This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Initialize some inversion lists at start-up.
[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)";
fb7e7255
KW
39
40/* Be sure to synchronize this message with the similar one in regcomp.c */
760c7c2f 41static const char cp_above_legal_max[] =
d22ec717
KW
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
760c7c2f 44
d22ec717 45#define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
901b21bf 46
48ef279e 47/*
ccfc67b7 48=head1 Unicode Support
7fefc6c1 49These are various utility functions for manipulating UTF8-encoded
72d33970 50strings. For the uninitiated, this is a method of representing arbitrary
61296642 51Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
52characters in the ASCII range are unmodified, and a zero byte never appears
53within non-zero characters.
166f8a29 54
eaf7a4d2
CS
55=cut
56*/
57
9cbfb8ab
KW
58void
59Perl__force_out_malformed_utf8_message(pTHX_
60 const U8 *const p, /* First byte in UTF-8 sequence */
61 const U8 * const e, /* Final byte in sequence (may include
62 multiple chars */
63 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
64 usually 0, or some DISALLOW flags */
65 const bool die_here) /* If TRUE, this function does not return */
66{
67 /* This core-only function is to be called when a malformed UTF-8 character
68 * is found, in order to output the detailed information about the
69 * malformation before dieing. The reason it exists is for the occasions
70 * when such a malformation is fatal, but warnings might be turned off, so
71 * that normally they would not be actually output. This ensures that they
72 * do get output. Because a sequence may be malformed in more than one
73 * way, multiple messages may be generated, so we can't make them fatal, as
74 * that would cause the first one to die.
75 *
76 * Instead we pretend -W was passed to perl, then die afterwards. The
77 * flexibility is here to return to the caller so they can finish up and
78 * die themselves */
79 U32 errors;
80
81 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
82
83 ENTER;
c15a80f3 84 SAVEI8(PL_dowarn);
9cbfb8ab
KW
85 SAVESPTR(PL_curcop);
86
87 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
88 if (PL_curcop) {
89 PL_curcop->cop_warnings = pWARN_ALL;
90 }
91
92 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
93
94 LEAVE;
95
96 if (! errors) {
97 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
98 " be called only when there are errors found");
99 }
100
101 if (die_here) {
102 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
103 }
104}
105
bb07812e
KW
106STATIC HV *
107S_new_msg_hv(pTHX_ const char * const message, /* The message text */
108 U32 categories, /* Packed warning categories */
109 U32 flag) /* Flag associated with this message */
110{
111 /* Creates, populates, and returns an HV* that describes an error message
112 * for the translators between UTF8 and code point */
113
114 SV* msg_sv = newSVpv(message, 0);
115 SV* category_sv = newSVuv(categories);
116 SV* flag_bit_sv = newSVuv(flag);
117
118 HV* msg_hv = newHV();
119
120 PERL_ARGS_ASSERT_NEW_MSG_HV;
121
2b672cf5
KW
122 (void) hv_stores(msg_hv, "text", msg_sv);
123 (void) hv_stores(msg_hv, "warn_categories", category_sv);
124 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
125
126 return msg_hv;
127}
128
eaf7a4d2 129/*
378516de 130=for apidoc uvoffuni_to_utf8_flags
eebe1485 131
a27992cc 132THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
133Instead, B<Almost all code should use L</uvchr_to_utf8> or
134L</uvchr_to_utf8_flags>>.
a27992cc 135
de69f3af
KW
136This function is like them, but the input is a strict Unicode
137(as opposed to native) code point. Only in very rare circumstances should code
138not be using the native code point.
949cf498 139
efa9cd84 140For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 141
eebe1485
SC
142=cut
143*/
144
33f38593
KW
145U8 *
146Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
147{
148 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
149
150 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
151}
152
c94c2f39
KW
153/* All these formats take a single UV code point argument */
154const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
155const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
156 " is not recommended for open interchange";
157const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
158 " may not be portable";
57ff5f59
KW
159const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
160 " Unicode, requires a Perl extension," \
161 " and so is not portable";
c94c2f39 162
33f38593 163#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
164 STMT_START { \
165 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
166 U32 category = packWARN(WARN_SURROGATE); \
167 const char * format = surrogate_cp_format; \
168 if (msgs) { \
169 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
170 category, \
171 UNICODE_GOT_SURROGATE); \
172 } \
173 else { \
174 Perl_ck_warner_d(aTHX_ category, format, uv); \
175 } \
8ee1cdcb
KW
176 } \
177 if (flags & UNICODE_DISALLOW_SURROGATE) { \
178 return NULL; \
179 } \
180 } STMT_END;
181
33f38593 182#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
183 STMT_START { \
184 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
185 U32 category = packWARN(WARN_NONCHAR); \
186 const char * format = nonchar_cp_format; \
187 if (msgs) { \
188 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
189 category, \
190 UNICODE_GOT_NONCHAR); \
191 } \
192 else { \
193 Perl_ck_warner_d(aTHX_ category, format, uv); \
194 } \
8ee1cdcb
KW
195 } \
196 if (flags & UNICODE_DISALLOW_NONCHAR) { \
197 return NULL; \
198 } \
199 } STMT_END;
200
ba6ed43c
KW
201/* Use shorter names internally in this file */
202#define SHIFT UTF_ACCUMULATION_SHIFT
203#undef MARK
204#define MARK UTF_CONTINUATION_MARK
205#define MASK UTF_CONTINUATION_MASK
206
33f38593
KW
207/*
208=for apidoc uvchr_to_utf8_flags_msgs
209
210THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
211
212Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
213
214This function is for code that wants any warning and/or error messages to be
215returned to the caller rather than be displayed. All messages that would have
884a31ee 216been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
217
218It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
219placed after all the others, C<msgs>. If this parameter is 0, this function
220behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
221be a pointer to an C<HV *> variable, in which this function creates a new HV to
222contain any appropriate messages. The hash has three key-value pairs, as
223follows:
224
225=over 4
226
227=item C<text>
228
229The text of the message as a C<SVpv>.
230
231=item C<warn_categories>
232
233The warning category (or categories) packed into a C<SVuv>.
234
235=item C<flag>
236
237A single flag bit associated with this message, in a C<SVuv>.
238The bit corresponds to some bit in the C<*errors> return value,
239such as C<UNICODE_GOT_SURROGATE>.
240
241=back
242
243It's important to note that specifying this parameter as non-null will cause
244any warnings this function would otherwise generate to be suppressed, and
245instead be placed in C<*msgs>. The caller can check the lexical warnings state
246(or not) when choosing what to do with the returned messages.
247
248The caller, of course, is responsible for freeing any returned HV.
249
250=cut
251*/
252
253/* Undocumented; we don't want people using this. Instead they should use
254 * uvchr_to_utf8_flags_msgs() */
dfe13c55 255U8 *
33f38593 256Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 257{
33f38593
KW
258 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
259
260 if (msgs) {
261 *msgs = NULL;
262 }
7918f24d 263
2d1545e5 264 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 265 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
266 return d;
267 }
facc1dc2 268
3ea68d71 269 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
270 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
271 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
272 return d;
273 }
d9432125 274
ba6ed43c
KW
275 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
276 * below, the 16 is for start bytes E0-EF (which are all the possible ones
277 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
278 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
279 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
280 * 0x800-0xFFFF on ASCII */
281 if (uv < (16 * (1U << (2 * SHIFT)))) {
282 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
283 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
284 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
285
286#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
287 aren't tested here */
288 /* The most likely code points in this range are below the surrogates.
289 * Do an extra test to quickly exclude those. */
290 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
291 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
292 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
293 {
33f38593 294 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
295 }
296 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 297 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 298 }
ba6ed43c
KW
299 }
300#endif
301 return d;
302 }
303
304 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
305 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
306 * happen starting with 4-byte characters on ASCII platforms. We unify the
307 * code for these with EBCDIC, even though some of them require 5-bytes on
308 * those, because khw believes the code saving is worth the very slight
309 * performance hit on these high EBCDIC code points. */
310
311 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
d22ec717
KW
312 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
313 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
a5bf80e0 314 }
33f38593
KW
315 if ( (flags & UNICODE_WARN_SUPER)
316 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 317 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 318 {
33f38593
KW
319 const char * format = super_cp_format;
320 U32 category = packWARN(WARN_NON_UNICODE);
321 U32 flag = UNICODE_GOT_SUPER;
322
323 /* Choose the more dire applicable warning */
324 if (UNICODE_IS_PERL_EXTENDED(uv)) {
325 format = perl_extended_cp_format;
326 if (flags & (UNICODE_WARN_PERL_EXTENDED
327 |UNICODE_DISALLOW_PERL_EXTENDED))
328 {
329 flag = UNICODE_GOT_PERL_EXTENDED;
330 }
331 }
a5bf80e0 332
33f38593
KW
333 if (msgs) {
334 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
335 category, flag);
336 }
337 else {
338 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
339 }
a5bf80e0 340 }
56576a04 341 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
342 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
343 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
344 {
345 return NULL;
346 }
347 }
ba6ed43c 348 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 349 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 350 }
d9432125 351
ba6ed43c
KW
352 /* Test for and handle 4-byte result. In the test immediately below, the
353 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
354 * characters). The 3 is for 3 continuation bytes; these each contribute
355 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
356 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
357 * 0x1_0000-0x1F_FFFF on ASCII */
358 if (uv < (8 * (1U << (3 * SHIFT)))) {
359 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
360 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
362 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
363
364#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
365 characters. The end-plane non-characters for EBCDIC were
366 handled just above */
367 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 368 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 369 }
ba6ed43c 370 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 371 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
372 }
373#endif
374
375 return d;
376 }
377
378 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
379 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
380 * format. The unrolled version above turns out to not save all that much
381 * time, and at these high code points (well above the legal Unicode range
382 * on ASCII platforms, and well above anything in common use in EBCDIC),
383 * khw believes that less code outweighs slight performance gains. */
384
d9432125 385 {
5aaebcb3 386 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
387 U8 *p = d+len-1;
388 while (p > d) {
957a9e81
KW
389 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
390 uv >>= SHIFT;
1d72bdf6 391 }
4c8cd605 392 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
393 return d+len;
394 }
a0ed51b3 395}
a5bf80e0 396
646ca15d 397/*
07693fe6
KW
398=for apidoc uvchr_to_utf8
399
bcb1a2d4 400Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 401of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
402C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
403the byte after the end of the new character. In other words,
07693fe6
KW
404
405 d = uvchr_to_utf8(d, uv);
406
407is the recommended wide native character-aware way of saying
408
409 *(d++) = uv;
410
d22ec717
KW
411This function accepts any code point from 0..C<IV_MAX> as input.
412C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
413
414It is possible to forbid or warn on non-Unicode code points, or those that may
415be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 416
07693fe6
KW
417=cut
418*/
419
de69f3af
KW
420/* This is also a macro */
421PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
422
07693fe6
KW
423U8 *
424Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
425{
de69f3af 426 return uvchr_to_utf8(d, uv);
07693fe6
KW
427}
428
de69f3af
KW
429/*
430=for apidoc uvchr_to_utf8_flags
431
432Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 433of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
434C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
435the byte after the end of the new character. In other words,
de69f3af
KW
436
437 d = uvchr_to_utf8_flags(d, uv, flags);
438
439or, in most cases,
440
441 d = uvchr_to_utf8_flags(d, uv, 0);
442
443This is the Unicode-aware way of saying
444
445 *(d++) = uv;
446
d22ec717
KW
447If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
448input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
449
450Specifying C<flags> can further restrict what is allowed and not warned on, as
451follows:
de69f3af 452
796b6530 453If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
454the function will raise a warning, provided UTF8 warnings are enabled. If
455instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
456NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 457
760c7c2f
KW
458Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
459affect how the function handles a Unicode non-character.
93e6dbd6 460
760c7c2f
KW
461And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
462affect the handling of code points that are above the Unicode maximum of
4630x10FFFF. Languages other than Perl may not be able to accept files that
464contain these.
93e6dbd6
KW
465
466The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
467the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
468three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
469allowed inputs to the strict UTF-8 traditionally defined by Unicode.
470Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
471C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
472above-Unicode and surrogate flags, but not the non-character ones, as
473defined in
474L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
475See L<perlunicode/Noncharacter code points>.
93e6dbd6 476
57ff5f59
KW
477Extremely high code points were never specified in any standard, and require an
478extension to UTF-8 to express, which Perl does. It is likely that programs
479written in something other than Perl would not be able to read files that
480contain these; nor would Perl understand files written by something that uses a
481different extension. For these reasons, there is a separate set of flags that
482can warn and/or disallow these extremely high code points, even if other
483above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
484and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
485L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
486treat all above-Unicode code points, including these, as malformations. (Note
487that the Unicode standard considers anything above 0x10FFFF to be illegal, but
488there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
489
490A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
491retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
492C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
493C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
494platforms,these flags can apply to code points that actually do fit in 31 bits.
495The new names accurately describe the situation in all cases.
de69f3af 496
de69f3af
KW
497=cut
498*/
499
500/* This is also a macro */
501PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
502
07693fe6
KW
503U8 *
504Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
505{
de69f3af 506 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
507}
508
57ff5f59
KW
509#ifndef UV_IS_QUAD
510
e050c007
KW
511STATIC int
512S_is_utf8_cp_above_31_bits(const U8 * const s,
513 const U8 * const e,
514 const bool consider_overlongs)
83dc0f42
KW
515{
516 /* Returns TRUE if the first code point represented by the Perl-extended-
517 * UTF-8-encoded string starting at 's', and looking no further than 'e -
518 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
519 *
520 * The function handles the case where the input bytes do not include all
521 * the ones necessary to represent a full character. That is, they may be
522 * the intial bytes of the representation of a code point, but possibly
523 * the final ones necessary for the complete representation may be beyond
524 * 'e - 1'.
