This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Move and slightly change comment block
[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[] =
76513bdc 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf ". This will be fatal in Perl 5.28";
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
c15a80f3 81 SAVEI8(PL_dowarn);
9cbfb8ab
KW
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
c94c2f39
KW
119/* All these formats take a single UV code point argument */
120const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
121const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
122 " is not recommended for open interchange";
123const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
124 " may not be portable";
57ff5f59
KW
125const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
126 " Unicode, requires a Perl extension," \
127 " and so is not portable";
c94c2f39 128
8ee1cdcb
KW
129#define HANDLE_UNICODE_SURROGATE(uv, flags) \
130 STMT_START { \
131 if (flags & UNICODE_WARN_SURROGATE) { \
132 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
c94c2f39 133 surrogate_cp_format, uv); \
8ee1cdcb
KW
134 } \
135 if (flags & UNICODE_DISALLOW_SURROGATE) { \
136 return NULL; \
137 } \
138 } STMT_END;
139
140#define HANDLE_UNICODE_NONCHAR(uv, flags) \
141 STMT_START { \
142 if (flags & UNICODE_WARN_NONCHAR) { \
143 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
c94c2f39 144 nonchar_cp_format, uv); \
8ee1cdcb
KW
145 } \
146 if (flags & UNICODE_DISALLOW_NONCHAR) { \
147 return NULL; \
148 } \
149 } STMT_END;
150
ba6ed43c
KW
151/* Use shorter names internally in this file */
152#define SHIFT UTF_ACCUMULATION_SHIFT
153#undef MARK
154#define MARK UTF_CONTINUATION_MARK
155#define MASK UTF_CONTINUATION_MASK
156
dfe13c55 157U8 *
4b31b634 158Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
a0ed51b3 159{
378516de 160 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 161
2d1545e5 162 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 163 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
164 return d;
165 }
facc1dc2 166
3ea68d71 167 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
168 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
169 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
170 return d;
171 }
d9432125 172
ba6ed43c
KW
173 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
174 * below, the 16 is for start bytes E0-EF (which are all the possible ones
175 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
176 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
177 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
178 * 0x800-0xFFFF on ASCII */
179 if (uv < (16 * (1U << (2 * SHIFT)))) {
180 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
181 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
182 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
183
184#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
185 aren't tested here */
186 /* The most likely code points in this range are below the surrogates.
187 * Do an extra test to quickly exclude those. */
188 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
189 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
190 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
191 {
8ee1cdcb
KW
192 HANDLE_UNICODE_NONCHAR(uv, flags);
193 }
194 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
195 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 196 }
ba6ed43c
KW
197 }
198#endif
199 return d;
200 }
201
202 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
203 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
204 * happen starting with 4-byte characters on ASCII platforms. We unify the
205 * code for these with EBCDIC, even though some of them require 5-bytes on
206 * those, because khw believes the code saving is worth the very slight
207 * performance hit on these high EBCDIC code points. */
208
209 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
76513bdc
KW
210 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
211 && ckWARN_d(WARN_DEPRECATED))
212 {
213 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
214 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
a5bf80e0 215 }
0a8a1a5b
KW
216 if ( (flags & UNICODE_WARN_SUPER)
217 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
218 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
219 {
220 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
221
222 /* Choose the more dire applicable warning */
d044b7a7 223 (UNICODE_IS_PERL_EXTENDED(uv))
57ff5f59 224 ? perl_extended_cp_format
c94c2f39 225 : super_cp_format,
a5bf80e0
KW
226 uv);
227 }
56576a04 228 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
229 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
230 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
231 {
232 return NULL;
233 }
234 }
ba6ed43c
KW
235 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
236 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 237 }
d9432125 238
ba6ed43c
KW
239 /* Test for and handle 4-byte result. In the test immediately below, the
240 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
241 * characters). The 3 is for 3 continuation bytes; these each contribute
242 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
243 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
244 * 0x1_0000-0x1F_FFFF on ASCII */
245 if (uv < (8 * (1U << (3 * SHIFT)))) {
246 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
247 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
248 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
249 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
250
251#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
252 characters. The end-plane non-characters for EBCDIC were
253 handled just above */
254 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
255 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 256 }
ba6ed43c
KW
257 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
258 HANDLE_UNICODE_SURROGATE(uv, flags);
259 }
260#endif
261
262 return d;
263 }
264
265 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
266 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
267 * format. The unrolled version above turns out to not save all that much
268 * time, and at these high code points (well above the legal Unicode range
269 * on ASCII platforms, and well above anything in common use in EBCDIC),
270 * khw believes that less code outweighs slight performance gains. */
271
d9432125 272 {
5aaebcb3 273 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
274 U8 *p = d+len-1;
275 while (p > d) {
4c8cd605 276 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
277 uv >>= UTF_ACCUMULATION_SHIFT;
278 }
4c8cd605 279 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
280 return d+len;
281 }
a0ed51b3 282}
a5bf80e0 283
646ca15d 284/*
07693fe6
KW
285=for apidoc uvchr_to_utf8
286
bcb1a2d4 287Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 288of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
289C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
290the byte after the end of the new character. In other words,
07693fe6
KW
291
292 d = uvchr_to_utf8(d, uv);
293
294is the recommended wide native character-aware way of saying
295
296 *(d++) = uv;
297
760c7c2f
KW
298This function accepts any UV as input, but very high code points (above
299C<IV_MAX> on the platform) will raise a deprecation warning. This is
300typically 0x7FFF_FFFF in a 32-bit word.
301
302It is possible to forbid or warn on non-Unicode code points, or those that may
303be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 304
07693fe6
KW
305=cut
306*/
307
de69f3af
KW
308/* This is also a macro */
309PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
310
07693fe6
KW
311U8 *
312Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
313{
de69f3af 314 return uvchr_to_utf8(d, uv);
07693fe6
KW
315}
316
de69f3af
KW
317/*
318=for apidoc uvchr_to_utf8_flags
319
320Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 321of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
322C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
323the byte after the end of the new character. In other words,
de69f3af
KW
324
325 d = uvchr_to_utf8_flags(d, uv, flags);
326
327or, in most cases,
328
329 d = uvchr_to_utf8_flags(d, uv, 0);
330
331This is the Unicode-aware way of saying
332
333 *(d++) = uv;
334
760c7c2f
KW
335If C<flags> is 0, this function accepts any UV as input, but very high code
336points (above C<IV_MAX> for the platform) will raise a deprecation warning.
337This is typically 0x7FFF_FFFF in a 32-bit word.
338
339Specifying C<flags> can further restrict what is allowed and not warned on, as
340follows:
de69f3af 341
796b6530 342If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
343the function will raise a warning, provided UTF8 warnings are enabled. If
344instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
345NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 346
760c7c2f
KW
347Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
348affect how the function handles a Unicode non-character.
93e6dbd6 349
760c7c2f
KW
350And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
351affect the handling of code points that are above the Unicode maximum of
3520x10FFFF. Languages other than Perl may not be able to accept files that
353contain these.
93e6dbd6
KW
354
355The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
356the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
357three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
358allowed inputs to the strict UTF-8 traditionally defined by Unicode.
359Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
360C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
361above-Unicode and surrogate flags, but not the non-character ones, as
362defined in
363L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
364See L<perlunicode/Noncharacter code points>.
93e6dbd6 365
57ff5f59
KW
366Extremely high code points were never specified in any standard, and require an
367extension to UTF-8 to express, which Perl does. It is likely that programs
368written in something other than Perl would not be able to read files that
369contain these; nor would Perl understand files written by something that uses a
370different extension. For these reasons, there is a separate set of flags that
371can warn and/or disallow these extremely high code points, even if other
372above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
373and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
374L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
375treat all above-Unicode code points, including these, as malformations. (Note
376that the Unicode standard considers anything above 0x10FFFF to be illegal, but
377there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
378
379A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
380retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
381C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
382C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because these
383flags can apply to code points that actually do fit in 31 bits. This happens
384on EBCDIC platforms, and sometimes when the L<overlong
385malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
386describe the situation in all cases.
de69f3af 387
de69f3af
KW
388=cut
389*/
390
391/* This is also a macro */
392PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
393
07693fe6
KW
394U8 *
395Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
396{
de69f3af 397 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
398}
399
57ff5f59
KW
400#ifndef UV_IS_QUAD
401
83dc0f42
KW
402PERL_STATIC_INLINE bool
403S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
404{
405 /* Returns TRUE if the first code point represented by the Perl-extended-
406 * UTF-8-encoded string starting at 's', and looking no further than 'e -
407 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
408 *
409 * The function handles the case where the input bytes do not include all
410 * the ones necessary to represent a full character. That is, they may be
411 * the intial bytes of the representation of a code point, but possibly
412 * the final ones necessary for the complete representation may be beyond
413 * 'e - 1'.
414 *
415 * The function assumes that the sequence is well-formed UTF-8 as far as it
416 * goes, and is for a UTF-8 variant code point. If the sequence is
417 * incomplete, the function returns FALSE if there is any well-formed
418 * UTF-8 byte sequence that can complete it in such a way that a code point
419 * < 2**31 is produced; otherwise it returns TRUE.
420 *
421 * Getting this exactly right is slightly tricky, and has to be done in
422 * several places in this file, so is centralized here. It is based on the
423 * following table:
424 *
83dc0f42
KW
425 */
426
427#ifdef EBCDIC
428
37086697
KW
429 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
430 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
431 const STRLEN prefix_len = sizeof(prefix) - 1;
432 const STRLEN len = e - s;
f880f78a 433 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
434
435#else
436
437 PERL_UNUSED_ARG(e);
438
439#endif
440
441 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
442
443 assert(! UTF8_IS_INVARIANT(*s));
444
445#ifndef EBCDIC
446
447 /* Technically, a start byte of FE can be for a code point that fits into
448 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
449 * malformation. */
450 return (*s >= 0xFE);
451
452#else
453
56576a04
KW
454 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
455 * mean a 32-bit or larger code point (0xFF is an invariant). For 0xFE, we
456 * need at least 2 bytes, and maybe up through 8 bytes, to be sure that the
457 * value is above 31 bits. */
83dc0f42
KW
458 if (*s != 0xFE || len == 1) {
459 return FALSE;
460 }
461
5f995336
KW
462 /* U+7FFFFFFF (2 ** 31 - 1)
463 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
464 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
465 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
466 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
467 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
468 * U+80000000 (2 ** 31):
469 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
470 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
471 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
472 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
473 */
474 /* (Note that in UTF-EBCDIC, the two lowest possible continuation bytes
475 * are \x41 and \x42.) If we have enough bytes available to determine
476 * the answer, or the bytes we do have differ from the UTF-8 prefix of
477 * the highest 30-bit code point, we can compare them to get a
478 * definitive answer */
479
83dc0f42
KW
480 return cBOOL(memGT(s + 1, prefix, cmp_len));
481
482#endif
483
484}
485
57ff5f59
KW
486#endif
487
d6be65ae 488PERL_STATIC_INLINE int
12a4bed3
KW
489S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
490{
d6be65ae
KW
491 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
492 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
493 * it isn't, and -1 if there isn't enough information to tell. This last
494 * return value can happen if the sequence is incomplete, missing some
495 * trailing bytes that would form a complete character. If there are
496 * enough bytes to make a definitive decision, this function does so.
497 * Usually 2 bytes sufficient.
498 *
499 * Overlongs can occur whenever the number of continuation bytes changes.
500 * That means whenever the number of leading 1 bits in a start byte
501 * increases from the next lower start byte. That happens for start bytes
502 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
503 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
504 * ASCII platforms: C0, C1
505 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 506 */
12a4bed3
KW
507
508 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
509 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
510
511 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
512 assert(len > 1 && UTF8_IS_START(*s));
513
514 /* Each platform has overlongs after the start bytes given above (expressed
515 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
516 * the logic is the same, except the E0 overlong has already been excluded
517 * on EBCDIC platforms. The values below were found by manually
518 * inspecting the UTF-8 patterns. See the tables in utf8.h and
519 * utfebcdic.h. */
520
521# ifdef EBCDIC
522# define F0_ABOVE_OVERLONG 0xB0
523# define F8_ABOVE_OVERLONG 0xA8
524# define FC_ABOVE_OVERLONG 0xA4
525# define FE_ABOVE_OVERLONG 0xA2
526# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
527 /* I8(0xfe) is FF */
528# else
529
530 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 531 return 1;
12a4bed3
KW
532 }
533
534# define F0_ABOVE_OVERLONG 0x90
535# define F8_ABOVE_OVERLONG 0x88
536# define FC_ABOVE_OVERLONG 0x84
537# define FE_ABOVE_OVERLONG 0x82
538# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
539# endif
540
541
542 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
543 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
544 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
545 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
546 {
d6be65ae 547 return 1;
12a4bed3
KW
548 }
549
b0b342d4 550 /* Check for the FF overlong */
d6be65ae 551 return isFF_OVERLONG(s, len);
b0b342d4
KW
552}
553
8d6204cc 554PERL_STATIC_INLINE int
b0b342d4
KW
555S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
556{
8d6204cc
KW
557 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
558 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
559 * it isn't, and -1 if there isn't enough information to tell. This last
560 * return value can happen if the sequence is incomplete, missing some
561 * trailing bytes that would form a complete character. If there are
562 * enough bytes to make a definitive decision, this function does so. */
563
b0b342d4 564 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 565
8d6204cc
KW
566 /* To be an FF overlong, all the available bytes must match */
567 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
568 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
569 {
570 return 0;
571 }
572
573 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
574 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 575 * utfebcdic.h. */
8d6204cc
KW
576 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
577 return 1;
578 }
12a4bed3 579
8d6204cc
KW
580 /* The missing bytes could cause the result to go one way or the other, so
581 * the result is indeterminate */
582 return -1;
12a4bed3
KW
583}
584
a77c906e
KW
585/* Anything larger than this will overflow the word if it were converted into a UV */
586#if defined(UV_IS_QUAD)
587# ifdef EBCDIC /* Actually is I8 */
588# define HIGHEST_REPRESENTABLE_UTF8 \
589 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
590# else
591# define HIGHEST_REPRESENTABLE_UTF8 \
592 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
593# endif
594#else /* 32-bit */
595# ifdef EBCDIC
596# define HIGHEST_REPRESENTABLE_UTF8 \
597 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
598# else
599# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
600# endif
601#endif
602
c285bbc4 603PERL_STATIC_INLINE int
a77c906e
KW
604S_does_utf8_overflow(const U8 * const s, const U8 * e)
605{
c285bbc4
KW
606 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
607 * 'e' - 1 would overflow a UV on this platform; that is if it represents a
608 * code point larger than the highest representable code point. It returns
609 * 1 if it does overflow; 0 if it doesn't, and -1 if there isn't enough
610 * information to tell. This last return value can happen if the sequence
611 * is incomplete, missing some trailing bytes that would form a complete
612 * character. If there are enough bytes to make a definitive decision,
613 * this function does so.
