This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Add Ken Brown to AUTHORS
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
d22ec717
KW
40 "Use of code point 0x%" UVXf " is not allowed; the"
41 " permissible max is 0x%" UVXf;
760c7c2f 42
d22ec717 43#define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
901b21bf 44
48ef279e 45/*
ccfc67b7 46=head1 Unicode Support
7fefc6c1 47These are various utility functions for manipulating UTF8-encoded
72d33970 48strings. For the uninitiated, this is a method of representing arbitrary
61296642 49Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
50characters in the ASCII range are unmodified, and a zero byte never appears
51within non-zero characters.
166f8a29 52
eaf7a4d2
CS
53=cut
54*/
55
9cbfb8ab
KW
56void
57Perl__force_out_malformed_utf8_message(pTHX_
58 const U8 *const p, /* First byte in UTF-8 sequence */
59 const U8 * const e, /* Final byte in sequence (may include
60 multiple chars */
61 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
62 usually 0, or some DISALLOW flags */
63 const bool die_here) /* If TRUE, this function does not return */
64{
65 /* This core-only function is to be called when a malformed UTF-8 character
66 * is found, in order to output the detailed information about the
67 * malformation before dieing. The reason it exists is for the occasions
68 * when such a malformation is fatal, but warnings might be turned off, so
69 * that normally they would not be actually output. This ensures that they
70 * do get output. Because a sequence may be malformed in more than one
71 * way, multiple messages may be generated, so we can't make them fatal, as
72 * that would cause the first one to die.
73 *
74 * Instead we pretend -W was passed to perl, then die afterwards. The
75 * flexibility is here to return to the caller so they can finish up and
76 * die themselves */
77 U32 errors;
78
79 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
80
81 ENTER;
c15a80f3 82 SAVEI8(PL_dowarn);
9cbfb8ab
KW
83 SAVESPTR(PL_curcop);
84
85 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
86 if (PL_curcop) {
87 PL_curcop->cop_warnings = pWARN_ALL;
88 }
89
90 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
91
92 LEAVE;
93
94 if (! errors) {
95 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
96 " be called only when there are errors found");
97 }
98
99 if (die_here) {
100 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
101 }
102}
103
eaf7a4d2 104/*
378516de 105=for apidoc uvoffuni_to_utf8_flags
eebe1485 106
a27992cc 107THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
108Instead, B<Almost all code should use L</uvchr_to_utf8> or
109L</uvchr_to_utf8_flags>>.
a27992cc 110
de69f3af
KW
111This function is like them, but the input is a strict Unicode
112(as opposed to native) code point. Only in very rare circumstances should code
113not be using the native code point.
949cf498 114
efa9cd84 115For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 116
eebe1485
SC
117=cut
118*/
119
c94c2f39
KW
120/* All these formats take a single UV code point argument */
121const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
122const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
123 " is not recommended for open interchange";
124const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
125 " may not be portable";
57ff5f59
KW
126const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
127 " Unicode, requires a Perl extension," \
128 " and so is not portable";
c94c2f39 129
8ee1cdcb
KW
130#define HANDLE_UNICODE_SURROGATE(uv, flags) \
131 STMT_START { \
132 if (flags & UNICODE_WARN_SURROGATE) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
c94c2f39 134 surrogate_cp_format, uv); \
8ee1cdcb
KW
135 } \
136 if (flags & UNICODE_DISALLOW_SURROGATE) { \
137 return NULL; \
138 } \
139 } STMT_END;
140
141#define HANDLE_UNICODE_NONCHAR(uv, flags) \
142 STMT_START { \
143 if (flags & UNICODE_WARN_NONCHAR) { \
144 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
c94c2f39 145 nonchar_cp_format, uv); \
8ee1cdcb
KW
146 } \
147 if (flags & UNICODE_DISALLOW_NONCHAR) { \
148 return NULL; \
149 } \
150 } STMT_END;
151
ba6ed43c
KW
152/* Use shorter names internally in this file */
153#define SHIFT UTF_ACCUMULATION_SHIFT
154#undef MARK
155#define MARK UTF_CONTINUATION_MARK
156#define MASK UTF_CONTINUATION_MASK
157
dfe13c55 158U8 *
4b31b634 159Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
a0ed51b3 160{
378516de 161 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 162
2d1545e5 163 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 164 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
165 return d;
166 }
facc1dc2 167
3ea68d71 168 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
169 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
170 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
171 return d;
172 }
d9432125 173
ba6ed43c
KW
174 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
175 * below, the 16 is for start bytes E0-EF (which are all the possible ones
176 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
177 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
178 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
179 * 0x800-0xFFFF on ASCII */
180 if (uv < (16 * (1U << (2 * SHIFT)))) {
181 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
182 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
183 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
184
185#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
186 aren't tested here */
187 /* The most likely code points in this range are below the surrogates.
188 * Do an extra test to quickly exclude those. */
189 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
190 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
191 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
192 {
8ee1cdcb
KW
193 HANDLE_UNICODE_NONCHAR(uv, flags);
194 }
195 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
196 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 197 }
ba6ed43c
KW
198 }
199#endif
200 return d;
201 }
202
203 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
204 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
205 * happen starting with 4-byte characters on ASCII platforms. We unify the
206 * code for these with EBCDIC, even though some of them require 5-bytes on
207 * those, because khw believes the code saving is worth the very slight
208 * performance hit on these high EBCDIC code points. */
209
210 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
d22ec717
KW
211 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
212 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
a5bf80e0 213 }
0a8a1a5b
KW
214 if ( (flags & UNICODE_WARN_SUPER)
215 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
216 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
217 {
218 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
219
220 /* Choose the more dire applicable warning */
d044b7a7 221 (UNICODE_IS_PERL_EXTENDED(uv))
57ff5f59 222 ? perl_extended_cp_format
c94c2f39 223 : super_cp_format,
a5bf80e0
KW
224 uv);
225 }
56576a04 226 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
227 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
228 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
229 {
230 return NULL;
231 }
232 }
ba6ed43c
KW
233 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
234 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 235 }
d9432125 236
ba6ed43c
KW
237 /* Test for and handle 4-byte result. In the test immediately below, the
238 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
239 * characters). The 3 is for 3 continuation bytes; these each contribute
240 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
241 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
242 * 0x1_0000-0x1F_FFFF on ASCII */
243 if (uv < (8 * (1U << (3 * SHIFT)))) {
244 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
245 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
246 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
247 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
248
249#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
250 characters. The end-plane non-characters for EBCDIC were
251 handled just above */
252 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
253 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 254 }
ba6ed43c
KW
255 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
256 HANDLE_UNICODE_SURROGATE(uv, flags);
257 }
258#endif
259
260 return d;
261 }
262
263 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
264 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
265 * format. The unrolled version above turns out to not save all that much
266 * time, and at these high code points (well above the legal Unicode range
267 * on ASCII platforms, and well above anything in common use in EBCDIC),
268 * khw believes that less code outweighs slight performance gains. */
269
d9432125 270 {
5aaebcb3 271 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
272 U8 *p = d+len-1;
273 while (p > d) {
4c8cd605 274 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
275 uv >>= UTF_ACCUMULATION_SHIFT;
276 }
4c8cd605 277 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
278 return d+len;
279 }
a0ed51b3 280}
a5bf80e0 281
646ca15d 282/*
07693fe6
KW
283=for apidoc uvchr_to_utf8
284
bcb1a2d4 285Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 286of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
287C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
288the byte after the end of the new character. In other words,
07693fe6
KW
289
290 d = uvchr_to_utf8(d, uv);
291
292is the recommended wide native character-aware way of saying
293
294 *(d++) = uv;
295
d22ec717
KW
296This function accepts any code point from 0..C<IV_MAX> as input.
297C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
298
299It is possible to forbid or warn on non-Unicode code points, or those that may
300be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 301
07693fe6
KW
302=cut
303*/
304
de69f3af
KW
305/* This is also a macro */
306PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
307
07693fe6
KW
308U8 *
309Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
310{
de69f3af 311 return uvchr_to_utf8(d, uv);
07693fe6
KW
312}
313
de69f3af
KW
314/*
315=for apidoc uvchr_to_utf8_flags
316
317Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 318of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
319C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
320the byte after the end of the new character. In other words,
de69f3af
KW
321
322 d = uvchr_to_utf8_flags(d, uv, flags);
323
324or, in most cases,
325
326 d = uvchr_to_utf8_flags(d, uv, 0);
327
328This is the Unicode-aware way of saying
329
330 *(d++) = uv;
331
d22ec717
KW
332If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
333input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
334
335Specifying C<flags> can further restrict what is allowed and not warned on, as
336follows:
de69f3af 337
796b6530 338If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
339the function will raise a warning, provided UTF8 warnings are enabled. If
340instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
341NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 342
760c7c2f
KW
343Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
344affect how the function handles a Unicode non-character.
93e6dbd6 345
760c7c2f
KW
346And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
347affect the handling of code points that are above the Unicode maximum of
3480x10FFFF. Languages other than Perl may not be able to accept files that
349contain these.
93e6dbd6
KW
350
351The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
352the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
353three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
354allowed inputs to the strict UTF-8 traditionally defined by Unicode.
355Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
356C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
357above-Unicode and surrogate flags, but not the non-character ones, as
358defined in
359L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
360See L<perlunicode/Noncharacter code points>.
93e6dbd6 361
57ff5f59
KW
362Extremely high code points were never specified in any standard, and require an
363extension to UTF-8 to express, which Perl does. It is likely that programs
364written in something other than Perl would not be able to read files that
365contain these; nor would Perl understand files written by something that uses a
366different extension. For these reasons, there is a separate set of flags that
367can warn and/or disallow these extremely high code points, even if other
368above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
369and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
370L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
371treat all above-Unicode code points, including these, as malformations. (Note
372that the Unicode standard considers anything above 0x10FFFF to be illegal, but
373there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
374
375A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
376retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
377C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
378C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because these
379flags can apply to code points that actually do fit in 31 bits. This happens
380on EBCDIC platforms, and sometimes when the L<overlong
381malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
382describe the situation in all cases.
de69f3af 383
de69f3af
KW
384=cut
385*/
386
387/* This is also a macro */
388PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
389
07693fe6
KW
390U8 *
391Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
392{
de69f3af 393 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
394}
395
57ff5f59
KW
396#ifndef UV_IS_QUAD
397
e050c007
KW
398STATIC int
399S_is_utf8_cp_above_31_bits(const U8 * const s,
400 const U8 * const e,
401 const bool consider_overlongs)
83dc0f42
KW
402{
403 /* Returns TRUE if the first code point represented by the Perl-extended-
404 * UTF-8-encoded string starting at 's', and looking no further than 'e -
405 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
406 *
407 * The function handles the case where the input bytes do not include all
408 * the ones necessary to represent a full character. That is, they may be
409 * the intial bytes of the representation of a code point, but possibly
410 * the final ones necessary for the complete representation may be beyond
411 * 'e - 1'.
412 *
e050c007
KW
413 * The function also can handle the case where the input is an overlong
414 * sequence. If 'consider_overlongs' is 0, the function assumes the
415 * input is not overlong, without checking, and will return based on that
416 * assumption. If this parameter is 1, the function will go to the trouble
417 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 418 *
e050c007 419 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
420 */
421
e050c007
KW
422 const STRLEN len = e - s;
423 int is_overlong;
424
425 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
426
427 assert(! UTF8_IS_INVARIANT(*s) && e > s);
428
83dc0f42
KW
429#ifdef EBCDIC
430
e050c007 431 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 432
e050c007
KW
433 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
434 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
435 * also be the start byte for a 31-bit code point; we need at least 2
436 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
437 * the start byte for an overlong sequence, but for 30-bit or smaller code
438 * points, so we don't have to worry about overlongs on EBCDIC.) */
439 if (*s != 0xFE) {
440 return 0;
441 }
83dc0f42 442
e050c007
KW
443 if (len == 1) {
444 return -1;
445 }
83dc0f42 446
e050c007 447#else
83dc0f42 448
e050c007
KW
449 /* On ASCII, FE and FF are the only start bytes that can evaluate to
450 * needing more than 31 bits. */
451 if (LIKELY(*s < 0xFE)) {
452 return 0;
453 }
83dc0f42 454
e050c007
KW
455 /* What we have left are FE and FF. Both of these require more than 31
456 * bits unless they are for overlongs. */
457 if (! consider_overlongs) {
458 return 1;
459 }
83dc0f42 460
e050c007
KW
461 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
462 * above 31 bits. But we need more than one byte to discern this, so if
463 * passed just the start byte, it could be an overlong evaluating to
464 * smaller */
465 if (len == 1) {
466 return -1;
467 }
83dc0f42 468
e050c007
KW
469 /* Having excluded len==1, and knowing that FE and FF are both valid start
470 * bytes, we can call the function below to see if the sequence is
471 * overlong. (We don't need the full generality of the called function,
472 * but for these huge code points, speed shouldn't be a consideration, and
473 * the compiler does have enough information, since it's static to this
474 * file, to optimize to just the needed parts.) */
475 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 476
e050c007
KW
477 /* If it isn't overlong, more than 31 bits are required. */
478 if (is_overlong == 0) {
479 return 1;
480 }
83dc0f42 481
e050c007
KW
482 /* If it is indeterminate if it is overlong, return that */
483 if (is_overlong < 0) {
484 return -1;
485 }
486
487 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
488 * the max it can be is 2**31 - 1 */
489 if (*s == 0xFE) {
490 return 0;
83dc0f42
KW
491 }
492
e050c007
KW
493#endif
494
495 /* Here, ASCII and EBCDIC rejoin:
496 * On ASCII: We have an overlong sequence starting with FF
497 * On EBCDIC: We have a sequence starting with FE. */
498
499 { /* For C89, use a block so the declaration can be close to its use */
500
501#ifdef EBCDIC
502
5f995336
KW
503 /* U+7FFFFFFF (2 ** 31 - 1)
504 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
505 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
506 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
507 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
508 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
509 * U+80000000 (2 ** 31):
510 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
511 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
512 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
513 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
514 *
515 * and since we know that *s = \xfe, any continuation sequcence
516 * following it that is gt the below is above 31 bits
517 [0] [1] [2] [3] [4] [5] [6] */
518 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
519
520#else
521
522 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
523 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
524 * FF overlong for U+80000000 (2 ** 31):
525 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
526 * and since we know that *s = \xff, any continuation sequcence
527 * following it that is gt the below is above 30 bits
528 [0] [1] [2] [3] [4] [5] [6] */
529 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 530
83dc0f42
KW
531
532#endif
e050c007
KW
533 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
534 const STRLEN cmp_len = MIN(conts_len, len - 1);
535
536 /* Now compare the continuation bytes in s with the ones we have
537 * compiled in that are for the largest 30 bit code point. If we have
538 * enough bytes available to determine the answer, or the bytes we do
539 * have differ from them, we can compare the two to get a definitive
540 * answer (Note that in UTF-EBCDIC, the two lowest possible
541 * continuation bytes are \x41 and \x42.) */
542 if (cmp_len >= conts_len || memNE(s + 1,
543 conts_for_highest_30_bit,
544 cmp_len))
545 {
546 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
547 }
83dc0f42 548
e050c007
KW
549 /* Here, all the bytes we have are the same as the highest 30-bit code
550 * point, but we are missing so many bytes that we can't make the
551 * determination */
552 return -1;
553 }
83dc0f42
KW
554}
555
57ff5f59
KW
556#endif
557
d6be65ae 558PERL_STATIC_INLINE int
12a4bed3
KW
559S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
560{
d6be65ae
KW
561 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
562 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
563 * it isn't, and -1 if there isn't enough information to tell. This last
564 * return value can happen if the sequence is incomplete, missing some
565 * trailing bytes that would form a complete character. If there are
566 * enough bytes to make a definitive decision, this function does so.
