This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regcomp.c: Silence compiler maybe uninit warnings
[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)
37657a5b
KW
1170are when an error is found. If you also need to know the generated warning
1171messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1172
1173It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1174all the others, C<errors>. If this parameter is 0, this function behaves
1175identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1176to a C<U32> variable, which this function sets to indicate any errors found.
1177Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1178C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1179of these bits will be set if a malformation is found, even if the input
7a65503b 1180C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1181exceptions are noted:
1182
1183=over 4
1184
57ff5f59 1185=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1186
57ff5f59
KW
1187The input sequence is not standard UTF-8, but a Perl extension. This bit is
1188set only if the input C<flags> parameter contains either the
1189C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1190
1191Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1192and so some extension must be used to express them. Perl uses a natural
1193extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1194extension to represent even higher ones, so that any code point that fits in a
119564-bit word can be represented. Text using these extensions is not likely to
1196be portable to non-Perl code. We lump both of these extensions together and
1197refer to them as Perl extended UTF-8. There exist other extensions that people
1198have invented, incompatible with Perl's.
1199
1200On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1201extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1202than on ASCII. Prior to that, code points 2**31 and higher were simply
1203unrepresentable, and a different, incompatible method was used to represent
1204code points between 2**30 and 2**31 - 1.
1205
1206On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1207Perl extended UTF-8 is used.
1208
1209In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1210may use for backward compatibility. That name is misleading, as this flag may
1211be set when the code point actually does fit in 31 bits. This happens on
1212EBCDIC platforms, and sometimes when the L<overlong
1213malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1214describes the situation in all cases.
f9380377
KW
1215
1216=item C<UTF8_GOT_CONTINUATION>
1217
1218The input sequence was malformed in that the first byte was a a UTF-8
1219continuation byte.
1220
1221=item C<UTF8_GOT_EMPTY>
1222
1223The input C<curlen> parameter was 0.
1224
1225=item C<UTF8_GOT_LONG>
1226
1227The input sequence was malformed in that there is some other sequence that
1228evaluates to the same code point, but that sequence is shorter than this one.
1229
fecaf136
KW
1230Until Unicode 3.1, it was legal for programs to accept this malformation, but
1231it was discovered that this created security issues.
1232
f9380377
KW
1233=item C<UTF8_GOT_NONCHAR>
1234
1235The code point represented by the input UTF-8 sequence is for a Unicode
1236non-character code point.
1237This bit is set only if the input C<flags> parameter contains either the
1238C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1239
1240=item C<UTF8_GOT_NON_CONTINUATION>
1241
1242The input sequence was malformed in that a non-continuation type byte was found
1243in a position where only a continuation type one should be.
1244
1245=item C<UTF8_GOT_OVERFLOW>
1246
1247The input sequence was malformed in that it is for a code point that is not
d22ec717 1248representable in the number of bits available in an IV on the current platform.
f9380377
KW
1249
1250=item C<UTF8_GOT_SHORT>
1251
1252The input sequence was malformed in that C<curlen> is smaller than required for
1253a complete sequence. In other words, the input is for a partial character
1254sequence.
1255
1256=item C<UTF8_GOT_SUPER>
1257
1258The input sequence was malformed in that it is for a non-Unicode code point;
1259that is, one above the legal Unicode maximum.
1260This bit is set only if the input C<flags> parameter contains either the
1261C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1262
1263=item C<UTF8_GOT_SURROGATE>
1264
1265The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1266code point.
1267This bit is set only if the input C<flags> parameter contains either the
1268C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1269
1270=back
1271
133551d8
KW
1272To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1273flag to suppress any warnings, and then examine the C<*errors> return.
1274
f9380377 1275=cut
37657a5b
KW
1276
1277Also implemented as a macro in utf8.h
37607a96 1278*/
67e989fb 1279
a0ed51b3 1280UV
f9380377 1281Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
37657a5b
KW
1282 STRLEN curlen,
1283 STRLEN *retlen,
1284 const U32 flags,
1285 U32 * errors)
1286{
1287 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1288
1289 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1290}
1291
1292/*
1293
1294=for apidoc utf8n_to_uvchr_msgs
1295
1296THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1297Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1298
1299This function is for code that needs to know what the precise malformation(s)
1300are when an error is found, and wants the corresponding warning and/or error
1301messages to be returned to the caller rather than be displayed. All messages
1302that would have been displayed if all lexcial warnings are enabled will be
1303returned.
1304
1305It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1306placed after all the others, C<msgs>. If this parameter is 0, this function
1307behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1308be a pointer to an C<AV *> variable, in which this function creates a new AV to
1309contain any appropriate messages. The elements of the array are ordered so
1310that the first message that would have been displayed is in the 0th element,
1311and so on. Each element is a hash with three key-value pairs, as follows:
1312
1313=over 4
1314
1315=item C<text>
1316
1317The text of the message as a C<SVpv>.
1318
1319=item C<warn_categories>
1320
1321The warning category (or categories) packed into a C<SVuv>.
1322
1323=item C<flag>
1324
1325A single flag bit associated with this message, in a C<SVuv>.
1326The bit corresponds to some bit in the C<*errors> return value,
1327such as C<UTF8_GOT_LONG>.
1328
1329=back
1330
1331It's important to note that specifying this parameter as non-null will cause
1332any warnings this function would otherwise generate to be suppressed, and
1333instead be placed in C<*msgs>. The caller can check the lexical warnings state
1334(or not) when choosing what to do with the returned messages.
1335
1336If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1337no AV is created.
1338
1339The caller, of course, is responsible for freeing any returned AV.
1340
1341=cut
1342*/
1343
1344UV
1345Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1346 STRLEN curlen,
1347 STRLEN *retlen,
1348 const U32 flags,
1349 U32 * errors,
1350 AV ** msgs)
a0ed51b3 1351{
d4c19fe8 1352 const U8 * const s0 = s;
2b5e7bc2
KW
1353 U8 * send = NULL; /* (initialized to silence compilers' wrong
1354 warning) */
1355 U32 possible_problems = 0; /* A bit is set here for each potential problem
1356 found as we go along */
eb83ed87 1357 UV uv = *s;
2b5e7bc2
KW
1358 STRLEN expectlen = 0; /* How long should this sequence be?
1359 (initialized to silence compilers' wrong
1360 warning) */
e308b348 1361 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1362 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1363 this gets set and discarded */
a0dbb045 1364
2b5e7bc2
KW
1365 /* The below are used only if there is both an overlong malformation and a
1366 * too short one. Otherwise the first two are set to 's0' and 'send', and
1367 * the third not used at all */
1368 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1369 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1370 routine; see [perl #130921] */
2b5e7bc2 1371 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1372
37657a5b 1373 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
f9380377
KW
1374
1375 if (errors) {
1376 *errors = 0;
1377 }
1378 else {
1379 errors = &discard_errors;
1380 }
a0dbb045 1381
eb83ed87
KW
1382 /* The order of malformation tests here is important. We should consume as
1383 * few bytes as possible in order to not skip any valid character. This is
1384 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1385 * http://unicode.org/reports/tr36 for more discussion as to why. For
1386 * example, once we've done a UTF8SKIP, we can tell the expected number of
1387 * bytes, and could fail right off the bat if the input parameters indicate
1388 * that there are too few available. But it could be that just that first
1389 * byte is garbled, and the intended character occupies fewer bytes. If we
1390 * blindly assumed that the first byte is correct, and skipped based on
1391 * that number, we could skip over a valid input character. So instead, we
1392 * always examine the sequence byte-by-byte.
1393 *
1394 * We also should not consume too few bytes, otherwise someone could inject
1395 * things. For example, an input could be deliberately designed to
1396 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1397 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1398 * which is actually part of of the overflowing sequence), that could look
1399 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1400 * sequence and process the rest, inappropriately.
1401 *
1402 * Some possible input sequences are malformed in more than one way. This
1403 * function goes to lengths to try to find all of them. This is necessary
1404 * for correctness, as the inputs may allow one malformation but not
1405 * another, and if we abandon searching for others after finding the
1406 * allowed one, we could allow in something that shouldn't have been.
1407 */
eb83ed87 1408
b5b9af04 1409 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1410 possible_problems |= UTF8_GOT_EMPTY;
1411 curlen = 0;
5a48568d 1412 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1413 goto ready_to_handle_errors;
0c443dc2
JH
1414 }
1415
eb83ed87
KW
1416 expectlen = UTF8SKIP(s);
1417
1418 /* A well-formed UTF-8 character, as the vast majority of calls to this
1419 * function will be for, has this expected length. For efficiency, set
1420 * things up here to return it. It will be overriden only in those rare
1421 * cases where a malformation is found */
1422 if (retlen) {
1423 *retlen = expectlen;
1424 }
1425
1426 /* An invariant is trivially well-formed */
1d72bdf6 1427 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1428 return uv;
a0ed51b3 1429 }
67e989fb 1430
eb83ed87 1431 /* A continuation character can't start a valid sequence */
b5b9af04 1432 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1433 possible_problems |= UTF8_GOT_CONTINUATION;
1434 curlen = 1;
1435 uv = UNICODE_REPLACEMENT;
1436 goto ready_to_handle_errors;
ba210ebe 1437 }
9041c2e3 1438
dcd27b3c 1439 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1440 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1441 * because it excludes start bytes like \xC0 that always lead to
1442 * overlongs.) */
dcd27b3c 1443
534752c1
KW
1444 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1445 * that indicate the number of bytes in the character's whole UTF-8
1446 * sequence, leaving just the bits that are part of the value. */
1447 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1448
e308b348
KW
1449 /* Setup the loop end point, making sure to not look past the end of the
1450 * input string, and flag it as too short if the size isn't big enough. */
1451 send = (U8*) s0;
1452 if (UNLIKELY(curlen < expectlen)) {
1453 possible_problems |= UTF8_GOT_SHORT;
1454 avail_len = curlen;
1455 send += curlen;
1456 }
1457 else {
1458 send += expectlen;
1459 }
e308b348 1460
eb83ed87 1461 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1462 * accumulating each into the working value as we go. */
eb83ed87 1463 for (s = s0 + 1; s < send; s++) {
b5b9af04 1464 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1465 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1466 continue;
1467 }
1468
1469 /* Here, found a non-continuation before processing all expected bytes.
