This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Comments, white-space only
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
76513bdc 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf ". This will be fatal in Perl 5.28";
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
c15a80f3 81 SAVEI8(PL_dowarn);
9cbfb8ab
KW
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
c94c2f39
KW
119/* All these formats take a single UV code point argument */
120const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
121const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
122 " is not recommended for open interchange";
123const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
124 " may not be portable";
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
2b5e7bc2
KW
1307 /* isn't problematic if < this */
1308 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1309 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1310
1311 /* if overflow, we know without looking further
1312 * precisely which of the problematic types it is,
1313 * and we deal with those in the overflow handling
1314 * code */
1315 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1316 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1317 && ((flags & ( UTF8_DISALLOW_NONCHAR
1318 |UTF8_DISALLOW_SURROGATE
1319 |UTF8_DISALLOW_SUPER
1320 |UTF8_DISALLOW_ABOVE_31_BIT
1321 |UTF8_WARN_NONCHAR
1322 |UTF8_WARN_SURROGATE
1323 |UTF8_WARN_SUPER
1324 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1325 /* In case of a malformation, 'uv' is not valid, and has
1326 * been changed to something in the Unicode range.
1327 * Currently we don't output a deprecation message if there
1328 * is already a malformation, so we don't have to special
1329 * case the test immediately below */
760c7c2f
KW
1330 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1331 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1332 {
2b5e7bc2
KW
1333 /* If there were no malformations, or the only malformation is an
1334 * overlong, 'uv' is valid */
1335 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1336 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1337 possible_problems |= UTF8_GOT_SURROGATE;
1338 }
1339 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1340 possible_problems |= UTF8_GOT_SUPER;
1341 }
1342 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1343 possible_problems |= UTF8_GOT_NONCHAR;
1344 }
1345 }
1346 else { /* Otherwise, need to look at the source UTF-8, possibly
1347 adjusted to be non-overlong */
1348
1349 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1350 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1351 {
2b5e7bc2
KW
1352 possible_problems |= UTF8_GOT_SUPER;
1353 }
1354 else if (curlen > 1) {
1355 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1356 NATIVE_UTF8_TO_I8(*adjusted_s0),
1357 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1358 {
2b5e7bc2 1359 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1360 }
2b5e7bc2
KW
1361 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1362 NATIVE_UTF8_TO_I8(*adjusted_s0),
1363 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1364 {
1365 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1366 }
1367 }
c0236afe 1368
2b5e7bc2
KW
1369 /* We need a complete well-formed UTF-8 character to discern
1370 * non-characters, so can't look for them here */
1371 }
1372 }
949cf498 1373
2b5e7bc2
KW
1374 ready_to_handle_errors:
1375
1376 /* At this point:
1377 * curlen contains the number of bytes in the sequence that
1378 * this call should advance the input by.
e308b348
KW
1379 * avail_len gives the available number of bytes passed in, but
1380 * only if this is less than the expected number of
1381 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1382 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1383 * is set in it for each potential problem found.
1384 * uv contains the code point the input sequence
1385 * represents; or if there is a problem that prevents
1386 * a well-defined value from being computed, it is
1387 * some subsitute value, typically the REPLACEMENT
1388 * CHARACTER.
1389 * s0 points to the first byte of the character
56576a04
KW
1390 * s points to just after were we left off processing
1391 * the character
1392 * send points to just after where that character should
1393 * end, based on how many bytes the start byte tells
1394 * us should be in it, but no further than s0 +
1395 * avail_len
2b5e7bc2
KW
1396 * adjusted_s0 normally is the same as s0, but in case of an
1397 * overlong for which the UTF-8 matters below, it is
1398 * the first byte of the shortest form representation
1399 * of the input.
1400 * adjusted_send normally is the same as 'send', but if adjusted_s0
1401 * is set to something other than s0, this points one
1402 * beyond its end
1403 */
eb83ed87 1404
2b5e7bc2
KW
1405 if (UNLIKELY(possible_problems)) {
1406 bool disallowed = FALSE;
1407 const U32 orig_problems = possible_problems;
1408
1409 while (possible_problems) { /* Handle each possible problem */
1410 UV pack_warn = 0;
1411 char * message = NULL;
1412
1413 /* Each 'if' clause handles one problem. They are ordered so that
1414 * the first ones' messages will be displayed before the later
1415 * ones; this is kinda in decreasing severity order */
1416 if (possible_problems & UTF8_GOT_OVERFLOW) {
1417
56576a04
KW
1418 /* Overflow means also got a super and are using Perl's
1419 * extended UTF-8, but we handle all three cases here */
2b5e7bc2
KW
1420 possible_problems
1421 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1422 *errors |= UTF8_GOT_OVERFLOW;
1423
1424 /* But the API says we flag all errors found */
1425 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1426 *errors |= UTF8_GOT_SUPER;
1427 }
ddb65933
KW
1428 if (flags
1429 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1430 {
f9380377
KW
1431 *errors |= UTF8_GOT_ABOVE_31_BIT;
1432 }
2b5e7bc2 1433
d60baaa7 1434 /* Disallow if any of the three categories say to */
56576a04 1435 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7
KW
1436 || (flags & ( UTF8_DISALLOW_SUPER
1437 |UTF8_DISALLOW_ABOVE_31_BIT)))
1438 {
1439 disallowed = TRUE;
1440 }
1441
d60baaa7
KW
1442 /* Likewise, warn if any say to, plus if deprecation warnings
1443 * are on, because this code point is above IV_MAX */
56576a04 1444 if ( ckWARN_d(WARN_DEPRECATED)
d60baaa7
KW
1445 || ! (flags & UTF8_ALLOW_OVERFLOW)
1446 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1447 {
2b5e7bc2 1448
ddb65933
KW
1449 /* The warnings code explicitly says it doesn't handle the
1450 * case of packWARN2 and two categories which have
1451 * parent-child relationship. Even if it works now to
1452 * raise the warning if either is enabled, it wouldn't
1453 * necessarily do so in the future. We output (only) the
56576a04 1454 * most dire warning */
ddb65933
KW
1455 if (! (flags & UTF8_CHECK_ONLY)) {
1456 if (ckWARN_d(WARN_UTF8)) {
1457 pack_warn = packWARN(WARN_UTF8);
1458 }
1459 else if (ckWARN_d(WARN_NON_UNICODE)) {
1460 pack_warn = packWARN(WARN_NON_UNICODE);
1461 }
1462 if (pack_warn) {
1463 message = Perl_form(aTHX_ "%s: %s (overflows)",
1464 malformed_text,
05b9033b 1465 _byte_dump_string(s0, curlen, 0));
ddb65933 1466 }
2b5e7bc2
KW
1467 }
1468 }
1469 }
1470 else if (possible_problems & UTF8_GOT_EMPTY) {
1471 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1472 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1473
1474 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1475
1476 /* This so-called malformation is now treated as a bug in
1477 * the caller. If you have nothing to decode, skip calling
1478 * this function */
1479 assert(0);
1480
2b5e7bc2
KW
1481 disallowed = TRUE;
1482 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1483 pack_warn = packWARN(WARN_UTF8);
1484 message = Perl_form(aTHX_ "%s (empty string)",
1485 malformed_text);
1486 }
1487 }
1488 }
1489 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1490 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1491 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1492
1493 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1494 disallowed = TRUE;
1495 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1496 pack_warn = packWARN(WARN_UTF8);
1497 message = Perl_form(aTHX_
1498 "%s: %s (unexpected continuation byte 0x%02x,"
1499 " with no preceding start byte)",
1500 malformed_text,
7e2f38b2 1501 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1502 }
1503 }
1504 }
2b5e7bc2
KW
1505 else if (possible_problems & UTF8_GOT_SHORT) {
1506 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1507 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1508
1509 if (! (flags & UTF8_ALLOW_SHORT)) {
1510 disallowed = TRUE;
1511 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1512 pack_warn = packWARN(WARN_UTF8);
1513 message = Perl_form(aTHX_
56576a04
KW
1514 "%s: %s (too short; %d byte%s available, need %d)",
1515 malformed_text,
1516 _byte_dump_string(s0, send - s0, 0),
1517 (int)avail_len,
1518 avail_len == 1 ? "" : "s",
1519 (int)expectlen);
2b5e7bc2
KW
1520 }
1521 }
ba210ebe 1522
2b5e7bc2 1523 }
e308b348
KW
1524 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1525 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1526 *errors |= UTF8_GOT_NON_CONTINUATION;
1527
1528 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1529 disallowed = TRUE;
1530 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1531
1532 /* If we don't know for sure that the input length is
1533 * valid, avoid as much as possible reading past the
1534 * end of the buffer */
1535 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1536 ? s - s0
1537 : send - s0;
e308b348
KW
1538 pack_warn = packWARN(WARN_UTF8);
1539 message = Perl_form(aTHX_ "%s",
1540 unexpected_non_continuation_text(s0,
99a765e9 1541 printlen,
e308b348
KW
1542 s - s0,
1543 (int) expectlen));
1544 }
1545 }
1546 }
2b5e7bc2
KW
1547 else if (possible_problems & UTF8_GOT_LONG) {
1548 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1549 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1550
