This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: A byte count should be Size_t, not I32
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
13f4dd34
A
40 "Use of code point 0x%" UVXf " is not allowed; "
41 "the permissible max is 0x%" UVXf;
760c7c2f 42
114d9c4d 43#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 44
48ef279e 45/*
ccfc67b7 46=head1 Unicode Support
7fefc6c1 47These are various utility functions for manipulating UTF8-encoded
72d33970 48strings. For the uninitiated, this is a method of representing arbitrary
61296642 49Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
50characters in the ASCII range are unmodified, and a zero byte never appears
51within non-zero characters.
166f8a29 52
eaf7a4d2
CS
53=cut
54*/
55
9cbfb8ab
KW
56void
57Perl__force_out_malformed_utf8_message(pTHX_
58 const U8 *const p, /* First byte in UTF-8 sequence */
59 const U8 * const e, /* Final byte in sequence (may include
60 multiple chars */
61 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
62 usually 0, or some DISALLOW flags */
63 const bool die_here) /* If TRUE, this function does not return */
64{
65 /* This core-only function is to be called when a malformed UTF-8 character
66 * is found, in order to output the detailed information about the
67 * malformation before dieing. The reason it exists is for the occasions
68 * when such a malformation is fatal, but warnings might be turned off, so
69 * that normally they would not be actually output. This ensures that they
70 * do get output. Because a sequence may be malformed in more than one
71 * way, multiple messages may be generated, so we can't make them fatal, as
72 * that would cause the first one to die.
73 *
74 * Instead we pretend -W was passed to perl, then die afterwards. The
75 * flexibility is here to return to the caller so they can finish up and
76 * die themselves */
77 U32 errors;
78
79 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
80
81 ENTER;
c15a80f3 82 SAVEI8(PL_dowarn);
9cbfb8ab
KW
83 SAVESPTR(PL_curcop);
84
85 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
86 if (PL_curcop) {
87 PL_curcop->cop_warnings = pWARN_ALL;
88 }
89
90 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
91
92 LEAVE;
93
94 if (! errors) {
95 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
96 " be called only when there are errors found");
97 }
98
99 if (die_here) {
100 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
101 }
102}
103
eaf7a4d2 104/*
378516de 105=for apidoc uvoffuni_to_utf8_flags
eebe1485 106
a27992cc 107THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
108Instead, B<Almost all code should use L</uvchr_to_utf8> or
109L</uvchr_to_utf8_flags>>.
a27992cc 110
de69f3af
KW
111This function is like them, but the input is a strict Unicode
112(as opposed to native) code point. Only in very rare circumstances should code
113not be using the native code point.
949cf498 114
efa9cd84 115For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 116
eebe1485
SC
117=cut
118*/
119
8ee1cdcb
KW
120#define HANDLE_UNICODE_SURROGATE(uv, flags) \
121 STMT_START { \
122 if (flags & UNICODE_WARN_SURROGATE) { \
123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
147e3846 124 "UTF-16 surrogate U+%04" UVXf, uv); \
8ee1cdcb
KW
125 } \
126 if (flags & UNICODE_DISALLOW_SURROGATE) { \
127 return NULL; \
128 } \
129 } STMT_END;
130
131#define HANDLE_UNICODE_NONCHAR(uv, flags) \
132 STMT_START { \
133 if (flags & UNICODE_WARN_NONCHAR) { \
134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
147e3846 135 "Unicode non-character U+%04" UVXf " is not " \
8ee1cdcb
KW
136 "recommended for open interchange", uv); \
137 } \
138 if (flags & UNICODE_DISALLOW_NONCHAR) { \
139 return NULL; \
140 } \
141 } STMT_END;
142
ba6ed43c
KW
143/* Use shorter names internally in this file */
144#define SHIFT UTF_ACCUMULATION_SHIFT
145#undef MARK
146#define MARK UTF_CONTINUATION_MARK
147#define MASK UTF_CONTINUATION_MASK
148
dfe13c55 149U8 *
4b31b634 150Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
a0ed51b3 151{
378516de 152 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 153
2d1545e5 154 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 155 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
156 return d;
157 }
facc1dc2 158
3ea68d71 159 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
160 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
161 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
162 return d;
163 }
d9432125 164
ba6ed43c
KW
165 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
166 * below, the 16 is for start bytes E0-EF (which are all the possible ones
167 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
168 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
169 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
170 * 0x800-0xFFFF on ASCII */
171 if (uv < (16 * (1U << (2 * SHIFT)))) {
172 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
173 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
174 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
175
176#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
177 aren't tested here */
178 /* The most likely code points in this range are below the surrogates.
179 * Do an extra test to quickly exclude those. */
180 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
181 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
182 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
183 {
8ee1cdcb
KW
184 HANDLE_UNICODE_NONCHAR(uv, flags);
185 }
186 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
187 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 188 }
ba6ed43c
KW
189 }
190#endif
191 return d;
192 }
193
194 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
195 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
196 * happen starting with 4-byte characters on ASCII platforms. We unify the
197 * code for these with EBCDIC, even though some of them require 5-bytes on
198 * those, because khw believes the code saving is worth the very slight
199 * performance hit on these high EBCDIC code points. */
200
201 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
13f4dd34
A
202 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP)) {
203 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
a5bf80e0
KW
204 }
205 if ( (flags & UNICODE_WARN_SUPER)
206 || ( UNICODE_IS_ABOVE_31_BIT(uv)
207 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
208 {
209 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
210
211 /* Choose the more dire applicable warning */
212 (UNICODE_IS_ABOVE_31_BIT(uv))
147e3846
KW
213 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
214 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
a5bf80e0
KW
215 uv);
216 }
217 if (flags & UNICODE_DISALLOW_SUPER
218 || ( UNICODE_IS_ABOVE_31_BIT(uv)
219 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
220 {
221 return NULL;
222 }
223 }
ba6ed43c
KW
224 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
225 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 226 }
d9432125 227
ba6ed43c
KW
228 /* Test for and handle 4-byte result. In the test immediately below, the
229 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
230 * characters). The 3 is for 3 continuation bytes; these each contribute
231 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
232 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
233 * 0x1_0000-0x1F_FFFF on ASCII */
234 if (uv < (8 * (1U << (3 * SHIFT)))) {
235 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
236 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
237 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
238 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
239
240#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
241 characters. The end-plane non-characters for EBCDIC were
242 handled just above */
243 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
244 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 245 }
ba6ed43c
KW
246 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
247 HANDLE_UNICODE_SURROGATE(uv, flags);
248 }
249#endif
250
251 return d;
252 }
253
254 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
255 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
256 * format. The unrolled version above turns out to not save all that much
257 * time, and at these high code points (well above the legal Unicode range
258 * on ASCII platforms, and well above anything in common use in EBCDIC),
259 * khw believes that less code outweighs slight performance gains. */
260
d9432125 261 {
5aaebcb3 262 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
263 U8 *p = d+len-1;
264 while (p > d) {
4c8cd605 265 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
266 uv >>= UTF_ACCUMULATION_SHIFT;
267 }
4c8cd605 268 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
269 return d+len;
270 }
a0ed51b3 271}
a5bf80e0 272
646ca15d 273/*
07693fe6
KW
274=for apidoc uvchr_to_utf8
275
bcb1a2d4 276Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 277of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
278C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
279the byte after the end of the new character. In other words,
07693fe6
KW
280
281 d = uvchr_to_utf8(d, uv);
282
283is the recommended wide native character-aware way of saying
284
285 *(d++) = uv;
286
760c7c2f
KW
287This function accepts any UV as input, but very high code points (above
288C<IV_MAX> on the platform) will raise a deprecation warning. This is
289typically 0x7FFF_FFFF in a 32-bit word.
290
291It is possible to forbid or warn on non-Unicode code points, or those that may
292be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 293
07693fe6
KW
294=cut
295*/
296
de69f3af
KW
297/* This is also a macro */
298PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
299
07693fe6
KW
300U8 *
301Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
302{
de69f3af 303 return uvchr_to_utf8(d, uv);
07693fe6
KW
304}
305
de69f3af
KW
306/*
307=for apidoc uvchr_to_utf8_flags
308
309Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 310of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
311C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
312the byte after the end of the new character. In other words,
de69f3af
KW
313
314 d = uvchr_to_utf8_flags(d, uv, flags);
315
316or, in most cases,
317
318 d = uvchr_to_utf8_flags(d, uv, 0);
319
320This is the Unicode-aware way of saying
321
322 *(d++) = uv;
323
760c7c2f
KW
324If C<flags> is 0, this function accepts any UV as input, but very high code
325points (above C<IV_MAX> for the platform) will raise a deprecation warning.
326This is typically 0x7FFF_FFFF in a 32-bit word.
327
328Specifying C<flags> can further restrict what is allowed and not warned on, as
329follows:
de69f3af 330
796b6530 331If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
332the function will raise a warning, provided UTF8 warnings are enabled. If
333instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
334NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 335
760c7c2f
KW
336Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
337affect how the function handles a Unicode non-character.
93e6dbd6 338
760c7c2f
KW
339And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
340affect the handling of code points that are above the Unicode maximum of
3410x10FFFF. Languages other than Perl may not be able to accept files that
342contain these.
93e6dbd6
KW
343
344The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
345the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
346three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
347allowed inputs to the strict UTF-8 traditionally defined by Unicode.
348Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
349C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
350above-Unicode and surrogate flags, but not the non-character ones, as
351defined in
352L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
353See L<perlunicode/Noncharacter code points>.
93e6dbd6 354
ab8e6d41
KW
355Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
356so using them is more problematic than other above-Unicode code points. Perl
357invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
358likely that non-Perl languages will not be able to read files that contain
359these that written by the perl interpreter; nor would Perl understand files
360written by something that uses a different extension. For these reasons, there
361is a separate set of flags that can warn and/or disallow these extremely high
362code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
363C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
364are entirely independent from the deprecation warning for code points above
365C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
366code point that needs more than 31 bits to represent. When that happens,
367effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
36832-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
369above-Unicode code points, including these, as malformations; and
370C<UNICODE_WARN_SUPER> warns on these.)
371
372On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
373extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
374than on ASCII. Prior to that, code points 2**31 and higher were simply
375unrepresentable, and a different, incompatible method was used to represent
376code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
377and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
378platforms, warning and disallowing 2**31 and higher.
de69f3af 379
de69f3af
KW
380=cut
381*/
382
383/* This is also a macro */
384PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
385
07693fe6
KW
386U8 *
387Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
388{
de69f3af 389 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
390}
391
83dc0f42
KW
392PERL_STATIC_INLINE bool
393S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
394{
395 /* Returns TRUE if the first code point represented by the Perl-extended-
396 * UTF-8-encoded string starting at 's', and looking no further than 'e -
397 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
398 *
399 * The function handles the case where the input bytes do not include all
400 * the ones necessary to represent a full character. That is, they may be
401 * the intial bytes of the representation of a code point, but possibly
402 * the final ones necessary for the complete representation may be beyond
403 * 'e - 1'.
404 *
405 * The function assumes that the sequence is well-formed UTF-8 as far as it
406 * goes, and is for a UTF-8 variant code point. If the sequence is
407 * incomplete, the function returns FALSE if there is any well-formed
408 * UTF-8 byte sequence that can complete it in such a way that a code point
409 * < 2**31 is produced; otherwise it returns TRUE.
410 *
411 * Getting this exactly right is slightly tricky, and has to be done in
412 * several places in this file, so is centralized here. It is based on the
413 * following table:
414 *
415 * U+7FFFFFFF (2 ** 31 - 1)
416 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
417 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
418 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
419 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
420 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
421 * U+80000000 (2 ** 31):
422 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
423 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
424 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
425 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
426 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
428 */
429
430#ifdef EBCDIC
431
37086697
KW
432 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
433 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
434 const STRLEN prefix_len = sizeof(prefix) - 1;
435 const STRLEN len = e - s;
f880f78a 436 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
437
438#else
439
440 PERL_UNUSED_ARG(e);
441
442#endif
443
444 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
445
446 assert(! UTF8_IS_INVARIANT(*s));
447
448#ifndef EBCDIC
449
450 /* Technically, a start byte of FE can be for a code point that fits into
451 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
452 * malformation. */
453 return (*s >= 0xFE);
454
455#else
456
457 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
458 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
459 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
460 * bits. */
461 if (*s != 0xFE || len == 1) {
462 return FALSE;
463 }
464
465 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
466 * \x41 and \x42. */
467 return cBOOL(memGT(s + 1, prefix, cmp_len));
468
469#endif
470
471}
472
12a4bed3
KW
473PERL_STATIC_INLINE bool
474S_does_utf8_overflow(const U8 * const s, const U8 * e)
475{
476 const U8 *x;
477 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
478
c551bb01
KW
479#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
480
481 const STRLEN len = e - s;
482
483#endif
484
12a4bed3
KW
485 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
486 * platform, that is if it represents a code point larger than the highest
487 * representable code point. (For ASCII platforms, we could use memcmp()
488 * because we don't have to convert each byte to I8, but it's very rare
489 * input indeed that would approach overflow, so the loop below will likely
490 * only get executed once.
