This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Use strict dfa to translate from UTF-8 to code point
[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"
afde5508 35#include "uni_keywords.h"
a0ed51b3 36
806547a7 37static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 38static const char unees[] =
806547a7 39 "Malformed UTF-8 character (unexpected end of string)";
fb7e7255
KW
40
41/* Be sure to synchronize this message with the similar one in regcomp.c */
760c7c2f 42static const char cp_above_legal_max[] =
d22ec717
KW
43 "Use of code point 0x%" UVXf " is not allowed; the"
44 " permissible max is 0x%" UVXf;
760c7c2f 45
d22ec717 46#define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
901b21bf 47
48ef279e 48/*
ccfc67b7 49=head1 Unicode Support
7fefc6c1 50These are various utility functions for manipulating UTF8-encoded
72d33970 51strings. For the uninitiated, this is a method of representing arbitrary
61296642 52Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
53characters in the ASCII range are unmodified, and a zero byte never appears
54within non-zero characters.
166f8a29 55
eaf7a4d2
CS
56=cut
57*/
58
9cbfb8ab
KW
59void
60Perl__force_out_malformed_utf8_message(pTHX_
61 const U8 *const p, /* First byte in UTF-8 sequence */
62 const U8 * const e, /* Final byte in sequence (may include
63 multiple chars */
64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
65 usually 0, or some DISALLOW flags */
66 const bool die_here) /* If TRUE, this function does not return */
67{
68 /* This core-only function is to be called when a malformed UTF-8 character
69 * is found, in order to output the detailed information about the
70 * malformation before dieing. The reason it exists is for the occasions
71 * when such a malformation is fatal, but warnings might be turned off, so
72 * that normally they would not be actually output. This ensures that they
73 * do get output. Because a sequence may be malformed in more than one
74 * way, multiple messages may be generated, so we can't make them fatal, as
75 * that would cause the first one to die.
76 *
77 * Instead we pretend -W was passed to perl, then die afterwards. The
78 * flexibility is here to return to the caller so they can finish up and
79 * die themselves */
80 U32 errors;
81
82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
83
84 ENTER;
c15a80f3 85 SAVEI8(PL_dowarn);
9cbfb8ab
KW
86 SAVESPTR(PL_curcop);
87
88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
89 if (PL_curcop) {
90 PL_curcop->cop_warnings = pWARN_ALL;
91 }
92
93 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
94
95 LEAVE;
96
97 if (! errors) {
98 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
99 " be called only when there are errors found");
100 }
101
102 if (die_here) {
103 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
104 }
105}
106
bb07812e
KW
107STATIC HV *
108S_new_msg_hv(pTHX_ const char * const message, /* The message text */
109 U32 categories, /* Packed warning categories */
110 U32 flag) /* Flag associated with this message */
111{
112 /* Creates, populates, and returns an HV* that describes an error message
113 * for the translators between UTF8 and code point */
114
115 SV* msg_sv = newSVpv(message, 0);
116 SV* category_sv = newSVuv(categories);
117 SV* flag_bit_sv = newSVuv(flag);
118
119 HV* msg_hv = newHV();
120
121 PERL_ARGS_ASSERT_NEW_MSG_HV;
122
2b672cf5
KW
123 (void) hv_stores(msg_hv, "text", msg_sv);
124 (void) hv_stores(msg_hv, "warn_categories", category_sv);
125 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
126
127 return msg_hv;
128}
129
eaf7a4d2 130/*
378516de 131=for apidoc uvoffuni_to_utf8_flags
eebe1485 132
a27992cc 133THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
134Instead, B<Almost all code should use L</uvchr_to_utf8> or
135L</uvchr_to_utf8_flags>>.
a27992cc 136
de69f3af
KW
137This function is like them, but the input is a strict Unicode
138(as opposed to native) code point. Only in very rare circumstances should code
139not be using the native code point.
949cf498 140
efa9cd84 141For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 142
eebe1485
SC
143=cut
144*/
145
33f38593
KW
146U8 *
147Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
148{
149 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
150
151 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
152}
153
c94c2f39
KW
154/* All these formats take a single UV code point argument */
155const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
156const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
157 " is not recommended for open interchange";
158const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
159 " may not be portable";
57ff5f59
KW
160const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
161 " Unicode, requires a Perl extension," \
162 " and so is not portable";
c94c2f39 163
33f38593 164#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
165 STMT_START { \
166 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
167 U32 category = packWARN(WARN_SURROGATE); \
168 const char * format = surrogate_cp_format; \
169 if (msgs) { \
170 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
171 category, \
172 UNICODE_GOT_SURROGATE); \
173 } \
174 else { \
175 Perl_ck_warner_d(aTHX_ category, format, uv); \
176 } \
8ee1cdcb
KW
177 } \
178 if (flags & UNICODE_DISALLOW_SURROGATE) { \
179 return NULL; \
180 } \
181 } STMT_END;
182
33f38593 183#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
184 STMT_START { \
185 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
186 U32 category = packWARN(WARN_NONCHAR); \
187 const char * format = nonchar_cp_format; \
188 if (msgs) { \
189 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
190 category, \
191 UNICODE_GOT_NONCHAR); \
192 } \
193 else { \
194 Perl_ck_warner_d(aTHX_ category, format, uv); \
195 } \
8ee1cdcb
KW
196 } \
197 if (flags & UNICODE_DISALLOW_NONCHAR) { \
198 return NULL; \
199 } \
200 } STMT_END;
201
ba6ed43c
KW
202/* Use shorter names internally in this file */
203#define SHIFT UTF_ACCUMULATION_SHIFT
204#undef MARK
205#define MARK UTF_CONTINUATION_MARK
206#define MASK UTF_CONTINUATION_MASK
207
33f38593
KW
208/*
209=for apidoc uvchr_to_utf8_flags_msgs
210
211THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
212
213Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
214
215This function is for code that wants any warning and/or error messages to be
216returned to the caller rather than be displayed. All messages that would have
884a31ee 217been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
218
219It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
220placed after all the others, C<msgs>. If this parameter is 0, this function
221behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
222be a pointer to an C<HV *> variable, in which this function creates a new HV to
223contain any appropriate messages. The hash has three key-value pairs, as
224follows:
225
226=over 4
227
228=item C<text>
229
230The text of the message as a C<SVpv>.
231
232=item C<warn_categories>
233
234The warning category (or categories) packed into a C<SVuv>.
235
236=item C<flag>
237
238A single flag bit associated with this message, in a C<SVuv>.
239The bit corresponds to some bit in the C<*errors> return value,
240such as C<UNICODE_GOT_SURROGATE>.
241
242=back
243
244It's important to note that specifying this parameter as non-null will cause
245any warnings this function would otherwise generate to be suppressed, and
246instead be placed in C<*msgs>. The caller can check the lexical warnings state
247(or not) when choosing what to do with the returned messages.
248
249The caller, of course, is responsible for freeing any returned HV.
250
251=cut
252*/
253
254/* Undocumented; we don't want people using this. Instead they should use
255 * uvchr_to_utf8_flags_msgs() */
dfe13c55 256U8 *
33f38593 257Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 258{
33f38593
KW
259 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
260
261 if (msgs) {
262 *msgs = NULL;
263 }
7918f24d 264
2d1545e5 265 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 266 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
267 return d;
268 }
facc1dc2 269
3ea68d71 270 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
271 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
272 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
273 return d;
274 }
d9432125 275
ba6ed43c
KW
276 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
277 * below, the 16 is for start bytes E0-EF (which are all the possible ones
278 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
279 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
280 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
281 * 0x800-0xFFFF on ASCII */
282 if (uv < (16 * (1U << (2 * SHIFT)))) {
283 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
284 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
285 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
286
287#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
288 aren't tested here */
289 /* The most likely code points in this range are below the surrogates.
290 * Do an extra test to quickly exclude those. */
291 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
292 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
293 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
294 {
33f38593 295 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
296 }
297 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 298 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 299 }
ba6ed43c
KW
300 }
301#endif
302 return d;
303 }
304
305 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
306 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
307 * happen starting with 4-byte characters on ASCII platforms. We unify the
308 * code for these with EBCDIC, even though some of them require 5-bytes on
309 * those, because khw believes the code saving is worth the very slight
310 * performance hit on these high EBCDIC code points. */
311
312 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
d22ec717
KW
313 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
314 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
a5bf80e0 315 }
33f38593
KW
316 if ( (flags & UNICODE_WARN_SUPER)
317 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 318 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 319 {
33f38593
KW
320 const char * format = super_cp_format;
321 U32 category = packWARN(WARN_NON_UNICODE);
322 U32 flag = UNICODE_GOT_SUPER;
323
324 /* Choose the more dire applicable warning */
325 if (UNICODE_IS_PERL_EXTENDED(uv)) {
326 format = perl_extended_cp_format;
327 if (flags & (UNICODE_WARN_PERL_EXTENDED
328 |UNICODE_DISALLOW_PERL_EXTENDED))
329 {
330 flag = UNICODE_GOT_PERL_EXTENDED;
331 }
332 }
a5bf80e0 333
33f38593
KW
334 if (msgs) {
335 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
336 category, flag);
337 }
338 else {
339 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
340 }
a5bf80e0 341 }
56576a04 342 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
343 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
344 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
345 {
346 return NULL;
347 }
348 }
ba6ed43c 349 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 350 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 351 }
d9432125 352
ba6ed43c
KW
353 /* Test for and handle 4-byte result. In the test immediately below, the
354 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
355 * characters). The 3 is for 3 continuation bytes; these each contribute
356 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
357 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
358 * 0x1_0000-0x1F_FFFF on ASCII */
359 if (uv < (8 * (1U << (3 * SHIFT)))) {
360 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
362 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
363 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
364
365#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
366 characters. The end-plane non-characters for EBCDIC were
367 handled just above */
368 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 369 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 370 }
ba6ed43c 371 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 372 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
373 }
374#endif
375
376 return d;
377 }
378
379 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
380 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
381 * format. The unrolled version above turns out to not save all that much
382 * time, and at these high code points (well above the legal Unicode range
383 * on ASCII platforms, and well above anything in common use in EBCDIC),
384 * khw believes that less code outweighs slight performance gains. */
385
d9432125 386 {
5aaebcb3 387 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
388 U8 *p = d+len-1;
389 while (p > d) {
957a9e81
KW
390 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
391 uv >>= SHIFT;
1d72bdf6 392 }
4c8cd605 393 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
394 return d+len;
395 }
a0ed51b3 396}
a5bf80e0 397
646ca15d 398/*
07693fe6
KW
399=for apidoc uvchr_to_utf8
400
bcb1a2d4 401Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 402of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
403C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
404the byte after the end of the new character. In other words,
07693fe6
KW
405
406 d = uvchr_to_utf8(d, uv);
407
408is the recommended wide native character-aware way of saying
409
410 *(d++) = uv;
411
d22ec717
KW
412This function accepts any code point from 0..C<IV_MAX> as input.
413C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
414
415It is possible to forbid or warn on non-Unicode code points, or those that may
416be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 417
07693fe6
KW
418=cut
419*/
420
de69f3af
KW
421/* This is also a macro */
422PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
423
07693fe6
KW
424U8 *
425Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
426{
de69f3af 427 return uvchr_to_utf8(d, uv);
07693fe6
KW
428}
429
de69f3af
KW
430/*
431=for apidoc uvchr_to_utf8_flags
432
433Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 434of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
435C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
436the byte after the end of the new character. In other words,
de69f3af
KW
437
438 d = uvchr_to_utf8_flags(d, uv, flags);
439
440or, in most cases,
441
442 d = uvchr_to_utf8_flags(d, uv, 0);
443
444This is the Unicode-aware way of saying
445
446 *(d++) = uv;
447
d22ec717
KW
448If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
449input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
450
451Specifying C<flags> can further restrict what is allowed and not warned on, as
452follows:
de69f3af 453
796b6530 454If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
455the function will raise a warning, provided UTF8 warnings are enabled. If
456instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
457NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 458
760c7c2f
KW
459Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
460affect how the function handles a Unicode non-character.
93e6dbd6 461
760c7c2f
KW
462And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
463affect the handling of code points that are above the Unicode maximum of
4640x10FFFF. Languages other than Perl may not be able to accept files that
465contain these.
93e6dbd6
KW
466
467The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
468the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
469three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
470allowed inputs to the strict UTF-8 traditionally defined by Unicode.
471Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
472C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
473above-Unicode and surrogate flags, but not the non-character ones, as
474defined in
475L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
476See L<perlunicode/Noncharacter code points>.
93e6dbd6 477
57ff5f59
KW
478Extremely high code points were never specified in any standard, and require an
479extension to UTF-8 to express, which Perl does. It is likely that programs
480written in something other than Perl would not be able to read files that
481contain these; nor would Perl understand files written by something that uses a
482different extension. For these reasons, there is a separate set of flags that
483can warn and/or disallow these extremely high code points, even if other
484above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
485and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
486L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
487treat all above-Unicode code points, including these, as malformations. (Note
488that the Unicode standard considers anything above 0x10FFFF to be illegal, but
489there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
490
491A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
492retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
493C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
494C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
495platforms,these flags can apply to code points that actually do fit in 31 bits.
496The new names accurately describe the situation in all cases.
de69f3af 497
de69f3af
KW
498=cut
499*/
500
501/* This is also a macro */
502PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
503
07693fe6
KW
504U8 *
505Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
506{
de69f3af 507 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
508}
509
57ff5f59
KW
510#ifndef UV_IS_QUAD
511
e050c007
KW
512STATIC int
513S_is_utf8_cp_above_31_bits(const U8 * const s,
514 const U8 * const e,
515 const bool consider_overlongs)
83dc0f42
KW
516{
517 /* Returns TRUE if the first code point represented by the Perl-extended-
518 * UTF-8-encoded string starting at 's', and looking no further than 'e -
519 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
520 *
521 * The function handles the case where the input bytes do not include all
522 * the ones necessary to represent a full character. That is, they may be
523 * the intial bytes of the representation of a code point, but possibly
524 * the final ones necessary for the complete representation may be beyond
525 * 'e - 1'.
