This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Generalize static fcn return for indeterminate result
[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 *
425 * U+7FFFFFFF (2 ** 31 - 1)
426 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
427 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
428 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
429 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
430 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
431 * U+80000000 (2 ** 31):
432 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
433 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
434 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
435 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
436 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
437 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
438 */
439
440#ifdef EBCDIC
441
37086697
KW
442 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
443 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
444 const STRLEN prefix_len = sizeof(prefix) - 1;
445 const STRLEN len = e - s;
f880f78a 446 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
447
448#else
449
450 PERL_UNUSED_ARG(e);
451
452#endif
453
454 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
455
456 assert(! UTF8_IS_INVARIANT(*s));
457
458#ifndef EBCDIC
459
460 /* Technically, a start byte of FE can be for a code point that fits into
461 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
462 * malformation. */
463 return (*s >= 0xFE);
464
465#else
466
56576a04
KW
467 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
468 * mean a 32-bit or larger code point (0xFF is an invariant). For 0xFE, we
469 * need at least 2 bytes, and maybe up through 8 bytes, to be sure that the
470 * value is above 31 bits. */
83dc0f42
KW
471 if (*s != 0xFE || len == 1) {
472 return FALSE;
473 }
474
475 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
476 * \x41 and \x42. */
477 return cBOOL(memGT(s + 1, prefix, cmp_len));
478
479#endif
480
481}
482
57ff5f59
KW
483#endif
484
d6be65ae 485PERL_STATIC_INLINE int
12a4bed3
KW
486S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
487{
d6be65ae
KW
488 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
489 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
490 * it isn't, and -1 if there isn't enough information to tell. This last
491 * return value can happen if the sequence is incomplete, missing some
492 * trailing bytes that would form a complete character. If there are
493 * enough bytes to make a definitive decision, this function does so.
494 * Usually 2 bytes sufficient.
495 *
496 * Overlongs can occur whenever the number of continuation bytes changes.
497 * That means whenever the number of leading 1 bits in a start byte
498 * increases from the next lower start byte. That happens for start bytes
499 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
500 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
501 * ASCII platforms: C0, C1
502 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 503 */
12a4bed3
KW
504
505 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
506 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
507
508 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
509 assert(len > 1 && UTF8_IS_START(*s));
510
511 /* Each platform has overlongs after the start bytes given above (expressed
512 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
513 * the logic is the same, except the E0 overlong has already been excluded
514 * on EBCDIC platforms. The values below were found by manually
515 * inspecting the UTF-8 patterns. See the tables in utf8.h and
516 * utfebcdic.h. */
517
518# ifdef EBCDIC
519# define F0_ABOVE_OVERLONG 0xB0
520# define F8_ABOVE_OVERLONG 0xA8
521# define FC_ABOVE_OVERLONG 0xA4
522# define FE_ABOVE_OVERLONG 0xA2
523# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
524 /* I8(0xfe) is FF */
525# else
526
527 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 528 return 1;
12a4bed3
KW
529 }
530
531# define F0_ABOVE_OVERLONG 0x90
532# define F8_ABOVE_OVERLONG 0x88
533# define FC_ABOVE_OVERLONG 0x84
534# define FE_ABOVE_OVERLONG 0x82
535# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
536# endif
537
538
539 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
540 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
541 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
542 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
543 {
d6be65ae 544 return 1;
12a4bed3
KW
545 }
546
b0b342d4 547 /* Check for the FF overlong */
d6be65ae 548 return isFF_OVERLONG(s, len);
b0b342d4
KW
549}
550
8d6204cc 551PERL_STATIC_INLINE int
b0b342d4
KW
552S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
553{
8d6204cc
KW
554 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
555 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
556 * it isn't, and -1 if there isn't enough information to tell. This last
557 * return value can happen if the sequence is incomplete, missing some
558 * trailing bytes that would form a complete character. If there are
559 * enough bytes to make a definitive decision, this function does so. */
560
b0b342d4 561 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 562
8d6204cc
KW
563 /* To be an FF overlong, all the available bytes must match */
564 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
565 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
566 {
567 return 0;
568 }
569
570 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
571 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 572 * utfebcdic.h. */
8d6204cc
KW
573 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
574 return 1;
575 }
12a4bed3 576
8d6204cc
KW
577 /* The missing bytes could cause the result to go one way or the other, so
578 * the result is indeterminate */
579 return -1;
12a4bed3
KW
580}
581
a77c906e
KW
582/* Anything larger than this will overflow the word if it were converted into a UV */
583#if defined(UV_IS_QUAD)
584# ifdef EBCDIC /* Actually is I8 */
585# define HIGHEST_REPRESENTABLE_UTF8 \
586 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
587# else
588# define HIGHEST_REPRESENTABLE_UTF8 \
589 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
590# endif
591#else /* 32-bit */
592# ifdef EBCDIC
593# define HIGHEST_REPRESENTABLE_UTF8 \
594 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
595# else
596# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
597# endif
598#endif
599
c285bbc4 600PERL_STATIC_INLINE int
a77c906e
KW
601S_does_utf8_overflow(const U8 * const s, const U8 * e)
602{
c285bbc4
KW
603 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
604 * 'e' - 1 would overflow a UV on this platform; that is if it represents a
605 * code point larger than the highest representable code point. It returns
606 * 1 if it does overflow; 0 if it doesn't, and -1 if there isn't enough
607 * information to tell. This last return value can happen if the sequence
608 * is incomplete, missing some trailing bytes that would form a complete
609 * character. If there are enough bytes to make a definitive decision,
610 * this function does so.
611 *
612 * (For ASCII platforms, we could use memcmp() because we don't have to
613 * convert each byte to I8, but it's very rare input indeed that would
614 * approach overflow, so the loop below will likely only get executed once.)
615 *
616 * 'e' - 1 must not be beyond a full character. */
a77c906e
KW
617
618 const STRLEN len = e - s;
c285bbc4
KW
619 const U8 *x;
620 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
a77c906e
KW
621
622 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
623 assert(s <= e && s + UTF8SKIP(s) >= e);
624
625#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
626
627 /* On 32 bit ASCII machines, many overlongs that start with FF don't
628 * overflow */
a77c906e 629 if (isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
630
631 /* To be such an overlong, the first bytes of 's' must match
632 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
633 * don't have any additional bytes available, the sequence, when
634 * completed might or might not fit in 32 bits. But if we have that
635 * next byte, we can tell for sure. If it is <= 0x83, then it does
636 * fit. */
637 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
638 return -1;
639 }
640
641 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
642 }
643
644#endif
645
646 for (x = s; x < e; x++, y++) {
647
648 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
649 continue;
650 }
651
652 /* If this byte is larger than the corresponding highest UTF-8 byte,
653 * the sequence overflow; otherwise the byte is less than, and so the
654 * sequence doesn't overflow */
655 return NATIVE_UTF8_TO_I8(*x) > *y;
656
657 }
658
659 /* Got to the end and all bytes are the same. If the input is a whole
660 * character, it doesn't overflow. And if it is a partial character,
c285bbc4
KW
661 * there's not enough information to tell */
662 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
663 return -1;
664 }
665
666 return 0;
a77c906e
KW
667}
668
12a4bed3
KW
669#undef F0_ABOVE_OVERLONG
670#undef F8_ABOVE_OVERLONG
671#undef FC_ABOVE_OVERLONG
672#undef FE_ABOVE_OVERLONG
673#undef FF_OVERLONG_PREFIX
674
35f8c9bd 675STRLEN
edc2c47a 676Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 677{
2b479609 678 STRLEN len;
12a4bed3 679 const U8 *x;
35f8c9bd 680
2b479609
KW
681 /* A helper function that should not be called directly.
682 *
683 * This function returns non-zero if the string beginning at 's' and
684 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
685 * code point; otherwise it returns 0. The examination stops after the
686 * first code point in 's' is validated, not looking at the rest of the
687 * input. If 'e' is such that there are not enough bytes to represent a
688 * complete code point, this function will return non-zero anyway, if the
689 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
690 * excluded by 'flags'.
691 *
692 * A non-zero return gives the number of bytes required to represent the
693 * code point. Be aware that if the input is for a partial character, the
694 * return will be larger than 'e - s'.
695 *
696 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
697 * The caller should have excluded the possibility of it being invariant
698 * before calling this function.
2b479609
KW
699 *
700 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
701 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
702 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
703 * disallowed by the flags. If the input is only for a partial character,
704 * the function will return non-zero if there is any sequence of
705 * well-formed UTF-8 that, when appended to the input sequence, could
706 * result in an allowed code point; otherwise it returns 0. Non characters
707 * cannot be determined based on partial character input. But many of the
708 * other excluded types can be determined with just the first one or two
709 * bytes.
710 *
711 */
712
713 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
714
715 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 716 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 717 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 718
2b479609 719 /* A variant char must begin with a start byte */
35f8c9bd
KW
720 if (UNLIKELY(! UTF8_IS_START(*s))) {
721 return 0;
722 }
723
edc2c47a
KW
724 /* Examine a maximum of a single whole code point */
725 if (e - s > UTF8SKIP(s)) {
726 e = s + UTF8SKIP(s);
727 }
728
2b479609
KW
729 len = e - s;
730
731 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
732 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 733
56576a04
KW
734 /* Here, we are disallowing some set of largish code points, and the
735 * first byte indicates the sequence is for a code point that could be
736 * in the excluded set. We generally don't have to look beyond this or
737 * the second byte to see if the sequence is actually for one of the
738 * excluded classes. The code below is derived from this table:
739 *
2b479609
KW
740 * UTF-8 UTF-EBCDIC I8
741 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
742 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
743 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
744 *
56576a04
KW
745 * Keep in mind that legal continuation bytes range between \x80..\xBF
746 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
747 * continuation bytes. Hence, we don't have to test the upper edge
748 * because if any of those is encountered, the sequence is malformed,
749 * and would fail elsewhere in this function.
750 *
751 * The code here likewise assumes that there aren't other
752 * malformations; again the function should fail elsewhere because of
753 * these. For example, an overlong beginning with FC doesn't actually
754 * have to be a super; it could actually represent a small code point,
755 * even U+0000. But, since overlongs (and other malformations) are
756 * illegal, the function should return FALSE in either case.
2b479609
KW
757 */
758
759#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
760# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 761# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 762
19794540
KW
763# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
764 /* B6 and B7 */ \
765 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 766# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
767#else
768# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
769# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
770# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 771# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
772#endif
773
774 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
775 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
776 {
2b479609
KW
777 return 0; /* Above Unicode */
778 }
779
d044b7a7 780 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 781 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 782 {
57ff5f59 783 return 0;
2b479609
KW
784 }
785
786 if (len > 1) {
787 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
788
789 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 790 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
791 {
792 return 0; /* Above Unicode */
793 }
794
795 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 796 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
797 {
798 return 0; /* Surrogate */
799 }
800
801 if ( (flags & UTF8_DISALLOW_NONCHAR)
802 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
803 {
804 return 0; /* Noncharacter code point */
805 }
806 }
807 }
808
809 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
810 for (x = s + 1; x < e; x++) {
811 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
812 return 0;
813 }
814 }
815
af13dd8a 816 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 817 * overlong. */
d6be65ae 818 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 819 return 0;
af13dd8a
KW
820 }
821
12a4bed3
KW
822 /* And finally, that the code point represented fits in a word on this
823 * platform */
c285bbc4 824 if (does_utf8_overflow(s, e) > 0) {
12a4bed3 825 return 0;
35f8c9bd
KW
826 }
827
2b479609 828 return UTF8SKIP(s);
35f8c9bd
KW
829}
830
7e2f38b2
KW
831char *
832Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
833{
834 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
835 * bytes starting at 's'. 'format' gives how to display each byte.