525 *
e050c007
KW
526 * The function also can handle the case where the input is an overlong
527 * sequence. If 'consider_overlongs' is 0, the function assumes the
528 * input is not overlong, without checking, and will return based on that
529 * assumption. If this parameter is 1, the function will go to the trouble
530 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 531 *
e050c007 532 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
533 */
534
e050c007
KW
535 const STRLEN len = e - s;
536 int is_overlong;
537
538 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
539
540 assert(! UTF8_IS_INVARIANT(*s) && e > s);
541
83dc0f42
KW
542#ifdef EBCDIC
543
e050c007 544 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 545
e050c007
KW
546 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
547 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
548 * also be the start byte for a 31-bit code point; we need at least 2
549 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
550 * the start byte for an overlong sequence, but for 30-bit or smaller code
551 * points, so we don't have to worry about overlongs on EBCDIC.) */
552 if (*s != 0xFE) {
553 return 0;
554 }
83dc0f42 555
e050c007
KW
556 if (len == 1) {
557 return -1;
558 }
83dc0f42 559
e050c007 560#else
83dc0f42 561
e050c007
KW
562 /* On ASCII, FE and FF are the only start bytes that can evaluate to
563 * needing more than 31 bits. */
564 if (LIKELY(*s < 0xFE)) {
565 return 0;
566 }
83dc0f42 567
e050c007
KW
568 /* What we have left are FE and FF. Both of these require more than 31
569 * bits unless they are for overlongs. */
570 if (! consider_overlongs) {
571 return 1;
572 }
83dc0f42 573
e050c007
KW
574 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
575 * above 31 bits. But we need more than one byte to discern this, so if
576 * passed just the start byte, it could be an overlong evaluating to
577 * smaller */
578 if (len == 1) {
579 return -1;
580 }
83dc0f42 581
e050c007
KW
582 /* Having excluded len==1, and knowing that FE and FF are both valid start
583 * bytes, we can call the function below to see if the sequence is
584 * overlong. (We don't need the full generality of the called function,
585 * but for these huge code points, speed shouldn't be a consideration, and
586 * the compiler does have enough information, since it's static to this
587 * file, to optimize to just the needed parts.) */
588 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 589
e050c007
KW
590 /* If it isn't overlong, more than 31 bits are required. */
591 if (is_overlong == 0) {
592 return 1;
593 }
83dc0f42 594
e050c007
KW
595 /* If it is indeterminate if it is overlong, return that */
596 if (is_overlong < 0) {
597 return -1;
598 }
599
600 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
601 * the max it can be is 2**31 - 1 */
602 if (*s == 0xFE) {
603 return 0;
83dc0f42
KW
604 }
605
e050c007
KW
606#endif
607
608 /* Here, ASCII and EBCDIC rejoin:
609 * On ASCII: We have an overlong sequence starting with FF
610 * On EBCDIC: We have a sequence starting with FE. */
611
612 { /* For C89, use a block so the declaration can be close to its use */
613
614#ifdef EBCDIC
615
5f995336
KW
616 /* U+7FFFFFFF (2 ** 31 - 1)
617 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
618 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
619 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
620 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
621 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
622 * U+80000000 (2 ** 31):
623 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
624 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
625 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
626 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
627 *
628 * and since we know that *s = \xfe, any continuation sequcence
629 * following it that is gt the below is above 31 bits
630 [0] [1] [2] [3] [4] [5] [6] */
631 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
632
633#else
634
635 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
636 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
637 * FF overlong for U+80000000 (2 ** 31):
638 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
639 * and since we know that *s = \xff, any continuation sequcence
640 * following it that is gt the below is above 30 bits
641 [0] [1] [2] [3] [4] [5] [6] */
642 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 643
83dc0f42
KW
644
645#endif
e050c007
KW
646 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
647 const STRLEN cmp_len = MIN(conts_len, len - 1);
648
649 /* Now compare the continuation bytes in s with the ones we have
650 * compiled in that are for the largest 30 bit code point. If we have
651 * enough bytes available to determine the answer, or the bytes we do
652 * have differ from them, we can compare the two to get a definitive
653 * answer (Note that in UTF-EBCDIC, the two lowest possible
654 * continuation bytes are \x41 and \x42.) */
655 if (cmp_len >= conts_len || memNE(s + 1,
656 conts_for_highest_30_bit,
657 cmp_len))
658 {
659 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
660 }
83dc0f42 661
e050c007
KW
662 /* Here, all the bytes we have are the same as the highest 30-bit code
663 * point, but we are missing so many bytes that we can't make the
664 * determination */
665 return -1;
666 }
83dc0f42
KW
667}
668
57ff5f59
KW
669#endif
670
d6be65ae 671PERL_STATIC_INLINE int
12a4bed3
KW
672S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
673{
d6be65ae
KW
674 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
675 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
676 * it isn't, and -1 if there isn't enough information to tell. This last
677 * return value can happen if the sequence is incomplete, missing some
678 * trailing bytes that would form a complete character. If there are
679 * enough bytes to make a definitive decision, this function does so.
680 * Usually 2 bytes sufficient.
681 *
682 * Overlongs can occur whenever the number of continuation bytes changes.
683 * That means whenever the number of leading 1 bits in a start byte
684 * increases from the next lower start byte. That happens for start bytes
685 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
686 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
687 * ASCII platforms: C0, C1
688 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 689 */
12a4bed3
KW
690
691 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
692 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
693
694 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
695 assert(len > 1 && UTF8_IS_START(*s));
696
697 /* Each platform has overlongs after the start bytes given above (expressed
698 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
699 * the logic is the same, except the E0 overlong has already been excluded
700 * on EBCDIC platforms. The values below were found by manually
701 * inspecting the UTF-8 patterns. See the tables in utf8.h and
702 * utfebcdic.h. */
703
704# ifdef EBCDIC
705# define F0_ABOVE_OVERLONG 0xB0
706# define F8_ABOVE_OVERLONG 0xA8
707# define FC_ABOVE_OVERLONG 0xA4
708# define FE_ABOVE_OVERLONG 0xA2
709# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
710 /* I8(0xfe) is FF */
711# else
712
713 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 714 return 1;
12a4bed3
KW
715 }
716
717# define F0_ABOVE_OVERLONG 0x90
718# define F8_ABOVE_OVERLONG 0x88
719# define FC_ABOVE_OVERLONG 0x84
720# define FE_ABOVE_OVERLONG 0x82
721# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
722# endif
723
724
725 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
726 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
727 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
728 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
729 {
d6be65ae 730 return 1;
12a4bed3
KW
731 }
732
b0b342d4 733 /* Check for the FF overlong */
d6be65ae 734 return isFF_OVERLONG(s, len);
b0b342d4
KW
735}
736
8d6204cc 737PERL_STATIC_INLINE int
b0b342d4
KW
738S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
739{
8d6204cc
KW
740 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
741 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
742 * it isn't, and -1 if there isn't enough information to tell. This last
743 * return value can happen if the sequence is incomplete, missing some
744 * trailing bytes that would form a complete character. If there are
745 * enough bytes to make a definitive decision, this function does so. */
746
b0b342d4 747 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 748
8d6204cc
KW
749 /* To be an FF overlong, all the available bytes must match */
750 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
751 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
752 {
753 return 0;
754 }
755
756 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
757 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 758 * utfebcdic.h. */
8d6204cc
KW
759 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
760 return 1;
761 }
12a4bed3 762
8d6204cc
KW
763 /* The missing bytes could cause the result to go one way or the other, so
764 * the result is indeterminate */
765 return -1;
12a4bed3
KW
766}
767
d22ec717 768#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
769# ifdef EBCDIC /* Actually is I8 */
770# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 771 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
772# else
773# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 774 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
775# endif
776#endif
777
c285bbc4 778PERL_STATIC_INLINE int
e050c007
KW
779S_does_utf8_overflow(const U8 * const s,
780 const U8 * e,
781 const bool consider_overlongs)
a77c906e 782{
c285bbc4 783 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
784 * 'e' - 1 would overflow an IV on this platform; that is if it represents
785 * a code point larger than the highest representable code point. It
786 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
787 * enough information to tell. This last return value can happen if the
788 * sequence is incomplete, missing some trailing bytes that would form a
789 * complete character. If there are enough bytes to make a definitive
790 * decision, this function does so.
c285bbc4 791 *
e050c007
KW
792 * If 'consider_overlongs' is TRUE, the function checks for the possibility
793 * that the sequence is an overlong that doesn't overflow. Otherwise, it
794 * assumes the sequence is not an overlong. This can give different
795 * results only on ASCII 32-bit platforms.
796 *
c285bbc4
KW
797 * (For ASCII platforms, we could use memcmp() because we don't have to
798 * convert each byte to I8, but it's very rare input indeed that would
799 * approach overflow, so the loop below will likely only get executed once.)
800 *
801 * 'e' - 1 must not be beyond a full character. */
a77c906e 802
a77c906e
KW
803
804 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
805 assert(s <= e && s + UTF8SKIP(s) >= e);
806
d22ec717
KW
807#if ! defined(UV_IS_QUAD)
808
809 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
810
811#else
812
813 PERL_UNUSED_ARG(consider_overlongs);
814
815 {
816 const STRLEN len = e - s;
817 const U8 *x;
818 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
819
820 for (x = s; x < e; x++, y++) {
821
822 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
823 continue;
824 }
825
826 /* If this byte is larger than the corresponding highest UTF-8
827 * byte, the sequence overflow; otherwise the byte is less than,
828 * and so the sequence doesn't overflow */
829 return NATIVE_UTF8_TO_I8(*x) > *y;
830
831 }
832
833 /* Got to the end and all bytes are the same. If the input is a whole
834 * character, it doesn't overflow. And if it is a partial character,
835 * there's not enough information to tell */
836 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
837 return -1;
838 }
839
840 return 0;
841 }
842
843#endif
844
845}
846
847#if 0
848
849/* This is the portions of the above function that deal with UV_MAX instead of
850 * IV_MAX. They are left here in case we want to combine them so that internal
851 * uses can have larger code points. The only logic difference is that the
852 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
853 * different logic.
854 */
855
856/* Anything larger than this will overflow the word if it were converted into a UV */
857#if defined(UV_IS_QUAD)
858# ifdef EBCDIC /* Actually is I8 */
859# define HIGHEST_REPRESENTABLE_UTF8 \
860 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
861# else
862# define HIGHEST_REPRESENTABLE_UTF8 \
863 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
864# endif
865#else /* 32-bit */
866# ifdef EBCDIC
867# define HIGHEST_REPRESENTABLE_UTF8 \
868 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
869# else
870# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
871# endif
872#endif
873
a77c906e
KW
874#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
875
876 /* On 32 bit ASCII machines, many overlongs that start with FF don't
877 * overflow */
e050c007 878 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
879
880 /* To be such an overlong, the first bytes of 's' must match
881 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
882 * don't have any additional bytes available, the sequence, when
883 * completed might or might not fit in 32 bits. But if we have that
884 * next byte, we can tell for sure. If it is <= 0x83, then it does
885 * fit. */
886 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
887 return -1;
888 }
889
890 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
891 }
892
d22ec717
KW
893/* Starting with the #else, the rest of the function is identical except
894 * 1. we need to move the 'len' declaration to be global to the function
895 * 2. the endif move to just after the UNUSED_ARG.
896 * An empty endif is given just below to satisfy the preprocessor
897 */
a77c906e
KW
898#endif
899
d22ec717 900#endif
a77c906e 901
12a4bed3
KW
902#undef F0_ABOVE_OVERLONG
903#undef F8_ABOVE_OVERLONG
904#undef FC_ABOVE_OVERLONG
905#undef FE_ABOVE_OVERLONG
906#undef FF_OVERLONG_PREFIX
907
35f8c9bd 908STRLEN
edc2c47a 909Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 910{
2b479609 911 STRLEN len;
12a4bed3 912 const U8 *x;
35f8c9bd 913
2b479609
KW
914 /* A helper function that should not be called directly.
915 *
916 * This function returns non-zero if the string beginning at 's' and
917 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
918 * code point; otherwise it returns 0. The examination stops after the
919 * first code point in 's' is validated, not looking at the rest of the
920 * input. If 'e' is such that there are not enough bytes to represent a
921 * complete code point, this function will return non-zero anyway, if the
922 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
923 * excluded by 'flags'.
924 *
925 * A non-zero return gives the number of bytes required to represent the
926 * code point. Be aware that if the input is for a partial character, the
927 * return will be larger than 'e - s'.
928 *
929 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
930 * The caller should have excluded the possibility of it being invariant
931 * before calling this function.
2b479609
KW
932 *
933 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
934 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
935 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
936 * disallowed by the flags. If the input is only for a partial character,
937 * the function will return non-zero if there is any sequence of
938 * well-formed UTF-8 that, when appended to the input sequence, could
939 * result in an allowed code point; otherwise it returns 0. Non characters
940 * cannot be determined based on partial character input. But many of the
941 * other excluded types can be determined with just the first one or two
942 * bytes.
943 *
944 */
945
946 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
947
948 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 949 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 950 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 951
2b479609 952 /* A variant char must begin with a start byte */
35f8c9bd
KW
953 if (UNLIKELY(! UTF8_IS_START(*s))) {
954 return 0;
955 }
956
edc2c47a
KW
957 /* Examine a maximum of a single whole code point */
958 if (e - s > UTF8SKIP(s)) {
959 e = s + UTF8SKIP(s);
960 }
961
2b479609
KW
962 len = e - s;
963
964 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
965 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 966
56576a04
KW
967 /* Here, we are disallowing some set of largish code points, and the
968 * first byte indicates the sequence is for a code point that could be
969 * in the excluded set. We generally don't have to look beyond this or
970 * the second byte to see if the sequence is actually for one of the
971 * excluded classes. The code below is derived from this table:
972 *
2b479609
KW
973 * UTF-8 UTF-EBCDIC I8
974 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
975 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
976 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
977 *
56576a04
KW
978 * Keep in mind that legal continuation bytes range between \x80..\xBF
979 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
980 * continuation bytes. Hence, we don't have to test the upper edge
981 * because if any of those is encountered, the sequence is malformed,
982 * and would fail elsewhere in this function.
