This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perl.h: Clarify comment
[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
216been displayed if all lexcial warnings are enabled will be returned.
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) {
4c8cd605 389 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
390 uv >>= UTF_ACCUMULATION_SHIFT;
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
1126 /* How many 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);
806547a7
KW
1142
1143 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1144
7cf8d05d
KW
1145 /* We don't need to pass this parameter, but since it has already been
1146 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1147 assert(expect_len == UTF8SKIP(s));
1148
1149 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1150 " %s after start byte 0x%02x; need %d bytes, got %d)",
1151 malformed_text,
7e2f38b2 1152 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
1153 *(s + non_cont_byte_pos),
1154 where,
1155 *s,
1156 (int) expect_len,
1157 (int) non_cont_byte_pos);
806547a7
KW
1158}
1159
35f8c9bd
KW
1160/*
1161
de69f3af 1162=for apidoc utf8n_to_uvchr
378516de
KW
1163
1164THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 1165Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 1166
9041c2e3 1167Bottom level UTF-8 decode routine.
de69f3af 1168Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1169which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1170C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1171the length, in bytes, of that character.
949cf498
KW
1172
1173The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1174well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1175causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1176is the next possible position in C<s> that could begin a non-malformed
1177character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1178is raised. Some UTF-8 input sequences may contain multiple malformations.
1179This function tries to find every possible one in each call, so multiple
56576a04 1180warnings can be raised for the same sequence.
949cf498
KW
1181
1182Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1183individual types of malformations, such as the sequence being overlong (that
1184is, when there is a shorter sequence that can express the same code point;
1185overlong sequences are expressly forbidden in the UTF-8 standard due to
1186potential security issues). Another malformation example is the first byte of
1187a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1188flags. Even if allowed, this function generally returns the Unicode
1189REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1190F<utf8.h> to override this behavior for the overlong malformations, but don't
1191do that except for very specialized purposes.
949cf498 1192
796b6530 1193The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1194flags) malformation is found. If this flag is set, the routine assumes that
1195the caller will raise a warning, and this function will silently just set
d088425d
KW
1196C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1197
75200dff 1198Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1199character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1200in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1201be set to 1. To disambiguate, upon a zero return, see if the first byte of
1202C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1203error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1204
1205Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1206Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1207By default these are considered regular code points, but certain situations
ecc1615f
KW
1208warrant special handling for them, which can be specified using the C<flags>
1209parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1210three classes are treated as malformations and handled as such. The flags
1211C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1212C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1213disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1214restricts the allowed inputs to the strict UTF-8 traditionally defined by
1215Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1216definition given by
1217L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1218The difference between traditional strictness and C9 strictness is that the
1219latter does not forbid non-character code points. (They are still discouraged,
1220however.) For more discussion see L<perlunicode/Noncharacter code points>.
1221
1222The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1223C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1224C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1225raised for their respective categories, but otherwise the code points are
1226considered valid (not malformations). To get a category to both be treated as
1227a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1228(But note that warnings are not raised if lexically disabled nor if
796b6530 1229C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1230
57ff5f59
KW
1231Extremely high code points were never specified in any standard, and require an
1232extension to UTF-8 to express, which Perl does. It is likely that programs
1233written in something other than Perl would not be able to read files that
1234contain these; nor would Perl understand files written by something that uses a
1235different extension. For these reasons, there is a separate set of flags that
1236can warn and/or disallow these extremely high code points, even if other
1237above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1238C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1239L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1240above-Unicode code points, including these, as malformations.
1241(Note that the Unicode standard considers anything above 0x10FFFF to be
1242illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1243(2**31 -1))
1244
1245A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1246retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1247C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1248C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1249can apply to code points that actually do fit in 31 bits. This happens on
1250EBCDIC platforms, and sometimes when the L<overlong
1251malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1252describe the situation in all cases.
1253
ab8e6d41 1254
949cf498
KW
1255All other code points corresponding to Unicode characters, including private
1256use and those yet to be assigned, are never considered malformed and never
1257warn.
67e989fb 1258
37607a96 1259=cut
f9380377
KW
1260
1261Also implemented as a macro in utf8.h
1262*/
1263
1264UV
1265Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1266 STRLEN curlen,
1267 STRLEN *retlen,
1268 const U32 flags)
1269{
1270 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1271
1272 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1273}
1274
2b9519f0
KW
1275/* The tables below come from http://bjoern.hoehrmann.de/utf-8/decoder/dfa/,
1276 * which requires this copyright notice */
1277
1278/* Copyright (c) 2008-2009 Bjoern Hoehrmann <bjoern@hoehrmann.de>
1279
1280Permission is hereby granted, free of charge, to any person obtaining a copy of
1281this software and associated documentation files (the "Software"), to deal in
1282the Software without restriction, including without limitation the rights to
1283use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
1284of the Software, and to permit persons to whom the Software is furnished to do
1285so, subject to the following conditions:
1286
1287The above copyright notice and this permission notice shall be included in all
1288copies or substantial portions of the Software.
1289
1290THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
1291IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
1292FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
1293AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
1294LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
1295OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
1296SOFTWARE.
1297
1298*/
1299
1300#if 0
1301static U8 utf8d_C9[] = {
1302 /* The first part of the table maps bytes to character classes that
1303 * to reduce the size of the transition table and create bitmasks. */
1304 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1305 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1306 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1307 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1308 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1309 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1310 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1311 10,3,3,3,3,3,3,3,3,3,3,3,3,4,3,3, 11,6,6,6,5,8,8,8,8,8,8,8,8,8,8,8, /*-FF*/
1312
1313 /* The second part is a transition table that maps a combination
1314 * of a state of the automaton and a character class to a state. */
1315 0,12,24,36,60,96,84,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,
1316 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,
1317 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,
1318 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,
1319 12,36,12,12,12,12,12,12,12,12,12,12
1320};
1321
1322#endif
1323
1324#ifndef EBCDIC
1325
1326/* This is a version of the above table customized for Perl that doesn't
1327 * exclude surrogates and accepts start bytes up through F7 (representing
1328 * 2**21 - 1). */
1329static U8 dfa_tab_for_perl[] = {
1330 /* The first part of the table maps bytes to character classes to reduce
1331 * the size of the transition table and create bitmasks. */
1332 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1333 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1334 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1335 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1336 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1337 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1338 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1339 10,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, 11,4,4,4,4,4,4,4,8,8,8,8,8,8,8,8, /*-FF*/
1340
1341 /* The second part is a transition table that maps a combination
1342 * of a state of the automaton and a character class to a state. */
1343 0,12,24,36,96,12,12,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,/*23*/
1344 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,/*47*/
1345 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,/*71*/
1346 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,/*95*/
1347 12,36,12,12,12,12,12,36,12,36,12,12 /* 96- 107 */
1348
1349 /* The customization was to repurpose the surrogates type '4' to instead be
1350 * for start bytes F1-F7. Types 5 and 6 are now unused, and their entries in
1351 * the transition part of the table are set to 12, so are illegal.
1352 *
1353 * To do higher code points would require expansion and some rearrangement of
1354 * the table. The type '1' entries for continuation bytes 80-8f would have to
1355 * be split into several types, because they aren't treated uniformly for
1356 * higher start bytes, since overlongs for F8 are 80-87; FC: 80-83; and FE:
1357 * 80-81. We start needing to worry about overflow if FE is included.
1358 * Ignoring, FE and FF, we could use type 5 for F9-FB, and 6 for FD (remember
1359 * from the web site that these are used to right shift). FE would
1360 * necessarily be type 7; and FF, type 8. And new states would have to be
1361 * created for F8 and FC (and FE and FF if used), so quite a bit of work would
1362 * be involved.
1363 *
1364 * XXX Better would be to customize the table so that the noncharacters are
1365 * excluded. This again is non trivial, but doing so would simplify the code
1366 * that uses this, and might make it small enough to make it inlinable */
1367};
1368
1369#endif
1370
f9380377
KW
1371/*
1372
1373=for apidoc utf8n_to_uvchr_error
1374
1375THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1376Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1377
1378This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1379are when an error is found. If you also need to know the generated warning
1380messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1381
1382It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1383all the others, C<errors>. If this parameter is 0, this function behaves
1384identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1385to a C<U32> variable, which this function sets to indicate any errors found.
1386Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1387C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1388of these bits will be set if a malformation is found, even if the input
7a65503b 1389C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1390exceptions are noted:
1391
1392=over 4
1393
57ff5f59 1394=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1395
57ff5f59
KW
1396The input sequence is not standard UTF-8, but a Perl extension. This bit is
1397set only if the input C<flags> parameter contains either the
1398C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1399
1400Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1401and so some extension must be used to express them. Perl uses a natural
1402extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1403extension to represent even higher ones, so that any code point that fits in a
140464-bit word can be represented. Text using these extensions is not likely to
1405be portable to non-Perl code. We lump both of these extensions together and
1406refer to them as Perl extended UTF-8. There exist other extensions that people
1407have invented, incompatible with Perl's.
1408
1409On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1410extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1411than on ASCII. Prior to that, code points 2**31 and higher were simply
1412unrepresentable, and a different, incompatible method was used to represent
1413code points between 2**30 and 2**31 - 1.
1414
1415On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1416Perl extended UTF-8 is used.
1417
1418In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1419may use for backward compatibility. That name is misleading, as this flag may
1420be set when the code point actually does fit in 31 bits. This happens on
1421EBCDIC platforms, and sometimes when the L<overlong
1422malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1423describes the situation in all cases.
f9380377
KW
1424
1425=item C<UTF8_GOT_CONTINUATION>
1426
1427The input sequence was malformed in that the first byte was a a UTF-8
1428continuation byte.
1429
1430=item C<UTF8_GOT_EMPTY>
1431
1432The input C<curlen> parameter was 0.
1433
1434=item C<UTF8_GOT_LONG>
1435
1436The input sequence was malformed in that there is some other sequence that
1437evaluates to the same code point, but that sequence is shorter than this one.
1438
fecaf136
KW
1439Until Unicode 3.1, it was legal for programs to accept this malformation, but
1440it was discovered that this created security issues.
1441
f9380377
KW
1442=item C<UTF8_GOT_NONCHAR>
1443
1444The code point represented by the input UTF-8 sequence is for a Unicode
1445non-character code point.