836 * Currently, there are only two formats, so it is currently a bool:
837 * 0 \xab
838 * 1 ab (that is a space between two hex digit bytes)
839 */
7cf8d05d
KW
840
841 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
842 trailing NUL */
843 const U8 * const e = s + len;
844 char * output;
845 char * d;
846
847 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
848
849 Newx(output, output_len, char);
850 SAVEFREEPV(output);
851
852 d = output;
853 for (; s < e; s++) {
854 const unsigned high_nibble = (*s & 0xF0) >> 4;
855 const unsigned low_nibble = (*s & 0x0F);
856
7e2f38b2
KW
857 if (format) {
858 *d++ = ' ';
859 }
860 else {
861 *d++ = '\\';
862 *d++ = 'x';
863 }
7cf8d05d
KW
864
865 if (high_nibble < 10) {
866 *d++ = high_nibble + '0';
867 }
868 else {
869 *d++ = high_nibble - 10 + 'a';
870 }
871
872 if (low_nibble < 10) {
873 *d++ = low_nibble + '0';
874 }
875 else {
876 *d++ = low_nibble - 10 + 'a';
877 }
878 }
879
880 *d = '\0';
881 return output;
882}
883
806547a7 884PERL_STATIC_INLINE char *
7cf8d05d
KW
885S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
886
887 /* How many bytes to print */
3cc6a05e 888 STRLEN print_len,
7cf8d05d
KW
889
890 /* Which one is the non-continuation */
891 const STRLEN non_cont_byte_pos,
892
893 /* How many bytes should there be? */
894 const STRLEN expect_len)
806547a7
KW
895{
896 /* Return the malformation warning text for an unexpected continuation
897 * byte. */
898
7cf8d05d 899 const char * const where = (non_cont_byte_pos == 1)
806547a7 900 ? "immediately"
7cf8d05d
KW
901 : Perl_form(aTHX_ "%d bytes",
902 (int) non_cont_byte_pos);
806547a7
KW
903
904 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
905
7cf8d05d
KW
906 /* We don't need to pass this parameter, but since it has already been
907 * calculated, it's likely faster to pass it; verify under DEBUGGING */
908 assert(expect_len == UTF8SKIP(s));
909
910 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
911 " %s after start byte 0x%02x; need %d bytes, got %d)",
912 malformed_text,
7e2f38b2 913 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
914 *(s + non_cont_byte_pos),
915 where,
916 *s,
917 (int) expect_len,
918 (int) non_cont_byte_pos);
806547a7
KW
919}
920
35f8c9bd
KW
921/*
922
de69f3af 923=for apidoc utf8n_to_uvchr
378516de
KW
924
925THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 926Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 927
9041c2e3 928Bottom level UTF-8 decode routine.
de69f3af 929Returns the native code point value of the first character in the string C<s>,
746afd53
KW
930which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
931C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
932the length, in bytes, of that character.
949cf498
KW
933
934The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
935well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
936causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
937is the next possible position in C<s> that could begin a non-malformed
938character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
939is raised. Some UTF-8 input sequences may contain multiple malformations.
940This function tries to find every possible one in each call, so multiple
56576a04 941warnings can be raised for the same sequence.
949cf498
KW
942
943Various ALLOW flags can be set in C<flags> to allow (and not warn on)
944individual types of malformations, such as the sequence being overlong (that
945is, when there is a shorter sequence that can express the same code point;
946overlong sequences are expressly forbidden in the UTF-8 standard due to
947potential security issues). Another malformation example is the first byte of
948a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
949flags. Even if allowed, this function generally returns the Unicode
950REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
951F<utf8.h> to override this behavior for the overlong malformations, but don't
952do that except for very specialized purposes.
949cf498 953
796b6530 954The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
955flags) malformation is found. If this flag is set, the routine assumes that
956the caller will raise a warning, and this function will silently just set
d088425d
KW
957C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
958
75200dff 959Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 960character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
961in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
962be set to 1. To disambiguate, upon a zero return, see if the first byte of
963C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 964error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
965
966Certain code points are considered problematic. These are Unicode surrogates,
746afd53 967Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 968By default these are considered regular code points, but certain situations
ecc1615f
KW
969warrant special handling for them, which can be specified using the C<flags>
970parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
971three classes are treated as malformations and handled as such. The flags
972C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
973C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
974disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
975restricts the allowed inputs to the strict UTF-8 traditionally defined by
976Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
977definition given by
978L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
979The difference between traditional strictness and C9 strictness is that the
980latter does not forbid non-character code points. (They are still discouraged,
981however.) For more discussion see L<perlunicode/Noncharacter code points>.
982
983The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
984C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
985C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
986raised for their respective categories, but otherwise the code points are
987considered valid (not malformations). To get a category to both be treated as
988a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 989(But note that warnings are not raised if lexically disabled nor if
796b6530 990C<UTF8_CHECK_ONLY> is also specified.)
949cf498 991
57ff5f59
KW
992Extremely high code points were never specified in any standard, and require an
993extension to UTF-8 to express, which Perl does. It is likely that programs
994written in something other than Perl would not be able to read files that
995contain these; nor would Perl understand files written by something that uses a
996different extension. For these reasons, there is a separate set of flags that
997can warn and/or disallow these extremely high code points, even if other
998above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
999C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1000L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1001above-Unicode code points, including these, as malformations.
1002(Note that the Unicode standard considers anything above 0x10FFFF to be
1003illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1004(2**31 -1))
1005
1006A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1007retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1008C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1009C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1010can apply to code points that actually do fit in 31 bits. This happens on
1011EBCDIC platforms, and sometimes when the L<overlong
1012malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1013describe the situation in all cases.
1014
760c7c2f
KW
1015It is now deprecated to have very high code points (above C<IV_MAX> on the
1016platforms) and this function will raise a deprecation warning for these (unless
d5944cab 1017such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 1018in a 32-bit word.
ab8e6d41 1019
949cf498
KW
1020All other code points corresponding to Unicode characters, including private
1021use and those yet to be assigned, are never considered malformed and never
1022warn.
67e989fb 1023
37607a96 1024=cut
f9380377
KW
1025
1026Also implemented as a macro in utf8.h
1027*/
1028
1029UV
1030Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1031 STRLEN curlen,
1032 STRLEN *retlen,
1033 const U32 flags)
1034{
1035 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1036
1037 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1038}
1039
1040/*
1041
1042=for apidoc utf8n_to_uvchr_error
1043
1044THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1045Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1046
1047This function is for code that needs to know what the precise malformation(s)
1048are when an error is found.
1049
1050It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1051all the others, C<errors>. If this parameter is 0, this function behaves
1052identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1053to a C<U32> variable, which this function sets to indicate any errors found.
1054Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1055C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1056of these bits will be set if a malformation is found, even if the input
7a65503b 1057C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1058exceptions are noted:
1059
1060=over 4
1061
57ff5f59 1062=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1063
57ff5f59
KW
1064The input sequence is not standard UTF-8, but a Perl extension. This bit is
1065set only if the input C<flags> parameter contains either the
1066C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1067
1068Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1069and so some extension must be used to express them. Perl uses a natural
1070extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1071extension to represent even higher ones, so that any code point that fits in a
107264-bit word can be represented. Text using these extensions is not likely to
1073be portable to non-Perl code. We lump both of these extensions together and
1074refer to them as Perl extended UTF-8. There exist other extensions that people
1075have invented, incompatible with Perl's.
1076
1077On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1078extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1079than on ASCII. Prior to that, code points 2**31 and higher were simply
1080unrepresentable, and a different, incompatible method was used to represent
1081code points between 2**30 and 2**31 - 1.
1082
1083On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1084Perl extended UTF-8 is used.
1085
1086In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1087may use for backward compatibility. That name is misleading, as this flag may
1088be set when the code point actually does fit in 31 bits. This happens on
1089EBCDIC platforms, and sometimes when the L<overlong
1090malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1091describes the situation in all cases.
f9380377
KW
1092
1093=item C<UTF8_GOT_CONTINUATION>
1094
1095The input sequence was malformed in that the first byte was a a UTF-8
1096continuation byte.
1097
1098=item C<UTF8_GOT_EMPTY>
1099
1100The input C<curlen> parameter was 0.
1101
1102=item C<UTF8_GOT_LONG>
1103
1104The input sequence was malformed in that there is some other sequence that
1105evaluates to the same code point, but that sequence is shorter than this one.
1106
fecaf136
KW
1107Until Unicode 3.1, it was legal for programs to accept this malformation, but
1108it was discovered that this created security issues.
1109
f9380377
KW
1110=item C<UTF8_GOT_NONCHAR>
1111
1112The code point represented by the input UTF-8 sequence is for a Unicode
1113non-character code point.
1114This bit is set only if the input C<flags> parameter contains either the
1115C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1116
1117=item C<UTF8_GOT_NON_CONTINUATION>
1118
1119The input sequence was malformed in that a non-continuation type byte was found
1120in a position where only a continuation type one should be.
1121
1122=item C<UTF8_GOT_OVERFLOW>
1123
1124The input sequence was malformed in that it is for a code point that is not
1125representable in the number of bits available in a UV on the current platform.
1126
1127=item C<UTF8_GOT_SHORT>
1128
1129The input sequence was malformed in that C<curlen> is smaller than required for
1130a complete sequence. In other words, the input is for a partial character
1131sequence.
1132
1133=item C<UTF8_GOT_SUPER>
1134
1135The input sequence was malformed in that it is for a non-Unicode code point;
1136that is, one above the legal Unicode maximum.
1137This bit is set only if the input C<flags> parameter contains either the
1138C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1139
1140=item C<UTF8_GOT_SURROGATE>
1141
1142The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1143code point.
1144This bit is set only if the input C<flags> parameter contains either the
1145C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1146
1147=back
1148
133551d8
KW
1149To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1150flag to suppress any warnings, and then examine the C<*errors> return.
1151
f9380377 1152=cut
37607a96 1153*/
67e989fb 1154
a0ed51b3 1155UV
f9380377
KW
1156Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1157 STRLEN curlen,
1158 STRLEN *retlen,
1159 const U32 flags,
1160 U32 * errors)
a0ed51b3 1161{
d4c19fe8 1162 const U8 * const s0 = s;
2b5e7bc2
KW
1163 U8 * send = NULL; /* (initialized to silence compilers' wrong
1164 warning) */
1165 U32 possible_problems = 0; /* A bit is set here for each potential problem
1166 found as we go along */
eb83ed87 1167 UV uv = *s;
2b5e7bc2
KW
1168 STRLEN expectlen = 0; /* How long should this sequence be?
1169 (initialized to silence compilers' wrong
1170 warning) */
e308b348 1171 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1172 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1173 this gets set and discarded */
a0dbb045 1174
2b5e7bc2
KW
1175 /* The below are used only if there is both an overlong malformation and a
1176 * too short one. Otherwise the first two are set to 's0' and 'send', and
1177 * the third not used at all */
1178 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1179 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1180 routine; see [perl #130921] */
2b5e7bc2 1181 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1182
f9380377
KW
1183 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1184
1185 if (errors) {
1186 *errors = 0;
1187 }
1188 else {
1189 errors = &discard_errors;
1190 }
a0dbb045 1191
eb83ed87
KW
1192 /* The order of malformation tests here is important. We should consume as
1193 * few bytes as possible in order to not skip any valid character. This is
1194 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1195 * http://unicode.org/reports/tr36 for more discussion as to why. For
1196 * example, once we've done a UTF8SKIP, we can tell the expected number of
1197 * bytes, and could fail right off the bat if the input parameters indicate
1198 * that there are too few available. But it could be that just that first
1199 * byte is garbled, and the intended character occupies fewer bytes. If we
1200 * blindly assumed that the first byte is correct, and skipped based on
1201 * that number, we could skip over a valid input character. So instead, we
1202 * always examine the sequence byte-by-byte.
1203 *
1204 * We also should not consume too few bytes, otherwise someone could inject
1205 * things. For example, an input could be deliberately designed to
1206 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1207 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1208 * which is actually part of of the overflowing sequence), that could look
1209 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1210 * sequence and process the rest, inappropriately.
1211 *
1212 * Some possible input sequences are malformed in more than one way. This
1213 * function goes to lengths to try to find all of them. This is necessary
1214 * for correctness, as the inputs may allow one malformation but not
1215 * another, and if we abandon searching for others after finding the
1216 * allowed one, we could allow in something that shouldn't have been.
