This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.h: Comments only
[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
c179379d
KW
485/* Anything larger than this will overflow the word if it were converted into a UV */
486#if defined(UV_IS_QUAD)
487# ifdef EBCDIC /* Actually is I8 */
488# define HIGHEST_REPRESENTABLE_UTF8 \
489 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
490# else
491# define HIGHEST_REPRESENTABLE_UTF8 \
492 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
493# endif
494#else /* 32-bit */
495# ifdef EBCDIC
496# define HIGHEST_REPRESENTABLE_UTF8 \
497 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
498# else
499# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
500# endif
501#endif
502
12a4bed3
KW
503PERL_STATIC_INLINE bool
504S_does_utf8_overflow(const U8 * const s, const U8 * e)
505{
506 const U8 *x;
507 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
508
c551bb01
KW
509#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
510
511 const STRLEN len = e - s;
512
513#endif
514
12a4bed3
KW
515 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
516 * platform, that is if it represents a code point larger than the highest
517 * representable code point. (For ASCII platforms, we could use memcmp()
518 * because we don't have to convert each byte to I8, but it's very rare
519 * input indeed that would approach overflow, so the loop below will likely
520 * only get executed once.
521 *
522 * 'e' must not be beyond a full character. If it is less than a full
523 * character, the function returns FALSE if there is any input beyond 'e'
524 * that could result in a non-overflowing code point */
525
526 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
527 assert(s <= e && s + UTF8SKIP(s) >= e);
528
529#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
530
531 /* On 32 bit ASCII machines, many overlongs that start with FF don't
532 * overflow */
533
c551bb01 534 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
535 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
536 return memGE(s, max_32_bit_overlong,
c551bb01 537 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
538 }
539
540#endif
12a4bed3
KW
541
542 for (x = s; x < e; x++, y++) {
543
10dabba9
KW
544 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
545 continue;
12a4bed3 546 }
12a4bed3 547
10dabba9
KW
548 /* If this byte is larger than the corresponding highest UTF-8 byte,
549 * the sequence overflow; otherwise the byte is less than, and so the
550 * sequence doesn't overflow */
551 return NATIVE_UTF8_TO_I8(*x) > *y;
552
553 }
554
12a4bed3
KW
555 /* Got to the end and all bytes are the same. If the input is a whole
556 * character, it doesn't overflow. And if it is a partial character,
557 * there's not enough information to tell, so assume doesn't overflow */
558 return FALSE;
559}
560
561PERL_STATIC_INLINE bool
562S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
563{
564 /* Overlongs can occur whenever the number of continuation bytes
565 * changes. That means whenever the number of leading 1 bits in a start
566 * byte increases from the next lower start byte. That happens for start
567 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
568 * illegal start bytes have already been excluded, so don't need to be
569 * tested here;
570 * ASCII platforms: C0, C1
571 * EBCDIC platforms C0, C1, C2, C3, C4, E0
572 *
573 * At least a second byte is required to determine if other sequences will
574 * be an overlong. */
575
576 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
577 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
578
579 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
580 assert(len > 1 && UTF8_IS_START(*s));
581
582 /* Each platform has overlongs after the start bytes given above (expressed
583 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
584 * the logic is the same, except the E0 overlong has already been excluded
585 * on EBCDIC platforms. The values below were found by manually
586 * inspecting the UTF-8 patterns. See the tables in utf8.h and
587 * utfebcdic.h. */
588
589# ifdef EBCDIC
590# define F0_ABOVE_OVERLONG 0xB0
591# define F8_ABOVE_OVERLONG 0xA8
592# define FC_ABOVE_OVERLONG 0xA4
593# define FE_ABOVE_OVERLONG 0xA2
594# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
595 /* I8(0xfe) is FF */
596# else
597
598 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
599 return TRUE;
600 }
601
602# define F0_ABOVE_OVERLONG 0x90
603# define F8_ABOVE_OVERLONG 0x88
604# define FC_ABOVE_OVERLONG 0x84
605# define FE_ABOVE_OVERLONG 0x82
606# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
607# endif
608
609
610 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
611 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
612 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
613 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
614 {
615 return TRUE;
616 }
617
b0b342d4
KW
618 /* Check for the FF overlong */
619 return isFF_OVERLONG(s, len);
620}
621
622PERL_STATIC_INLINE bool
623S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
624{
625 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
626
627 /* Check for the FF overlong. This happens only if all these bytes match;
628 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 629 * utfebcdic.h. */
12a4bed3 630
b0b342d4
KW
631 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
632 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
633 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
634}
635
636#undef F0_ABOVE_OVERLONG
637#undef F8_ABOVE_OVERLONG
638#undef FC_ABOVE_OVERLONG
639#undef FE_ABOVE_OVERLONG
640#undef FF_OVERLONG_PREFIX
641
35f8c9bd 642STRLEN
edc2c47a 643Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 644{
2b479609 645 STRLEN len;
12a4bed3 646 const U8 *x;
35f8c9bd 647
2b479609
KW
648 /* A helper function that should not be called directly.
649 *
650 * This function returns non-zero if the string beginning at 's' and
651 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
652 * code point; otherwise it returns 0. The examination stops after the
653 * first code point in 's' is validated, not looking at the rest of the
654 * input. If 'e' is such that there are not enough bytes to represent a
655 * complete code point, this function will return non-zero anyway, if the
656 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
657 * excluded by 'flags'.
658 *
659 * A non-zero return gives the number of bytes required to represent the
660 * code point. Be aware that if the input is for a partial character, the
661 * return will be larger than 'e - s'.
662 *
663 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
664 * The caller should have excluded the possibility of it being invariant
665 * before calling this function.
2b479609
KW
666 *
667 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
668 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
669 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
670 * disallowed by the flags. If the input is only for a partial character,
671 * the function will return non-zero if there is any sequence of
672 * well-formed UTF-8 that, when appended to the input sequence, could
673 * result in an allowed code point; otherwise it returns 0. Non characters
674 * cannot be determined based on partial character input. But many of the
675 * other excluded types can be determined with just the first one or two
676 * bytes.
677 *
678 */
679
680 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
681
682 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 683 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 684 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 685
2b479609 686 /* A variant char must begin with a start byte */
35f8c9bd
KW
687 if (UNLIKELY(! UTF8_IS_START(*s))) {
688 return 0;
689 }
690
edc2c47a
KW
691 /* Examine a maximum of a single whole code point */
692 if (e - s > UTF8SKIP(s)) {
693 e = s + UTF8SKIP(s);
694 }
695
2b479609
KW
696 len = e - s;
697
698 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
699 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 700
56576a04
KW
701 /* Here, we are disallowing some set of largish code points, and the
702 * first byte indicates the sequence is for a code point that could be
703 * in the excluded set. We generally don't have to look beyond this or
704 * the second byte to see if the sequence is actually for one of the
705 * excluded classes. The code below is derived from this table:
706 *
2b479609
KW
707 * UTF-8 UTF-EBCDIC I8
708 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
709 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
710 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
711 *
56576a04
KW
712 * Keep in mind that legal continuation bytes range between \x80..\xBF
713 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
714 * continuation bytes. Hence, we don't have to test the upper edge
715 * because if any of those is encountered, the sequence is malformed,
716 * and would fail elsewhere in this function.
717 *
718 * The code here likewise assumes that there aren't other
719 * malformations; again the function should fail elsewhere because of
720 * these. For example, an overlong beginning with FC doesn't actually
721 * have to be a super; it could actually represent a small code point,
722 * even U+0000. But, since overlongs (and other malformations) are
723 * illegal, the function should return FALSE in either case.
2b479609
KW
724 */
725
726#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
727# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 728# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 729
19794540
KW
730# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
731 /* B6 and B7 */ \
732 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 733# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
734#else
735# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
736# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
737# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 738# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
739#endif
740
741 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
742 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
743 {
2b479609
KW
744 return 0; /* Above Unicode */
745 }
746
d044b7a7 747 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 748 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 749 {
57ff5f59 750 return 0;
2b479609
KW
751 }
752
753 if (len > 1) {
754 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
755
756 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 757 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
758 {
759 return 0; /* Above Unicode */
760 }
761
762 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 763 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
764 {
765 return 0; /* Surrogate */
766 }
767
768 if ( (flags & UTF8_DISALLOW_NONCHAR)
769 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
770 {
771 return 0; /* Noncharacter code point */
772 }
773 }
774 }
775
776 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
777 for (x = s + 1; x < e; x++) {
778 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
779 return 0;
780 }
781 }
782
af13dd8a 783 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
784 * overlong. */
785 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
786 return 0;
af13dd8a
KW
787 }
788
12a4bed3
KW
789 /* And finally, that the code point represented fits in a word on this
790 * platform */
791 if (does_utf8_overflow(s, e)) {
792 return 0;
35f8c9bd
KW
793 }
794
2b479609 795 return UTF8SKIP(s);
35f8c9bd
KW
796}
797
7e2f38b2
KW
798char *
799Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
800{
801 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
802 * bytes starting at 's'. 'format' gives how to display each byte.
803 * Currently, there are only two formats, so it is currently a bool:
804 * 0 \xab
805 * 1 ab (that is a space between two hex digit bytes)
806 */
7cf8d05d
KW
807
808 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
809 trailing NUL */
810 const U8 * const e = s + len;
811 char * output;
812 char * d;
813
814 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
815
816 Newx(output, output_len, char);
817 SAVEFREEPV(output);
818
819 d = output;
820 for (; s < e; s++) {
821 const unsigned high_nibble = (*s & 0xF0) >> 4;
822 const unsigned low_nibble = (*s & 0x0F);
823
7e2f38b2
KW
824 if (format) {
825 *d++ = ' ';
826 }
827 else {
828 *d++ = '\\';
829 *d++ = 'x';
830 }
7cf8d05d
KW
831
832 if (high_nibble < 10) {
833 *d++ = high_nibble + '0';
834 }
835 else {
836 *d++ = high_nibble - 10 + 'a';
837 }
838
839 if (low_nibble < 10) {
840 *d++ = low_nibble + '0';
841 }
842 else {
843 *d++ = low_nibble - 10 + 'a';
844 }
845 }
846
847 *d = '\0';
848 return output;
849}
850
806547a7 851PERL_STATIC_INLINE char *
7cf8d05d
KW
852S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
853
854 /* How many bytes to print */
3cc6a05e 855 STRLEN print_len,
7cf8d05d
KW
856
857 /* Which one is the non-continuation */
858 const STRLEN non_cont_byte_pos,
859
860 /* How many bytes should there be? */
861 const STRLEN expect_len)
806547a7
KW
862{
863 /* Return the malformation warning text for an unexpected continuation
864 * byte. */
865
7cf8d05d 866 const char * const where = (non_cont_byte_pos == 1)
806547a7 867 ? "immediately"
7cf8d05d
KW
868 : Perl_form(aTHX_ "%d bytes",
869 (int) non_cont_byte_pos);
806547a7
KW
870
871 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
872
7cf8d05d
KW
873 /* We don't need to pass this parameter, but since it has already been
874 * calculated, it's likely faster to pass it; verify under DEBUGGING */
875 assert(expect_len == UTF8SKIP(s));
876
877 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
878 " %s after start byte 0x%02x; need %d bytes, got %d)",
879 malformed_text,
7e2f38b2 880 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
881 *(s + non_cont_byte_pos),
882 where,
883 *s,
884 (int) expect_len,
885 (int) non_cont_byte_pos);
806547a7
KW
886}
887
35f8c9bd
KW
888/*
889
de69f3af 890=for apidoc utf8n_to_uvchr
378516de
KW
891
892THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 893Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 894
9041c2e3 895Bottom level UTF-8 decode routine.
de69f3af 896Returns the native code point value of the first character in the string C<s>,
746afd53
KW
897which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
898C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
899the length, in bytes, of that character.
949cf498
KW
900
901The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
902well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
903causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
904is the next possible position in C<s> that could begin a non-malformed
905character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
906is raised. Some UTF-8 input sequences may contain multiple malformations.
907This function tries to find every possible one in each call, so multiple
56576a04 908warnings can be raised for the same sequence.
949cf498
KW
909
910Various ALLOW flags can be set in C<flags> to allow (and not warn on)
911individual types of malformations, such as the sequence being overlong (that
912is, when there is a shorter sequence that can express the same code point;
913overlong sequences are expressly forbidden in the UTF-8 standard due to
914potential security issues). Another malformation example is the first byte of
915a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
916flags. Even if allowed, this function generally returns the Unicode
917REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
918F<utf8.h> to override this behavior for the overlong malformations, but don't
919do that except for very specialized purposes.