1470 * This byte indicates the beginning of a new character, so quit, even
1471 * if allowing this malformation. */
2b5e7bc2 1472 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1473 break;
eb83ed87
KW
1474 } /* End of loop through the character's bytes */
1475
1476 /* Save how many bytes were actually in the character */
1477 curlen = s - s0;
1478
2b5e7bc2
KW
1479 /* Note that there are two types of too-short malformation. One is when
1480 * there is actual wrong data before the normal termination of the
1481 * sequence. The other is that the sequence wasn't complete before the end
1482 * of the data we are allowed to look at, based on the input 'curlen'.
1483 * This means that we were passed data for a partial character, but it is
1484 * valid as far as we saw. The other is definitely invalid. This
1485 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1486 * separate.
1487 *
1488 * A convenience macro that matches either of the too-short conditions. */
1489# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1490
1491 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1492 uv_so_far = uv;
1493 uv = UNICODE_REPLACEMENT;
1494 }
2b5e7bc2 1495
08e73697
KW
1496 /* Check for overflow. The algorithm requires us to not look past the end
1497 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1498 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1499 1 /* Do consider overlongs */
1500 )))
1501 {
2b5e7bc2
KW
1502 possible_problems |= UTF8_GOT_OVERFLOW;
1503 uv = UNICODE_REPLACEMENT;
eb83ed87 1504 }
eb83ed87 1505
2b5e7bc2
KW
1506 /* Check for overlong. If no problems so far, 'uv' is the correct code
1507 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1508 * we must look at the UTF-8 byte sequence itself to see if it is for an
1509 * overlong */
1510 if ( ( LIKELY(! possible_problems)
1511 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1512 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1513 && ( UNLIKELY(! UTF8_IS_START(*s0))
1514 || ( curlen > 1
d6be65ae 1515 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1516 s - s0))))))
2f8f112e 1517 {
2b5e7bc2
KW
1518 possible_problems |= UTF8_GOT_LONG;
1519
abc28b54 1520 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1521
abc28b54
KW
1522 /* The calculation in the 'true' branch of this 'if'
1523 * below won't work if overflows, and isn't needed
1524 * anyway. Further below we handle all overflow
1525 * cases */
1526 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1527 {
2b5e7bc2
KW
1528 UV min_uv = uv_so_far;
1529 STRLEN i;
1530
1531 /* Here, the input is both overlong and is missing some trailing
1532 * bytes. There is no single code point it could be for, but there
1533 * may be enough information present to determine if what we have
1534 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1535 * The code further below has the intelligence to determine this,
1536 * but just for non-overlong UTF-8 sequences. What we do here is
1537 * calculate the smallest code point the input could represent if
1538 * there were no too short malformation. Then we compute and save
1539 * the UTF-8 for that, which is what the code below looks at
1540 * instead of the raw input. It turns out that the smallest such
1541 * code point is all we need. */
2b5e7bc2
KW
1542 for (i = curlen; i < expectlen; i++) {
1543 min_uv = UTF8_ACCUMULATE(min_uv,
1544 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1545 }
1546
e9f2c446 1547 adjusted_s0 = temp_char_buf;
57ff5f59 1548 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1549 }
eb83ed87
KW
1550 }
1551
56576a04
KW
1552 /* Here, we have found all the possible problems, except for when the input
1553 * is for a problematic code point not allowed by the input parameters. */
1554
06188866
KW
1555 /* uv is valid for overlongs */
1556 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1557
1558 /* isn't problematic if < this */
1559 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1560 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1561
1562 /* if overflow, we know without looking further
1563 * precisely which of the problematic types it is,
1564 * and we deal with those in the overflow handling
1565 * code */
1566 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1567 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1568 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1569 && ((flags & ( UTF8_DISALLOW_NONCHAR
1570 |UTF8_DISALLOW_SURROGATE
1571 |UTF8_DISALLOW_SUPER
d044b7a7 1572 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1573 |UTF8_WARN_NONCHAR
1574 |UTF8_WARN_SURROGATE
1575 |UTF8_WARN_SUPER
d22ec717 1576 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1577 {
2b5e7bc2
KW
1578 /* If there were no malformations, or the only malformation is an
1579 * overlong, 'uv' is valid */
1580 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1581 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1582 possible_problems |= UTF8_GOT_SURROGATE;
1583 }
1584 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1585 possible_problems |= UTF8_GOT_SUPER;
1586 }
1587 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1588 possible_problems |= UTF8_GOT_NONCHAR;
1589 }
1590 }
1591 else { /* Otherwise, need to look at the source UTF-8, possibly
1592 adjusted to be non-overlong */
1593
1594 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1595 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1596 {
2b5e7bc2
KW
1597 possible_problems |= UTF8_GOT_SUPER;
1598 }
1599 else if (curlen > 1) {
1600 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1601 NATIVE_UTF8_TO_I8(*adjusted_s0),
1602 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1603 {
2b5e7bc2 1604 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1605 }
2b5e7bc2
KW
1606 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1607 NATIVE_UTF8_TO_I8(*adjusted_s0),
1608 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1609 {
1610 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1611 }
1612 }
c0236afe 1613
2b5e7bc2
KW
1614 /* We need a complete well-formed UTF-8 character to discern
1615 * non-characters, so can't look for them here */
1616 }
1617 }
949cf498 1618
2b5e7bc2
KW
1619 ready_to_handle_errors:
1620
1621 /* At this point:
1622 * curlen contains the number of bytes in the sequence that
1623 * this call should advance the input by.
e308b348
KW
1624 * avail_len gives the available number of bytes passed in, but
1625 * only if this is less than the expected number of
1626 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1627 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1628 * is set in it for each potential problem found.
1629 * uv contains the code point the input sequence
1630 * represents; or if there is a problem that prevents
1631 * a well-defined value from being computed, it is
1632 * some subsitute value, typically the REPLACEMENT
1633 * CHARACTER.
1634 * s0 points to the first byte of the character
56576a04
KW
1635 * s points to just after were we left off processing
1636 * the character
1637 * send points to just after where that character should
1638 * end, based on how many bytes the start byte tells
1639 * us should be in it, but no further than s0 +
1640 * avail_len
2b5e7bc2 1641 */
eb83ed87 1642
2b5e7bc2
KW
1643 if (UNLIKELY(possible_problems)) {
1644 bool disallowed = FALSE;
1645 const U32 orig_problems = possible_problems;
1646
37657a5b
KW
1647 if (msgs) {
1648 *msgs = NULL;
1649 }
1650
2b5e7bc2
KW
1651 while (possible_problems) { /* Handle each possible problem */
1652 UV pack_warn = 0;
1653 char * message = NULL;
37657a5b 1654 U32 this_flag_bit = 0;
2b5e7bc2
KW
1655
1656 /* Each 'if' clause handles one problem. They are ordered so that
1657 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1658 * ones; this is kinda in decreasing severity order. But the
1659 * overlong must come last, as it changes 'uv' looked at by the
1660 * others */
2b5e7bc2
KW
1661 if (possible_problems & UTF8_GOT_OVERFLOW) {
1662
56576a04
KW
1663 /* Overflow means also got a super and are using Perl's
1664 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1665 possible_problems
d044b7a7 1666 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1667 *errors |= UTF8_GOT_OVERFLOW;
1668
1669 /* But the API says we flag all errors found */
1670 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1671 *errors |= UTF8_GOT_SUPER;
1672 }
ddb65933 1673 if (flags
d044b7a7 1674 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1675 {
d044b7a7 1676 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1677 }
2b5e7bc2 1678
d60baaa7 1679 /* Disallow if any of the three categories say to */
56576a04 1680 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1681 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1682 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1683 {
1684 disallowed = TRUE;
1685 }
1686
d22ec717
KW
1687 /* Likewise, warn if any say to */
1688 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1689 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1690 {
2b5e7bc2 1691
ddb65933
KW
1692 /* The warnings code explicitly says it doesn't handle the
1693 * case of packWARN2 and two categories which have
1694 * parent-child relationship. Even if it works now to
1695 * raise the warning if either is enabled, it wouldn't
1696 * necessarily do so in the future. We output (only) the
56576a04 1697 * most dire warning */
ddb65933 1698 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1699 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1700 pack_warn = packWARN(WARN_UTF8);
1701 }
37657a5b 1702 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1703 pack_warn = packWARN(WARN_NON_UNICODE);
1704 }
1705 if (pack_warn) {
1706 message = Perl_form(aTHX_ "%s: %s (overflows)",
1707 malformed_text,
05b9033b 1708 _byte_dump_string(s0, curlen, 0));
37657a5b 1709 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1710 }
2b5e7bc2
KW
1711 }
1712 }
1713 }
1714 else if (possible_problems & UTF8_GOT_EMPTY) {
1715 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1716 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1717
1718 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1719
1720 /* This so-called malformation is now treated as a bug in
1721 * the caller. If you have nothing to decode, skip calling
1722 * this function */
1723 assert(0);
1724
2b5e7bc2 1725 disallowed = TRUE;
37657a5b
KW
1726 if ( (msgs
1727 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1728 {
2b5e7bc2
KW
