This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
perlapi: Consolidate some references to perlguts
[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 39
48ef279e 40/*
ccfc67b7 41=head1 Unicode Support
7fefc6c1 42These are various utility functions for manipulating UTF8-encoded
72d33970 43strings. For the uninitiated, this is a method of representing arbitrary
61296642 44Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
45characters in the ASCII range are unmodified, and a zero byte never appears
46within non-zero characters.
166f8a29 47
eaf7a4d2
CS
48=cut
49*/
50
dd051059
DM
51/* helper for Perl__force_out_malformed_utf8_message(). Like
52 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
53 * PL_compiling */
54
55static void
56S_restore_cop_warnings(pTHX_ void *p)
57{
1943af61 58 free_and_set_cop_warnings(PL_curcop, (STRLEN*) p);
dd051059
DM
59}
60
61
9cbfb8ab
KW
62void
63Perl__force_out_malformed_utf8_message(pTHX_
64 const U8 *const p, /* First byte in UTF-8 sequence */
65 const U8 * const e, /* Final byte in sequence (may include
66 multiple chars */
67 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
68 usually 0, or some DISALLOW flags */
69 const bool die_here) /* If TRUE, this function does not return */
70{
71 /* This core-only function is to be called when a malformed UTF-8 character
72 * is found, in order to output the detailed information about the
73 * malformation before dieing. The reason it exists is for the occasions
74 * when such a malformation is fatal, but warnings might be turned off, so
75 * that normally they would not be actually output. This ensures that they
76 * do get output. Because a sequence may be malformed in more than one
77 * way, multiple messages may be generated, so we can't make them fatal, as
78 * that would cause the first one to die.
79 *
80 * Instead we pretend -W was passed to perl, then die afterwards. The
81 * flexibility is here to return to the caller so they can finish up and
82 * die themselves */
83 U32 errors;
84
85 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
86
87 ENTER;
c15a80f3 88 SAVEI8(PL_dowarn);
9cbfb8ab
KW
89 SAVESPTR(PL_curcop);
90
91 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
92 if (PL_curcop) {
dd051059
DM
93 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
94 * than PL_compiling */
95 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
96 (void*)PL_curcop->cop_warnings);
9cbfb8ab
KW
97 PL_curcop->cop_warnings = pWARN_ALL;
98 }
99
100 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
101
102 LEAVE;
103
104 if (! errors) {
105 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
106 " be called only when there are errors found");
107 }
108
109 if (die_here) {
110 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
111 }
112}
113
bb07812e
KW
114STATIC HV *
115S_new_msg_hv(pTHX_ const char * const message, /* The message text */
116 U32 categories, /* Packed warning categories */
117 U32 flag) /* Flag associated with this message */
118{
119 /* Creates, populates, and returns an HV* that describes an error message
120 * for the translators between UTF8 and code point */
121
122 SV* msg_sv = newSVpv(message, 0);
123 SV* category_sv = newSVuv(categories);
124 SV* flag_bit_sv = newSVuv(flag);
125
126 HV* msg_hv = newHV();
127
128 PERL_ARGS_ASSERT_NEW_MSG_HV;
129
2b672cf5
KW
130 (void) hv_stores(msg_hv, "text", msg_sv);
131 (void) hv_stores(msg_hv, "warn_categories", category_sv);
132 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
bb07812e
KW
133
134 return msg_hv;
135}
136
eaf7a4d2 137/*
378516de 138=for apidoc uvoffuni_to_utf8_flags
eebe1485 139
a27992cc 140THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
141Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
142L<perlapi/uvchr_to_utf8_flags>>.
a27992cc 143
de69f3af
KW
144This function is like them, but the input is a strict Unicode
145(as opposed to native) code point. Only in very rare circumstances should code
146not be using the native code point.
949cf498 147
09232555 148For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
949cf498 149
eebe1485
SC
150=cut
151*/
152
33f38593
KW
153U8 *
154Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
155{
156 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
157
158 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
159}
160
c94c2f39
KW
161/* All these formats take a single UV code point argument */
162const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
163const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
164 " is not recommended for open interchange";
165const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
166 " may not be portable";
c94c2f39 167
33f38593 168#define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
8ee1cdcb
KW
169 STMT_START { \
170 if (flags & UNICODE_WARN_SURROGATE) { \
33f38593
KW
171 U32 category = packWARN(WARN_SURROGATE); \
172 const char * format = surrogate_cp_format; \
173 if (msgs) { \
174 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
175 category, \
176 UNICODE_GOT_SURROGATE); \
177 } \
178 else { \
179 Perl_ck_warner_d(aTHX_ category, format, uv); \
180 } \
8ee1cdcb
KW
181 } \
182 if (flags & UNICODE_DISALLOW_SURROGATE) { \
183 return NULL; \
184 } \
185 } STMT_END;
186
33f38593 187#define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
8ee1cdcb
KW
188 STMT_START { \
189 if (flags & UNICODE_WARN_NONCHAR) { \
33f38593
KW
190 U32 category = packWARN(WARN_NONCHAR); \
191 const char * format = nonchar_cp_format; \
192 if (msgs) { \
193 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
194 category, \
195 UNICODE_GOT_NONCHAR); \
196 } \
197 else { \
198 Perl_ck_warner_d(aTHX_ category, format, uv); \
199 } \
8ee1cdcb
KW
200 } \
201 if (flags & UNICODE_DISALLOW_NONCHAR) { \
202 return NULL; \
203 } \
204 } STMT_END;
205
ba6ed43c
KW
206/* Use shorter names internally in this file */
207#define SHIFT UTF_ACCUMULATION_SHIFT
208#undef MARK
209#define MARK UTF_CONTINUATION_MARK
210#define MASK UTF_CONTINUATION_MASK
211
33f38593
KW
212/*
213=for apidoc uvchr_to_utf8_flags_msgs
214
215THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
216
217Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
218
219This function is for code that wants any warning and/or error messages to be
220returned to the caller rather than be displayed. All messages that would have
884a31ee 221been displayed if all lexical warnings are enabled will be returned.
33f38593
KW
222
223It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
224placed after all the others, C<msgs>. If this parameter is 0, this function
225behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
226be a pointer to an C<HV *> variable, in which this function creates a new HV to
227contain any appropriate messages. The hash has three key-value pairs, as
228follows:
229
230=over 4
231
232=item C<text>
233
234The text of the message as a C<SVpv>.
235
236=item C<warn_categories>
237
238The warning category (or categories) packed into a C<SVuv>.
239
240=item C<flag>
241
242A single flag bit associated with this message, in a C<SVuv>.
243The bit corresponds to some bit in the C<*errors> return value,
244such as C<UNICODE_GOT_SURROGATE>.
245
246=back
247
248It's important to note that specifying this parameter as non-null will cause
249any warnings this function would otherwise generate to be suppressed, and
250instead be placed in C<*msgs>. The caller can check the lexical warnings state
251(or not) when choosing what to do with the returned messages.
252
253The caller, of course, is responsible for freeing any returned HV.
254
255=cut
256*/
257
258/* Undocumented; we don't want people using this. Instead they should use
259 * uvchr_to_utf8_flags_msgs() */
dfe13c55 260U8 *
33f38593 261Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
a0ed51b3 262{
33f38593
KW
263 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
264
265 if (msgs) {
266 *msgs = NULL;
267 }
7918f24d 268
2d1545e5 269 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 270 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
271 return d;
272 }
facc1dc2 273
3ea68d71 274 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
275 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
276 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
277 return d;
278 }
d9432125 279
ba6ed43c
KW
280 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
281 * below, the 16 is for start bytes E0-EF (which are all the possible ones
282 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
283 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
284 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
285 * 0x800-0xFFFF on ASCII */
286 if (uv < (16 * (1U << (2 * SHIFT)))) {
287 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
288 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
289 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
290
291#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
292 aren't tested here */
293 /* The most likely code points in this range are below the surrogates.
294 * Do an extra test to quickly exclude those. */
295 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
296 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
297 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
298 {
33f38593 299 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
8ee1cdcb
KW
300 }
301 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 302 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
760c7c2f 303 }
ba6ed43c
KW
304 }
305#endif
306 return d;
307 }
308
309 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
310 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
311 * happen starting with 4-byte characters on ASCII platforms. We unify the
312 * code for these with EBCDIC, even though some of them require 5-bytes on
313 * those, because khw believes the code saving is worth the very slight
314 * performance hit on these high EBCDIC code points. */
315
316 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
24b4c303
KW
317 if (UNLIKELY( uv > MAX_LEGAL_CP
318 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
319 {
fb2f0a6a 320 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv));
a5bf80e0 321 }
33f38593
KW
322 if ( (flags & UNICODE_WARN_SUPER)
323 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
0a8a1a5b 324 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0 325 {
33f38593
KW
326 const char * format = super_cp_format;
327 U32 category = packWARN(WARN_NON_UNICODE);
328 U32 flag = UNICODE_GOT_SUPER;
329
330 /* Choose the more dire applicable warning */
331 if (UNICODE_IS_PERL_EXTENDED(uv)) {
8911f9b0 332 format = PL_extended_cp_format;
dc4a6683 333 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
33f38593
KW
334 if (flags & (UNICODE_WARN_PERL_EXTENDED
335 |UNICODE_DISALLOW_PERL_EXTENDED))
336 {
337 flag = UNICODE_GOT_PERL_EXTENDED;
338 }
339 }
a5bf80e0 340
33f38593
KW
341 if (msgs) {
342 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
343 category, flag);
344 }
dc4a6683
KW
345 else if ( ckWARN_d(WARN_NON_UNICODE)
346 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
347 && ckWARN(WARN_PORTABLE)))
348 {
349 Perl_warner(aTHX_ category, format, uv);
33f38593 350 }
a5bf80e0 351 }
56576a04 352 if ( (flags & UNICODE_DISALLOW_SUPER)
0a8a1a5b
KW
353 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
354 && UNICODE_IS_PERL_EXTENDED(uv)))
a5bf80e0
KW
355 {
356 return NULL;
357 }
358 }
ba6ed43c 359 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
33f38593 360 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
507b9800 361 }
d9432125 362
ba6ed43c
KW
363 /* Test for and handle 4-byte result. In the test immediately below, the
364 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
365 * characters). The 3 is for 3 continuation bytes; these each contribute
366 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
367 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
368 * 0x1_0000-0x1F_FFFF on ASCII */
369 if (uv < (8 * (1U << (3 * SHIFT)))) {
370 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
371 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
372 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
373 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
374
375#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
376 characters. The end-plane non-characters for EBCDIC were
377 handled just above */
378 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
33f38593 379 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
d528804a 380 }
ba6ed43c 381 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
33f38593 382 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
ba6ed43c
KW
383 }
384#endif
385
386 return d;
387 }
388
389 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
390 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
391 * format. The unrolled version above turns out to not save all that much
392 * time, and at these high code points (well above the legal Unicode range
393 * on ASCII platforms, and well above anything in common use in EBCDIC),
394 * khw believes that less code outweighs slight performance gains. */
395
d9432125 396 {
5aaebcb3 397 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
398 U8 *p = d+len-1;
399 while (p > d) {
957a9e81
KW
400 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
401 uv >>= SHIFT;
1d72bdf6 402 }
4c8cd605 403 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
404 return d+len;
405 }
a0ed51b3 406}
a5bf80e0 407
646ca15d 408/*
07693fe6
KW
409=for apidoc uvchr_to_utf8
410
bcb1a2d4 411Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 412of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
413C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
414the byte after the end of the new character. In other words,
07693fe6
KW
415
416 d = uvchr_to_utf8(d, uv);
417
418is the recommended wide native character-aware way of saying
419
420 *(d++) = uv;
421
d22ec717
KW
422This function accepts any code point from 0..C<IV_MAX> as input.
423C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
424
425It is possible to forbid or warn on non-Unicode code points, or those that may
426be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 427
07693fe6
KW
428=cut
429*/
430
de69f3af
KW
431/* This is also a macro */
432PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
433
07693fe6
KW
434U8 *
435Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
436{
de69f3af 437 return uvchr_to_utf8(d, uv);
07693fe6
KW
438}
439
de69f3af
KW
440/*
441=for apidoc uvchr_to_utf8_flags
442
443Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 444of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
445C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
446the byte after the end of the new character. In other words,
de69f3af
KW
447
448 d = uvchr_to_utf8_flags(d, uv, flags);
449
450or, in most cases,
451
452 d = uvchr_to_utf8_flags(d, uv, 0);
453
454This is the Unicode-aware way of saying
455
456 *(d++) = uv;
457
d22ec717
KW
458If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
459input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
760c7c2f
KW
460
461Specifying C<flags> can further restrict what is allowed and not warned on, as
462follows:
de69f3af 463
796b6530 464If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
465the function will raise a warning, provided UTF8 warnings are enabled. If
466instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
467NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 468
760c7c2f
KW
469Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
470affect how the function handles a Unicode non-character.
93e6dbd6 471
760c7c2f
KW
472And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
473affect the handling of code points that are above the Unicode maximum of
4740x10FFFF. Languages other than Perl may not be able to accept files that
475contain these.
93e6dbd6
KW
476
477The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
478the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
479three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
480allowed inputs to the strict UTF-8 traditionally defined by Unicode.
481Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
482C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
483above-Unicode and surrogate flags, but not the non-character ones, as
484defined in
e2176993 485L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f 486See L<perlunicode/Noncharacter code points>.
93e6dbd6 487
57ff5f59
KW
488Extremely high code points were never specified in any standard, and require an
489extension to UTF-8 to express, which Perl does. It is likely that programs
490written in something other than Perl would not be able to read files that
491contain these; nor would Perl understand files written by something that uses a
492different extension. For these reasons, there is a separate set of flags that
493can warn and/or disallow these extremely high code points, even if other
494above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
495and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
eb992c6f 496C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
57ff5f59
KW
497treat all above-Unicode code points, including these, as malformations. (Note
498that the Unicode standard considers anything above 0x10FFFF to be illegal, but
499there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
500
501A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
502retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
503C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
7c4a22ed
KW
504C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
505platforms,these flags can apply to code points that actually do fit in 31 bits.
506The new names accurately describe the situation in all cases.
de69f3af 507
de69f3af
KW
508=cut
509*/
510
511/* This is also a macro */
512PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
513
07693fe6
KW
514U8 *
515Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
516{
de69f3af 517 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
518}
519
57ff5f59
KW
520#ifndef UV_IS_QUAD
521
e050c007
KW
522STATIC int
523S_is_utf8_cp_above_31_bits(const U8 * const s,
524 const U8 * const e,
525 const bool consider_overlongs)
83dc0f42
KW
526{
527 /* Returns TRUE if the first code point represented by the Perl-extended-
528 * UTF-8-encoded string starting at 's', and looking no further than 'e -
529 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
530 *
531 * The function handles the case where the input bytes do not include all
532 * the ones necessary to represent a full character. That is, they may be
533 * the intial bytes of the representation of a code point, but possibly
534 * the final ones necessary for the complete representation may be beyond
535 * 'e - 1'.