949cf498 920
796b6530 921The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
922flags) malformation is found. If this flag is set, the routine assumes that
923the caller will raise a warning, and this function will silently just set
d088425d
KW
924C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
925
75200dff 926Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 927character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
928in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
929be set to 1. To disambiguate, upon a zero return, see if the first byte of
930C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 931error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
932
933Certain code points are considered problematic. These are Unicode surrogates,
746afd53 934Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 935By default these are considered regular code points, but certain situations
ecc1615f
KW
936warrant special handling for them, which can be specified using the C<flags>
937parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
938three classes are treated as malformations and handled as such. The flags
939C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
940C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
941disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
942restricts the allowed inputs to the strict UTF-8 traditionally defined by
943Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
944definition given by
945L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
946The difference between traditional strictness and C9 strictness is that the
947latter does not forbid non-character code points. (They are still discouraged,
948however.) For more discussion see L<perlunicode/Noncharacter code points>.
949
950The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
951C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
952C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
953raised for their respective categories, but otherwise the code points are
954considered valid (not malformations). To get a category to both be treated as
955a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 956(But note that warnings are not raised if lexically disabled nor if
796b6530 957C<UTF8_CHECK_ONLY> is also specified.)
949cf498 958
57ff5f59
KW
959Extremely high code points were never specified in any standard, and require an
960extension to UTF-8 to express, which Perl does. It is likely that programs
961written in something other than Perl would not be able to read files that
962contain these; nor would Perl understand files written by something that uses a
963different extension. For these reasons, there is a separate set of flags that
964can warn and/or disallow these extremely high code points, even if other
965above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
966C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
967L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
968above-Unicode code points, including these, as malformations.
969(Note that the Unicode standard considers anything above 0x10FFFF to be
970illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
971(2**31 -1))
972
973A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
974retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
975C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
976C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
977can apply to code points that actually do fit in 31 bits. This happens on
978EBCDIC platforms, and sometimes when the L<overlong
979malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
980describe the situation in all cases.
981
760c7c2f
KW
982It is now deprecated to have very high code points (above C<IV_MAX> on the
983platforms) and this function will raise a deprecation warning for these (unless
d5944cab 984such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 985in a 32-bit word.
ab8e6d41 986
949cf498
KW
987All other code points corresponding to Unicode characters, including private
988use and those yet to be assigned, are never considered malformed and never
989warn.
67e989fb 990
37607a96 991=cut
f9380377
KW
992
993Also implemented as a macro in utf8.h
994*/
995
996UV
997Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
998 STRLEN curlen,
999 STRLEN *retlen,
1000 const U32 flags)
1001{
1002 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1003
1004 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1005}
1006
1007/*
1008
1009=for apidoc utf8n_to_uvchr_error
1010
1011THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1012Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1013
1014This function is for code that needs to know what the precise malformation(s)
1015are when an error is found.
1016
1017It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1018all the others, C<errors>. If this parameter is 0, this function behaves
1019identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1020to a C<U32> variable, which this function sets to indicate any errors found.
1021Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1022C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1023of these bits will be set if a malformation is found, even if the input
7a65503b 1024C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1025exceptions are noted:
1026
1027=over 4
1028
57ff5f59 1029=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1030
57ff5f59
KW
1031The input sequence is not standard UTF-8, but a Perl extension. This bit is
1032set only if the input C<flags> parameter contains either the
1033C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1034
1035Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1036and so some extension must be used to express them. Perl uses a natural
1037extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1038extension to represent even higher ones, so that any code point that fits in a
103964-bit word can be represented. Text using these extensions is not likely to
1040be portable to non-Perl code. We lump both of these extensions together and
1041refer to them as Perl extended UTF-8. There exist other extensions that people
1042have invented, incompatible with Perl's.
1043
1044On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1045extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1046than on ASCII. Prior to that, code points 2**31 and higher were simply
1047unrepresentable, and a different, incompatible method was used to represent
1048code points between 2**30 and 2**31 - 1.
1049
1050On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1051Perl extended UTF-8 is used.
1052
1053In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1054may use for backward compatibility. That name is misleading, as this flag may
1055be set when the code point actually does fit in 31 bits. This happens on
1056EBCDIC platforms, and sometimes when the L<overlong
1057malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1058describes the situation in all cases.
f9380377
KW
1059
1060=item C<UTF8_GOT_CONTINUATION>
1061
1062The input sequence was malformed in that the first byte was a a UTF-8
1063continuation byte.
1064
1065=item C<UTF8_GOT_EMPTY>
1066
1067The input C<curlen> parameter was 0.
1068
1069=item C<UTF8_GOT_LONG>
1070
1071The input sequence was malformed in that there is some other sequence that
1072evaluates to the same code point, but that sequence is shorter than this one.
1073
fecaf136
KW
1074Until Unicode 3.1, it was legal for programs to accept this malformation, but
1075it was discovered that this created security issues.
1076
f9380377
KW
1077=item C<UTF8_GOT_NONCHAR>
1078
1079The code point represented by the input UTF-8 sequence is for a Unicode
1080non-character code point.
1081This bit is set only if the input C<flags> parameter contains either the
1082C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1083
1084=item C<UTF8_GOT_NON_CONTINUATION>
1085
1086The input sequence was malformed in that a non-continuation type byte was found
1087in a position where only a continuation type one should be.
1088
1089=item C<UTF8_GOT_OVERFLOW>
1090
1091The input sequence was malformed in that it is for a code point that is not
1092representable in the number of bits available in a UV on the current platform.
1093
1094=item C<UTF8_GOT_SHORT>
1095
1096The input sequence was malformed in that C<curlen> is smaller than required for
1097a complete sequence. In other words, the input is for a partial character
1098sequence.
1099
1100=item C<UTF8_GOT_SUPER>
1101
1102The input sequence was malformed in that it is for a non-Unicode code point;
1103that is, one above the legal Unicode maximum.
1104This bit is set only if the input C<flags> parameter contains either the
1105C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1106
1107=item C<UTF8_GOT_SURROGATE>
1108
1109The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1110code point.
1111This bit is set only if the input C<flags> parameter contains either the
1112C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1113
1114=back
1115
133551d8
KW
1116To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1117flag to suppress any warnings, and then examine the C<*errors> return.
1118
f9380377 1119=cut
37607a96 1120*/
67e989fb 1121
a0ed51b3 1122UV
f9380377
KW
1123Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1124 STRLEN curlen,
1125 STRLEN *retlen,
1126 const U32 flags,
1127 U32 * errors)
a0ed51b3 1128{
d4c19fe8 1129 const U8 * const s0 = s;
2b5e7bc2
KW
1130 U8 * send = NULL; /* (initialized to silence compilers' wrong
1131 warning) */
1132 U32 possible_problems = 0; /* A bit is set here for each potential problem
1133 found as we go along */
eb83ed87 1134 UV uv = *s;
2b5e7bc2
KW
1135 STRLEN expectlen = 0; /* How long should this sequence be?
1136 (initialized to silence compilers' wrong
1137 warning) */
e308b348 1138 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1139 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1140 this gets set and discarded */
a0dbb045 1141
2b5e7bc2
KW
1142 /* The below are used only if there is both an overlong malformation and a
1143 * too short one. Otherwise the first two are set to 's0' and 'send', and
1144 * the third not used at all */
1145 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1146 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1147 routine; see [perl #130921] */
2b5e7bc2 1148 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1149
f9380377
KW
1150 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1151
1152 if (errors) {
1153 *errors = 0;
1154 }
1155 else {
1156 errors = &discard_errors;
1157 }
a0dbb045 1158
eb83ed87
KW
1159 /* The order of malformation tests here is important. We should consume as
1160 * few bytes as possible in order to not skip any valid character. This is
1161 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1162 * http://unicode.org/reports/tr36 for more discussion as to why. For
1163 * example, once we've done a UTF8SKIP, we can tell the expected number of
1164 * bytes, and could fail right off the bat if the input parameters indicate
1165 * that there are too few available. But it could be that just that first
1166 * byte is garbled, and the intended character occupies fewer bytes. If we
1167 * blindly assumed that the first byte is correct, and skipped based on
1168 * that number, we could skip over a valid input character. So instead, we
1169 * always examine the sequence byte-by-byte.
1170 *
1171 * We also should not consume too few bytes, otherwise someone could inject
1172 * things. For example, an input could be deliberately designed to
1173 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1174 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1175 * which is actually part of of the overflowing sequence), that could look
1176 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1177 * sequence and process the rest, inappropriately.
1178 *
1179 * Some possible input sequences are malformed in more than one way. This
1180 * function goes to lengths to try to find all of them. This is necessary
1181 * for correctness, as the inputs may allow one malformation but not
1182 * another, and if we abandon searching for others after finding the
1183 * allowed one, we could allow in something that shouldn't have been.
1184 */
eb83ed87 1185
b5b9af04 1186 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1187 possible_problems |= UTF8_GOT_EMPTY;
1188 curlen = 0;
5a48568d 1189 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1190 goto ready_to_handle_errors;
0c443dc2
JH
1191 }
1192
eb83ed87
KW
1193 expectlen = UTF8SKIP(s);
1194
1195 /* A well-formed UTF-8 character, as the vast majority of calls to this
1196 * function will be for, has this expected length. For efficiency, set
1197 * things up here to return it. It will be overriden only in those rare
1198 * cases where a malformation is found */
1199 if (retlen) {
1200 *retlen = expectlen;
1201 }
1202
1203 /* An invariant is trivially well-formed */
1d72bdf6 1204 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1205 return uv;
a0ed51b3 1206 }
67e989fb 1207
eb83ed87 1208 /* A continuation character can't start a valid sequence */
b5b9af04 1209 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1210 possible_problems |= UTF8_GOT_CONTINUATION;
1211 curlen = 1;
1212 uv = UNICODE_REPLACEMENT;
1213 goto ready_to_handle_errors;
ba210ebe 1214 }
9041c2e3 1215
dcd27b3c 1216 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1217 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1218 * because it excludes start bytes like \xC0 that always lead to
1219 * overlongs.) */
dcd27b3c 1220
534752c1
KW
1221 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1222 * that indicate the number of bytes in the character's whole UTF-8
1223 * sequence, leaving just the bits that are part of the value. */
1224 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1225
e308b348
KW
1226 /* Setup the loop end point, making sure to not look past the end of the
1227 * input string, and flag it as too short if the size isn't big enough. */
1228 send = (U8*) s0;
1229 if (UNLIKELY(curlen < expectlen)) {
1230 possible_problems |= UTF8_GOT_SHORT;
1231 avail_len = curlen;
1232 send += curlen;
1233 }
1234 else {
1235 send += expectlen;
1236 }
e308b348 1237
eb83ed87 1238 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1239 * accumulating each into the working value as we go. */
eb83ed87 1240 for (s = s0 + 1; s < send; s++) {
b5b9af04 1241 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1242 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1243 continue;
1244 }
1245
1246 /* Here, found a non-continuation before processing all expected bytes.
1247 * This byte indicates the beginning of a new character, so quit, even
1248 * if allowing this malformation. */
2b5e7bc2 1249 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1250 break;
eb83ed87
KW
1251 } /* End of loop through the character's bytes */
1252
1253 /* Save how many bytes were actually in the character */
1254 curlen = s - s0;
1255
2b5e7bc2
KW
1256 /* Note that there are two types of too-short malformation. One is when
1257 * there is actual wrong data before the normal termination of the
1258 * sequence. The other is that the sequence wasn't complete before the end
1259 * of the data we are allowed to look at, based on the input 'curlen'.
1260 * This means that we were passed data for a partial character, but it is
1261 * valid as far as we saw. The other is definitely invalid. This
1262 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1263 * separate.
