This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Generalize static fcn return for indeterminate result
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
76513bdc 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf ". This will be fatal in Perl 5.28";
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
c15a80f3 81 SAVEI8(PL_dowarn);
9cbfb8ab
KW
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
c94c2f39
KW
119/* All these formats take a single UV code point argument */
120const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
121const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
122 " is not recommended for open interchange";
123const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
124 " may not be portable";
57ff5f59
KW
125const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
126 " Unicode, requires a Perl extension," \
127 " and so is not portable";
c94c2f39 128
8ee1cdcb
KW
129#define HANDLE_UNICODE_SURROGATE(uv, flags) \
130 STMT_START { \
131 if (flags & UNICODE_WARN_SURROGATE) { \
132 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
c94c2f39 133 surrogate_cp_format, uv); \
8ee1cdcb
KW
134 } \
135 if (flags & UNICODE_DISALLOW_SURROGATE) { \
136 return NULL; \
137 } \
138 } STMT_END;
139
140#define HANDLE_UNICODE_NONCHAR(uv, flags) \
141 STMT_START { \
142 if (flags & UNICODE_WARN_NONCHAR) { \
143 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
c94c2f39 144 nonchar_cp_format, uv); \
8ee1cdcb
KW
145 } \
146 if (flags & UNICODE_DISALLOW_NONCHAR) { \
147 return NULL; \
148 } \
149 } STMT_END;
150
ba6ed43c
KW
151/* Use shorter names internally in this file */
152#define SHIFT UTF_ACCUMULATION_SHIFT
153#undef MARK
154#define MARK UTF_CONTINUATION_MARK
155#define MASK UTF_CONTINUATION_MASK
156
dfe13c55 157U8 *
4b31b634 158Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
a0ed51b3 159{
378516de 160 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 161
2d1545e5 162 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 163 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
164 return d;
165 }
facc1dc2 166
3ea68d71 167 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
168 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
169 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
170 return d;
171 }
d9432125 172
ba6ed43c
KW
173 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
174 * below, the 16 is for start bytes E0-EF (which are all the possible ones
175 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
176 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
177 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
178 * 0x800-0xFFFF on ASCII */
179 if (uv < (16 * (1U << (2 * SHIFT)))) {
180 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
181 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
182 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
183
184#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
185 aren't tested here */
186 /* The most likely code points in this range are below the surrogates.
187 * Do an extra test to quickly exclude those. */
188 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
189 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
190 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
191 {
8ee1cdcb
KW
192 HANDLE_UNICODE_NONCHAR(uv, flags);
193 }
194 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
195 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 196 }
ba6ed43c
KW
197 }
198#endif
199 return d;
200 }
201
202 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
203 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
204 * happen starting with 4-byte characters on ASCII platforms. We unify the
205 * code for these with EBCDIC, even though some of them require 5-bytes on
206 * those, because khw believes the code saving is worth the very slight
207 * performance hit on these high EBCDIC code points. */
208
209 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
76513bdc
KW
210 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
211 && ckWARN_d(WARN_DEPRECATED))
212 {
213 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
214 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
a5bf80e0 215 }
0a8a1a5b
KW
216 if ( (flags & UNICODE_WARN_SUPER)
217 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
218 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
219 {
220 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
221
222 /* Choose the more dire applicable warning */
d044b7a7 223 (UNICODE_IS_PERL_EXTENDED(uv))
57ff5f59 224 ? perl_extended_cp_format
c94c2f39 225 : super_cp_format,
a5bf80e0
KW
226 uv);
227 }
56576a04 228 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
229 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
230 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
231 {
232 return NULL;
233 }
234 }
ba6ed43c
KW
235 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
236 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 237 }
d9432125 238
ba6ed43c
KW
239 /* Test for and handle 4-byte result. In the test immediately below, the
240 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
241 * characters). The 3 is for 3 continuation bytes; these each contribute
242 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
243 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
244 * 0x1_0000-0x1F_FFFF on ASCII */
245 if (uv < (8 * (1U << (3 * SHIFT)))) {
246 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
247 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
248 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
249 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
250
251#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
252 characters. The end-plane non-characters for EBCDIC were
253 handled just above */
254 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
255 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 256 }
ba6ed43c
KW
257 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
258 HANDLE_UNICODE_SURROGATE(uv, flags);
259 }
260#endif
261
262 return d;
263 }
264
265 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
266 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
267 * format. The unrolled version above turns out to not save all that much
268 * time, and at these high code points (well above the legal Unicode range
269 * on ASCII platforms, and well above anything in common use in EBCDIC),
270 * khw believes that less code outweighs slight performance gains. */
271
d9432125 272 {
5aaebcb3 273 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
274 U8 *p = d+len-1;
275 while (p > d) {
4c8cd605 276 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
277 uv >>= UTF_ACCUMULATION_SHIFT;
278 }
4c8cd605 279 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
280 return d+len;
281 }
a0ed51b3 282}
a5bf80e0 283
646ca15d 284/*
07693fe6
KW
285=for apidoc uvchr_to_utf8
286
bcb1a2d4 287Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 288of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
289C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
290the byte after the end of the new character. In other words,
07693fe6
KW
291
292 d = uvchr_to_utf8(d, uv);
293
294is the recommended wide native character-aware way of saying
295
296 *(d++) = uv;
297
760c7c2f
KW
298This function accepts any UV as input, but very high code points (above
299C<IV_MAX> on the platform) will raise a deprecation warning. This is
300typically 0x7FFF_FFFF in a 32-bit word.
301
302It is possible to forbid or warn on non-Unicode code points, or those that may
303be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 304
07693fe6
KW
305=cut
306*/
307
de69f3af
KW
308/* This is also a macro */
309PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
310
07693fe6
KW
311U8 *
312Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
313{
de69f3af 314 return uvchr_to_utf8(d, uv);
07693fe6
KW
315}
316
de69f3af
KW
317/*
318=for apidoc uvchr_to_utf8_flags
319
320Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 321of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
322C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
323the byte after the end of the new character. In other words,
de69f3af
KW
324
325 d = uvchr_to_utf8_flags(d, uv, flags);
326
327or, in most cases,
328
329 d = uvchr_to_utf8_flags(d, uv, 0);
330
331This is the Unicode-aware way of saying
332
333 *(d++) = uv;
334
760c7c2f
KW
335If C<flags> is 0, this function accepts any UV as input, but very high code
336points (above C<IV_MAX> for the platform) will raise a deprecation warning.
337This is typically 0x7FFF_FFFF in a 32-bit word.
338
339Specifying C<flags> can further restrict what is allowed and not warned on, as
340follows:
de69f3af 341
796b6530 342If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
343the function will raise a warning, provided UTF8 warnings are enabled. If
344instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
345NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 346
760c7c2f
KW
347Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
348affect how the function handles a Unicode non-character.
93e6dbd6 349
760c7c2f
KW
350And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
351affect the handling of code points that are above the Unicode maximum of
3520x10FFFF. Languages other than Perl may not be able to accept files that
353contain these.
93e6dbd6
KW
354
355The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
356the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
357three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
358allowed inputs to the strict UTF-8 traditionally defined by Unicode.
359Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
360C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
361above-Unicode and surrogate flags, but not the non-character ones, as
362defined in
363L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
364See L<perlunicode/Noncharacter code points>.
93e6dbd6 365
57ff5f59
KW
366Extremely high code points were never specified in any standard, and require an
367extension to UTF-8 to express, which Perl does. It is likely that programs
368written in something other than Perl would not be able to read files that
369contain these; nor would Perl understand files written by something that uses a
370different extension. For these reasons, there is a separate set of flags that
371can warn and/or disallow these extremely high code points, even if other
372above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
373and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
374L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
375treat all above-Unicode code points, including these, as malformations. (Note
376that the Unicode standard considers anything above 0x10FFFF to be illegal, but
377there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
378
379A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
380retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
381C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
382C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because these
383flags can apply to code points that actually do fit in 31 bits. This happens
384on EBCDIC platforms, and sometimes when the L<overlong
385malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
386describe the situation in all cases.
de69f3af 387
de69f3af
KW
388=cut
389*/
390
391/* This is also a macro */
392PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
393
07693fe6
KW
394U8 *
395Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
396{
de69f3af 397 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
398}
399
57ff5f59
KW
400#ifndef UV_IS_QUAD
401
83dc0f42
KW
402PERL_STATIC_INLINE bool
403S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
404{
405 /* Returns TRUE if the first code point represented by the Perl-extended-
406 * UTF-8-encoded string starting at 's', and looking no further than 'e -
407 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
408 *
409 * The function handles the case where the input bytes do not include all
410 * the ones necessary to represent a full character. That is, they may be
411 * the intial bytes of the representation of a code point, but possibly
412 * the final ones necessary for the complete representation may be beyond
413 * 'e - 1'.
414 *
415 * The function assumes that the sequence is well-formed UTF-8 as far as it
416 * goes, and is for a UTF-8 variant code point. If the sequence is
417 * incomplete, the function returns FALSE if there is any well-formed
418 * UTF-8 byte sequence that can complete it in such a way that a code point
419 * < 2**31 is produced; otherwise it returns TRUE.
420 *
421 * Getting this exactly right is slightly tricky, and has to be done in
422 * several places in this file, so is centralized here. It is based on the
423 * following table:
424 *
425 * U+7FFFFFFF (2 ** 31 - 1)
426 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
427 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
428 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
429 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
430 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
431 * U+80000000 (2 ** 31):
432 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
433 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
434 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
435 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
436 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
437 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
438 */
439
440#ifdef EBCDIC
441
37086697
KW
442 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
443 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
444 const STRLEN prefix_len = sizeof(prefix) - 1;
445 const STRLEN len = e - s;
f880f78a 446 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
447
448#else
449
450 PERL_UNUSED_ARG(e);
451
452#endif
453
454 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
455
456 assert(! UTF8_IS_INVARIANT(*s));
457
458#ifndef EBCDIC
459
460 /* Technically, a start byte of FE can be for a code point that fits into
461 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
462 * malformation. */
463 return (*s >= 0xFE);
464
465#else
466
56576a04
KW
467 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
468 * mean a 32-bit or larger code point (0xFF is an invariant). For 0xFE, we
469 * need at least 2 bytes, and maybe up through 8 bytes, to be sure that the
470 * value is above 31 bits. */
83dc0f42
KW
471 if (*s != 0xFE || len == 1) {
472 return FALSE;
473 }
474
475 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
476 * \x41 and \x42. */
477 return cBOOL(memGT(s + 1, prefix, cmp_len));
478
479#endif
480
481}
482
57ff5f59
KW
483#endif
484
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
8d6204cc 534 if (isFF_OVERLONG(s, len) > 0) {
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 618 /* Check for the FF overlong */
8d6204cc 619 return isFF_OVERLONG(s, len) > 0;
b0b342d4
KW
620}
621
8d6204cc 622PERL_STATIC_INLINE int
b0b342d4
KW
623S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
624{
8d6204cc
KW
625 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
626 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
627 * it isn't, and -1 if there isn't enough information to tell. This last
628 * return value can happen if the sequence is incomplete, missing some
629 * trailing bytes that would form a complete character. If there are
630 * enough bytes to make a definitive decision, this function does so. */
631
b0b342d4 632 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 633
8d6204cc
KW
634 /* To be an FF overlong, all the available bytes must match */
635 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
636 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
637 {
638 return 0;
639 }
640
641 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
642 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 643 * utfebcdic.h. */
8d6204cc
KW
644 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
645 return 1;
646 }
12a4bed3 647
8d6204cc
KW
648 /* The missing bytes could cause the result to go one way or the other, so
649 * the result is indeterminate */
650 return -1;
12a4bed3
KW
651}
652
653#undef F0_ABOVE_OVERLONG
654#undef F8_ABOVE_OVERLONG
655#undef FC_ABOVE_OVERLONG
656#undef FE_ABOVE_OVERLONG
657#undef FF_OVERLONG_PREFIX
658
35f8c9bd 659STRLEN
edc2c47a 660Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 661{
2b479609 662 STRLEN len;
12a4bed3 663 const U8 *x;
35f8c9bd 664
2b479609
KW
665 /* A helper function that should not be called directly.
666 *
667 * This function returns non-zero if the string beginning at 's' and
668 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
669 * code point; otherwise it returns 0. The examination stops after the
670 * first code point in 's' is validated, not looking at the rest of the
671 * input. If 'e' is such that there are not enough bytes to represent a
672 * complete code point, this function will return non-zero anyway, if the
673 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
674 * excluded by 'flags'.
675 *
676 * A non-zero return gives the number of bytes required to represent the
677 * code point. Be aware that if the input is for a partial character, the
678 * return will be larger than 'e - s'.
679 *
680 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
681 * The caller should have excluded the possibility of it being invariant
682 * before calling this function.
2b479609
KW
683 *
684 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
685 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
686 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
687 * disallowed by the flags. If the input is only for a partial character,
688 * the function will return non-zero if there is any sequence of
689 * well-formed UTF-8 that, when appended to the input sequence, could
690 * result in an allowed code point; otherwise it returns 0. Non characters
691 * cannot be determined based on partial character input. But many of the
692 * other excluded types can be determined with just the first one or two
693 * bytes.