1446This bit is set only if the input C<flags> parameter contains either the
1447C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1448
1449=item C<UTF8_GOT_NON_CONTINUATION>
1450
1451The input sequence was malformed in that a non-continuation type byte was found
1452in a position where only a continuation type one should be.
1453
1454=item C<UTF8_GOT_OVERFLOW>
1455
1456The input sequence was malformed in that it is for a code point that is not
d22ec717 1457representable in the number of bits available in an IV on the current platform.
f9380377
KW
1458
1459=item C<UTF8_GOT_SHORT>
1460
1461The input sequence was malformed in that C<curlen> is smaller than required for
1462a complete sequence. In other words, the input is for a partial character
1463sequence.
1464
1465=item C<UTF8_GOT_SUPER>
1466
1467The input sequence was malformed in that it is for a non-Unicode code point;
1468that is, one above the legal Unicode maximum.
1469This bit is set only if the input C<flags> parameter contains either the
1470C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1471
1472=item C<UTF8_GOT_SURROGATE>
1473
1474The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1475code point.
1476This bit is set only if the input C<flags> parameter contains either the
1477C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1478
1479=back
1480
133551d8
KW
1481To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1482flag to suppress any warnings, and then examine the C<*errors> return.
1483
f9380377 1484=cut
37657a5b
KW
1485
1486Also implemented as a macro in utf8.h
37607a96 1487*/
67e989fb 1488
a0ed51b3 1489UV
f9380377 1490Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
37657a5b
KW
1491 STRLEN curlen,
1492 STRLEN *retlen,
1493 const U32 flags,
1494 U32 * errors)
1495{
1496 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1497
1498 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1499}
1500
1501/*
1502
1503=for apidoc utf8n_to_uvchr_msgs
1504
1505THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1506Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1507
1508This function is for code that needs to know what the precise malformation(s)
1509are when an error is found, and wants the corresponding warning and/or error
1510messages to be returned to the caller rather than be displayed. All messages
1511that would have been displayed if all lexcial warnings are enabled will be
1512returned.
1513
1514It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1515placed after all the others, C<msgs>. If this parameter is 0, this function
1516behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1517be a pointer to an C<AV *> variable, in which this function creates a new AV to
1518contain any appropriate messages. The elements of the array are ordered so
1519that the first message that would have been displayed is in the 0th element,
1520and so on. Each element is a hash with three key-value pairs, as follows:
1521
1522=over 4
1523
1524=item C<text>
1525
1526The text of the message as a C<SVpv>.
1527
1528=item C<warn_categories>
1529
1530The warning category (or categories) packed into a C<SVuv>.
1531
1532=item C<flag>
1533
1534A single flag bit associated with this message, in a C<SVuv>.
1535The bit corresponds to some bit in the C<*errors> return value,
1536such as C<UTF8_GOT_LONG>.
1537
1538=back
1539
1540It's important to note that specifying this parameter as non-null will cause
1541any warnings this function would otherwise generate to be suppressed, and
1542instead be placed in C<*msgs>. The caller can check the lexical warnings state
1543(or not) when choosing what to do with the returned messages.
1544
1545If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1546no AV is created.
1547
1548The caller, of course, is responsible for freeing any returned AV.
1549
1550=cut
1551*/
1552
1553UV
1554Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1555 STRLEN curlen,
1556 STRLEN *retlen,
1557 const U32 flags,
1558 U32 * errors,
1559 AV ** msgs)
a0ed51b3 1560{
d4c19fe8 1561 const U8 * const s0 = s;
2b9519f0 1562 const U8 * send = s0 + curlen;
2b5e7bc2
KW
1563 U32 possible_problems = 0; /* A bit is set here for each potential problem
1564 found as we go along */
2b672cf5 1565 UV uv = (UV) -1;
2b5e7bc2
KW
1566 STRLEN expectlen = 0; /* How long should this sequence be?
1567 (initialized to silence compilers' wrong
1568 warning) */
e308b348 1569 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1570 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1571 this gets set and discarded */
a0dbb045 1572
2b5e7bc2
KW
1573 /* The below are used only if there is both an overlong malformation and a
1574 * too short one. Otherwise the first two are set to 's0' and 'send', and
1575 * the third not used at all */
1576 U8 * adjusted_s0 = (U8 *) s0;
e9f2c446
KW
1577 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1578 routine; see [perl #130921] */
2b5e7bc2 1579 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1580
2b9519f0
KW
1581 UV state = 0;
1582
37657a5b 1583 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
f9380377
KW
1584
1585 if (errors) {
1586 *errors = 0;
1587 }
1588 else {
1589 errors = &discard_errors;
1590 }
a0dbb045 1591
eb83ed87
KW
1592 /* The order of malformation tests here is important. We should consume as
1593 * few bytes as possible in order to not skip any valid character. This is
1594 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1595 * http://unicode.org/reports/tr36 for more discussion as to why. For
1596 * example, once we've done a UTF8SKIP, we can tell the expected number of
1597 * bytes, and could fail right off the bat if the input parameters indicate
1598 * that there are too few available. But it could be that just that first
1599 * byte is garbled, and the intended character occupies fewer bytes. If we
1600 * blindly assumed that the first byte is correct, and skipped based on
1601 * that number, we could skip over a valid input character. So instead, we
1602 * always examine the sequence byte-by-byte.
1603 *
1604 * We also should not consume too few bytes, otherwise someone could inject
1605 * things. For example, an input could be deliberately designed to
1606 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1607 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1608 * which is actually part of of the overflowing sequence), that could look
1609 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1610 * sequence and process the rest, inappropriately.
1611 *
1612 * Some possible input sequences are malformed in more than one way. This
1613 * function goes to lengths to try to find all of them. This is necessary
1614 * for correctness, as the inputs may allow one malformation but not
1615 * another, and if we abandon searching for others after finding the
1616 * allowed one, we could allow in something that shouldn't have been.
1617 */
eb83ed87 1618
b5b9af04 1619 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1620 possible_problems |= UTF8_GOT_EMPTY;
1621 curlen = 0;
5a48568d 1622 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1623 goto ready_to_handle_errors;
0c443dc2
JH
1624 }
1625
eb83ed87
KW
1626 expectlen = UTF8SKIP(s);
1627
1628 /* A well-formed UTF-8 character, as the vast majority of calls to this
1629 * function will be for, has this expected length. For efficiency, set
1630 * things up here to return it. It will be overriden only in those rare
1631 * cases where a malformation is found */
1632 if (retlen) {
1633 *retlen = expectlen;
1634 }
1635
1636 /* An invariant is trivially well-formed */
2b9519f0
KW
1637 if (UTF8_IS_INVARIANT(*s0)) {
1638 return *s0;
a0ed51b3 1639 }
67e989fb 1640
2b9519f0
KW
1641#ifndef EBCDIC
1642
1643 /* Measurements show that this dfa is somewhat faster than the regular code
1644 * below, so use it first, dropping down for the non-normal cases. */
1645
1646# define PERL_UTF8_DECODE_REJECT 12
1647
1648 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
1649 UV type = dfa_tab_for_perl[*s];
1650
1651 if (state != 0) {
1652 uv = (*s & 0x3fu) | (uv << UTF_ACCUMULATION_SHIFT);
1653 state = dfa_tab_for_perl[256 + state + type];
1654 }
1655 else {
1656 uv = (0xff >> type) & (*s);
1657 state = dfa_tab_for_perl[256 + type];
1658 }
1659
1660 if (state == 0) {
1661
1662 /* If this could be a code point that the flags don't allow (the first
1663 * surrogate is the first such possible one), delve further, but we already
1664 * have calculated 'uv' */
1665 if ( (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1666 |UTF8_WARN_ILLEGAL_INTERCHANGE))
1667 && uv >= UNICODE_SURROGATE_FIRST)
1668 {
1669 curlen = s + 1 - s0;
1670 goto got_uv;
1671 }
1672
1673 return uv;
1674 }
1675
1676 s++;
1677 }
1678
1679 /* Here, is some sort of failure. Use the full mechanism */
1680
1681 uv = *s0;
1682
1683#endif
1684
eb83ed87 1685 /* A continuation character can't start a valid sequence */
b5b9af04 1686 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1687 possible_problems |= UTF8_GOT_CONTINUATION;
1688 curlen = 1;
1689 uv = UNICODE_REPLACEMENT;
1690 goto ready_to_handle_errors;
ba210ebe 1691 }
9041c2e3 1692
dcd27b3c 1693 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1694 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1695 * because it excludes start bytes like \xC0 that always lead to
1696 * overlongs.) */
dcd27b3c 1697
534752c1
KW
1698 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1699 * that indicate the number of bytes in the character's whole UTF-8
1700 * sequence, leaving just the bits that are part of the value. */
1701 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1702
e308b348
KW
1703 /* Setup the loop end point, making sure to not look past the end of the
1704 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1705 if (UNLIKELY(curlen < expectlen)) {
1706 possible_problems |= UTF8_GOT_SHORT;
1707 avail_len = curlen;
e308b348
KW
1708 }
1709 else {
2b9519f0 1710 send = (U8*) s0 + expectlen;
e308b348 1711 }
e308b348 1712
eb83ed87 1713 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1714 * accumulating each into the working value as we go. */
eb83ed87 1715 for (s = s0 + 1; s < send; s++) {
b5b9af04 1716 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1717 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1718 continue;
1719 }
1720
1721 /* Here, found a non-continuation before processing all expected bytes.
1722 * This byte indicates the beginning of a new character, so quit, even
1723 * if allowing this malformation. */
2b5e7bc2 1724 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1725 break;
eb83ed87
KW
1726 } /* End of loop through the character's bytes */
1727
1728 /* Save how many bytes were actually in the character */
1729 curlen = s - s0;
1730
2b5e7bc2
KW
1731 /* Note that there are two types of too-short malformation. One is when
1732 * there is actual wrong data before the normal termination of the
1733 * sequence. The other is that the sequence wasn't complete before the end
1734 * of the data we are allowed to look at, based on the input 'curlen'.
1735 * This means that we were passed data for a partial character, but it is
1736 * valid as far as we saw. The other is definitely invalid. This
1737 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1738 * separate.