536 *
e050c007
KW
537 * The function also can handle the case where the input is an overlong
538 * sequence. If 'consider_overlongs' is 0, the function assumes the
539 * input is not overlong, without checking, and will return based on that
540 * assumption. If this parameter is 1, the function will go to the trouble
541 * of figuring out if it actually evaluates to above or below 31 bits.
83dc0f42 542 *
e050c007 543 * The sequence is otherwise assumed to be well-formed, without checking.
83dc0f42
KW
544 */
545
e050c007
KW
546 const STRLEN len = e - s;
547 int is_overlong;
548
549 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
550
551 assert(! UTF8_IS_INVARIANT(*s) && e > s);
552
83dc0f42
KW
553#ifdef EBCDIC
554
e050c007 555 PERL_UNUSED_ARG(consider_overlongs);
83dc0f42 556
e050c007
KW
557 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
558 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
559 * also be the start byte for a 31-bit code point; we need at least 2
560 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
561 * the start byte for an overlong sequence, but for 30-bit or smaller code
562 * points, so we don't have to worry about overlongs on EBCDIC.) */
563 if (*s != 0xFE) {
564 return 0;
565 }
83dc0f42 566
e050c007
KW
567 if (len == 1) {
568 return -1;
569 }
83dc0f42 570
e050c007 571#else
83dc0f42 572
e050c007
KW
573 /* On ASCII, FE and FF are the only start bytes that can evaluate to
574 * needing more than 31 bits. */
575 if (LIKELY(*s < 0xFE)) {
576 return 0;
577 }
83dc0f42 578
e050c007
KW
579 /* What we have left are FE and FF. Both of these require more than 31
580 * bits unless they are for overlongs. */
581 if (! consider_overlongs) {
582 return 1;
583 }
83dc0f42 584
e050c007
KW
585 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
586 * above 31 bits. But we need more than one byte to discern this, so if
587 * passed just the start byte, it could be an overlong evaluating to
588 * smaller */
589 if (len == 1) {
590 return -1;
591 }
83dc0f42 592
e050c007
KW
593 /* Having excluded len==1, and knowing that FE and FF are both valid start
594 * bytes, we can call the function below to see if the sequence is
595 * overlong. (We don't need the full generality of the called function,
596 * but for these huge code points, speed shouldn't be a consideration, and
597 * the compiler does have enough information, since it's static to this
598 * file, to optimize to just the needed parts.) */
599 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
83dc0f42 600
e050c007
KW
601 /* If it isn't overlong, more than 31 bits are required. */
602 if (is_overlong == 0) {
603 return 1;
604 }
83dc0f42 605
e050c007
KW
606 /* If it is indeterminate if it is overlong, return that */
607 if (is_overlong < 0) {
608 return -1;
609 }
610
611 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
612 * the max it can be is 2**31 - 1 */
613 if (*s == 0xFE) {
614 return 0;
83dc0f42
KW
615 }
616
e050c007
KW
617#endif
618
619 /* Here, ASCII and EBCDIC rejoin:
620 * On ASCII: We have an overlong sequence starting with FF
621 * On EBCDIC: We have a sequence starting with FE. */
622
623 { /* For C89, use a block so the declaration can be close to its use */
624
625#ifdef EBCDIC
626
5f995336
KW
627 /* U+7FFFFFFF (2 ** 31 - 1)
628 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
629 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
630 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
631 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
632 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
633 * U+80000000 (2 ** 31):
634 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
635 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
636 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
637 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
e050c007
KW
638 *
639 * and since we know that *s = \xfe, any continuation sequcence
640 * following it that is gt the below is above 31 bits
641 [0] [1] [2] [3] [4] [5] [6] */
642 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
643
644#else
645
646 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
647 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
648 * FF overlong for U+80000000 (2 ** 31):
649 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
650 * and since we know that *s = \xff, any continuation sequcence
651 * following it that is gt the below is above 30 bits
652 [0] [1] [2] [3] [4] [5] [6] */
653 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
5f995336 654
83dc0f42
KW
655
656#endif
e050c007
KW
657 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
658 const STRLEN cmp_len = MIN(conts_len, len - 1);
659
660 /* Now compare the continuation bytes in s with the ones we have
661 * compiled in that are for the largest 30 bit code point. If we have
662 * enough bytes available to determine the answer, or the bytes we do
663 * have differ from them, we can compare the two to get a definitive
664 * answer (Note that in UTF-EBCDIC, the two lowest possible
665 * continuation bytes are \x41 and \x42.) */
666 if (cmp_len >= conts_len || memNE(s + 1,
667 conts_for_highest_30_bit,
668 cmp_len))
669 {
670 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
671 }
83dc0f42 672
e050c007
KW
673 /* Here, all the bytes we have are the same as the highest 30-bit code
674 * point, but we are missing so many bytes that we can't make the
675 * determination */
676 return -1;
677 }
83dc0f42
KW
678}
679
57ff5f59
KW
680#endif
681
d6be65ae 682PERL_STATIC_INLINE int
12a4bed3
KW
683S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
684{
d6be65ae
KW
685 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
686 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
687 * it isn't, and -1 if there isn't enough information to tell. This last
688 * return value can happen if the sequence is incomplete, missing some
689 * trailing bytes that would form a complete character. If there are
690 * enough bytes to make a definitive decision, this function does so.
691 * Usually 2 bytes sufficient.
692 *
693 * Overlongs can occur whenever the number of continuation bytes changes.
694 * That means whenever the number of leading 1 bits in a start byte
695 * increases from the next lower start byte. That happens for start bytes
696 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
697 * start bytes have already been excluded, so don't need to be tested here;
12a4bed3
KW
698 * ASCII platforms: C0, C1
699 * EBCDIC platforms C0, C1, C2, C3, C4, E0
d6be65ae 700 */
12a4bed3
KW
701
702 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
703 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
704
705 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
706 assert(len > 1 && UTF8_IS_START(*s));
707
708 /* Each platform has overlongs after the start bytes given above (expressed
709 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
710 * the logic is the same, except the E0 overlong has already been excluded
711 * on EBCDIC platforms. The values below were found by manually
712 * inspecting the UTF-8 patterns. See the tables in utf8.h and
713 * utfebcdic.h. */
714
715# ifdef EBCDIC
716# define F0_ABOVE_OVERLONG 0xB0
717# define F8_ABOVE_OVERLONG 0xA8
718# define FC_ABOVE_OVERLONG 0xA4
719# define FE_ABOVE_OVERLONG 0xA2
720# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
721 /* I8(0xfe) is FF */
722# else
723
724 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
d6be65ae 725 return 1;
12a4bed3
KW
726 }
727
728# define F0_ABOVE_OVERLONG 0x90
729# define F8_ABOVE_OVERLONG 0x88
730# define FC_ABOVE_OVERLONG 0x84
731# define FE_ABOVE_OVERLONG 0x82
732# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
733# endif
734
735
736 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
737 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
738 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
739 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
740 {
d6be65ae 741 return 1;
12a4bed3
KW
742 }
743
b0b342d4 744 /* Check for the FF overlong */
d6be65ae 745 return isFF_OVERLONG(s, len);
b0b342d4
KW
746}
747
8d6204cc 748PERL_STATIC_INLINE int
b0b342d4
KW
749S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
750{
8d6204cc
KW
751 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
752 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
753 * it isn't, and -1 if there isn't enough information to tell. This last
754 * return value can happen if the sequence is incomplete, missing some
755 * trailing bytes that would form a complete character. If there are
756 * enough bytes to make a definitive decision, this function does so. */
757
b0b342d4 758 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3 759
8d6204cc
KW
760 /* To be an FF overlong, all the available bytes must match */
761 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
762 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
763 {
764 return 0;
765 }
766
767 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
768 * be there; what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 769 * utfebcdic.h. */
8d6204cc
KW
770 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
771 return 1;
772 }
12a4bed3 773
8d6204cc
KW
774 /* The missing bytes could cause the result to go one way or the other, so
775 * the result is indeterminate */
776 return -1;
12a4bed3
KW
777}
778
d22ec717 779#if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
a77c906e
KW
780# ifdef EBCDIC /* Actually is I8 */
781# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 782 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
783# else
784# define HIGHEST_REPRESENTABLE_UTF8 \
d22ec717 785 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
a77c906e
KW
786# endif
787#endif
788
c285bbc4 789PERL_STATIC_INLINE int
e050c007
KW
790S_does_utf8_overflow(const U8 * const s,
791 const U8 * e,
792 const bool consider_overlongs)
a77c906e 793{
c285bbc4 794 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
d22ec717
KW
795 * 'e' - 1 would overflow an IV on this platform; that is if it represents
796 * a code point larger than the highest representable code point. It
797 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
798 * enough information to tell. This last return value can happen if the
799 * sequence is incomplete, missing some trailing bytes that would form a
800 * complete character. If there are enough bytes to make a definitive
801 * decision, this function does so.
c285bbc4 802 *
e050c007
KW
803 * If 'consider_overlongs' is TRUE, the function checks for the possibility
804 * that the sequence is an overlong that doesn't overflow. Otherwise, it
805 * assumes the sequence is not an overlong. This can give different
806 * results only on ASCII 32-bit platforms.
807 *
c285bbc4
KW
808 * (For ASCII platforms, we could use memcmp() because we don't have to
809 * convert each byte to I8, but it's very rare input indeed that would
810 * approach overflow, so the loop below will likely only get executed once.)
811 *
812 * 'e' - 1 must not be beyond a full character. */
a77c906e 813
a77c906e
KW
814
815 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
816 assert(s <= e && s + UTF8SKIP(s) >= e);
817
d22ec717
KW
818#if ! defined(UV_IS_QUAD)
819
820 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
821
822#else
823
824 PERL_UNUSED_ARG(consider_overlongs);
825
826 {
827 const STRLEN len = e - s;
828 const U8 *x;
829 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
830
831 for (x = s; x < e; x++, y++) {
832
833 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
834 continue;
835 }
836
837 /* If this byte is larger than the corresponding highest UTF-8
838 * byte, the sequence overflow; otherwise the byte is less than,
839 * and so the sequence doesn't overflow */
840 return NATIVE_UTF8_TO_I8(*x) > *y;
841
842 }
843
844 /* Got to the end and all bytes are the same. If the input is a whole
845 * character, it doesn't overflow. And if it is a partial character,
846 * there's not enough information to tell */
847 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
848 return -1;
849 }
850
851 return 0;
852 }
853
854#endif
855
856}
857
858#if 0
859
860/* This is the portions of the above function that deal with UV_MAX instead of
861 * IV_MAX. They are left here in case we want to combine them so that internal
862 * uses can have larger code points. The only logic difference is that the
863 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
864 * different logic.
865 */
866
867/* Anything larger than this will overflow the word if it were converted into a UV */
868#if defined(UV_IS_QUAD)
869# ifdef EBCDIC /* Actually is I8 */
870# define HIGHEST_REPRESENTABLE_UTF8 \
871 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
872# else
873# define HIGHEST_REPRESENTABLE_UTF8 \
874 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
875# endif
876#else /* 32-bit */
877# ifdef EBCDIC
878# define HIGHEST_REPRESENTABLE_UTF8 \
879 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
880# else
881# define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
882# endif
883#endif
884
a77c906e
KW
885#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
886
887 /* On 32 bit ASCII machines, many overlongs that start with FF don't
888 * overflow */
e050c007 889 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
c285bbc4
KW
890
891 /* To be such an overlong, the first bytes of 's' must match
892 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
893 * don't have any additional bytes available, the sequence, when
894 * completed might or might not fit in 32 bits. But if we have that
895 * next byte, we can tell for sure. If it is <= 0x83, then it does
896 * fit. */
897 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
898 return -1;
899 }
900
901 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
a77c906e
KW
902 }
903
d22ec717
KW
904/* Starting with the #else, the rest of the function is identical except
905 * 1. we need to move the 'len' declaration to be global to the function
906 * 2. the endif move to just after the UNUSED_ARG.
907 * An empty endif is given just below to satisfy the preprocessor
908 */
a77c906e
KW
909#endif
910
d22ec717 911#endif
a77c906e 912
12a4bed3
KW
913#undef F0_ABOVE_OVERLONG
914#undef F8_ABOVE_OVERLONG
915#undef FC_ABOVE_OVERLONG
916#undef FE_ABOVE_OVERLONG
917#undef FF_OVERLONG_PREFIX
918
35f8c9bd 919STRLEN
1376b35c 920Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 921{
2b479609 922 STRLEN len;
12a4bed3 923 const U8 *x;
35f8c9bd 924
2b479609
KW
925 /* A helper function that should not be called directly.
926 *
927 * This function returns non-zero if the string beginning at 's' and
928 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
929 * code point; otherwise it returns 0. The examination stops after the
930 * first code point in 's' is validated, not looking at the rest of the
931 * input. If 'e' is such that there are not enough bytes to represent a
932 * complete code point, this function will return non-zero anyway, if the
933 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
934 * excluded by 'flags'.
935 *
936 * A non-zero return gives the number of bytes required to represent the
937 * code point. Be aware that if the input is for a partial character, the
938 * return will be larger than 'e - s'.
939 *
940 * This function assumes that the code point represented is UTF-8 variant.
56576a04
KW
941 * The caller should have excluded the possibility of it being invariant
942 * before calling this function.
2b479609
KW
943 *
944 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
945 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
946 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
947 * disallowed by the flags. If the input is only for a partial character,
948 * the function will return non-zero if there is any sequence of
949 * well-formed UTF-8 that, when appended to the input sequence, could
950 * result in an allowed code point; otherwise it returns 0. Non characters
951 * cannot be determined based on partial character input. But many of the
952 * other excluded types can be determined with just the first one or two
953 * bytes.
954 *
955 */
956
1376b35c 957 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
2b479609
KW
958
959 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
d044b7a7 960 |UTF8_DISALLOW_PERL_EXTENDED)));
2b479609 961 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 962
2b479609 963 /* A variant char must begin with a start byte */
35f8c9bd
KW
964 if (UNLIKELY(! UTF8_IS_START(*s))) {
965 return 0;
966 }
967
edc2c47a
KW
968 /* Examine a maximum of a single whole code point */
969 if (e - s > UTF8SKIP(s)) {
970 e = s + UTF8SKIP(s);
971 }
972
2b479609
KW
973 len = e - s;
974
975 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
976 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 977
56576a04
KW
978 /* Here, we are disallowing some set of largish code points, and the
979 * first byte indicates the sequence is for a code point that could be
980 * in the excluded set. We generally don't have to look beyond this or
981 * the second byte to see if the sequence is actually for one of the
982 * excluded classes. The code below is derived from this table:
983 *
2b479609
KW
984 * UTF-8 UTF-EBCDIC I8
985 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
986 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
987 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
988 *
56576a04
KW
989 * Keep in mind that legal continuation bytes range between \x80..\xBF
990 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
991 * continuation bytes. Hence, we don't have to test the upper edge
992 * because if any of those is encountered, the sequence is malformed,
993 * and would fail elsewhere in this function.
994 *
995 * The code here likewise assumes that there aren't other
996 * malformations; again the function should fail elsewhere because of
997 * these. For example, an overlong beginning with FC doesn't actually
998 * have to be a super; it could actually represent a small code point,
999 * even U+0000. But, since overlongs (and other malformations) are
1000 * illegal, the function should return FALSE in either case.
