This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
toke.c: Don’t leak memory
[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 954char *
63ab03b3 955Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
956{
957 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 958 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
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 */
63ab03b3
KW
966 const U8 * s = start;
967 const U8 * const e = start + len;
7cf8d05d
KW
968 char * output;
969 char * d;
970
971 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
972
973 Newx(output, output_len, char);
974 SAVEFREEPV(output);
975
976 d = output;
63ab03b3 977 for (s = start; s < e; s++) {
7cf8d05d
KW
978 const unsigned high_nibble = (*s & 0xF0) >> 4;
979 const unsigned low_nibble = (*s & 0x0F);
980
7e2f38b2 981 if (format) {
63ab03b3
KW
982 if (s > start) {
983 *d++ = ' ';
984 }
7e2f38b2
KW
985 }
986 else {
987 *d++ = '\\';
988 *d++ = 'x';
989 }
7cf8d05d
KW
990
991 if (high_nibble < 10) {
992 *d++ = high_nibble + '0';
993 }
994 else {
995 *d++ = high_nibble - 10 + 'a';
996 }
997
998 if (low_nibble < 10) {
999 *d++ = low_nibble + '0';
1000 }
1001 else {
1002 *d++ = low_nibble - 10 + 'a';
1003 }
1004 }
1005
1006 *d = '\0';
1007 return output;
1008}
1009
806547a7 1010PERL_STATIC_INLINE char *
7cf8d05d
KW
1011S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1012
1013 /* How many bytes to print */
3cc6a05e 1014 STRLEN print_len,
7cf8d05d
KW
1015
1016 /* Which one is the non-continuation */
1017 const STRLEN non_cont_byte_pos,
1018
1019 /* How many bytes should there be? */
1020 const STRLEN expect_len)
806547a7
KW
1021{
1022 /* Return the malformation warning text for an unexpected continuation
1023 * byte. */
1024
7cf8d05d 1025 const char * const where = (non_cont_byte_pos == 1)
806547a7 1026 ? "immediately"
7cf8d05d
KW
1027 : Perl_form(aTHX_ "%d bytes",
1028 (int) non_cont_byte_pos);
806547a7
KW
1029
1030 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1031
7cf8d05d
KW
1032 /* We don't need to pass this parameter, but since it has already been
1033 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1034 assert(expect_len == UTF8SKIP(s));
1035
1036 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1037 " %s after start byte 0x%02x; need %d bytes, got %d)",
1038 malformed_text,
7e2f38b2 1039 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
1040 *(s + non_cont_byte_pos),
1041 where,
1042 *s,
1043 (int) expect_len,
1044 (int) non_cont_byte_pos);
806547a7
KW
1045}
1046
35f8c9bd
KW
1047/*
1048
de69f3af 1049=for apidoc utf8n_to_uvchr
378516de
KW
1050
1051THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1052Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1053
9041c2e3 1054Bottom level UTF-8 decode routine.
de69f3af 1055Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1056which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1057C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1058the length, in bytes, of that character.
949cf498
KW
1059
1060The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1061well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1062causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1063is the next possible position in C<s> that could begin a non-malformed
1064character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1065is raised. Some UTF-8 input sequences may contain multiple malformations.
1066This function tries to find every possible one in each call, so multiple
56576a04 1067warnings can be raised for the same sequence.
949cf498
KW
1068
1069Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1070individual types of malformations, such as the sequence being overlong (that
1071is, when there is a shorter sequence that can express the same code point;
1072overlong sequences are expressly forbidden in the UTF-8 standard due to
1073potential security issues). Another malformation example is the first byte of
1074a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1075flags. Even if allowed, this function generally returns the Unicode
1076REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1077F<utf8.h> to override this behavior for the overlong malformations, but don't
1078do that except for very specialized purposes.
949cf498 1079
796b6530 1080The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1081flags) malformation is found. If this flag is set, the routine assumes that
1082the caller will raise a warning, and this function will silently just set
d088425d
KW
1083C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1084
75200dff 1085Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1086character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1087in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1088be set to 1. To disambiguate, upon a zero return, see if the first byte of
1089C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1090error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1091
1092Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1093Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1094By default these are considered regular code points, but certain situations
ecc1615f
KW
1095warrant special handling for them, which can be specified using the C<flags>
1096parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1097three classes are treated as malformations and handled as such. The flags
1098C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1099C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1100disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1101restricts the allowed inputs to the strict UTF-8 traditionally defined by
1102Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1103definition given by
1104L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1105The difference between traditional strictness and C9 strictness is that the
1106latter does not forbid non-character code points. (They are still discouraged,
1107however.) For more discussion see L<perlunicode/Noncharacter code points>.
1108
1109The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1110C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1111C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1112raised for their respective categories, but otherwise the code points are
1113considered valid (not malformations). To get a category to both be treated as
1114a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1115(But note that warnings are not raised if lexically disabled nor if
796b6530 1116C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1117
57ff5f59
KW
1118Extremely high code points were never specified in any standard, and require an
1119extension to UTF-8 to express, which Perl does. It is likely that programs
1120written in something other than Perl would not be able to read files that
1121contain these; nor would Perl understand files written by something that uses a
1122different extension. For these reasons, there is a separate set of flags that
1123can warn and/or disallow these extremely high code points, even if other
1124above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1125C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1126L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1127above-Unicode code points, including these, as malformations.
1128(Note that the Unicode standard considers anything above 0x10FFFF to be
1129illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1130(2**31 -1))
1131
1132A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1133retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1134C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1135C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1136can apply to code points that actually do fit in 31 bits. This happens on
1137EBCDIC platforms, and sometimes when the L<overlong
1138malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1139describe the situation in all cases.
1140
ab8e6d41 1141
949cf498
KW
1142All other code points corresponding to Unicode characters, including private
1143use and those yet to be assigned, are never considered malformed and never
1144warn.
67e989fb 1145
37607a96 1146=cut
f9380377
KW
1147
1148Also implemented as a macro in utf8.h
1149*/
1150
1151UV
1152Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1153 STRLEN curlen,
1154 STRLEN *retlen,
1155 const U32 flags)
1156{
1157 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1158
1159 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1160}
1161
1162/*
1163
1164=for apidoc utf8n_to_uvchr_error
1165
1166THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1167Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1168
1169This function is for code that needs to know what the precise malformation(s)
1170are when an error is found.
1171
1172It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1173all the others, C<errors>. If this parameter is 0, this function behaves
1174identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1175to a C<U32> variable, which this function sets to indicate any errors found.
1176Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1177C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1178of these bits will be set if a malformation is found, even if the input
7a65503b 1179C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1180exceptions are noted:
1181
1182=over 4
1183
57ff5f59 1184=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1185
57ff5f59
KW
1186The input sequence is not standard UTF-8, but a Perl extension. This bit is
1187set only if the input C<flags> parameter contains either the
1188C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1189
1190Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1191and so some extension must be used to express them. Perl uses a natural
1192extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1193extension to represent even higher ones, so that any code point that fits in a
119464-bit word can be represented. Text using these extensions is not likely to
1195be portable to non-Perl code. We lump both of these extensions together and
1196refer to them as Perl extended UTF-8. There exist other extensions that people
1197have invented, incompatible with Perl's.
1198
1199On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1200extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1201than on ASCII. Prior to that, code points 2**31 and higher were simply
1202unrepresentable, and a different, incompatible method was used to represent
1203code points between 2**30 and 2**31 - 1.
1204
1205On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1206Perl extended UTF-8 is used.
1207
1208In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1209may use for backward compatibility. That name is misleading, as this flag may
1210be set when the code point actually does fit in 31 bits. This happens on
1211EBCDIC platforms, and sometimes when the L<overlong
1212malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1213describes the situation in all cases.
f9380377
KW
1214
1215=item C<UTF8_GOT_CONTINUATION>
1216
1217The input sequence was malformed in that the first byte was a a UTF-8
1218continuation byte.
1219
1220=item C<UTF8_GOT_EMPTY>
1221
1222The input C<curlen> parameter was 0.
1223
1224=item C<UTF8_GOT_LONG>
1225
1226The input sequence was malformed in that there is some other sequence that
1227evaluates to the same code point, but that sequence is shorter than this one.
1228
fecaf136
KW
1229Until Unicode 3.1, it was legal for programs to accept this malformation, but
1230it was discovered that this created security issues.
1231
f9380377
KW
1232=item C<UTF8_GOT_NONCHAR>
1233
1234The code point represented by the input UTF-8 sequence is for a Unicode
1235non-character code point.
1236This bit is set only if the input C<flags> parameter contains either the
1237C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1238
1239=item C<UTF8_GOT_NON_CONTINUATION>
1240
1241The input sequence was malformed in that a non-continuation type byte was found
1242in a position where only a continuation type one should be.
1243
1244=item C<UTF8_GOT_OVERFLOW>
1245
1246The input sequence was malformed in that it is for a code point that is not
d22ec717 1247representable in the number of bits available in an IV on the current platform.
f9380377
KW
1248
1249=item C<UTF8_GOT_SHORT>
1250
1251The input sequence was malformed in that C<curlen> is smaller than required for
1252a complete sequence. In other words, the input is for a partial character
1253sequence.
1254
1255=item C<UTF8_GOT_SUPER>
1256
1257The input sequence was malformed in that it is for a non-Unicode code point;
1258that is, one above the legal Unicode maximum.
1259This bit is set only if the input C<flags> parameter contains either the
1260C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1261
1262=item C<UTF8_GOT_SURROGATE>
1263
1264The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1265code point.
1266This bit is set only if the input C<flags> parameter contains either the
1267C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1268
1269=back
1270
133551d8
KW
1271To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1272flag to suppress any warnings, and then examine the C<*errors> return.