1264 *
1265 * A convenience macro that matches either of the too-short conditions. */
1266# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1267
1268 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1269 uv_so_far = uv;
1270 uv = UNICODE_REPLACEMENT;
1271 }
2b5e7bc2 1272
08e73697
KW
1273 /* Check for overflow. The algorithm requires us to not look past the end
1274 * of the current character, even if partial, so the upper limit is 's' */
1275 if (UNLIKELY(does_utf8_overflow(s0, s))) {
2b5e7bc2
KW
1276 possible_problems |= UTF8_GOT_OVERFLOW;
1277 uv = UNICODE_REPLACEMENT;
eb83ed87 1278 }
eb83ed87 1279
2b5e7bc2
KW
1280 /* Check for overlong. If no problems so far, 'uv' is the correct code
1281 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1282 * we must look at the UTF-8 byte sequence itself to see if it is for an
1283 * overlong */
1284 if ( ( LIKELY(! possible_problems)
1285 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1286 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1287 && ( UNLIKELY(! UTF8_IS_START(*s0))
1288 || ( curlen > 1
1289 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1290 s - s0))))))
2f8f112e 1291 {
2b5e7bc2
KW
1292 possible_problems |= UTF8_GOT_LONG;
1293
abc28b54 1294 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1295
abc28b54
KW
1296 /* The calculation in the 'true' branch of this 'if'
1297 * below won't work if overflows, and isn't needed
1298 * anyway. Further below we handle all overflow
1299 * cases */
1300 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1301 {
2b5e7bc2
KW
1302 UV min_uv = uv_so_far;
1303 STRLEN i;
1304
1305 /* Here, the input is both overlong and is missing some trailing
1306 * bytes. There is no single code point it could be for, but there
1307 * may be enough information present to determine if what we have
1308 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1309 * The code further below has the intelligence to determine this,
1310 * but just for non-overlong UTF-8 sequences. What we do here is
1311 * calculate the smallest code point the input could represent if
1312 * there were no too short malformation. Then we compute and save
1313 * the UTF-8 for that, which is what the code below looks at
1314 * instead of the raw input. It turns out that the smallest such
1315 * code point is all we need. */
2b5e7bc2
KW
1316 for (i = curlen; i < expectlen; i++) {
1317 min_uv = UTF8_ACCUMULATE(min_uv,
1318 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1319 }
1320
e9f2c446 1321 adjusted_s0 = temp_char_buf;
57ff5f59 1322 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1323 }
eb83ed87
KW
1324 }
1325
56576a04
KW
1326 /* Here, we have found all the possible problems, except for when the input
1327 * is for a problematic code point not allowed by the input parameters. */
1328
06188866
KW
1329 /* uv is valid for overlongs */
1330 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1331
1332 /* isn't problematic if < this */
1333 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1334 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1335
1336 /* if overflow, we know without looking further
1337 * precisely which of the problematic types it is,
1338 * and we deal with those in the overflow handling
1339 * code */
1340 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1341 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1342 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1343 && ((flags & ( UTF8_DISALLOW_NONCHAR
1344 |UTF8_DISALLOW_SURROGATE
1345 |UTF8_DISALLOW_SUPER
d044b7a7 1346 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1347 |UTF8_WARN_NONCHAR
1348 |UTF8_WARN_SURROGATE
1349 |UTF8_WARN_SUPER
d044b7a7 1350 |UTF8_WARN_PERL_EXTENDED))
2b5e7bc2
KW
1351 /* In case of a malformation, 'uv' is not valid, and has
1352 * been changed to something in the Unicode range.
1353 * Currently we don't output a deprecation message if there
1354 * is already a malformation, so we don't have to special
1355 * case the test immediately below */
760c7c2f
KW
1356 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1357 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1358 {
2b5e7bc2
KW
1359 /* If there were no malformations, or the only malformation is an
1360 * overlong, 'uv' is valid */
1361 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1362 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1363 possible_problems |= UTF8_GOT_SURROGATE;
1364 }
1365 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1366 possible_problems |= UTF8_GOT_SUPER;
1367 }
1368 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1369 possible_problems |= UTF8_GOT_NONCHAR;
1370 }
1371 }
1372 else { /* Otherwise, need to look at the source UTF-8, possibly
1373 adjusted to be non-overlong */
1374
1375 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1376 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1377 {
2b5e7bc2
KW
1378 possible_problems |= UTF8_GOT_SUPER;
1379 }
1380 else if (curlen > 1) {
1381 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1382 NATIVE_UTF8_TO_I8(*adjusted_s0),
1383 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1384 {
2b5e7bc2 1385 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1386 }
2b5e7bc2
KW
1387 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1388 NATIVE_UTF8_TO_I8(*adjusted_s0),
1389 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1390 {
1391 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1392 }
1393 }
c0236afe 1394
2b5e7bc2
KW
1395 /* We need a complete well-formed UTF-8 character to discern
1396 * non-characters, so can't look for them here */
1397 }
1398 }
949cf498 1399
2b5e7bc2
KW
1400 ready_to_handle_errors:
1401
1402 /* At this point:
1403 * curlen contains the number of bytes in the sequence that
1404 * this call should advance the input by.
e308b348
KW
1405 * avail_len gives the available number of bytes passed in, but
1406 * only if this is less than the expected number of
1407 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1408 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1409 * is set in it for each potential problem found.
1410 * uv contains the code point the input sequence
1411 * represents; or if there is a problem that prevents
1412 * a well-defined value from being computed, it is
1413 * some subsitute value, typically the REPLACEMENT
1414 * CHARACTER.
1415 * s0 points to the first byte of the character
56576a04
KW
1416 * s points to just after were we left off processing
1417 * the character
1418 * send points to just after where that character should
1419 * end, based on how many bytes the start byte tells
1420 * us should be in it, but no further than s0 +
1421 * avail_len
2b5e7bc2 1422 */
eb83ed87 1423
2b5e7bc2
KW
1424 if (UNLIKELY(possible_problems)) {
1425 bool disallowed = FALSE;
1426 const U32 orig_problems = possible_problems;
1427
1428 while (possible_problems) { /* Handle each possible problem */
1429 UV pack_warn = 0;
1430 char * message = NULL;
1431
1432 /* Each 'if' clause handles one problem. They are ordered so that
1433 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1434 * ones; this is kinda in decreasing severity order. But the
1435 * overlong must come last, as it changes 'uv' looked at by the
1436 * others */
2b5e7bc2
KW
1437 if (possible_problems & UTF8_GOT_OVERFLOW) {
1438
56576a04
KW
1439 /* Overflow means also got a super and are using Perl's
1440 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1441 possible_problems
d044b7a7 1442 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1443 *errors |= UTF8_GOT_OVERFLOW;
1444
1445 /* But the API says we flag all errors found */
1446 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1447 *errors |= UTF8_GOT_SUPER;
1448 }
ddb65933 1449 if (flags
d044b7a7 1450 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1451 {
d044b7a7 1452 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1453 }
2b5e7bc2 1454
d60baaa7 1455 /* Disallow if any of the three categories say to */
56576a04 1456 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1457 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1458 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1459 {
1460 disallowed = TRUE;
1461 }
1462
d60baaa7
KW
1463 /* Likewise, warn if any say to, plus if deprecation warnings
1464 * are on, because this code point is above IV_MAX */
56576a04 1465 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7 1466 || ! (flags & UTF8_ALLOW_OVERFLOW)
d044b7a7 1467 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1468 {
2b5e7bc2 1469
ddb65933
KW
1470 /* The warnings code explicitly says it doesn't handle the
1471 * case of packWARN2 and two categories which have
1472 * parent-child relationship. Even if it works now to
1473 * raise the warning if either is enabled, it wouldn't
1474 * necessarily do so in the future. We output (only) the
56576a04 1475 * most dire warning */
ddb65933
KW
1476 if (! (flags & UTF8_CHECK_ONLY)) {
1477 if (ckWARN_d(WARN_UTF8)) {
1478 pack_warn = packWARN(WARN_UTF8);
1479 }
1480 else if (ckWARN_d(WARN_NON_UNICODE)) {
1481 pack_warn = packWARN(WARN_NON_UNICODE);
1482 }
1483 if (pack_warn) {
1484 message = Perl_form(aTHX_ "%s: %s (overflows)",
1485 malformed_text,
05b9033b 1486 _byte_dump_string(s0, curlen, 0));
ddb65933 1487 }
2b5e7bc2
KW
1488 }
1489 }
1490 }
1491 else if (possible_problems & UTF8_GOT_EMPTY) {
1492 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1493 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1494
1495 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1496
1497 /* This so-called malformation is now treated as a bug in
1498 * the caller. If you have nothing to decode, skip calling
1499 * this function */
1500 assert(0);
1501
2b5e7bc2
KW
1502 disallowed = TRUE;
1503 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1504 pack_warn = packWARN(WARN_UTF8);
1505 message = Perl_form(aTHX_ "%s (empty string)",
1506 malformed_text);
1507 }
1508 }
1509 }
1510 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1511 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1512 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1513
1514 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1515 disallowed = TRUE;
1516 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1517 pack_warn = packWARN(WARN_UTF8);
1518 message = Perl_form(aTHX_
1519 "%s: %s (unexpected continuation byte 0x%02x,"
1520 " with no preceding start byte)",
1521 malformed_text,
7e2f38b2 1522 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1523 }
1524 }
1525 }
2b5e7bc2
KW
1526 else if (possible_problems & UTF8_GOT_SHORT) {
1527 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1528 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1529
1530 if (! (flags & UTF8_ALLOW_SHORT)) {
1531 disallowed = TRUE;
1532 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1533 pack_warn = packWARN(WARN_UTF8);
1534 message = Perl_form(aTHX_
56576a04
KW
1535 "%s: %s (too short; %d byte%s available, need %d)",
1536 malformed_text,
1537 _byte_dump_string(s0, send - s0, 0),
1538 (int)avail_len,
1539 avail_len == 1 ? "" : "s",
1540 (int)expectlen);
2b5e7bc2
KW
1541 }
1542 }
ba210ebe 1543
2b5e7bc2 1544 }
e308b348
KW
1545 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1546 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1547 *errors |= UTF8_GOT_NON_CONTINUATION;
1548
1549 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1550 disallowed = TRUE;
1551 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1552
1553 /* If we don't know for sure that the input length is
1554 * valid, avoid as much as possible reading past the
1555 * end of the buffer */
1556 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1557 ? s - s0
1558 : send - s0;
e308b348
KW
1559 pack_warn = packWARN(WARN_UTF8);
1560 message = Perl_form(aTHX_ "%s",
1561 unexpected_non_continuation_text(s0,
99a765e9 1562 printlen,
e308b348
KW
1563 s - s0,
1564 (int) expectlen));
1565 }
1566 }
1567 }
2b5e7bc2
KW
1568 else if (possible_problems & UTF8_GOT_SURROGATE) {
1569 possible_problems &= ~UTF8_GOT_SURROGATE;
1570
f9380377
KW
1571 if (flags & UTF8_WARN_SURROGATE) {
1572 *errors |= UTF8_GOT_SURROGATE;
1573
1574 if ( ! (flags & UTF8_CHECK_ONLY)
1575 && ckWARN_d(WARN_SURROGATE))
1576 {
2b5e7bc2
KW
1577 pack_warn = packWARN(WARN_SURROGATE);
1578
1579 /* These are the only errors that can occur with a
1580 * surrogate when the 'uv' isn't valid */
1581 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1582 message = Perl_form(aTHX_
1583 "UTF-16 surrogate (any UTF-8 sequence that"
1584 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1585 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1586 }
1587 else {
c94c2f39 1588 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1589 }
f9380377 1590 }
2b5e7bc2 1591 }
ba210ebe 1592
2b5e7bc2
KW
1593 if (flags & UTF8_DISALLOW_SURROGATE) {
1594 disallowed = TRUE;
f9380377 1595 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1596 }
1597 }
1598 else if (possible_problems & UTF8_GOT_SUPER) {
1599 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1600
f9380377
KW
1601 if (flags & UTF8_WARN_SUPER) {
1602 *errors |= UTF8_GOT_SUPER;
1603
1604 if ( ! (flags & UTF8_CHECK_ONLY)
1605 && ckWARN_d(WARN_NON_UNICODE))
1606 {
2b5e7bc2
KW
1607 pack_warn = packWARN(WARN_NON_UNICODE);
1608
1609 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1610 message = Perl_form(aTHX_
1611 "Any UTF-8 sequence that starts with"
1612 " \"%s\" is for a non-Unicode code point,"
1613 " may not be portable",
7e2f38b2 1614 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1615 }
1616 else {
c94c2f39 1617 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1618 }
f9380377 1619 }
2b5e7bc2 1620 }
ba210ebe 1621
57ff5f59
KW
1622 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1623 * and before possibly bailing out, so that the more dire
1624 * warning will override the regular one. */
1625 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1626 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1627 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1628 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1629 {
db0f09e6 1630 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1631
57ff5f59
KW
1632 /* If it is an overlong that evaluates to a code point
1633 * that doesn't have to use the Perl extended UTF-8, it
1634 * still used it, and so we output a message that
1635 * doesn't refer to the code point. The same is true
1636 * if there was a SHORT malformation where the code
1637 * point is not valid. In that case, 'uv' will have
1638 * been set to the REPLACEMENT CHAR, and the message
1639 * below without the code point in it will be selected
1640 * */
1641 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1642 message = Perl_form(aTHX_
57ff5f59 1643 perl_extended_cp_format, uv);
2b5e7bc2
KW
1644 }
1645 else {
1646 message = Perl_form(aTHX_
57ff5f59
KW
1647 "Any UTF-8 sequence that starts with"
1648 " \"%s\" is a Perl extension, and"
1649 " so is not portable",
1650 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1651 }
1652 }
1653
d044b7a7
KW
1654 if (flags & ( UTF8_WARN_PERL_EXTENDED
1655 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1656 {
d044b7a7 1657 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1658
d044b7a7 1659 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1660 disallowed = TRUE;
1661 }
2b5e7bc2
KW
1662 }
1663 }
eb83ed87 1664
2b5e7bc2 1665 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1666 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1667 disallowed = TRUE;
1668 }
eb83ed87 1669
2b5e7bc2
KW
1670 /* The deprecated warning overrides any non-deprecated one. If
1671 * there are other problems, a deprecation message is not
1672 * really helpful, so don't bother to raise it in that case.