2b479609
KW
1001 */
1002
1003#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1004# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 1005# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 1006
19794540
KW
1007# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1008 /* B6 and B7 */ \
1009 && ((s1) & 0xFE ) == 0xB6)
57ff5f59 1010# define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
2b479609
KW
1011#else
1012# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
1013# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1014# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
57ff5f59 1015# define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
2b479609
KW
1016#endif
1017
1018 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
1019 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1020 {
2b479609
KW
1021 return 0; /* Above Unicode */
1022 }
1023
d044b7a7 1024 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
57ff5f59 1025 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
2b479609 1026 {
57ff5f59 1027 return 0;
2b479609
KW
1028 }
1029
1030 if (len > 1) {
1031 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1032
1033 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 1034 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
1035 {
1036 return 0; /* Above Unicode */
1037 }
1038
1039 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 1040 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
1041 {
1042 return 0; /* Surrogate */
1043 }
1044
1045 if ( (flags & UTF8_DISALLOW_NONCHAR)
1046 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1047 {
1048 return 0; /* Noncharacter code point */
1049 }
1050 }
1051 }
1052
1053 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
1054 for (x = s + 1; x < e; x++) {
1055 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1056 return 0;
1057 }
1058 }
1059
af13dd8a 1060 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3 1061 * overlong. */
d6be65ae 1062 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
12a4bed3 1063 return 0;
af13dd8a
KW
1064 }
1065
12a4bed3
KW
1066 /* And finally, that the code point represented fits in a word on this
1067 * platform */
e050c007
KW
1068 if (0 < does_utf8_overflow(s, e,
1069 0 /* Don't consider overlongs */
1070 ))
1071 {
12a4bed3 1072 return 0;
35f8c9bd
KW
1073 }
1074
2b479609 1075 return UTF8SKIP(s);
35f8c9bd
KW
1076}
1077
7e2f38b2 1078char *
63ab03b3 1079Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
7cf8d05d
KW
1080{
1081 /* Returns a mortalized C string that is a displayable copy of the 'len'
63ab03b3 1082 * bytes starting at 'start'. 'format' gives how to display each byte.
7e2f38b2
KW
1083 * Currently, there are only two formats, so it is currently a bool:
1084 * 0 \xab
1085 * 1 ab (that is a space between two hex digit bytes)
1086 */
7cf8d05d
KW
1087
1088 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1089 trailing NUL */
63ab03b3
KW
1090 const U8 * s = start;
1091 const U8 * const e = start + len;
7cf8d05d
KW
1092 char * output;
1093 char * d;
1094
1095 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1096
1097 Newx(output, output_len, char);
1098 SAVEFREEPV(output);
1099
1100 d = output;
63ab03b3 1101 for (s = start; s < e; s++) {
7cf8d05d
KW
1102 const unsigned high_nibble = (*s & 0xF0) >> 4;
1103 const unsigned low_nibble = (*s & 0x0F);
1104
7e2f38b2 1105 if (format) {
63ab03b3
KW
1106 if (s > start) {
1107 *d++ = ' ';
1108 }
7e2f38b2
KW
1109 }
1110 else {
1111 *d++ = '\\';
1112 *d++ = 'x';
1113 }
7cf8d05d
KW
1114
1115 if (high_nibble < 10) {
1116 *d++ = high_nibble + '0';
1117 }
1118 else {
1119 *d++ = high_nibble - 10 + 'a';
1120 }
1121
1122 if (low_nibble < 10) {
1123 *d++ = low_nibble + '0';
1124 }
1125 else {
1126 *d++ = low_nibble - 10 + 'a';
1127 }
1128 }
1129
1130 *d = '\0';
1131 return output;
1132}
1133
806547a7 1134PERL_STATIC_INLINE char *
7cf8d05d
KW
1135S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1136
421da25c 1137 /* Max number of bytes to print */
3cc6a05e 1138 STRLEN print_len,
7cf8d05d
KW
1139
1140 /* Which one is the non-continuation */
1141 const STRLEN non_cont_byte_pos,
1142
1143 /* How many bytes should there be? */
1144 const STRLEN expect_len)
806547a7
KW
1145{
1146 /* Return the malformation warning text for an unexpected continuation
1147 * byte. */
1148
7cf8d05d 1149 const char * const where = (non_cont_byte_pos == 1)
806547a7 1150 ? "immediately"
7cf8d05d
KW
1151 : Perl_form(aTHX_ "%d bytes",
1152 (int) non_cont_byte_pos);
421da25c
KW
1153 const U8 * x = s + non_cont_byte_pos;
1154 const U8 * e = s + print_len;
806547a7
KW
1155
1156 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1157
7cf8d05d
KW
1158 /* We don't need to pass this parameter, but since it has already been
1159 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1160 assert(expect_len == UTF8SKIP(s));
1161
421da25c
KW
1162 /* As a defensive coding measure, don't output anything past a NUL. Such
1163 * bytes shouldn't be in the middle of a malformation, and could mark the
1164 * end of the allocated string, and what comes after is undefined */
1165 for (; x < e; x++) {
1166 if (*x == '\0') {
1167 x++; /* Output this particular NUL */
1168 break;
1169 }
1170 }
1171
7cf8d05d
KW
1172 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1173 " %s after start byte 0x%02x; need %d bytes, got %d)",
1174 malformed_text,
421da25c 1175 _byte_dump_string(s, x - s, 0),
7cf8d05d
KW
1176 *(s + non_cont_byte_pos),
1177 where,
1178 *s,
1179 (int) expect_len,
1180 (int) non_cont_byte_pos);
806547a7
KW
1181}
1182
35f8c9bd
KW
1183/*
1184
de69f3af 1185=for apidoc utf8n_to_uvchr
378516de
KW
1186
1187THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1188Most code should use L</utf8_to_uvchr_buf>() rather than call this
1189directly.
67e989fb 1190
9041c2e3 1191Bottom level UTF-8 decode routine.
de69f3af 1192Returns the native code point value of the first character in the string C<s>,
746afd53
KW
1193which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1194C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1195the length, in bytes, of that character.
949cf498
KW
1196
1197The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
1198well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1199causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1200is the next possible position in C<s> that could begin a non-malformed
1201character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1202is raised. Some UTF-8 input sequences may contain multiple malformations.
1203This function tries to find every possible one in each call, so multiple
56576a04 1204warnings can be raised for the same sequence.
949cf498
KW
1205
1206Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1207individual types of malformations, such as the sequence being overlong (that
1208is, when there is a shorter sequence that can express the same code point;
1209overlong sequences are expressly forbidden in the UTF-8 standard due to
1210potential security issues). Another malformation example is the first byte of
1211a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
1212flags. Even if allowed, this function generally returns the Unicode
1213REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1214F<utf8.h> to override this behavior for the overlong malformations, but don't
1215do that except for very specialized purposes.
949cf498 1216
796b6530 1217The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
1218flags) malformation is found. If this flag is set, the routine assumes that
1219the caller will raise a warning, and this function will silently just set
d088425d
KW
1220C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1221
75200dff 1222Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 1223character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
1224in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1225be set to 1. To disambiguate, upon a zero return, see if the first byte of
1226C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 1227error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
1228
1229Certain code points are considered problematic. These are Unicode surrogates,
746afd53 1230Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 1231By default these are considered regular code points, but certain situations
ecc1615f
KW
1232warrant special handling for them, which can be specified using the C<flags>
1233parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1234three classes are treated as malformations and handled as such. The flags
1235C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1236C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1237disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1238restricts the allowed inputs to the strict UTF-8 traditionally defined by
1239Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1240definition given by
e2176993 1241L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
ecc1615f
KW
1242The difference between traditional strictness and C9 strictness is that the
1243latter does not forbid non-character code points. (They are still discouraged,
1244however.) For more discussion see L<perlunicode/Noncharacter code points>.
1245
1246The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1247C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
1248C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1249raised for their respective categories, but otherwise the code points are
1250considered valid (not malformations). To get a category to both be treated as
1251a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 1252(But note that warnings are not raised if lexically disabled nor if
796b6530 1253C<UTF8_CHECK_ONLY> is also specified.)
949cf498 1254
57ff5f59
KW
1255Extremely high code points were never specified in any standard, and require an
1256extension to UTF-8 to express, which Perl does. It is likely that programs
1257written in something other than Perl would not be able to read files that
1258contain these; nor would Perl understand files written by something that uses a
1259different extension. For these reasons, there is a separate set of flags that
1260can warn and/or disallow these extremely high code points, even if other
1261above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1262C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
eb992c6f 1263C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
57ff5f59
KW
1264above-Unicode code points, including these, as malformations.
1265(Note that the Unicode standard considers anything above 0x10FFFF to be
1266illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1267(2**31 -1))
1268
1269A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1270retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1271C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1272C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1273can apply to code points that actually do fit in 31 bits. This happens on
1274EBCDIC platforms, and sometimes when the L<overlong
1275malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1276describe the situation in all cases.
1277
ab8e6d41 1278
949cf498
KW
1279All other code points corresponding to Unicode characters, including private
1280use and those yet to be assigned, are never considered malformed and never
1281warn.
67e989fb 1282
5af38e47
KW
1283=for apidoc Amnh||UTF8_CHECK_ONLY
1284=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1285=for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1286=for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1287=for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1288=for apidoc Amnh||UTF8_DISALLOW_SUPER
1289=for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1290=for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1291=for apidoc Amnh||UTF8_WARN_SURROGATE
1292=for apidoc Amnh||UTF8_WARN_NONCHAR
1293=for apidoc Amnh||UTF8_WARN_SUPER
1294=for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1295=for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1296
37607a96 1297=cut
f9380377
KW
1298
1299Also implemented as a macro in utf8.h
1300*/
1301
1302UV
e6a4ffc3
KW
1303Perl_utf8n_to_uvchr(const U8 *s,
1304 STRLEN curlen,
1305 STRLEN *retlen,
1306 const U32 flags)
f9380377
KW
1307{
1308 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1309
1310 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1311}
1312
1313/*
1314
1315=for apidoc utf8n_to_uvchr_error
1316
1317THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1318Most code should use L</utf8_to_uvchr_buf>() rather than call this
1319directly.
f9380377
KW
1320
1321This function is for code that needs to know what the precise malformation(s)
37657a5b
KW
1322are when an error is found. If you also need to know the generated warning
1323messages, use L</utf8n_to_uvchr_msgs>() instead.
f9380377
KW
1324
1325It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1326all the others, C<errors>. If this parameter is 0, this function behaves
1327identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1328to a C<U32> variable, which this function sets to indicate any errors found.
1329Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1330C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1331of these bits will be set if a malformation is found, even if the input
7a65503b 1332C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
1333exceptions are noted:
1334
1335=over 4
1336
57ff5f59 1337=item C<UTF8_GOT_PERL_EXTENDED>
f9380377 1338
57ff5f59
KW
1339The input sequence is not standard UTF-8, but a Perl extension. This bit is
1340set only if the input C<flags> parameter contains either the
1341C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1342
1343Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1344and so some extension must be used to express them. Perl uses a natural
1345extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1346extension to represent even higher ones, so that any code point that fits in a
134764-bit word can be represented. Text using these extensions is not likely to
1348be portable to non-Perl code. We lump both of these extensions together and
1349refer to them as Perl extended UTF-8. There exist other extensions that people
1350have invented, incompatible with Perl's.
1351
1352On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1353extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1354than on ASCII. Prior to that, code points 2**31 and higher were simply
1355unrepresentable, and a different, incompatible method was used to represent
1356code points between 2**30 and 2**31 - 1.
1357
1358On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1359Perl extended UTF-8 is used.
1360
1361In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1362may use for backward compatibility. That name is misleading, as this flag may
1363be set when the code point actually does fit in 31 bits. This happens on
1364EBCDIC platforms, and sometimes when the L<overlong
1365malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1366describes the situation in all cases.
f9380377
KW
1367
1368=item C<UTF8_GOT_CONTINUATION>
1369
a3815e44 1370The input sequence was malformed in that the first byte was a UTF-8
f9380377
KW
1371continuation byte.
1372
1373=item C<UTF8_GOT_EMPTY>
1374
1375The input C<curlen> parameter was 0.
1376
1377=item C<UTF8_GOT_LONG>
1378
1379The input sequence was malformed in that there is some other sequence that
1380evaluates to the same code point, but that sequence is shorter than this one.
1381
fecaf136
KW
1382Until Unicode 3.1, it was legal for programs to accept this malformation, but
1383it was discovered that this created security issues.
1384
f9380377
KW
1385=item C<UTF8_GOT_NONCHAR>
1386
1387The code point represented by the input UTF-8 sequence is for a Unicode
1388non-character code point.
1389This bit is set only if the input C<flags> parameter contains either the
1390C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1391
1392=item C<UTF8_GOT_NON_CONTINUATION>
1393
1394The input sequence was malformed in that a non-continuation type byte was found
00d976bb 1395in a position where only a continuation type one should be. See also
eb992c6f 1396C<L</UTF8_GOT_SHORT>>.
f9380377
KW
1397
1398=item C<UTF8_GOT_OVERFLOW>
1399
1400The input sequence was malformed in that it is for a code point that is not
d22ec717 1401representable in the number of bits available in an IV on the current platform.
f9380377
KW
1402
1403=item C<UTF8_GOT_SHORT>
1404
1405The input sequence was malformed in that C<curlen> is smaller than required for
1406a complete sequence. In other words, the input is for a partial character
1407sequence.
1408
00d976bb
KW
1409
1410C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1411sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1412that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1413sequence was looked at. If no other flags are present, it means that the
1414sequence was valid as far as it went. Depending on the application, this could
1415mean one of three things:
1416
1417=over
1418
1419=item *
1420
1421The C<curlen> length parameter passed in was too small, and the function was
1422prevented from examining all the necessary bytes.
1423
1424=item *
1425
1426The buffer being looked at is based on reading data, and the data received so
1427far stopped in the middle of a character, so that the next read will
1428read the remainder of this character. (It is up to the caller to deal with the
1429split bytes somehow.)
1430
1431=item *
1432
1433This is a real error, and the partial sequence is all we're going to get.
1434
1435=back
1436
f9380377
KW
1437=item C<UTF8_GOT_SUPER>
1438
1439The input sequence was malformed in that it is for a non-Unicode code point;
1440that is, one above the legal Unicode maximum.
1441This bit is set only if the input C<flags> parameter contains either the
1442C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1443
1444=item C<UTF8_GOT_SURROGATE>
1445
1446The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1447code point.
1448This bit is set only if the input C<flags> parameter contains either the
1449C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1450
1451=back
1452
133551d8
KW
1453To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1454flag to suppress any warnings, and then examine the C<*errors> return.
1455
f9380377 1456=cut
37657a5b
KW
1457
1458Also implemented as a macro in utf8.h
37607a96 1459*/
67e989fb 1460
a0ed51b3 1461UV
e6a4ffc3 1462Perl_utf8n_to_uvchr_error(const U8 *s,
37657a5b
KW
1463 STRLEN curlen,
1464 STRLEN *retlen,
1465 const U32 flags,
1466 U32 * errors)
1467{
1468 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1469
1470 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1471}
1472
1473/*
1474
1475=for apidoc utf8n_to_uvchr_msgs
1476
1477THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
09232555
KW
1478Most code should use L</utf8_to_uvchr_buf>() rather than call this
1479directly.