694 *
695 */
696
697 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
698
699 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 700 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 701 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 702
2b479609 703 /* A variant char must begin with a start byte */
35f8c9bd
KW
704 if (UNLIKELY(! UTF8_IS_START(*s))) {
705 return 0;
706 }
707
edc2c47a
KW
708 /* Examine a maximum of a single whole code point */
709 if (e - s > UTF8SKIP(s)) {
710 e = s + UTF8SKIP(s);
711 }
712
2b479609
KW
713 len = e - s;
714
715 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
716 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 717
56576a04
KW
718 /* Here, we are disallowing some set of largish code points, and the
719 * first byte indicates the sequence is for a code point that could be
720 * in the excluded set. We generally don't have to look beyond this or
721 * the second byte to see if the sequence is actually for one of the
722 * excluded classes. The code below is derived from this table:
723 *
2b479609
KW
724 * UTF-8 UTF-EBCDIC I8
725 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
726 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
727 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
728 *
56576a04
KW
729 * Keep in mind that legal continuation bytes range between \x80..\xBF
730 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
731 * continuation bytes. Hence, we don't have to test the upper edge
732 * because if any of those is encountered, the sequence is malformed,
733 * and would fail elsewhere in this function.
734 *
735 * The code here likewise assumes that there aren't other
736 * malformations; again the function should fail elsewhere because of
737 * these. For example, an overlong beginning with FC doesn't actually
738 * have to be a super; it could actually represent a small code point,
739 * even U+0000. But, since overlongs (and other malformations) are
740 * illegal, the function should return FALSE in either case.
2b479609
KW
741 */
742
743#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
744# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 745# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 746
19794540
KW
747# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
748 /* B6 and B7 */ \
749 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 750# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
751#else
752# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
753# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
754# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 755# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
756#endif
757
758 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
759 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
760 {
2b479609
KW
761 return 0; /* Above Unicode */
762 }
763
d044b7a7 764 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 765 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 766 {
57ff5f59 767 return 0;
2b479609
KW
768 }
769
770 if (len > 1) {
771 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
772
773 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 774 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
775 {
776 return 0; /* Above Unicode */
777 }
778
779 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 780 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
781 {
782 return 0; /* Surrogate */
783 }
784
785 if ( (flags & UTF8_DISALLOW_NONCHAR)
786 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
787 {
788 return 0; /* Noncharacter code point */
789 }
790 }
791 }
792
793 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
794 for (x = s + 1; x < e; x++) {
795 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
796 return 0;
797 }
798 }
799
af13dd8a 800 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
801 * overlong. */
802 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
803 return 0;
af13dd8a
KW
804 }
805
12a4bed3
KW
806 /* And finally, that the code point represented fits in a word on this
807 * platform */
808 if (does_utf8_overflow(s, e)) {
809 return 0;
35f8c9bd
KW
810 }
811
2b479609 812 return UTF8SKIP(s);
35f8c9bd
KW
813}
814
7e2f38b2
KW
815char *
816Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
817{
818 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
819 * bytes starting at 's'. 'format' gives how to display each byte.
820 * Currently, there are only two formats, so it is currently a bool:
821 * 0 \xab
822 * 1 ab (that is a space between two hex digit bytes)
823 */
7cf8d05d
KW
824
825 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
826 trailing NUL */
827 const U8 * const e = s + len;
828 char * output;
829 char * d;
830
831 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
832
833 Newx(output, output_len, char);
834 SAVEFREEPV(output);
835
836 d = output;
837 for (; s < e; s++) {
838 const unsigned high_nibble = (*s & 0xF0) >> 4;
839 const unsigned low_nibble = (*s & 0x0F);
840
7e2f38b2
KW
841 if (format) {
842 *d++ = ' ';
843 }
844 else {
845 *d++ = '\\';
846 *d++ = 'x';
847 }
7cf8d05d
KW
848
849 if (high_nibble < 10) {
850 *d++ = high_nibble + '0';
851 }
852 else {
853 *d++ = high_nibble - 10 + 'a';
854 }
855
856 if (low_nibble < 10) {
857 *d++ = low_nibble + '0';
858 }
859 else {
860 *d++ = low_nibble - 10 + 'a';
861 }
862 }
863
864 *d = '\0';
865 return output;
866}
867
806547a7 868PERL_STATIC_INLINE char *
7cf8d05d
KW
869S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
870
871 /* How many bytes to print */
3cc6a05e 872 STRLEN print_len,
7cf8d05d
KW
873
874 /* Which one is the non-continuation */
875 const STRLEN non_cont_byte_pos,
876
877 /* How many bytes should there be? */
878 const STRLEN expect_len)
806547a7
KW
879{
880 /* Return the malformation warning text for an unexpected continuation
881 * byte. */
882
7cf8d05d 883 const char * const where = (non_cont_byte_pos == 1)
806547a7 884 ? "immediately"
7cf8d05d
KW
885 : Perl_form(aTHX_ "%d bytes",
886 (int) non_cont_byte_pos);
806547a7
KW
887
888 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
889
7cf8d05d
KW
890 /* We don't need to pass this parameter, but since it has already been
891 * calculated, it's likely faster to pass it; verify under DEBUGGING */
892 assert(expect_len == UTF8SKIP(s));
893
894 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
895 " %s after start byte 0x%02x; need %d bytes, got %d)",
896 malformed_text,
7e2f38b2 897 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
898 *(s + non_cont_byte_pos),
899 where,
900 *s,
901 (int) expect_len,
902 (int) non_cont_byte_pos);
806547a7
KW
903}
904
35f8c9bd
KW
905/*
906
de69f3af 907=for apidoc utf8n_to_uvchr
378516de
KW
908
909THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 910Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 911
9041c2e3 912Bottom level UTF-8 decode routine.
de69f3af 913Returns the native code point value of the first character in the string C<s>,
746afd53
KW
914which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
915C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
916the length, in bytes, of that character.
949cf498
KW
917
918The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
919well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
920causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
921is the next possible position in C<s> that could begin a non-malformed
922character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
923is raised. Some UTF-8 input sequences may contain multiple malformations.
924This function tries to find every possible one in each call, so multiple
56576a04 925warnings can be raised for the same sequence.
949cf498
KW
926
927Various ALLOW flags can be set in C<flags> to allow (and not warn on)
928individual types of malformations, such as the sequence being overlong (that
929is, when there is a shorter sequence that can express the same code point;
930overlong sequences are expressly forbidden in the UTF-8 standard due to
931potential security issues). Another malformation example is the first byte of
932a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
933flags. Even if allowed, this function generally returns the Unicode
934REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
935F<utf8.h> to override this behavior for the overlong malformations, but don't
936do that except for very specialized purposes.
949cf498 937
796b6530 938The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
939flags) malformation is found. If this flag is set, the routine assumes that
940the caller will raise a warning, and this function will silently just set
d088425d
KW
941C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
942
75200dff 943Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 944character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
945in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
946be set to 1. To disambiguate, upon a zero return, see if the first byte of
947C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 948error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
949
950Certain code points are considered problematic. These are Unicode surrogates,
746afd53 951Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 952By default these are considered regular code points, but certain situations
ecc1615f
KW
953warrant special handling for them, which can be specified using the C<flags>
954parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
955three classes are treated as malformations and handled as such. The flags
956C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
957C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
958disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
959restricts the allowed inputs to the strict UTF-8 traditionally defined by
960Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
961definition given by
962L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
963The difference between traditional strictness and C9 strictness is that the
964latter does not forbid non-character code points. (They are still discouraged,
965however.) For more discussion see L<perlunicode/Noncharacter code points>.
966
967The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
968C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
969C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
970raised for their respective categories, but otherwise the code points are
971considered valid (not malformations). To get a category to both be treated as
972a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 973(But note that warnings are not raised if lexically disabled nor if
796b6530 974C<UTF8_CHECK_ONLY> is also specified.)
949cf498 975
57ff5f59
KW
976Extremely high code points were never specified in any standard, and require an
977extension to UTF-8 to express, which Perl does. It is likely that programs
978written in something other than Perl would not be able to read files that
979contain these; nor would Perl understand files written by something that uses a
980different extension. For these reasons, there is a separate set of flags that
981can warn and/or disallow these extremely high code points, even if other
982above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
983C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
984L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
985above-Unicode code points, including these, as malformations.
986(Note that the Unicode standard considers anything above 0x10FFFF to be
987illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
988(2**31 -1))
989
990A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
991retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
992C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
993C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
994can apply to code points that actually do fit in 31 bits. This happens on
995EBCDIC platforms, and sometimes when the L<overlong
996malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
997describe the situation in all cases.
998
760c7c2f
KW
999It is now deprecated to have very high code points (above C<IV_MAX> on the
1000platforms) and this function will raise a deprecation warning for these (unless
d5944cab 1001such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 1002in a 32-bit word.
ab8e6d41 1003
949cf498
KW
1004All other code points corresponding to Unicode characters, including private
1005use and those yet to be assigned, are never considered malformed and never
1006warn.
67e989fb 1007
37607a96 1008=cut
f9380377
KW
1009
1010Also implemented as a macro in utf8.h
1011*/
1012
1013UV
1014Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1015 STRLEN curlen,
1016 STRLEN *retlen,
1017 const U32 flags)
1018{
1019 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1020
1021 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1022}
1023
1024/*
1025
1026=for apidoc utf8n_to_uvchr_error
1027
1028THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1029Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1030
1031This function is for code that needs to know what the precise malformation(s)
1032are when an error is found.
1033
1034It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1035all the others, C<errors>. If this parameter is 0, this function behaves
1036identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1037to a C<U32> variable, which this function sets to indicate any errors found.
1038Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1039C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1040of these bits will be set if a malformation is found, even if the input
7a65503b 1041C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1042exceptions are noted:
1043
1044=over 4
1045
57ff5f59 1046=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1047
57ff5f59
KW
1048The input sequence is not standard UTF-8, but a Perl extension. This bit is
1049set only if the input C<flags> parameter contains either the
1050C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1051
1052Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1053and so some extension must be used to express them. Perl uses a natural
1054extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1055extension to represent even higher ones, so that any code point that fits in a
105664-bit word can be represented. Text using these extensions is not likely to
1057be portable to non-Perl code. We lump both of these extensions together and
1058refer to them as Perl extended UTF-8. There exist other extensions that people
1059have invented, incompatible with Perl's.
1060
1061On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1062extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1063than on ASCII. Prior to that, code points 2**31 and higher were simply
1064unrepresentable, and a different, incompatible method was used to represent
1065code points between 2**30 and 2**31 - 1.
1066
1067On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1068Perl extended UTF-8 is used.
1069
1070In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1071may use for backward compatibility. That name is misleading, as this flag may
1072be set when the code point actually does fit in 31 bits. This happens on
1073EBCDIC platforms, and sometimes when the L<overlong
1074malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1075describes the situation in all cases.
f9380377
KW
1076
1077=item C<UTF8_GOT_CONTINUATION>
1078
1079The input sequence was malformed in that the first byte was a a UTF-8
1080continuation byte.
1081
1082=item C<UTF8_GOT_EMPTY>
1083
1084The input C<curlen> parameter was 0.
1085
1086=item C<UTF8_GOT_LONG>
1087
1088The input sequence was malformed in that there is some other sequence that
1089evaluates to the same code point, but that sequence is shorter than this one.
1090
fecaf136
KW
1091Until Unicode 3.1, it was legal for programs to accept this malformation, but
1092it was discovered that this created security issues.
1093
f9380377
KW
1094=item C<UTF8_GOT_NONCHAR>
1095
1096The code point represented by the input UTF-8 sequence is for a Unicode
1097non-character code point.
1098This bit is set only if the input C<flags> parameter contains either the
1099C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1100
1101=item C<UTF8_GOT_NON_CONTINUATION>
1102
1103The input sequence was malformed in that a non-continuation type byte was found
1104in a position where only a continuation type one should be.
1105
1106=item C<UTF8_GOT_OVERFLOW>
1107
1108The input sequence was malformed in that it is for a code point that is not
1109representable in the number of bits available in a UV on the current platform.
1110
1111=item C<UTF8_GOT_SHORT>
1112
1113The input sequence was malformed in that C<curlen> is smaller than required for
1114a complete sequence. In other words, the input is for a partial character
1115sequence.
1116
1117=item C<UTF8_GOT_SUPER>
1118
1119The input sequence was malformed in that it is for a non-Unicode code point;
1120that is, one above the legal Unicode maximum.
1121This bit is set only if the input C<flags> parameter contains either the
1122C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1123
1124=item C<UTF8_GOT_SURROGATE>
1125
1126The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1127code point.
1128This bit is set only if the input C<flags> parameter contains either the
1129C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1130
1131=back
1132
133551d8
KW
1133To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1134flag to suppress any warnings, and then examine the C<*errors> return.