526 *
e050c007
KW
527 * The function also can handle the case where the input is an overlong
528 * sequence. If 'consider_overlongs' is 0, the function assumes the
529 * input is not overlong, without checking, and will return based on that
530 * assumption. If this parameter is 1, the function will go to the trouble
531 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 532 *
e050c007 533 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
534 */
535
e050c007
KW
536 const STRLEN len = e - s;
537 int is_overlong;
538
539 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
540
541 assert(! UTF8_IS_INVARIANT(*s) && e > s);
542
83dc0f42
KW
543#ifdef EBCDIC
544
e050c007 545 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 546
e050c007
KW
547 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
548 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
549 * also be the start byte for a 31-bit code point; we need at least 2
550 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
551 * the start byte for an overlong sequence, but for 30-bit or smaller code
552 * points, so we don't have to worry about overlongs on EBCDIC.) */
553 if (*s != 0xFE) {
554 return 0;
555 }
83dc0f42 556
e050c007
KW
557 if (len == 1) {
558 return -1;
559 }
83dc0f42 560
e050c007 561#else
83dc0f42 562
e050c007
KW
563 /* On ASCII, FE and FF are the only start bytes that can evaluate to
564 * needing more than 31 bits. */
565 if (LIKELY(*s < 0xFE)) {
566 return 0;
567 }
83dc0f42 568
e050c007
KW
569 /* What we have left are FE and FF. Both of these require more than 31
570 * bits unless they are for overlongs. */
571 if (! consider_overlongs) {
572 return 1;
573 }
83dc0f42 574
e050c007
KW
575 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
576 * above 31 bits. But we need more than one byte to discern this, so if
577 * passed just the start byte, it could be an overlong evaluating to
578 * smaller */
579 if (len == 1) {
580 return -1;
581 }
83dc0f42 582
e050c007
KW
583 /* Having excluded len==1, and knowing that FE and FF are both valid start
584 * bytes, we can call the function below to see if the sequence is
585 * overlong. (We don't need the full generality of the called function,
586 * but for these huge code points, speed shouldn't be a consideration, and
587 * the compiler does have enough information, since it's static to this
588 * file, to optimize to just the needed parts.) */
589 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 590
e050c007
KW
591 /* If it isn't overlong, more than 31 bits are required. */
592 if (is_overlong == 0) {
593 return 1;
594 }
83dc0f42 595
e050c007
KW
596 /* If it is indeterminate if it is overlong, return that */
597 if (is_overlong < 0) {
598 return -1;
599 }
600
601 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
602 * the max it can be is 2**31 - 1 */
603 if (*s == 0xFE) {
604 return 0;
83dc0f42
KW
605 }
606
e050c007
KW
607#endif
608
609 /* Here, ASCII and EBCDIC rejoin:
610 * On ASCII: We have an overlong sequence starting with FF
611 * On EBCDIC: We have a sequence starting with FE. */
612
613 { /* For C89, use a block so the declaration can be close to its use */
614
615#ifdef EBCDIC
616
5f995336
KW
617 /* U+7FFFFFFF (2 ** 31 - 1)
618 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
619 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
620 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
621 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
622 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
623 * U+80000000 (2 ** 31):
624 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
625 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
626 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
627 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
628 *
629 * and since we know that *s = \xfe, any continuation sequcence
630 * following it that is gt the below is above 31 bits
631 [0] [1] [2] [3] [4] [5] [6] */
632 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
633
634#else
635
636 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
637 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
638 * FF overlong for U+80000000 (2 ** 31):
639 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
640 * and since we know that *s = \xff, any continuation sequcence
641 * following it that is gt the below is above 30 bits
642 [0] [1] [2] [3] [4] [5] [6] */
643 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 644
83dc0f42
KW
645
646#endif
e050c007
KW
647 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
648 const STRLEN cmp_len = MIN(conts_len, len - 1);
649
650 /* Now compare the continuation bytes in s with the ones we have
651 * compiled in that are for the largest 30 bit code point. If we have
652 * enough bytes available to determine the answer, or the bytes we do
653 * have differ from them, we can compare the two to get a definitive
654 * answer (Note that in UTF-EBCDIC, the two lowest possible
655 * continuation bytes are \x41 and \x42.) */
656 if (cmp_len >= conts_len || memNE(s + 1,
657 conts_for_highest_30_bit,
658 cmp_len))
659 {
660 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
661 }
83dc0f42 662
e050c007
KW
663 /* Here, all the bytes we have are the same as the highest 30-bit code
664 * point, but we are missing so many bytes that we can't make the
665 * determination */
666 return -1;
667 }
83dc0f42
KW
668}
669
57ff5f59
KW
670#endif
671
d6be65ae 672PERL_STATIC_INLINE int
12a4bed3
KW
673S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
674{
d6be65ae
KW
675 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
676 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
677 * it isn't, and -1 if there isn't enough information to tell. This last
678 * return value can happen if the sequence is incomplete, missing some
679 * trailing bytes that would form a complete character. If there are
680 * enough bytes to make a definitive decision, this function does so.
681 * Usually 2 bytes sufficient.
682 *
683 * Overlongs can occur whenever the number of continuation bytes changes.
684 * That means whenever the number of leading 1 bits in a start byte
685 * increases from the next lower start byte. That happens for start bytes
686 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
687 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
688 * ASCII platforms: C0, C1
689 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 690 */
12a4bed3
KW
691
692 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
693 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
694
695 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
696 assert(len > 1 && UTF8_IS_START(*s));
697
698 /* Each platform has overlongs after the start bytes given above (expressed
699 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
700 * the logic is the same, except the E0 overlong has already been excluded
701 * on EBCDIC platforms. The values below were found by manually
702 * inspecting the UTF-8 patterns. See the tables in utf8.h and
703 * utfebcdic.h. */
704
705# ifdef EBCDIC
706# define F0_ABOVE_OVERLONG 0xB0
707# define F8_ABOVE_OVERLONG 0xA8
708# define FC_ABOVE_OVERLONG 0xA4
709# define FE_ABOVE_OVERLONG 0xA2
710# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
711 /* I8(0xfe) is FF */
712# else
713
714 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 715 return 1;
12a4bed3
KW
716 }
717
718# define F0_ABOVE_OVERLONG 0x90
719# define F8_ABOVE_OVERLONG 0x88
720# define FC_ABOVE_OVERLONG 0x84
721# define FE_ABOVE_OVERLONG 0x82
722# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
723# endif
724
725
726 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
727 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
728 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
729 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
730 {
d6be65ae 731 return 1;
12a4bed3
KW
732 }
733
b0b342d4 734 /* Check for the FF overlong */
d6be65ae 735 return isFF_OVERLONG(s, len);
b0b342d4
KW
736}
737
8d6204cc 738PERL_STATIC_INLINE int
b0b342d4
KW
739S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
740{
8d6204cc
KW
741 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
742 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
743 * it isn't, and -1 if there isn't enough information to tell. This last
744 * return value can happen if the sequence is incomplete, missing some
745 * trailing bytes that would form a complete character. If there are
746 * enough bytes to make a definitive decision, this function does so. */
747
b0b342d4 748 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 749
8d6204cc
KW
750 /* To be an FF overlong, all the available bytes must match */
751 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
752 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
753 {
754 return 0;
755 }
756
757 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
758 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 759 * utfebcdic.h. */
8d6204cc
KW
760 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
761 return 1;
762 }
12a4bed3 763
8d6204cc
KW
764 /* The missing bytes could cause the result to go one way or the other, so
765 * the result is indeterminate */
766 return -1;
12a4bed3
KW
767}
768
d22ec717 769#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
770# ifdef EBCDIC /* Actually is I8 */
771# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 772 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
773# else
774# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 775 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
776# endif
777#endif
778
c285bbc4 779PERL_STATIC_INLINE int
e050c007
KW
780S_does_utf8_overflow(const U8 * const s,
781 const U8 * e,
782 const bool consider_overlongs)
a77c906e 783{
c285bbc4 784 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
785 * 'e' - 1 would overflow an IV on this platform; that is if it represents
786 * a code point larger than the highest representable code point. It
787 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
788 * enough information to tell. This last return value can happen if the
789 * sequence is incomplete, missing some trailing bytes that would form a
790 * complete character. If there are enough bytes to make a definitive
791 * decision, this function does so.
c285bbc4 792 *
e050c007
KW
793 * If 'consider_overlongs' is TRUE, the function checks for the possibility
794 * that the sequence is an overlong that doesn't overflow. Otherwise, it
795 * assumes the sequence is not an overlong. This can give different
796 * results only on ASCII 32-bit platforms.
797 *
c285bbc4
KW
798 * (For ASCII platforms, we could use memcmp() because we don't have to
799 * convert each byte to I8, but it's very rare input indeed that would
800 * approach overflow, so the loop below will likely only get executed once.)
801 *
802 * 'e' - 1 must not be beyond a full character. */
a77c906e 803
a77c906e
KW
804
805 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
806 assert(s <= e && s + UTF8SKIP(s) >= e);
807
d22ec717
KW
808#if ! defined(UV_IS_QUAD)
809
810 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
811
812#else
813
814 PERL_UNUSED_ARG(consider_overlongs);
815
816 {
817 const STRLEN len = e - s;
818 const U8 *x;
819 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
820
821 for (x = s; x < e; x++, y++) {
822
823 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
824 continue;
825 }
826
827 /* If this byte is larger than the corresponding highest UTF-8
828 * byte, the sequence overflow; otherwise the byte is less than,
829 * and so the sequence doesn't overflow */
830 return NATIVE_UTF8_TO_I8(*x) > *y;
831
832 }
833
834 /* Got to the end and all bytes are the same. If the input is a whole
835 * character, it doesn't overflow. And if it is a partial character,
836 * there's not enough information to tell */
837 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
838 return -1;
839 }
840
841 return 0;
842 }
843
844#endif
845
846}
847
848#if 0
849
850/* This is the portions of the above function that deal with UV_MAX instead of
851 * IV_MAX. They are left here in case we want to combine them so that internal
852 * uses can have larger code points. The only logic difference is that the
853 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
854 * different logic.
855 */
856
857/* Anything larger than this will overflow the word if it were converted into a UV */
858#if defined(UV_IS_QUAD)
859# ifdef EBCDIC /* Actually is I8 */
860# define HIGHEST_REPRESENTABLE_UTF8 \
861 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
862# else
863# define HIGHEST_REPRESENTABLE_UTF8 \
864 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
865# endif
866#else /* 32-bit */
867# ifdef EBCDIC
868# define HIGHEST_REPRESENTABLE_UTF8 \
869 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
870# else
871# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
872# endif
873#endif
874
a77c906e
KW
875#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
876
877 /* On 32 bit ASCII machines, many overlongs that start with FF don't
878 * overflow */
e050c007 879 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
880
881 /* To be such an overlong, the first bytes of 's' must match
882 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
883 * don't have any additional bytes available, the sequence, when
884 * completed might or might not fit in 32 bits. But if we have that
885 * next byte, we can tell for sure. If it is <= 0x83, then it does
886 * fit. */
887 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
888 return -1;
889 }
890
891 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
892 }
893
d22ec717
KW
894/* Starting with the #else, the rest of the function is identical except
895 * 1. we need to move the 'len' declaration to be global to the function
896 * 2. the endif move to just after the UNUSED_ARG.
897 * An empty endif is given just below to satisfy the preprocessor
898 */
a77c906e
KW
899#endif
900
d22ec717 901#endif
a77c906e 902
12a4bed3
KW
903#undef F0_ABOVE_OVERLONG
904#undef F8_ABOVE_OVERLONG
905#undef FC_ABOVE_OVERLONG
906#undef FE_ABOVE_OVERLONG
907#undef FF_OVERLONG_PREFIX
908
35f8c9bd 909STRLEN
edc2c47a 910Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 911{
2b479609 912 STRLEN len;
12a4bed3 913 const U8 *x;
35f8c9bd 914
2b479609
KW
915 /* A helper function that should not be called directly.
916 *
917 * This function returns non-zero if the string beginning at 's' and
918 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
919 * code point; otherwise it returns 0. The examination stops after the
920 * first code point in 's' is validated, not looking at the rest of the
921 * input. If 'e' is such that there are not enough bytes to represent a
922 * complete code point, this function will return non-zero anyway, if the
923 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
924 * excluded by 'flags'.
925 *
926 * A non-zero return gives the number of bytes required to represent the
927 * code point. Be aware that if the input is for a partial character, the
928 * return will be larger than 'e - s'.
929 *
930 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
931 * The caller should have excluded the possibility of it being invariant
932 * before calling this function.
2b479609
KW
933 *
934 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
935 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
936 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
937 * disallowed by the flags. If the input is only for a partial character,
938 * the function will return non-zero if there is any sequence of
939 * well-formed UTF-8 that, when appended to the input sequence, could
940 * result in an allowed code point; otherwise it returns 0. Non characters
941 * cannot be determined based on partial character input. But many of the
942 * other excluded types can be determined with just the first one or two
943 * bytes.
944 *
945 */
946
947 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
948
949 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 950 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 951 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 952
2b479609 953 /* A variant char must begin with a start byte */
35f8c9bd
KW
954 if (UNLIKELY(! UTF8_IS_START(*s))) {
955 return 0;
956 }
957
edc2c47a
KW
958 /* Examine a maximum of a single whole code point */
959 if (e - s > UTF8SKIP(s)) {
960 e = s + UTF8SKIP(s);
961 }
962
2b479609
KW
963 len = e - s;
964
965 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
966 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 967
56576a04
KW
968 /* Here, we are disallowing some set of largish code points, and the
969 * first byte indicates the sequence is for a code point that could be
970 * in the excluded set. We generally don't have to look beyond this or
971 * the second byte to see if the sequence is actually for one of the
972 * excluded classes. The code below is derived from this table:
973 *
2b479609
KW
974 * UTF-8 UTF-EBCDIC I8
975 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
976 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
977 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
978 *
56576a04
KW
979 * Keep in mind that legal continuation bytes range between \x80..\xBF
980 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
981 * continuation bytes. Hence, we don't have to test the upper edge
982 * because if any of those is encountered, the sequence is malformed,
983 * and would fail elsewhere in this function.