1673 * This also keeps the code from having to handle the case
1674 * where 'uv' is not valid. */
1675 if ( ! (orig_problems
1676 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1677 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1678 && ckWARN_d(WARN_DEPRECATED))
1679 {
1680 message = Perl_form(aTHX_ cp_above_legal_max,
1681 uv, MAX_NON_DEPRECATED_CP);
1682 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1683 }
1684 }
1685 else if (possible_problems & UTF8_GOT_NONCHAR) {
1686 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1687
f9380377
KW
1688 if (flags & UTF8_WARN_NONCHAR) {
1689 *errors |= UTF8_GOT_NONCHAR;
1690
1691 if ( ! (flags & UTF8_CHECK_ONLY)
1692 && ckWARN_d(WARN_NONCHAR))
1693 {
2b5e7bc2
KW
1694 /* The code above should have guaranteed that we don't
1695 * get here with errors other than overlong */
1696 assert (! (orig_problems
1697 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1698
1699 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1700 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1701 }
2b5e7bc2 1702 }
5b311467 1703
2b5e7bc2
KW
1704 if (flags & UTF8_DISALLOW_NONCHAR) {
1705 disallowed = TRUE;
f9380377 1706 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1707 }
6c64cd9d
KW
1708 }
1709 else if (possible_problems & UTF8_GOT_LONG) {
1710 possible_problems &= ~UTF8_GOT_LONG;
1711 *errors |= UTF8_GOT_LONG;
1712
1713 if (flags & UTF8_ALLOW_LONG) {
1714
1715 /* We don't allow the actual overlong value, unless the
1716 * special extra bit is also set */
1717 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1718 & ~UTF8_ALLOW_LONG)))
1719 {
1720 uv = UNICODE_REPLACEMENT;
1721 }
1722 }
1723 else {
1724 disallowed = TRUE;
1725
1726 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1727 pack_warn = packWARN(WARN_UTF8);
1728
1729 /* These error types cause 'uv' to be something that
1730 * isn't what was intended, so can't use it in the
1731 * message. The other error types either can't
1732 * generate an overlong, or else the 'uv' is valid */
1733 if (orig_problems &
1734 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1735 {
1736 message = Perl_form(aTHX_
1737 "%s: %s (any UTF-8 sequence that starts"
1738 " with \"%s\" is overlong which can and"
1739 " should be represented with a"
1740 " different, shorter sequence)",
1741 malformed_text,
1742 _byte_dump_string(s0, send - s0, 0),
1743 _byte_dump_string(s0, curlen, 0));
1744 }
1745 else {
1746 U8 tmpbuf[UTF8_MAXBYTES+1];
1747 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1748 uv, 0);
1749 const char * preface = (uv <= PERL_UNICODE_MAX)
1750 ? "U+"
1751 : "0x";
1752 message = Perl_form(aTHX_
1753 "%s: %s (overlong; instead use %s to represent"
1754 " %s%0*" UVXf ")",
1755 malformed_text,
1756 _byte_dump_string(s0, send - s0, 0),
1757 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1758 preface,
1759 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1760 small code points */
1761 uv);
1762 }
1763 }
1764 }
2b5e7bc2
KW
1765 } /* End of looking through the possible flags */
1766
1767 /* Display the message (if any) for the problem being handled in
1768 * this iteration of the loop */
1769 if (message) {
1770 if (PL_op)
1771 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1772 OP_DESC(PL_op));
1773 else
1774 Perl_warner(aTHX_ pack_warn, "%s", message);
1775 }
ddb65933 1776 } /* End of 'while (possible_problems)' */
a0dbb045 1777
2b5e7bc2
KW
1778 /* Since there was a possible problem, the returned length may need to
1779 * be changed from the one stored at the beginning of this function.
1780 * Instead of trying to figure out if that's needed, just do it. */
1781 if (retlen) {
1782 *retlen = curlen;
1783 }
a0dbb045 1784
2b5e7bc2
KW
1785 if (disallowed) {
1786 if (flags & UTF8_CHECK_ONLY && retlen) {
1787 *retlen = ((STRLEN) -1);
1788 }
1789 return 0;
1790 }
eb83ed87 1791 }
ba210ebe 1792
2b5e7bc2 1793 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1794}
1795
8e84507e 1796/*
ec5f19d0
KW
1797=for apidoc utf8_to_uvchr_buf
1798
1799Returns the native code point of the first character in the string C<s> which
1800is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1801C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1802
524080c4
KW
1803If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1804enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1805C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1806(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1807C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1808the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1809See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1810returned.
ec5f19d0 1811
760c7c2f
KW
1812Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1813unless those are turned off.
1814
ec5f19d0 1815=cut
52be2536
KW
1816
1817Also implemented as a macro in utf8.h
1818
ec5f19d0
KW
1819*/
1820
1821
1822UV
1823Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1824{
7f974d7e
KW
1825 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1826
ec5f19d0
KW
1827 assert(s < send);
1828
1829 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1830 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1831}
1832
52be2536
KW
1833/* This is marked as deprecated
1834 *
ec5f19d0
KW
1835=for apidoc utf8_to_uvuni_buf
1836
de69f3af
KW
1837Only in very rare circumstances should code need to be dealing in Unicode
1838(as opposed to native) code points. In those few cases, use
1839C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1840
1841Returns the Unicode (not-native) code point of the first character in the
1842string C<s> which
ec5f19d0
KW
1843is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1844C<retlen> will be set to the length, in bytes, of that character.
1845
524080c4
KW
1846If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1847enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1848NULL) to -1. If those warnings are off, the computed value if well-defined (or
1849the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1850is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1851next possible position in C<s> that could begin a non-malformed character.
de69f3af 1852See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1853
760c7c2f
KW
1854Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1855unless those are turned off.
1856
ec5f19d0
KW
1857=cut
1858*/
1859
1860UV
1861Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1862{
1863 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1864
1865 assert(send > s);
1866
5962d97e
KW
1867 /* Call the low level routine, asking for checks */
1868 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1869}
1870
b76347f2 1871/*
87cea99e 1872=for apidoc utf8_length
b76347f2
JH
1873
1874Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1875Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1876up past C<e>, croaks.
b76347f2
JH
1877
1878=cut
1879*/
1880
1881STRLEN
35a4481c 1882Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1883{
1884 STRLEN len = 0;
1885
7918f24d
NC
1886 PERL_ARGS_ASSERT_UTF8_LENGTH;
1887
8850bf83
JH
1888 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1889 * the bitops (especially ~) can create illegal UTF-8.
1890 * In other words: in Perl UTF-8 is not just for Unicode. */
1891
a3b680e6
AL
1892 if (e < s)
1893 goto warn_and_return;
b76347f2 1894 while (s < e) {
4cbf4130 1895 s += UTF8SKIP(s);
8e91ec7f
AV
1896 len++;
1897 }
1898
1899 if (e != s) {
1900 len--;
1901 warn_and_return:
9b387841
NC
1902 if (PL_op)
1903 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1904 "%s in %s", unees, OP_DESC(PL_op));
1905 else
61a12c31 1906 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1907 }
1908
1909 return len;
1910}
1911
b06226ff 1912/*
fed3ba5d
NC
1913=for apidoc bytes_cmp_utf8
1914
a1433954 1915Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1916sequence of characters (stored as UTF-8)
1917in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1918equal, -1 or -2 if the first string is less than the second string, +1 or +2
1919if the first string is greater than the second string.
1920
1921-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1922longer string. -2 or +2 is returned if
1923there was a difference between characters
fed3ba5d
NC
1924within the strings.
1925
1926=cut
1927*/
1928
1929int
1930Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1931{
1932 const U8 *const bend = b + blen;
1933 const U8 *const uend = u + ulen;
1934
1935 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1936
1937 while (b < bend && u < uend) {
1938 U8 c = *u++;
1939 if (!UTF8_IS_INVARIANT(c)) {
1940 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1941 if (u < uend) {
1942 U8 c1 = *u++;
1943 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1944 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1945 } else {
2b5e7bc2 1946 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1947 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1948 "%s %s%s",
1949 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1950 PL_op ? " in " : "",
1951 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1952 return -2;
1953 }
1954 } else {
1955 if (PL_op)
1956 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1957 "%s in %s", unees, OP_DESC(PL_op));
1958 else
61a12c31 1959 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1960 return -2; /* Really want to return undef :-) */
1961 }
1962 } else {
1963 return -2;
1964 }
1965 }
1966 if (*b != c) {
1967 return *b < c ? -2 : +2;
1968 }
1969 ++b;
1970 }
1971
1972 if (b == bend && u == uend)
1973 return 0;
1974
1975 return b < bend ? +1 : -1;
1976}
1977
1978/*
87cea99e 1979=for apidoc utf8_to_bytes
6940069f 1980
3bc0c78c 1981Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 1982Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 1983updates C<*lenp> to contain the new length.
3bc0c78c
KW
1984Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
1985
1986Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
1987having saved the value of C<*lenp> before the call, and subtracting the
1988after-call value of C<*lenp> from it.
6940069f 1989
a1433954 1990If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1991
6940069f
GS
1992=cut
1993*/
1994
1995U8 *
09af0336 1996Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 1997{
9fe0d3c2 1998 U8 * first_variant;
246fae53 1999
7918f24d 2000 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2001 PERL_UNUSED_CONTEXT;
7918f24d 2002
9fe0d3c2 2003 /* This is a no-op if no variants at all in the input */
09af0336 2004 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2005 return s;
2006 }
2007
2008 {
3c5aa262 2009 U8 * const save = s;
09af0336 2010 U8 * const send = s + *lenp;
3c5aa262
KW
2011 U8 * d;
2012
2013 /* Nothing before the first variant needs to be changed, so start the real
2014 * work there */
2015 s = first_variant;
2016 while (s < send) {
2017 if (! UTF8_IS_INVARIANT(*s)) {
2018 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2019 *lenp = ((STRLEN) -1);
3c5aa262
KW
2020 return 0;
2021 }
2022 s++;
d59937ca
KW
2023 }
2024 s++;
dcad2880 2025 }
dcad2880 2026
3c5aa262
KW
2027 /* Is downgradable, so do it */
2028 d = s = first_variant;
2029 while (s < send) {
2030 U8 c = *s++;
2031 if (! UVCHR_IS_INVARIANT(c)) {
2032 /* Then it is two-byte encoded */
2033 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2034 s++;
2035 }
2036 *d++ = c;
2037 }
2038 *d = '\0';
09af0336 2039 *lenp = d - save;
3c5aa262
KW
2040
2041 return save;
9fe0d3c2 2042 }
6940069f
GS
2043}
2044
2045/*
87cea99e 2046=for apidoc bytes_from_utf8
f9a63242 2047
09af0336 2048Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2049byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2050actually encoded in UTF-8.
2051
2052Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2053the input string.