567 * Usually 2 bytes sufficient.
568 *
569 * Overlongs can occur whenever the number of continuation bytes changes.
570 * That means whenever the number of leading 1 bits in a start byte
571 * increases from the next lower start byte. That happens for start bytes
572 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
573 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
574 * ASCII platforms: C0, C1
575 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 576 */
12a4bed3
KW
577
578 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
579 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
580
581 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
582 assert(len > 1 && UTF8_IS_START(*s));
583
584 /* Each platform has overlongs after the start bytes given above (expressed
585 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
586 * the logic is the same, except the E0 overlong has already been excluded
587 * on EBCDIC platforms. The values below were found by manually
588 * inspecting the UTF-8 patterns. See the tables in utf8.h and
589 * utfebcdic.h. */
590
591# ifdef EBCDIC
592# define F0_ABOVE_OVERLONG 0xB0
593# define F8_ABOVE_OVERLONG 0xA8
594# define FC_ABOVE_OVERLONG 0xA4
595# define FE_ABOVE_OVERLONG 0xA2
596# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
597 /* I8(0xfe) is FF */
598# else
599
600 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 601 return 1;
12a4bed3
KW
602 }
603
604# define F0_ABOVE_OVERLONG 0x90
605# define F8_ABOVE_OVERLONG 0x88
606# define FC_ABOVE_OVERLONG 0x84
607# define FE_ABOVE_OVERLONG 0x82
608# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
609# endif
610
611
612 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
613 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
614 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
615 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
616 {
d6be65ae 617 return 1;
12a4bed3
KW
618 }
619
b0b342d4 620 /* Check for the FF overlong */
d6be65ae 621 return isFF_OVERLONG(s, len);
b0b342d4
KW
622}
623
8d6204cc 624PERL_STATIC_INLINE int
b0b342d4
KW
625S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
626{
8d6204cc
KW
627 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
628 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
629 * it isn't, and -1 if there isn't enough information to tell. This last
630 * return value can happen if the sequence is incomplete, missing some
631 * trailing bytes that would form a complete character. If there are
632 * enough bytes to make a definitive decision, this function does so. */
633
b0b342d4 634 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 635
8d6204cc
KW
636 /* To be an FF overlong, all the available bytes must match */
637 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
638 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
639 {
640 return 0;
641 }
642
643 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
644 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 645 * utfebcdic.h. */
8d6204cc
KW
646 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
647 return 1;
648 }
12a4bed3 649
8d6204cc
KW
650 /* The missing bytes could cause the result to go one way or the other, so
651 * the result is indeterminate */
652 return -1;
12a4bed3
KW
653}
654
d22ec717 655#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
656# ifdef EBCDIC /* Actually is I8 */
657# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 658 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
659# else
660# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 661 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
662# endif
663#endif
664
c285bbc4 665PERL_STATIC_INLINE int
e050c007
KW
666S_does_utf8_overflow(const U8 * const s,
667 const U8 * e,
668 const bool consider_overlongs)
a77c906e 669{
c285bbc4 670 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
671 * 'e' - 1 would overflow an IV on this platform; that is if it represents
672 * a code point larger than the highest representable code point. It
673 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
674 * enough information to tell. This last return value can happen if the
675 * sequence is incomplete, missing some trailing bytes that would form a
676 * complete character. If there are enough bytes to make a definitive
677 * decision, this function does so.
c285bbc4 678 *
e050c007
KW
679 * If 'consider_overlongs' is TRUE, the function checks for the possibility
680 * that the sequence is an overlong that doesn't overflow. Otherwise, it
681 * assumes the sequence is not an overlong. This can give different
682 * results only on ASCII 32-bit platforms.
683 *
c285bbc4
KW
684 * (For ASCII platforms, we could use memcmp() because we don't have to
685 * convert each byte to I8, but it's very rare input indeed that would
686 * approach overflow, so the loop below will likely only get executed once.)
687 *
688 * 'e' - 1 must not be beyond a full character. */
a77c906e 689
a77c906e
KW
690
691 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
692 assert(s <= e && s + UTF8SKIP(s) >= e);
693
d22ec717
KW
694#if ! defined(UV_IS_QUAD)
695
696 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
697
698#else
699
700 PERL_UNUSED_ARG(consider_overlongs);
701
702 {
703 const STRLEN len = e - s;
704 const U8 *x;
705 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
706
707 for (x = s; x < e; x++, y++) {
708
709 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
710 continue;
711 }
712
713 /* If this byte is larger than the corresponding highest UTF-8
714 * byte, the sequence overflow; otherwise the byte is less than,
715 * and so the sequence doesn't overflow */
716 return NATIVE_UTF8_TO_I8(*x) > *y;
717
718 }
719
720 /* Got to the end and all bytes are the same. If the input is a whole
721 * character, it doesn't overflow. And if it is a partial character,
722 * there's not enough information to tell */
723 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
724 return -1;
725 }
726
727 return 0;
728 }
729
730#endif
731
732}
733
734#if 0
735
736/* This is the portions of the above function that deal with UV_MAX instead of
737 * IV_MAX. They are left here in case we want to combine them so that internal
738 * uses can have larger code points. The only logic difference is that the
739 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
740 * different logic.
741 */
742
743/* Anything larger than this will overflow the word if it were converted into a UV */
744#if defined(UV_IS_QUAD)
745# ifdef EBCDIC /* Actually is I8 */
746# define HIGHEST_REPRESENTABLE_UTF8 \
747 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
748# else
749# define HIGHEST_REPRESENTABLE_UTF8 \
750 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
751# endif
752#else /* 32-bit */
753# ifdef EBCDIC
754# define HIGHEST_REPRESENTABLE_UTF8 \
755 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
756# else
757# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
758# endif
759#endif
760
a77c906e
KW
761#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
762
763 /* On 32 bit ASCII machines, many overlongs that start with FF don't
764 * overflow */
e050c007 765 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
766
767 /* To be such an overlong, the first bytes of 's' must match
768 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
769 * don't have any additional bytes available, the sequence, when
770 * completed might or might not fit in 32 bits. But if we have that
771 * next byte, we can tell for sure. If it is <= 0x83, then it does
772 * fit. */
773 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
774 return -1;
775 }
776
777 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
778 }
779
d22ec717
KW
780/* Starting with the #else, the rest of the function is identical except
781 * 1. we need to move the 'len' declaration to be global to the function
782 * 2. the endif move to just after the UNUSED_ARG.
783 * An empty endif is given just below to satisfy the preprocessor
784 */
a77c906e
KW
785#endif
786
d22ec717 787#endif
a77c906e 788
12a4bed3
KW
789#undef F0_ABOVE_OVERLONG
790#undef F8_ABOVE_OVERLONG
791#undef FC_ABOVE_OVERLONG
792#undef FE_ABOVE_OVERLONG
793#undef FF_OVERLONG_PREFIX
794
35f8c9bd 795STRLEN
edc2c47a 796Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 797{
2b479609 798 STRLEN len;
12a4bed3 799 const U8 *x;
35f8c9bd 800
2b479609
KW
801 /* A helper function that should not be called directly.
802 *
803 * This function returns non-zero if the string beginning at 's' and
804 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
805 * code point; otherwise it returns 0. The examination stops after the
806 * first code point in 's' is validated, not looking at the rest of the
807 * input. If 'e' is such that there are not enough bytes to represent a
808 * complete code point, this function will return non-zero anyway, if the
809 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
810 * excluded by 'flags'.
811 *
812 * A non-zero return gives the number of bytes required to represent the
813 * code point. Be aware that if the input is for a partial character, the
814 * return will be larger than 'e - s'.
815 *
816 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
817 * The caller should have excluded the possibility of it being invariant
818 * before calling this function.
2b479609
KW
819 *
820 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
821 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
822 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
823 * disallowed by the flags. If the input is only for a partial character,
824 * the function will return non-zero if there is any sequence of
825 * well-formed UTF-8 that, when appended to the input sequence, could
826 * result in an allowed code point; otherwise it returns 0. Non characters
827 * cannot be determined based on partial character input. But many of the
828 * other excluded types can be determined with just the first one or two
829 * bytes.
830 *
831 */
832
833 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
834
835 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 836 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 837 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 838
2b479609 839 /* A variant char must begin with a start byte */
35f8c9bd
KW
840 if (UNLIKELY(! UTF8_IS_START(*s))) {
841 return 0;
842 }
843
edc2c47a
KW
844 /* Examine a maximum of a single whole code point */
845 if (e - s > UTF8SKIP(s)) {
846 e = s + UTF8SKIP(s);
847 }
848
2b479609
KW
849 len = e - s;
850
851 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
852 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 853
56576a04
KW
854 /* Here, we are disallowing some set of largish code points, and the
855 * first byte indicates the sequence is for a code point that could be
856 * in the excluded set. We generally don't have to look beyond this or
857 * the second byte to see if the sequence is actually for one of the
858 * excluded classes. The code below is derived from this table:
859 *
2b479609
KW
860 * UTF-8 UTF-EBCDIC I8
861 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
862 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
863 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
864 *
56576a04
KW
865 * Keep in mind that legal continuation bytes range between \x80..\xBF
866 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
867 * continuation bytes. Hence, we don't have to test the upper edge
868 * because if any of those is encountered, the sequence is malformed,
869 * and would fail elsewhere in this function.
870 *
871 * The code here likewise assumes that there aren't other
872 * malformations; again the function should fail elsewhere because of
873 * these. For example, an overlong beginning with FC doesn't actually
874 * have to be a super; it could actually represent a small code point,
875 * even U+0000. But, since overlongs (and other malformations) are
876 * illegal, the function should return FALSE in either case.
2b479609
KW
877 */
878
879#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
880# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 881# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 882
19794540
KW
883# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
884 /* B6 and B7 */ \
885 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 886# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
887#else
888# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
889# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
890# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 891# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
892#endif
893
894 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
895 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
896 {
2b479609
KW
897 return 0; /* Above Unicode */
898 }
899
d044b7a7 900 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 901 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 902 {
57ff5f59 903 return 0;
2b479609
KW
904 }
905
906 if (len > 1) {
907 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
908
909 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 910 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
911 {
912 return 0; /* Above Unicode */
913 }
914
915 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 916 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
917 {
918 return 0; /* Surrogate */
919 }
920
921 if ( (flags & UTF8_DISALLOW_NONCHAR)
922 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
923 {
924 return 0; /* Noncharacter code point */
925 }
926 }
927 }
928
929 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
930 for (x = s + 1; x < e; x++) {
931 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
932 return 0;
933 }
934 }
935
af13dd8a 936 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 937 * overlong. */
d6be65ae 938 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 939 return 0;
af13dd8a
KW
940 }
941
12a4bed3
KW
942 /* And finally, that the code point represented fits in a word on this
943 * platform */
e050c007
KW
944 if (0 < does_utf8_overflow(s, e,
945 0 /* Don't consider overlongs */
946 ))
947 {
12a4bed3 948 return 0;
35f8c9bd
KW
949 }
950
2b479609 951 return UTF8SKIP(s);
35f8c9bd
KW
952}
953
7e2f38b2
KW
954char *
955Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
956{
957 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
958 * bytes starting at 's'. 'format' gives how to display each byte.