983 *
984 * The code here likewise assumes that there aren't other
985 * malformations; again the function should fail elsewhere because of
986 * these. For example, an overlong beginning with FC doesn't actually
987 * have to be a super; it could actually represent a small code point,
988 * even U+0000. But, since overlongs (and other malformations) are
989 * illegal, the function should return FALSE in either case.
2b479609
KW
990 */
991
992#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
993# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 994# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 995
19794540
KW
996# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
997 /* B6 and B7 */ \
998 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 999# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1000#else
1001# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1002# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1003# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1004# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1005#endif
1006
1007 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1008 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1009 {
2b479609
KW
1010 return 0; /* Above Unicode */
1011 }
1012
d044b7a7 1013 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1014 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1015 {
57ff5f59 1016 return 0;
2b479609
KW
1017 }
1018
1019 if (len > 1) {
1020 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1021
1022 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1023 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1024 {
1025 return 0; /* Above Unicode */
1026 }
1027
1028 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1029 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1030 {
1031 return 0; /* Surrogate */
1032 }
1033
1034 if ( (flags & UTF8_DISALLOW_NONCHAR)
1035 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1036 {
1037 return 0; /* Noncharacter code point */
1038 }
1039 }
1040 }
1041
1042 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1043 for (x = s + 1; x < e; x++) {
1044 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1045 return 0;
1046 }
1047 }
1048
af13dd8a 1049 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1050 * overlong. */
d6be65ae 1051 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1052 return 0;
af13dd8a
KW
1053 }
1054
12a4bed3
KW
1055 /* And finally, that the code point represented fits in a word on this
1056 * platform */
e050c007
KW
1057 if (0 < does_utf8_overflow(s, e,
1058 0 /* Don't consider overlongs */
1059 ))
1060 {
12a4bed3 1061 return 0;
35f8c9bd
KW
1062 }
1063
2b479609 1064 return UTF8SKIP(s);
35f8c9bd
KW
1065}
1066
7e2f38b2 1067char *
63ab03b3 1068Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1069{
1070 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1071 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1072 * Currently, there are only two formats, so it is currently a bool:
1073 * 0 \xab
1074 * 1 ab (that is a space between two hex digit bytes)
1075 */
7cf8d05d
KW
1076
1077 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1078 trailing NUL */
63ab03b3
KW
1079 const U8 * s = start;
1080 const U8 * const e = start + len;
7cf8d05d
KW
1081 char * output;
1082 char * d;
1083
1084 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1085
1086 Newx(output, output_len, char);
1087 SAVEFREEPV(output);
1088
1089 d = output;
63ab03b3 1090 for (s = start; s < e; s++) {
7cf8d05d
KW
1091 const unsigned high_nibble = (*s & 0xF0) >> 4;
1092 const unsigned low_nibble = (*s & 0x0F);
1093
7e2f38b2 1094 if (format) {
63ab03b3
KW
1095 if (s > start) {
1096 *d++ = ' ';
1097 }
7e2f38b2
KW
1098 }
1099 else {
1100 *d++ = '\\';
1101 *d++ = 'x';
1102 }
7cf8d05d
KW
1103
1104 if (high_nibble < 10) {
1105 *d++ = high_nibble + '0';
1106 }
1107 else {
1108 *d++ = high_nibble - 10 + 'a';
1109 }
1110
1111 if (low_nibble < 10) {
1112 *d++ = low_nibble + '0';
1113 }
1114 else {
1115 *d++ = low_nibble - 10 + 'a';
1116 }
1117 }
1118
1119 *d = '\0';
1120 return output;
1121}
1122
806547a7 1123PERL_STATIC_INLINE char *
7cf8d05d
KW
1124S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1125
421da25c 1126 /* Max number of bytes to print */
3cc6a05e 1127 STRLEN print_len,
7cf8d05d
KW
1128
1129 /* Which one is the non-continuation */
1130 const STRLEN non_cont_byte_pos,
1131
1132 /* How many bytes should there be? */
1133 const STRLEN expect_len)
806547a7
KW
1134{
1135 /* Return the malformation warning text for an unexpected continuation
1136 * byte. */
1137
7cf8d05d 1138 const char * const where = (non_cont_byte_pos == 1)
806547a7 1139 ? "immediately"
7cf8d05d
KW
1140 : Perl_form(aTHX_ "%d bytes",
1141 (int) non_cont_byte_pos);
421da25c
KW
1142 const U8 * x = s + non_cont_byte_pos;
1143 const U8 * e = s + print_len;
806547a7
KW
1144
1145 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1146
7cf8d05d
KW
1147 /* We don't need to pass this parameter, but since it has already been
1148 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1149 assert(expect_len == UTF8SKIP(s));
1150
421da25c
KW
1151 /* As a defensive coding measure, don't output anything past a NUL. Such
1152 * bytes shouldn't be in the middle of a malformation, and could mark the
1153 * end of the allocated string, and what comes after is undefined */
1154 for (; x < e; x++) {
1155 if (*x == '\0') {
1156 x++; /* Output this particular NUL */
1157 break;
1158 }
1159 }
1160
7cf8d05d
KW
1161 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1162 " %s after start byte 0x%02x; need %d bytes, got %d)",
1163 malformed_text,
421da25c 1164 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1165 *(s + non_cont_byte_pos),
1166 where,
1167 *s,
1168 (int) expect_len,
1169 (int) non_cont_byte_pos);
806547a7
KW
1170}
1171
35f8c9bd
KW
1172/*
1173
de69f3af 1174=for apidoc utf8n_to_uvchr
378516de
KW
1175
1176THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1177Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1178
9041c2e3 1179Bottom level UTF-8 decode routine.
de69f3af 1180Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1181which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1182C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1183the length, in bytes, of that character.
949cf498
KW
1184
1185The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1186well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1187causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1188is the next possible position in C<s> that could begin a non-malformed
1189character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1190is raised. Some UTF-8 input sequences may contain multiple malformations.
1191This function tries to find every possible one in each call, so multiple
56576a04 1192warnings can be raised for the same sequence.
949cf498
KW
1193
1194Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1195individual types of malformations, such as the sequence being overlong (that
1196is, when there is a shorter sequence that can express the same code point;
1197overlong sequences are expressly forbidden in the UTF-8 standard due to
1198potential security issues). Another malformation example is the first byte of
1199a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1200flags. Even if allowed, this function generally returns the Unicode
1201REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1202F<utf8.h> to override this behavior for the overlong malformations, but don't
1203do that except for very specialized purposes.
949cf498 1204
796b6530 1205The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1206flags) malformation is found. If this flag is set, the routine assumes that
1207the caller will raise a warning, and this function will silently just set
d088425d
KW
1208C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1209
75200dff 1210Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1211character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1212in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1213be set to 1. To disambiguate, upon a zero return, see if the first byte of
1214C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1215error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1216
1217Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1218Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1219By default these are considered regular code points, but certain situations
ecc1615f
KW
1220warrant special handling for them, which can be specified using the C<flags>
1221parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1222three classes are treated as malformations and handled as such. The flags
1223C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1224C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1225disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1226restricts the allowed inputs to the strict UTF-8 traditionally defined by
1227Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1228definition given by
1229L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1230The difference between traditional strictness and C9 strictness is that the
1231latter does not forbid non-character code points. (They are still discouraged,
1232however.) For more discussion see L<perlunicode/Noncharacter code points>.
1233
1234The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1235C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1236C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1237raised for their respective categories, but otherwise the code points are
1238considered valid (not malformations). To get a category to both be treated as
1239a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1240(But note that warnings are not raised if lexically disabled nor if
796b6530 1241C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1242
57ff5f59
KW
1243Extremely high code points were never specified in any standard, and require an
1244extension to UTF-8 to express, which Perl does. It is likely that programs
1245written in something other than Perl would not be able to read files that
1246contain these; nor would Perl understand files written by something that uses a
1247different extension. For these reasons, there is a separate set of flags that
1248can warn and/or disallow these extremely high code points, even if other
1249above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1250C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1251L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1252above-Unicode code points, including these, as malformations.
1253(Note that the Unicode standard considers anything above 0x10FFFF to be
1254illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1255(2**31 -1))
1256
1257A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1258retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1259C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1260C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1261can apply to code points that actually do fit in 31 bits. This happens on
1262EBCDIC platforms, and sometimes when the L<overlong
1263malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1264describe the situation in all cases.
1265
ab8e6d41 1266
949cf498
KW
1267All other code points corresponding to Unicode characters, including private
1268use and those yet to be assigned, are never considered malformed and never
1269warn.
67e989fb 1270
37607a96 1271=cut
f9380377
KW
1272
1273Also implemented as a macro in utf8.h
1274*/
1275
1276UV
e6a4ffc3
KW
1277Perl_utf8n_to_uvchr(const U8 *s,
1278 STRLEN curlen,
1279 STRLEN *retlen,
1280 const U32 flags)
f9380377
KW
1281{
1282 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1283
1284 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1285}
1286
1287/*
1288
1289=for apidoc utf8n_to_uvchr_error
1290
1291THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1292Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1293
1294This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1295are when an error is found. If you also need to know the generated warning
1296messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1297
1298It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1299all the others, C<errors>. If this parameter is 0, this function behaves
1300identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1301to a C<U32> variable, which this function sets to indicate any errors found.
1302Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1303C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1304of these bits will be set if a malformation is found, even if the input
7a65503b 1305C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1306exceptions are noted:
1307
1308=over 4
1309
57ff5f59 1310=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1311
57ff5f59
KW
1312The input sequence is not standard UTF-8, but a Perl extension. This bit is
1313set only if the input C<flags> parameter contains either the
1314C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1315
1316Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1317and so some extension must be used to express them. Perl uses a natural
1318extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1319extension to represent even higher ones, so that any code point that fits in a
132064-bit word can be represented. Text using these extensions is not likely to
1321be portable to non-Perl code. We lump both of these extensions together and
1322refer to them as Perl extended UTF-8. There exist other extensions that people
1323have invented, incompatible with Perl's.
1324
1325On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1326extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1327than on ASCII. Prior to that, code points 2**31 and higher were simply
1328unrepresentable, and a different, incompatible method was used to represent
1329code points between 2**30 and 2**31 - 1.
1330
1331On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1332Perl extended UTF-8 is used.
1333
1334In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1335may use for backward compatibility. That name is misleading, as this flag may
1336be set when the code point actually does fit in 31 bits. This happens on
1337EBCDIC platforms, and sometimes when the L<overlong
1338malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1339describes the situation in all cases.
f9380377
KW
1340
1341=item C<UTF8_GOT_CONTINUATION>
1342
1343The input sequence was malformed in that the first byte was a a UTF-8
1344continuation byte.
1345
1346=item C<UTF8_GOT_EMPTY>
1347
1348The input C<curlen> parameter was 0.
1349
1350=item C<UTF8_GOT_LONG>
1351
1352The input sequence was malformed in that there is some other sequence that
1353evaluates to the same code point, but that sequence is shorter than this one.
1354
fecaf136
KW
1355Until Unicode 3.1, it was legal for programs to accept this malformation, but
1356it was discovered that this created security issues.
1357
f9380377
KW
1358=item C<UTF8_GOT_NONCHAR>
1359
1360The code point represented by the input UTF-8 sequence is for a Unicode
1361non-character code point.
1362This bit is set only if the input C<flags> parameter contains either the
1363C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1364
1365=item C<UTF8_GOT_NON_CONTINUATION>
1366
1367The input sequence was malformed in that a non-continuation type byte was found
1368in a position where only a continuation type one should be.
1369
1370=item C<UTF8_GOT_OVERFLOW>
1371
1372The input sequence was malformed in that it is for a code point that is not
d22ec717 1373representable in the number of bits available in an IV on the current platform.
f9380377
KW
1374
1375=item C<UTF8_GOT_SHORT>
1376
1377The input sequence was malformed in that C<curlen> is smaller than required for
1378a complete sequence. In other words, the input is for a partial character
1379sequence.
1380
1381=item C<UTF8_GOT_SUPER>
1382
1383The input sequence was malformed in that it is for a non-Unicode code point;
1384that is, one above the legal Unicode maximum.
1385This bit is set only if the input C<flags> parameter contains either the
1386C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1387
1388=item C<UTF8_GOT_SURROGATE>
1389
1390The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1391code point.
1392This bit is set only if the input C<flags> parameter contains either the
1393C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1394
1395=back
1396
133551d8
KW
1397To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1398flag to suppress any warnings, and then examine the C<*errors> return.
1399
f9380377 1400=cut
37657a5b
KW
1401
1402Also implemented as a macro in utf8.h
37607a96 1403*/
67e989fb 1404
a0ed51b3 1405UV
e6a4ffc3 1406Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1407 STRLEN curlen,
1408 STRLEN *retlen,
1409 const U32 flags,
1410 U32 * errors)
1411{
1412 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1413
1414 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1415}
1416
1417/*
1418
1419=for apidoc utf8n_to_uvchr_msgs
1420
1421THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1422Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1423
1424This function is for code that needs to know what the precise malformation(s)
1425are when an error is found, and wants the corresponding warning and/or error
1426messages to be returned to the caller rather than be displayed. All messages
1427that would have been displayed if all lexcial warnings are enabled will be
1428returned.
1429
1430It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1431placed after all the others, C<msgs>. If this parameter is 0, this function
1432behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1433be a pointer to an C<AV *> variable, in which this function creates a new AV to
1434contain any appropriate messages. The elements of the array are ordered so
1435that the first message that would have been displayed is in the 0th element,
1436and so on. Each element is a hash with three key-value pairs, as follows:
1437
1438=over 4
1439
1440=item C<text>
1441
1442The text of the message as a C<SVpv>.
1443
1444=item C<warn_categories>
1445
1446The warning category (or categories) packed into a C<SVuv>.
1447
1448=item C<flag>
1449
1450A single flag bit associated with this message, in a C<SVuv>.
1451The bit corresponds to some bit in the C<*errors> return value,
1452such as C<UTF8_GOT_LONG>.