491 *
492 * 'e' must not be beyond a full character. If it is less than a full
493 * character, the function returns FALSE if there is any input beyond 'e'
494 * that could result in a non-overflowing code point */
495
496 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
497 assert(s <= e && s + UTF8SKIP(s) >= e);
498
499#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
500
501 /* On 32 bit ASCII machines, many overlongs that start with FF don't
502 * overflow */
503
c551bb01 504 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
505 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
506 return memGE(s, max_32_bit_overlong,
c551bb01 507 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
508 }
509
510#endif
12a4bed3
KW
511
512 for (x = s; x < e; x++, y++) {
513
514 /* If this byte is larger than the corresponding highest UTF-8 byte, it
515 * overflows */
516 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
517 return TRUE;
518 }
519
520 /* If not the same as this byte, it must be smaller, doesn't overflow */
521 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
522 return FALSE;
523 }
524 }
525
526 /* Got to the end and all bytes are the same. If the input is a whole
527 * character, it doesn't overflow. And if it is a partial character,
528 * there's not enough information to tell, so assume doesn't overflow */
529 return FALSE;
530}
531
532PERL_STATIC_INLINE bool
533S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
534{
535 /* Overlongs can occur whenever the number of continuation bytes
536 * changes. That means whenever the number of leading 1 bits in a start
537 * byte increases from the next lower start byte. That happens for start
538 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
539 * illegal start bytes have already been excluded, so don't need to be
540 * tested here;
541 * ASCII platforms: C0, C1
542 * EBCDIC platforms C0, C1, C2, C3, C4, E0
543 *
544 * At least a second byte is required to determine if other sequences will
545 * be an overlong. */
546
547 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
548 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
549
550 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
551 assert(len > 1 && UTF8_IS_START(*s));
552
553 /* Each platform has overlongs after the start bytes given above (expressed
554 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
555 * the logic is the same, except the E0 overlong has already been excluded
556 * on EBCDIC platforms. The values below were found by manually
557 * inspecting the UTF-8 patterns. See the tables in utf8.h and
558 * utfebcdic.h. */
559
560# ifdef EBCDIC
561# define F0_ABOVE_OVERLONG 0xB0
562# define F8_ABOVE_OVERLONG 0xA8
563# define FC_ABOVE_OVERLONG 0xA4
564# define FE_ABOVE_OVERLONG 0xA2
565# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
566 /* I8(0xfe) is FF */
567# else
568
569 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
570 return TRUE;
571 }
572
573# define F0_ABOVE_OVERLONG 0x90
574# define F8_ABOVE_OVERLONG 0x88
575# define FC_ABOVE_OVERLONG 0x84
576# define FE_ABOVE_OVERLONG 0x82
577# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
578# endif
579
580
581 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
582 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
583 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
584 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
585 {
586 return TRUE;
587 }
588
b0b342d4
KW
589 /* Check for the FF overlong */
590 return isFF_OVERLONG(s, len);
591}
592
593PERL_STATIC_INLINE bool
594S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
595{
596 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
597
598 /* Check for the FF overlong. This happens only if all these bytes match;
599 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 600 * utfebcdic.h. */
12a4bed3 601
b0b342d4
KW
602 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
603 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
604 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
605}
606
607#undef F0_ABOVE_OVERLONG
608#undef F8_ABOVE_OVERLONG
609#undef FC_ABOVE_OVERLONG
610#undef FE_ABOVE_OVERLONG
611#undef FF_OVERLONG_PREFIX
612
35f8c9bd 613STRLEN
edc2c47a 614Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 615{
2b479609 616 STRLEN len;
12a4bed3 617 const U8 *x;
35f8c9bd 618
2b479609
KW
619 /* A helper function that should not be called directly.
620 *
621 * This function returns non-zero if the string beginning at 's' and
622 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
623 * code point; otherwise it returns 0. The examination stops after the
624 * first code point in 's' is validated, not looking at the rest of the
625 * input. If 'e' is such that there are not enough bytes to represent a
626 * complete code point, this function will return non-zero anyway, if the
627 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
628 * excluded by 'flags'.
629 *
630 * A non-zero return gives the number of bytes required to represent the
631 * code point. Be aware that if the input is for a partial character, the
632 * return will be larger than 'e - s'.
633 *
634 * This function assumes that the code point represented is UTF-8 variant.
635 * The caller should have excluded this possibility before calling this
636 * function.
637 *
638 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
639 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
640 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
641 * disallowed by the flags. If the input is only for a partial character,
642 * the function will return non-zero if there is any sequence of
643 * well-formed UTF-8 that, when appended to the input sequence, could
644 * result in an allowed code point; otherwise it returns 0. Non characters
645 * cannot be determined based on partial character input. But many of the
646 * other excluded types can be determined with just the first one or two
647 * bytes.
648 *
649 */
650
651 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
652
653 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
654 |UTF8_DISALLOW_ABOVE_31_BIT)));
655 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 656
2b479609 657 /* A variant char must begin with a start byte */
35f8c9bd
KW
658 if (UNLIKELY(! UTF8_IS_START(*s))) {
659 return 0;
660 }
661
edc2c47a
KW
662 /* Examine a maximum of a single whole code point */
663 if (e - s > UTF8SKIP(s)) {
664 e = s + UTF8SKIP(s);
665 }
666
2b479609
KW
667 len = e - s;
668
669 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
670 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 671
2b479609
KW
672 /* The code below is derived from this table. Keep in mind that legal
673 * continuation bytes range between \x80..\xBF for UTF-8, and
674 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
675 * Hence, we don't have to test the upper edge because if any of those
676 * are encountered, the sequence is malformed, and will fail elsewhere
677 * in this function.
678 * UTF-8 UTF-EBCDIC I8
679 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
680 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
681 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
682 *
683 */
684
685#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
686# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 687# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 688
19794540
KW
689# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
690 /* B6 and B7 */ \
691 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
692#else
693# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
694# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
695# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
696#endif
697
698 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
699 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
700 {
2b479609
KW
701 return 0; /* Above Unicode */
702 }
703
704 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
705 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
706 {
707 return 0; /* Above 31 bits */
708 }
709
710 if (len > 1) {
711 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
712
713 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 714 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
715 {
716 return 0; /* Above Unicode */
717 }
718
719 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 720 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
721 {
722 return 0; /* Surrogate */
723 }
724
725 if ( (flags & UTF8_DISALLOW_NONCHAR)
726 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
727 {
728 return 0; /* Noncharacter code point */
729 }
730 }
731 }
732
733 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
734 for (x = s + 1; x < e; x++) {
735 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
736 return 0;
737 }
738 }
739
af13dd8a 740 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
741 * overlong. */
742 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
743 return 0;
af13dd8a
KW
744 }
745
12a4bed3
KW
746 /* And finally, that the code point represented fits in a word on this
747 * platform */
748 if (does_utf8_overflow(s, e)) {
749 return 0;
35f8c9bd
KW
750 }
751
2b479609 752 return UTF8SKIP(s);
35f8c9bd
KW
753}
754
7e2f38b2
KW
755char *
756Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
7cf8d05d
KW
757{
758 /* Returns a mortalized C string that is a displayable copy of the 'len'
7e2f38b2
KW
759 * bytes starting at 's'. 'format' gives how to display each byte.
760 * Currently, there are only two formats, so it is currently a bool:
761 * 0 \xab
762 * 1 ab (that is a space between two hex digit bytes)
763 */
7cf8d05d
KW
764
765 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
766 trailing NUL */
767 const U8 * const e = s + len;
768 char * output;
769 char * d;
770
771 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
772
773 Newx(output, output_len, char);
774 SAVEFREEPV(output);
775
776 d = output;
777 for (; s < e; s++) {
778 const unsigned high_nibble = (*s & 0xF0) >> 4;
779 const unsigned low_nibble = (*s & 0x0F);
780
7e2f38b2
KW
781 if (format) {
782 *d++ = ' ';
783 }
784 else {
785 *d++ = '\\';
786 *d++ = 'x';
787 }
7cf8d05d
KW
788
789 if (high_nibble < 10) {
790 *d++ = high_nibble + '0';
791 }
792 else {
793 *d++ = high_nibble - 10 + 'a';
794 }
795
796 if (low_nibble < 10) {
797 *d++ = low_nibble + '0';
798 }
799 else {
800 *d++ = low_nibble - 10 + 'a';
801 }
802 }
803
804 *d = '\0';
805 return output;
806}
807
806547a7 808PERL_STATIC_INLINE char *
7cf8d05d
KW
809S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
810
811 /* How many bytes to print */
3cc6a05e 812 STRLEN print_len,
7cf8d05d
KW
813
814 /* Which one is the non-continuation */
815 const STRLEN non_cont_byte_pos,
816
817 /* How many bytes should there be? */
818 const STRLEN expect_len)
806547a7
KW
819{
820 /* Return the malformation warning text for an unexpected continuation
821 * byte. */
822
7cf8d05d 823 const char * const where = (non_cont_byte_pos == 1)
806547a7 824 ? "immediately"
7cf8d05d
KW
825 : Perl_form(aTHX_ "%d bytes",
826 (int) non_cont_byte_pos);
806547a7
KW
827
828 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
829
7cf8d05d
KW
830 /* We don't need to pass this parameter, but since it has already been
831 * calculated, it's likely faster to pass it; verify under DEBUGGING */
832 assert(expect_len == UTF8SKIP(s));
833
834 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
835 " %s after start byte 0x%02x; need %d bytes, got %d)",
836 malformed_text,
7e2f38b2 837 _byte_dump_string(s, print_len, 0),
7cf8d05d
KW
838 *(s + non_cont_byte_pos),
839 where,
840 *s,
841 (int) expect_len,
842 (int) non_cont_byte_pos);
806547a7
KW
843}
844
35f8c9bd
KW
845/*
846
de69f3af 847=for apidoc utf8n_to_uvchr
378516de
KW
848
849THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 850Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 851
9041c2e3 852Bottom level UTF-8 decode routine.
de69f3af 853Returns the native code point value of the first character in the string C<s>,
746afd53
KW
854which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
855C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
856the length, in bytes, of that character.
949cf498
KW
857
858The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
859well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
860causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
861is the next possible position in C<s> that could begin a non-malformed
862character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
863is raised. Some UTF-8 input sequences may contain multiple malformations.
864This function tries to find every possible one in each call, so multiple
865warnings can be raised for each sequence.
949cf498
KW
866
867Various ALLOW flags can be set in C<flags> to allow (and not warn on)
868individual types of malformations, such as the sequence being overlong (that
869is, when there is a shorter sequence that can express the same code point;
870overlong sequences are expressly forbidden in the UTF-8 standard due to
871potential security issues). Another malformation example is the first byte of
872a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
873flags. Even if allowed, this function generally returns the Unicode
874REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
875F<utf8.h> to override this behavior for the overlong malformations, but don't
876do that except for very specialized purposes.
949cf498 877
796b6530 878The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
879flags) malformation is found. If this flag is set, the routine assumes that
880the caller will raise a warning, and this function will silently just set
d088425d
KW
881C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
882
75200dff 883Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 884character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
885in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
886be set to 1. To disambiguate, upon a zero return, see if the first byte of
887C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 888error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
889
890Certain code points are considered problematic. These are Unicode surrogates,
746afd53 891Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 892By default these are considered regular code points, but certain situations
ecc1615f
KW
893warrant special handling for them, which can be specified using the C<flags>
894parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
895three classes are treated as malformations and handled as such. The flags
896C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
897C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
898disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
899restricts the allowed inputs to the strict UTF-8 traditionally defined by
900Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
901definition given by
902L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
903The difference between traditional strictness and C9 strictness is that the
904latter does not forbid non-character code points. (They are still discouraged,
905however.) For more discussion see L<perlunicode/Noncharacter code points>.
906
907The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
908C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
909C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
910raised for their respective categories, but otherwise the code points are
911considered valid (not malformations). To get a category to both be treated as
912a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 913(But note that warnings are not raised if lexically disabled nor if
796b6530 914C<UTF8_CHECK_ONLY> is also specified.)
949cf498 915
760c7c2f
KW
916It is now deprecated to have very high code points (above C<IV_MAX> on the
917platforms) and this function will raise a deprecation warning for these (unless
d5944cab 918such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 919in a 32-bit word.
ab8e6d41
KW
920
921Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
922so using them is more problematic than other above-Unicode code points. Perl
923invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
924likely that non-Perl languages will not be able to read files that contain
aff2be59 925these; nor would Perl understand files
ab8e6d41
KW
926written by something that uses a different extension. For these reasons, there
927is a separate set of flags that can warn and/or disallow these extremely high
928code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
929C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
930are entirely independent from the deprecation warning for code points above
931C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
932code point that needs more than 31 bits to represent. When that happens,
933effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
93432-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
935above-Unicode code points, including these, as malformations; and
936C<UTF8_WARN_SUPER> warns on these.)
937
938On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
939extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
940than on ASCII. Prior to that, code points 2**31 and higher were simply
941unrepresentable, and a different, incompatible method was used to represent
942code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
943and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
944platforms, warning and disallowing 2**31 and higher.
949cf498
KW
945
946All other code points corresponding to Unicode characters, including private
947use and those yet to be assigned, are never considered malformed and never
948warn.
67e989fb 949
37607a96 950=cut
f9380377
KW
951
952Also implemented as a macro in utf8.h
953*/
954
955UV
956Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
957 STRLEN curlen,
958 STRLEN *retlen,
959 const U32 flags)
960{
961 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
962
963 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
964}
965
966/*
967
968=for apidoc utf8n_to_uvchr_error
969
970THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
971Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
972
973This function is for code that needs to know what the precise malformation(s)
974are when an error is found.
975
976It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
977all the others, C<errors>. If this parameter is 0, this function behaves
978identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
979to a C<U32> variable, which this function sets to indicate any errors found.
980Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
981C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
982of these bits will be set if a malformation is found, even if the input
7a65503b 983C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
984exceptions are noted:
985
986=over 4
987
988=item C<UTF8_GOT_ABOVE_31_BIT>
989
990The code point represented by the input UTF-8 sequence occupies more than 31
991bits.
992This bit is set only if the input C<flags> parameter contains either the
993C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
994
995=item C<UTF8_GOT_CONTINUATION>
996
997The input sequence was malformed in that the first byte was a a UTF-8
998continuation byte.
999
1000=item C<UTF8_GOT_EMPTY>
1001
1002The input C<curlen> parameter was 0.
1003
1004=item C<UTF8_GOT_LONG>
1005
1006The input sequence was malformed in that there is some other sequence that
1007evaluates to the same code point, but that sequence is shorter than this one.
1008
1009=item C<UTF8_GOT_NONCHAR>
1010
1011The code point represented by the input UTF-8 sequence is for a Unicode
1012non-character code point.
1013This bit is set only if the input C<flags> parameter contains either the
1014C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1015
1016=item C<UTF8_GOT_NON_CONTINUATION>
1017
1018The input sequence was malformed in that a non-continuation type byte was found
1019in a position where only a continuation type one should be.
1020
1021=item C<UTF8_GOT_OVERFLOW>
1022
1023The input sequence was malformed in that it is for a code point that is not
1024representable in the number of bits available in a UV on the current platform.
1025
1026=item C<UTF8_GOT_SHORT>
1027
1028The input sequence was malformed in that C<curlen> is smaller than required for
1029a complete sequence. In other words, the input is for a partial character
1030sequence.
1031
1032=item C<UTF8_GOT_SUPER>
1033
1034The input sequence was malformed in that it is for a non-Unicode code point;
1035that is, one above the legal Unicode maximum.