37657a5b
KW
1480
1481This function is for code that needs to know what the precise malformation(s)
1482are when an error is found, and wants the corresponding warning and/or error
1483messages to be returned to the caller rather than be displayed. All messages
1484that would have been displayed if all lexcial warnings are enabled will be
1485returned.
1486
1487It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1488placed after all the others, C<msgs>. If this parameter is 0, this function
1489behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1490be a pointer to an C<AV *> variable, in which this function creates a new AV to
1491contain any appropriate messages. The elements of the array are ordered so
1492that the first message that would have been displayed is in the 0th element,
1493and so on. Each element is a hash with three key-value pairs, as follows:
1494
1495=over 4
1496
1497=item C<text>
1498
1499The text of the message as a C<SVpv>.
1500
1501=item C<warn_categories>
1502
1503The warning category (or categories) packed into a C<SVuv>.
1504
1505=item C<flag>
1506
1507A single flag bit associated with this message, in a C<SVuv>.
1508The bit corresponds to some bit in the C<*errors> return value,
1509such as C<UTF8_GOT_LONG>.
1510
1511=back
1512
1513It's important to note that specifying this parameter as non-null will cause
1514any warnings this function would otherwise generate to be suppressed, and
1515instead be placed in C<*msgs>. The caller can check the lexical warnings state
1516(or not) when choosing what to do with the returned messages.
1517
1518If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1519no AV is created.
1520
1521The caller, of course, is responsible for freeing any returned AV.
1522
1523=cut
1524*/
1525
1526UV
e6a4ffc3 1527Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
37657a5b
KW
1528 STRLEN curlen,
1529 STRLEN *retlen,
1530 const U32 flags,
1531 U32 * errors,
1532 AV ** msgs)
a0ed51b3 1533{
d4c19fe8 1534 const U8 * const s0 = s;
2b9519f0 1535 const U8 * send = s0 + curlen;
5af9f822
KW
1536 U32 possible_problems; /* A bit is set here for each potential problem
1537 found as we go along */
1538 UV uv;
1539 STRLEN expectlen; /* How long should this sequence be? */
1540 STRLEN avail_len; /* When input is too short, gives what that is */
1541 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1542 gets set and discarded */
a0dbb045 1543
2b5e7bc2
KW
1544 /* The below are used only if there is both an overlong malformation and a
1545 * too short one. Otherwise the first two are set to 's0' and 'send', and
1546 * the third not used at all */
5af9f822 1547 U8 * adjusted_s0;
e9f2c446
KW
1548 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1549 routine; see [perl #130921] */
5af9f822 1550 UV uv_so_far;
e6a4ffc3 1551 dTHX;
5af9f822 1552
e6a4ffc3 1553 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
5af9f822
KW
1554
1555 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1556 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1557 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1558 * final case. */
1559
1560#ifndef EBCDIC
1561
1562 /* Each of the affected Hanguls starts with \xED */
1563
1564 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1565 if (retlen) {
1566 *retlen = 3;
1567 }
1568 if (errors) {
1569 *errors = 0;
1570 }
1571 if (msgs) {
1572 *msgs = NULL;
1573 }
1574
1575 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1576 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1577 | (s0[2] & UTF_CONTINUATION_MASK);
1578 }
1579
1580#endif
1581
1582 /* In conjunction with the exhaustive tests that can be enabled in
1583 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1584 * what it is intended to do, and that no flaws in it are masked by
1585 * dropping down and executing the code below
1586 assert(! isUTF8_CHAR(s0, send)
1587 || UTF8_IS_SURROGATE(s0, send)
1588 || UTF8_IS_SUPER(s0, send)
1589 || UTF8_IS_NONCHAR(s0,send));
1590 */
1591
1592 s = s0;
1593 uv = *s0;
1594 possible_problems = 0;
1595 expectlen = 0;
1596 avail_len = 0;
1597 discard_errors = 0;
1598 adjusted_s0 = (U8 *) s0;
1599 uv_so_far = 0;
1600
f9380377
KW
1601 if (errors) {
1602 *errors = 0;
1603 }
1604 else {
1605 errors = &discard_errors;
1606 }
a0dbb045 1607
eb83ed87
KW
1608 /* The order of malformation tests here is important. We should consume as
1609 * few bytes as possible in order to not skip any valid character. This is
1610 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
e2176993 1611 * https://unicode.org/reports/tr36 for more discussion as to why. For
eb83ed87
KW
1612 * example, once we've done a UTF8SKIP, we can tell the expected number of
1613 * bytes, and could fail right off the bat if the input parameters indicate
1614 * that there are too few available. But it could be that just that first
1615 * byte is garbled, and the intended character occupies fewer bytes. If we
1616 * blindly assumed that the first byte is correct, and skipped based on
1617 * that number, we could skip over a valid input character. So instead, we
1618 * always examine the sequence byte-by-byte.
1619 *
1620 * We also should not consume too few bytes, otherwise someone could inject
1621 * things. For example, an input could be deliberately designed to
1622 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1623 * returning to the caller C<*retlen> pointing to the very next byte (one
a3815e44 1624 * which is actually part of the overflowing sequence), that could look
eb83ed87 1625 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1626 * sequence and process the rest, inappropriately.
1627 *
1628 * Some possible input sequences are malformed in more than one way. This
1629 * function goes to lengths to try to find all of them. This is necessary
1630 * for correctness, as the inputs may allow one malformation but not
1631 * another, and if we abandon searching for others after finding the
1632 * allowed one, we could allow in something that shouldn't have been.
1633 */
eb83ed87 1634
b5b9af04 1635 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1636 possible_problems |= UTF8_GOT_EMPTY;
1637 curlen = 0;
5a48568d 1638 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1639 goto ready_to_handle_errors;
0c443dc2
JH
1640 }
1641
eb83ed87
KW
1642 expectlen = UTF8SKIP(s);
1643
1644 /* A well-formed UTF-8 character, as the vast majority of calls to this
1645 * function will be for, has this expected length. For efficiency, set
1646 * things up here to return it. It will be overriden only in those rare
1647 * cases where a malformation is found */
1648 if (retlen) {
1649 *retlen = expectlen;
1650 }
1651
eb83ed87 1652 /* A continuation character can't start a valid sequence */
b5b9af04 1653 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1654 possible_problems |= UTF8_GOT_CONTINUATION;
1655 curlen = 1;
1656 uv = UNICODE_REPLACEMENT;
1657 goto ready_to_handle_errors;
ba210ebe 1658 }
9041c2e3 1659
dcd27b3c 1660 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1661 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1662 * because it excludes start bytes like \xC0 that always lead to
1663 * overlongs.) */
dcd27b3c 1664
534752c1
KW
1665 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1666 * that indicate the number of bytes in the character's whole UTF-8
1667 * sequence, leaving just the bits that are part of the value. */
1668 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1669
e308b348
KW
1670 /* Setup the loop end point, making sure to not look past the end of the
1671 * input string, and flag it as too short if the size isn't big enough. */
e308b348
KW
1672 if (UNLIKELY(curlen < expectlen)) {
1673 possible_problems |= UTF8_GOT_SHORT;
1674 avail_len = curlen;
e308b348
KW
1675 }
1676 else {
2b9519f0 1677 send = (U8*) s0 + expectlen;
e308b348 1678 }
e308b348 1679
eb83ed87 1680 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1681 * accumulating each into the working value as we go. */
eb83ed87 1682 for (s = s0 + 1; s < send; s++) {
b5b9af04 1683 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1684 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1685 continue;
1686 }
1687
1688 /* Here, found a non-continuation before processing all expected bytes.
1689 * This byte indicates the beginning of a new character, so quit, even
1690 * if allowing this malformation. */
2b5e7bc2 1691 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1692 break;
eb83ed87
KW
1693 } /* End of loop through the character's bytes */
1694
1695 /* Save how many bytes were actually in the character */
1696 curlen = s - s0;
1697
2b5e7bc2
KW
1698 /* Note that there are two types of too-short malformation. One is when
1699 * there is actual wrong data before the normal termination of the
1700 * sequence. The other is that the sequence wasn't complete before the end
1701 * of the data we are allowed to look at, based on the input 'curlen'.
1702 * This means that we were passed data for a partial character, but it is
1703 * valid as far as we saw. The other is definitely invalid. This
1704 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1705 * separate.
1706 *
1707 * A convenience macro that matches either of the too-short conditions. */
1708# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1709
1710 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1711 uv_so_far = uv;
1712 uv = UNICODE_REPLACEMENT;
1713 }
2b5e7bc2 1714
08e73697
KW
1715 /* Check for overflow. The algorithm requires us to not look past the end
1716 * of the current character, even if partial, so the upper limit is 's' */
e050c007
KW
1717 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1718 1 /* Do consider overlongs */
1719 )))
1720 {
2b5e7bc2
KW
1721 possible_problems |= UTF8_GOT_OVERFLOW;
1722 uv = UNICODE_REPLACEMENT;
eb83ed87 1723 }
eb83ed87 1724
2b5e7bc2
KW
1725 /* Check for overlong. If no problems so far, 'uv' is the correct code
1726 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1727 * we must look at the UTF-8 byte sequence itself to see if it is for an
1728 * overlong */
1729 if ( ( LIKELY(! possible_problems)
1730 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
56576a04 1731 || ( UNLIKELY(possible_problems)
2b5e7bc2
KW
1732 && ( UNLIKELY(! UTF8_IS_START(*s0))
1733 || ( curlen > 1
d6be65ae 1734 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
08e73697 1735 s - s0))))))
2f8f112e 1736 {
2b5e7bc2
KW
1737 possible_problems |= UTF8_GOT_LONG;
1738
abc28b54 1739 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
56576a04 1740
abc28b54
KW
1741 /* The calculation in the 'true' branch of this 'if'
1742 * below won't work if overflows, and isn't needed
1743 * anyway. Further below we handle all overflow
1744 * cases */
1745 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1746 {
2b5e7bc2
KW
1747 UV min_uv = uv_so_far;
1748 STRLEN i;
1749
1750 /* Here, the input is both overlong and is missing some trailing
1751 * bytes. There is no single code point it could be for, but there
1752 * may be enough information present to determine if what we have
1753 * so far is for an unallowed code point, such as for a surrogate.
56576a04
KW
1754 * The code further below has the intelligence to determine this,
1755 * but just for non-overlong UTF-8 sequences. What we do here is
1756 * calculate the smallest code point the input could represent if
1757 * there were no too short malformation. Then we compute and save
1758 * the UTF-8 for that, which is what the code below looks at
1759 * instead of the raw input. It turns out that the smallest such
1760 * code point is all we need. */
2b5e7bc2
KW
1761 for (i = curlen; i < expectlen; i++) {
1762 min_uv = UTF8_ACCUMULATE(min_uv,
1763 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1764 }
1765
e9f2c446 1766 adjusted_s0 = temp_char_buf;
57ff5f59 1767 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
2b5e7bc2 1768 }
eb83ed87
KW
1769 }
1770
56576a04
KW
1771 /* Here, we have found all the possible problems, except for when the input
1772 * is for a problematic code point not allowed by the input parameters. */
1773
06188866
KW
1774 /* uv is valid for overlongs */
1775 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1776
1777 /* isn't problematic if < this */
1778 && uv >= UNICODE_SURROGATE_FIRST)
2b5e7bc2 1779 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1780
1781 /* if overflow, we know without looking further
1782 * precisely which of the problematic types it is,
1783 * and we deal with those in the overflow handling
1784 * code */
1785 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
57ff5f59
KW
1786 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1787 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
760c7c2f
KW
1788 && ((flags & ( UTF8_DISALLOW_NONCHAR
1789 |UTF8_DISALLOW_SURROGATE
1790 |UTF8_DISALLOW_SUPER
d044b7a7 1791 |UTF8_DISALLOW_PERL_EXTENDED
760c7c2f
KW
1792 |UTF8_WARN_NONCHAR
1793 |UTF8_WARN_SURROGATE
1794 |UTF8_WARN_SUPER
d22ec717 1795 |UTF8_WARN_PERL_EXTENDED))))
eb83ed87 1796 {
2b5e7bc2
KW
1797 /* If there were no malformations, or the only malformation is an
1798 * overlong, 'uv' is valid */
1799 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1800 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1801 possible_problems |= UTF8_GOT_SURROGATE;
1802 }
1803 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1804 possible_problems |= UTF8_GOT_SUPER;
1805 }
1806 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1807 possible_problems |= UTF8_GOT_NONCHAR;
1808 }
1809 }
1810 else { /* Otherwise, need to look at the source UTF-8, possibly
1811 adjusted to be non-overlong */
1812
1813 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1814 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1815 {
2b5e7bc2
KW
1816 possible_problems |= UTF8_GOT_SUPER;
1817 }
1818 else if (curlen > 1) {
1819 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1820 NATIVE_UTF8_TO_I8(*adjusted_s0),
1821 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1822 {
2b5e7bc2 1823 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1824 }
2b5e7bc2
KW
1825 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1826 NATIVE_UTF8_TO_I8(*adjusted_s0),
1827 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1828 {
1829 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1830 }
1831 }
c0236afe 1832
2b5e7bc2
KW
1833 /* We need a complete well-formed UTF-8 character to discern
1834 * non-characters, so can't look for them here */
1835 }
1836 }
949cf498 1837
2b5e7bc2
KW
1838 ready_to_handle_errors:
1839
1840 /* At this point:
1841 * curlen contains the number of bytes in the sequence that
1842 * this call should advance the input by.
e308b348
KW
1843 * avail_len gives the available number of bytes passed in, but
1844 * only if this is less than the expected number of
1845 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1846 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1847 * is set in it for each potential problem found.
1848 * uv contains the code point the input sequence
1849 * represents; or if there is a problem that prevents
1850 * a well-defined value from being computed, it is
1851 * some subsitute value, typically the REPLACEMENT
1852 * CHARACTER.