614 *
615 * (For ASCII platforms, we could use memcmp() because we don't have to
616 * convert each byte to I8, but it's very rare input indeed that would
617 * approach overflow, so the loop below will likely only get executed once.)
618 *
619 * 'e' - 1 must not be beyond a full character. */
a77c906e
KW
620
621 const STRLEN len = e - s;
c285bbc4
KW
622 const U8 *x;
623 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
a77c906e
KW
624
625 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
626 assert(s <= e && s + UTF8SKIP(s) >= e);
627
628#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
629
630 /* On 32 bit ASCII machines, many overlongs that start with FF don't
631 * overflow */
a77c906e 632 if (isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
633
634 /* To be such an overlong, the first bytes of 's' must match
635 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
636 * don't have any additional bytes available, the sequence, when
637 * completed might or might not fit in 32 bits. But if we have that
638 * next byte, we can tell for sure. If it is <= 0x83, then it does
639 * fit. */
640 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
641 return -1;
642 }
643
644 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
645 }
646
647#endif
648
649 for (x = s; x < e; x++, y++) {
650
651 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
652 continue;
653 }
654
655 /* If this byte is larger than the corresponding highest UTF-8 byte,
656 * the sequence overflow; otherwise the byte is less than, and so the
657 * sequence doesn't overflow */
658 return NATIVE_UTF8_TO_I8(*x) > *y;
659
660 }
661
662 /* Got to the end and all bytes are the same. If the input is a whole
663 * character, it doesn't overflow. And if it is a partial character,
c285bbc4
KW
664 * there's not enough information to tell */
665 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
666 return -1;
667 }
668
669 return 0;
a77c906e
KW
670}
671
12a4bed3
KW
672#undef F0_ABOVE_OVERLONG
673#undef F8_ABOVE_OVERLONG
674#undef FC_ABOVE_OVERLONG
675#undef FE_ABOVE_OVERLONG
676#undef FF_OVERLONG_PREFIX
677
35f8c9bd 678STRLEN
edc2c47a 679Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 680{
2b479609 681 STRLEN len;
12a4bed3 682 const U8 *x;
35f8c9bd 683
2b479609
KW
684 /* A helper function that should not be called directly.
685 *
686 * This function returns non-zero if the string beginning at 's' and
687 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
688 * code point; otherwise it returns 0. The examination stops after the
689 * first code point in 's' is validated, not looking at the rest of the
690 * input. If 'e' is such that there are not enough bytes to represent a
691 * complete code point, this function will return non-zero anyway, if the
692 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
693 * excluded by 'flags'.
694 *
695 * A non-zero return gives the number of bytes required to represent the
696 * code point. Be aware that if the input is for a partial character, the
697 * return will be larger than 'e - s'.
698 *
699 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
700 * The caller should have excluded the possibility of it being invariant
701 * before calling this function.
2b479609
KW
702 *
703 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
704 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
705 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
706 * disallowed by the flags. If the input is only for a partial character,
707 * the function will return non-zero if there is any sequence of
708 * well-formed UTF-8 that, when appended to the input sequence, could
709 * result in an allowed code point; otherwise it returns 0. Non characters
710 * cannot be determined based on partial character input. But many of the
711 * other excluded types can be determined with just the first one or two
712 * bytes.
713 *
714 */
715
716 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
717
718 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 719 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 720 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 721
2b479609 722 /* A variant char must begin with a start byte */
35f8c9bd
KW
723 if (UNLIKELY(! UTF8_IS_START(*s))) {
724 return 0;
725 }
726
edc2c47a
KW
727 /* Examine a maximum of a single whole code point */
728 if (e - s > UTF8SKIP(s)) {
729 e = s + UTF8SKIP(s);
730 }
731
2b479609
KW
732 len = e - s;
733
734 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
735 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 736
56576a04
KW
737 /* Here, we are disallowing some set of largish code points, and the
738 * first byte indicates the sequence is for a code point that could be
739 * in the excluded set. We generally don't have to look beyond this or
740 * the second byte to see if the sequence is actually for one of the
741 * excluded classes. The code below is derived from this table:
742 *
2b479609
KW
743 * UTF-8 UTF-EBCDIC I8
744 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
745 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
746 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
747 *
56576a04
KW
748 * Keep in mind that legal continuation bytes range between \x80..\xBF
749 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
750 * continuation bytes. Hence, we don't have to test the upper edge
751 * because if any of those is encountered, the sequence is malformed,
752 * and would fail elsewhere in this function.
753 *
754 * The code here likewise assumes that there aren't other
755 * malformations; again the function should fail elsewhere because of
756 * these. For example, an overlong beginning with FC doesn't actually
757 * have to be a super; it could actually represent a small code point,
758 * even U+0000. But, since overlongs (and other malformations) are
759 * illegal, the function should return FALSE in either case.
2b479609
KW
760 */
761
762#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
763# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 764# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 765
19794540
KW
766# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
767 /* B6 and B7 */ \
768 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 769# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
770#else
771# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
772# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
773# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 774# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
775#endif
776
777 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
778 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
779 {
2b479609
KW
780 return 0; /* Above Unicode */
781 }
782
d044b7a7 783 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 784 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 785 {
57ff5f59 786 return 0;
2b479609
KW
787 }
788
789 if (len > 1) {
790 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
791
792 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 793 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
794 {
795 return 0; /* Above Unicode */
796 }
797
798 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 799 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
800 {
801 return 0; /* Surrogate */
802 }
803
804 if ( (flags & UTF8_DISALLOW_NONCHAR)
805 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
806 {
807 return 0; /* Noncharacter code point */
808 }
809 }
810 }
811
812 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
813 for (x = s + 1; x < e; x++) {
814 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
815 return 0;
816 }
817 }
818
af13dd8a 819 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 820 * overlong. */
d6be65ae 821 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 822 return 0;
af13dd8a
KW
823 }
824
12a4bed3
KW
825 /* And finally, that the code point represented fits in a word on this
826 * platform */
c285bbc4 827 if (does_utf8_overflow(s, e) > 0) {
12a4bed3 828 return 0;
35f8c9bd
KW
829 }
830
2b479609 831 return UTF8SKIP(s);
35f8c9bd
KW
832}
833
7e2f38b2
KW
834char *
835Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
836{
837 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
838 * bytes starting at 's'. 'format' gives how to display each byte.
839 * Currently, there are only two formats, so it is currently a bool:
840 * 0 \xab
841 * 1 ab (that is a space between two hex digit bytes)
842 */
7cf8d05d
KW
843
844 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
845 trailing NUL */
846 const U8 * const e = s + len;
847 char * output;
848 char * d;
849
850 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
851
852 Newx(output, output_len, char);
853 SAVEFREEPV(output);
854
855 d = output;
856 for (; s < e; s++) {
857 const unsigned high_nibble = (*s & 0xF0) >> 4;
858 const unsigned low_nibble = (*s & 0x0F);
859
7e2f38b2
KW
860 if (format) {
861 *d++ = ' ';
862 }
863 else {
864 *d++ = '\\';
865 *d++ = 'x';
866 }
7cf8d05d
KW
867
868 if (high_nibble < 10) {
869 *d++ = high_nibble + '0';
870 }
871 else {
872 *d++ = high_nibble - 10 + 'a';
873 }
874
875 if (low_nibble < 10) {
876 *d++ = low_nibble + '0';
877 }
878 else {
879 *d++ = low_nibble - 10 + 'a';
880 }
881 }
882
883 *d = '\0';
884 return output;
885}
886
806547a7 887PERL_STATIC_INLINE char *
7cf8d05d
KW
888S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
889
890 /* How many bytes to print */
3cc6a05e 891 STRLEN print_len,
7cf8d05d
KW
892
893 /* Which one is the non-continuation */
894 const STRLEN non_cont_byte_pos,
895
896 /* How many bytes should there be? */
897 const STRLEN expect_len)
806547a7
KW
898{
899 /* Return the malformation warning text for an unexpected continuation
900 * byte. */
901
7cf8d05d 902 const char * const where = (non_cont_byte_pos == 1)
806547a7 903 ? "immediately"
7cf8d05d
KW
904 : Perl_form(aTHX_ "%d bytes",
905 (int) non_cont_byte_pos);
806547a7
KW
906
907 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
908
7cf8d05d
KW
909 /* We don't need to pass this parameter, but since it has already been
910 * calculated, it's likely faster to pass it; verify under DEBUGGING */
911 assert(expect_len == UTF8SKIP(s));
912
913 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
914 " %s after start byte 0x%02x; need %d bytes, got %d)",
915 malformed_text,
7e2f38b2 916 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
917 *(s + non_cont_byte_pos),
918 where,
919 *s,
920 (int) expect_len,
921 (int) non_cont_byte_pos);
806547a7
KW
922}
923
35f8c9bd
KW
924/*
925
de69f3af 926=for apidoc utf8n_to_uvchr
378516de
KW
927
928THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 929Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 930
9041c2e3 931Bottom level UTF-8 decode routine.
de69f3af 932Returns the native code point value of the first character in the string C<s>,
746afd53
KW
933which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
934C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
935the length, in bytes, of that character.
949cf498
KW
936
937The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
938well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
939causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
940is the next possible position in C<s> that could begin a non-malformed
941character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
942is raised. Some UTF-8 input sequences may contain multiple malformations.
943This function tries to find every possible one in each call, so multiple
56576a04 944warnings can be raised for the same sequence.
949cf498
KW
945
946Various ALLOW flags can be set in C<flags> to allow (and not warn on)
947individual types of malformations, such as the sequence being overlong (that
948is, when there is a shorter sequence that can express the same code point;
949overlong sequences are expressly forbidden in the UTF-8 standard due to
950potential security issues). Another malformation example is the first byte of
951a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
952flags. Even if allowed, this function generally returns the Unicode
953REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
954F<utf8.h> to override this behavior for the overlong malformations, but don't
955do that except for very specialized purposes.
949cf498 956
796b6530 957The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
958flags) malformation is found. If this flag is set, the routine assumes that
959the caller will raise a warning, and this function will silently just set
d088425d
KW
960C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
961
75200dff 962Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 963character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
964in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
965be set to 1. To disambiguate, upon a zero return, see if the first byte of
966C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 967error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
968
969Certain code points are considered problematic. These are Unicode surrogates,
746afd53 970Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 971By default these are considered regular code points, but certain situations
ecc1615f
KW
972warrant special handling for them, which can be specified using the C<flags>
973parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
974three classes are treated as malformations and handled as such. The flags
975C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
976C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
977disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
978restricts the allowed inputs to the strict UTF-8 traditionally defined by
979Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
980definition given by
981L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
982The difference between traditional strictness and C9 strictness is that the
983latter does not forbid non-character code points. (They are still discouraged,
984however.) For more discussion see L<perlunicode/Noncharacter code points>.
985
986The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
987C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
988C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
989raised for their respective categories, but otherwise the code points are
990considered valid (not malformations). To get a category to both be treated as
991a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 992(But note that warnings are not raised if lexically disabled nor if
796b6530 993C<UTF8_CHECK_ONLY> is also specified.)
949cf498 994
57ff5f59
KW
995Extremely high code points were never specified in any standard, and require an
996extension to UTF-8 to express, which Perl does. It is likely that programs
997written in something other than Perl would not be able to read files that
998contain these; nor would Perl understand files written by something that uses a
999different extension. For these reasons, there is a separate set of flags that
1000can warn and/or disallow these extremely high code points, even if other
1001above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1002C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1003L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1004above-Unicode code points, including these, as malformations.
1005(Note that the Unicode standard considers anything above 0x10FFFF to be
1006illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1007(2**31 -1))
1008
1009A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1010retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1011C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1012C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1013can apply to code points that actually do fit in 31 bits. This happens on
1014EBCDIC platforms, and sometimes when the L<overlong
1015malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1016describe the situation in all cases.
1017
760c7c2f
KW
1018It is now deprecated to have very high code points (above C<IV_MAX> on the
1019platforms) and this function will raise a deprecation warning for these (unless
d5944cab 1020such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 1021in a 32-bit word.
ab8e6d41 1022
949cf498
KW
1023All other code points corresponding to Unicode characters, including private
1024use and those yet to be assigned, are never considered malformed and never
1025warn.
67e989fb 1026
37607a96 1027=cut
f9380377
KW
1028
1029Also implemented as a macro in utf8.h
1030*/
1031
1032UV
1033Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1034 STRLEN curlen,
1035 STRLEN *retlen,
1036 const U32 flags)
1037{
1038 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1039
1040 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1041}
1042
1043/*
1044
1045=for apidoc utf8n_to_uvchr_error
1046
1047THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1048Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1049
1050This function is for code that needs to know what the precise malformation(s)
1051are when an error is found.