959 * Currently, there are only two formats, so it is currently a bool:
960 * 0 \xab
961 * 1 ab (that is a space between two hex digit bytes)
962 */
7cf8d05d
KW
963
964 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
965 trailing NUL */
966 const U8 * const e = s + len;
967 char * output;
968 char * d;
969
970 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
971
972 Newx(output, output_len, char);
973 SAVEFREEPV(output);
974
975 d = output;
976 for (; s < e; s++) {
977 const unsigned high_nibble = (*s & 0xF0) >> 4;
978 const unsigned low_nibble = (*s & 0x0F);
979
7e2f38b2
KW
980 if (format) {
981 *d++ = ' ';
982 }
983 else {
984 *d++ = '\\';
985 *d++ = 'x';
986 }
7cf8d05d
KW
987
988 if (high_nibble < 10) {
989 *d++ = high_nibble + '0';
990 }
991 else {
992 *d++ = high_nibble - 10 + 'a';
993 }
994
995 if (low_nibble < 10) {
996 *d++ = low_nibble + '0';
997 }
998 else {
999 *d++ = low_nibble - 10 + 'a';
1000 }
1001 }
1002
1003 *d = '\0';
1004 return output;
1005}
1006
806547a7 1007PERL_STATIC_INLINE char *
7cf8d05d
KW
1008S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1009
1010 /* How many bytes to print */
3cc6a05e 1011 STRLEN print_len,
7cf8d05d
KW
1012
1013 /* Which one is the non-continuation */
1014 const STRLEN non_cont_byte_pos,
1015
1016 /* How many bytes should there be? */
1017 const STRLEN expect_len)
806547a7
KW
1018{
1019 /* Return the malformation warning text for an unexpected continuation
1020 * byte. */
1021
7cf8d05d 1022 const char * const where = (non_cont_byte_pos == 1)
806547a7 1023 ? "immediately"
7cf8d05d
KW
1024 : Perl_form(aTHX_ "%d bytes",
1025 (int) non_cont_byte_pos);
806547a7
KW
1026
1027 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1028
7cf8d05d
KW
1029 /* We don't need to pass this parameter, but since it has already been
1030 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1031 assert(expect_len == UTF8SKIP(s));
1032
1033 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1034 " %s after start byte 0x%02x; need %d bytes, got %d)",
1035 malformed_text,
7e2f38b2 1036 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
1037 *(s + non_cont_byte_pos),
1038 where,
1039 *s,
1040 (int) expect_len,
1041 (int) non_cont_byte_pos);
806547a7
KW
1042}
1043
35f8c9bd
KW
1044/*
1045
de69f3af 1046=for apidoc utf8n_to_uvchr
378516de
KW
1047
1048THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1049Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1050
9041c2e3 1051Bottom level UTF-8 decode routine.
de69f3af 1052Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1053which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1054C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1055the length, in bytes, of that character.
949cf498
KW
1056
1057The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1058well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1059causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1060is the next possible position in C<s> that could begin a non-malformed
1061character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1062is raised. Some UTF-8 input sequences may contain multiple malformations.
1063This function tries to find every possible one in each call, so multiple
56576a04 1064warnings can be raised for the same sequence.
949cf498
KW
1065
1066Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1067individual types of malformations, such as the sequence being overlong (that
1068is, when there is a shorter sequence that can express the same code point;
1069overlong sequences are expressly forbidden in the UTF-8 standard due to
1070potential security issues). Another malformation example is the first byte of
1071a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1072flags. Even if allowed, this function generally returns the Unicode
1073REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1074F<utf8.h> to override this behavior for the overlong malformations, but don't
1075do that except for very specialized purposes.
949cf498 1076
796b6530 1077The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1078flags) malformation is found. If this flag is set, the routine assumes that
1079the caller will raise a warning, and this function will silently just set
d088425d
KW
1080C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1081
75200dff 1082Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1083character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1084in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1085be set to 1. To disambiguate, upon a zero return, see if the first byte of
1086C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1087error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1088
1089Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1090Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1091By default these are considered regular code points, but certain situations
ecc1615f
KW
1092warrant special handling for them, which can be specified using the C<flags>
1093parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1094three classes are treated as malformations and handled as such. The flags
1095C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1096C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1097disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1098restricts the allowed inputs to the strict UTF-8 traditionally defined by
1099Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1100definition given by
1101L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1102The difference between traditional strictness and C9 strictness is that the
1103latter does not forbid non-character code points. (They are still discouraged,
1104however.) For more discussion see L<perlunicode/Noncharacter code points>.
1105
1106The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1107C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1108C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1109raised for their respective categories, but otherwise the code points are
1110considered valid (not malformations). To get a category to both be treated as
1111a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1112(But note that warnings are not raised if lexically disabled nor if
796b6530 1113C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1114
57ff5f59
KW
1115Extremely high code points were never specified in any standard, and require an
1116extension to UTF-8 to express, which Perl does. It is likely that programs
1117written in something other than Perl would not be able to read files that
1118contain these; nor would Perl understand files written by something that uses a
1119different extension. For these reasons, there is a separate set of flags that
1120can warn and/or disallow these extremely high code points, even if other
1121above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1122C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1123L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1124above-Unicode code points, including these, as malformations.
1125(Note that the Unicode standard considers anything above 0x10FFFF to be
1126illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1127(2**31 -1))
1128
1129A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1130retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1131C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1132C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1133can apply to code points that actually do fit in 31 bits. This happens on
1134EBCDIC platforms, and sometimes when the L<overlong
1135malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1136describe the situation in all cases.
1137
ab8e6d41 1138
949cf498
KW
1139All other code points corresponding to Unicode characters, including private
1140use and those yet to be assigned, are never considered malformed and never
1141warn.
67e989fb 1142
37607a96 1143=cut
f9380377
KW
1144
1145Also implemented as a macro in utf8.h
1146*/
1147
1148UV
1149Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1150 STRLEN curlen,
1151 STRLEN *retlen,
1152 const U32 flags)
1153{
1154 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1155
1156 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1157}
1158
1159/*
1160
1161=for apidoc utf8n_to_uvchr_error
1162
1163THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1164Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1165
1166This function is for code that needs to know what the precise malformation(s)
1167are when an error is found.
1168
1169It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1170all the others, C<errors>. If this parameter is 0, this function behaves
1171identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1172to a C<U32> variable, which this function sets to indicate any errors found.
1173Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1174C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1175of these bits will be set if a malformation is found, even if the input
7a65503b 1176C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1177exceptions are noted:
1178
1179=over 4
1180
57ff5f59 1181=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1182
57ff5f59
KW
1183The input sequence is not standard UTF-8, but a Perl extension. This bit is
1184set only if the input C<flags> parameter contains either the
1185C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1186
1187Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1188and so some extension must be used to express them. Perl uses a natural
1189extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1190extension to represent even higher ones, so that any code point that fits in a
119164-bit word can be represented. Text using these extensions is not likely to
1192be portable to non-Perl code. We lump both of these extensions together and
1193refer to them as Perl extended UTF-8. There exist other extensions that people
1194have invented, incompatible with Perl's.
1195
1196On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1197extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1198than on ASCII. Prior to that, code points 2**31 and higher were simply
1199unrepresentable, and a different, incompatible method was used to represent
1200code points between 2**30 and 2**31 - 1.
1201
1202On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1203Perl extended UTF-8 is used.
1204
1205In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1206may use for backward compatibility. That name is misleading, as this flag may
1207be set when the code point actually does fit in 31 bits. This happens on
1208EBCDIC platforms, and sometimes when the L<overlong
1209malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1210describes the situation in all cases.
f9380377
KW
1211
1212=item C<UTF8_GOT_CONTINUATION>
1213
1214The input sequence was malformed in that the first byte was a a UTF-8
1215continuation byte.
1216
1217=item C<UTF8_GOT_EMPTY>
1218
1219The input C<curlen> parameter was 0.
1220
1221=item C<UTF8_GOT_LONG>
1222
1223The input sequence was malformed in that there is some other sequence that
1224evaluates to the same code point, but that sequence is shorter than this one.
1225
fecaf136
KW
1226Until Unicode 3.1, it was legal for programs to accept this malformation, but
1227it was discovered that this created security issues.
1228
f9380377
KW
1229=item C<UTF8_GOT_NONCHAR>
1230
1231The code point represented by the input UTF-8 sequence is for a Unicode
1232non-character code point.
1233This bit is set only if the input C<flags> parameter contains either the
1234C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1235
1236=item C<UTF8_GOT_NON_CONTINUATION>
1237
1238The input sequence was malformed in that a non-continuation type byte was found
1239in a position where only a continuation type one should be.
1240
1241=item C<UTF8_GOT_OVERFLOW>
1242
1243The input sequence was malformed in that it is for a code point that is not
d22ec717 1244representable in the number of bits available in an IV on the current platform.
f9380377
KW
1245
1246=item C<UTF8_GOT_SHORT>
1247
1248The input sequence was malformed in that C<curlen> is smaller than required for
1249a complete sequence. In other words, the input is for a partial character
1250sequence.
1251
1252=item C<UTF8_GOT_SUPER>
1253
1254The input sequence was malformed in that it is for a non-Unicode code point;
1255that is, one above the legal Unicode maximum.
1256This bit is set only if the input C<flags> parameter contains either the
1257C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1258
1259=item C<UTF8_GOT_SURROGATE>
1260
1261The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1262code point.
1263This bit is set only if the input C<flags> parameter contains either the
1264C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1265
1266=back
1267
133551d8
KW
1268To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1269flag to suppress any warnings, and then examine the C<*errors> return.
1270
f9380377 1271=cut
37607a96 1272*/
67e989fb 1273
a0ed51b3 1274UV
f9380377
KW
1275Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1276 STRLEN curlen,
1277 STRLEN *retlen,
1278 const U32 flags,
1279 U32 * errors)
a0ed51b3 1280{
d4c19fe8 1281 const U8 * const s0 = s;
2b5e7bc2
KW
1282 U8 * send = NULL; /* (initialized to silence compilers' wrong
1283 warning) */
1284 U32 possible_problems = 0; /* A bit is set here for each potential problem
1285 found as we go along */
eb83ed87 1286 UV uv = *s;
2b5e7bc2
KW
1287 STRLEN expectlen = 0; /* How long should this sequence be?
1288 (initialized to silence compilers' wrong
1289 warning) */
e308b348 1290 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1291 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1292 this gets set and discarded */
a0dbb045 1293
2b5e7bc2
KW
1294 /* The below are used only if there is both an overlong malformation and a
1295 * too short one. Otherwise the first two are set to 's0' and 'send', and
1296 * the third not used at all */
1297 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1298 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1299 routine; see [perl #130921] */
2b5e7bc2 1300 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1301
f9380377
KW
1302 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1303
1304 if (errors) {
1305 *errors = 0;
1306 }
1307 else {
1308 errors = &discard_errors;
1309 }
a0dbb045 1310
eb83ed87
KW
1311 /* The order of malformation tests here is important. We should consume as
1312 * few bytes as possible in order to not skip any valid character. This is
1313 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1314 * http://unicode.org/reports/tr36 for more discussion as to why. For
1315 * example, once we've done a UTF8SKIP, we can tell the expected number of
1316 * bytes, and could fail right off the bat if the input parameters indicate
1317 * that there are too few available. But it could be that just that first
1318 * byte is garbled, and the intended character occupies fewer bytes. If we
1319 * blindly assumed that the first byte is correct, and skipped based on
1320 * that number, we could skip over a valid input character. So instead, we
1321 * always examine the sequence byte-by-byte.
1322 *
1323 * We also should not consume too few bytes, otherwise someone could inject
1324 * things. For example, an input could be deliberately designed to
1325 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1326 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1327 * which is actually part of of the overflowing sequence), that could look
1328 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1329 * sequence and process the rest, inappropriately.
1330 *
1331 * Some possible input sequences are malformed in more than one way. This
1332 * function goes to lengths to try to find all of them. This is necessary
1333 * for correctness, as the inputs may allow one malformation but not
1334 * another, and if we abandon searching for others after finding the
1335 * allowed one, we could allow in something that shouldn't have been.
1336 */
eb83ed87 1337
b5b9af04 1338 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1339 possible_problems |= UTF8_GOT_EMPTY;
1340 curlen = 0;
5a48568d 1341 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1342 goto ready_to_handle_errors;
0c443dc2
JH
1343 }
1344
eb83ed87
KW
1345 expectlen = UTF8SKIP(s);
1346
1347 /* A well-formed UTF-8 character, as the vast majority of calls to this
1348 * function will be for, has this expected length. For efficiency, set
1349 * things up here to return it. It will be overriden only in those rare
1350 * cases where a malformation is found */
1351 if (retlen) {
1352 *retlen = expectlen;
1353 }
1354
1355 /* An invariant is trivially well-formed */
1d72bdf6 1356 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1357 return uv;
a0ed51b3 1358 }
67e989fb 1359
eb83ed87 1360 /* A continuation character can't start a valid sequence */
b5b9af04 1361 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1362 possible_problems |= UTF8_GOT_CONTINUATION;
1363 curlen = 1;
1364 uv = UNICODE_REPLACEMENT;
1365 goto ready_to_handle_errors;
ba210ebe 1366 }
9041c2e3 1367
dcd27b3c 1368 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1369 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1370 * because it excludes start bytes like \xC0 that always lead to
1371 * overlongs.) */
dcd27b3c 1372
534752c1
KW
1373 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1374 * that indicate the number of bytes in the character's whole UTF-8
1375 * sequence, leaving just the bits that are part of the value. */
1376 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1377
e308b348
KW
1378 /* Setup the loop end point, making sure to not look past the end of the
1379 * input string, and flag it as too short if the size isn't big enough. */
1380 send = (U8*) s0;
1381 if (UNLIKELY(curlen < expectlen)) {
1382 possible_problems |= UTF8_GOT_SHORT;
1383 avail_len = curlen;
1384 send += curlen;
1385 }
1386 else {
1387 send += expectlen;
1388 }
e308b348 1389
eb83ed87 1390 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1391 * accumulating each into the working value as we go. */
eb83ed87 1392 for (s = s0 + 1; s < send; s++) {
b5b9af04 1393 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1394 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1395 continue;
1396 }
1397
1398 /* Here, found a non-continuation before processing all expected bytes.
1399 * This byte indicates the beginning of a new character, so quit, even
1400 * if allowing this malformation. */
2b5e7bc2 1401 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1402 break;
eb83ed87
KW
1403 } /* End of loop through the character's bytes */
1404
1405 /* Save how many bytes were actually in the character */
1406 curlen = s - s0;
1407
2b5e7bc2
KW
1408 /* Note that there are two types of too-short malformation. One is when
1409 * there is actual wrong data before the normal termination of the
1410 * sequence. The other is that the sequence wasn't complete before the end
1411 * of the data we are allowed to look at, based on the input 'curlen'.