1217 */
eb83ed87 1218
b5b9af04 1219 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1220 possible_problems |= UTF8_GOT_EMPTY;
1221 curlen = 0;
5a48568d 1222 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1223 goto ready_to_handle_errors;
0c443dc2
JH
1224 }
1225
eb83ed87
KW
1226 expectlen = UTF8SKIP(s);
1227
1228 /* A well-formed UTF-8 character, as the vast majority of calls to this
1229 * function will be for, has this expected length. For efficiency, set
1230 * things up here to return it. It will be overriden only in those rare
1231 * cases where a malformation is found */
1232 if (retlen) {
1233 *retlen = expectlen;
1234 }
1235
1236 /* An invariant is trivially well-formed */
1d72bdf6 1237 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1238 return uv;
a0ed51b3 1239 }
67e989fb 1240
eb83ed87 1241 /* A continuation character can't start a valid sequence */
b5b9af04 1242 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1243 possible_problems |= UTF8_GOT_CONTINUATION;
1244 curlen = 1;
1245 uv = UNICODE_REPLACEMENT;
1246 goto ready_to_handle_errors;
ba210ebe 1247 }
9041c2e3 1248
dcd27b3c 1249 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1250 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1251 * because it excludes start bytes like \xC0 that always lead to
1252 * overlongs.) */
dcd27b3c 1253
534752c1
KW
1254 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1255 * that indicate the number of bytes in the character's whole UTF-8
1256 * sequence, leaving just the bits that are part of the value. */
1257 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1258
e308b348
KW
1259 /* Setup the loop end point, making sure to not look past the end of the
1260 * input string, and flag it as too short if the size isn't big enough. */
1261 send = (U8*) s0;
1262 if (UNLIKELY(curlen < expectlen)) {
1263 possible_problems |= UTF8_GOT_SHORT;
1264 avail_len = curlen;
1265 send += curlen;
1266 }
1267 else {
1268 send += expectlen;
1269 }
e308b348 1270
eb83ed87 1271 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1272 * accumulating each into the working value as we go. */
eb83ed87 1273 for (s = s0 + 1; s < send; s++) {
b5b9af04 1274 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1275 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1276 continue;
1277 }
1278
1279 /* Here, found a non-continuation before processing all expected bytes.
1280 * This byte indicates the beginning of a new character, so quit, even
1281 * if allowing this malformation. */
2b5e7bc2 1282 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1283 break;
eb83ed87
KW
1284 } /* End of loop through the character's bytes */
1285
1286 /* Save how many bytes were actually in the character */
1287 curlen = s - s0;
1288
2b5e7bc2
KW
1289 /* Note that there are two types of too-short malformation. One is when
1290 * there is actual wrong data before the normal termination of the
1291 * sequence. The other is that the sequence wasn't complete before the end
1292 * of the data we are allowed to look at, based on the input 'curlen'.
1293 * This means that we were passed data for a partial character, but it is
1294 * valid as far as we saw. The other is definitely invalid. This
1295 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1296 * separate.
1297 *
1298 * A convenience macro that matches either of the too-short conditions. */
1299# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1300
1301 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1302 uv_so_far = uv;
1303 uv = UNICODE_REPLACEMENT;
1304 }
2b5e7bc2 1305
08e73697
KW
1306 /* Check for overflow. The algorithm requires us to not look past the end
1307 * of the current character, even if partial, so the upper limit is 's' */
c285bbc4 1308 if (UNLIKELY(does_utf8_overflow(s0, s) > 0)) {
2b5e7bc2
KW
1309 possible_problems |= UTF8_GOT_OVERFLOW;
1310 uv = UNICODE_REPLACEMENT;
eb83ed87 1311 }
eb83ed87 1312
2b5e7bc2
KW
1313 /* Check for overlong. If no problems so far, 'uv' is the correct code
1314 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1315 * we must look at the UTF-8 byte sequence itself to see if it is for an
1316 * overlong */
1317 if ( ( LIKELY(! possible_problems)
1318 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1319 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1320 && ( UNLIKELY(! UTF8_IS_START(*s0))
1321 || ( curlen > 1
d6be65ae 1322 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1323 s - s0))))))
2f8f112e 1324 {
2b5e7bc2
KW
1325 possible_problems |= UTF8_GOT_LONG;
1326
abc28b54 1327 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1328
abc28b54
KW
1329 /* The calculation in the 'true' branch of this 'if'
1330 * below won't work if overflows, and isn't needed
1331 * anyway. Further below we handle all overflow
1332 * cases */
1333 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1334 {
2b5e7bc2
KW
1335 UV min_uv = uv_so_far;
1336 STRLEN i;
1337
1338 /* Here, the input is both overlong and is missing some trailing
1339 * bytes. There is no single code point it could be for, but there
1340 * may be enough information present to determine if what we have
1341 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1342 * The code further below has the intelligence to determine this,
1343 * but just for non-overlong UTF-8 sequences. What we do here is
1344 * calculate the smallest code point the input could represent if
1345 * there were no too short malformation. Then we compute and save
1346 * the UTF-8 for that, which is what the code below looks at
1347 * instead of the raw input. It turns out that the smallest such
1348 * code point is all we need. */
2b5e7bc2
KW
1349 for (i = curlen; i < expectlen; i++) {
1350 min_uv = UTF8_ACCUMULATE(min_uv,
1351 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1352 }
1353
e9f2c446 1354 adjusted_s0 = temp_char_buf;
57ff5f59 1355 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1356 }
eb83ed87
KW
1357 }
1358
56576a04
KW
1359 /* Here, we have found all the possible problems, except for when the input
1360 * is for a problematic code point not allowed by the input parameters. */
1361
06188866
KW
1362 /* uv is valid for overlongs */
1363 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1364
1365 /* isn't problematic if < this */
1366 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1367 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1368
1369 /* if overflow, we know without looking further
1370 * precisely which of the problematic types it is,
1371 * and we deal with those in the overflow handling
1372 * code */
1373 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1374 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1375 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1376 && ((flags & ( UTF8_DISALLOW_NONCHAR
1377 |UTF8_DISALLOW_SURROGATE
1378 |UTF8_DISALLOW_SUPER
d044b7a7 1379 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1380 |UTF8_WARN_NONCHAR
1381 |UTF8_WARN_SURROGATE
1382 |UTF8_WARN_SUPER
d044b7a7 1383 |UTF8_WARN_PERL_EXTENDED))
2b5e7bc2
KW
1384 /* In case of a malformation, 'uv' is not valid, and has
1385 * been changed to something in the Unicode range.
1386 * Currently we don't output a deprecation message if there
1387 * is already a malformation, so we don't have to special
1388 * case the test immediately below */
760c7c2f
KW
1389 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1390 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1391 {
2b5e7bc2
KW
1392 /* If there were no malformations, or the only malformation is an
1393 * overlong, 'uv' is valid */
1394 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1395 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1396 possible_problems |= UTF8_GOT_SURROGATE;
1397 }
1398 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1399 possible_problems |= UTF8_GOT_SUPER;
1400 }
1401 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1402 possible_problems |= UTF8_GOT_NONCHAR;
1403 }
1404 }
1405 else { /* Otherwise, need to look at the source UTF-8, possibly
1406 adjusted to be non-overlong */
1407
1408 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1409 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1410 {
2b5e7bc2
KW
1411 possible_problems |= UTF8_GOT_SUPER;
1412 }
1413 else if (curlen > 1) {
1414 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1415 NATIVE_UTF8_TO_I8(*adjusted_s0),
1416 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1417 {
2b5e7bc2 1418 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1419 }
2b5e7bc2
KW
1420 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1421 NATIVE_UTF8_TO_I8(*adjusted_s0),
1422 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1423 {
1424 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1425 }
1426 }
c0236afe 1427
2b5e7bc2
KW
1428 /* We need a complete well-formed UTF-8 character to discern
1429 * non-characters, so can't look for them here */
1430 }
1431 }
949cf498 1432
2b5e7bc2
KW
1433 ready_to_handle_errors:
1434
1435 /* At this point:
1436 * curlen contains the number of bytes in the sequence that
1437 * this call should advance the input by.
e308b348
KW
1438 * avail_len gives the available number of bytes passed in, but
1439 * only if this is less than the expected number of
1440 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1441 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1442 * is set in it for each potential problem found.
1443 * uv contains the code point the input sequence
1444 * represents; or if there is a problem that prevents
1445 * a well-defined value from being computed, it is
1446 * some subsitute value, typically the REPLACEMENT
1447 * CHARACTER.
1448 * s0 points to the first byte of the character
56576a04
KW
1449 * s points to just after were we left off processing
1450 * the character
1451 * send points to just after where that character should
1452 * end, based on how many bytes the start byte tells
1453 * us should be in it, but no further than s0 +
1454 * avail_len
2b5e7bc2 1455 */
eb83ed87 1456
2b5e7bc2
KW
1457 if (UNLIKELY(possible_problems)) {
1458 bool disallowed = FALSE;
1459 const U32 orig_problems = possible_problems;
1460
1461 while (possible_problems) { /* Handle each possible problem */
1462 UV pack_warn = 0;
1463 char * message = NULL;
1464
1465 /* Each 'if' clause handles one problem. They are ordered so that
1466 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1467 * ones; this is kinda in decreasing severity order. But the
1468 * overlong must come last, as it changes 'uv' looked at by the
1469 * others */
2b5e7bc2
KW
1470 if (possible_problems & UTF8_GOT_OVERFLOW) {
1471
56576a04
KW
1472 /* Overflow means also got a super and are using Perl's
1473 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1474 possible_problems
d044b7a7 1475 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1476 *errors |= UTF8_GOT_OVERFLOW;
1477
1478 /* But the API says we flag all errors found */
1479 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1480 *errors |= UTF8_GOT_SUPER;
1481 }
ddb65933 1482 if (flags
d044b7a7 1483 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1484 {
d044b7a7 1485 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1486 }
2b5e7bc2 1487
d60baaa7 1488 /* Disallow if any of the three categories say to */
56576a04 1489 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1490 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1491 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1492 {
1493 disallowed = TRUE;
1494 }
1495
d60baaa7
KW
1496 /* Likewise, warn if any say to, plus if deprecation warnings
1497 * are on, because this code point is above IV_MAX */
56576a04 1498 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7 1499 || ! (flags & UTF8_ALLOW_OVERFLOW)
d044b7a7 1500 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1501 {
2b5e7bc2 1502
ddb65933
KW
1503 /* The warnings code explicitly says it doesn't handle the
1504 * case of packWARN2 and two categories which have
1505 * parent-child relationship. Even if it works now to
1506 * raise the warning if either is enabled, it wouldn't
1507 * necessarily do so in the future. We output (only) the
56576a04 1508 * most dire warning */
ddb65933
KW
1509 if (! (flags & UTF8_CHECK_ONLY)) {
1510 if (ckWARN_d(WARN_UTF8)) {
1511 pack_warn = packWARN(WARN_UTF8);
1512 }
1513 else if (ckWARN_d(WARN_NON_UNICODE)) {
1514 pack_warn = packWARN(WARN_NON_UNICODE);
1515 }
1516 if (pack_warn) {
1517 message = Perl_form(aTHX_ "%s: %s (overflows)",
1518 malformed_text,
05b9033b 1519 _byte_dump_string(s0, curlen, 0));
ddb65933 1520 }
2b5e7bc2
KW
1521 }
1522 }
1523 }
1524 else if (possible_problems & UTF8_GOT_EMPTY) {
1525 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1526 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1527
1528 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1529
1530 /* This so-called malformation is now treated as a bug in
1531 * the caller. If you have nothing to decode, skip calling
1532 * this function */
1533 assert(0);
1534
2b5e7bc2
KW
1535 disallowed = TRUE;
1536 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1537 pack_warn = packWARN(WARN_UTF8);
1538 message = Perl_form(aTHX_ "%s (empty string)",
1539 malformed_text);
1540 }
1541 }
1542 }
1543 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1544 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1545 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1546
1547 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1548 disallowed = TRUE;
1549 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1550 pack_warn = packWARN(WARN_UTF8);
1551 message = Perl_form(aTHX_
1552 "%s: %s (unexpected continuation byte 0x%02x,"
1553 " with no preceding start byte)",
1554 malformed_text,
7e2f38b2 1555 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1556 }
1557 }
1558 }
2b5e7bc2
KW
1559 else if (possible_problems & UTF8_GOT_SHORT) {
1560 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1561 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1562
1563 if (! (flags & UTF8_ALLOW_SHORT)) {
1564 disallowed = TRUE;
1565 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1566 pack_warn = packWARN(WARN_UTF8);
1567 message = Perl_form(aTHX_
56576a04
KW
1568 "%s: %s (too short; %d byte%s available, need %d)",
1569 malformed_text,
1570 _byte_dump_string(s0, send - s0, 0),
1571 (int)avail_len,
1572 avail_len == 1 ? "" : "s",
1573 (int)expectlen);
2b5e7bc2
KW
1574 }
1575 }
ba210ebe 1576
2b5e7bc2 1577 }
e308b348
KW
1578 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1579 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1580 *errors |= UTF8_GOT_NON_CONTINUATION;
1581
1582 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1583 disallowed = TRUE;
1584 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1585
1586 /* If we don't know for sure that the input length is
1587 * valid, avoid as much as possible reading past the
1588 * end of the buffer */
1589 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1590 ? s - s0
1591 : send - s0;
e308b348
KW
1592 pack_warn = packWARN(WARN_UTF8);
1593 message = Perl_form(aTHX_ "%s",
1594 unexpected_non_continuation_text(s0,
99a765e9 1595 printlen,
e308b348
KW
1596 s - s0,
1597 (int) expectlen));
1598 }
1599 }
1600 }
2b5e7bc2
KW
1601 else if (possible_problems & UTF8_GOT_SURROGATE) {
1602 possible_problems &= ~UTF8_GOT_SURROGATE;
1603
f9380377
KW
1604 if (flags & UTF8_WARN_SURROGATE) {
1605 *errors |= UTF8_GOT_SURROGATE;
1606
1607 if ( ! (flags & UTF8_CHECK_ONLY)
1608 && ckWARN_d(WARN_SURROGATE))
1609 {
2b5e7bc2
KW
1610 pack_warn = packWARN(WARN_SURROGATE);
1611
1612 /* These are the only errors that can occur with a
1613 * surrogate when the 'uv' isn't valid */
1614 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1615 message = Perl_form(aTHX_
1616 "UTF-16 surrogate (any UTF-8 sequence that"
1617 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1618 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1619 }
1620 else {
c94c2f39 1621 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1622 }
f9380377 1623 }
2b5e7bc2 1624 }
ba210ebe 1625
2b5e7bc2
KW
1626 if (flags & UTF8_DISALLOW_SURROGATE) {
1627 disallowed = TRUE;
f9380377 1628 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1629 }
1630 }
1631 else if (possible_problems & UTF8_GOT_SUPER) {
1632 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1633
f9380377
KW
1634 if (flags & UTF8_WARN_SUPER) {
1635 *errors |= UTF8_GOT_SUPER;
1636
1637 if ( ! (flags & UTF8_CHECK_ONLY)
1638 && ckWARN_d(WARN_NON_UNICODE))
1639 {
2b5e7bc2
KW
1640 pack_warn = packWARN(WARN_NON_UNICODE);
1641
1642 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1643 message = Perl_form(aTHX_
1644 "Any UTF-8 sequence that starts with"
1645 " \"%s\" is for a non-Unicode code point,"
1646 " may not be portable",
7e2f38b2 1647 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1648 }
1649 else {
c94c2f39 1650 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1651 }
f9380377 1652 }
2b5e7bc2 1653 }
ba210ebe 1654
57ff5f59
KW
1655 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1656 * and before possibly bailing out, so that the more dire
1657 * warning will override the regular one. */
1658 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1659 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1660 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1661 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1662 {
db0f09e6 1663 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1664
57ff5f59
KW
1665 /* If it is an overlong that evaluates to a code point
1666 * that doesn't have to use the Perl extended UTF-8, it
1667 * still used it, and so we output a message that
1668 * doesn't refer to the code point. The same is true
1669 * if there was a SHORT malformation where the code
1670 * point is not valid. In that case, 'uv' will have
1671 * been set to the REPLACEMENT CHAR, and the message
1672 * below without the code point in it will be selected
1673 * */
1674 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1675 message = Perl_form(aTHX_
57ff5f59 1676 perl_extended_cp_format, uv);
2b5e7bc2
KW
1677 }
1678 else {
1679 message = Perl_form(aTHX_
57ff5f59
KW
1680 "Any UTF-8 sequence that starts with"
1681 " \"%s\" is a Perl extension, and"
1682 " so is not portable",
1683 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1684 }
1685 }
1686
d044b7a7
KW
1687 if (flags & ( UTF8_WARN_PERL_EXTENDED
1688 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1689 {
d044b7a7 1690 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1691
d044b7a7 1692 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1693 disallowed = TRUE;
1694 }
2b5e7bc2
KW
1695 }
1696 }
eb83ed87 1697
2b5e7bc2 1698 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1699 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1700 disallowed = TRUE;
1701 }
eb83ed87 1702
2b5e7bc2
KW
1703 /* The deprecated warning overrides any non-deprecated one. If
1704 * there are other problems, a deprecation message is not
1705 * really helpful, so don't bother to raise it in that case.