1273
f9380377 1274=cut
37607a96 1275*/
67e989fb 1276
a0ed51b3 1277UV
f9380377
KW
1278Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1279 STRLEN curlen,
1280 STRLEN *retlen,
1281 const U32 flags,
1282 U32 * errors)
a0ed51b3 1283{
d4c19fe8 1284 const U8 * const s0 = s;
2b5e7bc2
KW
1285 U8 * send = NULL; /* (initialized to silence compilers' wrong
1286 warning) */
1287 U32 possible_problems = 0; /* A bit is set here for each potential problem
1288 found as we go along */
eb83ed87 1289 UV uv = *s;
2b5e7bc2
KW
1290 STRLEN expectlen = 0; /* How long should this sequence be?
1291 (initialized to silence compilers' wrong
1292 warning) */
e308b348 1293 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1294 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1295 this gets set and discarded */
a0dbb045 1296
2b5e7bc2
KW
1297 /* The below are used only if there is both an overlong malformation and a
1298 * too short one. Otherwise the first two are set to 's0' and 'send', and
1299 * the third not used at all */
1300 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1301 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1302 routine; see [perl #130921] */
2b5e7bc2 1303 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1304
f9380377
KW
1305 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1306
1307 if (errors) {
1308 *errors = 0;
1309 }
1310 else {
1311 errors = &discard_errors;
1312 }
a0dbb045 1313
eb83ed87
KW
1314 /* The order of malformation tests here is important. We should consume as
1315 * few bytes as possible in order to not skip any valid character. This is
1316 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1317 * http://unicode.org/reports/tr36 for more discussion as to why. For
1318 * example, once we've done a UTF8SKIP, we can tell the expected number of
1319 * bytes, and could fail right off the bat if the input parameters indicate
1320 * that there are too few available. But it could be that just that first
1321 * byte is garbled, and the intended character occupies fewer bytes. If we
1322 * blindly assumed that the first byte is correct, and skipped based on
1323 * that number, we could skip over a valid input character. So instead, we
1324 * always examine the sequence byte-by-byte.
1325 *
1326 * We also should not consume too few bytes, otherwise someone could inject
1327 * things. For example, an input could be deliberately designed to
1328 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1329 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1330 * which is actually part of of the overflowing sequence), that could look
1331 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1332 * sequence and process the rest, inappropriately.
1333 *
1334 * Some possible input sequences are malformed in more than one way. This
1335 * function goes to lengths to try to find all of them. This is necessary
1336 * for correctness, as the inputs may allow one malformation but not
1337 * another, and if we abandon searching for others after finding the
1338 * allowed one, we could allow in something that shouldn't have been.
1339 */
eb83ed87 1340
b5b9af04 1341 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1342 possible_problems |= UTF8_GOT_EMPTY;
1343 curlen = 0;
5a48568d 1344 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1345 goto ready_to_handle_errors;
0c443dc2
JH
1346 }
1347
eb83ed87
KW
1348 expectlen = UTF8SKIP(s);
1349
1350 /* A well-formed UTF-8 character, as the vast majority of calls to this
1351 * function will be for, has this expected length. For efficiency, set
1352 * things up here to return it. It will be overriden only in those rare
1353 * cases where a malformation is found */
1354 if (retlen) {
1355 *retlen = expectlen;
1356 }
1357
1358 /* An invariant is trivially well-formed */
1d72bdf6 1359 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1360 return uv;
a0ed51b3 1361 }
67e989fb 1362
eb83ed87 1363 /* A continuation character can't start a valid sequence */
b5b9af04 1364 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1365 possible_problems |= UTF8_GOT_CONTINUATION;
1366 curlen = 1;
1367 uv = UNICODE_REPLACEMENT;
1368 goto ready_to_handle_errors;
ba210ebe 1369 }
9041c2e3 1370
dcd27b3c 1371 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1372 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1373 * because it excludes start bytes like \xC0 that always lead to
1374 * overlongs.) */
dcd27b3c 1375
534752c1
KW
1376 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1377 * that indicate the number of bytes in the character's whole UTF-8
1378 * sequence, leaving just the bits that are part of the value. */
1379 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1380
e308b348
KW
1381 /* Setup the loop end point, making sure to not look past the end of the
1382 * input string, and flag it as too short if the size isn't big enough. */
1383 send = (U8*) s0;
1384 if (UNLIKELY(curlen < expectlen)) {
1385 possible_problems |= UTF8_GOT_SHORT;
1386 avail_len = curlen;
1387 send += curlen;
1388 }
1389 else {
1390 send += expectlen;
1391 }
e308b348 1392
eb83ed87 1393 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1394 * accumulating each into the working value as we go. */
eb83ed87 1395 for (s = s0 + 1; s < send; s++) {
b5b9af04 1396 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1397 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1398 continue;
1399 }
1400
1401 /* Here, found a non-continuation before processing all expected bytes.
1402 * This byte indicates the beginning of a new character, so quit, even
1403 * if allowing this malformation. */
2b5e7bc2 1404 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1405 break;
eb83ed87
KW
1406 } /* End of loop through the character's bytes */
1407
1408 /* Save how many bytes were actually in the character */
1409 curlen = s - s0;
1410
2b5e7bc2
KW
1411 /* Note that there are two types of too-short malformation. One is when
1412 * there is actual wrong data before the normal termination of the
1413 * sequence. The other is that the sequence wasn't complete before the end
1414 * of the data we are allowed to look at, based on the input 'curlen'.
1415 * This means that we were passed data for a partial character, but it is
1416 * valid as far as we saw. The other is definitely invalid. This
1417 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1418 * separate.
1419 *
1420 * A convenience macro that matches either of the too-short conditions. */
1421# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1422
1423 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1424 uv_so_far = uv;
1425 uv = UNICODE_REPLACEMENT;
1426 }
2b5e7bc2 1427
08e73697
KW
1428 /* Check for overflow. The algorithm requires us to not look past the end
1429 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1430 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1431 1 /* Do consider overlongs */
1432 )))
1433 {
2b5e7bc2
KW
1434 possible_problems |= UTF8_GOT_OVERFLOW;
1435 uv = UNICODE_REPLACEMENT;
eb83ed87 1436 }
eb83ed87 1437
2b5e7bc2
KW
1438 /* Check for overlong. If no problems so far, 'uv' is the correct code
1439 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1440 * we must look at the UTF-8 byte sequence itself to see if it is for an
1441 * overlong */
1442 if ( ( LIKELY(! possible_problems)
1443 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1444 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1445 && ( UNLIKELY(! UTF8_IS_START(*s0))
1446 || ( curlen > 1
d6be65ae 1447 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1448 s - s0))))))
2f8f112e 1449 {
2b5e7bc2
KW
1450 possible_problems |= UTF8_GOT_LONG;
1451
abc28b54 1452 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1453
abc28b54
KW
1454 /* The calculation in the 'true' branch of this 'if'
1455 * below won't work if overflows, and isn't needed
1456 * anyway. Further below we handle all overflow
1457 * cases */
1458 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1459 {
2b5e7bc2
KW
1460 UV min_uv = uv_so_far;
1461 STRLEN i;
1462
1463 /* Here, the input is both overlong and is missing some trailing
1464 * bytes. There is no single code point it could be for, but there
1465 * may be enough information present to determine if what we have
1466 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1467 * The code further below has the intelligence to determine this,
1468 * but just for non-overlong UTF-8 sequences. What we do here is
1469 * calculate the smallest code point the input could represent if
1470 * there were no too short malformation. Then we compute and save
1471 * the UTF-8 for that, which is what the code below looks at
1472 * instead of the raw input. It turns out that the smallest such
1473 * code point is all we need. */
2b5e7bc2
KW
1474 for (i = curlen; i < expectlen; i++) {
1475 min_uv = UTF8_ACCUMULATE(min_uv,
1476 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1477 }
1478
e9f2c446 1479 adjusted_s0 = temp_char_buf;
57ff5f59 1480 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1481 }
eb83ed87
KW
1482 }
1483
56576a04
KW
1484 /* Here, we have found all the possible problems, except for when the input
1485 * is for a problematic code point not allowed by the input parameters. */
1486
06188866
KW
1487 /* uv is valid for overlongs */
1488 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1489
1490 /* isn't problematic if < this */
1491 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1492 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1493
1494 /* if overflow, we know without looking further
1495 * precisely which of the problematic types it is,
1496 * and we deal with those in the overflow handling
1497 * code */
1498 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1499 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1500 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1501 && ((flags & ( UTF8_DISALLOW_NONCHAR
1502 |UTF8_DISALLOW_SURROGATE
1503 |UTF8_DISALLOW_SUPER
d044b7a7 1504 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1505 |UTF8_WARN_NONCHAR
1506 |UTF8_WARN_SURROGATE
1507 |UTF8_WARN_SUPER
d22ec717 1508 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1509 {
2b5e7bc2
KW
1510 /* If there were no malformations, or the only malformation is an
1511 * overlong, 'uv' is valid */
1512 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1513 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1514 possible_problems |= UTF8_GOT_SURROGATE;
1515 }
1516 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1517 possible_problems |= UTF8_GOT_SUPER;
1518 }
1519 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1520 possible_problems |= UTF8_GOT_NONCHAR;
1521 }
1522 }
1523 else { /* Otherwise, need to look at the source UTF-8, possibly
1524 adjusted to be non-overlong */
1525
1526 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1527 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1528 {
2b5e7bc2
KW
1529 possible_problems |= UTF8_GOT_SUPER;
1530 }
1531 else if (curlen > 1) {
1532 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1533 NATIVE_UTF8_TO_I8(*adjusted_s0),
1534 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1535 {
2b5e7bc2 1536 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1537 }
2b5e7bc2
KW
1538 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1539 NATIVE_UTF8_TO_I8(*adjusted_s0),
1540 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1541 {
1542 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1543 }
1544 }
c0236afe 1545
2b5e7bc2
KW
1546 /* We need a complete well-formed UTF-8 character to discern
1547 * non-characters, so can't look for them here */
1548 }
1549 }
949cf498 1550
2b5e7bc2
KW
1551 ready_to_handle_errors:
1552
1553 /* At this point:
1554 * curlen contains the number of bytes in the sequence that
1555 * this call should advance the input by.
e308b348
KW
1556 * avail_len gives the available number of bytes passed in, but
1557 * only if this is less than the expected number of
1558 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1559 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1560 * is set in it for each potential problem found.
1561 * uv contains the code point the input sequence
1562 * represents; or if there is a problem that prevents
1563 * a well-defined value from being computed, it is
1564 * some subsitute value, typically the REPLACEMENT
1565 * CHARACTER.