1412 * This means that we were passed data for a partial character, but it is
1413 * valid as far as we saw. The other is definitely invalid. This
1414 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1415 * separate.
1416 *
1417 * A convenience macro that matches either of the too-short conditions. */
1418# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1419
1420 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1421 uv_so_far = uv;
1422 uv = UNICODE_REPLACEMENT;
1423 }
2b5e7bc2 1424
08e73697
KW
1425 /* Check for overflow. The algorithm requires us to not look past the end
1426 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1427 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1428 1 /* Do consider overlongs */
1429 )))
1430 {
2b5e7bc2
KW
1431 possible_problems |= UTF8_GOT_OVERFLOW;
1432 uv = UNICODE_REPLACEMENT;
eb83ed87 1433 }
eb83ed87 1434
2b5e7bc2
KW
1435 /* Check for overlong. If no problems so far, 'uv' is the correct code
1436 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1437 * we must look at the UTF-8 byte sequence itself to see if it is for an
1438 * overlong */
1439 if ( ( LIKELY(! possible_problems)
1440 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1441 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1442 && ( UNLIKELY(! UTF8_IS_START(*s0))
1443 || ( curlen > 1
d6be65ae 1444 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1445 s - s0))))))
2f8f112e 1446 {
2b5e7bc2
KW
1447 possible_problems |= UTF8_GOT_LONG;
1448
abc28b54 1449 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1450
abc28b54
KW
1451 /* The calculation in the 'true' branch of this 'if'
1452 * below won't work if overflows, and isn't needed
1453 * anyway. Further below we handle all overflow
1454 * cases */
1455 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1456 {
2b5e7bc2
KW
1457 UV min_uv = uv_so_far;
1458 STRLEN i;
1459
1460 /* Here, the input is both overlong and is missing some trailing
1461 * bytes. There is no single code point it could be for, but there
1462 * may be enough information present to determine if what we have
1463 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1464 * The code further below has the intelligence to determine this,
1465 * but just for non-overlong UTF-8 sequences. What we do here is
1466 * calculate the smallest code point the input could represent if
1467 * there were no too short malformation. Then we compute and save
1468 * the UTF-8 for that, which is what the code below looks at
1469 * instead of the raw input. It turns out that the smallest such
1470 * code point is all we need. */
2b5e7bc2
KW
1471 for (i = curlen; i < expectlen; i++) {
1472 min_uv = UTF8_ACCUMULATE(min_uv,
1473 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1474 }
1475
e9f2c446 1476 adjusted_s0 = temp_char_buf;
57ff5f59 1477 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1478 }
eb83ed87
KW
1479 }
1480
56576a04
KW
1481 /* Here, we have found all the possible problems, except for when the input
1482 * is for a problematic code point not allowed by the input parameters. */
1483
06188866
KW
1484 /* uv is valid for overlongs */
1485 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1486
1487 /* isn't problematic if < this */
1488 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1489 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1490
1491 /* if overflow, we know without looking further
1492 * precisely which of the problematic types it is,
1493 * and we deal with those in the overflow handling
1494 * code */
1495 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1496 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1497 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1498 && ((flags & ( UTF8_DISALLOW_NONCHAR
1499 |UTF8_DISALLOW_SURROGATE
1500 |UTF8_DISALLOW_SUPER
d044b7a7 1501 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1502 |UTF8_WARN_NONCHAR
1503 |UTF8_WARN_SURROGATE
1504 |UTF8_WARN_SUPER
d22ec717 1505 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1506 {
2b5e7bc2
KW
1507 /* If there were no malformations, or the only malformation is an
1508 * overlong, 'uv' is valid */
1509 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1510 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1511 possible_problems |= UTF8_GOT_SURROGATE;
1512 }
1513 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1514 possible_problems |= UTF8_GOT_SUPER;
1515 }
1516 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1517 possible_problems |= UTF8_GOT_NONCHAR;
1518 }
1519 }
1520 else { /* Otherwise, need to look at the source UTF-8, possibly
1521 adjusted to be non-overlong */
1522
1523 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1524 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1525 {
2b5e7bc2
KW
1526 possible_problems |= UTF8_GOT_SUPER;
1527 }
1528 else if (curlen > 1) {
1529 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1530 NATIVE_UTF8_TO_I8(*adjusted_s0),
1531 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1532 {
2b5e7bc2 1533 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1534 }
2b5e7bc2
KW
1535 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1536 NATIVE_UTF8_TO_I8(*adjusted_s0),
1537 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1538 {
1539 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1540 }
1541 }
c0236afe 1542
2b5e7bc2
KW
1543 /* We need a complete well-formed UTF-8 character to discern
1544 * non-characters, so can't look for them here */
1545 }
1546 }
949cf498 1547
2b5e7bc2
KW
1548 ready_to_handle_errors:
1549
1550 /* At this point:
1551 * curlen contains the number of bytes in the sequence that
1552 * this call should advance the input by.
e308b348
KW
1553 * avail_len gives the available number of bytes passed in, but
1554 * only if this is less than the expected number of
1555 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1556 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1557 * is set in it for each potential problem found.
1558 * uv contains the code point the input sequence
1559 * represents; or if there is a problem that prevents
1560 * a well-defined value from being computed, it is
1561 * some subsitute value, typically the REPLACEMENT
1562 * CHARACTER.
1563 * s0 points to the first byte of the character
56576a04
KW
1564 * s points to just after were we left off processing
1565 * the character
1566 * send points to just after where that character should
1567 * end, based on how many bytes the start byte tells
1568 * us should be in it, but no further than s0 +
1569 * avail_len
2b5e7bc2 1570 */
eb83ed87 1571
2b5e7bc2
KW
1572 if (UNLIKELY(possible_problems)) {
1573 bool disallowed = FALSE;
1574 const U32 orig_problems = possible_problems;
1575
1576 while (possible_problems) { /* Handle each possible problem */
1577 UV pack_warn = 0;
1578 char * message = NULL;
1579
1580 /* Each 'if' clause handles one problem. They are ordered so that
1581 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1582 * ones; this is kinda in decreasing severity order. But the
1583 * overlong must come last, as it changes 'uv' looked at by the
1584 * others */
2b5e7bc2
KW
1585 if (possible_problems & UTF8_GOT_OVERFLOW) {
1586
56576a04
KW
1587 /* Overflow means also got a super and are using Perl's
1588 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1589 possible_problems
d044b7a7 1590 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1591 *errors |= UTF8_GOT_OVERFLOW;
1592
1593 /* But the API says we flag all errors found */
1594 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1595 *errors |= UTF8_GOT_SUPER;
1596 }
ddb65933 1597 if (flags
d044b7a7 1598 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1599 {
d044b7a7 1600 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1601 }
2b5e7bc2 1602
d60baaa7 1603 /* Disallow if any of the three categories say to */
56576a04 1604 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1605 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1606 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1607 {
1608 disallowed = TRUE;
1609 }
1610
d22ec717
KW
1611 /* Likewise, warn if any say to */
1612 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1613 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1614 {
2b5e7bc2 1615
ddb65933
KW
1616 /* The warnings code explicitly says it doesn't handle the
1617 * case of packWARN2 and two categories which have
1618 * parent-child relationship. Even if it works now to
1619 * raise the warning if either is enabled, it wouldn't
1620 * necessarily do so in the future. We output (only) the
56576a04 1621 * most dire warning */
ddb65933
KW
1622 if (! (flags & UTF8_CHECK_ONLY)) {
1623 if (ckWARN_d(WARN_UTF8)) {
1624 pack_warn = packWARN(WARN_UTF8);
1625 }
1626 else if (ckWARN_d(WARN_NON_UNICODE)) {
1627 pack_warn = packWARN(WARN_NON_UNICODE);
1628 }
1629 if (pack_warn) {
1630 message = Perl_form(aTHX_ "%s: %s (overflows)",
1631 malformed_text,
05b9033b 1632 _byte_dump_string(s0, curlen, 0));
ddb65933 1633 }
2b5e7bc2
KW
1634 }
1635 }
1636 }
1637 else if (possible_problems & UTF8_GOT_EMPTY) {
1638 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1639 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1640
1641 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1642
1643 /* This so-called malformation is now treated as a bug in
1644 * the caller. If you have nothing to decode, skip calling
1645 * this function */
1646 assert(0);
1647
2b5e7bc2
KW
1648 disallowed = TRUE;
1649 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1650 pack_warn = packWARN(WARN_UTF8);
1651 message = Perl_form(aTHX_ "%s (empty string)",
1652 malformed_text);
1653 }
1654 }
1655 }
1656 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1657 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1658 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1659
1660 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1661 disallowed = TRUE;
1662 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1663 pack_warn = packWARN(WARN_UTF8);
1664 message = Perl_form(aTHX_
1665 "%s: %s (unexpected continuation byte 0x%02x,"
1666 " with no preceding start byte)",
1667 malformed_text,
7e2f38b2 1668 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1669 }
1670 }
1671 }
2b5e7bc2
KW
1672 else if (possible_problems & UTF8_GOT_SHORT) {
1673 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1674 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1675
1676 if (! (flags & UTF8_ALLOW_SHORT)) {
1677 disallowed = TRUE;
1678 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1679 pack_warn = packWARN(WARN_UTF8);
1680 message = Perl_form(aTHX_
56576a04
KW
1681 "%s: %s (too short; %d byte%s available, need %d)",
1682 malformed_text,
1683 _byte_dump_string(s0, send - s0, 0),
1684 (int)avail_len,
1685 avail_len == 1 ? "" : "s",
1686 (int)expectlen);
2b5e7bc2
KW
1687 }
1688 }
ba210ebe 1689
2b5e7bc2 1690 }
e308b348
KW
1691 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1692 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1693 *errors |= UTF8_GOT_NON_CONTINUATION;
1694
1695 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1696 disallowed = TRUE;
1697 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1698
1699 /* If we don't know for sure that the input length is
1700 * valid, avoid as much as possible reading past the
1701 * end of the buffer */
1702 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1703 ? s - s0
1704 : send - s0;
e308b348
KW
1705 pack_warn = packWARN(WARN_UTF8);
1706 message = Perl_form(aTHX_ "%s",
1707 unexpected_non_continuation_text(s0,
99a765e9 1708 printlen,
e308b348
KW
1709 s - s0,
1710 (int) expectlen));
1711 }
1712 }
1713 }
2b5e7bc2
KW
1714 else if (possible_problems & UTF8_GOT_SURROGATE) {
1715 possible_problems &= ~UTF8_GOT_SURROGATE;
1716
f9380377
KW
1717 if (flags & UTF8_WARN_SURROGATE) {
1718 *errors |= UTF8_GOT_SURROGATE;
1719
1720 if ( ! (flags & UTF8_CHECK_ONLY)
1721 && ckWARN_d(WARN_SURROGATE))
1722 {
2b5e7bc2
KW
1723 pack_warn = packWARN(WARN_SURROGATE);
1724
1725 /* These are the only errors that can occur with a
1726 * surrogate when the 'uv' isn't valid */
1727 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1728 message = Perl_form(aTHX_
1729 "UTF-16 surrogate (any UTF-8 sequence that"
1730 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1731 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1732 }
1733 else {
c94c2f39 1734 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1735 }
f9380377 1736 }
2b5e7bc2 1737 }
ba210ebe 1738
2b5e7bc2
KW
1739 if (flags & UTF8_DISALLOW_SURROGATE) {
1740 disallowed = TRUE;
f9380377 1741 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1742 }
1743 }
1744 else if (possible_problems & UTF8_GOT_SUPER) {
1745 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1746
f9380377
KW
1747 if (flags & UTF8_WARN_SUPER) {
1748 *errors |= UTF8_GOT_SUPER;
1749
1750 if ( ! (flags & UTF8_CHECK_ONLY)
1751 && ckWARN_d(WARN_NON_UNICODE))
1752 {
2b5e7bc2
KW
1753 pack_warn = packWARN(WARN_NON_UNICODE);
1754
1755 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1756 message = Perl_form(aTHX_
1757 "Any UTF-8 sequence that starts with"
1758 " \"%s\" is for a non-Unicode code point,"
1759 " may not be portable",
7e2f38b2 1760 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1761 }
1762 else {
c94c2f39 1763 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1764 }
f9380377 1765 }
2b5e7bc2 1766 }
ba210ebe 1767
57ff5f59
KW
1768 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1769 * and before possibly bailing out, so that the more dire
1770 * warning will override the regular one. */
1771 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1772 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1773 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1774 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1775 {
db0f09e6 1776 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1777
57ff5f59
KW
1778 /* If it is an overlong that evaluates to a code point
1779 * that doesn't have to use the Perl extended UTF-8, it
1780 * still used it, and so we output a message that
1781 * doesn't refer to the code point. The same is true
1782 * if there was a SHORT malformation where the code
1783 * point is not valid. In that case, 'uv' will have
1784 * been set to the REPLACEMENT CHAR, and the message
1785 * below without the code point in it will be selected
1786 * */
1787 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1788 message = Perl_form(aTHX_
57ff5f59 1789 perl_extended_cp_format, uv);
2b5e7bc2
KW
1790 }
1791 else {
1792 message = Perl_form(aTHX_
57ff5f59
KW
1793 "Any UTF-8 sequence that starts with"
1794 " \"%s\" is a Perl extension, and"
1795 " so is not portable",
1796 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1797 }
1798 }
1799
d044b7a7
KW
1800 if (flags & ( UTF8_WARN_PERL_EXTENDED
1801 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1802 {
d044b7a7 1803 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1804
d044b7a7 1805 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1806 disallowed = TRUE;
1807 }
2b5e7bc2
KW
1808 }
1809 }
eb83ed87 1810
2b5e7bc2 1811 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1812 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1813 disallowed = TRUE;
1814 }
2b5e7bc2
KW
1815 }
1816 else if (possible_problems & UTF8_GOT_NONCHAR) {
1817 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1818
f9380377
KW
1819 if (flags & UTF8_WARN_NONCHAR) {
1820 *errors |= UTF8_GOT_NONCHAR;
1821
1822 if ( ! (flags & UTF8_CHECK_ONLY)
1823 && ckWARN_d(WARN_NONCHAR))
1824 {
2b5e7bc2
KW
1825 /* The code above should have guaranteed that we don't
1826 * get here with errors other than overlong */
1827 assert (! (orig_problems
1828 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1829
1830 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1831 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1832 }
2b5e7bc2 1833 }
5b311467 1834
2b5e7bc2
KW
1835 if (flags & UTF8_DISALLOW_NONCHAR) {
1836 disallowed = TRUE;
f9380377 1837 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1838 }
6c64cd9d
KW
1839 }
1840 else if (possible_problems & UTF8_GOT_LONG) {
1841 possible_problems &= ~UTF8_GOT_LONG;
1842 *errors |= UTF8_GOT_LONG;
1843
1844 if (flags & UTF8_ALLOW_LONG) {
1845
1846 /* We don't allow the actual overlong value, unless the
1847 * special extra bit is also set */
1848 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1849 & ~UTF8_ALLOW_LONG)))
1850 {
1851 uv = UNICODE_REPLACEMENT;
1852 }
1853 }
1854 else {
1855 disallowed = TRUE;
1856
1857 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1858 pack_warn = packWARN(WARN_UTF8);
1859
1860 /* These error types cause 'uv' to be something that
1861 * isn't what was intended, so can't use it in the
1862 * message. The other error types either can't
1863 * generate an overlong, or else the 'uv' is valid */
1864 if (orig_problems &
1865 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1866 {
1867 message = Perl_form(aTHX_
1868 "%s: %s (any UTF-8 sequence that starts"
1869 " with \"%s\" is overlong which can and"
1870 " should be represented with a"
1871 " different, shorter sequence)",
1872 malformed_text,
1873 _byte_dump_string(s0, send - s0, 0),
1874 _byte_dump_string(s0, curlen, 0));
1875 }
1876 else {
1877 U8 tmpbuf[UTF8_MAXBYTES+1];
d819dc50
KW
1878 const U8 * const e = uvchr_to_utf8_flags(tmpbuf,
1879 uv, 0);
1880 /* Don't use U+ for non-Unicode code points, which
1881 * includes those in the Latin1 range */
1882 const char * preface = ( uv > PERL_UNICODE_MAX
1883#ifdef EBCDIC
1884 || uv <= 0xFF
1885#endif
1886 )
1887 ? "0x"
1888 : "U+";
6c64cd9d
KW
1889 message = Perl_form(aTHX_
1890 "%s: %s (overlong; instead use %s to represent"
1891 " %s%0*" UVXf ")",
1892 malformed_text,
1893 _byte_dump_string(s0, send - s0, 0),
1894 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1895 preface,
1896 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1897 small code points */
1898 uv);
1899 }
1900 }
1901 }
2b5e7bc2
KW
1902 } /* End of looking through the possible flags */
1903
1904 /* Display the message (if any) for the problem being handled in
1905 * this iteration of the loop */
1906 if (message) {
1907 if (PL_op)
1908 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1909 OP_DESC(PL_op));
1910 else
1911 Perl_warner(aTHX_ pack_warn, "%s", message);
1912 }
ddb65933 1913 } /* End of 'while (possible_problems)' */
a0dbb045 1914
2b5e7bc2
KW
1915 /* Since there was a possible problem, the returned length may need to
1916 * be changed from the one stored at the beginning of this function.