1036This bit is set only if the input C<flags> parameter contains either the
1037C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1038
1039=item C<UTF8_GOT_SURROGATE>
1040
1041The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1042code point.
1043This bit is set only if the input C<flags> parameter contains either the
1044C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1045
1046=back
1047
133551d8
KW
1048To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1049flag to suppress any warnings, and then examine the C<*errors> return.
1050
f9380377 1051=cut
37607a96 1052*/
67e989fb 1053
a0ed51b3 1054UV
f9380377
KW
1055Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1056 STRLEN curlen,
1057 STRLEN *retlen,
1058 const U32 flags,
1059 U32 * errors)
a0ed51b3 1060{
d4c19fe8 1061 const U8 * const s0 = s;
2b5e7bc2
KW
1062 U8 * send = NULL; /* (initialized to silence compilers' wrong
1063 warning) */
1064 U32 possible_problems = 0; /* A bit is set here for each potential problem
1065 found as we go along */
eb83ed87 1066 UV uv = *s;
2b5e7bc2
KW
1067 STRLEN expectlen = 0; /* How long should this sequence be?
1068 (initialized to silence compilers' wrong
1069 warning) */
e308b348 1070 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1071 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1072 this gets set and discarded */
a0dbb045 1073
2b5e7bc2
KW
1074 /* The below are used only if there is both an overlong malformation and a
1075 * too short one. Otherwise the first two are set to 's0' and 'send', and
1076 * the third not used at all */
1077 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1078 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1079 warning) */
e9f2c446
KW
1080 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1081 routine; see [perl #130921] */
2b5e7bc2 1082 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1083
f9380377
KW
1084 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1085
1086 if (errors) {
1087 *errors = 0;
1088 }
1089 else {
1090 errors = &discard_errors;
1091 }
a0dbb045 1092
eb83ed87
KW
1093 /* The order of malformation tests here is important. We should consume as
1094 * few bytes as possible in order to not skip any valid character. This is
1095 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1096 * http://unicode.org/reports/tr36 for more discussion as to why. For
1097 * example, once we've done a UTF8SKIP, we can tell the expected number of
1098 * bytes, and could fail right off the bat if the input parameters indicate
1099 * that there are too few available. But it could be that just that first
1100 * byte is garbled, and the intended character occupies fewer bytes. If we
1101 * blindly assumed that the first byte is correct, and skipped based on
1102 * that number, we could skip over a valid input character. So instead, we
1103 * always examine the sequence byte-by-byte.
1104 *
1105 * We also should not consume too few bytes, otherwise someone could inject
1106 * things. For example, an input could be deliberately designed to
1107 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1108 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1109 * which is actually part of of the overflowing sequence), that could look
1110 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1111 * sequence and process the rest, inappropriately.
1112 *
1113 * Some possible input sequences are malformed in more than one way. This
1114 * function goes to lengths to try to find all of them. This is necessary
1115 * for correctness, as the inputs may allow one malformation but not
1116 * another, and if we abandon searching for others after finding the
1117 * allowed one, we could allow in something that shouldn't have been.
1118 */
eb83ed87 1119
b5b9af04 1120 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1121 possible_problems |= UTF8_GOT_EMPTY;
1122 curlen = 0;
5a48568d 1123 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1124 goto ready_to_handle_errors;
0c443dc2
JH
1125 }
1126
eb83ed87
KW
1127 expectlen = UTF8SKIP(s);
1128
1129 /* A well-formed UTF-8 character, as the vast majority of calls to this
1130 * function will be for, has this expected length. For efficiency, set
1131 * things up here to return it. It will be overriden only in those rare
1132 * cases where a malformation is found */
1133 if (retlen) {
1134 *retlen = expectlen;
1135 }
1136
1137 /* An invariant is trivially well-formed */
1d72bdf6 1138 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1139 return uv;
a0ed51b3 1140 }
67e989fb 1141
eb83ed87 1142 /* A continuation character can't start a valid sequence */
b5b9af04 1143 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1144 possible_problems |= UTF8_GOT_CONTINUATION;
1145 curlen = 1;
1146 uv = UNICODE_REPLACEMENT;
1147 goto ready_to_handle_errors;
ba210ebe 1148 }
9041c2e3 1149
dcd27b3c 1150 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1151 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1152 * because it excludes start bytes like \xC0 that always lead to
1153 * overlongs.) */
dcd27b3c 1154
534752c1
KW
1155 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1156 * that indicate the number of bytes in the character's whole UTF-8
1157 * sequence, leaving just the bits that are part of the value. */
1158 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1159
e308b348
KW
1160 /* Setup the loop end point, making sure to not look past the end of the
1161 * input string, and flag it as too short if the size isn't big enough. */
1162 send = (U8*) s0;
1163 if (UNLIKELY(curlen < expectlen)) {
1164 possible_problems |= UTF8_GOT_SHORT;
1165 avail_len = curlen;
1166 send += curlen;
1167 }
1168 else {
1169 send += expectlen;
1170 }
1171 adjusted_send = send;
1172
eb83ed87 1173 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1174 * accumulating each into the working value as we go. */
eb83ed87 1175 for (s = s0 + 1; s < send; s++) {
b5b9af04 1176 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1177 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1178 continue;
1179 }
1180
1181 /* Here, found a non-continuation before processing all expected bytes.
1182 * This byte indicates the beginning of a new character, so quit, even
1183 * if allowing this malformation. */
2b5e7bc2 1184 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1185 break;
eb83ed87
KW
1186 } /* End of loop through the character's bytes */
1187
1188 /* Save how many bytes were actually in the character */
1189 curlen = s - s0;
1190
2b5e7bc2
KW
1191 /* Note that there are two types of too-short malformation. One is when
1192 * there is actual wrong data before the normal termination of the
1193 * sequence. The other is that the sequence wasn't complete before the end
1194 * of the data we are allowed to look at, based on the input 'curlen'.
1195 * This means that we were passed data for a partial character, but it is
1196 * valid as far as we saw. The other is definitely invalid. This
1197 * distinction could be important to a caller, so the two types are kept
15b010f0
KW
1198 * separate.
1199 *
1200 * A convenience macro that matches either of the too-short conditions. */
1201# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1202
1203 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1204 uv_so_far = uv;
1205 uv = UNICODE_REPLACEMENT;
1206 }
2b5e7bc2
KW
1207
1208 /* Check for overflow */
1209 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1210 possible_problems |= UTF8_GOT_OVERFLOW;
1211 uv = UNICODE_REPLACEMENT;
eb83ed87 1212 }
eb83ed87 1213
2b5e7bc2
KW
1214 /* Check for overlong. If no problems so far, 'uv' is the correct code
1215 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1216 * we must look at the UTF-8 byte sequence itself to see if it is for an
1217 * overlong */
1218 if ( ( LIKELY(! possible_problems)
1219 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1220 || ( UNLIKELY( possible_problems)
1221 && ( UNLIKELY(! UTF8_IS_START(*s0))
1222 || ( curlen > 1
1223 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1224 send - s0))))))
2f8f112e 1225 {
2b5e7bc2
KW
1226 possible_problems |= UTF8_GOT_LONG;
1227
2b5e7bc2
KW
1228 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1229 UV min_uv = uv_so_far;
1230 STRLEN i;
1231
1232 /* Here, the input is both overlong and is missing some trailing
1233 * bytes. There is no single code point it could be for, but there
1234 * may be enough information present to determine if what we have
1235 * so far is for an unallowed code point, such as for a surrogate.
1236 * The code below has the intelligence to determine this, but just
1237 * for non-overlong UTF-8 sequences. What we do here is calculate
1238 * the smallest code point the input could represent if there were
1239 * no too short malformation. Then we compute and save the UTF-8
1240 * for that, which is what the code below looks at instead of the
1241 * raw input. It turns out that the smallest such code point is
1242 * all we need. */
1243 for (i = curlen; i < expectlen; i++) {
1244 min_uv = UTF8_ACCUMULATE(min_uv,
1245 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1246 }
1247
e9f2c446 1248 adjusted_s0 = temp_char_buf;
2b5e7bc2
KW
1249 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1250 }
eb83ed87
KW
1251 }
1252
2b5e7bc2
KW
1253 /* Now check that the input isn't for a problematic code point not allowed
1254 * by the input parameters. */
1255 /* isn't problematic if < this */
1256 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1257 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1258
1259 /* if overflow, we know without looking further
1260 * precisely which of the problematic types it is,
1261 * and we deal with those in the overflow handling
1262 * code */
1263 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1264 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1265 && ((flags & ( UTF8_DISALLOW_NONCHAR
1266 |UTF8_DISALLOW_SURROGATE
1267 |UTF8_DISALLOW_SUPER
1268 |UTF8_DISALLOW_ABOVE_31_BIT
1269 |UTF8_WARN_NONCHAR
1270 |UTF8_WARN_SURROGATE
1271 |UTF8_WARN_SUPER
1272 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1273 /* In case of a malformation, 'uv' is not valid, and has
1274 * been changed to something in the Unicode range.
1275 * Currently we don't output a deprecation message if there
1276 * is already a malformation, so we don't have to special
1277 * case the test immediately below */
760c7c2f
KW
1278 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1279 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1280 {
2b5e7bc2
KW
1281 /* If there were no malformations, or the only malformation is an
1282 * overlong, 'uv' is valid */
1283 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1284 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1285 possible_problems |= UTF8_GOT_SURROGATE;
1286 }
1287 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1288 possible_problems |= UTF8_GOT_SUPER;
1289 }
1290 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1291 possible_problems |= UTF8_GOT_NONCHAR;
1292 }
1293 }
1294 else { /* Otherwise, need to look at the source UTF-8, possibly
1295 adjusted to be non-overlong */
1296
1297 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1298 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1299 {
2b5e7bc2
KW
1300 possible_problems |= UTF8_GOT_SUPER;
1301 }
1302 else if (curlen > 1) {
1303 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1304 NATIVE_UTF8_TO_I8(*adjusted_s0),
1305 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1306 {
2b5e7bc2 1307 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1308 }
2b5e7bc2
KW
1309 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1310 NATIVE_UTF8_TO_I8(*adjusted_s0),
1311 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1312 {
1313 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1314 }
1315 }
c0236afe 1316
2b5e7bc2
KW
1317 /* We need a complete well-formed UTF-8 character to discern
1318 * non-characters, so can't look for them here */
1319 }
1320 }
949cf498 1321
2b5e7bc2
KW
1322 ready_to_handle_errors:
1323
1324 /* At this point:
1325 * curlen contains the number of bytes in the sequence that
1326 * this call should advance the input by.
e308b348
KW
1327 * avail_len gives the available number of bytes passed in, but
1328 * only if this is less than the expected number of
1329 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1330 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1331 * is set in it for each potential problem found.
1332 * uv contains the code point the input sequence
1333 * represents; or if there is a problem that prevents
1334 * a well-defined value from being computed, it is
1335 * some subsitute value, typically the REPLACEMENT
1336 * CHARACTER.
1337 * s0 points to the first byte of the character
1338 * send points to just after where that (potentially
1339 * partial) character ends
1340 * adjusted_s0 normally is the same as s0, but in case of an
1341 * overlong for which the UTF-8 matters below, it is
1342 * the first byte of the shortest form representation
1343 * of the input.