1729 pack_warn = packWARN(WARN_UTF8);
1730 message = Perl_form(aTHX_ "%s (empty string)",
1731 malformed_text);
37657a5b 1732 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1733 }
1734 }
1735 }
1736 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1737 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1738 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1739
1740 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1741 disallowed = TRUE;
37657a5b
KW
1742 if (( msgs
1743 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1744 {
2b5e7bc2
KW
1745 pack_warn = packWARN(WARN_UTF8);
1746 message = Perl_form(aTHX_
1747 "%s: %s (unexpected continuation byte 0x%02x,"
1748 " with no preceding start byte)",
1749 malformed_text,
7e2f38b2 1750 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1751 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1752 }
1753 }
1754 }
2b5e7bc2
KW
1755 else if (possible_problems & UTF8_GOT_SHORT) {
1756 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1757 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1758
1759 if (! (flags & UTF8_ALLOW_SHORT)) {
1760 disallowed = TRUE;
37657a5b
KW
1761 if (( msgs
1762 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1763 {
2b5e7bc2
KW
1764 pack_warn = packWARN(WARN_UTF8);
1765 message = Perl_form(aTHX_
56576a04
KW
1766 "%s: %s (too short; %d byte%s available, need %d)",
1767 malformed_text,
1768 _byte_dump_string(s0, send - s0, 0),
1769 (int)avail_len,
1770 avail_len == 1 ? "" : "s",
1771 (int)expectlen);
37657a5b 1772 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1773 }
1774 }
ba210ebe 1775
2b5e7bc2 1776 }
e308b348
KW
1777 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1778 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1779 *errors |= UTF8_GOT_NON_CONTINUATION;
1780
1781 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1782 disallowed = TRUE;
37657a5b
KW
1783 if (( msgs
1784 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1785 {
99a765e9
KW
1786
1787 /* If we don't know for sure that the input length is
1788 * valid, avoid as much as possible reading past the
1789 * end of the buffer */
1790 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1791 ? s - s0
1792 : send - s0;
e308b348
KW
1793 pack_warn = packWARN(WARN_UTF8);
1794 message = Perl_form(aTHX_ "%s",
1795 unexpected_non_continuation_text(s0,
99a765e9 1796 printlen,
e308b348
KW
1797 s - s0,
1798 (int) expectlen));
37657a5b 1799 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
1800 }
1801 }
1802 }
2b5e7bc2
KW
1803 else if (possible_problems & UTF8_GOT_SURROGATE) {
1804 possible_problems &= ~UTF8_GOT_SURROGATE;
1805
f9380377
KW
1806 if (flags & UTF8_WARN_SURROGATE) {
1807 *errors |= UTF8_GOT_SURROGATE;
1808
1809 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 1810 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 1811 {
2b5e7bc2
KW
1812 pack_warn = packWARN(WARN_SURROGATE);
1813
1814 /* These are the only errors that can occur with a
1815 * surrogate when the 'uv' isn't valid */
1816 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1817 message = Perl_form(aTHX_
1818 "UTF-16 surrogate (any UTF-8 sequence that"
1819 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1820 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1821 }
1822 else {
c94c2f39 1823 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1824 }
37657a5b 1825 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 1826 }
2b5e7bc2 1827 }
ba210ebe 1828
2b5e7bc2
KW
1829 if (flags & UTF8_DISALLOW_SURROGATE) {
1830 disallowed = TRUE;
f9380377 1831 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1832 }
1833 }
1834 else if (possible_problems & UTF8_GOT_SUPER) {
1835 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1836
f9380377
KW
1837 if (flags & UTF8_WARN_SUPER) {
1838 *errors |= UTF8_GOT_SUPER;
1839
1840 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 1841 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 1842 {
2b5e7bc2
KW
1843 pack_warn = packWARN(WARN_NON_UNICODE);
1844
1845 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1846 message = Perl_form(aTHX_
1847 "Any UTF-8 sequence that starts with"
1848 " \"%s\" is for a non-Unicode code point,"
1849 " may not be portable",
7e2f38b2 1850 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1851 }
1852 else {
c94c2f39 1853 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1854 }
37657a5b 1855 this_flag_bit = UTF8_GOT_SUPER;
f9380377 1856 }
2b5e7bc2 1857 }
ba210ebe 1858
57ff5f59
KW
1859 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1860 * and before possibly bailing out, so that the more dire
1861 * warning will override the regular one. */
1862 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1863 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1864 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 1865 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 1866 {
db0f09e6 1867 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1868
57ff5f59
KW
1869 /* If it is an overlong that evaluates to a code point
1870 * that doesn't have to use the Perl extended UTF-8, it
1871 * still used it, and so we output a message that
1872 * doesn't refer to the code point. The same is true
1873 * if there was a SHORT malformation where the code
1874 * point is not valid. In that case, 'uv' will have
1875 * been set to the REPLACEMENT CHAR, and the message
1876 * below without the code point in it will be selected
1877 * */
1878 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1879 message = Perl_form(aTHX_
57ff5f59 1880 perl_extended_cp_format, uv);
2b5e7bc2
KW
1881 }
1882 else {
1883 message = Perl_form(aTHX_
57ff5f59
KW
1884 "Any UTF-8 sequence that starts with"
1885 " \"%s\" is a Perl extension, and"
1886 " so is not portable",
1887 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 1888 }
37657a5b 1889 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
1890 }
1891
d044b7a7
KW
1892 if (flags & ( UTF8_WARN_PERL_EXTENDED
1893 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1894 {
d044b7a7 1895 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1896
d044b7a7 1897 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1898 disallowed = TRUE;
1899 }
2b5e7bc2
KW
1900 }
1901 }
eb83ed87 1902
2b5e7bc2 1903 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1904 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1905 disallowed = TRUE;
1906 }
2b5e7bc2
KW
1907 }
1908 else if (possible_problems & UTF8_GOT_NONCHAR) {
1909 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1910
f9380377
KW
1911 if (flags & UTF8_WARN_NONCHAR) {
1912 *errors |= UTF8_GOT_NONCHAR;
1913
1914 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 1915 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 1916 {
2b5e7bc2
KW
1917 /* The code above should have guaranteed that we don't
1918 * get here with errors other than overlong */
1919 assert (! (orig_problems
1920 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1921
1922 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1923 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 1924 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 1925 }
2b5e7bc2 1926 }
5b311467 1927
2b5e7bc2
KW
1928 if (flags & UTF8_DISALLOW_NONCHAR) {
1929 disallowed = TRUE;
f9380377 1930 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1931 }
6c64cd9d
KW
1932 }
1933 else if (possible_problems & UTF8_GOT_LONG) {
1934 possible_problems &= ~UTF8_GOT_LONG;
1935 *errors |= UTF8_GOT_LONG;
1936
1937 if (flags & UTF8_ALLOW_LONG) {
1938
1939 /* We don't allow the actual overlong value, unless the
1940 * special extra bit is also set */
1941 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1942 & ~UTF8_ALLOW_LONG)))
1943 {
1944 uv = UNICODE_REPLACEMENT;
1945 }
1946 }
1947 else {
1948 disallowed = TRUE;
1949
37657a5b
KW
1950 if (( msgs
1951 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1952 {
6c64cd9d
KW
1953 pack_warn = packWARN(WARN_UTF8);
1954
1955 /* These error types cause 'uv' to be something that
1956 * isn't what was intended, so can't use it in the
1957 * message. The other error types either can't
1958 * generate an overlong, or else the 'uv' is valid */
1959 if (orig_problems &
1960 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1961 {
1962 message = Perl_form(aTHX_
1963 "%s: %s (any UTF-8 sequence that starts"
1964 " with \"%s\" is overlong which can and"
1965 " should be represented with a"
1966 " different, shorter sequence)",
1967 malformed_text,
1968 _byte_dump_string(s0, send - s0, 0),
1969 _byte_dump_string(s0, curlen, 0));
1970 }
1971 else {
1972 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
1973 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1974 uv, 0);
d819dc50
KW
1975 /* Don't use U+ for non-Unicode code points, which
1976 * includes those in the Latin1 range */
1977 const char * preface = ( uv > PERL_UNICODE_MAX
1978#ifdef EBCDIC
1979 || uv <= 0xFF
1980#endif
1981 )
1982 ? "0x"
1983 : "U+";
6c64cd9d
KW
1984 message = Perl_form(aTHX_
1985 "%s: %s (overlong; instead use %s to represent"
1986 " %s%0*" UVXf ")",
1987 malformed_text,
1988 _byte_dump_string(s0, send - s0, 0),
1989 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1990 preface,
1991 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1992 small code points */
1be62ab9 1993 UNI_TO_NATIVE(uv));
6c64cd9d 1994 }
37657a5b 1995 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
1996 }
1997 }
2b5e7bc2
KW
1998 } /* End of looking through the possible flags */
1999
2000 /* Display the message (if any) for the problem being handled in
2001 * this iteration of the loop */
2002 if (message) {
37657a5b
KW
2003 if (msgs) {
2004 SV* msg_sv = newSVpv(message, 0);
2005 SV* category_sv = newSVuv(pack_warn);
2006 SV* flag_bit_sv = newSVuv(this_flag_bit);
2007 HV* msg_hv = newHV();
2008
2009 assert(this_flag_bit);
2010
2011 if (*msgs == NULL) {
2012 *msgs = newAV();
2013 }
2014
2015 hv_stores(msg_hv, "text", msg_sv);
2016 hv_stores(msg_hv, "warn_categories", category_sv);
2017 hv_stores(msg_hv, "flag_bit", flag_bit_sv);
2018
2019 av_push(*msgs, newRV_noinc((SV*)msg_hv));
2020 }
2021 else if (PL_op)
2b5e7bc2
KW
2022 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2023 OP_DESC(PL_op));
2024 else
2025 Perl_warner(aTHX_ pack_warn, "%s", message);
2026 }
ddb65933 2027 } /* End of 'while (possible_problems)' */
a0dbb045 2028
2b5e7bc2
KW
2029 /* Since there was a possible problem, the returned length may need to
2030 * be changed from the one stored at the beginning of this function.