1052
1053It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1054all the others, C<errors>. If this parameter is 0, this function behaves
1055identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1056to a C<U32> variable, which this function sets to indicate any errors found.
1057Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1058C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1059of these bits will be set if a malformation is found, even if the input
7a65503b 1060C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1061exceptions are noted:
1062
1063=over 4
1064
57ff5f59 1065=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1066
57ff5f59
KW
1067The input sequence is not standard UTF-8, but a Perl extension. This bit is
1068set only if the input C<flags> parameter contains either the
1069C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1070
1071Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1072and so some extension must be used to express them. Perl uses a natural
1073extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1074extension to represent even higher ones, so that any code point that fits in a
107564-bit word can be represented. Text using these extensions is not likely to
1076be portable to non-Perl code. We lump both of these extensions together and
1077refer to them as Perl extended UTF-8. There exist other extensions that people
1078have invented, incompatible with Perl's.
1079
1080On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1081extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1082than on ASCII. Prior to that, code points 2**31 and higher were simply
1083unrepresentable, and a different, incompatible method was used to represent
1084code points between 2**30 and 2**31 - 1.
1085
1086On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1087Perl extended UTF-8 is used.
1088
1089In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1090may use for backward compatibility. That name is misleading, as this flag may
1091be set when the code point actually does fit in 31 bits. This happens on
1092EBCDIC platforms, and sometimes when the L<overlong
1093malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1094describes the situation in all cases.
f9380377
KW
1095
1096=item C<UTF8_GOT_CONTINUATION>
1097
1098The input sequence was malformed in that the first byte was a a UTF-8
1099continuation byte.
1100
1101=item C<UTF8_GOT_EMPTY>
1102
1103The input C<curlen> parameter was 0.
1104
1105=item C<UTF8_GOT_LONG>
1106
1107The input sequence was malformed in that there is some other sequence that
1108evaluates to the same code point, but that sequence is shorter than this one.
1109
fecaf136
KW
1110Until Unicode 3.1, it was legal for programs to accept this malformation, but
1111it was discovered that this created security issues.
1112
f9380377
KW
1113=item C<UTF8_GOT_NONCHAR>
1114
1115The code point represented by the input UTF-8 sequence is for a Unicode
1116non-character code point.
1117This bit is set only if the input C<flags> parameter contains either the
1118C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1119
1120=item C<UTF8_GOT_NON_CONTINUATION>
1121
1122The input sequence was malformed in that a non-continuation type byte was found
1123in a position where only a continuation type one should be.
1124
1125=item C<UTF8_GOT_OVERFLOW>
1126
1127The input sequence was malformed in that it is for a code point that is not
1128representable in the number of bits available in a UV on the current platform.
1129
1130=item C<UTF8_GOT_SHORT>
1131
1132The input sequence was malformed in that C<curlen> is smaller than required for
1133a complete sequence. In other words, the input is for a partial character
1134sequence.
1135
1136=item C<UTF8_GOT_SUPER>
1137
1138The input sequence was malformed in that it is for a non-Unicode code point;
1139that is, one above the legal Unicode maximum.
1140This bit is set only if the input C<flags> parameter contains either the
1141C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1142
1143=item C<UTF8_GOT_SURROGATE>
1144
1145The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1146code point.
1147This bit is set only if the input C<flags> parameter contains either the
1148C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1149
1150=back
1151
133551d8
KW
1152To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1153flag to suppress any warnings, and then examine the C<*errors> return.
1154
f9380377 1155=cut
37607a96 1156*/
67e989fb 1157
a0ed51b3 1158UV
f9380377
KW
1159Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1160 STRLEN curlen,
1161 STRLEN *retlen,
1162 const U32 flags,
1163 U32 * errors)
a0ed51b3 1164{
d4c19fe8 1165 const U8 * const s0 = s;
2b5e7bc2
KW
1166 U8 * send = NULL; /* (initialized to silence compilers' wrong
1167 warning) */
1168 U32 possible_problems = 0; /* A bit is set here for each potential problem
1169 found as we go along */
eb83ed87 1170 UV uv = *s;
2b5e7bc2
KW
1171 STRLEN expectlen = 0; /* How long should this sequence be?
1172 (initialized to silence compilers' wrong
1173 warning) */
e308b348 1174 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1175 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1176 this gets set and discarded */
a0dbb045 1177
2b5e7bc2
KW
1178 /* The below are used only if there is both an overlong malformation and a
1179 * too short one. Otherwise the first two are set to 's0' and 'send', and
1180 * the third not used at all */
1181 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1182 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1183 routine; see [perl #130921] */
2b5e7bc2 1184 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1185
f9380377
KW
1186 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1187
1188 if (errors) {
1189 *errors = 0;
1190 }
1191 else {
1192 errors = &discard_errors;
1193 }
a0dbb045 1194
eb83ed87
KW
1195 /* The order of malformation tests here is important. We should consume as
1196 * few bytes as possible in order to not skip any valid character. This is
1197 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1198 * http://unicode.org/reports/tr36 for more discussion as to why. For
1199 * example, once we've done a UTF8SKIP, we can tell the expected number of
1200 * bytes, and could fail right off the bat if the input parameters indicate
1201 * that there are too few available. But it could be that just that first
1202 * byte is garbled, and the intended character occupies fewer bytes. If we
1203 * blindly assumed that the first byte is correct, and skipped based on
1204 * that number, we could skip over a valid input character. So instead, we
1205 * always examine the sequence byte-by-byte.
1206 *
1207 * We also should not consume too few bytes, otherwise someone could inject
1208 * things. For example, an input could be deliberately designed to
1209 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1210 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1211 * which is actually part of of the overflowing sequence), that could look
1212 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1213 * sequence and process the rest, inappropriately.
1214 *
1215 * Some possible input sequences are malformed in more than one way. This
1216 * function goes to lengths to try to find all of them. This is necessary
1217 * for correctness, as the inputs may allow one malformation but not
1218 * another, and if we abandon searching for others after finding the
1219 * allowed one, we could allow in something that shouldn't have been.
1220 */
eb83ed87 1221
b5b9af04 1222 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1223 possible_problems |= UTF8_GOT_EMPTY;
1224 curlen = 0;
5a48568d 1225 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1226 goto ready_to_handle_errors;
0c443dc2
JH
1227 }
1228
eb83ed87
KW
1229 expectlen = UTF8SKIP(s);
1230
1231 /* A well-formed UTF-8 character, as the vast majority of calls to this
1232 * function will be for, has this expected length. For efficiency, set
1233 * things up here to return it. It will be overriden only in those rare
1234 * cases where a malformation is found */
1235 if (retlen) {
1236 *retlen = expectlen;
1237 }
1238
1239 /* An invariant is trivially well-formed */
1d72bdf6 1240 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1241 return uv;
a0ed51b3 1242 }
67e989fb 1243
eb83ed87 1244 /* A continuation character can't start a valid sequence */
b5b9af04 1245 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1246 possible_problems |= UTF8_GOT_CONTINUATION;
1247 curlen = 1;
1248 uv = UNICODE_REPLACEMENT;
1249 goto ready_to_handle_errors;
ba210ebe 1250 }
9041c2e3 1251
dcd27b3c 1252 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1253 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1254 * because it excludes start bytes like \xC0 that always lead to
1255 * overlongs.) */
dcd27b3c 1256
534752c1
KW
1257 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1258 * that indicate the number of bytes in the character's whole UTF-8
1259 * sequence, leaving just the bits that are part of the value. */
1260 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1261
e308b348
KW
1262 /* Setup the loop end point, making sure to not look past the end of the
1263 * input string, and flag it as too short if the size isn't big enough. */
1264 send = (U8*) s0;
1265 if (UNLIKELY(curlen < expectlen)) {
1266 possible_problems |= UTF8_GOT_SHORT;
1267 avail_len = curlen;
1268 send += curlen;
1269 }
1270 else {
1271 send += expectlen;
1272 }
e308b348 1273
eb83ed87 1274 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1275 * accumulating each into the working value as we go. */
eb83ed87 1276 for (s = s0 + 1; s < send; s++) {
b5b9af04 1277 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1278 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1279 continue;
1280 }
1281
1282 /* Here, found a non-continuation before processing all expected bytes.
1283 * This byte indicates the beginning of a new character, so quit, even
1284 * if allowing this malformation. */
2b5e7bc2 1285 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1286 break;
eb83ed87
KW
1287 } /* End of loop through the character's bytes */
1288
1289 /* Save how many bytes were actually in the character */
1290 curlen = s - s0;
1291
2b5e7bc2
KW
1292 /* Note that there are two types of too-short malformation. One is when
1293 * there is actual wrong data before the normal termination of the
1294 * sequence. The other is that the sequence wasn't complete before the end
1295 * of the data we are allowed to look at, based on the input 'curlen'.
1296 * This means that we were passed data for a partial character, but it is
1297 * valid as far as we saw. The other is definitely invalid. This
1298 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1299 * separate.
1300 *
1301 * A convenience macro that matches either of the too-short conditions. */
1302# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1303
1304 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1305 uv_so_far = uv;
1306 uv = UNICODE_REPLACEMENT;
1307 }
2b5e7bc2 1308
08e73697
KW
1309 /* Check for overflow. The algorithm requires us to not look past the end
1310 * of the current character, even if partial, so the upper limit is 's' */
c285bbc4 1311 if (UNLIKELY(does_utf8_overflow(s0, s) > 0)) {
2b5e7bc2
KW
1312 possible_problems |= UTF8_GOT_OVERFLOW;
1313 uv = UNICODE_REPLACEMENT;
eb83ed87 1314 }
eb83ed87 1315
2b5e7bc2
KW
1316 /* Check for overlong. If no problems so far, 'uv' is the correct code
1317 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1318 * we must look at the UTF-8 byte sequence itself to see if it is for an
1319 * overlong */
1320 if ( ( LIKELY(! possible_problems)
1321 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1322 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1323 && ( UNLIKELY(! UTF8_IS_START(*s0))
1324 || ( curlen > 1
d6be65ae 1325 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1326 s - s0))))))
2f8f112e 1327 {
2b5e7bc2
KW
1328 possible_problems |= UTF8_GOT_LONG;
1329
abc28b54 1330 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1331
abc28b54
KW
1332 /* The calculation in the 'true' branch of this 'if'
1333 * below won't work if overflows, and isn't needed
1334 * anyway. Further below we handle all overflow
1335 * cases */
1336 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1337 {
2b5e7bc2
KW
1338 UV min_uv = uv_so_far;
1339 STRLEN i;
1340
1341 /* Here, the input is both overlong and is missing some trailing
1342 * bytes. There is no single code point it could be for, but there
1343 * may be enough information present to determine if what we have
1344 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1345 * The code further below has the intelligence to determine this,
1346 * but just for non-overlong UTF-8 sequences. What we do here is
1347 * calculate the smallest code point the input could represent if
1348 * there were no too short malformation. Then we compute and save
1349 * the UTF-8 for that, which is what the code below looks at
1350 * instead of the raw input. It turns out that the smallest such
1351 * code point is all we need. */
2b5e7bc2
KW
1352 for (i = curlen; i < expectlen; i++) {
1353 min_uv = UTF8_ACCUMULATE(min_uv,
1354 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1355 }
1356
e9f2c446 1357 adjusted_s0 = temp_char_buf;
57ff5f59 1358 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1359 }
eb83ed87
KW
1360 }
1361
56576a04
KW
1362 /* Here, we have found all the possible problems, except for when the input
1363 * is for a problematic code point not allowed by the input parameters. */
1364
06188866
KW
1365 /* uv is valid for overlongs */
1366 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1367
1368 /* isn't problematic if < this */
1369 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1370 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1371
1372 /* if overflow, we know without looking further
1373 * precisely which of the problematic types it is,
1374 * and we deal with those in the overflow handling
1375 * code */
1376 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1377 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1378 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1379 && ((flags & ( UTF8_DISALLOW_NONCHAR
1380 |UTF8_DISALLOW_SURROGATE
1381 |UTF8_DISALLOW_SUPER
d044b7a7 1382 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1383 |UTF8_WARN_NONCHAR
1384 |UTF8_WARN_SURROGATE
1385 |UTF8_WARN_SUPER
d044b7a7 1386 |UTF8_WARN_PERL_EXTENDED))
2b5e7bc2
KW
1387 /* In case of a malformation, 'uv' is not valid, and has
1388 * been changed to something in the Unicode range.
1389 * Currently we don't output a deprecation message if there
1390 * is already a malformation, so we don't have to special
1391 * case the test immediately below */
760c7c2f
KW
1392 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1393 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1394 {
2b5e7bc2
KW
1395 /* If there were no malformations, or the only malformation is an
1396 * overlong, 'uv' is valid */
1397 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1398 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1399 possible_problems |= UTF8_GOT_SURROGATE;
1400 }
1401 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1402 possible_problems |= UTF8_GOT_SUPER;
1403 }
1404 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1405 possible_problems |= UTF8_GOT_NONCHAR;
1406 }
1407 }
1408 else { /* Otherwise, need to look at the source UTF-8, possibly
1409 adjusted to be non-overlong */
1410
1411 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1412 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1413 {
2b5e7bc2
KW
1414 possible_problems |= UTF8_GOT_SUPER;
1415 }
1416 else if (curlen > 1) {
1417 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1418 NATIVE_UTF8_TO_I8(*adjusted_s0),
1419 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1420 {
2b5e7bc2 1421 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1422 }
2b5e7bc2
KW
1423 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1424 NATIVE_UTF8_TO_I8(*adjusted_s0),
1425 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1426 {
1427 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1428 }
1429 }
c0236afe 1430
2b5e7bc2
KW
1431 /* We need a complete well-formed UTF-8 character to discern
1432 * non-characters, so can't look for them here */
1433 }
1434 }
949cf498 1435
2b5e7bc2
KW
1436 ready_to_handle_errors:
1437
1438 /* At this point:
1439 * curlen contains the number of bytes in the sequence that
1440 * this call should advance the input by.
e308b348
KW
1441 * avail_len gives the available number of bytes passed in, but
1442 * only if this is less than the expected number of
1443 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1444 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1445 * is set in it for each potential problem found.
