This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Return REPLACEMENT for UTF-8 overlong malformation
[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)
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) {
702 return 0; /* Above Unicode */
703 }
704
705 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
706 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
707 {
708 return 0; /* Above 31 bits */
709 }
710
711 if (len > 1) {
712 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
713
714 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 715 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
716 {
717 return 0; /* Above Unicode */
718 }
719
720 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 721 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
722 {
723 return 0; /* Surrogate */
724 }
725
726 if ( (flags & UTF8_DISALLOW_NONCHAR)
727 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
728 {
729 return 0; /* Noncharacter code point */
730 }
731 }
732 }
733
734 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
735 for (x = s + 1; x < e; x++) {
736 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
737 return 0;
738 }
739 }
740
af13dd8a 741 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
742 * overlong. */
743 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
744 return 0;
af13dd8a
KW
745 }
746
12a4bed3
KW
747 /* And finally, that the code point represented fits in a word on this
748 * platform */
749 if (does_utf8_overflow(s, e)) {
750 return 0;
35f8c9bd
KW
751 }
752
2b479609 753 return UTF8SKIP(s);
35f8c9bd
KW
754}
755
7cf8d05d
KW
756STATIC char *
757S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
758{
759 /* Returns a mortalized C string that is a displayable copy of the 'len'
760 * bytes starting at 's', each in a \xXY format. */
761
762 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
763 trailing NUL */
764 const U8 * const e = s + len;
765 char * output;
766 char * d;
767
768 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
769
770 Newx(output, output_len, char);
771 SAVEFREEPV(output);
772
773 d = output;
774 for (; s < e; s++) {
775 const unsigned high_nibble = (*s & 0xF0) >> 4;
776 const unsigned low_nibble = (*s & 0x0F);
777
778 *d++ = '\\';
779 *d++ = 'x';
780
781 if (high_nibble < 10) {
782 *d++ = high_nibble + '0';
783 }
784 else {
785 *d++ = high_nibble - 10 + 'a';
786 }
787
788 if (low_nibble < 10) {
789 *d++ = low_nibble + '0';
790 }
791 else {
792 *d++ = low_nibble - 10 + 'a';
793 }
794 }
795
796 *d = '\0';
797 return output;
798}
799
806547a7 800PERL_STATIC_INLINE char *
7cf8d05d
KW
801S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
802
803 /* How many bytes to print */
3cc6a05e 804 STRLEN print_len,
7cf8d05d
KW
805
806 /* Which one is the non-continuation */
807 const STRLEN non_cont_byte_pos,
808
809 /* How many bytes should there be? */
810 const STRLEN expect_len)
806547a7
KW
811{
812 /* Return the malformation warning text for an unexpected continuation
813 * byte. */
814
7cf8d05d 815 const char * const where = (non_cont_byte_pos == 1)
806547a7 816 ? "immediately"
7cf8d05d
KW
817 : Perl_form(aTHX_ "%d bytes",
818 (int) non_cont_byte_pos);
3cc6a05e 819 unsigned int i;
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
3cc6a05e
KW
827 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
828 * length that is larger than is actually available in the buffer. If we
829 * print all the bytes based on that length, we will read past the buffer
830 * end. Often, the strings are NUL terminated, so to lower the chances of
831 * this happening, print the malformed bytes only up through any NUL. */
832 for (i = 1; i < print_len; i++) {
833 if (*(s + i) == '\0') {
834 print_len = i + 1; /* +1 gets the NUL printed */
835 break;
836 }
837 }
838
7cf8d05d
KW
839 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
840 " %s after start byte 0x%02x; need %d bytes, got %d)",
841 malformed_text,
842 _byte_dump_string(s, print_len),
843 *(s + non_cont_byte_pos),
844 where,
845 *s,
846 (int) expect_len,
847 (int) non_cont_byte_pos);
806547a7
KW
848}
849
35f8c9bd
KW
850/*
851
de69f3af 852=for apidoc utf8n_to_uvchr
378516de
KW
853
854THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 855Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 856
9041c2e3 857Bottom level UTF-8 decode routine.
de69f3af 858Returns the native code point value of the first character in the string C<s>,
746afd53
KW
859which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
860C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
861the length, in bytes, of that character.
949cf498
KW
862
863The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
864well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
865causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
866is the next possible position in C<s> that could begin a non-malformed
867character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
868is raised. Some UTF-8 input sequences may contain multiple malformations.