2031 * Instead of trying to figure out if that's needed, just do it. */
2032 if (retlen) {
2033 *retlen = curlen;
2034 }
a0dbb045 2035
2b5e7bc2
KW
2036 if (disallowed) {
2037 if (flags & UTF8_CHECK_ONLY && retlen) {
2038 *retlen = ((STRLEN) -1);
2039 }
2040 return 0;
2041 }
eb83ed87 2042 }
ba210ebe 2043
2b5e7bc2 2044 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2045}
2046
8e84507e 2047/*
ec5f19d0
KW
2048=for apidoc utf8_to_uvchr_buf
2049
2050Returns the native code point of the first character in the string C<s> which
2051is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2052C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2053
524080c4
KW
2054If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2055enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2056C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2057(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2058C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2059the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2060See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2061returned.
ec5f19d0
KW
2062
2063=cut
52be2536
KW
2064
2065Also implemented as a macro in utf8.h
2066
ec5f19d0
KW
2067*/
2068
2069
2070UV
2071Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2072{
7f974d7e
KW
2073 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2074
ec5f19d0
KW
2075 assert(s < send);
2076
2077 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 2078 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
2079}
2080
52be2536
KW
2081/* This is marked as deprecated
2082 *
ec5f19d0
KW
2083=for apidoc utf8_to_uvuni_buf
2084
de69f3af
KW
2085Only in very rare circumstances should code need to be dealing in Unicode
2086(as opposed to native) code points. In those few cases, use
2087C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
2088
2089Returns the Unicode (not-native) code point of the first character in the
2090string C<s> which
ec5f19d0
KW
2091is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2092C<retlen> will be set to the length, in bytes, of that character.
2093
524080c4
KW
2094If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2095enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2096NULL) to -1. If those warnings are off, the computed value if well-defined (or
2097the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2098is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2099next possible position in C<s> that could begin a non-malformed character.
de69f3af 2100See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2101
2102=cut
2103*/
2104
2105UV
2106Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2107{
2108 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2109
2110 assert(send > s);
2111
5962d97e 2112 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2113}
2114
b76347f2 2115/*
87cea99e 2116=for apidoc utf8_length
b76347f2 2117
b2e7ed74
KW
2118Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2119at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2120same place, it returns 0 with no warning raised.
2121
2122If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2123and returns the number of valid characters.
b76347f2
JH
2124
2125=cut
2126*/
2127
2128STRLEN
35a4481c 2129Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2130{
2131 STRLEN len = 0;
2132
7918f24d
NC
2133 PERL_ARGS_ASSERT_UTF8_LENGTH;
2134
8850bf83
JH
2135 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2136 * the bitops (especially ~) can create illegal UTF-8.
2137 * In other words: in Perl UTF-8 is not just for Unicode. */
2138
a3b680e6
AL
2139 if (e < s)
2140 goto warn_and_return;
b76347f2 2141 while (s < e) {
4cbf4130 2142 s += UTF8SKIP(s);
8e91ec7f
AV
2143 len++;
2144 }
2145
2146 if (e != s) {
2147 len--;
2148 warn_and_return:
9b387841
NC
2149 if (PL_op)
2150 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2151 "%s in %s", unees, OP_DESC(PL_op));
2152 else
61a12c31 2153 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2154 }
2155
2156 return len;
2157}
2158
b06226ff 2159/*
fed3ba5d
NC
2160=for apidoc bytes_cmp_utf8
2161
a1433954 2162Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2163sequence of characters (stored as UTF-8)
2164in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2165equal, -1 or -2 if the first string is less than the second string, +1 or +2
2166if the first string is greater than the second string.
2167
2168-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2169longer string. -2 or +2 is returned if
2170there was a difference between characters
fed3ba5d
NC
2171within the strings.
2172
2173=cut
2174*/
2175
2176int
2177Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2178{
2179 const U8 *const bend = b + blen;
2180 const U8 *const uend = u + ulen;
2181
2182 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2183
2184 while (b < bend && u < uend) {
2185 U8 c = *u++;
2186 if (!UTF8_IS_INVARIANT(c)) {
2187 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2188 if (u < uend) {
2189 U8 c1 = *u++;
2190 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2191 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2192 } else {
2b5e7bc2 2193 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2194 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2195 "%s %s%s",
2196 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2197 PL_op ? " in " : "",
2198 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2199 return -2;
2200 }
2201 } else {
2202 if (PL_op)
2203 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2204 "%s in %s", unees, OP_DESC(PL_op));
2205 else
61a12c31 2206 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2207 return -2; /* Really want to return undef :-) */
2208 }
2209 } else {
2210 return -2;
2211 }
2212 }
2213 if (*b != c) {
2214 return *b < c ? -2 : +2;
2215 }
2216 ++b;
2217 }
2218
2219 if (b == bend && u == uend)
2220 return 0;
2221
2222 return b < bend ? +1 : -1;
2223}
2224
2225/*
87cea99e 2226=for apidoc utf8_to_bytes
6940069f 2227
3bc0c78c 2228Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2229Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2230updates C<*lenp> to contain the new length.
3bc0c78c
KW
2231Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2232
2233Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2234having saved the value of C<*lenp> before the call, and subtracting the
2235after-call value of C<*lenp> from it.
6940069f 2236
a1433954 2237If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2238
6940069f
GS
2239=cut
2240*/
2241
2242U8 *
09af0336 2243Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2244{
9fe0d3c2 2245 U8 * first_variant;
246fae53 2246
7918f24d 2247 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2248 PERL_UNUSED_CONTEXT;
7918f24d 2249
9fe0d3c2 2250 /* This is a no-op if no variants at all in the input */
09af0336 2251 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2252 return s;
2253 }
2254
2255 {
3c5aa262 2256 U8 * const save = s;
09af0336 2257 U8 * const send = s + *lenp;
3c5aa262
KW
2258 U8 * d;
2259
2260 /* Nothing before the first variant needs to be changed, so start the real
2261 * work there */
2262 s = first_variant;
2263 while (s < send) {
2264 if (! UTF8_IS_INVARIANT(*s)) {
2265 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2266 *lenp = ((STRLEN) -1);
3c5aa262
KW
2267 return 0;
2268 }
2269 s++;
d59937ca
KW
2270 }
2271 s++;
dcad2880 2272 }
dcad2880 2273
3c5aa262
KW
2274 /* Is downgradable, so do it */
2275 d = s = first_variant;
2276 while (s < send) {
2277 U8 c = *s++;
2278 if (! UVCHR_IS_INVARIANT(c)) {
2279 /* Then it is two-byte encoded */
2280 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2281 s++;
2282 }
2283 *d++ = c;
2284 }
2285 *d = '\0';
09af0336 2286 *lenp = d - save;
3c5aa262
KW
2287
2288 return save;
9fe0d3c2 2289 }
6940069f
GS
2290}
2291
2292/*
87cea99e 2293=for apidoc bytes_from_utf8
f9a63242 2294
09af0336 2295Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2296byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2297actually encoded in UTF-8.
2298
2299Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2300the input string.
2301
41ae6089
KW
2302Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2303not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2304C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2305
41ae6089 2306Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2307newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2308returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2309
3bc0c78c 2310Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2311having saved the value of C<*lenp> before the call, and subtracting the
2312after-call value of C<*lenp> from it.
3bc0c78c 2313
37607a96 2314=cut
976c1b08
KW
2315
2316There is a macro that avoids this function call, but this is retained for
2317anyone who calls it with the Perl_ prefix */
f9a63242
JH
2318
2319U8 *
41ae6089 2320Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2321{
7918f24d 2322 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2323 PERL_UNUSED_CONTEXT;
f9a63242 2324
976c1b08
KW
2325 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2326}
2327
2328/*
2329No = here because currently externally undocumented
2330for apidoc bytes_from_utf8_loc
2331
2332Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2333to store the location of the first character in C<"s"> that cannot be
2334converted to non-UTF8.
2335
2336If that parameter is C<NULL>, this function behaves identically to
2337C<bytes_from_utf8>.
2338
2339Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2340C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2341
2342Otherwise, the function returns a newly created C<NUL>-terminated string
2343containing the non-UTF8 equivalent of the convertible first portion of
2344C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2345If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2346and C<*first_non_downgradable> is set to C<NULL>.
2347
2348Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2349first character in the original string that wasn't converted. C<*is_utf8p> is
2350unchanged. Note that the new string may have length 0.
2351
2352Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2353C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2354converts as many characters in it as possible stopping at the first one it
385b74be 2355finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2356set to point to that. The function returns the portion that could be converted
2357in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2358not including the terminating C<NUL>. If the very first character in the
2359original could not be converted, C<*lenp> will be 0, and the new string will
2360contain just a single C<NUL>. If the entire input string was converted,
2361C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2362
2363Upon successful return, the number of variants in the converted portion of the
2364string can be computed by having saved the value of C<*lenp> before the call,
2365and subtracting the after-call value of C<*lenp> from it.
2366
2367=cut
2368
2369
2370*/
2371
2372U8 *
2373Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2374{
2375 U8 *d;
2376 const U8 *original = s;
2377 U8 *converted_start;
2378 const U8 *send = s + *lenp;
f9a63242 2379
976c1b08 2380 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2381
976c1b08
KW
2382 if (! *is_utf8p) {
2383 if (first_unconverted) {
2384 *first_unconverted = NULL;
2385 }
2386
2387 return (U8 *) original;
2388 }
2389
2390 Newx(d, (*lenp) + 1, U8);
2391
2392 converted_start = d;
7299a045
KW
2393 while (s < send) {
2394 U8 c = *s++;
2395 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2396
2397 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2398 * have to stop now */
2399 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2400 if (first_unconverted) {
2401 *first_unconverted = s - 1;
2402 goto finish_and_return;
2403 }
2404 else {
2405 Safefree(converted_start);
2406 return (U8 *) original;
2407 }
2408 }
2409
7299a045
KW
2410 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2411 s++;
38af28cf 2412 }
7299a045
KW
2413 *d++ = c;
2414 }
170a1c22 2415
976c1b08
KW
2416 /* Here, converted the whole of the input */
2417 *is_utf8p = FALSE;
2418 if (first_unconverted) {
2419 *first_unconverted = NULL;
170a1c22 2420 }
976c1b08
KW
2421
2422 finish_and_return:
46a08a6f
KW
2423 *d = '\0';
2424 *lenp = d - converted_start;
976c1b08
KW
2425
2426 /* Trim unused space */
2427 Renew(converted_start, *lenp + 1, U8);
2428
2429 return converted_start;
f9a63242
JH
2430}
2431
2432/*
87cea99e 2433=for apidoc bytes_to_utf8
6940069f 2434
09af0336 2435Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2436UTF-8.