1446 * uv contains the code point the input sequence
1447 * represents; or if there is a problem that prevents
1448 * a well-defined value from being computed, it is
1449 * some subsitute value, typically the REPLACEMENT
1450 * CHARACTER.
1451 * s0 points to the first byte of the character
56576a04
KW
1452 * s points to just after were we left off processing
1453 * the character
1454 * send points to just after where that character should
1455 * end, based on how many bytes the start byte tells
1456 * us should be in it, but no further than s0 +
1457 * avail_len
2b5e7bc2 1458 */
eb83ed87 1459
2b5e7bc2
KW
1460 if (UNLIKELY(possible_problems)) {
1461 bool disallowed = FALSE;
1462 const U32 orig_problems = possible_problems;
1463
1464 while (possible_problems) { /* Handle each possible problem */
1465 UV pack_warn = 0;
1466 char * message = NULL;
1467
1468 /* Each 'if' clause handles one problem. They are ordered so that
1469 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1470 * ones; this is kinda in decreasing severity order. But the
1471 * overlong must come last, as it changes 'uv' looked at by the
1472 * others */
2b5e7bc2
KW
1473 if (possible_problems & UTF8_GOT_OVERFLOW) {
1474
56576a04
KW
1475 /* Overflow means also got a super and are using Perl's
1476 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1477 possible_problems
d044b7a7 1478 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1479 *errors |= UTF8_GOT_OVERFLOW;
1480
1481 /* But the API says we flag all errors found */
1482 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1483 *errors |= UTF8_GOT_SUPER;
1484 }
ddb65933 1485 if (flags
d044b7a7 1486 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1487 {
d044b7a7 1488 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1489 }
2b5e7bc2 1490
d60baaa7 1491 /* Disallow if any of the three categories say to */
56576a04 1492 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1493 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1494 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1495 {
1496 disallowed = TRUE;
1497 }
1498
d60baaa7
KW
1499 /* Likewise, warn if any say to, plus if deprecation warnings
1500 * are on, because this code point is above IV_MAX */
56576a04 1501 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7 1502 || ! (flags & UTF8_ALLOW_OVERFLOW)
d044b7a7 1503 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1504 {
2b5e7bc2 1505
ddb65933
KW
1506 /* The warnings code explicitly says it doesn't handle the
1507 * case of packWARN2 and two categories which have
1508 * parent-child relationship. Even if it works now to
1509 * raise the warning if either is enabled, it wouldn't
1510 * necessarily do so in the future. We output (only) the
56576a04 1511 * most dire warning */
ddb65933
KW
1512 if (! (flags & UTF8_CHECK_ONLY)) {
1513 if (ckWARN_d(WARN_UTF8)) {
1514 pack_warn = packWARN(WARN_UTF8);
1515 }
1516 else if (ckWARN_d(WARN_NON_UNICODE)) {
1517 pack_warn = packWARN(WARN_NON_UNICODE);
1518 }
1519 if (pack_warn) {
1520 message = Perl_form(aTHX_ "%s: %s (overflows)",
1521 malformed_text,
05b9033b 1522 _byte_dump_string(s0, curlen, 0));
ddb65933 1523 }
2b5e7bc2
KW
1524 }
1525 }
1526 }
1527 else if (possible_problems & UTF8_GOT_EMPTY) {
1528 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1529 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1530
1531 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1532
1533 /* This so-called malformation is now treated as a bug in
1534 * the caller. If you have nothing to decode, skip calling
1535 * this function */
1536 assert(0);
1537
2b5e7bc2
KW
1538 disallowed = TRUE;
1539 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1540 pack_warn = packWARN(WARN_UTF8);
1541 message = Perl_form(aTHX_ "%s (empty string)",
1542 malformed_text);
1543 }
1544 }
1545 }
1546 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1547 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1548 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1549
1550 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1551 disallowed = TRUE;
1552 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1553 pack_warn = packWARN(WARN_UTF8);
1554 message = Perl_form(aTHX_
1555 "%s: %s (unexpected continuation byte 0x%02x,"
1556 " with no preceding start byte)",
1557 malformed_text,
7e2f38b2 1558 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1559 }
1560 }
1561 }
2b5e7bc2
KW
1562 else if (possible_problems & UTF8_GOT_SHORT) {
1563 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1564 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1565
1566 if (! (flags & UTF8_ALLOW_SHORT)) {
1567 disallowed = TRUE;
1568 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1569 pack_warn = packWARN(WARN_UTF8);
1570 message = Perl_form(aTHX_
56576a04
KW
1571 "%s: %s (too short; %d byte%s available, need %d)",
1572 malformed_text,
1573 _byte_dump_string(s0, send - s0, 0),
1574 (int)avail_len,
1575 avail_len == 1 ? "" : "s",
1576 (int)expectlen);
2b5e7bc2
KW
1577 }
1578 }
ba210ebe 1579
2b5e7bc2 1580 }
e308b348
KW
1581 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1582 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1583 *errors |= UTF8_GOT_NON_CONTINUATION;
1584
1585 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1586 disallowed = TRUE;
1587 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1588
1589 /* If we don't know for sure that the input length is
1590 * valid, avoid as much as possible reading past the
1591 * end of the buffer */
1592 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1593 ? s - s0
1594 : send - s0;
e308b348
KW
1595 pack_warn = packWARN(WARN_UTF8);
1596 message = Perl_form(aTHX_ "%s",
1597 unexpected_non_continuation_text(s0,
99a765e9 1598 printlen,
e308b348
KW
1599 s - s0,
1600 (int) expectlen));
1601 }
1602 }
1603 }
2b5e7bc2
KW
1604 else if (possible_problems & UTF8_GOT_SURROGATE) {
1605 possible_problems &= ~UTF8_GOT_SURROGATE;
1606
f9380377
KW
1607 if (flags & UTF8_WARN_SURROGATE) {
1608 *errors |= UTF8_GOT_SURROGATE;
1609
1610 if ( ! (flags & UTF8_CHECK_ONLY)
1611 && ckWARN_d(WARN_SURROGATE))
1612 {
2b5e7bc2
KW
1613 pack_warn = packWARN(WARN_SURROGATE);
1614
1615 /* These are the only errors that can occur with a
1616 * surrogate when the 'uv' isn't valid */
1617 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1618 message = Perl_form(aTHX_
1619 "UTF-16 surrogate (any UTF-8 sequence that"
1620 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1621 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1622 }
1623 else {
c94c2f39 1624 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1625 }
f9380377 1626 }
2b5e7bc2 1627 }
ba210ebe 1628
2b5e7bc2
KW
1629 if (flags & UTF8_DISALLOW_SURROGATE) {
1630 disallowed = TRUE;
f9380377 1631 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1632 }
1633 }
1634 else if (possible_problems & UTF8_GOT_SUPER) {
1635 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1636
f9380377
KW
1637 if (flags & UTF8_WARN_SUPER) {
1638 *errors |= UTF8_GOT_SUPER;
1639
1640 if ( ! (flags & UTF8_CHECK_ONLY)
1641 && ckWARN_d(WARN_NON_UNICODE))
1642 {
2b5e7bc2
KW
1643 pack_warn = packWARN(WARN_NON_UNICODE);
1644
1645 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1646 message = Perl_form(aTHX_
1647 "Any UTF-8 sequence that starts with"
1648 " \"%s\" is for a non-Unicode code point,"
1649 " may not be portable",
7e2f38b2 1650 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1651 }
1652 else {
c94c2f39 1653 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1654 }
f9380377 1655 }
2b5e7bc2 1656 }
ba210ebe 1657
57ff5f59
KW
1658 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1659 * and before possibly bailing out, so that the more dire
1660 * warning will override the regular one. */
1661 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1662 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1663 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1664 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1665 {
db0f09e6 1666 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1667
57ff5f59
KW
1668 /* If it is an overlong that evaluates to a code point
1669 * that doesn't have to use the Perl extended UTF-8, it
1670 * still used it, and so we output a message that
1671 * doesn't refer to the code point. The same is true
1672 * if there was a SHORT malformation where the code
1673 * point is not valid. In that case, 'uv' will have
1674 * been set to the REPLACEMENT CHAR, and the message
1675 * below without the code point in it will be selected
1676 * */
1677 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1678 message = Perl_form(aTHX_
57ff5f59 1679 perl_extended_cp_format, uv);
2b5e7bc2
KW
1680 }
1681 else {
1682 message = Perl_form(aTHX_
57ff5f59
KW
1683 "Any UTF-8 sequence that starts with"
1684 " \"%s\" is a Perl extension, and"
1685 " so is not portable",
1686 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1687 }
1688 }
1689
d044b7a7
KW
1690 if (flags & ( UTF8_WARN_PERL_EXTENDED
1691 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1692 {
d044b7a7 1693 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1694
d044b7a7 1695 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1696 disallowed = TRUE;
1697 }
2b5e7bc2
KW
1698 }
1699 }
eb83ed87 1700
2b5e7bc2 1701 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1702 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1703 disallowed = TRUE;
1704 }
eb83ed87 1705
2b5e7bc2
KW
1706 /* The deprecated warning overrides any non-deprecated one. If
1707 * there are other problems, a deprecation message is not
1708 * really helpful, so don't bother to raise it in that case.
1709 * This also keeps the code from having to handle the case
1710 * where 'uv' is not valid. */
1711 if ( ! (orig_problems
1712 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1713 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1714 && ckWARN_d(WARN_DEPRECATED))
1715 {
1716 message = Perl_form(aTHX_ cp_above_legal_max,
1717 uv, MAX_NON_DEPRECATED_CP);
1718 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1719 }
1720 }
1721 else if (possible_problems & UTF8_GOT_NONCHAR) {
1722 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1723
f9380377
KW
1724 if (flags & UTF8_WARN_NONCHAR) {
1725 *errors |= UTF8_GOT_NONCHAR;
1726
1727 if ( ! (flags & UTF8_CHECK_ONLY)
1728 && ckWARN_d(WARN_NONCHAR))
1729 {
2b5e7bc2
KW
1730 /* The code above should have guaranteed that we don't
1731 * get here with errors other than overlong */
1732 assert (! (orig_problems
1733 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1734
1735 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1736 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1737 }
2b5e7bc2 1738 }
5b311467 1739
2b5e7bc2
KW
1740 if (flags & UTF8_DISALLOW_NONCHAR) {
1741 disallowed = TRUE;
f9380377 1742 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1743 }
6c64cd9d
KW
1744 }
1745 else if (possible_problems & UTF8_GOT_LONG) {
1746 possible_problems &= ~UTF8_GOT_LONG;
1747 *errors |= UTF8_GOT_LONG;
1748
1749 if (flags & UTF8_ALLOW_LONG) {
1750
1751 /* We don't allow the actual overlong value, unless the
1752 * special extra bit is also set */
1753 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1754 & ~UTF8_ALLOW_LONG)))
1755 {
1756 uv = UNICODE_REPLACEMENT;
1757 }
1758 }
1759 else {
1760 disallowed = TRUE;
1761
1762 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1763 pack_warn = packWARN(WARN_UTF8);
1764
1765 /* These error types cause 'uv' to be something that
1766 * isn't what was intended, so can't use it in the
1767 * message. The other error types either can't
1768 * generate an overlong, or else the 'uv' is valid */
1769 if (orig_problems &
1770 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1771 {
1772 message = Perl_form(aTHX_
1773 "%s: %s (any UTF-8 sequence that starts"
1774 " with \"%s\" is overlong which can and"
1775 " should be represented with a"
1776 " different, shorter sequence)",
1777 malformed_text,
1778 _byte_dump_string(s0, send - s0, 0),
1779 _byte_dump_string(s0, curlen, 0));
1780 }
1781 else {
1782 U8 tmpbuf[UTF8_MAXBYTES+1];
1783 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1784 uv, 0);
1785 const char * preface = (uv <= PERL_UNICODE_MAX)
1786 ? "U+"
1787 : "0x";
1788 message = Perl_form(aTHX_
1789 "%s: %s (overlong; instead use %s to represent"
1790 " %s%0*" UVXf ")",
1791 malformed_text,
1792 _byte_dump_string(s0, send - s0, 0),
1793 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1794 preface,
1795 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1796 small code points */
1797 uv);
1798 }
1799 }
1800 }
2b5e7bc2
KW
1801 } /* End of looking through the possible flags */
1802
1803 /* Display the message (if any) for the problem being handled in
1804 * this iteration of the loop */
1805 if (message) {
1806 if (PL_op)
1807 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1808 OP_DESC(PL_op));
1809 else
1810 Perl_warner(aTHX_ pack_warn, "%s", message);
1811 }
ddb65933 1812 } /* End of 'while (possible_problems)' */
a0dbb045 1813
2b5e7bc2
KW
1814 /* Since there was a possible problem, the returned length may need to
1815 * be changed from the one stored at the beginning of this function.