94953955
KW
1551 if (flags & UTF8_ALLOW_LONG) {
1552
1553 /* We don't allow the actual overlong value, unless the
1554 * special extra bit is also set */
1555 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1556 & ~UTF8_ALLOW_LONG)))
1557 {
1558 uv = UNICODE_REPLACEMENT;
1559 }
1560 }
1561 else {
2b5e7bc2
KW
1562 disallowed = TRUE;
1563
1564 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1565 pack_warn = packWARN(WARN_UTF8);
1566
1567 /* These error types cause 'uv' to be something that
1568 * isn't what was intended, so can't use it in the
1569 * message. The other error types either can't
1570 * generate an overlong, or else the 'uv' is valid */
1571 if (orig_problems &
1572 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1573 {
1574 message = Perl_form(aTHX_
1575 "%s: %s (any UTF-8 sequence that starts"
1576 " with \"%s\" is overlong which can and"
1577 " should be represented with a"
1578 " different, shorter sequence)",
1579 malformed_text,
7e2f38b2
KW
1580 _byte_dump_string(s0, send - s0, 0),
1581 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1582 }
1583 else {
1584 U8 tmpbuf[UTF8_MAXBYTES+1];
1585 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1586 uv, 0);
1587 message = Perl_form(aTHX_
1588 "%s: %s (overlong; instead use %s to represent"
147e3846 1589 " U+%0*" UVXf ")",
2b5e7bc2 1590 malformed_text,
05b9033b 1591 _byte_dump_string(s0, curlen, 0),
7e2f38b2 1592 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2b5e7bc2
KW
1593 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1594 small code points */
1595 uv);
1596 }
1597 }
1598 }
1599 }
1600 else if (possible_problems & UTF8_GOT_SURROGATE) {
1601 possible_problems &= ~UTF8_GOT_SURROGATE;
1602
f9380377
KW
1603 if (flags & UTF8_WARN_SURROGATE) {
1604 *errors |= UTF8_GOT_SURROGATE;
1605
1606 if ( ! (flags & UTF8_CHECK_ONLY)
1607 && ckWARN_d(WARN_SURROGATE))
1608 {
2b5e7bc2
KW
1609 pack_warn = packWARN(WARN_SURROGATE);
1610
1611 /* These are the only errors that can occur with a
1612 * surrogate when the 'uv' isn't valid */
1613 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1614 message = Perl_form(aTHX_
1615 "UTF-16 surrogate (any UTF-8 sequence that"
1616 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1617 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1618 }
1619 else {
c94c2f39 1620 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1621 }
f9380377 1622 }
2b5e7bc2 1623 }
ba210ebe 1624
2b5e7bc2
KW
1625 if (flags & UTF8_DISALLOW_SURROGATE) {
1626 disallowed = TRUE;
f9380377 1627 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1628 }
1629 }
1630 else if (possible_problems & UTF8_GOT_SUPER) {
1631 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1632
f9380377
KW
1633 if (flags & UTF8_WARN_SUPER) {
1634 *errors |= UTF8_GOT_SUPER;
1635
1636 if ( ! (flags & UTF8_CHECK_ONLY)
1637 && ckWARN_d(WARN_NON_UNICODE))
1638 {
2b5e7bc2
KW
1639 pack_warn = packWARN(WARN_NON_UNICODE);
1640
1641 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1642 message = Perl_form(aTHX_
1643 "Any UTF-8 sequence that starts with"
1644 " \"%s\" is for a non-Unicode code point,"
1645 " may not be portable",
7e2f38b2 1646 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1647 }
1648 else {
c94c2f39 1649 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 1650 }
f9380377 1651 }
2b5e7bc2 1652 }
ba210ebe 1653
2b5e7bc2
KW
1654 /* The maximum code point ever specified by a standard was
1655 * 2**31 - 1. Anything larger than that is a Perl extension
1656 * that very well may not be understood by other applications
1657 * (including earlier perl versions on EBCDIC platforms). We
1658 * test for these after the regular SUPER ones, and before
1659 * possibly bailing out, so that the slightly more dire warning
1660 * will override the regular one. */
1661 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1662 |UTF8_WARN_SUPER
1663 |UTF8_DISALLOW_ABOVE_31_BIT))
1664 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1665 && UNLIKELY(is_utf8_cp_above_31_bits(
56576a04
KW
1666 adjusted_s0,
1667 adjusted_send)))
2b5e7bc2
KW
1668 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1669 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1670 {
1671 if ( ! (flags & UTF8_CHECK_ONLY)
1672 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1673 && ckWARN_d(WARN_UTF8))
1674 {
1675 pack_warn = packWARN(WARN_UTF8);
1676
1677 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1678 message = Perl_form(aTHX_
1679 "Any UTF-8 sequence that starts with"
1680 " \"%s\" is for a non-Unicode code"
1681 " point, and is not portable",
7e2f38b2 1682 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1683 }
1684 else {
1685 message = Perl_form(aTHX_
c94c2f39 1686 above_31_bit_cp_format, uv);
2b5e7bc2
KW
1687 }
1688 }
1689
ddb65933
KW
1690 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1691 |UTF8_DISALLOW_ABOVE_31_BIT))
1692 {
f9380377
KW
1693 *errors |= UTF8_GOT_ABOVE_31_BIT;
1694
1695 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1696 disallowed = TRUE;
1697 }
2b5e7bc2
KW
1698 }
1699 }
eb83ed87 1700
2b5e7bc2 1701 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1702 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1703 disallowed = TRUE;
1704 }
eb83ed87 1705
2b5e7bc2
KW
1706 /* The deprecated warning overrides any non-deprecated one. If
1707 * there are other problems, a deprecation message is not
1708 * really helpful, so don't bother to raise it in that case.
1709 * This also keeps the code from having to handle the case
1710 * where 'uv' is not valid. */
1711 if ( ! (orig_problems
1712 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
76513bdc
KW
1713 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1714 && ckWARN_d(WARN_DEPRECATED))
1715 {
1716 message = Perl_form(aTHX_ cp_above_legal_max,
1717 uv, MAX_NON_DEPRECATED_CP);
1718 pack_warn = packWARN(WARN_DEPRECATED);
2b5e7bc2
KW
1719 }
1720 }
1721 else if (possible_problems & UTF8_GOT_NONCHAR) {
1722 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1723
f9380377
KW
1724 if (flags & UTF8_WARN_NONCHAR) {
1725 *errors |= UTF8_GOT_NONCHAR;
1726
1727 if ( ! (flags & UTF8_CHECK_ONLY)
1728 && ckWARN_d(WARN_NONCHAR))
1729 {
2b5e7bc2
KW
1730 /* The code above should have guaranteed that we don't
1731 * get here with errors other than overlong */
1732 assert (! (orig_problems
1733 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1734
1735 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 1736 message = Perl_form(aTHX_ nonchar_cp_format, uv);
f9380377 1737 }
2b5e7bc2 1738 }
5b311467 1739
2b5e7bc2
KW
1740 if (flags & UTF8_DISALLOW_NONCHAR) {
1741 disallowed = TRUE;
f9380377 1742 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1743 }
1744 } /* End of looking through the possible flags */
1745
1746 /* Display the message (if any) for the problem being handled in
1747 * this iteration of the loop */
1748 if (message) {
1749 if (PL_op)
1750 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1751 OP_DESC(PL_op));
1752 else
1753 Perl_warner(aTHX_ pack_warn, "%s", message);
1754 }
ddb65933 1755 } /* End of 'while (possible_problems)' */
a0dbb045 1756
2b5e7bc2
KW
1757 /* Since there was a possible problem, the returned length may need to
1758 * be changed from the one stored at the beginning of this function.
1759 * Instead of trying to figure out if that's needed, just do it. */
1760 if (retlen) {
1761 *retlen = curlen;
1762 }
a0dbb045 1763
2b5e7bc2
KW
1764 if (disallowed) {
1765 if (flags & UTF8_CHECK_ONLY && retlen) {
1766 *retlen = ((STRLEN) -1);
1767 }
1768 return 0;
1769 }
eb83ed87 1770 }
ba210ebe 1771
2b5e7bc2 1772 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1773}
1774
8e84507e 1775/*
ec5f19d0
KW
1776=for apidoc utf8_to_uvchr_buf
1777
1778Returns the native code point of the first character in the string C<s> which
1779is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1780C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1781
524080c4
KW
1782If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1783enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1784C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1785(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1786C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1787the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1788See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1789returned.
ec5f19d0 1790
760c7c2f
KW
1791Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1792unless those are turned off.
1793
ec5f19d0 1794=cut
52be2536
KW
1795
1796Also implemented as a macro in utf8.h
1797
ec5f19d0
KW
1798*/
1799
1800
1801UV
1802Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1803{
7f974d7e
KW
1804 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1805
ec5f19d0
KW
1806 assert(s < send);
1807
1808 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1809 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1810}
1811
52be2536
KW
1812/* This is marked as deprecated
1813 *
ec5f19d0
KW
1814=for apidoc utf8_to_uvuni_buf
1815
de69f3af
KW
1816Only in very rare circumstances should code need to be dealing in Unicode
1817(as opposed to native) code points. In those few cases, use
1818C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1819
1820Returns the Unicode (not-native) code point of the first character in the
1821string C<s> which
ec5f19d0
KW
1822is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1823C<retlen> will be set to the length, in bytes, of that character.
1824
524080c4
KW
1825If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1826enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1827NULL) to -1. If those warnings are off, the computed value if well-defined (or
1828the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1829is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1830next possible position in C<s> that could begin a non-malformed character.
de69f3af 1831See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1832
760c7c2f
KW
1833Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1834unless those are turned off.
