This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
APItest/t/utf8_warn_base.pl: Add some tests
[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";
125const char above_31_bit_cp_format[] = "Code point 0x%" UVXf " is not"
126 " Unicode, and not portable";
127
8ee1cdcb
KW
128#define HANDLE_UNICODE_SURROGATE(uv, flags) \
129 STMT_START { \
130 if (flags & UNICODE_WARN_SURROGATE) { \
131 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
c94c2f39 132 surrogate_cp_format, uv); \
8ee1cdcb
KW
133 } \
134 if (flags & UNICODE_DISALLOW_SURROGATE) { \
135 return NULL; \
136 } \
137 } STMT_END;
138
139#define HANDLE_UNICODE_NONCHAR(uv, flags) \
140 STMT_START { \
141 if (flags & UNICODE_WARN_NONCHAR) { \
142 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
c94c2f39 143 nonchar_cp_format, uv); \
8ee1cdcb
KW
144 } \
145 if (flags & UNICODE_DISALLOW_NONCHAR) { \
146 return NULL; \
147 } \
148 } STMT_END;
149
ba6ed43c
KW
150/* Use shorter names internally in this file */
151#define SHIFT UTF_ACCUMULATION_SHIFT
152#undef MARK
153#define MARK UTF_CONTINUATION_MARK
154#define MASK UTF_CONTINUATION_MASK
155
dfe13c55 156U8 *
4b31b634 157Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
a0ed51b3 158{
378516de 159 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 160
2d1545e5 161 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 162 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
163 return d;
164 }
facc1dc2 165
3ea68d71 166 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
167 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
168 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
169 return d;
170 }
d9432125 171
ba6ed43c
KW
172 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
173 * below, the 16 is for start bytes E0-EF (which are all the possible ones
174 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
175 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
176 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
177 * 0x800-0xFFFF on ASCII */
178 if (uv < (16 * (1U << (2 * SHIFT)))) {
179 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
180 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
181 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
182
183#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
184 aren't tested here */
185 /* The most likely code points in this range are below the surrogates.
186 * Do an extra test to quickly exclude those. */
187 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
188 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
189 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
190 {
8ee1cdcb
KW
191 HANDLE_UNICODE_NONCHAR(uv, flags);
192 }
193 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
194 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 195 }
ba6ed43c
KW
196 }
197#endif
198 return d;
199 }
200
201 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
202 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
203 * happen starting with 4-byte characters on ASCII platforms. We unify the
204 * code for these with EBCDIC, even though some of them require 5-bytes on
205 * those, because khw believes the code saving is worth the very slight
206 * performance hit on these high EBCDIC code points. */
207
208 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
76513bdc
KW
209 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
210 && ckWARN_d(WARN_DEPRECATED))
211 {
212 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
213 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
a5bf80e0 214 }
56576a04 215 if ( (flags & UNICODE_WARN_SUPER)
a5bf80e0
KW
216 || ( UNICODE_IS_ABOVE_31_BIT(uv)
217 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
218 {
219 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
220
221 /* Choose the more dire applicable warning */
222 (UNICODE_IS_ABOVE_31_BIT(uv))
c94c2f39
KW
223 ? above_31_bit_cp_format
224 : super_cp_format,
a5bf80e0
KW
225 uv);
226 }
56576a04 227 if ( (flags & UNICODE_DISALLOW_SUPER)
a5bf80e0
KW
228 || ( UNICODE_IS_ABOVE_31_BIT(uv)
229 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
230 {
231 return NULL;
232 }
233 }
ba6ed43c
KW
234 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
235 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 236 }
d9432125 237
ba6ed43c
KW
238 /* Test for and handle 4-byte result. In the test immediately below, the
239 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
240 * characters). The 3 is for 3 continuation bytes; these each contribute
241 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
242 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
243 * 0x1_0000-0x1F_FFFF on ASCII */
244 if (uv < (8 * (1U << (3 * SHIFT)))) {
245 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
246 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
247 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
248 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
249
250#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
251 characters. The end-plane non-characters for EBCDIC were
252 handled just above */
253 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
254 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 255 }
ba6ed43c
KW
256 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
257 HANDLE_UNICODE_SURROGATE(uv, flags);
258 }
259#endif
260
261 return d;
262 }
263
264 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
265 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
266 * format. The unrolled version above turns out to not save all that much
267 * time, and at these high code points (well above the legal Unicode range
268 * on ASCII platforms, and well above anything in common use in EBCDIC),
269 * khw believes that less code outweighs slight performance gains. */
270
d9432125 271 {
5aaebcb3 272 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
273 U8 *p = d+len-1;
274 while (p > d) {
4c8cd605 275 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
276 uv >>= UTF_ACCUMULATION_SHIFT;
277 }
4c8cd605 278 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
279 return d+len;
280 }
a0ed51b3 281}
a5bf80e0 282
646ca15d 283/*
07693fe6
KW
284=for apidoc uvchr_to_utf8
285
bcb1a2d4 286Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 287of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
288C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
289the byte after the end of the new character. In other words,
07693fe6
KW
290
291 d = uvchr_to_utf8(d, uv);
292
293is the recommended wide native character-aware way of saying
294
295 *(d++) = uv;
296
760c7c2f
KW
297This function accepts any UV as input, but very high code points (above
298C<IV_MAX> on the platform) will raise a deprecation warning. This is
299typically 0x7FFF_FFFF in a 32-bit word.
300
301It is possible to forbid or warn on non-Unicode code points, or those that may
302be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 303
07693fe6
KW
304=cut
305*/
306
de69f3af
KW
307/* This is also a macro */
308PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
309
07693fe6
KW
310U8 *
311Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
312{
de69f3af 313 return uvchr_to_utf8(d, uv);
07693fe6
KW
314}
315
de69f3af
KW
316/*
317=for apidoc uvchr_to_utf8_flags
318
319Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 320of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
321C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
322the byte after the end of the new character. In other words,
de69f3af
KW
323
324 d = uvchr_to_utf8_flags(d, uv, flags);
325
326or, in most cases,
327
328 d = uvchr_to_utf8_flags(d, uv, 0);
329
330This is the Unicode-aware way of saying
331
332 *(d++) = uv;
333
760c7c2f
KW
334If C<flags> is 0, this function accepts any UV as input, but very high code
335points (above C<IV_MAX> for the platform) will raise a deprecation warning.
336This is typically 0x7FFF_FFFF in a 32-bit word.
337
338Specifying C<flags> can further restrict what is allowed and not warned on, as
339follows:
de69f3af 340
796b6530 341If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
342the function will raise a warning, provided UTF8 warnings are enabled. If
343instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
344NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 345
760c7c2f
KW
346Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
347affect how the function handles a Unicode non-character.
93e6dbd6 348
760c7c2f
KW
349And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
350affect the handling of code points that are above the Unicode maximum of
3510x10FFFF. Languages other than Perl may not be able to accept files that
352contain these.
93e6dbd6
KW
353
354The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
355the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
356three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
357allowed inputs to the strict UTF-8 traditionally defined by Unicode.
358Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
359C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
360above-Unicode and surrogate flags, but not the non-character ones, as
361defined in
362L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
363See L<perlunicode/Noncharacter code points>.
93e6dbd6 364
ab8e6d41
KW
365Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
366so using them is more problematic than other above-Unicode code points. Perl
367invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
368likely that non-Perl languages will not be able to read files that contain
369these that written by the perl interpreter; nor would Perl understand files
370written by something that uses a different extension. For these reasons, there
371is a separate set of flags that can warn and/or disallow these extremely high
372code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
373C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
374are entirely independent from the deprecation warning for code points above
375C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
376code point that needs more than 31 bits to represent. When that happens,
377effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
37832-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
379above-Unicode code points, including these, as malformations; and
380C<UNICODE_WARN_SUPER> warns on these.)
381
382On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
383extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
384than on ASCII. Prior to that, code points 2**31 and higher were simply
385unrepresentable, and a different, incompatible method was used to represent
386code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
387and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
388platforms, warning and disallowing 2**31 and higher.
de69f3af 389
de69f3af
KW
390=cut
391*/
392
393/* This is also a macro */
394PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
395
07693fe6
KW
396U8 *
397Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
398{
de69f3af 399 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
400}
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
c179379d
KW
483/* Anything larger than this will overflow the word if it were converted into a UV */
484#if defined(UV_IS_QUAD)
485# ifdef EBCDIC /* Actually is I8 */
486# define HIGHEST_REPRESENTABLE_UTF8 \
487 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
488# else
489# define HIGHEST_REPRESENTABLE_UTF8 \
490 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
491# endif
492#else /* 32-bit */
493# ifdef EBCDIC
494# define HIGHEST_REPRESENTABLE_UTF8 \
495 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
496# else
497# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
498# endif
499#endif
500
12a4bed3
KW
501PERL_STATIC_INLINE bool
502S_does_utf8_overflow(const U8 * const s, const U8 * e)
503{
504 const U8 *x;
505 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
506
c551bb01
KW
507#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
508
509 const STRLEN len = e - s;
510
511#endif
512
12a4bed3
KW
513 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
514 * platform, that is if it represents a code point larger than the highest
515 * representable code point. (For ASCII platforms, we could use memcmp()
516 * because we don't have to convert each byte to I8, but it's very rare
517 * input indeed that would approach overflow, so the loop below will likely
518 * only get executed once.
519 *
520 * 'e' must not be beyond a full character. If it is less than a full
521 * character, the function returns FALSE if there is any input beyond 'e'
522 * that could result in a non-overflowing code point */
523
524 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
525 assert(s <= e && s + UTF8SKIP(s) >= e);
526
527#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
528
529 /* On 32 bit ASCII machines, many overlongs that start with FF don't
530 * overflow */
531
c551bb01 532 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
533 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
534 return memGE(s, max_32_bit_overlong,
c551bb01 535 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
536 }
537
538#endif
12a4bed3
KW
539
540 for (x = s; x < e; x++, y++) {
541
56576a04
KW
542 /* If this byte is larger than the corresponding highest UTF-8
543 * byte, it overflows */
544 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
545 return TRUE;
546 }
12a4bed3 547
56576a04
KW
548 /* If not the same as this byte, it must be smaller, doesn't
549 * overflow */
550 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
551 return FALSE;
552 }
12a4bed3 553 }
12a4bed3
KW
554
555 /* Got to the end and all bytes are the same. If the input is a whole
556 * character, it doesn't overflow. And if it is a partial character,
557 * there's not enough information to tell, so assume doesn't overflow */
558 return FALSE;
559}
560
561PERL_STATIC_INLINE bool
562S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
563{
564 /* Overlongs can occur whenever the number of continuation bytes
565 * changes. That means whenever the number of leading 1 bits in a start
566 * byte increases from the next lower start byte. That happens for start
567 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
568 * illegal start bytes have already been excluded, so don't need to be
569 * tested here;
570 * ASCII platforms: C0, C1
571 * EBCDIC platforms C0, C1, C2, C3, C4, E0
572 *
573 * At least a second byte is required to determine if other sequences will
574 * be an overlong. */
575
576 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
577 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
578
579 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
580 assert(len > 1 && UTF8_IS_START(*s));
581
582 /* Each platform has overlongs after the start bytes given above (expressed
583 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
584 * the logic is the same, except the E0 overlong has already been excluded
585 * on EBCDIC platforms. The values below were found by manually
586 * inspecting the UTF-8 patterns. See the tables in utf8.h and
587 * utfebcdic.h. */
588
589# ifdef EBCDIC
590# define F0_ABOVE_OVERLONG 0xB0
591# define F8_ABOVE_OVERLONG 0xA8
592# define FC_ABOVE_OVERLONG 0xA4
593# define FE_ABOVE_OVERLONG 0xA2
594# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
595 /* I8(0xfe) is FF */
596# else
597
598 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
599 return TRUE;
600 }
601
602# define F0_ABOVE_OVERLONG 0x90
603# define F8_ABOVE_OVERLONG 0x88
604# define FC_ABOVE_OVERLONG 0x84
605# define FE_ABOVE_OVERLONG 0x82
606# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
607# endif
608
609
610 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
611 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
612 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
613 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
614 {
615 return TRUE;
616 }
617
b0b342d4
KW
618 /* Check for the FF overlong */
619 return isFF_OVERLONG(s, len);
620}
621
622PERL_STATIC_INLINE bool
623S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
624{
625 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
626
627 /* Check for the FF overlong. This happens only if all these bytes match;
628 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 629 * utfebcdic.h. */
12a4bed3 630
b0b342d4
KW
631 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
632 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
633 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
634}
635
636#undef F0_ABOVE_OVERLONG
637#undef F8_ABOVE_OVERLONG
638#undef FC_ABOVE_OVERLONG
639#undef FE_ABOVE_OVERLONG
640#undef FF_OVERLONG_PREFIX
641
35f8c9bd 642STRLEN
edc2c47a 643Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 644{
2b479609 645 STRLEN len;
12a4bed3 646 const U8 *x;
35f8c9bd 647
2b479609
KW
648 /* A helper function that should not be called directly.
649 *
650 * This function returns non-zero if the string beginning at 's' and
651 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
652 * code point; otherwise it returns 0. The examination stops after the
653 * first code point in 's' is validated, not looking at the rest of the
654 * input. If 'e' is such that there are not enough bytes to represent a
655 * complete code point, this function will return non-zero anyway, if the
656 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
657 * excluded by 'flags'.
658 *
659 * A non-zero return gives the number of bytes required to represent the
660 * code point. Be aware that if the input is for a partial character, the
661 * return will be larger than 'e - s'.