1566 * s0 points to the first byte of the character
56576a04
KW
1567 * s points to just after were we left off processing
1568 * the character
1569 * send points to just after where that character should
1570 * end, based on how many bytes the start byte tells
1571 * us should be in it, but no further than s0 +
1572 * avail_len
2b5e7bc2 1573 */
eb83ed87 1574
2b5e7bc2
KW
1575 if (UNLIKELY(possible_problems)) {
1576 bool disallowed = FALSE;
1577 const U32 orig_problems = possible_problems;
1578
1579 while (possible_problems) { /* Handle each possible problem */
1580 UV pack_warn = 0;
1581 char * message = NULL;
1582
1583 /* Each 'if' clause handles one problem. They are ordered so that
1584 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1585 * ones; this is kinda in decreasing severity order. But the
1586 * overlong must come last, as it changes 'uv' looked at by the
1587 * others */
2b5e7bc2
KW
1588 if (possible_problems & UTF8_GOT_OVERFLOW) {
1589
56576a04
KW
1590 /* Overflow means also got a super and are using Perl's
1591 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1592 possible_problems
d044b7a7 1593 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1594 *errors |= UTF8_GOT_OVERFLOW;
1595
1596 /* But the API says we flag all errors found */
1597 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1598 *errors |= UTF8_GOT_SUPER;
1599 }
ddb65933 1600 if (flags
d044b7a7 1601 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1602 {
d044b7a7 1603 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1604 }
2b5e7bc2 1605
d60baaa7 1606 /* Disallow if any of the three categories say to */
56576a04 1607 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1608 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1609 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1610 {
1611 disallowed = TRUE;
1612 }
1613
d22ec717
KW
1614 /* Likewise, warn if any say to */
1615 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1616 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1617 {
2b5e7bc2 1618
ddb65933
KW
1619 /* The warnings code explicitly says it doesn't handle the
1620 * case of packWARN2 and two categories which have
1621 * parent-child relationship. Even if it works now to
1622 * raise the warning if either is enabled, it wouldn't
1623 * necessarily do so in the future. We output (only) the
56576a04 1624 * most dire warning */
ddb65933
KW
1625 if (! (flags & UTF8_CHECK_ONLY)) {
1626 if (ckWARN_d(WARN_UTF8)) {
1627 pack_warn = packWARN(WARN_UTF8);
1628 }
1629 else if (ckWARN_d(WARN_NON_UNICODE)) {
1630 pack_warn = packWARN(WARN_NON_UNICODE);
1631 }
1632 if (pack_warn) {
1633 message = Perl_form(aTHX_ "%s: %s (overflows)",
1634 malformed_text,
05b9033b 1635 _byte_dump_string(s0, curlen, 0));
ddb65933 1636 }
2b5e7bc2
KW
1637 }
1638 }
1639 }
1640 else if (possible_problems & UTF8_GOT_EMPTY) {
1641 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1642 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1643
1644 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1645
1646 /* This so-called malformation is now treated as a bug in
1647 * the caller. If you have nothing to decode, skip calling
1648 * this function */
1649 assert(0);
1650
2b5e7bc2
KW
1651 disallowed = TRUE;
1652 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1653 pack_warn = packWARN(WARN_UTF8);
1654 message = Perl_form(aTHX_ "%s (empty string)",
1655 malformed_text);
1656 }
1657 }
1658 }
1659 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1660 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1661 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1662
1663 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1664 disallowed = TRUE;
1665 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1666 pack_warn = packWARN(WARN_UTF8);
1667 message = Perl_form(aTHX_
1668 "%s: %s (unexpected continuation byte 0x%02x,"
1669 " with no preceding start byte)",
1670 malformed_text,
7e2f38b2 1671 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1672 }
1673 }
1674 }
2b5e7bc2
KW
1675 else if (possible_problems & UTF8_GOT_SHORT) {
1676 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1677 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1678
1679 if (! (flags & UTF8_ALLOW_SHORT)) {
1680 disallowed = TRUE;
1681 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1682 pack_warn = packWARN(WARN_UTF8);
1683 message = Perl_form(aTHX_
56576a04
KW
1684 "%s: %s (too short; %d byte%s available, need %d)",
1685 malformed_text,
1686 _byte_dump_string(s0, send - s0, 0),
1687 (int)avail_len,
1688 avail_len == 1 ? "" : "s",
1689 (int)expectlen);
2b5e7bc2
KW
1690 }
1691 }
ba210ebe 1692
2b5e7bc2 1693 }
e308b348
KW
1694 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1695 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1696 *errors |= UTF8_GOT_NON_CONTINUATION;
1697
1698 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1699 disallowed = TRUE;
1700 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1701
1702 /* If we don't know for sure that the input length is
1703 * valid, avoid as much as possible reading past the
1704 * end of the buffer */
1705 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1706 ? s - s0
1707 : send - s0;
e308b348
KW
1708 pack_warn = packWARN(WARN_UTF8);
1709 message = Perl_form(aTHX_ "%s",
1710 unexpected_non_continuation_text(s0,
99a765e9 1711 printlen,
e308b348
KW
1712 s - s0,
1713 (int) expectlen));
1714 }
1715 }
1716 }
2b5e7bc2
KW
1717 else if (possible_problems & UTF8_GOT_SURROGATE) {
1718 possible_problems &= ~UTF8_GOT_SURROGATE;
1719
f9380377
KW
1720 if (flags & UTF8_WARN_SURROGATE) {
1721 *errors |= UTF8_GOT_SURROGATE;
1722
1723 if ( ! (flags & UTF8_CHECK_ONLY)
1724 && ckWARN_d(WARN_SURROGATE))
1725 {
2b5e7bc2
KW
1726 pack_warn = packWARN(WARN_SURROGATE);
1727
1728 /* These are the only errors that can occur with a
1729 * surrogate when the 'uv' isn't valid */
1730 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1731 message = Perl_form(aTHX_
1732 "UTF-16 surrogate (any UTF-8 sequence that"
1733 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1734 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1735 }
1736 else {
c94c2f39 1737 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1738 }
f9380377 1739 }
2b5e7bc2 1740 }
ba210ebe 1741
2b5e7bc2
KW
1742 if (flags & UTF8_DISALLOW_SURROGATE) {
1743 disallowed = TRUE;
f9380377 1744 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1745 }
1746 }
1747 else if (possible_problems & UTF8_GOT_SUPER) {
1748 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1749
f9380377
KW
1750 if (flags & UTF8_WARN_SUPER) {
1751 *errors |= UTF8_GOT_SUPER;
1752
1753 if ( ! (flags & UTF8_CHECK_ONLY)
1754 && ckWARN_d(WARN_NON_UNICODE))
1755 {
2b5e7bc2
KW
1756 pack_warn = packWARN(WARN_NON_UNICODE);
1757
1758 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1759 message = Perl_form(aTHX_
1760 "Any UTF-8 sequence that starts with"
1761 " \"%s\" is for a non-Unicode code point,"
1762 " may not be portable",
7e2f38b2 1763 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1764 }
1765 else {
c94c2f39 1766 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1767 }
f9380377 1768 }
2b5e7bc2 1769 }
ba210ebe 1770
57ff5f59
KW
1771 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1772 * and before possibly bailing out, so that the more dire
1773 * warning will override the regular one. */
1774 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1775 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1776 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1777 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1778 {
db0f09e6 1779 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1780
57ff5f59
KW
1781 /* If it is an overlong that evaluates to a code point
1782 * that doesn't have to use the Perl extended UTF-8, it
1783 * still used it, and so we output a message that
1784 * doesn't refer to the code point. The same is true
1785 * if there was a SHORT malformation where the code
1786 * point is not valid. In that case, 'uv' will have
1787 * been set to the REPLACEMENT CHAR, and the message
1788 * below without the code point in it will be selected
1789 * */
1790 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1791 message = Perl_form(aTHX_
57ff5f59 1792 perl_extended_cp_format, uv);
2b5e7bc2
KW
1793 }
1794 else {
1795 message = Perl_form(aTHX_
57ff5f59
KW
1796 "Any UTF-8 sequence that starts with"
1797 " \"%s\" is a Perl extension, and"
1798 " so is not portable",
1799 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1800 }
1801 }
1802
d044b7a7
KW
1803 if (flags & ( UTF8_WARN_PERL_EXTENDED
1804 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1805 {
d044b7a7 1806 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1807
d044b7a7 1808 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1809 disallowed = TRUE;
1810 }
2b5e7bc2
KW
1811 }
1812 }
eb83ed87 1813
2b5e7bc2 1814 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1815 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1816 disallowed = TRUE;
1817 }
2b5e7bc2
KW
1818 }
1819 else if (possible_problems & UTF8_GOT_NONCHAR) {
1820 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1821
f9380377
KW
1822 if (flags & UTF8_WARN_NONCHAR) {
1823 *errors |= UTF8_GOT_NONCHAR;
1824
1825 if ( ! (flags & UTF8_CHECK_ONLY)
1826 && ckWARN_d(WARN_NONCHAR))
1827 {
2b5e7bc2
KW
1828 /* The code above should have guaranteed that we don't
1829 * get here with errors other than overlong */
1830 assert (! (orig_problems
1831 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1832
1833 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1834 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1835 }
2b5e7bc2 1836 }
5b311467 1837
2b5e7bc2
KW
1838 if (flags & UTF8_DISALLOW_NONCHAR) {
1839 disallowed = TRUE;
f9380377 1840 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1841 }
6c64cd9d
KW
1842 }
1843 else if (possible_problems & UTF8_GOT_LONG) {
1844 possible_problems &= ~UTF8_GOT_LONG;
1845 *errors |= UTF8_GOT_LONG;
1846
1847 if (flags & UTF8_ALLOW_LONG) {
1848
1849 /* We don't allow the actual overlong value, unless the
1850 * special extra bit is also set */
1851 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1852 & ~UTF8_ALLOW_LONG)))
1853 {
1854 uv = UNICODE_REPLACEMENT;
1855 }
1856 }
1857 else {
1858 disallowed = TRUE;
1859
1860 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1861 pack_warn = packWARN(WARN_UTF8);
1862
1863 /* These error types cause 'uv' to be something that
1864 * isn't what was intended, so can't use it in the
1865 * message. The other error types either can't
1866 * generate an overlong, or else the 'uv' is valid */
1867 if (orig_problems &
1868 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1869 {
1870 message = Perl_form(aTHX_
1871 "%s: %s (any UTF-8 sequence that starts"
1872 " with \"%s\" is overlong which can and"
1873 " should be represented with a"
1874 " different, shorter sequence)",
1875 malformed_text,
1876 _byte_dump_string(s0, send - s0, 0),
1877 _byte_dump_string(s0, curlen, 0));
1878 }
1879 else {
1880 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
1881 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1882 uv, 0);
d819dc50
KW
1883 /* Don't use U+ for non-Unicode code points, which
1884 * includes those in the Latin1 range */
1885 const char * preface = ( uv > PERL_UNICODE_MAX
1886#ifdef EBCDIC
1887 || uv <= 0xFF
1888#endif
1889 )
1890 ? "0x"
1891 : "U+";
6c64cd9d
KW
1892 message = Perl_form(aTHX_
1893 "%s: %s (overlong; instead use %s to represent"
1894 " %s%0*" UVXf ")",
1895 malformed_text,
1896 _byte_dump_string(s0, send - s0, 0),
1897 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1898 preface,
1899 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1900 small code points */
1be62ab9 1901 UNI_TO_NATIVE(uv));
6c64cd9d
KW
1902 }
1903 }
1904 }
2b5e7bc2
KW
1905 } /* End of looking through the possible flags */
1906
1907 /* Display the message (if any) for the problem being handled in
1908 * this iteration of the loop */
1909 if (message) {
1910 if (PL_op)
1911 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1912 OP_DESC(PL_op));
1913 else
1914 Perl_warner(aTHX_ pack_warn, "%s", message);
1915 }
ddb65933 1916 } /* End of 'while (possible_problems)' */
a0dbb045 1917
2b5e7bc2
KW
1918 /* Since there was a possible problem, the returned length may need to
1919 * be changed from the one stored at the beginning of this function.