2054
41ae6089
KW
2055Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2056not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2057C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2058
41ae6089 2059Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2060newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2061returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2062
3bc0c78c 2063Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2064having saved the value of C<*lenp> before the call, and subtracting the
2065after-call value of C<*lenp> from it.
3bc0c78c 2066
37607a96 2067=cut
976c1b08
KW
2068
2069There is a macro that avoids this function call, but this is retained for
2070anyone who calls it with the Perl_ prefix */
f9a63242
JH
2071
2072U8 *
41ae6089 2073Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2074{
7918f24d 2075 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2076 PERL_UNUSED_CONTEXT;
f9a63242 2077
976c1b08
KW
2078 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2079}
2080
2081/*
2082No = here because currently externally undocumented
2083for apidoc bytes_from_utf8_loc
2084
2085Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2086to store the location of the first character in C<"s"> that cannot be
2087converted to non-UTF8.
2088
2089If that parameter is C<NULL>, this function behaves identically to
2090C<bytes_from_utf8>.
2091
2092Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2093C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2094
2095Otherwise, the function returns a newly created C<NUL>-terminated string
2096containing the non-UTF8 equivalent of the convertible first portion of
2097C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2098If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2099and C<*first_non_downgradable> is set to C<NULL>.
2100
2101Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2102first character in the original string that wasn't converted. C<*is_utf8p> is
2103unchanged. Note that the new string may have length 0.
2104
2105Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2106C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2107converts as many characters in it as possible stopping at the first one it
385b74be 2108finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2109set to point to that. The function returns the portion that could be converted
2110in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2111not including the terminating C<NUL>. If the very first character in the
2112original could not be converted, C<*lenp> will be 0, and the new string will
2113contain just a single C<NUL>. If the entire input string was converted,
2114C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2115
2116Upon successful return, the number of variants in the converted portion of the
2117string can be computed by having saved the value of C<*lenp> before the call,
2118and subtracting the after-call value of C<*lenp> from it.
2119
2120=cut
2121
2122
2123*/
2124
2125U8 *
2126Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2127{
2128 U8 *d;
2129 const U8 *original = s;
2130 U8 *converted_start;
2131 const U8 *send = s + *lenp;
f9a63242 2132
976c1b08 2133 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2134
976c1b08
KW
2135 if (! *is_utf8p) {
2136 if (first_unconverted) {
2137 *first_unconverted = NULL;
2138 }
2139
2140 return (U8 *) original;
2141 }
2142
2143 Newx(d, (*lenp) + 1, U8);
2144
2145 converted_start = d;
7299a045
KW
2146 while (s < send) {
2147 U8 c = *s++;
2148 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2149
2150 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2151 * have to stop now */
2152 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2153 if (first_unconverted) {
2154 *first_unconverted = s - 1;
2155 goto finish_and_return;
2156 }
2157 else {
2158 Safefree(converted_start);
2159 return (U8 *) original;
2160 }
2161 }
2162
7299a045
KW
2163 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2164 s++;
38af28cf 2165 }
7299a045
KW
2166 *d++ = c;
2167 }
170a1c22 2168
976c1b08
KW
2169 /* Here, converted the whole of the input */
2170 *is_utf8p = FALSE;
2171 if (first_unconverted) {
2172 *first_unconverted = NULL;
170a1c22 2173 }
976c1b08
KW
2174
2175 finish_and_return:
2176 *d = '\0';
2177 *lenp = d - converted_start;
2178
2179 /* Trim unused space */
2180 Renew(converted_start, *lenp + 1, U8);
2181
2182 return converted_start;
f9a63242
JH
2183}
2184
2185/*
87cea99e 2186=for apidoc bytes_to_utf8
6940069f 2187
09af0336 2188Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2189UTF-8.
09af0336 2190Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2191reflect the new length in bytes.
6940069f 2192
3bc0c78c 2193Upon successful return, the number of variants in the string can be computed by
23b37b12 2194having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2195after-call value of C<*lenp>.
2196
75200dff 2197A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2198
2199If you want to convert to UTF-8 from encodings other than
2200the native (Latin1 or EBCDIC),
a1433954 2201see L</sv_recode_to_utf8>().
c9ada85f 2202
497711e7 2203=cut
6940069f
GS
2204*/
2205
2206U8*
09af0336 2207Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2208{
09af0336 2209 const U8 * const send = s + (*lenp);
6940069f
GS
2210 U8 *d;
2211 U8 *dst;
7918f24d
NC
2212
2213 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2214 PERL_UNUSED_CONTEXT;
6940069f 2215
09af0336 2216 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2217 dst = d;
2218
2219 while (s < send) {
55d09dc8
KW
2220 append_utf8_from_native_byte(*s, &d);
2221 s++;
6940069f
GS
2222 }
2223 *d = '\0';
09af0336 2224 *lenp = d-dst;
6940069f
GS
2225 return dst;
2226}
2227
a0ed51b3 2228/*
dea0fc0b 2229 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2230 *
2231 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2232 * We optimize for native, for obvious reasons. */
2233
2234U8*
dea0fc0b 2235Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2236{
dea0fc0b
JH
2237 U8* pend;
2238 U8* dstart = d;
2239
7918f24d
NC
2240 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2241
dea0fc0b 2242 if (bytelen & 1)
56576a04
KW
2243 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2244 (UV)bytelen);
dea0fc0b
JH
2245
2246 pend = p + bytelen;
2247
a0ed51b3 2248 while (p < pend) {
dea0fc0b
JH
2249 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2250 p += 2;
2d1545e5 2251 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2252 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2253 continue;
2254 }
56d37426
KW
2255 if (uv <= MAX_UTF8_TWO_BYTE) {
2256 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2257 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2258 continue;
2259 }
46956fad
KW
2260#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2261#define LAST_HIGH_SURROGATE 0xDBFF
2262#define FIRST_LOW_SURROGATE 0xDC00
2263#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2264
2265 /* This assumes that most uses will be in the first Unicode plane, not
2266 * needing surrogates */
2267 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2268 && uv <= UNICODE_SURROGATE_LAST))
2269 {
2270 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2271 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2272 }
2273 else {
01ea242b 2274 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2275 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2276 || UNLIKELY(low > LAST_LOW_SURROGATE))
2277 {
01ea242b 2278 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2279 }
2280 p += 2;
46956fad
KW
2281 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2282 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2283 }
a0ed51b3 2284 }
56d37426
KW
2285#ifdef EBCDIC
2286 d = uvoffuni_to_utf8_flags(d, uv, 0);
2287#else
a0ed51b3 2288 if (uv < 0x10000) {
eb160463
GS
2289 *d++ = (U8)(( uv >> 12) | 0xe0);
2290 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2291 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2292 continue;
2293 }
2294 else {
eb160463
GS
2295 *d++ = (U8)(( uv >> 18) | 0xf0);
2296 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2297 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2298 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2299 continue;
2300 }
56d37426 2301#endif
a0ed51b3 2302 }
dea0fc0b 2303 *newlen = d - dstart;
a0ed51b3
LW
2304 return d;
2305}
2306
2307/* Note: this one is slightly destructive of the source. */
2308
2309U8*
dea0fc0b 2310Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2311{
2312 U8* s = (U8*)p;
d4c19fe8 2313 U8* const send = s + bytelen;
7918f24d
NC
2314
2315 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2316
e0ea5e2d 2317 if (bytelen & 1)
147e3846 2318 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2319 (UV)bytelen);
2320
a0ed51b3 2321 while (s < send) {
d4c19fe8 2322 const U8 tmp = s[0];
a0ed51b3
LW
2323 s[0] = s[1];
2324 s[1] = tmp;
2325 s += 2;
2326 }
dea0fc0b 2327 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2328}
2329
922e8cb4
KW
2330bool
2331Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2332{
2333 U8 tmpbuf[UTF8_MAXBYTES+1];
2334 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2335 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2336}
2337
f9ae8fb6
JD
2338/* Internal function so we can deprecate the external one, and call
2339 this one from other deprecated functions in this file */
2340
f2645549
KW
2341bool
2342Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2343{
f2645549 2344 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2345
2346 if (*p == '_')
2347 return TRUE;
f25ce844 2348 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2349}
2350
5092f92a 2351bool
eba68aa0
KW
2352Perl__is_uni_perl_idcont(pTHX_ UV c)
2353{
2354 U8 tmpbuf[UTF8_MAXBYTES+1];
2355 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2356 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2357}
2358
2359bool
f91dcd13
KW
2360Perl__is_uni_perl_idstart(pTHX_ UV c)
2361{
2362 U8 tmpbuf[UTF8_MAXBYTES+1];
2363 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2364 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2365}
2366
3a4c58c9 2367UV
56576a04
KW
2368Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2369 const char S_or_s)
3a4c58c9
KW
2370{
2371 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2372 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2373 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2374 * either "SS" or "Ss". Which one to use is passed into the routine in
2375 * 'S_or_s' to avoid a test */
2376
2377 UV converted = toUPPER_LATIN1_MOD(c);
2378
2379 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2380
2381 assert(S_or_s == 'S' || S_or_s == 's');
2382
6f2d5cbc 2383 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2384 characters in this range */
3a4c58c9
KW
2385 *p = (U8) converted;
2386 *lenp = 1;
2387 return converted;
2388 }
2389
2390 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2391 * which it maps to one of them, so as to only have to have one check for
2392 * it in the main case */
2393 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2394 switch (c) {
2395 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2396 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2397 break;
2398 case MICRO_SIGN:
2399 converted = GREEK_CAPITAL_LETTER_MU;
2400 break;
79e064b9
KW
2401#if UNICODE_MAJOR_VERSION > 2 \
2402 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2403 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2404 case LATIN_SMALL_LETTER_SHARP_S:
2405 *(p)++ = 'S';
2406 *p = S_or_s;
2407 *lenp = 2;
2408 return 'S';
79e064b9 2409#endif
3a4c58c9 2410 default:
56576a04
KW
2411 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2412 " '%c' to map to '%c'",
2413 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2414 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2415 }
2416 }
2417
2418 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2419 *p = UTF8_TWO_BYTE_LO(converted);
2420 *lenp = 2;
2421
2422 return converted;
2423}
2424
50bda2c3
KW
2425/* Call the function to convert a UTF-8 encoded character to the specified case.
2426 * Note that there may be more than one character in the result.
2427 * INP is a pointer to the first byte of the input character
2428 * OUTP will be set to the first byte of the string of changed characters. It
2429 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2430 * LENP will be set to the length in bytes of the string of changed characters
2431 *
56576a04
KW
2432 * The functions return the ordinal of the first character in the string of
2433 * OUTP */
2434#define CALL_UPPER_CASE(uv, s, d, lenp) \
2435 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2436#define CALL_TITLE_CASE(uv, s, d, lenp) \
2437 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2438#define CALL_LOWER_CASE(uv, s, d, lenp) \
2439 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2440
b9992569
KW
2441/* This additionally has the input parameter 'specials', which if non-zero will
2442 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2443 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2444#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2445_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2446
84afefe6
JH
2447UV
2448Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2449{
a1433954
KW
2450 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2451 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2452 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2453 * the changed version may be longer than the original character.