662 *
663 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
664 * The caller should have excluded the possibility of it being invariant
665 * before calling this function.
2b479609
KW
666 *
667 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
668 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
669 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
670 * disallowed by the flags. If the input is only for a partial character,
671 * the function will return non-zero if there is any sequence of
672 * well-formed UTF-8 that, when appended to the input sequence, could
673 * result in an allowed code point; otherwise it returns 0. Non characters
674 * cannot be determined based on partial character input. But many of the
675 * other excluded types can be determined with just the first one or two
676 * bytes.
677 *
678 */
679
680 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
681
682 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
683 |UTF8_DISALLOW_ABOVE_31_BIT)));
684 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 685
2b479609 686 /* A variant char must begin with a start byte */
35f8c9bd
KW
687 if (UNLIKELY(! UTF8_IS_START(*s))) {
688 return 0;
689 }
690
edc2c47a
KW
691 /* Examine a maximum of a single whole code point */
692 if (e - s > UTF8SKIP(s)) {
693 e = s + UTF8SKIP(s);
694 }
695
2b479609
KW
696 len = e - s;
697
698 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
699 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 700
56576a04
KW
701 /* Here, we are disallowing some set of largish code points, and the
702 * first byte indicates the sequence is for a code point that could be
703 * in the excluded set. We generally don't have to look beyond this or
704 * the second byte to see if the sequence is actually for one of the
705 * excluded classes. The code below is derived from this table:
706 *
2b479609
KW
707 * UTF-8 UTF-EBCDIC I8
708 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
709 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
710 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
711 *
56576a04
KW
712 * Keep in mind that legal continuation bytes range between \x80..\xBF
713 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
714 * continuation bytes. Hence, we don't have to test the upper edge
715 * because if any of those is encountered, the sequence is malformed,
716 * and would fail elsewhere in this function.
717 *
718 * The code here likewise assumes that there aren't other
719 * malformations; again the function should fail elsewhere because of
720 * these. For example, an overlong beginning with FC doesn't actually
721 * have to be a super; it could actually represent a small code point,
722 * even U+0000. But, since overlongs (and other malformations) are
723 * illegal, the function should return FALSE in either case.
2b479609
KW
724 */
725
726#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
727# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 728# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 729
19794540
KW
730# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
731 /* B6 and B7 */ \
732 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
733#else
734# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
735# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
736# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
737#endif
738
739 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
740 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
741 {
2b479609
KW
742 return 0; /* Above Unicode */
743 }
744
745 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
746 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
747 {
748 return 0; /* Above 31 bits */
749 }
750
751 if (len > 1) {
752 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
753
754 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 755 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
756 {
757 return 0; /* Above Unicode */
758 }
759
760 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 761 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
762 {
763 return 0; /* Surrogate */
764 }
765
766 if ( (flags & UTF8_DISALLOW_NONCHAR)
767 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
768 {
769 return 0; /* Noncharacter code point */
770 }
771 }
772 }
773
774 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
775 for (x = s + 1; x < e; x++) {
776 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
777 return 0;
778 }
779 }
780
af13dd8a 781 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
782 * overlong. */
783 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
784 return 0;
af13dd8a
KW
785 }
786
12a4bed3
KW
787 /* And finally, that the code point represented fits in a word on this
788 * platform */
789 if (does_utf8_overflow(s, e)) {
790 return 0;
35f8c9bd
KW
791 }
792
2b479609 793 return UTF8SKIP(s);
35f8c9bd
KW
794}
795
7e2f38b2
KW
796char *
797Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
798{
799 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
800 * bytes starting at 's'. 'format' gives how to display each byte.
801 * Currently, there are only two formats, so it is currently a bool:
802 * 0 \xab
803 * 1 ab (that is a space between two hex digit bytes)
804 */
7cf8d05d
KW
805
806 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
807 trailing NUL */
808 const U8 * const e = s + len;
809 char * output;
810 char * d;
811
812 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
813
814 Newx(output, output_len, char);
815 SAVEFREEPV(output);
816
817 d = output;
818 for (; s < e; s++) {
819 const unsigned high_nibble = (*s & 0xF0) >> 4;
820 const unsigned low_nibble = (*s & 0x0F);
821
7e2f38b2
KW
822 if (format) {
823 *d++ = ' ';
824 }
825 else {
826 *d++ = '\\';
827 *d++ = 'x';
828 }
7cf8d05d
KW
829
830 if (high_nibble < 10) {
831 *d++ = high_nibble + '0';
832 }
833 else {
834 *d++ = high_nibble - 10 + 'a';
835 }
836
837 if (low_nibble < 10) {
838 *d++ = low_nibble + '0';
839 }
840 else {
841 *d++ = low_nibble - 10 + 'a';
842 }
843 }
844
845 *d = '\0';
846 return output;
847}
848
806547a7 849PERL_STATIC_INLINE char *
7cf8d05d
KW
850S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
851
852 /* How many bytes to print */
3cc6a05e 853 STRLEN print_len,
7cf8d05d
KW
854
855 /* Which one is the non-continuation */
856 const STRLEN non_cont_byte_pos,
857
858 /* How many bytes should there be? */
859 const STRLEN expect_len)
806547a7
KW
860{
861 /* Return the malformation warning text for an unexpected continuation
862 * byte. */
863
7cf8d05d 864 const char * const where = (non_cont_byte_pos == 1)
806547a7 865 ? "immediately"
7cf8d05d
KW
866 : Perl_form(aTHX_ "%d bytes",
867 (int) non_cont_byte_pos);
806547a7
KW
868
869 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
870
7cf8d05d
KW
871 /* We don't need to pass this parameter, but since it has already been
872 * calculated, it's likely faster to pass it; verify under DEBUGGING */
873 assert(expect_len == UTF8SKIP(s));
874
875 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
876 " %s after start byte 0x%02x; need %d bytes, got %d)",
877 malformed_text,
7e2f38b2 878 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
879 *(s + non_cont_byte_pos),
880 where,
881 *s,
882 (int) expect_len,
883 (int) non_cont_byte_pos);
806547a7
KW
884}
885
35f8c9bd
KW
886/*
887
de69f3af 888=for apidoc utf8n_to_uvchr
378516de
KW
889
890THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 891Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 892
9041c2e3 893Bottom level UTF-8 decode routine.
de69f3af 894Returns the native code point value of the first character in the string C<s>,
746afd53
KW
895which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
896C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
897the length, in bytes, of that character.
949cf498
KW
898
899The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
900well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
901causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
902is the next possible position in C<s> that could begin a non-malformed
903character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
904is raised. Some UTF-8 input sequences may contain multiple malformations.
905This function tries to find every possible one in each call, so multiple
56576a04 906warnings can be raised for the same sequence.
949cf498
KW
907
908Various ALLOW flags can be set in C<flags> to allow (and not warn on)
909individual types of malformations, such as the sequence being overlong (that
910is, when there is a shorter sequence that can express the same code point;
911overlong sequences are expressly forbidden in the UTF-8 standard due to
912potential security issues). Another malformation example is the first byte of
913a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
914flags. Even if allowed, this function generally returns the Unicode
915REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
916F<utf8.h> to override this behavior for the overlong malformations, but don't
917do that except for very specialized purposes.
949cf498 918
796b6530 919The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
920flags) malformation is found. If this flag is set, the routine assumes that
921the caller will raise a warning, and this function will silently just set
d088425d
KW
922C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
923
75200dff 924Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 925character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
926in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
927be set to 1. To disambiguate, upon a zero return, see if the first byte of
928C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 929error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
930
931Certain code points are considered problematic. These are Unicode surrogates,
746afd53 932Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 933By default these are considered regular code points, but certain situations
ecc1615f
KW
934warrant special handling for them, which can be specified using the C<flags>
935parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
936three classes are treated as malformations and handled as such. The flags
937C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
938C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
939disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
940restricts the allowed inputs to the strict UTF-8 traditionally defined by
941Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
942definition given by
943L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
944The difference between traditional strictness and C9 strictness is that the
945latter does not forbid non-character code points. (They are still discouraged,
946however.) For more discussion see L<perlunicode/Noncharacter code points>.
947
948The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
949C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
950C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
951raised for their respective categories, but otherwise the code points are
952considered valid (not malformations). To get a category to both be treated as
953a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 954(But note that warnings are not raised if lexically disabled nor if
796b6530 955C<UTF8_CHECK_ONLY> is also specified.)
949cf498 956
760c7c2f
KW
957It is now deprecated to have very high code points (above C<IV_MAX> on the
958platforms) and this function will raise a deprecation warning for these (unless
d5944cab 959such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 960in a 32-bit word.
ab8e6d41
KW
961
962Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
963so using them is more problematic than other above-Unicode code points. Perl
964invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
965likely that non-Perl languages will not be able to read files that contain
aff2be59 966these; nor would Perl understand files
ab8e6d41
KW
967written by something that uses a different extension. For these reasons, there
968is a separate set of flags that can warn and/or disallow these extremely high
969code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
970C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
971are entirely independent from the deprecation warning for code points above
972C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
973code point that needs more than 31 bits to represent. When that happens,
974effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
97532-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
976above-Unicode code points, including these, as malformations; and
977C<UTF8_WARN_SUPER> warns on these.)
978
979On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
980extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
981than on ASCII. Prior to that, code points 2**31 and higher were simply
982unrepresentable, and a different, incompatible method was used to represent
983code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
984and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
985platforms, warning and disallowing 2**31 and higher.
949cf498
KW
986
987All other code points corresponding to Unicode characters, including private
988use and those yet to be assigned, are never considered malformed and never
989warn.
67e989fb 990
37607a96 991=cut
f9380377
KW
992
993Also implemented as a macro in utf8.h
994*/
995
996UV
997Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
998 STRLEN curlen,
999 STRLEN *retlen,
1000 const U32 flags)
1001{
1002 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1003
1004 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1005}
1006
1007/*
1008
1009=for apidoc utf8n_to_uvchr_error
1010
1011THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1012Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1013
1014This function is for code that needs to know what the precise malformation(s)
1015are when an error is found.
1016
1017It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1018all the others, C<errors>. If this parameter is 0, this function behaves
1019identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1020to a C<U32> variable, which this function sets to indicate any errors found.
1021Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1022C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1023of these bits will be set if a malformation is found, even if the input
7a65503b 1024C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1025exceptions are noted:
1026
1027=over 4
1028
1029=item C<UTF8_GOT_ABOVE_31_BIT>
1030
1031The code point represented by the input UTF-8 sequence occupies more than 31
1032bits.
1033This bit is set only if the input C<flags> parameter contains either the
1034C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
1035
1036=item C<UTF8_GOT_CONTINUATION>
1037
1038The input sequence was malformed in that the first byte was a a UTF-8
1039continuation byte.
1040
1041=item C<UTF8_GOT_EMPTY>
1042
1043The input C<curlen> parameter was 0.
1044
1045=item C<UTF8_GOT_LONG>
1046
1047The input sequence was malformed in that there is some other sequence that
1048evaluates to the same code point, but that sequence is shorter than this one.
1049
fecaf136
KW
1050Until Unicode 3.1, it was legal for programs to accept this malformation, but
1051it was discovered that this created security issues.
1052
f9380377
KW
1053=item C<UTF8_GOT_NONCHAR>
1054
1055The code point represented by the input UTF-8 sequence is for a Unicode
1056non-character code point.
1057This bit is set only if the input C<flags> parameter contains either the
1058C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1059
1060=item C<UTF8_GOT_NON_CONTINUATION>
1061
1062The input sequence was malformed in that a non-continuation type byte was found
1063in a position where only a continuation type one should be.
1064
1065=item C<UTF8_GOT_OVERFLOW>
1066
1067The input sequence was malformed in that it is for a code point that is not
1068representable in the number of bits available in a UV on the current platform.
1069
1070=item C<UTF8_GOT_SHORT>
1071
1072The input sequence was malformed in that C<curlen> is smaller than required for
1073a complete sequence. In other words, the input is for a partial character
1074sequence.
1075
1076=item C<UTF8_GOT_SUPER>
1077
1078The input sequence was malformed in that it is for a non-Unicode code point;
1079that is, one above the legal Unicode maximum.
1080This bit is set only if the input C<flags> parameter contains either the
1081C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1082
1083=item C<UTF8_GOT_SURROGATE>
1084
1085The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1086code point.
1087This bit is set only if the input C<flags> parameter contains either the
1088C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1089
1090=back
1091
133551d8
KW
1092To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1093flag to suppress any warnings, and then examine the C<*errors> return.