1739 *
1740 * A convenience macro that matches either of the too-short conditions. */
1741# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1742
1743 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1744 uv_so_far = uv;
1745 uv = UNICODE_REPLACEMENT;
1746 }
2b5e7bc2 1747
08e73697
KW
1748 /* Check for overflow. The algorithm requires us to not look past the end
1749 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1750 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1751 1 /* Do consider overlongs */
1752 )))
1753 {
2b5e7bc2
KW
1754 possible_problems |= UTF8_GOT_OVERFLOW;
1755 uv = UNICODE_REPLACEMENT;
eb83ed87 1756 }
eb83ed87 1757
2b5e7bc2
KW
1758 /* Check for overlong. If no problems so far, 'uv' is the correct code
1759 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1760 * we must look at the UTF-8 byte sequence itself to see if it is for an
1761 * overlong */
1762 if ( ( LIKELY(! possible_problems)
1763 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1764 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1765 && ( UNLIKELY(! UTF8_IS_START(*s0))
1766 || ( curlen > 1
d6be65ae 1767 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1768 s - s0))))))
2f8f112e 1769 {
2b5e7bc2
KW
1770 possible_problems |= UTF8_GOT_LONG;
1771
abc28b54 1772 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1773
abc28b54
KW
1774 /* The calculation in the 'true' branch of this 'if'
1775 * below won't work if overflows, and isn't needed
1776 * anyway. Further below we handle all overflow
1777 * cases */
1778 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1779 {
2b5e7bc2
KW
1780 UV min_uv = uv_so_far;
1781 STRLEN i;
1782
1783 /* Here, the input is both overlong and is missing some trailing
1784 * bytes. There is no single code point it could be for, but there
1785 * may be enough information present to determine if what we have
1786 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1787 * The code further below has the intelligence to determine this,
1788 * but just for non-overlong UTF-8 sequences. What we do here is
1789 * calculate the smallest code point the input could represent if
1790 * there were no too short malformation. Then we compute and save
1791 * the UTF-8 for that, which is what the code below looks at
1792 * instead of the raw input. It turns out that the smallest such
1793 * code point is all we need. */
2b5e7bc2
KW
1794 for (i = curlen; i < expectlen; i++) {
1795 min_uv = UTF8_ACCUMULATE(min_uv,
1796 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1797 }
1798
e9f2c446 1799 adjusted_s0 = temp_char_buf;
57ff5f59 1800 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1801 }
eb83ed87
KW
1802 }
1803
2b9519f0
KW
1804 got_uv:
1805
56576a04
KW
1806 /* Here, we have found all the possible problems, except for when the input
1807 * is for a problematic code point not allowed by the input parameters. */
1808
06188866
KW
1809 /* uv is valid for overlongs */
1810 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1811
1812 /* isn't problematic if < this */
1813 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1814 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1815
1816 /* if overflow, we know without looking further
1817 * precisely which of the problematic types it is,
1818 * and we deal with those in the overflow handling
1819 * code */
1820 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1821 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1822 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1823 && ((flags & ( UTF8_DISALLOW_NONCHAR
1824 |UTF8_DISALLOW_SURROGATE
1825 |UTF8_DISALLOW_SUPER
d044b7a7 1826 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1827 |UTF8_WARN_NONCHAR
1828 |UTF8_WARN_SURROGATE
1829 |UTF8_WARN_SUPER
d22ec717 1830 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1831 {
2b5e7bc2
KW
1832 /* If there were no malformations, or the only malformation is an
1833 * overlong, 'uv' is valid */
1834 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1835 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1836 possible_problems |= UTF8_GOT_SURROGATE;
1837 }
1838 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1839 possible_problems |= UTF8_GOT_SUPER;
1840 }
1841 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1842 possible_problems |= UTF8_GOT_NONCHAR;
1843 }
1844 }
1845 else { /* Otherwise, need to look at the source UTF-8, possibly
1846 adjusted to be non-overlong */
1847
1848 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1849 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1850 {
2b5e7bc2
KW
1851 possible_problems |= UTF8_GOT_SUPER;
1852 }
1853 else if (curlen > 1) {
1854 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1855 NATIVE_UTF8_TO_I8(*adjusted_s0),
1856 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1857 {
2b5e7bc2 1858 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1859 }
2b5e7bc2
KW
1860 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1861 NATIVE_UTF8_TO_I8(*adjusted_s0),
1862 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1863 {
1864 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1865 }
1866 }
c0236afe 1867
2b5e7bc2
KW
1868 /* We need a complete well-formed UTF-8 character to discern
1869 * non-characters, so can't look for them here */
1870 }
1871 }
949cf498 1872
2b5e7bc2
KW
1873 ready_to_handle_errors:
1874
1875 /* At this point:
1876 * curlen contains the number of bytes in the sequence that
1877 * this call should advance the input by.
e308b348
KW
1878 * avail_len gives the available number of bytes passed in, but
1879 * only if this is less than the expected number of
1880 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1881 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1882 * is set in it for each potential problem found.
1883 * uv contains the code point the input sequence
1884 * represents; or if there is a problem that prevents
1885 * a well-defined value from being computed, it is
1886 * some subsitute value, typically the REPLACEMENT
1887 * CHARACTER.
1888 * s0 points to the first byte of the character
56576a04
KW
1889 * s points to just after were we left off processing
1890 * the character
1891 * send points to just after where that character should
1892 * end, based on how many bytes the start byte tells
1893 * us should be in it, but no further than s0 +
1894 * avail_len
2b5e7bc2 1895 */
eb83ed87 1896
2b5e7bc2
KW
1897 if (UNLIKELY(possible_problems)) {
1898 bool disallowed = FALSE;
1899 const U32 orig_problems = possible_problems;
1900
37657a5b
KW
1901 if (msgs) {
1902 *msgs = NULL;
1903 }
1904
2b5e7bc2
KW
1905 while (possible_problems) { /* Handle each possible problem */
1906 UV pack_warn = 0;
1907 char * message = NULL;
37657a5b 1908 U32 this_flag_bit = 0;
2b5e7bc2
KW
1909
1910 /* Each 'if' clause handles one problem. They are ordered so that
1911 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1912 * ones; this is kinda in decreasing severity order. But the
1913 * overlong must come last, as it changes 'uv' looked at by the
1914 * others */
2b5e7bc2
KW
1915 if (possible_problems & UTF8_GOT_OVERFLOW) {
1916
56576a04
KW
1917 /* Overflow means also got a super and are using Perl's
1918 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1919 possible_problems
d044b7a7 1920 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1921 *errors |= UTF8_GOT_OVERFLOW;
1922
1923 /* But the API says we flag all errors found */
1924 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1925 *errors |= UTF8_GOT_SUPER;
1926 }
ddb65933 1927 if (flags
d044b7a7 1928 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1929 {
d044b7a7 1930 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1931 }
2b5e7bc2 1932
d60baaa7 1933 /* Disallow if any of the three categories say to */
56576a04 1934 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1935 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1936 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1937 {
1938 disallowed = TRUE;
1939 }
1940
d22ec717
KW
1941 /* Likewise, warn if any say to */
1942 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1943 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1944 {
2b5e7bc2 1945
ddb65933
KW
1946 /* The warnings code explicitly says it doesn't handle the
1947 * case of packWARN2 and two categories which have
1948 * parent-child relationship. Even if it works now to
1949 * raise the warning if either is enabled, it wouldn't
1950 * necessarily do so in the future. We output (only) the
56576a04 1951 * most dire warning */
ddb65933 1952 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1953 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1954 pack_warn = packWARN(WARN_UTF8);
1955 }
37657a5b 1956 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1957 pack_warn = packWARN(WARN_NON_UNICODE);
1958 }
1959 if (pack_warn) {
1960 message = Perl_form(aTHX_ "%s: %s (overflows)",
1961 malformed_text,
05b9033b 1962 _byte_dump_string(s0, curlen, 0));
37657a5b 1963 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1964 }
2b5e7bc2
KW
1965 }
1966 }
1967 }
1968 else if (possible_problems & UTF8_GOT_EMPTY) {
1969 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1970 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1971
1972 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1973
1974 /* This so-called malformation is now treated as a bug in
1975 * the caller. If you have nothing to decode, skip calling
1976 * this function */
1977 assert(0);
1978
2b5e7bc2 1979 disallowed = TRUE;
37657a5b
KW
1980 if ( (msgs
1981 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1982 {
2b5e7bc2
KW
1983 pack_warn = packWARN(WARN_UTF8);
1984 message = Perl_form(aTHX_ "%s (empty string)",
1985 malformed_text);
37657a5b 1986 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1987 }
1988 }
1989 }
1990 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1991 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1992 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1993
1994 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1995 disallowed = TRUE;
37657a5b
KW
1996 if (( msgs
1997 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1998 {
2b5e7bc2
KW
1999 pack_warn = packWARN(WARN_UTF8);
2000 message = Perl_form(aTHX_
2001 "%s: %s (unexpected continuation byte 0x%02x,"
2002 " with no preceding start byte)",
2003 malformed_text,
7e2f38b2 2004 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 2005 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
2006 }
2007 }
2008 }
2b5e7bc2
KW
2009 else if (possible_problems & UTF8_GOT_SHORT) {
2010 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 2011 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
2012
2013 if (! (flags & UTF8_ALLOW_SHORT)) {
2014 disallowed = TRUE;
37657a5b
KW
2015 if (( msgs
2016 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2017 {
2b5e7bc2
KW
2018 pack_warn = packWARN(WARN_UTF8);
2019 message = Perl_form(aTHX_
56576a04
KW
2020 "%s: %s (too short; %d byte%s available, need %d)",
2021 malformed_text,
2022 _byte_dump_string(s0, send - s0, 0),
2023 (int)avail_len,
2024 avail_len == 1 ? "" : "s",
2025 (int)expectlen);
37657a5b 2026 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
2027 }
2028 }
ba210ebe 2029
2b5e7bc2 2030 }
e308b348
KW
2031 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2032 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2033 *errors |= UTF8_GOT_NON_CONTINUATION;
2034
2035 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2036 disallowed = TRUE;
37657a5b
KW
2037 if (( msgs
2038 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2039 {