1853 * s0 points to the first byte of the character
56576a04
KW
1854 * s points to just after were we left off processing
1855 * the character
1856 * send points to just after where that character should
1857 * end, based on how many bytes the start byte tells
1858 * us should be in it, but no further than s0 +
1859 * avail_len
2b5e7bc2 1860 */
eb83ed87 1861
2b5e7bc2
KW
1862 if (UNLIKELY(possible_problems)) {
1863 bool disallowed = FALSE;
1864 const U32 orig_problems = possible_problems;
1865
37657a5b
KW
1866 if (msgs) {
1867 *msgs = NULL;
1868 }
1869
2b5e7bc2 1870 while (possible_problems) { /* Handle each possible problem */
9fde5914 1871 U32 pack_warn = 0;
2b5e7bc2 1872 char * message = NULL;
37657a5b 1873 U32 this_flag_bit = 0;
2b5e7bc2
KW
1874
1875 /* Each 'if' clause handles one problem. They are ordered so that
1876 * the first ones' messages will be displayed before the later
6c64cd9d
KW
1877 * ones; this is kinda in decreasing severity order. But the
1878 * overlong must come last, as it changes 'uv' looked at by the
1879 * others */
2b5e7bc2
KW
1880 if (possible_problems & UTF8_GOT_OVERFLOW) {
1881
56576a04
KW
1882 /* Overflow means also got a super and are using Perl's
1883 * extended UTF-8, but we handle all three cases here */
2b5e7bc2 1884 possible_problems
d044b7a7 1885 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
f9380377
KW
1886 *errors |= UTF8_GOT_OVERFLOW;
1887
1888 /* But the API says we flag all errors found */
1889 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1890 *errors |= UTF8_GOT_SUPER;
1891 }
ddb65933 1892 if (flags
d044b7a7 1893 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 1894 {
d044b7a7 1895 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 1896 }
2b5e7bc2 1897
d60baaa7 1898 /* Disallow if any of the three categories say to */
56576a04 1899 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
d60baaa7 1900 || (flags & ( UTF8_DISALLOW_SUPER
d044b7a7 1901 |UTF8_DISALLOW_PERL_EXTENDED)))
d60baaa7
KW
1902 {
1903 disallowed = TRUE;
1904 }
1905
d22ec717
KW
1906 /* Likewise, warn if any say to */
1907 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1908 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
d60baaa7 1909 {
2b5e7bc2 1910
ddb65933
KW
1911 /* The warnings code explicitly says it doesn't handle the
1912 * case of packWARN2 and two categories which have
1913 * parent-child relationship. Even if it works now to
1914 * raise the warning if either is enabled, it wouldn't
1915 * necessarily do so in the future. We output (only) the
56576a04 1916 * most dire warning */
ddb65933 1917 if (! (flags & UTF8_CHECK_ONLY)) {
37657a5b 1918 if (msgs || ckWARN_d(WARN_UTF8)) {
ddb65933
KW
1919 pack_warn = packWARN(WARN_UTF8);
1920 }
37657a5b 1921 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
ddb65933
KW
1922 pack_warn = packWARN(WARN_NON_UNICODE);
1923 }
1924 if (pack_warn) {
1925 message = Perl_form(aTHX_ "%s: %s (overflows)",
1926 malformed_text,
05b9033b 1927 _byte_dump_string(s0, curlen, 0));
37657a5b 1928 this_flag_bit = UTF8_GOT_OVERFLOW;
ddb65933 1929 }
2b5e7bc2
KW
1930 }
1931 }
1932 }
1933 else if (possible_problems & UTF8_GOT_EMPTY) {
1934 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1935 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1936
1937 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1938
1939 /* This so-called malformation is now treated as a bug in
1940 * the caller. If you have nothing to decode, skip calling
1941 * this function */
1942 assert(0);
1943
2b5e7bc2 1944 disallowed = TRUE;
37657a5b
KW
1945 if ( (msgs
1946 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1947 {
2b5e7bc2
KW
1948 pack_warn = packWARN(WARN_UTF8);
1949 message = Perl_form(aTHX_ "%s (empty string)",
1950 malformed_text);
37657a5b 1951 this_flag_bit = UTF8_GOT_EMPTY;
2b5e7bc2
KW
1952 }
1953 }
1954 }
1955 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1956 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1957 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1958
1959 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1960 disallowed = TRUE;
37657a5b
KW
1961 if (( msgs
1962 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1963 {
2b5e7bc2
KW
1964 pack_warn = packWARN(WARN_UTF8);
1965 message = Perl_form(aTHX_
1966 "%s: %s (unexpected continuation byte 0x%02x,"
1967 " with no preceding start byte)",
1968 malformed_text,
7e2f38b2 1969 _byte_dump_string(s0, 1, 0), *s0);
37657a5b 1970 this_flag_bit = UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1971 }
1972 }
1973 }
2b5e7bc2
KW
1974 else if (possible_problems & UTF8_GOT_SHORT) {
1975 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1976 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1977
1978 if (! (flags & UTF8_ALLOW_SHORT)) {
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_
56576a04
KW
1985 "%s: %s (too short; %d byte%s available, need %d)",
1986 malformed_text,
1987 _byte_dump_string(s0, send - s0, 0),
1988 (int)avail_len,
1989 avail_len == 1 ? "" : "s",
1990 (int)expectlen);
37657a5b 1991 this_flag_bit = UTF8_GOT_SHORT;
2b5e7bc2
KW
1992 }
1993 }
ba210ebe 1994
2b5e7bc2 1995 }
e308b348
KW
1996 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1997 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1998 *errors |= UTF8_GOT_NON_CONTINUATION;
1999
2000 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2001 disallowed = TRUE;
37657a5b
KW
2002 if (( msgs
2003 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2004 {
99a765e9
KW
2005
2006 /* If we don't know for sure that the input length is
2007 * valid, avoid as much as possible reading past the
2008 * end of the buffer */
2009 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
100de20c
KW
2010 ? (int) (s - s0)
2011 : (int) (send - s0);
e308b348
KW
2012 pack_warn = packWARN(WARN_UTF8);
2013 message = Perl_form(aTHX_ "%s",
2014 unexpected_non_continuation_text(s0,
99a765e9 2015 printlen,
e308b348
KW
2016 s - s0,
2017 (int) expectlen));
37657a5b 2018 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
e308b348
KW
2019 }
2020 }
2021 }
2b5e7bc2
KW
2022 else if (possible_problems & UTF8_GOT_SURROGATE) {
2023 possible_problems &= ~UTF8_GOT_SURROGATE;
2024
f9380377
KW
2025 if (flags & UTF8_WARN_SURROGATE) {
2026 *errors |= UTF8_GOT_SURROGATE;
2027
2028 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2029 && (msgs || ckWARN_d(WARN_SURROGATE)))
f9380377 2030 {
2b5e7bc2
KW
2031 pack_warn = packWARN(WARN_SURROGATE);
2032
2033 /* These are the only errors that can occur with a
2034 * surrogate when the 'uv' isn't valid */
2035 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2036 message = Perl_form(aTHX_
2037 "UTF-16 surrogate (any UTF-8 sequence that"
2038 " starts with \"%s\" is for a surrogate)",
7e2f38b2 2039 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2040 }
2041 else {
c94c2f39 2042 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2b5e7bc2 2043 }
37657a5b 2044 this_flag_bit = UTF8_GOT_SURROGATE;
f9380377 2045 }
2b5e7bc2 2046 }
ba210ebe 2047
2b5e7bc2
KW
2048 if (flags & UTF8_DISALLOW_SURROGATE) {
2049 disallowed = TRUE;
f9380377 2050 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
2051 }
2052 }
2053 else if (possible_problems & UTF8_GOT_SUPER) {
2054 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 2055
f9380377
KW
2056 if (flags & UTF8_WARN_SUPER) {
2057 *errors |= UTF8_GOT_SUPER;
2058
2059 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2060 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
f9380377 2061 {
2b5e7bc2
KW
2062 pack_warn = packWARN(WARN_NON_UNICODE);
2063
2064 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2065 message = Perl_form(aTHX_
2066 "Any UTF-8 sequence that starts with"
2067 " \"%s\" is for a non-Unicode code point,"
2068 " may not be portable",
7e2f38b2 2069 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
2070 }
2071 else {
c94c2f39 2072 message = Perl_form(aTHX_ super_cp_format, uv);
2b5e7bc2 2073 }
37657a5b 2074 this_flag_bit = UTF8_GOT_SUPER;
f9380377 2075 }
2b5e7bc2 2076 }
ba210ebe 2077
57ff5f59
KW
2078 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2079 * and before possibly bailing out, so that the more dire
2080 * warning will override the regular one. */
2081 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2b5e7bc2 2082 if ( ! (flags & UTF8_CHECK_ONLY)
d044b7a7 2083 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
dc4a6683
KW
2084 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2085 || ckWARN(WARN_PORTABLE))))
2b5e7bc2 2086 {
dc4a6683 2087 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2b5e7bc2 2088
57ff5f59
KW
2089 /* If it is an overlong that evaluates to a code point
2090 * that doesn't have to use the Perl extended UTF-8, it
2091 * still used it, and so we output a message that
2092 * doesn't refer to the code point. The same is true
2093 * if there was a SHORT malformation where the code
2094 * point is not valid. In that case, 'uv' will have
2095 * been set to the REPLACEMENT CHAR, and the message
2096 * below without the code point in it will be selected
2097 * */
2098 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2b5e7bc2 2099 message = Perl_form(aTHX_
8911f9b0 2100 PL_extended_cp_format, uv);
2b5e7bc2
KW
2101 }
2102 else {
2103 message = Perl_form(aTHX_
57ff5f59
KW
2104 "Any UTF-8 sequence that starts with"
2105 " \"%s\" is a Perl extension, and"
2106 " so is not portable",
2107 _byte_dump_string(s0, curlen, 0));
2b5e7bc2 2108 }
37657a5b 2109 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2b5e7bc2
KW
2110 }
2111
d044b7a7
KW
2112 if (flags & ( UTF8_WARN_PERL_EXTENDED
2113 |UTF8_DISALLOW_PERL_EXTENDED))
ddb65933 2114 {
d044b7a7 2115 *errors |= UTF8_GOT_PERL_EXTENDED;
f9380377 2116
d044b7a7 2117 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
f9380377
KW
2118 disallowed = TRUE;
2119 }
2b5e7bc2
KW
2120 }
2121 }
eb83ed87 2122
2b5e7bc2 2123 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 2124 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
2125 disallowed = TRUE;
2126 }
2b5e7bc2
KW
2127 }
2128 else if (possible_problems & UTF8_GOT_NONCHAR) {
2129 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 2130
f9380377
KW
2131 if (flags & UTF8_WARN_NONCHAR) {
2132 *errors |= UTF8_GOT_NONCHAR;
2133
2134 if ( ! (flags & UTF8_CHECK_ONLY)
37657a5b 2135 && (msgs || ckWARN_d(WARN_NONCHAR)))
f9380377 2136 {
2b5e7bc2
KW
2137 /* The code above should have guaranteed that we don't
2138 * get here with errors other than overlong */
2139 assert (! (orig_problems
2140 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2141
2142 pack_warn = packWARN(WARN_NONCHAR);
c94c2f39 2143 message = Perl_form(aTHX_ nonchar_cp_format, uv);
37657a5b 2144 this_flag_bit = UTF8_GOT_NONCHAR;
f9380377 2145 }
2b5e7bc2 2146 }
5b311467 2147
2b5e7bc2
KW
2148 if (flags & UTF8_DISALLOW_NONCHAR) {
2149 disallowed = TRUE;
f9380377 2150 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2 2151 }
6c64cd9d
KW
2152 }
2153 else if (possible_problems & UTF8_GOT_LONG) {
2154 possible_problems &= ~UTF8_GOT_LONG;
2155 *errors |= UTF8_GOT_LONG;
2156
2157 if (flags & UTF8_ALLOW_LONG) {
2158
2159 /* We don't allow the actual overlong value, unless the
2160 * special extra bit is also set */
2161 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2162 & ~UTF8_ALLOW_LONG)))
2163 {
2164 uv = UNICODE_REPLACEMENT;
2165 }
2166 }
2167 else {
2168 disallowed = TRUE;
2169
37657a5b
KW
2170 if (( msgs
2171 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2172 {
6c64cd9d
KW
2173 pack_warn = packWARN(WARN_UTF8);
2174
2175 /* These error types cause 'uv' to be something that
2176 * isn't what was intended, so can't use it in the
2177 * message. The other error types either can't
2178 * generate an overlong, or else the 'uv' is valid */
2179 if (orig_problems &
2180 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2181 {
2182 message = Perl_form(aTHX_
2183 "%s: %s (any UTF-8 sequence that starts"
2184 " with \"%s\" is overlong which can and"
2185 " should be represented with a"
2186 " different, shorter sequence)",
2187 malformed_text,
2188 _byte_dump_string(s0, send - s0, 0),
2189 _byte_dump_string(s0, curlen, 0));
2190 }
2191 else {
2192 U8 tmpbuf[UTF8_MAXBYTES+1];
1be62ab9
KW
2193 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2194 uv, 0);
d819dc50
KW
2195 /* Don't use U+ for non-Unicode code points, which
2196 * includes those in the Latin1 range */
2197 const char * preface = ( uv > PERL_UNICODE_MAX
2198#ifdef EBCDIC
2199 || uv <= 0xFF
2200#endif
2201 )
2202 ? "0x"
2203 : "U+";
6c64cd9d
KW
2204 message = Perl_form(aTHX_
2205 "%s: %s (overlong; instead use %s to represent"
2206 " %s%0*" UVXf ")",
2207 malformed_text,
2208 _byte_dump_string(s0, send - s0, 0),
2209 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2210 preface,
2211 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2212 small code points */
1be62ab9 2213 UNI_TO_NATIVE(uv));
6c64cd9d 2214 }
37657a5b 2215 this_flag_bit = UTF8_GOT_LONG;
6c64cd9d
KW
2216 }
2217 }
2b5e7bc2
KW
2218 } /* End of looking through the possible flags */
2219
2220 /* Display the message (if any) for the problem being handled in
2221 * this iteration of the loop */
2222 if (message) {
37657a5b 2223 if (msgs) {
37657a5b
KW
2224 assert(this_flag_bit);
2225
2226 if (*msgs == NULL) {
2227 *msgs = newAV();
2228 }
2229
bb07812e
KW
2230 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2231 pack_warn,
2232 this_flag_bit)));
37657a5b
KW
2233 }
2234 else if (PL_op)
2b5e7bc2
KW
2235 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2236 OP_DESC(PL_op));
2237 else
2238 Perl_warner(aTHX_ pack_warn, "%s", message);
2239 }
ddb65933 2240 } /* End of 'while (possible_problems)' */
a0dbb045 2241
2b5e7bc2
KW
2242 /* Since there was a possible problem, the returned length may need to
2243 * be changed from the one stored at the beginning of this function.
2244 * Instead of trying to figure out if that's needed, just do it. */
2245 if (retlen) {
2246 *retlen = curlen;
2247 }
a0dbb045 2248
2b5e7bc2
KW
2249 if (disallowed) {
2250 if (flags & UTF8_CHECK_ONLY && retlen) {
2251 *retlen = ((STRLEN) -1);
2252 }
2253 return 0;
2254 }
eb83ed87 2255 }
ba210ebe 2256
2b5e7bc2 2257 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
2258}
2259
8e84507e 2260/*
ec5f19d0
KW
2261=for apidoc utf8_to_uvchr_buf
2262
2263Returns the native code point of the first character in the string C<s> which
2264is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 2265C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 2266
524080c4
KW
2267If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2268enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 2269C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 2270(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 2271C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 2272the next possible position in C<s> that could begin a non-malformed character.
de69f3af 2273See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 2274returned.
ec5f19d0
KW
2275
2276=cut
52be2536
KW
2277
2278Also implemented as a macro in utf8.h
2279
ec5f19d0
KW
2280*/
2281
2282
2283UV
2284Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2285{
7f974d7e
KW
2286 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2287
9a9a6c98 2288 return utf8_to_uvchr_buf_helper(s, send, retlen);
ec5f19d0
KW
2289}
2290
52be2536
KW
2291/* This is marked as deprecated
2292 *
ec5f19d0
KW
2293=for apidoc utf8_to_uvuni_buf
2294
de69f3af
KW
2295Only in very rare circumstances should code need to be dealing in Unicode
2296(as opposed to native) code points. In those few cases, use
09232555
KW
2297C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2298If you are not absolutely sure this is one of those cases, then assume it isn't
2299and use plain C<utf8_to_uvchr_buf> instead.