1453
1454=back
1455
1456It's important to note that specifying this parameter as non-null will cause
1457any warnings this function would otherwise generate to be suppressed, and
1458instead be placed in C<*msgs>. The caller can check the lexical warnings state
1459(or not) when choosing what to do with the returned messages.
1460
1461If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1462no AV is created.
1463
1464The caller, of course, is responsible for freeing any returned AV.
1465
1466=cut
1467*/
1468
1469UV
e6a4ffc3 1470Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1471 STRLEN curlen,
1472 STRLEN *retlen,
1473 const U32 flags,
1474 U32 * errors,
1475 AV ** msgs)
a0ed51b3 1476{
d4c19fe8 1477 const U8 * const s0 = s;
2b9519f0 1478 const U8 * send = s0 + curlen;
5af9f822
KW
1479 U32 possible_problems; /* A bit is set here for each potential problem
1480 found as we go along */
1481 UV uv;
1482 STRLEN expectlen; /* How long should this sequence be? */
1483 STRLEN avail_len; /* When input is too short, gives what that is */
1484 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1485 gets set and discarded */
a0dbb045 1486
2b5e7bc2
KW
1487 /* The below are used only if there is both an overlong malformation and a
1488 * too short one. Otherwise the first two are set to 's0' and 'send', and
1489 * the third not used at all */
5af9f822 1490 U8 * adjusted_s0;
e9f2c446
KW
1491 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1492 routine; see [perl #130921] */
5af9f822 1493 UV uv_so_far;
e6a4ffc3 1494 dTHX;
5af9f822 1495
e6a4ffc3 1496 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1497
1498 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1499 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1500 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1501 * final case. */
1502
1503#ifndef EBCDIC
1504
1505 /* Each of the affected Hanguls starts with \xED */
1506
1507 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1508 if (retlen) {
1509 *retlen = 3;
1510 }
1511 if (errors) {
1512 *errors = 0;
1513 }
1514 if (msgs) {
1515 *msgs = NULL;
1516 }
1517
1518 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1519 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1520 | (s0[2] & UTF_CONTINUATION_MASK);
1521 }
1522
1523#endif
1524
1525 /* In conjunction with the exhaustive tests that can be enabled in
1526 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1527 * what it is intended to do, and that no flaws in it are masked by
1528 * dropping down and executing the code below
1529 assert(! isUTF8_CHAR(s0, send)
1530 || UTF8_IS_SURROGATE(s0, send)
1531 || UTF8_IS_SUPER(s0, send)
1532 || UTF8_IS_NONCHAR(s0,send));
1533 */
1534
1535 s = s0;
1536 uv = *s0;
1537 possible_problems = 0;
1538 expectlen = 0;
1539 avail_len = 0;
1540 discard_errors = 0;
1541 adjusted_s0 = (U8 *) s0;
1542 uv_so_far = 0;
1543
f9380377
KW
1544 if (errors) {
1545 *errors = 0;
1546 }
1547 else {
1548 errors = &discard_errors;
1549 }
a0dbb045 1550
eb83ed87
KW
1551 /* The order of malformation tests here is important. We should consume as
1552 * few bytes as possible in order to not skip any valid character. This is
1553 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1554 * http://unicode.org/reports/tr36 for more discussion as to why. For
1555 * example, once we've done a UTF8SKIP, we can tell the expected number of
1556 * bytes, and could fail right off the bat if the input parameters indicate
1557 * that there are too few available. But it could be that just that first
1558 * byte is garbled, and the intended character occupies fewer bytes. If we
1559 * blindly assumed that the first byte is correct, and skipped based on
1560 * that number, we could skip over a valid input character. So instead, we
1561 * always examine the sequence byte-by-byte.
1562 *
1563 * We also should not consume too few bytes, otherwise someone could inject
1564 * things. For example, an input could be deliberately designed to
1565 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1566 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1567 * which is actually part of of the overflowing sequence), that could look
1568 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1569 * sequence and process the rest, inappropriately.
1570 *
1571 * Some possible input sequences are malformed in more than one way. This
1572 * function goes to lengths to try to find all of them. This is necessary
1573 * for correctness, as the inputs may allow one malformation but not
1574 * another, and if we abandon searching for others after finding the
1575 * allowed one, we could allow in something that shouldn't have been.
1576 */
eb83ed87 1577
b5b9af04 1578 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1579 possible_problems |= UTF8_GOT_EMPTY;
1580 curlen = 0;
5a48568d 1581 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1582 goto ready_to_handle_errors;
0c443dc2
JH
1583 }
1584
eb83ed87
KW
1585 expectlen = UTF8SKIP(s);
1586
1587 /* A well-formed UTF-8 character, as the vast majority of calls to this
1588 * function will be for, has this expected length. For efficiency, set
1589 * things up here to return it. It will be overriden only in those rare
1590 * cases where a malformation is found */
1591 if (retlen) {
1592 *retlen = expectlen;
1593 }
1594
eb83ed87 1595 /* A continuation character can't start a valid sequence */
b5b9af04 1596 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1597 possible_problems |= UTF8_GOT_CONTINUATION;
1598 curlen = 1;
1599 uv = UNICODE_REPLACEMENT;
1600 goto ready_to_handle_errors;
ba210ebe 1601 }
9041c2e3 1602
dcd27b3c 1603 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1604 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1605 * because it excludes start bytes like \xC0 that always lead to
1606 * overlongs.) */
dcd27b3c 1607
534752c1
KW
1608 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1609 * that indicate the number of bytes in the character's whole UTF-8
1610 * sequence, leaving just the bits that are part of the value. */
1611 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1612
e308b348
KW
1613 /* Setup the loop end point, making sure to not look past the end of the
1614 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1615 if (UNLIKELY(curlen < expectlen)) {
1616 possible_problems |= UTF8_GOT_SHORT;
1617 avail_len = curlen;
e308b348
KW
1618 }
1619 else {
2b9519f0 1620 send = (U8*) s0 + expectlen;
e308b348 1621 }
e308b348 1622
eb83ed87 1623 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1624 * accumulating each into the working value as we go. */
eb83ed87 1625 for (s = s0 + 1; s < send; s++) {
b5b9af04 1626 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1627 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1628 continue;
1629 }
1630
1631 /* Here, found a non-continuation before processing all expected bytes.
1632 * This byte indicates the beginning of a new character, so quit, even
1633 * if allowing this malformation. */
2b5e7bc2 1634 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1635 break;
eb83ed87
KW
1636 } /* End of loop through the character's bytes */
1637
1638 /* Save how many bytes were actually in the character */
1639 curlen = s - s0;
1640
2b5e7bc2
KW
1641 /* Note that there are two types of too-short malformation. One is when
1642 * there is actual wrong data before the normal termination of the
1643 * sequence. The other is that the sequence wasn't complete before the end
1644 * of the data we are allowed to look at, based on the input 'curlen'.
1645 * This means that we were passed data for a partial character, but it is
1646 * valid as far as we saw. The other is definitely invalid. This
1647 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1648 * separate.
1649 *
1650 * A convenience macro that matches either of the too-short conditions. */
1651# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1652
1653 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1654 uv_so_far = uv;
1655 uv = UNICODE_REPLACEMENT;
1656 }
2b5e7bc2 1657
08e73697
KW
1658 /* Check for overflow. The algorithm requires us to not look past the end
1659 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1660 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1661 1 /* Do consider overlongs */
1662 )))
1663 {
2b5e7bc2
KW
1664 possible_problems |= UTF8_GOT_OVERFLOW;
1665 uv = UNICODE_REPLACEMENT;
eb83ed87 1666 }
eb83ed87 1667
2b5e7bc2
KW
1668 /* Check for overlong. If no problems so far, 'uv' is the correct code
1669 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1670 * we must look at the UTF-8 byte sequence itself to see if it is for an
1671 * overlong */
1672 if ( ( LIKELY(! possible_problems)
1673 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1674 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1675 && ( UNLIKELY(! UTF8_IS_START(*s0))
1676 || ( curlen > 1
d6be65ae 1677 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1678 s - s0))))))
2f8f112e 1679 {
2b5e7bc2
KW
1680 possible_problems |= UTF8_GOT_LONG;
1681
abc28b54 1682 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1683
abc28b54
KW
1684 /* The calculation in the 'true' branch of this 'if'
1685 * below won't work if overflows, and isn't needed
1686 * anyway. Further below we handle all overflow
1687 * cases */
1688 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1689 {
2b5e7bc2
KW
1690 UV min_uv = uv_so_far;
1691 STRLEN i;
1692
1693 /* Here, the input is both overlong and is missing some trailing
1694 * bytes. There is no single code point it could be for, but there
1695 * may be enough information present to determine if what we have
1696 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1697 * The code further below has the intelligence to determine this,
1698 * but just for non-overlong UTF-8 sequences. What we do here is
1699 * calculate the smallest code point the input could represent if
1700 * there were no too short malformation. Then we compute and save
1701 * the UTF-8 for that, which is what the code below looks at
1702 * instead of the raw input. It turns out that the smallest such
1703 * code point is all we need. */
2b5e7bc2
KW
1704 for (i = curlen; i < expectlen; i++) {
1705 min_uv = UTF8_ACCUMULATE(min_uv,
1706 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1707 }
1708
e9f2c446 1709 adjusted_s0 = temp_char_buf;
57ff5f59 1710 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1711 }
eb83ed87
KW
1712 }
1713
56576a04
KW
1714 /* Here, we have found all the possible problems, except for when the input
1715 * is for a problematic code point not allowed by the input parameters. */
1716
06188866
KW
1717 /* uv is valid for overlongs */
1718 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1719
1720 /* isn't problematic if < this */
1721 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1722 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1723
1724 /* if overflow, we know without looking further
1725 * precisely which of the problematic types it is,
1726 * and we deal with those in the overflow handling
1727 * code */
1728 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1729 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1730 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1731 && ((flags & ( UTF8_DISALLOW_NONCHAR
1732 |UTF8_DISALLOW_SURROGATE
1733 |UTF8_DISALLOW_SUPER
d044b7a7 1734 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1735 |UTF8_WARN_NONCHAR
1736 |UTF8_WARN_SURROGATE
1737 |UTF8_WARN_SUPER
d22ec717 1738 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1739 {
2b5e7bc2
KW
1740 /* If there were no malformations, or the only malformation is an
1741 * overlong, 'uv' is valid */
1742 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1743 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1744 possible_problems |= UTF8_GOT_SURROGATE;
1745 }
1746 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1747 possible_problems |= UTF8_GOT_SUPER;
1748 }
1749 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1750 possible_problems |= UTF8_GOT_NONCHAR;
1751 }
1752 }
1753 else { /* Otherwise, need to look at the source UTF-8, possibly
1754 adjusted to be non-overlong */
1755
1756 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1757 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1758 {
2b5e7bc2
KW
1759 possible_problems |= UTF8_GOT_SUPER;
1760 }
1761 else if (curlen > 1) {
1762 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1763 NATIVE_UTF8_TO_I8(*adjusted_s0),
1764 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1765 {
2b5e7bc2 1766 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1767 }
2b5e7bc2
KW
1768 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1769 NATIVE_UTF8_TO_I8(*adjusted_s0),
1770 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1771 {
1772 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1773 }
1774 }
c0236afe 1775
2b5e7bc2
KW
1776 /* We need a complete well-formed UTF-8 character to discern
1777 * non-characters, so can't look for them here */
1778 }
1779 }
949cf498 1780
2b5e7bc2
KW
1781 ready_to_handle_errors:
1782
1783 /* At this point:
1784 * curlen contains the number of bytes in the sequence that
1785 * this call should advance the input by.
e308b348
KW
1786 * avail_len gives the available number of bytes passed in, but
1787 * only if this is less than the expected number of
1788 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1789 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1790 * is set in it for each potential problem found.
1791 * uv contains the code point the input sequence
1792 * represents; or if there is a problem that prevents
1793 * a well-defined value from being computed, it is
1794 * some subsitute value, typically the REPLACEMENT
1795 * CHARACTER.