1917 * Instead of trying to figure out if that's needed, just do it. */
1918 if (retlen) {
1919 *retlen = curlen;
1920 }
a0dbb045 1921
2b5e7bc2
KW
1922 if (disallowed) {
1923 if (flags & UTF8_CHECK_ONLY && retlen) {
1924 *retlen = ((STRLEN) -1);
1925 }
1926 return 0;
1927 }
eb83ed87 1928 }
ba210ebe 1929
2b5e7bc2 1930 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1931}
1932
8e84507e 1933/*
ec5f19d0
KW
1934=for apidoc utf8_to_uvchr_buf
1935
1936Returns the native code point of the first character in the string C<s> which
1937is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1938C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1939
524080c4
KW
1940If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1941enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1942C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1943(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1944C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1945the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1946See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1947returned.
ec5f19d0
KW
1948
1949=cut
52be2536
KW
1950
1951Also implemented as a macro in utf8.h
1952
ec5f19d0
KW
1953*/
1954
1955
1956UV
1957Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1958{
7f974d7e
KW
1959 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1960
ec5f19d0
KW
1961 assert(s < send);
1962
1963 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1964 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1965}
1966
52be2536
KW
1967/* This is marked as deprecated
1968 *
ec5f19d0
KW
1969=for apidoc utf8_to_uvuni_buf
1970
de69f3af
KW
1971Only in very rare circumstances should code need to be dealing in Unicode
1972(as opposed to native) code points. In those few cases, use
1973C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1974
1975Returns the Unicode (not-native) code point of the first character in the
1976string C<s> which
ec5f19d0
KW
1977is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1978C<retlen> will be set to the length, in bytes, of that character.
1979
524080c4
KW
1980If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1981enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1982NULL) to -1. If those warnings are off, the computed value if well-defined (or
1983the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1984is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1985next possible position in C<s> that could begin a non-malformed character.
de69f3af 1986See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
1987
1988=cut
1989*/
1990
1991UV
1992Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1993{
1994 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1995
1996 assert(send > s);
1997
5962d97e
KW
1998 /* Call the low level routine, asking for checks */
1999 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2000}
2001
b76347f2 2002/*
87cea99e 2003=for apidoc utf8_length
b76347f2
JH
2004
2005Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
2006Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
2007up past C<e>, croaks.
b76347f2
JH
2008
2009=cut
2010*/
2011
2012STRLEN
35a4481c 2013Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2014{
2015 STRLEN len = 0;
2016
7918f24d
NC
2017 PERL_ARGS_ASSERT_UTF8_LENGTH;
2018
8850bf83
JH
2019 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2020 * the bitops (especially ~) can create illegal UTF-8.
2021 * In other words: in Perl UTF-8 is not just for Unicode. */
2022
a3b680e6
AL
2023 if (e < s)
2024 goto warn_and_return;
b76347f2 2025 while (s < e) {
4cbf4130 2026 s += UTF8SKIP(s);
8e91ec7f
AV
2027 len++;
2028 }
2029
2030 if (e != s) {
2031 len--;
2032 warn_and_return:
9b387841
NC
2033 if (PL_op)
2034 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2035 "%s in %s", unees, OP_DESC(PL_op));
2036 else
61a12c31 2037 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2038 }
2039
2040 return len;
2041}
2042
b06226ff 2043/*
fed3ba5d
NC
2044=for apidoc bytes_cmp_utf8
2045
a1433954 2046Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2047sequence of characters (stored as UTF-8)
2048in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2049equal, -1 or -2 if the first string is less than the second string, +1 or +2
2050if the first string is greater than the second string.
2051
2052-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2053longer string. -2 or +2 is returned if
2054there was a difference between characters
fed3ba5d
NC
2055within the strings.
2056
2057=cut
2058*/
2059
2060int
2061Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2062{
2063 const U8 *const bend = b + blen;
2064 const U8 *const uend = u + ulen;
2065
2066 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2067
2068 while (b < bend && u < uend) {
2069 U8 c = *u++;
2070 if (!UTF8_IS_INVARIANT(c)) {
2071 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2072 if (u < uend) {
2073 U8 c1 = *u++;
2074 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2075 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2076 } else {
2b5e7bc2 2077 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2078 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2079 "%s %s%s",
2080 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2081 PL_op ? " in " : "",
2082 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2083 return -2;
2084 }
2085 } else {
2086 if (PL_op)
2087 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2088 "%s in %s", unees, OP_DESC(PL_op));
2089 else
61a12c31 2090 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2091 return -2; /* Really want to return undef :-) */
2092 }
2093 } else {
2094 return -2;
2095 }
2096 }
2097 if (*b != c) {
2098 return *b < c ? -2 : +2;
2099 }
2100 ++b;
2101 }
2102
2103 if (b == bend && u == uend)
2104 return 0;
2105
2106 return b < bend ? +1 : -1;
2107}
2108
2109/*
87cea99e 2110=for apidoc utf8_to_bytes
6940069f 2111
3bc0c78c 2112Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2113Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2114updates C<*lenp> to contain the new length.
3bc0c78c
KW
2115Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2116
2117Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2118having saved the value of C<*lenp> before the call, and subtracting the
2119after-call value of C<*lenp> from it.
6940069f 2120
a1433954 2121If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2122
6940069f
GS
2123=cut
2124*/
2125
2126U8 *
09af0336 2127Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2128{
9fe0d3c2 2129 U8 * first_variant;
246fae53 2130
7918f24d 2131 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2132 PERL_UNUSED_CONTEXT;
7918f24d 2133
9fe0d3c2 2134 /* This is a no-op if no variants at all in the input */
09af0336 2135 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2136 return s;
2137 }
2138
2139 {
3c5aa262 2140 U8 * const save = s;
09af0336 2141 U8 * const send = s + *lenp;
3c5aa262
KW
2142 U8 * d;
2143
2144 /* Nothing before the first variant needs to be changed, so start the real
2145 * work there */
2146 s = first_variant;
2147 while (s < send) {
2148 if (! UTF8_IS_INVARIANT(*s)) {
2149 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2150 *lenp = ((STRLEN) -1);
3c5aa262
KW
2151 return 0;
2152 }
2153 s++;
d59937ca
KW
2154 }
2155 s++;
dcad2880 2156 }
dcad2880 2157
3c5aa262
KW
2158 /* Is downgradable, so do it */
2159 d = s = first_variant;
2160 while (s < send) {
2161 U8 c = *s++;
2162 if (! UVCHR_IS_INVARIANT(c)) {
2163 /* Then it is two-byte encoded */
2164 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2165 s++;
2166 }
2167 *d++ = c;
2168 }
2169 *d = '\0';
09af0336 2170 *lenp = d - save;
3c5aa262
KW
2171
2172 return save;
9fe0d3c2 2173 }
6940069f
GS
2174}
2175
2176/*
87cea99e 2177=for apidoc bytes_from_utf8
f9a63242 2178
09af0336 2179Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2180byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2181actually encoded in UTF-8.
2182
2183Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2184the input string.
2185
41ae6089
KW
2186Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2187not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2188C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2189
41ae6089 2190Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2191newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2192returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2193
3bc0c78c 2194Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2195having saved the value of C<*lenp> before the call, and subtracting the
2196after-call value of C<*lenp> from it.
3bc0c78c 2197
37607a96 2198=cut
976c1b08
KW
2199
2200There is a macro that avoids this function call, but this is retained for
2201anyone who calls it with the Perl_ prefix */
f9a63242
JH
2202
2203U8 *
41ae6089 2204Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2205{
7918f24d 2206 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2207 PERL_UNUSED_CONTEXT;
f9a63242 2208
976c1b08
KW
2209 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2210}
2211
2212/*
2213No = here because currently externally undocumented
2214for apidoc bytes_from_utf8_loc
2215
2216Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2217to store the location of the first character in C<"s"> that cannot be
2218converted to non-UTF8.
2219
2220If that parameter is C<NULL>, this function behaves identically to
2221C<bytes_from_utf8>.
2222
2223Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2224C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2225
2226Otherwise, the function returns a newly created C<NUL>-terminated string
2227containing the non-UTF8 equivalent of the convertible first portion of
2228C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2229If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2230and C<*first_non_downgradable> is set to C<NULL>.
2231
2232Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2233first character in the original string that wasn't converted. C<*is_utf8p> is
2234unchanged. Note that the new string may have length 0.
2235
2236Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2237C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2238converts as many characters in it as possible stopping at the first one it
385b74be 2239finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2240set to point to that. The function returns the portion that could be converted
2241in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2242not including the terminating C<NUL>. If the very first character in the
2243original could not be converted, C<*lenp> will be 0, and the new string will
2244contain just a single C<NUL>. If the entire input string was converted,
2245C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2246
2247Upon successful return, the number of variants in the converted portion of the
2248string can be computed by having saved the value of C<*lenp> before the call,
2249and subtracting the after-call value of C<*lenp> from it.
2250
2251=cut
2252
2253
2254*/
2255
2256U8 *
2257Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2258{
2259 U8 *d;
2260 const U8 *original = s;
2261 U8 *converted_start;
2262 const U8 *send = s + *lenp;
f9a63242 2263
976c1b08 2264 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2265
976c1b08
KW
2266 if (! *is_utf8p) {
2267 if (first_unconverted) {
2268 *first_unconverted = NULL;
2269 }
2270
2271 return (U8 *) original;
2272 }
2273
2274 Newx(d, (*lenp) + 1, U8);
2275
2276 converted_start = d;
7299a045
KW
2277 while (s < send) {
2278 U8 c = *s++;
2279 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2280
2281 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2282 * have to stop now */
2283 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2284 if (first_unconverted) {
2285 *first_unconverted = s - 1;
2286 goto finish_and_return;
2287 }
2288 else {
2289 Safefree(converted_start);
2290 return (U8 *) original;
2291 }
2292 }
2293
7299a045
KW
2294 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2295 s++;
38af28cf 2296 }
7299a045
KW
2297 *d++ = c;
2298 }
170a1c22 2299
976c1b08
KW
2300 /* Here, converted the whole of the input */
2301 *is_utf8p = FALSE;
2302 if (first_unconverted) {
2303 *first_unconverted = NULL;
170a1c22 2304 }
976c1b08
KW
2305
2306 finish_and_return:
2307 *d = '\0';
2308 *lenp = d - converted_start;
2309
2310 /* Trim unused space */
2311 Renew(converted_start, *lenp + 1, U8);
2312
2313 return converted_start;
f9a63242
JH
2314}
2315
2316/*
87cea99e 2317=for apidoc bytes_to_utf8
6940069f 2318
09af0336 2319Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2320UTF-8.