09af0336 2437Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2438reflect the new length in bytes.
6940069f 2439
3bc0c78c 2440Upon successful return, the number of variants in the string can be computed by
23b37b12 2441having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2442after-call value of C<*lenp>.
2443
75200dff 2444A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2445
2446If you want to convert to UTF-8 from encodings other than
2447the native (Latin1 or EBCDIC),
a1433954 2448see L</sv_recode_to_utf8>().
c9ada85f 2449
497711e7 2450=cut
6940069f
GS
2451*/
2452
2453U8*
09af0336 2454Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2455{
09af0336 2456 const U8 * const send = s + (*lenp);
6940069f
GS
2457 U8 *d;
2458 U8 *dst;
7918f24d
NC
2459
2460 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2461 PERL_UNUSED_CONTEXT;
6940069f 2462
09af0336 2463 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2464 dst = d;
2465
2466 while (s < send) {
55d09dc8
KW
2467 append_utf8_from_native_byte(*s, &d);
2468 s++;
6940069f 2469 }
2e11cf67 2470
6940069f 2471 *d = '\0';
09af0336 2472 *lenp = d-dst;
2e11cf67
KW
2473
2474 /* Trim unused space */
2475 Renew(dst, *lenp + 1, U8);
2476
6940069f
GS
2477 return dst;
2478}
2479
a0ed51b3 2480/*
624504c5
KW
2481 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2482 * use utf16_to_utf8_reversed().
a0ed51b3 2483 *
624504c5
KW
2484 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2485 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2486 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2487 *
2488 * These functions don't check for overflow. The worst case is every code
2489 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2490 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2491 * destination must be pre-extended to 2 times the source length.
2492 *
2493 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2494
2495U8*
dea0fc0b 2496Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2497{
dea0fc0b
JH
2498 U8* pend;
2499 U8* dstart = d;
2500
7918f24d
NC
2501 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2502
dea0fc0b 2503 if (bytelen & 1)
56576a04
KW
2504 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2505 (UV)bytelen);
dea0fc0b
JH
2506
2507 pend = p + bytelen;
2508
a0ed51b3 2509 while (p < pend) {
dea0fc0b
JH
2510 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2511 p += 2;
2d1545e5 2512 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2513 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2514 continue;
2515 }
56d37426
KW
2516 if (uv <= MAX_UTF8_TWO_BYTE) {
2517 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2518 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2519 continue;
2520 }
ffd0a9d3 2521
46956fad
KW
2522#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2523#define LAST_HIGH_SURROGATE 0xDBFF
2524#define FIRST_LOW_SURROGATE 0xDC00
2525#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2526#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2527
2528 /* This assumes that most uses will be in the first Unicode plane, not
2529 * needing surrogates */
2530 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2531 && uv <= UNICODE_SURROGATE_LAST))
2532 {
2533 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2534 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2535 }
2536 else {
01ea242b 2537 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2538 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2539 || UNLIKELY(low > LAST_LOW_SURROGATE))
2540 {
01ea242b 2541 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2542 }
2543 p += 2;
46956fad 2544 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2545 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2546 }
a0ed51b3 2547 }
56d37426
KW
2548#ifdef EBCDIC
2549 d = uvoffuni_to_utf8_flags(d, uv, 0);
2550#else
ffd0a9d3 2551 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2552 *d++ = (U8)(( uv >> 12) | 0xe0);
2553 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2554 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2555 continue;
2556 }
2557 else {
eb160463
GS
2558 *d++ = (U8)(( uv >> 18) | 0xf0);
2559 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2560 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2561 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2562 continue;
2563 }
56d37426 2564#endif
a0ed51b3 2565 }
dea0fc0b 2566 *newlen = d - dstart;
a0ed51b3
LW
2567 return d;
2568}
2569
2570/* Note: this one is slightly destructive of the source. */
2571
2572U8*
dea0fc0b 2573Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2574{
2575 U8* s = (U8*)p;
d4c19fe8 2576 U8* const send = s + bytelen;
7918f24d
NC
2577
2578 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2579
e0ea5e2d 2580 if (bytelen & 1)
147e3846 2581 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2582 (UV)bytelen);
2583
a0ed51b3 2584 while (s < send) {
d4c19fe8 2585 const U8 tmp = s[0];
a0ed51b3
LW
2586 s[0] = s[1];
2587 s[1] = tmp;
2588 s += 2;
2589 }
dea0fc0b 2590 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2591}
2592
922e8cb4
KW
2593bool
2594Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2595{
2596 U8 tmpbuf[UTF8_MAXBYTES+1];
2597 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2598 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2599}
2600
f9ae8fb6
JD
2601/* Internal function so we can deprecate the external one, and call
2602 this one from other deprecated functions in this file */
2603
f2645549
KW
2604bool
2605Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2606{
f2645549 2607 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2608
2609 if (*p == '_')
2610 return TRUE;
f25ce844 2611 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2612}
2613
5092f92a 2614bool
eba68aa0
KW
2615Perl__is_uni_perl_idcont(pTHX_ UV c)
2616{
2617 U8 tmpbuf[UTF8_MAXBYTES+1];
2618 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2619 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2620}
2621
2622bool
f91dcd13
KW
2623Perl__is_uni_perl_idstart(pTHX_ UV c)
2624{
2625 U8 tmpbuf[UTF8_MAXBYTES+1];
2626 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2627 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2628}
2629
3a4c58c9 2630UV
56576a04
KW
2631Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2632 const char S_or_s)
3a4c58c9
KW
2633{
2634 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2635 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2636 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2637 * either "SS" or "Ss". Which one to use is passed into the routine in
2638 * 'S_or_s' to avoid a test */
2639
2640 UV converted = toUPPER_LATIN1_MOD(c);
2641
2642 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2643
2644 assert(S_or_s == 'S' || S_or_s == 's');
2645
6f2d5cbc 2646 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2647 characters in this range */
3a4c58c9
KW
2648 *p = (U8) converted;
2649 *lenp = 1;
2650 return converted;
2651 }
2652
2653 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2654 * which it maps to one of them, so as to only have to have one check for
2655 * it in the main case */
2656 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2657 switch (c) {
2658 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2659 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2660 break;
2661 case MICRO_SIGN:
2662 converted = GREEK_CAPITAL_LETTER_MU;
2663 break;
79e064b9
KW
2664#if UNICODE_MAJOR_VERSION > 2 \
2665 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2666 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2667 case LATIN_SMALL_LETTER_SHARP_S:
2668 *(p)++ = 'S';
2669 *p = S_or_s;
2670 *lenp = 2;
2671 return 'S';
79e064b9 2672#endif
3a4c58c9 2673 default:
56576a04
KW
2674 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2675 " '%c' to map to '%c'",
2676 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2677 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2678 }
2679 }
2680
2681 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2682 *p = UTF8_TWO_BYTE_LO(converted);
2683 *lenp = 2;
2684
2685 return converted;
2686}
2687
50bda2c3
KW
2688/* Call the function to convert a UTF-8 encoded character to the specified case.
2689 * Note that there may be more than one character in the result.
2690 * INP is a pointer to the first byte of the input character
2691 * OUTP will be set to the first byte of the string of changed characters. It
2692 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2693 * LENP will be set to the length in bytes of the string of changed characters
2694 *
56576a04
KW
2695 * The functions return the ordinal of the first character in the string of
2696 * OUTP */
2697#define CALL_UPPER_CASE(uv, s, d, lenp) \
2698 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2699#define CALL_TITLE_CASE(uv, s, d, lenp) \
2700 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2701#define CALL_LOWER_CASE(uv, s, d, lenp) \
2702 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2703
b9992569
KW
2704/* This additionally has the input parameter 'specials', which if non-zero will
2705 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2706 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2707#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2708_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2709
84afefe6
JH
2710UV
2711Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2712{
a1433954
KW
2713 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2714 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2715 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2716 * the changed version may be longer than the original character.
