
3447 *first_folds_to = 0;
3448 *remaining_folds_to = NULL;
3449 return 0;
3450 }
3451
3452#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3453
3454 assert(base > 0);
3455
3456#else
3457
3458 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3459
3460 /* The data structure is set up so that the absolute value of 'base' is
3461 * an index into a table of pointers to arrays, with the array
3462 * corresponding to the index being the list of code points that fold
3463 * to 'cp', and the parallel array containing the length of the list
3464 * array */
3465 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
99f30495
KW
3466 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3467 /* +1 excludes first_folds_to */
b74fe592
KW
3468 return IVCF_AUX_TABLE_lengths[-base];
3469 }
3470
3471#endif
3472
3473 /* Only the single code point. This works like 'fc(G) = G - A + a' */
40d2776f
KW
3474 *first_folds_to = (U32) (base + cp
3475 - invlist_array(PL_utf8_foldclosures)[index]);
b74fe592
KW
3476 *remaining_folds_to = NULL;
3477 return 1;
3478}
3479
051a06d4 3480STATIC UV
56576a04
KW
3481S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3482 U8* const ustrp, STRLEN *lenp)
051a06d4 3483{
4a4088c4 3484 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3485 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3486 * result contains a character that crosses the 255/256 boundary, disallow
3487 * the change, and return the original code point. See L<perlfunc/lc> for
3488 * why;
051a06d4 3489 *
a1433954
KW
3490 * p points to the original string whose case was changed; assumed
3491 * by this routine to be well-formed
051a06d4 3492 * result the code point of the first character in the changed-case string
56576a04
KW
3493 * ustrp points to the changed-case string (<result> represents its
3494 * first char)
051a06d4
KW
3495 * lenp points to the length of <ustrp> */
3496
3497 UV original; /* To store the first code point of <p> */
3498
3499 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3500
a4f12ed7 3501 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3502
3503 /* We know immediately if the first character in the string crosses the
5e45c680 3504 * boundary, so can skip testing */
051a06d4
KW
3505 if (result > 255) {
3506
3507 /* Look at every character in the result; if any cross the
3508 * boundary, the whole thing is disallowed */
3509 U8* s = ustrp + UTF8SKIP(ustrp);
3510 U8* e = ustrp + *lenp;
3511 while (s < e) {
a4f12ed7 3512 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3513 goto bad_crossing;
3514 }
3515 s += UTF8SKIP(s);
3516 }
3517
613abc6d
KW
3518 /* Here, no characters crossed, result is ok as-is, but we warn. */
3519 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3520 return result;
3521 }
3522
7b52d656 3523 bad_crossing:
051a06d4
KW
3524
3525 /* Failed, have to return the original */
4b88fb76 3526 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3527
3528 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3529 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3530 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3531 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3532 OP_DESC(PL_op),
ab0b796c
KW
3533 original,
3534 original);
051a06d4
KW
3535 Copy(p, ustrp, *lenp, char);
3536 return original;
3537}
3538
b257a28c
KW
3539STATIC UV
3540S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3541 U8 * ustrp, STRLEN *lenp)
3542{
3543 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3544 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3545 * Otherwise, it returns the first code point of the Turkic foldcased
3546 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3547 * contain *lenp bytes
3548 *
3549 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3550 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3551 * DOTLESS I */
3552
3553 PERL_ARGS_ASSERT_TURKIC_FC;
3554 assert(e > p);
3555
3556 if (UNLIKELY(*p == 'I')) {
3557 *lenp = 2;
3558 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3559 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3560 return LATIN_SMALL_LETTER_DOTLESS_I;
3561 }
3562
3563 if (UNLIKELY(memBEGINs(p, e - p,
3564 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3565 {
3566 *lenp = 1;
3567 *ustrp = 'i';
3568 return 'i';
3569 }
3570
3571 return 0;
3572}
3573
3574STATIC UV
3575S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3576 U8 * ustrp, STRLEN *lenp)
3577{
3578 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3579 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3580 * Otherwise, it returns the first code point of the Turkic lowercased
3581 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3582 * contain *lenp bytes */
3583
1565c085 3584 dVAR;
b257a28c
KW
3585 PERL_ARGS_ASSERT_TURKIC_LC;
3586 assert(e > p0);
3587
3588 /* A 'I' requires context as to what to do */
3589 if (UNLIKELY(*p0 == 'I')) {
3590 const U8 * p = p0 + 1;
3591