1706 * This also keeps the code from having to handle the case
1707 * where 'uv' is not valid. */
1708 if ( ! (orig_problems
1709 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1710 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1711 && ckWARN_d(WARN_DEPRECATED))
1712 {
1713 message = Perl_form(aTHX_ cp_above_legal_max,
1714 uv, MAX_NON_DEPRECATED_CP);
1715 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1716 }
1717 }
1718 else if (possible_problems & UTF8_GOT_NONCHAR) {
1719 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1720
f9380377
KW
1721 if (flags & UTF8_WARN_NONCHAR) {
1722 *errors |= UTF8_GOT_NONCHAR;
1723
1724 if ( ! (flags & UTF8_CHECK_ONLY)
1725 && ckWARN_d(WARN_NONCHAR))
1726 {
2b5e7bc2
KW
1727 /* The code above should have guaranteed that we don't
1728 * get here with errors other than overlong */
1729 assert (! (orig_problems
1730 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1731
1732 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1733 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1734 }
2b5e7bc2 1735 }
5b311467 1736
2b5e7bc2
KW
1737 if (flags & UTF8_DISALLOW_NONCHAR) {
1738 disallowed = TRUE;
f9380377 1739 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1740 }
6c64cd9d
KW
1741 }
1742 else if (possible_problems & UTF8_GOT_LONG) {
1743 possible_problems &= ~UTF8_GOT_LONG;
1744 *errors |= UTF8_GOT_LONG;
1745
1746 if (flags & UTF8_ALLOW_LONG) {
1747
1748 /* We don't allow the actual overlong value, unless the
1749 * special extra bit is also set */
1750 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1751 & ~UTF8_ALLOW_LONG)))
1752 {
1753 uv = UNICODE_REPLACEMENT;
1754 }
1755 }
1756 else {
1757 disallowed = TRUE;
1758
1759 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1760 pack_warn = packWARN(WARN_UTF8);
1761
1762 /* These error types cause 'uv' to be something that
1763 * isn't what was intended, so can't use it in the
1764 * message. The other error types either can't
1765 * generate an overlong, or else the 'uv' is valid */
1766 if (orig_problems &
1767 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1768 {
1769 message = Perl_form(aTHX_
1770 "%s: %s (any UTF-8 sequence that starts"
1771 " with \"%s\" is overlong which can and"
1772 " should be represented with a"
1773 " different, shorter sequence)",
1774 malformed_text,
1775 _byte_dump_string(s0, send - s0, 0),
1776 _byte_dump_string(s0, curlen, 0));
1777 }
1778 else {
1779 U8 tmpbuf[UTF8_MAXBYTES+1];
1780 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1781 uv, 0);
1782 const char * preface = (uv <= PERL_UNICODE_MAX)
1783 ? "U+"
1784 : "0x";
1785 message = Perl_form(aTHX_
1786 "%s: %s (overlong; instead use %s to represent"
1787 " %s%0*" UVXf ")",
1788 malformed_text,
1789 _byte_dump_string(s0, send - s0, 0),
1790 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1791 preface,
1792 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1793 small code points */
1794 uv);
1795 }
1796 }
1797 }
2b5e7bc2
KW
1798 } /* End of looking through the possible flags */
1799
1800 /* Display the message (if any) for the problem being handled in
1801 * this iteration of the loop */
1802 if (message) {
1803 if (PL_op)
1804 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1805 OP_DESC(PL_op));
1806 else
1807 Perl_warner(aTHX_ pack_warn, "%s", message);
1808 }
ddb65933 1809 } /* End of 'while (possible_problems)' */
a0dbb045 1810
2b5e7bc2
KW
1811 /* Since there was a possible problem, the returned length may need to
1812 * be changed from the one stored at the beginning of this function.
1813 * Instead of trying to figure out if that's needed, just do it. */
1814 if (retlen) {
1815 *retlen = curlen;
1816 }
a0dbb045 1817
2b5e7bc2
KW
1818 if (disallowed) {
1819 if (flags & UTF8_CHECK_ONLY && retlen) {
1820 *retlen = ((STRLEN) -1);
1821 }
1822 return 0;
1823 }
eb83ed87 1824 }
ba210ebe 1825
2b5e7bc2 1826 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1827}
1828
8e84507e 1829/*
ec5f19d0
KW
1830=for apidoc utf8_to_uvchr_buf
1831
1832Returns the native code point of the first character in the string C<s> which
1833is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1834C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1835
524080c4
KW
1836If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1837enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1838C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1839(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1840C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1841the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1842See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1843returned.
ec5f19d0 1844
760c7c2f
KW
1845Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1846unless those are turned off.
1847
ec5f19d0 1848=cut
52be2536
KW
1849
1850Also implemented as a macro in utf8.h
1851
ec5f19d0
KW
1852*/
1853
1854
1855UV
1856Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1857{
7f974d7e
KW
1858 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1859
ec5f19d0
KW
1860 assert(s < send);
1861
1862 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1863 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1864}
1865
52be2536
KW
1866/* This is marked as deprecated
1867 *
ec5f19d0
KW
1868=for apidoc utf8_to_uvuni_buf
1869
de69f3af
KW
1870Only in very rare circumstances should code need to be dealing in Unicode
1871(as opposed to native) code points. In those few cases, use
1872C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1873
1874Returns the Unicode (not-native) code point of the first character in the
1875string C<s> which
ec5f19d0
KW
1876is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1877C<retlen> will be set to the length, in bytes, of that character.
1878
524080c4
KW
1879If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1880enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1881NULL) to -1. If those warnings are off, the computed value if well-defined (or
1882the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1883is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1884next possible position in C<s> that could begin a non-malformed character.
de69f3af 1885See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1886
760c7c2f
KW
1887Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1888unless those are turned off.
1889
ec5f19d0
KW
1890=cut
1891*/
1892
1893UV
1894Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1895{
1896 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1897
1898 assert(send > s);
1899
5962d97e
KW
1900 /* Call the low level routine, asking for checks */
1901 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1902}
1903
b76347f2 1904/*
87cea99e 1905=for apidoc utf8_length
b76347f2
JH
1906
1907Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1908Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1909up past C<e>, croaks.
b76347f2
JH
1910
1911=cut
1912*/
1913
1914STRLEN
35a4481c 1915Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1916{
1917 STRLEN len = 0;
1918
7918f24d
NC
1919 PERL_ARGS_ASSERT_UTF8_LENGTH;
1920
8850bf83
JH
1921 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1922 * the bitops (especially ~) can create illegal UTF-8.
1923 * In other words: in Perl UTF-8 is not just for Unicode. */
1924
a3b680e6
AL
1925 if (e < s)
1926 goto warn_and_return;
b76347f2 1927 while (s < e) {
4cbf4130 1928 s += UTF8SKIP(s);
8e91ec7f
AV
1929 len++;
1930 }
1931
1932 if (e != s) {
1933 len--;
1934 warn_and_return:
9b387841
NC
1935 if (PL_op)
1936 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1937 "%s in %s", unees, OP_DESC(PL_op));
1938 else
61a12c31 1939 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1940 }
1941
1942 return len;
1943}
1944
b06226ff 1945/*
fed3ba5d
NC
1946=for apidoc bytes_cmp_utf8
1947
a1433954 1948Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1949sequence of characters (stored as UTF-8)
1950in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1951equal, -1 or -2 if the first string is less than the second string, +1 or +2
1952if the first string is greater than the second string.
1953
1954-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1955longer string. -2 or +2 is returned if
1956there was a difference between characters
fed3ba5d
NC
1957within the strings.
1958
1959=cut
1960*/
1961
1962int
1963Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1964{
1965 const U8 *const bend = b + blen;
1966 const U8 *const uend = u + ulen;
1967
1968 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1969
1970 while (b < bend && u < uend) {
1971 U8 c = *u++;
1972 if (!UTF8_IS_INVARIANT(c)) {
1973 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1974 if (u < uend) {
1975 U8 c1 = *u++;
1976 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1977 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1978 } else {
2b5e7bc2 1979 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1980 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1981 "%s %s%s",
1982 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1983 PL_op ? " in " : "",
1984 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1985 return -2;
1986 }
1987 } else {
1988 if (PL_op)
1989 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1990 "%s in %s", unees, OP_DESC(PL_op));
1991 else
61a12c31 1992 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1993 return -2; /* Really want to return undef :-) */
1994 }
1995 } else {
1996 return -2;
1997 }
1998 }
1999 if (*b != c) {
2000 return *b < c ? -2 : +2;
2001 }
2002 ++b;
2003 }
2004
2005 if (b == bend && u == uend)
2006 return 0;
2007
2008 return b < bend ? +1 : -1;
2009}
2010
2011/*
87cea99e 2012=for apidoc utf8_to_bytes
6940069f 2013
3bc0c78c 2014Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2015Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2016updates C<*lenp> to contain the new length.
3bc0c78c
KW
2017Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2018
2019Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2020having saved the value of C<*lenp> before the call, and subtracting the
2021after-call value of C<*lenp> from it.