1094
f9380377 1095=cut
37607a96 1096*/
67e989fb 1097
a0ed51b3 1098UV
f9380377
KW
1099Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1100 STRLEN curlen,
1101 STRLEN *retlen,
1102 const U32 flags,
1103 U32 * errors)
a0ed51b3 1104{
d4c19fe8 1105 const U8 * const s0 = s;
2b5e7bc2
KW
1106 U8 * send = NULL; /* (initialized to silence compilers' wrong
1107 warning) */
1108 U32 possible_problems = 0; /* A bit is set here for each potential problem
1109 found as we go along */
eb83ed87 1110 UV uv = *s;
2b5e7bc2
KW
1111 STRLEN expectlen = 0; /* How long should this sequence be?
1112 (initialized to silence compilers' wrong
1113 warning) */
e308b348 1114 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1115 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1116 this gets set and discarded */
a0dbb045 1117
2b5e7bc2
KW
1118 /* The below are used only if there is both an overlong malformation and a
1119 * too short one. Otherwise the first two are set to 's0' and 'send', and
1120 * the third not used at all */
1121 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1122 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1123 warning) */
e9f2c446
KW
1124 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1125 routine; see [perl #130921] */
2b5e7bc2 1126 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1127
f9380377
KW
1128 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1129
1130 if (errors) {
1131 *errors = 0;
1132 }
1133 else {
1134 errors = &discard_errors;
1135 }
a0dbb045 1136
eb83ed87
KW
1137 /* The order of malformation tests here is important. We should consume as
1138 * few bytes as possible in order to not skip any valid character. This is
1139 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1140 * http://unicode.org/reports/tr36 for more discussion as to why. For
1141 * example, once we've done a UTF8SKIP, we can tell the expected number of
1142 * bytes, and could fail right off the bat if the input parameters indicate
1143 * that there are too few available. But it could be that just that first
1144 * byte is garbled, and the intended character occupies fewer bytes. If we
1145 * blindly assumed that the first byte is correct, and skipped based on
1146 * that number, we could skip over a valid input character. So instead, we
1147 * always examine the sequence byte-by-byte.
1148 *
1149 * We also should not consume too few bytes, otherwise someone could inject
1150 * things. For example, an input could be deliberately designed to
1151 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1152 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1153 * which is actually part of of the overflowing sequence), that could look
1154 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1155 * sequence and process the rest, inappropriately.
1156 *
1157 * Some possible input sequences are malformed in more than one way. This
1158 * function goes to lengths to try to find all of them. This is necessary
1159 * for correctness, as the inputs may allow one malformation but not
1160 * another, and if we abandon searching for others after finding the
1161 * allowed one, we could allow in something that shouldn't have been.
1162 */
eb83ed87 1163
b5b9af04 1164 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1165 possible_problems |= UTF8_GOT_EMPTY;
1166 curlen = 0;
5a48568d 1167 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1168 goto ready_to_handle_errors;
0c443dc2
JH
1169 }
1170
eb83ed87
KW
1171 expectlen = UTF8SKIP(s);
1172
1173 /* A well-formed UTF-8 character, as the vast majority of calls to this
1174 * function will be for, has this expected length. For efficiency, set
1175 * things up here to return it. It will be overriden only in those rare
1176 * cases where a malformation is found */
1177 if (retlen) {
1178 *retlen = expectlen;
1179 }
1180
1181 /* An invariant is trivially well-formed */
1d72bdf6 1182 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1183 return uv;
a0ed51b3 1184 }
67e989fb 1185
eb83ed87 1186 /* A continuation character can't start a valid sequence */
b5b9af04 1187 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1188 possible_problems |= UTF8_GOT_CONTINUATION;
1189 curlen = 1;
1190 uv = UNICODE_REPLACEMENT;
1191 goto ready_to_handle_errors;
ba210ebe 1192 }
9041c2e3 1193
dcd27b3c 1194 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1195 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1196 * because it excludes start bytes like \xC0 that always lead to
1197 * overlongs.) */
dcd27b3c 1198
534752c1
KW
1199 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1200 * that indicate the number of bytes in the character's whole UTF-8
1201 * sequence, leaving just the bits that are part of the value. */
1202 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1203
e308b348
KW
1204 /* Setup the loop end point, making sure to not look past the end of the
1205 * input string, and flag it as too short if the size isn't big enough. */
1206 send = (U8*) s0;
1207 if (UNLIKELY(curlen < expectlen)) {
1208 possible_problems |= UTF8_GOT_SHORT;
1209 avail_len = curlen;
1210 send += curlen;
1211 }
1212 else {
1213 send += expectlen;
1214 }
1215 adjusted_send = send;
1216
eb83ed87 1217 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1218 * accumulating each into the working value as we go. */
eb83ed87 1219 for (s = s0 + 1; s < send; s++) {
b5b9af04 1220 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1221 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1222 continue;
1223 }
1224
1225 /* Here, found a non-continuation before processing all expected bytes.
1226 * This byte indicates the beginning of a new character, so quit, even
1227 * if allowing this malformation. */
2b5e7bc2 1228 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1229 break;
eb83ed87
KW
1230 } /* End of loop through the character's bytes */
1231
1232 /* Save how many bytes were actually in the character */
1233 curlen = s - s0;
1234
2b5e7bc2
KW
1235 /* Note that there are two types of too-short malformation. One is when
1236 * there is actual wrong data before the normal termination of the
1237 * sequence. The other is that the sequence wasn't complete before the end
1238 * of the data we are allowed to look at, based on the input 'curlen'.
1239 * This means that we were passed data for a partial character, but it is
1240 * valid as far as we saw. The other is definitely invalid. This
1241 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1242 * separate.
1243 *
1244 * A convenience macro that matches either of the too-short conditions. */
1245# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1246
1247 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1248 uv_so_far = uv;
1249 uv = UNICODE_REPLACEMENT;
1250 }
2b5e7bc2 1251
56576a04 1252 /* Check for overflow. */
2b5e7bc2
KW
1253 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1254 possible_problems |= UTF8_GOT_OVERFLOW;
1255 uv = UNICODE_REPLACEMENT;
eb83ed87 1256 }
eb83ed87 1257
2b5e7bc2
KW
1258 /* Check for overlong. If no problems so far, 'uv' is the correct code
1259 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1260 * we must look at the UTF-8 byte sequence itself to see if it is for an
1261 * overlong */
1262 if ( ( LIKELY(! possible_problems)
1263 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1264 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1265 && ( UNLIKELY(! UTF8_IS_START(*s0))
1266 || ( curlen > 1
1267 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1268 send - s0))))))
2f8f112e 1269 {
2b5e7bc2
KW
1270 possible_problems |= UTF8_GOT_LONG;
1271
abc28b54 1272 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1273
abc28b54
KW
1274 /* The calculation in the 'true' branch of this 'if'
1275 * below won't work if overflows, and isn't needed
1276 * anyway. Further below we handle all overflow
1277 * cases */
1278 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1279 {
2b5e7bc2
KW
1280 UV min_uv = uv_so_far;
1281 STRLEN i;
1282
1283 /* Here, the input is both overlong and is missing some trailing
1284 * bytes. There is no single code point it could be for, but there
1285 * may be enough information present to determine if what we have
1286 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1287 * The code further below has the intelligence to determine this,
1288 * but just for non-overlong UTF-8 sequences. What we do here is
1289 * calculate the smallest code point the input could represent if
1290 * there were no too short malformation. Then we compute and save
1291 * the UTF-8 for that, which is what the code below looks at
1292 * instead of the raw input. It turns out that the smallest such
1293 * code point is all we need. */
2b5e7bc2
KW
1294 for (i = curlen; i < expectlen; i++) {
1295 min_uv = UTF8_ACCUMULATE(min_uv,
1296 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1297 }
1298
e9f2c446 1299 adjusted_s0 = temp_char_buf;
2b5e7bc2
KW
1300 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1301 }
eb83ed87
KW
1302 }
1303
56576a04
KW
1304 /* Here, we have found all the possible problems, except for when the input
1305 * is for a problematic code point not allowed by the input parameters. */
1306
06188866
KW
1307 /* uv is valid for overlongs */
1308 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1309
1310 /* isn't problematic if < this */
1311 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1312 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1313
1314 /* if overflow, we know without looking further
1315 * precisely which of the problematic types it is,
1316 * and we deal with those in the overflow handling
1317 * code */
1318 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1319 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1320 && ((flags & ( UTF8_DISALLOW_NONCHAR
1321 |UTF8_DISALLOW_SURROGATE
1322 |UTF8_DISALLOW_SUPER
1323 |UTF8_DISALLOW_ABOVE_31_BIT
1324 |UTF8_WARN_NONCHAR
1325 |UTF8_WARN_SURROGATE
1326 |UTF8_WARN_SUPER
1327 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1328 /* In case of a malformation, 'uv' is not valid, and has
1329 * been changed to something in the Unicode range.
1330 * Currently we don't output a deprecation message if there
1331 * is already a malformation, so we don't have to special
1332 * case the test immediately below */
760c7c2f
KW
1333 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1334 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1335 {
2b5e7bc2
KW
1336 /* If there were no malformations, or the only malformation is an
1337 * overlong, 'uv' is valid */
1338 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1339 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1340 possible_problems |= UTF8_GOT_SURROGATE;
1341 }
1342 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1343 possible_problems |= UTF8_GOT_SUPER;
1344 }
1345 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1346 possible_problems |= UTF8_GOT_NONCHAR;
1347 }
1348 }
1349 else { /* Otherwise, need to look at the source UTF-8, possibly
1350 adjusted to be non-overlong */
1351
1352 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1353 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1354 {
2b5e7bc2
KW
1355 possible_problems |= UTF8_GOT_SUPER;
1356 }
1357 else if (curlen > 1) {
1358 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1359 NATIVE_UTF8_TO_I8(*adjusted_s0),
1360 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1361 {
2b5e7bc2 1362 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1363 }
2b5e7bc2
KW
1364 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1365 NATIVE_UTF8_TO_I8(*adjusted_s0),
1366 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1367 {
1368 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1369 }
1370 }
c0236afe 1371
2b5e7bc2
KW
1372 /* We need a complete well-formed UTF-8 character to discern
1373 * non-characters, so can't look for them here */
1374 }
1375 }
949cf498 1376
2b5e7bc2
KW
1377 ready_to_handle_errors:
1378
1379 /* At this point:
1380 * curlen contains the number of bytes in the sequence that
1381 * this call should advance the input by.
e308b348
KW
1382 * avail_len gives the available number of bytes passed in, but
1383 * only if this is less than the expected number of
1384 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1385 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1386 * is set in it for each potential problem found.
1387 * uv contains the code point the input sequence
1388 * represents; or if there is a problem that prevents
1389 * a well-defined value from being computed, it is
1390 * some subsitute value, typically the REPLACEMENT
1391 * CHARACTER.
1392 * s0 points to the first byte of the character
56576a04
KW
1393 * s points to just after were we left off processing
1394 * the character
1395 * send points to just after where that character should
1396 * end, based on how many bytes the start byte tells
1397 * us should be in it, but no further than s0 +
1398 * avail_len
2b5e7bc2
KW
1399 * adjusted_s0 normally is the same as s0, but in case of an
1400 * overlong for which the UTF-8 matters below, it is
1401 * the first byte of the shortest form representation
1402 * of the input.