2717 *
2718 * The ordinal of the first character of the changed version is returned
2719 * (but note, as explained above, that there may be more.) */
2720
7918f24d
NC
2721 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2722
3a4c58c9
KW
2723 if (c < 256) {
2724 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2725 }
2726
0ebc6274 2727 uvchr_to_utf8(p, c);
b9992569 2728 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2729}
2730
84afefe6
JH
2731UV
2732Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2733{
7918f24d
NC
2734 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2735
3a4c58c9
KW
2736 if (c < 256) {
2737 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2738 }
2739
0ebc6274 2740 uvchr_to_utf8(p, c);
b9992569 2741 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2742}
2743
afc16117 2744STATIC U8
eaf412bf 2745S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2746{
2747 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2748 * those, converting the result to UTF-8. Since the result is always just
a1433954 2749 * one character, we allow <p> to be NULL */
afc16117
KW
2750
2751 U8 converted = toLOWER_LATIN1(c);
2752
eaf412bf
KW
2753 PERL_UNUSED_ARG(dummy);
2754
afc16117 2755 if (p != NULL) {
6f2d5cbc 2756 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2757 *p = converted;
2758 *lenp = 1;
2759 }
2760 else {
430c9760
KW
2761 /* Result is known to always be < 256, so can use the EIGHT_BIT
2762 * macros */
2763 *p = UTF8_EIGHT_BIT_HI(converted);
2764 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2765 *lenp = 2;
2766 }
2767 }
2768 return converted;
2769}
2770
84afefe6
JH
2771UV
2772Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2773{
7918f24d
NC
2774 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2775
afc16117 2776 if (c < 256) {
eaf412bf 2777 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2778 }
2779
afc16117 2780 uvchr_to_utf8(p, c);
b9992569 2781 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2782}
2783
84afefe6 2784UV
56576a04
KW
2785Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2786 const unsigned int flags)
a1dde8de 2787{
51910141 2788 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2789 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2790 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2791 *
2792 * Not to be used for locale folds
51910141 2793 */
f673fad4 2794
a1dde8de
KW
2795 UV converted;
2796
2797 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2798 PERL_UNUSED_CONTEXT;
a1dde8de 2799
1ca267a5
KW
2800 assert (! (flags & FOLD_FLAGS_LOCALE));
2801
659a7c2d 2802 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2803 converted = GREEK_SMALL_LETTER_MU;
2804 }
9b63e895
KW
2805#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2806 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2807 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2808 else if ( (flags & FOLD_FLAGS_FULL)
2809 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2810 {
1ca267a5
KW
2811 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2812 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2813 * under those circumstances. */
2814 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2815 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2816 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2817 p, *lenp, U8);
2818 return LATIN_SMALL_LETTER_LONG_S;
2819 }
2820 else {
4f489194
KW
2821 *(p)++ = 's';
2822 *p = 's';
2823 *lenp = 2;
2824 return 's';
1ca267a5 2825 }
a1dde8de 2826 }
9b63e895 2827#endif
a1dde8de
KW
2828 else { /* In this range the fold of all other characters is their lower
2829 case */
2830 converted = toLOWER_LATIN1(c);
2831 }
2832
6f2d5cbc 2833 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2834 *p = (U8) converted;
2835 *lenp = 1;
2836 }
2837 else {
2838 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2839 *p = UTF8_TWO_BYTE_LO(converted);
2840 *lenp = 2;
2841 }
2842
2843 return converted;
2844}
2845
2846UV
31f05a37 2847Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2848{
4b593389 2849
a0270393
KW
2850 /* Not currently externally documented, and subject to change
2851 * <flags> bits meanings:
2852 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2853 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2854 * locale are to be used.
a0270393
KW
2855 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2856 */
4b593389 2857
36bb2ab6 2858 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2859
780fcc9f
KW
2860 if (flags & FOLD_FLAGS_LOCALE) {
2861 /* Treat a UTF-8 locale as not being in locale at all */
2862 if (IN_UTF8_CTYPE_LOCALE) {
2863 flags &= ~FOLD_FLAGS_LOCALE;
2864 }
2865 else {
2866 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2867 goto needs_full_generality;
780fcc9f 2868 }
31f05a37
KW
2869 }
2870
a1dde8de 2871 if (c < 256) {
e7b7ac46 2872 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2873 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2874 }
2875
2f306ab9 2876 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2877 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2878 uvchr_to_utf8(p, c);
b9992569 2879 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2880 }
567b353c 2881 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2882 the special flags. */
2883 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2884
2885 needs_full_generality:
a0270393 2886 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2887 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2888 p, lenp, flags);
a0270393 2889 }
84afefe6
JH
2890}
2891
26483009 2892PERL_STATIC_INLINE bool
5141f98e 2893S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2894 const char *const swashname, SV* const invlist)
bde6a22d 2895{
ea317ccb
KW
2896 /* returns a boolean giving whether or not the UTF8-encoded character that
2897 * starts at <p> is in the swash indicated by <swashname>. <swash>
2898 * contains a pointer to where the swash indicated by <swashname>
2899 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2900 * look at <*swash> and only generate a swash if it is not null. <invlist>
2901 * is NULL or an inversion list that defines the swash. If not null, it
2902 * saves time during initialization of the swash.
ea317ccb
KW
2903 *
2904 * Note that it is assumed that the buffer length of <p> is enough to
2905 * contain all the bytes that comprise the character. Thus, <*p> should
2906 * have been checked before this call for mal-formedness enough to assure
2907 * that. */
2908
7918f24d
NC
2909 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2910
492a624f 2911 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2912 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2913 * as far as there being enough bytes available in it to accommodate the
2914 * character without reading beyond the end, and pass that number on to the
2915 * validating routine */
6302f837 2916 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2917 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2918 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2919 1 /* Die */ );
2920 NOT_REACHED; /* NOTREACHED */
28123549 2921 }
86ae6e94 2922
87367d5f
KW
2923 if (!*swash) {
2924 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2925 *swash = _core_swash_init("utf8",
2926
2927 /* Only use the name if there is no inversion
2928 * list; otherwise will go out to disk */
2929 (invlist) ? "" : swashname,
2930
2931 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2932 }
28123549 2933
bde6a22d
NC
2934 return swash_fetch(*swash, p, TRUE) != 0;
2935}
2936
da8c1a98 2937PERL_STATIC_INLINE bool
56576a04
KW
2938S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2939 SV **swash, const char *const swashname,
2940 SV* const invlist)
da8c1a98
KW
2941{
2942 /* returns a boolean giving whether or not the UTF8-encoded character that
2943 * starts at <p>, and extending no further than <e - 1> is in the swash
2944 * indicated by <swashname>. <swash> contains a pointer to where the swash
2945 * indicated by <swashname> is to be stored; which this routine will do, so
2946 * that future calls will look at <*swash> and only generate a swash if it
2947 * is not null. <invlist> is NULL or an inversion list that defines the
2948 * swash. If not null, it saves time during initialization of the swash.
2949 */
2950
2951 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2952
2953 if (! isUTF8_CHAR(p, e)) {
2954 _force_out_malformed_utf8_message(p, e, 0, 1);
2955 NOT_REACHED; /* NOTREACHED */
2956 }
2957
2958 if (!*swash) {
2959 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2960 *swash = _core_swash_init("utf8",
2961
2962 /* Only use the name if there is no inversion
2963 * list; otherwise will go out to disk */
2964 (invlist) ? "" : swashname,
2965
2966 &PL_sv_undef, 1, 0, invlist, &flags);
2967 }
2968
2969 return swash_fetch(*swash, p, TRUE) != 0;
2970}
2971
34aeb2e9
KW
2972STATIC void
2973S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2974 const char * const alternative,
2975 const bool use_locale,
2976 const char * const file,
2977 const unsigned line)
2978{
2979 const char * key;
2980
2981 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2982
2983 if (ckWARN_d(WARN_DEPRECATED)) {
2984
2985 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2986 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2987 if (! PL_seen_deprecated_macro) {
2988 PL_seen_deprecated_macro = newHV();
2989 }
2990 if (! hv_store(PL_seen_deprecated_macro, key,
2991 strlen(key), &PL_sv_undef, 0))
2992 {
2993 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2994 }
2995
c44e9413 2996 if (instr(file, "mathoms.c")) {
607313a1
KW
2997 Perl_warner(aTHX_ WARN_DEPRECATED,
2998 "In %s, line %d, starting in Perl v5.30, %s()"
2999 " will be removed. Avoid this message by"
3000 " converting to use %s().\n",
3001 file, line, name, alternative);
3002 }
3003 else {
34aeb2e9
KW
3004 Perl_warner(aTHX_ WARN_DEPRECATED,
3005 "In %s, line %d, starting in Perl v5.30, %s() will"
3006 " require an additional parameter. Avoid this"
3007 " message by converting to use %s().\n",
3008 file, line, name, alternative);
607313a1 3009 }
34aeb2e9
KW
3010 }
3011 }
3012}
3013
bde6a22d 3014bool
34aeb2e9 3015Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3016 const U8 * const p,
34aeb2e9
KW
3017 const char * const name,
3018 const char * const alternative,
3019 const bool use_utf8,
3020 const bool use_locale,
3021 const char * const file,
3022 const unsigned line)
922e8cb4 3023{
922e8cb4
KW
3024 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3025
34aeb2e9
KW
3026 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3027
3028 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3029
3030 switch (classnum) {
3031 case _CC_WORDCHAR:
3032 case _CC_DIGIT:
3033 case _CC_ALPHA:
3034 case _CC_LOWER:
3035 case _CC_UPPER:
3036 case _CC_PUNCT:
3037 case _CC_PRINT:
3038 case _CC_ALPHANUMERIC:
3039 case _CC_GRAPH:
3040 case _CC_CASED:
3041
3042 return is_utf8_common(p,
3043 &PL_utf8_swash_ptrs[classnum],
3044 swash_property_names[classnum],
3045 PL_XPosix_ptrs[classnum]);
3046
3047 case _CC_SPACE:
3048 return is_XPERLSPACE_high(p);
3049 case _CC_BLANK:
3050 return is_HORIZWS_high(p);
3051 case _CC_XDIGIT:
3052 return is_XDIGIT_high(p);
3053 case _CC_CNTRL:
3054 return 0;
3055 case _CC_ASCII:
3056 return 0;
3057 case _CC_VERTSPACE:
3058 return is_VERTWS_high(p);
3059 case _CC_IDFIRST:
3060 if (! PL_utf8_perl_idstart) {
22f0498f
KW
3061 PL_utf8_perl_idstart
3062 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 3063 }
22f0498f
KW
3064 return is_utf8_common(p, &PL_utf8_perl_idstart,
3065 "_Perl_IDStart", NULL);
34aeb2e9
KW
3066 case _CC_IDCONT:
3067 if (! PL_utf8_perl_idcont) {
22f0498f
KW
3068 PL_utf8_perl_idcont