4f83cdcd
KW
2300
2301Returns the Unicode (not-native) code point of the first character in the
2302string C<s> which
ec5f19d0
KW
2303is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2304C<retlen> will be set to the length, in bytes, of that character.
2305
524080c4
KW
2306If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2307enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2308NULL) to -1. If those warnings are off, the computed value if well-defined (or
2309the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2310is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2311next possible position in C<s> that could begin a non-malformed character.
09232555
KW
2312See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2313returned.
ec5f19d0
KW
2314
2315=cut
2316*/
2317
2318UV
2319Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2320{
2321 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2322
2323 assert(send > s);
2324
5962d97e 2325 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
2326}
2327
b76347f2 2328/*
87cea99e 2329=for apidoc utf8_length
b76347f2 2330
b2e7ed74
KW
2331Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2332at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2333same place, it returns 0 with no warning raised.
2334
2335If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2336and returns the number of valid characters.
b76347f2
JH
2337
2338=cut
2339*/
2340
2341STRLEN
35a4481c 2342Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
2343{
2344 STRLEN len = 0;
2345
7918f24d
NC
2346 PERL_ARGS_ASSERT_UTF8_LENGTH;
2347
8850bf83
JH
2348 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2349 * the bitops (especially ~) can create illegal UTF-8.
2350 * In other words: in Perl UTF-8 is not just for Unicode. */
2351
12c43b0a 2352 if (UNLIKELY(e < s))
a3b680e6 2353 goto warn_and_return;
b76347f2 2354 while (s < e) {
4cbf4130 2355 s += UTF8SKIP(s);
8e91ec7f
AV
2356 len++;
2357 }
2358
12c43b0a 2359 if (UNLIKELY(e != s)) {
8e91ec7f
AV
2360 len--;
2361 warn_and_return:
9b387841
NC
2362 if (PL_op)
2363 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2364 "%s in %s", unees, OP_DESC(PL_op));
2365 else
61a12c31 2366 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
2367 }
2368
2369 return len;
2370}
2371
b06226ff 2372/*
fed3ba5d
NC
2373=for apidoc bytes_cmp_utf8
2374
a1433954 2375Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
2376sequence of characters (stored as UTF-8)
2377in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
2378equal, -1 or -2 if the first string is less than the second string, +1 or +2
2379if the first string is greater than the second string.
2380
2381-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
2382longer string. -2 or +2 is returned if
2383there was a difference between characters
fed3ba5d
NC
2384within the strings.
2385
2386=cut
2387*/
2388
2389int
2390Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2391{
2392 const U8 *const bend = b + blen;
2393 const U8 *const uend = u + ulen;
2394
2395 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
2396
2397 while (b < bend && u < uend) {
2398 U8 c = *u++;
2399 if (!UTF8_IS_INVARIANT(c)) {
2400 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2401 if (u < uend) {
2402 U8 c1 = *u++;
2403 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 2404 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 2405 } else {
2b5e7bc2 2406 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 2407 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
56576a04
KW
2408 "%s %s%s",
2409 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2410 PL_op ? " in " : "",
2411 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
2412 return -2;
2413 }
2414 } else {
2415 if (PL_op)
2416 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2417 "%s in %s", unees, OP_DESC(PL_op));
2418 else
61a12c31 2419 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
2420 return -2; /* Really want to return undef :-) */
2421 }
2422 } else {
2423 return -2;
2424 }
2425 }
2426 if (*b != c) {
2427 return *b < c ? -2 : +2;
2428 }
2429 ++b;
2430 }
2431
2432 if (b == bend && u == uend)
2433 return 0;
2434
2435 return b < bend ? +1 : -1;
2436}
2437
2438/*
87cea99e 2439=for apidoc utf8_to_bytes
6940069f 2440
3bc0c78c 2441Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
a1433954 2442Unlike L</bytes_to_utf8>, this over-writes the original string, and
09af0336 2443updates C<*lenp> to contain the new length.
3bc0c78c
KW
2444Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2445
2446Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2447having saved the value of C<*lenp> before the call, and subtracting the
2448after-call value of C<*lenp> from it.
6940069f 2449
a1433954 2450If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 2451
6940069f
GS
2452=cut
2453*/
2454
2455U8 *
09af0336 2456Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
6940069f 2457{
9fe0d3c2 2458 U8 * first_variant;
246fae53 2459
7918f24d 2460 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 2461 PERL_UNUSED_CONTEXT;
7918f24d 2462
9fe0d3c2 2463 /* This is a no-op if no variants at all in the input */
09af0336 2464 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
9fe0d3c2
KW
2465 return s;
2466 }
2467
2468 {
3c5aa262 2469 U8 * const save = s;
09af0336 2470 U8 * const send = s + *lenp;
3c5aa262
KW
2471 U8 * d;
2472
2473 /* Nothing before the first variant needs to be changed, so start the real
2474 * work there */
2475 s = first_variant;
2476 while (s < send) {
2477 if (! UTF8_IS_INVARIANT(*s)) {
2478 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
09af0336 2479 *lenp = ((STRLEN) -1);
3c5aa262
KW
2480 return 0;
2481 }
2482 s++;
d59937ca
KW
2483 }
2484 s++;
dcad2880 2485 }
dcad2880 2486
3c5aa262
KW
2487 /* Is downgradable, so do it */
2488 d = s = first_variant;
2489 while (s < send) {
2490 U8 c = *s++;
2491 if (! UVCHR_IS_INVARIANT(c)) {
2492 /* Then it is two-byte encoded */
2493 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2494 s++;
2495 }
2496 *d++ = c;
2497 }
2498 *d = '\0';
09af0336 2499 *lenp = d - save;
3c5aa262
KW
2500
2501 return save;
9fe0d3c2 2502 }
6940069f
GS
2503}
2504
2505/*
87cea99e 2506=for apidoc bytes_from_utf8
f9a63242 2507
09af0336 2508Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
41ae6089 2509byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
4f3d592d
KW
2510actually encoded in UTF-8.
2511
2512Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2513the input string.
2514
41ae6089
KW
2515Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2516not expressible in native byte encoding. In these cases, C<*is_utf8p> and
09af0336 2517C<*lenp> are unchanged, and the return value is the original C<s>.
4f3d592d 2518
41ae6089 2519Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
4f3d592d 2520newly created string containing a downgraded copy of C<s>, and whose length is
9ff99fb3
KW
2521returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2522caller is responsible for arranging for the memory used by this string to get
2523freed.
f9a63242 2524
3bc0c78c 2525Upon successful return, the number of variants in the string can be computed by
23b37b12
KW
2526having saved the value of C<*lenp> before the call, and subtracting the
2527after-call value of C<*lenp> from it.
3bc0c78c 2528
37607a96 2529=cut
976c1b08
KW
2530
2531There is a macro that avoids this function call, but this is retained for
2532anyone who calls it with the Perl_ prefix */
f9a63242
JH
2533
2534U8 *
41ae6089 2535Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
f9a63242 2536{
7918f24d 2537 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 2538 PERL_UNUSED_CONTEXT;
f9a63242 2539
976c1b08
KW
2540 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2541}
2542
2543/*
df6bd76f 2544=for apidoc bytes_from_utf8_loc
976c1b08
KW
2545
2546Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2547to store the location of the first character in C<"s"> that cannot be
2548converted to non-UTF8.
2549
2550If that parameter is C<NULL>, this function behaves identically to
2551C<bytes_from_utf8>.
2552
2553Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2554C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2555
2556Otherwise, the function returns a newly created C<NUL>-terminated string
2557containing the non-UTF8 equivalent of the convertible first portion of
2558C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2559If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2560and C<*first_non_downgradable> is set to C<NULL>.
2561
8505db87 2562Otherwise, C<*first_non_downgradable> is set to point to the first byte of the
976c1b08
KW
2563first character in the original string that wasn't converted. C<*is_utf8p> is
2564unchanged. Note that the new string may have length 0.
2565
2566Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2567C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2568converts as many characters in it as possible stopping at the first one it
385b74be 2569finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
976c1b08
KW
2570set to point to that. The function returns the portion that could be converted
2571in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2572not including the terminating C<NUL>. If the very first character in the
2573original could not be converted, C<*lenp> will be 0, and the new string will
2574contain just a single C<NUL>. If the entire input string was converted,
2575C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2576
2577Upon successful return, the number of variants in the converted portion of the
2578string can be computed by having saved the value of C<*lenp> before the call,
2579and subtracting the after-call value of C<*lenp> from it.
2580
2581=cut
2582
2583
2584*/
2585
2586U8 *
2587Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2588{
2589 U8 *d;
2590 const U8 *original = s;
2591 U8 *converted_start;
2592 const U8 *send = s + *lenp;
f9a63242 2593
976c1b08 2594 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
170a1c22 2595
976c1b08
KW
2596 if (! *is_utf8p) {
2597 if (first_unconverted) {
2598 *first_unconverted = NULL;
2599 }
2600
2601 return (U8 *) original;
2602 }
2603
2604 Newx(d, (*lenp) + 1, U8);
2605
2606 converted_start = d;
7299a045
KW
2607 while (s < send) {
2608 U8 c = *s++;
2609 if (! UTF8_IS_INVARIANT(c)) {
976c1b08
KW
2610
2611 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2612 * have to stop now */
2613 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2614 if (first_unconverted) {
2615 *first_unconverted = s - 1;
2616 goto finish_and_return;
2617 }
2618 else {
2619 Safefree(converted_start);
2620 return (U8 *) original;
2621 }
2622 }
2623
7299a045
KW
2624 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2625 s++;
38af28cf 2626 }
7299a045
KW
2627 *d++ = c;
2628 }
170a1c22 2629
976c1b08
KW
2630 /* Here, converted the whole of the input */
2631 *is_utf8p = FALSE;
2632 if (first_unconverted) {
2633 *first_unconverted = NULL;
170a1c22 2634 }
976c1b08
KW
2635
2636 finish_and_return:
46a08a6f
KW
2637 *d = '\0';
2638 *lenp = d - converted_start;
976c1b08
KW
2639
2640 /* Trim unused space */
2641 Renew(converted_start, *lenp + 1, U8);
2642
2643 return converted_start;
f9a63242
JH
2644}
2645
2646/*
87cea99e 2647=for apidoc bytes_to_utf8
6940069f 2648
09af0336 2649Converts a string C<s> of length C<*lenp> bytes from the native encoding into
ff97e5cf 2650UTF-8.
09af0336 2651Returns a pointer to the newly-created string, and sets C<*lenp> to
9ff99fb3
KW
2652reflect the new length in bytes. The caller is responsible for arranging for
2653the memory used by this string to get freed.
6940069f 2654
3bc0c78c 2655Upon successful return, the number of variants in the string can be computed by
23b37b12 2656having saved the value of C<*lenp> before the call, and subtracting it from the
3bc0c78c
KW
2657after-call value of C<*lenp>.
2658
75200dff 2659A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2660
2661If you want to convert to UTF-8 from encodings other than
2662the native (Latin1 or EBCDIC),
a1433954 2663see L</sv_recode_to_utf8>().
c9ada85f 2664
497711e7 2665=cut
6940069f
GS
2666*/
2667
2668U8*
09af0336 2669Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
6940069f 2670{
09af0336 2671 const U8 * const send = s + (*lenp);
6940069f
GS
2672 U8 *d;
2673 U8 *dst;
7918f24d
NC
2674
2675 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2676 PERL_UNUSED_CONTEXT;
6940069f 2677
d4662719
KW
2678 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2679 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
6940069f
GS
2680 dst = d;
2681
2682 while (s < send) {
55d09dc8
KW
2683 append_utf8_from_native_byte(*s, &d);
2684 s++;
6940069f 2685 }
2e11cf67 2686
6940069f 2687 *d = '\0';
09af0336 2688 *lenp = d-dst;
2e11cf67 2689
6940069f
GS
2690 return dst;
2691}
2692
a0ed51b3 2693/*
624504c5
KW
2694 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2695 * use utf16_to_utf8_reversed().
a0ed51b3 2696 *
624504c5
KW
2697 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2698 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2699 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2700 *
2701 * These functions don't check for overflow. The worst case is every code
2702 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2703 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2704 * destination must be pre-extended to 2 times the source length.