99a765e9
KW
2040
2041 /* If we don't know for sure that the input length is
2042 * valid, avoid as much as possible reading past the
2043 * end of the buffer */
2044 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2045 ? s - s0
2046 : send - s0;
e308b348
KW
2047 pack_warn = packWARN(WARN_UTF8);
2048 message = Perl_form(aTHX_ "%s",
2049 unexpected_non_continuation_text(s0,
99a765e9 2050 printlen,
e308b348
KW
2051 s - s0,
2052 (int) expectlen));
37657a5b 2053 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2054 }
2055 }
2056 }
2b5e7bc2
KW
2057 else if (possible_problems & UTF8_GOT_SURROGATE) {
2058 possible_problems &= ~UTF8_GOT_SURROGATE;
2059
f9380377
KW
2060 if (flags & UTF8_WARN_SURROGATE) {
2061 *errors |= UTF8_GOT_SURROGATE;
2062
2063 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2064 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2065 {
2b5e7bc2
KW
2066 pack_warn = packWARN(WARN_SURROGATE);
2067
2068 /* These are the only errors that can occur with a
2069 * surrogate when the 'uv' isn't valid */
2070 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2071 message = Perl_form(aTHX_
2072 "UTF-16 surrogate (any UTF-8 sequence that"
2073 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2074 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2075 }
2076 else {
c94c2f39 2077 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2078 }
37657a5b 2079 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2080 }
2b5e7bc2 2081 }
ba210ebe 2082
2b5e7bc2
KW
2083 if (flags & UTF8_DISALLOW_SURROGATE) {
2084 disallowed = TRUE;
f9380377 2085 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2086 }
2087 }
2088 else if (possible_problems & UTF8_GOT_SUPER) {
2089 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2090
f9380377
KW
2091 if (flags & UTF8_WARN_SUPER) {
2092 *errors |= UTF8_GOT_SUPER;
2093
2094 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2095 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2096 {
2b5e7bc2
KW
2097 pack_warn = packWARN(WARN_NON_UNICODE);
2098
2099 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2100 message = Perl_form(aTHX_
2101 "Any UTF-8 sequence that starts with"
2102 " \"%s\" is for a non-Unicode code point,"
2103 " may not be portable",
7e2f38b2 2104 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2105 }
2106 else {
c94c2f39 2107 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2108 }
37657a5b 2109 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2110 }
2b5e7bc2 2111 }
ba210ebe 2112
57ff5f59
KW
2113 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2114 * and before possibly bailing out, so that the more dire
2115 * warning will override the regular one. */
2116 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2117 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2118 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
37657a5b 2119 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2b5e7bc2 2120 {
db0f09e6 2121 pack_warn = packWARN(WARN_NON_UNICODE);
2b5e7bc2 2122
57ff5f59
KW
2123 /* If it is an overlong that evaluates to a code point
2124 * that doesn't have to use the Perl extended UTF-8, it
2125 * still used it, and so we output a message that
2126 * doesn't refer to the code point. The same is true
2127 * if there was a SHORT malformation where the code
2128 * point is not valid. In that case, 'uv' will have
2129 * been set to the REPLACEMENT CHAR, and the message
2130 * below without the code point in it will be selected
2131 * */
2132 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2133 message = Perl_form(aTHX_
57ff5f59 2134 perl_extended_cp_format, uv);
2b5e7bc2
KW
2135 }
2136 else {
2137 message = Perl_form(aTHX_
57ff5f59
KW
2138 "Any UTF-8 sequence that starts with"
2139 " \"%s\" is a Perl extension, and"
2140 " so is not portable",
2141 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2142 }
37657a5b 2143 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2144 }
2145
d044b7a7
KW
2146 if (flags & ( UTF8_WARN_PERL_EXTENDED
2147 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2148 {
d044b7a7 2149 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2150
d044b7a7 2151 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2152 disallowed = TRUE;
2153 }
2b5e7bc2
KW
2154 }
2155 }
eb83ed87 2156
2b5e7bc2 2157 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2158 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2159 disallowed = TRUE;
2160 }
2b5e7bc2
KW
2161 }
2162 else if (possible_problems & UTF8_GOT_NONCHAR) {
2163 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2164
f9380377
KW
2165 if (flags & UTF8_WARN_NONCHAR) {
2166 *errors |= UTF8_GOT_NONCHAR;
2167
2168 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2169 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2170 {
2b5e7bc2
KW
2171 /* The code above should have guaranteed that we don't
2172 * get here with errors other than overlong */
2173 assert (! (orig_problems
2174 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2175
2176 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2177 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2178 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2179 }
2b5e7bc2 2180 }
5b311467 2181
2b5e7bc2
KW
2182 if (flags & UTF8_DISALLOW_NONCHAR) {
2183 disallowed = TRUE;
f9380377 2184 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2185 }
6c64cd9d
KW
2186 }
2187 else if (possible_problems & UTF8_GOT_LONG) {
2188 possible_problems &= ~UTF8_GOT_LONG;
2189 *errors |= UTF8_GOT_LONG;
2190
2191 if (flags & UTF8_ALLOW_LONG) {
2192
2193 /* We don't allow the actual overlong value, unless the
2194 * special extra bit is also set */
2195 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2196 & ~UTF8_ALLOW_LONG)))
2197 {
2198 uv = UNICODE_REPLACEMENT;
2199 }
2200 }
2201 else {
2202 disallowed = TRUE;
2203
37657a5b
KW
2204 if (( msgs
2205 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2206 {
6c64cd9d
KW
2207 pack_warn = packWARN(WARN_UTF8);
2208
2209 /* These error types cause 'uv' to be something that
2210 * isn't what was intended, so can't use it in the
2211 * message. The other error types either can't
2212 * generate an overlong, or else the 'uv' is valid */
2213 if (orig_problems &
2214 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2215 {
2216 message = Perl_form(aTHX_
2217 "%s: %s (any UTF-8 sequence that starts"
2218 " with \"%s\" is overlong which can and"
2219 " should be represented with a"
2220 " different, shorter sequence)",
2221 malformed_text,
2222 _byte_dump_string(s0, send - s0, 0),
2223 _byte_dump_string(s0, curlen, 0));
2224 }
2225 else {
2226 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2227 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2228 uv, 0);
d819dc50
KW
2229 /* Don't use U+ for non-Unicode code points, which
2230 * includes those in the Latin1 range */
2231 const char * preface = ( uv > PERL_UNICODE_MAX
2232#ifdef EBCDIC
2233 || uv <= 0xFF
2234#endif
2235 )
2236 ? "0x"
2237 : "U+";
6c64cd9d
KW
2238 message = Perl_form(aTHX_
2239 "%s: %s (overlong; instead use %s to represent"
2240 " %s%0*" UVXf ")",
2241 malformed_text,
2242 _byte_dump_string(s0, send - s0, 0),
2243 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2244 preface,
2245 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2246 small code points */
1be62ab9 2247 UNI_TO_NATIVE(uv));
6c64cd9d 2248 }
37657a5b 2249 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2250 }
2251 }
2b5e7bc2
KW
2252 } /* End of looking through the possible flags */
2253
2254 /* Display the message (if any) for the problem being handled in
2255 * this iteration of the loop */
2256 if (message) {
37657a5b 2257 if (msgs) {
37657a5b
KW
2258 assert(this_flag_bit);
2259
2260 if (*msgs == NULL) {
2261 *msgs = newAV();
2262 }
2263
bb07812e
KW
2264 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2265 pack_warn,
2266 this_flag_bit)));
37657a5b
KW
2267 }
2268 else if (PL_op)
2b5e7bc2
KW
2269 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2270 OP_DESC(PL_op));
2271 else
2272 Perl_warner(aTHX_ pack_warn, "%s", message);
2273 }
ddb65933 2274 } /* End of 'while (possible_problems)' */
a0dbb045 2275
2b5e7bc2
KW
2276 /* Since there was a possible problem, the returned length may need to
2277 * be changed from the one stored at the beginning of this function.
2278 * Instead of trying to figure out if that's needed, just do it. */
2279 if (retlen) {
2280 *retlen = curlen;
2281 }
a0dbb045 2282
2b5e7bc2
KW
2283 if (disallowed) {
2284 if (flags & UTF8_CHECK_ONLY && retlen) {
2285 *retlen = ((STRLEN) -1);
2286 }
2287 return 0;
2288 }
eb83ed87 2289 }
ba210ebe 2290
2b5e7bc2 2291 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2292}
2293
8e84507e 2294/*
ec5f19d0
KW
2295=for apidoc utf8_to_uvchr_buf
2296
2297Returns the native code point of the first character in the string C<s> which
2298is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2299C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2300
524080c4
KW
2301If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2302enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2303C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2304(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2305C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2306the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2307See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2308returned.
ec5f19d0
KW
2309
2310=cut
52be2536
KW
2311
2312Also implemented as a macro in utf8.h
2313
ec5f19d0
KW
2314*/
2315
2316
2317UV
2318Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2319{
7f974d7e
KW
2320 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2321
ec5f19d0
KW
2322 assert(s < send);
2323
2324 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 2325 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
2326}
2327
52be2536
KW
2328/* This is marked as deprecated
2329 *
ec5f19d0
KW
2330=for apidoc utf8_to_uvuni_buf
2331
de69f3af
KW
2332Only in very rare circumstances should code need to be dealing in Unicode
2333(as opposed to native) code points. In those few cases, use
07dfe0a4
KW
2334C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2335are not absolutely sure this is one of those cases, then assume it isn't and
2336use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2337
2338Returns the Unicode (not-native) code point of the first character in the
2339string C<s> which
ec5f19d0
KW
2340is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2341C<retlen> will be set to the length, in bytes, of that character.
2342
524080c4
KW
2343If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2344enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2345NULL) to -1. If those warnings are off, the computed value if well-defined (or
2346the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2347is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2348next possible position in C<s> that could begin a non-malformed character.
de69f3af 2349See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0
KW
2350
2351=cut
2352*/
2353
2354UV
2355Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2356{
2357 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2358
2359 assert(send > s);
2360
5962d97e 2361 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2362}
2363
b76347f2 2364/*
87cea99e 2365=for apidoc utf8_length
b76347f2 2366
b2e7ed74
KW
2367Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2368at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2369same place, it returns 0 with no warning raised.