1920 * Instead of trying to figure out if that's needed, just do it. */
1921 if (retlen) {
1922 *retlen = curlen;
1923 }
a0dbb045 1924
2b5e7bc2
KW
1925 if (disallowed) {
1926 if (flags & UTF8_CHECK_ONLY && retlen) {
1927 *retlen = ((STRLEN) -1);
1928 }
1929 return 0;
1930 }
eb83ed87 1931 }
ba210ebe 1932
2b5e7bc2 1933 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1934}
1935
8e84507e 1936/*
ec5f19d0
KW
1937=for apidoc utf8_to_uvchr_buf
1938
1939Returns the native code point of the first character in the string C<s> which
1940is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1941C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1942
524080c4
KW
1943If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1944enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1945C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1946(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1947C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1948the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1949See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1950returned.
ec5f19d0
KW
1951
1952=cut
52be2536
KW
1953
1954Also implemented as a macro in utf8.h
1955
ec5f19d0
KW
1956*/
1957
1958
1959UV
1960Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1961{
7f974d7e
KW
1962 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1963
ec5f19d0
KW
1964 assert(s < send);
1965
1966 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1967 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1968}
1969
52be2536
KW
1970/* This is marked as deprecated
1971 *
ec5f19d0
KW
1972=for apidoc utf8_to_uvuni_buf
1973
de69f3af
KW
1974Only in very rare circumstances should code need to be dealing in Unicode
1975(as opposed to native) code points. In those few cases, use
1976C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1977
1978Returns the Unicode (not-native) code point of the first character in the
1979string C<s> which
ec5f19d0
KW
1980is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1981C<retlen> will be set to the length, in bytes, of that character.
1982
524080c4
KW
1983If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1984enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1985NULL) to -1. If those warnings are off, the computed value if well-defined (or
1986the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1987is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1988next possible position in C<s> that could begin a non-malformed character.
de69f3af 1989See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
1990
1991=cut
1992*/
1993
1994UV
1995Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1996{
1997 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1998
1999 assert(send > s);
2000
5962d97e 2001 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2002}
2003
b76347f2 2004/*
87cea99e 2005=for apidoc utf8_length
b76347f2
JH
2006
2007Return the length of the UTF-8 char encoded string C<s> in characters.
647672aa
DM
2008Stops at C<e> (i.e. the C<*e> byte does not form part of the character).
2009If C<e E<lt> s> or if the scan would end up past C<e>, it croaks.
b76347f2
JH
2010
2011=cut
2012*/
2013
2014STRLEN
35a4481c 2015Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2016{
2017 STRLEN len = 0;
2018
7918f24d
NC
2019 PERL_ARGS_ASSERT_UTF8_LENGTH;
2020
8850bf83
JH
2021 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2022 * the bitops (especially ~) can create illegal UTF-8.
2023 * In other words: in Perl UTF-8 is not just for Unicode. */
2024
a3b680e6
AL
2025 if (e < s)
2026 goto warn_and_return;
b76347f2 2027 while (s < e) {
4cbf4130 2028 s += UTF8SKIP(s);
8e91ec7f
AV
2029 len++;
2030 }
2031
2032 if (e != s) {
2033 len--;
2034 warn_and_return:
9b387841
NC
2035 if (PL_op)
2036 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2037 "%s in %s", unees, OP_DESC(PL_op));
2038 else
61a12c31 2039 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2040 }
2041
2042 return len;
2043}
2044
b06226ff 2045/*
fed3ba5d
NC
2046=for apidoc bytes_cmp_utf8
2047
a1433954 2048Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2049sequence of characters (stored as UTF-8)
2050in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2051equal, -1 or -2 if the first string is less than the second string, +1 or +2
2052if the first string is greater than the second string.
2053
2054-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2055longer string. -2 or +2 is returned if
2056there was a difference between characters
fed3ba5d
NC
2057within the strings.
2058
2059=cut
2060*/
2061
2062int
2063Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2064{
2065 const U8 *const bend = b + blen;
2066 const U8 *const uend = u + ulen;
2067
2068 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2069
2070 while (b < bend && u < uend) {
2071 U8 c = *u++;
2072 if (!UTF8_IS_INVARIANT(c)) {
2073 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2074 if (u < uend) {
2075 U8 c1 = *u++;
2076 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2077 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2078 } else {
2b5e7bc2 2079 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2080 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2081 "%s %s%s",
2082 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2083 PL_op ? " in " : "",
2084 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2085 return -2;
2086 }
2087 } else {
2088 if (PL_op)
2089 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2090 "%s in %s", unees, OP_DESC(PL_op));
2091 else
61a12c31 2092 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2093 return -2; /* Really want to return undef :-) */
2094 }
2095 } else {
2096 return -2;
2097 }
2098 }
2099 if (*b != c) {
2100 return *b < c ? -2 : +2;
2101 }
2102 ++b;
2103 }
2104
2105 if (b == bend && u == uend)
2106 return 0;
2107
2108 return b < bend ? +1 : -1;
2109}
2110
2111/*
87cea99e 2112=for apidoc utf8_to_bytes
6940069f 2113
3bc0c78c 2114Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2115Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2116updates C<*lenp> to contain the new length.
3bc0c78c
KW
2117Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2118
2119Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2120having saved the value of C<*lenp> before the call, and subtracting the
2121after-call value of C<*lenp> from it.
6940069f 2122
a1433954 2123If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2124
6940069f
GS
2125=cut
2126*/
2127
2128U8 *
09af0336 2129Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2130{
9fe0d3c2 2131 U8 * first_variant;
246fae53 2132
7918f24d 2133 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2134 PERL_UNUSED_CONTEXT;
7918f24d 2135
9fe0d3c2 2136 /* This is a no-op if no variants at all in the input */
09af0336 2137 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2138 return s;
2139 }
2140
2141 {
3c5aa262 2142 U8 * const save = s;
09af0336 2143 U8 * const send = s + *lenp;
3c5aa262
KW
2144 U8 * d;
2145
2146 /* Nothing before the first variant needs to be changed, so start the real
2147 * work there */
2148 s = first_variant;
2149 while (s < send) {
2150 if (! UTF8_IS_INVARIANT(*s)) {
2151 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2152 *lenp = ((STRLEN) -1);
3c5aa262
KW
2153 return 0;
2154 }
2155 s++;
d59937ca
KW
2156 }
2157 s++;
dcad2880 2158 }
dcad2880 2159
3c5aa262
KW
2160 /* Is downgradable, so do it */
2161 d = s = first_variant;
2162 while (s < send) {
2163 U8 c = *s++;
2164 if (! UVCHR_IS_INVARIANT(c)) {
2165 /* Then it is two-byte encoded */
2166 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2167 s++;
2168 }
2169 *d++ = c;
2170 }
2171 *d = '\0';
09af0336 2172 *lenp = d - save;
3c5aa262
KW
2173
2174 return save;
9fe0d3c2 2175 }
6940069f
GS
2176}
2177
2178/*
87cea99e 2179=for apidoc bytes_from_utf8
f9a63242 2180
09af0336 2181Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2182byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2183actually encoded in UTF-8.
2184
2185Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2186the input string.
2187
41ae6089
KW
2188Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2189not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2190C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2191
41ae6089 2192Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2193newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2194returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2195
3bc0c78c 2196Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2197having saved the value of C<*lenp> before the call, and subtracting the
2198after-call value of C<*lenp> from it.
3bc0c78c 2199
37607a96 2200=cut
976c1b08
KW
2201
2202There is a macro that avoids this function call, but this is retained for
2203anyone who calls it with the Perl_ prefix */
f9a63242
JH
2204
2205U8 *
41ae6089 2206Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2207{
7918f24d 2208 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2209 PERL_UNUSED_CONTEXT;
f9a63242 2210
976c1b08
KW
2211 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2212}
2213
2214/*
2215No = here because currently externally undocumented
2216for apidoc bytes_from_utf8_loc
2217
2218Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2219to store the location of the first character in C<"s"> that cannot be
2220converted to non-UTF8.
2221
2222If that parameter is C<NULL>, this function behaves identically to
2223C<bytes_from_utf8>.
2224
2225Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2226C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2227
2228Otherwise, the function returns a newly created C<NUL>-terminated string
2229containing the non-UTF8 equivalent of the convertible first portion of
2230C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2231If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2232and C<*first_non_downgradable> is set to C<NULL>.
2233
2234Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2235first character in the original string that wasn't converted. C<*is_utf8p> is
2236unchanged. Note that the new string may have length 0.
2237
2238Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2239C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2240converts as many characters in it as possible stopping at the first one it
385b74be 2241finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2242set to point to that. The function returns the portion that could be converted
2243in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2244not including the terminating C<NUL>. If the very first character in the
2245original could not be converted, C<*lenp> will be 0, and the new string will
2246contain just a single C<NUL>. If the entire input string was converted,
2247C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2248
2249Upon successful return, the number of variants in the converted portion of the
2250string can be computed by having saved the value of C<*lenp> before the call,
2251and subtracting the after-call value of C<*lenp> from it.
2252
2253=cut
2254
2255
2256*/
2257
2258U8 *
2259Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2260{
2261 U8 *d;
2262 const U8 *original = s;
2263 U8 *converted_start;
2264 const U8 *send = s + *lenp;
f9a63242 2265
976c1b08 2266 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2267
976c1b08
KW
2268 if (! *is_utf8p) {
2269 if (first_unconverted) {
2270 *first_unconverted = NULL;
2271 }
2272
2273 return (U8 *) original;
2274 }
2275
2276 Newx(d, (*lenp) + 1, U8);
2277
2278 converted_start = d;
7299a045
KW
2279 while (s < send) {
2280 U8 c = *s++;
2281 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2282
2283 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2284 * have to stop now */
2285 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2286 if (first_unconverted) {
2287 *first_unconverted = s - 1;
2288 goto finish_and_return;
2289 }
2290 else {
2291 Safefree(converted_start);
2292 return (U8 *) original;
2293 }
2294 }
2295
7299a045
KW
2296 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2297 s++;
38af28cf 2298 }
7299a045
KW
2299 *d++ = c;
2300 }
170a1c22 2301
976c1b08
KW
2302 /* Here, converted the whole of the input */
2303 *is_utf8p = FALSE;
2304 if (first_unconverted) {
2305 *first_unconverted = NULL;
170a1c22 2306 }
976c1b08
KW
2307
2308 finish_and_return:
46a08a6f
KW
2309 *d = '\0';
2310 *lenp = d - converted_start;
976c1b08
KW
2311
2312 /* Trim unused space */
2313 Renew(converted_start, *lenp + 1, U8);
2314
2315 return converted_start;
f9a63242
JH
2316}
2317
2318/*
87cea99e 2319=for apidoc bytes_to_utf8
6940069f 2320
09af0336 2321Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2322UTF-8.