1344 * adjusted_send normally is the same as 'send', but if adjusted_s0
1345 * is set to something other than s0, this points one
1346 * beyond its end
1347 */
eb83ed87 1348
2b5e7bc2
KW
1349 if (UNLIKELY(possible_problems)) {
1350 bool disallowed = FALSE;
1351 const U32 orig_problems = possible_problems;
1352
1353 while (possible_problems) { /* Handle each possible problem */
1354 UV pack_warn = 0;
1355 char * message = NULL;
1356
1357 /* Each 'if' clause handles one problem. They are ordered so that
1358 * the first ones' messages will be displayed before the later
1359 * ones; this is kinda in decreasing severity order */
1360 if (possible_problems & UTF8_GOT_OVERFLOW) {
1361
1362 /* Overflow means also got a super and above 31 bits, but we
1363 * handle all three cases here */
1364 possible_problems
1365 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1366 *errors |= UTF8_GOT_OVERFLOW;
1367
1368 /* But the API says we flag all errors found */
1369 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1370 *errors |= UTF8_GOT_SUPER;
1371 }
ddb65933
KW
1372 if (flags
1373 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1374 {
f9380377
KW
1375 *errors |= UTF8_GOT_ABOVE_31_BIT;
1376 }
2b5e7bc2 1377
d60baaa7
KW
1378 /* Disallow if any of the three categories say to */
1379 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1380 || (flags & ( UTF8_DISALLOW_SUPER
1381 |UTF8_DISALLOW_ABOVE_31_BIT)))
1382 {
1383 disallowed = TRUE;
1384 }
1385
1386
1387 /* Likewise, warn if any say to, plus if deprecation warnings
1388 * are on, because this code point is above IV_MAX */
1389 if ( ckWARN_d(WARN_DEPRECATED)
1390 || ! (flags & UTF8_ALLOW_OVERFLOW)
1391 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1392 {
2b5e7bc2 1393
ddb65933
KW
1394 /* The warnings code explicitly says it doesn't handle the
1395 * case of packWARN2 and two categories which have
1396 * parent-child relationship. Even if it works now to
1397 * raise the warning if either is enabled, it wouldn't
1398 * necessarily do so in the future. We output (only) the
1399 * most dire warning*/
1400 if (! (flags & UTF8_CHECK_ONLY)) {
1401 if (ckWARN_d(WARN_UTF8)) {
1402 pack_warn = packWARN(WARN_UTF8);
1403 }
1404 else if (ckWARN_d(WARN_NON_UNICODE)) {
1405 pack_warn = packWARN(WARN_NON_UNICODE);
1406 }
1407 if (pack_warn) {
1408 message = Perl_form(aTHX_ "%s: %s (overflows)",
1409 malformed_text,
7e2f38b2 1410 _byte_dump_string(s0, send - s0, 0));
ddb65933 1411 }
2b5e7bc2
KW
1412 }
1413 }
1414 }
1415 else if (possible_problems & UTF8_GOT_EMPTY) {
1416 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1417 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1418
1419 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1420
1421 /* This so-called malformation is now treated as a bug in
1422 * the caller. If you have nothing to decode, skip calling
1423 * this function */
1424 assert(0);
1425
2b5e7bc2
KW
1426 disallowed = TRUE;
1427 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1428 pack_warn = packWARN(WARN_UTF8);
1429 message = Perl_form(aTHX_ "%s (empty string)",
1430 malformed_text);
1431 }
1432 }
1433 }
1434 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1435 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1436 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1437
1438 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1439 disallowed = TRUE;
1440 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1441 pack_warn = packWARN(WARN_UTF8);
1442 message = Perl_form(aTHX_
1443 "%s: %s (unexpected continuation byte 0x%02x,"
1444 " with no preceding start byte)",
1445 malformed_text,
7e2f38b2 1446 _byte_dump_string(s0, 1, 0), *s0);
2b5e7bc2
KW
1447 }
1448 }
1449 }
2b5e7bc2
KW
1450 else if (possible_problems & UTF8_GOT_SHORT) {
1451 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1452 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1453
1454 if (! (flags & UTF8_ALLOW_SHORT)) {
1455 disallowed = TRUE;
1456 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1457 pack_warn = packWARN(WARN_UTF8);
1458 message = Perl_form(aTHX_
9a6c9c81 1459 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2 1460 malformed_text,
7e2f38b2 1461 _byte_dump_string(s0, send - s0, 0),
e308b348
KW
1462 (int)avail_len,
1463 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1464 (int)expectlen);
1465 }
1466 }
ba210ebe 1467
2b5e7bc2 1468 }
e308b348
KW
1469 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1470 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1471 *errors |= UTF8_GOT_NON_CONTINUATION;
1472
1473 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1474 disallowed = TRUE;
1475 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1476
1477 /* If we don't know for sure that the input length is
1478 * valid, avoid as much as possible reading past the
1479 * end of the buffer */
1480 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1481 ? s - s0
1482 : send - s0;
e308b348
KW
1483 pack_warn = packWARN(WARN_UTF8);
1484 message = Perl_form(aTHX_ "%s",
1485 unexpected_non_continuation_text(s0,
99a765e9 1486 printlen,
e308b348
KW
1487 s - s0,
1488 (int) expectlen));
1489 }
1490 }
1491 }
2b5e7bc2
KW
1492 else if (possible_problems & UTF8_GOT_LONG) {
1493 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1494 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1495
94953955
KW
1496 if (flags & UTF8_ALLOW_LONG) {
1497
1498 /* We don't allow the actual overlong value, unless the
1499 * special extra bit is also set */
1500 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1501 & ~UTF8_ALLOW_LONG)))
1502 {
1503 uv = UNICODE_REPLACEMENT;
1504 }
1505 }
1506 else {
2b5e7bc2
KW
1507 disallowed = TRUE;
1508
1509 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1510 pack_warn = packWARN(WARN_UTF8);
1511
1512 /* These error types cause 'uv' to be something that
1513 * isn't what was intended, so can't use it in the
1514 * message. The other error types either can't
1515 * generate an overlong, or else the 'uv' is valid */
1516 if (orig_problems &
1517 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1518 {
1519 message = Perl_form(aTHX_
1520 "%s: %s (any UTF-8 sequence that starts"
1521 " with \"%s\" is overlong which can and"
1522 " should be represented with a"
1523 " different, shorter sequence)",
1524 malformed_text,
7e2f38b2
KW
1525 _byte_dump_string(s0, send - s0, 0),
1526 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1527 }
1528 else {
1529 U8 tmpbuf[UTF8_MAXBYTES+1];
1530 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1531 uv, 0);
1532 message = Perl_form(aTHX_
1533 "%s: %s (overlong; instead use %s to represent"
147e3846 1534 " U+%0*" UVXf ")",
2b5e7bc2 1535 malformed_text,
7e2f38b2
KW
1536 _byte_dump_string(s0, send - s0, 0),
1537 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2b5e7bc2
KW
1538 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1539 small code points */
1540 uv);
1541 }
1542 }
1543 }
1544 }
1545 else if (possible_problems & UTF8_GOT_SURROGATE) {
1546 possible_problems &= ~UTF8_GOT_SURROGATE;
1547
f9380377
KW
1548 if (flags & UTF8_WARN_SURROGATE) {
1549 *errors |= UTF8_GOT_SURROGATE;
1550
1551 if ( ! (flags & UTF8_CHECK_ONLY)
1552 && ckWARN_d(WARN_SURROGATE))
1553 {
2b5e7bc2
KW
1554 pack_warn = packWARN(WARN_SURROGATE);
1555
1556 /* These are the only errors that can occur with a
1557 * surrogate when the 'uv' isn't valid */
1558 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1559 message = Perl_form(aTHX_
1560 "UTF-16 surrogate (any UTF-8 sequence that"
1561 " starts with \"%s\" is for a surrogate)",
7e2f38b2 1562 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1563 }
1564 else {
1565 message = Perl_form(aTHX_
147e3846 1566 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1567 }
f9380377 1568 }
2b5e7bc2 1569 }
ba210ebe 1570
2b5e7bc2
KW
1571 if (flags & UTF8_DISALLOW_SURROGATE) {
1572 disallowed = TRUE;
f9380377 1573 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1574 }
1575 }
1576 else if (possible_problems & UTF8_GOT_SUPER) {
1577 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1578
f9380377
KW
1579 if (flags & UTF8_WARN_SUPER) {
1580 *errors |= UTF8_GOT_SUPER;
1581
1582 if ( ! (flags & UTF8_CHECK_ONLY)
1583 && ckWARN_d(WARN_NON_UNICODE))
1584 {
2b5e7bc2
KW
1585 pack_warn = packWARN(WARN_NON_UNICODE);
1586
1587 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1588 message = Perl_form(aTHX_
1589 "Any UTF-8 sequence that starts with"
1590 " \"%s\" is for a non-Unicode code point,"
1591 " may not be portable",
7e2f38b2 1592 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1593 }
1594 else {
1595 message = Perl_form(aTHX_
147e3846 1596 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1597 " Unicode, may not be portable",
1598 uv);
1599 }
f9380377 1600 }
2b5e7bc2 1601 }
ba210ebe 1602
2b5e7bc2
KW
1603 /* The maximum code point ever specified by a standard was
1604 * 2**31 - 1. Anything larger than that is a Perl extension
1605 * that very well may not be understood by other applications
1606 * (including earlier perl versions on EBCDIC platforms). We
1607 * test for these after the regular SUPER ones, and before
1608 * possibly bailing out, so that the slightly more dire warning
1609 * will override the regular one. */
1610 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1611 |UTF8_WARN_SUPER
1612 |UTF8_DISALLOW_ABOVE_31_BIT))
1613 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1614 && UNLIKELY(is_utf8_cp_above_31_bits(
1615 adjusted_s0,
1616 adjusted_send)))
1617 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1618 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1619 {
1620 if ( ! (flags & UTF8_CHECK_ONLY)
1621 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1622 && ckWARN_d(WARN_UTF8))
1623 {
1624 pack_warn = packWARN(WARN_UTF8);
1625
1626 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1627 message = Perl_form(aTHX_
1628 "Any UTF-8 sequence that starts with"
1629 " \"%s\" is for a non-Unicode code"
1630 " point, and is not portable",
7e2f38b2 1631 _byte_dump_string(s0, curlen, 0));
2b5e7bc2
KW
1632 }
1633 else {
1634 message = Perl_form(aTHX_
147e3846 1635 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1636 " and not portable",
1637 uv);
1638 }
1639 }
1640
ddb65933
KW
1641 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1642 |UTF8_DISALLOW_ABOVE_31_BIT))
1643 {
f9380377
KW
1644 *errors |= UTF8_GOT_ABOVE_31_BIT;
1645
1646 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1647 disallowed = TRUE;
1648 }
2b5e7bc2
KW
1649 }
1650 }
eb83ed87 1651
2b5e7bc2 1652 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1653 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1654 disallowed = TRUE;
1655 }
eb83ed87 1656
2b5e7bc2
KW
1657 /* The deprecated warning overrides any non-deprecated one. If
1658 * there are other problems, a deprecation message is not
1659 * really helpful, so don't bother to raise it in that case.
1660 * This also keeps the code from having to handle the case
1661 * where 'uv' is not valid. */
1662 if ( ! (orig_problems
1663 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
13f4dd34
A
1664 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)) {
1665 Perl_croak(aTHX_ cp_above_legal_max, uv,
1666 MAX_NON_DEPRECATED_CP);
2b5e7bc2
KW
1667 }
1668 }
1669 else if (possible_problems & UTF8_GOT_NONCHAR) {
1670 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1671
f9380377
KW
1672 if (flags & UTF8_WARN_NONCHAR) {
1673 *errors |= UTF8_GOT_NONCHAR;
1674
1675 if ( ! (flags & UTF8_CHECK_ONLY)
1676 && ckWARN_d(WARN_NONCHAR))
1677 {
2b5e7bc2
KW
1678 /* The code above should have guaranteed that we don't
1679 * get here with errors other than overlong */
1680 assert (! (orig_problems
1681 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1682
1683 pack_warn = packWARN(WARN_NONCHAR);
1684 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1685 " U+%04" UVXf " is not recommended"
2b5e7bc2 1686 " for open interchange", uv);
f9380377 1687 }
2b5e7bc2 1688 }
5b311467 1689
2b5e7bc2
KW
1690 if (flags & UTF8_DISALLOW_NONCHAR) {
1691 disallowed = TRUE;
f9380377 1692 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1693 }
1694 } /* End of looking through the possible flags */
1695
1696 /* Display the message (if any) for the problem being handled in
1697 * this iteration of the loop */
1698 if (message) {
1699 if (PL_op)
1700 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1701 OP_DESC(PL_op));
1702 else
1703 Perl_warner(aTHX_ pack_warn, "%s", message);
1704 }
ddb65933 1705 } /* End of 'while (possible_problems)' */
a0dbb045 1706
2b5e7bc2
KW
1707 /* Since there was a possible problem, the returned length may need to
1708 * be changed from the one stored at the beginning of this function.
1709 * Instead of trying to figure out if that's needed, just do it. */
1710 if (retlen) {
1711 *retlen = curlen;
1712 }
a0dbb045 1713
2b5e7bc2
KW
1714 if (disallowed) {
1715 if (flags & UTF8_CHECK_ONLY && retlen) {
1716 *retlen = ((STRLEN) -1);
1717 }
1718 return 0;
1719 }
eb83ed87 1720 }
ba210ebe 1721
2b5e7bc2 1722 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1723}
1724
8e84507e 1725/*
ec5f19d0
KW
1726=for apidoc utf8_to_uvchr_buf
1727
1728Returns the native code point of the first character in the string C<s> which
1729is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1730C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1731
524080c4
KW
1732If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1733enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1734C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1735(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1736C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1737the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1738See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1739returned.
ec5f19d0 1740
760c7c2f
KW
1741Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1742unless those are turned off.
1743
ec5f19d0 1744=cut
52be2536
KW
1745
1746Also implemented as a macro in utf8.h
1747
ec5f19d0
KW
1748*/
1749
1750
1751UV
1752Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1753{
7f974d7e
KW
1754 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1755
ec5f19d0
KW
1756 assert(s < send);
1757
1758 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1759 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1760}
1761
52be2536
KW
1762/* This is marked as deprecated
1763 *
ec5f19d0
KW
1764=for apidoc utf8_to_uvuni_buf
1765
de69f3af
KW
1766Only in very rare circumstances should code need to be dealing in Unicode
1767(as opposed to native) code points. In those few cases, use
1768C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1769
1770Returns the Unicode (not-native) code point of the first character in the
1771string C<s> which
ec5f19d0
KW
1772is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1773C<retlen> will be set to the length, in bytes, of that character.
1774
524080c4
KW
1775If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1776enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1777NULL) to -1. If those warnings are off, the computed value if well-defined (or
1778the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1779is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1780next possible position in C<s> that could begin a non-malformed character.
de69f3af 1781See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1782
760c7c2f
KW
1783Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1784unless those are turned off.
1785
ec5f19d0
KW
1786=cut
1787*/
1788
1789UV
1790Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1791{
1792 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1793
1794 assert(send > s);
1795
5962d97e
KW
1796 /* Call the low level routine, asking for checks */
1797 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1798}
1799
b76347f2 1800/*
87cea99e 1801=for apidoc utf8_length
b76347f2
JH
1802
1803Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1804Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1805up past C<e>, croaks.
b76347f2
JH
1806
1807=cut
1808*/
1809
1810STRLEN
35a4481c 1811Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1812{
1813 STRLEN len = 0;
1814
7918f24d
NC
1815 PERL_ARGS_ASSERT_UTF8_LENGTH;
1816
8850bf83
JH
1817 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1818 * the bitops (especially ~) can create illegal UTF-8.
1819 * In other words: in Perl UTF-8 is not just for Unicode. */
1820
a3b680e6
AL
1821 if (e < s)
1822 goto warn_and_return;
b76347f2 1823 while (s < e) {
4cbf4130 1824 s += UTF8SKIP(s);
8e91ec7f
AV
1825 len++;
1826 }
1827
1828 if (e != s) {
1829 len--;
1830 warn_and_return:
9b387841
NC
1831 if (PL_op)
1832 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1833 "%s in %s", unees, OP_DESC(PL_op));
1834 else
61a12c31 1835 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1836 }
1837
1838 return len;
1839}
1840
b06226ff 1841/*
fed3ba5d
NC
1842=for apidoc bytes_cmp_utf8
1843
a1433954 1844Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1845sequence of characters (stored as UTF-8)
1846in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1847equal, -1 or -2 if the first string is less than the second string, +1 or +2
1848if the first string is greater than the second string.
1849
1850-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1851longer string. -2 or +2 is returned if
1852there was a difference between characters
fed3ba5d
NC
1853within the strings.