984 *
985 * The code here likewise assumes that there aren't other
986 * malformations; again the function should fail elsewhere because of
987 * these. For example, an overlong beginning with FC doesn't actually
988 * have to be a super; it could actually represent a small code point,
989 * even U+0000. But, since overlongs (and other malformations) are
990 * illegal, the function should return FALSE in either case.
2b479609
KW
991 */
992
993#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
994# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 995# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 996
19794540
KW
997# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
998 /* B6 and B7 */ \
999 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1000# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1001#else
1002# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1003# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1004# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1005# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1006#endif
1007
1008 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1009 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1010 {
2b479609
KW
1011 return 0; /* Above Unicode */
1012 }
1013
d044b7a7 1014 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1015 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1016 {
57ff5f59 1017 return 0;
2b479609
KW
1018 }
1019
1020 if (len > 1) {
1021 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1022
1023 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1024 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1025 {
1026 return 0; /* Above Unicode */
1027 }
1028
1029 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1030 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1031 {
1032 return 0; /* Surrogate */
1033 }
1034
1035 if ( (flags & UTF8_DISALLOW_NONCHAR)
1036 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1037 {
1038 return 0; /* Noncharacter code point */
1039 }
1040 }
1041 }
1042
1043 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1044 for (x = s + 1; x < e; x++) {
1045 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1046 return 0;
1047 }
1048 }
1049
af13dd8a 1050 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1051 * overlong. */
d6be65ae 1052 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1053 return 0;
af13dd8a
KW
1054 }
1055
12a4bed3
KW
1056 /* And finally, that the code point represented fits in a word on this
1057 * platform */
e050c007
KW
1058 if (0 < does_utf8_overflow(s, e,
1059 0 /* Don't consider overlongs */
1060 ))
1061 {
12a4bed3 1062 return 0;
35f8c9bd
KW
1063 }
1064
2b479609 1065 return UTF8SKIP(s);
35f8c9bd
KW
1066}
1067
7e2f38b2 1068char *
63ab03b3 1069Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1070{
1071 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1072 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1073 * Currently, there are only two formats, so it is currently a bool:
1074 * 0 \xab
1075 * 1 ab (that is a space between two hex digit bytes)
1076 */
7cf8d05d
KW
1077
1078 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1079 trailing NUL */
63ab03b3
KW
1080 const U8 * s = start;
1081 const U8 * const e = start + len;
7cf8d05d
KW
1082 char * output;
1083 char * d;
1084
1085 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1086
1087 Newx(output, output_len, char);
1088 SAVEFREEPV(output);
1089
1090 d = output;
63ab03b3 1091 for (s = start; s < e; s++) {
7cf8d05d
KW
1092 const unsigned high_nibble = (*s & 0xF0) >> 4;
1093 const unsigned low_nibble = (*s & 0x0F);
1094
7e2f38b2 1095 if (format) {
63ab03b3
KW
1096 if (s > start) {
1097 *d++ = ' ';
1098 }
7e2f38b2
KW
1099 }
1100 else {
1101 *d++ = '\\';
1102 *d++ = 'x';
1103 }
7cf8d05d
KW
1104
1105 if (high_nibble < 10) {
1106 *d++ = high_nibble + '0';
1107 }
1108 else {
1109 *d++ = high_nibble - 10 + 'a';
1110 }
1111
1112 if (low_nibble < 10) {
1113 *d++ = low_nibble + '0';
1114 }
1115 else {
1116 *d++ = low_nibble - 10 + 'a';
1117 }
1118 }
1119
1120 *d = '\0';
1121 return output;
1122}
1123
806547a7 1124PERL_STATIC_INLINE char *
7cf8d05d
KW
1125S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1126
421da25c 1127 /* Max number of bytes to print */
3cc6a05e 1128 STRLEN print_len,
7cf8d05d
KW
1129
1130 /* Which one is the non-continuation */
1131 const STRLEN non_cont_byte_pos,
1132
1133 /* How many bytes should there be? */
1134 const STRLEN expect_len)
806547a7
KW
1135{
1136 /* Return the malformation warning text for an unexpected continuation
1137 * byte. */
1138
7cf8d05d 1139 const char * const where = (non_cont_byte_pos == 1)
806547a7 1140 ? "immediately"
7cf8d05d
KW
1141 : Perl_form(aTHX_ "%d bytes",
1142 (int) non_cont_byte_pos);
421da25c
KW
1143 const U8 * x = s + non_cont_byte_pos;
1144 const U8 * e = s + print_len;
806547a7
KW
1145
1146 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1147
7cf8d05d
KW
1148 /* We don't need to pass this parameter, but since it has already been
1149 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1150 assert(expect_len == UTF8SKIP(s));
1151
421da25c
KW
1152 /* As a defensive coding measure, don't output anything past a NUL. Such
1153 * bytes shouldn't be in the middle of a malformation, and could mark the
1154 * end of the allocated string, and what comes after is undefined */
1155 for (; x < e; x++) {
1156 if (*x == '\0') {
1157 x++; /* Output this particular NUL */
1158 break;
1159 }
1160 }
1161
7cf8d05d
KW
1162 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1163 " %s after start byte 0x%02x; need %d bytes, got %d)",
1164 malformed_text,
421da25c 1165 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1166 *(s + non_cont_byte_pos),
1167 where,
1168 *s,
1169 (int) expect_len,
1170 (int) non_cont_byte_pos);
806547a7
KW
1171}
1172
35f8c9bd
KW
1173/*
1174
de69f3af 1175=for apidoc utf8n_to_uvchr
378516de
KW
1176
1177THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1178Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1179
9041c2e3 1180Bottom level UTF-8 decode routine.
de69f3af 1181Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1182which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1183C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1184the length, in bytes, of that character.
949cf498
KW
1185
1186The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1187well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1188causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1189is the next possible position in C<s> that could begin a non-malformed
1190character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1191is raised. Some UTF-8 input sequences may contain multiple malformations.
1192This function tries to find every possible one in each call, so multiple
56576a04 1193warnings can be raised for the same sequence.
949cf498
KW
1194
1195Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1196individual types of malformations, such as the sequence being overlong (that
1197is, when there is a shorter sequence that can express the same code point;
1198overlong sequences are expressly forbidden in the UTF-8 standard due to
1199potential security issues). Another malformation example is the first byte of
1200a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1201flags. Even if allowed, this function generally returns the Unicode
1202REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1203F<utf8.h> to override this behavior for the overlong malformations, but don't
1204do that except for very specialized purposes.
949cf498 1205
796b6530 1206The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1207flags) malformation is found. If this flag is set, the routine assumes that
1208the caller will raise a warning, and this function will silently just set
d088425d
KW
1209C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1210
75200dff 1211Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1212character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1213in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1214be set to 1. To disambiguate, upon a zero return, see if the first byte of
1215C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1216error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1217
1218Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1219Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1220By default these are considered regular code points, but certain situations
ecc1615f
KW
1221warrant special handling for them, which can be specified using the C<flags>
1222parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1223three classes are treated as malformations and handled as such. The flags
1224C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1225C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1226disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1227restricts the allowed inputs to the strict UTF-8 traditionally defined by
1228Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1229definition given by
1230L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1231The difference between traditional strictness and C9 strictness is that the
1232latter does not forbid non-character code points. (They are still discouraged,
1233however.) For more discussion see L<perlunicode/Noncharacter code points>.
1234
1235The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1236C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1237C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1238raised for their respective categories, but otherwise the code points are
1239considered valid (not malformations). To get a category to both be treated as
1240a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1241(But note that warnings are not raised if lexically disabled nor if
796b6530 1242C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1243
57ff5f59
KW
1244Extremely high code points were never specified in any standard, and require an
1245extension to UTF-8 to express, which Perl does. It is likely that programs
1246written in something other than Perl would not be able to read files that
1247contain these; nor would Perl understand files written by something that uses a
1248different extension. For these reasons, there is a separate set of flags that
1249can warn and/or disallow these extremely high code points, even if other
1250above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1251C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1252L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1253above-Unicode code points, including these, as malformations.
1254(Note that the Unicode standard considers anything above 0x10FFFF to be
1255illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1256(2**31 -1))
1257
1258A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1259retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1260C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1261C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1262can apply to code points that actually do fit in 31 bits. This happens on
1263EBCDIC platforms, and sometimes when the L<overlong
1264malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1265describe the situation in all cases.
1266
ab8e6d41 1267
949cf498
KW
1268All other code points corresponding to Unicode characters, including private
1269use and those yet to be assigned, are never considered malformed and never
1270warn.
67e989fb 1271
37607a96 1272=cut
f9380377
KW
1273
1274Also implemented as a macro in utf8.h
1275*/
1276
1277UV
1278Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1279 STRLEN curlen,
1280 STRLEN *retlen,
1281 const U32 flags)
1282{
1283 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1284
1285 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1286}
1287
1288/*
1289
1290=for apidoc utf8n_to_uvchr_error
1291
1292THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1293Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1294
1295This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1296are when an error is found. If you also need to know the generated warning
1297messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1298
1299It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1300all the others, C<errors>. If this parameter is 0, this function behaves
1301identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1302to a C<U32> variable, which this function sets to indicate any errors found.
1303Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1304C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1305of these bits will be set if a malformation is found, even if the input
7a65503b 1306C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1307exceptions are noted:
1308
1309=over 4
1310
57ff5f59 1311=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1312
57ff5f59
KW
1313The input sequence is not standard UTF-8, but a Perl extension. This bit is
1314set only if the input C<flags> parameter contains either the
1315C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1316
1317Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1318and so some extension must be used to express them. Perl uses a natural
1319extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1320extension to represent even higher ones, so that any code point that fits in a
132164-bit word can be represented. Text using these extensions is not likely to
1322be portable to non-Perl code. We lump both of these extensions together and
1323refer to them as Perl extended UTF-8. There exist other extensions that people
1324have invented, incompatible with Perl's.
1325
1326On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1327extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1328than on ASCII. Prior to that, code points 2**31 and higher were simply
1329unrepresentable, and a different, incompatible method was used to represent
1330code points between 2**30 and 2**31 - 1.
1331
1332On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1333Perl extended UTF-8 is used.
1334
1335In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1336may use for backward compatibility. That name is misleading, as this flag may
1337be set when the code point actually does fit in 31 bits. This happens on
1338EBCDIC platforms, and sometimes when the L<overlong
1339malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1340describes the situation in all cases.
f9380377
KW
1341
1342=item C<UTF8_GOT_CONTINUATION>
1343
1344The input sequence was malformed in that the first byte was a a UTF-8
1345continuation byte.
1346
1347=item C<UTF8_GOT_EMPTY>
1348
1349The input C<curlen> parameter was 0.
1350
1351=item C<UTF8_GOT_LONG>
1352
1353The input sequence was malformed in that there is some other sequence that
1354evaluates to the same code point, but that sequence is shorter than this one.
1355
fecaf136
KW
1356Until Unicode 3.1, it was legal for programs to accept this malformation, but
1357it was discovered that this created security issues.
1358
f9380377
KW
1359=item C<UTF8_GOT_NONCHAR>
1360
1361The code point represented by the input UTF-8 sequence is for a Unicode
1362non-character code point.
1363This bit is set only if the input C<flags> parameter contains either the
1364C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1365
1366=item C<UTF8_GOT_NON_CONTINUATION>
1367
1368The input sequence was malformed in that a non-continuation type byte was found
1369in a position where only a continuation type one should be.
1370
1371=item C<UTF8_GOT_OVERFLOW>
1372
1373The input sequence was malformed in that it is for a code point that is not
d22ec717 1374representable in the number of bits available in an IV on the current platform.
f9380377
KW
1375
1376=item C<UTF8_GOT_SHORT>
1377
1378The input sequence was malformed in that C<curlen> is smaller than required for
1379a complete sequence. In other words, the input is for a partial character
1380sequence.
1381
1382=item C<UTF8_GOT_SUPER>
1383
1384The input sequence was malformed in that it is for a non-Unicode code point;
1385that is, one above the legal Unicode maximum.
1386This bit is set only if the input C<flags> parameter contains either the
1387C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1388
1389=item C<UTF8_GOT_SURROGATE>
1390
1391The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1392code point.
1393This bit is set only if the input C<flags> parameter contains either the
1394C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1395
1396=back
1397
133551d8
KW
1398To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1399flag to suppress any warnings, and then examine the C<*errors> return.
1400
f9380377 1401=cut
37657a5b
KW
1402
1403Also implemented as a macro in utf8.h
37607a96 1404*/
67e989fb 1405
a0ed51b3 1406UV
f9380377 1407Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
37657a5b
KW
1408 STRLEN curlen,
1409 STRLEN *retlen,
1410 const U32 flags,
1411 U32 * errors)
1412{
1413 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1414
1415 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1416}
1417
1418/*
1419
1420=for apidoc utf8n_to_uvchr_msgs
1421
1422THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1423Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1424
1425This function is for code that needs to know what the precise malformation(s)
1426are when an error is found, and wants the corresponding warning and/or error
1427messages to be returned to the caller rather than be displayed. All messages
1428that would have been displayed if all lexcial warnings are enabled will be
1429returned.
1430
1431It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1432placed after all the others, C<msgs>. If this parameter is 0, this function
1433behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1434be a pointer to an C<AV *> variable, in which this function creates a new AV to
1435contain any appropriate messages. The elements of the array are ordered so
1436that the first message that would have been displayed is in the 0th element,
1437and so on. Each element is a hash with three key-value pairs, as follows:
1438
1439=over 4
1440
1441=item C<text>
1442
1443The text of the message as a C<SVpv>.
1444
1445=item C<warn_categories>
1446
1447The warning category (or categories) packed into a C<SVuv>.