1835
ec5f19d0
KW
1836=cut
1837*/
1838
1839UV
1840Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1841{
1842 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1843
1844 assert(send > s);
1845
5962d97e
KW
1846 /* Call the low level routine, asking for checks */
1847 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1848}
1849
b76347f2 1850/*
87cea99e 1851=for apidoc utf8_length
b76347f2
JH
1852
1853Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1854Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1855up past C<e>, croaks.
b76347f2
JH
1856
1857=cut
1858*/
1859
1860STRLEN
35a4481c 1861Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1862{
1863 STRLEN len = 0;
1864
7918f24d
NC
1865 PERL_ARGS_ASSERT_UTF8_LENGTH;
1866
8850bf83
JH
1867 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1868 * the bitops (especially ~) can create illegal UTF-8.
1869 * In other words: in Perl UTF-8 is not just for Unicode. */
1870
a3b680e6
AL
1871 if (e < s)
1872 goto warn_and_return;
b76347f2 1873 while (s < e) {
4cbf4130 1874 s += UTF8SKIP(s);
8e91ec7f
AV
1875 len++;
1876 }
1877
1878 if (e != s) {
1879 len--;
1880 warn_and_return:
9b387841
NC
1881 if (PL_op)
1882 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1883 "%s in %s", unees, OP_DESC(PL_op));
1884 else
61a12c31 1885 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1886 }
1887
1888 return len;
1889}
1890
b06226ff 1891/*
fed3ba5d
NC
1892=for apidoc bytes_cmp_utf8
1893
a1433954 1894Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1895sequence of characters (stored as UTF-8)
1896in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1897equal, -1 or -2 if the first string is less than the second string, +1 or +2
1898if the first string is greater than the second string.
1899
1900-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1901longer string. -2 or +2 is returned if
1902there was a difference between characters
fed3ba5d
NC
1903within the strings.
1904
1905=cut
1906*/
1907
1908int
1909Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1910{
1911 const U8 *const bend = b + blen;
1912 const U8 *const uend = u + ulen;
1913
1914 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1915
1916 while (b < bend && u < uend) {
1917 U8 c = *u++;
1918 if (!UTF8_IS_INVARIANT(c)) {
1919 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1920 if (u < uend) {
1921 U8 c1 = *u++;
1922 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1923 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1924 } else {
2b5e7bc2 1925 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1926 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
1927 "%s %s%s",
1928 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1929 PL_op ? " in " : "",
1930 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1931 return -2;
1932 }
1933 } else {
1934 if (PL_op)
1935 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1936 "%s in %s", unees, OP_DESC(PL_op));
1937 else
61a12c31 1938 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1939 return -2; /* Really want to return undef :-) */
1940 }
1941 } else {
1942 return -2;
1943 }
1944 }
1945 if (*b != c) {
1946 return *b < c ? -2 : +2;
1947 }
1948 ++b;
1949 }
1950
1951 if (b == bend && u == uend)
1952 return 0;
1953
1954 return b < bend ? +1 : -1;
1955}
1956
1957/*
87cea99e 1958=for apidoc utf8_to_bytes
6940069f 1959
3bc0c78c 1960Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 1961Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 1962updates C<*lenp> to contain the new length.
3bc0c78c
KW
1963Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
1964
1965Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
1966having saved the value of C<*lenp> before the call, and subtracting the
1967after-call value of C<*lenp> from it.
6940069f 1968
a1433954 1969If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1970
6940069f
GS
1971=cut
1972*/
1973
1974U8 *
09af0336 1975Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 1976{
9fe0d3c2 1977 U8 * first_variant;
246fae53 1978
7918f24d 1979 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1980 PERL_UNUSED_CONTEXT;
7918f24d 1981
9fe0d3c2 1982 /* This is a no-op if no variants at all in the input */
09af0336 1983 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
1984 return s;
1985 }
1986
1987 {
3c5aa262 1988 U8 * const save = s;
09af0336 1989 U8 * const send = s + *lenp;
3c5aa262
KW
1990 U8 * d;
1991
1992 /* Nothing before the first variant needs to be changed, so start the real
1993 * work there */
1994 s = first_variant;
1995 while (s < send) {
1996 if (! UTF8_IS_INVARIANT(*s)) {
1997 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 1998 *lenp = ((STRLEN) -1);
3c5aa262
KW
1999 return 0;
2000 }
2001 s++;
d59937ca
KW
2002 }
2003 s++;
dcad2880 2004 }
dcad2880 2005
3c5aa262
KW
2006 /* Is downgradable, so do it */
2007 d = s = first_variant;
2008 while (s < send) {
2009 U8 c = *s++;
2010 if (! UVCHR_IS_INVARIANT(c)) {
2011 /* Then it is two-byte encoded */
2012 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2013 s++;
2014 }
2015 *d++ = c;
2016 }
2017 *d = '\0';
09af0336 2018 *lenp = d - save;
3c5aa262
KW
2019
2020 return save;
9fe0d3c2 2021 }
6940069f
GS
2022}
2023
2024/*
87cea99e 2025=for apidoc bytes_from_utf8
f9a63242 2026
09af0336 2027Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2028byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2029actually encoded in UTF-8.
2030
2031Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2032the input string.
2033
41ae6089
KW
2034Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2035not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2036C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2037
41ae6089 2038Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2039newly created string containing a downgraded copy of C<s>, and whose length is
23b37b12 2040returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
f9a63242 2041
3bc0c78c 2042Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2043having saved the value of C<*lenp> before the call, and subtracting the
2044after-call value of C<*lenp> from it.
3bc0c78c 2045
37607a96 2046=cut
976c1b08
KW
2047
2048There is a macro that avoids this function call, but this is retained for
2049anyone who calls it with the Perl_ prefix */
f9a63242
JH
2050
2051U8 *
41ae6089 2052Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2053{
7918f24d 2054 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2055 PERL_UNUSED_CONTEXT;
f9a63242 2056
976c1b08
KW
2057 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2058}
2059
2060/*
2061No = here because currently externally undocumented
2062for apidoc bytes_from_utf8_loc
2063
2064Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2065to store the location of the first character in C<"s"> that cannot be
2066converted to non-UTF8.
2067
2068If that parameter is C<NULL>, this function behaves identically to
2069C<bytes_from_utf8>.
2070
2071Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2072C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2073
2074Otherwise, the function returns a newly created C<NUL>-terminated string
2075containing the non-UTF8 equivalent of the convertible first portion of
2076C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2077If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2078and C<*first_non_downgradable> is set to C<NULL>.
2079
2080Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2081first character in the original string that wasn't converted. C<*is_utf8p> is
2082unchanged. Note that the new string may have length 0.
2083
2084Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2085C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2086converts as many characters in it as possible stopping at the first one it
385b74be 2087finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2088set to point to that. The function returns the portion that could be converted
2089in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2090not including the terminating C<NUL>. If the very first character in the
2091original could not be converted, C<*lenp> will be 0, and the new string will
2092contain just a single C<NUL>. If the entire input string was converted,
2093C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2094
2095Upon successful return, the number of variants in the converted portion of the
2096string can be computed by having saved the value of C<*lenp> before the call,
2097and subtracting the after-call value of C<*lenp> from it.
2098
2099=cut
2100
2101
2102*/
2103
2104U8 *
2105Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2106{
2107 U8 *d;
2108 const U8 *original = s;
2109 U8 *converted_start;
2110 const U8 *send = s + *lenp;
f9a63242 2111
976c1b08 2112 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2113
976c1b08
KW
2114 if (! *is_utf8p) {
2115 if (first_unconverted) {
2116 *first_unconverted = NULL;
2117 }
2118
2119 return (U8 *) original;
2120 }
2121
2122 Newx(d, (*lenp) + 1, U8);
2123
2124 converted_start = d;
7299a045
KW
2125 while (s < send) {
2126 U8 c = *s++;
2127 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2128
2129 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2130 * have to stop now */
2131 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2132 if (first_unconverted) {
2133 *first_unconverted = s - 1;
2134 goto finish_and_return;
2135 }
2136 else {
2137 Safefree(converted_start);
2138 return (U8 *) original;
2139 }
2140 }
2141
7299a045
KW
2142 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2143 s++;
38af28cf 2144 }
7299a045
KW
2145 *d++ = c;
2146 }
170a1c22 2147
976c1b08
KW
2148 /* Here, converted the whole of the input */
2149 *is_utf8p = FALSE;
2150 if (first_unconverted) {
2151 *first_unconverted = NULL;
170a1c22 2152 }
976c1b08
KW
2153
2154 finish_and_return:
2155 *d = '\0';
2156 *lenp = d - converted_start;
2157
2158 /* Trim unused space */
2159 Renew(converted_start, *lenp + 1, U8);
2160
2161 return converted_start;
f9a63242
JH
2162}
2163
2164/*
87cea99e 2165=for apidoc bytes_to_utf8
6940069f 2166
09af0336 2167Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2168UTF-8.
09af0336 2169Returns a pointer to the newly-created string, and sets C<*lenp> to
ff97e5cf 2170reflect the new length in bytes.
6940069f 2171
3bc0c78c 2172Upon successful return, the number of variants in the string can be computed by
23b37b12 2173having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2174after-call value of C<*lenp>.