1796 * s0 points to the first byte of the character
56576a04
KW
1797 * s points to just after were we left off processing
1798 * the character
1799 * send points to just after where that character should
1800 * end, based on how many bytes the start byte tells
1801 * us should be in it, but no further than s0 +
1802 * avail_len
2b5e7bc2 1803 */
eb83ed87 1804
2b5e7bc2
KW
1805 if (UNLIKELY(possible_problems)) {
1806 bool disallowed = FALSE;
1807 const U32 orig_problems = possible_problems;
1808
37657a5b
KW
1809 if (msgs) {
1810 *msgs = NULL;
1811 }
1812
2b5e7bc2
KW
1813 while (possible_problems) { /* Handle each possible problem */
1814 UV pack_warn = 0;
1815 char * message = NULL;
37657a5b 1816 U32 this_flag_bit = 0;
2b5e7bc2
KW
1817
1818 /* Each 'if' clause handles one problem. They are ordered so that
1819 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1820 * ones; this is kinda in decreasing severity order. But the
1821 * overlong must come last, as it changes 'uv' looked at by the
1822 * others */
2b5e7bc2
KW
1823 if (possible_problems & UTF8_GOT_OVERFLOW) {
1824
56576a04
KW
1825 /* Overflow means also got a super and are using Perl's
1826 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1827 possible_problems
d044b7a7 1828 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1829 *errors |= UTF8_GOT_OVERFLOW;
1830
1831 /* But the API says we flag all errors found */
1832 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1833 *errors |= UTF8_GOT_SUPER;
1834 }
ddb65933 1835 if (flags
d044b7a7 1836 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1837 {
d044b7a7 1838 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1839 }
2b5e7bc2 1840
d60baaa7 1841 /* Disallow if any of the three categories say to */
56576a04 1842 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1843 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1844 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1845 {
1846 disallowed = TRUE;
1847 }
1848
d22ec717
KW
1849 /* Likewise, warn if any say to */
1850 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1851 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1852 {
2b5e7bc2 1853
ddb65933
KW
1854 /* The warnings code explicitly says it doesn't handle the
1855 * case of packWARN2 and two categories which have
1856 * parent-child relationship. Even if it works now to
1857 * raise the warning if either is enabled, it wouldn't
1858 * necessarily do so in the future. We output (only) the
56576a04 1859 * most dire warning */
ddb65933 1860 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1861 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1862 pack_warn = packWARN(WARN_UTF8);
1863 }
37657a5b 1864 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1865 pack_warn = packWARN(WARN_NON_UNICODE);
1866 }
1867 if (pack_warn) {
1868 message = Perl_form(aTHX_ "%s: %s (overflows)",
1869 malformed_text,
05b9033b 1870 _byte_dump_string(s0, curlen, 0));
37657a5b 1871 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1872 }
2b5e7bc2
KW
1873 }
1874 }
1875 }
1876 else if (possible_problems & UTF8_GOT_EMPTY) {
1877 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1878 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1879
1880 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1881
1882 /* This so-called malformation is now treated as a bug in
1883 * the caller. If you have nothing to decode, skip calling
1884 * this function */
1885 assert(0);
1886
2b5e7bc2 1887 disallowed = TRUE;
37657a5b
KW
1888 if ( (msgs
1889 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1890 {
2b5e7bc2
KW
1891 pack_warn = packWARN(WARN_UTF8);
1892 message = Perl_form(aTHX_ "%s (empty string)",
1893 malformed_text);
37657a5b 1894 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1895 }
1896 }
1897 }
1898 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1899 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1900 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1901
1902 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1903 disallowed = TRUE;
37657a5b
KW
1904 if (( msgs
1905 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1906 {
2b5e7bc2
KW
1907 pack_warn = packWARN(WARN_UTF8);
1908 message = Perl_form(aTHX_
1909 "%s: %s (unexpected continuation byte 0x%02x,"
1910 " with no preceding start byte)",
1911 malformed_text,
7e2f38b2 1912 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1913 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1914 }
1915 }
1916 }
2b5e7bc2
KW
1917 else if (possible_problems & UTF8_GOT_SHORT) {
1918 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1919 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1920
1921 if (! (flags & UTF8_ALLOW_SHORT)) {
1922 disallowed = TRUE;
37657a5b
KW
1923 if (( msgs
1924 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1925 {
2b5e7bc2
KW
1926 pack_warn = packWARN(WARN_UTF8);
1927 message = Perl_form(aTHX_
56576a04
KW
1928 "%s: %s (too short; %d byte%s available, need %d)",
1929 malformed_text,
1930 _byte_dump_string(s0, send - s0, 0),
1931 (int)avail_len,
1932 avail_len == 1 ? "" : "s",
1933 (int)expectlen);
37657a5b 1934 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1935 }
1936 }
ba210ebe 1937
2b5e7bc2 1938 }
e308b348
KW
1939 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1940 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1941 *errors |= UTF8_GOT_NON_CONTINUATION;
1942
1943 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1944 disallowed = TRUE;
37657a5b
KW
1945 if (( msgs
1946 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1947 {
99a765e9
KW
1948
1949 /* If we don't know for sure that the input length is
1950 * valid, avoid as much as possible reading past the
1951 * end of the buffer */
1952 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1953 ? s - s0
1954 : send - s0;
e308b348
KW
1955 pack_warn = packWARN(WARN_UTF8);
1956 message = Perl_form(aTHX_ "%s",
1957 unexpected_non_continuation_text(s0,
99a765e9 1958 printlen,
e308b348
KW
1959 s - s0,
1960 (int) expectlen));
37657a5b 1961 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
1962 }
1963 }
1964 }
2b5e7bc2
KW
1965 else if (possible_problems & UTF8_GOT_SURROGATE) {
1966 possible_problems &= ~UTF8_GOT_SURROGATE;
1967
f9380377
KW
1968 if (flags & UTF8_WARN_SURROGATE) {
1969 *errors |= UTF8_GOT_SURROGATE;
1970
1971 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 1972 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 1973 {
2b5e7bc2
KW
1974 pack_warn = packWARN(WARN_SURROGATE);
1975
1976 /* These are the only errors that can occur with a
1977 * surrogate when the 'uv' isn't valid */
1978 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1979 message = Perl_form(aTHX_
1980 "UTF-16 surrogate (any UTF-8 sequence that"
1981 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1982 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1983 }
1984 else {
c94c2f39 1985 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 1986 }
37657a5b 1987 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 1988 }
2b5e7bc2 1989 }
ba210ebe 1990
2b5e7bc2
KW
1991 if (flags & UTF8_DISALLOW_SURROGATE) {
1992 disallowed = TRUE;
f9380377 1993 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1994 }
1995 }
1996 else if (possible_problems & UTF8_GOT_SUPER) {
1997 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1998
f9380377
KW
1999 if (flags & UTF8_WARN_SUPER) {
2000 *errors |= UTF8_GOT_SUPER;
2001
2002 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2003 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2004 {
2b5e7bc2
KW
2005 pack_warn = packWARN(WARN_NON_UNICODE);
2006
2007 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2008 message = Perl_form(aTHX_
2009 "Any UTF-8 sequence that starts with"
2010 " \"%s\" is for a non-Unicode code point,"
2011 " may not be portable",
7e2f38b2 2012 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2013 }
2014 else {
c94c2f39 2015 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2016 }
37657a5b 2017 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2018 }
2b5e7bc2 2019 }
ba210ebe 2020
57ff5f59
KW
2021 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2022 * and before possibly bailing out, so that the more dire
2023 * warning will override the regular one. */
2024 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2025 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2026 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2027 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2028 {
db0f09e6 2029 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2030
57ff5f59
KW
2031 /* If it is an overlong that evaluates to a code point
2032 * that doesn't have to use the Perl extended UTF-8, it
2033 * still used it, and so we output a message that
2034 * doesn't refer to the code point. The same is true
2035 * if there was a SHORT malformation where the code
2036 * point is not valid. In that case, 'uv' will have
2037 * been set to the REPLACEMENT CHAR, and the message
2038 * below without the code point in it will be selected
2039 * */
2040 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2041 message = Perl_form(aTHX_
57ff5f59 2042 perl_extended_cp_format, uv);
2b5e7bc2
KW
2043 }
2044 else {
2045 message = Perl_form(aTHX_
57ff5f59
KW
2046 "Any UTF-8 sequence that starts with"
2047 " \"%s\" is a Perl extension, and"
2048 " so is not portable",
2049 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2050 }
37657a5b 2051 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2052 }
2053
d044b7a7
KW
2054 if (flags & ( UTF8_WARN_PERL_EXTENDED
2055 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2056 {
d044b7a7 2057 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2058
d044b7a7 2059 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2060 disallowed = TRUE;
2061 }
2b5e7bc2
KW
2062 }
2063 }
eb83ed87 2064
2b5e7bc2 2065 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2066 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2067 disallowed = TRUE;
2068 }
2b5e7bc2
KW
2069 }
2070 else if (possible_problems & UTF8_GOT_NONCHAR) {
2071 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2072
f9380377
KW
2073 if (flags & UTF8_WARN_NONCHAR) {
2074 *errors |= UTF8_GOT_NONCHAR;
2075
2076 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2077 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2078 {
2b5e7bc2
KW
2079 /* The code above should have guaranteed that we don't
2080 * get here with errors other than overlong */
2081 assert (! (orig_problems
2082 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2083
2084 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2085 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2086 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2087 }
2b5e7bc2 2088 }
5b311467 2089
2b5e7bc2
KW
2090 if (flags & UTF8_DISALLOW_NONCHAR) {
2091 disallowed = TRUE;
f9380377 2092 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2093 }
6c64cd9d
KW
2094 }
2095 else if (possible_problems & UTF8_GOT_LONG) {
2096 possible_problems &= ~UTF8_GOT_LONG;
2097 *errors |= UTF8_GOT_LONG;
2098
2099 if (flags & UTF8_ALLOW_LONG) {
2100
2101 /* We don't allow the actual overlong value, unless the
2102 * special extra bit is also set */
2103 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2104 & ~UTF8_ALLOW_LONG)))
2105 {
2106 uv = UNICODE_REPLACEMENT;
2107 }
2108 }
2109 else {
2110 disallowed = TRUE;
2111
37657a5b
KW
2112 if (( msgs
2113 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2114 {
6c64cd9d
KW
2115 pack_warn = packWARN(WARN_UTF8);
2116
2117 /* These error types cause 'uv' to be something that
2118 * isn't what was intended, so can't use it in the
2119 * message. The other error types either can't
2120 * generate an overlong, or else the 'uv' is valid */
2121 if (orig_problems &
2122 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2123 {
2124 message = Perl_form(aTHX_
2125 "%s: %s (any UTF-8 sequence that starts"
2126 " with \"%s\" is overlong which can and"
2127 " should be represented with a"
2128 " different, shorter sequence)",
2129 malformed_text,
2130 _byte_dump_string(s0, send - s0, 0),
2131 _byte_dump_string(s0, curlen, 0));
2132 }
2133 else {
2134 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2135 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2136 uv, 0);
d819dc50
KW
2137 /* Don't use U+ for non-Unicode code points, which
2138 * includes those in the Latin1 range */
2139 const char * preface = ( uv > PERL_UNICODE_MAX
2140#ifdef EBCDIC
2141 || uv <= 0xFF
2142#endif
2143 )
2144 ? "0x"
2145 : "U+";
6c64cd9d
KW
2146 message = Perl_form(aTHX_
2147 "%s: %s (overlong; instead use %s to represent"
2148 " %s%0*" UVXf ")",
2149 malformed_text,
2150 _byte_dump_string(s0, send - s0, 0),
2151 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2152 preface,
2153 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2154 small code points */
1be62ab9 2155 UNI_TO_NATIVE(uv));
6c64cd9d 2156 }
37657a5b 2157 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2158 }
2159 }
2b5e7bc2
KW
2160 } /* End of looking through the possible flags */
2161
2162 /* Display the message (if any) for the problem being handled in
2163 * this iteration of the loop */
2164 if (message) {
37657a5b 2165 if (msgs) {
37657a5b
KW
2166 assert(this_flag_bit);
2167
2168 if (*msgs == NULL) {
2169 *msgs = newAV();
2170 }
2171
bb07812e
KW
2172 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2173 pack_warn,
2174 this_flag_bit)));
37657a5b
KW
2175 }
2176 else if (PL_op)
2b5e7bc2
KW
2177 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2178 OP_DESC(PL_op));
2179 else
2180 Perl_warner(aTHX_ pack_warn, "%s", message);
2181 }
ddb65933 2182 } /* End of 'while (possible_problems)' */
a0dbb045 2183
2b5e7bc2
KW
2184 /* Since there was a possible problem, the returned length may need to
2185 * be changed from the one stored at the beginning of this function.
2186 * Instead of trying to figure out if that's needed, just do it. */
2187 if (retlen) {
2188 *retlen = curlen;
2189 }
a0dbb045 2190
2b5e7bc2
KW
2191 if (disallowed) {
2192 if (flags & UTF8_CHECK_ONLY && retlen) {
2193 *retlen = ((STRLEN) -1);
2194 }
2195 return 0;
2196 }
eb83ed87 2197 }
ba210ebe 2198
2b5e7bc2 2199 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2200}
2201
8e84507e 2202/*
ec5f19d0
KW
2203=for apidoc utf8_to_uvchr_buf
2204
2205Returns the native code point of the first character in the string C<s> which
2206is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2207C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2208
524080c4
KW
2209If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2210enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2211C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2212(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2213C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2214the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2215See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2216returned.
ec5f19d0
KW
2217
2218=cut
52be2536
KW
2219
2220Also implemented as a macro in utf8.h
2221
ec5f19d0
KW
2222*/
2223
2224
2225UV
2226Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2227{
7f974d7e
KW
2228 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2229
ec5f19d0
KW
2230 assert(s < send);
2231
2232 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 2233 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
2234}
2235
52be2536
KW
2236/* This is marked as deprecated
2237 *
ec5f19d0
KW
2238=for apidoc utf8_to_uvuni_buf
2239
de69f3af
KW
2240Only in very rare circumstances should code need to be dealing in Unicode
2241(as opposed to native) code points. In those few cases, use
07dfe0a4
KW
2242C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2243are not absolutely sure this is one of those cases, then assume it isn't and
2244use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2245
2246Returns the Unicode (not-native) code point of the first character in the
2247string C<s> which
ec5f19d0
KW
2248is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2249C<retlen> will be set to the length, in bytes, of that character.
2250
524080c4
KW
2251If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2252enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2253NULL) to -1. If those warnings are off, the computed value if well-defined (or
2254the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2255is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2256next possible position in C<s> that could begin a non-malformed character.
de69f3af 2257See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2258
2259=cut
2260*/
2261
2262UV
2263Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2264{
2265 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2266
2267 assert(send > s);
2268
5962d97e 2269 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2270}
2271
b76347f2 2272/*
87cea99e 2273=for apidoc utf8_length
b76347f2 2274
b2e7ed74
KW
2275Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2276at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2277same place, it returns 0 with no warning raised.
2278
2279If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2280and returns the number of valid characters.
b76347f2
JH
2281
2282=cut
2283*/
2284
2285STRLEN
35a4481c 2286Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2287{
2288 STRLEN len = 0;
2289
7918f24d
NC
2290 PERL_ARGS_ASSERT_UTF8_LENGTH;
2291
8850bf83
JH
2292 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2293 * the bitops (especially ~) can create illegal UTF-8.
2294 * In other words: in Perl UTF-8 is not just for Unicode. */
2295
a3b680e6
AL
2296 if (e < s)
2297 goto warn_and_return;
b76347f2 2298 while (s < e) {
4cbf4130 2299 s += UTF8SKIP(s);
8e91ec7f
AV
2300 len++;
2301 }
2302
2303 if (e != s) {
2304 len--;
2305 warn_and_return:
9b387841
NC
2306 if (PL_op)
2307 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2308 "%s in %s", unees, OP_DESC(PL_op));
2309 else
61a12c31 2310 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2311 }
2312
2313 return len;
2314}
2315
b06226ff 2316/*
fed3ba5d
NC
2317=for apidoc bytes_cmp_utf8
2318
a1433954 2319Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2320sequence of characters (stored as UTF-8)
2321in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2322equal, -1 or -2 if the first string is less than the second string, +1 or +2
2323if the first string is greater than the second string.