869This function tries to find every possible one in each call, so multiple
870warnings can be raised for each sequence.
949cf498
KW
871
872Various ALLOW flags can be set in C<flags> to allow (and not warn on)
873individual types of malformations, such as the sequence being overlong (that
874is, when there is a shorter sequence that can express the same code point;
875overlong sequences are expressly forbidden in the UTF-8 standard due to
876potential security issues). Another malformation example is the first byte of
877a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
878flags. Even if allowed, this function generally returns the Unicode
879REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
880F<utf8.h> to override this behavior for the overlong malformations, but don't
881do that except for very specialized purposes.
949cf498 882
796b6530 883The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
884flags) malformation is found. If this flag is set, the routine assumes that
885the caller will raise a warning, and this function will silently just set
d088425d
KW
886C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
887
75200dff 888Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 889character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
890in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
891be set to 1. To disambiguate, upon a zero return, see if the first byte of
892C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 893error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
894
895Certain code points are considered problematic. These are Unicode surrogates,
746afd53 896Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 897By default these are considered regular code points, but certain situations
ecc1615f
KW
898warrant special handling for them, which can be specified using the C<flags>
899parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
900three classes are treated as malformations and handled as such. The flags
901C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
902C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
903disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
904restricts the allowed inputs to the strict UTF-8 traditionally defined by
905Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
906definition given by
907L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
908The difference between traditional strictness and C9 strictness is that the
909latter does not forbid non-character code points. (They are still discouraged,
910however.) For more discussion see L<perlunicode/Noncharacter code points>.
911
912The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
913C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
914C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
915raised for their respective categories, but otherwise the code points are
916considered valid (not malformations). To get a category to both be treated as
917a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 918(But note that warnings are not raised if lexically disabled nor if
796b6530 919C<UTF8_CHECK_ONLY> is also specified.)
949cf498 920
760c7c2f
KW
921It is now deprecated to have very high code points (above C<IV_MAX> on the
922platforms) and this function will raise a deprecation warning for these (unless
d5944cab 923such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 924in a 32-bit word.
ab8e6d41
KW
925
926Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
927so using them is more problematic than other above-Unicode code points. Perl
928invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
929likely that non-Perl languages will not be able to read files that contain
aff2be59 930these; nor would Perl understand files
ab8e6d41
KW
931written by something that uses a different extension. For these reasons, there
932is a separate set of flags that can warn and/or disallow these extremely high
933code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
934C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
935are entirely independent from the deprecation warning for code points above
936C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
937code point that needs more than 31 bits to represent. When that happens,
938effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
93932-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
940above-Unicode code points, including these, as malformations; and
941C<UTF8_WARN_SUPER> warns on these.)
942
943On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
944extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
945than on ASCII. Prior to that, code points 2**31 and higher were simply
946unrepresentable, and a different, incompatible method was used to represent
947code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
948and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
949platforms, warning and disallowing 2**31 and higher.
949cf498
KW
950
951All other code points corresponding to Unicode characters, including private
952use and those yet to be assigned, are never considered malformed and never
953warn.
67e989fb 954
37607a96 955=cut
f9380377
KW
956
957Also implemented as a macro in utf8.h
958*/
959
960UV
961Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
962 STRLEN curlen,
963 STRLEN *retlen,
964 const U32 flags)
965{
966 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
967
968 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
969}
970
971/*
972
973=for apidoc utf8n_to_uvchr_error
974
975THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
976Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
977
978This function is for code that needs to know what the precise malformation(s)
979are when an error is found.
980
981It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
982all the others, C<errors>. If this parameter is 0, this function behaves
983identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
984to a C<U32> variable, which this function sets to indicate any errors found.
985Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
986C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
987of these bits will be set if a malformation is found, even if the input
7a65503b 988C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
989exceptions are noted:
990
991=over 4
992
993=item C<UTF8_GOT_ABOVE_31_BIT>
994
995The code point represented by the input UTF-8 sequence occupies more than 31
996bits.
997This bit is set only if the input C<flags> parameter contains either the
998C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
999
1000=item C<UTF8_GOT_CONTINUATION>
1001
1002The input sequence was malformed in that the first byte was a a UTF-8
1003continuation byte.
1004
1005=item C<UTF8_GOT_EMPTY>
1006
1007The input C<curlen> parameter was 0.