1403 * adjusted_send normally is the same as 'send', but if adjusted_s0
1404 * is set to something other than s0, this points one
1405 * beyond its end
1406 */
eb83ed87 1407
2b5e7bc2
KW
1408 if (UNLIKELY(possible_problems)) {
1409 bool disallowed = FALSE;
1410 const U32 orig_problems = possible_problems;
1411
1412 while (possible_problems) { /* Handle each possible problem */
1413 UV pack_warn = 0;
1414 char * message = NULL;
1415
1416 /* Each 'if' clause handles one problem. They are ordered so that
1417 * the first ones' messages will be displayed before the later
1418 * ones; this is kinda in decreasing severity order */
1419 if (possible_problems & UTF8_GOT_OVERFLOW) {
1420
56576a04
KW
1421 /* Overflow means also got a super and are using Perl's
1422 * extended UTF-8, but we handle all three cases here */
2b5e7bc2
KW
1423 possible_problems
1424 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1425 *errors |= UTF8_GOT_OVERFLOW;
1426
1427 /* But the API says we flag all errors found */
1428 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1429 *errors |= UTF8_GOT_SUPER;
1430 }
ddb65933
KW
1431 if (flags
1432 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1433 {
f9380377
KW
1434 *errors |= UTF8_GOT_ABOVE_31_BIT;
1435 }
2b5e7bc2 1436
d60baaa7 1437 /* Disallow if any of the three categories say to */
56576a04 1438 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7
KW
1439 || (flags & ( UTF8_DISALLOW_SUPER
1440 |UTF8_DISALLOW_ABOVE_31_BIT)))
1441 {
1442 disallowed = TRUE;
1443 }
1444
d60baaa7
KW
1445 /* Likewise, warn if any say to, plus if deprecation warnings
1446 * are on, because this code point is above IV_MAX */
56576a04 1447 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7
KW
1448 || ! (flags & UTF8_ALLOW_OVERFLOW)
1449 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1450 {
2b5e7bc2 1451
ddb65933
KW
1452 /* The warnings code explicitly says it doesn't handle the
1453 * case of packWARN2 and two categories which have
1454 * parent-child relationship. Even if it works now to
1455 * raise the warning if either is enabled, it wouldn't
1456 * necessarily do so in the future. We output (only) the
56576a04 1457 * most dire warning */
ddb65933
KW
1458 if (! (flags & UTF8_CHECK_ONLY)) {
1459 if (ckWARN_d(WARN_UTF8)) {
1460 pack_warn = packWARN(WARN_UTF8);
1461 }
1462 else if (ckWARN_d(WARN_NON_UNICODE)) {
1463 pack_warn = packWARN(WARN_NON_UNICODE);
1464 }
1465 if (pack_warn) {
1466 message = Perl_form(aTHX_ "%s: %s (overflows)",
1467 malformed_text,
05b9033b 1468 _byte_dump_string(s0, curlen, 0));
ddb65933 1469 }
2b5e7bc2
KW
1470 }
1471 }
1472 }
1473 else if (possible_problems & UTF8_GOT_EMPTY) {
1474 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1475 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1476
1477 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1478
1479 /* This so-called malformation is now treated as a bug in
1480 * the caller. If you have nothing to decode, skip calling
1481 * this function */
1482 assert(0);
1483
2b5e7bc2
KW
1484 disallowed = TRUE;
1485 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1486 pack_warn = packWARN(WARN_UTF8);
1487 message = Perl_form(aTHX_ "%s (empty string)",
1488 malformed_text);
1489 }
1490 }
1491 }
1492 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1493 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1494 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1495
1496 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1497 disallowed = TRUE;
1498 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1499 pack_warn = packWARN(WARN_UTF8);
1500 message = Perl_form(aTHX_
1501 "%s: %s (unexpected continuation byte 0x%02x,"
1502 " with no preceding start byte)",
1503 malformed_text,
7e2f38b2 1504 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1505 }
1506 }
1507 }
2b5e7bc2
KW
1508 else if (possible_problems & UTF8_GOT_SHORT) {
1509 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1510 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1511
1512 if (! (flags & UTF8_ALLOW_SHORT)) {
1513 disallowed = TRUE;
1514 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1515 pack_warn = packWARN(WARN_UTF8);
1516 message = Perl_form(aTHX_
56576a04
KW
1517 "%s: %s (too short; %d byte%s available, need %d)",
1518 malformed_text,
1519 _byte_dump_string(s0, send - s0, 0),
1520 (int)avail_len,
1521 avail_len == 1 ? "" : "s",
1522 (int)expectlen);
2b5e7bc2
KW
1523 }
1524 }
ba210ebe 1525
2b5e7bc2 1526 }
e308b348
KW
1527 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1528 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1529 *errors |= UTF8_GOT_NON_CONTINUATION;
1530
1531 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1532 disallowed = TRUE;
1533 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1534
1535 /* If we don't know for sure that the input length is
1536 * valid, avoid as much as possible reading past the
1537 * end of the buffer */
1538 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1539 ? s - s0
1540 : send - s0;
e308b348
KW
1541 pack_warn = packWARN(WARN_UTF8);
1542 message = Perl_form(aTHX_ "%s",
1543 unexpected_non_continuation_text(s0,
99a765e9 1544 printlen,
e308b348
KW
1545 s - s0,
1546 (int) expectlen));
1547 }
1548 }
1549 }
2b5e7bc2
KW
1550 else if (possible_problems & UTF8_GOT_LONG) {
1551 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1552 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1553
94953955
KW
1554 if (flags & UTF8_ALLOW_LONG) {
1555
1556 /* We don't allow the actual overlong value, unless the
1557 * special extra bit is also set */
1558 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1559 & ~UTF8_ALLOW_LONG)))
1560 {
1561 uv = UNICODE_REPLACEMENT;
1562 }
1563 }
1564 else {
2b5e7bc2
KW
1565 disallowed = TRUE;
1566
1567 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1568 pack_warn = packWARN(WARN_UTF8);
1569
1570 /* These error types cause 'uv' to be something that
1571 * isn't what was intended, so can't use it in the
1572 * message. The other error types either can't
1573 * generate an overlong, or else the 'uv' is valid */
1574 if (orig_problems &
1575 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1576 {
1577 message = Perl_form(aTHX_
1578 "%s: %s (any UTF-8 sequence that starts"
1579 " with \"%s\" is overlong which can and"
1580 " should be represented with a"
1581 " different, shorter sequence)",
1582 malformed_text,
7e2f38b2
KW
1583 _byte_dump_string(s0, send - s0, 0),
1584 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1585 }
1586 else {
1587 U8 tmpbuf[UTF8_MAXBYTES+1];
1588 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1589 uv, 0);
56d0c377
KW
1590 const char * preface = (uv <= PERL_UNICODE_MAX)
1591 ? "U+"
1592 : "0x";
2b5e7bc2
KW
1593 message = Perl_form(aTHX_
1594 "%s: %s (overlong; instead use %s to represent"
56d0c377 1595 " %s%0*" UVXf ")",
2b5e7bc2 1596 malformed_text,
05b9033b 1597 _byte_dump_string(s0, curlen, 0),
7e2f38b2 1598 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
56d0c377 1599 preface,
2b5e7bc2
KW
1600 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1601 small code points */
1602 uv);
1603 }
1604 }
1605 }
1606 }
1607 else if (possible_problems & UTF8_GOT_SURROGATE) {
1608 possible_problems &= ~UTF8_GOT_SURROGATE;
1609
f9380377
KW
1610 if (flags & UTF8_WARN_SURROGATE) {
1611 *errors |= UTF8_GOT_SURROGATE;
1612
1613 if ( ! (flags & UTF8_CHECK_ONLY)
1614 && ckWARN_d(WARN_SURROGATE))
1615 {
2b5e7bc2
KW
1616 pack_warn = packWARN(WARN_SURROGATE);
1617
1618 /* These are the only errors that can occur with a
1619 * surrogate when the 'uv' isn't valid */
1620 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1621 message = Perl_form(aTHX_
1622 "UTF-16 surrogate (any UTF-8 sequence that"
1623 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1624 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1625 }
1626 else {
c94c2f39 1627 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1628 }
f9380377 1629 }
2b5e7bc2 1630 }
ba210ebe 1631
2b5e7bc2
KW
1632 if (flags & UTF8_DISALLOW_SURROGATE) {
1633 disallowed = TRUE;
f9380377 1634 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1635 }
1636 }
1637 else if (possible_problems & UTF8_GOT_SUPER) {
1638 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1639
f9380377
KW
1640 if (flags & UTF8_WARN_SUPER) {
1641 *errors |= UTF8_GOT_SUPER;
1642
1643 if ( ! (flags & UTF8_CHECK_ONLY)
1644 && ckWARN_d(WARN_NON_UNICODE))
1645 {
2b5e7bc2
KW
1646 pack_warn = packWARN(WARN_NON_UNICODE);
1647
1648 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1649 message = Perl_form(aTHX_
1650 "Any UTF-8 sequence that starts with"
1651 " \"%s\" is for a non-Unicode code point,"
1652 " may not be portable",
7e2f38b2 1653 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1654 }
1655 else {
c94c2f39 1656 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1657 }
f9380377 1658 }
2b5e7bc2 1659 }
ba210ebe 1660
2b5e7bc2
KW
1661 /* The maximum code point ever specified by a standard was
1662 * 2**31 - 1. Anything larger than that is a Perl extension
1663 * that very well may not be understood by other applications
1664 * (including earlier perl versions on EBCDIC platforms). We
1665 * test for these after the regular SUPER ones, and before
1666 * possibly bailing out, so that the slightly more dire warning
1667 * will override the regular one. */
1668 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1669 |UTF8_WARN_SUPER
1670 |UTF8_DISALLOW_ABOVE_31_BIT))
1671 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1672 && UNLIKELY(is_utf8_cp_above_31_bits(
56576a04
KW
1673 adjusted_s0,
1674 adjusted_send)))
2b5e7bc2
KW
1675 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1676 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1677 {
1678 if ( ! (flags & UTF8_CHECK_ONLY)
1679 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
db0f09e6 1680 && ckWARN_d(WARN_NON_UNICODE))
2b5e7bc2 1681 {
db0f09e6 1682 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2
KW
1683
1684 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1685 message = Perl_form(aTHX_
1686 "Any UTF-8 sequence that starts with"
1687 " \"%s\" is for a non-Unicode code"
1688 " point, and is not portable",
7e2f38b2 1689 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1690 }
1691 else {
1692 message = Perl_form(aTHX_
c94c2f39 1693 above_31_bit_cp_format, uv);
2b5e7bc2
KW
1694 }
1695 }
1696
ddb65933
KW
1697 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1698 |UTF8_DISALLOW_ABOVE_31_BIT))
1699 {
f9380377
KW
1700 *errors |= UTF8_GOT_ABOVE_31_BIT;
1701
1702 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1703 disallowed = TRUE;
1704 }
2b5e7bc2
KW
1705 }
1706 }
eb83ed87 1707
2b5e7bc2 1708 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1709 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1710 disallowed = TRUE;
1711 }
eb83ed87 1712
2b5e7bc2
KW
1713 /* The deprecated warning overrides any non-deprecated one. If
1714 * there are other problems, a deprecation message is not
1715 * really helpful, so don't bother to raise it in that case.
1716 * This also keeps the code from having to handle the case
1717 * where 'uv' is not valid. */
1718 if ( ! (orig_problems
1719 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1720 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1721 && ckWARN_d(WARN_DEPRECATED))
1722 {
1723 message = Perl_form(aTHX_ cp_above_legal_max,
1724 uv, MAX_NON_DEPRECATED_CP);
1725 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1726 }
1727 }
1728 else if (possible_problems & UTF8_GOT_NONCHAR) {
1729 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1730
f9380377
KW
1731 if (flags & UTF8_WARN_NONCHAR) {
1732 *errors |= UTF8_GOT_NONCHAR;
1733
1734 if ( ! (flags & UTF8_CHECK_ONLY)
1735 && ckWARN_d(WARN_NONCHAR))
1736 {
2b5e7bc2
KW
1737 /* The code above should have guaranteed that we don't
1738 * get here with errors other than overlong */
1739 assert (! (orig_problems
1740 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1741
1742 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1743 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1744 }
2b5e7bc2 1745 }
5b311467 1746
2b5e7bc2
KW
1747 if (flags & UTF8_DISALLOW_NONCHAR) {
1748 disallowed = TRUE;
f9380377 1749 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1750 }
1751 } /* End of looking through the possible flags */
1752
1753 /* Display the message (if any) for the problem being handled in
1754 * this iteration of the loop */
1755 if (message) {
1756 if (PL_op)
1757 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1758 OP_DESC(PL_op));
1759 else
1760 Perl_warner(aTHX_ pack_warn, "%s", message);
1761 }
ddb65933 1762 } /* End of 'while (possible_problems)' */
a0dbb045 1763
2b5e7bc2
KW
1764 /* Since there was a possible problem, the returned length may need to
1765 * be changed from the one stored at the beginning of this function.
1766 * Instead of trying to figure out if that's needed, just do it. */
1767 if (retlen) {
1768 *retlen = curlen;
1769 }
a0dbb045 1770
2b5e7bc2
KW
1771 if (disallowed) {
1772 if (flags & UTF8_CHECK_ONLY && retlen) {
1773 *retlen = ((STRLEN) -1);
1774 }
1775 return 0;
1776 }
eb83ed87 1777 }
ba210ebe 1778
2b5e7bc2 1779 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1780}
1781
8e84507e 1782/*
ec5f19d0
KW
1783=for apidoc utf8_to_uvchr_buf
1784
1785Returns the native code point of the first character in the string C<s> which
1786is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1787C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1788
524080c4
KW
1789If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1790enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1791C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1792(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1793C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1794the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1795See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1796returned.
ec5f19d0 1797
760c7c2f
KW
1798Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1799unless those are turned off.
1800
ec5f19d0 1801=cut
52be2536
KW
1802
1803Also implemented as a macro in utf8.h
1804
ec5f19d0
KW
1805*/
1806
1807
1808UV
1809Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1810{
7f974d7e
KW
1811 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1812
ec5f19d0
KW
1813 assert(s < send);
1814
1815 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1816 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1817}
1818
52be2536
KW
1819/* This is marked as deprecated
1820 *
ec5f19d0
KW
1821=for apidoc utf8_to_uvuni_buf
1822
de69f3af
KW
1823Only in very rare circumstances should code need to be dealing in Unicode
1824(as opposed to native) code points. In those few cases, use
1825C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1826
1827Returns the Unicode (not-native) code point of the first character in the
1828string C<s> which
ec5f19d0
KW
1829is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1830C<retlen> will be set to the length, in bytes, of that character.
1831
524080c4
KW
1832If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1833enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1834NULL) to -1. If those warnings are off, the computed value if well-defined (or
1835the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1836is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1837next possible position in C<s> that could begin a non-malformed character.
de69f3af 1838See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1839
760c7c2f
KW
1840Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1841unless those are turned off.
1842
ec5f19d0
KW
1843=cut
1844*/
1845
1846UV
1847Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1848{
1849 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1850
1851 assert(send > s);
1852
5962d97e
KW
1853 /* Call the low level routine, asking for checks */
1854 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1855}
1856
b76347f2 1857/*
87cea99e 1858=for apidoc utf8_length
b76347f2
JH
1859
1860Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1861Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1862up past C<e>, croaks.
b76347f2
JH
1863
1864=cut
1865*/
1866
1867STRLEN
35a4481c 1868Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1869{
1870 STRLEN len = 0;
1871
7918f24d
NC
1872 PERL_ARGS_ASSERT_UTF8_LENGTH;
1873
8850bf83
JH
1874 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1875 * the bitops (especially ~) can create illegal UTF-8.
1876 * In other words: in Perl UTF-8 is not just for Unicode. */
1877
a3b680e6
AL
1878 if (e < s)
1879 goto warn_and_return;
b76347f2 1880 while (s < e) {
4cbf4130 1881 s += UTF8SKIP(s);
8e91ec7f
AV
1882 len++;
1883 }
1884
1885 if (e != s) {
1886 len--;
1887 warn_and_return:
9b387841
NC
1888 if (PL_op)
1889 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1890 "%s in %s", unees, OP_DESC(PL_op));
1891 else
61a12c31 1892 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1893 }
1894
1895 return len;
1896}
1897
b06226ff 1898/*
fed3ba5d
NC
1899=for apidoc bytes_cmp_utf8
1900
a1433954 1901Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1902sequence of characters (stored as UTF-8)
1903in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1904equal, -1 or -2 if the first string is less than the second string, +1 or +2
1905if the first string is greater than the second string.