1135
f9380377 1136=cut
37607a96 1137*/
67e989fb 1138
a0ed51b3 1139UV
f9380377
KW
1140Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1141 STRLEN curlen,
1142 STRLEN *retlen,
1143 const U32 flags,
1144 U32 * errors)
a0ed51b3 1145{
d4c19fe8 1146 const U8 * const s0 = s;
2b5e7bc2
KW
1147 U8 * send = NULL; /* (initialized to silence compilers' wrong
1148 warning) */
1149 U32 possible_problems = 0; /* A bit is set here for each potential problem
1150 found as we go along */
eb83ed87 1151 UV uv = *s;
2b5e7bc2
KW
1152 STRLEN expectlen = 0; /* How long should this sequence be?
1153 (initialized to silence compilers' wrong
1154 warning) */
e308b348 1155 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1156 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1157 this gets set and discarded */
a0dbb045 1158
2b5e7bc2
KW
1159 /* The below are used only if there is both an overlong malformation and a
1160 * too short one. Otherwise the first two are set to 's0' and 'send', and
1161 * the third not used at all */
1162 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1163 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1164 routine; see [perl #130921] */
2b5e7bc2 1165 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1166
f9380377
KW
1167 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1168
1169 if (errors) {
1170 *errors = 0;
1171 }
1172 else {
1173 errors = &discard_errors;
1174 }
a0dbb045 1175
eb83ed87
KW
1176 /* The order of malformation tests here is important. We should consume as
1177 * few bytes as possible in order to not skip any valid character. This is
1178 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1179 * http://unicode.org/reports/tr36 for more discussion as to why. For
1180 * example, once we've done a UTF8SKIP, we can tell the expected number of
1181 * bytes, and could fail right off the bat if the input parameters indicate
1182 * that there are too few available. But it could be that just that first
1183 * byte is garbled, and the intended character occupies fewer bytes. If we
1184 * blindly assumed that the first byte is correct, and skipped based on
1185 * that number, we could skip over a valid input character. So instead, we
1186 * always examine the sequence byte-by-byte.
1187 *
1188 * We also should not consume too few bytes, otherwise someone could inject
1189 * things. For example, an input could be deliberately designed to
1190 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1191 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1192 * which is actually part of of the overflowing sequence), that could look
1193 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1194 * sequence and process the rest, inappropriately.
1195 *
1196 * Some possible input sequences are malformed in more than one way. This
1197 * function goes to lengths to try to find all of them. This is necessary
1198 * for correctness, as the inputs may allow one malformation but not
1199 * another, and if we abandon searching for others after finding the
1200 * allowed one, we could allow in something that shouldn't have been.
1201 */
eb83ed87 1202
b5b9af04 1203 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1204 possible_problems |= UTF8_GOT_EMPTY;
1205 curlen = 0;
5a48568d 1206 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1207 goto ready_to_handle_errors;
0c443dc2
JH
1208 }
1209
eb83ed87
KW
1210 expectlen = UTF8SKIP(s);
1211
1212 /* A well-formed UTF-8 character, as the vast majority of calls to this
1213 * function will be for, has this expected length. For efficiency, set
1214 * things up here to return it. It will be overriden only in those rare
1215 * cases where a malformation is found */
1216 if (retlen) {
1217 *retlen = expectlen;
1218 }
1219
1220 /* An invariant is trivially well-formed */
1d72bdf6 1221 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1222 return uv;
a0ed51b3 1223 }
67e989fb 1224
eb83ed87 1225 /* A continuation character can't start a valid sequence */
b5b9af04 1226 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1227 possible_problems |= UTF8_GOT_CONTINUATION;
1228 curlen = 1;
1229 uv = UNICODE_REPLACEMENT;
1230 goto ready_to_handle_errors;
ba210ebe 1231 }
9041c2e3 1232
dcd27b3c 1233 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1234 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1235 * because it excludes start bytes like \xC0 that always lead to
1236 * overlongs.) */
dcd27b3c 1237
534752c1
KW
1238 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1239 * that indicate the number of bytes in the character's whole UTF-8
1240 * sequence, leaving just the bits that are part of the value. */
1241 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1242
e308b348
KW
1243 /* Setup the loop end point, making sure to not look past the end of the
1244 * input string, and flag it as too short if the size isn't big enough. */
1245 send = (U8*) s0;
1246 if (UNLIKELY(curlen < expectlen)) {
1247 possible_problems |= UTF8_GOT_SHORT;
1248 avail_len = curlen;
1249 send += curlen;
1250 }
1251 else {
1252 send += expectlen;
1253 }
e308b348 1254
eb83ed87 1255 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1256 * accumulating each into the working value as we go. */
eb83ed87 1257 for (s = s0 + 1; s < send; s++) {
b5b9af04 1258 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1259 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1260 continue;
1261 }
1262
1263 /* Here, found a non-continuation before processing all expected bytes.
1264 * This byte indicates the beginning of a new character, so quit, even
1265 * if allowing this malformation. */
2b5e7bc2 1266 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1267 break;
eb83ed87
KW
1268 } /* End of loop through the character's bytes */
1269
1270 /* Save how many bytes were actually in the character */
1271 curlen = s - s0;
1272
2b5e7bc2
KW
1273 /* Note that there are two types of too-short malformation. One is when
1274 * there is actual wrong data before the normal termination of the
1275 * sequence. The other is that the sequence wasn't complete before the end
1276 * of the data we are allowed to look at, based on the input 'curlen'.
1277 * This means that we were passed data for a partial character, but it is
1278 * valid as far as we saw. The other is definitely invalid. This
1279 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1280 * separate.
1281 *
1282 * A convenience macro that matches either of the too-short conditions. */
1283# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1284
1285 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1286 uv_so_far = uv;
1287 uv = UNICODE_REPLACEMENT;
1288 }
2b5e7bc2 1289
08e73697
KW
1290 /* Check for overflow. The algorithm requires us to not look past the end
1291 * of the current character, even if partial, so the upper limit is 's' */
1292 if (UNLIKELY(does_utf8_overflow(s0, s))) {
2b5e7bc2
KW
1293 possible_problems |= UTF8_GOT_OVERFLOW;
1294 uv = UNICODE_REPLACEMENT;
eb83ed87 1295 }
eb83ed87 1296
2b5e7bc2
KW
1297 /* Check for overlong. If no problems so far, 'uv' is the correct code
1298 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1299 * we must look at the UTF-8 byte sequence itself to see if it is for an
1300 * overlong */
1301 if ( ( LIKELY(! possible_problems)
1302 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1303 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1304 && ( UNLIKELY(! UTF8_IS_START(*s0))
1305 || ( curlen > 1
1306 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1307 s - s0))))))
2f8f112e 1308 {
2b5e7bc2
KW
1309 possible_problems |= UTF8_GOT_LONG;
1310
abc28b54 1311 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1312
abc28b54
KW
1313 /* The calculation in the 'true' branch of this 'if'
1314 * below won't work if overflows, and isn't needed
1315 * anyway. Further below we handle all overflow
1316 * cases */
1317 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1318 {
2b5e7bc2
KW
1319 UV min_uv = uv_so_far;
1320 STRLEN i;
1321
1322 /* Here, the input is both overlong and is missing some trailing
1323 * bytes. There is no single code point it could be for, but there
1324 * may be enough information present to determine if what we have
1325 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1326 * The code further below has the intelligence to determine this,
1327 * but just for non-overlong UTF-8 sequences. What we do here is
1328 * calculate the smallest code point the input could represent if
1329 * there were no too short malformation. Then we compute and save
1330 * the UTF-8 for that, which is what the code below looks at
1331 * instead of the raw input. It turns out that the smallest such
1332 * code point is all we need. */
2b5e7bc2
KW
1333 for (i = curlen; i < expectlen; i++) {
1334 min_uv = UTF8_ACCUMULATE(min_uv,
1335 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1336 }
1337
e9f2c446 1338 adjusted_s0 = temp_char_buf;
57ff5f59 1339 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1340 }
eb83ed87
KW
1341 }
1342
56576a04
KW
1343 /* Here, we have found all the possible problems, except for when the input
1344 * is for a problematic code point not allowed by the input parameters. */
1345
06188866
KW
1346 /* uv is valid for overlongs */
1347 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1348
1349 /* isn't problematic if < this */
1350 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1351 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1352
1353 /* if overflow, we know without looking further
1354 * precisely which of the problematic types it is,
1355 * and we deal with those in the overflow handling
1356 * code */
1357 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1358 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1359 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1360 && ((flags & ( UTF8_DISALLOW_NONCHAR
1361 |UTF8_DISALLOW_SURROGATE
1362 |UTF8_DISALLOW_SUPER
d044b7a7 1363 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1364 |UTF8_WARN_NONCHAR
1365 |UTF8_WARN_SURROGATE
1366 |UTF8_WARN_SUPER
d044b7a7 1367 |UTF8_WARN_PERL_EXTENDED))
2b5e7bc2
KW
1368 /* In case of a malformation, 'uv' is not valid, and has
1369 * been changed to something in the Unicode range.
1370 * Currently we don't output a deprecation message if there
1371 * is already a malformation, so we don't have to special
1372 * case the test immediately below */
760c7c2f
KW
1373 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1374 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1375 {
2b5e7bc2
KW
1376 /* If there were no malformations, or the only malformation is an
1377 * overlong, 'uv' is valid */
1378 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1379 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1380 possible_problems |= UTF8_GOT_SURROGATE;
1381 }
1382 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1383 possible_problems |= UTF8_GOT_SUPER;
1384 }
1385 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1386 possible_problems |= UTF8_GOT_NONCHAR;
1387 }
1388 }
1389 else { /* Otherwise, need to look at the source UTF-8, possibly
1390 adjusted to be non-overlong */
1391
1392 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1393 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1394 {
2b5e7bc2
KW
1395 possible_problems |= UTF8_GOT_SUPER;
1396 }
1397 else if (curlen > 1) {
1398 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1399 NATIVE_UTF8_TO_I8(*adjusted_s0),
1400 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1401 {
2b5e7bc2 1402 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1403 }
2b5e7bc2
KW
1404 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1405 NATIVE_UTF8_TO_I8(*adjusted_s0),
1406 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1407 {
1408 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1409 }
1410 }
c0236afe 1411
2b5e7bc2
KW
1412 /* We need a complete well-formed UTF-8 character to discern
1413 * non-characters, so can't look for them here */
1414 }
1415 }
949cf498 1416
2b5e7bc2
KW
1417 ready_to_handle_errors:
1418
1419 /* At this point:
1420 * curlen contains the number of bytes in the sequence that
1421 * this call should advance the input by.
e308b348
KW
1422 * avail_len gives the available number of bytes passed in, but
1423 * only if this is less than the expected number of
1424 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1425 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1426 * is set in it for each potential problem found.
1427 * uv contains the code point the input sequence
1428 * represents; or if there is a problem that prevents
1429 * a well-defined value from being computed, it is
1430 * some subsitute value, typically the REPLACEMENT
1431 * CHARACTER.