2454 *
2455 * The ordinal of the first character of the changed version is returned
2456 * (but note, as explained above, that there may be more.) */
2457
7918f24d
NC
2458 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2459
3a4c58c9
KW
2460 if (c < 256) {
2461 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2462 }
2463
0ebc6274 2464 uvchr_to_utf8(p, c);
b9992569 2465 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2466}
2467
84afefe6
JH
2468UV
2469Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2470{
7918f24d
NC
2471 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2472
3a4c58c9
KW
2473 if (c < 256) {
2474 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2475 }
2476
0ebc6274 2477 uvchr_to_utf8(p, c);
b9992569 2478 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2479}
2480
afc16117 2481STATIC U8
eaf412bf 2482S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2483{
2484 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2485 * those, converting the result to UTF-8. Since the result is always just
a1433954 2486 * one character, we allow <p> to be NULL */
afc16117
KW
2487
2488 U8 converted = toLOWER_LATIN1(c);
2489
eaf412bf
KW
2490 PERL_UNUSED_ARG(dummy);
2491
afc16117 2492 if (p != NULL) {
6f2d5cbc 2493 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2494 *p = converted;
2495 *lenp = 1;
2496 }
2497 else {
430c9760
KW
2498 /* Result is known to always be < 256, so can use the EIGHT_BIT
2499 * macros */
2500 *p = UTF8_EIGHT_BIT_HI(converted);
2501 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2502 *lenp = 2;
2503 }
2504 }
2505 return converted;
2506}
2507
84afefe6
JH
2508UV
2509Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2510{
7918f24d
NC
2511 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2512
afc16117 2513 if (c < 256) {
eaf412bf 2514 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2515 }
2516
afc16117 2517 uvchr_to_utf8(p, c);
b9992569 2518 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2519}
2520
84afefe6 2521UV
56576a04
KW
2522Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2523 const unsigned int flags)
a1dde8de 2524{
51910141 2525 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2526 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2527 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2528 *
2529 * Not to be used for locale folds
51910141 2530 */
f673fad4 2531
a1dde8de
KW
2532 UV converted;
2533
2534 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2535 PERL_UNUSED_CONTEXT;
a1dde8de 2536
1ca267a5
KW
2537 assert (! (flags & FOLD_FLAGS_LOCALE));
2538
659a7c2d 2539 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2540 converted = GREEK_SMALL_LETTER_MU;
2541 }
9b63e895
KW
2542#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2543 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2544 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2545 else if ( (flags & FOLD_FLAGS_FULL)
2546 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2547 {
1ca267a5
KW
2548 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2549 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2550 * under those circumstances. */
2551 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2552 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2553 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2554 p, *lenp, U8);
2555 return LATIN_SMALL_LETTER_LONG_S;
2556 }
2557 else {
4f489194
KW
2558 *(p)++ = 's';
2559 *p = 's';
2560 *lenp = 2;
2561 return 's';
1ca267a5 2562 }
a1dde8de 2563 }
9b63e895 2564#endif
a1dde8de
KW
2565 else { /* In this range the fold of all other characters is their lower
2566 case */
2567 converted = toLOWER_LATIN1(c);
2568 }
2569
6f2d5cbc 2570 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2571 *p = (U8) converted;
2572 *lenp = 1;
2573 }
2574 else {
2575 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2576 *p = UTF8_TWO_BYTE_LO(converted);
2577 *lenp = 2;
2578 }
2579
2580 return converted;
2581}
2582
2583UV
31f05a37 2584Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2585{
4b593389 2586
a0270393
KW
2587 /* Not currently externally documented, and subject to change
2588 * <flags> bits meanings:
2589 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2590 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2591 * locale are to be used.
a0270393
KW
2592 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2593 */
4b593389 2594
36bb2ab6 2595 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2596
780fcc9f
KW
2597 if (flags & FOLD_FLAGS_LOCALE) {
2598 /* Treat a UTF-8 locale as not being in locale at all */
2599 if (IN_UTF8_CTYPE_LOCALE) {
2600 flags &= ~FOLD_FLAGS_LOCALE;
2601 }
2602 else {
2603 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2604 goto needs_full_generality;
780fcc9f 2605 }
31f05a37
KW
2606 }
2607
a1dde8de 2608 if (c < 256) {
e7b7ac46 2609 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2610 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2611 }
2612
2f306ab9 2613 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2614 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2615 uvchr_to_utf8(p, c);
b9992569 2616 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2617 }
567b353c 2618 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2619 the special flags. */
2620 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2621
2622 needs_full_generality:
a0270393 2623 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2624 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2625 p, lenp, flags);
a0270393 2626 }
84afefe6
JH
2627}
2628
26483009 2629PERL_STATIC_INLINE bool
5141f98e 2630S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2631 const char *const swashname, SV* const invlist)
bde6a22d 2632{
ea317ccb
KW
2633 /* returns a boolean giving whether or not the UTF8-encoded character that
2634 * starts at <p> is in the swash indicated by <swashname>. <swash>
2635 * contains a pointer to where the swash indicated by <swashname>
2636 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2637 * look at <*swash> and only generate a swash if it is not null. <invlist>
2638 * is NULL or an inversion list that defines the swash. If not null, it
2639 * saves time during initialization of the swash.
ea317ccb
KW
2640 *
2641 * Note that it is assumed that the buffer length of <p> is enough to
2642 * contain all the bytes that comprise the character. Thus, <*p> should
2643 * have been checked before this call for mal-formedness enough to assure
2644 * that. */
2645
7918f24d
NC
2646 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2647
492a624f 2648 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2649 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2650 * as far as there being enough bytes available in it to accommodate the
2651 * character without reading beyond the end, and pass that number on to the
2652 * validating routine */
6302f837 2653 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2654 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2655 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2656 1 /* Die */ );
2657 NOT_REACHED; /* NOTREACHED */
28123549 2658 }
86ae6e94 2659
87367d5f
KW
2660 if (!*swash) {
2661 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2662 *swash = _core_swash_init("utf8",
2663
2664 /* Only use the name if there is no inversion
2665 * list; otherwise will go out to disk */
2666 (invlist) ? "" : swashname,
2667
2668 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2669 }
28123549 2670
bde6a22d
NC
2671 return swash_fetch(*swash, p, TRUE) != 0;
2672}
2673
da8c1a98 2674PERL_STATIC_INLINE bool
56576a04
KW
2675S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2676 SV **swash, const char *const swashname,
2677 SV* const invlist)
da8c1a98
KW
2678{
2679 /* returns a boolean giving whether or not the UTF8-encoded character that
2680 * starts at <p>, and extending no further than <e - 1> is in the swash
2681 * indicated by <swashname>. <swash> contains a pointer to where the swash
2682 * indicated by <swashname> is to be stored; which this routine will do, so
2683 * that future calls will look at <*swash> and only generate a swash if it
2684 * is not null. <invlist> is NULL or an inversion list that defines the
2685 * swash. If not null, it saves time during initialization of the swash.
2686 */
2687
2688 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2689
2690 if (! isUTF8_CHAR(p, e)) {
2691 _force_out_malformed_utf8_message(p, e, 0, 1);
2692 NOT_REACHED; /* NOTREACHED */
2693 }
2694
2695 if (!*swash) {
2696 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2697 *swash = _core_swash_init("utf8",
2698
2699 /* Only use the name if there is no inversion
2700 * list; otherwise will go out to disk */
2701 (invlist) ? "" : swashname,
2702
2703 &PL_sv_undef, 1, 0, invlist, &flags);
2704 }
2705
2706 return swash_fetch(*swash, p, TRUE) != 0;
2707}
2708
34aeb2e9
KW
2709STATIC void
2710S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2711 const char * const alternative,
2712 const bool use_locale,
2713 const char * const file,
2714 const unsigned line)
2715{
2716 const char * key;
2717
2718 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2719
2720 if (ckWARN_d(WARN_DEPRECATED)) {
2721
2722 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2723 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2724 if (! PL_seen_deprecated_macro) {
2725 PL_seen_deprecated_macro = newHV();
2726 }
2727 if (! hv_store(PL_seen_deprecated_macro, key,
2728 strlen(key), &PL_sv_undef, 0))
2729 {
2730 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2731 }
2732
c44e9413 2733 if (instr(file, "mathoms.c")) {
607313a1
KW
2734 Perl_warner(aTHX_ WARN_DEPRECATED,
2735 "In %s, line %d, starting in Perl v5.30, %s()"
2736 " will be removed. Avoid this message by"
2737 " converting to use %s().\n",
2738 file, line, name, alternative);
2739 }
2740 else {
34aeb2e9
KW
2741 Perl_warner(aTHX_ WARN_DEPRECATED,
2742 "In %s, line %d, starting in Perl v5.30, %s() will"
2743 " require an additional parameter. Avoid this"
2744 " message by converting to use %s().\n",
2745 file, line, name, alternative);
607313a1 2746 }
34aeb2e9
KW
2747 }
2748 }
2749}
2750
bde6a22d 2751bool
34aeb2e9 2752Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2753 const U8 * const p,
34aeb2e9
KW
2754 const char * const name,
2755 const char * const alternative,
2756 const bool use_utf8,
2757 const bool use_locale,
2758 const char * const file,
2759 const unsigned line)
922e8cb4 2760{
922e8cb4
KW
2761 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2762
34aeb2e9
KW
2763 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2764
2765 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2766
2767 switch (classnum) {
2768 case _CC_WORDCHAR:
2769 case _CC_DIGIT:
2770 case _CC_ALPHA:
2771 case _CC_LOWER:
2772 case _CC_UPPER:
2773 case _CC_PUNCT:
2774 case _CC_PRINT:
2775 case _CC_ALPHANUMERIC:
2776 case _CC_GRAPH:
2777 case _CC_CASED:
2778
2779 return is_utf8_common(p,
2780 &PL_utf8_swash_ptrs[classnum],
2781 swash_property_names[classnum],
2782 PL_XPosix_ptrs[classnum]);
2783
2784 case _CC_SPACE:
2785 return is_XPERLSPACE_high(p);
2786 case _CC_BLANK:
2787 return is_HORIZWS_high(p);
2788 case _CC_XDIGIT:
2789 return is_XDIGIT_high(p);
2790 case _CC_CNTRL:
2791 return 0;
2792 case _CC_ASCII:
2793 return 0;
2794 case _CC_VERTSPACE:
2795 return is_VERTWS_high(p);
2796 case _CC_IDFIRST:
2797 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2798 PL_utf8_perl_idstart
2799 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2800 }
22f0498f
KW
2801 return is_utf8_common(p, &PL_utf8_perl_idstart,
2802 "_Perl_IDStart", NULL);
34aeb2e9
KW
2803 case _CC_IDCONT:
2804 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2805 PL_utf8_perl_idcont
2806 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2807 }
22f0498f
KW
2808 return is_utf8_common(p, &PL_utf8_perl_idcont,
2809 "_Perl_IDCont", NULL);
34aeb2e9
KW
2810 }
2811 }
2812
2813 /* idcont is the same as wordchar below 256 */
2814 if (classnum == _CC_IDCONT) {
2815 classnum = _CC_WORDCHAR;
2816 }
2817 else if (classnum == _CC_IDFIRST) {
2818 if (*p == '_') {
2819 return TRUE;
2820 }
2821 classnum = _CC_ALPHA;
2822 }
2823
2824 if (! use_locale) {
2825 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2826 return _generic_isCC(*p, classnum);
2827 }
922e8cb4 2828
34aeb2e9
KW
2829 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2830 }
2831 else {
2832 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2833 return isFOO_lc(classnum, *p);
2834 }
2835
2836 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2837 }
2838
2839 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2840}
2841
2842bool
da8c1a98
KW
2843Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2844 const U8 * const e)
2845{
2846 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2847
2848 assert(classnum < _FIRST_NON_SWASH_CC);
2849
2850 return is_utf8_common_with_len(p,
2851 e,
2852 &PL_utf8_swash_ptrs[classnum],
2853 swash_property_names[classnum],
2854 PL_XPosix_ptrs[classnum]);
2855}
2856
2857bool
da8c1a98
KW
2858Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2859{
2860 SV* invlist = NULL;
2861
2862 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2863
2864 if (! PL_utf8_perl_idstart) {
2865 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2866 }
2867 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2868 "_Perl_IDStart", invlist);
2869}
2870
2871bool
f2645549 2872Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2873{
f2645549 2874 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2875
2876 if (*p == '_')
2877 return TRUE;
f25ce844 2878 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2879}
2880
2881bool
da8c1a98
KW
2882Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2883{
2884 SV* invlist = NULL;