1854
1855=cut
1856*/
1857
1858int
1859Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1860{
1861 const U8 *const bend = b + blen;
1862 const U8 *const uend = u + ulen;
1863
1864 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1865
1866 while (b < bend && u < uend) {
1867 U8 c = *u++;
1868 if (!UTF8_IS_INVARIANT(c)) {
1869 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1870 if (u < uend) {
1871 U8 c1 = *u++;
1872 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1873 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1874 } else {
2b5e7bc2 1875 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1876 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1877 "%s %s%s",
7cf8d05d 1878 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1879 PL_op ? " in " : "",
1880 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1881 return -2;
1882 }
1883 } else {
1884 if (PL_op)
1885 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1886 "%s in %s", unees, OP_DESC(PL_op));
1887 else
61a12c31 1888 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1889 return -2; /* Really want to return undef :-) */
1890 }
1891 } else {
1892 return -2;
1893 }
1894 }
1895 if (*b != c) {
1896 return *b < c ? -2 : +2;
1897 }
1898 ++b;
1899 }
1900
1901 if (b == bend && u == uend)
1902 return 0;
1903
1904 return b < bend ? +1 : -1;
1905}
1906
1907/*
87cea99e 1908=for apidoc utf8_to_bytes
6940069f 1909
2bbc8d55 1910Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1911Unlike L</bytes_to_utf8>, this over-writes the original string, and
1912updates C<len> to contain the new length.
67e989fb 1913Returns zero on failure, setting C<len> to -1.
6940069f 1914
a1433954 1915If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1916
6940069f
GS
1917=cut
1918*/
1919
1920U8 *
37607a96 1921Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1922{
d4c19fe8
AL
1923 U8 * const save = s;
1924 U8 * const send = s + *len;
6940069f 1925 U8 *d;
246fae53 1926
7918f24d 1927 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1928 PERL_UNUSED_CONTEXT;
7918f24d 1929
1e54db1a 1930 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1931 while (s < send) {
d59937ca
KW
1932 if (! UTF8_IS_INVARIANT(*s)) {
1933 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1934 *len = ((STRLEN) -1);
1935 return 0;
1936 }
1937 s++;
dcad2880 1938 }
d59937ca 1939 s++;
246fae53 1940 }
dcad2880
JH
1941
1942 d = s = save;
6940069f 1943 while (s < send) {
80e0b38f
KW
1944 U8 c = *s++;
1945 if (! UTF8_IS_INVARIANT(c)) {
1946 /* Then it is two-byte encoded */
a62b247b 1947 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1948 s++;
1949 }
1950 *d++ = c;
6940069f
GS
1951 }
1952 *d = '\0';
246fae53 1953 *len = d - save;
6940069f
GS
1954 return save;
1955}
1956
1957/*
87cea99e 1958=for apidoc bytes_from_utf8
f9a63242 1959
4f3d592d
KW
1960Converts a potentially UTF-8 encoded string C<s> of length C<len> into native
1961byte encoding. On input, the boolean C<*is_utf8> gives whether or not C<s> is
1962actually encoded in UTF-8.
1963
1964Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
1965the input string.
1966
1967Do nothing if C<*is_utf8> is 0, or if there are code points in the string
1968not expressible in native byte encoding. In these cases, C<*is_utf8> and
1969C<*len> are unchanged, and the return value is the original C<s>.
1970
1971Otherwise, C<*is_utf8> is set to 0, and the return value is a pointer to a
1972newly created string containing a downgraded copy of C<s>, and whose length is
1973returned in C<*len>, updated.
f9a63242 1974
37607a96
PK
1975=cut
1976*/
f9a63242
JH
1977
1978U8 *
e1ec3a88 1979Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1980{
f9a63242 1981 U8 *d;
e1ec3a88
AL
1982 const U8 *start = s;
1983 const U8 *send;
2a67563b 1984 Size_t count = 0;
f9a63242 1985
7918f24d 1986 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1987 PERL_UNUSED_CONTEXT;
f9a63242 1988 if (!*is_utf8)
73d840c0 1989 return (U8 *)start;
f9a63242 1990
1e54db1a 1991 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1992 for (send = s + *len; s < send;) {
d59937ca
KW
1993 if (! UTF8_IS_INVARIANT(*s)) {
1994 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1995 return (U8 *)start;
d59937ca
KW
1996 }
1997 count++;
1998 s++;
db42d148 1999 }
d59937ca 2000 s++;
f9a63242
JH
2001 }
2002
35da51f7 2003 *is_utf8 = FALSE;
f9a63242 2004
212542aa 2005 Newx(d, (*len) - count + 1, U8);
ef9edfd0 2006 s = start; start = d;
f9a63242
JH
2007 while (s < send) {
2008 U8 c = *s++;
1a91c45d 2009 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 2010 /* Then it is two-byte encoded */
a62b247b 2011 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 2012 s++;
c4d5f83a
NIS
2013 }
2014 *d++ = c;
f9a63242
JH
2015 }
2016 *d = '\0';
2017 *len = d - start;
73d840c0 2018 return (U8 *)start;
f9a63242
JH
2019}
2020
2021/*
87cea99e 2022=for apidoc bytes_to_utf8
6940069f 2023
ff97e5cf
KW
2024Converts a string C<s> of length C<len> bytes from the native encoding into
2025UTF-8.
6662521e 2026Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 2027reflect the new length in bytes.
6940069f 2028
75200dff 2029A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2030
2031If you want to convert to UTF-8 from encodings other than
2032the native (Latin1 or EBCDIC),
a1433954 2033see L</sv_recode_to_utf8>().
c9ada85f 2034
497711e7 2035=cut
6940069f
GS
2036*/
2037
c682ebef
FC
2038/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2039 likewise need duplication. */
2040
6940069f 2041U8*
35a4481c 2042Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 2043{
35a4481c 2044 const U8 * const send = s + (*len);
6940069f
GS
2045 U8 *d;
2046 U8 *dst;
7918f24d
NC
2047
2048 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2049 PERL_UNUSED_CONTEXT;
6940069f 2050
212542aa 2051 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2052 dst = d;
2053
2054 while (s < send) {
55d09dc8
KW
2055 append_utf8_from_native_byte(*s, &d);
2056 s++;
6940069f
GS
2057 }
2058 *d = '\0';
6662521e 2059 *len = d-dst;
6940069f
GS
2060 return dst;
2061}
2062
a0ed51b3 2063/*
dea0fc0b 2064 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2065 *
2066 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2067 * We optimize for native, for obvious reasons. */
2068
2069U8*
dea0fc0b 2070Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2071{
dea0fc0b
JH
2072 U8* pend;
2073 U8* dstart = d;
2074
7918f24d
NC
2075 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2076
dea0fc0b 2077 if (bytelen & 1)
147e3846 2078 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2079
2080 pend = p + bytelen;
2081
a0ed51b3 2082 while (p < pend) {
dea0fc0b
JH
2083 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2084 p += 2;
2d1545e5 2085 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2086 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2087 continue;
2088 }
56d37426
KW
2089 if (uv <= MAX_UTF8_TWO_BYTE) {
2090 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2091 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2092 continue;
2093 }
46956fad
KW
2094#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2095#define LAST_HIGH_SURROGATE 0xDBFF
2096#define FIRST_LOW_SURROGATE 0xDC00
2097#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2098
2099 /* This assumes that most uses will be in the first Unicode plane, not
2100 * needing surrogates */
2101 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2102 && uv <= UNICODE_SURROGATE_LAST))
2103 {
2104 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2105 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2106 }
2107 else {
01ea242b 2108 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2109 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2110 || UNLIKELY(low > LAST_LOW_SURROGATE))
2111 {
01ea242b 2112 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2113 }
2114 p += 2;
46956fad
KW
2115 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2116 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2117 }
a0ed51b3 2118 }
56d37426
KW
2119#ifdef EBCDIC
2120 d = uvoffuni_to_utf8_flags(d, uv, 0);
2121#else
a0ed51b3 2122 if (uv < 0x10000) {
eb160463
GS
2123 *d++ = (U8)(( uv >> 12) | 0xe0);
2124 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2125 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2126 continue;
2127 }
2128 else {
eb160463
GS
2129 *d++ = (U8)(( uv >> 18) | 0xf0);
2130 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2131 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2132 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2133 continue;
2134 }
56d37426 2135#endif
a0ed51b3 2136 }
dea0fc0b 2137 *newlen = d - dstart;
a0ed51b3
LW
2138 return d;
2139}
2140
2141/* Note: this one is slightly destructive of the source. */
2142
2143U8*
dea0fc0b 2144Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2145{
2146 U8* s = (U8*)p;
d4c19fe8 2147 U8* const send = s + bytelen;
7918f24d
NC
2148
2149 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2150
e0ea5e2d 2151 if (bytelen & 1)
147e3846 2152 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2153 (UV)bytelen);
2154
a0ed51b3 2155 while (s < send) {
d4c19fe8 2156 const U8 tmp = s[0];
a0ed51b3
LW
2157 s[0] = s[1];
2158 s[1] = tmp;
2159 s += 2;
2160 }
dea0fc0b 2161 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2162}
2163
922e8cb4
KW
2164bool
2165Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2166{
2167 U8 tmpbuf[UTF8_MAXBYTES+1];
2168 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2169 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2170}
2171
f9ae8fb6
JD
2172/* Internal function so we can deprecate the external one, and call
2173 this one from other deprecated functions in this file */
2174
f2645549
KW
2175bool
2176Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2177{
f2645549 2178 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2179
2180 if (*p == '_')
2181 return TRUE;
f25ce844 2182 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2183}
2184
5092f92a 2185bool
eba68aa0
KW
2186Perl__is_uni_perl_idcont(pTHX_ UV c)
2187{
2188 U8 tmpbuf[UTF8_MAXBYTES+1];
2189 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2190 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2191}
2192
2193bool
f91dcd13
KW
2194Perl__is_uni_perl_idstart(pTHX_ UV c)
2195{
2196 U8 tmpbuf[UTF8_MAXBYTES+1];
2197 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2198 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2199}
2200
3a4c58c9
KW
2201UV
2202Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2203{
2204 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2205 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2206 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2207 * either "SS" or "Ss". Which one to use is passed into the routine in
2208 * 'S_or_s' to avoid a test */
2209
2210 UV converted = toUPPER_LATIN1_MOD(c);
2211
2212 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2213
2214 assert(S_or_s == 'S' || S_or_s == 's');
2215
6f2d5cbc 2216 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2217 characters in this range */
3a4c58c9
KW
2218 *p = (U8) converted;
2219 *lenp = 1;
2220 return converted;
2221 }
2222
2223 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2224 * which it maps to one of them, so as to only have to have one check for
2225 * it in the main case */
2226 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2227 switch (c) {
2228 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2229 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2230 break;
2231 case MICRO_SIGN:
2232 converted = GREEK_CAPITAL_LETTER_MU;
2233 break;
79e064b9
KW
2234#if UNICODE_MAJOR_VERSION > 2 \
2235 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2236 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2237 case LATIN_SMALL_LETTER_SHARP_S:
2238 *(p)++ = 'S';
2239 *p = S_or_s;
2240 *lenp = 2;
2241 return 'S';
79e064b9 2242#endif
3a4c58c9
KW
2243 default:
2244 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2245 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2246 }
2247 }
2248
2249 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2250 *p = UTF8_TWO_BYTE_LO(converted);
2251 *lenp = 2;
2252
2253 return converted;
2254}
2255
50bda2c3
KW
2256/* Call the function to convert a UTF-8 encoded character to the specified case.
2257 * Note that there may be more than one character in the result.
2258 * INP is a pointer to the first byte of the input character
2259 * OUTP will be set to the first byte of the string of changed characters. It
2260 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2261 * LENP will be set to the length in bytes of the string of changed characters
2262 *
2263 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2264#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2265#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2266#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2267
b9992569
KW
2268/* This additionally has the input parameter 'specials', which if non-zero will
2269 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2270 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2271#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2272
84afefe6
JH
2273UV
2274Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2275{
a1433954
KW
2276 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2277 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2278 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2279 * the changed version may be longer than the original character.
2280 *
2281 * The ordinal of the first character of the changed version is returned
2282 * (but note, as explained above, that there may be more.) */
2283
7918f24d
NC
2284 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2285
3a4c58c9
KW
2286 if (c < 256) {
2287 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2288 }
2289
0ebc6274 2290 uvchr_to_utf8(p, c);
b9992569 2291 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2292}
2293
84afefe6
JH
2294UV
2295Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2296{
7918f24d
NC
2297 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2298
3a4c58c9
KW
2299 if (c < 256) {
2300 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2301 }
2302
0ebc6274 2303 uvchr_to_utf8(p, c);
b9992569 2304 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2305}
2306
afc16117 2307STATIC U8
eaf412bf 2308S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2309{
2310 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2311 * those, converting the result to UTF-8. Since the result is always just
a1433954 2312 * one character, we allow <p> to be NULL */
afc16117
KW
2313
2314 U8 converted = toLOWER_LATIN1(c);
2315
eaf412bf
KW
2316 PERL_UNUSED_ARG(dummy);
2317
afc16117 2318 if (p != NULL) {
6f2d5cbc 2319 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2320 *p = converted;
2321 *lenp = 1;
2322 }
2323 else {
430c9760
KW
2324 /* Result is known to always be < 256, so can use the EIGHT_BIT
2325 * macros */
2326 *p = UTF8_EIGHT_BIT_HI(converted);
2327 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2328 *lenp = 2;
2329 }
2330 }
2331 return converted;
2332}
2333
84afefe6
JH
2334UV
2335Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2336{
7918f24d
NC
2337 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2338
afc16117 2339 if (c < 256) {
eaf412bf 2340 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2341 }
2342
afc16117 2343 uvchr_to_utf8(p, c);
b9992569 2344 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2345}
2346
84afefe6 2347UV
51910141 2348Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2349{
51910141 2350 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2351 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2352 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2353 *
2354 * Not to be used for locale folds
51910141 2355 */
f673fad4 2356
a1dde8de
KW
2357 UV converted;
2358
2359 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2360 PERL_UNUSED_CONTEXT;
a1dde8de 2361
1ca267a5
KW
2362 assert (! (flags & FOLD_FLAGS_LOCALE));
2363
659a7c2d 2364 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2365 converted = GREEK_SMALL_LETTER_MU;
2366 }
9b63e895
KW
2367#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2368 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2369 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2370 else if ( (flags & FOLD_FLAGS_FULL)
2371 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2372 {
1ca267a5
KW
2373 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2374 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2375 * under those circumstances. */
2376 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2377 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2378 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2379 p, *lenp, U8);
2380 return LATIN_SMALL_LETTER_LONG_S;
2381 }
2382 else {
4f489194
KW
2383 *(p)++ = 's';
2384 *p = 's';
2385 *lenp = 2;
2386 return 's';
1ca267a5 2387 }
a1dde8de 2388 }
9b63e895 2389#endif
a1dde8de
KW
2390 else { /* In this range the fold of all other characters is their lower
2391 case */
2392 converted = toLOWER_LATIN1(c);
2393 }
2394
6f2d5cbc 2395 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2396 *p = (U8) converted;
2397 *lenp = 1;
2398 }
2399 else {
2400 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2401 *p = UTF8_TWO_BYTE_LO(converted);
2402 *lenp = 2;
2403 }
2404
2405 return converted;
2406}
2407
2408UV
31f05a37 2409Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2410{
4b593389 2411
a0270393
KW
2412 /* Not currently externally documented, and subject to change
2413 * <flags> bits meanings:
2414 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2415 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2416 * locale are to be used.