2324
2325-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2326longer string. -2 or +2 is returned if
2327there was a difference between characters
fed3ba5d
NC
2328within the strings.
2329
2330=cut
2331*/
2332
2333int
2334Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2335{
2336 const U8 *const bend = b + blen;
2337 const U8 *const uend = u + ulen;
2338
2339 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2340
2341 while (b < bend && u < uend) {
2342 U8 c = *u++;
2343 if (!UTF8_IS_INVARIANT(c)) {
2344 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2345 if (u < uend) {
2346 U8 c1 = *u++;
2347 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2348 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2349 } else {
2b5e7bc2 2350 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2351 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2352 "%s %s%s",
2353 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2354 PL_op ? " in " : "",
2355 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2356 return -2;
2357 }
2358 } else {
2359 if (PL_op)
2360 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2361 "%s in %s", unees, OP_DESC(PL_op));
2362 else
61a12c31 2363 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2364 return -2; /* Really want to return undef :-) */
2365 }
2366 } else {
2367 return -2;
2368 }
2369 }
2370 if (*b != c) {
2371 return *b < c ? -2 : +2;
2372 }
2373 ++b;
2374 }
2375
2376 if (b == bend && u == uend)
2377 return 0;
2378
2379 return b < bend ? +1 : -1;
2380}
2381
2382/*
87cea99e 2383=for apidoc utf8_to_bytes
6940069f 2384
3bc0c78c 2385Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2386Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2387updates C<*lenp> to contain the new length.
3bc0c78c
KW
2388Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2389
2390Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2391having saved the value of C<*lenp> before the call, and subtracting the
2392after-call value of C<*lenp> from it.
6940069f 2393
a1433954 2394If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2395
6940069f
GS
2396=cut
2397*/
2398
2399U8 *
09af0336 2400Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2401{
9fe0d3c2 2402 U8 * first_variant;
246fae53 2403
7918f24d 2404 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2405 PERL_UNUSED_CONTEXT;
7918f24d 2406
9fe0d3c2 2407 /* This is a no-op if no variants at all in the input */
09af0336 2408 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2409 return s;
2410 }
2411
2412 {
3c5aa262 2413 U8 * const save = s;
09af0336 2414 U8 * const send = s + *lenp;
3c5aa262
KW
2415 U8 * d;
2416
2417 /* Nothing before the first variant needs to be changed, so start the real
2418 * work there */
2419 s = first_variant;
2420 while (s < send) {
2421 if (! UTF8_IS_INVARIANT(*s)) {
2422 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2423 *lenp = ((STRLEN) -1);
3c5aa262
KW
2424 return 0;
2425 }
2426 s++;
d59937ca
KW
2427 }
2428 s++;
dcad2880 2429 }
dcad2880 2430
3c5aa262
KW
2431 /* Is downgradable, so do it */
2432 d = s = first_variant;
2433 while (s < send) {
2434 U8 c = *s++;
2435 if (! UVCHR_IS_INVARIANT(c)) {
2436 /* Then it is two-byte encoded */
2437 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2438 s++;
2439 }
2440 *d++ = c;
2441 }
2442 *d = '\0';
09af0336 2443 *lenp = d - save;
3c5aa262
KW
2444
2445 return save;
9fe0d3c2 2446 }
6940069f
GS
2447}
2448
2449/*
87cea99e 2450=for apidoc bytes_from_utf8
f9a63242 2451
09af0336 2452Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2453byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2454actually encoded in UTF-8.
2455
2456Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2457the input string.
2458
41ae6089
KW
2459Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2460not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2461C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2462
41ae6089 2463Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2464newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2465returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2466caller is responsible for arranging for the memory used by this string to get
2467freed.
f9a63242 2468
3bc0c78c 2469Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2470having saved the value of C<*lenp> before the call, and subtracting the
2471after-call value of C<*lenp> from it.
3bc0c78c 2472
37607a96 2473=cut
976c1b08
KW
2474
2475There is a macro that avoids this function call, but this is retained for
2476anyone who calls it with the Perl_ prefix */
f9a63242
JH
2477
2478U8 *
41ae6089 2479Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2480{
7918f24d 2481 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2482 PERL_UNUSED_CONTEXT;
f9a63242 2483
976c1b08
KW
2484 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2485}
2486
2487/*
2488No = here because currently externally undocumented
2489for apidoc bytes_from_utf8_loc
2490
2491Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2492to store the location of the first character in C<"s"> that cannot be
2493converted to non-UTF8.
2494
2495If that parameter is C<NULL>, this function behaves identically to
2496C<bytes_from_utf8>.
2497
2498Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2499C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2500
2501Otherwise, the function returns a newly created C<NUL>-terminated string
2502containing the non-UTF8 equivalent of the convertible first portion of
2503C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2504If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2505and C<*first_non_downgradable> is set to C<NULL>.
2506
2507Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2508first character in the original string that wasn't converted. C<*is_utf8p> is
2509unchanged. Note that the new string may have length 0.
2510
2511Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2512C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2513converts as many characters in it as possible stopping at the first one it
385b74be 2514finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2515set to point to that. The function returns the portion that could be converted
2516in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2517not including the terminating C<NUL>. If the very first character in the
2518original could not be converted, C<*lenp> will be 0, and the new string will
2519contain just a single C<NUL>. If the entire input string was converted,
2520C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2521
2522Upon successful return, the number of variants in the converted portion of the
2523string can be computed by having saved the value of C<*lenp> before the call,
2524and subtracting the after-call value of C<*lenp> from it.
2525
2526=cut
2527
2528
2529*/
2530
2531U8 *
2532Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2533{
2534 U8 *d;
2535 const U8 *original = s;
2536 U8 *converted_start;
2537 const U8 *send = s + *lenp;
f9a63242 2538
976c1b08 2539 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2540
976c1b08
KW
2541 if (! *is_utf8p) {
2542 if (first_unconverted) {
2543 *first_unconverted = NULL;
2544 }
2545
2546 return (U8 *) original;
2547 }
2548
2549 Newx(d, (*lenp) + 1, U8);
2550
2551 converted_start = d;
7299a045
KW
2552 while (s < send) {
2553 U8 c = *s++;
2554 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2555
2556 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2557 * have to stop now */
2558 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2559 if (first_unconverted) {
2560 *first_unconverted = s - 1;
2561 goto finish_and_return;
2562 }
2563 else {
2564 Safefree(converted_start);
2565 return (U8 *) original;
2566 }
2567 }
2568
7299a045
KW
2569 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2570 s++;
38af28cf 2571 }
7299a045
KW
2572 *d++ = c;
2573 }
170a1c22 2574
976c1b08
KW
2575 /* Here, converted the whole of the input */
2576 *is_utf8p = FALSE;
2577 if (first_unconverted) {
2578 *first_unconverted = NULL;
170a1c22 2579 }
976c1b08
KW
2580
2581 finish_and_return:
46a08a6f
KW
2582 *d = '\0';
2583 *lenp = d - converted_start;
976c1b08
KW
2584
2585 /* Trim unused space */
2586 Renew(converted_start, *lenp + 1, U8);
2587
2588 return converted_start;
f9a63242
JH
2589}
2590
2591/*
87cea99e 2592=for apidoc bytes_to_utf8
6940069f 2593
09af0336 2594Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2595UTF-8.
09af0336 2596Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2597reflect the new length in bytes. The caller is responsible for arranging for
2598the memory used by this string to get freed.
6940069f 2599
3bc0c78c 2600Upon successful return, the number of variants in the string can be computed by
23b37b12 2601having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2602after-call value of C<*lenp>.
2603
75200dff 2604A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2605
2606If you want to convert to UTF-8 from encodings other than
2607the native (Latin1 or EBCDIC),
a1433954 2608see L</sv_recode_to_utf8>().
c9ada85f 2609
497711e7 2610=cut
6940069f
GS
2611*/
2612
2613U8*
09af0336 2614Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2615{
09af0336 2616 const U8 * const send = s + (*lenp);
6940069f
GS
2617 U8 *d;
2618 U8 *dst;
7918f24d
NC
2619
2620 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2621 PERL_UNUSED_CONTEXT;
6940069f 2622
09af0336 2623 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2624 dst = d;
2625
2626 while (s < send) {
55d09dc8
KW
2627 append_utf8_from_native_byte(*s, &d);
2628 s++;
6940069f 2629 }
2e11cf67 2630
6940069f 2631 *d = '\0';
09af0336 2632 *lenp = d-dst;
2e11cf67
KW
2633
2634 /* Trim unused space */
2635 Renew(dst, *lenp + 1, U8);
2636
6940069f
GS
2637 return dst;
2638}
2639
a0ed51b3 2640/*
624504c5
KW
2641 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2642 * use utf16_to_utf8_reversed().
a0ed51b3 2643 *
624504c5
KW
2644 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2645 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2646 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2647 *
2648 * These functions don't check for overflow. The worst case is every code
2649 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2650 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2651 * destination must be pre-extended to 2 times the source length.
2652 *
2653 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2654
2655U8*
dea0fc0b 2656Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2657{
dea0fc0b
JH
2658 U8* pend;
2659 U8* dstart = d;
2660
7918f24d
NC
2661 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2662
dea0fc0b 2663 if (bytelen & 1)
56576a04
KW
2664 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2665 (UV)bytelen);
dea0fc0b
JH
2666
2667 pend = p + bytelen;
2668
a0ed51b3 2669 while (p < pend) {
dea0fc0b
JH
2670 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2671 p += 2;
2d1545e5 2672 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2673 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2674 continue;
2675 }
56d37426
KW
2676 if (uv <= MAX_UTF8_TWO_BYTE) {
2677 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2678 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2679 continue;
2680 }
ffd0a9d3 2681
46956fad
KW
2682#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2683#define LAST_HIGH_SURROGATE 0xDBFF
2684#define FIRST_LOW_SURROGATE 0xDC00
2685#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2686#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2687
2688 /* This assumes that most uses will be in the first Unicode plane, not
2689 * needing surrogates */
2690 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2691 && uv <= UNICODE_SURROGATE_LAST))
2692 {
2693 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2694 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2695 }
2696 else {
01ea242b 2697 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2698 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2699 || UNLIKELY(low > LAST_LOW_SURROGATE))
2700 {
01ea242b 2701 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2702 }
2703 p += 2;
46956fad 2704 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2705 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2706 }
a0ed51b3 2707 }
56d37426
KW
2708#ifdef EBCDIC
2709 d = uvoffuni_to_utf8_flags(d, uv, 0);
2710#else
ffd0a9d3 2711 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2712 *d++ = (U8)(( uv >> 12) | 0xe0);
2713 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2714 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2715 continue;
2716 }
2717 else {
eb160463
GS
2718 *d++ = (U8)(( uv >> 18) | 0xf0);
2719 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2720 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2721 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2722 continue;
2723 }
56d37426 2724#endif
a0ed51b3 2725 }
dea0fc0b 2726 *newlen = d - dstart;
a0ed51b3
LW
2727 return d;
2728}
2729
2730/* Note: this one is slightly destructive of the source. */
2731
2732U8*
dea0fc0b 2733Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2734{
2735 U8* s = (U8*)p;
d4c19fe8 2736 U8* const send = s + bytelen;
7918f24d
NC
2737
2738 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2739
e0ea5e2d 2740 if (bytelen & 1)
147e3846 2741 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2742 (UV)bytelen);
2743
a0ed51b3 2744 while (s < send) {
d4c19fe8 2745 const U8 tmp = s[0];
a0ed51b3
LW
2746 s[0] = s[1];
2747 s[1] = tmp;
2748 s += 2;
2749 }
dea0fc0b 2750 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2751}
2752
922e8cb4
KW
2753bool
2754Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2755{
dc31b55c 2756 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2757}
2758
f9ae8fb6
JD
2759/* Internal function so we can deprecate the external one, and call
2760 this one from other deprecated functions in this file */
2761
f2645549
KW
2762bool
2763Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2764{
f2645549 2765 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2766
2767 if (*p == '_')
2768 return TRUE;
af0d7385
KW
2769 return is_utf8_common(p, NULL,
2770 "This is buggy if this gets used",
2771 PL_utf8_idstart);
61b19385
KW
2772}
2773
5092f92a 2774bool
eba68aa0
KW
2775Perl__is_uni_perl_idcont(pTHX_ UV c)
2776{
c12658c9 2777 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2778}
2779
2780bool
f91dcd13
KW
2781Perl__is_uni_perl_idstart(pTHX_ UV c)
2782{
c12658c9 2783 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2784}
2785
3a4c58c9 2786UV
56576a04
KW
2787Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2788 const char S_or_s)
3a4c58c9
KW
2789{
2790 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2791 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2792 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2793 * either "SS" or "Ss". Which one to use is passed into the routine in
2794 * 'S_or_s' to avoid a test */
2795
2796 UV converted = toUPPER_LATIN1_MOD(c);
2797
2798 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2799
2800 assert(S_or_s == 'S' || S_or_s == 's');
2801
6f2d5cbc 2802 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2803 characters in this range */
3a4c58c9
KW
2804 *p = (U8) converted;
2805 *lenp = 1;
2806 return converted;
2807 }
2808
2809 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2810 * which it maps to one of them, so as to only have to have one check for
2811 * it in the main case */
2812 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2813 switch (c) {
2814 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2815 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2816 break;
2817 case MICRO_SIGN:
2818 converted = GREEK_CAPITAL_LETTER_MU;
2819 break;
79e064b9
KW
2820#if UNICODE_MAJOR_VERSION > 2 \
2821 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2822 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2823 case LATIN_SMALL_LETTER_SHARP_S:
2824 *(p)++ = 'S';
2825 *p = S_or_s;
2826 *lenp = 2;
2827 return 'S';
79e064b9 2828#endif
3a4c58c9 2829 default:
56576a04
KW
2830 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2831 " '%c' to map to '%c'",
2832 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2833 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2834 }
2835 }
2836
2837 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2838 *p = UTF8_TWO_BYTE_LO(converted);
2839 *lenp = 2;
2840
2841 return converted;
2842}
2843
fe63c520
KW
2844/* If compiled on an early Unicode version, there may not be auxiliary tables
2845 * */
2846#ifndef HAS_UC_AUX_TABLES
2847# define UC_AUX_TABLE_ptrs NULL
2848# define UC_AUX_TABLE_lengths NULL
2849#endif
2850#ifndef HAS_TC_AUX_TABLES
2851# define TC_AUX_TABLE_ptrs NULL
2852# define TC_AUX_TABLE_lengths NULL
2853#endif
2854#ifndef HAS_LC_AUX_TABLES
2855# define LC_AUX_TABLE_ptrs NULL
2856# define LC_AUX_TABLE_lengths NULL
2857#endif
2858#ifndef HAS_CF_AUX_TABLES
2859# define CF_AUX_TABLE_ptrs NULL
2860# define CF_AUX_TABLE_lengths NULL
2861#endif
2862#ifndef HAS_UC_AUX_TABLES
2863# define UC_AUX_TABLE_ptrs NULL
2864# define UC_AUX_TABLE_lengths NULL
2865#endif
2866
50bda2c3
KW
2867/* Call the function to convert a UTF-8 encoded character to the specified case.