09af0336 2323Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2324reflect the new length in bytes.
6940069f 2325
3bc0c78c 2326Upon successful return, the number of variants in the string can be computed by
23b37b12 2327having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2328after-call value of C<*lenp>.
2329
75200dff 2330A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2331
2332If you want to convert to UTF-8 from encodings other than
2333the native (Latin1 or EBCDIC),
a1433954 2334see L</sv_recode_to_utf8>().
c9ada85f 2335
497711e7 2336=cut
6940069f
GS
2337*/
2338
2339U8*
09af0336 2340Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2341{
09af0336 2342 const U8 * const send = s + (*lenp);
6940069f
GS
2343 U8 *d;
2344 U8 *dst;
7918f24d
NC
2345
2346 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2347 PERL_UNUSED_CONTEXT;
6940069f 2348
09af0336 2349 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2350 dst = d;
2351
2352 while (s < send) {
55d09dc8
KW
2353 append_utf8_from_native_byte(*s, &d);
2354 s++;
6940069f 2355 }
2e11cf67 2356
6940069f 2357 *d = '\0';
09af0336 2358 *lenp = d-dst;
2e11cf67
KW
2359
2360 /* Trim unused space */
2361 Renew(dst, *lenp + 1, U8);
2362
6940069f
GS
2363 return dst;
2364}
2365
a0ed51b3 2366/*
624504c5
KW
2367 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2368 * use utf16_to_utf8_reversed().
a0ed51b3 2369 *
624504c5
KW
2370 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2371 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2372 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2373 *
2374 * These functions don't check for overflow. The worst case is every code
2375 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2376 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2377 * destination must be pre-extended to 2 times the source length.
2378 *
2379 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2380
2381U8*
dea0fc0b 2382Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2383{
dea0fc0b
JH
2384 U8* pend;
2385 U8* dstart = d;
2386
7918f24d
NC
2387 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2388
dea0fc0b 2389 if (bytelen & 1)
56576a04
KW
2390 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2391 (UV)bytelen);
dea0fc0b
JH
2392
2393 pend = p + bytelen;
2394
a0ed51b3 2395 while (p < pend) {
dea0fc0b
JH
2396 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2397 p += 2;
2d1545e5 2398 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2399 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2400 continue;
2401 }
56d37426
KW
2402 if (uv <= MAX_UTF8_TWO_BYTE) {
2403 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2404 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2405 continue;
2406 }
ffd0a9d3 2407
46956fad
KW
2408#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2409#define LAST_HIGH_SURROGATE 0xDBFF
2410#define FIRST_LOW_SURROGATE 0xDC00
2411#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2412#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2413
2414 /* This assumes that most uses will be in the first Unicode plane, not
2415 * needing surrogates */
2416 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2417 && uv <= UNICODE_SURROGATE_LAST))
2418 {
2419 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2420 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2421 }
2422 else {
01ea242b 2423 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2424 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2425 || UNLIKELY(low > LAST_LOW_SURROGATE))
2426 {
01ea242b 2427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2428 }
2429 p += 2;
46956fad 2430 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2431 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2432 }
a0ed51b3 2433 }
56d37426
KW
2434#ifdef EBCDIC
2435 d = uvoffuni_to_utf8_flags(d, uv, 0);
2436#else
ffd0a9d3 2437 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2438 *d++ = (U8)(( uv >> 12) | 0xe0);
2439 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2440 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2441 continue;
2442 }
2443 else {
eb160463
GS
2444 *d++ = (U8)(( uv >> 18) | 0xf0);
2445 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2446 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2447 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2448 continue;
2449 }
56d37426 2450#endif
a0ed51b3 2451 }
dea0fc0b 2452 *newlen = d - dstart;
a0ed51b3
LW
2453 return d;
2454}
2455
2456/* Note: this one is slightly destructive of the source. */
2457
2458U8*
dea0fc0b 2459Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2460{
2461 U8* s = (U8*)p;
d4c19fe8 2462 U8* const send = s + bytelen;
7918f24d
NC
2463
2464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2465
e0ea5e2d 2466 if (bytelen & 1)
147e3846 2467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2468 (UV)bytelen);
2469
a0ed51b3 2470 while (s < send) {
d4c19fe8 2471 const U8 tmp = s[0];
a0ed51b3
LW
2472 s[0] = s[1];
2473 s[1] = tmp;
2474 s += 2;
2475 }
dea0fc0b 2476 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2477}
2478
922e8cb4
KW
2479bool
2480Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2481{
2482 U8 tmpbuf[UTF8_MAXBYTES+1];
2483 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2484 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2485}
2486
f9ae8fb6
JD
2487/* Internal function so we can deprecate the external one, and call
2488 this one from other deprecated functions in this file */
2489
f2645549
KW
2490bool
2491Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2492{
f2645549 2493 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2494
2495 if (*p == '_')
2496 return TRUE;
f25ce844 2497 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2498}
2499
5092f92a 2500bool
eba68aa0
KW
2501Perl__is_uni_perl_idcont(pTHX_ UV c)
2502{
2503 U8 tmpbuf[UTF8_MAXBYTES+1];
2504 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2505 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2506}
2507
2508bool
f91dcd13
KW
2509Perl__is_uni_perl_idstart(pTHX_ UV c)
2510{
2511 U8 tmpbuf[UTF8_MAXBYTES+1];
2512 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2513 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2514}
2515
3a4c58c9 2516UV
56576a04
KW
2517Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2518 const char S_or_s)
3a4c58c9
KW
2519{
2520 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2521 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2522 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2523 * either "SS" or "Ss". Which one to use is passed into the routine in
2524 * 'S_or_s' to avoid a test */
2525
2526 UV converted = toUPPER_LATIN1_MOD(c);
2527
2528 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2529
2530 assert(S_or_s == 'S' || S_or_s == 's');
2531
6f2d5cbc 2532 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2533 characters in this range */
3a4c58c9
KW
2534 *p = (U8) converted;
2535 *lenp = 1;
2536 return converted;
2537 }
2538
2539 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2540 * which it maps to one of them, so as to only have to have one check for
2541 * it in the main case */
2542 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2543 switch (c) {
2544 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2545 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2546 break;
2547 case MICRO_SIGN:
2548 converted = GREEK_CAPITAL_LETTER_MU;
2549 break;
79e064b9
KW
2550#if UNICODE_MAJOR_VERSION > 2 \
2551 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2552 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2553 case LATIN_SMALL_LETTER_SHARP_S:
2554 *(p)++ = 'S';
2555 *p = S_or_s;
2556 *lenp = 2;
2557 return 'S';
79e064b9 2558#endif
3a4c58c9 2559 default:
56576a04
KW
2560 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2561 " '%c' to map to '%c'",
2562 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2563 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2564 }
2565 }
2566
2567 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2568 *p = UTF8_TWO_BYTE_LO(converted);
2569 *lenp = 2;
2570
2571 return converted;
2572}
2573
50bda2c3
KW
2574/* Call the function to convert a UTF-8 encoded character to the specified case.
2575 * Note that there may be more than one character in the result.
2576 * INP is a pointer to the first byte of the input character
2577 * OUTP will be set to the first byte of the string of changed characters. It
2578 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2579 * LENP will be set to the length in bytes of the string of changed characters
2580 *
56576a04
KW
2581 * The functions return the ordinal of the first character in the string of
2582 * OUTP */
2583#define CALL_UPPER_CASE(uv, s, d, lenp) \
2584 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2585#define CALL_TITLE_CASE(uv, s, d, lenp) \
2586 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2587#define CALL_LOWER_CASE(uv, s, d, lenp) \
2588 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2589
b9992569
KW
2590/* This additionally has the input parameter 'specials', which if non-zero will
2591 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2592 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2593#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2594_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2595
84afefe6
JH
2596UV
2597Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2598{
a1433954
KW
2599 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2600 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2601 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2602 * the changed version may be longer than the original character.
2603 *
2604 * The ordinal of the first character of the changed version is returned
2605 * (but note, as explained above, that there may be more.) */
2606
7918f24d
NC
2607 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2608
3a4c58c9
KW
2609 if (c < 256) {
2610 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2611 }
2612
0ebc6274 2613 uvchr_to_utf8(p, c);
b9992569 2614 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2615}
2616
84afefe6
JH
2617UV
2618Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2619{
7918f24d
NC
2620 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2621
3a4c58c9
KW
2622 if (c < 256) {
2623 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2624 }
2625
0ebc6274 2626 uvchr_to_utf8(p, c);
b9992569 2627 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2628}
2629
afc16117 2630STATIC U8
eaf412bf 2631S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2632{
2633 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2634 * those, converting the result to UTF-8. Since the result is always just
a1433954 2635 * one character, we allow <p> to be NULL */
afc16117
KW
2636
2637 U8 converted = toLOWER_LATIN1(c);
2638
eaf412bf
KW
2639 PERL_UNUSED_ARG(dummy);
2640
afc16117 2641 if (p != NULL) {
6f2d5cbc 2642 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2643 *p = converted;
2644 *lenp = 1;
2645 }
2646 else {
430c9760
KW
2647 /* Result is known to always be < 256, so can use the EIGHT_BIT
2648 * macros */
2649 *p = UTF8_EIGHT_BIT_HI(converted);
2650 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2651 *lenp = 2;
2652 }
2653 }
2654 return converted;
2655}
2656
84afefe6
JH
2657UV
2658Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2659{
7918f24d
NC
2660 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2661
afc16117 2662 if (c < 256) {
eaf412bf 2663 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2664 }
2665
afc16117 2666 uvchr_to_utf8(p, c);
b9992569 2667 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2668}
2669
84afefe6 2670UV
56576a04
KW
2671Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2672 const unsigned int flags)
a1dde8de 2673{
51910141 2674 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2675 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2676 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2677 *
2678 * Not to be used for locale folds
51910141 2679 */
f673fad4 2680
a1dde8de
KW
2681 UV converted;
2682
2683 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2684 PERL_UNUSED_CONTEXT;
a1dde8de 2685
1ca267a5
KW
2686 assert (! (flags & FOLD_FLAGS_LOCALE));
2687
659a7c2d 2688 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2689 converted = GREEK_SMALL_LETTER_MU;
2690 }
9b63e895
KW
2691#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2692 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2693 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2694 else if ( (flags & FOLD_FLAGS_FULL)
2695 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2696 {
1ca267a5
KW
2697 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2698 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2699 * under those circumstances. */
2700 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2701 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2702 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2703 p, *lenp, U8);
2704 return LATIN_SMALL_LETTER_LONG_S;
2705 }
2706 else {
4f489194
KW
2707 *(p)++ = 's';
2708 *p = 's';
2709 *lenp = 2;
2710 return 's';
1ca267a5 2711 }
a1dde8de 2712 }
9b63e895 2713#endif
a1dde8de
KW
2714 else { /* In this range the fold of all other characters is their lower
2715 case */
2716 converted = toLOWER_LATIN1(c);
2717 }
2718
6f2d5cbc 2719 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2720 *p = (U8) converted;
2721 *lenp = 1;
2722 }
2723 else {
2724 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2725 *p = UTF8_TWO_BYTE_LO(converted);
2726 *lenp = 2;
2727 }
2728
2729 return converted;
2730}
2731
2732UV
31f05a37 2733Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2734{
4b593389 2735
a0270393
KW
2736 /* Not currently externally documented, and subject to change
2737 * <flags> bits meanings:
2738 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2739 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2740 * locale are to be used.