2370
2371If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2372and returns the number of valid characters.
b76347f2
JH
2373
2374=cut
2375*/
2376
2377STRLEN
35a4481c 2378Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2379{
2380 STRLEN len = 0;
2381
7918f24d
NC
2382 PERL_ARGS_ASSERT_UTF8_LENGTH;
2383
8850bf83
JH
2384 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2385 * the bitops (especially ~) can create illegal UTF-8.
2386 * In other words: in Perl UTF-8 is not just for Unicode. */
2387
a3b680e6
AL
2388 if (e < s)
2389 goto warn_and_return;
b76347f2 2390 while (s < e) {
4cbf4130 2391 s += UTF8SKIP(s);
8e91ec7f
AV
2392 len++;
2393 }
2394
2395 if (e != s) {
2396 len--;
2397 warn_and_return:
9b387841
NC
2398 if (PL_op)
2399 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2400 "%s in %s", unees, OP_DESC(PL_op));
2401 else
61a12c31 2402 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2403 }
2404
2405 return len;
2406}
2407
b06226ff 2408/*
fed3ba5d
NC
2409=for apidoc bytes_cmp_utf8
2410
a1433954 2411Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2412sequence of characters (stored as UTF-8)
2413in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2414equal, -1 or -2 if the first string is less than the second string, +1 or +2
2415if the first string is greater than the second string.
2416
2417-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2418longer string. -2 or +2 is returned if
2419there was a difference between characters
fed3ba5d
NC
2420within the strings.
2421
2422=cut
2423*/
2424
2425int
2426Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2427{
2428 const U8 *const bend = b + blen;
2429 const U8 *const uend = u + ulen;
2430
2431 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2432
2433 while (b < bend && u < uend) {
2434 U8 c = *u++;
2435 if (!UTF8_IS_INVARIANT(c)) {
2436 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2437 if (u < uend) {
2438 U8 c1 = *u++;
2439 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2440 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2441 } else {
2b5e7bc2 2442 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2443 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2444 "%s %s%s",
2445 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2446 PL_op ? " in " : "",
2447 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2448 return -2;
2449 }
2450 } else {
2451 if (PL_op)
2452 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2453 "%s in %s", unees, OP_DESC(PL_op));
2454 else
61a12c31 2455 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2456 return -2; /* Really want to return undef :-) */
2457 }
2458 } else {
2459 return -2;
2460 }
2461 }
2462 if (*b != c) {
2463 return *b < c ? -2 : +2;
2464 }
2465 ++b;
2466 }
2467
2468 if (b == bend && u == uend)
2469 return 0;
2470
2471 return b < bend ? +1 : -1;
2472}
2473
2474/*
87cea99e 2475=for apidoc utf8_to_bytes
6940069f 2476
3bc0c78c 2477Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2478Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2479updates C<*lenp> to contain the new length.
3bc0c78c
KW
2480Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2481
2482Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2483having saved the value of C<*lenp> before the call, and subtracting the
2484after-call value of C<*lenp> from it.
6940069f 2485
a1433954 2486If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2487
6940069f
GS
2488=cut
2489*/
2490
2491U8 *
09af0336 2492Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2493{
9fe0d3c2 2494 U8 * first_variant;
246fae53 2495
7918f24d 2496 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2497 PERL_UNUSED_CONTEXT;
7918f24d 2498
9fe0d3c2 2499 /* This is a no-op if no variants at all in the input */
09af0336 2500 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2501 return s;
2502 }
2503
2504 {
3c5aa262 2505 U8 * const save = s;
09af0336 2506 U8 * const send = s + *lenp;
3c5aa262
KW
2507 U8 * d;
2508
2509 /* Nothing before the first variant needs to be changed, so start the real
2510 * work there */
2511 s = first_variant;
2512 while (s < send) {
2513 if (! UTF8_IS_INVARIANT(*s)) {
2514 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2515 *lenp = ((STRLEN) -1);
3c5aa262
KW
2516 return 0;
2517 }
2518 s++;
d59937ca
KW
2519 }
2520 s++;
dcad2880 2521 }
dcad2880 2522
3c5aa262
KW
2523 /* Is downgradable, so do it */
2524 d = s = first_variant;
2525 while (s < send) {
2526 U8 c = *s++;
2527 if (! UVCHR_IS_INVARIANT(c)) {
2528 /* Then it is two-byte encoded */
2529 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2530 s++;
2531 }
2532 *d++ = c;
2533 }
2534 *d = '\0';
09af0336 2535 *lenp = d - save;
3c5aa262
KW
2536
2537 return save;
9fe0d3c2 2538 }
6940069f
GS
2539}
2540
2541/*
87cea99e 2542=for apidoc bytes_from_utf8
f9a63242 2543
09af0336 2544Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2545byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2546actually encoded in UTF-8.
2547
2548Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2549the input string.
2550
41ae6089
KW
2551Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2552not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2553C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2554
41ae6089 2555Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2556newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2557returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2558caller is responsible for arranging for the memory used by this string to get
2559freed.
f9a63242 2560
3bc0c78c 2561Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2562having saved the value of C<*lenp> before the call, and subtracting the
2563after-call value of C<*lenp> from it.
3bc0c78c 2564
37607a96 2565=cut
976c1b08
KW
2566
2567There is a macro that avoids this function call, but this is retained for
2568anyone who calls it with the Perl_ prefix */
f9a63242
JH
2569
2570U8 *
41ae6089 2571Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2572{
7918f24d 2573 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2574 PERL_UNUSED_CONTEXT;
f9a63242 2575
976c1b08
KW
2576 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2577}
2578
2579/*
2580No = here because currently externally undocumented
2581for apidoc bytes_from_utf8_loc
2582
2583Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2584to store the location of the first character in C<"s"> that cannot be
2585converted to non-UTF8.
2586
2587If that parameter is C<NULL>, this function behaves identically to
2588C<bytes_from_utf8>.
2589
2590Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2591C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2592
2593Otherwise, the function returns a newly created C<NUL>-terminated string
2594containing the non-UTF8 equivalent of the convertible first portion of
2595C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2596If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2597and C<*first_non_downgradable> is set to C<NULL>.
2598
2599Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2600first character in the original string that wasn't converted. C<*is_utf8p> is
2601unchanged. Note that the new string may have length 0.
2602
2603Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2604C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2605converts as many characters in it as possible stopping at the first one it
385b74be 2606finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2607set to point to that. The function returns the portion that could be converted
2608in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2609not including the terminating C<NUL>. If the very first character in the
2610original could not be converted, C<*lenp> will be 0, and the new string will
2611contain just a single C<NUL>. If the entire input string was converted,
2612C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2613
2614Upon successful return, the number of variants in the converted portion of the
2615string can be computed by having saved the value of C<*lenp> before the call,
2616and subtracting the after-call value of C<*lenp> from it.
2617
2618=cut
2619
2620
2621*/
2622
2623U8 *
2624Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2625{
2626 U8 *d;
2627 const U8 *original = s;
2628 U8 *converted_start;
2629 const U8 *send = s + *lenp;
f9a63242 2630
976c1b08 2631 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2632
976c1b08
KW
2633 if (! *is_utf8p) {
2634 if (first_unconverted) {
2635 *first_unconverted = NULL;
2636 }
2637
2638 return (U8 *) original;
2639 }
2640
2641 Newx(d, (*lenp) + 1, U8);
2642
2643 converted_start = d;
7299a045
KW
2644 while (s < send) {
2645 U8 c = *s++;
2646 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2647
2648 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2649 * have to stop now */
2650 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2651 if (first_unconverted) {
2652 *first_unconverted = s - 1;
2653 goto finish_and_return;
2654 }
2655 else {
2656 Safefree(converted_start);
2657 return (U8 *) original;
2658 }
2659 }
2660
7299a045
KW
2661 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2662 s++;
38af28cf 2663 }
7299a045
KW
2664 *d++ = c;
2665 }
170a1c22 2666
976c1b08
KW
2667 /* Here, converted the whole of the input */
2668 *is_utf8p = FALSE;
2669 if (first_unconverted) {
2670 *first_unconverted = NULL;
170a1c22 2671 }
976c1b08
KW
2672
2673 finish_and_return:
46a08a6f
KW
2674 *d = '\0';
2675 *lenp = d - converted_start;
976c1b08
KW
2676
2677 /* Trim unused space */
2678 Renew(converted_start, *lenp + 1, U8);
2679
2680 return converted_start;
f9a63242
JH
2681}
2682
2683/*
87cea99e 2684=for apidoc bytes_to_utf8
6940069f 2685
09af0336 2686Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2687UTF-8.
09af0336 2688Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2689reflect the new length in bytes. The caller is responsible for arranging for
2690the memory used by this string to get freed.
6940069f 2691
3bc0c78c 2692Upon successful return, the number of variants in the string can be computed by
23b37b12 2693having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2694after-call value of C<*lenp>.
2695
75200dff 2696A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2697
2698If you want to convert to UTF-8 from encodings other than
2699the native (Latin1 or EBCDIC),
a1433954 2700see L</sv_recode_to_utf8>().
c9ada85f 2701
497711e7 2702=cut
6940069f
GS
2703*/
2704
2705U8*
09af0336 2706Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2707{
09af0336 2708 const U8 * const send = s + (*lenp);
6940069f
GS
2709 U8 *d;
2710 U8 *dst;
7918f24d
NC
2711
2712 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2713 PERL_UNUSED_CONTEXT;
6940069f 2714
09af0336 2715 Newx(d, (*lenp) * 2 + 1, U8);
6940069f
GS
2716 dst = d;
2717
2718 while (s < send) {
55d09dc8
KW
2719 append_utf8_from_native_byte(*s, &d);
2720 s++;
6940069f 2721 }
2e11cf67 2722
6940069f 2723 *d = '\0';
09af0336 2724 *lenp = d-dst;
2e11cf67
KW
2725
2726 /* Trim unused space */
2727 Renew(dst, *lenp + 1, U8);
2728
6940069f
GS
2729 return dst;
2730}
2731
a0ed51b3 2732/*
624504c5
KW
2733 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2734 * use utf16_to_utf8_reversed().
a0ed51b3 2735 *
624504c5
KW
2736 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2737 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2738 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2739 *
2740 * These functions don't check for overflow. The worst case is every code
2741 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2742 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2743 * destination must be pre-extended to 2 times the source length.