2705 *
2706 * Do not use in-place. We optimize for native, for obvious reasons. */
a0ed51b3
LW
2707
2708U8*
f46dcac2 2709Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3 2710{
dea0fc0b
JH
2711 U8* pend;
2712 U8* dstart = d;
2713
7918f24d
NC
2714 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2715
dea0fc0b 2716 if (bytelen & 1)
56576a04
KW
2717 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2718 (UV)bytelen);
dea0fc0b
JH
2719
2720 pend = p + bytelen;
2721
a0ed51b3 2722 while (p < pend) {
dea0fc0b
JH
2723 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2724 p += 2;
2d1545e5 2725 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2726 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2727 continue;
2728 }
56d37426
KW
2729 if (uv <= MAX_UTF8_TWO_BYTE) {
2730 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2731 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2732 continue;
2733 }
ffd0a9d3 2734
46956fad
KW
2735#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2736#define LAST_HIGH_SURROGATE 0xDBFF
2737#define FIRST_LOW_SURROGATE 0xDC00
2738#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
ffd0a9d3 2739#define FIRST_IN_PLANE1 0x10000
e23c50db
KW
2740
2741 /* This assumes that most uses will be in the first Unicode plane, not
2742 * needing surrogates */
b497502c
KW
2743 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2744 UNICODE_SURROGATE_LAST)))
e23c50db
KW
2745 {
2746 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2747 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2748 }
2749 else {
01ea242b 2750 UV low = (p[0] << 8) + p[1];
b497502c
KW
2751 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2752 LAST_LOW_SURROGATE)))
e23c50db 2753 {
01ea242b 2754 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2755 }
2756 p += 2;
46956fad 2757 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
ffd0a9d3 2758 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
01ea242b 2759 }
a0ed51b3 2760 }
56d37426
KW
2761#ifdef EBCDIC
2762 d = uvoffuni_to_utf8_flags(d, uv, 0);
2763#else
ffd0a9d3 2764 if (uv < FIRST_IN_PLANE1) {
eb160463
GS
2765 *d++ = (U8)(( uv >> 12) | 0xe0);
2766 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2767 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2768 continue;
2769 }
2770 else {
eb160463
GS
2771 *d++ = (U8)(( uv >> 18) | 0xf0);
2772 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2773 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2774 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2775 continue;
2776 }
56d37426 2777#endif
a0ed51b3 2778 }
dea0fc0b 2779 *newlen = d - dstart;
a0ed51b3
LW
2780 return d;
2781}
2782
2783/* Note: this one is slightly destructive of the source. */
2784
2785U8*
f46dcac2 2786Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
a0ed51b3
LW
2787{
2788 U8* s = (U8*)p;
d4c19fe8 2789 U8* const send = s + bytelen;
7918f24d
NC
2790
2791 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2792
e0ea5e2d 2793 if (bytelen & 1)
147e3846 2794 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2795 (UV)bytelen);
2796
a0ed51b3 2797 while (s < send) {
d4c19fe8 2798 const U8 tmp = s[0];
a0ed51b3
LW
2799 s[0] = s[1];
2800 s[1] = tmp;
2801 s += 2;
2802 }
dea0fc0b 2803 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2804}
2805
922e8cb4
KW
2806bool
2807Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2808{
dc31b55c 2809 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
922e8cb4
KW
2810}
2811
5092f92a 2812bool
eba68aa0
KW
2813Perl__is_uni_perl_idcont(pTHX_ UV c)
2814{
c12658c9 2815 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
eba68aa0
KW
2816}
2817
2818bool
f91dcd13
KW
2819Perl__is_uni_perl_idstart(pTHX_ UV c)
2820{
c12658c9 2821 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
f91dcd13
KW
2822}
2823
3a4c58c9 2824UV
56576a04
KW
2825Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2826 const char S_or_s)
3a4c58c9
KW
2827{
2828 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2829 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2830 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2831 * either "SS" or "Ss". Which one to use is passed into the routine in
2832 * 'S_or_s' to avoid a test */
2833
2834 UV converted = toUPPER_LATIN1_MOD(c);
2835
2836 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2837
2838 assert(S_or_s == 'S' || S_or_s == 's');
2839
6f2d5cbc 2840 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2841 characters in this range */
3a4c58c9
KW
2842 *p = (U8) converted;
2843 *lenp = 1;
2844 return converted;
2845 }
2846
2847 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2848 * which it maps to one of them, so as to only have to have one check for
2849 * it in the main case */
2850 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2851 switch (c) {
2852 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2853 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2854 break;
2855 case MICRO_SIGN:
2856 converted = GREEK_CAPITAL_LETTER_MU;
2857 break;
79e064b9
KW
2858#if UNICODE_MAJOR_VERSION > 2 \
2859 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2860 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2861 case LATIN_SMALL_LETTER_SHARP_S:
2862 *(p)++ = 'S';
2863 *p = S_or_s;
2864 *lenp = 2;
2865 return 'S';
79e064b9 2866#endif
3a4c58c9 2867 default:
56576a04
KW
2868 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2869 " '%c' to map to '%c'",
2870 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2871 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2872 }
2873 }
2874
2875 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2876 *p = UTF8_TWO_BYTE_LO(converted);
2877 *lenp = 2;
2878
2879 return converted;
2880}
2881
fe63c520
KW
2882/* If compiled on an early Unicode version, there may not be auxiliary tables
2883 * */
2884#ifndef HAS_UC_AUX_TABLES
2885# define UC_AUX_TABLE_ptrs NULL
2886# define UC_AUX_TABLE_lengths NULL
2887#endif
2888#ifndef HAS_TC_AUX_TABLES
2889# define TC_AUX_TABLE_ptrs NULL
2890# define TC_AUX_TABLE_lengths NULL
2891#endif
2892#ifndef HAS_LC_AUX_TABLES
2893# define LC_AUX_TABLE_ptrs NULL
2894# define LC_AUX_TABLE_lengths NULL
2895#endif
2896#ifndef HAS_CF_AUX_TABLES
2897# define CF_AUX_TABLE_ptrs NULL
2898# define CF_AUX_TABLE_lengths NULL
2899#endif
2900#ifndef HAS_UC_AUX_TABLES
2901# define UC_AUX_TABLE_ptrs NULL
2902# define UC_AUX_TABLE_lengths NULL
2903#endif
2904
50bda2c3
KW
2905/* Call the function to convert a UTF-8 encoded character to the specified case.
2906 * Note that there may be more than one character in the result.
6fa2f9bc
KW
2907 * 's' is a pointer to the first byte of the input character
2908 * 'd' will be set to the first byte of the string of changed characters. It
50bda2c3 2909 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
6fa2f9bc 2910 * 'lenp' will be set to the length in bytes of the string of changed characters
50bda2c3 2911 *
56576a04 2912 * The functions return the ordinal of the first character in the string of
6fa2f9bc 2913 * 'd' */
56576a04 2914#define CALL_UPPER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2915 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2916 Uppercase_Mapping_invmap, \
2917 UC_AUX_TABLE_ptrs, \
2918 UC_AUX_TABLE_lengths, \
2919 "uppercase")
56576a04 2920#define CALL_TITLE_CASE(uv, s, d, lenp) \
8946fcd9
KW
2921 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2922 Titlecase_Mapping_invmap, \
2923 TC_AUX_TABLE_ptrs, \
2924 TC_AUX_TABLE_lengths, \
2925 "titlecase")
56576a04 2926#define CALL_LOWER_CASE(uv, s, d, lenp) \
8946fcd9
KW
2927 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2928 Lowercase_Mapping_invmap, \
2929 LC_AUX_TABLE_ptrs, \
2930 LC_AUX_TABLE_lengths, \
2931 "lowercase")
2932
50bda2c3 2933
b9992569
KW
2934/* This additionally has the input parameter 'specials', which if non-zero will
2935 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2936 * folding); otherwise, when zero, this implies a simple case fold */
56576a04 2937#define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
8946fcd9
KW
2938 (specials) \
2939 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2940 Case_Folding_invmap, \
2941 CF_AUX_TABLE_ptrs, \
2942 CF_AUX_TABLE_lengths, \
2943 "foldcase") \
2944 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2945 Simple_Case_Folding_invmap, \
2946 NULL, NULL, \
2947 "foldcase")
c3fd2246 2948
84afefe6
JH
2949UV
2950Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2951{
a1433954
KW
2952 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2953 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2954 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2955 * the changed version may be longer than the original character.
2956 *
2957 * The ordinal of the first character of the changed version is returned
2958 * (but note, as explained above, that there may be more.) */
2959
7918f24d
NC
2960 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2961
3a4c58c9
KW
2962 if (c < 256) {
2963 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2964 }
2965
a13f1de4 2966 return CALL_UPPER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2967}
2968
84afefe6
JH
2969UV
2970Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2971{
7918f24d
NC
2972 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2973
3a4c58c9
KW
2974 if (c < 256) {
2975 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2976 }
2977
a13f1de4 2978 return CALL_TITLE_CASE(c, NULL, p, lenp);
a0ed51b3
LW
2979}
2980
afc16117 2981STATIC U8
eaf412bf 2982S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2983{
2984 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2985 * those, converting the result to UTF-8. Since the result is always just
a1433954 2986 * one character, we allow <p> to be NULL */
afc16117
KW
2987
2988 U8 converted = toLOWER_LATIN1(c);
2989
eaf412bf
KW
2990 PERL_UNUSED_ARG(dummy);
2991
afc16117 2992 if (p != NULL) {
6f2d5cbc 2993 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2994 *p = converted;
2995 *lenp = 1;
2996 }
2997 else {
430c9760
KW
2998 /* Result is known to always be < 256, so can use the EIGHT_BIT
2999 * macros */
3000 *p = UTF8_EIGHT_BIT_HI(converted);
3001 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
3002 *lenp = 2;
3003 }
3004 }
3005 return converted;
3006}
3007
84afefe6
JH
3008UV
3009Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 3010{
7918f24d
NC
3011 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3012
afc16117 3013 if (c < 256) {
eaf412bf 3014 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
3015 }
3016
a13f1de4 3017 return CALL_LOWER_CASE(c, NULL, p, lenp);
a0ed51b3
LW
3018}
3019
84afefe6 3020UV
7c0ab950 3021Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 3022{
51910141 3023 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 3024 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 3025 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
3026 *
3027 * Not to be used for locale folds
51910141 3028 */
f673fad4 3029
a1dde8de
KW
3030 UV converted;
3031
3032 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3033
1ca267a5
KW
3034 assert (! (flags & FOLD_FLAGS_LOCALE));
3035
659a7c2d 3036 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
3037 converted = GREEK_SMALL_LETTER_MU;
3038 }
9b63e895
KW
3039#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3040 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3041 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
3042 else if ( (flags & FOLD_FLAGS_FULL)
3043 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3044 {
1ca267a5
KW
3045 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3046 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3047 * under those circumstances. */
3048 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3049 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3050 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3051 p, *lenp, U8);
3052 return LATIN_SMALL_LETTER_LONG_S;
3053 }
3054 else {
4f489194
KW
3055 *(p)++ = 's';
3056 *p = 's';
3057 *lenp = 2;
3058 return 's';
1ca267a5 3059 }
a1dde8de 3060 }
9b63e895 3061#endif
a1dde8de
KW
3062 else { /* In this range the fold of all other characters is their lower
3063 case */
3064 converted = toLOWER_LATIN1(c);
3065 }
3066
6f2d5cbc 3067 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
3068 *p = (U8) converted;
3069 *lenp = 1;
3070 }
3071 else {
3072 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3073 *p = UTF8_TWO_BYTE_LO(converted);
3074 *lenp = 2;
3075 }
3076
3077 return converted;
3078}
3079
3080UV
31f05a37 3081Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 3082{
4b593389 3083
a0270393
KW
3084 /* Not currently externally documented, and subject to change
3085 * <flags> bits meanings:
3086 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
3087 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3088 * locale are to be used.
a0270393
KW
3089 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3090 */
4b593389 3091
36bb2ab6 3092 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 3093
780fcc9f 3094 if (flags & FOLD_FLAGS_LOCALE) {
b257a28c
KW
3095 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3096 * except for potentially warning */
8b7358b9 3097 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
b257a28c 3098 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
780fcc9f
KW
3099 flags &= ~FOLD_FLAGS_LOCALE;
3100 }
3101 else {
e7b7ac46 3102 goto needs_full_generality;
780fcc9f 3103 }
31f05a37
KW
3104 }
3105
a1dde8de 3106 if (c < 256) {
e7b7ac46 3107 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 3108 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
3109 }
3110
2f306ab9 3111 /* Here, above 255. If no special needs, just use the macro */
a0270393 3112 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
a13f1de4 3113 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 3114 }
567b353c 3115 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
3116 the special flags. */
3117 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
3118
3119 needs_full_generality:
a0270393 3120 uvchr_to_utf8(utf8_c, c);
56576a04
KW
3121 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3122 p, lenp, flags);
a0270393 3123 }
84afefe6
JH
3124}
3125
26483009 3126PERL_STATIC_INLINE bool
dd1a3ba7
KW
3127S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3128 SV* const invlist)
da8c1a98
KW
3129{
3130 /* returns a boolean giving whether or not the UTF8-encoded character that
eb1f4bb4
KW
3131 * starts at <p>, and extending no further than <e - 1> is in the inversion
3132 * list <invlist>. */
da8c1a98 3133
b68ffe0c
KW
3134 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3135
dd1a3ba7 3136 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
da8c1a98 3137
b68ffe0c 3138 if (cp == 0 && (p >= e || *p != '\0')) {
da8c1a98
KW
3139 _force_out_malformed_utf8_message(p, e, 0, 1);
3140 NOT_REACHED; /* NOTREACHED */
3141 }
3142
eb1f4bb4 3143 assert(invlist);
b68ffe0c 3144 return _invlist_contains_cp(invlist, cp);
da8c1a98
KW
3145}
3146
059703b0 3147#if 0 /* Not currently used, but may be needed in the future */
dd1a3ba7
KW
3148PERLVAR(I, seen_deprecated_macro, HV *)
3149
34aeb2e9
KW
3150STATIC void
3151S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3152 const char * const alternative,
3153 const bool use_locale,
3154 const char * const file,
3155 const unsigned line)
3156{
3157 const char * key;
3158
3159 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3160
3161 if (ckWARN_d(WARN_DEPRECATED)) {
3162
3163 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3164 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3165 if (! PL_seen_deprecated_macro) {
3166 PL_seen_deprecated_macro = newHV();
3167 }
3168 if (! hv_store(PL_seen_deprecated_macro, key,
3169 strlen(key), &PL_sv_undef, 0))
3170 {
3171 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3172 }
3173
c44e9413 3174 if (instr(file, "mathoms.c")) {
607313a1 3175 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3176 "In %s, line %d, starting in Perl v5.32, %s()"
607313a1
KW
3177 " will be removed. Avoid this message by"
3178 " converting to use %s().\n",
3179 file, line, name, alternative);
3180 }
3181 else {
34aeb2e9 3182 Perl_warner(aTHX_ WARN_DEPRECATED,
5203d63d 3183 "In %s, line %d, starting in Perl v5.32, %s() will"
34aeb2e9
KW
3184 " require an additional parameter. Avoid this"
3185 " message by converting to use %s().\n",
3186 file, line, name, alternative);
607313a1 3187 }
34aeb2e9
KW
3188 }
3189 }
3190}
059703b0 3191#endif
922e8cb4
KW
3192
3193bool
dd1a3ba7 3194Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
da8c1a98 3195{
dd1a3ba7 3196 PERL_ARGS_ASSERT__IS_UTF8_FOO;
da8c1a98 3197
dd1a3ba7 3198 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
da8c1a98
KW
3199}
3200
3201bool
dd1a3ba7 3202Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
da8c1a98 3203{
dd1a3ba7 3204 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
da8c1a98 3205
dd1a3ba7 3206 return is_utf8_common(p, e, PL_utf8_perl_idstart);
da8c1a98
KW
3207}
3208
3209bool
dd1a3ba7 3210Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
c11ff943 3211{
dd1a3ba7 3212 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
7dbf68d2 3213
dd1a3ba7 3214 return is_utf8_common(p, e, PL_utf8_perl_idcont);
7dbf68d2
KW
3215}
3216
6a4a25f4 3217STATIC UV
30613bdc
KW
3218S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3219 U8* ustrp, STRLEN *lenp,
40d2776f
KW
3220 SV *invlist, const I32 * const invmap,
3221 const U32 * const * const aux_tables,
30613bdc
KW
3222 const U8 * const aux_table_lengths,
3223 const char * const normal)
b9992569 3224{
0134edef 3225 STRLEN len = 0;
7918f24d 3226
30613bdc
KW
3227 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3228 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3229 * to name the new case in any generated messages, as a fallback if the
3230 * operation being used is not available. The new case is given by the
3231 * data structures in the remaining arguments.