6940069f 2022
a1433954 2023If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2024
6940069f
GS
2025=cut
2026*/
2027
2028U8 *
09af0336 2029Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2030{
9fe0d3c2 2031 U8 * first_variant;
246fae53 2032
7918f24d 2033 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2034 PERL_UNUSED_CONTEXT;
7918f24d 2035
9fe0d3c2 2036 /* This is a no-op if no variants at all in the input */
09af0336 2037 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2038 return s;
2039 }
2040
2041 {
3c5aa262 2042 U8 * const save = s;
09af0336 2043 U8 * const send = s + *lenp;
3c5aa262
KW
2044 U8 * d;
2045
2046 /* Nothing before the first variant needs to be changed, so start the real
2047 * work there */
2048 s = first_variant;
2049 while (s < send) {
2050 if (! UTF8_IS_INVARIANT(*s)) {
2051 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2052 *lenp = ((STRLEN) -1);
3c5aa262
KW
2053 return 0;
2054 }
2055 s++;
d59937ca
KW
2056 }
2057 s++;
dcad2880 2058 }
dcad2880 2059
3c5aa262
KW
2060 /* Is downgradable, so do it */
2061 d = s = first_variant;
2062 while (s < send) {
2063 U8 c = *s++;
2064 if (! UVCHR_IS_INVARIANT(c)) {
2065 /* Then it is two-byte encoded */
2066 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2067 s++;
2068 }
2069 *d++ = c;
2070 }
2071 *d = '\0';
09af0336 2072 *lenp = d - save;
3c5aa262
KW
2073
2074 return save;
9fe0d3c2 2075 }
6940069f
GS
2076}
2077
2078/*
87cea99e 2079=for apidoc bytes_from_utf8
f9a63242 2080
09af0336 2081Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2082byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2083actually encoded in UTF-8.
2084
2085Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2086the input string.
2087
41ae6089
KW
2088Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2089not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2090C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2091
41ae6089 2092Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2093newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2094returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2095
3bc0c78c 2096Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2097having saved the value of C<*lenp> before the call, and subtracting the
2098after-call value of C<*lenp> from it.
3bc0c78c 2099
37607a96 2100=cut
976c1b08
KW
2101
2102There is a macro that avoids this function call, but this is retained for
2103anyone who calls it with the Perl_ prefix */
f9a63242
JH
2104
2105U8 *
41ae6089 2106Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2107{
7918f24d 2108 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2109 PERL_UNUSED_CONTEXT;
f9a63242 2110
976c1b08
KW
2111 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2112}
2113
2114/*
2115No = here because currently externally undocumented
2116for apidoc bytes_from_utf8_loc
2117
2118Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2119to store the location of the first character in C<"s"> that cannot be
2120converted to non-UTF8.
2121
2122If that parameter is C<NULL>, this function behaves identically to
2123C<bytes_from_utf8>.
2124
2125Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2126C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2127
2128Otherwise, the function returns a newly created C<NUL>-terminated string
2129containing the non-UTF8 equivalent of the convertible first portion of
2130C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2131If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2132and C<*first_non_downgradable> is set to C<NULL>.
2133
2134Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2135first character in the original string that wasn't converted. C<*is_utf8p> is
2136unchanged. Note that the new string may have length 0.
2137
2138Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2139C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2140converts as many characters in it as possible stopping at the first one it
385b74be 2141finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2142set to point to that. The function returns the portion that could be converted
2143in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2144not including the terminating C<NUL>. If the very first character in the
2145original could not be converted, C<*lenp> will be 0, and the new string will
2146contain just a single C<NUL>. If the entire input string was converted,
2147C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2148
2149Upon successful return, the number of variants in the converted portion of the
2150string can be computed by having saved the value of C<*lenp> before the call,
2151and subtracting the after-call value of C<*lenp> from it.
2152
2153=cut
2154
2155
2156*/
2157
2158U8 *
2159Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2160{
2161 U8 *d;
2162 const U8 *original = s;
2163 U8 *converted_start;
2164 const U8 *send = s + *lenp;
f9a63242 2165
976c1b08 2166 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2167
976c1b08
KW
2168 if (! *is_utf8p) {
2169 if (first_unconverted) {
2170 *first_unconverted = NULL;
2171 }
2172
2173 return (U8 *) original;
2174 }
2175
2176 Newx(d, (*lenp) + 1, U8);
2177
2178 converted_start = d;
7299a045
KW
2179 while (s < send) {
2180 U8 c = *s++;
2181 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2182
2183 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2184 * have to stop now */
2185 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2186 if (first_unconverted) {
2187 *first_unconverted = s - 1;
2188 goto finish_and_return;
2189 }
2190 else {
2191 Safefree(converted_start);
2192 return (U8 *) original;
2193 }
2194 }
2195
7299a045
KW
2196 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2197 s++;
38af28cf 2198 }
7299a045
KW
2199 *d++ = c;
2200 }
170a1c22 2201
976c1b08
KW
2202 /* Here, converted the whole of the input */
2203 *is_utf8p = FALSE;
2204 if (first_unconverted) {
2205 *first_unconverted = NULL;
170a1c22 2206 }
976c1b08
KW
2207
2208 finish_and_return:
2209 *d = '\0';
2210 *lenp = d - converted_start;
2211
2212 /* Trim unused space */
2213 Renew(converted_start, *lenp + 1, U8);
2214
2215 return converted_start;
f9a63242
JH
2216}
2217
2218/*
87cea99e 2219=for apidoc bytes_to_utf8
6940069f 2220
09af0336 2221Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2222UTF-8.
09af0336 2223Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2224reflect the new length in bytes.
6940069f 2225
3bc0c78c 2226Upon successful return, the number of variants in the string can be computed by
23b37b12 2227having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2228after-call value of C<*lenp>.
2229
75200dff 2230A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2231
2232If you want to convert to UTF-8 from encodings other than
2233the native (Latin1 or EBCDIC),
a1433954 2234see L</sv_recode_to_utf8>().
c9ada85f 2235
497711e7 2236=cut
6940069f
GS
2237*/
2238
2239U8*
09af0336 2240Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2241{
09af0336 2242 const U8 * const send = s + (*lenp);
6940069f
GS
2243 U8 *d;
2244 U8 *dst;
7918f24d
NC
2245
2246 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2247 PERL_UNUSED_CONTEXT;
6940069f 2248
09af0336 2249 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2250 dst = d;
2251
2252 while (s < send) {
55d09dc8
KW
2253 append_utf8_from_native_byte(*s, &d);
2254 s++;
6940069f
GS
2255 }
2256 *d = '\0';
09af0336 2257 *lenp = d-dst;
6940069f
GS
2258 return dst;
2259}
2260
a0ed51b3 2261/*
dea0fc0b 2262 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2263 *
2264 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2265 * We optimize for native, for obvious reasons. */
2266
2267U8*
dea0fc0b 2268Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2269{
dea0fc0b
JH
2270 U8* pend;
2271 U8* dstart = d;
2272
7918f24d
NC
2273 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2274
dea0fc0b 2275 if (bytelen & 1)
56576a04
KW
2276 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2277 (UV)bytelen);
dea0fc0b
JH
2278
2279 pend = p + bytelen;
2280
a0ed51b3 2281 while (p < pend) {
dea0fc0b
JH
2282 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2283 p += 2;
2d1545e5 2284 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2285 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2286 continue;
2287 }
56d37426
KW
2288 if (uv <= MAX_UTF8_TWO_BYTE) {
2289 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2290 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2291 continue;
2292 }
46956fad
KW
2293#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2294#define LAST_HIGH_SURROGATE 0xDBFF
2295#define FIRST_LOW_SURROGATE 0xDC00
2296#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2297
2298 /* This assumes that most uses will be in the first Unicode plane, not
2299 * needing surrogates */
2300 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2301 && uv <= UNICODE_SURROGATE_LAST))
2302 {
2303 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2304 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2305 }
2306 else {
01ea242b 2307 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2308 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2309 || UNLIKELY(low > LAST_LOW_SURROGATE))
2310 {
01ea242b 2311 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2312 }
2313 p += 2;
46956fad
KW
2314 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2315 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2316 }
a0ed51b3 2317 }
56d37426
KW
2318#ifdef EBCDIC
2319 d = uvoffuni_to_utf8_flags(d, uv, 0);
2320#else
a0ed51b3 2321 if (uv < 0x10000) {
eb160463
GS
2322 *d++ = (U8)(( uv >> 12) | 0xe0);
2323 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2324 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2325 continue;
2326 }
2327 else {
eb160463
GS
2328 *d++ = (U8)(( uv >> 18) | 0xf0);
2329 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2330 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2331 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2332 continue;
2333 }
56d37426 2334#endif
a0ed51b3 2335 }
dea0fc0b 2336 *newlen = d - dstart;
a0ed51b3
LW
2337 return d;
2338}
2339
2340/* Note: this one is slightly destructive of the source. */
2341
2342U8*
dea0fc0b 2343Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2344{
2345 U8* s = (U8*)p;
d4c19fe8 2346 U8* const send = s + bytelen;
7918f24d
NC
2347
2348 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2349
e0ea5e2d 2350 if (bytelen & 1)
147e3846 2351 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2352 (UV)bytelen);
2353
a0ed51b3 2354 while (s < send) {
d4c19fe8 2355 const U8 tmp = s[0];
a0ed51b3
LW
2356 s[0] = s[1];
2357 s[1] = tmp;
2358 s += 2;
2359 }
dea0fc0b 2360 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2361}
2362
922e8cb4
KW
2363bool
2364Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2365{
2366 U8 tmpbuf[UTF8_MAXBYTES+1];
2367 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2368 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2369}
2370
f9ae8fb6
JD
2371/* Internal function so we can deprecate the external one, and call
2372 this one from other deprecated functions in this file */
2373
f2645549
KW
2374bool
2375Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2376{
f2645549 2377 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2378
2379 if (*p == '_')
2380 return TRUE;
f25ce844 2381 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2382}
2383
5092f92a 2384bool
eba68aa0
KW
2385Perl__is_uni_perl_idcont(pTHX_ UV c)
2386{
2387 U8 tmpbuf[UTF8_MAXBYTES+1];
2388 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2389 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2390}
2391
2392bool
f91dcd13
KW
2393Perl__is_uni_perl_idstart(pTHX_ UV c)
2394{
2395 U8 tmpbuf[UTF8_MAXBYTES+1];
2396 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2397 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2398}
2399
3a4c58c9 2400UV
56576a04
KW
2401Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2402 const char S_or_s)
3a4c58c9
KW
2403{
2404 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2405 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2406 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2407 * either "SS" or "Ss". Which one to use is passed into the routine in
2408 * 'S_or_s' to avoid a test */
2409
2410 UV converted = toUPPER_LATIN1_MOD(c);
2411
2412 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2413
2414 assert(S_or_s == 'S' || S_or_s == 's');
2415
6f2d5cbc 2416 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2417 characters in this range */
3a4c58c9
KW
2418 *p = (U8) converted;
2419 *lenp = 1;
2420 return converted;
2421 }
2422
2423 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2424 * which it maps to one of them, so as to only have to have one check for
2425 * it in the main case */
2426 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2427 switch (c) {
2428 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2429 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2430 break;
2431 case MICRO_SIGN:
2432 converted = GREEK_CAPITAL_LETTER_MU;
2433 break;
79e064b9
KW
2434#if UNICODE_MAJOR_VERSION > 2 \
2435 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2436 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2437 case LATIN_SMALL_LETTER_SHARP_S:
2438 *(p)++ = 'S';
2439 *p = S_or_s;
2440 *lenp = 2;
2441 return 'S';
79e064b9 2442#endif
3a4c58c9 2443 default:
56576a04
KW
2444 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2445 " '%c' to map to '%c'",
2446 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2447 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2448 }
2449 }
2450
2451 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2452 *p = UTF8_TWO_BYTE_LO(converted);
2453 *lenp = 2;
2454
2455 return converted;
2456}
2457
50bda2c3
KW
2458/* Call the function to convert a UTF-8 encoded character to the specified case.
2459 * Note that there may be more than one character in the result.
2460 * INP is a pointer to the first byte of the input character
2461 * OUTP will be set to the first byte of the string of changed characters. It
2462 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2463 * LENP will be set to the length in bytes of the string of changed characters
2464 *
56576a04
KW
2465 * The functions return the ordinal of the first character in the string of
2466 * OUTP */
2467#define CALL_UPPER_CASE(uv, s, d, lenp) \
2468 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2469#define CALL_TITLE_CASE(uv, s, d, lenp) \
2470 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2471#define CALL_LOWER_CASE(uv, s, d, lenp) \
2472 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2473
b9992569
KW
2474/* This additionally has the input parameter 'specials', which if non-zero will
2475 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2476 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2477#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2478_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2479
84afefe6
JH
2480UV
2481Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2482{
a1433954
KW
2483 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2484 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2485 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2486 * the changed version may be longer than the original character.