2868 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2869 * 's' is a pointer to the first byte of the input character
2870 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2871 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2872 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2873 *
56576a04 2874 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2875 * 'd' */
56576a04 2876#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2877 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2878 Uppercase_Mapping_invmap, \
2879 UC_AUX_TABLE_ptrs, \
2880 UC_AUX_TABLE_lengths, \
2881 "uppercase")
56576a04 2882#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2883 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2884 Titlecase_Mapping_invmap, \
2885 TC_AUX_TABLE_ptrs, \
2886 TC_AUX_TABLE_lengths, \
2887 "titlecase")
56576a04 2888#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2889 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2890 Lowercase_Mapping_invmap, \
2891 LC_AUX_TABLE_ptrs, \
2892 LC_AUX_TABLE_lengths, \
2893 "lowercase")
2894
50bda2c3 2895
b9992569
KW
2896/* This additionally has the input parameter 'specials', which if non-zero will
2897 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2898 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2899#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2900 (specials) \
2901 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2902 Case_Folding_invmap, \
2903 CF_AUX_TABLE_ptrs, \
2904 CF_AUX_TABLE_lengths, \
2905 "foldcase") \
2906 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2907 Simple_Case_Folding_invmap, \
2908 NULL, NULL, \
2909 "foldcase")
c3fd2246 2910
84afefe6
JH
2911UV
2912Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2913{
a1433954
KW
2914 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2915 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2916 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2917 * the changed version may be longer than the original character.
2918 *
2919 * The ordinal of the first character of the changed version is returned
2920 * (but note, as explained above, that there may be more.) */
2921
7918f24d
NC
2922 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2923
3a4c58c9
KW
2924 if (c < 256) {
2925 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2926 }
2927
a13f1de4 2928 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2929}
2930
84afefe6
JH
2931UV
2932Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2933{
7918f24d
NC
2934 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2935
3a4c58c9
KW
2936 if (c < 256) {
2937 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2938 }
2939
a13f1de4 2940 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2941}
2942
afc16117 2943STATIC U8
eaf412bf 2944S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2945{
2946 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2947 * those, converting the result to UTF-8. Since the result is always just
a1433954 2948 * one character, we allow <p> to be NULL */
afc16117
KW
2949
2950 U8 converted = toLOWER_LATIN1(c);
2951
eaf412bf
KW
2952 PERL_UNUSED_ARG(dummy);
2953
afc16117 2954 if (p != NULL) {
6f2d5cbc 2955 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2956 *p = converted;
2957 *lenp = 1;
2958 }
2959 else {
430c9760
KW
2960 /* Result is known to always be < 256, so can use the EIGHT_BIT
2961 * macros */
2962 *p = UTF8_EIGHT_BIT_HI(converted);
2963 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2964 *lenp = 2;
2965 }
2966 }
2967 return converted;
2968}
2969
84afefe6
JH
2970UV
2971Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2972{
7918f24d
NC
2973 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2974
afc16117 2975 if (c < 256) {
eaf412bf 2976 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2977 }
2978
a13f1de4 2979 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2980}
2981
84afefe6 2982UV
7c0ab950 2983Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2984{
51910141 2985 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2986 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2987 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2988 *
2989 * Not to be used for locale folds
51910141 2990 */
f673fad4 2991
a1dde8de
KW
2992 UV converted;
2993
2994 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2995
1ca267a5
KW
2996 assert (! (flags & FOLD_FLAGS_LOCALE));
2997
659a7c2d 2998 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2999 converted = GREEK_SMALL_LETTER_MU;
3000 }
9b63e895
KW
3001#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3002 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3003 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3004 else if ( (flags & FOLD_FLAGS_FULL)
3005 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3006 {
1ca267a5
KW
3007 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3008 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3009 * under those circumstances. */
3010 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3011 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3012 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3013 p, *lenp, U8);
3014 return LATIN_SMALL_LETTER_LONG_S;
3015 }
3016 else {
4f489194
KW
3017 *(p)++ = 's';
3018 *p = 's';
3019 *lenp = 2;
3020 return 's';
1ca267a5 3021 }
a1dde8de 3022 }
9b63e895 3023#endif
a1dde8de
KW
3024 else { /* In this range the fold of all other characters is their lower
3025 case */
3026 converted = toLOWER_LATIN1(c);
3027 }
3028
6f2d5cbc 3029 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3030 *p = (U8) converted;
3031 *lenp = 1;
3032 }
3033 else {
3034 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3035 *p = UTF8_TWO_BYTE_LO(converted);
3036 *lenp = 2;
3037 }
3038
3039 return converted;
3040}
3041
3042UV
31f05a37 3043Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3044{
4b593389 3045
a0270393
KW
3046 /* Not currently externally documented, and subject to change
3047 * <flags> bits meanings:
3048 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3049 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3050 * locale are to be used.
a0270393
KW
3051 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3052 */
4b593389 3053
36bb2ab6 3054 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3055
780fcc9f 3056 if (flags & FOLD_FLAGS_LOCALE) {
8b7358b9
KW
3057 /* Treat a UTF-8 locale as not being in locale at all, except for
3058 * potentially warning */
3059 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
780fcc9f
KW
3060 if (IN_UTF8_CTYPE_LOCALE) {
3061 flags &= ~FOLD_FLAGS_LOCALE;
3062 }
3063 else {
e7b7ac46 3064 goto needs_full_generality;
780fcc9f 3065 }
31f05a37
KW
3066 }
3067
a1dde8de 3068 if (c < 256) {
e7b7ac46 3069 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3070 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3071 }
3072
2f306ab9 3073 /* Here, above 255. If no special needs, just use the macro */
a0270393 3074 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3075 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3076 }
567b353c 3077 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3078 the special flags. */
3079 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3080
3081 needs_full_generality:
a0270393 3082 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3083 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3084 p, lenp, flags);
a0270393 3085 }
84afefe6
JH
3086}
3087
26483009 3088PERL_STATIC_INLINE bool
5141f98e 3089S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 3090 const char *const swashname, SV* const invlist)
bde6a22d 3091{
ea317ccb
KW
3092 /* returns a boolean giving whether or not the UTF8-encoded character that
3093 * starts at <p> is in the swash indicated by <swashname>. <swash>
3094 * contains a pointer to where the swash indicated by <swashname>
3095 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
3096 * look at <*swash> and only generate a swash if it is not null. <invlist>
3097 * is NULL or an inversion list that defines the swash. If not null, it
3098 * saves time during initialization of the swash.
ea317ccb
KW
3099 *
3100 * Note that it is assumed that the buffer length of <p> is enough to
3101 * contain all the bytes that comprise the character. Thus, <*p> should
3102 * have been checked before this call for mal-formedness enough to assure
3103 * that. */
3104
7918f24d
NC
3105 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3106
492a624f 3107 /* The API should have included a length for the UTF-8 character in <p>,
28123549 3108 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
3109 * as far as there being enough bytes available in it to accommodate the
3110 * character without reading beyond the end, and pass that number on to the
3111 * validating routine */
6302f837 3112 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 3113 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 3114 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
3115 1 /* Die */ );
3116 NOT_REACHED; /* NOTREACHED */
28123549 3117 }
86ae6e94 3118
dc31b55c
KW
3119 if (invlist) {
3120 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3121 }
3122
af0d7385
KW
3123 assert(swash);
3124
87367d5f
KW
3125 if (!*swash) {
3126 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
3127 *swash = _core_swash_init("utf8",
3128
3129 /* Only use the name if there is no inversion
3130 * list; otherwise will go out to disk */
3131 (invlist) ? "" : swashname,
3132
3133 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 3134 }
28123549 3135
bde6a22d
NC
3136 return swash_fetch(*swash, p, TRUE) != 0;
3137}
3138
da8c1a98 3139PERL_STATIC_INLINE bool
56576a04
KW
3140S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3141 SV **swash, const char *const swashname,
3142 SV* const invlist)
da8c1a98
KW
3143{
3144 /* returns a boolean giving whether or not the UTF8-encoded character that
3145 * starts at <p>, and extending no further than <e - 1> is in the swash
3146 * indicated by <swashname>. <swash> contains a pointer to where the swash
3147 * indicated by <swashname> is to be stored; which this routine will do, so
3148 * that future calls will look at <*swash> and only generate a swash if it
3149 * is not null. <invlist> is NULL or an inversion list that defines the
3150 * swash. If not null, it saves time during initialization of the swash.
3151 */
3152
3153 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3154
3155 if (! isUTF8_CHAR(p, e)) {
3156 _force_out_malformed_utf8_message(p, e, 0, 1);
3157 NOT_REACHED; /* NOTREACHED */
3158 }
3159
dc31b55c
KW
3160 if (invlist) {
3161 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3162 }
3163
af0d7385
KW
3164 assert(swash);
3165
da8c1a98
KW
3166 if (!*swash) {
3167 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3168 *swash = _core_swash_init("utf8",
3169
3170 /* Only use the name if there is no inversion
3171 * list; otherwise will go out to disk */
3172 (invlist) ? "" : swashname,
3173
3174 &PL_sv_undef, 1, 0, invlist, &flags);
3175 }
3176
3177 return swash_fetch(*swash, p, TRUE) != 0;
3178}
3179
34aeb2e9
KW
3180STATIC void
3181S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3182 const char * const alternative,
3183 const bool use_locale,
3184 const char * const file,
3185 const unsigned line)
3186{
3187 const char * key;
3188
3189 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3190
3191 if (ckWARN_d(WARN_DEPRECATED)) {
3192
3193 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3194 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3195 if (! PL_seen_deprecated_macro) {
3196 PL_seen_deprecated_macro = newHV();
3197 }
3198 if (! hv_store(PL_seen_deprecated_macro, key,
3199 strlen(key), &PL_sv_undef, 0))
3200 {
3201 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3202 }
3203
c44e9413 3204 if (instr(file, "mathoms.c")) {
607313a1
KW
3205 Perl_warner(aTHX_ WARN_DEPRECATED,
3206 "In %s, line %d, starting in Perl v5.30, %s()"
3207 " will be removed. Avoid this message by"
3208 " converting to use %s().\n",
3209 file, line, name, alternative);
3210 }
3211 else {
34aeb2e9
KW
3212 Perl_warner(aTHX_ WARN_DEPRECATED,
3213 "In %s, line %d, starting in Perl v5.30, %s() will"
3214 " require an additional parameter. Avoid this"
3215 " message by converting to use %s().\n",
3216 file, line, name, alternative);
607313a1 3217 }
34aeb2e9
KW
3218 }
3219 }
3220}
3221
bde6a22d 3222bool
34aeb2e9 3223Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3224 const U8 * const p,
34aeb2e9
KW
3225 const char * const name,
3226 const char * const alternative,
3227 const bool use_utf8,
3228 const bool use_locale,
3229 const char * const file,
3230 const unsigned line)
922e8cb4 3231{
922e8cb4
KW
3232 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3233
34aeb2e9
KW
3234 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3235
3236 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3237
3238 switch (classnum) {
3239 case _CC_WORDCHAR:
3240 case _CC_DIGIT:
3241 case _CC_ALPHA:
3242 case _CC_LOWER:
3243 case _CC_UPPER:
3244 case _CC_PUNCT:
3245 case _CC_PRINT:
3246 case _CC_ALPHANUMERIC:
3247 case _CC_GRAPH:
3248 case _CC_CASED:
3249
3250 return is_utf8_common(p,
af0d7385
KW
3251 NULL,
3252 "This is buggy if this gets used",
34aeb2e9
KW
3253 PL_XPosix_ptrs[classnum]);
3254
3255 case _CC_SPACE:
3256 return is_XPERLSPACE_high(p);
3257 case _CC_BLANK:
3258 return is_HORIZWS_high(p);
3259 case _CC_XDIGIT:
3260 return is_XDIGIT_high(p);
3261 case _CC_CNTRL:
3262 return 0;
3263 case _CC_ASCII:
3264 return 0;
3265 case _CC_VERTSPACE:
3266 return is_VERTWS_high(p);
3267 case _CC_IDFIRST:
af0d7385
KW
3268 return is_utf8_common(p, NULL,
3269 "This is buggy if this gets used",
3270 PL_utf8_perl_idstart);
34aeb2e9 3271 case _CC_IDCONT:
af0d7385
KW
3272 return is_utf8_common(p, NULL,
3273 "This is buggy if this gets used",
3274 PL_utf8_perl_idcont);
34aeb2e9
KW
3275 }
3276 }
3277
3278 /* idcont is the same as wordchar below 256 */
3279 if (classnum == _CC_IDCONT) {
3280 classnum = _CC_WORDCHAR;
3281 }
3282 else if (classnum == _CC_IDFIRST) {
3283 if (*p == '_') {
3284 return TRUE;
3285 }
3286 classnum = _CC_ALPHA;
3287 }
3288
3289 if (! use_locale) {
3290 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3291 return _generic_isCC(*p, classnum);
3292 }
922e8cb4 3293
34aeb2e9
KW
3294 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3295 }
3296 else {
3297 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3298 return isFOO_lc(classnum, *p);
3299 }
3300
3301 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3302 }
3303
3304 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3305}
3306
3307bool
da8c1a98
KW
3308Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3309 const U8 * const e)
3310{
3311 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3312
af0d7385
KW
3313 return is_utf8_common_with_len(p, e, NULL,
3314 "This is buggy if this gets used",
da8c1a98
KW
3315 PL_XPosix_ptrs[classnum]);
3316}
3317
3318bool
da8c1a98
KW
3319Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3320{
da8c1a98
KW
3321 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3322
af0d7385
KW
3323 return is_utf8_common_with_len(p, e, NULL,
3324 "This is buggy if this gets used",
3325 PL_utf8_perl_idstart);
da8c1a98
KW
3326}
3327
3328bool
f2645549 3329Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3330{
f2645549 3331 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3332
3333 if (*p == '_')
3334 return TRUE;
f5fb7dc5
KW
3335
3336 return is_utf8_common(p, NULL,
3337 "This is buggy if this gets used",
3338 PL_utf8_xidstart);
c11ff943
KW
3339}
3340
3341bool
da8c1a98
KW
3342Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3343{
da8c1a98
KW
3344 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3345
af0d7385
KW
3346 return is_utf8_common_with_len(p, e, NULL,
3347 "This is buggy if this gets used",
3348 PL_utf8_perl_idcont);
da8c1a98
KW
3349}
3350
3351bool
f2645549 3352Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3353{
f2645549 3354 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3355
f5fb7dc5
KW
3356 return is_utf8_common(p, NULL,
3357 "This is buggy if this gets used",
3358 PL_utf8_idcont);
a0ed51b3
LW
3359}
3360
3361bool
f2645549 3362Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3363{
f2645549 3364 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3365
f5fb7dc5
KW
3366 return is_utf8_common(p, NULL,
3367 "This is buggy if this gets used",
3368 PL_utf8_xidcont);
c11ff943
KW
3369}
3370
3371bool
7dbf68d2
KW
3372Perl__is_utf8_mark(pTHX_ const U8 *p)
3373{
7dbf68d2
KW
3374 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3375
7258295b 3376 return is_utf8_common(p, NULL, "IsM", PL_utf8_mark);
7dbf68d2
KW
3377}
3378
6a4a25f4 3379STATIC UV
30613bdc
KW
3380S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3381 U8* ustrp, STRLEN *lenp,
341bb5b7 3382 SV *invlist, const int * const invmap,
e39a4130 3383 const unsigned int * const * const aux_tables,
30613bdc
KW
3384 const U8 * const aux_table_lengths,
3385 const char * const normal)
b9992569 3386{
0134edef 3387 STRLEN len = 0;
7918f24d 3388
30613bdc
KW
3389 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3390 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3391 * to name the new case in any generated messages, as a fallback if the
3392 * operation being used is not available. The new case is given by the
3393 * data structures in the remaining arguments.