2175
75200dff 2176A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2177
2178If you want to convert to UTF-8 from encodings other than
2179the native (Latin1 or EBCDIC),
a1433954 2180see L</sv_recode_to_utf8>().
c9ada85f 2181
497711e7 2182=cut
6940069f
GS
2183*/
2184
2185U8*
09af0336 2186Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2187{
09af0336 2188 const U8 * const send = s + (*lenp);
6940069f
GS
2189 U8 *d;
2190 U8 *dst;
7918f24d
NC
2191
2192 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2193 PERL_UNUSED_CONTEXT;
6940069f 2194
09af0336 2195 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2196 dst = d;
2197
2198 while (s < send) {
55d09dc8
KW
2199 append_utf8_from_native_byte(*s, &d);
2200 s++;
6940069f
GS
2201 }
2202 *d = '\0';
09af0336 2203 *lenp = d-dst;
6940069f
GS
2204 return dst;
2205}
2206
a0ed51b3 2207/*
dea0fc0b 2208 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2209 *
2210 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2211 * We optimize for native, for obvious reasons. */
2212
2213U8*
dea0fc0b 2214Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2215{
dea0fc0b
JH
2216 U8* pend;
2217 U8* dstart = d;
2218
7918f24d
NC
2219 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2220
dea0fc0b 2221 if (bytelen & 1)
56576a04
KW
2222 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2223 (UV)bytelen);
dea0fc0b
JH
2224
2225 pend = p + bytelen;
2226
a0ed51b3 2227 while (p < pend) {
dea0fc0b
JH
2228 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2229 p += 2;
2d1545e5 2230 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2231 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2232 continue;
2233 }
56d37426
KW
2234 if (uv <= MAX_UTF8_TWO_BYTE) {
2235 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2236 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2237 continue;
2238 }
46956fad
KW
2239#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2240#define LAST_HIGH_SURROGATE 0xDBFF
2241#define FIRST_LOW_SURROGATE 0xDC00
2242#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2243
2244 /* This assumes that most uses will be in the first Unicode plane, not
2245 * needing surrogates */
2246 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2247 && uv <= UNICODE_SURROGATE_LAST))
2248 {
2249 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2250 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2251 }
2252 else {
01ea242b 2253 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2254 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2255 || UNLIKELY(low > LAST_LOW_SURROGATE))
2256 {
01ea242b 2257 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2258 }
2259 p += 2;
46956fad
KW
2260 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2261 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2262 }
a0ed51b3 2263 }
56d37426
KW
2264#ifdef EBCDIC
2265 d = uvoffuni_to_utf8_flags(d, uv, 0);
2266#else
a0ed51b3 2267 if (uv < 0x10000) {
eb160463
GS
2268 *d++ = (U8)(( uv >> 12) | 0xe0);
2269 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2270 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2271 continue;
2272 }
2273 else {
eb160463
GS
2274 *d++ = (U8)(( uv >> 18) | 0xf0);
2275 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2276 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2277 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2278 continue;
2279 }
56d37426 2280#endif
a0ed51b3 2281 }
dea0fc0b 2282 *newlen = d - dstart;
a0ed51b3
LW
2283 return d;
2284}
2285
2286/* Note: this one is slightly destructive of the source. */
2287
2288U8*
dea0fc0b 2289Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2290{
2291 U8* s = (U8*)p;
d4c19fe8 2292 U8* const send = s + bytelen;
7918f24d
NC
2293
2294 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2295
e0ea5e2d 2296 if (bytelen & 1)
147e3846 2297 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2298 (UV)bytelen);
2299
a0ed51b3 2300 while (s < send) {
d4c19fe8 2301 const U8 tmp = s[0];
a0ed51b3
LW
2302 s[0] = s[1];
2303 s[1] = tmp;
2304 s += 2;
2305 }
dea0fc0b 2306 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2307}
2308
922e8cb4
KW
2309bool
2310Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2311{
2312 U8 tmpbuf[UTF8_MAXBYTES+1];
2313 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2314 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2315}
2316
f9ae8fb6
JD
2317/* Internal function so we can deprecate the external one, and call
2318 this one from other deprecated functions in this file */
2319
f2645549
KW
2320bool
2321Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2322{
f2645549 2323 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2324
2325 if (*p == '_')
2326 return TRUE;
f25ce844 2327 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2328}
2329
5092f92a 2330bool
eba68aa0
KW
2331Perl__is_uni_perl_idcont(pTHX_ UV c)
2332{
2333 U8 tmpbuf[UTF8_MAXBYTES+1];
2334 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2335 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2336}
2337
2338bool
f91dcd13
KW
2339Perl__is_uni_perl_idstart(pTHX_ UV c)
2340{
2341 U8 tmpbuf[UTF8_MAXBYTES+1];
2342 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2343 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2344}
2345
3a4c58c9 2346UV
56576a04
KW
2347Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2348 const char S_or_s)
3a4c58c9
KW
2349{
2350 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2351 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2352 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2353 * either "SS" or "Ss". Which one to use is passed into the routine in
2354 * 'S_or_s' to avoid a test */
2355
2356 UV converted = toUPPER_LATIN1_MOD(c);
2357
2358 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2359
2360 assert(S_or_s == 'S' || S_or_s == 's');
2361
6f2d5cbc 2362 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2363 characters in this range */
3a4c58c9
KW
2364 *p = (U8) converted;
2365 *lenp = 1;
2366 return converted;
2367 }
2368
2369 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2370 * which it maps to one of them, so as to only have to have one check for
2371 * it in the main case */
2372 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2373 switch (c) {
2374 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2375 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2376 break;
2377 case MICRO_SIGN:
2378 converted = GREEK_CAPITAL_LETTER_MU;
2379 break;
79e064b9
KW
2380#if UNICODE_MAJOR_VERSION > 2 \
2381 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2382 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2383 case LATIN_SMALL_LETTER_SHARP_S:
2384 *(p)++ = 'S';
2385 *p = S_or_s;
2386 *lenp = 2;
2387 return 'S';
79e064b9 2388#endif
3a4c58c9 2389 default:
56576a04
KW
2390 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2391 " '%c' to map to '%c'",
2392 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2393 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2394 }
2395 }
2396
2397 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2398 *p = UTF8_TWO_BYTE_LO(converted);
2399 *lenp = 2;
2400
2401 return converted;
2402}
2403
50bda2c3
KW
2404/* Call the function to convert a UTF-8 encoded character to the specified case.
2405 * Note that there may be more than one character in the result.
2406 * INP is a pointer to the first byte of the input character
2407 * OUTP will be set to the first byte of the string of changed characters. It
2408 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2409 * LENP will be set to the length in bytes of the string of changed characters
2410 *
56576a04
KW
2411 * The functions return the ordinal of the first character in the string of
2412 * OUTP */
2413#define CALL_UPPER_CASE(uv, s, d, lenp) \
2414 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2415#define CALL_TITLE_CASE(uv, s, d, lenp) \
2416 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2417#define CALL_LOWER_CASE(uv, s, d, lenp) \
2418 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2419
b9992569
KW
2420/* This additionally has the input parameter 'specials', which if non-zero will
2421 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2422 * folding); otherwise, when zero, this implies a simple case fold */
56576a04
KW
2423#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2424_to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2425
84afefe6
JH
2426UV
2427Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2428{
a1433954
KW
2429 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2430 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2431 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2432 * the changed version may be longer than the original character.
2433 *
2434 * The ordinal of the first character of the changed version is returned
2435 * (but note, as explained above, that there may be more.) */
2436
7918f24d
NC
2437 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2438
3a4c58c9
KW
2439 if (c < 256) {
2440 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2441 }
2442
0ebc6274 2443 uvchr_to_utf8(p, c);
b9992569 2444 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2445}
2446
84afefe6
JH
2447UV
2448Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2449{
7918f24d
NC
2450 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2451
3a4c58c9
KW
2452 if (c < 256) {
2453 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2454 }
2455
0ebc6274 2456 uvchr_to_utf8(p, c);
b9992569 2457 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2458}
2459
afc16117 2460STATIC U8
eaf412bf 2461S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2462{
2463 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2464 * those, converting the result to UTF-8. Since the result is always just
a1433954 2465 * one character, we allow <p> to be NULL */
afc16117
KW
2466
2467 U8 converted = toLOWER_LATIN1(c);
2468
eaf412bf
KW
2469 PERL_UNUSED_ARG(dummy);
2470
afc16117 2471 if (p != NULL) {
6f2d5cbc 2472 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2473 *p = converted;
2474 *lenp = 1;
2475 }
2476 else {
430c9760
KW
2477 /* Result is known to always be < 256, so can use the EIGHT_BIT
2478 * macros */
2479 *p = UTF8_EIGHT_BIT_HI(converted);
2480 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2481 *lenp = 2;
2482 }
2483 }
2484 return converted;
2485}
2486
84afefe6
JH
2487UV
2488Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2489{
7918f24d
NC
2490 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2491
afc16117 2492 if (c < 256) {
eaf412bf 2493 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2494 }
2495
afc16117 2496 uvchr_to_utf8(p, c);
b9992569 2497 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2498}
2499
84afefe6 2500UV
56576a04
KW
2501Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2502 const unsigned int flags)
a1dde8de 2503{
51910141 2504 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2505 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2506 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2507 *
2508 * Not to be used for locale folds
51910141 2509 */
f673fad4 2510
a1dde8de
KW
2511 UV converted;
2512
2513 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2514 PERL_UNUSED_CONTEXT;
a1dde8de 2515
1ca267a5
KW
2516 assert (! (flags & FOLD_FLAGS_LOCALE));
2517
659a7c2d 2518 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2519 converted = GREEK_SMALL_LETTER_MU;
2520 }
9b63e895
KW
2521#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2522 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2523 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2524 else if ( (flags & FOLD_FLAGS_FULL)
2525 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2526 {
1ca267a5
KW
2527 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2528 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2529 * under those circumstances. */
2530 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2531 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2532 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2533 p, *lenp, U8);
2534 return LATIN_SMALL_LETTER_LONG_S;
2535 }
2536 else {
4f489194
KW
2537 *(p)++ = 's';
2538 *p = 's';
2539 *lenp = 2;
2540 return 's';
1ca267a5 2541 }
a1dde8de 2542 }
9b63e895 2543#endif
a1dde8de
KW
2544 else { /* In this range the fold of all other characters is their lower
2545 case */
2546 converted = toLOWER_LATIN1(c);
2547 }
2548
6f2d5cbc 2549 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2550 *p = (U8) converted;
2551 *lenp = 1;
2552 }
2553 else {
2554 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2555 *p = UTF8_TWO_BYTE_LO(converted);
2556 *lenp = 2;
2557 }
2558
2559 return converted;
2560}
2561
2562UV
31f05a37 2563Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2564{
4b593389 2565
a0270393
KW
2566 /* Not currently externally documented, and subject to change
2567 * <flags> bits meanings:
2568 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2569 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2570 * locale are to be used.