1816 * Instead of trying to figure out if that's needed, just do it. */
1817 if (retlen) {
1818 *retlen = curlen;
1819 }
a0dbb045 1820
2b5e7bc2
KW
1821 if (disallowed) {
1822 if (flags & UTF8_CHECK_ONLY && retlen) {
1823 *retlen = ((STRLEN) -1);
1824 }
1825 return 0;
1826 }
eb83ed87 1827 }
ba210ebe 1828
2b5e7bc2 1829 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1830}
1831
8e84507e 1832/*
ec5f19d0
KW
1833=for apidoc utf8_to_uvchr_buf
1834
1835Returns the native code point of the first character in the string C<s> which
1836is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1837C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1838
524080c4
KW
1839If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1840enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1841C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1842(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1843C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1844the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1845See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1846returned.
ec5f19d0 1847
760c7c2f
KW
1848Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1849unless those are turned off.
1850
ec5f19d0 1851=cut
52be2536
KW
1852
1853Also implemented as a macro in utf8.h
1854
ec5f19d0
KW
1855*/
1856
1857
1858UV
1859Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1860{
7f974d7e
KW
1861 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1862
ec5f19d0
KW
1863 assert(s < send);
1864
1865 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1866 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1867}
1868
52be2536
KW
1869/* This is marked as deprecated
1870 *
ec5f19d0
KW
1871=for apidoc utf8_to_uvuni_buf
1872
de69f3af
KW
1873Only in very rare circumstances should code need to be dealing in Unicode
1874(as opposed to native) code points. In those few cases, use
1875C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1876
1877Returns the Unicode (not-native) code point of the first character in the
1878string C<s> which
ec5f19d0
KW
1879is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1880C<retlen> will be set to the length, in bytes, of that character.
1881
524080c4
KW
1882If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1883enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1884NULL) to -1. If those warnings are off, the computed value if well-defined (or
1885the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1886is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1887next possible position in C<s> that could begin a non-malformed character.
de69f3af 1888See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1889
760c7c2f
KW
1890Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1891unless those are turned off.
1892
ec5f19d0
KW
1893=cut
1894*/
1895
1896UV
1897Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1898{
1899 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1900
1901 assert(send > s);
1902
5962d97e
KW
1903 /* Call the low level routine, asking for checks */
1904 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1905}
1906
b76347f2 1907/*
87cea99e 1908=for apidoc utf8_length
b76347f2
JH
1909
1910Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1911Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1912up past C<e>, croaks.
b76347f2
JH
1913
1914=cut
1915*/
1916
1917STRLEN
35a4481c 1918Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1919{
1920 STRLEN len = 0;
1921
7918f24d
NC
1922 PERL_ARGS_ASSERT_UTF8_LENGTH;
1923
8850bf83
JH
1924 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1925 * the bitops (especially ~) can create illegal UTF-8.
1926 * In other words: in Perl UTF-8 is not just for Unicode. */
1927
a3b680e6
AL
1928 if (e < s)
1929 goto warn_and_return;
b76347f2 1930 while (s < e) {
4cbf4130 1931 s += UTF8SKIP(s);
8e91ec7f
AV
1932 len++;
1933 }
1934
1935 if (e != s) {
1936 len--;
1937 warn_and_return:
9b387841
NC
1938 if (PL_op)
1939 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1940 "%s in %s", unees, OP_DESC(PL_op));
1941 else
61a12c31 1942 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1943 }
1944
1945 return len;
1946}
1947
b06226ff 1948/*
fed3ba5d
NC
1949=for apidoc bytes_cmp_utf8
1950
a1433954 1951Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1952sequence of characters (stored as UTF-8)
1953in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1954equal, -1 or -2 if the first string is less than the second string, +1 or +2
1955if the first string is greater than the second string.
1956
1957-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1958longer string. -2 or +2 is returned if
1959there was a difference between characters
fed3ba5d
NC
1960within the strings.
1961
1962=cut
1963*/
1964
1965int
1966Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1967{
1968 const U8 *const bend = b + blen;
1969 const U8 *const uend = u + ulen;
1970
1971 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1972
1973 while (b < bend && u < uend) {
1974 U8 c = *u++;
1975 if (!UTF8_IS_INVARIANT(c)) {
1976 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1977 if (u < uend) {
1978 U8 c1 = *u++;
1979 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1980 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1981 } else {
2b5e7bc2 1982 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1983 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1984 "%s %s%s",
1985 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1986 PL_op ? " in " : "",
1987 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1988 return -2;
1989 }
1990 } else {
1991 if (PL_op)
1992 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1993 "%s in %s", unees, OP_DESC(PL_op));
1994 else
61a12c31 1995 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1996 return -2; /* Really want to return undef :-) */
1997 }
1998 } else {
1999 return -2;
2000 }
2001 }
2002 if (*b != c) {
2003 return *b < c ? -2 : +2;
2004 }
2005 ++b;
2006 }
2007
2008 if (b == bend && u == uend)
2009 return 0;
2010
2011 return b < bend ? +1 : -1;
2012}
2013
2014/*
87cea99e 2015=for apidoc utf8_to_bytes
6940069f 2016
3bc0c78c 2017Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2018Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2019updates C<*lenp> to contain the new length.
3bc0c78c
KW
2020Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2021
2022Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2023having saved the value of C<*lenp> before the call, and subtracting the
2024after-call value of C<*lenp> from it.
6940069f 2025
a1433954 2026If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2027
6940069f
GS
2028=cut
2029*/
2030
2031U8 *
09af0336 2032Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2033{
9fe0d3c2 2034 U8 * first_variant;
246fae53 2035
7918f24d 2036 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2037 PERL_UNUSED_CONTEXT;
7918f24d 2038
9fe0d3c2 2039 /* This is a no-op if no variants at all in the input */
09af0336 2040 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2041 return s;
2042 }
2043
2044 {
3c5aa262 2045 U8 * const save = s;
09af0336 2046 U8 * const send = s + *lenp;
3c5aa262
KW
2047 U8 * d;
2048
2049 /* Nothing before the first variant needs to be changed, so start the real
2050 * work there */
2051 s = first_variant;
2052 while (s < send) {
2053 if (! UTF8_IS_INVARIANT(*s)) {
2054 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2055 *lenp = ((STRLEN) -1);
3c5aa262
KW
2056 return 0;
2057 }
2058 s++;
d59937ca
KW
2059 }
2060 s++;
dcad2880 2061 }
dcad2880 2062
3c5aa262
KW
2063 /* Is downgradable, so do it */
2064 d = s = first_variant;
2065 while (s < send) {
2066 U8 c = *s++;
2067 if (! UVCHR_IS_INVARIANT(c)) {
2068 /* Then it is two-byte encoded */
2069 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2070 s++;
2071 }
2072 *d++ = c;
2073 }
2074 *d = '\0';
09af0336 2075 *lenp = d - save;
3c5aa262
KW
2076
2077 return save;
9fe0d3c2 2078 }
6940069f
GS
2079}
2080
2081/*
87cea99e 2082=for apidoc bytes_from_utf8
f9a63242 2083
09af0336 2084Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2085byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2086actually encoded in UTF-8.
2087
2088Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2089the input string.
2090
41ae6089
KW
2091Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2092not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2093C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2094
41ae6089 2095Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2096newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2097returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2098
3bc0c78c 2099Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2100having saved the value of C<*lenp> before the call, and subtracting the
2101after-call value of C<*lenp> from it.
3bc0c78c 2102
37607a96 2103=cut
976c1b08
KW
2104
2105There is a macro that avoids this function call, but this is retained for
2106anyone who calls it with the Perl_ prefix */
f9a63242
JH
2107
2108U8 *
41ae6089 2109Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2110{
7918f24d 2111 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2112 PERL_UNUSED_CONTEXT;
f9a63242 2113
976c1b08
KW
2114 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2115}
2116
2117/*
2118No = here because currently externally undocumented
2119for apidoc bytes_from_utf8_loc
2120
2121Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2122to store the location of the first character in C<"s"> that cannot be
2123converted to non-UTF8.
2124
2125If that parameter is C<NULL>, this function behaves identically to
2126C<bytes_from_utf8>.
2127
2128Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2129C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2130
2131Otherwise, the function returns a newly created C<NUL>-terminated string
2132containing the non-UTF8 equivalent of the convertible first portion of
2133C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2134If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2135and C<*first_non_downgradable> is set to C<NULL>.
2136
2137Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2138first character in the original string that wasn't converted. C<*is_utf8p> is
2139unchanged. Note that the new string may have length 0.
2140
2141Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2142C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2143converts as many characters in it as possible stopping at the first one it
385b74be 2144finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2145set to point to that. The function returns the portion that could be converted
2146in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2147not including the terminating C<NUL>. If the very first character in the
2148original could not be converted, C<*lenp> will be 0, and the new string will
2149contain just a single C<NUL>. If the entire input string was converted,
2150C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2151
2152Upon successful return, the number of variants in the converted portion of the
2153string can be computed by having saved the value of C<*lenp> before the call,
2154and subtracting the after-call value of C<*lenp> from it.
2155
2156=cut
2157
2158
2159*/
2160
2161U8 *
2162Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2163{
2164 U8 *d;
2165 const U8 *original = s;
2166 U8 *converted_start;
2167 const U8 *send = s + *lenp;
f9a63242 2168
976c1b08 2169 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2170
976c1b08
KW
2171 if (! *is_utf8p) {
2172 if (first_unconverted) {
2173 *first_unconverted = NULL;
2174 }
2175
2176 return (U8 *) original;
2177 }
2178
2179 Newx(d, (*lenp) + 1, U8);
2180
2181 converted_start = d;
7299a045
KW
2182 while (s < send) {
2183 U8 c = *s++;
2184 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2185
2186 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2187 * have to stop now */
2188 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2189 if (first_unconverted) {
2190 *first_unconverted = s - 1;
2191 goto finish_and_return;
2192 }
2193 else {
2194 Safefree(converted_start);
2195 return (U8 *) original;
2196 }
2197 }
2198
7299a045
KW
2199 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2200 s++;
38af28cf 2201 }
7299a045
KW
2202 *d++ = c;
2203 }
170a1c22 2204
976c1b08
KW
2205 /* Here, converted the whole of the input */
2206 *is_utf8p = FALSE;
2207 if (first_unconverted) {
2208 *first_unconverted = NULL;
170a1c22 2209 }
976c1b08
KW
2210
2211 finish_and_return:
2212 *d = '\0';
2213 *lenp = d - converted_start;
2214
2215 /* Trim unused space */
2216 Renew(converted_start, *lenp + 1, U8);
2217
2218 return converted_start;
f9a63242
JH
2219}
2220
2221/*
87cea99e 2222=for apidoc bytes_to_utf8
6940069f 2223
09af0336 2224Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2225UTF-8.
09af0336 2226Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2227reflect the new length in bytes.
6940069f 2228
3bc0c78c 2229Upon successful return, the number of variants in the string can be computed by
23b37b12 2230having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2231after-call value of C<*lenp>.
2232
75200dff 2233A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2234
2235If you want to convert to UTF-8 from encodings other than
2236the native (Latin1 or EBCDIC),
a1433954 2237see L</sv_recode_to_utf8>().