1008
1009=item C<UTF8_GOT_LONG>
1010
1011The input sequence was malformed in that there is some other sequence that
1012evaluates to the same code point, but that sequence is shorter than this one.
1013
1014=item C<UTF8_GOT_NONCHAR>
1015
1016The code point represented by the input UTF-8 sequence is for a Unicode
1017non-character code point.
1018This bit is set only if the input C<flags> parameter contains either the
1019C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1020
1021=item C<UTF8_GOT_NON_CONTINUATION>
1022
1023The input sequence was malformed in that a non-continuation type byte was found
1024in a position where only a continuation type one should be.
1025
1026=item C<UTF8_GOT_OVERFLOW>
1027
1028The input sequence was malformed in that it is for a code point that is not
1029representable in the number of bits available in a UV on the current platform.
1030
1031=item C<UTF8_GOT_SHORT>
1032
1033The input sequence was malformed in that C<curlen> is smaller than required for
1034a complete sequence. In other words, the input is for a partial character
1035sequence.
1036
1037=item C<UTF8_GOT_SUPER>
1038
1039The input sequence was malformed in that it is for a non-Unicode code point;
1040that is, one above the legal Unicode maximum.
1041This bit is set only if the input C<flags> parameter contains either the
1042C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1043
1044=item C<UTF8_GOT_SURROGATE>
1045
1046The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1047code point.
1048This bit is set only if the input C<flags> parameter contains either the
1049C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1050
1051=back
1052
133551d8
KW
1053To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1054flag to suppress any warnings, and then examine the C<*errors> return.
1055
f9380377 1056=cut
37607a96 1057*/
67e989fb 1058
a0ed51b3 1059UV
f9380377
KW
1060Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1061 STRLEN curlen,
1062 STRLEN *retlen,
1063 const U32 flags,
1064 U32 * errors)
a0ed51b3 1065{
d4c19fe8 1066 const U8 * const s0 = s;
2b5e7bc2
KW
1067 U8 * send = NULL; /* (initialized to silence compilers' wrong
1068 warning) */
1069 U32 possible_problems = 0; /* A bit is set here for each potential problem
1070 found as we go along */
eb83ed87 1071 UV uv = *s;
2b5e7bc2
KW
1072 STRLEN expectlen = 0; /* How long should this sequence be?
1073 (initialized to silence compilers' wrong
1074 warning) */
e308b348 1075 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1076 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1077 this gets set and discarded */
a0dbb045 1078
2b5e7bc2
KW
1079 /* The below are used only if there is both an overlong malformation and a
1080 * too short one. Otherwise the first two are set to 's0' and 'send', and
1081 * the third not used at all */
1082 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1083 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1084 warning) */
2b5e7bc2 1085 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1086
f9380377
KW
1087 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1088
1089 if (errors) {
1090 *errors = 0;
1091 }
1092 else {
1093 errors = &discard_errors;
1094 }
a0dbb045 1095
eb83ed87
KW
1096 /* The order of malformation tests here is important. We should consume as
1097 * few bytes as possible in order to not skip any valid character. This is
1098 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1099 * http://unicode.org/reports/tr36 for more discussion as to why. For
1100 * example, once we've done a UTF8SKIP, we can tell the expected number of
1101 * bytes, and could fail right off the bat if the input parameters indicate
1102 * that there are too few available. But it could be that just that first
1103 * byte is garbled, and the intended character occupies fewer bytes. If we
1104 * blindly assumed that the first byte is correct, and skipped based on
1105 * that number, we could skip over a valid input character. So instead, we
1106 * always examine the sequence byte-by-byte.
1107 *
1108 * We also should not consume too few bytes, otherwise someone could inject
1109 * things. For example, an input could be deliberately designed to
1110 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1111 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1112 * which is actually part of of the overflowing sequence), that could look
1113 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1114 * sequence and process the rest, inappropriately.
1115 *
1116 * Some possible input sequences are malformed in more than one way. This
1117 * function goes to lengths to try to find all of them. This is necessary
1118 * for correctness, as the inputs may allow one malformation but not
1119 * another, and if we abandon searching for others after finding the
1120 * allowed one, we could allow in something that shouldn't have been.
1121 */
eb83ed87 1122
b5b9af04 1123 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1124 possible_problems |= UTF8_GOT_EMPTY;
1125 curlen = 0;
5a48568d 1126 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1127 goto ready_to_handle_errors;
0c443dc2
JH
1128 }
1129
eb83ed87
KW
1130 expectlen = UTF8SKIP(s);
1131
1132 /* A well-formed UTF-8 character, as the vast majority of calls to this
1133 * function will be for, has this expected length. For efficiency, set
1134 * things up here to return it. It will be overriden only in those rare
1135 * cases where a malformation is found */
1136 if (retlen) {
1137 *retlen = expectlen;
1138 }
1139
1140 /* An invariant is trivially well-formed */
1d72bdf6 1141 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1142 return uv;
a0ed51b3 1143 }
67e989fb 1144
eb83ed87 1145 /* A continuation character can't start a valid sequence */
b5b9af04 1146 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1147 possible_problems |= UTF8_GOT_CONTINUATION;
1148 curlen = 1;
1149 uv = UNICODE_REPLACEMENT;
1150 goto ready_to_handle_errors;
ba210ebe 1151 }
9041c2e3 1152
dcd27b3c
KW
1153 /* Here is not a continuation byte, nor an invariant. The only thing left
1154 * is a start byte (possibly for an overlong) */
1155
534752c1
KW
1156 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1157 * that indicate the number of bytes in the character's whole UTF-8
1158 * sequence, leaving just the bits that are part of the value. */
1159 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1160
e308b348
KW
1161 /* Setup the loop end point, making sure to not look past the end of the
1162 * input string, and flag it as too short if the size isn't big enough. */
1163 send = (U8*) s0;
1164 if (UNLIKELY(curlen < expectlen)) {
1165 possible_problems |= UTF8_GOT_SHORT;
1166 avail_len = curlen;
1167 send += curlen;
1168 }
1169 else {
1170 send += expectlen;
1171 }
1172 adjusted_send = send;
1173
eb83ed87 1174 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1175 * accumulating each into the working value as we go. */
eb83ed87 1176 for (s = s0 + 1; s < send; s++) {
b5b9af04 1177 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1178 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1179 continue;
1180 }
1181
1182 /* Here, found a non-continuation before processing all expected bytes.
1183 * This byte indicates the beginning of a new character, so quit, even
1184 * if allowing this malformation. */
2b5e7bc2 1185 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1186 break;
eb83ed87
KW
1187 } /* End of loop through the character's bytes */
1188
1189 /* Save how many bytes were actually in the character */
1190 curlen = s - s0;
1191
e308b348
KW
1192 /* A convenience macro that matches either of the too-short conditions. */
1193# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1194
e308b348 1195 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1196 uv_so_far = uv;
1197 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1198 }
1199
2b5e7bc2
KW
1200 /* Note that there are two types of too-short malformation. One is when
1201 * there is actual wrong data before the normal termination of the
1202 * sequence. The other is that the sequence wasn't complete before the end
1203 * of the data we are allowed to look at, based on the input 'curlen'.
1204 * This means that we were passed data for a partial character, but it is
1205 * valid as far as we saw. The other is definitely invalid. This
1206 * distinction could be important to a caller, so the two types are kept
1207 * separate. */
1208
1209 /* Check for overflow */
1210 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1211 possible_problems |= UTF8_GOT_OVERFLOW;
1212 uv = UNICODE_REPLACEMENT;
eb83ed87 1213 }
eb83ed87 1214
2b5e7bc2
KW
1215 /* Check for overlong. If no problems so far, 'uv' is the correct code
1216 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1217 * we must look at the UTF-8 byte sequence itself to see if it is for an
1218 * overlong */
1219 if ( ( LIKELY(! possible_problems)
1220 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1221 || ( UNLIKELY( possible_problems)
1222 && ( UNLIKELY(! UTF8_IS_START(*s0))
1223 || ( curlen > 1
1224 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1225 send - s0))))))
2f8f112e 1226 {
2b5e7bc2
KW
1227 possible_problems |= UTF8_GOT_LONG;
1228
2b5e7bc2
KW
1229 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1230 UV min_uv = uv_so_far;
1231 STRLEN i;
1232
1233 /* Here, the input is both overlong and is missing some trailing
1234 * bytes. There is no single code point it could be for, but there
1235 * may be enough information present to determine if what we have
1236 * so far is for an unallowed code point, such as for a surrogate.
1237 * The code below has the intelligence to determine this, but just
1238 * for non-overlong UTF-8 sequences. What we do here is calculate
1239 * the smallest code point the input could represent if there were
1240 * no too short malformation. Then we compute and save the UTF-8
1241 * for that, which is what the code below looks at instead of the
1242 * raw input. It turns out that the smallest such code point is
1243 * all we need. */
1244 for (i = curlen; i < expectlen; i++) {
1245 min_uv = UTF8_ACCUMULATE(min_uv,
1246 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1247 }
1248
1249 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1250 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1251 to free it ourselves if
1252 warnings are made fatal */
1253 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1254 }
eb83ed87
KW
1255 }
1256
2b5e7bc2
KW
1257 /* Now check that the input isn't for a problematic code point not allowed
1258 * by the input parameters. */
1259 /* isn't problematic if < this */
1260 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1261 || ( UNLIKELY(possible_problems)
1262 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1263 && ((flags & ( UTF8_DISALLOW_NONCHAR
1264 |UTF8_DISALLOW_SURROGATE
1265 |UTF8_DISALLOW_SUPER
1266 |UTF8_DISALLOW_ABOVE_31_BIT
1267 |UTF8_WARN_NONCHAR
1268 |UTF8_WARN_SURROGATE
1269 |UTF8_WARN_SUPER
1270 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1271 /* In case of a malformation, 'uv' is not valid, and has
1272 * been changed to something in the Unicode range.
1273 * Currently we don't output a deprecation message if there
1274 * is already a malformation, so we don't have to special
1275 * case the test immediately below */
760c7c2f
KW
1276 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1277 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1278 {
2b5e7bc2
KW
1279 /* If there were no malformations, or the only malformation is an
1280 * overlong, 'uv' is valid */
1281 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1282 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1283 possible_problems |= UTF8_GOT_SURROGATE;
1284 }
1285 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1286 possible_problems |= UTF8_GOT_SUPER;
1287 }
1288 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1289 possible_problems |= UTF8_GOT_NONCHAR;
1290 }
1291 }
1292 else { /* Otherwise, need to look at the source UTF-8, possibly
1293 adjusted to be non-overlong */
1294
1295 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1296 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1297 {
2b5e7bc2
KW
1298 possible_problems |= UTF8_GOT_SUPER;
1299 }
1300 else if (curlen > 1) {
1301 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1302 NATIVE_UTF8_TO_I8(*adjusted_s0),
1303 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1304 {
2b5e7bc2 1305 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1306 }
2b5e7bc2
KW
1307 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1308 NATIVE_UTF8_TO_I8(*adjusted_s0),
1309 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1310 {
1311 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1312 }
1313 }
c0236afe 1314
2b5e7bc2
KW
1315 /* We need a complete well-formed UTF-8 character to discern
1316 * non-characters, so can't look for them here */
1317 }
1318 }
949cf498 1319
2b5e7bc2
KW
1320 ready_to_handle_errors:
1321
1322 /* At this point:
1323 * curlen contains the number of bytes in the sequence that
1324 * this call should advance the input by.
e308b348
KW
1325 * avail_len gives the available number of bytes passed in, but
1326 * only if this is less than the expected number of
1327 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1328 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1329 * is set in it for each potential problem found.
1330 * uv contains the code point the input sequence
1331 * represents; or if there is a problem that prevents
1332 * a well-defined value from being computed, it is
1333 * some subsitute value, typically the REPLACEMENT
1334 * CHARACTER.
1335 * s0 points to the first byte of the character
1336 * send points to just after where that (potentially
1337 * partial) character ends
1338 * adjusted_s0 normally is the same as s0, but in case of an
1339 * overlong for which the UTF-8 matters below, it is
1340 * the first byte of the shortest form representation
1341 * of the input.
1342 * adjusted_send normally is the same as 'send', but if adjusted_s0
1343 * is set to something other than s0, this points one
1344 * beyond its end
1345 */
eb83ed87 1346
2b5e7bc2
KW
1347 if (UNLIKELY(possible_problems)) {
1348 bool disallowed = FALSE;
1349 const U32 orig_problems = possible_problems;
1350
1351 while (possible_problems) { /* Handle each possible problem */
1352 UV pack_warn = 0;
1353 char * message = NULL;
1354
1355 /* Each 'if' clause handles one problem. They are ordered so that
1356 * the first ones' messages will be displayed before the later
1357 * ones; this is kinda in decreasing severity order */
1358 if (possible_problems & UTF8_GOT_OVERFLOW) {
1359
1360 /* Overflow means also got a super and above 31 bits, but we
1361 * handle all three cases here */
1362 possible_problems
1363 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1364 *errors |= UTF8_GOT_OVERFLOW;
1365
1366 /* But the API says we flag all errors found */
1367 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1368 *errors |= UTF8_GOT_SUPER;
1369 }
1370 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1371 *errors |= UTF8_GOT_ABOVE_31_BIT;
1372 }
2b5e7bc2
KW
1373
1374 disallowed = TRUE;
1375
1376 /* The warnings code explicitly says it doesn't handle the case
1377 * of packWARN2 and two categories which have parent-child
1378 * relationship. Even if it works now to raise the warning if
1379 * either is enabled, it wouldn't necessarily do so in the
1380 * future. We output (only) the most dire warning*/
1381 if (! (flags & UTF8_CHECK_ONLY)) {
1382 if (ckWARN_d(WARN_UTF8)) {
1383 pack_warn = packWARN(WARN_UTF8);
1384 }
1385 else if (ckWARN_d(WARN_NON_UNICODE)) {
1386 pack_warn = packWARN(WARN_NON_UNICODE);
1387 }
1388 if (pack_warn) {
1389 message = Perl_form(aTHX_ "%s: %s (overflows)",
1390 malformed_text,
1391 _byte_dump_string(s0, send - s0));
1392 }
1393 }
1394 }
1395 else if (possible_problems & UTF8_GOT_EMPTY) {
1396 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1397 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1398
1399 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1400
1401 /* This so-called malformation is now treated as a bug in
1402 * the caller. If you have nothing to decode, skip calling
1403 * this function */
1404 assert(0);
1405
2b5e7bc2
KW
1406 disallowed = TRUE;
1407 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1408 pack_warn = packWARN(WARN_UTF8);
1409 message = Perl_form(aTHX_ "%s (empty string)",
1410 malformed_text);
1411 }
1412 }
1413 }
1414 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1415 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1416 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1417
1418 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1419 disallowed = TRUE;
1420 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1421 pack_warn = packWARN(WARN_UTF8);
1422 message = Perl_form(aTHX_
1423 "%s: %s (unexpected continuation byte 0x%02x,"
1424 " with no preceding start byte)",
1425 malformed_text,
1426 _byte_dump_string(s0, 1), *s0);
1427 }
1428 }
1429 }
2b5e7bc2
KW
1430 else if (possible_problems & UTF8_GOT_SHORT) {
1431 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1432 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1433
1434 if (! (flags & UTF8_ALLOW_SHORT)) {
1435 disallowed = TRUE;
1436 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1437 pack_warn = packWARN(WARN_UTF8);
1438 message = Perl_form(aTHX_
9a6c9c81 1439 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1440 malformed_text,
1441 _byte_dump_string(s0, send - s0),
e308b348
KW
1442 (int)avail_len,
1443 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1444 (int)expectlen);
1445 }
1446 }
ba210ebe 1447
2b5e7bc2 1448 }
e308b348
KW
1449 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1450 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1451 *errors |= UTF8_GOT_NON_CONTINUATION;
1452
1453 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1454 disallowed = TRUE;
1455 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1456 pack_warn = packWARN(WARN_UTF8);
1457 message = Perl_form(aTHX_ "%s",
1458 unexpected_non_continuation_text(s0,
1459 send - s0,
1460 s - s0,
1461 (int) expectlen));
1462 }
1463 }
1464 }
2b5e7bc2
KW
1465 else if (possible_problems & UTF8_GOT_LONG) {
1466 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1467 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1468