2487 *
2488 * The ordinal of the first character of the changed version is returned
2489 * (but note, as explained above, that there may be more.) */
2490
7918f24d
NC
2491 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2492
3a4c58c9
KW
2493 if (c < 256) {
2494 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2495 }
2496
0ebc6274 2497 uvchr_to_utf8(p, c);
b9992569 2498 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2499}
2500
84afefe6
JH
2501UV
2502Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2503{
7918f24d
NC
2504 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2505
3a4c58c9
KW
2506 if (c < 256) {
2507 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2508 }
2509
0ebc6274 2510 uvchr_to_utf8(p, c);
b9992569 2511 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2512}
2513
afc16117 2514STATIC U8
eaf412bf 2515S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2516{
2517 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2518 * those, converting the result to UTF-8. Since the result is always just
a1433954 2519 * one character, we allow <p> to be NULL */
afc16117
KW
2520
2521 U8 converted = toLOWER_LATIN1(c);
2522
eaf412bf
KW
2523 PERL_UNUSED_ARG(dummy);
2524
afc16117 2525 if (p != NULL) {
6f2d5cbc 2526 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2527 *p = converted;
2528 *lenp = 1;
2529 }
2530 else {
430c9760
KW
2531 /* Result is known to always be < 256, so can use the EIGHT_BIT
2532 * macros */
2533 *p = UTF8_EIGHT_BIT_HI(converted);
2534 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2535 *lenp = 2;
2536 }
2537 }
2538 return converted;
2539}
2540
84afefe6
JH
2541UV
2542Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2543{
7918f24d
NC
2544 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2545
afc16117 2546 if (c < 256) {
eaf412bf 2547 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2548 }
2549
afc16117 2550 uvchr_to_utf8(p, c);
b9992569 2551 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2552}
2553
84afefe6 2554UV
56576a04
KW
2555Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2556 const unsigned int flags)
a1dde8de 2557{
51910141 2558 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2559 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2560 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2561 *
2562 * Not to be used for locale folds
51910141 2563 */
f673fad4 2564
a1dde8de
KW
2565 UV converted;
2566
2567 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2568 PERL_UNUSED_CONTEXT;
a1dde8de 2569
1ca267a5
KW
2570 assert (! (flags & FOLD_FLAGS_LOCALE));
2571
659a7c2d 2572 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2573 converted = GREEK_SMALL_LETTER_MU;
2574 }
9b63e895
KW
2575#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2576 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2577 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2578 else if ( (flags & FOLD_FLAGS_FULL)
2579 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2580 {
1ca267a5
KW
2581 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2582 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2583 * under those circumstances. */
2584 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2585 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2586 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2587 p, *lenp, U8);
2588 return LATIN_SMALL_LETTER_LONG_S;
2589 }
2590 else {
4f489194
KW
2591 *(p)++ = 's';
2592 *p = 's';
2593 *lenp = 2;
2594 return 's';
1ca267a5 2595 }
a1dde8de 2596 }
9b63e895 2597#endif
a1dde8de
KW
2598 else { /* In this range the fold of all other characters is their lower
2599 case */
2600 converted = toLOWER_LATIN1(c);
2601 }
2602
6f2d5cbc 2603 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2604 *p = (U8) converted;
2605 *lenp = 1;
2606 }
2607 else {
2608 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2609 *p = UTF8_TWO_BYTE_LO(converted);
2610 *lenp = 2;
2611 }
2612
2613 return converted;
2614}
2615
2616UV
31f05a37 2617Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2618{
4b593389 2619
a0270393
KW
2620 /* Not currently externally documented, and subject to change
2621 * <flags> bits meanings:
2622 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2623 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2624 * locale are to be used.
a0270393
KW
2625 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2626 */
4b593389 2627
36bb2ab6 2628 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2629
780fcc9f
KW
2630 if (flags & FOLD_FLAGS_LOCALE) {
2631 /* Treat a UTF-8 locale as not being in locale at all */
2632 if (IN_UTF8_CTYPE_LOCALE) {
2633 flags &= ~FOLD_FLAGS_LOCALE;
2634 }
2635 else {
2636 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2637 goto needs_full_generality;
780fcc9f 2638 }
31f05a37
KW
2639 }
2640
a1dde8de 2641 if (c < 256) {
e7b7ac46 2642 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2643 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2644 }
2645
2f306ab9 2646 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2647 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2648 uvchr_to_utf8(p, c);
b9992569 2649 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2650 }
567b353c 2651 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2652 the special flags. */
2653 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2654
2655 needs_full_generality:
a0270393 2656 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2657 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2658 p, lenp, flags);
a0270393 2659 }
84afefe6
JH
2660}
2661
26483009 2662PERL_STATIC_INLINE bool
5141f98e 2663S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2664 const char *const swashname, SV* const invlist)
bde6a22d 2665{
ea317ccb
KW
2666 /* returns a boolean giving whether or not the UTF8-encoded character that
2667 * starts at <p> is in the swash indicated by <swashname>. <swash>
2668 * contains a pointer to where the swash indicated by <swashname>
2669 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2670 * look at <*swash> and only generate a swash if it is not null. <invlist>
2671 * is NULL or an inversion list that defines the swash. If not null, it
2672 * saves time during initialization of the swash.
ea317ccb
KW
2673 *
2674 * Note that it is assumed that the buffer length of <p> is enough to
2675 * contain all the bytes that comprise the character. Thus, <*p> should
2676 * have been checked before this call for mal-formedness enough to assure
2677 * that. */
2678
7918f24d
NC
2679 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2680
492a624f 2681 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2682 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2683 * as far as there being enough bytes available in it to accommodate the
2684 * character without reading beyond the end, and pass that number on to the
2685 * validating routine */
6302f837 2686 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2687 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2688 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2689 1 /* Die */ );
2690 NOT_REACHED; /* NOTREACHED */
28123549 2691 }
86ae6e94 2692
87367d5f
KW
2693 if (!*swash) {
2694 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2695 *swash = _core_swash_init("utf8",
2696
2697 /* Only use the name if there is no inversion
2698 * list; otherwise will go out to disk */
2699 (invlist) ? "" : swashname,
2700
2701 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2702 }
28123549 2703
bde6a22d
NC
2704 return swash_fetch(*swash, p, TRUE) != 0;
2705}
2706
da8c1a98 2707PERL_STATIC_INLINE bool
56576a04
KW
2708S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2709 SV **swash, const char *const swashname,
2710 SV* const invlist)
da8c1a98
KW
2711{
2712 /* returns a boolean giving whether or not the UTF8-encoded character that
2713 * starts at <p>, and extending no further than <e - 1> is in the swash
2714 * indicated by <swashname>. <swash> contains a pointer to where the swash
2715 * indicated by <swashname> is to be stored; which this routine will do, so
2716 * that future calls will look at <*swash> and only generate a swash if it
2717 * is not null. <invlist> is NULL or an inversion list that defines the
2718 * swash. If not null, it saves time during initialization of the swash.
2719 */
2720
2721 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2722
2723 if (! isUTF8_CHAR(p, e)) {
2724 _force_out_malformed_utf8_message(p, e, 0, 1);
2725 NOT_REACHED; /* NOTREACHED */
2726 }
2727
2728 if (!*swash) {
2729 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2730 *swash = _core_swash_init("utf8",
2731
2732 /* Only use the name if there is no inversion
2733 * list; otherwise will go out to disk */
2734 (invlist) ? "" : swashname,
2735
2736 &PL_sv_undef, 1, 0, invlist, &flags);
2737 }
2738
2739 return swash_fetch(*swash, p, TRUE) != 0;
2740}
2741
34aeb2e9
KW
2742STATIC void
2743S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2744 const char * const alternative,
2745 const bool use_locale,
2746 const char * const file,
2747 const unsigned line)
2748{
2749 const char * key;
2750
2751 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2752
2753 if (ckWARN_d(WARN_DEPRECATED)) {
2754
2755 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2756 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2757 if (! PL_seen_deprecated_macro) {
2758 PL_seen_deprecated_macro = newHV();
2759 }
2760 if (! hv_store(PL_seen_deprecated_macro, key,
2761 strlen(key), &PL_sv_undef, 0))
2762 {
2763 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2764 }
2765
c44e9413 2766 if (instr(file, "mathoms.c")) {
607313a1
KW
2767 Perl_warner(aTHX_ WARN_DEPRECATED,
2768 "In %s, line %d, starting in Perl v5.30, %s()"
2769 " will be removed. Avoid this message by"
2770 " converting to use %s().\n",
2771 file, line, name, alternative);
2772 }
2773 else {
34aeb2e9
KW
2774 Perl_warner(aTHX_ WARN_DEPRECATED,
2775 "In %s, line %d, starting in Perl v5.30, %s() will"
2776 " require an additional parameter. Avoid this"
2777 " message by converting to use %s().\n",
2778 file, line, name, alternative);
607313a1 2779 }
34aeb2e9
KW
2780 }
2781 }
2782}
2783
bde6a22d 2784bool
34aeb2e9 2785Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2786 const U8 * const p,
34aeb2e9
KW
2787 const char * const name,
2788 const char * const alternative,
2789 const bool use_utf8,
2790 const bool use_locale,
2791 const char * const file,
2792 const unsigned line)
922e8cb4 2793{
922e8cb4
KW
2794 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2795
34aeb2e9
KW
2796 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2797
2798 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2799
2800 switch (classnum) {
2801 case _CC_WORDCHAR:
2802 case _CC_DIGIT:
2803 case _CC_ALPHA:
2804 case _CC_LOWER:
2805 case _CC_UPPER:
2806 case _CC_PUNCT:
2807 case _CC_PRINT:
2808 case _CC_ALPHANUMERIC:
2809 case _CC_GRAPH:
2810 case _CC_CASED:
2811
2812 return is_utf8_common(p,
2813 &PL_utf8_swash_ptrs[classnum],
2814 swash_property_names[classnum],
2815 PL_XPosix_ptrs[classnum]);
2816
2817 case _CC_SPACE:
2818 return is_XPERLSPACE_high(p);
2819 case _CC_BLANK:
2820 return is_HORIZWS_high(p);
2821 case _CC_XDIGIT:
2822 return is_XDIGIT_high(p);
2823 case _CC_CNTRL:
2824 return 0;
2825 case _CC_ASCII:
2826 return 0;
2827 case _CC_VERTSPACE:
2828 return is_VERTWS_high(p);
2829 case _CC_IDFIRST:
2830 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2831 PL_utf8_perl_idstart
2832 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2833 }
22f0498f
KW
2834 return is_utf8_common(p, &PL_utf8_perl_idstart,
2835 "_Perl_IDStart", NULL);
34aeb2e9
KW
2836 case _CC_IDCONT:
2837 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2838 PL_utf8_perl_idcont
2839 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2840 }
22f0498f
KW
2841 return is_utf8_common(p, &PL_utf8_perl_idcont,
2842 "_Perl_IDCont", NULL);
34aeb2e9
KW
2843 }
2844 }
2845
2846 /* idcont is the same as wordchar below 256 */
2847 if (classnum == _CC_IDCONT) {
2848 classnum = _CC_WORDCHAR;
2849 }
2850 else if (classnum == _CC_IDFIRST) {
2851 if (*p == '_') {
2852 return TRUE;
2853 }
2854 classnum = _CC_ALPHA;
2855 }
2856
2857 if (! use_locale) {
2858 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2859 return _generic_isCC(*p, classnum);
2860 }
922e8cb4 2861
34aeb2e9
KW
2862 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2863 }
2864 else {
2865 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2866 return isFOO_lc(classnum, *p);
2867 }
2868
2869 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2870 }
2871
2872 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2873}
2874
2875bool
da8c1a98
KW
2876Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2877 const U8 * const e)
2878{
2879 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2880
2881 assert(classnum < _FIRST_NON_SWASH_CC);
2882
2883 return is_utf8_common_with_len(p,
2884 e,
2885 &PL_utf8_swash_ptrs[classnum],
2886 swash_property_names[classnum],
2887 PL_XPosix_ptrs[classnum]);
2888}
2889
2890bool
da8c1a98
KW
2891Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2892{
2893 SV* invlist = NULL;
2894
2895 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2896
2897 if (! PL_utf8_perl_idstart) {
2898 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2899 }
2900 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2901 "_Perl_IDStart", invlist);
2902}
2903
2904bool
f2645549 2905Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2906{
f2645549 2907 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2908
2909 if (*p == '_')
2910 return TRUE;
f25ce844 2911 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2912}
2913
2914bool
da8c1a98
KW
2915Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2916{
2917 SV* invlist = NULL;