1448
1449=item C<flag>
1450
1451A single flag bit associated with this message, in a C<SVuv>.
1452The bit corresponds to some bit in the C<*errors> return value,
1453such as C<UTF8_GOT_LONG>.
1454
1455=back
1456
1457It's important to note that specifying this parameter as non-null will cause
1458any warnings this function would otherwise generate to be suppressed, and
1459instead be placed in C<*msgs>. The caller can check the lexical warnings state
1460(or not) when choosing what to do with the returned messages.
1461
1462If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1463no AV is created.
1464
1465The caller, of course, is responsible for freeing any returned AV.
1466
1467=cut
1468*/
1469
1470UV
1471Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1472 STRLEN curlen,
1473 STRLEN *retlen,
1474 const U32 flags,
1475 U32 * errors,
1476 AV ** msgs)
a0ed51b3 1477{
d4c19fe8 1478 const U8 * const s0 = s;
2b9519f0 1479 const U8 * send = s0 + curlen;
2b5e7bc2
KW
1480 U32 possible_problems = 0; /* A bit is set here for each potential problem
1481 found as we go along */
2b672cf5 1482 UV uv = (UV) -1;
2b5e7bc2
KW
1483 STRLEN expectlen = 0; /* How long should this sequence be?
1484 (initialized to silence compilers' wrong
1485 warning) */
e308b348 1486 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1487 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1488 this gets set and discarded */
a0dbb045 1489
2b5e7bc2
KW
1490 /* The below are used only if there is both an overlong malformation and a
1491 * too short one. Otherwise the first two are set to 's0' and 'send', and
1492 * the third not used at all */
1493 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1494 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1495 routine; see [perl #130921] */
2b5e7bc2 1496 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1497
2b9519f0
KW
1498 UV state = 0;
1499
37657a5b 1500 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
f9380377
KW
1501
1502 if (errors) {
1503 *errors = 0;
1504 }
1505 else {
1506 errors = &discard_errors;
1507 }
a0dbb045 1508
eb83ed87
KW
1509 /* The order of malformation tests here is important. We should consume as
1510 * few bytes as possible in order to not skip any valid character. This is
1511 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1512 * http://unicode.org/reports/tr36 for more discussion as to why. For
1513 * example, once we've done a UTF8SKIP, we can tell the expected number of
1514 * bytes, and could fail right off the bat if the input parameters indicate
1515 * that there are too few available. But it could be that just that first
1516 * byte is garbled, and the intended character occupies fewer bytes. If we
1517 * blindly assumed that the first byte is correct, and skipped based on
1518 * that number, we could skip over a valid input character. So instead, we
1519 * always examine the sequence byte-by-byte.
1520 *
1521 * We also should not consume too few bytes, otherwise someone could inject
1522 * things. For example, an input could be deliberately designed to
1523 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1524 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1525 * which is actually part of of the overflowing sequence), that could look
1526 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1527 * sequence and process the rest, inappropriately.
1528 *
1529 * Some possible input sequences are malformed in more than one way. This
1530 * function goes to lengths to try to find all of them. This is necessary
1531 * for correctness, as the inputs may allow one malformation but not
1532 * another, and if we abandon searching for others after finding the
1533 * allowed one, we could allow in something that shouldn't have been.
1534 */
eb83ed87 1535
b5b9af04 1536 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1537 possible_problems |= UTF8_GOT_EMPTY;
1538 curlen = 0;
5a48568d 1539 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1540 goto ready_to_handle_errors;
0c443dc2
JH
1541 }
1542
eb83ed87
KW
1543 expectlen = UTF8SKIP(s);
1544
1545 /* A well-formed UTF-8 character, as the vast majority of calls to this
1546 * function will be for, has this expected length. For efficiency, set
1547 * things up here to return it. It will be overriden only in those rare
1548 * cases where a malformation is found */
1549 if (retlen) {
1550 *retlen = expectlen;
1551 }
1552
1553 /* An invariant is trivially well-formed */
2b9519f0
KW
1554 if (UTF8_IS_INVARIANT(*s0)) {
1555 return *s0;
a0ed51b3 1556 }
67e989fb 1557
2b9519f0
KW
1558 /* Measurements show that this dfa is somewhat faster than the regular code
1559 * below, so use it first, dropping down for the non-normal cases. */
1560
3de6d141 1561#define PERL_UTF8_DECODE_REJECT 1
2b9519f0
KW
1562
1563 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
6c12993c 1564 UV type = strict_utf8_dfa_tab[*s];
2b9519f0 1565
3720e27c
KW
1566 uv = (state == 0)
1567 ? ((0xff >> type) & NATIVE_UTF8_TO_I8(*s))
1568 : UTF8_ACCUMULATE(uv, *s);
6c12993c 1569 state = strict_utf8_dfa_tab[256 + state + type];
2b9519f0
KW
1570
1571 if (state == 0) {
6c12993c 1572 return uv;
2b9519f0
KW
1573 }
1574
1575 s++;
1576 }
1577
6c12993c
KW
1578 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1579 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1580 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1581 * final case.
1582 *
1583 * Each of the affected Hanguls starts with \xED */
1584
1585#ifndef EBCDIC
2b9519f0 1586
6c12993c
KW
1587 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1588 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1589 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1590 | (s0[2] & UTF_CONTINUATION_MASK);
1591 }
1592
1593#endif
1594
1595 /* Here is potentially problematic. Use the full mechanism */
2b9519f0
KW
1596 uv = *s0;
1597
6c12993c
KW
1598 /* In conjunction with the exhaustive tests that can be enabled in
1599 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1600 * what it is intended to do, and that no flaws in it are masked by
1601 * dropping down and executing the code below
1602
1603 assert(! isUTF8_CHAR(s0, send)
1604 || UTF8_IS_SURROGATE(s0, send)
1605 || UTF8_IS_SUPER(s0, send)
1606 || UTF8_IS_NONCHAR(s0,send));
1607 */
1608
eb83ed87 1609 /* A continuation character can't start a valid sequence */
b5b9af04 1610 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1611 possible_problems |= UTF8_GOT_CONTINUATION;
1612 curlen = 1;
1613 uv = UNICODE_REPLACEMENT;
1614 goto ready_to_handle_errors;
ba210ebe 1615 }
9041c2e3 1616
dcd27b3c 1617 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1618 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1619 * because it excludes start bytes like \xC0 that always lead to
1620 * overlongs.) */
dcd27b3c 1621
534752c1
KW
1622 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1623 * that indicate the number of bytes in the character's whole UTF-8
1624 * sequence, leaving just the bits that are part of the value. */
1625 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1626
e308b348
KW
1627 /* Setup the loop end point, making sure to not look past the end of the
1628 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1629 if (UNLIKELY(curlen < expectlen)) {
1630 possible_problems |= UTF8_GOT_SHORT;
1631 avail_len = curlen;
e308b348
KW
1632 }
1633 else {
2b9519f0 1634 send = (U8*) s0 + expectlen;
e308b348 1635 }
e308b348 1636
eb83ed87 1637 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1638 * accumulating each into the working value as we go. */
eb83ed87 1639 for (s = s0 + 1; s < send; s++) {
b5b9af04 1640 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1641 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1642 continue;
1643 }
1644
1645 /* Here, found a non-continuation before processing all expected bytes.
1646 * This byte indicates the beginning of a new character, so quit, even
1647 * if allowing this malformation. */
2b5e7bc2 1648 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1649 break;
eb83ed87
KW
1650 } /* End of loop through the character's bytes */
1651
1652 /* Save how many bytes were actually in the character */
1653 curlen = s - s0;
1654
2b5e7bc2
KW
1655 /* Note that there are two types of too-short malformation. One is when
1656 * there is actual wrong data before the normal termination of the
1657 * sequence. The other is that the sequence wasn't complete before the end
1658 * of the data we are allowed to look at, based on the input 'curlen'.
1659 * This means that we were passed data for a partial character, but it is
1660 * valid as far as we saw. The other is definitely invalid. This
1661 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1662 * separate.
1663 *
1664 * A convenience macro that matches either of the too-short conditions. */
1665# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1666
1667 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1668 uv_so_far = uv;
1669 uv = UNICODE_REPLACEMENT;
1670 }
2b5e7bc2 1671
08e73697
KW
1672 /* Check for overflow. The algorithm requires us to not look past the end
1673 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1674 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1675 1 /* Do consider overlongs */
1676 )))
1677 {
2b5e7bc2
KW
1678 possible_problems |= UTF8_GOT_OVERFLOW;
1679 uv = UNICODE_REPLACEMENT;
eb83ed87 1680 }
eb83ed87 1681
2b5e7bc2
KW
1682 /* Check for overlong. If no problems so far, 'uv' is the correct code
1683 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1684 * we must look at the UTF-8 byte sequence itself to see if it is for an
1685 * overlong */
1686 if ( ( LIKELY(! possible_problems)
1687 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1688 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1689 && ( UNLIKELY(! UTF8_IS_START(*s0))
1690 || ( curlen > 1
d6be65ae 1691 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1692 s - s0))))))
2f8f112e 1693 {
2b5e7bc2
KW
1694 possible_problems |= UTF8_GOT_LONG;
1695
abc28b54 1696 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1697
abc28b54
KW
1698 /* The calculation in the 'true' branch of this 'if'
1699 * below won't work if overflows, and isn't needed
1700 * anyway. Further below we handle all overflow
1701 * cases */
1702 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1703 {
2b5e7bc2
KW
1704 UV min_uv = uv_so_far;
1705 STRLEN i;
1706
1707 /* Here, the input is both overlong and is missing some trailing
1708 * bytes. There is no single code point it could be for, but there
1709 * may be enough information present to determine if what we have
1710 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1711 * The code further below has the intelligence to determine this,
1712 * but just for non-overlong UTF-8 sequences. What we do here is
1713 * calculate the smallest code point the input could represent if
1714 * there were no too short malformation. Then we compute and save
1715 * the UTF-8 for that, which is what the code below looks at
1716 * instead of the raw input. It turns out that the smallest such
1717 * code point is all we need. */
2b5e7bc2
KW
1718 for (i = curlen; i < expectlen; i++) {
1719 min_uv = UTF8_ACCUMULATE(min_uv,
1720 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1721 }
1722
e9f2c446 1723 adjusted_s0 = temp_char_buf;
57ff5f59 1724 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1725 }
eb83ed87
KW
1726 }
1727
56576a04
KW
1728 /* Here, we have found all the possible problems, except for when the input
1729 * is for a problematic code point not allowed by the input parameters. */
1730
06188866
KW
1731 /* uv is valid for overlongs */
1732 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1733
1734 /* isn't problematic if < this */
1735 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1736 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1737
1738 /* if overflow, we know without looking further
1739 * precisely which of the problematic types it is,
1740 * and we deal with those in the overflow handling
1741 * code */
1742 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1743 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1744 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1745 && ((flags & ( UTF8_DISALLOW_NONCHAR
1746 |UTF8_DISALLOW_SURROGATE
1747 |UTF8_DISALLOW_SUPER
d044b7a7 1748 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1749 |UTF8_WARN_NONCHAR
1750 |UTF8_WARN_SURROGATE
1751 |UTF8_WARN_SUPER
d22ec717 1752 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1753 {
2b5e7bc2
KW
1754 /* If there were no malformations, or the only malformation is an
1755 * overlong, 'uv' is valid */
1756 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1757 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1758 possible_problems |= UTF8_GOT_SURROGATE;
1759 }
1760 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1761 possible_problems |= UTF8_GOT_SUPER;
1762 }
1763 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1764 possible_problems |= UTF8_GOT_NONCHAR;
1765 }
1766 }
1767 else { /* Otherwise, need to look at the source UTF-8, possibly
1768 adjusted to be non-overlong */
1769
1770 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1771 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1772 {
2b5e7bc2
KW
1773 possible_problems |= UTF8_GOT_SUPER;
1774 }
1775 else if (curlen > 1) {
1776 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1777 NATIVE_UTF8_TO_I8(*adjusted_s0),
1778 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1779 {
2b5e7bc2 1780 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1781 }
2b5e7bc2
KW
1782 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1783 NATIVE_UTF8_TO_I8(*adjusted_s0),
1784 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1785 {
1786 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1787 }
1788 }
c0236afe 1789
2b5e7bc2
KW
1790 /* We need a complete well-formed UTF-8 character to discern
1791 * non-characters, so can't look for them here */
1792 }
1793 }
949cf498 1794
2b5e7bc2
KW
1795 ready_to_handle_errors:
1796
1797 /* At this point:
1798 * curlen contains the number of bytes in the sequence that
1799 * this call should advance the input by.
e308b348
KW
1800 * avail_len gives the available number of bytes passed in, but
1801 * only if this is less than the expected number of
1802 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1803 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1804 * is set in it for each potential problem found.
1805 * uv contains the code point the input sequence
1806 * represents; or if there is a problem that prevents
1807 * a well-defined value from being computed, it is
1808 * some subsitute value, typically the REPLACEMENT
1809 * CHARACTER.