c9ada85f 2238
497711e7 2239=cut
6940069f
GS
2240*/
2241
2242U8*
09af0336 2243Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2244{
09af0336 2245 const U8 * const send = s + (*lenp);
6940069f
GS
2246 U8 *d;
2247 U8 *dst;
7918f24d
NC
2248
2249 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2250 PERL_UNUSED_CONTEXT;
6940069f 2251
09af0336 2252 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2253 dst = d;
2254
2255 while (s < send) {
55d09dc8
KW
2256 append_utf8_from_native_byte(*s, &d);
2257 s++;
6940069f
GS
2258 }
2259 *d = '\0';
09af0336 2260 *lenp = d-dst;
6940069f
GS
2261 return dst;
2262}
2263
a0ed51b3 2264/*
dea0fc0b 2265 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2266 *
2267 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2268 * We optimize for native, for obvious reasons. */
2269
2270U8*
dea0fc0b 2271Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2272{
dea0fc0b
JH
2273 U8* pend;
2274 U8* dstart = d;
2275
7918f24d
NC
2276 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2277
dea0fc0b 2278 if (bytelen & 1)
56576a04
KW
2279 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2280 (UV)bytelen);
dea0fc0b
JH
2281
2282 pend = p + bytelen;
2283
a0ed51b3 2284 while (p < pend) {
dea0fc0b
JH
2285 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2286 p += 2;
2d1545e5 2287 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2288 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2289 continue;
2290 }
56d37426
KW
2291 if (uv <= MAX_UTF8_TWO_BYTE) {
2292 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2293 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2294 continue;
2295 }
46956fad
KW
2296#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2297#define LAST_HIGH_SURROGATE 0xDBFF
2298#define FIRST_LOW_SURROGATE 0xDC00
2299#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2300
2301 /* This assumes that most uses will be in the first Unicode plane, not
2302 * needing surrogates */
2303 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2304 && uv <= UNICODE_SURROGATE_LAST))
2305 {
2306 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2307 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2308 }
2309 else {
01ea242b 2310 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2311 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2312 || UNLIKELY(low > LAST_LOW_SURROGATE))
2313 {
01ea242b 2314 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2315 }
2316 p += 2;
46956fad
KW
2317 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2318 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2319 }
a0ed51b3 2320 }
56d37426
KW
2321#ifdef EBCDIC
2322 d = uvoffuni_to_utf8_flags(d, uv, 0);
2323#else
a0ed51b3 2324 if (uv < 0x10000) {
eb160463
GS
2325 *d++ = (U8)(( uv >> 12) | 0xe0);
2326 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2327 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2328 continue;
2329 }
2330 else {
eb160463
GS
2331 *d++ = (U8)(( uv >> 18) | 0xf0);
2332 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2333 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2334 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2335 continue;
2336 }
56d37426 2337#endif
a0ed51b3 2338 }
dea0fc0b 2339 *newlen = d - dstart;
a0ed51b3
LW
2340 return d;
2341}
2342
2343/* Note: this one is slightly destructive of the source. */
2344
2345U8*
dea0fc0b 2346Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2347{
2348 U8* s = (U8*)p;
d4c19fe8 2349 U8* const send = s + bytelen;
7918f24d
NC
2350
2351 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2352
e0ea5e2d 2353 if (bytelen & 1)
147e3846 2354 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2355 (UV)bytelen);
2356
a0ed51b3 2357 while (s < send) {
d4c19fe8 2358 const U8 tmp = s[0];
a0ed51b3
LW
2359 s[0] = s[1];
2360 s[1] = tmp;
2361 s += 2;
2362 }
dea0fc0b 2363 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2364}
2365
922e8cb4
KW
2366bool
2367Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2368{
2369 U8 tmpbuf[UTF8_MAXBYTES+1];
2370 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2371 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2372}
2373
f9ae8fb6
JD
2374/* Internal function so we can deprecate the external one, and call
2375 this one from other deprecated functions in this file */
2376
f2645549
KW
2377bool
2378Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2379{
f2645549 2380 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2381
2382 if (*p == '_')
2383 return TRUE;
f25ce844 2384 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2385}
2386
5092f92a 2387bool
eba68aa0
KW
2388Perl__is_uni_perl_idcont(pTHX_ UV c)
2389{
2390 U8 tmpbuf[UTF8_MAXBYTES+1];
2391 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2392 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2393}
2394
2395bool
f91dcd13
KW
2396Perl__is_uni_perl_idstart(pTHX_ UV c)
2397{
2398 U8 tmpbuf[UTF8_MAXBYTES+1];
2399 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2400 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2401}
2402
3a4c58c9 2403UV
56576a04
KW
2404Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2405 const char S_or_s)
3a4c58c9
KW
2406{
2407 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2408 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2409 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2410 * either "SS" or "Ss". Which one to use is passed into the routine in
2411 * 'S_or_s' to avoid a test */
2412
2413 UV converted = toUPPER_LATIN1_MOD(c);
2414
2415 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2416
2417 assert(S_or_s == 'S' || S_or_s == 's');
2418
6f2d5cbc 2419 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2420 characters in this range */
3a4c58c9
KW
2421 *p = (U8) converted;
2422 *lenp = 1;
2423 return converted;
2424 }
2425
2426 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2427 * which it maps to one of them, so as to only have to have one check for
2428 * it in the main case */
2429 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2430 switch (c) {
2431 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2432 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2433 break;
2434 case MICRO_SIGN:
2435 converted = GREEK_CAPITAL_LETTER_MU;
2436 break;
79e064b9
KW
2437#if UNICODE_MAJOR_VERSION > 2 \
2438 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2439 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2440 case LATIN_SMALL_LETTER_SHARP_S:
2441 *(p)++ = 'S';
2442 *p = S_or_s;
2443 *lenp = 2;
2444 return 'S';
79e064b9 2445#endif
3a4c58c9 2446 default:
56576a04
KW
2447 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2448 " '%c' to map to '%c'",
2449 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2450 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2451 }
2452 }
2453
2454 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2455 *p = UTF8_TWO_BYTE_LO(converted);
2456 *lenp = 2;
2457
2458 return converted;
2459}
2460
50bda2c3
KW
2461/* Call the function to convert a UTF-8 encoded character to the specified case.
2462 * Note that there may be more than one character in the result.
2463 * INP is a pointer to the first byte of the input character
2464 * OUTP will be set to the first byte of the string of changed characters. It
2465 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2466 * LENP will be set to the length in bytes of the string of changed characters
2467 *
56576a04
KW
2468 * The functions return the ordinal of the first character in the string of
2469 * OUTP */
2470#define CALL_UPPER_CASE(uv, s, d, lenp) \
2471 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2472#define CALL_TITLE_CASE(uv, s, d, lenp) \
2473 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2474#define CALL_LOWER_CASE(uv, s, d, lenp) \
2475 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2476
b9992569
KW
2477/* This additionally has the input parameter 'specials', which if non-zero will
2478 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2479 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2480#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2481_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2482
84afefe6
JH
2483UV
2484Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2485{
a1433954
KW
2486 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2487 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2488 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2489 * the changed version may be longer than the original character.
2490 *
2491 * The ordinal of the first character of the changed version is returned
2492 * (but note, as explained above, that there may be more.) */
2493
7918f24d
NC
2494 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2495
3a4c58c9
KW
2496 if (c < 256) {
2497 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2498 }
2499
0ebc6274 2500 uvchr_to_utf8(p, c);
b9992569 2501 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2502}
2503
84afefe6
JH
2504UV
2505Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2506{
7918f24d
NC
2507 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2508
3a4c58c9
KW
2509 if (c < 256) {
2510 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2511 }
2512
0ebc6274 2513 uvchr_to_utf8(p, c);
b9992569 2514 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2515}
2516
afc16117 2517STATIC U8
eaf412bf 2518S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2519{
2520 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2521 * those, converting the result to UTF-8. Since the result is always just
a1433954 2522 * one character, we allow <p> to be NULL */
afc16117
KW
2523
2524 U8 converted = toLOWER_LATIN1(c);
2525
eaf412bf
KW
2526 PERL_UNUSED_ARG(dummy);
2527
afc16117 2528 if (p != NULL) {
6f2d5cbc 2529 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2530 *p = converted;
2531 *lenp = 1;
2532 }
2533 else {
430c9760
KW
2534 /* Result is known to always be < 256, so can use the EIGHT_BIT
2535 * macros */
2536 *p = UTF8_EIGHT_BIT_HI(converted);
2537 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2538 *lenp = 2;
2539 }
2540 }
2541 return converted;
2542}
2543
84afefe6
JH
2544UV
2545Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2546{
7918f24d
NC
2547 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2548
afc16117 2549 if (c < 256) {
eaf412bf 2550 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2551 }
2552
afc16117 2553 uvchr_to_utf8(p, c);
b9992569 2554 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2555}
2556
84afefe6 2557UV
56576a04
KW
2558Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2559 const unsigned int flags)
a1dde8de 2560{
51910141 2561 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2562 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2563 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2564 *
2565 * Not to be used for locale folds
51910141 2566 */
f673fad4 2567
a1dde8de
KW
2568 UV converted;
2569
2570 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2571 PERL_UNUSED_CONTEXT;
a1dde8de 2572
1ca267a5
KW
2573 assert (! (flags & FOLD_FLAGS_LOCALE));
2574
659a7c2d 2575 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2576 converted = GREEK_SMALL_LETTER_MU;
2577 }
9b63e895
KW
2578#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2579 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2580 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2581 else if ( (flags & FOLD_FLAGS_FULL)
2582 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2583 {
1ca267a5
KW
2584 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2585 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2586 * under those circumstances. */
2587 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2588 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2589 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2590 p, *lenp, U8);
2591 return LATIN_SMALL_LETTER_LONG_S;
2592 }
2593 else {
4f489194
KW
2594 *(p)++ = 's';
2595 *p = 's';
2596 *lenp = 2;
2597 return 's';
1ca267a5 2598 }
a1dde8de 2599 }
9b63e895 2600#endif
a1dde8de
KW
2601 else { /* In this range the fold of all other characters is their lower
2602 case */
2603 converted = toLOWER_LATIN1(c);
2604 }
2605
6f2d5cbc 2606 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2607 *p = (U8) converted;
2608 *lenp = 1;
2609 }
2610 else {
2611 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2612 *p = UTF8_TWO_BYTE_LO(converted);
2613 *lenp = 2;
2614 }
2615
2616 return converted;
2617}
2618
2619UV
31f05a37 2620Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2621{
4b593389 2622
a0270393
KW
2623 /* Not currently externally documented, and subject to change
2624 * <flags> bits meanings:
2625 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2626 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2627 * locale are to be used.
a0270393
KW
2628 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2629 */
4b593389 2630
36bb2ab6 2631 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2632
780fcc9f
KW
2633 if (flags & FOLD_FLAGS_LOCALE) {
2634 /* Treat a UTF-8 locale as not being in locale at all */
2635 if (IN_UTF8_CTYPE_LOCALE) {
2636 flags &= ~FOLD_FLAGS_LOCALE;
2637 }
2638 else {
2639 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2640 goto needs_full_generality;
780fcc9f 2641 }
31f05a37
KW
2642 }
2643
a1dde8de 2644 if (c < 256) {
e7b7ac46 2645 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2646 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2647 }
2648
2f306ab9 2649 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2650 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2651 uvchr_to_utf8(p, c);
b9992569 2652 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2653 }
567b353c 2654 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2655 the special flags. */
2656 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2657
2658 needs_full_generality:
a0270393 2659 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2660 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2661 p, lenp, flags);
a0270393 2662 }
84afefe6
JH
2663}
2664
26483009 2665PERL_STATIC_INLINE bool
5141f98e 2666S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2667 const char *const swashname, SV* const invlist)
bde6a22d 2668{
ea317ccb
KW
2669 /* returns a boolean giving whether or not the UTF8-encoded character that
2670 * starts at <p> is in the swash indicated by <swashname>. <swash>
2671 * contains a pointer to where the swash indicated by <swashname>
2672 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2673 * look at <*swash> and only generate a swash if it is not null. <invlist>
2674 * is NULL or an inversion list that defines the swash. If not null, it
2675 * saves time during initialization of the swash.
ea317ccb
KW
2676 *
2677 * Note that it is assumed that the buffer length of <p> is enough to
2678 * contain all the bytes that comprise the character. Thus, <*p> should
2679 * have been checked before this call for mal-formedness enough to assure
2680 * that. */
2681
7918f24d
NC
2682 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2683
492a624f 2684 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2685 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2686 * as far as there being enough bytes available in it to accommodate the
2687 * character without reading beyond the end, and pass that number on to the
2688 * validating routine */
6302f837 2689 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2690 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2691 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2692 1 /* Die */ );
2693 NOT_REACHED; /* NOTREACHED */
28123549 2694 }
86ae6e94 2695
87367d5f
KW
2696 if (!*swash) {
2697 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2698 *swash = _core_swash_init("utf8",
2699
2700 /* Only use the name if there is no inversion
2701 * list; otherwise will go out to disk */
2702 (invlist) ? "" : swashname,
2703
2704 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2705 }
28123549 2706
bde6a22d
NC
2707 return swash_fetch(*swash, p, TRUE) != 0;
2708}
2709
da8c1a98 2710PERL_STATIC_INLINE bool
56576a04
KW
2711S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2712 SV **swash, const char *const swashname,
2713 SV* const invlist)
da8c1a98
KW
2714{
2715 /* returns a boolean giving whether or not the UTF8-encoded character that
2716 * starts at <p>, and extending no further than <e - 1> is in the swash
2717 * indicated by <swashname>. <swash> contains a pointer to where the swash
2718 * indicated by <swashname> is to be stored; which this routine will do, so
2719 * that future calls will look at <*swash> and only generate a swash if it
2720 * is not null. <invlist> is NULL or an inversion list that defines the
2721 * swash. If not null, it saves time during initialization of the swash.