3069 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 3070 }
22f0498f
KW
3071 return is_utf8_common(p, &PL_utf8_perl_idcont,
3072 "_Perl_IDCont", NULL);
34aeb2e9
KW
3073 }
3074 }
3075
3076 /* idcont is the same as wordchar below 256 */
3077 if (classnum == _CC_IDCONT) {
3078 classnum = _CC_WORDCHAR;
3079 }
3080 else if (classnum == _CC_IDFIRST) {
3081 if (*p == '_') {
3082 return TRUE;
3083 }
3084 classnum = _CC_ALPHA;
3085 }
3086
3087 if (! use_locale) {
3088 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3089 return _generic_isCC(*p, classnum);
3090 }
922e8cb4 3091
34aeb2e9
KW
3092 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3093 }
3094 else {
3095 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3096 return isFOO_lc(classnum, *p);
3097 }
3098
3099 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3100 }
3101
3102 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3103}
3104
3105bool
da8c1a98
KW
3106Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3107 const U8 * const e)
3108{
3109 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3110
3111 assert(classnum < _FIRST_NON_SWASH_CC);
3112
3113 return is_utf8_common_with_len(p,
3114 e,
3115 &PL_utf8_swash_ptrs[classnum],
3116 swash_property_names[classnum],
3117 PL_XPosix_ptrs[classnum]);
3118}
3119
3120bool
da8c1a98
KW
3121Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3122{
3123 SV* invlist = NULL;
3124
3125 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3126
3127 if (! PL_utf8_perl_idstart) {
3128 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
3129 }
3130 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
3131 "_Perl_IDStart", invlist);
3132}
3133
3134bool
f2645549 3135Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3136{
f2645549 3137 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3138
3139 if (*p == '_')
3140 return TRUE;
f25ce844 3141 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3142}
3143
3144bool
da8c1a98
KW
3145Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3146{
3147 SV* invlist = NULL;
3148
3149 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3150
3151 if (! PL_utf8_perl_idcont) {
3152 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3153 }
3154 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3155 "_Perl_IDCont", invlist);
3156}
3157
3158bool
f2645549 3159Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3160{
f2645549 3161 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3162
f25ce844 3163 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3164}
3165
3166bool
f2645549 3167Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3168{
f2645549 3169 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3170
f25ce844 3171 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
3172}
3173
3174bool
7dbf68d2
KW
3175Perl__is_utf8_mark(pTHX_ const U8 *p)
3176{
7dbf68d2
KW
3177 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3178
f25ce844 3179 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3180}
3181
b9992569 3182 /* change namve uv1 to 'from' */
6a4a25f4 3183STATIC UV
b9992569
KW
3184S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
3185 SV **swashp, const char *normal, const char *special)
3186{
0134edef 3187 STRLEN len = 0;
7918f24d 3188
b9992569 3189 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3190
36eaa811
KW
3191 /* For code points that don't change case, we already know that the output
3192 * of this function is the unchanged input, so we can skip doing look-ups
3193 * for them. Unfortunately the case-changing code points are scattered
3194 * around. But there are some long consecutive ranges where there are no
3195 * case changing code points. By adding tests, we can eliminate the lookup
3196 * for all the ones in such ranges. This is currently done here only for
3197 * just a few cases where the scripts are in common use in modern commerce
3198 * (and scripts adjacent to those which can be included without additional
3199 * tests). */
3200
3201 if (uv1 >= 0x0590) {
3202 /* This keeps from needing further processing the code points most
3203 * likely to be used in the following non-cased scripts: Hebrew,
3204 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3205 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3206 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3207 if (uv1 < 0x10A0) {
3208 goto cases_to_self;
3209 }
3210
3211 /* The following largish code point ranges also don't have case
3212 * changes, but khw didn't think they warranted extra tests to speed
3213 * them up (which would slightly slow down everything else above them):
3214 * 1100..139F Hangul Jamo, Ethiopic
3215 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3216 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3217 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3218 * Combining Diacritical Marks Extended, Balinese,
3219 * Sundanese, Batak, Lepcha, Ol Chiki
3220 * 2000..206F General Punctuation
3221 */
3222
3223 if (uv1 >= 0x2D30) {
3224
3225 /* This keeps the from needing further processing the code points
3226 * most likely to be used in the following non-cased major scripts:
3227 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3228 *
3229 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3230 * event that Unicode eventually allocates the unused block as of
3231 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3232 * that the test suite will start having failures to alert you
3233 * should that happen) */
3234 if (uv1 < 0xA640) {
3235 goto cases_to_self;
3236 }
3237
3238 if (uv1 >= 0xAC00) {
3239 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3240 if (ckWARN_d(WARN_SURROGATE)) {
3241 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3242 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3243 "Operation \"%s\" returns its argument for"
3244 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3245 }
3246 goto cases_to_self;
3247 }
36eaa811
KW
3248
3249 /* AC00..FAFF Catches Hangul syllables and private use, plus
3250 * some others */
3251 if (uv1 < 0xFB00) {
3252 goto cases_to_self;
3253
3254 }
3255
5af9bc97 3256 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3257 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3258 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3259 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3260 }
3261 if (ckWARN_d(WARN_NON_UNICODE)) {
3262 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3263 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3264 "Operation \"%s\" returns its argument for"
3265 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3266 }
3267 goto cases_to_self;
3268 }
3bfc1e70
KW
3269#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3270 if (UNLIKELY(uv1
3271 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3272 {
3273
56576a04
KW
3274 /* As of Unicode 10.0, this means we avoid swash creation
3275 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3276 goto cases_to_self;
3277 }
3278#endif
36eaa811
KW
3279 }
3280 }
9ae3ac1a 3281
36eaa811
KW
3282 /* Note that non-characters are perfectly legal, so no warning should
3283 * be given. There are so few of them, that it isn't worth the extra
3284 * tests to avoid swash creation */
9ae3ac1a
KW
3285 }
3286
0134edef 3287 if (!*swashp) /* load on-demand */
56576a04
KW
3288 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3289 4, 0, NULL, NULL);
0134edef 3290
a6f87d8c 3291 if (special) {
0134edef 3292 /* It might be "special" (sometimes, but not always,
2a37f04d 3293 * a multicharacter mapping) */
4a8240a3 3294 HV *hv = NULL;
b08cf34e
JH
3295 SV **svp;
3296
4a8240a3
KW
3297 /* If passed in the specials name, use that; otherwise use any
3298 * given in the swash */
3299 if (*special != '\0') {
3300 hv = get_hv(special, 0);
3301 }
3302 else {
3303 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3304 if (svp) {
3305 hv = MUTABLE_HV(SvRV(*svp));
3306 }
3307 }
3308
176fe009 3309 if (hv
5f560d8a 3310 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3311 && (*svp))
3312 {
cfd0369c 3313 const char *s;
47654450 3314
cfd0369c 3315 s = SvPV_const(*svp, len);
47654450 3316 if (len == 1)
f4cd282c 3317 /* EIGHTBIT */
c80e42f3 3318 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3319 else {
d2dcd0fb 3320 Copy(s, ustrp, len, U8);
29e98929 3321 }
983ffd37 3322 }
0134edef
JH
3323 }
3324
3325 if (!len && *swashp) {
4a4088c4 3326 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3327
0134edef
JH
3328 if (uv2) {
3329 /* It was "normal" (a single character mapping). */
f4cd282c 3330 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3331 }
3332 }
1feea2c7 3333
cbe07460
KW
3334 if (len) {
3335 if (lenp) {
3336 *lenp = len;
3337 }
3338 return valid_utf8_to_uvchr(ustrp, 0);
3339 }
3340
3341 /* Here, there was no mapping defined, which means that the code point maps
3342 * to itself. Return the inputs */
e24dfe9c 3343 cases_to_self:
bfdf22ec 3344 len = UTF8SKIP(p);
ca9fab46
KW
3345 if (p != ustrp) { /* Don't copy onto itself */
3346 Copy(p, ustrp, len, U8);
3347 }
0134edef 3348
2a37f04d
JH
3349 if (lenp)
3350 *lenp = len;
3351
f4cd282c 3352 return uv1;
cbe07460 3353
a0ed51b3
LW
3354}
3355
051a06d4 3356STATIC UV
56576a04
KW
3357S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3358 U8* const ustrp, STRLEN *lenp)
051a06d4 3359{
4a4088c4 3360 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3361 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3362 * result contains a character that crosses the 255/256 boundary, disallow
3363 * the change, and return the original code point. See L<perlfunc/lc> for
3364 * why;
051a06d4 3365 *
a1433954
KW
3366 * p points to the original string whose case was changed; assumed
3367 * by this routine to be well-formed
051a06d4 3368 * result the code point of the first character in the changed-case string
56576a04
KW
3369 * ustrp points to the changed-case string (<result> represents its
3370 * first char)
051a06d4
KW
3371 * lenp points to the length of <ustrp> */
3372
3373 UV original; /* To store the first code point of <p> */
3374
3375 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3376
a4f12ed7 3377 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3378
3379 /* We know immediately if the first character in the string crosses the
3380 * boundary, so can skip */
3381 if (result > 255) {
3382
3383 /* Look at every character in the result; if any cross the
3384 * boundary, the whole thing is disallowed */
3385 U8* s = ustrp + UTF8SKIP(ustrp);
3386 U8* e = ustrp + *lenp;
3387 while (s < e) {
a4f12ed7 3388 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3389 goto bad_crossing;
3390 }
3391 s += UTF8SKIP(s);
3392 }
3393
613abc6d
KW
3394 /* Here, no characters crossed, result is ok as-is, but we warn. */
3395 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3396 return result;
3397 }
3398
7b52d656 3399 bad_crossing:
051a06d4
KW
3400
3401 /* Failed, have to return the original */
4b88fb76 3402 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3403
3404 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3405 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3406 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3407 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3408 OP_DESC(PL_op),
ab0b796c
KW
3409 original,
3410 original);
051a06d4
KW
3411 Copy(p, ustrp, *lenp, char);
3412 return original;
3413}
3414
607313a1
KW
3415STATIC U32
3416S_check_and_deprecate(pTHX_ const U8 *p,
3417 const U8 **e,
3418 const unsigned int type, /* See below */
3419 const bool use_locale, /* Is this a 'LC_'
3420 macro call? */
3421 const char * const file,
3422 const unsigned line)
3423{
3424 /* This is a temporary function to deprecate the unsafe calls to the case
3425 * changing macros and functions. It keeps all the special stuff in just
3426 * one place.