3592 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3593 * modified by a dot above lowercases to 'i' even in turkic locales. */
3594 while (p < e) {
3595 UV cp;
3596
3597 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3598 ustrp[0] = 'i';
3599 *lenp = 1;
3600 return 'i';
3601 }
3602
3603 /* For the dot above to modify the 'I', it must be part of a
3604 * combining sequence immediately following the 'I', and no other
3605 * modifier with a ccc of 230 may intervene */
3606 cp = utf8_to_uvchr_buf(p, e, NULL);
3607 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3608 break;
3609 }
3610
3611 /* Here the combining sequence continues */
3612 p += UTF8SKIP(p);
3613 }
3614 }
3615
3616 /* In all other cases the lc is the same as the fold */
3617 return turkic_fc(p0, e, ustrp, lenp);
3618}
3619
3620STATIC UV
3621S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3622 U8 * ustrp, STRLEN *lenp)
3623{
3624 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3625 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3626 * Otherwise, it returns the first code point of the Turkic upper or
3627 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3628 * ustrp will contain *lenp bytes
3629 *
3630 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3631 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3632 * DOTLESS I */
3633
3634 PERL_ARGS_ASSERT_TURKIC_UC;
3635 assert(e > p);
3636
3637 if (*p == 'i') {
3638 *lenp = 2;
3639 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3640 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3641 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3642 }
3643
3644 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3645 *lenp = 1;
3646 *ustrp = 'I';
3647 return 'I';
3648 }
3649
3650 return 0;
3651}
3652
eaf412bf
KW
3653/* The process for changing the case is essentially the same for the four case
3654 * change types, except there are complications for folding. Otherwise the
3655 * difference is only which case to change to. To make sure that they all do
3656 * the same thing, the bodies of the functions are extracted out into the
3657 * following two macros. The functions are written with the same variable
3658 * names, and these are known and used inside these macros. It would be
3659 * better, of course, to have inline functions to do it, but since different
3660 * macros are called, depending on which case is being changed to, this is not
3661 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3662 * function can start with the common start macro, then finish with its special
3663 * handling; while the other three cases can just use the common end macro.
3664 *
3665 * The algorithm is to use the proper (passed in) macro or function to change
3666 * the case for code points that are below 256. The macro is used if using
3667 * locale rules for the case change; the function if not. If the code point is
3668 * above 255, it is computed from the input UTF-8, and another macro is called
3669 * to do the conversion. If necessary, the output is converted to UTF-8. If
3670 * using a locale, we have to check that the change did not cross the 255/256
3671 * boundary, see check_locale_boundary_crossing() for further details.
3672 *
3673 * The macros are split with the correct case change for the below-256 case
3674 * stored into 'result', and in the middle of an else clause for the above-255
3675 * case. At that point in the 'else', 'result' is not the final result, but is
3676 * the input code point calculated from the UTF-8. The fold code needs to
3677 * realize all this and take it from there.
3678 *
b257a28c
KW
3679 * To deal with Turkic locales, the function specified by the parameter
3680 * 'turkic' is called when appropriate.
3681 *
eaf412bf
KW
3682 * If you read the two macros as sequential, it's easier to understand what's
3683 * going on. */
3684#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
b257a28c 3685 L1_func_extra_param, turkic) \
a239b1e2 3686 \
eaf412bf 3687 if (flags & (locale_flags)) { \
8b7358b9 3688 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
eaf412bf 3689 if (IN_UTF8_CTYPE_LOCALE) { \
b257a28c
KW
3690 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3691 UV ret = turkic(p, e, ustrp, lenp); \
3692 if (ret) return ret; \
3693 } \
3694 \
3695 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3696 * all */ \
eaf412bf
KW
3697 flags &= ~(locale_flags); \
3698 } \
eaf412bf
KW
3699 } \
3700 \
3701 if (UTF8_IS_INVARIANT(*p)) { \
3702 if (flags & (locale_flags)) { \
3703 result = LC_L1_change_macro(*p); \
3704 } \
3705 else { \
3706 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3707 } \
3708 } \
a239b1e2 3709 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
1a751160 3710 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
eaf412bf 3711 if (flags & (locale_flags)) { \
1a751160 3712 result = LC_L1_change_macro(c); \
eaf412bf
KW
3713 } \
3714 else { \
1a751160