2722 */
2723
2724 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2725
2726 if (! isUTF8_CHAR(p, e)) {
2727 _force_out_malformed_utf8_message(p, e, 0, 1);
2728 NOT_REACHED; /* NOTREACHED */
2729 }
2730
2731 if (!*swash) {
2732 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2733 *swash = _core_swash_init("utf8",
2734
2735 /* Only use the name if there is no inversion
2736 * list; otherwise will go out to disk */
2737 (invlist) ? "" : swashname,
2738
2739 &PL_sv_undef, 1, 0, invlist, &flags);
2740 }
2741
2742 return swash_fetch(*swash, p, TRUE) != 0;
2743}
2744
34aeb2e9
KW
2745STATIC void
2746S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2747 const char * const alternative,
2748 const bool use_locale,
2749 const char * const file,
2750 const unsigned line)
2751{
2752 const char * key;
2753
2754 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2755
2756 if (ckWARN_d(WARN_DEPRECATED)) {
2757
2758 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2759 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2760 if (! PL_seen_deprecated_macro) {
2761 PL_seen_deprecated_macro = newHV();
2762 }
2763 if (! hv_store(PL_seen_deprecated_macro, key,
2764 strlen(key), &PL_sv_undef, 0))
2765 {
2766 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2767 }
2768
c44e9413 2769 if (instr(file, "mathoms.c")) {
607313a1
KW
2770 Perl_warner(aTHX_ WARN_DEPRECATED,
2771 "In %s, line %d, starting in Perl v5.30, %s()"
2772 " will be removed. Avoid this message by"
2773 " converting to use %s().\n",
2774 file, line, name, alternative);
2775 }
2776 else {
34aeb2e9
KW
2777 Perl_warner(aTHX_ WARN_DEPRECATED,
2778 "In %s, line %d, starting in Perl v5.30, %s() will"
2779 " require an additional parameter. Avoid this"
2780 " message by converting to use %s().\n",
2781 file, line, name, alternative);
607313a1 2782 }
34aeb2e9
KW
2783 }
2784 }
2785}
2786
bde6a22d 2787bool
34aeb2e9 2788Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2789 const U8 * const p,
34aeb2e9
KW
2790 const char * const name,
2791 const char * const alternative,
2792 const bool use_utf8,
2793 const bool use_locale,
2794 const char * const file,
2795 const unsigned line)
922e8cb4 2796{
922e8cb4
KW
2797 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2798
34aeb2e9
KW
2799 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2800
2801 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2802
2803 switch (classnum) {
2804 case _CC_WORDCHAR:
2805 case _CC_DIGIT:
2806 case _CC_ALPHA:
2807 case _CC_LOWER:
2808 case _CC_UPPER:
2809 case _CC_PUNCT:
2810 case _CC_PRINT:
2811 case _CC_ALPHANUMERIC:
2812 case _CC_GRAPH:
2813 case _CC_CASED:
2814
2815 return is_utf8_common(p,
2816 &PL_utf8_swash_ptrs[classnum],
2817 swash_property_names[classnum],
2818 PL_XPosix_ptrs[classnum]);
2819
2820 case _CC_SPACE:
2821 return is_XPERLSPACE_high(p);
2822 case _CC_BLANK:
2823 return is_HORIZWS_high(p);
2824 case _CC_XDIGIT:
2825 return is_XDIGIT_high(p);
2826 case _CC_CNTRL:
2827 return 0;
2828 case _CC_ASCII:
2829 return 0;
2830 case _CC_VERTSPACE:
2831 return is_VERTWS_high(p);
2832 case _CC_IDFIRST:
2833 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2834 PL_utf8_perl_idstart
2835 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2836 }
22f0498f
KW
2837 return is_utf8_common(p, &PL_utf8_perl_idstart,
2838 "_Perl_IDStart", NULL);
34aeb2e9
KW
2839 case _CC_IDCONT:
2840 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2841 PL_utf8_perl_idcont
2842 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2843 }
22f0498f
KW
2844 return is_utf8_common(p, &PL_utf8_perl_idcont,
2845 "_Perl_IDCont", NULL);
34aeb2e9
KW
2846 }
2847 }
2848
2849 /* idcont is the same as wordchar below 256 */
2850 if (classnum == _CC_IDCONT) {
2851 classnum = _CC_WORDCHAR;
2852 }
2853 else if (classnum == _CC_IDFIRST) {
2854 if (*p == '_') {
2855 return TRUE;
2856 }
2857 classnum = _CC_ALPHA;
2858 }
2859
2860 if (! use_locale) {
2861 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2862 return _generic_isCC(*p, classnum);
2863 }
922e8cb4 2864
34aeb2e9
KW
2865 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2866 }
2867 else {
2868 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2869 return isFOO_lc(classnum, *p);
2870 }
2871
2872 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2873 }
2874
2875 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2876}
2877
2878bool
da8c1a98
KW
2879Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2880 const U8 * const e)
2881{
2882 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2883
2884 assert(classnum < _FIRST_NON_SWASH_CC);
2885
2886 return is_utf8_common_with_len(p,
2887 e,
2888 &PL_utf8_swash_ptrs[classnum],
2889 swash_property_names[classnum],
2890 PL_XPosix_ptrs[classnum]);
2891}
2892
2893bool
da8c1a98
KW
2894Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2895{
2896 SV* invlist = NULL;
2897
2898 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2899
2900 if (! PL_utf8_perl_idstart) {
2901 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2902 }
2903 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2904 "_Perl_IDStart", invlist);
2905}
2906
2907bool
f2645549 2908Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2909{
f2645549 2910 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2911
2912 if (*p == '_')
2913 return TRUE;
f25ce844 2914 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2915}
2916
2917bool
da8c1a98
KW
2918Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2919{
2920 SV* invlist = NULL;
2921
2922 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2923
2924 if (! PL_utf8_perl_idcont) {
2925 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2926 }
2927 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2928 "_Perl_IDCont", invlist);
2929}
2930
2931bool
f2645549 2932Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2933{
f2645549 2934 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2935
f25ce844 2936 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2937}
2938
2939bool
f2645549 2940Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2941{
f2645549 2942 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2943
f25ce844 2944 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2945}
2946
2947bool
7dbf68d2
KW
2948Perl__is_utf8_mark(pTHX_ const U8 *p)
2949{
7dbf68d2
KW
2950 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2951
f25ce844 2952 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2953}
2954
b9992569 2955 /* change namve uv1 to 'from' */
6a4a25f4 2956STATIC UV
b9992569
KW
2957S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2958 SV **swashp, const char *normal, const char *special)
2959{
0134edef 2960 STRLEN len = 0;
7918f24d 2961
b9992569 2962 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2963
36eaa811
KW
2964 /* For code points that don't change case, we already know that the output
2965 * of this function is the unchanged input, so we can skip doing look-ups
2966 * for them. Unfortunately the case-changing code points are scattered
2967 * around. But there are some long consecutive ranges where there are no
2968 * case changing code points. By adding tests, we can eliminate the lookup
2969 * for all the ones in such ranges. This is currently done here only for
2970 * just a few cases where the scripts are in common use in modern commerce
2971 * (and scripts adjacent to those which can be included without additional
2972 * tests). */
2973
2974 if (uv1 >= 0x0590) {
2975 /* This keeps from needing further processing the code points most
2976 * likely to be used in the following non-cased scripts: Hebrew,
2977 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2978 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2979 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2980 if (uv1 < 0x10A0) {
2981 goto cases_to_self;
2982 }
2983
2984 /* The following largish code point ranges also don't have case
2985 * changes, but khw didn't think they warranted extra tests to speed
2986 * them up (which would slightly slow down everything else above them):
2987 * 1100..139F Hangul Jamo, Ethiopic
2988 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2989 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2990 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2991 * Combining Diacritical Marks Extended, Balinese,
2992 * Sundanese, Batak, Lepcha, Ol Chiki
2993 * 2000..206F General Punctuation
2994 */
2995
2996 if (uv1 >= 0x2D30) {
2997
2998 /* This keeps the from needing further processing the code points
2999 * most likely to be used in the following non-cased major scripts:
3000 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3001 *
3002 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3003 * event that Unicode eventually allocates the unused block as of
3004 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3005 * that the test suite will start having failures to alert you
3006 * should that happen) */
3007 if (uv1 < 0xA640) {
3008 goto cases_to_self;
3009 }
3010
3011 if (uv1 >= 0xAC00) {
3012 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3013 if (ckWARN_d(WARN_SURROGATE)) {
3014 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3015 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3016 "Operation \"%s\" returns its argument for"
3017 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3018 }
3019 goto cases_to_self;
3020 }
36eaa811
KW
3021
3022 /* AC00..FAFF Catches Hangul syllables and private use, plus
3023 * some others */
3024 if (uv1 < 0xFB00) {
3025 goto cases_to_self;
3026
3027 }
3028
5af9bc97 3029 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
3030 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
3031 && ckWARN_d(WARN_DEPRECATED))
3032 {
3033 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
3034 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
3035 }
3036 if (ckWARN_d(WARN_NON_UNICODE)) {
3037 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3038 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3039 "Operation \"%s\" returns its argument for"
3040 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3041 }
3042 goto cases_to_self;
3043 }
3bfc1e70
KW
3044#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3045 if (UNLIKELY(uv1
3046 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3047 {
3048
56576a04
KW
3049 /* As of Unicode 10.0, this means we avoid swash creation
3050 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3051 goto cases_to_self;
3052 }
3053#endif
36eaa811
KW
3054 }
3055 }
9ae3ac1a 3056
36eaa811
KW
3057 /* Note that non-characters are perfectly legal, so no warning should
3058 * be given. There are so few of them, that it isn't worth the extra
3059 * tests to avoid swash creation */
9ae3ac1a
KW
3060 }
3061
0134edef 3062 if (!*swashp) /* load on-demand */
56576a04
KW
3063 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3064 4, 0, NULL, NULL);
0134edef 3065
a6f87d8c 3066 if (special) {
0134edef 3067 /* It might be "special" (sometimes, but not always,
2a37f04d 3068 * a multicharacter mapping) */
4a8240a3 3069 HV *hv = NULL;
b08cf34e
JH
3070 SV **svp;
3071
4a8240a3
KW
3072 /* If passed in the specials name, use that; otherwise use any
3073 * given in the swash */
3074 if (*special != '\0') {
3075 hv = get_hv(special, 0);
3076 }
3077 else {
3078 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3079 if (svp) {
3080 hv = MUTABLE_HV(SvRV(*svp));
3081 }
3082 }
3083
176fe009 3084 if (hv
5f560d8a 3085 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3086 && (*svp))
3087 {
cfd0369c 3088 const char *s;
47654450 3089
cfd0369c 3090 s = SvPV_const(*svp, len);
47654450 3091 if (len == 1)
f4cd282c 3092 /* EIGHTBIT */
c80e42f3 3093 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3094 else {
d2dcd0fb 3095 Copy(s, ustrp, len, U8);
29e98929 3096 }
983ffd37 3097 }
0134edef
JH
3098 }
3099
3100 if (!len && *swashp) {
4a4088c4 3101 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3102
0134edef
JH
3103 if (uv2) {
3104 /* It was "normal" (a single character mapping). */
f4cd282c 3105 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3106 }
3107 }
1feea2c7 3108
cbe07460
KW
3109 if (len) {
3110 if (lenp) {
3111 *lenp = len;
3112 }
3113 return valid_utf8_to_uvchr(ustrp, 0);
3114 }
3115
3116 /* Here, there was no mapping defined, which means that the code point maps
3117 * to itself. Return the inputs */
e24dfe9c 3118 cases_to_self:
bfdf22ec 3119 len = UTF8SKIP(p);
ca9fab46
KW
3120 if (p != ustrp) { /* Don't copy onto itself */
3121 Copy(p, ustrp, len, U8);
3122 }
0134edef 3123
2a37f04d
JH
3124 if (lenp)
3125 *lenp = len;
3126
f4cd282c 3127 return uv1;
cbe07460 3128
a0ed51b3
LW
3129}
3130
051a06d4 3131STATIC UV
56576a04
KW
3132S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3133 U8* const ustrp, STRLEN *lenp)
051a06d4 3134{
4a4088c4 3135 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3136 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3137 * result contains a character that crosses the 255/256 boundary, disallow
3138 * the change, and return the original code point. See L<perlfunc/lc> for
3139 * why;
051a06d4 3140 *
a1433954
KW
3141 * p points to the original string whose case was changed; assumed
3142 * by this routine to be well-formed
051a06d4 3143 * result the code point of the first character in the changed-case string
56576a04
KW
3144 * ustrp points to the changed-case string (<result> represents its
3145 * first char)
051a06d4
KW
3146 * lenp points to the length of <ustrp> */
3147
3148 UV original; /* To store the first code point of <p> */
3149
3150 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3151
a4f12ed7 3152 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3153
3154 /* We know immediately if the first character in the string crosses the
3155 * boundary, so can skip */
3156 if (result > 255) {
3157
3158 /* Look at every character in the result; if any cross the
3159 * boundary, the whole thing is disallowed */
3160 U8* s = ustrp + UTF8SKIP(ustrp);
3161 U8* e = ustrp + *lenp;
3162 while (s < e) {
a4f12ed7 3163 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3164 goto bad_crossing;
3165 }
3166 s += UTF8SKIP(s);
3167 }
3168
613abc6d
KW
3169 /* Here, no characters crossed, result is ok as-is, but we warn. */
3170 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3171 return result;
3172 }
3173
7b52d656 3174 bad_crossing:
051a06d4
KW
3175
3176 /* Failed, have to return the original */
4b88fb76 3177 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3178
3179 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3180 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3181 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3182 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3183 OP_DESC(PL_op),
ab0b796c
KW
3184 original,
3185 original);
051a06d4
KW
3186 Copy(p, ustrp, *lenp, char);
3187 return original;
3188}
3189
607313a1
KW
3190STATIC U32
3191S_check_and_deprecate(pTHX_ const U8 *p,
3192 const U8 **e,
3193 const unsigned int type, /* See below */
3194 const bool use_locale, /* Is this a 'LC_'
3195 macro call? */
3196 const char * const file,
3197 const unsigned line)
3198{
3199 /* This is a temporary function to deprecate the unsafe calls to the case
3200 * changing macros and functions. It keeps all the special stuff in just
3201 * one place.
3202 *
3203 * It updates *e with the pointer to the end of the input string. If using
3204 * the old-style macros, *e is NULL on input, and so this function assumes
3205 * the input string is long enough to hold the entire UTF-8 sequence, and
3206 * sets *e accordingly, but it then returns a flag to pass the
3207 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3208 * using the full length if possible.
3209 *
3210 * It also does the assert that *e > p when *e is not NULL. This should be
3211 * migrated to the callers when this function gets deleted.
3212 *
3213 * The 'type' parameter is used for the caller to specify which case
3214 * changing function this is called from: */
3215
3216# define DEPRECATE_TO_UPPER 0
3217# define DEPRECATE_TO_TITLE 1
3218# define DEPRECATE_TO_LOWER 2
3219# define DEPRECATE_TO_FOLD 3
3220
3221 U32 utf8n_flags = 0;
3222 const char * name;
3223 const char * alternative;
3224
3225 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3226
3227 if (*e == NULL) {
3228 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3229 *e = p + UTF8SKIP(p);
3230
3231 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3232 * there. It could be part of a larger path */
607313a1 3233 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3234 name = instr(file, "mathoms.c")
607313a1
KW
3235 ? "to_utf8_upper"
3236 : "toUPPER_utf8";
3237 alternative = "toUPPER_utf8_safe";
3238 }
3239 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3240 name = instr(file, "mathoms.c")
607313a1
KW
3241 ? "to_utf8_title"
3242 : "toTITLE_utf8";
3243 alternative = "toTITLE_utf8_safe";
3244 }
3245 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3246 name = instr(file, "mathoms.c")
607313a1
KW
3247 ? "to_utf8_lower"
3248 : "toLOWER_utf8";
3249 alternative = "toLOWER_utf8_safe";
3250 }
3251 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3252 name = instr(file, "mathoms.c")
607313a1
KW
3253 ? "to_utf8_fold"
3254 : "toFOLD_utf8";
3255 alternative = "toFOLD_utf8_safe";
3256 }
3257 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3258
3259 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3260 }
3261 else {
3262 assert (p < *e);
3263 }
3264
3265 return utf8n_flags;
3266}
3267
eaf412bf
KW
3268/* The process for changing the case is essentially the same for the four case
3269 * change types, except there are complications for folding. Otherwise the
3270 * difference is only which case to change to. To make sure that they all do
3271 * the same thing, the bodies of the functions are extracted out into the
3272 * following two macros. The functions are written with the same variable
3273 * names, and these are known and used inside these macros. It would be
3274 * better, of course, to have inline functions to do it, but since different
3275 * macros are called, depending on which case is being changed to, this is not
3276 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3277 * function can start with the common start macro, then finish with its special
3278 * handling; while the other three cases can just use the common end macro.