09af0336 2321Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2322reflect the new length in bytes.
6940069f 2323
3bc0c78c 2324Upon successful return, the number of variants in the string can be computed by
23b37b12 2325having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2326after-call value of C<*lenp>.
2327
75200dff 2328A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2329
2330If you want to convert to UTF-8 from encodings other than
2331the native (Latin1 or EBCDIC),
a1433954 2332see L</sv_recode_to_utf8>().
c9ada85f 2333
497711e7 2334=cut
6940069f
GS
2335*/
2336
2337U8*
09af0336 2338Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2339{
09af0336 2340 const U8 * const send = s + (*lenp);
6940069f
GS
2341 U8 *d;
2342 U8 *dst;
7918f24d
NC
2343
2344 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2345 PERL_UNUSED_CONTEXT;
6940069f 2346
09af0336 2347 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2348 dst = d;
2349
2350 while (s < send) {
55d09dc8
KW
2351 append_utf8_from_native_byte(*s, &d);
2352 s++;
6940069f
GS
2353 }
2354 *d = '\0';
09af0336 2355 *lenp = d-dst;
6940069f
GS
2356 return dst;
2357}
2358
a0ed51b3 2359/*
dea0fc0b 2360 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2361 *
2362 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2363 * We optimize for native, for obvious reasons. */
2364
2365U8*
dea0fc0b 2366Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2367{
dea0fc0b
JH
2368 U8* pend;
2369 U8* dstart = d;
2370
7918f24d
NC
2371 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2372
dea0fc0b 2373 if (bytelen & 1)
56576a04
KW
2374 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2375 (UV)bytelen);
dea0fc0b
JH
2376
2377 pend = p + bytelen;
2378
a0ed51b3 2379 while (p < pend) {
dea0fc0b
JH
2380 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2381 p += 2;
2d1545e5 2382 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2383 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2384 continue;
2385 }
56d37426
KW
2386 if (uv <= MAX_UTF8_TWO_BYTE) {
2387 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2388 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2389 continue;
2390 }
46956fad
KW
2391#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2392#define LAST_HIGH_SURROGATE 0xDBFF
2393#define FIRST_LOW_SURROGATE 0xDC00
2394#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2395
2396 /* This assumes that most uses will be in the first Unicode plane, not
2397 * needing surrogates */
2398 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2399 && uv <= UNICODE_SURROGATE_LAST))
2400 {
2401 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2402 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2403 }
2404 else {
01ea242b 2405 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2406 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2407 || UNLIKELY(low > LAST_LOW_SURROGATE))
2408 {
01ea242b 2409 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2410 }
2411 p += 2;
46956fad
KW
2412 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2413 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2414 }
a0ed51b3 2415 }
56d37426
KW
2416#ifdef EBCDIC
2417 d = uvoffuni_to_utf8_flags(d, uv, 0);
2418#else
a0ed51b3 2419 if (uv < 0x10000) {
eb160463
GS
2420 *d++ = (U8)(( uv >> 12) | 0xe0);
2421 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2422 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2423 continue;
2424 }
2425 else {
eb160463
GS
2426 *d++ = (U8)(( uv >> 18) | 0xf0);
2427 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2428 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2429 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2430 continue;
2431 }
56d37426 2432#endif
a0ed51b3 2433 }
dea0fc0b 2434 *newlen = d - dstart;
a0ed51b3
LW
2435 return d;
2436}
2437
2438/* Note: this one is slightly destructive of the source. */
2439
2440U8*
dea0fc0b 2441Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2442{
2443 U8* s = (U8*)p;
d4c19fe8 2444 U8* const send = s + bytelen;
7918f24d
NC
2445
2446 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2447
e0ea5e2d 2448 if (bytelen & 1)
147e3846 2449 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2450 (UV)bytelen);
2451
a0ed51b3 2452 while (s < send) {
d4c19fe8 2453 const U8 tmp = s[0];
a0ed51b3
LW
2454 s[0] = s[1];
2455 s[1] = tmp;
2456 s += 2;
2457 }
dea0fc0b 2458 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2459}
2460
922e8cb4
KW
2461bool
2462Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2463{
2464 U8 tmpbuf[UTF8_MAXBYTES+1];
2465 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2466 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2467}
2468
f9ae8fb6
JD
2469/* Internal function so we can deprecate the external one, and call
2470 this one from other deprecated functions in this file */
2471
f2645549
KW
2472bool
2473Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2474{
f2645549 2475 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2476
2477 if (*p == '_')
2478 return TRUE;
f25ce844 2479 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2480}
2481
5092f92a 2482bool
eba68aa0
KW
2483Perl__is_uni_perl_idcont(pTHX_ UV c)
2484{
2485 U8 tmpbuf[UTF8_MAXBYTES+1];
2486 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2487 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2488}
2489
2490bool
f91dcd13
KW
2491Perl__is_uni_perl_idstart(pTHX_ UV c)
2492{
2493 U8 tmpbuf[UTF8_MAXBYTES+1];
2494 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2495 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2496}
2497
3a4c58c9 2498UV
56576a04
KW
2499Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2500 const char S_or_s)
3a4c58c9
KW
2501{
2502 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2503 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2504 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2505 * either "SS" or "Ss". Which one to use is passed into the routine in
2506 * 'S_or_s' to avoid a test */
2507
2508 UV converted = toUPPER_LATIN1_MOD(c);
2509
2510 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2511
2512 assert(S_or_s == 'S' || S_or_s == 's');
2513
6f2d5cbc 2514 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2515 characters in this range */
3a4c58c9
KW
2516 *p = (U8) converted;
2517 *lenp = 1;
2518 return converted;
2519 }
2520
2521 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2522 * which it maps to one of them, so as to only have to have one check for
2523 * it in the main case */
2524 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2525 switch (c) {
2526 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2527 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2528 break;
2529 case MICRO_SIGN:
2530 converted = GREEK_CAPITAL_LETTER_MU;
2531 break;
79e064b9
KW
2532#if UNICODE_MAJOR_VERSION > 2 \
2533 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2534 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2535 case LATIN_SMALL_LETTER_SHARP_S:
2536 *(p)++ = 'S';
2537 *p = S_or_s;
2538 *lenp = 2;
2539 return 'S';
79e064b9 2540#endif
3a4c58c9 2541 default:
56576a04
KW
2542 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2543 " '%c' to map to '%c'",
2544 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2545 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2546 }
2547 }
2548
2549 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2550 *p = UTF8_TWO_BYTE_LO(converted);
2551 *lenp = 2;
2552
2553 return converted;
2554}
2555
50bda2c3
KW
2556/* Call the function to convert a UTF-8 encoded character to the specified case.
2557 * Note that there may be more than one character in the result.
2558 * INP is a pointer to the first byte of the input character
2559 * OUTP will be set to the first byte of the string of changed characters. It
2560 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2561 * LENP will be set to the length in bytes of the string of changed characters
2562 *
56576a04
KW
2563 * The functions return the ordinal of the first character in the string of
2564 * OUTP */
2565#define CALL_UPPER_CASE(uv, s, d, lenp) \
2566 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2567#define CALL_TITLE_CASE(uv, s, d, lenp) \
2568 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2569#define CALL_LOWER_CASE(uv, s, d, lenp) \
2570 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2571
b9992569
KW
2572/* This additionally has the input parameter 'specials', which if non-zero will
2573 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2574 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2575#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2576_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2577
84afefe6
JH
2578UV
2579Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2580{
a1433954
KW
2581 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2582 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2583 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2584 * the changed version may be longer than the original character.
2585 *
2586 * The ordinal of the first character of the changed version is returned
2587 * (but note, as explained above, that there may be more.) */
2588
7918f24d
NC
2589 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2590
3a4c58c9
KW
2591 if (c < 256) {
2592 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2593 }
2594
0ebc6274 2595 uvchr_to_utf8(p, c);
b9992569 2596 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2597}
2598
84afefe6
JH
2599UV
2600Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2601{
7918f24d
NC
2602 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2603
3a4c58c9
KW
2604 if (c < 256) {
2605 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2606 }
2607
0ebc6274 2608 uvchr_to_utf8(p, c);
b9992569 2609 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2610}
2611
afc16117 2612STATIC U8
eaf412bf 2613S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2614{
2615 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2616 * those, converting the result to UTF-8. Since the result is always just
a1433954 2617 * one character, we allow <p> to be NULL */
afc16117
KW
2618
2619 U8 converted = toLOWER_LATIN1(c);
2620
eaf412bf
KW
2621 PERL_UNUSED_ARG(dummy);
2622
afc16117 2623 if (p != NULL) {
6f2d5cbc 2624 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2625 *p = converted;
2626 *lenp = 1;
2627 }
2628 else {
430c9760
KW
2629 /* Result is known to always be < 256, so can use the EIGHT_BIT
2630 * macros */
2631 *p = UTF8_EIGHT_BIT_HI(converted);
2632 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2633 *lenp = 2;
2634 }
2635 }
2636 return converted;
2637}
2638
84afefe6
JH
2639UV
2640Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2641{
7918f24d
NC
2642 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2643
afc16117 2644 if (c < 256) {
eaf412bf 2645 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2646 }
2647
afc16117 2648 uvchr_to_utf8(p, c);
b9992569 2649 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2650}
2651
84afefe6 2652UV
56576a04
KW
2653Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2654 const unsigned int flags)
a1dde8de 2655{
51910141 2656 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2657 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2658 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2659 *
2660 * Not to be used for locale folds
51910141 2661 */
f673fad4 2662
a1dde8de
KW
2663 UV converted;
2664
2665 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2666 PERL_UNUSED_CONTEXT;
a1dde8de 2667
1ca267a5
KW
2668 assert (! (flags & FOLD_FLAGS_LOCALE));
2669
659a7c2d 2670 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2671 converted = GREEK_SMALL_LETTER_MU;
2672 }
9b63e895
KW
2673#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2674 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2675 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2676 else if ( (flags & FOLD_FLAGS_FULL)
2677 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2678 {
1ca267a5
KW
2679 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2680 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2681 * under those circumstances. */
2682 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2683 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2684 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2685 p, *lenp, U8);
2686 return LATIN_SMALL_LETTER_LONG_S;
2687 }
2688 else {
4f489194
KW
2689 *(p)++ = 's';
2690 *p = 's';
2691 *lenp = 2;
2692 return 's';
1ca267a5 2693 }
a1dde8de 2694 }
9b63e895 2695#endif
a1dde8de
KW
2696 else { /* In this range the fold of all other characters is their lower
2697 case */
2698 converted = toLOWER_LATIN1(c);
2699 }
2700
6f2d5cbc 2701 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2702 *p = (U8) converted;
2703 *lenp = 1;
2704 }
2705 else {
2706 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2707 *p = UTF8_TWO_BYTE_LO(converted);
2708 *lenp = 2;
2709 }
2710
2711 return converted;
2712}
2713
2714UV
31f05a37 2715Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2716{
4b593389 2717
a0270393
KW
2718 /* Not currently externally documented, and subject to change
2719 * <flags> bits meanings:
2720 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2721 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2722 * locale are to be used.
a0270393
KW
2723 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2724 */
4b593389 2725
36bb2ab6 2726 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2727
780fcc9f
KW
2728 if (flags & FOLD_FLAGS_LOCALE) {
2729 /* Treat a UTF-8 locale as not being in locale at all */
2730 if (IN_UTF8_CTYPE_LOCALE) {
2731 flags &= ~FOLD_FLAGS_LOCALE;
2732 }
2733 else {
2734 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2735 goto needs_full_generality;
780fcc9f 2736 }
31f05a37
KW
2737 }
2738
a1dde8de 2739 if (c < 256) {
e7b7ac46 2740 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2741 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2742 }
2743
2f306ab9 2744 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2745 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2746 uvchr_to_utf8(p, c);
b9992569 2747 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2748 }
567b353c 2749 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2750 the special flags. */
2751 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2752
2753 needs_full_generality:
a0270393 2754 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2755 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2756 p, lenp, flags);
a0270393 2757 }
84afefe6
JH
2758}
2759
26483009 2760PERL_STATIC_INLINE bool
5141f98e 2761S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2762 const char *const swashname, SV* const invlist)
bde6a22d 2763{
ea317ccb
KW
2764 /* returns a boolean giving whether or not the UTF8-encoded character that
2765 * starts at <p> is in the swash indicated by <swashname>. <swash>
2766 * contains a pointer to where the swash indicated by <swashname>
2767 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2768 * look at <*swash> and only generate a swash if it is not null. <invlist>
2769 * is NULL or an inversion list that defines the swash. If not null, it
2770 * saves time during initialization of the swash.
ea317ccb
KW
2771 *
2772 * Note that it is assumed that the buffer length of <p> is enough to
2773 * contain all the bytes that comprise the character. Thus, <*p> should
2774 * have been checked before this call for mal-formedness enough to assure
2775 * that. */
2776
7918f24d
NC
2777 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2778
492a624f 2779 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2780 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2781 * as far as there being enough bytes available in it to accommodate the
2782 * character without reading beyond the end, and pass that number on to the
2783 * validating routine */
6302f837 2784 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2785 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2786 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2787 1 /* Die */ );
2788 NOT_REACHED; /* NOTREACHED */
28123549 2789 }
86ae6e94 2790
87367d5f
KW
2791 if (!*swash) {
2792 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2793 *swash = _core_swash_init("utf8",
2794
2795 /* Only use the name if there is no inversion
2796 * list; otherwise will go out to disk */
2797 (invlist) ? "" : swashname,
2798
2799 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2800 }
28123549 2801
bde6a22d
NC
2802 return swash_fetch(*swash, p, TRUE) != 0;
2803}
2804
da8c1a98 2805PERL_STATIC_INLINE bool
56576a04
KW
2806S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2807 SV **swash, const char *const swashname,
2808 SV* const invlist)
da8c1a98
KW
2809{
2810 /* returns a boolean giving whether or not the UTF8-encoded character that
2811 * starts at <p>, and extending no further than <e - 1> is in the swash
2812 * indicated by <swashname>. <swash> contains a pointer to where the swash
2813 * indicated by <swashname> is to be stored; which this routine will do, so
2814 * that future calls will look at <*swash> and only generate a swash if it
2815 * is not null. <invlist> is NULL or an inversion list that defines the
2816 * swash. If not null, it saves time during initialization of the swash.