2918
2919 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2920
2921 if (! PL_utf8_perl_idcont) {
2922 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2923 }
2924 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2925 "_Perl_IDCont", invlist);
2926}
2927
2928bool
f2645549 2929Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2930{
f2645549 2931 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2932
f25ce844 2933 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2934}
2935
2936bool
f2645549 2937Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2938{
f2645549 2939 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2940
f25ce844 2941 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2942}
2943
2944bool
7dbf68d2
KW
2945Perl__is_utf8_mark(pTHX_ const U8 *p)
2946{
7dbf68d2
KW
2947 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2948
f25ce844 2949 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2950}
2951
b9992569 2952 /* change namve uv1 to 'from' */
6a4a25f4 2953STATIC UV
b9992569
KW
2954S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2955 SV **swashp, const char *normal, const char *special)
2956{
0134edef 2957 STRLEN len = 0;
7918f24d 2958
b9992569 2959 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2960
36eaa811
KW
2961 /* For code points that don't change case, we already know that the output
2962 * of this function is the unchanged input, so we can skip doing look-ups
2963 * for them. Unfortunately the case-changing code points are scattered
2964 * around. But there are some long consecutive ranges where there are no
2965 * case changing code points. By adding tests, we can eliminate the lookup
2966 * for all the ones in such ranges. This is currently done here only for
2967 * just a few cases where the scripts are in common use in modern commerce
2968 * (and scripts adjacent to those which can be included without additional
2969 * tests). */
2970
2971 if (uv1 >= 0x0590) {
2972 /* This keeps from needing further processing the code points most
2973 * likely to be used in the following non-cased scripts: Hebrew,
2974 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2975 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2976 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2977 if (uv1 < 0x10A0) {
2978 goto cases_to_self;
2979 }
2980
2981 /* The following largish code point ranges also don't have case
2982 * changes, but khw didn't think they warranted extra tests to speed
2983 * them up (which would slightly slow down everything else above them):
2984 * 1100..139F Hangul Jamo, Ethiopic
2985 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2986 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2987 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2988 * Combining Diacritical Marks Extended, Balinese,
2989 * Sundanese, Batak, Lepcha, Ol Chiki
2990 * 2000..206F General Punctuation
2991 */
2992
2993 if (uv1 >= 0x2D30) {
2994
2995 /* This keeps the from needing further processing the code points
2996 * most likely to be used in the following non-cased major scripts:
2997 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2998 *
2999 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3000 * event that Unicode eventually allocates the unused block as of
3001 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3002 * that the test suite will start having failures to alert you
3003 * should that happen) */
3004 if (uv1 < 0xA640) {
3005 goto cases_to_self;
3006 }
3007
3008 if (uv1 >= 0xAC00) {
3009 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3010 if (ckWARN_d(WARN_SURROGATE)) {
3011 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3012 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3013 "Operation \"%s\" returns its argument for"
3014 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3015 }
3016 goto cases_to_self;
3017 }
36eaa811
KW
3018
3019 /* AC00..FAFF Catches Hangul syllables and private use, plus
3020 * some others */
3021 if (uv1 < 0xFB00) {
3022 goto cases_to_self;
3023
3024 }
3025
5af9bc97 3026 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
3027 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
3028 && ckWARN_d(WARN_DEPRECATED))
3029 {
3030 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
3031 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
3032 }
3033 if (ckWARN_d(WARN_NON_UNICODE)) {
3034 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3035 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3036 "Operation \"%s\" returns its argument for"
3037 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3038 }
3039 goto cases_to_self;
3040 }
3bfc1e70
KW
3041#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3042 if (UNLIKELY(uv1
3043 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3044 {
3045
56576a04
KW
3046 /* As of Unicode 10.0, this means we avoid swash creation
3047 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3048 goto cases_to_self;
3049 }
3050#endif
36eaa811
KW
3051 }
3052 }
9ae3ac1a 3053
36eaa811
KW
3054 /* Note that non-characters are perfectly legal, so no warning should
3055 * be given. There are so few of them, that it isn't worth the extra
3056 * tests to avoid swash creation */
9ae3ac1a
KW
3057 }
3058
0134edef 3059 if (!*swashp) /* load on-demand */
56576a04
KW
3060 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3061 4, 0, NULL, NULL);
0134edef 3062
a6f87d8c 3063 if (special) {
0134edef 3064 /* It might be "special" (sometimes, but not always,
2a37f04d 3065 * a multicharacter mapping) */
4a8240a3 3066 HV *hv = NULL;
b08cf34e
JH
3067 SV **svp;
3068
4a8240a3
KW
3069 /* If passed in the specials name, use that; otherwise use any
3070 * given in the swash */
3071 if (*special != '\0') {
3072 hv = get_hv(special, 0);
3073 }
3074 else {
3075 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3076 if (svp) {
3077 hv = MUTABLE_HV(SvRV(*svp));
3078 }
3079 }
3080
176fe009 3081 if (hv
5f560d8a 3082 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3083 && (*svp))
3084 {
cfd0369c 3085 const char *s;
47654450 3086
cfd0369c 3087 s = SvPV_const(*svp, len);
47654450 3088 if (len == 1)
f4cd282c 3089 /* EIGHTBIT */
c80e42f3 3090 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3091 else {
d2dcd0fb 3092 Copy(s, ustrp, len, U8);
29e98929 3093 }
983ffd37 3094 }
0134edef
JH
3095 }
3096
3097 if (!len && *swashp) {
4a4088c4 3098 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3099
0134edef
JH
3100 if (uv2) {
3101 /* It was "normal" (a single character mapping). */
f4cd282c 3102 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3103 }
3104 }
1feea2c7 3105
cbe07460
KW
3106 if (len) {
3107 if (lenp) {
3108 *lenp = len;
3109 }
3110 return valid_utf8_to_uvchr(ustrp, 0);
3111 }
3112
3113 /* Here, there was no mapping defined, which means that the code point maps
3114 * to itself. Return the inputs */
e24dfe9c 3115 cases_to_self:
bfdf22ec 3116 len = UTF8SKIP(p);
ca9fab46
KW
3117 if (p != ustrp) { /* Don't copy onto itself */
3118 Copy(p, ustrp, len, U8);
3119 }
0134edef 3120
2a37f04d
JH
3121 if (lenp)
3122 *lenp = len;
3123
f4cd282c 3124 return uv1;
cbe07460 3125
a0ed51b3
LW
3126}
3127
051a06d4 3128STATIC UV
56576a04
KW
3129S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3130 U8* const ustrp, STRLEN *lenp)
051a06d4 3131{
4a4088c4 3132 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3133 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3134 * result contains a character that crosses the 255/256 boundary, disallow
3135 * the change, and return the original code point. See L<perlfunc/lc> for
3136 * why;
051a06d4 3137 *
a1433954
KW
3138 * p points to the original string whose case was changed; assumed
3139 * by this routine to be well-formed
051a06d4 3140 * result the code point of the first character in the changed-case string
56576a04
KW
3141 * ustrp points to the changed-case string (<result> represents its
3142 * first char)
051a06d4
KW
3143 * lenp points to the length of <ustrp> */
3144
3145 UV original; /* To store the first code point of <p> */
3146
3147 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3148
a4f12ed7 3149 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3150
3151 /* We know immediately if the first character in the string crosses the
3152 * boundary, so can skip */
3153 if (result > 255) {
3154
3155 /* Look at every character in the result; if any cross the
3156 * boundary, the whole thing is disallowed */
3157 U8* s = ustrp + UTF8SKIP(ustrp);
3158 U8* e = ustrp + *lenp;
3159 while (s < e) {
a4f12ed7 3160 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3161 goto bad_crossing;
3162 }
3163 s += UTF8SKIP(s);
3164 }
3165
613abc6d
KW
3166 /* Here, no characters crossed, result is ok as-is, but we warn. */
3167 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3168 return result;
3169 }
3170
7b52d656 3171 bad_crossing:
051a06d4
KW
3172
3173 /* Failed, have to return the original */
4b88fb76 3174 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3175
3176 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3177 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3178 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3179 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3180 OP_DESC(PL_op),
ab0b796c
KW
3181 original,
3182 original);
051a06d4
KW
3183 Copy(p, ustrp, *lenp, char);
3184 return original;
3185}
3186
607313a1
KW
3187STATIC U32
3188S_check_and_deprecate(pTHX_ const U8 *p,
3189 const U8 **e,
3190 const unsigned int type, /* See below */
3191 const bool use_locale, /* Is this a 'LC_'
3192 macro call? */
3193 const char * const file,
3194 const unsigned line)
3195{
3196 /* This is a temporary function to deprecate the unsafe calls to the case
3197 * changing macros and functions. It keeps all the special stuff in just
3198 * one place.
3199 *
3200 * It updates *e with the pointer to the end of the input string. If using
3201 * the old-style macros, *e is NULL on input, and so this function assumes
3202 * the input string is long enough to hold the entire UTF-8 sequence, and
3203 * sets *e accordingly, but it then returns a flag to pass the
3204 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3205 * using the full length if possible.
3206 *
3207 * It also does the assert that *e > p when *e is not NULL. This should be
3208 * migrated to the callers when this function gets deleted.
3209 *
3210 * The 'type' parameter is used for the caller to specify which case
3211 * changing function this is called from: */
3212
3213# define DEPRECATE_TO_UPPER 0
3214# define DEPRECATE_TO_TITLE 1
3215# define DEPRECATE_TO_LOWER 2
3216# define DEPRECATE_TO_FOLD 3
3217
3218 U32 utf8n_flags = 0;
3219 const char * name;
3220 const char * alternative;
3221
3222 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3223
3224 if (*e == NULL) {
3225 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3226 *e = p + UTF8SKIP(p);
3227
3228 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3229 * there. It could be part of a larger path */
607313a1 3230 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3231 name = instr(file, "mathoms.c")
607313a1
KW
3232 ? "to_utf8_upper"
3233 : "toUPPER_utf8";
3234 alternative = "toUPPER_utf8_safe";
3235 }
3236 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3237 name = instr(file, "mathoms.c")
607313a1
KW
3238 ? "to_utf8_title"
3239 : "toTITLE_utf8";
3240 alternative = "toTITLE_utf8_safe";
3241 }
3242 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3243 name = instr(file, "mathoms.c")
607313a1
KW
3244 ? "to_utf8_lower"
3245 : "toLOWER_utf8";
3246 alternative = "toLOWER_utf8_safe";
3247 }
3248 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3249 name = instr(file, "mathoms.c")
607313a1
KW
3250 ? "to_utf8_fold"
3251 : "toFOLD_utf8";
3252 alternative = "toFOLD_utf8_safe";
3253 }
3254 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3255
3256 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3257 }
3258 else {
3259 assert (p < *e);
3260 }
3261
3262 return utf8n_flags;
3263}
3264
eaf412bf
KW
3265/* The process for changing the case is essentially the same for the four case
3266 * change types, except there are complications for folding. Otherwise the
3267 * difference is only which case to change to. To make sure that they all do
3268 * the same thing, the bodies of the functions are extracted out into the
3269 * following two macros. The functions are written with the same variable
3270 * names, and these are known and used inside these macros. It would be
3271 * better, of course, to have inline functions to do it, but since different
3272 * macros are called, depending on which case is being changed to, this is not
3273 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3274 * function can start with the common start macro, then finish with its special
3275 * handling; while the other three cases can just use the common end macro.
3276 *
3277 * The algorithm is to use the proper (passed in) macro or function to change
3278 * the case for code points that are below 256. The macro is used if using
3279 * locale rules for the case change; the function if not. If the code point is
3280 * above 255, it is computed from the input UTF-8, and another macro is called
3281 * to do the conversion. If necessary, the output is converted to UTF-8. If
3282 * using a locale, we have to check that the change did not cross the 255/256
3283 * boundary, see check_locale_boundary_crossing() for further details.
3284 *
3285 * The macros are split with the correct case change for the below-256 case
3286 * stored into 'result', and in the middle of an else clause for the above-255
3287 * case. At that point in the 'else', 'result' is not the final result, but is
3288 * the input code point calculated from the UTF-8. The fold code needs to
3289 * realize all this and take it from there.