1906
1907-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1908longer string. -2 or +2 is returned if
1909there was a difference between characters
fed3ba5d
NC
1910within the strings.
1911
1912=cut
1913*/
1914
1915int
1916Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1917{
1918 const U8 *const bend = b + blen;
1919 const U8 *const uend = u + ulen;
1920
1921 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1922
1923 while (b < bend && u < uend) {
1924 U8 c = *u++;
1925 if (!UTF8_IS_INVARIANT(c)) {
1926 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1927 if (u < uend) {
1928 U8 c1 = *u++;
1929 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1930 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1931 } else {
2b5e7bc2 1932 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1933 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1934 "%s %s%s",
1935 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1936 PL_op ? " in " : "",
1937 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1938 return -2;
1939 }
1940 } else {
1941 if (PL_op)
1942 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1943 "%s in %s", unees, OP_DESC(PL_op));
1944 else
61a12c31 1945 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1946 return -2; /* Really want to return undef :-) */
1947 }
1948 } else {
1949 return -2;
1950 }
1951 }
1952 if (*b != c) {
1953 return *b < c ? -2 : +2;
1954 }
1955 ++b;
1956 }
1957
1958 if (b == bend && u == uend)
1959 return 0;
1960
1961 return b < bend ? +1 : -1;
1962}
1963
1964/*
87cea99e 1965=for apidoc utf8_to_bytes
6940069f 1966
3bc0c78c 1967Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 1968Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 1969updates C<*lenp> to contain the new length.
3bc0c78c
KW
1970Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
1971
1972Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
1973having saved the value of C<*lenp> before the call, and subtracting the
1974after-call value of C<*lenp> from it.
6940069f 1975
a1433954 1976If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1977
6940069f
GS
1978=cut
1979*/
1980
1981U8 *
09af0336 1982Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 1983{
9fe0d3c2 1984 U8 * first_variant;
246fae53 1985
7918f24d 1986 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1987 PERL_UNUSED_CONTEXT;
7918f24d 1988
9fe0d3c2 1989 /* This is a no-op if no variants at all in the input */
09af0336 1990 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
1991 return s;
1992 }
1993
1994 {
3c5aa262 1995 U8 * const save = s;
09af0336 1996 U8 * const send = s + *lenp;
3c5aa262
KW
1997 U8 * d;
1998
1999 /* Nothing before the first variant needs to be changed, so start the real
2000 * work there */
2001 s = first_variant;
2002 while (s < send) {
2003 if (! UTF8_IS_INVARIANT(*s)) {
2004 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2005 *lenp = ((STRLEN) -1);
3c5aa262
KW
2006 return 0;
2007 }
2008 s++;
d59937ca
KW
2009 }
2010 s++;
dcad2880 2011 }
dcad2880 2012
3c5aa262
KW
2013 /* Is downgradable, so do it */
2014 d = s = first_variant;
2015 while (s < send) {
2016 U8 c = *s++;
2017 if (! UVCHR_IS_INVARIANT(c)) {
2018 /* Then it is two-byte encoded */
2019 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2020 s++;
2021 }
2022 *d++ = c;
2023 }
2024 *d = '\0';
09af0336 2025 *lenp = d - save;
3c5aa262
KW
2026
2027 return save;
9fe0d3c2 2028 }
6940069f
GS
2029}
2030
2031/*
87cea99e 2032=for apidoc bytes_from_utf8
f9a63242 2033
09af0336 2034Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2035byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2036actually encoded in UTF-8.
2037
2038Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2039the input string.
2040
41ae6089
KW
2041Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2042not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2043C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2044
41ae6089 2045Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2046newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2047returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2048
3bc0c78c 2049Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2050having saved the value of C<*lenp> before the call, and subtracting the
2051after-call value of C<*lenp> from it.
3bc0c78c 2052
37607a96 2053=cut
976c1b08
KW
2054
2055There is a macro that avoids this function call, but this is retained for
2056anyone who calls it with the Perl_ prefix */
f9a63242
JH
2057
2058U8 *
41ae6089 2059Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2060{
7918f24d 2061 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2062 PERL_UNUSED_CONTEXT;
f9a63242 2063
976c1b08
KW
2064 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2065}
2066
2067/*
2068No = here because currently externally undocumented
2069for apidoc bytes_from_utf8_loc
2070
2071Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2072to store the location of the first character in C<"s"> that cannot be
2073converted to non-UTF8.
2074
2075If that parameter is C<NULL>, this function behaves identically to
2076C<bytes_from_utf8>.
2077
2078Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2079C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2080
2081Otherwise, the function returns a newly created C<NUL>-terminated string
2082containing the non-UTF8 equivalent of the convertible first portion of
2083C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2084If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2085and C<*first_non_downgradable> is set to C<NULL>.
2086
2087Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2088first character in the original string that wasn't converted. C<*is_utf8p> is
2089unchanged. Note that the new string may have length 0.
2090
2091Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2092C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2093converts as many characters in it as possible stopping at the first one it
385b74be 2094finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2095set to point to that. The function returns the portion that could be converted
2096in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2097not including the terminating C<NUL>. If the very first character in the
2098original could not be converted, C<*lenp> will be 0, and the new string will
2099contain just a single C<NUL>. If the entire input string was converted,
2100C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2101
2102Upon successful return, the number of variants in the converted portion of the
2103string can be computed by having saved the value of C<*lenp> before the call,
2104and subtracting the after-call value of C<*lenp> from it.
2105
2106=cut
2107
2108
2109*/
2110
2111U8 *
2112Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2113{
2114 U8 *d;
2115 const U8 *original = s;
2116 U8 *converted_start;
2117 const U8 *send = s + *lenp;
f9a63242 2118
976c1b08 2119 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2120
976c1b08
KW
2121 if (! *is_utf8p) {
2122 if (first_unconverted) {
2123 *first_unconverted = NULL;
2124 }
2125
2126 return (U8 *) original;
2127 }
2128
2129 Newx(d, (*lenp) + 1, U8);
2130
2131 converted_start = d;
7299a045
KW
2132 while (s < send) {
2133 U8 c = *s++;
2134 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2135
2136 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2137 * have to stop now */
2138 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2139 if (first_unconverted) {
2140 *first_unconverted = s - 1;
2141 goto finish_and_return;
2142 }
2143 else {
2144 Safefree(converted_start);
2145 return (U8 *) original;
2146 }
2147 }
2148
7299a045
KW
2149 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2150 s++;
38af28cf 2151 }
7299a045
KW
2152 *d++ = c;
2153 }
170a1c22 2154
976c1b08
KW
2155 /* Here, converted the whole of the input */
2156 *is_utf8p = FALSE;
2157 if (first_unconverted) {
2158 *first_unconverted = NULL;
170a1c22 2159 }
976c1b08
KW
2160
2161 finish_and_return:
2162 *d = '\0';
2163 *lenp = d - converted_start;
2164
2165 /* Trim unused space */
2166 Renew(converted_start, *lenp + 1, U8);
2167
2168 return converted_start;
f9a63242
JH
2169}
2170
2171/*
87cea99e 2172=for apidoc bytes_to_utf8
6940069f 2173
09af0336 2174Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2175UTF-8.
09af0336 2176Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2177reflect the new length in bytes.
6940069f 2178
3bc0c78c 2179Upon successful return, the number of variants in the string can be computed by
23b37b12 2180having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2181after-call value of C<*lenp>.
2182
75200dff 2183A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2184
2185If you want to convert to UTF-8 from encodings other than
2186the native (Latin1 or EBCDIC),
a1433954 2187see L</sv_recode_to_utf8>().
c9ada85f 2188
497711e7 2189=cut
6940069f
GS
2190*/
2191
2192U8*
09af0336 2193Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2194{
09af0336 2195 const U8 * const send = s + (*lenp);
6940069f
GS
2196 U8 *d;
2197 U8 *dst;
7918f24d
NC
2198
2199 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2200 PERL_UNUSED_CONTEXT;
6940069f 2201
09af0336 2202 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2203 dst = d;
2204
2205 while (s < send) {
55d09dc8
KW
2206 append_utf8_from_native_byte(*s, &d);
2207 s++;
6940069f
GS
2208 }
2209 *d = '\0';
09af0336 2210 *lenp = d-dst;
6940069f
GS
2211 return dst;
2212}
2213
a0ed51b3 2214/*
dea0fc0b 2215 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2216 *
2217 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2218 * We optimize for native, for obvious reasons. */
2219
2220U8*
dea0fc0b 2221Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2222{
dea0fc0b
JH
2223 U8* pend;
2224 U8* dstart = d;
2225
7918f24d
NC
2226 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2227
dea0fc0b 2228 if (bytelen & 1)
56576a04
KW
2229 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2230 (UV)bytelen);
dea0fc0b
JH
2231
2232 pend = p + bytelen;
2233
a0ed51b3 2234 while (p < pend) {
dea0fc0b
JH
2235 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2236 p += 2;
2d1545e5 2237 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2238 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2239 continue;
2240 }
56d37426
KW
2241 if (uv <= MAX_UTF8_TWO_BYTE) {
2242 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2243 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2244 continue;
2245 }
46956fad
KW
2246#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2247#define LAST_HIGH_SURROGATE 0xDBFF
2248#define FIRST_LOW_SURROGATE 0xDC00
2249#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2250
2251 /* This assumes that most uses will be in the first Unicode plane, not
2252 * needing surrogates */
2253 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2254 && uv <= UNICODE_SURROGATE_LAST))
2255 {
2256 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2257 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2258 }
2259 else {
01ea242b 2260 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2261 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2262 || UNLIKELY(low > LAST_LOW_SURROGATE))
2263 {
01ea242b 2264 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2265 }
2266 p += 2;
46956fad
KW
2267 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2268 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2269 }
a0ed51b3 2270 }
56d37426
KW
2271#ifdef EBCDIC
2272 d = uvoffuni_to_utf8_flags(d, uv, 0);
2273#else
a0ed51b3 2274 if (uv < 0x10000) {
eb160463
GS
2275 *d++ = (U8)(( uv >> 12) | 0xe0);
2276 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2277 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2278 continue;
2279 }
2280 else {
eb160463
GS
2281 *d++ = (U8)(( uv >> 18) | 0xf0);
2282 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2283 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2284 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2285 continue;
2286 }
56d37426 2287#endif
a0ed51b3 2288 }
dea0fc0b 2289 *newlen = d - dstart;
a0ed51b3
LW
2290 return d;
2291}
2292
2293/* Note: this one is slightly destructive of the source. */
2294
2295U8*
dea0fc0b 2296Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2297{
2298 U8* s = (U8*)p;
d4c19fe8 2299 U8* const send = s + bytelen;
7918f24d
NC
2300
2301 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2302
e0ea5e2d 2303 if (bytelen & 1)
147e3846 2304 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2305 (UV)bytelen);
2306
a0ed51b3 2307 while (s < send) {
d4c19fe8 2308 const U8 tmp = s[0];
a0ed51b3
LW
2309 s[0] = s[1];
2310 s[1] = tmp;
2311 s += 2;
2312 }
dea0fc0b 2313 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2314}
2315
922e8cb4
KW
2316bool
2317Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2318{
2319 U8 tmpbuf[UTF8_MAXBYTES+1];
2320 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2321 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2322}
2323
f9ae8fb6
JD
2324/* Internal function so we can deprecate the external one, and call
2325 this one from other deprecated functions in this file */
2326
f2645549
KW
2327bool
2328Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2329{
f2645549 2330 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2331
2332 if (*p == '_')
2333 return TRUE;
f25ce844 2334 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2335}
2336
5092f92a 2337bool
eba68aa0
KW
2338Perl__is_uni_perl_idcont(pTHX_ UV c)
2339{
2340 U8 tmpbuf[UTF8_MAXBYTES+1];
2341 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2342 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2343}
2344
2345bool
f91dcd13
KW
2346Perl__is_uni_perl_idstart(pTHX_ UV c)
2347{
2348 U8 tmpbuf[UTF8_MAXBYTES+1];
2349 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2350 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2351}
2352
3a4c58c9 2353UV
56576a04
KW
2354Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2355 const char S_or_s)
3a4c58c9
KW
2356{
2357 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2358 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2359 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2360 * either "SS" or "Ss". Which one to use is passed into the routine in
2361 * 'S_or_s' to avoid a test */
2362
2363 UV converted = toUPPER_LATIN1_MOD(c);
2364
2365 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2366
2367 assert(S_or_s == 'S' || S_or_s == 's');
2368
6f2d5cbc 2369 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2370 characters in this range */
3a4c58c9
KW
2371 *p = (U8) converted;
2372 *lenp = 1;
2373 return converted;
2374 }
2375
2376 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2377 * which it maps to one of them, so as to only have to have one check for
2378 * it in the main case */
2379 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2380 switch (c) {
2381 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2382 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2383 break;
2384 case MICRO_SIGN:
2385 converted = GREEK_CAPITAL_LETTER_MU;
2386 break;
79e064b9
KW
2387#if UNICODE_MAJOR_VERSION > 2 \
2388 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2389 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2390 case LATIN_SMALL_LETTER_SHARP_S:
2391 *(p)++ = 'S';
2392 *p = S_or_s;
2393 *lenp = 2;
2394 return 'S';
79e064b9 2395#endif
3a4c58c9 2396 default:
56576a04
KW
2397 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2398 " '%c' to map to '%c'",
2399 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2400 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2401 }
2402 }
2403
2404 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2405 *p = UTF8_TWO_BYTE_LO(converted);
2406 *lenp = 2;
2407
2408 return converted;
2409}
2410
50bda2c3
KW
2411/* Call the function to convert a UTF-8 encoded character to the specified case.