2744 *
2745 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2746
2747U8*
dea0fc0b 2748Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2749{
dea0fc0b
JH
2750 U8* pend;
2751 U8* dstart = d;
2752
7918f24d
NC
2753 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2754
dea0fc0b 2755 if (bytelen & 1)
56576a04
KW
2756 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2757 (UV)bytelen);
dea0fc0b
JH
2758
2759 pend = p + bytelen;
2760
a0ed51b3 2761 while (p < pend) {
dea0fc0b
JH
2762 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2763 p += 2;
2d1545e5 2764 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2765 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2766 continue;
2767 }
56d37426
KW
2768 if (uv <= MAX_UTF8_TWO_BYTE) {
2769 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2770 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2771 continue;
2772 }
ffd0a9d3 2773
46956fad
KW
2774#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2775#define LAST_HIGH_SURROGATE 0xDBFF
2776#define FIRST_LOW_SURROGATE 0xDC00
2777#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2778#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2779
2780 /* This assumes that most uses will be in the first Unicode plane, not
2781 * needing surrogates */
2782 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2783 && uv <= UNICODE_SURROGATE_LAST))
2784 {
2785 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2786 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2787 }
2788 else {
01ea242b 2789 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2790 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2791 || UNLIKELY(low > LAST_LOW_SURROGATE))
2792 {
01ea242b 2793 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2794 }
2795 p += 2;
46956fad 2796 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2797 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2798 }
a0ed51b3 2799 }
56d37426
KW
2800#ifdef EBCDIC
2801 d = uvoffuni_to_utf8_flags(d, uv, 0);
2802#else
ffd0a9d3 2803 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2804 *d++ = (U8)(( uv >> 12) | 0xe0);
2805 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2806 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2807 continue;
2808 }
2809 else {
eb160463
GS
2810 *d++ = (U8)(( uv >> 18) | 0xf0);
2811 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2812 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2813 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2814 continue;
2815 }
56d37426 2816#endif
a0ed51b3 2817 }
dea0fc0b 2818 *newlen = d - dstart;
a0ed51b3
LW
2819 return d;
2820}
2821
2822/* Note: this one is slightly destructive of the source. */
2823
2824U8*
dea0fc0b 2825Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2826{
2827 U8* s = (U8*)p;
d4c19fe8 2828 U8* const send = s + bytelen;
7918f24d
NC
2829
2830 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2831
e0ea5e2d 2832 if (bytelen & 1)
147e3846 2833 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2834 (UV)bytelen);
2835
a0ed51b3 2836 while (s < send) {
d4c19fe8 2837 const U8 tmp = s[0];
a0ed51b3
LW
2838 s[0] = s[1];
2839 s[1] = tmp;
2840 s += 2;
2841 }
dea0fc0b 2842 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2843}
2844
922e8cb4
KW
2845bool
2846Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2847{
2848 U8 tmpbuf[UTF8_MAXBYTES+1];
2849 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2850 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2851}
2852
f9ae8fb6
JD
2853/* Internal function so we can deprecate the external one, and call
2854 this one from other deprecated functions in this file */
2855
f2645549
KW
2856bool
2857Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2858{
f2645549 2859 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2860
2861 if (*p == '_')
2862 return TRUE;
f25ce844 2863 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2864}
2865
5092f92a 2866bool
eba68aa0
KW
2867Perl__is_uni_perl_idcont(pTHX_ UV c)
2868{
2869 U8 tmpbuf[UTF8_MAXBYTES+1];
2870 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2871 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2872}
2873
2874bool
f91dcd13
KW
2875Perl__is_uni_perl_idstart(pTHX_ UV c)
2876{
2877 U8 tmpbuf[UTF8_MAXBYTES+1];
2878 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2879 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2880}
2881
3a4c58c9 2882UV
56576a04
KW
2883Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2884 const char S_or_s)
3a4c58c9
KW
2885{
2886 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2887 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2888 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2889 * either "SS" or "Ss". Which one to use is passed into the routine in
2890 * 'S_or_s' to avoid a test */
2891
2892 UV converted = toUPPER_LATIN1_MOD(c);
2893
2894 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2895
2896 assert(S_or_s == 'S' || S_or_s == 's');
2897
6f2d5cbc 2898 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2899 characters in this range */
3a4c58c9
KW
2900 *p = (U8) converted;
2901 *lenp = 1;
2902 return converted;
2903 }
2904
2905 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2906 * which it maps to one of them, so as to only have to have one check for
2907 * it in the main case */
2908 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2909 switch (c) {
2910 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2911 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2912 break;
2913 case MICRO_SIGN:
2914 converted = GREEK_CAPITAL_LETTER_MU;
2915 break;
79e064b9
KW
2916#if UNICODE_MAJOR_VERSION > 2 \
2917 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2918 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2919 case LATIN_SMALL_LETTER_SHARP_S:
2920 *(p)++ = 'S';
2921 *p = S_or_s;
2922 *lenp = 2;
2923 return 'S';
79e064b9 2924#endif
3a4c58c9 2925 default:
56576a04
KW
2926 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2927 " '%c' to map to '%c'",
2928 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2929 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2930 }
2931 }
2932
2933 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2934 *p = UTF8_TWO_BYTE_LO(converted);
2935 *lenp = 2;
2936
2937 return converted;
2938}
2939
fe63c520
KW
2940/* If compiled on an early Unicode version, there may not be auxiliary tables
2941 * */
2942#ifndef HAS_UC_AUX_TABLES
2943# define UC_AUX_TABLE_ptrs NULL
2944# define UC_AUX_TABLE_lengths NULL
2945#endif
2946#ifndef HAS_TC_AUX_TABLES
2947# define TC_AUX_TABLE_ptrs NULL
2948# define TC_AUX_TABLE_lengths NULL
2949#endif
2950#ifndef HAS_LC_AUX_TABLES
2951# define LC_AUX_TABLE_ptrs NULL
2952# define LC_AUX_TABLE_lengths NULL
2953#endif
2954#ifndef HAS_CF_AUX_TABLES
2955# define CF_AUX_TABLE_ptrs NULL
2956# define CF_AUX_TABLE_lengths NULL
2957#endif
2958#ifndef HAS_UC_AUX_TABLES
2959# define UC_AUX_TABLE_ptrs NULL
2960# define UC_AUX_TABLE_lengths NULL
2961#endif
2962
50bda2c3
KW
2963/* Call the function to convert a UTF-8 encoded character to the specified case.
2964 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2965 * 's' is a pointer to the first byte of the input character
2966 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2967 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2968 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2969 *
56576a04 2970 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2971 * 'd' */
56576a04 2972#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2973 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2974 Uppercase_Mapping_invmap, \
2975 UC_AUX_TABLE_ptrs, \
2976 UC_AUX_TABLE_lengths, \
2977 "uppercase")
56576a04 2978#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2979 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2980 Titlecase_Mapping_invmap, \
2981 TC_AUX_TABLE_ptrs, \
2982 TC_AUX_TABLE_lengths, \
2983 "titlecase")
56576a04 2984#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2985 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2986 Lowercase_Mapping_invmap, \
2987 LC_AUX_TABLE_ptrs, \
2988 LC_AUX_TABLE_lengths, \
2989 "lowercase")
2990
50bda2c3 2991
b9992569
KW
2992/* This additionally has the input parameter 'specials', which if non-zero will
2993 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2994 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2995#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2996 (specials) \
2997 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2998 Case_Folding_invmap, \
2999 CF_AUX_TABLE_ptrs, \
3000 CF_AUX_TABLE_lengths, \
3001 "foldcase") \
3002 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
3003 Simple_Case_Folding_invmap, \
3004 NULL, NULL, \
3005 "foldcase")
c3fd2246 3006
84afefe6
JH
3007UV
3008Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3009{
a1433954
KW
3010 /* Convert the Unicode character whose ordinal is <c> to its uppercase
3011 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
3012 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
3013 * the changed version may be longer than the original character.
3014 *
3015 * The ordinal of the first character of the changed version is returned
3016 * (but note, as explained above, that there may be more.) */
3017
7918f24d
NC
3018 PERL_ARGS_ASSERT_TO_UNI_UPPER;
3019
3a4c58c9
KW
3020 if (c < 256) {
3021 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
3022 }
3023
0ebc6274 3024 uvchr_to_utf8(p, c);
b9992569 3025 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
3026}
3027
84afefe6
JH
3028UV
3029Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3030{
7918f24d
NC
3031 PERL_ARGS_ASSERT_TO_UNI_TITLE;
3032
3a4c58c9
KW
3033 if (c < 256) {
3034 return _to_upper_title_latin1((U8) c, p, lenp, 's');
3035 }
3036
0ebc6274 3037 uvchr_to_utf8(p, c);
b9992569 3038 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
3039}
3040
afc16117 3041STATIC U8
eaf412bf 3042S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
3043{
3044 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 3045 * those, converting the result to UTF-8. Since the result is always just
a1433954 3046 * one character, we allow <p> to be NULL */
afc16117
KW
3047
3048 U8 converted = toLOWER_LATIN1(c);
3049
eaf412bf
KW
3050 PERL_UNUSED_ARG(dummy);
3051
afc16117 3052 if (p != NULL) {
6f2d5cbc 3053 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
3054 *p = converted;
3055 *lenp = 1;
3056 }
3057 else {
430c9760
KW
3058 /* Result is known to always be < 256, so can use the EIGHT_BIT
3059 * macros */
3060 *p = UTF8_EIGHT_BIT_HI(converted);
3061 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3062 *lenp = 2;
3063 }
3064 }
3065 return converted;
3066}
3067
84afefe6
JH
3068UV
3069Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3070{
7918f24d
NC
3071 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3072
afc16117 3073 if (c < 256) {
eaf412bf 3074 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3075 }
3076
afc16117 3077 uvchr_to_utf8(p, c);
b9992569 3078 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
3079}
3080
84afefe6 3081UV
56576a04
KW
3082Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
3083 const unsigned int flags)
a1dde8de 3084{
51910141 3085 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3086 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3087 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3088 *
3089 * Not to be used for locale folds
51910141 3090 */
f673fad4 3091
a1dde8de
KW
3092 UV converted;
3093
3094 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 3095 PERL_UNUSED_CONTEXT;
a1dde8de 3096
1ca267a5
KW
3097 assert (! (flags & FOLD_FLAGS_LOCALE));
3098
659a7c2d 3099 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3100 converted = GREEK_SMALL_LETTER_MU;
3101 }
9b63e895
KW
3102#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3103 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3104 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3105 else if ( (flags & FOLD_FLAGS_FULL)
3106 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3107 {
1ca267a5
KW
3108 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3109 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3110 * under those circumstances. */
3111 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3112 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3113 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3114 p, *lenp, U8);
3115 return LATIN_SMALL_LETTER_LONG_S;
3116 }
3117 else {
4f489194
KW
3118 *(p)++ = 's';
3119 *p = 's';
3120 *lenp = 2;
3121 return 's';
1ca267a5 3122 }
a1dde8de 3123 }
9b63e895 3124#endif
a1dde8de
KW
3125 else { /* In this range the fold of all other characters is their lower
3126 case */
3127 converted = toLOWER_LATIN1(c);
3128 }
3129
6f2d5cbc 3130 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3131 *p = (U8) converted;
3132 *lenp = 1;
3133 }
3134 else {
3135 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3136 *p = UTF8_TWO_BYTE_LO(converted);
3137 *lenp = 2;
3138 }
3139
3140 return converted;
3141}
3142
3143UV
31f05a37 3144Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3145{
4b593389 3146
a0270393
KW
3147 /* Not currently externally documented, and subject to change
3148 * <flags> bits meanings:
3149 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3150 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3151 * locale are to be used.