a0270393
KW
2741 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2742 */
4b593389 2743
36bb2ab6 2744 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2745
780fcc9f
KW
2746 if (flags & FOLD_FLAGS_LOCALE) {
2747 /* Treat a UTF-8 locale as not being in locale at all */
2748 if (IN_UTF8_CTYPE_LOCALE) {
2749 flags &= ~FOLD_FLAGS_LOCALE;
2750 }
2751 else {
2752 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2753 goto needs_full_generality;
780fcc9f 2754 }
31f05a37
KW
2755 }
2756
a1dde8de 2757 if (c < 256) {
e7b7ac46 2758 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2759 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2760 }
2761
2f306ab9 2762 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2763 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2764 uvchr_to_utf8(p, c);
b9992569 2765 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2766 }
567b353c 2767 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2768 the special flags. */
2769 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2770
2771 needs_full_generality:
a0270393 2772 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2773 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2774 p, lenp, flags);
a0270393 2775 }
84afefe6
JH
2776}
2777
26483009 2778PERL_STATIC_INLINE bool
5141f98e 2779S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2780 const char *const swashname, SV* const invlist)
bde6a22d 2781{
ea317ccb
KW
2782 /* returns a boolean giving whether or not the UTF8-encoded character that
2783 * starts at <p> is in the swash indicated by <swashname>. <swash>
2784 * contains a pointer to where the swash indicated by <swashname>
2785 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2786 * look at <*swash> and only generate a swash if it is not null. <invlist>
2787 * is NULL or an inversion list that defines the swash. If not null, it
2788 * saves time during initialization of the swash.
ea317ccb
KW
2789 *
2790 * Note that it is assumed that the buffer length of <p> is enough to
2791 * contain all the bytes that comprise the character. Thus, <*p> should
2792 * have been checked before this call for mal-formedness enough to assure
2793 * that. */
2794
7918f24d
NC
2795 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2796
492a624f 2797 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2798 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2799 * as far as there being enough bytes available in it to accommodate the
2800 * character without reading beyond the end, and pass that number on to the
2801 * validating routine */
6302f837 2802 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2803 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2804 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2805 1 /* Die */ );
2806 NOT_REACHED; /* NOTREACHED */
28123549 2807 }
86ae6e94 2808
87367d5f
KW
2809 if (!*swash) {
2810 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2811 *swash = _core_swash_init("utf8",
2812
2813 /* Only use the name if there is no inversion
2814 * list; otherwise will go out to disk */
2815 (invlist) ? "" : swashname,
2816
2817 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2818 }
28123549 2819
bde6a22d
NC
2820 return swash_fetch(*swash, p, TRUE) != 0;
2821}
2822
da8c1a98 2823PERL_STATIC_INLINE bool
56576a04
KW
2824S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2825 SV **swash, const char *const swashname,
2826 SV* const invlist)
da8c1a98
KW
2827{
2828 /* returns a boolean giving whether or not the UTF8-encoded character that
2829 * starts at <p>, and extending no further than <e - 1> is in the swash
2830 * indicated by <swashname>. <swash> contains a pointer to where the swash
2831 * indicated by <swashname> is to be stored; which this routine will do, so
2832 * that future calls will look at <*swash> and only generate a swash if it
2833 * is not null. <invlist> is NULL or an inversion list that defines the
2834 * swash. If not null, it saves time during initialization of the swash.
2835 */
2836
2837 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2838
2839 if (! isUTF8_CHAR(p, e)) {
2840 _force_out_malformed_utf8_message(p, e, 0, 1);
2841 NOT_REACHED; /* NOTREACHED */
2842 }
2843
2844 if (!*swash) {
2845 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2846 *swash = _core_swash_init("utf8",
2847
2848 /* Only use the name if there is no inversion
2849 * list; otherwise will go out to disk */
2850 (invlist) ? "" : swashname,
2851
2852 &PL_sv_undef, 1, 0, invlist, &flags);
2853 }
2854
2855 return swash_fetch(*swash, p, TRUE) != 0;
2856}
2857
34aeb2e9
KW
2858STATIC void
2859S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2860 const char * const alternative,
2861 const bool use_locale,
2862 const char * const file,
2863 const unsigned line)
2864{
2865 const char * key;
2866
2867 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2868
2869 if (ckWARN_d(WARN_DEPRECATED)) {
2870
2871 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2872 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2873 if (! PL_seen_deprecated_macro) {
2874 PL_seen_deprecated_macro = newHV();
2875 }
2876 if (! hv_store(PL_seen_deprecated_macro, key,
2877 strlen(key), &PL_sv_undef, 0))
2878 {
2879 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2880 }
2881
c44e9413 2882 if (instr(file, "mathoms.c")) {
607313a1
KW
2883 Perl_warner(aTHX_ WARN_DEPRECATED,
2884 "In %s, line %d, starting in Perl v5.30, %s()"
2885 " will be removed. Avoid this message by"
2886 " converting to use %s().\n",
2887 file, line, name, alternative);
2888 }
2889 else {
34aeb2e9
KW
2890 Perl_warner(aTHX_ WARN_DEPRECATED,
2891 "In %s, line %d, starting in Perl v5.30, %s() will"
2892 " require an additional parameter. Avoid this"
2893 " message by converting to use %s().\n",
2894 file, line, name, alternative);
607313a1 2895 }
34aeb2e9
KW
2896 }
2897 }
2898}
2899
bde6a22d 2900bool
34aeb2e9 2901Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2902 const U8 * const p,
34aeb2e9
KW
2903 const char * const name,
2904 const char * const alternative,
2905 const bool use_utf8,
2906 const bool use_locale,
2907 const char * const file,
2908 const unsigned line)
922e8cb4 2909{
922e8cb4
KW
2910 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2911
34aeb2e9
KW
2912 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2913
2914 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2915
2916 switch (classnum) {
2917 case _CC_WORDCHAR:
2918 case _CC_DIGIT:
2919 case _CC_ALPHA:
2920 case _CC_LOWER:
2921 case _CC_UPPER:
2922 case _CC_PUNCT:
2923 case _CC_PRINT:
2924 case _CC_ALPHANUMERIC:
2925 case _CC_GRAPH:
2926 case _CC_CASED:
2927
2928 return is_utf8_common(p,
2929 &PL_utf8_swash_ptrs[classnum],
2930 swash_property_names[classnum],
2931 PL_XPosix_ptrs[classnum]);
2932
2933 case _CC_SPACE:
2934 return is_XPERLSPACE_high(p);
2935 case _CC_BLANK:
2936 return is_HORIZWS_high(p);
2937 case _CC_XDIGIT:
2938 return is_XDIGIT_high(p);
2939 case _CC_CNTRL:
2940 return 0;
2941 case _CC_ASCII:
2942 return 0;
2943 case _CC_VERTSPACE:
2944 return is_VERTWS_high(p);
2945 case _CC_IDFIRST:
2946 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2947 PL_utf8_perl_idstart
2948 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2949 }
22f0498f
KW
2950 return is_utf8_common(p, &PL_utf8_perl_idstart,
2951 "_Perl_IDStart", NULL);
34aeb2e9
KW
2952 case _CC_IDCONT:
2953 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2954 PL_utf8_perl_idcont
2955 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2956 }
22f0498f
KW
2957 return is_utf8_common(p, &PL_utf8_perl_idcont,
2958 "_Perl_IDCont", NULL);
34aeb2e9
KW
2959 }
2960 }
2961
2962 /* idcont is the same as wordchar below 256 */
2963 if (classnum == _CC_IDCONT) {
2964 classnum = _CC_WORDCHAR;
2965 }
2966 else if (classnum == _CC_IDFIRST) {
2967 if (*p == '_') {
2968 return TRUE;
2969 }
2970 classnum = _CC_ALPHA;
2971 }
2972
2973 if (! use_locale) {
2974 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2975 return _generic_isCC(*p, classnum);
2976 }
922e8cb4 2977
34aeb2e9
KW
2978 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2979 }
2980 else {
2981 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2982 return isFOO_lc(classnum, *p);
2983 }
2984
2985 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2986 }
2987
2988 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2989}
2990
2991bool
da8c1a98
KW
2992Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2993 const U8 * const e)
2994{
2995 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2996
2997 assert(classnum < _FIRST_NON_SWASH_CC);
2998
2999 return is_utf8_common_with_len(p,
3000 e,
3001 &PL_utf8_swash_ptrs[classnum],
3002 swash_property_names[classnum],
3003 PL_XPosix_ptrs[classnum]);
3004}
3005
3006bool
da8c1a98
KW
3007Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3008{
3009 SV* invlist = NULL;
3010
3011 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3012
3013 if (! PL_utf8_perl_idstart) {
3014 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
3015 }
3016 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
3017 "_Perl_IDStart", invlist);
3018}
3019
3020bool
f2645549 3021Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3022{
f2645549 3023 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3024
3025 if (*p == '_')
3026 return TRUE;
f25ce844 3027 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3028}
3029
3030bool
da8c1a98
KW
3031Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3032{
3033 SV* invlist = NULL;
3034
3035 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3036
3037 if (! PL_utf8_perl_idcont) {
3038 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3039 }
3040 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3041 "_Perl_IDCont", invlist);
3042}
3043
3044bool
f2645549 3045Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3046{
f2645549 3047 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3048
f25ce844 3049 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3050}
3051
3052bool
f2645549 3053Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3054{
f2645549 3055 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3056
f25ce844 3057 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
3058}
3059
3060bool
7dbf68d2
KW
3061Perl__is_utf8_mark(pTHX_ const U8 *p)
3062{
7dbf68d2
KW
3063 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3064
f25ce844 3065 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3066}
3067
b9992569 3068 /* change namve uv1 to 'from' */
6a4a25f4 3069STATIC UV
b9992569
KW
3070S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
3071 SV **swashp, const char *normal, const char *special)
3072{
0134edef 3073 STRLEN len = 0;
7918f24d 3074
b9992569 3075 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3076
36eaa811
KW
3077 /* For code points that don't change case, we already know that the output
3078 * of this function is the unchanged input, so we can skip doing look-ups
3079 * for them. Unfortunately the case-changing code points are scattered
3080 * around. But there are some long consecutive ranges where there are no
3081 * case changing code points. By adding tests, we can eliminate the lookup