2412 * Note that there may be more than one character in the result.
2413 * INP is a pointer to the first byte of the input character
2414 * OUTP will be set to the first byte of the string of changed characters. It
2415 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2416 * LENP will be set to the length in bytes of the string of changed characters
2417 *
56576a04
KW
2418 * The functions return the ordinal of the first character in the string of
2419 * OUTP */
2420#define CALL_UPPER_CASE(uv, s, d, lenp) \
2421 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2422#define CALL_TITLE_CASE(uv, s, d, lenp) \
2423 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2424#define CALL_LOWER_CASE(uv, s, d, lenp) \
2425 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2426
b9992569
KW
2427/* This additionally has the input parameter 'specials', which if non-zero will
2428 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2429 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2430#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2431_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2432
84afefe6
JH
2433UV
2434Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2435{
a1433954
KW
2436 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2437 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2438 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2439 * the changed version may be longer than the original character.
2440 *
2441 * The ordinal of the first character of the changed version is returned
2442 * (but note, as explained above, that there may be more.) */
2443
7918f24d
NC
2444 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2445
3a4c58c9
KW
2446 if (c < 256) {
2447 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2448 }
2449
0ebc6274 2450 uvchr_to_utf8(p, c);
b9992569 2451 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2452}
2453
84afefe6
JH
2454UV
2455Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2456{
7918f24d
NC
2457 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2458
3a4c58c9
KW
2459 if (c < 256) {
2460 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2461 }
2462
0ebc6274 2463 uvchr_to_utf8(p, c);
b9992569 2464 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2465}
2466
afc16117 2467STATIC U8
eaf412bf 2468S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2469{
2470 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2471 * those, converting the result to UTF-8. Since the result is always just
a1433954 2472 * one character, we allow <p> to be NULL */
afc16117
KW
2473
2474 U8 converted = toLOWER_LATIN1(c);
2475
eaf412bf
KW
2476 PERL_UNUSED_ARG(dummy);
2477
afc16117 2478 if (p != NULL) {
6f2d5cbc 2479 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2480 *p = converted;
2481 *lenp = 1;
2482 }
2483 else {
430c9760
KW
2484 /* Result is known to always be < 256, so can use the EIGHT_BIT
2485 * macros */
2486 *p = UTF8_EIGHT_BIT_HI(converted);
2487 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2488 *lenp = 2;
2489 }
2490 }
2491 return converted;
2492}
2493
84afefe6
JH
2494UV
2495Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2496{
7918f24d
NC
2497 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2498
afc16117 2499 if (c < 256) {
eaf412bf 2500 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2501 }
2502
afc16117 2503 uvchr_to_utf8(p, c);
b9992569 2504 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2505}
2506
84afefe6 2507UV
56576a04
KW
2508Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2509 const unsigned int flags)
a1dde8de 2510{
51910141 2511 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2512 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2513 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2514 *
2515 * Not to be used for locale folds
51910141 2516 */
f673fad4 2517
a1dde8de
KW
2518 UV converted;
2519
2520 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2521 PERL_UNUSED_CONTEXT;
a1dde8de 2522
1ca267a5
KW
2523 assert (! (flags & FOLD_FLAGS_LOCALE));
2524
659a7c2d 2525 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2526 converted = GREEK_SMALL_LETTER_MU;
2527 }
9b63e895
KW
2528#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2529 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2530 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2531 else if ( (flags & FOLD_FLAGS_FULL)
2532 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2533 {
1ca267a5
KW
2534 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2535 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2536 * under those circumstances. */
2537 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2538 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2539 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2540 p, *lenp, U8);
2541 return LATIN_SMALL_LETTER_LONG_S;
2542 }
2543 else {
4f489194
KW
2544 *(p)++ = 's';
2545 *p = 's';
2546 *lenp = 2;
2547 return 's';
1ca267a5 2548 }
a1dde8de 2549 }
9b63e895 2550#endif
a1dde8de
KW
2551 else { /* In this range the fold of all other characters is their lower
2552 case */
2553 converted = toLOWER_LATIN1(c);
2554 }
2555
6f2d5cbc 2556 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2557 *p = (U8) converted;
2558 *lenp = 1;
2559 }
2560 else {
2561 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2562 *p = UTF8_TWO_BYTE_LO(converted);
2563 *lenp = 2;
2564 }
2565
2566 return converted;
2567}
2568
2569UV
31f05a37 2570Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2571{
4b593389 2572
a0270393
KW
2573 /* Not currently externally documented, and subject to change
2574 * <flags> bits meanings:
2575 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2576 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2577 * locale are to be used.
a0270393
KW
2578 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2579 */
4b593389 2580
36bb2ab6 2581 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2582
780fcc9f
KW
2583 if (flags & FOLD_FLAGS_LOCALE) {
2584 /* Treat a UTF-8 locale as not being in locale at all */
2585 if (IN_UTF8_CTYPE_LOCALE) {
2586 flags &= ~FOLD_FLAGS_LOCALE;
2587 }
2588 else {
2589 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2590 goto needs_full_generality;
780fcc9f 2591 }
31f05a37
KW
2592 }
2593
a1dde8de 2594 if (c < 256) {
e7b7ac46 2595 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2596 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2597 }
2598
2f306ab9 2599 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2600 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2601 uvchr_to_utf8(p, c);
b9992569 2602 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2603 }
567b353c 2604 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2605 the special flags. */
2606 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2607
2608 needs_full_generality:
a0270393 2609 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2610 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2611 p, lenp, flags);
a0270393 2612 }
84afefe6
JH
2613}
2614
26483009 2615PERL_STATIC_INLINE bool
5141f98e 2616S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2617 const char *const swashname, SV* const invlist)
bde6a22d 2618{
ea317ccb
KW
2619 /* returns a boolean giving whether or not the UTF8-encoded character that
2620 * starts at <p> is in the swash indicated by <swashname>. <swash>
2621 * contains a pointer to where the swash indicated by <swashname>
2622 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2623 * look at <*swash> and only generate a swash if it is not null. <invlist>
2624 * is NULL or an inversion list that defines the swash. If not null, it
2625 * saves time during initialization of the swash.
ea317ccb
KW
2626 *
2627 * Note that it is assumed that the buffer length of <p> is enough to
2628 * contain all the bytes that comprise the character. Thus, <*p> should
2629 * have been checked before this call for mal-formedness enough to assure
2630 * that. */
2631
7918f24d
NC
2632 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2633
492a624f 2634 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2635 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2636 * as far as there being enough bytes available in it to accommodate the
2637 * character without reading beyond the end, and pass that number on to the
2638 * validating routine */
6302f837 2639 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2640 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2641 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2642 1 /* Die */ );
2643 NOT_REACHED; /* NOTREACHED */
28123549 2644 }
86ae6e94 2645
87367d5f
KW
2646 if (!*swash) {
2647 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2648 *swash = _core_swash_init("utf8",
2649
2650 /* Only use the name if there is no inversion
2651 * list; otherwise will go out to disk */
2652 (invlist) ? "" : swashname,
2653
2654 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2655 }
28123549 2656
bde6a22d
NC
2657 return swash_fetch(*swash, p, TRUE) != 0;
2658}
2659
da8c1a98 2660PERL_STATIC_INLINE bool
56576a04
KW
2661S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2662 SV **swash, const char *const swashname,
2663 SV* const invlist)
da8c1a98
KW
2664{
2665 /* returns a boolean giving whether or not the UTF8-encoded character that
2666 * starts at <p>, and extending no further than <e - 1> is in the swash
2667 * indicated by <swashname>. <swash> contains a pointer to where the swash
2668 * indicated by <swashname> is to be stored; which this routine will do, so
2669 * that future calls will look at <*swash> and only generate a swash if it
2670 * is not null. <invlist> is NULL or an inversion list that defines the
2671 * swash. If not null, it saves time during initialization of the swash.
2672 */
2673
2674 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2675
2676 if (! isUTF8_CHAR(p, e)) {
2677 _force_out_malformed_utf8_message(p, e, 0, 1);
2678 NOT_REACHED; /* NOTREACHED */
2679 }
2680
2681 if (!*swash) {
2682 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2683 *swash = _core_swash_init("utf8",
2684
2685 /* Only use the name if there is no inversion
2686 * list; otherwise will go out to disk */
2687 (invlist) ? "" : swashname,
2688
2689 &PL_sv_undef, 1, 0, invlist, &flags);
2690 }
2691
2692 return swash_fetch(*swash, p, TRUE) != 0;
2693}
2694
34aeb2e9
KW
2695STATIC void
2696S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2697 const char * const alternative,
2698 const bool use_locale,
2699 const char * const file,
2700 const unsigned line)
2701{
2702 const char * key;
2703
2704 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2705
2706 if (ckWARN_d(WARN_DEPRECATED)) {
2707
2708 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2709 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2710 if (! PL_seen_deprecated_macro) {
2711 PL_seen_deprecated_macro = newHV();
2712 }
2713 if (! hv_store(PL_seen_deprecated_macro, key,
2714 strlen(key), &PL_sv_undef, 0))
2715 {
2716 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2717 }
2718
c44e9413 2719 if (instr(file, "mathoms.c")) {
607313a1
KW
2720 Perl_warner(aTHX_ WARN_DEPRECATED,
2721 "In %s, line %d, starting in Perl v5.30, %s()"
2722 " will be removed. Avoid this message by"
2723 " converting to use %s().\n",
2724 file, line, name, alternative);
2725 }
2726 else {
34aeb2e9
KW
2727 Perl_warner(aTHX_ WARN_DEPRECATED,
2728 "In %s, line %d, starting in Perl v5.30, %s() will"
2729 " require an additional parameter. Avoid this"
2730 " message by converting to use %s().\n",
2731 file, line, name, alternative);
607313a1 2732 }
34aeb2e9
KW
2733 }
2734 }
2735}
2736
bde6a22d 2737bool
34aeb2e9 2738Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2739 const U8 * const p,
34aeb2e9
KW
2740 const char * const name,
2741 const char * const alternative,
2742 const bool use_utf8,
2743 const bool use_locale,
2744 const char * const file,
2745 const unsigned line)
922e8cb4 2746{
922e8cb4
KW
2747 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2748
34aeb2e9
KW
2749 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2750
2751 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2752
2753 switch (classnum) {
2754 case _CC_WORDCHAR:
2755 case _CC_DIGIT:
2756 case _CC_ALPHA:
2757 case _CC_LOWER:
2758 case _CC_UPPER:
2759 case _CC_PUNCT:
2760 case _CC_PRINT:
2761 case _CC_ALPHANUMERIC:
2762 case _CC_GRAPH:
2763 case _CC_CASED:
2764
2765 return is_utf8_common(p,
2766 &PL_utf8_swash_ptrs[classnum],
2767 swash_property_names[classnum],
2768 PL_XPosix_ptrs[classnum]);
2769
2770 case _CC_SPACE:
2771 return is_XPERLSPACE_high(p);
2772 case _CC_BLANK:
2773 return is_HORIZWS_high(p);
2774 case _CC_XDIGIT:
2775 return is_XDIGIT_high(p);
2776 case _CC_CNTRL:
2777 return 0;
2778 case _CC_ASCII:
2779 return 0;
2780 case _CC_VERTSPACE:
2781 return is_VERTWS_high(p);
2782 case _CC_IDFIRST:
2783 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2784 PL_utf8_perl_idstart
2785 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2786 }
22f0498f
KW
2787 return is_utf8_common(p, &PL_utf8_perl_idstart,
2788 "_Perl_IDStart", NULL);
34aeb2e9
KW
2789 case _CC_IDCONT:
2790 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2791 PL_utf8_perl_idcont
2792 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2793 }
22f0498f
KW
2794 return is_utf8_common(p, &PL_utf8_perl_idcont,
2795 "_Perl_IDCont", NULL);
34aeb2e9
KW
2796 }
2797 }
2798
2799 /* idcont is the same as wordchar below 256 */
2800 if (classnum == _CC_IDCONT) {
2801 classnum = _CC_WORDCHAR;
2802 }
2803 else if (classnum == _CC_IDFIRST) {
2804 if (*p == '_') {
2805 return TRUE;
2806 }
2807 classnum = _CC_ALPHA;
2808 }
2809
2810 if (! use_locale) {
2811 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2812 return _generic_isCC(*p, classnum);
2813 }
922e8cb4 2814
34aeb2e9
KW
2815 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2816 }
2817 else {
2818 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2819 return isFOO_lc(classnum, *p);
2820 }
2821
2822 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2823 }
2824
2825 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2826}
2827
2828bool
da8c1a98
KW
2829Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2830 const U8 * const e)
2831{
2832 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2833
2834 assert(classnum < _FIRST_NON_SWASH_CC);
2835
2836 return is_utf8_common_with_len(p,
2837 e,
2838 &PL_utf8_swash_ptrs[classnum],
2839 swash_property_names[classnum],
2840 PL_XPosix_ptrs[classnum]);
2841}
2842
2843bool
da8c1a98
KW