a0270393
KW
2571 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2572 */
4b593389 2573
36bb2ab6 2574 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2575
780fcc9f
KW
2576 if (flags & FOLD_FLAGS_LOCALE) {
2577 /* Treat a UTF-8 locale as not being in locale at all */
2578 if (IN_UTF8_CTYPE_LOCALE) {
2579 flags &= ~FOLD_FLAGS_LOCALE;
2580 }
2581 else {
2582 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2583 goto needs_full_generality;
780fcc9f 2584 }
31f05a37
KW
2585 }
2586
a1dde8de 2587 if (c < 256) {
e7b7ac46 2588 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2589 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2590 }
2591
2f306ab9 2592 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2593 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2594 uvchr_to_utf8(p, c);
b9992569 2595 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2596 }
567b353c 2597 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2598 the special flags. */
2599 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2600
2601 needs_full_generality:
a0270393 2602 uvchr_to_utf8(utf8_c, c);
56576a04
KW
2603 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2604 p, lenp, flags);
a0270393 2605 }
84afefe6
JH
2606}
2607
26483009 2608PERL_STATIC_INLINE bool
5141f98e 2609S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2610 const char *const swashname, SV* const invlist)
bde6a22d 2611{
ea317ccb
KW
2612 /* returns a boolean giving whether or not the UTF8-encoded character that
2613 * starts at <p> is in the swash indicated by <swashname>. <swash>
2614 * contains a pointer to where the swash indicated by <swashname>
2615 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2616 * look at <*swash> and only generate a swash if it is not null. <invlist>
2617 * is NULL or an inversion list that defines the swash. If not null, it
2618 * saves time during initialization of the swash.
ea317ccb
KW
2619 *
2620 * Note that it is assumed that the buffer length of <p> is enough to
2621 * contain all the bytes that comprise the character. Thus, <*p> should
2622 * have been checked before this call for mal-formedness enough to assure
2623 * that. */
2624
7918f24d
NC
2625 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2626
492a624f 2627 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2628 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2629 * as far as there being enough bytes available in it to accommodate the
2630 * character without reading beyond the end, and pass that number on to the
2631 * validating routine */
6302f837 2632 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2633 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2634 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2635 1 /* Die */ );
2636 NOT_REACHED; /* NOTREACHED */
28123549 2637 }
86ae6e94 2638
87367d5f
KW
2639 if (!*swash) {
2640 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2641 *swash = _core_swash_init("utf8",
2642
2643 /* Only use the name if there is no inversion
2644 * list; otherwise will go out to disk */
2645 (invlist) ? "" : swashname,
2646
2647 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2648 }
28123549 2649
bde6a22d
NC
2650 return swash_fetch(*swash, p, TRUE) != 0;
2651}
2652
da8c1a98 2653PERL_STATIC_INLINE bool
56576a04
KW
2654S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2655 SV **swash, const char *const swashname,
2656 SV* const invlist)
da8c1a98
KW
2657{
2658 /* returns a boolean giving whether or not the UTF8-encoded character that
2659 * starts at <p>, and extending no further than <e - 1> is in the swash
2660 * indicated by <swashname>. <swash> contains a pointer to where the swash
2661 * indicated by <swashname> is to be stored; which this routine will do, so
2662 * that future calls will look at <*swash> and only generate a swash if it
2663 * is not null. <invlist> is NULL or an inversion list that defines the
2664 * swash. If not null, it saves time during initialization of the swash.
2665 */
2666
2667 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2668
2669 if (! isUTF8_CHAR(p, e)) {
2670 _force_out_malformed_utf8_message(p, e, 0, 1);
2671 NOT_REACHED; /* NOTREACHED */
2672 }
2673
2674 if (!*swash) {
2675 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2676 *swash = _core_swash_init("utf8",
2677
2678 /* Only use the name if there is no inversion
2679 * list; otherwise will go out to disk */
2680 (invlist) ? "" : swashname,
2681
2682 &PL_sv_undef, 1, 0, invlist, &flags);
2683 }
2684
2685 return swash_fetch(*swash, p, TRUE) != 0;
2686}
2687
34aeb2e9
KW
2688STATIC void
2689S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2690 const char * const alternative,
2691 const bool use_locale,
2692 const char * const file,
2693 const unsigned line)
2694{
2695 const char * key;
2696
2697 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2698
2699 if (ckWARN_d(WARN_DEPRECATED)) {
2700
2701 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2702 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2703 if (! PL_seen_deprecated_macro) {
2704 PL_seen_deprecated_macro = newHV();
2705 }
2706 if (! hv_store(PL_seen_deprecated_macro, key,
2707 strlen(key), &PL_sv_undef, 0))
2708 {
2709 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2710 }
2711
c44e9413 2712 if (instr(file, "mathoms.c")) {
607313a1
KW
2713 Perl_warner(aTHX_ WARN_DEPRECATED,
2714 "In %s, line %d, starting in Perl v5.30, %s()"
2715 " will be removed. Avoid this message by"
2716 " converting to use %s().\n",
2717 file, line, name, alternative);
2718 }
2719 else {
34aeb2e9
KW
2720 Perl_warner(aTHX_ WARN_DEPRECATED,
2721 "In %s, line %d, starting in Perl v5.30, %s() will"
2722 " require an additional parameter. Avoid this"
2723 " message by converting to use %s().\n",
2724 file, line, name, alternative);
607313a1 2725 }
34aeb2e9
KW
2726 }
2727 }
2728}
2729
bde6a22d 2730bool
34aeb2e9 2731Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 2732 const U8 * const p,
34aeb2e9
KW
2733 const char * const name,
2734 const char * const alternative,
2735 const bool use_utf8,
2736 const bool use_locale,
2737 const char * const file,
2738 const unsigned line)
922e8cb4 2739{
922e8cb4
KW
2740 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2741
34aeb2e9
KW
2742 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2743
2744 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2745
2746 switch (classnum) {
2747 case _CC_WORDCHAR:
2748 case _CC_DIGIT:
2749 case _CC_ALPHA:
2750 case _CC_LOWER:
2751 case _CC_UPPER:
2752 case _CC_PUNCT:
2753 case _CC_PRINT:
2754 case _CC_ALPHANUMERIC:
2755 case _CC_GRAPH:
2756 case _CC_CASED:
2757
2758 return is_utf8_common(p,
2759 &PL_utf8_swash_ptrs[classnum],
2760 swash_property_names[classnum],
2761 PL_XPosix_ptrs[classnum]);
2762
2763 case _CC_SPACE:
2764 return is_XPERLSPACE_high(p);
2765 case _CC_BLANK:
2766 return is_HORIZWS_high(p);
2767 case _CC_XDIGIT:
2768 return is_XDIGIT_high(p);
2769 case _CC_CNTRL:
2770 return 0;
2771 case _CC_ASCII:
2772 return 0;
2773 case _CC_VERTSPACE:
2774 return is_VERTWS_high(p);
2775 case _CC_IDFIRST:
2776 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2777 PL_utf8_perl_idstart
2778 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2779 }
22f0498f
KW
2780 return is_utf8_common(p, &PL_utf8_perl_idstart,
2781 "_Perl_IDStart", NULL);
34aeb2e9
KW
2782 case _CC_IDCONT:
2783 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2784 PL_utf8_perl_idcont
2785 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2786 }
22f0498f
KW
2787 return is_utf8_common(p, &PL_utf8_perl_idcont,
2788 "_Perl_IDCont", NULL);
34aeb2e9
KW
2789 }
2790 }
2791
2792 /* idcont is the same as wordchar below 256 */
2793 if (classnum == _CC_IDCONT) {
2794 classnum = _CC_WORDCHAR;
2795 }
2796 else if (classnum == _CC_IDFIRST) {
2797 if (*p == '_') {
2798 return TRUE;
2799 }
2800 classnum = _CC_ALPHA;
2801 }
2802
2803 if (! use_locale) {
2804 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2805 return _generic_isCC(*p, classnum);
2806 }
922e8cb4 2807
34aeb2e9
KW
2808 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2809 }
2810 else {
2811 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2812 return isFOO_lc(classnum, *p);
2813 }
2814
2815 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2816 }
2817
2818 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2819}
2820
2821bool
da8c1a98
KW
2822Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2823 const U8 * const e)
2824{
2825 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2826
2827 assert(classnum < _FIRST_NON_SWASH_CC);
2828
2829 return is_utf8_common_with_len(p,
2830 e,
2831 &PL_utf8_swash_ptrs[classnum],
2832 swash_property_names[classnum],
2833 PL_XPosix_ptrs[classnum]);
2834}
2835
2836bool
da8c1a98
KW
2837Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2838{
2839 SV* invlist = NULL;
2840
2841 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2842
2843 if (! PL_utf8_perl_idstart) {
2844 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2845 }
2846 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2847 "_Perl_IDStart", invlist);
2848}
2849
2850bool
f2645549 2851Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2852{
f2645549 2853 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2854
2855 if (*p == '_')
2856 return TRUE;
f25ce844 2857 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2858}
2859
2860bool
da8c1a98
KW
2861Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2862{
2863 SV* invlist = NULL;
2864
2865 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2866
2867 if (! PL_utf8_perl_idcont) {
2868 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2869 }