1810 * s0 points to the first byte of the character
56576a04
KW
1811 * s points to just after were we left off processing
1812 * the character
1813 * send points to just after where that character should
1814 * end, based on how many bytes the start byte tells
1815 * us should be in it, but no further than s0 +
1816 * avail_len
2b5e7bc2 1817 */
eb83ed87 1818
2b5e7bc2
KW
1819 if (UNLIKELY(possible_problems)) {
1820 bool disallowed = FALSE;
1821 const U32 orig_problems = possible_problems;
1822
37657a5b
KW
1823 if (msgs) {
1824 *msgs = NULL;
1825 }
1826
2b5e7bc2
KW
1827 while (possible_problems) { /* Handle each possible problem */
1828 UV pack_warn = 0;
1829 char * message = NULL;
37657a5b 1830 U32 this_flag_bit = 0;
2b5e7bc2
KW
1831
1832 /* Each 'if' clause handles one problem. They are ordered so that
1833 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1834 * ones; this is kinda in decreasing severity order. But the
1835 * overlong must come last, as it changes 'uv' looked at by the
1836 * others */
2b5e7bc2
KW
1837 if (possible_problems & UTF8_GOT_OVERFLOW) {
1838
56576a04
KW
1839 /* Overflow means also got a super and are using Perl's
1840 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1841 possible_problems
d044b7a7 1842 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1843 *errors |= UTF8_GOT_OVERFLOW;
1844
1845 /* But the API says we flag all errors found */
1846 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1847 *errors |= UTF8_GOT_SUPER;
1848 }
ddb65933 1849 if (flags
d044b7a7 1850 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1851 {
d044b7a7 1852 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1853 }
2b5e7bc2 1854
d60baaa7 1855 /* Disallow if any of the three categories say to */
56576a04 1856 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1857 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1858 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1859 {
1860 disallowed = TRUE;
1861 }
1862
d22ec717
KW
1863 /* Likewise, warn if any say to */
1864 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1865 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1866 {
2b5e7bc2 1867
ddb65933
KW
1868 /* The warnings code explicitly says it doesn't handle the
1869 * case of packWARN2 and two categories which have
1870 * parent-child relationship. Even if it works now to
1871 * raise the warning if either is enabled, it wouldn't
1872 * necessarily do so in the future. We output (only) the
56576a04 1873 * most dire warning */
ddb65933 1874 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1875 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1876 pack_warn = packWARN(WARN_UTF8);
1877 }
37657a5b 1878 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1879 pack_warn = packWARN(WARN_NON_UNICODE);
1880 }
1881 if (pack_warn) {
1882 message = Perl_form(aTHX_ "%s: %s (overflows)",
1883 malformed_text,
05b9033b 1884 _byte_dump_string(s0, curlen, 0));
37657a5b 1885 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1886 }
2b5e7bc2
KW
1887 }
1888 }
1889 }
1890 else if (possible_problems & UTF8_GOT_EMPTY) {
1891 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1892 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1893
1894 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1895
1896 /* This so-called malformation is now treated as a bug in
1897 * the caller. If you have nothing to decode, skip calling
1898 * this function */
1899 assert(0);
1900
2b5e7bc2 1901 disallowed = TRUE;
37657a5b
KW
1902 if ( (msgs
1903 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1904 {
2b5e7bc2
KW
1905 pack_warn = packWARN(WARN_UTF8);
1906 message = Perl_form(aTHX_ "%s (empty string)",
1907 malformed_text);
37657a5b 1908 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1909 }
1910 }
1911 }
1912 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1913 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1914 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1915
1916 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1917 disallowed = TRUE;
37657a5b
KW
1918 if (( msgs
1919 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1920 {
2b5e7bc2
KW
1921 pack_warn = packWARN(WARN_UTF8);
1922 message = Perl_form(aTHX_
1923 "%s: %s (unexpected continuation byte 0x%02x,"
1924 " with no preceding start byte)",
1925 malformed_text,
7e2f38b2 1926 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1927 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1928 }
1929 }
1930 }
2b5e7bc2
KW
1931 else if (possible_problems & UTF8_GOT_SHORT) {
1932 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1933 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1934
1935 if (! (flags & UTF8_ALLOW_SHORT)) {
1936 disallowed = TRUE;
37657a5b
KW
1937 if (( msgs
1938 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1939 {
2b5e7bc2
KW
1940 pack_warn = packWARN(WARN_UTF8);
1941 message = Perl_form(aTHX_
56576a04
KW
1942 "%s: %s (too short; %d byte%s available, need %d)",
1943 malformed_text,
1944 _byte_dump_string(s0, send - s0, 0),
1945 (int)avail_len,
1946 avail_len == 1 ? "" : "s",
1947 (int)expectlen);
37657a5b 1948 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1949 }
1950 }
ba210ebe 1951
2b5e7bc2 1952 }
e308b348
KW
1953 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1954 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1955 *errors |= UTF8_GOT_NON_CONTINUATION;
1956
1957 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1958 disallowed = TRUE;
37657a5b
KW
1959 if (( msgs
1960 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1961 {
99a765e9
KW
1962
1963 /* If we don't know for sure that the input length is
1964 * valid, avoid as much as possible reading past the
1965 * end of the buffer */
1966 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1967 ? s - s0
1968 : send - s0;
e308b348
KW
1969 pack_warn = packWARN(WARN_UTF8);
1970 message = Perl_form(aTHX_ "%s",
1971 unexpected_non_continuation_text(s0,
99a765e9 1972 printlen,
e308b348
KW
1973 s - s0,
1974 (int) expectlen));
37657a5b 1975 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
1976 }
1977 }
1978 }
2b5e7bc2
KW
1979 else if (possible_problems & UTF8_GOT_SURROGATE) {
1980 possible_problems &= ~UTF8_GOT_SURROGATE;
1981
f9380377
KW
1982 if (flags & UTF8_WARN_SURROGATE) {
1983 *errors |= UTF8_GOT_SURROGATE;
1984
1985 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 1986 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 1987 {
2b5e7bc2
KW
1988 pack_warn = packWARN(WARN_SURROGATE);
1989
1990 /* These are the only errors that can occur with a
1991 * surrogate when the 'uv' isn't valid */
1992 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1993 message = Perl_form(aTHX_
1994 "UTF-16 surrogate (any UTF-8 sequence that"
1995 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1996 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1997 }
1998 else {
c94c2f39 1999 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2000 }
37657a5b 2001 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2002 }
2b5e7bc2 2003 }
ba210ebe 2004
2b5e7bc2
KW
2005 if (flags & UTF8_DISALLOW_SURROGATE) {
2006 disallowed = TRUE;
f9380377 2007 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2008 }
2009 }
2010 else if (possible_problems & UTF8_GOT_SUPER) {
2011 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2012
f9380377
KW
2013 if (flags & UTF8_WARN_SUPER) {
2014 *errors |= UTF8_GOT_SUPER;
2015
2016 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2017 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2018 {
2b5e7bc2
KW
2019 pack_warn = packWARN(WARN_NON_UNICODE);
2020
2021 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2022 message = Perl_form(aTHX_
2023 "Any UTF-8 sequence that starts with"
2024 " \"%s\" is for a non-Unicode code point,"
2025 " may not be portable",
7e2f38b2 2026 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2027 }
2028 else {
c94c2f39 2029 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2030 }
37657a5b 2031 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2032 }
2b5e7bc2 2033 }
ba210ebe 2034
57ff5f59
KW
2035 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2036 * and before possibly bailing out, so that the more dire
2037 * warning will override the regular one. */
2038 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2039 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2040 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2041 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2042 {
db0f09e6 2043 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2044
57ff5f59
KW
2045 /* If it is an overlong that evaluates to a code point
2046 * that doesn't have to use the Perl extended UTF-8, it
2047 * still used it, and so we output a message that
2048 * doesn't refer to the code point. The same is true
2049 * if there was a SHORT malformation where the code
2050 * point is not valid. In that case, 'uv' will have
2051 * been set to the REPLACEMENT CHAR, and the message
2052 * below without the code point in it will be selected
2053 * */
2054 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2055 message = Perl_form(aTHX_
57ff5f59 2056 perl_extended_cp_format, uv);
2b5e7bc2
KW
2057 }
2058 else {
2059 message = Perl_form(aTHX_
57ff5f59
KW
2060 "Any UTF-8 sequence that starts with"
2061 " \"%s\" is a Perl extension, and"
2062 " so is not portable",
2063 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2064 }
37657a5b 2065 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2066 }
2067
d044b7a7
KW
2068 if (flags & ( UTF8_WARN_PERL_EXTENDED
2069 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2070 {
d044b7a7 2071 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2072
d044b7a7 2073 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2074 disallowed = TRUE;
2075 }
2b5e7bc2
KW
2076 }
2077 }
eb83ed87 2078
2b5e7bc2 2079 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2080 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2081 disallowed = TRUE;
2082 }
2b5e7bc2
KW
2083 }
2084 else if (possible_problems & UTF8_GOT_NONCHAR) {
2085 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2086
f9380377
KW
2087 if (flags & UTF8_WARN_NONCHAR) {
2088 *errors |= UTF8_GOT_NONCHAR;
2089
2090 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2091 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2092 {
2b5e7bc2
KW
2093 /* The code above should have guaranteed that we don't
2094 * get here with errors other than overlong */
2095 assert (! (orig_problems
2096 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2097
2098 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2099 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2100 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2101 }
2b5e7bc2 2102 }
5b311467 2103
2b5e7bc2
KW
2104 if (flags & UTF8_DISALLOW_NONCHAR) {
2105 disallowed = TRUE;
f9380377 2106 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2107 }
6c64cd9d
KW
2108 }
2109 else if (possible_problems & UTF8_GOT_LONG) {
2110 possible_problems &= ~UTF8_GOT_LONG;
2111 *errors |= UTF8_GOT_LONG;
2112
2113 if (flags & UTF8_ALLOW_LONG) {
2114
2115 /* We don't allow the actual overlong value, unless the
2116 * special extra bit is also set */
2117 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2118 & ~UTF8_ALLOW_LONG)))
2119 {
2120 uv = UNICODE_REPLACEMENT;
2121 }
2122 }
2123 else {
2124 disallowed = TRUE;
2125
37657a5b
KW
2126 if (( msgs
2127 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2128 {
6c64cd9d
KW
2129 pack_warn = packWARN(WARN_UTF8);
2130
2131 /* These error types cause 'uv' to be something that
2132 * isn't what was intended, so can't use it in the
2133 * message. The other error types either can't
2134 * generate an overlong, or else the 'uv' is valid */
2135 if (orig_problems &
2136 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2137 {
2138 message = Perl_form(aTHX_
2139 "%s: %s (any UTF-8 sequence that starts"
2140 " with \"%s\" is overlong which can and"
2141 " should be represented with a"
2142 " different, shorter sequence)",
2143 malformed_text,
2144 _byte_dump_string(s0, send - s0, 0),
2145 _byte_dump_string(s0, curlen, 0));
2146 }
2147 else {
2148 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2149 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2150 uv, 0);
d819dc50
KW
2151 /* Don't use U+ for non-Unicode code points, which
2152 * includes those in the Latin1 range */
2153 const char * preface = ( uv > PERL_UNICODE_MAX
2154#ifdef EBCDIC
2155 || uv <= 0xFF
2156#endif
2157 )
2158 ? "0x"
2159 : "U+";
6c64cd9d
KW
2160 message = Perl_form(aTHX_
2161 "%s: %s (overlong; instead use %s to represent"
2162 " %s%0*" UVXf ")",
2163 malformed_text,
2164 _byte_dump_string(s0, send - s0, 0),
2165 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2166 preface,
2167 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2168 small code points */
1be62ab9 2169 UNI_TO_NATIVE(uv));
6c64cd9d 2170 }
37657a5b 2171 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2172 }
2173 }
2b5e7bc2
KW
2174 } /* End of looking through the possible flags */
2175
2176 /* Display the message (if any) for the problem being handled in
2177 * this iteration of the loop */
2178 if (message) {
37657a5b 2179 if (msgs) {
37657a5b
KW
2180 assert(this_flag_bit);
2181
2182 if (*msgs == NULL) {
2183 *msgs = newAV();
2184 }
2185
bb07812e
KW
2186 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2187 pack_warn,
2188 this_flag_bit)));
37657a5b
KW
2189 }
2190 else if (PL_op)
2b5e7bc2
KW
2191 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2192 OP_DESC(PL_op));
2193 else
2194 Perl_warner(aTHX_ pack_warn, "%s", message);
2195 }
ddb65933 2196 } /* End of 'while (possible_problems)' */
a0dbb045 2197
2b5e7bc2
KW
2198 /* Since there was a possible problem, the returned length may need to
2199 * be changed from the one stored at the beginning of this function.
2200 * Instead of trying to figure out if that's needed, just do it. */
2201 if (retlen) {
2202 *retlen = curlen;
2203 }
a0dbb045 2204
2b5e7bc2
KW
2205 if (disallowed) {
2206 if (flags & UTF8_CHECK_ONLY && retlen) {
2207 *retlen = ((STRLEN) -1);
2208 }
2209 return 0;
2210 }
eb83ed87 2211 }
ba210ebe 2212
2b5e7bc2 2213 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2214}
2215
8e84507e 2216/*
ec5f19d0
KW
2217=for apidoc utf8_to_uvchr_buf
2218
2219Returns the native code point of the first character in the string C<s> which
2220is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2221C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2222
524080c4
KW
2223If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2224enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2225C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2226(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2227C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2228the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2229See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2230returned.
ec5f19d0
KW
2231
2232=cut
52be2536
KW
2233
2234Also implemented as a macro in utf8.h
2235
ec5f19d0
KW
2236*/
2237
2238
2239UV
2240Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2241{
7f974d7e
KW
2242 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2243
ec5f19d0
KW
2244 assert(s < send);
2245
2246 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 2247 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
2248}
2249
52be2536
KW
2250/* This is marked as deprecated
2251 *
ec5f19d0
KW
2252=for apidoc utf8_to_uvuni_buf
2253
de69f3af
KW
2254Only in very rare circumstances should code need to be dealing in Unicode
2255(as opposed to native) code points. In those few cases, use
07dfe0a4
KW
2256C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2257are not absolutely sure this is one of those cases, then assume it isn't and
2258use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2259
2260Returns the Unicode (not-native) code point of the first character in the
2261string C<s> which
ec5f19d0
KW
2262is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2263C<retlen> will be set to the length, in bytes, of that character.
2264
524080c4
KW
2265If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2266enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2267NULL) to -1. If those warnings are off, the computed value if well-defined (or
2268the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2269is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2270next possible position in C<s> that could begin a non-malformed character.
de69f3af 2271See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2272
2273=cut
2274*/
2275
2276UV
2277Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2278{
2279 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2280
2281 assert(send > s);
2282
5962d97e 2283 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2284}
2285
b76347f2 2286/*
87cea99e 2287=for apidoc utf8_length
b76347f2 2288
b2e7ed74
KW
2289Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2290at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2291same place, it returns 0 with no warning raised.