3082 * for all the ones in such ranges. This is currently done here only for
3083 * just a few cases where the scripts are in common use in modern commerce
3084 * (and scripts adjacent to those which can be included without additional
3085 * tests). */
3086
3087 if (uv1 >= 0x0590) {
3088 /* This keeps from needing further processing the code points most
3089 * likely to be used in the following non-cased scripts: Hebrew,
3090 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3091 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3092 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3093 if (uv1 < 0x10A0) {
3094 goto cases_to_self;
3095 }
3096
3097 /* The following largish code point ranges also don't have case
3098 * changes, but khw didn't think they warranted extra tests to speed
3099 * them up (which would slightly slow down everything else above them):
3100 * 1100..139F Hangul Jamo, Ethiopic
3101 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3102 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3103 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3104 * Combining Diacritical Marks Extended, Balinese,
3105 * Sundanese, Batak, Lepcha, Ol Chiki
3106 * 2000..206F General Punctuation
3107 */
3108
3109 if (uv1 >= 0x2D30) {
3110
3111 /* This keeps the from needing further processing the code points
3112 * most likely to be used in the following non-cased major scripts:
3113 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3114 *
3115 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3116 * event that Unicode eventually allocates the unused block as of
3117 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3118 * that the test suite will start having failures to alert you
3119 * should that happen) */
3120 if (uv1 < 0xA640) {
3121 goto cases_to_self;
3122 }
3123
3124 if (uv1 >= 0xAC00) {
3125 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3126 if (ckWARN_d(WARN_SURROGATE)) {
3127 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3128 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3129 "Operation \"%s\" returns its argument for"
3130 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3131 }
3132 goto cases_to_self;
3133 }
36eaa811
KW
3134
3135 /* AC00..FAFF Catches Hangul syllables and private use, plus
3136 * some others */
3137 if (uv1 < 0xFB00) {
3138 goto cases_to_self;
3139
3140 }
3141
5af9bc97 3142 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3143 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3144 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3145 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3146 }
3147 if (ckWARN_d(WARN_NON_UNICODE)) {
3148 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3149 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3150 "Operation \"%s\" returns its argument for"
3151 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3152 }
3153 goto cases_to_self;
3154 }
3bfc1e70
KW
3155#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3156 if (UNLIKELY(uv1
3157 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3158 {
3159
56576a04
KW
3160 /* As of Unicode 10.0, this means we avoid swash creation
3161 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3162 goto cases_to_self;
3163 }
3164#endif
36eaa811
KW
3165 }
3166 }
9ae3ac1a 3167
36eaa811
KW
3168 /* Note that non-characters are perfectly legal, so no warning should
3169 * be given. There are so few of them, that it isn't worth the extra
3170 * tests to avoid swash creation */
9ae3ac1a
KW
3171 }
3172
0134edef 3173 if (!*swashp) /* load on-demand */
56576a04
KW
3174 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3175 4, 0, NULL, NULL);
0134edef 3176
a6f87d8c 3177 if (special) {
0134edef 3178 /* It might be "special" (sometimes, but not always,
2a37f04d 3179 * a multicharacter mapping) */
4a8240a3 3180 HV *hv = NULL;
b08cf34e
JH
3181 SV **svp;
3182
4a8240a3
KW
3183 /* If passed in the specials name, use that; otherwise use any
3184 * given in the swash */
3185 if (*special != '\0') {
3186 hv = get_hv(special, 0);
3187 }
3188 else {
3189 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3190 if (svp) {
3191 hv = MUTABLE_HV(SvRV(*svp));
3192 }
3193 }
3194
176fe009 3195 if (hv
5f560d8a 3196 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3197 && (*svp))
3198 {
cfd0369c 3199 const char *s;
47654450 3200
cfd0369c 3201 s = SvPV_const(*svp, len);
47654450 3202 if (len == 1)
f4cd282c 3203 /* EIGHTBIT */
c80e42f3 3204 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3205 else {
d2dcd0fb 3206 Copy(s, ustrp, len, U8);
29e98929 3207 }
983ffd37 3208 }
0134edef
JH
3209 }
3210
3211 if (!len && *swashp) {
4a4088c4 3212 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3213
0134edef
JH
3214 if (uv2) {
3215 /* It was "normal" (a single character mapping). */
f4cd282c 3216 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3217 }
3218 }
1feea2c7 3219
cbe07460
KW
3220 if (len) {
3221 if (lenp) {
3222 *lenp = len;
3223 }
3224 return valid_utf8_to_uvchr(ustrp, 0);
3225 }
3226
3227 /* Here, there was no mapping defined, which means that the code point maps
3228 * to itself. Return the inputs */
e24dfe9c 3229 cases_to_self:
bfdf22ec 3230 len = UTF8SKIP(p);
ca9fab46
KW
3231 if (p != ustrp) { /* Don't copy onto itself */
3232 Copy(p, ustrp, len, U8);
3233 }
0134edef 3234
2a37f04d
JH
3235 if (lenp)
3236 *lenp = len;
3237
f4cd282c 3238 return uv1;
cbe07460 3239
a0ed51b3
LW
3240}
3241
051a06d4 3242STATIC UV
56576a04
KW
3243S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3244 U8* const ustrp, STRLEN *lenp)
051a06d4 3245{
4a4088c4 3246 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3247 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3248 * result contains a character that crosses the 255/256 boundary, disallow
3249 * the change, and return the original code point. See L<perlfunc/lc> for
3250 * why;
051a06d4 3251 *
a1433954
KW
3252 * p points to the original string whose case was changed; assumed
3253 * by this routine to be well-formed
051a06d4 3254 * result the code point of the first character in the changed-case string
56576a04
KW
3255 * ustrp points to the changed-case string (<result> represents its
3256 * first char)
051a06d4
KW
3257 * lenp points to the length of <ustrp> */
3258
3259 UV original; /* To store the first code point of <p> */
3260
3261 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3262
a4f12ed7 3263 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3264
3265 /* We know immediately if the first character in the string crosses the
3266 * boundary, so can skip */
3267 if (result > 255) {
3268
3269 /* Look at every character in the result; if any cross the
3270 * boundary, the whole thing is disallowed */
3271 U8* s = ustrp + UTF8SKIP(ustrp);
3272 U8* e = ustrp + *lenp;
3273 while (s < e) {
a4f12ed7 3274 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3275 goto bad_crossing;
3276 }
3277 s += UTF8SKIP(s);
3278 }
3279
613abc6d
KW
3280 /* Here, no characters crossed, result is ok as-is, but we warn. */
3281 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3282 return result;
3283 }
3284
7b52d656 3285 bad_crossing:
051a06d4
KW
3286
3287 /* Failed, have to return the original */
4b88fb76 3288 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3289
3290 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3291 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3292 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3293 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3294 OP_DESC(PL_op),
ab0b796c
KW
3295 original,
3296 original);
051a06d4
KW
3297 Copy(p, ustrp, *lenp, char);
3298 return original;
3299}
3300
607313a1
KW
3301STATIC U32
3302S_check_and_deprecate(pTHX_ const U8 *p,
3303 const U8 **e,
3304 const unsigned int type, /* See below */
3305 const bool use_locale, /* Is this a 'LC_'
3306 macro call? */
3307 const char * const file,
3308 const unsigned line)
3309{
3310 /* This is a temporary function to deprecate the unsafe calls to the case
3311 * changing macros and functions. It keeps all the special stuff in just
3312 * one place.
3313 *
3314 * It updates *e with the pointer to the end of the input string. If using
3315 * the old-style macros, *e is NULL on input, and so this function assumes
3316 * the input string is long enough to hold the entire UTF-8 sequence, and
3317 * sets *e accordingly, but it then returns a flag to pass the
3318 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3319 * using the full length if possible.
3320 *
3321 * It also does the assert that *e > p when *e is not NULL. This should be
3322 * migrated to the callers when this function gets deleted.
3323 *
3324 * The 'type' parameter is used for the caller to specify which case
3325 * changing function this is called from: */
3326
3327# define DEPRECATE_TO_UPPER 0
3328# define DEPRECATE_TO_TITLE 1
3329# define DEPRECATE_TO_LOWER 2
3330# define DEPRECATE_TO_FOLD 3
3331
3332 U32 utf8n_flags = 0;
3333 const char * name;
3334 const char * alternative;
3335
3336 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3337
3338 if (*e == NULL) {
3339 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3340 *e = p + UTF8SKIP(p);
3341
3342 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3343 * there. It could be part of a larger path */
607313a1 3344 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3345 name = instr(file, "mathoms.c")
607313a1
KW
3346 ? "to_utf8_upper"
3347 : "toUPPER_utf8";
3348 alternative = "toUPPER_utf8_safe";
3349 }
3350 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3351 name = instr(file, "mathoms.c")
607313a1
KW
3352 ? "to_utf8_title"
3353 : "toTITLE_utf8";
3354 alternative = "toTITLE_utf8_safe";
3355 }
3356 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3357 name = instr(file, "mathoms.c")
607313a1
KW
3358 ? "to_utf8_lower"
3359 : "toLOWER_utf8";
3360 alternative = "toLOWER_utf8_safe";
3361 }
3362 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3363 name = instr(file, "mathoms.c")
607313a1
KW
3364 ? "to_utf8_fold"
3365 : "toFOLD_utf8";
3366 alternative = "toFOLD_utf8_safe";
3367 }
3368 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3369
3370 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3371 }
3372 else {
3373 assert (p < *e);
3374 }
3375
3376 return utf8n_flags;
3377}
3378
eaf412bf
KW
3379/* The process for changing the case is essentially the same for the four case
3380 * change types, except there are complications for folding. Otherwise the
3381 * difference is only which case to change to. To make sure that they all do
3382 * the same thing, the bodies of the functions are extracted out into the
3383 * following two macros. The functions are written with the same variable
3384 * names, and these are known and used inside these macros. It would be
3385 * better, of course, to have inline functions to do it, but since different
3386 * macros are called, depending on which case is being changed to, this is not
3387 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3388 * function can start with the common start macro, then finish with its special
3389 * handling; while the other three cases can just use the common end macro.