a0270393
KW
2417 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2418 */
4b593389 2419
36bb2ab6 2420 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2421
780fcc9f
KW
2422 if (flags & FOLD_FLAGS_LOCALE) {
2423 /* Treat a UTF-8 locale as not being in locale at all */
2424 if (IN_UTF8_CTYPE_LOCALE) {
2425 flags &= ~FOLD_FLAGS_LOCALE;
2426 }
2427 else {
2428 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2429 goto needs_full_generality;
780fcc9f 2430 }
31f05a37
KW
2431 }
2432
a1dde8de 2433 if (c < 256) {
e7b7ac46 2434 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2435 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2436 }
2437
2f306ab9 2438 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2439 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2440 uvchr_to_utf8(p, c);
b9992569 2441 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2442 }
567b353c 2443 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2444 the special flags. */
2445 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2446
2447 needs_full_generality:
a0270393 2448 uvchr_to_utf8(utf8_c, c);
a1a5ec35 2449 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
a0270393 2450 }
84afefe6
JH
2451}
2452
26483009 2453PERL_STATIC_INLINE bool
5141f98e 2454S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2455 const char *const swashname, SV* const invlist)
bde6a22d 2456{
ea317ccb
KW
2457 /* returns a boolean giving whether or not the UTF8-encoded character that
2458 * starts at <p> is in the swash indicated by <swashname>. <swash>
2459 * contains a pointer to where the swash indicated by <swashname>
2460 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2461 * look at <*swash> and only generate a swash if it is not null. <invlist>
2462 * is NULL or an inversion list that defines the swash. If not null, it
2463 * saves time during initialization of the swash.
ea317ccb
KW
2464 *
2465 * Note that it is assumed that the buffer length of <p> is enough to
2466 * contain all the bytes that comprise the character. Thus, <*p> should
2467 * have been checked before this call for mal-formedness enough to assure
2468 * that. */
2469
7918f24d
NC
2470 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2471
492a624f 2472 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2473 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2474 * as far as there being enough bytes available in it to accommodate the
2475 * character without reading beyond the end, and pass that number on to the
2476 * validating routine */
6302f837 2477 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2478 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2479 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2480 1 /* Die */ );
2481 NOT_REACHED; /* NOTREACHED */
28123549 2482 }
86ae6e94 2483
87367d5f
KW
2484 if (!*swash) {
2485 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2486 *swash = _core_swash_init("utf8",
2487
2488 /* Only use the name if there is no inversion
2489 * list; otherwise will go out to disk */
2490 (invlist) ? "" : swashname,
2491
2492 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2493 }
28123549 2494
bde6a22d
NC
2495 return swash_fetch(*swash, p, TRUE) != 0;
2496}
2497
da8c1a98
KW
2498PERL_STATIC_INLINE bool
2499S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2500 const char *const swashname, SV* const invlist)
2501{
2502 /* returns a boolean giving whether or not the UTF8-encoded character that
2503 * starts at <p>, and extending no further than <e - 1> is in the swash
2504 * indicated by <swashname>. <swash> contains a pointer to where the swash
2505 * indicated by <swashname> is to be stored; which this routine will do, so
2506 * that future calls will look at <*swash> and only generate a swash if it
2507 * is not null. <invlist> is NULL or an inversion list that defines the
2508 * swash. If not null, it saves time during initialization of the swash.
2509 */
2510
2511 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2512
2513 if (! isUTF8_CHAR(p, e)) {
2514 _force_out_malformed_utf8_message(p, e, 0, 1);
2515 NOT_REACHED; /* NOTREACHED */
2516 }
2517
2518 if (!*swash) {
2519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2520 *swash = _core_swash_init("utf8",
2521
2522 /* Only use the name if there is no inversion
2523 * list; otherwise will go out to disk */
2524 (invlist) ? "" : swashname,
2525
2526 &PL_sv_undef, 1, 0, invlist, &flags);
2527 }
2528
2529 return swash_fetch(*swash, p, TRUE) != 0;
2530}
2531
34aeb2e9
KW
2532STATIC void
2533S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2534 const char * const alternative,
2535 const bool use_locale,
2536 const char * const file,
2537 const unsigned line)
2538{
2539 const char * key;
2540
2541 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2542
2543 if (ckWARN_d(WARN_DEPRECATED)) {
2544
2545 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2546 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2547 if (! PL_seen_deprecated_macro) {
2548 PL_seen_deprecated_macro = newHV();
2549 }
2550 if (! hv_store(PL_seen_deprecated_macro, key,
2551 strlen(key), &PL_sv_undef, 0))
2552 {
2553 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2554 }
2555
c44e9413 2556 if (instr(file, "mathoms.c")) {
607313a1
KW
2557 Perl_warner(aTHX_ WARN_DEPRECATED,
2558 "In %s, line %d, starting in Perl v5.30, %s()"
2559 " will be removed. Avoid this message by"
2560 " converting to use %s().\n",
2561 file, line, name, alternative);
2562 }
2563 else {
34aeb2e9
KW
2564 Perl_warner(aTHX_ WARN_DEPRECATED,
2565 "In %s, line %d, starting in Perl v5.30, %s() will"
2566 " require an additional parameter. Avoid this"
2567 " message by converting to use %s().\n",
2568 file, line, name, alternative);
607313a1 2569 }
34aeb2e9
KW
2570 }
2571 }
2572}
2573
bde6a22d 2574bool
34aeb2e9
KW
2575Perl__is_utf8_FOO(pTHX_ U8 classnum,
2576 const U8 *p,
2577 const char * const name,
2578 const char * const alternative,
2579 const bool use_utf8,
2580 const bool use_locale,
2581 const char * const file,
2582 const unsigned line)
922e8cb4 2583{
922e8cb4
KW
2584 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2585
34aeb2e9
KW
2586 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2587
2588 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
34aeb2e9
KW
2589
2590 switch (classnum) {
2591 case _CC_WORDCHAR:
2592 case _CC_DIGIT:
2593 case _CC_ALPHA:
2594 case _CC_LOWER:
2595 case _CC_UPPER:
2596 case _CC_PUNCT:
2597 case _CC_PRINT:
2598 case _CC_ALPHANUMERIC:
2599 case _CC_GRAPH:
2600 case _CC_CASED:
2601
2602 return is_utf8_common(p,
2603 &PL_utf8_swash_ptrs[classnum],
2604 swash_property_names[classnum],
2605 PL_XPosix_ptrs[classnum]);
2606
2607 case _CC_SPACE:
2608 return is_XPERLSPACE_high(p);
2609 case _CC_BLANK:
2610 return is_HORIZWS_high(p);
2611 case _CC_XDIGIT:
2612 return is_XDIGIT_high(p);
2613 case _CC_CNTRL:
2614 return 0;
2615 case _CC_ASCII:
2616 return 0;
2617 case _CC_VERTSPACE:
2618 return is_VERTWS_high(p);
2619 case _CC_IDFIRST:
2620 if (! PL_utf8_perl_idstart) {
22f0498f
KW
2621 PL_utf8_perl_idstart
2622 = _new_invlist_C_array(_Perl_IDStart_invlist);
34aeb2e9 2623 }
22f0498f
KW
2624 return is_utf8_common(p, &PL_utf8_perl_idstart,
2625 "_Perl_IDStart", NULL);
34aeb2e9
KW
2626 case _CC_IDCONT:
2627 if (! PL_utf8_perl_idcont) {
22f0498f
KW
2628 PL_utf8_perl_idcont
2629 = _new_invlist_C_array(_Perl_IDCont_invlist);
34aeb2e9 2630 }
22f0498f
KW
2631 return is_utf8_common(p, &PL_utf8_perl_idcont,
2632 "_Perl_IDCont", NULL);
34aeb2e9
KW
2633 }
2634 }
2635
2636 /* idcont is the same as wordchar below 256 */
2637 if (classnum == _CC_IDCONT) {
2638 classnum = _CC_WORDCHAR;
2639 }
2640 else if (classnum == _CC_IDFIRST) {
2641 if (*p == '_') {
2642 return TRUE;
2643 }
2644 classnum = _CC_ALPHA;
2645 }
2646
2647 if (! use_locale) {
2648 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2649 return _generic_isCC(*p, classnum);
2650 }
922e8cb4 2651
34aeb2e9
KW
2652 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2653 }
2654 else {
2655 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2656 return isFOO_lc(classnum, *p);
2657 }
2658
2659 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2660 }
2661
2662 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2663}
2664
2665bool
da8c1a98
KW
2666Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2667 const U8 * const e)
2668{
2669 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2670
2671 assert(classnum < _FIRST_NON_SWASH_CC);
2672
2673 return is_utf8_common_with_len(p,
2674 e,
2675 &PL_utf8_swash_ptrs[classnum],
2676 swash_property_names[classnum],
2677 PL_XPosix_ptrs[classnum]);
2678}
2679
2680bool
da8c1a98
KW
2681Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2682{
2683 SV* invlist = NULL;
2684
2685 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2686
2687 if (! PL_utf8_perl_idstart) {
2688 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2689 }
2690 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2691 "_Perl_IDStart", invlist);
2692}
2693
2694bool
f2645549 2695Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2696{
f2645549 2697 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2698
2699 if (*p == '_')
2700 return TRUE;
f25ce844 2701 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2702}
2703
2704bool
da8c1a98
KW
2705Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2706{
2707 SV* invlist = NULL;
2708
2709 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2710
2711 if (! PL_utf8_perl_idcont) {
2712 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2713 }
2714 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2715 "_Perl_IDCont", invlist);
2716}
2717
2718bool
f2645549 2719Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2720{
f2645549 2721 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2722
f25ce844 2723 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2724}
2725
2726bool
f2645549 2727Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2728{
f2645549 2729 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2730
f25ce844 2731 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2732}
2733
2734bool
7dbf68d2
KW
2735Perl__is_utf8_mark(pTHX_ const U8 *p)
2736{
7dbf68d2
KW
2737 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2738
f25ce844 2739 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2740}
2741
b9992569 2742 /* change namve uv1 to 'from' */
6a4a25f4 2743STATIC UV
b9992569
KW
2744S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2745 SV **swashp, const char *normal, const char *special)
2746{
0134edef 2747 STRLEN len = 0;
7918f24d 2748
b9992569 2749 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2750
36eaa811
KW
2751 /* For code points that don't change case, we already know that the output
2752 * of this function is the unchanged input, so we can skip doing look-ups
2753 * for them. Unfortunately the case-changing code points are scattered
2754 * around. But there are some long consecutive ranges where there are no
2755 * case changing code points. By adding tests, we can eliminate the lookup
2756 * for all the ones in such ranges. This is currently done here only for
2757 * just a few cases where the scripts are in common use in modern commerce
2758 * (and scripts adjacent to those which can be included without additional
2759 * tests). */
2760
2761 if (uv1 >= 0x0590) {
2762 /* This keeps from needing further processing the code points most
2763 * likely to be used in the following non-cased scripts: Hebrew,
2764 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2765 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2766 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2767 if (uv1 < 0x10A0) {
2768 goto cases_to_self;
2769 }
2770
2771 /* The following largish code point ranges also don't have case
2772 * changes, but khw didn't think they warranted extra tests to speed
2773 * them up (which would slightly slow down everything else above them):
2774 * 1100..139F Hangul Jamo, Ethiopic
2775 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2776 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2777 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2778 * Combining Diacritical Marks Extended, Balinese,
2779 * Sundanese, Batak, Lepcha, Ol Chiki
2780 * 2000..206F General Punctuation
2781 */
2782
2783 if (uv1 >= 0x2D30) {
2784
2785 /* This keeps the from needing further processing the code points
2786 * most likely to be used in the following non-cased major scripts:
2787 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2788 *
2789 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2790 * event that Unicode eventually allocates the unused block as of
2791 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2792 * that the test suite will start having failures to alert you
2793 * should that happen) */
2794 if (uv1 < 0xA640) {
2795 goto cases_to_self;
2796 }
2797
2798 if (uv1 >= 0xAC00) {
2799 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2800 if (ckWARN_d(WARN_SURROGATE)) {
2801 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2802 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2803 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2804 }
2805 goto cases_to_self;
2806 }
36eaa811
KW
2807
2808 /* AC00..FAFF Catches Hangul syllables and private use, plus
2809 * some others */
2810 if (uv1 < 0xFB00) {
2811 goto cases_to_self;
2812
2813 }
2814
5af9bc97 2815 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
13f4dd34
A
2816 if (UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)) {
2817 Perl_croak(aTHX_ cp_above_legal_max, uv1,
2818 MAX_NON_DEPRECATED_CP);
5af9bc97
KW
2819 }
2820 if (ckWARN_d(WARN_NON_UNICODE)) {
2821 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2822 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2823 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2824 }
2825 goto cases_to_self;
2826 }
3bfc1e70
KW
2827#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2828 if (UNLIKELY(uv1
2829 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2830 {
2831
2832 /* As of this writing, this means we avoid swash creation
2833 * for anything beyond low Plane 1 */
2834 goto cases_to_self;
2835 }
2836#endif
36eaa811
KW
2837 }
2838 }
9ae3ac1a 2839
36eaa811
KW
2840 /* Note that non-characters are perfectly legal, so no warning should
2841 * be given. There are so few of them, that it isn't worth the extra
2842 * tests to avoid swash creation */
9ae3ac1a
KW
2843 }
2844
0134edef 2845 if (!*swashp) /* load on-demand */
5ab9d2ef 2846 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2847
a6f87d8c 2848 if (special) {
0134edef 2849 /* It might be "special" (sometimes, but not always,
2a37f04d 2850 * a multicharacter mapping) */
4a8240a3 2851 HV *hv = NULL;
b08cf34e
JH
2852 SV **svp;
2853
4a8240a3
KW
2854 /* If passed in the specials name, use that; otherwise use any
2855 * given in the swash */
2856 if (*special != '\0') {
2857 hv = get_hv(special, 0);
2858 }
2859 else {
2860 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2861 if (svp) {
2862 hv = MUTABLE_HV(SvRV(*svp));
2863 }
2864 }
2865
176fe009 2866 if (hv
5f560d8a 2867 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2868 && (*svp))
2869 {
cfd0369c 2870 const char *s;
47654450 2871
cfd0369c 2872 s = SvPV_const(*svp, len);
47654450 2873 if (len == 1)
f4cd282c 2874 /* EIGHTBIT */
c80e42f3 2875 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2876 else {
d2dcd0fb 2877 Copy(s, ustrp, len, U8);
29e98929 2878 }
983ffd37 2879 }
0134edef
JH
2880 }
2881
2882 if (!len && *swashp) {
4a4088c4 2883 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2884
0134edef
JH
2885 if (uv2) {
2886 /* It was "normal" (a single character mapping). */
f4cd282c 2887 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2888 }
2889 }
1feea2c7 2890
cbe07460
KW
2891 if (len) {
2892 if (lenp) {
2893 *lenp = len;
2894 }
2895 return valid_utf8_to_uvchr(ustrp, 0);
2896 }
2897
2898 /* Here, there was no mapping defined, which means that the code point maps
2899 * to itself. Return the inputs */
e24dfe9c 2900 cases_to_self:
bfdf22ec 2901 len = UTF8SKIP(p);
ca9fab46
KW
2902 if (p != ustrp) { /* Don't copy onto itself */
2903 Copy(p, ustrp, len, U8);
2904 }
0134edef 2905
2a37f04d
JH
2906 if (lenp)
2907 *lenp = len;
2908
f4cd282c 2909 return uv1;
cbe07460 2910
a0ed51b3
LW
2911}
2912
051a06d4 2913STATIC UV
357aadde 2914S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2915{
4a4088c4 2916 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2917 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2918 * result contains a character that crosses the 255/256 boundary, disallow
2919 * the change, and return the original code point. See L<perlfunc/lc> for
2920 * why;
051a06d4 2921 *
a1433954
KW
2922 * p points to the original string whose case was changed; assumed
2923 * by this routine to be well-formed
051a06d4
KW
2924 * result the code point of the first character in the changed-case string
2925 * ustrp points to the changed-case string (<result> represents its first char)
2926 * lenp points to the length of <ustrp> */
2927
2928 UV original; /* To store the first code point of <p> */
2929
2930 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2931
a4f12ed7 2932 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2933
2934 /* We know immediately if the first character in the string crosses the
2935 * boundary, so can skip */
2936 if (result > 255) {
2937
2938 /* Look at every character in the result; if any cross the
2939 * boundary, the whole thing is disallowed */
2940 U8* s = ustrp + UTF8SKIP(ustrp);
2941 U8* e = ustrp + *lenp;
2942 while (s < e) {
a4f12ed7 2943 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2944 goto bad_crossing;
2945 }
2946 s += UTF8SKIP(s);
2947 }
2948
613abc6d
KW
2949 /* Here, no characters crossed, result is ok as-is, but we warn. */
2950 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2951 return result;
2952 }
2953
7b52d656 2954 bad_crossing:
051a06d4
KW
2955
2956 /* Failed, have to return the original */
4b88fb76 2957 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2958
2959 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2960 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2961 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2962 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2963 OP_DESC(PL_op),
ab0b796c
KW
2964 original,
2965 original);
051a06d4
KW
2966 Copy(p, ustrp, *lenp, char);
2967 return original;
2968}
2969
607313a1
KW
2970STATIC U32
2971S_check_and_deprecate(pTHX_ const U8 *p,
2972 const U8 **e,
2973 const unsigned int type, /* See below */
2974 const bool use_locale, /* Is this a 'LC_'
2975 macro call? */
2976 const char * const file,
2977 const unsigned line)
2978{
2979 /* This is a temporary function to deprecate the unsafe calls to the case
2980 * changing macros and functions. It keeps all the special stuff in just
2981 * one place.