2292
2293If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2294and returns the number of valid characters.
b76347f2
JH
2295
2296=cut
2297*/
2298
2299STRLEN
35a4481c 2300Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2301{
2302 STRLEN len = 0;
2303
7918f24d
NC
2304 PERL_ARGS_ASSERT_UTF8_LENGTH;
2305
8850bf83
JH
2306 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2307 * the bitops (especially ~) can create illegal UTF-8.
2308 * In other words: in Perl UTF-8 is not just for Unicode. */
2309
a3b680e6
AL
2310 if (e < s)
2311 goto warn_and_return;
b76347f2 2312 while (s < e) {
4cbf4130 2313 s += UTF8SKIP(s);
8e91ec7f
AV
2314 len++;
2315 }
2316
2317 if (e != s) {
2318 len--;
2319 warn_and_return:
9b387841
NC
2320 if (PL_op)
2321 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2322 "%s in %s", unees, OP_DESC(PL_op));
2323 else
61a12c31 2324 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2325 }
2326
2327 return len;
2328}
2329
b06226ff 2330/*
fed3ba5d
NC
2331=for apidoc bytes_cmp_utf8
2332
a1433954 2333Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2334sequence of characters (stored as UTF-8)
2335in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2336equal, -1 or -2 if the first string is less than the second string, +1 or +2
2337if the first string is greater than the second string.
2338
2339-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2340longer string. -2 or +2 is returned if
2341there was a difference between characters
fed3ba5d
NC
2342within the strings.
2343
2344=cut
2345*/
2346
2347int
2348Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2349{
2350 const U8 *const bend = b + blen;
2351 const U8 *const uend = u + ulen;
2352
2353 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2354
2355 while (b < bend && u < uend) {
2356 U8 c = *u++;
2357 if (!UTF8_IS_INVARIANT(c)) {
2358 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2359 if (u < uend) {
2360 U8 c1 = *u++;
2361 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2362 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2363 } else {
2b5e7bc2 2364 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2365 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2366 "%s %s%s",
2367 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2368 PL_op ? " in " : "",
2369 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2370 return -2;
2371 }
2372 } else {
2373 if (PL_op)
2374 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2375 "%s in %s", unees, OP_DESC(PL_op));
2376 else
61a12c31 2377 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2378 return -2; /* Really want to return undef :-) */
2379 }
2380 } else {
2381 return -2;
2382 }
2383 }
2384 if (*b != c) {
2385 return *b < c ? -2 : +2;
2386 }
2387 ++b;
2388 }
2389
2390 if (b == bend && u == uend)
2391 return 0;
2392
2393 return b < bend ? +1 : -1;
2394}
2395
2396/*
87cea99e 2397=for apidoc utf8_to_bytes
6940069f 2398
3bc0c78c 2399Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2400Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2401updates C<*lenp> to contain the new length.
3bc0c78c
KW
2402Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2403
2404Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2405having saved the value of C<*lenp> before the call, and subtracting the
2406after-call value of C<*lenp> from it.
6940069f 2407
a1433954 2408If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2409
6940069f
GS
2410=cut
2411*/
2412
2413U8 *
09af0336 2414Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2415{
9fe0d3c2 2416 U8 * first_variant;
246fae53 2417
7918f24d 2418 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2419 PERL_UNUSED_CONTEXT;
7918f24d 2420
9fe0d3c2 2421 /* This is a no-op if no variants at all in the input */
09af0336 2422 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2423 return s;
2424 }
2425
2426 {
3c5aa262 2427 U8 * const save = s;
09af0336 2428 U8 * const send = s + *lenp;
3c5aa262
KW
2429 U8 * d;
2430
2431 /* Nothing before the first variant needs to be changed, so start the real
2432 * work there */
2433 s = first_variant;
2434 while (s < send) {
2435 if (! UTF8_IS_INVARIANT(*s)) {
2436 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2437 *lenp = ((STRLEN) -1);
3c5aa262
KW
2438 return 0;
2439 }
2440 s++;
d59937ca
KW
2441 }
2442 s++;
dcad2880 2443 }
dcad2880 2444
3c5aa262
KW
2445 /* Is downgradable, so do it */
2446 d = s = first_variant;
2447 while (s < send) {
2448 U8 c = *s++;
2449 if (! UVCHR_IS_INVARIANT(c)) {
2450 /* Then it is two-byte encoded */
2451 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2452 s++;
2453 }
2454 *d++ = c;
2455 }
2456 *d = '\0';
09af0336 2457 *lenp = d - save;
3c5aa262
KW
2458
2459 return save;
9fe0d3c2 2460 }
6940069f
GS
2461}
2462
2463/*
87cea99e 2464=for apidoc bytes_from_utf8
f9a63242 2465
09af0336 2466Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2467byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2468actually encoded in UTF-8.
2469
2470Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2471the input string.
2472
41ae6089
KW
2473Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2474not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2475C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2476
41ae6089 2477Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2478newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2479returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2480caller is responsible for arranging for the memory used by this string to get
2481freed.
f9a63242 2482
3bc0c78c 2483Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2484having saved the value of C<*lenp> before the call, and subtracting the
2485after-call value of C<*lenp> from it.
3bc0c78c 2486
37607a96 2487=cut
976c1b08
KW
2488
2489There is a macro that avoids this function call, but this is retained for
2490anyone who calls it with the Perl_ prefix */
f9a63242
JH
2491
2492U8 *
41ae6089 2493Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2494{
7918f24d 2495 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2496 PERL_UNUSED_CONTEXT;
f9a63242 2497
976c1b08
KW
2498 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2499}
2500
2501/*
2502No = here because currently externally undocumented
2503for apidoc bytes_from_utf8_loc
2504
2505Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2506to store the location of the first character in C<"s"> that cannot be
2507converted to non-UTF8.
2508
2509If that parameter is C<NULL>, this function behaves identically to
2510C<bytes_from_utf8>.
2511
2512Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2513C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2514
2515Otherwise, the function returns a newly created C<NUL>-terminated string
2516containing the non-UTF8 equivalent of the convertible first portion of
2517C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2518If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2519and C<*first_non_downgradable> is set to C<NULL>.
2520
2521Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2522first character in the original string that wasn't converted. C<*is_utf8p> is
2523unchanged. Note that the new string may have length 0.
2524
2525Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2526C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2527converts as many characters in it as possible stopping at the first one it
385b74be 2528finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2529set to point to that. The function returns the portion that could be converted
2530in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2531not including the terminating C<NUL>. If the very first character in the
2532original could not be converted, C<*lenp> will be 0, and the new string will
2533contain just a single C<NUL>. If the entire input string was converted,
2534C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2535
2536Upon successful return, the number of variants in the converted portion of the
2537string can be computed by having saved the value of C<*lenp> before the call,
2538and subtracting the after-call value of C<*lenp> from it.
2539
2540=cut
2541
2542
2543*/
2544
2545U8 *
2546Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2547{
2548 U8 *d;
2549 const U8 *original = s;
2550 U8 *converted_start;
2551 const U8 *send = s + *lenp;
f9a63242 2552
976c1b08 2553 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2554
976c1b08
KW
2555 if (! *is_utf8p) {
2556 if (first_unconverted) {
2557 *first_unconverted = NULL;
2558 }
2559
2560 return (U8 *) original;
2561 }
2562
2563 Newx(d, (*lenp) + 1, U8);
2564
2565 converted_start = d;
7299a045
KW
2566 while (s < send) {
2567 U8 c = *s++;
2568 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2569
2570 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2571 * have to stop now */
2572 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2573 if (first_unconverted) {
2574 *first_unconverted = s - 1;
2575 goto finish_and_return;
2576 }
2577 else {
2578 Safefree(converted_start);
2579 return (U8 *) original;
2580 }
2581 }
2582
7299a045
KW
2583 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2584 s++;
38af28cf 2585 }
7299a045
KW
2586 *d++ = c;
2587 }
170a1c22 2588
976c1b08
KW
2589 /* Here, converted the whole of the input */
2590 *is_utf8p = FALSE;
2591 if (first_unconverted) {
2592 *first_unconverted = NULL;
170a1c22 2593 }
976c1b08
KW
2594
2595 finish_and_return:
46a08a6f
KW
2596 *d = '\0';
2597 *lenp = d - converted_start;
976c1b08
KW
2598
2599 /* Trim unused space */
2600 Renew(converted_start, *lenp + 1, U8);
2601
2602 return converted_start;
f9a63242
JH
2603}
2604
2605/*
87cea99e 2606=for apidoc bytes_to_utf8
6940069f 2607
09af0336 2608Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2609UTF-8.
09af0336 2610Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2611reflect the new length in bytes. The caller is responsible for arranging for
2612the memory used by this string to get freed.
6940069f 2613
3bc0c78c 2614Upon successful return, the number of variants in the string can be computed by
23b37b12 2615having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2616after-call value of C<*lenp>.
2617
75200dff 2618A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2619
2620If you want to convert to UTF-8 from encodings other than
2621the native (Latin1 or EBCDIC),
a1433954 2622see L</sv_recode_to_utf8>().
c9ada85f 2623
497711e7 2624=cut
6940069f
GS
2625*/
2626
2627U8*
09af0336 2628Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2629{
09af0336 2630 const U8 * const send = s + (*lenp);
6940069f
GS
2631 U8 *d;
2632 U8 *dst;
7918f24d
NC
2633
2634 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2635 PERL_UNUSED_CONTEXT;
6940069f 2636
09af0336 2637 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2638 dst = d;
2639
2640 while (s < send) {
55d09dc8
KW
2641 append_utf8_from_native_byte(*s, &d);
2642 s++;
6940069f 2643 }
2e11cf67 2644
6940069f 2645 *d = '\0';
09af0336 2646 *lenp = d-dst;
2e11cf67
KW
2647
2648 /* Trim unused space */
2649 Renew(dst, *lenp + 1, U8);
2650
6940069f
GS
2651 return dst;
2652}
2653
a0ed51b3 2654/*
624504c5
KW
2655 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2656 * use utf16_to_utf8_reversed().
a0ed51b3 2657 *
624504c5
KW
2658 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2659 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2660 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2661 *
2662 * These functions don't check for overflow. The worst case is every code
2663 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2664 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2665 * destination must be pre-extended to 2 times the source length.
2666 *
2667 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2668
2669U8*
dea0fc0b 2670Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2671{
dea0fc0b
JH
2672 U8* pend;
2673 U8* dstart = d;
2674
7918f24d
NC
2675 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2676
dea0fc0b 2677 if (bytelen & 1)
56576a04
KW
2678 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2679 (UV)bytelen);
dea0fc0b
JH
2680
2681 pend = p + bytelen;
2682
a0ed51b3 2683 while (p < pend) {
dea0fc0b
JH
2684 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2685 p += 2;
2d1545e5 2686 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2687 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2688 continue;
2689 }
56d37426
KW
2690 if (uv <= MAX_UTF8_TWO_BYTE) {
2691 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2692 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2693 continue;
2694 }
ffd0a9d3 2695
46956fad
KW
2696#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2697#define LAST_HIGH_SURROGATE 0xDBFF
2698#define FIRST_LOW_SURROGATE 0xDC00
2699#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2700#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2701
2702 /* This assumes that most uses will be in the first Unicode plane, not
2703 * needing surrogates */
2704 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2705 && uv <= UNICODE_SURROGATE_LAST))
2706 {
2707 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2708 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2709 }
2710 else {
01ea242b 2711 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2712 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2713 || UNLIKELY(low > LAST_LOW_SURROGATE))
2714 {
01ea242b 2715 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2716 }
2717 p += 2;
46956fad 2718 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2719 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2720 }
a0ed51b3 2721 }
56d37426
KW
2722#ifdef EBCDIC
2723 d = uvoffuni_to_utf8_flags(d, uv, 0);
2724#else
ffd0a9d3 2725 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2726 *d++ = (U8)(( uv >> 12) | 0xe0);
2727 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2728 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2729 continue;
2730 }
2731 else {
eb160463
GS
2732 *d++ = (U8)(( uv >> 18) | 0xf0);
2733 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2734 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2735 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2736 continue;
2737 }
56d37426 2738#endif
a0ed51b3 2739 }
dea0fc0b 2740 *newlen = d - dstart;
a0ed51b3
LW
2741 return d;
2742}
2743
2744/* Note: this one is slightly destructive of the source. */
2745
2746U8*
dea0fc0b 2747Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2748{
2749 U8* s = (U8*)p;
d4c19fe8 2750 U8* const send = s + bytelen;
7918f24d
NC
2751
2752 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2753
e0ea5e2d 2754 if (bytelen & 1)
147e3846 2755 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2756 (UV)bytelen);
2757
a0ed51b3 2758 while (s < send) {
d4c19fe8 2759 const U8 tmp = s[0];
a0ed51b3
LW
2760 s[0] = s[1];
2761 s[1] = tmp;
2762 s += 2;
2763 }
dea0fc0b 2764 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2765}
2766
922e8cb4
KW
2767bool
2768Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2769{
dc31b55c 2770 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2771}
2772
f9ae8fb6
JD
2773/* Internal function so we can deprecate the external one, and call
2774 this one from other deprecated functions in this file */
2775
f2645549
KW
2776bool
2777Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2778{
f2645549 2779 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2780
2781 if (*p == '_')
2782 return TRUE;
af0d7385
KW
2783 return is_utf8_common(p, NULL,
2784 "This is buggy if this gets used",
2785 PL_utf8_idstart);
61b19385
KW
2786}
2787
5092f92a 2788bool
eba68aa0
KW
2789Perl__is_uni_perl_idcont(pTHX_ UV c)
2790{
c12658c9 2791 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2792}
2793
2794bool
f91dcd13
KW
2795Perl__is_uni_perl_idstart(pTHX_ UV c)
2796{
c12658c9 2797 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2798}
2799
3a4c58c9 2800UV
56576a04
KW
2801Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2802 const char S_or_s)
3a4c58c9
KW
2803{
2804 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2805 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2806 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2807 * either "SS" or "Ss". Which one to use is passed into the routine in
2808 * 'S_or_s' to avoid a test */
2809
2810 UV converted = toUPPER_LATIN1_MOD(c);
2811
2812 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2813
2814 assert(S_or_s == 'S' || S_or_s == 's');
2815
6f2d5cbc 2816 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2817 characters in this range */
3a4c58c9
KW
2818 *p = (U8) converted;
2819 *lenp = 1;
2820 return converted;
2821 }
2822
2823 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2824 * which it maps to one of them, so as to only have to have one check for
2825 * it in the main case */
2826 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2827 switch (c) {
2828 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2829 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2830 break;
2831 case MICRO_SIGN:
2832 converted = GREEK_CAPITAL_LETTER_MU;
2833 break;
79e064b9
KW
2834#if UNICODE_MAJOR_VERSION > 2 \
2835 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2836 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2837 case LATIN_SMALL_LETTER_SHARP_S:
2838 *(p)++ = 'S';
2839 *p = S_or_s;
2840 *lenp = 2;
2841 return 'S';
79e064b9 2842#endif
3a4c58c9 2843 default:
56576a04
KW
2844 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2845 " '%c' to map to '%c'",
2846 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2847 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2848 }
2849 }
2850
2851 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2852 *p = UTF8_TWO_BYTE_LO(converted);
2853 *lenp = 2;
2854
2855 return converted;
2856}
2857
fe63c520
KW
2858/* If compiled on an early Unicode version, there may not be auxiliary tables
2859 * */
2860#ifndef HAS_UC_AUX_TABLES
2861# define UC_AUX_TABLE_ptrs NULL
2862# define UC_AUX_TABLE_lengths NULL
2863#endif
2864#ifndef HAS_TC_AUX_TABLES
2865# define TC_AUX_TABLE_ptrs NULL
2866# define TC_AUX_TABLE_lengths NULL
2867#endif
2868#ifndef HAS_LC_AUX_TABLES
2869# define LC_AUX_TABLE_ptrs NULL
2870# define LC_AUX_TABLE_lengths NULL
2871#endif
2872#ifndef HAS_CF_AUX_TABLES
2873# define CF_AUX_TABLE_ptrs NULL
2874# define CF_AUX_TABLE_lengths NULL
2875#endif
2876#ifndef HAS_UC_AUX_TABLES
2877# define UC_AUX_TABLE_ptrs NULL
2878# define UC_AUX_TABLE_lengths NULL
2879#endif
2880
50bda2c3
KW
2881/* Call the function to convert a UTF-8 encoded character to the specified case.