3232 *
3233 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3234 * entire changed case string, and the return value is the first code point
3235 * in that string */
3236
b9992569 3237 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 3238
36eaa811
KW
3239 /* For code points that don't change case, we already know that the output
3240 * of this function is the unchanged input, so we can skip doing look-ups
3241 * for them. Unfortunately the case-changing code points are scattered
3242 * around. But there are some long consecutive ranges where there are no
3243 * case changing code points. By adding tests, we can eliminate the lookup
3244 * for all the ones in such ranges. This is currently done here only for
3245 * just a few cases where the scripts are in common use in modern commerce
3246 * (and scripts adjacent to those which can be included without additional
3247 * tests). */
3248
3249 if (uv1 >= 0x0590) {
3250 /* This keeps from needing further processing the code points most
3251 * likely to be used in the following non-cased scripts: Hebrew,
3252 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3253 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3254 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3255 if (uv1 < 0x10A0) {
3256 goto cases_to_self;
3257 }
3258
3259 /* The following largish code point ranges also don't have case
3260 * changes, but khw didn't think they warranted extra tests to speed
3261 * them up (which would slightly slow down everything else above them):
3262 * 1100..139F Hangul Jamo, Ethiopic
3263 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3264 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3265 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3266 * Combining Diacritical Marks Extended, Balinese,
3267 * Sundanese, Batak, Lepcha, Ol Chiki
3268 * 2000..206F General Punctuation
3269 */
3270
3271 if (uv1 >= 0x2D30) {
3272
3273 /* This keeps the from needing further processing the code points
3274 * most likely to be used in the following non-cased major scripts:
3275 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3276 *
3277 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3278 * event that Unicode eventually allocates the unused block as of
3279 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3280 * that the test suite will start having failures to alert you
3281 * should that happen) */
3282 if (uv1 < 0xA640) {
3283 goto cases_to_self;
3284 }
3285
3286 if (uv1 >= 0xAC00) {
3287 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
3288 if (ckWARN_d(WARN_SURROGATE)) {
3289 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3290 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
56576a04
KW
3291 "Operation \"%s\" returns its argument for"
3292 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
3293 }
3294 goto cases_to_self;
3295 }
36eaa811
KW
3296
3297 /* AC00..FAFF Catches Hangul syllables and private use, plus
3298 * some others */
3299 if (uv1 < 0xFB00) {
3300 goto cases_to_self;
36eaa811
KW
3301 }
3302
5af9bc97 3303 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
40606899 3304 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
fb2f0a6a 3305 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1));
5af9bc97
KW
3306 }
3307 if (ckWARN_d(WARN_NON_UNICODE)) {
3308 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3309 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
56576a04
KW
3310 "Operation \"%s\" returns its argument for"
3311 " non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
3312 }
3313 goto cases_to_self;
3314 }
3bfc1e70
KW
3315#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3316 if (UNLIKELY(uv1
3317 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3318 {
3319
3bfc1e70
KW
3320 goto cases_to_self;
3321 }
3322#endif
36eaa811
KW
3323 }
3324 }
9ae3ac1a 3325
36eaa811 3326 /* Note that non-characters are perfectly legal, so no warning should
8946fcd9 3327 * be given. */
9ae3ac1a
KW
3328 }
3329
8946fcd9
KW
3330 {
3331 unsigned int i;
22197324 3332 const U32 * cp_list;
8946fcd9 3333 U8 * d;
69352d88
KW
3334
3335 /* 'index' is guaranteed to be non-negative, as this is an inversion
3336 * map that covers all possible inputs. See [perl #133365] */
8946fcd9 3337 SSize_t index = _invlist_search(invlist, uv1);
22197324 3338 I32 base = invmap[index];
0134edef 3339
30613bdc
KW
3340 /* The data structures are set up so that if 'base' is non-negative,
3341 * the case change is 1-to-1; and if 0, the change is to itself */
8946fcd9
KW
3342 if (base >= 0) {
3343 IV lc;
b08cf34e 3344
8946fcd9
KW
3345 if (base == 0) {
3346 goto cases_to_self;
4a8240a3 3347 }
4a8240a3 3348
30613bdc 3349 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
8946fcd9
KW
3350 lc = base + uv1 - invlist_array(invlist)[index];
3351 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3352 return lc;
3353 }
1feea2c7 3354
30613bdc
KW
3355 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3356 * requires an auxiliary table look up. abs(base) gives the index into
3357 * a list of such tables which points to the proper aux table. And a
3358 * parallel list gives the length of each corresponding aux table. */
8946fcd9 3359 cp_list = aux_tables[-base];
30613bdc
KW
3360
3361 /* Create the string of UTF-8 from the mapped-to code points */
8946fcd9
KW
3362 d = ustrp;
3363 for (i = 0; i < aux_table_lengths[-base]; i++) {
3364 d = uvchr_to_utf8(d, cp_list[i]);
cbe07460 3365 }
8946fcd9
KW
3366 *d = '\0';
3367 *lenp = d - ustrp;
3368
3369 return cp_list[0];
cbe07460
KW
3370 }
3371
3372 /* Here, there was no mapping defined, which means that the code point maps
3373 * to itself. Return the inputs */
e24dfe9c 3374 cases_to_self:
a13f1de4
KW
3375 if (p) {
3376 len = UTF8SKIP(p);
3377 if (p != ustrp) { /* Don't copy onto itself */
3378 Copy(p, ustrp, len, U8);
3379 }
3380 *lenp = len;
3381 }
3382 else {
3383 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
ca9fab46 3384 }
2a37f04d 3385
f4cd282c 3386 return uv1;
cbe07460 3387
a0ed51b3
LW
3388}
3389
b74fe592 3390Size_t
1b292063 3391Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
40d2776f 3392 const U32 ** remaining_folds_to)
b74fe592
KW
3393{
3394 /* Returns the count of the number of code points that fold to the input
3395 * 'cp' (besides itself).
3396 *
3397 * If the return is 0, there is nothing else that folds to it, and
3398 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3399 *
3400 * If the return is 1, '*first_folds_to' is set to the single code point,
3401 * and '*remaining_folds_to' is set to NULL.
3402 *
3403 * Otherwise, '*first_folds_to' is set to a code point, and
3404 * '*remaining_fold_to' is set to an array that contains the others. The
3405 * length of this array is the returned count minus 1.
3406 *
3407 * The reason for this convolution is to avoid having to deal with
3408 * allocating and freeing memory. The lists are already constructed, so
3409 * the return can point to them, but single code points aren't, so would
1b292063
KW
3410 * need to be constructed if we didn't employ something like this API
3411 *
3412 * The code points returned by this function are all legal Unicode, which
3413 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3414 * constructed with this size (to save space and memory), and we return
3415 * pointers, so they must be this size */
b74fe592 3416
69352d88
KW
3417 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3418 * that covers all possible inputs. See [perl #133365] */
b74fe592 3419 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
40d2776f 3420 I32 base = _Perl_IVCF_invmap[index];
b74fe592
KW
3421
3422 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3423
3424 if (base == 0) { /* No fold */
3425 *first_folds_to = 0;
3426 *remaining_folds_to = NULL;
3427 return 0;
3428 }
3429
3430#ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3431
3432 assert(base > 0);
3433
3434#else
3435
3436 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3437
3438 /* The data structure is set up so that the absolute value of 'base' is
3439 * an index into a table of pointers to arrays, with the array
3440 * corresponding to the index being the list of code points that fold
3441 * to 'cp', and the parallel array containing the length of the list
3442 * array */
3443 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
99f30495
KW
3444 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3445 /* +1 excludes first_folds_to */
b74fe592
KW
3446 return IVCF_AUX_TABLE_lengths[-base];
3447 }
3448
3449#endif
3450
3451 /* Only the single code point. This works like 'fc(G) = G - A + a' */
40d2776f
KW
3452 *first_folds_to = (U32) (base + cp
3453 - invlist_array(PL_utf8_foldclosures)[index]);
b74fe592
KW
3454 *remaining_folds_to = NULL;
3455 return 1;
3456}
3457
051a06d4 3458STATIC UV
56576a04
KW
3459S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3460 U8* const ustrp, STRLEN *lenp)
051a06d4 3461{
4a4088c4 3462 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
3463 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3464 * result contains a character that crosses the 255/256 boundary, disallow
3465 * the change, and return the original code point. See L<perlfunc/lc> for
3466 * why;
051a06d4 3467 *
a1433954
KW
3468 * p points to the original string whose case was changed; assumed
3469 * by this routine to be well-formed
051a06d4 3470 * result the code point of the first character in the changed-case string
56576a04
KW
3471 * ustrp points to the changed-case string (<result> represents its
3472 * first char)
051a06d4
KW
3473 * lenp points to the length of <ustrp> */
3474
3475 UV original; /* To store the first code point of <p> */
3476
3477 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3478
a4f12ed7 3479 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
3480
3481 /* We know immediately if the first character in the string crosses the
5e45c680 3482 * boundary, so can skip testing */
051a06d4
KW
3483 if (result > 255) {
3484
3485 /* Look at every character in the result; if any cross the
3486 * boundary, the whole thing is disallowed */
3487 U8* s = ustrp + UTF8SKIP(ustrp);
3488 U8* e = ustrp + *lenp;
3489 while (s < e) {
a4f12ed7 3490 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
3491 goto bad_crossing;
3492 }
3493 s += UTF8SKIP(s);
3494 }
3495
613abc6d
KW
3496 /* Here, no characters crossed, result is ok as-is, but we warn. */
3497 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
3498 return result;
3499 }
3500
7b52d656 3501 bad_crossing:
051a06d4
KW
3502
3503 /* Failed, have to return the original */
4b88fb76 3504 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
3505
3506 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3507 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
56576a04
KW
3508 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3509 " locale; resolved to \"\\x{%" UVXf "}\".",
357aadde 3510 OP_DESC(PL_op),
ab0b796c
KW
3511 original,
3512 original);
051a06d4
KW
3513 Copy(p, ustrp, *lenp, char);
3514 return original;
3515}
3516
b257a28c
KW
3517STATIC UV
3518S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3519 U8 * ustrp, STRLEN *lenp)
3520{
3521 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3522 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3523 * Otherwise, it returns the first code point of the Turkic foldcased
3524 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3525 * contain *lenp bytes
3526 *
3527 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3528 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3529 * DOTLESS I */
3530
3531 PERL_ARGS_ASSERT_TURKIC_FC;
3532 assert(e > p);
3533
3534 if (UNLIKELY(*p == 'I')) {
3535 *lenp = 2;
3536 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3537 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3538 return LATIN_SMALL_LETTER_DOTLESS_I;
3539 }
3540
3541 if (UNLIKELY(memBEGINs(p, e - p,
3542 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3543 {
3544 *lenp = 1;
3545 *ustrp = 'i';
3546 return 'i';
3547 }
3548
3549 return 0;
3550}
3551
3552STATIC UV
3553S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3554 U8 * ustrp, STRLEN *lenp)
3555{
3556 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3557 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3558 * Otherwise, it returns the first code point of the Turkic lowercased
3559 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3560 * contain *lenp bytes */
3561
3562 PERL_ARGS_ASSERT_TURKIC_LC;
3563 assert(e > p0);
3564
3565 /* A 'I' requires context as to what to do */
3566 if (UNLIKELY(*p0 == 'I')) {
3567 const U8 * p = p0 + 1;
3568
3569 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3570 * modified by a dot above lowercases to 'i' even in turkic locales. */
3571 while (p < e) {
3572 UV cp;
3573
3574 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3575 ustrp[0] = 'i';
3576 *lenp = 1;
3577 return 'i';
3578 }
3579
3580 /* For the dot above to modify the 'I', it must be part of a
3581 * combining sequence immediately following the 'I', and no other
3582 * modifier with a ccc of 230 may intervene */
3583 cp = utf8_to_uvchr_buf(p, e, NULL);
3584 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3585 break;
3586 }
3587
3588 /* Here the combining sequence continues */
3589 p += UTF8SKIP(p);
3590 }
3591 }
3592
3593 /* In all other cases the lc is the same as the fold */
3594 return turkic_fc(p0, e, ustrp, lenp);
3595}
3596
3597STATIC UV
3598S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3599 U8 * ustrp, STRLEN *lenp)
3600{
3601 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3602 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3603 * Otherwise, it returns the first code point of the Turkic upper or
3604 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3605 * ustrp will contain *lenp bytes
3606 *
3607 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
a3815e44 3608 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
b257a28c
KW
3609 * DOTLESS I */
3610
3611 PERL_ARGS_ASSERT_TURKIC_UC;
3612 assert(e > p);
3613
3614 if (*p == 'i') {
3615 *lenp = 2;
3616 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3617 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3618 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3619 }
3620
3621 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3622 *lenp = 1;
3623 *ustrp = 'I';
3624 return 'I';
3625 }
3626
3627 return 0;
3628}
3629
eaf412bf
KW
3630/* The process for changing the case is essentially the same for the four case
3631 * change types, except there are complications for folding. Otherwise the
3632 * difference is only which case to change to. To make sure that they all do
3633 * the same thing, the bodies of the functions are extracted out into the
3634 * following two macros. The functions are written with the same variable
3635 * names, and these are known and used inside these macros. It would be
3636 * better, of course, to have inline functions to do it, but since different
3637 * macros are called, depending on which case is being changed to, this is not
3638 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3639 * function can start with the common start macro, then finish with its special
3640 * handling; while the other three cases can just use the common end macro.
3641 *
3642 * The algorithm is to use the proper (passed in) macro or function to change
3643 * the case for code points that are below 256. The macro is used if using
3644 * locale rules for the case change; the function if not. If the code point is
3645 * above 255, it is computed from the input UTF-8, and another macro is called
3646 * to do the conversion. If necessary, the output is converted to UTF-8. If
3647 * using a locale, we have to check that the change did not cross the 255/256
3648 * boundary, see check_locale_boundary_crossing() for further details.
3649 *
3650 * The macros are split with the correct case change for the below-256 case
3651 * stored into 'result', and in the middle of an else clause for the above-255
3652 * case. At that point in the 'else', 'result' is not the final result, but is
3653 * the input code point calculated from the UTF-8. The fold code needs to
3654 * realize all this and take it from there.
3655 *
b257a28c
KW
3656 * To deal with Turkic locales, the function specified by the parameter
3657 * 'turkic' is called when appropriate.
3658 *
eaf412bf
KW
3659 * If you read the two macros as sequential, it's easier to understand what's
3660 * going on. */
3661#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
b257a28c 3662 L1_func_extra_param, turkic) \
a239b1e2 3663 \
eaf412bf 3664 if (flags & (locale_flags)) { \
8b7358b9 3665 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
eaf412bf 3666 if (IN_UTF8_CTYPE_LOCALE) { \
b257a28c
KW
3667 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3668 UV ret = turkic(p, e, ustrp, lenp); \
3669 if (ret) return ret; \
3670 } \
3671 \
3672 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3673 * all */ \
eaf412bf
KW
3674 flags &= ~(locale_flags); \
3675 } \
eaf412bf
KW
3676 } \
3677 \
3678 if (UTF8_IS_INVARIANT(*p)) { \
3679 if (flags & (locale_flags)) { \
3680 result = LC_L1_change_macro(*p); \
3681 } \
3682 else { \
3683 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3684 } \
3685 } \
a239b1e2 3686 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
1a751160 3687 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
eaf412bf 3688 if (flags & (locale_flags)) { \
1a751160 3689 result = LC_L1_change_macro(c); \
eaf412bf
KW
3690 } \
3691 else { \
1a751160 3692 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
eaf412bf
KW
3693 } \
3694 } \
fa8ab374
KW
3695 else { /* malformed UTF-8 or ord above 255 */ \
3696 STRLEN len_result; \
fa8ab374
KW
3697 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3698 if (len_result == (STRLEN) -1) { \
059703b0 3699 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
fa8ab374 3700 }
eaf412bf
KW
3701
3702#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3703 result = change_macro(result, p, ustrp, lenp); \
3704 \
3705 if (flags & (locale_flags)) { \