2982 *
2983 * It updates *e with the pointer to the end of the input string. If using
2984 * the old-style macros, *e is NULL on input, and so this function assumes
2985 * the input string is long enough to hold the entire UTF-8 sequence, and
2986 * sets *e accordingly, but it then returns a flag to pass the
2987 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
2988 * using the full length if possible.
2989 *
2990 * It also does the assert that *e > p when *e is not NULL. This should be
2991 * migrated to the callers when this function gets deleted.
2992 *
2993 * The 'type' parameter is used for the caller to specify which case
2994 * changing function this is called from: */
2995
2996# define DEPRECATE_TO_UPPER 0
2997# define DEPRECATE_TO_TITLE 1
2998# define DEPRECATE_TO_LOWER 2
2999# define DEPRECATE_TO_FOLD 3
3000
3001 U32 utf8n_flags = 0;
3002 const char * name;
3003 const char * alternative;
3004
3005 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3006
3007 if (*e == NULL) {
3008 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3009 *e = p + UTF8SKIP(p);
3010
3011 /* For mathoms.c calls, we use the function name we know is stored
c44e9413 3012 * there. It could be part of a larger path */
607313a1 3013 if (type == DEPRECATE_TO_UPPER) {
c44e9413 3014 name = instr(file, "mathoms.c")
607313a1
KW
3015 ? "to_utf8_upper"
3016 : "toUPPER_utf8";
3017 alternative = "toUPPER_utf8_safe";
3018 }
3019 else if (type == DEPRECATE_TO_TITLE) {
c44e9413 3020 name = instr(file, "mathoms.c")
607313a1
KW
3021 ? "to_utf8_title"
3022 : "toTITLE_utf8";
3023 alternative = "toTITLE_utf8_safe";
3024 }
3025 else if (type == DEPRECATE_TO_LOWER) {
c44e9413 3026 name = instr(file, "mathoms.c")
607313a1
KW
3027 ? "to_utf8_lower"
3028 : "toLOWER_utf8";
3029 alternative = "toLOWER_utf8_safe";
3030 }
3031 else if (type == DEPRECATE_TO_FOLD) {
c44e9413 3032 name = instr(file, "mathoms.c")
607313a1
KW
3033 ? "to_utf8_fold"
3034 : "toFOLD_utf8";
3035 alternative = "toFOLD_utf8_safe";
3036 }
3037 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3038
3039 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3040 }
3041 else {
3042 assert (p < *e);
3043 }
3044
3045 return utf8n_flags;
3046}
3047
eaf412bf
KW
3048/* The process for changing the case is essentially the same for the four case
3049 * change types, except there are complications for folding. Otherwise the
3050 * difference is only which case to change to. To make sure that they all do
3051 * the same thing, the bodies of the functions are extracted out into the
3052 * following two macros. The functions are written with the same variable
3053 * names, and these are known and used inside these macros. It would be
3054 * better, of course, to have inline functions to do it, but since different
3055 * macros are called, depending on which case is being changed to, this is not
3056 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3057 * function can start with the common start macro, then finish with its special
3058 * handling; while the other three cases can just use the common end macro.
3059 *
3060 * The algorithm is to use the proper (passed in) macro or function to change
3061 * the case for code points that are below 256. The macro is used if using
3062 * locale rules for the case change; the function if not. If the code point is
3063 * above 255, it is computed from the input UTF-8, and another macro is called
3064 * to do the conversion. If necessary, the output is converted to UTF-8. If
3065 * using a locale, we have to check that the change did not cross the 255/256
3066 * boundary, see check_locale_boundary_crossing() for further details.
3067 *
3068 * The macros are split with the correct case change for the below-256 case
3069 * stored into 'result', and in the middle of an else clause for the above-255
3070 * case. At that point in the 'else', 'result' is not the final result, but is
3071 * the input code point calculated from the UTF-8. The fold code needs to
3072 * realize all this and take it from there.
3073 *
3074 * If you read the two macros as sequential, it's easier to understand what's
3075 * going on. */
3076#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3077 L1_func_extra_param) \
a239b1e2 3078 \
eaf412bf
KW
3079 if (flags & (locale_flags)) { \
3080 /* Treat a UTF-8 locale as not being in locale at all */ \
3081 if (IN_UTF8_CTYPE_LOCALE) { \
3082 flags &= ~(locale_flags); \
3083 } \
3084 else { \
3085 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3086 } \
3087 } \
3088 \
3089 if (UTF8_IS_INVARIANT(*p)) { \
3090 if (flags & (locale_flags)) { \
3091 result = LC_L1_change_macro(*p); \
3092 } \
3093 else { \
3094 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3095 } \
3096 } \
a239b1e2 3097 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3098 if (flags & (locale_flags)) { \
3099 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3100 *(p+1))); \
3101 } \
3102 else { \
3103 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3104 ustrp, lenp, L1_func_extra_param); \
3105 } \
3106 } \
fa8ab374
KW
3107 else { /* malformed UTF-8 or ord above 255 */ \
3108 STRLEN len_result; \
fa8ab374
KW
3109 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3110 if (len_result == (STRLEN) -1) { \
607313a1
KW
3111 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3112 1 /* Die */ ); \
fa8ab374 3113 }
eaf412bf
KW
3114
3115#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3116 result = change_macro(result, p, ustrp, lenp); \
3117 \
3118 if (flags & (locale_flags)) { \
3119 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3120 } \
3121 return result; \
3122 } \
3123 \
3124 /* Here, used locale rules. Convert back to UTF-8 */ \
3125 if (UTF8_IS_INVARIANT(result)) { \
3126 *ustrp = (U8) result; \
3127 *lenp = 1; \
3128 } \
3129 else { \
3130 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3131 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3132 *lenp = 2; \
3133 } \
3134 \
3135 return result;
3136
d3e79532 3137/*
87cea99e 3138=for apidoc to_utf8_upper
d3e79532 3139
a239b1e2 3140Instead use L</toUPPER_utf8_safe>.
a1433954 3141
d3e79532
JH
3142=cut */
3143
051a06d4 3144/* Not currently externally documented, and subject to change:
31f05a37
KW
3145 * <flags> is set iff iff the rules from the current underlying locale are to
3146 * be used. */
051a06d4 3147
2104c8d9 3148UV
607313a1
KW
3149Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3150 const U8 *e,
3151 U8* ustrp,
3152 STRLEN *lenp,
3153 bool flags,
3154 const char * const file,
3155 const int line)
a0ed51b3 3156{
051a06d4 3157 UV result;
607313a1
KW
3158 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3159 cBOOL(flags), file, line);
051a06d4
KW
3160
3161 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3162
eaf412bf
KW
3163 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3164 /* 2nd char of uc(U+DF) is 'S' */
3165 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3166 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3167}
a0ed51b3 3168
d3e79532 3169/*
87cea99e 3170=for apidoc to_utf8_title
d3e79532 3171
a239b1e2 3172Instead use L</toTITLE_utf8_safe>.
a1433954 3173
d3e79532
JH
3174=cut */
3175
051a06d4 3176/* Not currently externally documented, and subject to change:
31f05a37
KW
3177 * <flags> is set iff the rules from the current underlying locale are to be
3178 * used. Since titlecase is not defined in POSIX, for other than a
3179 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3180 */
051a06d4 3181
983ffd37 3182UV
607313a1
KW
3183Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3184 const U8 *e,
3185 U8* ustrp,
3186 STRLEN *lenp,
3187 bool flags,
3188 const char * const file,
3189 const int line)
983ffd37 3190{
051a06d4 3191 UV result;
607313a1
KW
3192 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3193 cBOOL(flags), file, line);
051a06d4
KW
3194
3195 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3196
eaf412bf
KW
3197 /* 2nd char of ucfirst(U+DF) is 's' */
3198 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3199 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3200}
3201
d3e79532 3202/*
87cea99e 3203=for apidoc to_utf8_lower
d3e79532 3204
a239b1e2 3205Instead use L</toLOWER_utf8_safe>.
a1433954 3206
d3e79532
JH
3207=cut */
3208
051a06d4 3209/* Not currently externally documented, and subject to change:
31f05a37
KW
3210 * <flags> is set iff iff the rules from the current underlying locale are to
3211 * be used.
3212 */
051a06d4 3213
2104c8d9 3214UV
607313a1
KW
3215Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3216 const U8 *e,
3217 U8* ustrp,
3218 STRLEN *lenp,
3219 bool flags,
3220 const char * const file,
3221 const int line)
a0ed51b3 3222{
051a06d4 3223 UV result;
607313a1
KW
3224 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3225 cBOOL(flags), file, line);
051a06d4 3226
051a06d4 3227 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3228
eaf412bf
KW
3229 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3230 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3231}
3232
d3e79532 3233/*
87cea99e 3234=for apidoc to_utf8_fold
d3e79532 3235
a239b1e2 3236Instead use L</toFOLD_utf8_safe>.
a1433954 3237
d3e79532
JH
3238=cut */
3239
051a06d4
KW
3240/* Not currently externally documented, and subject to change,
3241 * in <flags>
31f05a37
KW
3242 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3243 * locale are to be used.