2882 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2883 * 's' is a pointer to the first byte of the input character
2884 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2885 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2886 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2887 *
56576a04 2888 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2889 * 'd' */
56576a04 2890#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2891 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2892 Uppercase_Mapping_invmap, \
2893 UC_AUX_TABLE_ptrs, \
2894 UC_AUX_TABLE_lengths, \
2895 "uppercase")
56576a04 2896#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2897 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2898 Titlecase_Mapping_invmap, \
2899 TC_AUX_TABLE_ptrs, \
2900 TC_AUX_TABLE_lengths, \
2901 "titlecase")
56576a04 2902#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2903 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2904 Lowercase_Mapping_invmap, \
2905 LC_AUX_TABLE_ptrs, \
2906 LC_AUX_TABLE_lengths, \
2907 "lowercase")
2908
50bda2c3 2909
b9992569
KW
2910/* This additionally has the input parameter 'specials', which if non-zero will
2911 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2912 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2913#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2914 (specials) \
2915 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2916 Case_Folding_invmap, \
2917 CF_AUX_TABLE_ptrs, \
2918 CF_AUX_TABLE_lengths, \
2919 "foldcase") \
2920 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2921 Simple_Case_Folding_invmap, \
2922 NULL, NULL, \
2923 "foldcase")
c3fd2246 2924
84afefe6
JH
2925UV
2926Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2927{
a1433954
KW
2928 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2929 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2930 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2931 * the changed version may be longer than the original character.
2932 *
2933 * The ordinal of the first character of the changed version is returned
2934 * (but note, as explained above, that there may be more.) */
2935
7918f24d
NC
2936 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2937
3a4c58c9
KW
2938 if (c < 256) {
2939 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2940 }
2941
0ebc6274 2942 uvchr_to_utf8(p, c);
b9992569 2943 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2944}
2945
84afefe6
JH
2946UV
2947Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2948{
7918f24d
NC
2949 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2950
3a4c58c9
KW
2951 if (c < 256) {
2952 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2953 }
2954
0ebc6274 2955 uvchr_to_utf8(p, c);
b9992569 2956 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2957}
2958
afc16117 2959STATIC U8
eaf412bf 2960S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2961{
2962 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2963 * those, converting the result to UTF-8. Since the result is always just
a1433954 2964 * one character, we allow <p> to be NULL */
afc16117
KW
2965
2966 U8 converted = toLOWER_LATIN1(c);
2967
eaf412bf
KW
2968 PERL_UNUSED_ARG(dummy);
2969
afc16117 2970 if (p != NULL) {
6f2d5cbc 2971 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2972 *p = converted;
2973 *lenp = 1;
2974 }
2975 else {
430c9760
KW
2976 /* Result is known to always be < 256, so can use the EIGHT_BIT
2977 * macros */
2978 *p = UTF8_EIGHT_BIT_HI(converted);
2979 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2980 *lenp = 2;
2981 }
2982 }
2983 return converted;
2984}
2985
84afefe6
JH
2986UV
2987Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2988{
7918f24d
NC
2989 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2990
afc16117 2991 if (c < 256) {
eaf412bf 2992 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2993 }
2994
afc16117 2995 uvchr_to_utf8(p, c);
b9992569 2996 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2997}
2998
84afefe6 2999UV
7c0ab950 3000Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3001{
51910141 3002 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3003 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3004 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3005 *
3006 * Not to be used for locale folds
51910141 3007 */
f673fad4 3008
a1dde8de
KW
3009 UV converted;
3010
3011 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3012
1ca267a5
KW
3013 assert (! (flags & FOLD_FLAGS_LOCALE));
3014
659a7c2d 3015 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3016 converted = GREEK_SMALL_LETTER_MU;
3017 }
9b63e895
KW
3018#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3019 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3020 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3021 else if ( (flags & FOLD_FLAGS_FULL)
3022 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3023 {
1ca267a5
KW
3024 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3025 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3026 * under those circumstances. */
3027 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3028 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3029 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3030 p, *lenp, U8);
3031 return LATIN_SMALL_LETTER_LONG_S;
3032 }
3033 else {
4f489194
KW
3034 *(p)++ = 's';
3035 *p = 's';
3036 *lenp = 2;
3037 return 's';
1ca267a5 3038 }
a1dde8de 3039 }
9b63e895 3040#endif
a1dde8de
KW
3041 else { /* In this range the fold of all other characters is their lower
3042 case */
3043 converted = toLOWER_LATIN1(c);
3044 }
3045
6f2d5cbc 3046 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3047 *p = (U8) converted;
3048 *lenp = 1;
3049 }
3050 else {
3051 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3052 *p = UTF8_TWO_BYTE_LO(converted);
3053 *lenp = 2;
3054 }
3055
3056 return converted;
3057}
3058
3059UV
31f05a37 3060Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3061{
4b593389 3062
a0270393
KW
3063 /* Not currently externally documented, and subject to change
3064 * <flags> bits meanings:
3065 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3066 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3067 * locale are to be used.
a0270393
KW
3068 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3069 */
4b593389 3070
36bb2ab6 3071 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3072
780fcc9f 3073 if (flags & FOLD_FLAGS_LOCALE) {
8b7358b9
KW
3074 /* Treat a UTF-8 locale as not being in locale at all, except for
3075 * potentially warning */
3076 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
780fcc9f
KW
3077 if (IN_UTF8_CTYPE_LOCALE) {
3078 flags &= ~FOLD_FLAGS_LOCALE;
3079 }
3080 else {
e7b7ac46 3081 goto needs_full_generality;
780fcc9f 3082 }
31f05a37
KW
3083 }
3084
a1dde8de 3085 if (c < 256) {
e7b7ac46 3086 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3087 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3088 }
3089
2f306ab9 3090 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
3091 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3092 uvchr_to_utf8(p, c);
b9992569 3093 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3094 }
567b353c 3095 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3096 the special flags. */
3097 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3098
3099 needs_full_generality:
a0270393 3100 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3101 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3102 p, lenp, flags);
a0270393 3103 }
84afefe6
JH
3104}
3105
26483009 3106PERL_STATIC_INLINE bool
5141f98e 3107S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 3108 const char *const swashname, SV* const invlist)
bde6a22d 3109{
ea317ccb
KW
3110 /* returns a boolean giving whether or not the UTF8-encoded character that
3111 * starts at <p> is in the swash indicated by <swashname>. <swash>
3112 * contains a pointer to where the swash indicated by <swashname>
3113 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
3114 * look at <*swash> and only generate a swash if it is not null. <invlist>
3115 * is NULL or an inversion list that defines the swash. If not null, it
3116 * saves time during initialization of the swash.
ea317ccb
KW
3117 *
3118 * Note that it is assumed that the buffer length of <p> is enough to
3119 * contain all the bytes that comprise the character. Thus, <*p> should
3120 * have been checked before this call for mal-formedness enough to assure
3121 * that. */
3122
7918f24d
NC
3123 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3124
492a624f 3125 /* The API should have included a length for the UTF-8 character in <p>,
28123549 3126 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
3127 * as far as there being enough bytes available in it to accommodate the
3128 * character without reading beyond the end, and pass that number on to the
3129 * validating routine */
6302f837 3130 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 3131 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 3132 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
3133 1 /* Die */ );
3134 NOT_REACHED; /* NOTREACHED */
28123549 3135 }
86ae6e94 3136
dc31b55c
KW
3137 if (invlist) {
3138 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3139 }
3140
af0d7385
KW
3141 assert(swash);
3142
87367d5f
KW
3143 if (!*swash) {
3144 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
3145 *swash = _core_swash_init("utf8",
3146
3147 /* Only use the name if there is no inversion
3148 * list; otherwise will go out to disk */
3149 (invlist) ? "" : swashname,
3150
3151 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 3152 }
28123549 3153
bde6a22d
NC
3154 return swash_fetch(*swash, p, TRUE) != 0;
3155}
3156
da8c1a98 3157PERL_STATIC_INLINE bool
56576a04
KW
3158S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3159 SV **swash, const char *const swashname,
3160 SV* const invlist)
da8c1a98
KW
3161{
3162 /* returns a boolean giving whether or not the UTF8-encoded character that
3163 * starts at <p>, and extending no further than <e - 1> is in the swash
3164 * indicated by <swashname>. <swash> contains a pointer to where the swash
3165 * indicated by <swashname> is to be stored; which this routine will do, so
3166 * that future calls will look at <*swash> and only generate a swash if it
3167 * is not null. <invlist> is NULL or an inversion list that defines the
3168 * swash. If not null, it saves time during initialization of the swash.