2844Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2845{
2846 SV* invlist = NULL;
2847
2848 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2849
2850 if (! PL_utf8_perl_idstart) {
2851 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2852 }
2853 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2854 "_Perl_IDStart", invlist);
2855}
2856
2857bool
f2645549 2858Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2859{
f2645549 2860 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2861
2862 if (*p == '_')
2863 return TRUE;
f25ce844 2864 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2865}
2866
2867bool
da8c1a98
KW
2868Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2869{
2870 SV* invlist = NULL;
2871
2872 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2873
2874 if (! PL_utf8_perl_idcont) {
2875 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2876 }
2877 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2878 "_Perl_IDCont", invlist);
2879}
2880
2881bool
f2645549 2882Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2883{
f2645549 2884 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2885
f25ce844 2886 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2887}
2888
2889bool
f2645549 2890Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2891{
f2645549 2892 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2893
f25ce844 2894 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2895}
2896
2897bool
7dbf68d2
KW
2898Perl__is_utf8_mark(pTHX_ const U8 *p)
2899{
7dbf68d2
KW
2900 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2901
f25ce844 2902 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2903}
2904
b9992569 2905 /* change namve uv1 to 'from' */
6a4a25f4 2906STATIC UV
b9992569
KW
2907S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2908 SV **swashp, const char *normal, const char *special)
2909{
0134edef 2910 STRLEN len = 0;
7918f24d 2911
b9992569 2912 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2913
36eaa811
KW
2914 /* For code points that don't change case, we already know that the output
2915 * of this function is the unchanged input, so we can skip doing look-ups
2916 * for them. Unfortunately the case-changing code points are scattered
2917 * around. But there are some long consecutive ranges where there are no
2918 * case changing code points. By adding tests, we can eliminate the lookup
2919 * for all the ones in such ranges. This is currently done here only for
2920 * just a few cases where the scripts are in common use in modern commerce
2921 * (and scripts adjacent to those which can be included without additional
2922 * tests). */
2923
2924 if (uv1 >= 0x0590) {
2925 /* This keeps from needing further processing the code points most
2926 * likely to be used in the following non-cased scripts: Hebrew,
2927 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2928 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2929 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2930 if (uv1 < 0x10A0) {
2931 goto cases_to_self;
2932 }
2933
2934 /* The following largish code point ranges also don't have case
2935 * changes, but khw didn't think they warranted extra tests to speed
2936 * them up (which would slightly slow down everything else above them):
2937 * 1100..139F Hangul Jamo, Ethiopic
2938 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2939 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2940 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2941 * Combining Diacritical Marks Extended, Balinese,
2942 * Sundanese, Batak, Lepcha, Ol Chiki
2943 * 2000..206F General Punctuation
2944 */
2945
2946 if (uv1 >= 0x2D30) {
2947
2948 /* This keeps the from needing further processing the code points
2949 * most likely to be used in the following non-cased major scripts:
2950 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2951 *
2952 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2953 * event that Unicode eventually allocates the unused block as of
2954 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2955 * that the test suite will start having failures to alert you
2956 * should that happen) */
2957 if (uv1 < 0xA640) {
2958 goto cases_to_self;
2959 }
2960
2961 if (uv1 >= 0xAC00) {
2962 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2963 if (ckWARN_d(WARN_SURROGATE)) {
2964 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2965 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
2966 "Operation \"%s\" returns its argument for"
2967 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2968 }
2969 goto cases_to_self;
2970 }
36eaa811
KW
2971
2972 /* AC00..FAFF Catches Hangul syllables and private use, plus
2973 * some others */
2974 if (uv1 < 0xFB00) {
2975 goto cases_to_self;
2976
2977 }
2978
5af9bc97 2979 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
2980 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2981 && ckWARN_d(WARN_DEPRECATED))
2982 {
2983 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2984 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
2985 }
2986 if (ckWARN_d(WARN_NON_UNICODE)) {
2987 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2988 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
2989 "Operation \"%s\" returns its argument for"
2990 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2991 }
2992 goto cases_to_self;
2993 }
3bfc1e70
KW
2994#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2995 if (UNLIKELY(uv1
2996 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2997 {
2998
56576a04
KW
2999 /* As of Unicode 10.0, this means we avoid swash creation
3000 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3001 goto cases_to_self;
3002 }
3003#endif
36eaa811
KW
3004 }
3005 }
9ae3ac1a 3006
36eaa811
KW
3007 /* Note that non-characters are perfectly legal, so no warning should
3008 * be given. There are so few of them, that it isn't worth the extra
3009 * tests to avoid swash creation */
9ae3ac1a
KW
3010 }
3011
0134edef 3012 if (!*swashp) /* load on-demand */
56576a04
KW
3013 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3014 4, 0, NULL, NULL);
0134edef 3015
a6f87d8c 3016 if (special) {
0134edef 3017 /* It might be "special" (sometimes, but not always,
2a37f04d 3018 * a multicharacter mapping) */
4a8240a3 3019 HV *hv = NULL;
b08cf34e
JH
3020 SV **svp;
3021
4a8240a3
KW
3022 /* If passed in the specials name, use that; otherwise use any
3023 * given in the swash */
3024 if (*special != '\0') {
3025 hv = get_hv(special, 0);
3026 }
3027 else {
3028 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3029 if (svp) {
3030 hv = MUTABLE_HV(SvRV(*svp));
3031 }
3032 }
3033
176fe009 3034 if (hv
5f560d8a 3035 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3036 && (*svp))
3037 {
cfd0369c 3038 const char *s;
47654450 3039
cfd0369c 3040 s = SvPV_const(*svp, len);
47654450 3041 if (len == 1)
f4cd282c 3042 /* EIGHTBIT */
c80e42f3 3043 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3044 else {
d2dcd0fb 3045 Copy(s, ustrp, len, U8);
29e98929 3046 }
983ffd37 3047 }
0134edef
JH
3048 }
3049
3050 if (!len && *swashp) {
4a4088c4 3051 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3052
0134edef
JH
3053 if (uv2) {
3054 /* It was "normal" (a single character mapping). */
f4cd282c 3055 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3056 }
3057 }
1feea2c7 3058
cbe07460
KW
3059 if (len) {
3060 if (lenp) {
3061 *lenp = len;
3062 }
3063 return valid_utf8_to_uvchr(ustrp, 0);
3064 }
3065
3066 /* Here, there was no mapping defined, which means that the code point maps
3067 * to itself. Return the inputs */
e24dfe9c 3068 cases_to_self:
bfdf22ec 3069 len = UTF8SKIP(p);
ca9fab46
KW
3070 if (p != ustrp) { /* Don't copy onto itself */
3071 Copy(p, ustrp, len, U8);
3072 }
0134edef 3073
2a37f04d
JH
3074 if (lenp)
3075 *lenp = len;
3076
f4cd282c 3077 return uv1;
cbe07460 3078
a0ed51b3
LW
3079}
3080
051a06d4 3081STATIC UV
56576a04
KW
3082S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3083 U8* const ustrp, STRLEN *lenp)
051a06d4 3084{
4a4088c4 3085 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3086 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3087 * result contains a character that crosses the 255/256 boundary, disallow
3088 * the change, and return the original code point. See L<perlfunc/lc> for
3089 * why;
051a06d4 3090 *
a1433954
KW
3091 * p points to the original string whose case was changed; assumed
3092 * by this routine to be well-formed
051a06d4 3093 * result the code point of the first character in the changed-case string
56576a04
KW
3094 * ustrp points to the changed-case string (<result> represents its
3095 * first char)
051a06d4
KW
3096 * lenp points to the length of <ustrp> */
3097
3098 UV original; /* To store the first code point of <p> */
3099
3100 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3101
a4f12ed7 3102 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3103
3104 /* We know immediately if the first character in the string crosses the
3105 * boundary, so can skip */
3106 if (result > 255) {
3107
3108 /* Look at every character in the result; if any cross the
3109 * boundary, the whole thing is disallowed */
3110 U8* s = ustrp + UTF8SKIP(ustrp);
3111 U8* e = ustrp + *lenp;
3112 while (s < e) {
a4f12ed7 3113 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3114 goto bad_crossing;
3115 }
3116 s += UTF8SKIP(s);
3117 }
3118
613abc6d
KW
3119 /* Here, no characters crossed, result is ok as-is, but we warn. */
3120 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3121 return result;
3122 }
3123
7b52d656 3124 bad_crossing:
051a06d4
KW
3125
3126 /* Failed, have to return the original */
4b88fb76 3127 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3128
3129 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3130 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3131 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3132 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3133 OP_DESC(PL_op),
ab0b796c
KW
3134 original,
3135 original);
051a06d4
KW
3136 Copy(p, ustrp, *lenp, char);
3137 return original;
3138}
3139
607313a1
KW
3140STATIC U32
3141S_check_and_deprecate(pTHX_ const U8 *p,
3142 const U8 **e,
3143 const unsigned int type, /* See below */
3144 const bool use_locale, /* Is this a 'LC_'
3145 macro call? */
3146 const char * const file,
3147 const unsigned line)
3148{
3149 /* This is a temporary function to deprecate the unsafe calls to the case
3150 * changing macros and functions. It keeps all the special stuff in just
3151 * one place.
3152 *
3153 * It updates *e with the pointer to the end of the input string. If using
3154 * the old-style macros, *e is NULL on input, and so this function assumes
3155 * the input string is long enough to hold the entire UTF-8 sequence, and
3156 * sets *e accordingly, but it then returns a flag to pass the
3157 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3158 * using the full length if possible.
3159 *
3160 * It also does the assert that *e > p when *e is not NULL. This should be
3161 * migrated to the callers when this function gets deleted.
3162 *
3163 * The 'type' parameter is used for the caller to specify which case
3164 * changing function this is called from: */
3165
3166# define DEPRECATE_TO_UPPER 0
3167# define DEPRECATE_TO_TITLE 1
3168# define DEPRECATE_TO_LOWER 2
3169# define DEPRECATE_TO_FOLD 3
3170
3171 U32 utf8n_flags = 0;
3172 const char * name;
3173 const char * alternative;
3174
3175 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3176
3177 if (*e == NULL) {
3178 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3179 *e = p + UTF8SKIP(p);
3180
3181 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3182 * there. It could be part of a larger path */
607313a1 3183 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3184 name = instr(file, "mathoms.c")
607313a1
KW
3185 ? "to_utf8_upper"
3186 : "toUPPER_utf8";
3187 alternative = "toUPPER_utf8_safe";
3188 }
3189 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3190 name = instr(file, "mathoms.c")
607313a1
KW
3191 ? "to_utf8_title"
3192 : "toTITLE_utf8";
3193 alternative = "toTITLE_utf8_safe";
3194 }
3195 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3196 name = instr(file, "mathoms.c")
607313a1
KW
3197 ? "to_utf8_lower"
3198 : "toLOWER_utf8";
3199 alternative = "toLOWER_utf8_safe";
3200 }
3201 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3202 name = instr(file, "mathoms.c")
607313a1
KW
3203 ? "to_utf8_fold"
3204 : "toFOLD_utf8";
3205 alternative = "toFOLD_utf8_safe";
3206 }
3207 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3208
3209 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3210 }
3211 else {
3212 assert (p < *e);
3213 }
3214
3215 return utf8n_flags;
3216}
3217
eaf412bf
KW
3218/* The process for changing the case is essentially the same for the four case
3219 * change types, except there are complications for folding. Otherwise the
3220 * difference is only which case to change to. To make sure that they all do
3221 * the same thing, the bodies of the functions are extracted out into the
3222 * following two macros. The functions are written with the same variable
3223 * names, and these are known and used inside these macros. It would be
3224 * better, of course, to have inline functions to do it, but since different
3225 * macros are called, depending on which case is being changed to, this is not
3226 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3227 * function can start with the common start macro, then finish with its special
3228 * handling; while the other three cases can just use the common end macro.
3229 *
3230 * The algorithm is to use the proper (passed in) macro or function to change
3231 * the case for code points that are below 256. The macro is used if using
3232 * locale rules for the case change; the function if not. If the code point is
3233 * above 255, it is computed from the input UTF-8, and another macro is called
3234 * to do the conversion. If necessary, the output is converted to UTF-8. If
3235 * using a locale, we have to check that the change did not cross the 255/256
3236 * boundary, see check_locale_boundary_crossing() for further details.
3237 *
3238 * The macros are split with the correct case change for the below-256 case
3239 * stored into 'result', and in the middle of an else clause for the above-255
3240 * case. At that point in the 'else', 'result' is not the final result, but is
3241 * the input code point calculated from the UTF-8. The fold code needs to
3242 * realize all this and take it from there.