051a06d4
KW
3244 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3245 * otherwise simple folds
a0270393
KW
3246 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3247 * prohibited
445bf929 3248 */
36bb2ab6 3249
b4e400f9 3250UV
607313a1
KW
3251Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3252 const U8 *e,
3253 U8* ustrp,
3254 STRLEN *lenp,
3255 U8 flags,
3256 const char * const file,
3257 const int line)
b4e400f9 3258{
051a06d4 3259 UV result;
607313a1
KW
3260 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3261 cBOOL(flags), file, line);
051a06d4 3262
36bb2ab6 3263 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3264
a0270393
KW
3265 /* These are mutually exclusive */
3266 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3267
50ba90ff
KW
3268 assert(p != ustrp); /* Otherwise overwrites */
3269
eaf412bf
KW
3270 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3271 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3272
eaf412bf 3273 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3274
1ca267a5
KW
3275 if (flags & FOLD_FLAGS_LOCALE) {
3276
76f2ffcd 3277# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3278 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3279
3280# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3281# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3282
3283 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3284
538e84ed
KW
3285 /* Special case these two characters, as what normally gets
3286 * returned under locale doesn't work */
76f2ffcd
KW
3287 if (UTF8SKIP(p) == cap_sharp_s_len
3288 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3289 {
ab0b796c
KW
3290 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3291 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3292 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3293 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3294 goto return_long_s;
3295 }
0766489e
KW
3296 else
3297#endif
3298 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3299 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3300 {
ab0b796c
KW
3301 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3302 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3303 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3304 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3305 goto return_ligature_st;
3306 }
74894415
KW
3307
3308#if UNICODE_MAJOR_VERSION == 3 \
3309 && UNICODE_DOT_VERSION == 0 \
3310 && UNICODE_DOT_DOT_VERSION == 1
3311# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3312
3313 /* And special case this on this Unicode version only, for the same
3314 * reaons the other two are special cased. They would cross the
3315 * 255/256 boundary which is forbidden under /l, and so the code
3316 * wouldn't catch that they are equivalent (which they are only in
3317 * this release) */
3318 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3319 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3320 {
3321 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3322 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3323 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3324 "resolved to \"\\x{0131}\".");
3325 goto return_dotless_i;
3326 }
3327#endif
3328
357aadde 3329 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3330 }
a0270393
KW
3331 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3332 return result;
3333 }
3334 else {
4a4088c4 3335 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3336 * character above the ASCII range, and the result should not
3337 * contain an ASCII character. */
a0270393
KW
3338
3339 UV original; /* To store the first code point of <p> */
3340
3341 /* Look at every character in the result; if any cross the
3342 * boundary, the whole thing is disallowed */
3343 U8* s = ustrp;
3344 U8* e = ustrp + *lenp;
3345 while (s < e) {
3346 if (isASCII(*s)) {
3347 /* Crossed, have to return the original */
3348 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3349
9fc2026f 3350 /* But in these instances, there is an alternative we can
1ca267a5 3351 * return that is valid */
0766489e
KW
3352 if (original == LATIN_SMALL_LETTER_SHARP_S
3353#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3354 || original == LATIN_CAPITAL_LETTER_SHARP_S
3355#endif
3356 ) {
1ca267a5
KW
3357 goto return_long_s;
3358 }
9fc2026f
KW
3359 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3360 goto return_ligature_st;
3361 }
74894415
KW
3362#if UNICODE_MAJOR_VERSION == 3 \
3363 && UNICODE_DOT_VERSION == 0 \
3364 && UNICODE_DOT_DOT_VERSION == 1
3365
3366 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3367 goto return_dotless_i;
3368 }
3369#endif
a0270393
KW
3370 Copy(p, ustrp, *lenp, char);
3371 return original;
3372 }
3373 s += UTF8SKIP(s);
3374 }
051a06d4 3375
a0270393
KW
3376 /* Here, no characters crossed, result is ok as-is */
3377 return result;
3378 }
051a06d4
KW
3379 }
3380
4a4088c4 3381 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3382 if (UTF8_IS_INVARIANT(result)) {
3383 *ustrp = (U8) result;
3384 *lenp = 1;
3385 }
3386 else {
62cb07ea
KW
3387 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3388 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3389 *lenp = 2;
3390 }
3391
051a06d4 3392 return result;
1ca267a5
KW
3393
3394 return_long_s:
3395 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3396 * folds to a string of two of these characters. By returning this
3397 * instead, then, e.g.,
3398 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3399 * works. */
3400
3401 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3402 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3403 ustrp, *lenp, U8);
3404 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3405
3406 return_ligature_st:
3407 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3408 * have the other one fold to it */
3409
3410 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3411 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3412 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3413
3414#if UNICODE_MAJOR_VERSION == 3 \
3415 && UNICODE_DOT_VERSION == 0 \
3416 && UNICODE_DOT_DOT_VERSION == 1
3417
3418 return_dotless_i:
3419 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3420 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3421 return LATIN_SMALL_LETTER_DOTLESS_I;
3422
3423#endif
3424
a0ed51b3
LW
3425}
3426
711a919c 3427/* Note:
f90a9a02 3428 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3429 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3430 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3431 */
c4a5db0c 3432
a0ed51b3 3433SV*
7fc63493 3434Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3435{
c4a5db0c
KW
3436 PERL_ARGS_ASSERT_SWASH_INIT;
3437
3438 /* Returns a copy of a swash initiated by the called function. This is the
3439 * public interface, and returning a copy prevents others from doing
3440 * mischief on the original */
3441
5d3d13d1 3442 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3443}
3444
3445SV*
5d3d13d1 3446Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3447{
2c1f00b9
YO
3448
3449 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3450 * use the following define */
3451
3452#define CORE_SWASH_INIT_RETURN(x) \
3453 PL_curpm= old_PL_curpm; \
3454 return x
3455
c4a5db0c 3456 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3457 * by calling utf8_heavy.pl in the general case. The returned value may be
3458 * the swash's inversion list instead if the input parameters allow it.
3459 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3460 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3461 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3462 *
3463 * This interface should only be used by functions that won't destroy or
3464 * adversely change the swash, as doing so affects all other uses of the
3465 * swash in the program; the general public should use 'Perl_swash_init'
3466 * instead.
3467 *
3468 * pkg is the name of the package that <name> should be in.
3469 * name is the name of the swash to find. Typically it is a Unicode
3470 * property name, including user-defined ones
3471 * listsv is a string to initialize the swash with. It must be of the form
3472 * documented as the subroutine return value in
3473 * L<perlunicode/User-Defined Character Properties>
3474 * minbits is the number of bits required to represent each data element.
3475 * It is '1' for binary properties.
3476 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3477 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3478 * flags_p if non-NULL is the address of various input and output flag bits
3479 * to the routine, as follows: ('I' means is input to the routine;
3480 * 'O' means output from the routine. Only flags marked O are
3481 * meaningful on return.)
3482 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3483 * came from a user-defined property. (I O)
5d3d13d1
KW
3484 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3485 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3486 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3487 * return of an inversion list instead of a swash hash if this routine
3488 * thinks that would result in faster execution of swash_fetch() later
3489 * on. (I)
9a53f6cf
KW
3490 *
3491 * Thus there are three possible inputs to find the swash: <name>,
3492 * <listsv>, and <invlist>. At least one must be specified. The result
3493 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3494 * actions can intersect, etc. what <name> gives. To avoid going out to
3495 * disk at all, <invlist> should specify completely what the swash should
3496 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3497 *
3498 * <invlist> is only valid for binary properties */
c4a5db0c 3499
2c1f00b9
YO
3500 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3501
c4a5db0c 3502 SV* retval = &PL_sv_undef;
83199d38 3503 HV* swash_hv = NULL;
87367d5f
KW
3504 const int invlist_swash_boundary =
3505 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3506 ? 512 /* Based on some benchmarking, but not extensive, see commit
3507 message */
3508 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3509
3510 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3511 assert(! invlist || minbits == 1);
3512
2c1f00b9
YO
3513 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3514 that triggered the swash init and the swash init perl logic itself.
3515 See perl #122747 */
3516
9a53f6cf
KW
3517 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3518 * so */
3519 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3520 dSP;
3521 const size_t pkg_len = strlen(pkg);
3522 const size_t name_len = strlen(name);
3523 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3524 SV* errsv_save;
3525 GV *method;
3526
3527 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3528
3529 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3530 ENTER;
69794297 3531 SAVEHINTS();
2782061f 3532 save_re_context();
650f067c
JL
3533 /* We might get here via a subroutine signature which uses a utf8
3534 * parameter name, at which point PL_subname will have been set
3535 * but not yet used. */
3536 save_item(PL_subname);
69794297
KW
3537 if (PL_parser && PL_parser->error_count)
3538 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3539 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3540 if (!method) { /* demand load UTF-8 */
69794297 3541 ENTER;
db2c6cb3
FC
3542 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3543 GvSV(PL_errgv) = NULL;
1a419e6b 3544#ifndef NO_TAINT_SUPPORT
69794297
KW
3545 /* It is assumed that callers of this routine are not passing in
3546 * any user derived data. */
2782061f
DM
3547 /* Need to do this after save_re_context() as it will set
3548 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3549 * in Perl_magic_get). Even line to create errsv_save can turn on
3550 * PL_tainted. */
284167a5
SM
3551 SAVEBOOL(TAINT_get);
3552 TAINT_NOT;
3553#endif
69794297
KW
3554 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3555 NULL);
eed484f9 3556 {
db2c6cb3
FC
3557 /* Not ERRSV, as there is no need to vivify a scalar we are
3558 about to discard. */
3559 SV * const errsv = GvSV(PL_errgv);
3560 if (!SvTRUE(errsv)) {
3561 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3562 SvREFCNT_dec(errsv);
3563 }
eed484f9 3564 }
69794297
KW
3565 LEAVE;
3566 }
3567 SPAGAIN;
3568 PUSHMARK(SP);
3569 EXTEND(SP,5);
3570 mPUSHp(pkg, pkg_len);
3571 mPUSHp(name, name_len);
3572 PUSHs(listsv);
3573 mPUSHi(minbits);
3574 mPUSHi(none);
3575 PUTBACK;
db2c6cb3
FC
3576 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3577 GvSV(PL_errgv) = NULL;
69794297
KW
3578 /* If we already have a pointer to the method, no need to use
3579 * call_method() to repeat the lookup. */
c41800a8
KW
3580 if (method
3581 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3582 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3583 {
3584 retval = *PL_stack_sp--;
3585 SvREFCNT_inc(retval);
3586 }
eed484f9 3587 {
db2c6cb3
FC
3588 /* Not ERRSV. See above. */
3589 SV * const errsv = GvSV(PL_errgv);
3590 if (!SvTRUE(errsv)) {
3591 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3592 SvREFCNT_dec(errsv);
3593 }
eed484f9 3594 }
ce3b816e 3595 LEAVE;
69794297
KW
3596 POPSTACK;
3597 if (IN_PERL_COMPILETIME) {
3598 CopHINTS_set(PL_curcop, PL_hints);
3599 }
3600 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3601 if (SvPOK(retval)) {
69794297
KW
3602
3603 /* If caller wants to handle missing properties, let them */
5d3d13d1 3604 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3605 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3606 }
3607 Perl_croak(aTHX_
147e3846 3608 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3609 SVfARG(retval));
a25b5927 3610 NOT_REACHED; /* NOTREACHED */
d95e4a00 3611 }
69794297 3612 }
9a53f6cf 3613 } /* End of calling the module to find the swash */
36eb48b4 3614
83199d38
KW
3615 /* If this operation fetched a swash, and we will need it later, get it */
3616 if (retval != &PL_sv_undef
3617 && (minbits == 1 || (flags_p
3618 && ! (*flags_p
3619 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3620 {
3621 swash_hv = MUTABLE_HV(SvRV(retval));
3622
3623 /* If we don't already know that there is a user-defined component to
3624 * this swash, and the user has indicated they wish to know if there is
3625 * one (by passing <flags_p>), find out */
3626 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3627 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3628 if (user_defined && SvUV(*user_defined)) {
3629 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3630 }
3631 }
3632 }
3633
36eb48b4
KW
3634 /* Make sure there is an inversion list for binary properties */
3635 if (minbits == 1) {
3636 SV** swash_invlistsvp = NULL;
3637 SV* swash_invlist = NULL;
9a53f6cf 3638 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3639 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3640 an unclaimed reference count */
36eb48b4 3641
9a53f6cf 3642 /* If this operation fetched a swash, get its already existing
83199d38 3643 * inversion list, or create one for it */
36eb48b4 3644
83199d38 3645 if (swash_hv) {
5c9f4bd2 3646 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3647 if (swash_invlistsvp) {
3648 swash_invlist = *swash_invlistsvp;
3649 invlist_in_swash_is_valid = TRUE;
3650 }
3651 else {
36eb48b4 3652 swash_invlist = _swash_to_invlist(retval);
02c85471 3653 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3654 }
3655 }
3656
3657 /* If an inversion list was passed in, have to include it */
3658 if (invlist) {
3659
3660 /* Any fetched swash will by now have an inversion list in it;
3661 * otherwise <swash_invlist> will be NULL, indicating that we
3662 * didn't fetch a swash */
3663 if (swash_invlist) {
3664
3665 /* Add the passed-in inversion list, which invalidates the one
3666 * already stored in the swash */
3667 invlist_in_swash_is_valid = FALSE;
eee4c920 3668 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3669 _invlist_union(invlist, swash_invlist, &swash_invlist);
3670 }
3671 else {
3672
87367d5f
KW
3673 /* Here, there is no swash already. Set up a minimal one, if
3674 * we are going to return a swash */
3675 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3676 swash_hv = newHV();
4aca0fe6 3677 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3678 }
9a53f6cf
KW
3679 swash_invlist = invlist;
3680 }
9a53f6cf
KW
3681 }
3682
3683 /* Here, we have computed the union of all the passed-in data. It may
3684 * be that there was an inversion list in the swash which didn't get
538e84ed 3685 * touched; otherwise save the computed one */
87367d5f
KW
3686 if (! invlist_in_swash_is_valid
3687 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3688 {
5c9f4bd2 3689 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3690 {
3691 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3692 }
cc34d8c5
FC
3693 /* We just stole a reference count. */
3694 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3695 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3696 }
87367d5f 3697
ee3222e3 3698 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3699 SvREADONLY_on(swash_invlist);
3700
c41800a8 3701 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3702 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3703 SvREFCNT_dec(retval);
02c85471
FC
3704 if (!swash_invlist_unclaimed)
3705 SvREFCNT_inc_simple_void_NN(swash_invlist);
3706 retval = newRV_noinc(swash_invlist);
87367d5f 3707 }
36eb48b4
KW
3708 }
3709
2c1f00b9
YO
3710 CORE_SWASH_INIT_RETURN(retval);
3711#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3712}
3713
035d37be
JH
3714
3715/* This API is wrong for special case conversions since we may need to
3716 * return several Unicode characters for a single Unicode character
3717 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3718 * the lower-level routine, and it is similarly broken for returning
38684baa 3719 * multiple values. --jhi
b9992569 3720 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3721/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3722
3723/* Note:
3724 * Returns the value of property/mapping C<swash> for the first character
3725 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3726 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3727 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3728 *
3729 * A "swash" is a hash which contains initially the keys/values set up by
3730 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3731 * property for all possible code points. Things are stored in a compact form
3732 * (see utf8_heavy.pl) so that calculation is required to find the actual
3733 * property value for a given code point. As code points are looked up, new
3734 * key/value pairs are added to the hash, so that the calculation doesn't have
3735 * to ever be re-done. Further, each calculation is done, not just for the