3279 *
3280 * The algorithm is to use the proper (passed in) macro or function to change
3281 * the case for code points that are below 256. The macro is used if using
3282 * locale rules for the case change; the function if not. If the code point is
3283 * above 255, it is computed from the input UTF-8, and another macro is called
3284 * to do the conversion. If necessary, the output is converted to UTF-8. If
3285 * using a locale, we have to check that the change did not cross the 255/256
3286 * boundary, see check_locale_boundary_crossing() for further details.
3287 *
3288 * The macros are split with the correct case change for the below-256 case
3289 * stored into 'result', and in the middle of an else clause for the above-255
3290 * case. At that point in the 'else', 'result' is not the final result, but is
3291 * the input code point calculated from the UTF-8. The fold code needs to
3292 * realize all this and take it from there.
3293 *
3294 * If you read the two macros as sequential, it's easier to understand what's
3295 * going on. */
3296#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3297 L1_func_extra_param) \
a239b1e2 3298 \
eaf412bf
KW
3299 if (flags & (locale_flags)) { \
3300 /* Treat a UTF-8 locale as not being in locale at all */ \
3301 if (IN_UTF8_CTYPE_LOCALE) { \
3302 flags &= ~(locale_flags); \
3303 } \
3304 else { \
3305 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3306 } \
3307 } \
3308 \
3309 if (UTF8_IS_INVARIANT(*p)) { \
3310 if (flags & (locale_flags)) { \
3311 result = LC_L1_change_macro(*p); \
3312 } \
3313 else { \
3314 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3315 } \
3316 } \
a239b1e2 3317 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3318 if (flags & (locale_flags)) { \
3319 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3320 *(p+1))); \
3321 } \
3322 else { \
3323 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3324 ustrp, lenp, L1_func_extra_param); \
3325 } \
3326 } \
fa8ab374
KW
3327 else { /* malformed UTF-8 or ord above 255 */ \
3328 STRLEN len_result; \
fa8ab374
KW
3329 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3330 if (len_result == (STRLEN) -1) { \
607313a1
KW
3331 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3332 1 /* Die */ ); \
fa8ab374 3333 }
eaf412bf
KW
3334
3335#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3336 result = change_macro(result, p, ustrp, lenp); \
3337 \
3338 if (flags & (locale_flags)) { \
3339 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3340 } \
3341 return result; \
3342 } \
3343 \
3344 /* Here, used locale rules. Convert back to UTF-8 */ \
3345 if (UTF8_IS_INVARIANT(result)) { \
3346 *ustrp = (U8) result; \
3347 *lenp = 1; \
3348 } \
3349 else { \
3350 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3351 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3352 *lenp = 2; \
3353 } \
3354 \
3355 return result;
3356
d3e79532 3357/*
87cea99e 3358=for apidoc to_utf8_upper
d3e79532 3359
a239b1e2 3360Instead use L</toUPPER_utf8_safe>.
a1433954 3361
d3e79532
JH
3362=cut */
3363
051a06d4 3364/* Not currently externally documented, and subject to change:
31f05a37
KW
3365 * <flags> is set iff iff the rules from the current underlying locale are to
3366 * be used. */
051a06d4 3367
2104c8d9 3368UV
607313a1
KW
3369Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3370 const U8 *e,
3371 U8* ustrp,
3372 STRLEN *lenp,
3373 bool flags,
3374 const char * const file,
3375 const int line)
a0ed51b3 3376{
051a06d4 3377 UV result;
607313a1
KW
3378 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3379 cBOOL(flags), file, line);
051a06d4
KW
3380
3381 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3382
eaf412bf
KW
3383 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3384 /* 2nd char of uc(U+DF) is 'S' */
3385 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3386 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3387}
a0ed51b3 3388
d3e79532 3389/*
87cea99e 3390=for apidoc to_utf8_title
d3e79532 3391
a239b1e2 3392Instead use L</toTITLE_utf8_safe>.
a1433954 3393
d3e79532
JH
3394=cut */
3395
051a06d4 3396/* Not currently externally documented, and subject to change:
31f05a37
KW
3397 * <flags> is set iff the rules from the current underlying locale are to be
3398 * used. Since titlecase is not defined in POSIX, for other than a
3399 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3400 */
051a06d4 3401
983ffd37 3402UV
607313a1
KW
3403Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3404 const U8 *e,
3405 U8* ustrp,
3406 STRLEN *lenp,
3407 bool flags,
3408 const char * const file,
3409 const int line)
983ffd37 3410{
051a06d4 3411 UV result;
607313a1
KW
3412 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3413 cBOOL(flags), file, line);
051a06d4
KW
3414
3415 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3416
eaf412bf
KW
3417 /* 2nd char of ucfirst(U+DF) is 's' */
3418 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3419 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3420}
3421
d3e79532 3422/*
87cea99e 3423=for apidoc to_utf8_lower
d3e79532 3424
a239b1e2 3425Instead use L</toLOWER_utf8_safe>.
a1433954 3426
d3e79532
JH
3427=cut */
3428
051a06d4 3429/* Not currently externally documented, and subject to change:
31f05a37
KW
3430 * <flags> is set iff iff the rules from the current underlying locale are to
3431 * be used.
3432 */
051a06d4 3433
2104c8d9 3434UV
607313a1
KW
3435Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3436 const U8 *e,
3437 U8* ustrp,
3438 STRLEN *lenp,
3439 bool flags,
3440 const char * const file,
3441 const int line)
a0ed51b3 3442{
051a06d4 3443 UV result;
607313a1
KW
3444 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3445 cBOOL(flags), file, line);
051a06d4 3446
051a06d4 3447 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3448
eaf412bf
KW
3449 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3450 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3451}
3452
d3e79532 3453/*
87cea99e 3454=for apidoc to_utf8_fold
d3e79532 3455
a239b1e2 3456Instead use L</toFOLD_utf8_safe>.
a1433954 3457
d3e79532
JH
3458=cut */
3459
051a06d4
KW
3460/* Not currently externally documented, and subject to change,
3461 * in <flags>
31f05a37
KW
3462 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3463 * locale are to be used.
051a06d4
KW
3464 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3465 * otherwise simple folds
a0270393
KW
3466 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3467 * prohibited
445bf929 3468 */
36bb2ab6 3469
b4e400f9 3470UV
607313a1
KW
3471Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3472 const U8 *e,
3473 U8* ustrp,
3474 STRLEN *lenp,
3475 U8 flags,
3476 const char * const file,
3477 const int line)
b4e400f9 3478{
051a06d4 3479 UV result;
607313a1
KW
3480 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3481 cBOOL(flags), file, line);
051a06d4 3482
36bb2ab6 3483 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3484
a0270393
KW
3485 /* These are mutually exclusive */
3486 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3487
50ba90ff
KW
3488 assert(p != ustrp); /* Otherwise overwrites */
3489
eaf412bf
KW
3490 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3491 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3492
eaf412bf 3493 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3494
1ca267a5
KW
3495 if (flags & FOLD_FLAGS_LOCALE) {
3496
76f2ffcd 3497# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3498 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3499
3500# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3501# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3502
3503 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3504
538e84ed
KW
3505 /* Special case these two characters, as what normally gets
3506 * returned under locale doesn't work */
76f2ffcd
KW
3507 if (UTF8SKIP(p) == cap_sharp_s_len
3508 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3509 {
ab0b796c
KW
3510 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3511 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3512 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3513 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3514 goto return_long_s;
3515 }
0766489e
KW
3516 else
3517#endif
3518 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3519 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3520 {
ab0b796c
KW
3521 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3522 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3523 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3524 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3525 goto return_ligature_st;
3526 }
74894415
KW
3527
3528#if UNICODE_MAJOR_VERSION == 3 \
3529 && UNICODE_DOT_VERSION == 0 \
3530 && UNICODE_DOT_DOT_VERSION == 1
3531# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3532
3533 /* And special case this on this Unicode version only, for the same
3534 * reaons the other two are special cased. They would cross the
3535 * 255/256 boundary which is forbidden under /l, and so the code
3536 * wouldn't catch that they are equivalent (which they are only in
3537 * this release) */
3538 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3539 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3540 {
3541 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3542 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3543 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3544 "resolved to \"\\x{0131}\".");
3545 goto return_dotless_i;
3546 }
3547#endif
3548
357aadde 3549 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3550 }
a0270393
KW
3551 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3552 return result;
3553 }
3554 else {
4a4088c4 3555 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3556 * character above the ASCII range, and the result should not
3557 * contain an ASCII character. */
a0270393
KW
3558
3559 UV original; /* To store the first code point of <p> */
3560
3561 /* Look at every character in the result; if any cross the
3562 * boundary, the whole thing is disallowed */
3563 U8* s = ustrp;
3564 U8* e = ustrp + *lenp;
3565 while (s < e) {
3566 if (isASCII(*s)) {
3567 /* Crossed, have to return the original */
3568 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3569
9fc2026f 3570 /* But in these instances, there is an alternative we can
1ca267a5 3571 * return that is valid */
0766489e
KW
3572 if (original == LATIN_SMALL_LETTER_SHARP_S
3573#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3574 || original == LATIN_CAPITAL_LETTER_SHARP_S
3575#endif
3576 ) {
1ca267a5
KW
3577 goto return_long_s;
3578 }
9fc2026f
KW
3579 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3580 goto return_ligature_st;
3581 }
74894415
KW
3582#if UNICODE_MAJOR_VERSION == 3 \
3583 && UNICODE_DOT_VERSION == 0 \
3584 && UNICODE_DOT_DOT_VERSION == 1
3585
3586 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3587 goto return_dotless_i;
3588 }
3589#endif
a0270393
KW
3590 Copy(p, ustrp, *lenp, char);
3591 return original;
3592 }
3593 s += UTF8SKIP(s);
3594 }
051a06d4 3595
a0270393
KW
3596 /* Here, no characters crossed, result is ok as-is */
3597 return result;
3598 }
051a06d4
KW
3599 }
3600
4a4088c4 3601 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3602 if (UTF8_IS_INVARIANT(result)) {
3603 *ustrp = (U8) result;
3604 *lenp = 1;
3605 }
3606 else {
62cb07ea
KW
3607 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3608 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3609 *lenp = 2;
3610 }
3611
051a06d4 3612 return result;
1ca267a5
KW
3613
3614 return_long_s:
3615 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3616 * folds to a string of two of these characters. By returning this
3617 * instead, then, e.g.,
3618 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3619 * works. */
3620
3621 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3622 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3623 ustrp, *lenp, U8);
3624 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3625
3626 return_ligature_st:
3627 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3628 * have the other one fold to it */
3629
3630 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3631 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3632 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3633
3634#if UNICODE_MAJOR_VERSION == 3 \
3635 && UNICODE_DOT_VERSION == 0 \
3636 && UNICODE_DOT_DOT_VERSION == 1
3637
3638 return_dotless_i:
3639 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3640 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3641 return LATIN_SMALL_LETTER_DOTLESS_I;
3642
3643#endif
3644
a0ed51b3
LW
3645}
3646
711a919c 3647/* Note:
f90a9a02 3648 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3649 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3650 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3651 */
c4a5db0c 3652
a0ed51b3 3653SV*
56576a04
KW
3654Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3655 I32 minbits, I32 none)
a0ed51b3 3656{
c4a5db0c
KW
3657 PERL_ARGS_ASSERT_SWASH_INIT;
3658
3659 /* Returns a copy of a swash initiated by the called function. This is the
3660 * public interface, and returning a copy prevents others from doing
3661 * mischief on the original */
3662
56576a04
KW
3663 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3664 NULL, NULL));
c4a5db0c
KW
3665}
3666
3667SV*
56576a04
KW
3668Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3669 I32 minbits, I32 none, SV* invlist,
3670 U8* const flags_p)
c4a5db0c 3671{
2c1f00b9
YO
3672
3673 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3674 * use the following define */
3675
3676#define CORE_SWASH_INIT_RETURN(x) \
3677 PL_curpm= old_PL_curpm; \
3678 return x
3679
c4a5db0c 3680 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3681 * by calling utf8_heavy.pl in the general case. The returned value may be
3682 * the swash's inversion list instead if the input parameters allow it.
3683 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3684 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3685 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3686 *
3687 * This interface should only be used by functions that won't destroy or
3688 * adversely change the swash, as doing so affects all other uses of the
3689 * swash in the program; the general public should use 'Perl_swash_init'
3690 * instead.
3691 *
3692 * pkg is the name of the package that <name> should be in.
3693 * name is the name of the swash to find. Typically it is a Unicode
3694 * property name, including user-defined ones
3695 * listsv is a string to initialize the swash with. It must be of the form
3696 * documented as the subroutine return value in
3697 * L<perlunicode/User-Defined Character Properties>
3698 * minbits is the number of bits required to represent each data element.
3699 * It is '1' for binary properties.
3700 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3701 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3702 * flags_p if non-NULL is the address of various input and output flag bits
3703 * to the routine, as follows: ('I' means is input to the routine;
3704 * 'O' means output from the routine. Only flags marked O are
3705 * meaningful on return.)
3706 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3707 * came from a user-defined property. (I O)
5d3d13d1
KW
3708 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3709 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3710 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3711 * return of an inversion list instead of a swash hash if this routine
3712 * thinks that would result in faster execution of swash_fetch() later
3713 * on. (I)
9a5