3290 *
3291 * If you read the two macros as sequential, it's easier to understand what's
3292 * going on. */
3293#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3294 L1_func_extra_param) \
a239b1e2 3295 \
eaf412bf
KW
3296 if (flags & (locale_flags)) { \
3297 /* Treat a UTF-8 locale as not being in locale at all */ \
3298 if (IN_UTF8_CTYPE_LOCALE) { \
3299 flags &= ~(locale_flags); \
3300 } \
3301 else { \
3302 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3303 } \
3304 } \
3305 \
3306 if (UTF8_IS_INVARIANT(*p)) { \
3307 if (flags & (locale_flags)) { \
3308 result = LC_L1_change_macro(*p); \
3309 } \
3310 else { \
3311 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3312 } \
3313 } \
a239b1e2 3314 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3315 if (flags & (locale_flags)) { \
3316 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3317 *(p+1))); \
3318 } \
3319 else { \
3320 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3321 ustrp, lenp, L1_func_extra_param); \
3322 } \
3323 } \
fa8ab374
KW
3324 else { /* malformed UTF-8 or ord above 255 */ \
3325 STRLEN len_result; \
fa8ab374
KW
3326 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3327 if (len_result == (STRLEN) -1) { \
607313a1
KW
3328 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3329 1 /* Die */ ); \
fa8ab374 3330 }
eaf412bf
KW
3331
3332#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3333 result = change_macro(result, p, ustrp, lenp); \
3334 \
3335 if (flags & (locale_flags)) { \
3336 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3337 } \
3338 return result; \
3339 } \
3340 \
3341 /* Here, used locale rules. Convert back to UTF-8 */ \
3342 if (UTF8_IS_INVARIANT(result)) { \
3343 *ustrp = (U8) result; \
3344 *lenp = 1; \
3345 } \
3346 else { \
3347 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3348 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3349 *lenp = 2; \
3350 } \
3351 \
3352 return result;
3353
d3e79532 3354/*
87cea99e 3355=for apidoc to_utf8_upper
d3e79532 3356
a239b1e2 3357Instead use L</toUPPER_utf8_safe>.
a1433954 3358
d3e79532
JH
3359=cut */
3360
051a06d4 3361/* Not currently externally documented, and subject to change:
31f05a37
KW
3362 * <flags> is set iff iff the rules from the current underlying locale are to
3363 * be used. */
051a06d4 3364
2104c8d9 3365UV
607313a1
KW
3366Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3367 const U8 *e,
3368 U8* ustrp,
3369 STRLEN *lenp,
3370 bool flags,
3371 const char * const file,
3372 const int line)
a0ed51b3 3373{
051a06d4 3374 UV result;
607313a1
KW
3375 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3376 cBOOL(flags), file, line);
051a06d4
KW
3377
3378 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3379
eaf412bf
KW
3380 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3381 /* 2nd char of uc(U+DF) is 'S' */
3382 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3383 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3384}
a0ed51b3 3385
d3e79532 3386/*
87cea99e 3387=for apidoc to_utf8_title
d3e79532 3388
a239b1e2 3389Instead use L</toTITLE_utf8_safe>.
a1433954 3390
d3e79532
JH
3391=cut */
3392
051a06d4 3393/* Not currently externally documented, and subject to change:
31f05a37
KW
3394 * <flags> is set iff the rules from the current underlying locale are to be
3395 * used. Since titlecase is not defined in POSIX, for other than a
3396 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3397 */
051a06d4 3398
983ffd37 3399UV
607313a1
KW
3400Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3401 const U8 *e,
3402 U8* ustrp,
3403 STRLEN *lenp,
3404 bool flags,
3405 const char * const file,
3406 const int line)
983ffd37 3407{
051a06d4 3408 UV result;
607313a1
KW
3409 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3410 cBOOL(flags), file, line);
051a06d4
KW
3411
3412 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3413
eaf412bf
KW
3414 /* 2nd char of ucfirst(U+DF) is 's' */
3415 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3416 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3417}
3418
d3e79532 3419/*
87cea99e 3420=for apidoc to_utf8_lower
d3e79532 3421
a239b1e2 3422Instead use L</toLOWER_utf8_safe>.
a1433954 3423
d3e79532
JH
3424=cut */
3425
051a06d4 3426/* Not currently externally documented, and subject to change:
31f05a37
KW
3427 * <flags> is set iff iff the rules from the current underlying locale are to
3428 * be used.
3429 */
051a06d4 3430
2104c8d9 3431UV
607313a1
KW
3432Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3433 const U8 *e,
3434 U8* ustrp,
3435 STRLEN *lenp,
3436 bool flags,
3437 const char * const file,
3438 const int line)
a0ed51b3 3439{
051a06d4 3440 UV result;
607313a1
KW
3441 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3442 cBOOL(flags), file, line);
051a06d4 3443
051a06d4 3444 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3445
eaf412bf
KW
3446 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3447 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3448}
3449
d3e79532 3450/*
87cea99e 3451=for apidoc to_utf8_fold
d3e79532 3452
a239b1e2 3453Instead use L</toFOLD_utf8_safe>.
a1433954 3454
d3e79532
JH
3455=cut */
3456
051a06d4
KW
3457/* Not currently externally documented, and subject to change,
3458 * in <flags>
31f05a37
KW
3459 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3460 * locale are to be used.
051a06d4
KW
3461 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3462 * otherwise simple folds
a0270393
KW
3463 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3464 * prohibited
445bf929 3465 */
36bb2ab6 3466
b4e400f9 3467UV
607313a1
KW
3468Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3469 const U8 *e,
3470 U8* ustrp,
3471 STRLEN *lenp,
3472 U8 flags,
3473 const char * const file,
3474 const int line)
b4e400f9 3475{
051a06d4 3476 UV result;
607313a1
KW
3477 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3478 cBOOL(flags), file, line);
051a06d4 3479
36bb2ab6 3480 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3481
a0270393
KW
3482 /* These are mutually exclusive */
3483 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3484
50ba90ff
KW
3485 assert(p != ustrp); /* Otherwise overwrites */
3486
eaf412bf
KW
3487 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3488 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3489
eaf412bf 3490 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3491
1ca267a5
KW
3492 if (flags & FOLD_FLAGS_LOCALE) {
3493
76f2ffcd 3494# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3495 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3496
3497# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3498# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3499
3500 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3501
538e84ed
KW
3502 /* Special case these two characters, as what normally gets
3503 * returned under locale doesn't work */
76f2ffcd
KW
3504 if (UTF8SKIP(p) == cap_sharp_s_len
3505 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3506 {
ab0b796c
KW
3507 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3508 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3509 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3510 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3511 goto return_long_s;
3512 }
0766489e
KW
3513 else
3514#endif
3515 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3516 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3517 {
ab0b796c
KW
3518 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3519 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3520 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3521 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3522 goto return_ligature_st;
3523 }
74894415
KW
3524
3525#if UNICODE_MAJOR_VERSION == 3 \
3526 && UNICODE_DOT_VERSION == 0 \
3527 && UNICODE_DOT_DOT_VERSION == 1
3528# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3529
3530 /* And special case this on this Unicode version only, for the same
3531 * reaons the other two are special cased. They would cross the
3532 * 255/256 boundary which is forbidden under /l, and so the code
3533 * wouldn't catch that they are equivalent (which they are only in
3534 * this release) */
3535 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3536 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3537 {
3538 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3539 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3540 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3541 "resolved to \"\\x{0131}\".");
3542 goto return_dotless_i;
3543 }
3544#endif
3545
357aadde 3546 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3547 }
a0270393
KW
3548 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3549 return result;
3550 }
3551 else {
4a4088c4 3552 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3553 * character above the ASCII range, and the result should not
3554 * contain an ASCII character. */
a0270393
KW
3555
3556 UV original; /* To store the first code point of <p> */
3557
3558 /* Look at every character in the result; if any cross the
3559 * boundary, the whole thing is disallowed */
3560 U8* s = ustrp;
3561 U8* e = ustrp + *lenp;
3562 while (s < e) {
3563 if (isASCII(*s)) {
3564 /* Crossed, have to return the original */
3565 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3566
9fc2026f 3567 /* But in these instances, there is an alternative we can
1ca267a5 3568 * return that is valid */
0766489e
KW
3569 if (original == LATIN_SMALL_LETTER_SHARP_S
3570#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3571 || original == LATIN_CAPITAL_LETTER_SHARP_S
3572#endif
3573 ) {
1ca267a5
KW
3574 goto return_long_s;
3575 }
9fc2026f
KW
3576 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3577 goto return_ligature_st;
3578 }
74894415
KW
3579#if UNICODE_MAJOR_VERSION == 3 \
3580 && UNICODE_DOT_VERSION == 0 \
3581 && UNICODE_DOT_DOT_VERSION == 1
3582
3583 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3584 goto return_dotless_i;
3585 }
3586#endif
a0270393
KW
3587 Copy(p, ustrp, *lenp, char);
3588 return original;
3589 }
3590 s += UTF8SKIP(s);
3591 }
051a06d4 3592
a0270393
KW
3593 /* Here, no characters crossed, result is ok as-is */
3594 return result;
3595 }
051a06d4
KW
3596 }
3597
4a4088c4 3598 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3599 if (UTF8_IS_INVARIANT(result)) {
3600 *ustrp = (U8) result;
3601 *lenp = 1;
3602 }
3603 else {
62cb07ea
KW
3604 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3605 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3606 *lenp = 2;
3607 }
3608
051a06d4 3609 return result;
1ca267a5
KW
3610
3611 return_long_s:
3612 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3613 * folds to a string of two of these characters. By returning this
3614 * instead, then, e.g.,
3615 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3616 * works. */
3617
3618 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3619 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3620 ustrp, *lenp, U8);
3621 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3622
3623 return_ligature_st:
3624 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3625 * have the other one fold to it */
3626
3627 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3628 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3629 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3630
3631#if UNICODE_MAJOR_VERSION == 3 \
3632 && UNICODE_DOT_VERSION == 0 \
3633 && UNICODE_DOT_DOT_VERSION == 1
3634
3635 return_dotless_i:
3636 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3637 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3638 return LATIN_SMALL_LETTER_DOTLESS_I;
3639
3640#endif
3641
a0ed51b3
LW
3642}
3643
711a919c 3644/* Note:
f90a9a02 3645 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3646 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3647 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3648 */
c4a5db0c 3649
a0ed51b3 3650SV*
56576a04
KW
3651Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3652 I32 minbits, I32 none)
a0ed51b3 3653{
c4a5db0c
KW
3654 PERL_ARGS_ASSERT_SWASH_INIT;
3655
3656 /* Returns a copy of a swash initiated by the called function. This is the
3657 * public interface, and returning a copy prevents others from doing
3658 * mischief on the original */
3659
56576a04
KW
3660 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3661 NULL, NULL));
c4a5db0c
KW
3662}
3663
3664SV*
56576a04
KW
3665Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3666 I32 minbits, I32 none, SV* invlist,
3667 U8* const flags_p)
c4a5db0c 3668{
2c1f00b9
YO
3669
3670 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3671 * use the following define */
3672
3673#define CORE_SWASH_INIT_RETURN(x) \
3674 PL_curpm= old_PL_curpm; \
3675 return x
3676
c4a5db0c 3677 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3678 * by calling utf8_heavy.pl in the general case. The returned value may be
3679 * the swash's inversion list instead if the input parameters allow it.
3680 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3681 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3682 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3683 *
3684 * This interface should only be used by functions that won't destroy or
3685 * adversely change the swash, as doing so affects all other uses of the
3686 * swash in the program; the general public should use 'Perl_swash_init'
3687 * instead.
3688 *
3689 * pkg is the name of the package that <name> should be in.
3690 * name is the name of the swash to find. Typically it is a Unicode
3691 * property name, including user-defined ones
3692 * listsv is a string to initialize the swash with. It must be of the form
3693 * documented as the subroutine return value in
3694 * L<perlunicode/User-Defined Character Properties>
3695 * minbits is the number of bits required to represent each data element.
3696 * It is '1' for binary properties.
3697 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3698 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3699 * flags_p if non-NULL is the address of various input and output flag bits
3700 * to the routine, as follows: ('I' means is input to the routine;
3701 * 'O' means output from the routine. Only flags marked O are
3702 * meaningful on return.)
3703 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3704 * came from a user-defined property. (I O)
5d3d13d1
KW
3705 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3706 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3707 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3708 * return of an inversion list instead of a swash hash if this routine
3709 * thinks that would result in faster execution of swash_fetch() later
3710 * on. (I)
9a53f6cf
KW
3711 *
3712 * Thus there are three possible inputs to find the swash: <name>,
3713 * <listsv>, and <invlist>. At least one must be specified. The result
3714 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW