This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Convert core to use toFOO_utf8_safe()
[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[] =
147e3846 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf;
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
81 SAVESPTR(PL_dowarn);
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
8ee1cdcb
KW
119#define HANDLE_UNICODE_SURROGATE(uv, flags) \
120 STMT_START { \
121 if (flags & UNICODE_WARN_SURROGATE) { \
122 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
147e3846 123 "UTF-16 surrogate U+%04" UVXf, uv); \
8ee1cdcb
KW
124 } \
125 if (flags & UNICODE_DISALLOW_SURROGATE) { \
126 return NULL; \
127 } \
128 } STMT_END;
129
130#define HANDLE_UNICODE_NONCHAR(uv, flags) \
131 STMT_START { \
132 if (flags & UNICODE_WARN_NONCHAR) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
147e3846 134 "Unicode non-character U+%04" UVXf " is not " \
8ee1cdcb
KW
135 "recommended for open interchange", uv); \
136 } \
137 if (flags & UNICODE_DISALLOW_NONCHAR) { \
138 return NULL; \
139 } \
140 } STMT_END;
141
ba6ed43c
KW
142/* Use shorter names internally in this file */
143#define SHIFT UTF_ACCUMULATION_SHIFT
144#undef MARK
145#define MARK UTF_CONTINUATION_MARK
146#define MASK UTF_CONTINUATION_MASK
147
dfe13c55 148U8 *
378516de 149Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 150{
378516de 151 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 152
2d1545e5 153 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 154 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
155 return d;
156 }
facc1dc2 157
3ea68d71 158 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
159 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
160 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
161 return d;
162 }
d9432125 163
ba6ed43c
KW
164 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
165 * below, the 16 is for start bytes E0-EF (which are all the possible ones
166 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
167 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
168 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
169 * 0x800-0xFFFF on ASCII */
170 if (uv < (16 * (1U << (2 * SHIFT)))) {
171 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
172 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
173 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
174
175#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
176 aren't tested here */
177 /* The most likely code points in this range are below the surrogates.
178 * Do an extra test to quickly exclude those. */
179 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
180 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
181 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
182 {
8ee1cdcb
KW
183 HANDLE_UNICODE_NONCHAR(uv, flags);
184 }
185 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
186 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 187 }
ba6ed43c
KW
188 }
189#endif
190 return d;
191 }
192
193 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
194 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
195 * happen starting with 4-byte characters on ASCII platforms. We unify the
196 * code for these with EBCDIC, even though some of them require 5-bytes on
197 * those, because khw believes the code saving is worth the very slight
198 * performance hit on these high EBCDIC code points. */
199
200 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
201 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
202 && ckWARN_d(WARN_DEPRECATED))
203 {
204 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
205 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
206 }
207 if ( (flags & UNICODE_WARN_SUPER)
208 || ( UNICODE_IS_ABOVE_31_BIT(uv)
209 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
210 {
211 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
212
213 /* Choose the more dire applicable warning */
214 (UNICODE_IS_ABOVE_31_BIT(uv))
147e3846
KW
215 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
216 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
a5bf80e0
KW
217 uv);
218 }
219 if (flags & UNICODE_DISALLOW_SUPER
220 || ( UNICODE_IS_ABOVE_31_BIT(uv)
221 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
222 {
223 return NULL;
224 }
225 }
ba6ed43c
KW
226 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
227 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 228 }
d9432125 229
ba6ed43c
KW
230 /* Test for and handle 4-byte result. In the test immediately below, the
231 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
232 * characters). The 3 is for 3 continuation bytes; these each contribute
233 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
234 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
235 * 0x1_0000-0x1F_FFFF on ASCII */
236 if (uv < (8 * (1U << (3 * SHIFT)))) {
237 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
238 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
239 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
240 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
241
242#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
243 characters. The end-plane non-characters for EBCDIC were
244 handled just above */
245 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
246 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 247 }
ba6ed43c
KW
248 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
249 HANDLE_UNICODE_SURROGATE(uv, flags);
250 }
251#endif
252
253 return d;
254 }
255
256 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
257 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
258 * format. The unrolled version above turns out to not save all that much
259 * time, and at these high code points (well above the legal Unicode range
260 * on ASCII platforms, and well above anything in common use in EBCDIC),
261 * khw believes that less code outweighs slight performance gains. */
262
d9432125 263 {
5aaebcb3 264 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
265 U8 *p = d+len-1;
266 while (p > d) {
4c8cd605 267 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
268 uv >>= UTF_ACCUMULATION_SHIFT;
269 }
4c8cd605 270 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
271 return d+len;
272 }
a0ed51b3 273}
a5bf80e0 274
646ca15d 275/*
07693fe6
KW
276=for apidoc uvchr_to_utf8
277
bcb1a2d4 278Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 279of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
280C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
281the byte after the end of the new character. In other words,
07693fe6
KW
282
283 d = uvchr_to_utf8(d, uv);
284
285is the recommended wide native character-aware way of saying
286
287 *(d++) = uv;
288
760c7c2f
KW
289This function accepts any UV as input, but very high code points (above
290C<IV_MAX> on the platform) will raise a deprecation warning. This is
291typically 0x7FFF_FFFF in a 32-bit word.
292
293It is possible to forbid or warn on non-Unicode code points, or those that may
294be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 295
07693fe6
KW
296=cut
297*/
298
de69f3af
KW
299/* This is also a macro */
300PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
301
07693fe6
KW
302U8 *
303Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
304{
de69f3af 305 return uvchr_to_utf8(d, uv);
07693fe6
KW
306}
307
de69f3af
KW
308/*
309=for apidoc uvchr_to_utf8_flags
310
311Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 312of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
313C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
314the byte after the end of the new character. In other words,
de69f3af
KW
315
316 d = uvchr_to_utf8_flags(d, uv, flags);
317
318or, in most cases,
319
320 d = uvchr_to_utf8_flags(d, uv, 0);
321
322This is the Unicode-aware way of saying
323
324 *(d++) = uv;
325
760c7c2f
KW
326If C<flags> is 0, this function accepts any UV as input, but very high code
327points (above C<IV_MAX> for the platform) will raise a deprecation warning.
328This is typically 0x7FFF_FFFF in a 32-bit word.
329
330Specifying C<flags> can further restrict what is allowed and not warned on, as
331follows:
de69f3af 332
796b6530 333If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
334the function will raise a warning, provided UTF8 warnings are enabled. If
335instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
336NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 337
760c7c2f
KW
338Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
339affect how the function handles a Unicode non-character.
93e6dbd6 340
760c7c2f
KW
341And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
342affect the handling of code points that are above the Unicode maximum of
3430x10FFFF. Languages other than Perl may not be able to accept files that
344contain these.
93e6dbd6
KW
345
346The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
347the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
348three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
349allowed inputs to the strict UTF-8 traditionally defined by Unicode.
350Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
351C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
352above-Unicode and surrogate flags, but not the non-character ones, as
353defined in
354L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
355See L<perlunicode/Noncharacter code points>.
93e6dbd6 356
ab8e6d41
KW
357Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
358so using them is more problematic than other above-Unicode code points. Perl
359invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
360likely that non-Perl languages will not be able to read files that contain
361these that written by the perl interpreter; nor would Perl understand files
362written by something that uses a different extension. For these reasons, there
363is a separate set of flags that can warn and/or disallow these extremely high
364code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
365C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
366are entirely independent from the deprecation warning for code points above
367C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
368code point that needs more than 31 bits to represent. When that happens,
369effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
37032-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
371above-Unicode code points, including these, as malformations; and
372C<UNICODE_WARN_SUPER> warns on these.)
373
374On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
375extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
376than on ASCII. Prior to that, code points 2**31 and higher were simply
377unrepresentable, and a different, incompatible method was used to represent
378code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
379and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
380platforms, warning and disallowing 2**31 and higher.
de69f3af 381
de69f3af
KW
382=cut
383*/
384
385/* This is also a macro */
386PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
387
07693fe6
KW
388U8 *
389Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
390{
de69f3af 391 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
392}
393
83dc0f42
KW
394PERL_STATIC_INLINE bool
395S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
396{
397 /* Returns TRUE if the first code point represented by the Perl-extended-
398 * UTF-8-encoded string starting at 's', and looking no further than 'e -
399 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
400 *
401 * The function handles the case where the input bytes do not include all
402 * the ones necessary to represent a full character. That is, they may be
403 * the intial bytes of the representation of a code point, but possibly
404 * the final ones necessary for the complete representation may be beyond
405 * 'e - 1'.
406 *
407 * The function assumes that the sequence is well-formed UTF-8 as far as it
408 * goes, and is for a UTF-8 variant code point. If the sequence is
409 * incomplete, the function returns FALSE if there is any well-formed
410 * UTF-8 byte sequence that can complete it in such a way that a code point
411 * < 2**31 is produced; otherwise it returns TRUE.
412 *
413 * Getting this exactly right is slightly tricky, and has to be done in
414 * several places in this file, so is centralized here. It is based on the
415 * following table:
416 *
417 * U+7FFFFFFF (2 ** 31 - 1)
418 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
419 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
420 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
421 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
422 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
423 * U+80000000 (2 ** 31):
424 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
425 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
426 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
428 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
429 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
430 */
431
432#ifdef EBCDIC
433
37086697
KW
434 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
435 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
436 const STRLEN prefix_len = sizeof(prefix) - 1;
437 const STRLEN len = e - s;
f880f78a 438 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
439
440#else
441
442 PERL_UNUSED_ARG(e);
443
444#endif
445
446 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
447
448 assert(! UTF8_IS_INVARIANT(*s));
449
450#ifndef EBCDIC
451
452 /* Technically, a start byte of FE can be for a code point that fits into
453 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
454 * malformation. */
455 return (*s >= 0xFE);
456
457#else
458
459 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
460 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
461 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
462 * bits. */
463 if (*s != 0xFE || len == 1) {
464 return FALSE;
465 }
466
467 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
468 * \x41 and \x42. */
469 return cBOOL(memGT(s + 1, prefix, cmp_len));
470
471#endif
472
473}
474
12a4bed3
KW
475PERL_STATIC_INLINE bool
476S_does_utf8_overflow(const U8 * const s, const U8 * e)
477{
478 const U8 *x;
479 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
480
c551bb01
KW
481#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
482
483 const STRLEN len = e - s;
484
485#endif
486
12a4bed3
KW
487 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
488 * platform, that is if it represents a code point larger than the highest
489 * representable code point. (For ASCII platforms, we could use memcmp()
490 * because we don't have to convert each byte to I8, but it's very rare
491 * input indeed that would approach overflow, so the loop below will likely
492 * only get executed once.
493 *
494 * 'e' must not be beyond a full character. If it is less than a full
495 * character, the function returns FALSE if there is any input beyond 'e'
496 * that could result in a non-overflowing code point */
497
498 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
499 assert(s <= e && s + UTF8SKIP(s) >= e);
500
501#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
502
503 /* On 32 bit ASCII machines, many overlongs that start with FF don't
504 * overflow */
505
c551bb01 506 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
507 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
508 return memGE(s, max_32_bit_overlong,
c551bb01 509 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
510 }
511
512#endif
12a4bed3
KW
513
514 for (x = s; x < e; x++, y++) {
515
516 /* If this byte is larger than the corresponding highest UTF-8 byte, it
517 * overflows */
518 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
519 return TRUE;
520 }
521
522 /* If not the same as this byte, it must be smaller, doesn't overflow */
523 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
524 return FALSE;
525 }
526 }
527
528 /* Got to the end and all bytes are the same. If the input is a whole
529 * character, it doesn't overflow. And if it is a partial character,
530 * there's not enough information to tell, so assume doesn't overflow */
531 return FALSE;
532}
533
534PERL_STATIC_INLINE bool
535S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
536{
537 /* Overlongs can occur whenever the number of continuation bytes
538 * changes. That means whenever the number of leading 1 bits in a start
539 * byte increases from the next lower start byte. That happens for start
540 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
541 * illegal start bytes have already been excluded, so don't need to be
542 * tested here;
543 * ASCII platforms: C0, C1
544 * EBCDIC platforms C0, C1, C2, C3, C4, E0
545 *
546 * At least a second byte is required to determine if other sequences will
547 * be an overlong. */
548
549 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
550 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
551
552 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
553 assert(len > 1 && UTF8_IS_START(*s));
554
555 /* Each platform has overlongs after the start bytes given above (expressed
556 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
557 * the logic is the same, except the E0 overlong has already been excluded
558 * on EBCDIC platforms. The values below were found by manually
559 * inspecting the UTF-8 patterns. See the tables in utf8.h and
560 * utfebcdic.h. */
561
562# ifdef EBCDIC
563# define F0_ABOVE_OVERLONG 0xB0
564# define F8_ABOVE_OVERLONG 0xA8
565# define FC_ABOVE_OVERLONG 0xA4
566# define FE_ABOVE_OVERLONG 0xA2
567# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
568 /* I8(0xfe) is FF */
569# else
570
571 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
572 return TRUE;
573 }
574
575# define F0_ABOVE_OVERLONG 0x90
576# define F8_ABOVE_OVERLONG 0x88
577# define FC_ABOVE_OVERLONG 0x84
578# define FE_ABOVE_OVERLONG 0x82
579# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
580# endif
581
582
583 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
584 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
585 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
586 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
587 {
588 return TRUE;
589 }
590
b0b342d4
KW
591 /* Check for the FF overlong */
592 return isFF_OVERLONG(s, len);
593}
594
595PERL_STATIC_INLINE bool
596S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
597{
598 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
599
600 /* Check for the FF overlong. This happens only if all these bytes match;
601 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 602 * utfebcdic.h. */
12a4bed3 603
b0b342d4
KW
604 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
605 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
606 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
607}
608
609#undef F0_ABOVE_OVERLONG
610#undef F8_ABOVE_OVERLONG
611#undef FC_ABOVE_OVERLONG
612#undef FE_ABOVE_OVERLONG
613#undef FF_OVERLONG_PREFIX
614
35f8c9bd 615STRLEN
edc2c47a 616Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 617{
2b479609 618 STRLEN len;
12a4bed3 619 const U8 *x;
35f8c9bd 620
2b479609
KW
621 /* A helper function that should not be called directly.
622 *
623 * This function returns non-zero if the string beginning at 's' and
624 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
625 * code point; otherwise it returns 0. The examination stops after the
626 * first code point in 's' is validated, not looking at the rest of the
627 * input. If 'e' is such that there are not enough bytes to represent a
628 * complete code point, this function will return non-zero anyway, if the
629 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
630 * excluded by 'flags'.
631 *
632 * A non-zero return gives the number of bytes required to represent the
633 * code point. Be aware that if the input is for a partial character, the
634 * return will be larger than 'e - s'.
635 *
636 * This function assumes that the code point represented is UTF-8 variant.
637 * The caller should have excluded this possibility before calling this
638 * function.
639 *
640 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
641 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
642 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
643 * disallowed by the flags. If the input is only for a partial character,
644 * the function will return non-zero if there is any sequence of
645 * well-formed UTF-8 that, when appended to the input sequence, could
646 * result in an allowed code point; otherwise it returns 0. Non characters
647 * cannot be determined based on partial character input. But many of the
648 * other excluded types can be determined with just the first one or two
649 * bytes.
650 *
651 */
652
653 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
654
655 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
656 |UTF8_DISALLOW_ABOVE_31_BIT)));
657 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 658
2b479609 659 /* A variant char must begin with a start byte */
35f8c9bd
KW
660 if (UNLIKELY(! UTF8_IS_START(*s))) {
661 return 0;
662 }
663
edc2c47a
KW
664 /* Examine a maximum of a single whole code point */
665 if (e - s > UTF8SKIP(s)) {
666 e = s + UTF8SKIP(s);
667 }
668
2b479609
KW
669 len = e - s;
670
671 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
672 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 673
2b479609
KW
674 /* The code below is derived from this table. Keep in mind that legal
675 * continuation bytes range between \x80..\xBF for UTF-8, and
676 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
677 * Hence, we don't have to test the upper edge because if any of those
678 * are encountered, the sequence is malformed, and will fail elsewhere
679 * in this function.
680 * UTF-8 UTF-EBCDIC I8
681 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
682 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
683 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
684 *
685 */
686
687#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
688# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 689# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 690
19794540
KW
691# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
692 /* B6 and B7 */ \
693 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
694#else
695# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
696# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
697# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
698#endif
699
700 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
702 {
2b479609
KW
703 return 0; /* Above Unicode */
704 }
705
706 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
707 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
708 {
709 return 0; /* Above 31 bits */
710 }
711
712 if (len > 1) {
713 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
714
715 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 716 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
717 {
718 return 0; /* Above Unicode */
719 }
720
721 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 722 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
723 {
724 return 0; /* Surrogate */
725 }
726
727 if ( (flags & UTF8_DISALLOW_NONCHAR)
728 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
729 {
730 return 0; /* Noncharacter code point */
731 }
732 }
733 }
734
735 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
736 for (x = s + 1; x < e; x++) {
737 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
738 return 0;
739 }
740 }
741
af13dd8a 742 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
743 * overlong. */
744 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
745 return 0;
af13dd8a
KW
746 }
747
12a4bed3
KW
748 /* And finally, that the code point represented fits in a word on this
749 * platform */
750 if (does_utf8_overflow(s, e)) {
751 return 0;
35f8c9bd
KW
752 }
753
2b479609 754 return UTF8SKIP(s);
35f8c9bd
KW
755}
756
7cf8d05d
KW
757STATIC char *
758S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
759{
760 /* Returns a mortalized C string that is a displayable copy of the 'len'
761 * bytes starting at 's', each in a \xXY format. */
762
763 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
764 trailing NUL */
765 const U8 * const e = s + len;
766 char * output;
767 char * d;
768
769 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
770
771 Newx(output, output_len, char);
772 SAVEFREEPV(output);
773
774 d = output;
775 for (; s < e; s++) {
776 const unsigned high_nibble = (*s & 0xF0) >> 4;
777 const unsigned low_nibble = (*s & 0x0F);
778
779 *d++ = '\\';
780 *d++ = 'x';
781
782 if (high_nibble < 10) {
783 *d++ = high_nibble + '0';
784 }
785 else {
786 *d++ = high_nibble - 10 + 'a';
787 }
788
789 if (low_nibble < 10) {
790 *d++ = low_nibble + '0';
791 }
792 else {
793 *d++ = low_nibble - 10 + 'a';
794 }
795 }
796
797 *d = '\0';
798 return output;
799}
800
806547a7 801PERL_STATIC_INLINE char *
7cf8d05d
KW
802S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
803
804 /* How many bytes to print */
3cc6a05e 805 STRLEN print_len,
7cf8d05d
KW
806
807 /* Which one is the non-continuation */
808 const STRLEN non_cont_byte_pos,
809
810 /* How many bytes should there be? */
811 const STRLEN expect_len)
806547a7
KW
812{
813 /* Return the malformation warning text for an unexpected continuation
814 * byte. */
815
7cf8d05d 816 const char * const where = (non_cont_byte_pos == 1)
806547a7 817 ? "immediately"
7cf8d05d
KW
818 : Perl_form(aTHX_ "%d bytes",
819 (int) non_cont_byte_pos);
806547a7
KW
820
821 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
822
7cf8d05d
KW
823 /* We don't need to pass this parameter, but since it has already been
824 * calculated, it's likely faster to pass it; verify under DEBUGGING */
825 assert(expect_len == UTF8SKIP(s));
826
827 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
828 " %s after start byte 0x%02x; need %d bytes, got %d)",
829 malformed_text,
830 _byte_dump_string(s, print_len),
831 *(s + non_cont_byte_pos),
832 where,
833 *s,
834 (int) expect_len,
835 (int) non_cont_byte_pos);
806547a7
KW
836}
837
35f8c9bd
KW
838/*
839
de69f3af 840=for apidoc utf8n_to_uvchr
378516de
KW
841
842THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 843Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 844
9041c2e3 845Bottom level UTF-8 decode routine.
de69f3af 846Returns the native code point value of the first character in the string C<s>,
746afd53
KW
847which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
848C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
849the length, in bytes, of that character.
949cf498
KW
850
851The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
852well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
853causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
854is the next possible position in C<s> that could begin a non-malformed
855character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
856is raised. Some UTF-8 input sequences may contain multiple malformations.
857This function tries to find every possible one in each call, so multiple
858warnings can be raised for each sequence.
949cf498
KW
859
860Various ALLOW flags can be set in C<flags> to allow (and not warn on)
861individual types of malformations, such as the sequence being overlong (that
862is, when there is a shorter sequence that can express the same code point;
863overlong sequences are expressly forbidden in the UTF-8 standard due to
864potential security issues). Another malformation example is the first byte of
865a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
866flags. Even if allowed, this function generally returns the Unicode
867REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
868F<utf8.h> to override this behavior for the overlong malformations, but don't
869do that except for very specialized purposes.
949cf498 870
796b6530 871The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
872flags) malformation is found. If this flag is set, the routine assumes that
873the caller will raise a warning, and this function will silently just set
d088425d
KW
874C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
875
75200dff 876Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 877character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
878in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
879be set to 1. To disambiguate, upon a zero return, see if the first byte of
880C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 881error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
882
883Certain code points are considered problematic. These are Unicode surrogates,
746afd53 884Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 885By default these are considered regular code points, but certain situations
ecc1615f
KW
886warrant special handling for them, which can be specified using the C<flags>
887parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
888three classes are treated as malformations and handled as such. The flags
889C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
890C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
891disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
892restricts the allowed inputs to the strict UTF-8 traditionally defined by
893Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
894definition given by
895L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
896The difference between traditional strictness and C9 strictness is that the
897latter does not forbid non-character code points. (They are still discouraged,
898however.) For more discussion see L<perlunicode/Noncharacter code points>.
899
900The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
901C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
902C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
903raised for their respective categories, but otherwise the code points are
904considered valid (not malformations). To get a category to both be treated as
905a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 906(But note that warnings are not raised if lexically disabled nor if
796b6530 907C<UTF8_CHECK_ONLY> is also specified.)
949cf498 908
760c7c2f
KW
909It is now deprecated to have very high code points (above C<IV_MAX> on the
910platforms) and this function will raise a deprecation warning for these (unless
d5944cab 911such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 912in a 32-bit word.
ab8e6d41
KW
913
914Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
915so using them is more problematic than other above-Unicode code points. Perl
916invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
917likely that non-Perl languages will not be able to read files that contain
aff2be59 918these; nor would Perl understand files
ab8e6d41
KW
919written by something that uses a different extension. For these reasons, there
920is a separate set of flags that can warn and/or disallow these extremely high
921code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
922C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
923are entirely independent from the deprecation warning for code points above
924C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
925code point that needs more than 31 bits to represent. When that happens,
926effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
92732-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
928above-Unicode code points, including these, as malformations; and
929C<UTF8_WARN_SUPER> warns on these.)
930
931On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
932extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
933than on ASCII. Prior to that, code points 2**31 and higher were simply
934unrepresentable, and a different, incompatible method was used to represent
935code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
936and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
937platforms, warning and disallowing 2**31 and higher.
949cf498
KW
938
939All other code points corresponding to Unicode characters, including private
940use and those yet to be assigned, are never considered malformed and never
941warn.
67e989fb 942
37607a96 943=cut
f9380377
KW
944
945Also implemented as a macro in utf8.h
946*/
947
948UV
949Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
950 STRLEN curlen,
951 STRLEN *retlen,
952 const U32 flags)
953{
954 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
955
956 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
957}
958
959/*
960
961=for apidoc utf8n_to_uvchr_error
962
963THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
964Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
965
966This function is for code that needs to know what the precise malformation(s)
967are when an error is found.
968
969It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
970all the others, C<errors>. If this parameter is 0, this function behaves
971identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
972to a C<U32> variable, which this function sets to indicate any errors found.
973Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
974C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
975of these bits will be set if a malformation is found, even if the input
7a65503b 976C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
977exceptions are noted:
978
979=over 4
980
981=item C<UTF8_GOT_ABOVE_31_BIT>
982
983The code point represented by the input UTF-8 sequence occupies more than 31
984bits.
985This bit is set only if the input C<flags> parameter contains either the
986C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
987
988=item C<UTF8_GOT_CONTINUATION>
989
990The input sequence was malformed in that the first byte was a a UTF-8
991continuation byte.
992
993=item C<UTF8_GOT_EMPTY>
994
995The input C<curlen> parameter was 0.
996
997=item C<UTF8_GOT_LONG>
998
999The input sequence was malformed in that there is some other sequence that
1000evaluates to the same code point, but that sequence is shorter than this one.
1001
1002=item C<UTF8_GOT_NONCHAR>
1003
1004The code point represented by the input UTF-8 sequence is for a Unicode
1005non-character code point.
1006This bit is set only if the input C<flags> parameter contains either the
1007C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1008
1009=item C<UTF8_GOT_NON_CONTINUATION>
1010
1011The input sequence was malformed in that a non-continuation type byte was found
1012in a position where only a continuation type one should be.
1013
1014=item C<UTF8_GOT_OVERFLOW>
1015
1016The input sequence was malformed in that it is for a code point that is not
1017representable in the number of bits available in a UV on the current platform.
1018
1019=item C<UTF8_GOT_SHORT>
1020
1021The input sequence was malformed in that C<curlen> is smaller than required for
1022a complete sequence. In other words, the input is for a partial character
1023sequence.
1024
1025=item C<UTF8_GOT_SUPER>
1026
1027The input sequence was malformed in that it is for a non-Unicode code point;
1028that is, one above the legal Unicode maximum.
1029This bit is set only if the input C<flags> parameter contains either the
1030C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1031
1032=item C<UTF8_GOT_SURROGATE>
1033
1034The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1035code point.
1036This bit is set only if the input C<flags> parameter contains either the
1037C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1038
1039=back
1040
133551d8
KW
1041To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1042flag to suppress any warnings, and then examine the C<*errors> return.
1043
f9380377 1044=cut
37607a96 1045*/
67e989fb 1046
a0ed51b3 1047UV
f9380377
KW
1048Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1049 STRLEN curlen,
1050 STRLEN *retlen,
1051 const U32 flags,
1052 U32 * errors)
a0ed51b3 1053{
d4c19fe8 1054 const U8 * const s0 = s;
2b5e7bc2
KW
1055 U8 * send = NULL; /* (initialized to silence compilers' wrong
1056 warning) */
1057 U32 possible_problems = 0; /* A bit is set here for each potential problem
1058 found as we go along */
eb83ed87 1059 UV uv = *s;
2b5e7bc2
KW
1060 STRLEN expectlen = 0; /* How long should this sequence be?
1061 (initialized to silence compilers' wrong
1062 warning) */
e308b348 1063 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1064 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1065 this gets set and discarded */
a0dbb045 1066
2b5e7bc2
KW
1067 /* The below are used only if there is both an overlong malformation and a
1068 * too short one. Otherwise the first two are set to 's0' and 'send', and
1069 * the third not used at all */
1070 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1071 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1072 warning) */
2b5e7bc2 1073 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1074
f9380377
KW
1075 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1076
1077 if (errors) {
1078 *errors = 0;
1079 }
1080 else {
1081 errors = &discard_errors;
1082 }
a0dbb045 1083
eb83ed87
KW
1084 /* The order of malformation tests here is important. We should consume as
1085 * few bytes as possible in order to not skip any valid character. This is
1086 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1087 * http://unicode.org/reports/tr36 for more discussion as to why. For
1088 * example, once we've done a UTF8SKIP, we can tell the expected number of
1089 * bytes, and could fail right off the bat if the input parameters indicate
1090 * that there are too few available. But it could be that just that first
1091 * byte is garbled, and the intended character occupies fewer bytes. If we
1092 * blindly assumed that the first byte is correct, and skipped based on
1093 * that number, we could skip over a valid input character. So instead, we
1094 * always examine the sequence byte-by-byte.
1095 *
1096 * We also should not consume too few bytes, otherwise someone could inject
1097 * things. For example, an input could be deliberately designed to
1098 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1099 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1100 * which is actually part of of the overflowing sequence), that could look
1101 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1102 * sequence and process the rest, inappropriately.
1103 *
1104 * Some possible input sequences are malformed in more than one way. This
1105 * function goes to lengths to try to find all of them. This is necessary
1106 * for correctness, as the inputs may allow one malformation but not
1107 * another, and if we abandon searching for others after finding the
1108 * allowed one, we could allow in something that shouldn't have been.
1109 */
eb83ed87 1110
b5b9af04 1111 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1112 possible_problems |= UTF8_GOT_EMPTY;
1113 curlen = 0;
5a48568d 1114 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1115 goto ready_to_handle_errors;
0c443dc2
JH
1116 }
1117
eb83ed87
KW
1118 expectlen = UTF8SKIP(s);
1119
1120 /* A well-formed UTF-8 character, as the vast majority of calls to this
1121 * function will be for, has this expected length. For efficiency, set
1122 * things up here to return it. It will be overriden only in those rare
1123 * cases where a malformation is found */
1124 if (retlen) {
1125 *retlen = expectlen;
1126 }
1127
1128 /* An invariant is trivially well-formed */
1d72bdf6 1129 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1130 return uv;
a0ed51b3 1131 }
67e989fb 1132
eb83ed87 1133 /* A continuation character can't start a valid sequence */
b5b9af04 1134 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1135 possible_problems |= UTF8_GOT_CONTINUATION;
1136 curlen = 1;
1137 uv = UNICODE_REPLACEMENT;
1138 goto ready_to_handle_errors;
ba210ebe 1139 }
9041c2e3 1140
dcd27b3c 1141 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1142 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1143 * because it excludes start bytes like \xC0 that always lead to
1144 * overlongs.) */
dcd27b3c 1145
534752c1
KW
1146 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1147 * that indicate the number of bytes in the character's whole UTF-8
1148 * sequence, leaving just the bits that are part of the value. */
1149 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1150
e308b348
KW
1151 /* Setup the loop end point, making sure to not look past the end of the
1152 * input string, and flag it as too short if the size isn't big enough. */
1153 send = (U8*) s0;
1154 if (UNLIKELY(curlen < expectlen)) {
1155 possible_problems |= UTF8_GOT_SHORT;
1156 avail_len = curlen;
1157 send += curlen;
1158 }
1159 else {
1160 send += expectlen;
1161 }
1162 adjusted_send = send;
1163
eb83ed87 1164 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1165 * accumulating each into the working value as we go. */
eb83ed87 1166 for (s = s0 + 1; s < send; s++) {
b5b9af04 1167 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1168 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1169 continue;
1170 }
1171
1172 /* Here, found a non-continuation before processing all expected bytes.
1173 * This byte indicates the beginning of a new character, so quit, even
1174 * if allowing this malformation. */
2b5e7bc2 1175 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1176 break;
eb83ed87
KW
1177 } /* End of loop through the character's bytes */
1178
1179 /* Save how many bytes were actually in the character */
1180 curlen = s - s0;
1181
e308b348
KW
1182 /* A convenience macro that matches either of the too-short conditions. */
1183# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1184
e308b348 1185 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1186 uv_so_far = uv;
1187 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1188 }
1189
2b5e7bc2
KW
1190 /* Note that there are two types of too-short malformation. One is when
1191 * there is actual wrong data before the normal termination of the
1192 * sequence. The other is that the sequence wasn't complete before the end
1193 * of the data we are allowed to look at, based on the input 'curlen'.
1194 * This means that we were passed data for a partial character, but it is
1195 * valid as far as we saw. The other is definitely invalid. This
1196 * distinction could be important to a caller, so the two types are kept
1197 * separate. */
1198
1199 /* Check for overflow */
1200 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1201 possible_problems |= UTF8_GOT_OVERFLOW;
1202 uv = UNICODE_REPLACEMENT;
eb83ed87 1203 }
eb83ed87 1204
2b5e7bc2
KW
1205 /* Check for overlong. If no problems so far, 'uv' is the correct code
1206 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1207 * we must look at the UTF-8 byte sequence itself to see if it is for an
1208 * overlong */
1209 if ( ( LIKELY(! possible_problems)
1210 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1211 || ( UNLIKELY( possible_problems)
1212 && ( UNLIKELY(! UTF8_IS_START(*s0))
1213 || ( curlen > 1
1214 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1215 send - s0))))))
2f8f112e 1216 {
2b5e7bc2
KW
1217 possible_problems |= UTF8_GOT_LONG;
1218
2b5e7bc2
KW
1219 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1220 UV min_uv = uv_so_far;
1221 STRLEN i;
1222
1223 /* Here, the input is both overlong and is missing some trailing
1224 * bytes. There is no single code point it could be for, but there
1225 * may be enough information present to determine if what we have
1226 * so far is for an unallowed code point, such as for a surrogate.
1227 * The code below has the intelligence to determine this, but just
1228 * for non-overlong UTF-8 sequences. What we do here is calculate
1229 * the smallest code point the input could represent if there were
1230 * no too short malformation. Then we compute and save the UTF-8
1231 * for that, which is what the code below looks at instead of the
1232 * raw input. It turns out that the smallest such code point is
1233 * all we need. */
1234 for (i = curlen; i < expectlen; i++) {
1235 min_uv = UTF8_ACCUMULATE(min_uv,
1236 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1237 }
1238
1239 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1240 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1241 to free it ourselves if
1242 warnings are made fatal */
1243 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1244 }
eb83ed87
KW
1245 }
1246
2b5e7bc2
KW
1247 /* Now check that the input isn't for a problematic code point not allowed
1248 * by the input parameters. */
1249 /* isn't problematic if < this */
1250 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1251 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1252
1253 /* if overflow, we know without looking further
1254 * precisely which of the problematic types it is,
1255 * and we deal with those in the overflow handling
1256 * code */
1257 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1258 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1259 && ((flags & ( UTF8_DISALLOW_NONCHAR
1260 |UTF8_DISALLOW_SURROGATE
1261 |UTF8_DISALLOW_SUPER
1262 |UTF8_DISALLOW_ABOVE_31_BIT
1263 |UTF8_WARN_NONCHAR
1264 |UTF8_WARN_SURROGATE
1265 |UTF8_WARN_SUPER
1266 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1267 /* In case of a malformation, 'uv' is not valid, and has
1268 * been changed to something in the Unicode range.
1269 * Currently we don't output a deprecation message if there
1270 * is already a malformation, so we don't have to special
1271 * case the test immediately below */
760c7c2f
KW
1272 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1273 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1274 {
2b5e7bc2
KW
1275 /* If there were no malformations, or the only malformation is an
1276 * overlong, 'uv' is valid */
1277 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1278 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1279 possible_problems |= UTF8_GOT_SURROGATE;
1280 }
1281 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1282 possible_problems |= UTF8_GOT_SUPER;
1283 }
1284 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1285 possible_problems |= UTF8_GOT_NONCHAR;
1286 }
1287 }
1288 else { /* Otherwise, need to look at the source UTF-8, possibly
1289 adjusted to be non-overlong */
1290
1291 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1292 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1293 {
2b5e7bc2
KW
1294 possible_problems |= UTF8_GOT_SUPER;
1295 }
1296 else if (curlen > 1) {
1297 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1298 NATIVE_UTF8_TO_I8(*adjusted_s0),
1299 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1300 {
2b5e7bc2 1301 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1302 }
2b5e7bc2
KW
1303 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1304 NATIVE_UTF8_TO_I8(*adjusted_s0),
1305 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1306 {
1307 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1308 }
1309 }
c0236afe 1310
2b5e7bc2
KW
1311 /* We need a complete well-formed UTF-8 character to discern
1312 * non-characters, so can't look for them here */
1313 }
1314 }
949cf498 1315
2b5e7bc2
KW
1316 ready_to_handle_errors:
1317
1318 /* At this point:
1319 * curlen contains the number of bytes in the sequence that
1320 * this call should advance the input by.
e308b348
KW
1321 * avail_len gives the available number of bytes passed in, but
1322 * only if this is less than the expected number of
1323 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1324 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1325 * is set in it for each potential problem found.
1326 * uv contains the code point the input sequence
1327 * represents; or if there is a problem that prevents
1328 * a well-defined value from being computed, it is
1329 * some subsitute value, typically the REPLACEMENT
1330 * CHARACTER.
1331 * s0 points to the first byte of the character
1332 * send points to just after where that (potentially
1333 * partial) character ends
1334 * adjusted_s0 normally is the same as s0, but in case of an
1335 * overlong for which the UTF-8 matters below, it is
1336 * the first byte of the shortest form representation
1337 * of the input.
1338 * adjusted_send normally is the same as 'send', but if adjusted_s0
1339 * is set to something other than s0, this points one
1340 * beyond its end
1341 */
eb83ed87 1342
2b5e7bc2
KW
1343 if (UNLIKELY(possible_problems)) {
1344 bool disallowed = FALSE;
1345 const U32 orig_problems = possible_problems;
1346
1347 while (possible_problems) { /* Handle each possible problem */
1348 UV pack_warn = 0;
1349 char * message = NULL;
1350
1351 /* Each 'if' clause handles one problem. They are ordered so that
1352 * the first ones' messages will be displayed before the later
1353 * ones; this is kinda in decreasing severity order */
1354 if (possible_problems & UTF8_GOT_OVERFLOW) {
1355
1356 /* Overflow means also got a super and above 31 bits, but we
1357 * handle all three cases here */
1358 possible_problems
1359 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1360 *errors |= UTF8_GOT_OVERFLOW;
1361
1362 /* But the API says we flag all errors found */
1363 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1364 *errors |= UTF8_GOT_SUPER;
1365 }
ddb65933
KW
1366 if (flags
1367 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1368 {
f9380377
KW
1369 *errors |= UTF8_GOT_ABOVE_31_BIT;
1370 }
2b5e7bc2 1371
d60baaa7
KW
1372 /* Disallow if any of the three categories say to */
1373 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1374 || (flags & ( UTF8_DISALLOW_SUPER
1375 |UTF8_DISALLOW_ABOVE_31_BIT)))
1376 {
1377 disallowed = TRUE;
1378 }
1379
1380
1381 /* Likewise, warn if any say to, plus if deprecation warnings
1382 * are on, because this code point is above IV_MAX */
1383 if ( ckWARN_d(WARN_DEPRECATED)
1384 || ! (flags & UTF8_ALLOW_OVERFLOW)
1385 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1386 {
2b5e7bc2 1387
ddb65933
KW
1388 /* The warnings code explicitly says it doesn't handle the
1389 * case of packWARN2 and two categories which have
1390 * parent-child relationship. Even if it works now to
1391 * raise the warning if either is enabled, it wouldn't
1392 * necessarily do so in the future. We output (only) the
1393 * most dire warning*/
1394 if (! (flags & UTF8_CHECK_ONLY)) {
1395 if (ckWARN_d(WARN_UTF8)) {
1396 pack_warn = packWARN(WARN_UTF8);
1397 }
1398 else if (ckWARN_d(WARN_NON_UNICODE)) {
1399 pack_warn = packWARN(WARN_NON_UNICODE);
1400 }
1401 if (pack_warn) {
1402 message = Perl_form(aTHX_ "%s: %s (overflows)",
1403 malformed_text,
1404 _byte_dump_string(s0, send - s0));
1405 }
2b5e7bc2
KW
1406 }
1407 }
1408 }
1409 else if (possible_problems & UTF8_GOT_EMPTY) {
1410 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1411 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1412
1413 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1414
1415 /* This so-called malformation is now treated as a bug in
1416 * the caller. If you have nothing to decode, skip calling
1417 * this function */
1418 assert(0);
1419
2b5e7bc2
KW
1420 disallowed = TRUE;
1421 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1422 pack_warn = packWARN(WARN_UTF8);
1423 message = Perl_form(aTHX_ "%s (empty string)",
1424 malformed_text);
1425 }
1426 }
1427 }
1428 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1429 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1430 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1431
1432 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1433 disallowed = TRUE;
1434 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1435 pack_warn = packWARN(WARN_UTF8);
1436 message = Perl_form(aTHX_
1437 "%s: %s (unexpected continuation byte 0x%02x,"
1438 " with no preceding start byte)",
1439 malformed_text,
1440 _byte_dump_string(s0, 1), *s0);
1441 }
1442 }
1443 }
2b5e7bc2
KW
1444 else if (possible_problems & UTF8_GOT_SHORT) {
1445 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1446 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1447
1448 if (! (flags & UTF8_ALLOW_SHORT)) {
1449 disallowed = TRUE;
1450 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1451 pack_warn = packWARN(WARN_UTF8);
1452 message = Perl_form(aTHX_
9a6c9c81 1453 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1454 malformed_text,
1455 _byte_dump_string(s0, send - s0),
e308b348
KW
1456 (int)avail_len,
1457 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1458 (int)expectlen);
1459 }
1460 }
ba210ebe 1461
2b5e7bc2 1462 }
e308b348
KW
1463 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1464 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1465 *errors |= UTF8_GOT_NON_CONTINUATION;
1466
1467 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1468 disallowed = TRUE;
1469 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
99a765e9
KW
1470
1471 /* If we don't know for sure that the input length is
1472 * valid, avoid as much as possible reading past the
1473 * end of the buffer */
1474 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1475 ? s - s0
1476 : send - s0;
e308b348
KW
1477 pack_warn = packWARN(WARN_UTF8);
1478 message = Perl_form(aTHX_ "%s",
1479 unexpected_non_continuation_text(s0,
99a765e9 1480 printlen,
e308b348
KW
1481 s - s0,
1482 (int) expectlen));
1483 }
1484 }
1485 }
2b5e7bc2
KW
1486 else if (possible_problems & UTF8_GOT_LONG) {
1487 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1488 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1489
94953955
KW
1490 if (flags & UTF8_ALLOW_LONG) {
1491
1492 /* We don't allow the actual overlong value, unless the
1493 * special extra bit is also set */
1494 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1495 & ~UTF8_ALLOW_LONG)))
1496 {
1497 uv = UNICODE_REPLACEMENT;
1498 }
1499 }
1500 else {
2b5e7bc2
KW
1501 disallowed = TRUE;
1502
1503 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1504 pack_warn = packWARN(WARN_UTF8);
1505
1506 /* These error types cause 'uv' to be something that
1507 * isn't what was intended, so can't use it in the
1508 * message. The other error types either can't
1509 * generate an overlong, or else the 'uv' is valid */
1510 if (orig_problems &
1511 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1512 {
1513 message = Perl_form(aTHX_
1514 "%s: %s (any UTF-8 sequence that starts"
1515 " with \"%s\" is overlong which can and"
1516 " should be represented with a"
1517 " different, shorter sequence)",
1518 malformed_text,
1519 _byte_dump_string(s0, send - s0),
1520 _byte_dump_string(s0, curlen));
1521 }
1522 else {
1523 U8 tmpbuf[UTF8_MAXBYTES+1];
1524 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1525 uv, 0);
1526 message = Perl_form(aTHX_
1527 "%s: %s (overlong; instead use %s to represent"
147e3846 1528 " U+%0*" UVXf ")",
2b5e7bc2
KW
1529 malformed_text,
1530 _byte_dump_string(s0, send - s0),
1531 _byte_dump_string(tmpbuf, e - tmpbuf),
1532 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1533 small code points */
1534 uv);
1535 }
1536 }
1537 }
1538 }
1539 else if (possible_problems & UTF8_GOT_SURROGATE) {
1540 possible_problems &= ~UTF8_GOT_SURROGATE;
1541
f9380377
KW
1542 if (flags & UTF8_WARN_SURROGATE) {
1543 *errors |= UTF8_GOT_SURROGATE;
1544
1545 if ( ! (flags & UTF8_CHECK_ONLY)
1546 && ckWARN_d(WARN_SURROGATE))
1547 {
2b5e7bc2
KW
1548 pack_warn = packWARN(WARN_SURROGATE);
1549
1550 /* These are the only errors that can occur with a
1551 * surrogate when the 'uv' isn't valid */
1552 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1553 message = Perl_form(aTHX_
1554 "UTF-16 surrogate (any UTF-8 sequence that"
1555 " starts with \"%s\" is for a surrogate)",
1556 _byte_dump_string(s0, curlen));
1557 }
1558 else {
1559 message = Perl_form(aTHX_
147e3846 1560 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1561 }
f9380377 1562 }
2b5e7bc2 1563 }
ba210ebe 1564
2b5e7bc2
KW
1565 if (flags & UTF8_DISALLOW_SURROGATE) {
1566 disallowed = TRUE;
f9380377 1567 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1568 }
1569 }
1570 else if (possible_problems & UTF8_GOT_SUPER) {
1571 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1572
f9380377
KW
1573 if (flags & UTF8_WARN_SUPER) {
1574 *errors |= UTF8_GOT_SUPER;
1575
1576 if ( ! (flags & UTF8_CHECK_ONLY)
1577 && ckWARN_d(WARN_NON_UNICODE))
1578 {
2b5e7bc2
KW
1579 pack_warn = packWARN(WARN_NON_UNICODE);
1580
1581 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1582 message = Perl_form(aTHX_
1583 "Any UTF-8 sequence that starts with"
1584 " \"%s\" is for a non-Unicode code point,"
1585 " may not be portable",
1586 _byte_dump_string(s0, curlen));
1587 }
1588 else {
1589 message = Perl_form(aTHX_
147e3846 1590 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1591 " Unicode, may not be portable",
1592 uv);
1593 }
f9380377 1594 }
2b5e7bc2 1595 }
ba210ebe 1596
2b5e7bc2
KW
1597 /* The maximum code point ever specified by a standard was
1598 * 2**31 - 1. Anything larger than that is a Perl extension
1599 * that very well may not be understood by other applications
1600 * (including earlier perl versions on EBCDIC platforms). We
1601 * test for these after the regular SUPER ones, and before
1602 * possibly bailing out, so that the slightly more dire warning
1603 * will override the regular one. */
1604 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1605 |UTF8_WARN_SUPER
1606 |UTF8_DISALLOW_ABOVE_31_BIT))
1607 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1608 && UNLIKELY(is_utf8_cp_above_31_bits(
1609 adjusted_s0,
1610 adjusted_send)))
1611 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1612 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1613 {
1614 if ( ! (flags & UTF8_CHECK_ONLY)
1615 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1616 && ckWARN_d(WARN_UTF8))
1617 {
1618 pack_warn = packWARN(WARN_UTF8);
1619
1620 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1621 message = Perl_form(aTHX_
1622 "Any UTF-8 sequence that starts with"
1623 " \"%s\" is for a non-Unicode code"
1624 " point, and is not portable",
1625 _byte_dump_string(s0, curlen));
1626 }
1627 else {
1628 message = Perl_form(aTHX_
147e3846 1629 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1630 " and not portable",
1631 uv);
1632 }
1633 }
1634
ddb65933
KW
1635 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1636 |UTF8_DISALLOW_ABOVE_31_BIT))
1637 {
f9380377
KW
1638 *errors |= UTF8_GOT_ABOVE_31_BIT;
1639
1640 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1641 disallowed = TRUE;
1642 }
2b5e7bc2
KW
1643 }
1644 }
eb83ed87 1645
2b5e7bc2 1646 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1647 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1648 disallowed = TRUE;
1649 }
eb83ed87 1650
2b5e7bc2
KW
1651 /* The deprecated warning overrides any non-deprecated one. If
1652 * there are other problems, a deprecation message is not
1653 * really helpful, so don't bother to raise it in that case.
1654 * This also keeps the code from having to handle the case
1655 * where 'uv' is not valid. */
1656 if ( ! (orig_problems
1657 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1658 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1659 && ckWARN_d(WARN_DEPRECATED))
1660 {
1661 message = Perl_form(aTHX_ cp_above_legal_max,
1662 uv, MAX_NON_DEPRECATED_CP);
1663 pack_warn = packWARN(WARN_DEPRECATED);
1664 }
1665 }
1666 else if (possible_problems & UTF8_GOT_NONCHAR) {
1667 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1668
f9380377
KW
1669 if (flags & UTF8_WARN_NONCHAR) {
1670 *errors |= UTF8_GOT_NONCHAR;
1671
1672 if ( ! (flags & UTF8_CHECK_ONLY)
1673 && ckWARN_d(WARN_NONCHAR))
1674 {
2b5e7bc2
KW
1675 /* The code above should have guaranteed that we don't
1676 * get here with errors other than overlong */
1677 assert (! (orig_problems
1678 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1679
1680 pack_warn = packWARN(WARN_NONCHAR);
1681 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1682 " U+%04" UVXf " is not recommended"
2b5e7bc2 1683 " for open interchange", uv);
f9380377 1684 }
2b5e7bc2 1685 }
5b311467 1686
2b5e7bc2
KW
1687 if (flags & UTF8_DISALLOW_NONCHAR) {
1688 disallowed = TRUE;
f9380377 1689 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1690 }
1691 } /* End of looking through the possible flags */
1692
1693 /* Display the message (if any) for the problem being handled in
1694 * this iteration of the loop */
1695 if (message) {
1696 if (PL_op)
1697 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1698 OP_DESC(PL_op));
1699 else
1700 Perl_warner(aTHX_ pack_warn, "%s", message);
1701 }
ddb65933 1702 } /* End of 'while (possible_problems)' */
a0dbb045 1703
2b5e7bc2
KW
1704 /* Since there was a possible problem, the returned length may need to
1705 * be changed from the one stored at the beginning of this function.
1706 * Instead of trying to figure out if that's needed, just do it. */
1707 if (retlen) {
1708 *retlen = curlen;
1709 }
a0dbb045 1710
2b5e7bc2
KW
1711 if (disallowed) {
1712 if (flags & UTF8_CHECK_ONLY && retlen) {
1713 *retlen = ((STRLEN) -1);
1714 }
1715 return 0;
1716 }
eb83ed87 1717 }
ba210ebe 1718
2b5e7bc2 1719 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1720}
1721
8e84507e 1722/*
ec5f19d0
KW
1723=for apidoc utf8_to_uvchr_buf
1724
1725Returns the native code point of the first character in the string C<s> which
1726is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1727C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1728
524080c4
KW
1729If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1730enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1731C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1732(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1733C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1734the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1735See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1736returned.
ec5f19d0 1737
760c7c2f
KW
1738Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1739unless those are turned off.
1740
ec5f19d0 1741=cut
52be2536
KW
1742
1743Also implemented as a macro in utf8.h
1744
ec5f19d0
KW
1745*/
1746
1747
1748UV
1749Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1750{
ec5f19d0
KW
1751 assert(s < send);
1752
1753 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1754 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1755}
1756
52be2536
KW
1757/* This is marked as deprecated
1758 *
ec5f19d0
KW
1759=for apidoc utf8_to_uvuni_buf
1760
de69f3af
KW
1761Only in very rare circumstances should code need to be dealing in Unicode
1762(as opposed to native) code points. In those few cases, use
1763C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1764
1765Returns the Unicode (not-native) code point of the first character in the
1766string C<s> which
ec5f19d0
KW
1767is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1768C<retlen> will be set to the length, in bytes, of that character.
1769
524080c4
KW
1770If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1771enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1772NULL) to -1. If those warnings are off, the computed value if well-defined (or
1773the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1774is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1775next possible position in C<s> that could begin a non-malformed character.
de69f3af 1776See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1777
760c7c2f
KW
1778Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1779unless those are turned off.
1780
ec5f19d0
KW
1781=cut
1782*/
1783
1784UV
1785Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1786{
1787 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1788
1789 assert(send > s);
1790
5962d97e
KW
1791 /* Call the low level routine, asking for checks */
1792 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1793}
1794
b76347f2 1795/*
87cea99e 1796=for apidoc utf8_length
b76347f2
JH
1797
1798Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1799Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1800up past C<e>, croaks.
b76347f2
JH
1801
1802=cut
1803*/
1804
1805STRLEN
35a4481c 1806Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1807{
1808 STRLEN len = 0;
1809
7918f24d
NC
1810 PERL_ARGS_ASSERT_UTF8_LENGTH;
1811
8850bf83
JH
1812 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1813 * the bitops (especially ~) can create illegal UTF-8.
1814 * In other words: in Perl UTF-8 is not just for Unicode. */
1815
a3b680e6
AL
1816 if (e < s)
1817 goto warn_and_return;
b76347f2 1818 while (s < e) {
4cbf4130 1819 s += UTF8SKIP(s);
8e91ec7f
AV
1820 len++;
1821 }
1822
1823 if (e != s) {
1824 len--;
1825 warn_and_return:
9b387841
NC
1826 if (PL_op)
1827 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1828 "%s in %s", unees, OP_DESC(PL_op));
1829 else
61a12c31 1830 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1831 }
1832
1833 return len;
1834}
1835
b06226ff 1836/*
fed3ba5d
NC
1837=for apidoc bytes_cmp_utf8
1838
a1433954 1839Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1840sequence of characters (stored as UTF-8)
1841in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1842equal, -1 or -2 if the first string is less than the second string, +1 or +2
1843if the first string is greater than the second string.
1844
1845-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1846longer string. -2 or +2 is returned if
1847there was a difference between characters
fed3ba5d
NC
1848within the strings.
1849
1850=cut
1851*/
1852
1853int
1854Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1855{
1856 const U8 *const bend = b + blen;
1857 const U8 *const uend = u + ulen;
1858
1859 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1860
1861 while (b < bend && u < uend) {
1862 U8 c = *u++;
1863 if (!UTF8_IS_INVARIANT(c)) {
1864 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1865 if (u < uend) {
1866 U8 c1 = *u++;
1867 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1868 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1869 } else {
2b5e7bc2 1870 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1871 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1872 "%s %s%s",
7cf8d05d 1873 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1874 PL_op ? " in " : "",
1875 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1876 return -2;
1877 }
1878 } else {
1879 if (PL_op)
1880 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1881 "%s in %s", unees, OP_DESC(PL_op));
1882 else
61a12c31 1883 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1884 return -2; /* Really want to return undef :-) */
1885 }
1886 } else {
1887 return -2;
1888 }
1889 }
1890 if (*b != c) {
1891 return *b < c ? -2 : +2;
1892 }
1893 ++b;
1894 }
1895
1896 if (b == bend && u == uend)
1897 return 0;
1898
1899 return b < bend ? +1 : -1;
1900}
1901
1902/*
87cea99e 1903=for apidoc utf8_to_bytes
6940069f 1904
2bbc8d55 1905Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1906Unlike L</bytes_to_utf8>, this over-writes the original string, and
1907updates C<len> to contain the new length.
67e989fb 1908Returns zero on failure, setting C<len> to -1.
6940069f 1909
a1433954 1910If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1911
6940069f
GS
1912=cut
1913*/
1914
1915U8 *
37607a96 1916Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1917{
d4c19fe8
AL
1918 U8 * const save = s;
1919 U8 * const send = s + *len;
6940069f 1920 U8 *d;
246fae53 1921
7918f24d 1922 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1923 PERL_UNUSED_CONTEXT;
7918f24d 1924
1e54db1a 1925 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1926 while (s < send) {
d59937ca
KW
1927 if (! UTF8_IS_INVARIANT(*s)) {
1928 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1929 *len = ((STRLEN) -1);
1930 return 0;
1931 }
1932 s++;
dcad2880 1933 }
d59937ca 1934 s++;
246fae53 1935 }
dcad2880
JH
1936
1937 d = s = save;
6940069f 1938 while (s < send) {
80e0b38f
KW
1939 U8 c = *s++;
1940 if (! UTF8_IS_INVARIANT(c)) {
1941 /* Then it is two-byte encoded */
a62b247b 1942 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1943 s++;
1944 }
1945 *d++ = c;
6940069f
GS
1946 }
1947 *d = '\0';
246fae53 1948 *len = d - save;
6940069f
GS
1949 return save;
1950}
1951
1952/*
87cea99e 1953=for apidoc bytes_from_utf8
f9a63242 1954
2bbc8d55 1955Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1956Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1957the newly-created string, and updates C<len> to contain the new
1958length. Returns the original string if no conversion occurs, C<len>
72d33970 1959is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19600 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1961in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1962
37607a96
PK
1963=cut
1964*/
f9a63242
JH
1965
1966U8 *
e1ec3a88 1967Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1968{
f9a63242 1969 U8 *d;
e1ec3a88
AL
1970 const U8 *start = s;
1971 const U8 *send;
f9a63242
JH
1972 I32 count = 0;
1973
7918f24d 1974 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1975 PERL_UNUSED_CONTEXT;
f9a63242 1976 if (!*is_utf8)
73d840c0 1977 return (U8 *)start;
f9a63242 1978
1e54db1a 1979 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1980 for (send = s + *len; s < send;) {
d59937ca
KW
1981 if (! UTF8_IS_INVARIANT(*s)) {
1982 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1983 return (U8 *)start;
d59937ca
KW
1984 }
1985 count++;
1986 s++;
db42d148 1987 }
d59937ca 1988 s++;
f9a63242
JH
1989 }
1990
35da51f7 1991 *is_utf8 = FALSE;
f9a63242 1992
212542aa 1993 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1994 s = start; start = d;
f9a63242
JH
1995 while (s < send) {
1996 U8 c = *s++;
1a91c45d 1997 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1998 /* Then it is two-byte encoded */
a62b247b 1999 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 2000 s++;
c4d5f83a
NIS
2001 }
2002 *d++ = c;
f9a63242
JH
2003 }
2004 *d = '\0';
2005 *len = d - start;
73d840c0 2006 return (U8 *)start;
f9a63242
JH
2007}
2008
2009/*
87cea99e 2010=for apidoc bytes_to_utf8
6940069f 2011
ff97e5cf
KW
2012Converts a string C<s> of length C<len> bytes from the native encoding into
2013UTF-8.
6662521e 2014Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 2015reflect the new length in bytes.
6940069f 2016
75200dff 2017A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2018
2019If you want to convert to UTF-8 from encodings other than
2020the native (Latin1 or EBCDIC),
a1433954 2021see L</sv_recode_to_utf8>().
c9ada85f 2022
497711e7 2023=cut
6940069f
GS
2024*/
2025
c682ebef
FC
2026/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2027 likewise need duplication. */
2028
6940069f 2029U8*
35a4481c 2030Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 2031{
35a4481c 2032 const U8 * const send = s + (*len);
6940069f
GS
2033 U8 *d;
2034 U8 *dst;
7918f24d
NC
2035
2036 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2037 PERL_UNUSED_CONTEXT;
6940069f 2038
212542aa 2039 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2040 dst = d;
2041
2042 while (s < send) {
55d09dc8
KW
2043 append_utf8_from_native_byte(*s, &d);
2044 s++;
6940069f
GS
2045 }
2046 *d = '\0';
6662521e 2047 *len = d-dst;
6940069f
GS
2048 return dst;
2049}
2050
a0ed51b3 2051/*
dea0fc0b 2052 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2053 *
2054 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2055 * We optimize for native, for obvious reasons. */
2056
2057U8*
dea0fc0b 2058Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2059{
dea0fc0b
JH
2060 U8* pend;
2061 U8* dstart = d;
2062
7918f24d
NC
2063 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2064
dea0fc0b 2065 if (bytelen & 1)
147e3846 2066 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2067
2068 pend = p + bytelen;
2069
a0ed51b3 2070 while (p < pend) {
dea0fc0b
JH
2071 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2072 p += 2;
2d1545e5 2073 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2074 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2075 continue;
2076 }
56d37426
KW
2077 if (uv <= MAX_UTF8_TWO_BYTE) {
2078 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2079 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2080 continue;
2081 }
46956fad
KW
2082#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2083#define LAST_HIGH_SURROGATE 0xDBFF
2084#define FIRST_LOW_SURROGATE 0xDC00
2085#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2086
2087 /* This assumes that most uses will be in the first Unicode plane, not
2088 * needing surrogates */
2089 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2090 && uv <= UNICODE_SURROGATE_LAST))
2091 {
2092 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2093 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2094 }
2095 else {
01ea242b 2096 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2097 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2098 || UNLIKELY(low > LAST_LOW_SURROGATE))
2099 {
01ea242b 2100 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2101 }
2102 p += 2;
46956fad
KW
2103 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2104 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2105 }
a0ed51b3 2106 }
56d37426
KW
2107#ifdef EBCDIC
2108 d = uvoffuni_to_utf8_flags(d, uv, 0);
2109#else
a0ed51b3 2110 if (uv < 0x10000) {
eb160463
GS
2111 *d++ = (U8)(( uv >> 12) | 0xe0);
2112 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2113 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2114 continue;
2115 }
2116 else {
eb160463
GS
2117 *d++ = (U8)(( uv >> 18) | 0xf0);
2118 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2119 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2120 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2121 continue;
2122 }
56d37426 2123#endif
a0ed51b3 2124 }
dea0fc0b 2125 *newlen = d - dstart;
a0ed51b3
LW
2126 return d;
2127}
2128
2129/* Note: this one is slightly destructive of the source. */
2130
2131U8*
dea0fc0b 2132Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2133{
2134 U8* s = (U8*)p;
d4c19fe8 2135 U8* const send = s + bytelen;
7918f24d
NC
2136
2137 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2138
e0ea5e2d 2139 if (bytelen & 1)
147e3846 2140 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2141 (UV)bytelen);
2142
a0ed51b3 2143 while (s < send) {
d4c19fe8 2144 const U8 tmp = s[0];
a0ed51b3
LW
2145 s[0] = s[1];
2146 s[1] = tmp;
2147 s += 2;
2148 }
dea0fc0b 2149 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2150}
2151
922e8cb4
KW
2152bool
2153Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2154{
2155 U8 tmpbuf[UTF8_MAXBYTES+1];
2156 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2157 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2158}
2159
f9ae8fb6
JD
2160/* Internal function so we can deprecate the external one, and call
2161 this one from other deprecated functions in this file */
2162
f2645549
KW
2163bool
2164Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2165{
f2645549 2166 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2167
2168 if (*p == '_')
2169 return TRUE;
f25ce844 2170 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2171}
2172
5092f92a 2173bool
eba68aa0
KW
2174Perl__is_uni_perl_idcont(pTHX_ UV c)
2175{
2176 U8 tmpbuf[UTF8_MAXBYTES+1];
2177 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2178 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2179}
2180
2181bool
f91dcd13
KW
2182Perl__is_uni_perl_idstart(pTHX_ UV c)
2183{
2184 U8 tmpbuf[UTF8_MAXBYTES+1];
2185 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2186 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2187}
2188
3a4c58c9
KW
2189UV
2190Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2191{
2192 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2193 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2194 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2195 * either "SS" or "Ss". Which one to use is passed into the routine in
2196 * 'S_or_s' to avoid a test */
2197
2198 UV converted = toUPPER_LATIN1_MOD(c);
2199
2200 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2201
2202 assert(S_or_s == 'S' || S_or_s == 's');
2203
6f2d5cbc 2204 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2205 characters in this range */
3a4c58c9
KW
2206 *p = (U8) converted;
2207 *lenp = 1;
2208 return converted;
2209 }
2210
2211 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2212 * which it maps to one of them, so as to only have to have one check for
2213 * it in the main case */
2214 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2215 switch (c) {
2216 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2217 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2218 break;
2219 case MICRO_SIGN:
2220 converted = GREEK_CAPITAL_LETTER_MU;
2221 break;
79e064b9
KW
2222#if UNICODE_MAJOR_VERSION > 2 \
2223 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2224 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2225 case LATIN_SMALL_LETTER_SHARP_S:
2226 *(p)++ = 'S';
2227 *p = S_or_s;
2228 *lenp = 2;
2229 return 'S';
79e064b9 2230#endif
3a4c58c9
KW
2231 default:
2232 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2233 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2234 }
2235 }
2236
2237 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2238 *p = UTF8_TWO_BYTE_LO(converted);
2239 *lenp = 2;
2240
2241 return converted;
2242}
2243
50bda2c3
KW
2244/* Call the function to convert a UTF-8 encoded character to the specified case.
2245 * Note that there may be more than one character in the result.
2246 * INP is a pointer to the first byte of the input character
2247 * OUTP will be set to the first byte of the string of changed characters. It
2248 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2249 * LENP will be set to the length in bytes of the string of changed characters
2250 *
2251 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2252#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2253#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2254#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2255
b9992569
KW
2256/* This additionally has the input parameter 'specials', which if non-zero will
2257 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2258 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2259#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2260
84afefe6
JH
2261UV
2262Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2263{
a1433954
KW
2264 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2265 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2266 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2267 * the changed version may be longer than the original character.
2268 *
2269 * The ordinal of the first character of the changed version is returned
2270 * (but note, as explained above, that there may be more.) */
2271
7918f24d
NC
2272 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2273
3a4c58c9
KW
2274 if (c < 256) {
2275 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2276 }
2277
0ebc6274 2278 uvchr_to_utf8(p, c);
b9992569 2279 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2280}
2281
84afefe6
JH
2282UV
2283Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2284{
7918f24d
NC
2285 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2286
3a4c58c9
KW
2287 if (c < 256) {
2288 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2289 }
2290
0ebc6274 2291 uvchr_to_utf8(p, c);
b9992569 2292 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2293}
2294
afc16117 2295STATIC U8
eaf412bf 2296S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2297{
2298 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2299 * those, converting the result to UTF-8. Since the result is always just
a1433954 2300 * one character, we allow <p> to be NULL */
afc16117
KW
2301
2302 U8 converted = toLOWER_LATIN1(c);
2303
eaf412bf
KW
2304 PERL_UNUSED_ARG(dummy);
2305
afc16117 2306 if (p != NULL) {
6f2d5cbc 2307 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2308 *p = converted;
2309 *lenp = 1;
2310 }
2311 else {
430c9760
KW
2312 /* Result is known to always be < 256, so can use the EIGHT_BIT
2313 * macros */
2314 *p = UTF8_EIGHT_BIT_HI(converted);
2315 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2316 *lenp = 2;
2317 }
2318 }
2319 return converted;
2320}
2321
84afefe6
JH
2322UV
2323Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2324{
7918f24d
NC
2325 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2326
afc16117 2327 if (c < 256) {
eaf412bf 2328 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2329 }
2330
afc16117 2331 uvchr_to_utf8(p, c);
b9992569 2332 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2333}
2334
84afefe6 2335UV
51910141 2336Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2337{
51910141 2338 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2339 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2340 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2341 *
2342 * Not to be used for locale folds
51910141 2343 */
f673fad4 2344
a1dde8de
KW
2345 UV converted;
2346
2347 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2348 PERL_UNUSED_CONTEXT;
a1dde8de 2349
1ca267a5
KW
2350 assert (! (flags & FOLD_FLAGS_LOCALE));
2351
659a7c2d 2352 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2353 converted = GREEK_SMALL_LETTER_MU;
2354 }
9b63e895
KW
2355#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2356 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2357 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2358 else if ( (flags & FOLD_FLAGS_FULL)
2359 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2360 {
1ca267a5
KW
2361 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2362 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2363 * under those circumstances. */
2364 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2365 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2366 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2367 p, *lenp, U8);
2368 return LATIN_SMALL_LETTER_LONG_S;
2369 }
2370 else {
4f489194
KW
2371 *(p)++ = 's';
2372 *p = 's';
2373 *lenp = 2;
2374 return 's';
1ca267a5 2375 }
a1dde8de 2376 }
9b63e895 2377#endif
a1dde8de
KW
2378 else { /* In this range the fold of all other characters is their lower
2379 case */
2380 converted = toLOWER_LATIN1(c);
2381 }
2382
6f2d5cbc 2383 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2384 *p = (U8) converted;
2385 *lenp = 1;
2386 }
2387 else {
2388 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2389 *p = UTF8_TWO_BYTE_LO(converted);
2390 *lenp = 2;
2391 }
2392
2393 return converted;
2394}
2395
2396UV
31f05a37 2397Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2398{
4b593389 2399
a0270393
KW
2400 /* Not currently externally documented, and subject to change
2401 * <flags> bits meanings:
2402 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2403 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2404 * locale are to be used.
a0270393
KW
2405 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2406 */
4b593389 2407
36bb2ab6 2408 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2409
780fcc9f
KW
2410 if (flags & FOLD_FLAGS_LOCALE) {
2411 /* Treat a UTF-8 locale as not being in locale at all */
2412 if (IN_UTF8_CTYPE_LOCALE) {
2413 flags &= ~FOLD_FLAGS_LOCALE;
2414 }
2415 else {
2416 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2417 goto needs_full_generality;
780fcc9f 2418 }
31f05a37
KW
2419 }
2420
a1dde8de 2421 if (c < 256) {
e7b7ac46 2422 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2423 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2424 }
2425
2f306ab9 2426 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2427 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2428 uvchr_to_utf8(p, c);
b9992569 2429 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393 2430 }
567b353c 2431 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
a0270393
KW
2432 the special flags. */
2433 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2434
2435 needs_full_generality:
a0270393 2436 uvchr_to_utf8(utf8_c, c);
a1a5ec35 2437 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
a0270393 2438 }
84afefe6
JH
2439}
2440
26483009 2441PERL_STATIC_INLINE bool
5141f98e 2442S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2443 const char *const swashname, SV* const invlist)
bde6a22d 2444{
ea317ccb
KW
2445 /* returns a boolean giving whether or not the UTF8-encoded character that
2446 * starts at <p> is in the swash indicated by <swashname>. <swash>
2447 * contains a pointer to where the swash indicated by <swashname>
2448 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2449 * look at <*swash> and only generate a swash if it is not null. <invlist>
2450 * is NULL or an inversion list that defines the swash. If not null, it
2451 * saves time during initialization of the swash.
ea317ccb
KW
2452 *
2453 * Note that it is assumed that the buffer length of <p> is enough to
2454 * contain all the bytes that comprise the character. Thus, <*p> should
2455 * have been checked before this call for mal-formedness enough to assure
2456 * that. */
2457
7918f24d
NC
2458 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2459
492a624f 2460 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2461 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2462 * as far as there being enough bytes available in it to accommodate the
2463 * character without reading beyond the end, and pass that number on to the
2464 * validating routine */
6302f837 2465 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94 2466 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
99a765e9 2467 _UTF8_NO_CONFIDENCE_IN_CURLEN,
86ae6e94
KW
2468 1 /* Die */ );
2469 NOT_REACHED; /* NOTREACHED */
28123549 2470 }
86ae6e94 2471
87367d5f
KW
2472 if (!*swash) {
2473 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2474 *swash = _core_swash_init("utf8",
2475
2476 /* Only use the name if there is no inversion
2477 * list; otherwise will go out to disk */
2478 (invlist) ? "" : swashname,
2479
2480 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2481 }
28123549 2482
bde6a22d
NC
2483 return swash_fetch(*swash, p, TRUE) != 0;
2484}
2485
da8c1a98
KW
2486PERL_STATIC_INLINE bool
2487S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2488 const char *const swashname, SV* const invlist)
2489{
2490 /* returns a boolean giving whether or not the UTF8-encoded character that
2491 * starts at <p>, and extending no further than <e - 1> is in the swash
2492 * indicated by <swashname>. <swash> contains a pointer to where the swash
2493 * indicated by <swashname> is to be stored; which this routine will do, so
2494 * that future calls will look at <*swash> and only generate a swash if it
2495 * is not null. <invlist> is NULL or an inversion list that defines the
2496 * swash. If not null, it saves time during initialization of the swash.
2497 */
2498
2499 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2500
2501 if (! isUTF8_CHAR(p, e)) {
2502 _force_out_malformed_utf8_message(p, e, 0, 1);
2503 NOT_REACHED; /* NOTREACHED */
2504 }
2505
2506 if (!*swash) {
2507 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2508 *swash = _core_swash_init("utf8",
2509
2510 /* Only use the name if there is no inversion
2511 * list; otherwise will go out to disk */
2512 (invlist) ? "" : swashname,
2513
2514 &PL_sv_undef, 1, 0, invlist, &flags);
2515 }
2516
2517 return swash_fetch(*swash, p, TRUE) != 0;
2518}
2519
34aeb2e9
KW
2520STATIC void
2521S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2522 const char * const alternative,
2523 const bool use_locale,
2524 const char * const file,
2525 const unsigned line)
2526{
2527 const char * key;
2528
2529 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2530
2531 if (ckWARN_d(WARN_DEPRECATED)) {
2532
2533 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2534 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2535 if (! PL_seen_deprecated_macro) {
2536 PL_seen_deprecated_macro = newHV();
2537 }
2538 if (! hv_store(PL_seen_deprecated_macro, key,
2539 strlen(key), &PL_sv_undef, 0))
2540 {
2541 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2542 }
2543
2544 Perl_warner(aTHX_ WARN_DEPRECATED,
2545 "In %s, line %d, starting in Perl v5.30, %s() will"
2546 " require an additional parameter. Avoid this"
2547 " message by converting to use %s().\n",
2548 file, line, name, alternative);
2549 }
2550 }
2551}
2552
bde6a22d 2553bool
34aeb2e9
KW
2554Perl__is_utf8_FOO(pTHX_ U8 classnum,
2555 const U8 *p,
2556 const char * const name,
2557 const char * const alternative,
2558 const bool use_utf8,
2559 const bool use_locale,
2560 const char * const file,
2561 const unsigned line)
922e8cb4 2562{
922e8cb4
KW
2563 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2564
34aeb2e9
KW
2565 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2566
2567 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2568 SV * invlist;
2569
2570 switch (classnum) {
2571 case _CC_WORDCHAR:
2572 case _CC_DIGIT:
2573 case _CC_ALPHA:
2574 case _CC_LOWER:
2575 case _CC_UPPER:
2576 case _CC_PUNCT:
2577 case _CC_PRINT:
2578 case _CC_ALPHANUMERIC:
2579 case _CC_GRAPH:
2580 case _CC_CASED:
2581
2582 return is_utf8_common(p,
2583 &PL_utf8_swash_ptrs[classnum],
2584 swash_property_names[classnum],
2585 PL_XPosix_ptrs[classnum]);
2586
2587 case _CC_SPACE:
2588 return is_XPERLSPACE_high(p);
2589 case _CC_BLANK:
2590 return is_HORIZWS_high(p);
2591 case _CC_XDIGIT:
2592 return is_XDIGIT_high(p);
2593 case _CC_CNTRL:
2594 return 0;
2595 case _CC_ASCII:
2596 return 0;
2597 case _CC_VERTSPACE:
2598 return is_VERTWS_high(p);
2599 case _CC_IDFIRST:
2600 if (! PL_utf8_perl_idstart) {
2601 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2602 }
2603 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
2604 case _CC_IDCONT:
2605 if (! PL_utf8_perl_idcont) {
2606 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2607 }
2608 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
2609 }
2610 }
2611
2612 /* idcont is the same as wordchar below 256 */
2613 if (classnum == _CC_IDCONT) {
2614 classnum = _CC_WORDCHAR;
2615 }
2616 else if (classnum == _CC_IDFIRST) {
2617 if (*p == '_') {
2618 return TRUE;
2619 }
2620 classnum = _CC_ALPHA;
2621 }
2622
2623 if (! use_locale) {
2624 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2625 return _generic_isCC(*p, classnum);
2626 }
922e8cb4 2627
34aeb2e9
KW
2628 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2629 }
2630 else {
2631 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2632 return isFOO_lc(classnum, *p);
2633 }
2634
2635 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2636 }
2637
2638 NOT_REACHED; /* NOTREACHED */
922e8cb4
KW
2639}
2640
2641bool
da8c1a98
KW
2642Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2643 const U8 * const e)
2644{
2645 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2646
2647 assert(classnum < _FIRST_NON_SWASH_CC);
2648
2649 return is_utf8_common_with_len(p,
2650 e,
2651 &PL_utf8_swash_ptrs[classnum],
2652 swash_property_names[classnum],
2653 PL_XPosix_ptrs[classnum]);
2654}
2655
2656bool
da8c1a98
KW
2657Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2658{
2659 SV* invlist = NULL;
2660
2661 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2662
2663 if (! PL_utf8_perl_idstart) {
2664 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2665 }
2666 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2667 "_Perl_IDStart", invlist);
2668}
2669
2670bool
f2645549 2671Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2672{
f2645549 2673 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2674
2675 if (*p == '_')
2676 return TRUE;
f25ce844 2677 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2678}
2679
2680bool
da8c1a98
KW
2681Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2682{
2683 SV* invlist = NULL;
2684
2685 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2686
2687 if (! PL_utf8_perl_idcont) {
2688 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2689 }
2690 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2691 "_Perl_IDCont", invlist);
2692}
2693
2694bool
f2645549 2695Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2696{
f2645549 2697 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2698
f25ce844 2699 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2700}
2701
2702bool
f2645549 2703Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2704{
f2645549 2705 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2706
f25ce844 2707 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2708}
2709
2710bool
7dbf68d2
KW
2711Perl__is_utf8_mark(pTHX_ const U8 *p)
2712{
7dbf68d2
KW
2713 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2714
f25ce844 2715 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2716}
2717
6b5c0936 2718/*
87cea99e 2719=for apidoc to_utf8_case
6b5c0936 2720
a239b1e2
KW
2721Instead use the appropriate one of L</toUPPER_utf8_safe>,
2722L</toTITLE_utf8_safe>,
2723L</toLOWER_utf8_safe>,
2724or L</toFOLD_utf8_safe>.
9da1e7cb 2725
6fae5207 2726C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2727the character that is being converted. This routine assumes that the character
2728at C<p> is well-formed.
6b5c0936 2729
6fae5207
KW
2730C<ustrp> is a pointer to the character buffer to put the
2731conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2732of the result.
2733
6fae5207 2734C<swashp> is a pointer to the swash to use.
6b5c0936 2735
a1433954 2736Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2737and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2738but not always, a multicharacter mapping), is tried first.
6b5c0936 2739
4a8240a3
KW
2740C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2741any special mappings; C<""> means to use the special mappings. Values other
2742than these two are treated as the name of the hash containing the special
2743mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2744
796b6530
KW
2745C<normal> is a string like C<"ToLower"> which means the swash
2746C<%utf8::ToLower>.
0134edef 2747
760c7c2f
KW
2748Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2749unless those are turned off.
2750
0134edef 2751=cut */
6b5c0936 2752
2104c8d9 2753UV
9a957fbc
AL
2754Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2755 SV **swashp, const char *normal, const char *special)
a0ed51b3 2756{
b9992569
KW
2757 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2758
2759 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2760}
2761
2762 /* change namve uv1 to 'from' */
6a4a25f4 2763STATIC UV
b9992569
KW
2764S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2765 SV **swashp, const char *normal, const char *special)
2766{
0134edef 2767 STRLEN len = 0;
7918f24d 2768
b9992569 2769 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2770
36eaa811
KW
2771 /* For code points that don't change case, we already know that the output
2772 * of this function is the unchanged input, so we can skip doing look-ups
2773 * for them. Unfortunately the case-changing code points are scattered
2774 * around. But there are some long consecutive ranges where there are no
2775 * case changing code points. By adding tests, we can eliminate the lookup
2776 * for all the ones in such ranges. This is currently done here only for
2777 * just a few cases where the scripts are in common use in modern commerce
2778 * (and scripts adjacent to those which can be included without additional
2779 * tests). */
2780
2781 if (uv1 >= 0x0590) {
2782 /* This keeps from needing further processing the code points most
2783 * likely to be used in the following non-cased scripts: Hebrew,
2784 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2785 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2786 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2787 if (uv1 < 0x10A0) {
2788 goto cases_to_self;
2789 }
2790
2791 /* The following largish code point ranges also don't have case
2792 * changes, but khw didn't think they warranted extra tests to speed
2793 * them up (which would slightly slow down everything else above them):
2794 * 1100..139F Hangul Jamo, Ethiopic
2795 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2796 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2797 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2798 * Combining Diacritical Marks Extended, Balinese,
2799 * Sundanese, Batak, Lepcha, Ol Chiki
2800 * 2000..206F General Punctuation
2801 */
2802
2803 if (uv1 >= 0x2D30) {
2804
2805 /* This keeps the from needing further processing the code points
2806 * most likely to be used in the following non-cased major scripts:
2807 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2808 *
2809 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2810 * event that Unicode eventually allocates the unused block as of
2811 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2812 * that the test suite will start having failures to alert you
2813 * should that happen) */
2814 if (uv1 < 0xA640) {
2815 goto cases_to_self;
2816 }
2817
2818 if (uv1 >= 0xAC00) {
2819 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2820 if (ckWARN_d(WARN_SURROGATE)) {
2821 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2822 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2823 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2824 }
2825 goto cases_to_self;
2826 }
36eaa811
KW
2827
2828 /* AC00..FAFF Catches Hangul syllables and private use, plus
2829 * some others */
2830 if (uv1 < 0xFB00) {
2831 goto cases_to_self;
2832
2833 }
2834
5af9bc97
KW
2835 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2836 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2837 && ckWARN_d(WARN_DEPRECATED))
2838 {
2839 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2840 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2841 }
2842 if (ckWARN_d(WARN_NON_UNICODE)) {
2843 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2844 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2845 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2846 }
2847 goto cases_to_self;
2848 }
3bfc1e70
KW
2849#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2850 if (UNLIKELY(uv1
2851 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2852 {
2853
2854 /* As of this writing, this means we avoid swash creation
2855 * for anything beyond low Plane 1 */
2856 goto cases_to_self;
2857 }
2858#endif
36eaa811
KW
2859 }
2860 }
9ae3ac1a 2861
36eaa811
KW
2862 /* Note that non-characters are perfectly legal, so no warning should
2863 * be given. There are so few of them, that it isn't worth the extra
2864 * tests to avoid swash creation */
9ae3ac1a
KW
2865 }
2866
0134edef 2867 if (!*swashp) /* load on-demand */
5ab9d2ef 2868 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2869
a6f87d8c 2870 if (special) {
0134edef 2871 /* It might be "special" (sometimes, but not always,
2a37f04d 2872 * a multicharacter mapping) */
4a8240a3 2873 HV *hv = NULL;
b08cf34e
JH
2874 SV **svp;
2875
4a8240a3
KW
2876 /* If passed in the specials name, use that; otherwise use any
2877 * given in the swash */
2878 if (*special != '\0') {
2879 hv = get_hv(special, 0);
2880 }
2881 else {
2882 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2883 if (svp) {
2884 hv = MUTABLE_HV(SvRV(*svp));
2885 }
2886 }
2887
176fe009 2888 if (hv
5f560d8a 2889 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2890 && (*svp))
2891 {
cfd0369c 2892 const char *s;
47654450 2893
cfd0369c 2894 s = SvPV_const(*svp, len);
47654450 2895 if (len == 1)
f4cd282c 2896 /* EIGHTBIT */
c80e42f3 2897 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2898 else {
d2dcd0fb 2899 Copy(s, ustrp, len, U8);
29e98929 2900 }
983ffd37 2901 }
0134edef
JH
2902 }
2903
2904 if (!len && *swashp) {
4a4088c4 2905 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2906
0134edef
JH
2907 if (uv2) {
2908 /* It was "normal" (a single character mapping). */
f4cd282c 2909 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2910 }
2911 }
1feea2c7 2912
cbe07460
KW
2913 if (len) {
2914 if (lenp) {
2915 *lenp = len;
2916 }
2917 return valid_utf8_to_uvchr(ustrp, 0);
2918 }
2919
2920 /* Here, there was no mapping defined, which means that the code point maps
2921 * to itself. Return the inputs */
e24dfe9c 2922 cases_to_self:
bfdf22ec 2923 len = UTF8SKIP(p);
ca9fab46
KW
2924 if (p != ustrp) { /* Don't copy onto itself */
2925 Copy(p, ustrp, len, U8);
2926 }
0134edef 2927
2a37f04d
JH
2928 if (lenp)
2929 *lenp = len;
2930
f4cd282c 2931 return uv1;
cbe07460 2932
a0ed51b3
LW
2933}
2934
051a06d4 2935STATIC UV
357aadde 2936S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2937{
4a4088c4 2938 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2939 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2940 * result contains a character that crosses the 255/256 boundary, disallow
2941 * the change, and return the original code point. See L<perlfunc/lc> for
2942 * why;
051a06d4 2943 *
a1433954
KW
2944 * p points to the original string whose case was changed; assumed
2945 * by this routine to be well-formed
051a06d4
KW
2946 * result the code point of the first character in the changed-case string
2947 * ustrp points to the changed-case string (<result> represents its first char)
2948 * lenp points to the length of <ustrp> */
2949
2950 UV original; /* To store the first code point of <p> */
2951
2952 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2953
a4f12ed7 2954 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2955
2956 /* We know immediately if the first character in the string crosses the
2957 * boundary, so can skip */
2958 if (result > 255) {
2959
2960 /* Look at every character in the result; if any cross the
2961 * boundary, the whole thing is disallowed */
2962 U8* s = ustrp + UTF8SKIP(ustrp);
2963 U8* e = ustrp + *lenp;
2964 while (s < e) {
a4f12ed7 2965 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2966 goto bad_crossing;
2967 }
2968 s += UTF8SKIP(s);
2969 }
2970
613abc6d
KW
2971 /* Here, no characters crossed, result is ok as-is, but we warn. */
2972 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2973 return result;
2974 }
2975
7b52d656 2976 bad_crossing:
051a06d4
KW
2977
2978 /* Failed, have to return the original */
4b88fb76 2979 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2980
2981 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2982 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2983 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2984 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2985 OP_DESC(PL_op),
ab0b796c
KW
2986 original,
2987 original);
051a06d4
KW
2988 Copy(p, ustrp, *lenp, char);
2989 return original;
2990}
2991
eaf412bf
KW
2992/* The process for changing the case is essentially the same for the four case
2993 * change types, except there are complications for folding. Otherwise the
2994 * difference is only which case to change to. To make sure that they all do
2995 * the same thing, the bodies of the functions are extracted out into the
2996 * following two macros. The functions are written with the same variable
2997 * names, and these are known and used inside these macros. It would be
2998 * better, of course, to have inline functions to do it, but since different
2999 * macros are called, depending on which case is being changed to, this is not
3000 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3001 * function can start with the common start macro, then finish with its special
3002 * handling; while the other three cases can just use the common end macro.
3003 *
3004 * The algorithm is to use the proper (passed in) macro or function to change
3005 * the case for code points that are below 256. The macro is used if using
3006 * locale rules for the case change; the function if not. If the code point is
3007 * above 255, it is computed from the input UTF-8, and another macro is called
3008 * to do the conversion. If necessary, the output is converted to UTF-8. If
3009 * using a locale, we have to check that the change did not cross the 255/256
3010 * boundary, see check_locale_boundary_crossing() for further details.
3011 *
3012 * The macros are split with the correct case change for the below-256 case
3013 * stored into 'result', and in the middle of an else clause for the above-255
3014 * case. At that point in the 'else', 'result' is not the final result, but is
3015 * the input code point calculated from the UTF-8. The fold code needs to
3016 * realize all this and take it from there.
3017 *
3018 * If you read the two macros as sequential, it's easier to understand what's
3019 * going on. */
3020#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3021 L1_func_extra_param) \
a239b1e2
KW
3022 if (e == NULL) e = p + UTF8SKIP(p); \
3023 \
eaf412bf
KW
3024 if (flags & (locale_flags)) { \
3025 /* Treat a UTF-8 locale as not being in locale at all */ \
3026 if (IN_UTF8_CTYPE_LOCALE) { \
3027 flags &= ~(locale_flags); \
3028 } \
3029 else { \
3030 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3031 } \
3032 } \
3033 \
3034 if (UTF8_IS_INVARIANT(*p)) { \
3035 if (flags & (locale_flags)) { \
3036 result = LC_L1_change_macro(*p); \
3037 } \
3038 else { \
3039 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3040 } \
3041 } \
a239b1e2 3042 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
eaf412bf
KW
3043 if (flags & (locale_flags)) { \
3044 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3045 *(p+1))); \
3046 } \
3047 else { \
3048 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3049 ustrp, lenp, L1_func_extra_param); \
3050 } \
3051 } \
fa8ab374
KW
3052 else { /* malformed UTF-8 or ord above 255 */ \
3053 STRLEN len_result; \
fa8ab374
KW
3054 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3055 if (len_result == (STRLEN) -1) { \
3056 _force_out_malformed_utf8_message(p, e, \
3057 _UTF8_NO_CONFIDENCE_IN_CURLEN, \
3058 1 /* Die */ ); \
3059 }
eaf412bf
KW
3060
3061#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3062 result = change_macro(result, p, ustrp, lenp); \
3063 \
3064 if (flags & (locale_flags)) { \
3065 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3066 } \
3067 return result; \
3068 } \
3069 \
3070 /* Here, used locale rules. Convert back to UTF-8 */ \
3071 if (UTF8_IS_INVARIANT(result)) { \
3072 *ustrp = (U8) result; \
3073 *lenp = 1; \
3074 } \
3075 else { \
3076 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3077 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3078 *lenp = 2; \
3079 } \
3080 \
3081 return result;
3082
d3e79532 3083/*
87cea99e 3084=for apidoc to_utf8_upper
d3e79532 3085
a239b1e2 3086Instead use L</toUPPER_utf8_safe>.
a1433954 3087
d3e79532
JH
3088=cut */
3089
051a06d4 3090/* Not currently externally documented, and subject to change:
31f05a37
KW
3091 * <flags> is set iff iff the rules from the current underlying locale are to
3092 * be used. */
051a06d4 3093
2104c8d9 3094UV
a239b1e2 3095Perl__to_utf8_upper_flags(pTHX_ const U8 *p, const U8 *e, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3096{
051a06d4
KW
3097 UV result;
3098
3099 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3100
eaf412bf
KW
3101 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3102 /* 2nd char of uc(U+DF) is 'S' */
3103 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3104 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3105}
a0ed51b3 3106
d3e79532 3107/*
87cea99e 3108=for apidoc to_utf8_title
d3e79532 3109
a239b1e2 3110Instead use L</toTITLE_utf8_safe>.
a1433954 3111
d3e79532
JH
3112=cut */
3113
051a06d4 3114/* Not currently externally documented, and subject to change:
31f05a37
KW
3115 * <flags> is set iff the rules from the current underlying locale are to be
3116 * used. Since titlecase is not defined in POSIX, for other than a
3117 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3118 */
051a06d4 3119
983ffd37 3120UV
a239b1e2 3121Perl__to_utf8_title_flags(pTHX_ const U8 *p, const U8 *e, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 3122{
051a06d4
KW
3123 UV result;
3124
3125 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3126
eaf412bf
KW
3127 /* 2nd char of ucfirst(U+DF) is 's' */
3128 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3129 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3130}
3131
d3e79532 3132/*
87cea99e 3133=for apidoc to_utf8_lower
d3e79532 3134
a239b1e2 3135Instead use L</toLOWER_utf8_safe>.
a1433954 3136
d3e79532
JH
3137=cut */
3138
051a06d4 3139/* Not currently externally documented, and subject to change:
31f05a37
KW
3140 * <flags> is set iff iff the rules from the current underlying locale are to
3141 * be used.
3142 */
051a06d4 3143
2104c8d9 3144UV
a239b1e2 3145Perl__to_utf8_lower_flags(pTHX_ const U8 *p, const U8 *e, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3146{
051a06d4
KW
3147 UV result;
3148
051a06d4 3149 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3150
eaf412bf
KW
3151 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3152 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3153}
3154
d3e79532 3155/*
87cea99e 3156=for apidoc to_utf8_fold
d3e79532 3157
a239b1e2 3158Instead use L</toFOLD_utf8_safe>.
a1433954 3159
d3e79532
JH
3160=cut */
3161
051a06d4
KW
3162/* Not currently externally documented, and subject to change,
3163 * in <flags>
31f05a37
KW
3164 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3165 * locale are to be used.
051a06d4
KW
3166 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3167 * otherwise simple folds
a0270393
KW
3168 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3169 * prohibited
445bf929 3170 */
36bb2ab6 3171
b4e400f9 3172UV
a239b1e2 3173Perl__to_utf8_fold_flags(pTHX_ const U8 *p, const U8 *e, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3174{
051a06d4
KW
3175 UV result;
3176
36bb2ab6 3177 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3178
a0270393
KW
3179 /* These are mutually exclusive */
3180 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3181
50ba90ff
KW
3182 assert(p != ustrp); /* Otherwise overwrites */
3183
eaf412bf
KW
3184 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3185 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3186
eaf412bf 3187 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3188
1ca267a5
KW
3189 if (flags & FOLD_FLAGS_LOCALE) {
3190
76f2ffcd 3191# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3192 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3193
3194# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3195# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3196
3197 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3198
538e84ed
KW
3199 /* Special case these two characters, as what normally gets
3200 * returned under locale doesn't work */
76f2ffcd
KW
3201 if (UTF8SKIP(p) == cap_sharp_s_len
3202 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3203 {
ab0b796c
KW
3204 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3205 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3206 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3207 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3208 goto return_long_s;
3209 }
0766489e
KW
3210 else
3211#endif
3212 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3213 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3214 {
ab0b796c
KW
3215 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3216 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3217 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3218 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3219 goto return_ligature_st;
3220 }
74894415
KW
3221
3222#if UNICODE_MAJOR_VERSION == 3 \
3223 && UNICODE_DOT_VERSION == 0 \
3224 && UNICODE_DOT_DOT_VERSION == 1
3225# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3226
3227 /* And special case this on this Unicode version only, for the same
3228 * reaons the other two are special cased. They would cross the
3229 * 255/256 boundary which is forbidden under /l, and so the code
3230 * wouldn't catch that they are equivalent (which they are only in
3231 * this release) */
3232 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3233 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3234 {
3235 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3236 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3237 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3238 "resolved to \"\\x{0131}\".");
3239 goto return_dotless_i;
3240 }
3241#endif
3242
357aadde 3243 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3244 }
a0270393
KW
3245 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3246 return result;
3247 }
3248 else {
4a4088c4 3249 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3250 * character above the ASCII range, and the result should not
3251 * contain an ASCII character. */
a0270393
KW
3252
3253 UV original; /* To store the first code point of <p> */
3254
3255 /* Look at every character in the result; if any cross the
3256 * boundary, the whole thing is disallowed */
3257 U8* s = ustrp;
3258 U8* e = ustrp + *lenp;
3259 while (s < e) {
3260 if (isASCII(*s)) {
3261 /* Crossed, have to return the original */
3262 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3263
9fc2026f 3264 /* But in these instances, there is an alternative we can
1ca267a5 3265 * return that is valid */
0766489e
KW
3266 if (original == LATIN_SMALL_LETTER_SHARP_S
3267#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3268 || original == LATIN_CAPITAL_LETTER_SHARP_S
3269#endif
3270 ) {
1ca267a5
KW
3271 goto return_long_s;
3272 }
9fc2026f
KW
3273 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3274 goto return_ligature_st;
3275 }
74894415
KW
3276#if UNICODE_MAJOR_VERSION == 3 \
3277 && UNICODE_DOT_VERSION == 0 \
3278 && UNICODE_DOT_DOT_VERSION == 1
3279
3280 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3281 goto return_dotless_i;
3282 }
3283#endif
a0270393
KW
3284 Copy(p, ustrp, *lenp, char);
3285 return original;
3286 }
3287 s += UTF8SKIP(s);
3288 }
051a06d4 3289
a0270393
KW
3290 /* Here, no characters crossed, result is ok as-is */
3291 return result;
3292 }
051a06d4
KW
3293 }
3294
4a4088c4 3295 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3296 if (UTF8_IS_INVARIANT(result)) {
3297 *ustrp = (U8) result;
3298 *lenp = 1;
3299 }
3300 else {
62cb07ea
KW
3301 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3302 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3303 *lenp = 2;
3304 }
3305
051a06d4 3306 return result;
1ca267a5
KW
3307
3308 return_long_s:
3309 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3310 * folds to a string of two of these characters. By returning this
3311 * instead, then, e.g.,
3312 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3313 * works. */
3314
3315 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3316 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3317 ustrp, *lenp, U8);
3318 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3319
3320 return_ligature_st:
3321 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3322 * have the other one fold to it */
3323
3324 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3325 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3326 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3327
3328#if UNICODE_MAJOR_VERSION == 3 \
3329 && UNICODE_DOT_VERSION == 0 \
3330 && UNICODE_DOT_DOT_VERSION == 1
3331
3332 return_dotless_i:
3333 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3334 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3335 return LATIN_SMALL_LETTER_DOTLESS_I;
3336
3337#endif
3338
a0ed51b3
LW
3339}
3340
711a919c 3341/* Note:
f90a9a02 3342 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3343 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3344 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3345 */
c4a5db0c 3346
a0ed51b3 3347SV*
7fc63493 3348Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3349{
c4a5db0c
KW
3350 PERL_ARGS_ASSERT_SWASH_INIT;
3351
3352 /* Returns a copy of a swash initiated by the called function. This is the
3353 * public interface, and returning a copy prevents others from doing
3354 * mischief on the original */
3355
5d3d13d1 3356 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3357}
3358
3359SV*
5d3d13d1 3360Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3361{
2c1f00b9
YO
3362
3363 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3364 * use the following define */
3365
3366#define CORE_SWASH_INIT_RETURN(x) \
3367 PL_curpm= old_PL_curpm; \
3368 return x
3369
c4a5db0c 3370 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3371 * by calling utf8_heavy.pl in the general case. The returned value may be
3372 * the swash's inversion list instead if the input parameters allow it.