1432 * s0 points to the first byte of the character
56576a04
KW
1433 * s points to just after were we left off processing
1434 * the character
1435 * send points to just after where that character should
1436 * end, based on how many bytes the start byte tells
1437 * us should be in it, but no further than s0 +
1438 * avail_len
2b5e7bc2 1439 */
eb83ed87 1440
2b5e7bc2
KW
1441 if (UNLIKELY(possible_problems)) {
1442 bool disallowed = FALSE;
1443 const U32 orig_problems = possible_problems;
1444
1445 while (possible_problems) { /* Handle each possible problem */
1446 UV pack_warn = 0;
1447 char * message = NULL;
1448
1449 /* Each 'if' clause handles one problem. They are ordered so that
1450 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1451 * ones; this is kinda in decreasing severity order. But the
1452 * overlong must come last, as it changes 'uv' looked at by the
1453 * others */
2b5e7bc2
KW
1454 if (possible_problems & UTF8_GOT_OVERFLOW) {
1455
56576a04
KW
1456 /* Overflow means also got a super and are using Perl's
1457 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1458 possible_problems
d044b7a7 1459 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1460 *errors |= UTF8_GOT_OVERFLOW;
1461
1462 /* But the API says we flag all errors found */
1463 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1464 *errors |= UTF8_GOT_SUPER;
1465 }
ddb65933 1466 if (flags
d044b7a7 1467 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1468 {
d044b7a7 1469 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1470 }
2b5e7bc2 1471
d60baaa7 1472 /* Disallow if any of the three categories say to */
56576a04 1473 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1474 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1475 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1476 {
1477 disallowed = TRUE;
1478 }
1479
d60baaa7
KW
1480 /* Likewise, warn if any say to, plus if deprecation warnings
1481 * are on, because this code point is above IV_MAX */
56576a04 1482 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7 1483 || ! (flags & UTF8_ALLOW_OVERFLOW)
d044b7a7 1484 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1485 {
2b5e7bc2 1486
ddb65933
KW
1487 /* The warnings code explicitly says it doesn't handle the
1488 * case of packWARN2 and two categories which have
1489 * parent-child relationship. Even if it works now to
1490 * raise the warning if either is enabled, it wouldn't
1491 * necessarily do so in the future. We output (only) the
56576a04 1492 * most dire warning */
ddb65933
KW
1493 if (! (flags & UTF8_CHECK_ONLY)) {
1494 if (ckWARN_d(WARN_UTF8)) {
1495 pack_warn = packWARN(WARN_UTF8);
1496 }
1497 else if (ckWARN_d(WARN_NON_UNICODE)) {
1498 pack_warn = packWARN(WARN_NON_UNICODE);
1499 }
1500 if (pack_warn) {
1501 message = Perl_form(aTHX_ "%s: %s (overflows)",
1502 malformed_text,
05b9033b 1503 _byte_dump_string(s0, curlen, 0));
ddb65933 1504 }
2b5e7bc2
KW
1505 }
1506 }
1507 }
1508 else if (possible_problems & UTF8_GOT_EMPTY) {
1509 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1510 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1511
1512 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1513
1514 /* This so-called malformation is now treated as a bug in
1515 * the caller. If you have nothing to decode, skip calling
1516 * this function */
1517 assert(0);
1518
2b5e7bc2
KW
1519 disallowed = TRUE;
1520 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1521 pack_warn = packWARN(WARN_UTF8);
1522 message = Perl_form(aTHX_ "%s (empty string)",
1523 malformed_text);
1524 }
1525 }
1526 }
1527 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1528 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1529 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1530
1531 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1532 disallowed = TRUE;
1533 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1534 pack_warn = packWARN(WARN_UTF8);
1535 message = Perl_form(aTHX_
1536 "%s: %s (unexpected continuation byte 0x%02x,"
1537 " with no preceding start byte)",
1538 malformed_text,
7e2f38b2 1539 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1540 }
1541 }
1542 }
2b5e7bc2
KW
1543 else if (possible_problems & UTF8_GOT_SHORT) {
1544 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1545 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1546
1547 if (! (flags & UTF8_ALLOW_SHORT)) {
1548 disallowed = TRUE;
1549 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1550 pack_warn = packWARN(WARN_UTF8);
1551 message = Perl_form(aTHX_
56576a04
KW
1552 "%s: %s (too short; %d byte%s available, need %d)",
1553 malformed_text,
1554 _byte_dump_string(s0, send - s0, 0),
1555 (int)avail_len,
1556 avail_len == 1 ? "" : "s",
1557 (int)expectlen);
2b5e7bc2
KW
1558 }
1559 }
ba210ebe 1560
2b5e7bc2 1561 }
e308b348
KW
1562 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1563 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1564 *errors |= UTF8_GOT_NON_CONTINUATION;
1565
1566 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1567 disallowed = TRUE;
1568 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1569
1570 /* If we don't know for sure that the input length is
1571 * valid, avoid as much as possible reading past the
1572 * end of the buffer */
1573 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1574 ? s - s0
1575 : send - s0;
e308b348
KW
1576 pack_warn = packWARN(WARN_UTF8);
1577 message = Perl_form(aTHX_ "%s",
1578 unexpected_non_continuation_text(s0,
99a765e9 1579 printlen,
e308b348
KW
1580 s - s0,
1581 (int) expectlen));
1582 }
1583 }
1584 }
2b5e7bc2
KW
1585 else if (possible_problems & UTF8_GOT_SURROGATE) {
1586 possible_problems &= ~UTF8_GOT_SURROGATE;
1587
f9380377
KW
1588 if (flags & UTF8_WARN_SURROGATE) {
1589 *errors |= UTF8_GOT_SURROGATE;
1590
1591 if ( ! (flags & UTF8_CHECK_ONLY)
1592 && ckWARN_d(WARN_SURROGATE))
1593 {
2b5e7bc2
KW
1594 pack_warn = packWARN(WARN_SURROGATE);
1595
1596 /* These are the only errors that can occur with a
1597 * surrogate when the 'uv' isn't valid */
1598 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1599 message = Perl_form(aTHX_
1600 "UTF-16 surrogate (any UTF-8 sequence that"
1601 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1602 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1603 }
1604 else {
c94c2f39 1605 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1606 }
f9380377 1607 }
2b5e7bc2 1608 }
ba210ebe 1609
2b5e7bc2
KW
1610 if (flags & UTF8_DISALLOW_SURROGATE) {
1611 disallowed = TRUE;
f9380377 1612 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1613 }
1614 }
1615 else if (possible_problems & UTF8_GOT_SUPER) {
1616 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1617
f9380377
KW
1618 if (flags & UTF8_WARN_SUPER) {
1619 *errors |= UTF8_GOT_SUPER;
1620
1621 if ( ! (flags & UTF8_CHECK_ONLY)
1622 && ckWARN_d(WARN_NON_UNICODE))
1623 {
2b5e7bc2
KW
1624 pack_warn = packWARN(WARN_NON_UNICODE);
1625
1626 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1627 message = Perl_form(aTHX_
1628 "Any UTF-8 sequence that starts with"
1629 " \"%s\" is for a non-Unicode code point,"
1630 " may not be portable",
7e2f38b2 1631 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1632 }
1633 else {
c94c2f39 1634 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1635 }
f9380377 1636 }
2b5e7bc2 1637 }
ba210ebe 1638
57ff5f59
KW
1639 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1640 * and before possibly bailing out, so that the more dire
1641 * warning will override the regular one. */
1642 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 1643 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 1644 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
db0f09e6 1645 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1646 {
db0f09e6 1647 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 1648
57ff5f59
KW
1649 /* If it is an overlong that evaluates to a code point
1650 * that doesn't have to use the Perl extended UTF-8, it
1651 * still used it, and so we output a message that
1652 * doesn't refer to the code point. The same is true
1653 * if there was a SHORT malformation where the code
1654 * point is not valid. In that case, 'uv' will have
1655 * been set to the REPLACEMENT CHAR, and the message
1656 * below without the code point in it will be selected
1657 * */
1658 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 1659 message = Perl_form(aTHX_
57ff5f59 1660 perl_extended_cp_format, uv);
2b5e7bc2
KW
1661 }
1662 else {
1663 message = Perl_form(aTHX_
57ff5f59
KW
1664 "Any UTF-8 sequence that starts with"
1665 " \"%s\" is a Perl extension, and"
1666 " so is not portable",
1667 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1668 }
1669 }
1670
d044b7a7
KW
1671 if (flags & ( UTF8_WARN_PERL_EXTENDED
1672 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1673 {
d044b7a7 1674 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1675
d044b7a7 1676 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
1677 disallowed = TRUE;
1678 }
2b5e7bc2
KW
1679 }
1680 }
eb83ed87 1681
2b5e7bc2 1682 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1683 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1684 disallowed = TRUE;
1685 }
eb83ed87 1686
2b5e7bc2
KW
1687 /* The deprecated warning overrides any non-deprecated one. If
1688 * there are other problems, a deprecation message is not
1689 * really helpful, so don't bother to raise it in that case.
1690 * This also keeps the code from having to handle the case
1691 * where 'uv' is not valid. */
1692 if ( ! (orig_problems
1693 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1694 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1695 && ckWARN_d(WARN_DEPRECATED))
1696 {
1697 message = Perl_form(aTHX_ cp_above_legal_max,
1698 uv, MAX_NON_DEPRECATED_CP);
1699 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1700 }
1701 }
1702 else if (possible_problems & UTF8_GOT_NONCHAR) {
1703 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1704
f9380377
KW
1705 if (flags & UTF8_WARN_NONCHAR) {
1706 *errors |= UTF8_GOT_NONCHAR;
1707
1708 if ( ! (flags & UTF8_CHECK_ONLY)
1709 && ckWARN_d(WARN_NONCHAR))
1710 {
2b5e7bc2
KW
1711 /* The code above should have guaranteed that we don't
1712 * get here with errors other than overlong */
1713 assert (! (orig_problems
1714 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1715
1716 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1717 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1718 }
2b5e7bc2 1719 }
5b311467 1720
2b5e7bc2
KW
1721 if (flags & UTF8_DISALLOW_NONCHAR) {
1722 disallowed = TRUE;
f9380377 1723 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 1724 }
6c64cd9d
KW
1725 }
1726 else if (possible_problems & UTF8_GOT_LONG) {
1727 possible_problems &= ~UTF8_GOT_LONG;
1728 *errors |= UTF8_GOT_LONG;
1729
1730 if (flags & UTF8_ALLOW_LONG) {
1731
1732 /* We don't allow the actual overlong value, unless the
1733 * special extra bit is also set */
1734 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1735 & ~UTF8_ALLOW_LONG)))
1736 {
1737 uv = UNICODE_REPLACEMENT;
1738 }
1739 }
1740 else {
1741 disallowed = TRUE;
1742
1743 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1744 pack_warn = packWARN(WARN_UTF8);
1745
1746 /* These error types cause 'uv' to be something that
1747 * isn't what was intended, so can't use it in the
1748 * message. The other error types either can't
1749 * generate an overlong, or else the 'uv' is valid */
1750 if (orig_problems &
1751 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1752 {
1753 message = Perl_form(aTHX_
1754 "%s: %s (any UTF-8 sequence that starts"
1755 " with \"%s\" is overlong which can and"
1756 " should be represented with a"
1757 " different, shorter sequence)",
1758 malformed_text,
1759 _byte_dump_string(s0, send - s0, 0),
1760 _byte_dump_string(s0, curlen, 0));
1761 }
1762 else {
1763 U8 tmpbuf[UTF8_MAXBYTES+1];
1764 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1765 uv, 0);
1766 const char * preface = (uv <= PERL_UNICODE_MAX)
1767 ? "U+"
1768 : "0x";
1769 message = Perl_form(aTHX_
1770 "%s: %s (overlong; instead use %s to represent"
1771 " %s%0*" UVXf ")",
1772 malformed_text,
1773 _byte_dump_string(s0, send - s0, 0),
1774 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1775 preface,
1776 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1777 small code points */
1778 uv);
1779 }
1780 }
1781 }
2b5e7bc2
KW
1782 } /* End of looking through the possible flags */
1783
1784 /* Display the message (if any) for the problem being handled in
1785 * this iteration of the loop */
1786 if (message) {
1787 if (PL_op)
1788 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1789 OP_DESC(PL_op));
1790 else
1791 Perl_warner(aTHX_ pack_warn, "%s", message);
1792 }
ddb65933 1793 } /* End of 'while (possible_problems)' */
a0dbb045 1794
2b5e7bc2
KW
1795 /* Since there was a possible problem, the returned length may need to
1796 * be changed from the one stored at the beginning of this function.
1797 * Instead of trying to figure out if that's needed, just do it. */
1798 if (retlen) {
1799 *retlen = curlen;
1800 }
a0dbb045 1801
2b5e7bc2
KW
1802 if (disallowed) {
1803 if (flags & UTF8_CHECK_ONLY && retlen) {
1804 *retlen = ((STRLEN) -1);
1805 }
1806 return 0;
1807 }
eb83ed87 1808 }
ba210ebe 1809
2b5e7bc2 1810 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1811}
1812
8e84507e 1813/*
ec5f19d0
KW
1814=for apidoc utf8_to_uvchr_buf
1815
1816Returns the native code point of the first character in the string C<s> which
1817is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1818C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1819
524080c4
KW
1820If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1821enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1822C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1823(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1824C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1825the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1826See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1827returned.
ec5f19d0 1828
760c7c2f
KW
1829Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1830unless those are turned off.
1831
ec5f19d0 1832=cut
52be2536
KW
1833
1834Also implemented as a macro in utf8.h
1835
ec5f19d0
KW
1836*/
1837
1838
1839UV
1840Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1841{
7f974d7e
KW
1842 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1843
ec5f19d0
KW
1844 assert(s < send);
1845
1846 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1847 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1848}
1849
52be2536
KW
1850/* This is marked as deprecated
1851 *
ec5f19d0
KW
1852=for apidoc utf8_to_uvuni_buf
1853
de69f3af
KW
1854Only in very rare circumstances should code need to be dealing in Unicode
1855(as opposed to native) code points. In those few cases, use
1856C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1857
1858Returns the Unicode (not-native) code point of the first character in the
1859string C<s> which
ec5f19d0
KW
1860is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1861C<retlen> will be set to the length, in bytes, of that character.
1862
524080c4
KW
1863If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1864enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1865NULL) to -1. If those warnings are off, the computed value if well-defined (or
1866the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1867is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1868next possible position in C<s> that could begin a non-malformed character.
de69f3af 1869See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1870
760c7c2f
KW
1871Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1872unless those are turned off.
1873
ec5f19d0
KW
1874=cut
1875*/
1876
1877UV
1878Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1879{
1880 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1881
1882 assert(send > s);
1883
5962d97e
KW
1884 /* Call the low level routine, asking for checks */
1885 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1886}
1887
b76347f2 1888/*
87cea99e 1889=for apidoc utf8_length
b76347f2
JH
1890
1891Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1892Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1893up past C<e>, croaks.
b76347f2
JH
1894
1895=cut
1896*/
1897
1898STRLEN
35a4481c 1899Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1900{
1901 STRLEN len = 0;
1902
7918f24d
NC
1903 PERL_ARGS_ASSERT_UTF8_LENGTH;
1904
8850bf83
JH
1905 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1906 * the bitops (especially ~) can create illegal UTF-8.
1907 * In other words: in Perl UTF-8 is not just for Unicode. */
1908
a3b680e6
AL
1909 if (e < s)
1910 goto warn_and_return;
b76347f2 1911 while (s < e) {
4cbf4130 1912 s += UTF8SKIP(s);
8e91ec7f
AV
1913 len++;
1914 }
1915
1916 if (e != s) {
1917 len--;
1918 warn_and_return:
9b387841
NC
1919 if (PL_op)
1920 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1921 "%s in %s", unees, OP_DESC(PL_op));
1922 else
61a12c31 1923 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1924 }
1925
1926 return len;
1927}
1928
b06226ff 1929/*
fed3ba5d
NC
1930=for apidoc bytes_cmp_utf8
1931
a1433954 1932Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1933sequence of characters (stored as UTF-8)
1934in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1935equal, -1 or -2 if the first string is less than the second string, +1 or +2
1936if the first string is greater than the second string.
1937
1938-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1939longer string. -2 or +2 is returned if
1940there was a difference between characters
fed3ba5d
NC
1941within the strings.
1942
1943=cut
1944*/
1945
1946int
1947Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1948{
1949 const U8 *const bend = b + blen;
1950 const U8 *const uend = u + ulen;
1951
1952 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1953
1954 while (b < bend && u < uend) {
1955 U8 c = *u++;
1956 if (!UTF8_IS_INVARIANT(c)) {
1957 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1958 if (u < uend) {
1959 U8 c1 = *u++;
1960 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1961 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1962 } else {
2b5e7bc2 1963 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1964 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1965 "%s %s%s",
1966 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1967 PL_op ? " in " : "",
1968 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1969 return -2;
1970 }
1971 } else {
1972 if (PL_op)
1973 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1974 "%s in %s", unees, OP_DESC(PL_op));
1975 else
61a12c31 1976 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1977 return -2; /* Really want to return undef :-) */
1978 }
1979 } else {
1980 return -2;
1981 }
1982 }
1983 if (*b != c) {
1984 return *b < c ? -2 : +2;
1985 }
1986 ++b;
1987 }
1988
1989 if (b == bend && u == uend)
1990 return 0;
1991
1992 return b < bend ? +1 : -1;
1993}
1994
1995/*
87cea99e 1996=for apidoc utf8_to_bytes
6940069f 1997
3bc0c78c 1998Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 1999Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2000updates C<*lenp> to contain the new length.
3bc0c78c
KW
2001Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2002
2003Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2004having saved the value of C<*lenp> before the call, and subtracting the
2005after-call value of C<*lenp> from it.
6940069f 2006
a1433954 2007If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2008
6940069f
GS
2009=cut
2010*/
2011
2012U8 *
09af0336 2013Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2014{
9fe0d3c2 2015 U8 * first_variant;
246fae53 2016
7918f24d 2017 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2018 PERL_UNUSED_CONTEXT;
7918f24d 2019
9fe0d3c2 2020 /* This is a no-op if no variants at all in the input */
09af0336 2021 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2022 return s;
2023 }
2024
2025 {
3c5aa262 2026 U8 * const save = s;
09af0336 2027 U8 * const send = s + *lenp;
3c5aa262
KW
2028 U8 * d;
2029
2030 /* Nothing before the first variant needs to be changed, so start the real
2031 * work there */
2032 s = first_variant;
2033 while (s < send) {
2034 if (! UTF8_IS_INVARIANT(*s)) {
2035 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2036 *lenp = ((STRLEN) -1);
3c5aa262
KW
2037 return 0;
2038 }
2039 s++;
d59937ca
KW
2040 }
2041 s++;
dcad2880 2042 }
dcad2880 2043
3c5aa262
KW
2044 /* Is downgradable, so do it */
2045 d = s = first_variant;
2046 while (s < send) {
2047 U8 c = *s++;
2048 if (! UVCHR_IS_INVARIANT(c)) {
2049 /* Then it is two-byte encoded */
2050 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2051 s++;
2052 }
2053 *d++ = c;
2054 }
2055 *d = '\0';
09af0336 2056 *lenp = d - save;
3c5aa262
KW
2057
2058 return save;
9fe0d3c2 2059 }
6940069f
GS
2060}
2061
2062/*
87cea99e 2063=for apidoc bytes_from_utf8
f9a63242 2064
09af0336 2065Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2066byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2067actually encoded in UTF-8.
2068
2069Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2070the input string.
2071
41ae6089
KW
2072Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2073not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2074C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2075
41ae6089 2076Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2077newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2078returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2079
3bc0c78c 2080Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2081having saved the value of C<*lenp> before the call, and subtracting the
2082after-call value of C<*lenp> from it.
3bc0c78c 2083
37607a96 2084=cut
976c1b08
KW
2085
2086There is a macro that avoids this function call, but this is retained for
2087anyone who calls it with the Perl_ prefix */
f9a63242
JH
2088
2089U8 *
41ae6089 2090Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2091{
7918f24d 2092 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2093 PERL_UNUSED_CONTEXT;
f9a63242 2094
976c1b08
KW
2095 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2096}
2097
2098/*
2099No = here because currently externally undocumented
2100for apidoc bytes_from_utf8_loc
2101
2102Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2103to store the location of the first character in C<"s"> that cannot be
2104converted to non-UTF8.
2105
2106If that parameter is C<NULL>, this function behaves identically to
2107C<bytes_from_utf8>.
2108
2109Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2110C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2111
2112Otherwise, the function returns a newly created C<NUL>-terminated string
2113containing the non-UTF8 equivalent of the convertible first portion of
2114C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2115If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2116and C<*first_non_downgradable> is set to C<NULL>.
2117
2118Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2119first character in the original string that wasn't converted. C<*is_utf8p> is
2120unchanged. Note that the new string may have length 0.
2121
2122Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2123C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2124converts as many characters in it as possible stopping at the first one it
385b74be 2125finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2126set to point to that. The function returns the portion that could be converted
2127in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2128not including the terminating C<NUL>. If the very first character in the
2129original could not be converted, C<*lenp> will be 0, and the new string will
2130contain just a single C<NUL>. If the entire input string was converted,
2131C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2132
2133Upon successful return, the number of variants in the converted portion of the
2134string can be computed by having saved the value of C<*lenp> before the call,
2135and subtracting the after-call value of C<*lenp> from it.
2136
2137=cut
2138
2139
2140*/
2141
2142U8 *
2143Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2144{
2145 U8 *d;
2146 const U8 *original = s;
2147 U8 *converted_start;
2148 const U8 *send = s + *lenp;
f9a63242 2149
976c1b08 2150 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2151
976c1b08
KW
2152 if (! *is_utf8p) {
2153 if (first_unconverted) {
2154 *first_unconverted = NULL;
2155 }
2156
2157 return (U8 *) original;
2158 }
2159
2160 Newx(d, (*lenp) + 1, U8);
2161
2162 converted_start = d;
7299a045
KW
2163 while (s < send) {
2164 U8 c = *s++;
2165 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2166
2167 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2168 * have to stop now */
2169 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2170 if (first_unconverted) {
2171 *first_unconverted = s - 1;
2172 goto finish_and_return;
2173 }
2174 else {
2175 Safefree(converted_start);
2176 return (U8 *) original;
2177 }
2178 }
2179
7299a045
KW
2180 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2181 s++;
38af28cf 2182 }
7299a045
KW
2183 *d++ = c;
2184 }
170a1c22 2185
976c1b08
KW
2186 /* Here, converted the whole of the input */
2187 *is_utf8p = FALSE;
2188 if (first_unconverted) {
2189 *first_unconverted = NULL;
170a1c22 2190 }
976c1b08
KW
2191
2192 finish_and_return:
2193 *d = '\0';
2194 *lenp = d - converted_start;
2195
2196 /* Trim unused space */
2197 Renew(converted_start, *lenp + 1, U8);
2198
2199 return converted_start;
f9a63242
JH
2200}
2201
2202/*
87cea99e 2203=for apidoc bytes_to_utf8
6940069f 2204
09af0336 2205Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2206UTF-8.
09af0336 2207Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2208reflect the new length in bytes.
6940069f 2209
3bc0c78c 2210Upon successful return, the number of variants in the string can be computed by
23b37b12 2211having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2212after-call value of C<*lenp>.
2213
75200dff 2214A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2215
2216If you want to convert to UTF-8 from encodings other than
2217the native (Latin1 or EBCDIC),
a1433954 2218see L</sv_recode_to_utf8>().
c9ada85f 2219
497711e7 2220=cut
6940069f
GS
2221*/
2222
2223U8*
09af0336 2224Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2225{
09af0336 2226 const U8 * const send = s + (*lenp);
6940069f
GS
2227 U8 *d;
2228 U8 *dst;
7918f24d
NC
2229
2230 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2231 PERL_UNUSED_CONTEXT;
6940069f 2232
09af0336 2233 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2234 dst = d;
2235
2236 while (s < send) {
55d09dc8
KW
2237 append_utf8_from_native_byte(*s, &d);
2238 s++;
6940069f
GS
2239 }
2240 *d = '\0';
09af0336 2241 *lenp = d-dst;
6940069f
GS
2242 return dst;
2243}
2244
a0ed51b3 2245/*
dea0fc0b 2246 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2247 *
2248 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2249 * We optimize for native, for obvious reasons. */
2250
2251U8*
dea0fc0b 2252Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2253{
dea0fc0b
JH
2254 U8* pend;
2255 U8* dstart = d;
2256
7918f24d
NC
2257 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2258
dea0fc0b 2259 if (bytelen & 1)
56576a04
KW
2260 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2261 (UV)bytelen);
dea0fc0b
JH
2262
2263 pend = p + bytelen;
2264
a0ed51b3 2265 while (p < pend) {
dea0fc0b
JH
2266 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2267 p += 2;
2d1545e5 2268 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2269 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2270 continue;
2271 }
56d37426
KW
2272 if (uv <= MAX_UTF8_TWO_BYTE) {
2273 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2274 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2275 continue;
2276 }
46956fad
KW
2277#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2278#define LAST_HIGH_SURROGATE 0xDBFF
2279#define FIRST_LOW_SURROGATE 0xDC00
2280#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2281
2282 /* This assumes that most uses will be in the first Unicode plane, not
2283 * needing surrogates */
2284 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2285 && uv <= UNICODE_SURROGATE_LAST))
2286 {
2287 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2288 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2289 }
2290 else {
01ea242b 2291 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2292 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2293 || UNLIKELY(low > LAST_LOW_SURROGATE))
2294 {
01ea242b 2295 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2296 }
2297 p += 2;
46956fad
KW
2298 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2299 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2300 }
a0ed51b3 2301 }
56d37426
KW
2302#ifdef EBCDIC
2303 d = uvoffuni_to_utf8_flags(d, uv, 0);
2304#else
a0ed51b3 2305 if (uv < 0x10000) {
eb160463
GS
2306 *d++ = (U8)(( uv >> 12) | 0xe0);
2307 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2308 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2309 continue;
2310 }
2311 else {
eb160463
GS
2312 *d++ = (U8)(( uv >> 18) | 0xf0);
2313 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2314 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2315 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2316 continue;
2317 }
56d37426 2318#endif
a0ed51b3 2319 }
dea0fc0b 2320 *newlen = d - dstart;
a0ed51b3
LW
2321 return d;
2322}
2323
2324/* Note: this one is slightly destructive of the source. */
2325
2326U8*
dea0fc0b 2327Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2328{
2329 U8* s = (U8*)p;
d4c19fe8 2330 U8* const send = s + bytelen;
7918f24d
NC
2331
2332 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2333
e0ea5e2d 2334 if (bytelen & 1)
147e3846 2335 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2336 (UV)bytelen);
2337
a0ed51b3 2338 while (s < send) {
d4c19fe8 2339 const U8 tmp = s[0];
a0ed51b3
LW
2340 s[0] = s[1];
2341 s[1] = tmp;
2342 s += 2;
2343 }
dea0fc0b 2344 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2345}
2346
922e8cb4
KW
2347bool
2348Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2349{
2350 U8 tmpbuf[UTF8_MAXBYTES+1];
2351 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2352 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2353}
2354
f9ae8fb6
JD
2355/* Internal function so we can deprecate the external one, and call
2356 this one from other deprecated functions in this file */
2357
f2645549
KW
2358bool
2359Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2360{
f2645549 2361 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2362
2363 if (*p == '_')
2364 return TRUE;
f25ce844 2365 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2366}
2367
5092f92a 2368bool
eba68aa0
KW
2369Perl__is_uni_perl_idcont(pTHX_ UV c)
2370{
2371 U8 tmpbuf[UTF8_MAXBYTES+1];
2372 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2373 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2374}
2375
2376bool
f91dcd13
KW
2377Perl__is_uni_perl_idstart(pTHX_ UV c)
2378{
2379 U8 tmpbuf[UTF8_MAXBYTES+1];
2380 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2381 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2382}
2383
3a4c58c9 2384UV
56576a04
KW
2385Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2386 const char S_or_s)
3a4c58c9
KW
2387{
2388 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2389 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2390 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2391 * either "SS" or "Ss". Which one to use is passed into the routine in
2392 * 'S_or_s' to avoid a test */
2393
2394 UV converted = toUPPER_LATIN1_MOD(c);
2395
2396 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2397
2398 assert(S_or_s == 'S' || S_or_s == 's');
2399
6f2d5cbc 2400 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2401 characters in this range */
3a4c58c9
KW
2402 *p = (U8) converted;
2403 *lenp = 1;
2404 return converted;
2405 }
2406
2407 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2408 * which it maps to one of them, so as to only have to have one check for
2409 * it in the main case */
2410 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2411 switch (c) {
2412 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2413 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2414 break;
2415 case MICRO_SIGN:
2416 converted = GREEK_CAPITAL_LETTER_MU;
2417 break;
79e064b9
KW
2418#if UNICODE_MAJOR_VERSION > 2 \
2419 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2420 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2421 case LATIN_SMALL_LETTER_SHARP_S:
2422 *(p)++ = 'S';
2423 *p = S_or_s;
2424 *lenp = 2;
2425 return 'S';
79e064b9 2426#endif
3a4c58c9 2427 default:
56576a04
KW
2428 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2429 " '%c' to map to '%c'",
2430 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2431 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2432 }
2433 }
2434
2435 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2436 *p = UTF8_TWO_BYTE_LO(converted);
2437 *lenp = 2;
2438
2439 return converted;
2440}
2441
50bda2c3
KW
2442/* Call the function to convert a UTF-8 encoded character to the specified case.
2443 * Note that there may be more than one character in the result.
2444 * INP is a pointer to the first byte of the input character
2445 * OUTP will be set to the first byte of the string of changed characters. It
2446 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2447 * LENP will be set to the length in bytes of the string of changed characters
2448 *
56576a04
KW
2449 * The functions return the ordinal of the first character in the string of
2450 * OUTP */
2451#define CALL_UPPER_CASE(uv, s, d, lenp) \
2452 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2453#define CALL_TITLE_CASE(uv, s, d, lenp) \
2454 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2455#define CALL_LOWER_CASE(uv, s, d, lenp) \
2456 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2457
b9992569
KW
2458/* This additionally has the input parameter 'specials', which if non-zero will
2459 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2460 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2461#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2462_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2463
84afefe6
JH
2464UV
2465Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2466{
a1433954
KW
2467 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2468 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2469 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2470 * the changed version may be longer than the original character.
2471 *
2472 * The ordinal of the first character of the changed version is returned
2473 * (but note, as explained above, that there may be more.) */
2474
7918f24d
NC
2475 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2476
3a4c58c9
KW
2477 if (c < 256) {
2478 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2479 }
2480
0ebc6274 2481 uvchr_to_utf8(p, c);
b9992569 2482 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2483}
2484
84afefe6
JH
2485UV
2486Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2487{
7918f24d
NC
2488 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2489
3a4c58c9
KW
2490 if (c < 256) {
2491 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2492 }
2493
0ebc6274 2494 uvchr_to_utf8(p, c);
b9992569 2495 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2496}
2497
afc16117 2498STATIC U8
eaf412bf 2499S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2500{
2501 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2502 * those, converting the result to UTF-8. Since the result is always just
a1433954 2503 * one character, we allow <p> to be NULL */
afc16117
KW
2504
2505 U8 converted = toLOWER_LATIN1(c);
2506
eaf412bf
KW
2507 PERL_UNUSED_ARG(dummy);
2508
afc16117 2509 if (p != NULL) {
6f2d5cbc 2510 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2511 *p = converted;
2512 *lenp = 1;
2513 }
2514 else {
430c9760
KW
2515 /* Result is known to always be < 256, so can use the EIGHT_BIT
2516 * macros */
2517 *p = UTF8_EIGHT_BIT_HI(converted);
2518 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2519 *lenp = 2;
2520 }
2521 }
2522 return converted;
2523}
2524
84afefe6
JH
2525UV
2526Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2527{
7918f24d
NC
2528 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2529
afc16117 2530 if (c < 256) {
eaf412bf 2531 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2532 }
2533
afc16117 2534 uvchr_to_utf8(p, c);
b9992569 2535 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2536}
2537
84afefe6 2538UV
56576a04
KW
2539Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2540 const unsigned int flags)
a1dde8de 2541{
51910141 2542 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2543 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2544 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2545 *
2546 * Not to be used for locale folds
51910141 2547 */
f673fad4 2548
a1dde8de
KW
2549 UV converted;
2550
2551 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2552 PERL_UNUSED_CONTEXT;
a1dde8de 2553
1ca267a5
KW
2554 assert (! (flags & FOLD_FLAGS_LOCALE));
2555
659a7c2d 2556 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2557 converted = GREEK_SMALL_LETTER_MU;
2558 }
9b63e895
KW
2559#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2560 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2561 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2562 else if ( (flags & FOLD_FLAGS_FULL)
2563 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2564 {
1ca267a5
KW
2565 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2566 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2567 * under those circumstances. */
2568 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2569 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2570 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2571 p, *lenp, U8);
2572 return LATIN_SMALL_LETTER_LONG_S;
2573 }
2574 else {
4f489194
KW
2575 *(p)++ = 's';
2576 *p = 's';
2577 *lenp = 2;
2578 return 's';
1ca267a5 2579 }
a1dde8de 2580 }
9b63e895 2581#endif
a1dde8de
KW
2582 else { /* In this range the fold of all other characters is their lower
2583 case */
2584 converted = toLOWER_LATIN1(c);
2585 }
2586
6f2d5cbc 2587 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2588 *p = (U8) converted;
2589 *lenp = 1;
2590 }
2591 else {
2592 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2593 *p = UTF8_TWO_BYTE_LO(converted);
2594 *lenp = 2;
2595 }
2596
2597 return converted;
2598}
2599
2600UV
31f05a37 2601Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2602{
4b593389 2603
a0270393
KW
2604 /* Not currently externally documented, and subject to change
2605 * <flags> bits meanings:
2606 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2607 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2608 * locale are to be used.
a0270393
KW
2609 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2610 */
4b593389 2611
36bb2ab6 2612 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2613
780fcc9f
KW
2614 if (flags & FOLD_FLAGS_LOCALE) {
2615 /* Treat a UTF-8 locale as not being in locale at all */
2616 if (IN_UTF8_CTYPE_LOCALE) {
2617 flags &= ~FOLD_FLAGS_LOCALE;
2618 }
2619 else {
2620 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2621 goto needs_full_generality;
780fcc9f 2622 }
31f05a37
KW
2623 }
2624
a1dde8de 2625 if (c < 256) {
e7b7ac46 2626 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2627 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2628 }
2629
2f306ab9 2630 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2631 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2632 uvchr_to_utf8(p, c);
b9992569 2633 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2634 }
567b353c 2635 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2636 the special flags. */
2637 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2638
2639 needs_full_generality:
a0270393 2640 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2641 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2642 p, lenp, flags);
a0270393 2643 }
84afefe6
JH
2644}
2645
26483009 2646PERL_STATIC_INLINE bool
5141f98e 2647S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2648 const char *const swashname, SV* const invlist)
bde6a22d 2649{
ea317ccb
KW
2650 /* returns a boolean giving whether or not the UTF8-encoded character that
2651 * starts at <p> is in the swash indicated by <swashname>. <swash>
2652 * contains a pointer to where the swash indicated by <swashname>
2653 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2654 * look at <*swash> and only generate a swash if it is not null. <invlist>
2655 * is NULL or an inversion list that defines the swash. If not null, it
2656 * saves time during initialization of the swash.
ea317ccb
KW
2657 *
2658 * Note that it is assumed that the buffer length of <p> is enough to
2659 * contain all the bytes that comprise the character. Thus, <*p> should
2660 * have been checked before this call for mal-formedness enough to assure
2661 * that. */
2662
7918f24d
NC
2663 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2664
492a624f 2665 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2666 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2667 * as far as there being enough bytes available in it to accommodate the
2668 * character without reading beyond the end, and pass that number on to the
2669 * validating routine */
6302f837 2670 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2671 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2672 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2673 1 /* Die */ );
2674 NOT_REACHED; /* NOTREACHED */
28123549 2675 }
86ae6e94 2676
87367d5f
KW
2677 if (!*swash) {
2678 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2679 *swash = _core_swash_init("utf8",
2680
2681 /* Only use the name if there is no inversion
2682 * list; otherwise will go out to disk */
2683 (invlist) ? "" : swashname,
2684
2685 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2686 }
28123549 2687
bde6a22d
NC
2688 return swash_fetch(*swash, p, TRUE) != 0;
2689}
2690
da8c1a98 2691PERL_STATIC_INLINE bool
56576a04
KW
2692S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2693 SV **swash, const char *const swashname,
2694 SV* const invlist)
da8c1a98
KW
2695{
2696 /* returns a boolean giving whether or not the UTF8-encoded character that
2697 * starts at <p>, and extending no further than <e - 1> is in the swash
2698 * indicated by <swashname>. <swash> contains a pointer to where the swash
2699 * indicated by <swashname> is to be stored; which this routine will do, so
2700 * that future calls will look at <*swash> and only generate a swash if it
2701 * is not null. <invlist> is NULL or an inversion list that defines the
2702 * swash. If not null, it saves time during initialization of the swash.
2703 */
2704
2705 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2706
2707 if (! isUTF8_CHAR(p, e)) {
2708 _force_out_malformed_utf8_message(p, e, 0, 1);
2709 NOT_REACHED; /* NOTREACHED */
2710 }
2711
2712 if (!*swash) {
2713 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2714 *swash = _core_swash_init("utf8",
2715
2716 /* Only use the name if there is no inversion
2717 * list; otherwise will go out to disk */
2718 (invlist) ? "" : swashname,
2719
2720 &PL_sv_undef, 1, 0, invlist, &flags);
2721 }
2722
2723 return swash_fetch(*swash, p, TRUE) != 0;
2724}
2725
34aeb2e9
KW
2726STATIC void
2727S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2728 const char * const alternative,
2729 const bool use_locale,
2730 const char * const file,
2731 const unsigned line)
2732{
2733 const char * key;
2734
2735 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2736
2737 if (ckWARN_d(WARN_DEPRECATED)) {
2738
2739 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2740 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2741 if (! PL_seen_deprecated_macro) {
2742 PL_seen_deprecated_macro = newHV();
2743 }
2744 if (! hv_store(PL_seen_deprecated_macro, key,
2745 strlen(key), &PL_sv_undef, 0))
2746 {
2747 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2748 }
2749
c44e9413 2750 if (instr(file, "mathoms.c")) {
607313a1
KW
2751 Perl_warner(aTHX_ WARN_DEPRECATED,
2752 "In %s, line %d, starting in Perl v5.30, %s()"
2753 " will be removed. Avoid this message by"
2754 " converting to use %s().\n",
2755 file, line, name, alternative);
2756 }
2757 else {
34aeb2e9
KW
2758 Perl_warner(aTHX_ WARN_DEPRECATED,
2759 "In %s, line %d, starting in Perl v5.30, %s() will"
2760 " require an additional parameter. Avoid this"
2761 " message by converting to use %s().\n",
2762 file, line, name, alternative);
607313a1 2763 }
34aeb2e9
KW
2764 }
2765 }
2766}
2767
bde6a22d 2768bool
34aeb2e9 2769Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2770 const U8 * const p,
34aeb2e9
KW
2771 const char * const name,
2772 const char * const alternative,
2773 const bool use_utf8,
2774 const bool use_locale,
2775 const char * const file,
2776 const unsigned line)
922e8cb4 2777{
922e8cb4
KW
2778 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2779
34aeb2e9
KW
2780 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2781
2782 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2783
2784 switch (classnum) {
2785 case _CC_WORDCHAR:
2786 case _CC_DIGIT:
2787 case _CC_ALPHA:
2788 case _CC_LOWER:
2789 case _CC_UPPER:
2790 case _CC_PUNCT:
2791 case _CC_PRINT:
2792 case _CC_ALPHANUMERIC:
2793 case _CC_GRAPH:
2794 case _CC_CASED:
2795
2796 return is_utf8_common(p,
2797 &PL_utf8_swash_ptrs[classnum],
2798 swash_property_names[classnum],
2799 PL_XPosix_ptrs[classnum]);
2800
2801 case _CC_SPACE:
2802 return is_XPERLSPACE_high(p);
2803 case _CC_BLANK:
2804 return is_HORIZWS_high(p);
2805 case _CC_XDIGIT:
2806 return is_XDIGIT_high(p);
2807 case _CC_CNTRL:
2808 return 0;
2809 case _CC_ASCII:
2810 return 0;
2811 case _CC_VERTSPACE:
2812 return is_VERTWS_high(p);
2813 case _CC_IDFIRST:
2814 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2815 PL_utf8_perl_idstart
2816 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2817 }
22f0498f
KW
2818 return is_utf8_common(p, &PL_utf8_perl_idstart,
2819 "_Perl_IDStart", NULL);
34aeb2e9
KW
2820 case _CC_IDCONT:
2821 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2822 PL_utf8_perl_idcont
2823 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2824 }
22f0498f
KW
2825 return is_utf8_common(p, &PL_utf8_perl_idcont,
2826 "_Perl_IDCont", NULL);
34aeb2e9
KW
2827 }
2828 }
2829
2830 /* idcont is the same as wordchar below 256 */
2831 if (classnum == _CC_IDCONT) {
2832 classnum = _CC_WORDCHAR;
2833 }
2834 else if (classnum == _CC_IDFIRST) {
2835 if (*p == '_') {
2836 return TRUE;
2837 }
2838 classnum = _CC_ALPHA;
2839 }
2840
2841 if (! use_locale) {
2842 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2843 return _generic_isCC(*p, classnum);
2844 }
922e8cb4 2845
34aeb2e9
KW
2846 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2847 }
2848 else {
2849 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2850 return isFOO_lc(classnum, *p);
2851 }
2852
2853 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2854 }
2855
2856 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2857}
2858
2859bool
da8c1a98
KW
2860Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2861 const U8 * const e)
2862{
2863 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2864
2865 assert(classnum < _FIRST_NON_SWASH_CC);
2866
2867 return is_utf8_common_with_len(p,
2868 e,
2869 &PL_utf8_swash_ptrs[classnum],
2870 swash_property_names[classnum],
2871 PL_XPosix_ptrs[classnum]);
2872}
2873
2874bool
da8c1a98
KW
2875Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2876{
2877 SV* invlist = NULL;
2878
2879 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2880
2881 if (! PL_utf8_perl_idstart) {
2882 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2883 }
2884 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2885 "_Perl_IDStart", invlist);
2886}
2887
2888bool
f2645549 2889Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2890{
f2645549 2891 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2892
2893 if (*p == '_')
2894 return TRUE;
f25ce844 2895 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2896}
2897
2898bool
da8c1a98
KW
2899Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2900{
2901 SV* invlist = NULL;
2902
2903 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2904
2905 if (! PL_utf8_perl_idcont) {
2906 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2907 }
2908 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2909 "_Perl_IDCont", invlist);
2910}
2911
2912bool
f2645549 2913Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2914{
f2645549 2915 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2916
f25ce844 2917 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2918}
2919
2920bool
f2645549 2921Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2922{
f2645549 2923 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2924
f25ce844 2925 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2926}
2927
2928bool
7dbf68d2
KW
2929Perl__is_utf8_mark(pTHX_ const U8 *p)
2930{
7dbf68d2
KW
2931 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2932
f25ce844 2933 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2934}
2935
b9992569 2936 /* change namve uv1 to 'from' */
6a4a25f4 2937STATIC UV
b9992569
KW
2938S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2939 SV **swashp, const char *normal, const char *special)
2940{
0134edef 2941 STRLEN len = 0;
7918f24d 2942
b9992569 2943 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2944
36eaa811
KW
2945 /* For code points that don't change case, we already know that the output
2946 * of this function is the unchanged input, so we can skip doing look-ups
2947 * for them. Unfortunately the case-changing code points are scattered
2948 * around. But there are some long consecutive ranges where there are no
2949 * case changing code points. By adding tests, we can eliminate the lookup
2950 * for all the ones in such ranges. This is currently done here only for
2951 * just a few cases where the scripts are in common use in modern commerce
2952 * (and scripts adjacent to those which can be included without additional
2953 * tests). */
2954
2955 if (uv1 >= 0x0590) {
2956 /* This keeps from needing further processing the code points most
2957 * likely to be used in the following non-cased scripts: Hebrew,
2958 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2959 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2960 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2961 if (uv1 < 0x10A0) {
2962 goto cases_to_self;
2963 }
2964
2965 /* The following largish code point ranges also don't have case
2966 * changes, but khw didn't think they warranted extra tests to speed
2967 * them up (which would slightly slow down everything else above them):
2968 * 1100..139F Hangul Jamo, Ethiopic
2969 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2970 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2971 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2972 * Combining Diacritical Marks Extended, Balinese,
2973 * Sundanese, Batak, Lepcha, Ol Chiki
2974 * 2000..206F General Punctuation
2975 */
2976
2977 if (uv1 >= 0x2D30) {
2978
2979 /* This keeps the from needing further processing the code points
2980 * most likely to be used in the following non-cased major scripts:
2981 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2982 *
2983 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2984 * event that Unicode eventually allocates the unused block as of
2985 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2986 * that the test suite will start having failures to alert you
2987 * should that happen) */
2988 if (uv1 < 0xA640) {
2989 goto cases_to_self;
2990 }
2991
2992 if (uv1 >= 0xAC00) {
2993 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2994 if (ckWARN_d(WARN_SURROGATE)) {
2995 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2996 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
2997 "Operation \"%s\" returns its argument for"
2998 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2999 }
3000 goto cases_to_self;
3001 }
36eaa811
KW
3002
3003 /* AC00..FAFF Catches Hangul syllables and private use, plus
3004 * some others */
3005 if (uv1 < 0xFB00) {
3006 goto cases_to_self;
3007
3008 }
3009
5af9bc97 3010 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
3011 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
3012 && ckWARN_d(WARN_DEPRECATED))
3013 {
3014 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
3015 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
3016 }
3017 if (ckWARN_d(WARN_NON_UNICODE)) {
3018 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3019 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3020 "Operation \"%s\" returns its argument for"
3021 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3022 }
3023 goto cases_to_self;
3024 }
3bfc1e70
KW
3025#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3026 if (UNLIKELY(uv1
3027 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3028 {
3029
56576a04
KW
3030 /* As of Unicode 10.0, this means we avoid swash creation
3031 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3032 goto cases_to_self;
3033 }
3034#endif
36eaa811
KW
3035 }
3036 }
9ae3ac1a 3037
36eaa811
KW
3038 /* Note that non-characters are perfectly legal, so no warning should
3039 * be given. There are so few of them, that it isn't worth the extra
3040 * tests to avoid swash creation */
9ae3ac1a
KW
3041 }
3042
0134edef 3043 if (!*swashp) /* load on-demand */
56576a04
KW
3044 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3045 4, 0, NULL, NULL);
0134edef 3046
a6f87d8c 3047 if (special) {
0134edef 3048 /* It might be "special" (sometimes, but not always,
2a37f04d 3049 * a multicharacter mapping) */
4a8240a3 3050 HV *hv = NULL;
b08cf34e
JH
3051 SV **svp;
3052
4a8240a3
KW
3053 /* If passed in the specials name, use that; otherwise use any
3054 * given in the swash */
3055 if (*special != '\0') {
3056 hv = get_hv(special, 0);
3057 }
3058 else {
3059 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3060 if (svp) {
3061 hv = MUTABLE_HV(SvRV(*svp));
3062 }
3063 }
3064
176fe009 3065 if (hv
5f560d8a 3066 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3067 && (*svp))
3068 {
cfd0369c 3069 const char *s;
47654450 3070
cfd0369c 3071 s = SvPV_const(*svp, len);
47654450 3072 if (len == 1)
f4cd282c 3073 /* EIGHTBIT */
c80e42f3 3074 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3075 else {
d2dcd0fb 3076 Copy(s, ustrp, len, U8);
29e98929 3077 }
983ffd37 3078 }
0134edef
JH
3079 }
3080
3081 if (!len && *swashp) {
4a4088c4 3082 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3083
0134edef
JH
3084 if (uv2) {
3085 /* It was "normal" (a single character mapping). */
f4cd282c 3086 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3087 }
3088 }
1feea2c7 3089
cbe07460
KW
3090 if (len) {
3091 if (lenp) {
3092 *lenp = len;
3093 }
3094 return valid_utf8_to_uvchr(ustrp, 0);
3095 }
3096
3097 /* Here, there was no mapping defined, which means that the code point maps
3098 * to itself. Return the inputs */
e24dfe9c 3099 cases_to_self:
bfdf22ec 3100 len = UTF8SKIP(p);
ca9fab46
KW
3101 if (p != ustrp) { /* Don't copy onto itself */
3102 Copy(p, ustrp, len, U8);
3103 }
0134edef 3104
2a37f04d
JH
3105 if (lenp)
3106 *lenp = len;
3107
f4cd282c 3108 return uv1;
cbe07460 3109
a0ed51b3
LW
3110}
3111
051a06d4 3112STATIC UV
56576a04
KW
3113S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3114 U8* const ustrp, STRLEN *lenp)
051a06d4 3115{
4a4088c4 3116 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3117 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3118 * result contains a character that crosses the 255/256 boundary, disallow
3119 * the change, and return the original code point. See L<perlfunc/lc> for
3120 * why;
051a06d4 3121 *
a1433954
KW
3122 * p points to the original string whose case was changed; assumed
3123 * by this routine to be well-formed
051a06d4 3124 * result the code point of the first character in the changed-case string
56576a04
KW
3125 * ustrp points to the changed-case string (<result> represents its
3126 * first char)
051a06d4
KW
3127 * lenp points to the length of <ustrp> */
3128
3129 UV original; /* To store the first code point of <p> */
3130
3131 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3132
a4f12ed7 3133 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3134
3135 /* We know immediately if the first character in the string crosses the
3136 * boundary, so can skip */
3137 if (result > 255) {
3138
3139 /* Look at every character in the result; if any cross the
3140 * boundary, the whole thing is disallowed */
3141 U8* s = ustrp + UTF8SKIP(ustrp);
3142 U8* e = ustrp + *lenp;
3143 while (s < e) {
a4f12ed7 3144 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3145 goto bad_crossing;
3146 }
3147 s += UTF8SKIP(s);
3148 }
3149
613abc6d
KW
3150 /* Here, no characters crossed, result is ok as-is, but we warn. */
3151 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3152 return result;
3153 }
3154
7b52d656 3155 bad_crossing:
051a06d4
KW
3156
3157 /* Failed, have to return the original */
4b88fb76 3158 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3159
3160 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3161 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3162 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3163 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3164 OP_DESC(PL_op),
ab0b796c
KW
3165 original,
3166 original);
051a06d4
KW
3167 Copy(p, ustrp, *lenp, char);
3168 return original;
3169}
3170
607313a1
KW
3171STATIC U32
3172S_check_and_deprecate(pTHX_ const U8 *p,
3173 const U8 **e,
3174 const unsigned int type, /* See below */
3175 const bool use_locale, /* Is this a 'LC_'
3176 macro call? */
3177 const char * const file,
3178 const unsigned line)
3179{
3180 /* This is a temporary function to deprecate the unsafe calls to the case
3181 * changing macros and functions. It keeps all the special stuff in just
3182 * one place.
3183 *
3184 * It updates *e with the pointer to the end of the input string. If using
3185 * the old-style macros, *e is NULL on input, and so this function assumes
3186 * the input string is long enough to hold the entire UTF-8 sequence, and
3187 * sets *e accordingly, but it then returns a flag to pass the
3188 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3189 * using the full length if possible.
3190 *
3191 * It also does the assert that *e > p when *e is not NULL. This should be
3192 * migrated to the callers when this function gets deleted.
3193 *
3194 * The 'type' parameter is used for the caller to specify which case
3195 * changing function this is called from: */
3196
3197# define DEPRECATE_TO_UPPER 0
3198# define DEPRECATE_TO_TITLE 1
3199# define DEPRECATE_TO_LOWER 2
3200# define DEPRECATE_TO_FOLD 3
3201
3202 U32 utf8n_flags = 0;
3203 const char * name;
3204 const char * alternative;
3205
3206 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3207
3208 if (*e == NULL) {
3209 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3210 *e = p + UTF8SKIP(p);
3211
3212 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3213 * there. It could be part of a larger path */
607313a1 3214 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3215 name = instr(file, "mathoms.c")
607313a1
KW
3216 ? "to_utf8_upper"
3217 : "toUPPER_utf8";
3218 alternative = "toUPPER_utf8_safe";
3219 }
3220 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3221 name = instr(file, "mathoms.c")
607313a1
KW
3222 ? "to_utf8_title"
3223 : "toTITLE_utf8";
3224 alternative = "toTITLE_utf8_safe";
3225 }
3226 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3227 name = instr(file, "mathoms.c")
607313a1
KW
3228 ? "to_utf8_lower"
3229 : "toLOWER_utf8";
3230 alternative = "toLOWER_utf8_safe";
3231 }
3232 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3233 name = instr(file, "mathoms.c")
607313a1
KW
3234 ? "to_utf8_fold"
3235 : "toFOLD_utf8";
3236 alternative = "toFOLD_utf8_safe";
3237 }
3238 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3239
3240 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3241 }
3242 else {
3243 assert (p < *e);
3244 }
3245
3246 return utf8n_flags;
3247}
3248
eaf412bf
KW
3249/* The process for changing the case is essentially the same for the four case
3250 * change types, except there are complications for folding. Otherwise the
3251 * difference is only which case to change to. To make sure that they all do
3252 * the same thing, the bodies of the functions are extracted out into the
3253 * following two macros. The functions are written with the same variable
3254 * names, and these are known and used inside these macros. It would be
3255 * better, of course, to have inline functions to do it, but since different
3256 * macros are called, depending on which case is being changed to, this is not
3257 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3258 * function can start with the common start macro, then finish with its special
3259 * handling; while the other three cases can just use the common end macro.
3260 *
3261 * The algorithm is to use the proper (passed in) macro or function to change
3262 * the case for code points that are below 256. The macro is used if using
3263 * locale rules for the case change; the function if not. If the code point is
3264 * above 255, it is computed from the input UTF-8, and another macro is called
3265 * to do the conversion. If necessary, the output is converted to UTF-8. If
3266 * using a locale, we have to check that the change did not cross the 255/256
3267 * boundary, see check_locale_boundary_crossing() for further details.
3268 *
3269 * The macros are split with the correct case change for the below-256 case
3270 * stored into 'result', and in the middle of an else clause for the above-255
3271 * case. At that point in the 'else', 'result' is not the final result, but is
3272 * the input code point calculated from the UTF-8. The fold code needs to
3273 * realize all this and take it from there.