2817 */
2818
2819 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2820
2821 if (! isUTF8_CHAR(p, e)) {
2822 _force_out_malformed_utf8_message(p, e, 0, 1);
2823 NOT_REACHED; /* NOTREACHED */
2824 }
2825
2826 if (!*swash) {
2827 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2828 *swash = _core_swash_init("utf8",
2829
2830 /* Only use the name if there is no inversion
2831 * list; otherwise will go out to disk */
2832 (invlist) ? "" : swashname,
2833
2834 &PL_sv_undef, 1, 0, invlist, &flags);
2835 }
2836
2837 return swash_fetch(*swash, p, TRUE) != 0;
2838}
2839
34aeb2e9
KW
2840STATIC void
2841S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2842 const char * const alternative,
2843 const bool use_locale,
2844 const char * const file,
2845 const unsigned line)
2846{
2847 const char * key;
2848
2849 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2850
2851 if (ckWARN_d(WARN_DEPRECATED)) {
2852
2853 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2854 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2855 if (! PL_seen_deprecated_macro) {
2856 PL_seen_deprecated_macro = newHV();
2857 }
2858 if (! hv_store(PL_seen_deprecated_macro, key,
2859 strlen(key), &PL_sv_undef, 0))
2860 {
2861 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2862 }
2863
c44e9413 2864 if (instr(file, "mathoms.c")) {
607313a1
KW
2865 Perl_warner(aTHX_ WARN_DEPRECATED,
2866 "In %s, line %d, starting in Perl v5.30, %s()"
2867 " will be removed. Avoid this message by"
2868 " converting to use %s().\n",
2869 file, line, name, alternative);
2870 }
2871 else {
34aeb2e9
KW
2872 Perl_warner(aTHX_ WARN_DEPRECATED,
2873 "In %s, line %d, starting in Perl v5.30, %s() will"
2874 " require an additional parameter. Avoid this"
2875 " message by converting to use %s().\n",
2876 file, line, name, alternative);
607313a1 2877 }
34aeb2e9
KW
2878 }
2879 }
2880}
2881
bde6a22d 2882bool
34aeb2e9 2883Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2884 const U8 * const p,
34aeb2e9
KW
2885 const char * const name,
2886 const char * const alternative,
2887 const bool use_utf8,
2888 const bool use_locale,
2889 const char * const file,
2890 const unsigned line)
922e8cb4 2891{
922e8cb4
KW
2892 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2893
34aeb2e9
KW
2894 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2895
2896 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2897
2898 switch (classnum) {
2899 case _CC_WORDCHAR:
2900 case _CC_DIGIT:
2901 case _CC_ALPHA:
2902 case _CC_LOWER:
2903 case _CC_UPPER:
2904 case _CC_PUNCT:
2905 case _CC_PRINT:
2906 case _CC_ALPHANUMERIC:
2907 case _CC_GRAPH:
2908 case _CC_CASED:
2909
2910 return is_utf8_common(p,
2911 &PL_utf8_swash_ptrs[classnum],
2912 swash_property_names[classnum],
2913 PL_XPosix_ptrs[classnum]);
2914
2915 case _CC_SPACE:
2916 return is_XPERLSPACE_high(p);
2917 case _CC_BLANK:
2918 return is_HORIZWS_high(p);
2919 case _CC_XDIGIT:
2920 return is_XDIGIT_high(p);
2921 case _CC_CNTRL:
2922 return 0;
2923 case _CC_ASCII:
2924 return 0;
2925 case _CC_VERTSPACE:
2926 return is_VERTWS_high(p);
2927 case _CC_IDFIRST:
2928 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2929 PL_utf8_perl_idstart
2930 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2931 }
22f0498f
KW
2932 return is_utf8_common(p, &PL_utf8_perl_idstart,
2933 "_Perl_IDStart", NULL);
34aeb2e9
KW
2934 case _CC_IDCONT:
2935 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2936 PL_utf8_perl_idcont
2937 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2938 }
22f0498f
KW
2939 return is_utf8_common(p, &PL_utf8_perl_idcont,
2940 "_Perl_IDCont", NULL);
34aeb2e9
KW
2941 }
2942 }
2943
2944 /* idcont is the same as wordchar below 256 */
2945 if (classnum == _CC_IDCONT) {
2946 classnum = _CC_WORDCHAR;
2947 }
2948 else if (classnum == _CC_IDFIRST) {
2949 if (*p == '_') {
2950 return TRUE;
2951 }
2952 classnum = _CC_ALPHA;
2953 }
2954
2955 if (! use_locale) {
2956 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2957 return _generic_isCC(*p, classnum);
2958 }
922e8cb4 2959
34aeb2e9
KW
2960 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2961 }
2962 else {
2963 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2964 return isFOO_lc(classnum, *p);
2965 }
2966
2967 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2968 }
2969
2970 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2971}
2972
2973bool
da8c1a98
KW
2974Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2975 const U8 * const e)
2976{
2977 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2978
2979 assert(classnum < _FIRST_NON_SWASH_CC);
2980
2981 return is_utf8_common_with_len(p,
2982 e,
2983 &PL_utf8_swash_ptrs[classnum],
2984 swash_property_names[classnum],
2985 PL_XPosix_ptrs[classnum]);
2986}
2987
2988bool
da8c1a98
KW
2989Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2990{
2991 SV* invlist = NULL;
2992
2993 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2994
2995 if (! PL_utf8_perl_idstart) {
2996 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2997 }
2998 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2999 "_Perl_IDStart", invlist);
3000}
3001
3002bool
f2645549 3003Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3004{
f2645549 3005 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3006
3007 if (*p == '_')
3008 return TRUE;
f25ce844 3009 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3010}
3011
3012bool
da8c1a98
KW
3013Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3014{
3015 SV* invlist = NULL;
3016
3017 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3018
3019 if (! PL_utf8_perl_idcont) {
3020 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3021 }
3022 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3023 "_Perl_IDCont", invlist);
3024}
3025
3026bool
f2645549 3027Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3028{
f2645549 3029 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3030
f25ce844 3031 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3032}
3033
3034bool
f2645549 3035Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3036{
f2645549 3037 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3038
f25ce844 3039 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
3040}
3041
3042bool
7dbf68d2
KW
3043Perl__is_utf8_mark(pTHX_ const U8 *p)
3044{
7dbf68d2
KW
3045 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3046
f25ce844 3047 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3048}
3049
b9992569 3050 /* change namve uv1 to 'from' */
6a4a25f4 3051STATIC UV
b9992569
KW
3052S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
3053 SV **swashp, const char *normal, const char *special)
3054{
0134edef 3055 STRLEN len = 0;
7918f24d 3056
b9992569 3057 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3058
36eaa811
KW
3059 /* For code points that don't change case, we already know that the output
3060 * of this function is the unchanged input, so we can skip doing look-ups
3061 * for them. Unfortunately the case-changing code points are scattered
3062 * around. But there are some long consecutive ranges where there are no
3063 * case changing code points. By adding tests, we can eliminate the lookup
3064 * for all the ones in such ranges. This is currently done here only for
3065 * just a few cases where the scripts are in common use in modern commerce
3066 * (and scripts adjacent to those which can be included without additional
3067 * tests). */
3068
3069 if (uv1 >= 0x0590) {
3070 /* This keeps from needing further processing the code points most
3071 * likely to be used in the following non-cased scripts: Hebrew,
3072 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3073 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3074 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3075 if (uv1 < 0x10A0) {
3076 goto cases_to_self;
3077 }
3078
3079 /* The following largish code point ranges also don't have case
3080 * changes, but khw didn't think they warranted extra tests to speed
3081 * them up (which would slightly slow down everything else above them):
3082 * 1100..139F Hangul Jamo, Ethiopic
3083 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3084 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3085 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3086 * Combining Diacritical Marks Extended, Balinese,
3087 * Sundanese, Batak, Lepcha, Ol Chiki
3088 * 2000..206F General Punctuation
3089 */
3090
3091 if (uv1 >= 0x2D30) {
3092
3093 /* This keeps the from needing further processing the code points
3094 * most likely to be used in the following non-cased major scripts:
3095 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3096 *
3097 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3098 * event that Unicode eventually allocates the unused block as of
3099 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3100 * that the test suite will start having failures to alert you
3101 * should that happen) */
3102 if (uv1 < 0xA640) {
3103 goto cases_to_self;
3104 }
3105
3106 if (uv1 >= 0xAC00) {
3107 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3108 if (ckWARN_d(WARN_SURROGATE)) {
3109 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3110 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3111 "Operation \"%s\" returns its argument for"
3112 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3113 }
3114 goto cases_to_self;
3115 }
36eaa811
KW
3116
3117 /* AC00..FAFF Catches Hangul syllables and private use, plus
3118 * some others */
3119 if (uv1 < 0xFB00) {
3120 goto cases_to_self;
3121
3122 }
3123
5af9bc97 3124 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3125 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3126 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3127 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3128 }
3129 if (ckWARN_d(WARN_NON_UNICODE)) {
3130 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3131 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3132 "Operation \"%s\" returns its argument for"
3133 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3134 }
3135 goto cases_to_self;
3136 }
3bfc1e70
KW
3137#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3138 if (UNLIKELY(uv1
3139 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3140 {
3141
56576a04
KW
3142 /* As of Unicode 10.0, this means we avoid swash creation
3143 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3144 goto cases_to_self;
3145 }
3146#endif
36eaa811
KW
3147 }
3148 }
9ae3ac1a 3149
36eaa811
KW
3150 /* Note that non-characters are perfectly legal, so no warning should
3151 * be given. There are so few of them, that it isn't worth the extra
3152 * tests to avoid swash creation */
9ae3ac1a
KW
3153 }
3154
0134edef 3155 if (!*swashp) /* load on-demand */
56576a04
KW
3156 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3157 4, 0, NULL, NULL);
0134edef 3158
a6f87d8c 3159 if (special) {
0134edef 3160 /* It might be "special" (sometimes, but not always,
2a37f04d 3161 * a multicharacter mapping) */
4a8240a3 3162 HV *hv = NULL;
b08cf34e
JH
3163 SV **svp;
3164
4a8240a3
KW
3165 /* If passed in the specials name, use that; otherwise use any
3166 * given in the swash */
3167 if (*special != '\0') {
3168 hv = get_hv(special, 0);
3169 }
3170 else {
3171 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3172 if (svp) {
3173 hv = MUTABLE_HV(SvRV(*svp));
3174 }
3175 }
3176
176fe009 3177 if (hv
5f560d8a 3178 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3179 && (*svp))
3180 {
cfd0369c 3181 const char *s;
47654450 3182
cfd0369c 3183 s = SvPV_const(*svp, len);
47654450 3184 if (len == 1)
f4cd282c 3185 /* EIGHTBIT */
c80e42f3 3186 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3187 else {
d2dcd0fb 3188 Copy(s, ustrp, len, U8);
29e98929 3189 }
983ffd37 3190 }
0134edef
JH
3191 }
3192
3193 if (!len && *swashp) {
4a4088c4 3194 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3195
0134edef
JH
3196 if (uv2) {
3197 /* It was "normal" (a single character mapping). */
f4cd282c 3198 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3199 }
3200 }
1feea2c7 3201
cbe07460
KW
3202 if (len) {
3203 if (lenp) {
3204 *lenp = len;
3205 }
3206 return valid_utf8_to_uvchr(ustrp, 0);
3207 }
3208
3209 /* Here, there was no mapping defined, which means that the code point maps
3210 * to itself. Return the inputs */
e24dfe9c 3211 cases_to_self:
bfdf22ec 3212 len = UTF8SKIP(p);
ca9fab46
KW
3213 if (p != ustrp) { /* Don't copy onto itself */
3214 Copy(p, ustrp, len, U8);
3215 }
0134edef 3216
2a37f04d
JH
3217 if (lenp)
3218 *lenp = len;
3219
f4cd282c 3220 return uv1;
cbe07460 3221
a0ed51b3
LW
3222}
3223
051a06d4 3224STATIC UV
56576a04
KW
3225S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3226 U8* const ustrp, STRLEN *lenp)
051a06d4 3227{
4a4088c4 3228 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3229 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3230 * result contains a character that crosses the 255/256 boundary, disallow
3231 * the change, and return the original code point. See L<perlfunc/lc> for
3232 * why;
051a06d4 3233 *
a1433954
KW
3234 * p points to the original string whose case was changed; assumed
3235 * by this routine to be well-formed
051a06d4 3236 * result the code point of the first character in the changed-case string
56576a04
KW
3237 * ustrp points to the changed-case string (<result> represents its
3238 * first char)
051a06d4
KW
3239 * lenp points to the length of <ustrp> */
3240
3241 UV original; /* To store the first code point of <p> */
3242
3243 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3244
a4f12ed7 3245 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3246
3247 /* We know immediately if the first character in the string crosses the
3248 * boundary, so can skip */
3249 if (result > 255) {
3250
3251 /* Look at every character in the result; if any cross the
3252 * boundary, the whole thing is disallowed */
3253 U8* s = ustrp + UTF8SKIP(ustrp);
3254 U8* e = ustrp + *lenp;
3255 while (s < e) {
a4f12ed7 3256 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3257 goto bad_crossing;
3258 }
3259 s += UTF8SKIP(s);
3260 }
3261
613abc6d
KW
3262 /* Here, no characters crossed, result is ok as-is, but we warn. */
3263 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3264 return result;
3265 }
3266
7b52d656 3267 bad_crossing:
051a06d4
KW
3268
3269 /* Failed, have to return the original */
4b88fb76 3270 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3271
3272 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3273 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3274 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3275 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3276 OP_DESC(PL_op),
ab0b796c
KW
3277 original,
3278 original);
051a06d4
KW
3279 Copy(p, ustrp, *lenp, char);
3280 return original;
3281}
3282
607313a1
KW
3283STATIC U32
3284S_check_and_deprecate(pTHX_ const U8 *p,
3285 const U8 **e,
3286 const unsigned int type, /* See below */
3287 const bool use_locale, /* Is this a 'LC_'
3288 macro call? */
3289 const char * const file,
3290 const unsigned line)
3291{
3292 /* This is a temporary function to deprecate the unsafe calls to the case
3293 * changing macros and functions. It keeps all the special stuff in just
3294 * one place.