2885
2886 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2887
2888 if (! PL_utf8_perl_idcont) {
2889 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2890 }
2891 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2892 "_Perl_IDCont", invlist);
2893}
2894
2895bool
f2645549 2896Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2897{
f2645549 2898 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2899
f25ce844 2900 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2901}
2902
2903bool
f2645549 2904Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2905{
f2645549 2906 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2907
f25ce844 2908 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2909}
2910
2911bool
7dbf68d2
KW
2912Perl__is_utf8_mark(pTHX_ const U8 *p)
2913{
7dbf68d2
KW
2914 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2915
f25ce844 2916 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2917}
2918
b9992569 2919 /* change namve uv1 to 'from' */
6a4a25f4 2920STATIC UV
b9992569
KW
2921S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2922 SV **swashp, const char *normal, const char *special)
2923{
0134edef 2924 STRLEN len = 0;
7918f24d 2925
b9992569 2926 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2927
36eaa811
KW
2928 /* For code points that don't change case, we already know that the output
2929 * of this function is the unchanged input, so we can skip doing look-ups
2930 * for them. Unfortunately the case-changing code points are scattered
2931 * around. But there are some long consecutive ranges where there are no
2932 * case changing code points. By adding tests, we can eliminate the lookup
2933 * for all the ones in such ranges. This is currently done here only for
2934 * just a few cases where the scripts are in common use in modern commerce
2935 * (and scripts adjacent to those which can be included without additional
2936 * tests). */
2937
2938 if (uv1 >= 0x0590) {
2939 /* This keeps from needing further processing the code points most
2940 * likely to be used in the following non-cased scripts: Hebrew,
2941 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2942 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2943 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2944 if (uv1 < 0x10A0) {
2945 goto cases_to_self;
2946 }
2947
2948 /* The following largish code point ranges also don't have case
2949 * changes, but khw didn't think they warranted extra tests to speed
2950 * them up (which would slightly slow down everything else above them):
2951 * 1100..139F Hangul Jamo, Ethiopic
2952 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2953 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2954 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2955 * Combining Diacritical Marks Extended, Balinese,
2956 * Sundanese, Batak, Lepcha, Ol Chiki
2957 * 2000..206F General Punctuation
2958 */
2959
2960 if (uv1 >= 0x2D30) {
2961
2962 /* This keeps the from needing further processing the code points
2963 * most likely to be used in the following non-cased major scripts:
2964 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2965 *
2966 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2967 * event that Unicode eventually allocates the unused block as of
2968 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2969 * that the test suite will start having failures to alert you
2970 * should that happen) */
2971 if (uv1 < 0xA640) {
2972 goto cases_to_self;
2973 }
2974
2975 if (uv1 >= 0xAC00) {
2976 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2977 if (ckWARN_d(WARN_SURROGATE)) {
2978 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2979 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
2980 "Operation \"%s\" returns its argument for"
2981 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2982 }
2983 goto cases_to_self;
2984 }
36eaa811
KW
2985
2986 /* AC00..FAFF Catches Hangul syllables and private use, plus
2987 * some others */
2988 if (uv1 < 0xFB00) {
2989 goto cases_to_self;
2990
2991 }
2992
5af9bc97 2993 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
2994 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2995 && ckWARN_d(WARN_DEPRECATED))
2996 {
2997 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2998 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
2999 }
3000 if (ckWARN_d(WARN_NON_UNICODE)) {
3001 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3002 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3003 "Operation \"%s\" returns its argument for"
3004 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3005 }
3006 goto cases_to_self;
3007 }
3bfc1e70
KW
3008#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3009 if (UNLIKELY(uv1
3010 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3011 {
3012
56576a04
KW
3013 /* As of Unicode 10.0, this means we avoid swash creation
3014 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3015 goto cases_to_self;
3016 }
3017#endif
36eaa811
KW
3018 }
3019 }
9ae3ac1a 3020
36eaa811
KW
3021 /* Note that non-characters are perfectly legal, so no warning should
3022 * be given. There are so few of them, that it isn't worth the extra
3023 * tests to avoid swash creation */
9ae3ac1a
KW
3024 }
3025
0134edef 3026 if (!*swashp) /* load on-demand */
56576a04
KW
3027 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3028 4, 0, NULL, NULL);
0134edef 3029
a6f87d8c 3030 if (special) {
0134edef 3031 /* It might be "special" (sometimes, but not always,
2a37f04d 3032 * a multicharacter mapping) */
4a8240a3 3033 HV *hv = NULL;
b08cf34e
JH
3034 SV **svp;
3035
4a8240a3
KW
3036 /* If passed in the specials name, use that; otherwise use any
3037 * given in the swash */
3038 if (*special != '\0') {
3039 hv = get_hv(special, 0);
3040 }
3041 else {
3042 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3043 if (svp) {
3044 hv = MUTABLE_HV(SvRV(*svp));
3045 }
3046 }
3047
176fe009 3048 if (hv
5f560d8a 3049 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3050 && (*svp))
3051 {
cfd0369c 3052 const char *s;
47654450 3053
cfd0369c 3054 s = SvPV_const(*svp, len);
47654450 3055 if (len == 1)
f4cd282c 3056 /* EIGHTBIT */
c80e42f3 3057 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3058 else {
d2dcd0fb 3059 Copy(s, ustrp, len, U8);
29e98929 3060 }
983ffd37 3061 }
0134edef
JH
3062 }
3063
3064 if (!len && *swashp) {
4a4088c4 3065 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3066
0134edef
JH
3067 if (uv2) {
3068 /* It was "normal" (a single character mapping). */
f4cd282c 3069 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3070 }
3071 }
1feea2c7 3072
cbe07460
KW
3073 if (len) {
3074 if (lenp) {
3075 *lenp = len;
3076 }
3077 return valid_utf8_to_uvchr(ustrp, 0);
3078 }
3079
3080 /* Here, there was no mapping defined, which means that the code point maps
3081 * to itself. Return the inputs */
e24dfe9c 3082 cases_to_self:
bfdf22ec 3083 len = UTF8SKIP(p);
ca9fab46
KW
3084 if (p != ustrp) { /* Don't copy onto itself */
3085 Copy(p, ustrp, len, U8);
3086 }
0134edef 3087
2a37f04d
JH
3088 if (lenp)
3089 *lenp = len;
3090
f4cd282c 3091 return uv1;
cbe07460 3092
a0ed51b3
LW
3093}
3094
051a06d4 3095STATIC UV
56576a04
KW
3096S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3097 U8* const ustrp, STRLEN *lenp)
051a06d4 3098{
4a4088c4 3099 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3100 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3101 * result contains a character that crosses the 255/256 boundary, disallow
3102 * the change, and return the original code point. See L<perlfunc/lc> for
3103 * why;
051a06d4 3104 *
a1433954
KW
3105 * p points to the original string whose case was changed; assumed
3106 * by this routine to be well-formed
051a06d4 3107 * result the code point of the first character in the changed-case string
56576a04
KW
3108 * ustrp points to the changed-case string (<result> represents its
3109 * first char)
051a06d4
KW
3110 * lenp points to the length of <ustrp> */
3111
3112 UV original; /* To store the first code point of <p> */
3113
3114 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3115
a4f12ed7 3116 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3117
3118 /* We know immediately if the first character in the string crosses the
3119 * boundary, so can skip */
3120 if (result > 255) {
3121
3122 /* Look at every character in the result; if any cross the
3123 * boundary, the whole thing is disallowed */
3124 U8* s = ustrp + UTF8SKIP(ustrp);
3125 U8* e = ustrp + *lenp;
3126 while (s < e) {
a4f12ed7 3127 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3128 goto bad_crossing;
3129 }
3130 s += UTF8SKIP(s);
3131 }
3132
613abc6d
KW
3133 /* Here, no characters crossed, result is ok as-is, but we warn. */
3134 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3135 return result;
3136 }
3137
7b52d656 3138 bad_crossing:
051a06d4
KW
3139
3140 /* Failed, have to return the original */
4b88fb76 3141 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3142
3143 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3144 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3145 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3146 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3147 OP_DESC(PL_op),
ab0b796c
KW
3148 original,
3149 original);
051a06d4
KW
3150 Copy(p, ustrp, *lenp, char);
3151 return original;
3152}
3153
607313a1
KW
3154STATIC U32
3155S_check_and_deprecate(pTHX_ const U8 *p,
3156 const U8 **e,
3157 const unsigned int type, /* See below */
3158 const bool use_locale, /* Is this a 'LC_'
3159 macro call? */
3160 const char * const file,
3161 const unsigned line)
3162{
3163 /* This is a temporary function to deprecate the unsafe calls to the case
3164 * changing macros and functions. It keeps all the special stuff in just
3165 * one place.
3166 *
3167 * It updates *e with the pointer to the end of the input string. If using
3168 * the old-style macros, *e is NULL on input, and so this function assumes
3169 * the input string is long enough to hold the entire UTF-8 sequence, and
3170 * sets *e accordingly, but it then returns a flag to pass the
3171 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3172 * using the full length if possible.
3173 *
3174 * It also does the assert that *e > p when *e is not NULL. This should be
3175 * migrated to the callers when this function gets deleted.
3176 *
3177 * The 'type' parameter is used for the caller to specify which case
3178 * changing function this is called from: */
3179
3180# define DEPRECATE_TO_UPPER 0
3181# define DEPRECATE_TO_TITLE 1
3182# define DEPRECATE_TO_LOWER 2
3183# define DEPRECATE_TO_FOLD 3
3184
3185 U32 utf8n_flags = 0;
3186 const char * name;
3187 const char * alternative;
3188
3189 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3190
3191 if (*e == NULL) {
3192 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3193 *e = p + UTF8SKIP(p);
3194
3195 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3196 * there. It could be part of a larger path */
607313a1 3197 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3198 name = instr(file, "mathoms.c")
607313a1
KW
3199 ? "to_utf8_upper"
3200 : "toUPPER_utf8";
3201 alternative = "toUPPER_utf8_safe";
3202 }
3203 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3204 name = instr(file, "mathoms.c")
607313a1
KW
3205 ? "to_utf8_title"
3206 : "toTITLE_utf8";
3207 alternative = "toTITLE_utf8_safe";
3208 }
3209 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3210 name = instr(file, "mathoms.c")
607313a1
KW
3211 ? "to_utf8_lower"
3212 : "toLOWER_utf8";
3213 alternative = "toLOWER_utf8_safe";
3214 }
3215 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3216 name = instr(file, "mathoms.c")
607313a1
KW
3217 ? "to_utf8_fold"
3218 : "toFOLD_utf8";
3219 alternative = "toFOLD_utf8_safe";
3220 }
3221 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3222
3223 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3224 }
3225 else {
3226 assert (p < *e);
3227 }
3228
3229 return utf8n_flags;
3230}
3231
eaf412bf
KW
3232/* The process for changing the case is essentially the same for the four case
3233 * change types, except there are complications for folding. Otherwise the
3234 * difference is only which case to change to. To make sure that they all do
3235 * the same thing, the bodies of the functions are extracted out into the
3236 * following two macros. The functions are written with the same variable
3237 * names, and these are known and used inside these macros. It would be
3238 * better, of course, to have inline functions to do it, but since different
3239 * macros are called, depending on which case is being changed to, this is not
3240 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3241 * function can start with the common start macro, then finish with its special
3242 * handling; while the other three cases can just use the common end macro.
3243 *
3244 * The algorithm is to use the proper (passed in) macro or function to change
3245 * the case for code points that are below 256. The macro is used if using
3246 * locale rules for the case change; the function if not. If the code point is
3247 * above 255, it is computed from the input UTF-8, and another macro is called
3248 * to do the conversion. If necessary, the output is converted to UTF-8. If
3249 * using a locale, we have to check that the change did not cross the 255/256
3250 * boundary, see check_locale_boundary_crossing() for further details.
3251 *
3252 * The macros are split with the correct case change for the below-256 case
3253 * stored into 'result', and in the middle of an else clause for the above-255
3254 * case. At that point in the 'else', 'result' is not the final result, but is
3255 * the input code point calculated from the UTF-8. The fold code needs to
3256 * realize all this and take it from there.