3274 *
3275 * If you read the two macros as sequential, it's easier to understand what's
3276 * going on. */
3277#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3278 L1_func_extra_param) \
a239b1e2 3279 \
eaf412bf
KW
3280 if (flags & (locale_flags)) { \
3281 /* Treat a UTF-8 locale as not being in locale at all */ \
3282 if (IN_UTF8_CTYPE_LOCALE) { \
3283 flags &= ~(locale_flags); \
3284 } \
3285 else { \
3286 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3287 } \
3288 } \
3289 \
3290 if (UTF8_IS_INVARIANT(*p)) { \
3291 if (flags & (locale_flags)) { \
3292 result = LC_L1_change_macro(*p); \
3293 } \
3294 else { \
3295 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3296 } \
3297 } \
a239b1e2 3298 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3299 if (flags & (locale_flags)) { \
3300 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3301 *(p+1))); \
3302 } \
3303 else { \
3304 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3305 ustrp, lenp, L1_func_extra_param); \
3306 } \
3307 } \
fa8ab374
KW
3308 else { /* malformed UTF-8 or ord above 255 */ \
3309 STRLEN len_result; \
fa8ab374
KW
3310 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3311 if (len_result == (STRLEN) -1) { \
607313a1
KW
3312 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3313 1 /* Die */ ); \
fa8ab374 3314 }
eaf412bf
KW
3315
3316#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3317 result = change_macro(result, p, ustrp, lenp); \
3318 \
3319 if (flags & (locale_flags)) { \
3320 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3321 } \
3322 return result; \
3323 } \
3324 \
3325 /* Here, used locale rules. Convert back to UTF-8 */ \
3326 if (UTF8_IS_INVARIANT(result)) { \
3327 *ustrp = (U8) result; \
3328 *lenp = 1; \
3329 } \
3330 else { \
3331 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3332 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3333 *lenp = 2; \
3334 } \
3335 \
3336 return result;
3337
d3e79532 3338/*
87cea99e 3339=for apidoc to_utf8_upper
d3e79532 3340
a239b1e2 3341Instead use L</toUPPER_utf8_safe>.
a1433954 3342
d3e79532
JH
3343=cut */
3344
051a06d4 3345/* Not currently externally documented, and subject to change:
31f05a37
KW
3346 * <flags> is set iff iff the rules from the current underlying locale are to
3347 * be used. */
051a06d4 3348
2104c8d9 3349UV
607313a1
KW
3350Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3351 const U8 *e,
3352 U8* ustrp,
3353 STRLEN *lenp,
3354 bool flags,
3355 const char * const file,
3356 const int line)
a0ed51b3 3357{
051a06d4 3358 UV result;
607313a1
KW
3359 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3360 cBOOL(flags), file, line);
051a06d4
KW
3361
3362 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3363
eaf412bf
KW
3364 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3365 /* 2nd char of uc(U+DF) is 'S' */
3366 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3367 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3368}
a0ed51b3 3369
d3e79532 3370/*
87cea99e 3371=for apidoc to_utf8_title
d3e79532 3372
a239b1e2 3373Instead use L</toTITLE_utf8_safe>.
a1433954 3374
d3e79532
JH
3375=cut */
3376
051a06d4 3377/* Not currently externally documented, and subject to change:
31f05a37
KW
3378 * <flags> is set iff the rules from the current underlying locale are to be
3379 * used. Since titlecase is not defined in POSIX, for other than a
3380 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3381 */
051a06d4 3382
983ffd37 3383UV
607313a1
KW
3384Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3385 const U8 *e,
3386 U8* ustrp,
3387 STRLEN *lenp,
3388 bool flags,
3389 const char * const file,
3390 const int line)
983ffd37 3391{
051a06d4 3392 UV result;
607313a1
KW
3393 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3394 cBOOL(flags), file, line);
051a06d4
KW
3395
3396 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3397
eaf412bf
KW
3398 /* 2nd char of ucfirst(U+DF) is 's' */
3399 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3400 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3401}
3402
d3e79532 3403/*
87cea99e 3404=for apidoc to_utf8_lower
d3e79532 3405
a239b1e2 3406Instead use L</toLOWER_utf8_safe>.
a1433954 3407
d3e79532
JH
3408=cut */
3409
051a06d4 3410/* Not currently externally documented, and subject to change:
31f05a37
KW
3411 * <flags> is set iff iff the rules from the current underlying locale are to
3412 * be used.
3413 */
051a06d4 3414
2104c8d9 3415UV
607313a1
KW
3416Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3417 const U8 *e,
3418 U8* ustrp,
3419 STRLEN *lenp,
3420 bool flags,
3421 const char * const file,
3422 const int line)
a0ed51b3 3423{
051a06d4 3424 UV result;
607313a1
KW
3425 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3426 cBOOL(flags), file, line);
051a06d4 3427
051a06d4 3428 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3429
eaf412bf
KW
3430 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3431 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3432}
3433
d3e79532 3434/*
87cea99e 3435=for apidoc to_utf8_fold
d3e79532 3436
a239b1e2 3437Instead use L</toFOLD_utf8_safe>.
a1433954 3438
d3e79532
JH
3439=cut */
3440
051a06d4
KW
3441/* Not currently externally documented, and subject to change,
3442 * in <flags>
31f05a37
KW
3443 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3444 * locale are to be used.
051a06d4
KW
3445 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3446 * otherwise simple folds
a0270393
KW
3447 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3448 * prohibited
445bf929 3449 */
36bb2ab6 3450
b4e400f9 3451UV
607313a1
KW
3452Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3453 const U8 *e,
3454 U8* ustrp,
3455 STRLEN *lenp,
3456 U8 flags,
3457 const char * const file,
3458 const int line)
b4e400f9 3459{
051a06d4 3460 UV result;
607313a1
KW
3461 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3462 cBOOL(flags), file, line);
051a06d4 3463
36bb2ab6 3464 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3465
a0270393
KW
3466 /* These are mutually exclusive */
3467 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3468
50ba90ff
KW
3469 assert(p != ustrp); /* Otherwise overwrites */
3470
eaf412bf
KW
3471 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3472 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3473
eaf412bf 3474 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3475
1ca267a5
KW
3476 if (flags & FOLD_FLAGS_LOCALE) {
3477
76f2ffcd 3478# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3479 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3480
3481# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3482# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3483
3484 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3485
538e84ed
KW
3486 /* Special case these two characters, as what normally gets
3487 * returned under locale doesn't work */
76f2ffcd
KW
3488 if (UTF8SKIP(p) == cap_sharp_s_len
3489 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3490 {
ab0b796c
KW
3491 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3492 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3493 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3494 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3495 goto return_long_s;
3496 }
0766489e
KW
3497 else
3498#endif
3499 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3500 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3501 {
ab0b796c
KW
3502 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3503 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3504 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3505 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3506 goto return_ligature_st;
3507 }
74894415
KW
3508
3509#if UNICODE_MAJOR_VERSION == 3 \
3510 && UNICODE_DOT_VERSION == 0 \
3511 && UNICODE_DOT_DOT_VERSION == 1
3512# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3513
3514 /* And special case this on this Unicode version only, for the same
3515 * reaons the other two are special cased. They would cross the
3516 * 255/256 boundary which is forbidden under /l, and so the code
3517 * wouldn't catch that they are equivalent (which they are only in
3518 * this release) */
3519 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3520 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3521 {
3522 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3523 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3524 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3525 "resolved to \"\\x{0131}\".");
3526 goto return_dotless_i;
3527 }
3528#endif
3529
357aadde 3530 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3531 }
a0270393
KW
3532 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3533 return result;
3534 }
3535 else {
4a4088c4 3536 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3537 * character above the ASCII range, and the result should not
3538 * contain an ASCII character. */
a0270393
KW
3539
3540 UV original; /* To store the first code point of <p> */
3541
3542 /* Look at every character in the result; if any cross the
3543 * boundary, the whole thing is disallowed */
3544 U8* s = ustrp;
3545 U8* e = ustrp + *lenp;
3546 while (s < e) {
3547 if (isASCII(*s)) {
3548 /* Crossed, have to return the original */
3549 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3550
9fc2026f 3551 /* But in these instances, there is an alternative we can
1ca267a5 3552 * return that is valid */
0766489e
KW
3553 if (original == LATIN_SMALL_LETTER_SHARP_S
3554#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3555 || original == LATIN_CAPITAL_LETTER_SHARP_S
3556#endif
3557 ) {
1ca267a5
KW
3558 goto return_long_s;
3559 }
9fc2026f
KW
3560 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3561 goto return_ligature_st;
3562 }
74894415
KW
3563#if UNICODE_MAJOR_VERSION == 3 \
3564 && UNICODE_DOT_VERSION == 0 \
3565 && UNICODE_DOT_DOT_VERSION == 1
3566
3567 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3568 goto return_dotless_i;
3569 }
3570#endif
a0270393
KW
3571 Copy(p, ustrp, *lenp, char);
3572 return original;
3573 }
3574 s += UTF8SKIP(s);
3575 }
051a06d4 3576
a0270393
KW
3577 /* Here, no characters crossed, result is ok as-is */
3578 return result;
3579 }
051a06d4
KW
3580 }
3581
4a4088c4 3582 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3583 if (UTF8_IS_INVARIANT(result)) {
3584 *ustrp = (U8) result;
3585 *lenp = 1;
3586 }
3587 else {
62cb07ea
KW
3588 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3589 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3590 *lenp = 2;
3591 }
3592
051a06d4 3593 return result;
1ca267a5
KW
3594
3595 return_long_s:
3596 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3597 * folds to a string of two of these characters. By returning this
3598 * instead, then, e.g.,
3599 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3600 * works. */
3601
3602 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3603 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3604 ustrp, *lenp, U8);
3605 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3606
3607 return_ligature_st:
3608 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3609 * have the other one fold to it */
3610
3611 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3612 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3613 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3614
3615#if UNICODE_MAJOR_VERSION == 3 \
3616 && UNICODE_DOT_VERSION == 0 \
3617 && UNICODE_DOT_DOT_VERSION == 1
3618
3619 return_dotless_i:
3620 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3621 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3622 return LATIN_SMALL_LETTER_DOTLESS_I;
3623
3624#endif
3625
a0ed51b3
LW
3626}
3627
711a919c 3628/* Note:
f90a9a02 3629 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3630 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3631 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3632 */
c4a5db0c 3633
a0ed51b3 3634SV*
56576a04
KW
3635Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3636 I32 minbits, I32 none)
a0ed51b3 3637{
c4a5db0c
KW
3638 PERL_ARGS_ASSERT_SWASH_INIT;
3639
3640 /* Returns a copy of a swash initiated by the called function. This is the
3641 * public interface, and returning a copy prevents others from doing
3642 * mischief on the original */
3643
56576a04
KW
3644 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3645 NULL, NULL));
c4a5db0c
KW
3646}
3647
3648SV*
56576a04
KW
3649Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3650 I32 minbits, I32 none, SV* invlist,
3651 U8* const flags_p)
c4a5db0c 3652{
2c1f00b9
YO
3653
3654 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3655 * use the following define */
3656
3657#define CORE_SWASH_INIT_RETURN(x) \
3658 PL_curpm= old_PL_curpm; \
3659 return x
3660
c4a5db0c 3661 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3662 * by calling utf8_heavy.pl in the general case. The returned value may be
3663 * the swash's inversion list instead if the input parameters allow it.
3664 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3665 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3666 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3667 *
3668 * This interface should only be used by functions that won't destroy or
3669 * adversely change the swash, as doing so affects all other uses of the
3670 * swash in the program; the general public should use 'Perl_swash_init'
3671 * instead.
3672 *
3673 * pkg is the name of the package that <name> should be in.
3674 * name is the name of the swash to find. Typically it is a Unicode
3675 * property name, including user-defined ones
3676 * listsv is a string to initialize the swash with. It must be of the form
3677 * documented as the subroutine return value in
3678 * L<perlunicode/User-Defined Character Properties>
3679 * minbits is the number of bits required to represent each data element.
3680 * It is '1' for binary properties.
3681 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3682 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3683 * flags_p if non-NULL is the address of various input and output flag bits
3684 * to the routine, as follows: ('I' means is input to the routine;
3685 * 'O' means output from the routine. Only flags marked O are
3686 * meaningful on return.)
3687 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3688 * came from a user-defined property. (I O)
5d3d13d1
KW
3689 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3690 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3691 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3692 * return of an inversion list instead of a swash hash if this routine
3693 * thinks that would result in faster execution of swash_fetch() later
3694 * on. (I)
9a53f6cf
KW
3695 *
3696 * Thus there are three possible inputs to find the swash: <name>,
3697 * <listsv>, and <invlist>. At least one must be specified. The result
3698 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3699 * actions can intersect, etc. what <name> gives. To avoid going out to
3700 * disk at all, <invlist> should specify completely what the swash should
3701 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3702 *
3703 * <invlist> is only valid for binary properties */
c4a5db0c 3704
2c1f00b9
YO
3705 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3706
c4a5db0c 3707 SV* retval = &PL_sv_undef;
83199d38 3708 HV* swash_hv = NULL;
87367d5f
KW
3709 const int invlist_swash_boundary =
3710 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3711 ? 512 /* Based on some benchmarking, but not extensive, see commit
3712 message */
3713 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3714
3715 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3716 assert(! invlist || minbits == 1);
3717
56576a04
KW
3718 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3719 regex that triggered the swash init and the swash init