3295 *
3296 * It updates *e with the pointer to the end of the input string. If using
3297 * the old-style macros, *e is NULL on input, and so this function assumes
3298 * the input string is long enough to hold the entire UTF-8 sequence, and
3299 * sets *e accordingly, but it then returns a flag to pass the
3300 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3301 * using the full length if possible.
3302 *
3303 * It also does the assert that *e > p when *e is not NULL. This should be
3304 * migrated to the callers when this function gets deleted.
3305 *
3306 * The 'type' parameter is used for the caller to specify which case
3307 * changing function this is called from: */
3308
3309# define DEPRECATE_TO_UPPER 0
3310# define DEPRECATE_TO_TITLE 1
3311# define DEPRECATE_TO_LOWER 2
3312# define DEPRECATE_TO_FOLD 3
3313
3314 U32 utf8n_flags = 0;
3315 const char * name;
3316 const char * alternative;
3317
3318 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3319
3320 if (*e == NULL) {
3321 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3322 *e = p + UTF8SKIP(p);
3323
3324 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3325 * there. It could be part of a larger path */
607313a1 3326 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3327 name = instr(file, "mathoms.c")
607313a1
KW
3328 ? "to_utf8_upper"
3329 : "toUPPER_utf8";
3330 alternative = "toUPPER_utf8_safe";
3331 }
3332 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3333 name = instr(file, "mathoms.c")
607313a1
KW
3334 ? "to_utf8_title"
3335 : "toTITLE_utf8";
3336 alternative = "toTITLE_utf8_safe";
3337 }
3338 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3339 name = instr(file, "mathoms.c")
607313a1
KW
3340 ? "to_utf8_lower"
3341 : "toLOWER_utf8";
3342 alternative = "toLOWER_utf8_safe";
3343 }
3344 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3345 name = instr(file, "mathoms.c")
607313a1
KW
3346 ? "to_utf8_fold"
3347 : "toFOLD_utf8";
3348 alternative = "toFOLD_utf8_safe";
3349 }
3350 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3351
3352 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3353 }
3354 else {
3355 assert (p < *e);
3356 }
3357
3358 return utf8n_flags;
3359}
3360
eaf412bf
KW
3361/* The process for changing the case is essentially the same for the four case
3362 * change types, except there are complications for folding. Otherwise the
3363 * difference is only which case to change to. To make sure that they all do
3364 * the same thing, the bodies of the functions are extracted out into the
3365 * following two macros. The functions are written with the same variable
3366 * names, and these are known and used inside these macros. It would be
3367 * better, of course, to have inline functions to do it, but since different
3368 * macros are called, depending on which case is being changed to, this is not
3369 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3370 * function can start with the common start macro, then finish with its special
3371 * handling; while the other three cases can just use the common end macro.
3372 *
3373 * The algorithm is to use the proper (passed in) macro or function to change
3374 * the case for code points that are below 256. The macro is used if using
3375 * locale rules for the case change; the function if not. If the code point is
3376 * above 255, it is computed from the input UTF-8, and another macro is called
3377 * to do the conversion. If necessary, the output is converted to UTF-8. If
3378 * using a locale, we have to check that the change did not cross the 255/256
3379 * boundary, see check_locale_boundary_crossing() for further details.
3380 *
3381 * The macros are split with the correct case change for the below-256 case
3382 * stored into 'result', and in the middle of an else clause for the above-255
3383 * case. At that point in the 'else', 'result' is not the final result, but is
3384 * the input code point calculated from the UTF-8. The fold code needs to
3385 * realize all this and take it from there.
3386 *
3387 * If you read the two macros as sequential, it's easier to understand what's
3388 * going on. */
3389#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3390 L1_func_extra_param) \
a239b1e2 3391 \
eaf412bf
KW
3392 if (flags & (locale_flags)) { \
3393 /* Treat a UTF-8 locale as not being in locale at all */ \
3394 if (IN_UTF8_CTYPE_LOCALE) { \
3395 flags &= ~(locale_flags); \
3396 } \
3397 else { \
3398 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3399 } \
3400 } \
3401 \
3402 if (UTF8_IS_INVARIANT(*p)) { \
3403 if (flags & (locale_flags)) { \
3404 result = LC_L1_change_macro(*p); \
3405 } \
3406 else { \
3407 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3408 } \
3409 } \
a239b1e2 3410 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3411 if (flags & (locale_flags)) { \
3412 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3413 *(p+1))); \
3414 } \
3415 else { \
3416 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3417 ustrp, lenp, L1_func_extra_param); \
3418 } \
3419 } \
fa8ab374
KW
3420 else { /* malformed UTF-8 or ord above 255 */ \
3421 STRLEN len_result; \
fa8ab374
KW
3422 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3423 if (len_result == (STRLEN) -1) { \
607313a1
KW
3424 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3425 1 /* Die */ ); \
fa8ab374 3426 }
eaf412bf
KW
3427
3428#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3429 result = change_macro(result, p, ustrp, lenp); \
3430 \
3431 if (flags & (locale_flags)) { \
3432 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3433 } \
3434 return result; \
3435 } \
3436 \
3437 /* Here, used locale rules. Convert back to UTF-8 */ \
3438 if (UTF8_IS_INVARIANT(result)) { \
3439 *ustrp = (U8) result; \
3440 *lenp = 1; \
3441 } \
3442 else { \
3443 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3444 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3445 *lenp = 2; \
3446 } \
3447 \
3448 return result;
3449
d3e79532 3450/*
87cea99e 3451=for apidoc to_utf8_upper
d3e79532 3452
a239b1e2 3453Instead use L</toUPPER_utf8_safe>.
a1433954 3454
d3e79532
JH
3455=cut */
3456
051a06d4 3457/* Not currently externally documented, and subject to change:
31f05a37
KW
3458 * <flags> is set iff iff the rules from the current underlying locale are to
3459 * be used. */
051a06d4 3460
2104c8d9 3461UV
607313a1
KW
3462Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3463 const U8 *e,
3464 U8* ustrp,
3465 STRLEN *lenp,
3466 bool flags,
3467 const char * const file,
3468 const int line)
a0ed51b3 3469{
051a06d4 3470 UV result;
607313a1
KW
3471 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3472 cBOOL(flags), file, line);
051a06d4
KW
3473
3474 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3475
eaf412bf
KW
3476 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3477 /* 2nd char of uc(U+DF) is 'S' */
3478 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3479 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3480}
a0ed51b3 3481
d3e79532 3482/*
87cea99e 3483=for apidoc to_utf8_title
d3e79532 3484
a239b1e2 3485Instead use L</toTITLE_utf8_safe>.
a1433954 3486
d3e79532
JH
3487=cut */
3488
051a06d4 3489/* Not currently externally documented, and subject to change:
31f05a37
KW
3490 * <flags> is set iff the rules from the current underlying locale are to be
3491 * used. Since titlecase is not defined in POSIX, for other than a
3492 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3493 */
051a06d4 3494
983ffd37 3495UV
607313a1
KW
3496Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3497 const U8 *e,
3498 U8* ustrp,
3499 STRLEN *lenp,
3500 bool flags,
3501 const char * const file,
3502 const int line)
983ffd37 3503{
051a06d4 3504 UV result;
607313a1
KW
3505 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3506 cBOOL(flags), file, line);
051a06d4
KW
3507
3508 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3509
eaf412bf
KW
3510 /* 2nd char of ucfirst(U+DF) is 's' */
3511 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3512 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3513}
3514
d3e79532 3515/*
87cea99e 3516=for apidoc to_utf8_lower
d3e79532 3517
a239b1e2 3518Instead use L</toLOWER_utf8_safe>.
a1433954 3519
d3e79532
JH
3520=cut */
3521
051a06d4 3522/* Not currently externally documented, and subject to change:
31f05a37
KW
3523 * <flags> is set iff iff the rules from the current underlying locale are to
3524 * be used.
3525 */
051a06d4 3526
2104c8d9 3527UV
607313a1
KW
3528Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3529 const U8 *e,
3530 U8* ustrp,
3531 STRLEN *lenp,
3532 bool flags,
3533 const char * const file,
3534 const int line)
a0ed51b3 3535{
051a06d4 3536 UV result;
607313a1
KW
3537 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3538 cBOOL(flags), file, line);
051a06d4 3539
051a06d4 3540 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3541
eaf412bf
KW
3542 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3543 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3544}
3545
d3e79532 3546/*
87cea99e 3547=for apidoc to_utf8_fold
d3e79532 3548
a239b1e2 3549Instead use L</toFOLD_utf8_safe>.
a1433954 3550
d3e79532
JH
3551=cut */
3552
051a06d4
KW
3553/* Not currently externally documented, and subject to change,
3554 * in <flags>
31f05a37
KW
3555 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3556 * locale are to be used.
051a06d4
KW
3557 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3558 * otherwise simple folds
a0270393
KW
3559 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3560 * prohibited
445bf929 3561 */
36bb2ab6 3562
b4e400f9 3563UV
607313a1
KW
3564Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3565 const U8 *e,
3566 U8* ustrp,
3567 STRLEN *lenp,
3568 U8 flags,
3569 const char * const file,
3570 const int line)
b4e400f9 3571{
051a06d4 3572 UV result;
607313a1
KW
3573 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3574 cBOOL(flags), file, line);
051a06d4 3575
36bb2ab6 3576 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3577
a0270393
KW
3578 /* These are mutually exclusive */
3579 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3580
50ba90ff
KW
3581 assert(p != ustrp); /* Otherwise overwrites */
3582
eaf412bf
KW
3583 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3584 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3585
eaf412bf 3586 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3587
1ca267a5
KW
3588 if (flags & FOLD_FLAGS_LOCALE) {
3589
76f2ffcd 3590# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3591 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3592
3593# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3594# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3595
3596 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3597
538e84ed
KW
3598 /* Special case these two characters, as what normally gets
3599 * returned under locale doesn't work */
76f2ffcd
KW
3600 if (UTF8SKIP(p) == cap_sharp_s_len
3601 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3602 {
ab0b796c
KW
3603 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3604 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3605 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3606 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3607 goto return_long_s;
3608 }
0766489e
KW
3609 else
3610#endif
3611 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3612 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3613 {
ab0b796c
KW
3614 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3615 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3616 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3617 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3618 goto return_ligature_st;
3619 }
74894415
KW
3620
3621#if UNICODE_MAJOR_VERSION == 3 \
3622 && UNICODE_DOT_VERSION == 0 \
3623 && UNICODE_DOT_DOT_VERSION == 1
3624# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3625
3626 /* And special case this on this Unicode version only, for the same
3627 * reaons the other two are special cased. They would cross the
3628 * 255/256 boundary which is forbidden under /l, and so the code
3629 * wouldn't catch that they are equivalent (which they are only in
3630 * this release) */
3631 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3632 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3633 {
3634 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3635 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3636 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3637 "resolved to \"\\x{0131}\".");
3638 goto return_dotless_i;
3639 }
3640#endif
3641
357aadde 3642 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3643 }
a0270393
KW
3644 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3645 return result;
3646 }
3647 else {
4a4088c4 3648 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3649 * character above the ASCII range, and the result should not
3650 * contain an ASCII character. */
a0270393
KW
3651
3652 UV original; /* To store the first code point of <p> */
3653
3654 /* Look at every character in the result; if any cross the
3655 * boundary, the whole thing is disallowed */
3656 U8* s = ustrp;
3657 U8* e = ustrp + *lenp;
3658 while (s < e) {
3659 if (isASCII(*s)) {
3660 /* Crossed, have to return the original */
3661 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3662
9fc2026f 3663 /* But in these instances, there is an alternative we can
1ca267a5 3664 * return that is valid */
0766489e
KW
3665 if (original == LATIN_SMALL_LETTER_SHARP_S
3666#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3667 || original == LATIN_CAPITAL_LETTER_SHARP_S
3668#endif
3669 ) {
1ca267a5
KW
3670 goto return_long_s;
3671 }
9fc2026f
KW
3672 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3673 goto return_ligature_st;
3674 }
74894415
KW
3675#if UNICODE_MAJOR_VERSION == 3 \
3676 && UNICODE_DOT_VERSION == 0 \
3677 && UNICODE_DOT_DOT_VERSION == 1
3678
3679 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3680 goto return_dotless_i;
3681 }
3682#endif
a0270393
KW
3683 Copy(p, ustrp, *lenp, char);
3684 return original;
3685 }
3686 s += UTF8SKIP(s);
3687 }
051a06d4 3688
a0270393
KW
3689 /* Here, no characters crossed, result is ok as-is */
3690 return result;
3691 }
051a06d4
KW
3692 }
3693
4a4088c4 3694 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3695 if (UTF8_IS_INVARIANT(result)) {
3696 *ustrp = (U8) result;
3697 *lenp = 1;
3698 }
3699 else {
62cb07ea
KW
3700 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3701 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3702 *lenp = 2;
3703 }
3704
051a06d4 3705 return result;
1ca267a5
KW
3706
3707 return_long_s:
3708 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3709 * folds to a string of two of these characters. By returning this
3710 * instead, then, e.g.,
3711 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3712 * works. */
3713
3714 *lenp =