2870 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2871 "_Perl_IDCont", invlist);
2872}
2873
2874bool
f2645549 2875Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2876{
f2645549 2877 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2878
f25ce844 2879 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2880}
2881
2882bool
f2645549 2883Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2884{
f2645549 2885 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2886
f25ce844 2887 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2888}
2889
2890bool
7dbf68d2
KW
2891Perl__is_utf8_mark(pTHX_ const U8 *p)
2892{
7dbf68d2
KW
2893 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2894
f25ce844 2895 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2896}
2897
b9992569 2898 /* change namve uv1 to 'from' */
6a4a25f4 2899STATIC UV
b9992569
KW
2900S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2901 SV **swashp, const char *normal, const char *special)
2902{
0134edef 2903 STRLEN len = 0;
7918f24d 2904
b9992569 2905 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2906
36eaa811
KW
2907 /* For code points that don't change case, we already know that the output
2908 * of this function is the unchanged input, so we can skip doing look-ups
2909 * for them. Unfortunately the case-changing code points are scattered
2910 * around. But there are some long consecutive ranges where there are no
2911 * case changing code points. By adding tests, we can eliminate the lookup
2912 * for all the ones in such ranges. This is currently done here only for
2913 * just a few cases where the scripts are in common use in modern commerce
2914 * (and scripts adjacent to those which can be included without additional
2915 * tests). */
2916
2917 if (uv1 >= 0x0590) {
2918 /* This keeps from needing further processing the code points most
2919 * likely to be used in the following non-cased scripts: Hebrew,
2920 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2921 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2922 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2923 if (uv1 < 0x10A0) {
2924 goto cases_to_self;
2925 }
2926
2927 /* The following largish code point ranges also don't have case
2928 * changes, but khw didn't think they warranted extra tests to speed
2929 * them up (which would slightly slow down everything else above them):
2930 * 1100..139F Hangul Jamo, Ethiopic
2931 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2932 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2933 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2934 * Combining Diacritical Marks Extended, Balinese,
2935 * Sundanese, Batak, Lepcha, Ol Chiki
2936 * 2000..206F General Punctuation
2937 */
2938
2939 if (uv1 >= 0x2D30) {
2940
2941 /* This keeps the from needing further processing the code points
2942 * most likely to be used in the following non-cased major scripts:
2943 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2944 *
2945 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2946 * event that Unicode eventually allocates the unused block as of
2947 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2948 * that the test suite will start having failures to alert you
2949 * should that happen) */
2950 if (uv1 < 0xA640) {
2951 goto cases_to_self;
2952 }
2953
2954 if (uv1 >= 0xAC00) {
2955 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2956 if (ckWARN_d(WARN_SURROGATE)) {
2957 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2958 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
2959 "Operation \"%s\" returns its argument for"
2960 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2961 }
2962 goto cases_to_self;
2963 }
36eaa811
KW
2964
2965 /* AC00..FAFF Catches Hangul syllables and private use, plus
2966 * some others */
2967 if (uv1 < 0xFB00) {
2968 goto cases_to_self;
2969
2970 }
2971
5af9bc97 2972 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
76513bdc
KW
2973 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2974 && ckWARN_d(WARN_DEPRECATED))
2975 {
2976 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2977 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
5af9bc97
KW
2978 }
2979 if (ckWARN_d(WARN_NON_UNICODE)) {
2980 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2981 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
2982 "Operation \"%s\" returns its argument for"
2983 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2984 }
2985 goto cases_to_self;
2986 }
3bfc1e70
KW
2987#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2988 if (UNLIKELY(uv1
2989 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2990 {
2991
56576a04
KW
2992 /* As of Unicode 10.0, this means we avoid swash creation
2993 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
2994 goto cases_to_self;
2995 }
2996#endif
36eaa811
KW
2997 }
2998 }
9ae3ac1a 2999
36eaa811
KW
3000 /* Note that non-characters are perfectly legal, so no warning should
3001 * be given. There are so few of them, that it isn't worth the extra
3002 * tests to avoid swash creation */
9ae3ac1a
KW
3003 }
3004
0134edef 3005 if (!*swashp) /* load on-demand */
56576a04
KW
3006 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3007 4, 0, NULL, NULL);
0134edef 3008
a6f87d8c 3009 if (special) {
0134edef 3010 /* It might be "special" (sometimes, but not always,
2a37f04d 3011 * a multicharacter mapping) */
4a8240a3 3012 HV *hv = NULL;
b08cf34e
JH
3013 SV **svp;
3014
4a8240a3
KW
3015 /* If passed in the specials name, use that; otherwise use any
3016 * given in the swash */
3017 if (*special != '\0') {
3018 hv = get_hv(special, 0);
3019 }
3020 else {
3021 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3022 if (svp) {
3023 hv = MUTABLE_HV(SvRV(*svp));
3024 }
3025 }
3026
176fe009 3027 if (hv
5f560d8a 3028 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
3029 && (*svp))
3030 {
cfd0369c 3031 const char *s;
47654450 3032
cfd0369c 3033 s = SvPV_const(*svp, len);
47654450 3034 if (len == 1)
f4cd282c 3035 /* EIGHTBIT */
c80e42f3 3036 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 3037 else {
d2dcd0fb 3038 Copy(s, ustrp, len, U8);
29e98929 3039 }
983ffd37 3040 }
0134edef
JH
3041 }
3042
3043 if (!len && *swashp) {
4a4088c4 3044 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 3045
0134edef
JH
3046 if (uv2) {
3047 /* It was "normal" (a single character mapping). */
f4cd282c 3048 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
3049 }
3050 }
1feea2c7 3051
cbe07460
KW
3052 if (len) {
3053 if (lenp) {
3054 *lenp = len;
3055 }
3056 return valid_utf8_to_uvchr(ustrp, 0);
3057 }
3058
3059 /* Here, there was no mapping defined, which means that the code point maps
3060 * to itself. Return the inputs */
e24dfe9c 3061 cases_to_self:
bfdf22ec 3062 len = UTF8SKIP(p);
ca9fab46
KW
3063 if (p != ustrp) { /* Don't copy onto itself */
3064 Copy(p, ustrp, len, U8);
3065 }
0134edef 3066
2a37f04d
JH
3067 if (lenp)
3068 *lenp = len;
3069
f4cd282c 3070 return uv1;
cbe07460 3071
a0ed51b3
LW
3072}
3073
051a06d4 3074STATIC UV
56576a04
KW
3075S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3076 U8* const ustrp, STRLEN *lenp)
051a06d4 3077{
4a4088c4 3078 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3079 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3080 * result contains a character that crosses the 255/256 boundary, disallow
3081 * the change, and return the original code point. See L<perlfunc/lc> for
3082 * why;
051a06d4 3083 *
a1433954
KW
3084 * p points to the original string whose case was changed; assumed
3085 * by this routine to be well-formed
051a06d4 3086 * result the code point of the first character in the changed-case string
56576a04
KW
3087 * ustrp points to the changed-case string (<result> represents its
3088 * first char)
051a06d4
KW
3089 * lenp points to the length of <ustrp> */
3090
3091 UV original; /* To store the first code point of <p> */
3092
3093 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3094
a4f12ed7 3095 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3096
3097 /* We know immediately if the first character in the string crosses the
3098 * boundary, so can skip */
3099 if (result > 255) {
3100
3101 /* Look at every character in the result; if any cross the
3102 * boundary, the whole thing is disallowed */
3103 U8* s = ustrp + UTF8SKIP(ustrp);
3104 U8* e = ustrp + *lenp;
3105 while (s < e) {
a4f12ed7 3106 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3107 goto bad_crossing;
3108 }
3109 s += UTF8SKIP(s);
3110 }
3111
613abc6d
KW
3112 /* Here, no characters crossed, result is ok as-is, but we warn. */
3113 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3114 return result;
3115 }
3116
7b52d656 3117 bad_crossing:
051a06d4
KW
3118
3119 /* Failed, have to return the original */
4b88fb76 3120 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3121
3122 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3123 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3124 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3125 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3126 OP_DESC(PL_op),
ab0b796c
KW
3127 original,
3128 original);
051a06d4
KW
3129 Copy(p, ustrp, *lenp, char);
3130 return original;
3131}
3132
607313a1
KW
3133STATIC U32
3134S_check_and_deprecate(pTHX_ const U8 *p,
3135 const U8 **e,
3136 const unsigned int type, /* See below */
3137 const bool use_locale, /* Is this a 'LC_'
3138 macro call? */
3139 const char * const file,
3140 const unsigned line)
3141{
3142 /* This is a temporary function to deprecate the unsafe calls to the case
3143 * changing macros and functions. It keeps all the special stuff in just
3144 * one place.