a0270393
KW
3152 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3153 */
4b593389 3154
36bb2ab6 3155 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3156
780fcc9f 3157 if (flags & FOLD_FLAGS_LOCALE) {
8b7358b9
KW
3158 /* Treat a UTF-8 locale as not being in locale at all, except for
3159 * potentially warning */
3160 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
780fcc9f
KW
3161 if (IN_UTF8_CTYPE_LOCALE) {
3162 flags &= ~FOLD_FLAGS_LOCALE;
3163 }
3164 else {
e7b7ac46 3165 goto needs_full_generality;
780fcc9f 3166 }
31f05a37
KW
3167 }
3168
a1dde8de 3169 if (c < 256) {
e7b7ac46 3170 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3171 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3172 }
3173
2f306ab9 3174 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
3175 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3176 uvchr_to_utf8(p, c);
b9992569 3177 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3178 }
567b353c 3179 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3180 the special flags. */
3181 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3182
3183 needs_full_generality:
a0270393 3184 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3185 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3186 p, lenp, flags);
a0270393 3187 }
84afefe6
JH
3188}
3189
26483009 3190PERL_STATIC_INLINE bool
5141f98e 3191S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 3192 const char *const swashname, SV* const invlist)
bde6a22d 3193{
ea317ccb
KW
3194 /* returns a boolean giving whether or not the UTF8-encoded character that
3195 * starts at <p> is in the swash indicated by <swashname>. <swash>
3196 * contains a pointer to where the swash indicated by <swashname>
3197 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
3198 * look at <*swash> and only generate a swash if it is not null. <invlist>
3199 * is NULL or an inversion list that defines the swash. If not null, it
3200 * saves time during initialization of the swash.
ea317ccb
KW
3201 *
3202 * Note that it is assumed that the buffer length of <p> is enough to
3203 * contain all the bytes that comprise the character. Thus, <*p> should
3204 * have been checked before this call for mal-formedness enough to assure
3205 * that. */
3206
7918f24d
NC
3207 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3208
492a624f 3209 /* The API should have included a length for the UTF-8 character in <p>,
28123549 3210 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
3211 * as far as there being enough bytes available in it to accommodate the
3212 * character without reading beyond the end, and pass that number on to the
3213 * validating routine */
6302f837 3214 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 3215 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 3216 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
3217 1 /* Die */ );
3218 NOT_REACHED; /* NOTREACHED */
28123549 3219 }
86ae6e94 3220
87367d5f
KW
3221 if (!*swash) {
3222 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
3223 *swash = _core_swash_init("utf8",
3224
3225 /* Only use the name if there is no inversion
3226 * list; otherwise will go out to disk */
3227 (invlist) ? "" : swashname,
3228
3229 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 3230 }
28123549 3231
bde6a22d
NC
3232 return swash_fetch(*swash, p, TRUE) != 0;
3233}
3234
da8c1a98 3235PERL_STATIC_INLINE bool
56576a04
KW
3236S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3237 SV **swash, const char *const swashname,
3238 SV* const invlist)
da8c1a98
KW
3239{
3240 /* returns a boolean giving whether or not the UTF8-encoded character that
3241 * starts at <p>, and extending no further than <e - 1> is in the swash
3242 * indicated by <swashname>. <swash> contains a pointer to where the swash
3243 * indicated by <swashname> is to be stored; which this routine will do, so
3244 * that future calls will look at <*swash> and only generate a swash if it
3245 * is not null. <invlist> is NULL or an inversion list that defines the
3246 * swash. If not null, it saves time during initialization of the swash.
3247 */
3248
3249 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3250
3251 if (! isUTF8_CHAR(p, e)) {
3252 _force_out_malformed_utf8_message(p, e, 0, 1);
3253 NOT_REACHED; /* NOTREACHED */
3254 }
3255
3256 if (!*swash) {
3257 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3258 *swash = _core_swash_init("utf8",
3259
3260 /* Only use the name if there is no inversion
3261 * list; otherwise will go out to disk */
3262 (invlist) ? "" : swashname,
3263
3264 &PL_sv_undef, 1, 0, invlist, &flags);
3265 }
3266
3267 return swash_fetch(*swash, p, TRUE) != 0;
3268}
3269
34aeb2e9
KW
3270STATIC void
3271S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3272 const char * const alternative,
3273 const bool use_locale,
3274 const char * const file,
3275 const unsigned line)
3276{
3277 const char * key;
3278
3279 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3280
3281 if (ckWARN_d(WARN_DEPRECATED)) {
3282
3283 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3284 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3285 if (! PL_seen_deprecated_macro) {
3286 PL_seen_deprecated_macro = newHV();
3287 }
3288 if (! hv_store(PL_seen_deprecated_macro, key,
3289 strlen(key), &PL_sv_undef, 0))
3290 {
3291 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3292 }
3293
c44e9413 3294 if (instr(file, "mathoms.c")) {
607313a1
KW
3295 Perl_warner(aTHX_ WARN_DEPRECATED,
3296 "In %s, line %d, starting in Perl v5.30, %s()"
3297 " will be removed. Avoid this message by"
3298 " converting to use %s().\n",
3299 file, line, name, alternative);
3300 }
3301 else {
34aeb2e9
KW
3302 Perl_warner(aTHX_ WARN_DEPRECATED,
3303 "In %s, line %d, starting in Perl v5.30, %s() will"
3304 " require an additional parameter. Avoid this"
3305 " message by converting to use %s().\n",
3306 file, line, name, alternative);
607313a1 3307 }
34aeb2e9
KW
3308 }
3309 }
3310}
3311
bde6a22d 3312bool
34aeb2e9 3313Perl__is_utf8_FOO(pTHX_ U8 classnum,
be99e2c2 3314 const U8 * const p,
34aeb2e9
KW
3315 const char * const name,
3316 const char * const alternative,
3317 const bool use_utf8,
3318 const bool use_locale,
3319 const char * const file,
3320 const unsigned line)
922e8cb4 3321{
922e8cb4
KW
3322 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3323
34aeb2e9
KW
3324 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3325
3326 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
3327
3328 switch (classnum) {
3329 case _CC_WORDCHAR:
3330 case _CC_DIGIT:
3331 case _CC_ALPHA:
3332 case _CC_LOWER:
3333 case _CC_UPPER:
3334 case _CC_PUNCT:
3335 case _CC_PRINT:
3336 case _CC_ALPHANUMERIC:
3337 case _CC_GRAPH:
3338 case _CC_CASED:
3339
3340 return is_utf8_common(p,
3341 &PL_utf8_swash_ptrs[classnum],
3342 swash_property_names[classnum],
3343 PL_XPosix_ptrs[classnum]);
3344
3345 case _CC_SPACE:
3346 return is_XPERLSPACE_high(p);
3347 case _CC_BLANK:
3348 return is_HORIZWS_high(p);
3349 case _CC_XDIGIT:
3350 return is_XDIGIT_high(p);
3351 case _CC_CNTRL:
3352 return 0;
3353 case _CC_ASCII:
3354 return 0;
3355 case _CC_VERTSPACE:
3356 return is_VERTWS_high(p);
3357 case _CC_IDFIRST:
3358 if (! PL_utf8_perl_idstart) {
22f0498f
KW
3359 PL_utf8_perl_idstart
3360 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 3361 }
22f0498f
KW
3362 return is_utf8_common(p, &PL_utf8_perl_idstart,
3363 "_Perl_IDStart", NULL);
34aeb2e9
KW
3364 case _CC_IDCONT:
3365 if (! PL_utf8_perl_idcont) {
22f0498f
KW
3366 PL_utf8_perl_idcont
3367 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 3368 }
22f0498f
KW
3369 return is_utf8_common(p, &PL_utf8_perl_idcont,
3370 "_Perl_IDCont", NULL);
34aeb2e9
KW
3371 }
3372 }
3373
3374 /* idcont is the same as wordchar below 256 */
3375 if (classnum == _CC_IDCONT) {
3376 classnum = _CC_WORDCHAR;
3377 }
3378 else if (classnum == _CC_IDFIRST) {
3379 if (*p == '_') {
3380 return TRUE;
3381 }
3382 classnum = _CC_ALPHA;
3383 }
3384
3385 if (! use_locale) {
3386 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3387 return _generic_isCC(*p, classnum);
3388 }
922e8cb4 3389
34aeb2e9
KW
3390 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3391 }
3392 else {
3393 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3394 return isFOO_lc(classnum, *p);
3395 }
3396
3397 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3398 }
3399
3400 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
3401}
3402
3403bool
da8c1a98
KW
3404Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3405 const U8 * const e)
3406{
3407 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3408
3409 assert(classnum < _FIRST_NON_SWASH_CC);
3410
3411 return is_utf8_common_with_len(p,
3412 e,
3413 &PL_utf8_swash_ptrs[classnum],
3414 swash_property_names[classnum],
3415 PL_XPosix_ptrs[classnum]);
3416}
3417
3418bool
da8c1a98
KW
3419Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3420{
3421 SV* invlist = NULL;
3422
3423 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3424
3425 if (! PL_utf8_perl_idstart) {
3426 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
3427 }
3428 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
3429 "_Perl_IDStart", invlist);
3430}
3431
3432bool
f2645549 3433Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 3434{
f2645549 3435 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
3436
3437 if (*p == '_')
3438 return TRUE;
f25ce844 3439 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
3440}
3441
3442bool
da8c1a98
KW
3443Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3444{
3445 SV* invlist = NULL;
3446
3447 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3448
3449 if (! PL_utf8_perl_idcont) {
3450 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3451 }
3452 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3453 "_Perl_IDCont", invlist);
3454}
3455
3456bool
f2645549 3457Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 3458{
f2645549 3459 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 3460
f25ce844 3461 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
3462}
3463
3464bool
f2645549 3465Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 3466{
f2645549 3467 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 3468
f25ce844 3469 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
3470}
3471
3472bool
7dbf68d2
KW
3473Perl__is_utf8_mark(pTHX_ const U8 *p)
3474{
7dbf68d2
KW
3475 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3476
f25ce844 3477 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
3478}
3479
6a4a25f4 3480STATIC UV
30613bdc
KW
3481S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3482 U8* ustrp, STRLEN *lenp,
3483 SV *invlist, const IV * const invmap,
3484 const int * const * aux_tables,
3485 const U8 * const aux_table_lengths,
3486 const char * const normal)
b9992569 3487{
0134edef 3488 STRLEN len = 0;
7918f24d 3489
30613bdc
KW
3490 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3491 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3492 * to name the new case in any generated messages, as a fallback if the
3493 * operation being used is not available. The new case is given by the
3494 * data structures in the remaining arguments.