3390 *
3391 * The algorithm is to use the proper (passed in) macro or function to change
3392 * the case for code points that are below 256. The macro is used if using
3393 * locale rules for the case change; the function if not. If the code point is
3394 * above 255, it is computed from the input UTF-8, and another macro is called
3395 * to do the conversion. If necessary, the output is converted to UTF-8. If
3396 * using a locale, we have to check that the change did not cross the 255/256
3397 * boundary, see check_locale_boundary_crossing() for further details.
3398 *
3399 * The macros are split with the correct case change for the below-256 case
3400 * stored into 'result', and in the middle of an else clause for the above-255
3401 * case. At that point in the 'else', 'result' is not the final result, but is
3402 * the input code point calculated from the UTF-8. The fold code needs to
3403 * realize all this and take it from there.
3404 *
3405 * If you read the two macros as sequential, it's easier to understand what's
3406 * going on. */
3407#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3408 L1_func_extra_param) \
a239b1e2 3409 \
eaf412bf
KW
3410 if (flags & (locale_flags)) { \
3411 /* Treat a UTF-8 locale as not being in locale at all */ \
3412 if (IN_UTF8_CTYPE_LOCALE) { \
3413 flags &= ~(locale_flags); \
3414 } \
3415 else { \
3416 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3417 } \
3418 } \
3419 \
3420 if (UTF8_IS_INVARIANT(*p)) { \
3421 if (flags & (locale_flags)) { \
3422 result = LC_L1_change_macro(*p); \
3423 } \
3424 else { \
3425 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3426 } \
3427 } \
a239b1e2 3428 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3429 if (flags & (locale_flags)) { \
3430 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3431 *(p+1))); \
3432 } \
3433 else { \
3434 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3435 ustrp, lenp, L1_func_extra_param); \
3436 } \
3437 } \
fa8ab374
KW
3438 else { /* malformed UTF-8 or ord above 255 */ \
3439 STRLEN len_result; \
fa8ab374
KW
3440 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3441 if (len_result == (STRLEN) -1) { \
607313a1
KW
3442 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3443 1 /* Die */ ); \
fa8ab374 3444 }
eaf412bf
KW
3445
3446#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3447 result = change_macro(result, p, ustrp, lenp); \
3448 \
3449 if (flags & (locale_flags)) { \
3450 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3451 } \
3452 return result; \
3453 } \
3454 \
3455 /* Here, used locale rules. Convert back to UTF-8 */ \
3456 if (UTF8_IS_INVARIANT(result)) { \
3457 *ustrp = (U8) result; \
3458 *lenp = 1; \
3459 } \
3460 else { \
3461 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3462 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3463 *lenp = 2; \
3464 } \
3465 \
3466 return result;
3467
d3e79532 3468/*
87cea99e 3469=for apidoc to_utf8_upper
d3e79532 3470
a239b1e2 3471Instead use L</toUPPER_utf8_safe>.
a1433954 3472
d3e79532
JH
3473=cut */
3474
051a06d4 3475/* Not currently externally documented, and subject to change:
31f05a37
KW
3476 * <flags> is set iff iff the rules from the current underlying locale are to
3477 * be used. */
051a06d4 3478
2104c8d9 3479UV
607313a1
KW
3480Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3481 const U8 *e,
3482 U8* ustrp,
3483 STRLEN *lenp,
3484 bool flags,
3485 const char * const file,
3486 const int line)
a0ed51b3 3487{
051a06d4 3488 UV result;
607313a1
KW
3489 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3490 cBOOL(flags), file, line);
051a06d4
KW
3491
3492 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3493
eaf412bf
KW
3494 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3495 /* 2nd char of uc(U+DF) is 'S' */
3496 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3497 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3498}
a0ed51b3 3499
d3e79532 3500/*
87cea99e 3501=for apidoc to_utf8_title
d3e79532 3502
a239b1e2 3503Instead use L</toTITLE_utf8_safe>.
a1433954 3504
d3e79532
JH
3505=cut */
3506
051a06d4 3507/* Not currently externally documented, and subject to change:
31f05a37
KW
3508 * <flags> is set iff the rules from the current underlying locale are to be
3509 * used. Since titlecase is not defined in POSIX, for other than a
3510 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3511 */
051a06d4 3512
983ffd37 3513UV
607313a1
KW
3514Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3515 const U8 *e,
3516 U8* ustrp,
3517 STRLEN *lenp,
3518 bool flags,
3519 const char * const file,
3520 const int line)
983ffd37 3521{
051a06d4 3522 UV result;
607313a1
KW
3523 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3524 cBOOL(flags), file, line);
051a06d4
KW
3525
3526 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3527
eaf412bf
KW
3528 /* 2nd char of ucfirst(U+DF) is 's' */
3529 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3530 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3531}
3532
d3e79532 3533/*
87cea99e 3534=for apidoc to_utf8_lower
d3e79532 3535
a239b1e2 3536Instead use L</toLOWER_utf8_safe>.
a1433954 3537
d3e79532
JH
3538=cut */
3539
051a06d4 3540/* Not currently externally documented, and subject to change:
31f05a37
KW
3541 * <flags> is set iff iff the rules from the current underlying locale are to
3542 * be used.
3543 */
051a06d4 3544
2104c8d9 3545UV
607313a1
KW
3546Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3547 const U8 *e,
3548 U8* ustrp,
3549 STRLEN *lenp,
3550 bool flags,
3551 const char * const file,
3552 const int line)
a0ed51b3 3553{
051a06d4 3554 UV result;
607313a1
KW
3555 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3556 cBOOL(flags), file, line);
051a06d4 3557
051a06d4 3558 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3559
eaf412bf
KW
3560 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3561 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3562}
3563
d3e79532 3564/*
87cea99e 3565=for apidoc to_utf8_fold
d3e79532 3566
a239b1e2 3567Instead use L</toFOLD_utf8_safe>.
a1433954 3568
d3e79532
JH
3569=cut */
3570
051a06d4
KW
3571/* Not currently externally documented, and subject to change,
3572 * in <flags>
31f05a37
KW
3573 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3574 * locale are to be used.
051a06d4
KW
3575 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3576 * otherwise simple folds
a0270393
KW
3577 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3578 * prohibited
445bf929 3579 */
36bb2ab6 3580
b4e400f9 3581UV
607313a1
KW
3582Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3583 const U8 *e,
3584 U8* ustrp,
3585 STRLEN *lenp,
3586 U8 flags,
3587 const char * const file,
3588 const int line)
b4e400f9 3589{
051a06d4 3590 UV result;
607313a1
KW
3591 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3592 cBOOL(flags), file, line);
051a06d4 3593
36bb2ab6 3594 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3595
a0270393
KW
3596 /* These are mutually exclusive */
3597 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3598
50ba90ff
KW
3599 assert(p != ustrp); /* Otherwise overwrites */
3600
eaf412bf
KW
3601 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3602 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3603
eaf412bf 3604 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3605
1ca267a5
KW
3606 if (flags & FOLD_FLAGS_LOCALE) {
3607
76f2ffcd 3608# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3609# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3610# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd 3611
538e84ed
KW
3612 /* Special case these two characters, as what normally gets
3613 * returned under locale doesn't work */
b59bf0b2 3614 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S))
1ca267a5 3615 {
ab0b796c
KW
3616 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3617 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3618 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3619 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3620 goto return_long_s;
3621 }
0766489e
KW
3622 else
3623#endif
b59bf0b2 3624 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T))
9fc2026f 3625 {
ab0b796c
KW
3626 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3627 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3628 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3629 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3630 goto return_ligature_st;
3631 }
74894415
KW
3632
3633#if UNICODE_MAJOR_VERSION == 3 \
3634 && UNICODE_DOT_VERSION == 0 \
3635 && UNICODE_DOT_DOT_VERSION == 1
3636# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3637
3638 /* And special case this on this Unicode version only, for the same
3639 * reaons the other two are special cased. They would cross the
3640 * 255/256 boundary which is forbidden under /l, and so the code
3641 * wouldn't catch that they are equivalent (which they are only in
3642 * this release) */
b59bf0b2 3643 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) {
74894415
KW
3644 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3645 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3646 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3647 "resolved to \"\\x{0131}\".");
3648 goto return_dotless_i;
3649 }
3650#endif
3651
357aadde 3652 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3653 }
a0270393
KW
3654 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3655 return result;
3656 }
3657 else {
4a4088c4 3658 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3659 * character above the ASCII range, and the result should not
3660 * contain an ASCII character. */
a0270393
KW
3661
3662 UV original; /* To store the first code point of <p> */
3663
3664 /* Look at every character in the result; if any cross the
3665 * boundary, the whole thing is disallowed */
3666 U8* s = ustrp;
3667 U8* e = ustrp + *lenp;
3668 while (s < e) {
3669 if (isASCII(*s)) {
3670 /* Crossed, have to return the original */
3671 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3672
9fc2026f 3673 /* But in these instances, there is an alternative we can
1ca267a5 3674 * return that is valid */
0766489e
KW
3675 if (original == LATIN_SMALL_LETTER_SHARP_S
3676#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3677 || original == LATIN_CAPITAL_LETTER_SHARP_S
3678#endif
3679 ) {
1ca267a5
KW
3680 goto return_long_s;
3681 }
9fc2026f
KW
3682 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3683 goto return_ligature_st;
3684 }
74894415
KW
3685#if UNICODE_MAJOR_VERSION == 3 \
3686 && UNICODE_DOT_VERSION == 0 \
3687 && UNICODE_DOT_DOT_VERSION == 1
3688
3689 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3690 goto return_dotless_i;
3691 }
3692#endif
a0270393
KW
3693 Copy(p, ustrp, *lenp, char);
3694 return original;
3695 }
3696 s += UTF8SKIP(s);
3697 }
051a06d4 3698
a0270393
KW
3699 /* Here, no characters crossed, result is ok as-is */
3700 return result;
3701 }