3145 *
3146 * It updates *e with the pointer to the end of the input string. If using
3147 * the old-style macros, *e is NULL on input, and so this function assumes
3148 * the input string is long enough to hold the entire UTF-8 sequence, and
3149 * sets *e accordingly, but it then returns a flag to pass the
3150 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3151 * using the full length if possible.
3152 *
3153 * It also does the assert that *e > p when *e is not NULL. This should be
3154 * migrated to the callers when this function gets deleted.
3155 *
3156 * The 'type' parameter is used for the caller to specify which case
3157 * changing function this is called from: */
3158
3159# define DEPRECATE_TO_UPPER 0
3160# define DEPRECATE_TO_TITLE 1
3161# define DEPRECATE_TO_LOWER 2
3162# define DEPRECATE_TO_FOLD 3
3163
3164 U32 utf8n_flags = 0;
3165 const char * name;
3166 const char * alternative;
3167
3168 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3169
3170 if (*e == NULL) {
3171 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3172 *e = p + UTF8SKIP(p);
3173
3174 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3175 * there. It could be part of a larger path */
607313a1 3176 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3177 name = instr(file, "mathoms.c")
607313a1
KW
3178 ? "to_utf8_upper"
3179 : "toUPPER_utf8";
3180 alternative = "toUPPER_utf8_safe";
3181 }
3182 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3183 name = instr(file, "mathoms.c")
607313a1
KW
3184 ? "to_utf8_title"
3185 : "toTITLE_utf8";
3186 alternative = "toTITLE_utf8_safe";
3187 }
3188 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3189 name = instr(file, "mathoms.c")
607313a1
KW
3190 ? "to_utf8_lower"
3191 : "toLOWER_utf8";
3192 alternative = "toLOWER_utf8_safe";
3193 }
3194 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3195 name = instr(file, "mathoms.c")
607313a1
KW
3196 ? "to_utf8_fold"
3197 : "toFOLD_utf8";
3198 alternative = "toFOLD_utf8_safe";
3199 }
3200 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3201
3202 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3203 }
3204 else {
3205 assert (p < *e);
3206 }
3207
3208 return utf8n_flags;
3209}
3210
eaf412bf
KW
3211/* The process for changing the case is essentially the same for the four case
3212 * change types, except there are complications for folding. Otherwise the
3213 * difference is only which case to change to. To make sure that they all do
3214 * the same thing, the bodies of the functions are extracted out into the
3215 * following two macros. The functions are written with the same variable
3216 * names, and these are known and used inside these macros. It would be
3217 * better, of course, to have inline functions to do it, but since different
3218 * macros are called, depending on which case is being changed to, this is not
3219 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3220 * function can start with the common start macro, then finish with its special
3221 * handling; while the other three cases can just use the common end macro.
3222 *
3223 * The algorithm is to use the proper (passed in) macro or function to change
3224 * the case for code points that are below 256. The macro is used if using
3225 * locale rules for the case change; the function if not. If the code point is
3226 * above 255, it is computed from the input UTF-8, and another macro is called
3227 * to do the conversion. If necessary, the output is converted to UTF-8. If
3228 * using a locale, we have to check that the change did not cross the 255/256
3229 * boundary, see check_locale_boundary_crossing() for further details.
3230 *
3231 * The macros are split with the correct case change for the below-256 case
3232 * stored into 'result', and in the middle of an else clause for the above-255
3233 * case. At that point in the 'else', 'result' is not the final result, but is
3234 * the input code point calculated from the UTF-8. The fold code needs to
3235 * realize all this and take it from there.
3236 *
3237 * If you read the two macros as sequential, it's easier to understand what's
3238 * going on. */
3239#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3240 L1_func_extra_param) \
a239b1e2 3241 \
eaf412bf
KW
3242 if (flags & (locale_flags)) { \
3243 /* Treat a UTF-8 locale as not being in locale at all */ \
3244 if (IN_UTF8_CTYPE_LOCALE) { \
3245 flags &= ~(locale_flags); \
3246 } \
3247 else { \
3248 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3249 } \
3250 } \
3251 \
3252 if (UTF8_IS_INVARIANT(*p)) { \
3253 if (flags & (locale_flags)) { \
3254 result = LC_L1_change_macro(*p); \
3255 } \
3256 else { \
3257 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3258 } \
3259 } \
a239b1e2 3260 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3261 if (flags & (locale_flags)) { \
3262 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3263 *(p+1))); \
3264 } \
3265 else { \
3266 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3267 ustrp, lenp, L1_func_extra_param); \
3268 } \
3269 } \
fa8ab374
KW
3270 else { /* malformed UTF-8 or ord above 255 */ \
3271 STRLEN len_result; \
fa8ab374
KW
3272 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3273 if (len_result == (STRLEN) -1) { \
607313a1
KW
3274 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3275 1 /* Die */ ); \
fa8ab374 3276 }
eaf412bf
KW
3277
3278#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3279 result = change_macro(result, p, ustrp, lenp); \
3280 \
3281 if (flags & (locale_flags)) { \
3282 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3283 } \
3284 return result; \
3285 } \
3286 \
3287 /* Here, used locale rules. Convert back to UTF-8 */ \
3288 if (UTF8_IS_INVARIANT(result)) { \
3289 *ustrp = (U8) result; \
3290 *lenp = 1; \
3291 } \
3292 else { \
3293 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3294 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3295 *lenp = 2; \
3296 } \
3297 \
3298 return result;
3299
d3e79532 3300/*
87cea99e 3301=for apidoc to_utf8_upper
d3e79532 3302
a239b1e2 3303Instead use L</toUPPER_utf8_safe>.
a1433954 3304
d3e79532
JH
3305=cut */
3306
051a06d4 3307/* Not currently externally documented, and subject to change:
31f05a37
KW
3308 * <flags> is set iff iff the rules from the current underlying locale are to
3309 * be used. */
051a06d4 3310
2104c8d9 3311UV
607313a1
KW
3312Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3313 const U8 *e,
3314 U8* ustrp,
3315 STRLEN *lenp,
3316 bool flags,
3317 const char * const file,
3318 const int line)
a0ed51b3 3319{
051a06d4 3320 UV result;
607313a1
KW
3321 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3322 cBOOL(flags), file, line);
051a06d4
KW
3323
3324 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3325
eaf412bf
KW
3326 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3327 /* 2nd char of uc(U+DF) is 'S' */
3328 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3329 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3330}
a0ed51b3 3331
d3e79532 3332/*
87cea99e 3333=for apidoc to_utf8_title
d3e79532 3334
a239b1e2 3335Instead use L</toTITLE_utf8_safe>.
a1433954 3336
d3e79532
JH
3337=cut */
3338
051a06d4 3339/* Not currently externally documented, and subject to change:
31f05a37
KW
3340 * <flags> is set iff the rules from the current underlying locale are to be
3341 * used. Since titlecase is not defined in POSIX, for other than a
3342 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3343 */
051a06d4 3344
983ffd37 3345UV
607313a1
KW
3346Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3347 const U8 *e,
3348 U8* ustrp,
3349 STRLEN *lenp,
3350 bool flags,
3351 const char * const file,
3352 const int line)
983ffd37 3353{
051a06d4 3354 UV result;
607313a1
KW
3355 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3356 cBOOL(flags), file, line);
051a06d4
KW
3357
3358 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3359
eaf412bf
KW
3360 /* 2nd char of ucfirst(U+DF) is 's' */
3361 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3362 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3363}
3364
d3e79532 3365/*
87cea99e 3366=for apidoc to_utf8_lower
d3e79532 3367
a239b1e2 3368Instead use L</toLOWER_utf8_safe>.
a1433954 3369
d3e79532
JH
3370=cut */
3371
051a06d4 3372/* Not currently externally documented, and subject to change:
31f05a37
KW
3373 * <flags> is set iff iff the rules from the current underlying locale are to
3374 * be used.
3375 */
051a06d4 3376
2104c8d9 3377UV
607313a1
KW
3378Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3379 const U8 *e,
3380 U8* ustrp,
3381 STRLEN *lenp,
3382 bool flags,
3383 const char * const file,
3384 const int line)
a0ed51b3 3385{
051a06d4 3386 UV result;
607313a1
KW
3387 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3388 cBOOL(flags), file, line);
051a06d4 3389
051a06d4 3390 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3391
eaf412bf
KW
3392 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3393 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3394}
3395
d3e79532 3396/*
87cea99e 3397=for apidoc to_utf8_fold
d3e79532 3398
a239b1e2 3399Instead use L</toFOLD_utf8_safe>.
a1433954 3400
d3e79532
JH
3401=cut */
3402
051a06d4
KW
3403/* Not currently externally documented, and subject to change,
3404 * in <flags>
31f05a37
KW
3405 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3406 * locale are to be used.