3495 *
3496 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3497 * entire changed case string, and the return value is the first code point
3498 * in that string */
3499
b9992569 3500 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3501
36eaa811
KW
3502 /* For code points that don't change case, we already know that the output
3503 * of this function is the unchanged input, so we can skip doing look-ups
3504 * for them. Unfortunately the case-changing code points are scattered
3505 * around. But there are some long consecutive ranges where there are no
3506 * case changing code points. By adding tests, we can eliminate the lookup
3507 * for all the ones in such ranges. This is currently done here only for
3508 * just a few cases where the scripts are in common use in modern commerce
3509 * (and scripts adjacent to those which can be included without additional
3510 * tests). */
3511
3512 if (uv1 >= 0x0590) {
3513 /* This keeps from needing further processing the code points most
3514 * likely to be used in the following non-cased scripts: Hebrew,
3515 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3516 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3517 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3518 if (uv1 < 0x10A0) {
3519 goto cases_to_self;
3520 }
3521
3522 /* The following largish code point ranges also don't have case
3523 * changes, but khw didn't think they warranted extra tests to speed
3524 * them up (which would slightly slow down everything else above them):
3525 * 1100..139F Hangul Jamo, Ethiopic
3526 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3527 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3528 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3529 * Combining Diacritical Marks Extended, Balinese,
3530 * Sundanese, Batak, Lepcha, Ol Chiki
3531 * 2000..206F General Punctuation
3532 */
3533
3534 if (uv1 >= 0x2D30) {
3535
3536 /* This keeps the from needing further processing the code points
3537 * most likely to be used in the following non-cased major scripts:
3538 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3539 *
3540 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3541 * event that Unicode eventually allocates the unused block as of
3542 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3543 * that the test suite will start having failures to alert you
3544 * should that happen) */
3545 if (uv1 < 0xA640) {
3546 goto cases_to_self;
3547 }
3548
3549 if (uv1 >= 0xAC00) {
3550 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3551 if (ckWARN_d(WARN_SURROGATE)) {
3552 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3553 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3554 "Operation \"%s\" returns its argument for"
3555 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3556 }
3557 goto cases_to_self;
3558 }
36eaa811
KW
3559
3560 /* AC00..FAFF Catches Hangul syllables and private use, plus
3561 * some others */
3562 if (uv1 < 0xFB00) {
3563 goto cases_to_self;
36eaa811
KW
3564 }
3565
5af9bc97 3566 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
d22ec717
KW
3567 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3568 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3569 MAX_EXTERNALLY_LEGAL_CP);
5af9bc97
KW
3570 }
3571 if (ckWARN_d(WARN_NON_UNICODE)) {
3572 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3573 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3574 "Operation \"%s\" returns its argument for"
3575 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3576 }
3577 goto cases_to_self;
3578 }
3bfc1e70
KW
3579#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3580 if (UNLIKELY(uv1
3581 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3582 {
3583
56576a04
KW
3584 /* As of Unicode 10.0, this means we avoid swash creation
3585 * for anything beyond high Plane 1 (below emojis) */
3bfc1e70
KW
3586 goto cases_to_self;
3587 }
3588#endif
36eaa811
KW
3589 }
3590 }
9ae3ac1a 3591
36eaa811 3592 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3593 * be given. */
9ae3ac1a
KW
3594 }
3595
8946fcd9
KW
3596 {
3597 unsigned int i;
3598 const int * cp_list;
3599 U8 * d;
3600 SSize_t index = _invlist_search(invlist, uv1);
3601 IV base = invmap[index];
0134edef 3602
30613bdc
KW
3603 /* The data structures are set up so that if 'base' is non-negative,
3604 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3605 if (base >= 0) {
3606 IV lc;
b08cf34e 3607
8946fcd9
KW
3608 if (base == 0) {
3609 goto cases_to_self;
4a8240a3 3610 }
4a8240a3 3611
30613bdc 3612 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3613 lc = base + uv1 - invlist_array(invlist)[index];
3614 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3615 return lc;
3616 }
1feea2c7 3617
30613bdc
KW
3618 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3619 * requires an auxiliary table look up. abs(base) gives the index into
3620 * a list of such tables which points to the proper aux table. And a
3621 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3622 cp_list = aux_tables[-base];
30613bdc
KW
3623
3624 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3625 d = ustrp;
3626 for (i = 0; i < aux_table_lengths[-base]; i++) {
3627 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3628 }
8946fcd9
KW
3629 *d = '\0';
3630 *lenp = d - ustrp;
3631
3632 return cp_list[0];
cbe07460
KW
3633 }
3634
3635 /* Here, there was no mapping defined, which means that the code point maps
3636 * to itself. Return the inputs */
e24dfe9c 3637 cases_to_self:
bfdf22ec 3638 len = UTF8SKIP(p);
ca9fab46
KW
3639 if (p != ustrp) { /* Don't copy onto itself */
3640 Copy(p, ustrp, len, U8);
3641 }
0134edef 3642
2a37f04d
JH
3643 if (lenp)
3644 *lenp = len;
3645
f4cd282c 3646 return uv1;
cbe07460 3647
a0ed51b3
LW
3648}
3649
051a06d4 3650STATIC UV
56576a04
KW
3651S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3652 U8* const ustrp, STRLEN *lenp)
051a06d4 3653{
4a4088c4 3654 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3655 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3656 * result contains a character that crosses the 255/256 boundary, disallow
3657 * the change, and return the original code point. See L<perlfunc/lc> for
3658 * why;
051a06d4 3659 *
a1433954
KW
3660 * p points to the original string whose case was changed; assumed
3661 * by this routine to be well-formed
051a06d4 3662 * result the code point of the first character in the changed-case string
56576a04
KW
3663 * ustrp points to the changed-case string (<result> represents its
3664 * first char)
051a06d4
KW
3665 * lenp points to the length of <ustrp> */
3666
3667 UV original; /* To store the first code point of <p> */
3668
3669 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3670
a4f12ed7 3671 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3672
3673 /* We know immediately if the first character in the string crosses the
3674 * boundary, so can skip */
3675 if (result > 255) {
3676
3677 /* Look at every character in the result; if any cross the
3678 * boundary, the whole thing is disallowed */
3679 U8* s = ustrp + UTF8SKIP(ustrp);
3680 U8* e = ustrp + *lenp;
3681 while (s < e) {
a4f12ed7 3682 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3683 goto bad_crossing;
3684 }
3685 s += UTF8SKIP(s);
3686 }
3687
613abc6d
KW
3688 /* Here, no characters crossed, result is ok as-is, but we warn. */
3689 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3690 return result;
3691 }
3692
7b52d656 3693 bad_crossing:
051a06d4
KW
3694
3695 /* Failed, have to return the original */
4b88fb76 3696 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3697
3698 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3699 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3700 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3701 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3702 OP_DESC(PL_op),
ab0b796c
KW
3703 original,
3704 original);
051a06d4
KW
3705 Copy(p, ustrp, *lenp, char);
3706 return original;
3707}
3708
607313a1
KW
3709STATIC U32
3710S_check_and_deprecate(pTHX_ const U8 *p,
3711 const U8 **e,
3712 const unsigned int type, /* See below */
3713 const bool use_locale, /* Is this a 'LC_'
3714 macro call? */
3715 const char * const file,
3716 const unsigned line)
3717{
3718 /* This is a temporary function to deprecate the unsafe calls to the case
3719 * changing macros and functions. It keeps all the special stuff in just
3720 * one place.
3721 *
3722 * It updates *e with the pointer to the end of the input string. If using
3723 * the old-style macros, *e is NULL on input, and so this function assumes
3724 * the input string is long enough to hold the entire UTF-8 sequence, and
3725 * sets *e accordingly, but it then returns a flag to pass the
3726 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3727 * using the full length if possible.
3728 *
3729 * It also does the assert that *e > p when *e is not NULL. This should be
3730 * migrated to the callers when this function gets deleted.
3731 *
3732 * The 'type' parameter is used for the caller to specify which case
3733 * changing function this is called from: */
3734
3735# define DEPRECATE_TO_UPPER 0
3736# define DEPRECATE_TO_TITLE 1
3737# define DEPRECATE_TO_LOWER 2
3738# define DEPRECATE_TO_FOLD 3
3739
3740 U32 utf8n_flags = 0;
3741 const char * name;
3742 const char * alternative;
3743
3744 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3745
3746 if (*e == NULL) {
3747 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;