3394 *
3395 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3396 * entire changed case string, and the return value is the first code point
3397 * in that string */
3398
b9992569 3399 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3400
36eaa811
KW
3401 /* For code points that don't change case, we already know that the output
3402 * of this function is the unchanged input, so we can skip doing look-ups
3403 * for them. Unfortunately the case-changing code points are scattered
3404 * around. But there are some long consecutive ranges where there are no
3405 * case changing code points. By adding tests, we can eliminate the lookup
3406 * for all the ones in such ranges. This is currently done here only for
3407 * just a few cases where the scripts are in common use in modern commerce
3408 * (and scripts adjacent to those which can be included without additional
3409 * tests). */
3410
3411 if (uv1 >= 0x0590) {
3412 /* This keeps from needing further processing the code points most
3413 * likely to be used in the following non-cased scripts: Hebrew,
3414 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3415 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3416 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3417 if (uv1 < 0x10A0) {
3418 goto cases_to_self;
3419 }
3420
3421 /* The following largish code point ranges also don't have case
3422 * changes, but khw didn't think they warranted extra tests to speed
3423 * them up (which would slightly slow down everything else above them):
3424 * 1100..139F Hangul Jamo, Ethiopic
3425 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3426 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3427 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3428 * Combining Diacritical Marks Extended, Balinese,
3429 * Sundanese, Batak, Lepcha, Ol Chiki
3430 * 2000..206F General Punctuation
3431 */
3432
3433 if (uv1 >= 0x2D30) {
3434
3435 /* This keeps the from needing further processing the code points
3436 * most likely to be used in the following non-cased major scripts:
3437 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3438 *
3439 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3440 * event that Unicode eventually allocates the unused block as of
3441 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3442 * that the test suite will start having failures to alert you
3443 * should that happen) */
3444 if (uv1 < 0xA640) {
3445 goto cases_to_self;
3446 }
3447
3448 if (uv1 >= 0xAC00) {
3449 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3450 if (ckWARN_d(WARN_SURROGATE)) {
3451 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3452 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3453 "Operation \"%s\" returns its argument for"
3454 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3455 }
3456 goto cases_to_self;
3457 }
36eaa811
KW
3458
3459 /* AC00..FAFF Catches Hangul syllables and private use, plus
3460 * some others */
3461 if (uv1 < 0xFB00) {
3462 goto cases_to_self;
36eaa811
KW
3463 }
3464
5af9bc97 3465 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3466 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3467 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3468 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3469 }
3470 if (ckWARN_d(WARN_NON_UNICODE)) {
3471 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3472 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3473 "Operation \"%s\" returns its argument for"
3474 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3475 }
3476 goto cases_to_self;
3477 }
3bfc1e70
KW
3478#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3479 if (UNLIKELY(uv1
3480 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3481 {
3482
56576a04
KW
3483 /* As of Unicode 10.0, this means we avoid swash creation
3484 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3485 goto cases_to_self;
3486 }
3487#endif
36eaa811
KW
3488 }
3489 }
9ae3ac1a 3490
36eaa811 3491 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3492 * be given. */
9ae3ac1a
KW
3493 }
3494
8946fcd9
KW
3495 {
3496 unsigned int i;
e39a4130 3497 const unsigned int * cp_list;
8946fcd9 3498 U8 * d;
69352d88
KW
3499
3500 /* 'index' is guaranteed to be non-negative, as this is an inversion
3501 * map that covers all possible inputs. See [perl #133365] */
8946fcd9
KW
3502 SSize_t index = _invlist_search(invlist, uv1);
3503 IV base = invmap[index];
0134edef 3504
30613bdc
KW
3505 /* The data structures are set up so that if 'base' is non-negative,
3506 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3507 if (base >= 0) {
3508 IV lc;
b08cf34e 3509
8946fcd9
KW
3510 if (base == 0) {
3511 goto cases_to_self;
4a8240a3 3512 }
4a8240a3 3513
30613bdc 3514 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3515 lc = base + uv1 - invlist_array(invlist)[index];
3516 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3517 return lc;
3518 }
1feea2c7 3519
30613bdc
KW
3520 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3521 * requires an auxiliary table look up. abs(base) gives the index into
3522 * a list of such tables which points to the proper aux table. And a
3523 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3524 cp_list = aux_tables[-base];
30613bdc
KW
3525
3526 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3527 d = ustrp;
3528 for (i = 0; i < aux_table_lengths[-base]; i++) {
3529 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3530 }
8946fcd9
KW
3531 *d = '\0';
3532 *lenp = d - ustrp;
3533
3534 return cp_list[0];
cbe07460
KW
3535 }
3536
3537 /* Here, there was no mapping defined, which means that the code point maps
3538 * to itself. Return the inputs */
e24dfe9c 3539 cases_to_self:
a13f1de4
KW
3540 if (p) {
3541 len = UTF8SKIP(p);
3542 if (p != ustrp) { /* Don't copy onto itself */
3543 Copy(p, ustrp, len, U8);
3544 }
3545 *lenp = len;
3546 }
3547 else {
3548 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3549 }
2a37f04d 3550
f4cd282c 3551 return uv1;
cbe07460 3552
a0ed51b3
LW
3553}
3554
b74fe592 3555Size_t
e39a4130
KW
3556Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3557 const unsigned int ** remaining_folds_to)
b74fe592
KW
3558{
3559 /* Returns the count of the number of code points that fold to the input
3560 * 'cp' (besides itself).
3561 *
3562 * If the return is 0, there is nothing else that folds to it, and
3563 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3564 *
3565 * If the return is 1, '*first_folds_to' is set to the single code point,
3566 * and '*remaining_folds_to' is set to NULL.
3567 *
3568 * Otherwise, '*first_folds_to' is set to a code point, and
3569 * '*remaining_fold_to' is set to an array that contains the others. The
3570 * length of this array is the returned count minus 1.
3571 *
3572 * The reason for this convolution is to avoid having to deal with
3573 * allocating and freeing memory. The lists are already constructed, so
3574 * the return can point to them, but single code points aren't, so would
3575 * need to be constructed if we didn't employ something like this API */
3576
69352d88
KW
3577 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3578 * that covers all possible inputs. See [perl #133365] */
b74fe592
KW
3579 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3580 int base = _Perl_IVCF_invmap[index];
3581
3582 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3583
3584 if (base == 0) { /* No fold */
3585 *first_folds_to = 0;
3586 *remaining_folds_to = NULL;
3587 return 0;
3588 }
3589
3590#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3591
3592 assert(base > 0);
3593
3594#else
3595
3596 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3597
3598 /* The data structure is set up so that the absolute value of 'base' is
3599 * an index into a table of pointers to arrays, with the array
3600 * corresponding to the index being the list of code points that fold
3601 * to 'cp', and the parallel array containing the length of the list
3602 * array */
3603 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3604 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3605 *first_folds_to
3606 */
3607 return IVCF_AUX_TABLE_lengths[-base];
3608 }
3609
3610#endif
3611
3612 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3613 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3614 *remaining_folds_to = NULL;
3615 return 1;
3616}
3617
051a06d4 3618STATIC UV
56576a04
KW
3619S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3620 U8* const ustrp, STRLEN *lenp)
051a06d4 3621{
4a4088c4 3622 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3623 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3624 * result contains a character that crosses the 255/256 boundary, disallow
3625 * the change, and return the original code point. See L<perlfunc/lc> for
3626 * why;
051a06d4 3627 *
a1433954
KW
3628 * p points to the original string whose case was changed; assumed
3629 * by this routine to be well-formed
051a06d4 3630 * result the code point of the first character in the changed-case string
56576a04
KW
3631 * ustrp points to the changed-case string (<result> represents its
3632 * first char)
051a06d4
KW
3633 * lenp points to the length of <ustrp> */
3634
3635 UV original; /* To store the first code point of <p> */
3636
3637 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3638
a4f12ed7 3639 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3640
3641 /* We know immediately if the first character in the string crosses the
5e45c680 3642 * boundary, so can skip testing */
051a06d4
KW
3643 if (result > 255) {
3644
3645 /* Look at every character in the result; if any cross the
3646 * boundary, the whole thing is disallowed */
3647 U8* s = ustrp + UTF8SKIP(ustrp);
3648 U8* e = ustrp + *lenp;
3649 while (s < e) {
a4f12ed7 3650 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3651 goto bad_crossing;
3652 }
3653 s += UTF8SKIP(s);
3654 }
3655
613abc6d
KW
3656 /* Here, no characters crossed, result is ok as-is, but we warn. */
3657 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3658 return result;
3659 }
3660
7b52d656 3661 bad_crossing:
051a06d4
KW
3662
3663 /* Failed, have to return the original */
4b88fb76 3664 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3665
3666 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3667 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3668 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3669 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3670 OP_DESC(PL_op),
ab0b796c
KW
3671 original,
3672 original);
051a06d4
KW
3673 Copy(p, ustrp, *lenp, char);
3674 return original;
3675}
3676
607313a1
KW
3677STATIC U32
3678S_check_and_deprecate(pTHX_ const U8 *p,
3679 const U8 **e,
3680 const unsigned int type, /* See below */
3681 const bool use_locale, /* Is this a 'LC_'
3682 macro call? */
3683 const char * const file,
3684 const unsigned line)
3685{
3686 /* This is a temporary function to deprecate the unsafe calls to the case
3687 * changing macros and functions. It keeps all the special stuff in just
3688 * one place.
3689 *
3690 * It updates *e with the pointer to the end of the input string. If using
3691 * the old-style macros, *e is NULL on input, and so this function assumes
3692 * the input string is long enough to hold the entire UTF-8 sequence, and
3693 * sets *e accordingly, but it then returns a flag to pass the
3694 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3695 * using the full length if possible.
3696 *
3697 * It also does the assert that *e > p when *e is not NULL. This should be
3698 * migrated to the callers when this function gets deleted.
3699 *
3700 * The 'type' parameter is used for the caller to specify which case
3701 * changing function this is called from: */
3702
3703# define DEPRECATE_TO_UPPER 0
3704# define DEPRECATE_TO_TITLE 1
3705# define DEPRECATE_TO_LOWER 2
3706# define DEPRECATE_TO_FOLD 3
3707
3708 U32 utf8n_flags = 0;
3709 const char * name;
3710 const char * alternative;
3711
3712 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3713
3714 if (*e == NULL) {
3715 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3716 *e = p + UTF8SKIP(p);
3717
3718 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3719 * there. It could be part of a larger path */
607313a1 3720 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3721 name = instr(file, "mathoms.c")
607313a1
KW
3722 ? "to_utf8_upper"
3723 : "toUPPER_utf8";
3724 alternative = "toUPPER_utf8_safe";
3725 }
3726 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3727 name = instr(file, "mathoms.c")
607313a1
KW
3728 ? "to_utf8_title"
3729 : "toTITLE_utf8";
3730 alternative = "toTITLE_utf8_safe";
3731 }