3169 */
3170
3171 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3172
3173 if (! isUTF8_CHAR(p, e)) {
3174 _force_out_malformed_utf8_message(p, e, 0, 1);
3175 NOT_REACHED; /* NOTREACHED */
3176 }
3177
dc31b55c
KW
3178 if (invlist) {
3179 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3180 }
3181
af0d7385
KW
3182 assert(swash);
3183
da8c1a98
KW
3184 if (!*swash) {
3185 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3186 *swash = _core_swash_init("utf8",
3187
3188 /* Only use the name if there is no inversion
3189 * list; otherwise will go out to disk */
3190 (invlist) ? "" : swashname,
3191
3192 &PL_sv_undef, 1, 0, invlist, &flags);
3193 }
3194
3195 return swash_fetch(*swash, p, TRUE) != 0;
3196}
3197
34aeb2e9
KW
3198STATIC void
3199S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3200 const char * const alternative,
3201 const bool use_locale,
3202 const char * const file,
3203 const unsigned line)
3204{
3205 const char * key;
3206
3207 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3208
3209 if (ckWARN_d(WARN_DEPRECATED)) {
3210
3211 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3212 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3213 if (! PL_seen_deprecated_macro) {
3214 PL_seen_deprecated_macro = newHV();
3215 }
3216 if (! hv_store(PL_seen_deprecated_macro, key,
3217 strlen(key), &PL_sv_undef, 0))
3218 {
3219 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3220 }
3221
c44e9413 3222 if (instr(file, "mathoms.c")) {
607313a1
KW
3223 Perl_warner(aTHX_ WARN_DEPRECATED,
3224 "In %s, line %d, starting in Perl v5.30, %s()"
3225 " will be removed. Avoid this message by"
3226 " converting to use %s().\n",
3227 file, line, name, alternative);
3228 }
3229 else {
34aeb2e9
KW
3230 Perl_warner(aTHX_ WARN_DEPRECATED,
3231 "In %s, line %d, starting in Perl v5.30, %s() will"
3232 " require an additional parameter. Avoid this"
3233 " message by converting to use %s().\n",
3234 file, line, name, alternative);
607313a1 3235 }
34aeb2e9
KW
3236 }
3237 }
3238}
3239
bde6a22d 3240bool
34aeb2e9 3241Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3242 const U8 * const p,
34aeb2e9
KW
3243 const char * const name,
3244 const char * const alternative,
3245 const bool use_utf8,
3246 const bool use_locale,
3247 const char * const file,
3248 const unsigned line)
922e8cb4 3249{
922e8cb4
KW
3250 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3251
34aeb2e9
KW
3252 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3253
3254 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3255
3256 switch (classnum) {
3257 case _CC_WORDCHAR:
3258 case _CC_DIGIT:
3259 case _CC_ALPHA:
3260 case _CC_LOWER:
3261 case _CC_UPPER:
3262 case _CC_PUNCT:
3263 case _CC_PRINT:
3264 case _CC_ALPHANUMERIC:
3265 case _CC_GRAPH:
3266 case _CC_CASED:
3267
3268 return is_utf8_common(p,
af0d7385
KW
3269 NULL,
3270 "This is buggy if this gets used",
34aeb2e9
KW
3271 PL_XPosix_ptrs[classnum]);
3272
3273 case _CC_SPACE:
3274 return is_XPERLSPACE_high(p);
3275 case _CC_BLANK:
3276 return is_HORIZWS_high(p);
3277 case _CC_XDIGIT:
3278 return is_XDIGIT_high(p);
3279 case _CC_CNTRL:
3280 return 0;
3281 case _CC_ASCII:
3282 return 0;
3283 case _CC_VERTSPACE:
3284 return is_VERTWS_high(p);
3285 case _CC_IDFIRST:
af0d7385
KW
3286 return is_utf8_common(p, NULL,
3287 "This is buggy if this gets used",
3288 PL_utf8_perl_idstart);
34aeb2e9 3289 case _CC_IDCONT:
af0d7385
KW
3290 return is_utf8_common(p, NULL,
3291 "This is buggy if this gets used",
3292 PL_utf8_perl_idcont);
34aeb2e9
KW
3293 }
3294 }
3295
3296 /* idcont is the same as wordchar below 256 */
3297 if (classnum == _CC_IDCONT) {
3298 classnum = _CC_WORDCHAR;
3299 }
3300 else if (classnum == _CC_IDFIRST) {
3301 if (*p == '_') {
3302 return TRUE;
3303 }
3304 classnum = _CC_ALPHA;
3305 }
3306
3307 if (! use_locale) {
3308 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3309 return _generic_isCC(*p, classnum);
3310 }
922e8cb4 3311
34aeb2e9
KW
3312 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3313 }
3314 else {
3315 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3316 return isFOO_lc(classnum, *p);
3317 }
3318
3319 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3320 }
3321
3322 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3323}
3324
3325bool
da8c1a98
KW
3326Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3327 const U8 * const e)
3328{
3329 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3330
af0d7385
KW
3331 return is_utf8_common_with_len(p, e, NULL,
3332 "This is buggy if this gets used",
da8c1a98
KW
3333 PL_XPosix_ptrs[classnum]);
3334}
3335
3336bool
da8c1a98
KW
3337Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3338{
da8c1a98
KW
3339 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3340
af0d7385
KW
3341 return is_utf8_common_with_len(p, e, NULL,
3342 "This is buggy if this gets used",
3343 PL_utf8_perl_idstart);
da8c1a98
KW
3344}
3345
3346bool
f2645549 3347Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3348{
f2645549 3349 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3350
3351 if (*p == '_')
3352 return TRUE;
f25ce844 3353 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3354}
3355
3356bool
da8c1a98
KW
3357Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3358{
da8c1a98
KW
3359 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3360
af0d7385
KW
3361 return is_utf8_common_with_len(p, e, NULL,
3362 "This is buggy if this gets used",
3363 PL_utf8_perl_idcont);
da8c1a98
KW
3364}
3365
3366bool
f2645549 3367Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3368{
f2645549 3369 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3370
f25ce844 3371 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3372}
3373
3374bool
f2645549 3375Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3376{
f2645549 3377 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3378
5062866a 3379 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
c11ff943
KW
3380}
3381
3382bool
7dbf68d2
KW
3383Perl__is_utf8_mark(pTHX_ const U8 *p)
3384{
7dbf68d2
KW
3385 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3386
f25ce844 3387 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3388}
3389
6a4a25f4 3390STATIC UV
30613bdc
KW
3391S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3392 U8* ustrp, STRLEN *lenp,
341bb5b7 3393 SV *invlist, const int * const invmap,
e39a4130 3394 const unsigned int * const * const aux_tables,
30613bdc
KW
3395 const U8 * const aux_table_lengths,
3396 const char * const normal)
b9992569 3397{
0134edef 3398 STRLEN len = 0;
7918f24d 3399
30613bdc
KW
3400 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3401 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3402 * to name the new case in any generated messages, as a fallback if the
3403 * operation being used is not available. The new case is given by the
3404 * data structures in the remaining arguments.
3405 *
3406 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3407 * entire changed case string, and the return value is the first code point
3408 * in that string */
3409
b9992569 3410 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3411
36eaa811
KW
3412 /* For code points that don't change case, we already know that the output
3413 * of this function is the unchanged input, so we can skip doing look-ups
3414 * for them. Unfortunately the case-changing code points are scattered
3415 * around. But there are some long consecutive ranges where there are no
3416 * case changing code points. By adding tests, we can eliminate the lookup
3417 * for all the ones in such ranges. This is currently done here only for
3418 * just a few cases where the scripts are in common use in modern commerce
3419 * (and scripts adjacent to those which can be included without additional
3420 * tests). */
3421
3422 if (uv1 >= 0x0590) {
3423 /* This keeps from needing further processing the code points most
3424 * likely to be used in the following non-cased scripts: Hebrew,
3425 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3426 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3427 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3428 if (uv1 < 0x10A0) {
3429 goto cases_to_self;
3430 }
3431
3432 /* The following largish code point ranges also don't have case
3433 * changes, but khw didn't think they warranted extra tests to speed
3434 * them up (which would slightly slow down everything else above them):
3435 * 1100..139F Hangul Jamo, Ethiopic
3436 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3437 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3438 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3439 * Combining Diacritical Marks Extended, Balinese,
3440 * Sundanese, Batak, Lepcha, Ol Chiki
3441 * 2000..206F General Punctuation
3442 */
3443
3444 if (uv1 >= 0x2D30) {
3445
3446 /* This keeps the from needing further processing the code points
3447 * most likely to be used in the following non-cased major scripts:
3448 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3449 *
3450 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3451 * event that Unicode eventually allocates the unused block as of
3452 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3453 * that the test suite will start having failures to alert you
3454 * should that happen) */
3455 if (uv1 < 0xA640) {
3456 goto cases_to_self;
3457 }
3458
3459 if (uv1 >= 0xAC00) {
3460 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3461 if (ckWARN_d(WARN_SURROGATE)) {
3462 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3463 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3464 "Operation \"%s\" returns its argument for"
3465 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3466 }
3467 goto cases_to_self;
3468 }
36eaa811
KW
3469
3470 /* AC00..FAFF Catches Hangul syllables and private use, plus
3471 * some others */
3472 if (uv1 < 0xFB00) {
3473 goto cases_to_self;
36eaa811
KW
3474 }
3475
5af9bc97 3476 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3477 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3478 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3479 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3480 }
3481 if (ckWARN_d(WARN_NON_UNICODE)) {
3482 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3483 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3484 "Operation \"%s\" returns its argument for"
3485 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3486 }
3487 goto cases_to_self;
3488 }
3bfc1e70
KW
3489#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3490 if (UNLIKELY(uv1
3491 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3492 {
3493
56576a04
KW
3494 /* As of Unicode 10.0, this means we avoid swash creation
3495 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3496 goto cases_to_self;
3497 }
3498#endif
36eaa811
KW
3499 }
3500 }
9ae3ac1a 3501
36eaa811 3502 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3503 * be given. */
9ae3ac1a
KW
3504 }
3505
8946fcd9
KW
3506 {
3507 unsigned int i;
e39a4130 3508 const unsigned int * cp_list;
8946fcd9
KW
3509 U8 * d;
3510 SSize_t index = _invlist_search(invlist, uv1);
3511 IV base = invmap[index];
0134edef 3512
30613bdc
KW
3513 /* The data structures are set up so that if 'base' is non-negative,
3514 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3515 if (base >= 0) {
3516 IV lc;
b08cf34e 3517
8946fcd9
KW
3518 if (base == 0) {
3519 goto cases_to_self;
4a8240a3 3520 }
4a8240a3 3521
30613bdc 3522 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3523 lc = base + uv1 - invlist_array(invlist)[index];
3524 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3525 return lc;
3526 }
1feea2c7 3527
30613bdc
KW
3528 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3529 * requires an auxiliary table look up. abs(base) gives the index into
3530 * a list of such tables which points to the proper aux table. And a
3531 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3532 cp_list = aux_tables[-base];
30613bdc
KW
3533
3534 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3535 d = ustrp;
3536 for (i = 0; i < aux_table_lengths[-base]; i++) {
3537 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3538 }
8946fcd9
KW
3539 *d = '\0';
3540 *lenp = d - ustrp;
3541
3542 return cp_list[0];
cbe07460
KW
3543 }
3544
3545 /* Here, there was no mapping defined, which means that the code point maps
3546 * to itself. Return the inputs */
e24dfe9c 3547 cases_to_self:
bfdf22ec 3548 len = UTF8SKIP(p);
ca9fab46
KW
3549 if (p != ustrp) { /* Don't copy onto itself */
3550 Copy(p, ustrp, len, U8);
3551 }
0134edef 3552
2a37f04d
JH
3553 if (lenp)
3554 *lenp = len;
3555
f4cd282c 3556 return uv1;
cbe07460 3557
a0ed51b3
LW
3558}
3559
b74fe592 3560Size_t
e39a4130
KW
3561Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3562 const unsigned int ** remaining_folds_to)
b74fe592
KW
3563{
3564 /* Returns the count of the number of code points that fold to the input
3565 * 'cp' (besides itself).
3566 *
3567 * If the return is 0, there is nothing else that folds to it, and
3568 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3569 *
3570 * If the return is 1, '*first_folds_to' is set to the single code point,
3571 * and '*remaining_folds_to' is set to NULL.
3572 *
3573 * Otherwise, '*first_folds_to' is set to a code point, and
3574 * '*remaining_fold_to' is set to an array that contains the others. The
3575 * length of this array is the returned count minus 1.
3576 *
3577 * The reason for this convolution is to avoid having to deal with
3578 * allocating and freeing memory. The lists are already constructed, so
3579 * the return can point to them, but single code points aren't, so would
3580 * need to be constructed if we didn't employ something like this API */
3581
3582 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3583 int base = _Perl_IVCF_invmap[index];
3584
3585 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3586
3587 if (base == 0) { /* No fold */
3588 *first_folds_to = 0;
3589 *remaining_folds_to = NULL;
3590 return 0;
3591 }
3592
3593#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3594
3595 assert(base > 0);
3596
3597#else
3598
3599 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3600
3601 /* The data structure is set up so that the absolute value of 'base' is
3602 * an index into a table of pointers to arrays, with the array
3603 * corresponding to the index being the list of code points that fold
3604 * to 'cp', and the parallel array containing the length of the list
3605 * array */
3606 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3607 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3608 *first_folds_to
3609 */
3610 return IVCF_AUX_TABLE_lengths[-base];
3611 }
3612
3613#endif
3614
3615 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3616 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3617 *remaining_folds_to = NULL;
3618 return 1;
3619}
3620
051a06d4 3621STATIC UV
56576a04
KW
3622S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3623 U8* const ustrp, STRLEN *lenp)
051a06d4 3624{
4a4088c4 3625 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3626 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3627 * result contains a character that crosses the 255/256 boundary, disallow
3628 * the change, and return the original code point. See L<perlfunc/lc> for
3629 * why;
051a06d4 3630 *
a1433954
KW
3631 * p points to the original string whose case was changed; assumed
3632 * by this routine to be well-formed
051a06d4 3633 * result the code point of the first character in the changed-case string
56576a04
KW
3634 * ustrp points to the changed-case string (<result> represents its
3635 * first char)
051a06d4
KW
3636 * lenp points to the length of <ustrp> */
3637
3638 UV original; /* To store the first code point of <p> */
3639
3640 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3641
a4f12ed7 3642 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3643
3644 /* We know immediately if the first character in the string crosses the
5e45c680 3645 * boundary, so can skip testing */
051a06d4
KW
3646 if (result > 255) {
3647
3648 /* Look at every character in the result; if any cross the
3649 * boundary, the whole thing is disallowed */
3650 U8* s = ustrp + UTF8SKIP(ustrp);
3651 U8* e = ustrp + *lenp;
3652 while (s < e) {
a4f12ed7 3653 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3654 goto bad_crossing;
3655 }
3656 s += UTF8SKIP(s);
3657 }
3658
613abc6d
KW
3659 /* Here, no characters crossed, result is ok as-is, but we warn. */
3660 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3661 return result;
3662 }
3663
7b52d656 3664 bad_crossing:
051a06d4
KW
3665
3666 /* Failed, have to return the original */
4b88fb76 3667 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3668
3669 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3670 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3671 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3672 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3673 OP_DESC(PL_op),
ab0b796c
KW
3674 original,
3675 original);
051a06d4
KW
3676 Copy(p, ustrp, *lenp, char);
3677 return original;
3678}
3679
607313a1
KW
3680STATIC U32
3681S_check_and_deprecate(pTHX_ const U8 *p,
3682 const U8 **e,
3683 const unsigned int type, /* See below */
3684 const bool use_locale, /* Is this a 'LC_'
3685 macro call? */
3686 const char * const file,
3687 const unsigned line)
3688{
3689 /* This is a temporary function to deprecate the unsafe calls to the case
3690 * changing macros and functions. It keeps all the special stuff in just
3691 * one place.
3692 *
3693 * It updates *e with the pointer to the end of the input string. If using
3694 * the old-style macros, *e is NULL on input, and so this function assumes
3695 * the input string is long enough to hold the entire UTF-8 sequence, and
3696 * sets *e accordingly, but it then returns a flag to pass the
3697 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3698 * using the full length if possible.
3699 *
3700 * It also does the assert that *e > p when *e is not NULL. This should be
3701 * migrated to the callers when this function gets deleted.
3702 *
3703 * The 'type' parameter is used for the caller to specify which case
3704 * changing function this is called from: */
3705
3706# define DEPRECATE_TO_UPPER 0
3707# define DEPRECATE_TO_TITLE 1
3708# define DEPRECATE_TO_LOWER 2
3709# define DEPRECATE_TO_FOLD 3
3710
3711 U32 utf8n_flags = 0;
3712 const char * name;
3713 const char * alternative;
3714
3715 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3716
3717 if (*e == NULL) {
3718 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3719 *e = p + UTF8SKIP(p);
3720
3721 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3722 * there. It could be part of a larger path */