3257 *
3258 * If you read the two macros as sequential, it's easier to understand what's
3259 * going on. */
3260#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3261 L1_func_extra_param) \
a239b1e2 3262 \
eaf412bf
KW
3263 if (flags & (locale_flags)) { \
3264 /* Treat a UTF-8 locale as not being in locale at all */ \
3265 if (IN_UTF8_CTYPE_LOCALE) { \
3266 flags &= ~(locale_flags); \
3267 } \
3268 else { \
3269 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3270 } \
3271 } \
3272 \
3273 if (UTF8_IS_INVARIANT(*p)) { \
3274 if (flags & (locale_flags)) { \
3275 result = LC_L1_change_macro(*p); \
3276 } \
3277 else { \
3278 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3279 } \
3280 } \
a239b1e2 3281 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3282 if (flags & (locale_flags)) { \
3283 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3284 *(p+1))); \
3285 } \
3286 else { \
3287 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3288 ustrp, lenp, L1_func_extra_param); \
3289 } \
3290 } \
fa8ab374
KW
3291 else { /* malformed UTF-8 or ord above 255 */ \
3292 STRLEN len_result; \
fa8ab374
KW
3293 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3294 if (len_result == (STRLEN) -1) { \
607313a1
KW
3295 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3296 1 /* Die */ ); \
fa8ab374 3297 }
eaf412bf
KW
3298
3299#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3300 result = change_macro(result, p, ustrp, lenp); \
3301 \
3302 if (flags & (locale_flags)) { \
3303 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3304 } \
3305 return result; \
3306 } \
3307 \
3308 /* Here, used locale rules. Convert back to UTF-8 */ \
3309 if (UTF8_IS_INVARIANT(result)) { \
3310 *ustrp = (U8) result; \
3311 *lenp = 1; \
3312 } \
3313 else { \
3314 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3315 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3316 *lenp = 2; \
3317 } \
3318 \
3319 return result;
3320
d3e79532 3321/*
87cea99e 3322=for apidoc to_utf8_upper
d3e79532 3323
a239b1e2 3324Instead use L</toUPPER_utf8_safe>.
a1433954 3325
d3e79532
JH
3326=cut */
3327
051a06d4 3328/* Not currently externally documented, and subject to change:
31f05a37
KW
3329 * <flags> is set iff iff the rules from the current underlying locale are to
3330 * be used. */
051a06d4 3331
2104c8d9 3332UV
607313a1
KW
3333Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3334 const U8 *e,
3335 U8* ustrp,
3336 STRLEN *lenp,
3337 bool flags,
3338 const char * const file,
3339 const int line)
a0ed51b3 3340{
051a06d4 3341 UV result;
607313a1
KW
3342 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3343 cBOOL(flags), file, line);
051a06d4
KW
3344
3345 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3346
eaf412bf
KW
3347 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3348 /* 2nd char of uc(U+DF) is 'S' */
3349 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3350 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3351}
a0ed51b3 3352
d3e79532 3353/*
87cea99e 3354=for apidoc to_utf8_title
d3e79532 3355
a239b1e2 3356Instead use L</toTITLE_utf8_safe>.
a1433954 3357
d3e79532
JH
3358=cut */
3359
051a06d4 3360/* Not currently externally documented, and subject to change:
31f05a37
KW
3361 * <flags> is set iff the rules from the current underlying locale are to be
3362 * used. Since titlecase is not defined in POSIX, for other than a
3363 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3364 */
051a06d4 3365
983ffd37 3366UV
607313a1
KW
3367Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3368 const U8 *e,
3369 U8* ustrp,
3370 STRLEN *lenp,
3371 bool flags,
3372 const char * const file,
3373 const int line)
983ffd37 3374{
051a06d4 3375 UV result;
607313a1
KW
3376 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3377 cBOOL(flags), file, line);
051a06d4
KW
3378
3379 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3380
eaf412bf
KW
3381 /* 2nd char of ucfirst(U+DF) is 's' */
3382 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3383 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3384}
3385
d3e79532 3386/*
87cea99e 3387=for apidoc to_utf8_lower
d3e79532 3388
a239b1e2 3389Instead use L</toLOWER_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 iff the rules from the current underlying locale are to
3395 * be used.
3396 */
051a06d4 3397
2104c8d9 3398UV
607313a1
KW
3399Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3400 const U8 *e,
3401 U8* ustrp,
3402 STRLEN *lenp,
3403 bool flags,
3404 const char * const file,
3405 const int line)
a0ed51b3 3406{
051a06d4 3407 UV result;
607313a1
KW
3408 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3409 cBOOL(flags), file, line);
051a06d4 3410
051a06d4 3411 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3412
eaf412bf
KW
3413 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3414 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3415}
3416
d3e79532 3417/*
87cea99e 3418=for apidoc to_utf8_fold
d3e79532 3419
a239b1e2 3420Instead use L</toFOLD_utf8_safe>.
a1433954 3421
d3e79532
JH
3422=cut */
3423
051a06d4
KW
3424/* Not currently externally documented, and subject to change,
3425 * in <flags>
31f05a37
KW
3426 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3427 * locale are to be used.
051a06d4
KW
3428 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3429 * otherwise simple folds
a0270393
KW
3430 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3431 * prohibited
445bf929 3432 */
36bb2ab6 3433
b4e400f9 3434UV
607313a1
KW
3435Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3436 const U8 *e,
3437 U8* ustrp,
3438 STRLEN *lenp,
3439 U8 flags,
3440 const char * const file,
3441 const int line)
b4e400f9 3442{
051a06d4 3443 UV result;
607313a1
KW
3444 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3445 cBOOL(flags), file, line);
051a06d4 3446
36bb2ab6 3447 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3448
a0270393
KW
3449 /* These are mutually exclusive */
3450 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3451
50ba90ff
KW
3452 assert(p != ustrp); /* Otherwise overwrites */
3453
eaf412bf
KW
3454 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3455 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3456
eaf412bf 3457 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3458
1ca267a5
KW
3459 if (flags & FOLD_FLAGS_LOCALE) {
3460
76f2ffcd 3461# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3462 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3463
3464# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3465# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3466
3467 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3468
538e84ed
KW
3469 /* Special case these two characters, as what normally gets
3470 * returned under locale doesn't work */
76f2ffcd
KW
3471 if (UTF8SKIP(p) == cap_sharp_s_len
3472 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3473 {
ab0b796c
KW
3474 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3475 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3476 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3477 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3478 goto return_long_s;
3479 }
0766489e
KW
3480 else
3481#endif
3482 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3483 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3484 {
ab0b796c
KW
3485 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3486 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3487 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3488 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3489 goto return_ligature_st;
3490 }
74894415
KW
3491
3492#if UNICODE_MAJOR_VERSION == 3 \
3493 && UNICODE_DOT_VERSION == 0 \
3494 && UNICODE_DOT_DOT_VERSION == 1
3495# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3496
3497 /* And special case this on this Unicode version only, for the same
3498 * reaons the other two are special cased. They would cross the
3499 * 255/256 boundary which is forbidden under /l, and so the code
3500 * wouldn't catch that they are equivalent (which they are only in
3501 * this release) */
3502 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3503 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3504 {
3505 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3506 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3507 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3508 "resolved to \"\\x{0131}\".");
3509 goto return_dotless_i;
3510 }
3511#endif
3512
357aadde 3513 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3514 }
a0270393
KW
3515 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3516 return result;
3517 }
3518 else {
4a4088c4 3519 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3520 * character above the ASCII range, and the result should not
3521 * contain an ASCII character. */
a0270393
KW
3522
3523 UV original; /* To store the first code point of <p> */
3524
3525 /* Look at every character in the result; if any cross the
3526 * boundary, the whole thing is disallowed */
3527 U8* s = ustrp;
3528 U8* e = ustrp + *lenp;
3529 while (s < e) {
3530 if (isASCII(*s)) {
3531 /* Crossed, have to return the original */
3532 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3533
9fc2026f 3534 /* But in these instances, there is an alternative we can
1ca267a5 3535 * return that is valid */
0766489e
KW
3536 if (original == LATIN_SMALL_LETTER_SHARP_S
3537#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3538 || original == LATIN_CAPITAL_LETTER_SHARP_S
3539#endif
3540 ) {
1ca267a5
KW
3541 goto return_long_s;
3542 }
9fc2026f
KW
3543 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3544 goto return_ligature_st;
3545 }
74894415
KW
3546#if UNICODE_MAJOR_VERSION == 3 \
3547 && UNICODE_DOT_VERSION == 0 \
3548 && UNICODE_DOT_DOT_VERSION == 1
3549
3550 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3551 goto return_dotless_i;
3552 }
3553#endif
a0270393
KW
3554 Copy(p, ustrp, *lenp, char);
3555 return original;
3556 }
3557 s += UTF8SKIP(s);
3558 }
051a06d4 3559
a0270393
KW
3560 /* Here, no characters crossed, result is ok as-is */
3561 return result;
3562 }
051a06d4
KW
3563 }
3564
4a4088c4 3565 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3566 if (UTF8_IS_INVARIANT(result)) {
3567 *ustrp = (U8) result;
3568 *lenp = 1;
3569 }
3570 else {
62cb07ea
KW
3571 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3572 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3573 *lenp = 2;
3574 }
3575
051a06d4 3576 return result;
1ca267a5
KW
3577
3578 return_long_s:
3579 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3580 * folds to a string of two of these characters. By returning this
3581 * instead, then, e.g.,
3582 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3583 * works. */
3584
3585 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3586 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3587 ustrp, *lenp, U8);
3588 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3589
3590 return_ligature_st:
3591 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3592 * have the other one fold to it */
3593
3594 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3595 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3596 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3597
3598#if UNICODE_MAJOR_VERSION == 3 \
3599 && UNICODE_DOT_VERSION == 0 \
3600 && UNICODE_DOT_DOT_VERSION == 1
3601
3602 return_dotless_i:
3603 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3604 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3605 return LATIN_SMALL_LETTER_DOTLESS_I;
3606
3607#endif
3608
a0ed51b3
LW
3609}
3610
711a919c 3611/* Note:
f90a9a02 3612 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3613 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3614 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3615 */
c4a5db0c 3616
a0ed51b3 3617SV*
56576a04
KW
3618Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3619 I32 minbits, I32 none)
a0ed51b3 3620{
c4a5db0c
KW
3621 PERL_ARGS_ASSERT_SWASH_INIT;
3622
3623 /* Returns a copy of a swash initiated by the called function. This is the
3624 * public interface, and returning a copy prevents others from doing
3625 * mischief on the original */
3626
56576a04
KW
3627 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3628 NULL, NULL));
c4a5db0c
KW
3629}
3630
3631SV*
56576a04
KW
3632Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3633 I32 minbits, I32 none, SV* invlist,
3634 U8* const flags_p)
c4a5db0c 3635{
2c1f00b9
YO
3636
3637 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3638 * use the following define */
3639
3640#define CORE_SWASH_INIT_RETURN(x) \
3641 PL_curpm= old_PL_curpm; \
3642 return x
3643
c4a5db0c 3644 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3645 * by calling utf8_heavy.pl in the general case. The returned value may be
3646 * the swash's inversion list instead if the input parameters allow it.
3647 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3648 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3649 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3650 *
3651 * This interface should only be used by functions that won't destroy or
3652 * adversely change the swash, as doing so affects all other uses of the
3653 * swash in the program; the general public should use 'Perl_swash_init'
3654 * instead.
3655 *
3656 * pkg is the name of the package that <name> should be in.
3657 * name is the name of the swash to find. Typically it is a Unicode
3658 * property name, including user-defined ones
3659 * listsv is a string to initialize the swash with. It must be of the form
3660 * documented as the subroutine return value in
3661 * L<perlunicode/User-Defined Character Properties>
3662 * minbits is the number of bits required to represent each data element.
3663 * It is '1' for binary properties.
3664 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3665 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3666 * flags_p if non-NULL is the address of various input and output flag bits
3667 * to the routine, as follows: ('I' means is input to the routine;
3668 * 'O' means output from the routine. Only flags marked O are
3669 * meaningful on return.)
3670 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3671 * came from a user-defined property. (I O)
5d3d13d1
KW
3672 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3673 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3674 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3675 * return of an inversion list instead of a swash hash if this routine
3676 * thinks that would result in faster execution of swash_fetch() later
3677 * on. (I)
9a53f6cf
KW
3678 *
3679 * Thus there are three possible inputs to find the swash: <name>,
3680 * <listsv>, and <invlist>. At least one must be specified. The result
3681 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3682 * actions can intersect, etc. what <name> gives. To avoid going out to
3683 * disk at all, <invlist> should specify completely what the swash should
3684 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3685 *
3686 * <invlist> is only valid for binary properties */
c4a5db0c 3687
2c1f00b9
YO
3688 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3689
c4a5db0c 3690 SV* retval = &PL_sv_undef;
83199d38 3691 HV* swash_hv = NULL;
87367d5f
KW
3692 const int invlist_swash_boundary =
3693 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3694 ? 512 /* Based on some benchmarking, but not extensive, see commit
3695 message */
3696 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3697
3698 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3699 assert(! invlist || minbits == 1);
3700
56576a04
KW
3701 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3702 regex that triggered the swash init and the swash init
3703 perl logic itself. See perl #122747 */
2c1f00b9 3704
9a53f6cf
KW
3705 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3706 * so */
3707 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3708 dSP;
3709 const size_t pkg_len = strlen(pkg);
3710 const size_t name_len = strlen(name);
3711 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3712 SV* errsv_save;
3713 GV *method;
3714
3715 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3716
3717 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3718 ENTER;
69794297 3719 SAVEHINTS();
2782061f 3720 save_re_context();
650f067c
JL
3721 /* We might get here via a subroutine signature which uses a utf8