3243 *
3244 * If you read the two macros as sequential, it's easier to understand what's
3245 * going on. */
3246#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3247 L1_func_extra_param) \
a239b1e2 3248 \
eaf412bf
KW
3249 if (flags & (locale_flags)) { \
3250 /* Treat a UTF-8 locale as not being in locale at all */ \
3251 if (IN_UTF8_CTYPE_LOCALE) { \
3252 flags &= ~(locale_flags); \
3253 } \
3254 else { \
3255 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3256 } \
3257 } \
3258 \
3259 if (UTF8_IS_INVARIANT(*p)) { \
3260 if (flags & (locale_flags)) { \
3261 result = LC_L1_change_macro(*p); \
3262 } \
3263 else { \
3264 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3265 } \
3266 } \
a239b1e2 3267 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3268 if (flags & (locale_flags)) { \
3269 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3270 *(p+1))); \
3271 } \
3272 else { \
3273 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3274 ustrp, lenp, L1_func_extra_param); \
3275 } \
3276 } \
fa8ab374
KW
3277 else { /* malformed UTF-8 or ord above 255 */ \
3278 STRLEN len_result; \
fa8ab374
KW
3279 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3280 if (len_result == (STRLEN) -1) { \
607313a1
KW
3281 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3282 1 /* Die */ ); \
fa8ab374 3283 }
eaf412bf
KW
3284
3285#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3286 result = change_macro(result, p, ustrp, lenp); \
3287 \
3288 if (flags & (locale_flags)) { \
3289 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3290 } \
3291 return result; \
3292 } \
3293 \
3294 /* Here, used locale rules. Convert back to UTF-8 */ \
3295 if (UTF8_IS_INVARIANT(result)) { \
3296 *ustrp = (U8) result; \
3297 *lenp = 1; \
3298 } \
3299 else { \
3300 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3301 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3302 *lenp = 2; \
3303 } \
3304 \
3305 return result;
3306
d3e79532 3307/*
87cea99e 3308=for apidoc to_utf8_upper
d3e79532 3309
a239b1e2 3310Instead use L</toUPPER_utf8_safe>.
a1433954 3311
d3e79532
JH
3312=cut */
3313
051a06d4 3314/* Not currently externally documented, and subject to change:
31f05a37
KW
3315 * <flags> is set iff iff the rules from the current underlying locale are to
3316 * be used. */
051a06d4 3317
2104c8d9 3318UV
607313a1
KW
3319Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3320 const U8 *e,
3321 U8* ustrp,
3322 STRLEN *lenp,
3323 bool flags,
3324 const char * const file,
3325 const int line)
a0ed51b3 3326{
051a06d4 3327 UV result;
607313a1
KW
3328 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3329 cBOOL(flags), file, line);
051a06d4
KW
3330
3331 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3332
eaf412bf
KW
3333 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3334 /* 2nd char of uc(U+DF) is 'S' */
3335 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3336 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3337}
a0ed51b3 3338
d3e79532 3339/*
87cea99e 3340=for apidoc to_utf8_title
d3e79532 3341
a239b1e2 3342Instead use L</toTITLE_utf8_safe>.
a1433954 3343
d3e79532
JH
3344=cut */
3345
051a06d4 3346/* Not currently externally documented, and subject to change:
31f05a37
KW
3347 * <flags> is set iff the rules from the current underlying locale are to be
3348 * used. Since titlecase is not defined in POSIX, for other than a
3349 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3350 */
051a06d4 3351
983ffd37 3352UV
607313a1
KW
3353Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3354 const U8 *e,
3355 U8* ustrp,
3356 STRLEN *lenp,
3357 bool flags,
3358 const char * const file,
3359 const int line)
983ffd37 3360{
051a06d4 3361 UV result;
607313a1
KW
3362 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3363 cBOOL(flags), file, line);
051a06d4
KW
3364
3365 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3366
eaf412bf
KW
3367 /* 2nd char of ucfirst(U+DF) is 's' */
3368 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3369 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3370}
3371
d3e79532 3372/*
87cea99e 3373=for apidoc to_utf8_lower
d3e79532 3374
a239b1e2 3375Instead use L</toLOWER_utf8_safe>.
a1433954 3376
d3e79532
JH
3377=cut */
3378
051a06d4 3379/* Not currently externally documented, and subject to change:
31f05a37
KW
3380 * <flags> is set iff iff the rules from the current underlying locale are to
3381 * be used.
3382 */
051a06d4 3383
2104c8d9 3384UV
607313a1
KW
3385Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3386 const U8 *e,
3387 U8* ustrp,
3388 STRLEN *lenp,
3389 bool flags,
3390 const char * const file,
3391 const int line)
a0ed51b3 3392{
051a06d4 3393 UV result;
607313a1
KW
3394 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3395 cBOOL(flags), file, line);
051a06d4 3396
051a06d4 3397 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3398
eaf412bf
KW
3399 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3400 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3401}
3402
d3e79532 3403/*
87cea99e 3404=for apidoc to_utf8_fold
d3e79532 3405
a239b1e2 3406Instead use L</toFOLD_utf8_safe>.
a1433954 3407
d3e79532
JH
3408=cut */
3409
051a06d4
KW
3410/* Not currently externally documented, and subject to change,
3411 * in <flags>
31f05a37
KW
3412 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3413 * locale are to be used.
051a06d4
KW
3414 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3415 * otherwise simple folds
a0270393
KW
3416 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3417 * prohibited
445bf929 3418 */
36bb2ab6 3419
b4e400f9 3420UV
607313a1
KW
3421Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3422 const U8 *e,
3423 U8* ustrp,
3424 STRLEN *lenp,
3425 U8 flags,
3426 const char * const file,
3427 const int line)
b4e400f9 3428{
051a06d4 3429 UV result;
607313a1
KW
3430 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3431 cBOOL(flags), file, line);
051a06d4 3432
36bb2ab6 3433 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3434
a0270393
KW
3435 /* These are mutually exclusive */
3436 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3437
50ba90ff
KW
3438 assert(p != ustrp); /* Otherwise overwrites */
3439
eaf412bf
KW
3440 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3441 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3442
eaf412bf 3443 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3444
1ca267a5
KW
3445 if (flags & FOLD_FLAGS_LOCALE) {
3446
76f2ffcd 3447# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3448 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3449
3450# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3451# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3452
3453 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3454
538e84ed
KW
3455 /* Special case these two characters, as what normally gets
3456 * returned under locale doesn't work */
76f2ffcd
KW
3457 if (UTF8SKIP(p) == cap_sharp_s_len
3458 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3459 {
ab0b796c
KW
3460 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3461 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3462 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3463 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3464 goto return_long_s;
3465 }
0766489e
KW
3466 else
3467#endif
3468 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3469 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3470 {
ab0b796c
KW
3471 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3472 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3473 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3474 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3475 goto return_ligature_st;
3476 }
74894415
KW
3477
3478#if UNICODE_MAJOR_VERSION == 3 \
3479 && UNICODE_DOT_VERSION == 0 \
3480 && UNICODE_DOT_DOT_VERSION == 1
3481# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3482
3483 /* And special case this on this Unicode version only, for the same
3484 * reaons the other two are special cased. They would cross the
3485 * 255/256 boundary which is forbidden under /l, and so the code
3486 * wouldn't catch that they are equivalent (which they are only in
3487 * this release) */
3488 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3489 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3490 {
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{0130}\") on non-UTF-8 locale; "
3494 "resolved to \"\\x{0131}\".");
3495 goto return_dotless_i;
3496 }
3497#endif
3498
357aadde 3499 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3500 }
a0270393
KW
3501 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3502 return result;
3503 }
3504 else {
4a4088c4 3505 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3506 * character above the ASCII range, and the result should not
3507 * contain an ASCII character. */
a0270393
KW
3508
3509 UV original; /* To store the first code point of <p> */
3510
3511 /* Look at every character in the result; if any cross the
3512 * boundary, the whole thing is disallowed */
3513 U8* s = ustrp;
3514 U8* e = ustrp + *lenp;
3515 while (s < e) {
3516 if (isASCII(*s)) {
3517 /* Crossed, have to return the original */
3518 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3519
9fc2026f 3520 /* But in these instances, there is an alternative we can
1ca267a5 3521 * return that is valid */
0766489e
KW
3522 if (original == LATIN_SMALL_LETTER_SHARP_S
3523#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3524 || original == LATIN_CAPITAL_LETTER_SHARP_S
3525#endif
3526 ) {
1ca267a5
KW
3527 goto return_long_s;
3528 }
9fc2026f
KW
3529 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3530 goto return_ligature_st;
3531 }
74894415
KW
3532#if UNICODE_MAJOR_VERSION == 3 \
3533 && UNICODE_DOT_VERSION == 0 \
3534 && UNICODE_DOT_DOT_VERSION == 1
3535
3536 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3537 goto return_dotless_i;
3538 }
3539#endif
a0270393
KW
3540 Copy(p, ustrp, *lenp, char);
3541 return original;
3542 }
3543 s += UTF8SKIP(s);
3544 }
051a06d4 3545
a0270393
KW
3546 /* Here, no characters crossed, result is ok as-is */
3547 return result;
3548 }
051a06d4
KW
3549 }
3550
4a4088c4 3551 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3552 if (UTF8_IS_INVARIANT(result)) {
3553 *ustrp = (U8) result;
3554 *lenp = 1;
3555 }
3556 else {
62cb07ea
KW
3557 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3558 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3559 *lenp = 2;
3560 }
3561
051a06d4 3562 return result;
1ca267a5
KW
3563
3564 return_long_s:
3565 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3566 * folds to a string of two of these characters. By returning this
3567 * instead, then, e.g.,
3568 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3569 * works. */
3570
3571 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3572 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3573 ustrp, *lenp, U8);
3574 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3575
3576 return_ligature_st:
3577 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3578 * have the other one fold to it */
3579
3580 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3581 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3582 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3583
3584#if UNICODE_MAJOR_VERSION == 3 \
3585 && UNICODE_DOT_VERSION == 0 \
3586 && UNICODE_DOT_DOT_VERSION == 1
3587
3588 return_dotless_i:
3589 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3590 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3591 return LATIN_SMALL_LETTER_DOTLESS_I;
3592
3593#endif
3594
a0ed51b3
LW
3595}
3596
711a919c 3597/* Note:
f90a9a02 3598 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3599 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3600 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3601 */
c4a5db0c 3602
a0ed51b3 3603SV*
56576a04
KW
3604Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3605 I32 minbits, I32 none)
a0ed51b3 3606{
c4a5db0c
KW
3607 PERL_ARGS_ASSERT_SWASH_INIT;
3608
3609 /* Returns a copy of a swash initiated by the called function. This is the
3610 * public interface, and returning a copy prevents others from doing
3611 * mischief on the original */
3612
56576a04
KW
3613 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3614 NULL, NULL));
c4a5db0c
KW
3615}
3616
3617SV*
56576a04
KW
3618Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3619 I32 minbits, I32 none, SV* invlist,
3620 U8* const flags_p)
c4a5db0c 3621{
2c1f00b9
YO
3622
3623 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3624 * use the following define */
3625
3626#define CORE_SWASH_INIT_RETURN(x) \
3627 PL_curpm= old_PL_curpm; \
3628 return x
3629
c4a5db0c 3630 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3631 * by calling utf8_heavy.pl in the general case. The returned value may be
3632 * the swash's inversion list instead if the input parameters allow it.
3633 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3634 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3635 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3636 *
3637 * This interface should only be used by functions that won't destroy or
3638 * adversely change the swash, as doing so affects all other uses of the
3639 * swash in the program; the general public should use 'Perl_swash_init'
3640 * instead.
3641 *
3642 * pkg is the name of the package that <name> should be in.
3643 * name is the name of the swash to find. Typically it is a Unicode
3644 * property name, including user-defined ones
3645 * listsv is a string to initialize the swash with. It must be of the form
3646 * documented as the subroutine return value in
3647 * L<perlunicode/User-Defined Character Properties>
3648 * minbits is the number of bits required to represent each data element.
3649 * It is '1' for binary properties.
3650 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3651 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3652 * flags_p if non-NULL is the address of various input and output flag bits
3653 * to the routine, as follows: ('I' means is input to the routine;
3654 * 'O' means output from the routine. Only flags marked O are
3655 * meaningful on return.)
3656 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3657 * came from a user-defined property. (I O)
5d3d13d1
KW
3658 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3659 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3660 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3661 * return of an inversion list instead of a swash hash if this routine
3662 * thinks that would result in faster execution of swash_fetch() later
3663 * on. (I)
9a53f6cf
KW
3664 *
3665 * Thus there are three possible inputs to find the swash: <name>,
3666 * <listsv>, and <invlist>. At least one must be specified. The result
3667 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3668 * actions can intersect, etc. what <name> gives. To avoid going out to
3669 * disk at all, <invlist> should specify completely what the swash should
3670 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3671 *
3672 * <invlist> is only valid for binary properties */
c4a5db0c 3673
2c1f00b9
YO
3674 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3675
c4a5db0c 3676 SV* retval = &PL_sv_undef;
83199d38 3677 HV* swash_hv = NULL;
87367d5f
KW
3678 const int invlist_swash_boundary =
3679 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3680 ? 512 /* Based on some benchmarking, but not extensive, see commit
3681 message */
3682 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3683
3684 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3685 assert(! invlist || minbits == 1);
3686
56576a04
KW
3687 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3688 regex that triggered the swash init and the swash init
3689 perl logic itself. See perl #122747 */
2c1f00b9 3690
9a53f6cf
KW
3691 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3692 * so */
3693 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3694 dSP;
3695 const size_t pkg_len = strlen(pkg);
3696 const size_t name_len = strlen(name);
3697 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3698 SV* errsv_save;
3699 GV *method;
3700
3701 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3702
3703 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3704 ENTER;
69794297 3705 SAVEHINTS();
2782061f 3706 save_re_context();
650f067c
JL
3707 /* We might get here via a subroutine signature which uses a utf8
3708 * parameter name, at which point PL_subname will have been set
3709 * but not yet used. */
3710 save_item(PL_subname);
69794297
KW
3711 if (PL_parser && PL_parser->error_count)
3712 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3713 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3714 if (!method) { /* demand load UTF-8 */
69794297 3715 ENTER;
db2c6cb3
FC
3716 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3717 GvSV(PL_errgv) = NULL;
1a419e6b 3718#ifndef NO_TAINT_SUPPORT
69794297
KW
3719 /* It is assumed that callers of this routine are not passing in
3720 * any user derived data. */
2782061f
DM
3721 /* Need to do this after save_re_context() as it will set
3722 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3723 * in Perl_magic_get). Even line to create errsv_save can turn on
3724 * PL_tainted. */
284167a5
SM
3725 SAVEBOOL(TAINT_get);
3726 TAINT_NOT;
3727#endif
69794297
KW
3728 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3729 NULL);
eed484f9 3730 {
db2c6cb3
FC<