3373 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3374 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3375 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3376 *
3377 * This interface should only be used by functions that won't destroy or
3378 * adversely change the swash, as doing so affects all other uses of the
3379 * swash in the program; the general public should use 'Perl_swash_init'
3380 * instead.
3381 *
3382 * pkg is the name of the package that <name> should be in.
3383 * name is the name of the swash to find. Typically it is a Unicode
3384 * property name, including user-defined ones
3385 * listsv is a string to initialize the swash with. It must be of the form
3386 * documented as the subroutine return value in
3387 * L<perlunicode/User-Defined Character Properties>
3388 * minbits is the number of bits required to represent each data element.
3389 * It is '1' for binary properties.
3390 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3391 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3392 * flags_p if non-NULL is the address of various input and output flag bits
3393 * to the routine, as follows: ('I' means is input to the routine;
3394 * 'O' means output from the routine. Only flags marked O are
3395 * meaningful on return.)
3396 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3397 * came from a user-defined property. (I O)
5d3d13d1
KW
3398 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3399 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3400 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3401 * return of an inversion list instead of a swash hash if this routine
3402 * thinks that would result in faster execution of swash_fetch() later
3403 * on. (I)
9a53f6cf
KW
3404 *
3405 * Thus there are three possible inputs to find the swash: <name>,
3406 * <listsv>, and <invlist>. At least one must be specified. The result
3407 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3408 * actions can intersect, etc. what <name> gives. To avoid going out to
3409 * disk at all, <invlist> should specify completely what the swash should
3410 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3411 *
3412 * <invlist> is only valid for binary properties */
c4a5db0c 3413
2c1f00b9
YO
3414 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3415
c4a5db0c 3416 SV* retval = &PL_sv_undef;
83199d38 3417 HV* swash_hv = NULL;
87367d5f
KW
3418 const int invlist_swash_boundary =
3419 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3420 ? 512 /* Based on some benchmarking, but not extensive, see commit
3421 message */
3422 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3423
3424 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3425 assert(! invlist || minbits == 1);
3426
2c1f00b9
YO
3427 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3428 that triggered the swash init and the swash init perl logic itself.
3429 See perl #122747 */
3430
9a53f6cf
KW
3431 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3432 * so */
3433 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3434 dSP;
3435 const size_t pkg_len = strlen(pkg);
3436 const size_t name_len = strlen(name);
3437 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3438 SV* errsv_save;
3439 GV *method;
3440
3441 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3442
3443 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3444 ENTER;
69794297 3445 SAVEHINTS();
2782061f 3446 save_re_context();
650f067c
JL
3447 /* We might get here via a subroutine signature which uses a utf8
3448 * parameter name, at which point PL_subname will have been set
3449 * but not yet used. */
3450 save_item(PL_subname);
69794297
KW
3451 if (PL_parser && PL_parser->error_count)
3452 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3453 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3454 if (!method) { /* demand load UTF-8 */
69794297 3455 ENTER;
db2c6cb3
FC
3456 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3457 GvSV(PL_errgv) = NULL;
1a419e6b 3458#ifndef NO_TAINT_SUPPORT
69794297
KW
3459 /* It is assumed that callers of this routine are not passing in
3460 * any user derived data. */
2782061f
DM
3461 /* Need to do this after save_re_context() as it will set
3462 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3463 * in Perl_magic_get). Even line to create errsv_save can turn on
3464 * PL_tainted. */
284167a5
SM
3465 SAVEBOOL(TAINT_get);
3466 TAINT_NOT;
3467#endif
69794297
KW
3468 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3469 NULL);
eed484f9 3470 {
db2c6cb3
FC
3471 /* Not ERRSV, as there is no need to vivify a scalar we are
3472 about to discard. */
3473 SV * const errsv = GvSV(PL_errgv);
3474 if (!SvTRUE(errsv)) {
3475 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3476 SvREFCNT_dec(errsv);
3477 }
eed484f9 3478 }
69794297
KW
3479 LEAVE;
3480 }
3481 SPAGAIN;
3482 PUSHMARK(SP);
3483 EXTEND(SP,5);
3484 mPUSHp(pkg, pkg_len);
3485 mPUSHp(name, name_len);
3486 PUSHs(listsv);
3487 mPUSHi(minbits);
3488 mPUSHi(none);
3489 PUTBACK;
db2c6cb3
FC
3490 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3491 GvSV(PL_errgv) = NULL;
69794297
KW
3492 /* If we already have a pointer to the method, no need to use
3493 * call_method() to repeat the lookup. */
c41800a8
KW
3494 if (method
3495 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3496 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3497 {
3498 retval = *PL_stack_sp--;
3499 SvREFCNT_inc(retval);
3500 }
eed484f9 3501 {
db2c6cb3
FC
3502 /* Not ERRSV. See above. */
3503 SV * const errsv = GvSV(PL_errgv);
3504 if (!SvTRUE(errsv)) {
3505 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3506 SvREFCNT_dec(errsv);
3507 }
eed484f9 3508 }
ce3b816e 3509 LEAVE;
69794297
KW
3510 POPSTACK;
3511 if (IN_PERL_COMPILETIME) {
3512 CopHINTS_set(PL_curcop, PL_hints);
3513 }
3514 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3515 if (SvPOK(retval)) {
69794297
KW
3516
3517 /* If caller wants to handle missing properties, let them */
5d3d13d1 3518 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3519 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3520 }
3521 Perl_croak(aTHX_
147e3846 3522 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3523 SVfARG(retval));
a25b5927 3524 NOT_REACHED; /* NOTREACHED */
d95e4a00 3525 }
69794297 3526 }
9a53f6cf 3527 } /* End of calling the module to find the swash */
36eb48b4 3528
83199d38
KW
3529 /* If this operation fetched a swash, and we will need it later, get it */
3530 if (retval != &PL_sv_undef
3531 && (minbits == 1 || (flags_p
3532 && ! (*flags_p
3533 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3534 {
3535 swash_hv = MUTABLE_HV(SvRV(retval));
3536
3537 /* If we don't already know that there is a user-defined component to
3538 * this swash, and the user has indicated they wish to know if there is
3539 * one (by passing <flags_p>), find out */
3540 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3541 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3542 if (user_defined && SvUV(*user_defined)) {
3543 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3544 }
3545 }
3546 }
3547
36eb48b4
KW
3548 /* Make sure there is an inversion list for binary properties */
3549 if (minbits == 1) {
3550 SV** swash_invlistsvp = NULL;
3551 SV* swash_invlist = NULL;
9a53f6cf 3552 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3553 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3554 an unclaimed reference count */
36eb48b4 3555
9a53f6cf 3556 /* If this operation fetched a swash, get its already existing
83199d38 3557 * inversion list, or create one for it */
36eb48b4 3558
83199d38 3559 if (swash_hv) {
5c9f4bd2 3560 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3561 if (swash_invlistsvp) {
3562 swash_invlist = *swash_invlistsvp;
3563 invlist_in_swash_is_valid = TRUE;
3564 }
3565 else {
36eb48b4 3566 swash_invlist = _swash_to_invlist(retval);
02c85471 3567 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3568 }
3569 }
3570
3571 /* If an inversion list was passed in, have to include it */
3572 if (invlist) {
3573
3574 /* Any fetched swash will by now have an inversion list in it;
3575 * otherwise <swash_invlist> will be NULL, indicating that we
3576 * didn't fetch a swash */
3577 if (swash_invlist) {
3578
3579 /* Add the passed-in inversion list, which invalidates the one
3580 * already stored in the swash */
3581 invlist_in_swash_is_valid = FALSE;
eee4c920 3582 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3583 _invlist_union(invlist, swash_invlist, &swash_invlist);
3584 }
3585 else {
3586
87367d5f
KW
3587 /* Here, there is no swash already. Set up a minimal one, if
3588 * we are going to return a swash */
3589 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3590 swash_hv = newHV();
4aca0fe6 3591 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3592 }
9a53f6cf
KW
3593 swash_invlist = invlist;
3594 }
9a53f6cf
KW
3595 }
3596
3597 /* Here, we have computed the union of all the passed-in data. It may
3598 * be that there was an inversion list in the swash which didn't get
538e84ed 3599 * touched; otherwise save the computed one */
87367d5f
KW
3600 if (! invlist_in_swash_is_valid
3601 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3602 {
5c9f4bd2 3603 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3604 {
3605 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3606 }
cc34d8c5
FC
3607 /* We just stole a reference count. */
3608 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3609 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3610 }
87367d5f 3611
ee3222e3 3612 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3613 SvREADONLY_on(swash_invlist);
3614
c41800a8 3615 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3616 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3617 SvREFCNT_dec(retval);
02c85471
FC
3618 if (!swash_invlist_unclaimed)
3619 SvREFCNT_inc_simple_void_NN(swash_invlist);
3620 retval = newRV_noinc(swash_invlist);
87367d5f 3621 }
36eb48b4
KW
3622 }
3623
2c1f00b9
YO
3624 CORE_SWASH_INIT_RETURN(retval);
3625#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3626}
3627
035d37be
JH
3628
3629/* This API is wrong for special case conversions since we may need to
3630 * return several Unicode characters for a single Unicode character
3631 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3632 * the lower-level routine, and it is similarly broken for returning
38684baa 3633 * multiple values. --jhi
b9992569 3634 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3635/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3636
3637/* Note:
3638 * Returns the value of property/mapping C<swash> for the first character
3639 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3640 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3641 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3642 *
3643 * A "swash" is a hash which contains initially the keys/values set up by
3644 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3645 * property for all possible code points. Things are stored in a compact form
3646 * (see utf8_heavy.pl) so that calculation is required to find the actual
3647 * property value for a given code point. As code points are looked up, new
3648 * key/value pairs are added to the hash, so that the calculation doesn't have
3649 * to ever be re-done. Further, each calculation is done, not just for the
3650 * desired one, but for a whole block of code points adjacent to that one.
3651 * For binary properties on ASCII machines, the block is usually for 64 code
3652 * points, starting with a code point evenly divisible by 64. Thus if the
3653 * property value for code point 257 is requested, the code goes out and
3654 * calculates the property values for all 64 code points between 256 and 319,
3655 * and stores these as a single 64-bit long bit vector, called a "swatch",
3656 * under the key for code point 256. The key is the UTF-8 encoding for code
3657 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3658 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3659 * for code point 258 is then requested, this code realizes that it would be
3660 * stored under the key for 256, and would find that value and extract the
3661 * relevant bit, offset from 256.
3662 *
3663 * Non-binary properties are stored in as many bits as necessary to represent
3664 * their values (32 currently, though the code is more general than that), not
fc273927 3665 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3666 * vector that encompasses the property values for all code points whose UTF-8
3667 * representations are represented by the key. That is, for all code points
3668 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3669 * bytes of that.
680c470c 3670 */
a0ed51b3 3671UV
680c470c 3672Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3673{
ef8f7699 3674 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3675 U32 klen;
3676 U32 off;
9b56a019 3677 STRLEN slen = 0;
7d85a32c 3678 STRLEN needents;
cfd0369c 3679 const U8 *tmps = NULL;
979f2922 3680 SV *swatch;
08fb1ac5 3681 const U8 c = *ptr;
3568d838 3682
7918f24d
NC
3683 PERL_ARGS_ASSERT_SWASH_FETCH;
3684
87367d5f
KW
3685 /* If it really isn't a hash, it isn't really swash; must be an inversion
3686 * list */
3687 if (SvTYPE(hv) != SVt_PVHV) {
3688 return _invlist_contains_cp((SV*)hv,
3689 (do_utf8)
3690 ? valid_utf8_to_uvchr(ptr, NULL)
3691 : c);
3692 }
3693
08fb1ac5
KW
3694 /* We store the values in a "swatch" which is a vec() value in a swash
3695 * hash. Code points 0-255 are a single vec() stored with key length
3696 * (klen) 0. All other code points have a UTF-8 representation
3697 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3698 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3699 * length for them is the length of the encoded char - 1. ptr[klen] is the
3700 * final byte in the sequence representing the character */
3701 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3702 klen = 0;
3703 needents = 256;
3704 off = c;
3568d838 3705 }
08fb1ac5
KW
3706 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3707 klen = 0;
3708 needents = 256;
a62b247b 3709 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3710 }
3711 else {
08fb1ac5
KW
3712 klen = UTF8SKIP(ptr) - 1;
3713
3714 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3715 * the vec is the final byte in the sequence. (In EBCDIC this is
3716 * converted to I8 to get consecutive values.) To help you visualize
3717 * all this:
3718 * Straight 1047 After final byte
3719 * UTF-8 UTF-EBCDIC I8 transform
3720 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3721 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3722 * ...
3723 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3724 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3725 * ...
3726 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3727 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3728 * ...
3729 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3730 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3731 * ...
3732 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3733 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3734 *
3735 * (There are no discontinuities in the elided (...) entries.)
3736 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3737 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3738 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3739 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3740 * index into the vec() swatch (after subtracting 0x80, which we
3741 * actually do with an '&').
3742 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3743 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3744 * dicontinuities which g