051a06d4
KW
3407 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3408 * otherwise simple folds
a0270393
KW
3409 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3410 * prohibited
445bf929 3411 */
36bb2ab6 3412
b4e400f9 3413UV
607313a1
KW
3414Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3415 const U8 *e,
3416 U8* ustrp,
3417 STRLEN *lenp,
3418 U8 flags,
3419 const char * const file,
3420 const int line)
b4e400f9 3421{
051a06d4 3422 UV result;
607313a1
KW
3423 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3424 cBOOL(flags), file, line);
051a06d4 3425
36bb2ab6 3426 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3427
a0270393
KW
3428 /* These are mutually exclusive */
3429 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3430
50ba90ff
KW
3431 assert(p != ustrp); /* Otherwise overwrites */
3432
eaf412bf
KW
3433 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3434 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3435
eaf412bf 3436 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3437
1ca267a5
KW
3438 if (flags & FOLD_FLAGS_LOCALE) {
3439
76f2ffcd 3440# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3441 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3442
3443# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3444# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3445
3446 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3447
538e84ed
KW
3448 /* Special case these two characters, as what normally gets
3449 * returned under locale doesn't work */
76f2ffcd
KW
3450 if (UTF8SKIP(p) == cap_sharp_s_len
3451 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3452 {
ab0b796c
KW
3453 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3454 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3455 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3456 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3457 goto return_long_s;
3458 }
0766489e
KW
3459 else
3460#endif
3461 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3462 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3463 {
ab0b796c
KW
3464 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3465 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3466 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3467 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3468 goto return_ligature_st;
3469 }
74894415
KW
3470
3471#if UNICODE_MAJOR_VERSION == 3 \
3472 && UNICODE_DOT_VERSION == 0 \
3473 && UNICODE_DOT_DOT_VERSION == 1
3474# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3475
3476 /* And special case this on this Unicode version only, for the same
3477 * reaons the other two are special cased. They would cross the
3478 * 255/256 boundary which is forbidden under /l, and so the code
3479 * wouldn't catch that they are equivalent (which they are only in
3480 * this release) */
3481 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3482 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3483 {
3484 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3485 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3486 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3487 "resolved to \"\\x{0131}\".");
3488 goto return_dotless_i;
3489 }
3490#endif
3491
357aadde 3492 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3493 }
a0270393
KW
3494 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3495 return result;
3496 }
3497 else {
4a4088c4 3498 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3499 * character above the ASCII range, and the result should not
3500 * contain an ASCII character. */
a0270393
KW
3501
3502 UV original; /* To store the first code point of <p> */
3503
3504 /* Look at every character in the result; if any cross the
3505 * boundary, the whole thing is disallowed */
3506 U8* s = ustrp;
3507 U8* e = ustrp + *lenp;
3508 while (s < e) {
3509 if (isASCII(*s)) {
3510 /* Crossed, have to return the original */
3511 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3512
9fc2026f 3513 /* But in these instances, there is an alternative we can
1ca267a5 3514 * return that is valid */
0766489e
KW
3515 if (original == LATIN_SMALL_LETTER_SHARP_S
3516#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3517 || original == LATIN_CAPITAL_LETTER_SHARP_S
3518#endif
3519 ) {
1ca267a5
KW
3520 goto return_long_s;
3521 }
9fc2026f
KW
3522 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3523 goto return_ligature_st;
3524 }
74894415
KW
3525#if UNICODE_MAJOR_VERSION == 3 \
3526 && UNICODE_DOT_VERSION == 0 \
3527 && UNICODE_DOT_DOT_VERSION == 1
3528
3529 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3530 goto return_dotless_i;
3531 }
3532#endif
a0270393
KW
3533 Copy(p, ustrp, *lenp, char);
3534 return original;
3535 }
3536 s += UTF8SKIP(s);
3537 }
051a06d4 3538
a0270393
KW
3539 /* Here, no characters crossed, result is ok as-is */
3540 return result;
3541 }
051a06d4
KW
3542 }
3543
4a4088c4 3544 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3545 if (UTF8_IS_INVARIANT(result)) {
3546 *ustrp = (U8) result;
3547 *lenp = 1;
3548 }
3549 else {
62cb07ea
KW
3550 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3551 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3552 *lenp = 2;
3553 }
3554
051a06d4 3555 return result;
1ca267a5
KW
3556
3557 return_long_s:
3558 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3559 * folds to a string of two of these characters. By returning this
3560 * instead, then, e.g.,
3561 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3562 * works. */
3563
3564 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3565 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3566 ustrp, *lenp, U8);
3567 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3568
3569 return_ligature_st:
3570 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3571 * have the other one fold to it */
3572
3573 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3574 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3575 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3576
3577#if UNICODE_MAJOR_VERSION == 3 \
3578 && UNICODE_DOT_VERSION == 0 \
3579 && UNICODE_DOT_DOT_VERSION == 1
3580
3581 return_dotless_i:
3582 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3583 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3584 return LATIN_SMALL_LETTER_DOTLESS_I;
3585
3586#endif
3587
a0ed51b3
LW
3588}
3589
711a919c 3590/* Note:
f90a9a02 3591 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3592 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3593 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3594 */
c4a5db0c 3595
a0ed51b3 3596SV*
56576a04
KW
3597Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3598 I32 minbits, I32 none)
a0ed51b3 3599{
c4a5db0c
KW
3600 PERL_ARGS_ASSERT_SWASH_INIT;
3601
3602 /* Returns a copy of a swash initiated by the called function. This is the
3603 * public interface, and returning a copy prevents others from doing
3604 * mischief on the original */
3605
56576a04
KW
3606 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3607 NULL, NULL));
c4a5db0c
KW
3608}
3609
3610SV*
56576a04
KW
3611Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3612 I32 minbits, I32 none, SV* invlist,
3613 U8* const flags_p)
c4a5db0c 3614{
2c1f00b9
YO
3615
3616 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3617 * use the following define */
3618
3619#define CORE_SWASH_INIT_RETURN(x) \
3620 PL_curpm= old_PL_curpm; \
3621 return x
3622
c4a5db0c 3623 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3624 * by calling utf8_heavy.pl in the general case. The returned value may be
3625 * the swash's inversion list instead if the input parameters allow it.
3626 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3627 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3628 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3629 *
3630 * This interface should only be used by functions that won't destroy or
3631 * adversely change the swash, as doing so affects all other uses of the
3632 * swash in the program; the general public should use 'Perl_swash_init'
3633 * instead.
3634 *
3635 * pkg is the name of the package that <name> should be in.
3636 * name is the name of the swash to find. Typically it is a Unicode
3637 * property name, including user-defined ones
3638 * listsv is a string to initialize the swash with. It must be of the form
3639 * documented as the subroutine return value in
3640 * L<perlunicode/User-Defined Character Properties>
3641 * minbits is the number of bits required to represent each data element.
3642 * It is '1' for binary properties.
3643 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3644 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3645 * flags_p if non-NULL is the address of various input and output flag bits
3646 * to the routine, as follows: ('I' means is input to the routine;
3647 * 'O' means output from the routine. Only flags marked O are
3648 * meaningful on return.)
3649 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3650 * came from a user-defined property. (I O)
5d3d13d1
KW
3651 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3652 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3653 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3654 * return of an inversion list instead of a swash hash if this routine
3655 * thinks that would result in faster execution of swash_fetch() later
3656 * on. (I)
9a53f6cf
KW
3657 *
3658 * Thus there are three possible inputs to find the swash: <name>,
3659 * <listsv>, and <invlist>. At least one must be specified. The result
3660 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3661 * actions can intersect, etc. what <name> gives. To avoid going out to
3662 * disk at all, <invlist> should specify completely what the swash should
3663 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3664 *
3665 * <invlist> is only valid for binary properties */
c4a5db0c 3666
2c1f00b9
YO
3667 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3668
c4a5db0c 3669 SV* retval = &PL_sv_undef;
83199d38 3670 HV* swash_hv = NULL;
87367d5f
KW
3671 const int invlist_swash_boundary =
3672 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3673 ? 512 /* Based on some benchmarking, but not extensive, see commit
3674 message */
3675 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3676
3677 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3678 assert(! invlist || minbits == 1);
3679
56576a04
KW
3680 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3681 regex that triggered the swash init and the swash init
3682 perl logic itself. See perl #122747 */
2c1f00b9 3683
9a53f6cf
KW
3684 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3685 * so */
3686 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3687 dSP;
3688 const size_t pkg_len = strlen(pkg);
3689 const size_t name_len = strlen(name);
3690 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3691 SV* errsv_save;
3692 GV *method;
3693
3694 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3695
3696 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3697 ENTER;
69794297 3698 SAVEHINTS();
2782061f 3699 save_re_context();
650f067c
JL
3700 /* We might get here via a subroutine signature which uses a utf8
3701 * parameter name, at which point PL_subname will have been set
3702 * but not yet used. */
3703 save_item(PL_subname);
69794297
KW
3704 if (PL_parser && PL_parser->error_count)
3705 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3706 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3707 if (!method) { /* demand load UTF-8 */
69794297 3708 ENTER;
db2c6cb3
FC
3709 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3710 GvSV(PL_errgv) = NULL;
1a419e6b 3711#ifndef NO_TAINT_SUPPORT
69794297
KW
3712 /* It is assumed that callers of this routine are not passing in
3713 * any user derived data. */
2782061f
DM
3714 /* Need to do this after save_re_context() as it will set
3715 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3716 * in Perl_magic_get). Even line to create errsv_save can turn on
3717 * PL_tainted. */
284167a5
SM
3718 SAVEBOOL(TAINT_get);
3719 TAINT_NOT;
3720#endif
69794297
KW
3721 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3722 NULL);
eed484f9 3723 {
db2c6cb3
FC
3724 /* Not ERRSV, as there is no need to vivify a scalar we are
3725 about to discard. */
3726 SV * const errsv = GvSV(PL_errgv);
3727 if (!SvTRUE(errsv)) {
3728 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3729 SvREFCNT_dec(errsv);
3730 }
eed484f9 3731 }
69794297
KW
3732 LEAVE;
3733 }
3734 SPAGAIN;
3735 PUSHMARK(SP);
3736 EXTEND(SP,5);