3427 *
3428 * It updates *e with the pointer to the end of the input string. If using
3429 * the old-style macros, *e is NULL on input, and so this function assumes
3430 * the input string is long enough to hold the entire UTF-8 sequence, and
3431 * sets *e accordingly, but it then returns a flag to pass the
3432 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3433 * using the full length if possible.
3434 *
3435 * It also does the assert that *e > p when *e is not NULL. This should be
3436 * migrated to the callers when this function gets deleted.
3437 *
3438 * The 'type' parameter is used for the caller to specify which case
3439 * changing function this is called from: */
3440
3441# define DEPRECATE_TO_UPPER 0
3442# define DEPRECATE_TO_TITLE 1
3443# define DEPRECATE_TO_LOWER 2
3444# define DEPRECATE_TO_FOLD 3
3445
3446 U32 utf8n_flags = 0;
3447 const char * name;
3448 const char * alternative;
3449
3450 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3451
3452 if (*e == NULL) {
3453 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3454 *e = p + UTF8SKIP(p);
3455
3456 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3457 * there. It could be part of a larger path */
607313a1 3458 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3459 name = instr(file, "mathoms.c")
607313a1
KW
3460 ? "to_utf8_upper"
3461 : "toUPPER_utf8";
3462 alternative = "toUPPER_utf8_safe";
3463 }
3464 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3465 name = instr(file, "mathoms.c")
607313a1
KW
3466 ? "to_utf8_title"
3467 : "toTITLE_utf8";
3468 alternative = "toTITLE_utf8_safe";
3469 }
3470 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3471 name = instr(file, "mathoms.c")
607313a1
KW
3472 ? "to_utf8_lower"
3473 : "toLOWER_utf8";
3474 alternative = "toLOWER_utf8_safe";
3475 }
3476 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3477 name = instr(file, "mathoms.c")
607313a1
KW
3478 ? "to_utf8_fold"
3479 : "toFOLD_utf8";
3480 alternative = "toFOLD_utf8_safe";
3481 }
3482 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3483
3484 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3485 }
3486 else {
3487 assert (p < *e);
3488 }
3489
3490 return utf8n_flags;
3491}
3492
eaf412bf
KW
3493/* The process for changing the case is essentially the same for the four case
3494 * change types, except there are complications for folding. Otherwise the
3495 * difference is only which case to change to. To make sure that they all do
3496 * the same thing, the bodies of the functions are extracted out into the
3497 * following two macros. The functions are written with the same variable
3498 * names, and these are known and used inside these macros. It would be
3499 * better, of course, to have inline functions to do it, but since different
3500 * macros are called, depending on which case is being changed to, this is not
3501 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3502 * function can start with the common start macro, then finish with its special
3503 * handling; while the other three cases can just use the common end macro.
3504 *
3505 * The algorithm is to use the proper (passed in) macro or function to change
3506 * the case for code points that are below 256. The macro is used if using
3507 * locale rules for the case change; the function if not. If the code point is
3508 * above 255, it is computed from the input UTF-8, and another macro is called
3509 * to do the conversion. If necessary, the output is converted to UTF-8. If
3510 * using a locale, we have to check that the change did not cross the 255/256
3511 * boundary, see check_locale_boundary_crossing() for further details.
3512 *
3513 * The macros are split with the correct case change for the below-256 case
3514 * stored into 'result', and in the middle of an else clause for the above-255
3515 * case. At that point in the 'else', 'result' is not the final result, but is
3516 * the input code point calculated from the UTF-8. The fold code needs to
3517 * realize all this and take it from there.
3518 *
3519 * If you read the two macros as sequential, it's easier to understand what's
3520 * going on. */
3521#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3522 L1_func_extra_param) \
a239b1e2 3523 \
eaf412bf
KW
3524 if (flags & (locale_flags)) { \
3525 /* Treat a UTF-8 locale as not being in locale at all */ \
3526 if (IN_UTF8_CTYPE_LOCALE) { \
3527 flags &= ~(locale_flags); \
3528 } \
3529 else { \
3530 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3531 } \
3532 } \
3533 \
3534 if (UTF8_IS_INVARIANT(*p)) { \
3535 if (flags & (locale_flags)) { \
3536 result = LC_L1_change_macro(*p); \
3537 } \
3538 else { \
3539 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3540 } \
3541 } \
a239b1e2 3542 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3543 if (flags & (locale_flags)) { \
3544 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3545 *(p+1))); \
3546 } \
3547 else { \
3548 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3549 ustrp, lenp, L1_func_extra_param); \
3550 } \
3551 } \
fa8ab374
KW
3552 else { /* malformed UTF-8 or ord above 255 */ \
3553 STRLEN len_result; \
fa8ab374
KW
3554 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3555 if (len_result == (STRLEN) -1) { \
607313a1
KW
3556 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3557 1 /* Die */ ); \
fa8ab374 3558 }
eaf412bf
KW
3559
3560#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3561 result = change_macro(result, p, ustrp, lenp); \
3562 \
3563 if (flags & (locale_flags)) { \
3564 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3565 } \
3566 return result; \
3567 } \
3568 \
3569 /* Here, used locale rules. Convert back to UTF-8 */ \
3570 if (UTF8_IS_INVARIANT(result)) { \
3571 *ustrp = (U8) result; \
3572 *lenp = 1; \
3573 } \
3574 else { \
3575 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3576 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3577 *lenp = 2; \
3578 } \
3579 \
3580 return result;
3581
d3e79532 3582/*
87cea99e 3583=for apidoc to_utf8_upper
d3e79532 3584
a239b1e2 3585Instead use L</toUPPER_utf8_safe>.
a1433954 3586
d3e79532
JH
3587=cut */
3588
051a06d4 3589/* Not currently externally documented, and subject to change:
31f05a37
KW
3590 * <flags> is set iff iff the rules from the current underlying locale are to
3591 * be used. */
051a06d4 3592
2104c8d9 3593UV
607313a1
KW
3594Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3595 const U8 *e,
3596 U8* ustrp,
3597 STRLEN *lenp,
3598 bool flags,
3599 const char * const file,
3600 const int line)
a0ed51b3 3601{
051a06d4 3602 UV result;
607313a1
KW
3603 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3604 cBOOL(flags), file, line);
051a06d4
KW
3605
3606 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3607
eaf412bf
KW
3608 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3609 /* 2nd char of uc(U+DF) is 'S' */
3610 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3611 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3612}
a0ed51b3 3613
d3e79532 3614/*
87cea99e 3615=for apidoc to_utf8_title
d3e79532 3616
a239b1e2 3617Instead use L</toTITLE_utf8_safe>.
a1433954 3618
d3e79532
JH
3619=cut */
3620
051a06d4 3621/* Not currently externally documented, and subject to change:
31f05a37
KW
3622 * <flags> is set iff the rules from the current underlying locale are to be
3623 * used. Since titlecase is not defined in POSIX, for other than a
3624 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3625 */
051a06d4 3626
983ffd37 3627UV
607313a1
KW
3628Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3629 const U8 *e,
3630 U8* ustrp,
3631 STRLEN *lenp,
3632 bool flags,
3633 const char * const file,
3634 const int line)
983ffd37 3635{
051a06d4 3636 UV result;
607313a1
KW
3637 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3638 cBOOL(flags), file, line);
051a06d4
KW
3639
3640 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3641
eaf412bf
KW
3642 /* 2nd char of ucfirst(U+DF) is 's' */
3643 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3644 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3645}
3646
d3e79532 3647/*
87cea99e 3648=for apidoc to_utf8_lower
d3e79532 3649
a239b1e2 3650Instead use L</toLOWER_utf8_safe>.
a1433954 3651
d3e79532
JH
3652=cut */
3653
051a06d4 3654/* Not currently externally documented, and subject to change:
31f05a37
KW
3655 * <flags> is set iff iff the rules from the current underlying locale are to
3656 * be used.
3657 */
051a06d4 3658
2104c8d9 3659UV
607313a1
KW
3660Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3661 const U8 *e,
3662 U8* ustrp,
3663 STRLEN *lenp,
3664 bool flags,
3665 const char * const file,
3666 const int line)
a0ed51b3 3667{
051a06d4 3668 UV result;
607313a1
KW
3669 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3670 cBOOL(flags), file, line);
051a06d4 3671
051a06d4 3672 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3673
eaf412bf
KW
3674 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3675 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3676}
3677
d3e79532 3678/*
87cea99e 3679=for apidoc to_utf8_fold
d3e79532 3680
a239b1e2 3681Instead use L</toFOLD_utf8_safe>.
a1433954 3682
d3e79532
JH
3683=cut */
3684
051a06d4
KW
3685/* Not currently externally documented, and subject to change,
3686 * in <flags>
31f05a37
KW
3687 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3688 * locale are to be used.
051a06d4
KW
3689 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3690 * otherwise simple folds
a0270393
KW
3691 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3692 * prohibited
445bf929 3693 */
36bb2ab6 3694
b4e400f9 3695UV
607313a1
KW
3696Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3697 const U8 *e,
3698 U8* ustrp,
3699 STRLEN *lenp,
3700 U8 flags,
3701 const char * const file,
3702 const int line)
b4e400f9 3703{
051a06d4