94953955
KW
1469 if (flags & UTF8_ALLOW_LONG) {
1470
1471 /* We don't allow the actual overlong value, unless the
1472 * special extra bit is also set */
1473 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1474 & ~UTF8_ALLOW_LONG)))
1475 {
1476 uv = UNICODE_REPLACEMENT;
1477 }
1478 }
1479 else {
2b5e7bc2
KW
1480 disallowed = TRUE;
1481
1482 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1483 pack_warn = packWARN(WARN_UTF8);
1484
1485 /* These error types cause 'uv' to be something that
1486 * isn't what was intended, so can't use it in the
1487 * message. The other error types either can't
1488 * generate an overlong, or else the 'uv' is valid */
1489 if (orig_problems &
1490 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1491 {
1492 message = Perl_form(aTHX_
1493 "%s: %s (any UTF-8 sequence that starts"
1494 " with \"%s\" is overlong which can and"
1495 " should be represented with a"
1496 " different, shorter sequence)",
1497 malformed_text,
1498 _byte_dump_string(s0, send - s0),
1499 _byte_dump_string(s0, curlen));
1500 }
1501 else {
1502 U8 tmpbuf[UTF8_MAXBYTES+1];
1503 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1504 uv, 0);
1505 message = Perl_form(aTHX_
1506 "%s: %s (overlong; instead use %s to represent"
147e3846 1507 " U+%0*" UVXf ")",
2b5e7bc2
KW
1508 malformed_text,
1509 _byte_dump_string(s0, send - s0),
1510 _byte_dump_string(tmpbuf, e - tmpbuf),
1511 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1512 small code points */
1513 uv);
1514 }
1515 }
1516 }
1517 }
1518 else if (possible_problems & UTF8_GOT_SURROGATE) {
1519 possible_problems &= ~UTF8_GOT_SURROGATE;
1520
f9380377
KW
1521 if (flags & UTF8_WARN_SURROGATE) {
1522 *errors |= UTF8_GOT_SURROGATE;
1523
1524 if ( ! (flags & UTF8_CHECK_ONLY)
1525 && ckWARN_d(WARN_SURROGATE))
1526 {
2b5e7bc2
KW
1527 pack_warn = packWARN(WARN_SURROGATE);
1528
1529 /* These are the only errors that can occur with a
1530 * surrogate when the 'uv' isn't valid */
1531 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1532 message = Perl_form(aTHX_
1533 "UTF-16 surrogate (any UTF-8 sequence that"
1534 " starts with \"%s\" is for a surrogate)",
1535 _byte_dump_string(s0, curlen));
1536 }
1537 else {
1538 message = Perl_form(aTHX_
147e3846 1539 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1540 }
f9380377 1541 }
2b5e7bc2 1542 }
ba210ebe 1543
2b5e7bc2
KW
1544 if (flags & UTF8_DISALLOW_SURROGATE) {
1545 disallowed = TRUE;
f9380377 1546 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1547 }
1548 }
1549 else if (possible_problems & UTF8_GOT_SUPER) {
1550 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1551
f9380377
KW
1552 if (flags & UTF8_WARN_SUPER) {
1553 *errors |= UTF8_GOT_SUPER;
1554
1555 if ( ! (flags & UTF8_CHECK_ONLY)
1556 && ckWARN_d(WARN_NON_UNICODE))
1557 {
2b5e7bc2
KW
1558 pack_warn = packWARN(WARN_NON_UNICODE);
1559
1560 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1561 message = Perl_form(aTHX_
1562 "Any UTF-8 sequence that starts with"
1563 " \"%s\" is for a non-Unicode code point,"
1564 " may not be portable",
1565 _byte_dump_string(s0, curlen));
1566 }
1567 else {
1568 message = Perl_form(aTHX_
147e3846 1569 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1570 " Unicode, may not be portable",
1571 uv);
1572 }
f9380377 1573 }
2b5e7bc2 1574 }
ba210ebe 1575
2b5e7bc2
KW
1576 /* The maximum code point ever specified by a standard was
1577 * 2**31 - 1. Anything larger than that is a Perl extension
1578 * that very well may not be understood by other applications
1579 * (including earlier perl versions on EBCDIC platforms). We
1580 * test for these after the regular SUPER ones, and before
1581 * possibly bailing out, so that the slightly more dire warning
1582 * will override the regular one. */
1583 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1584 |UTF8_WARN_SUPER
1585 |UTF8_DISALLOW_ABOVE_31_BIT))
1586 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1587 && UNLIKELY(is_utf8_cp_above_31_bits(
1588 adjusted_s0,
1589 adjusted_send)))
1590 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1591 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1592 {
1593 if ( ! (flags & UTF8_CHECK_ONLY)
1594 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1595 && ckWARN_d(WARN_UTF8))
1596 {
1597 pack_warn = packWARN(WARN_UTF8);
1598
1599 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1600 message = Perl_form(aTHX_
1601 "Any UTF-8 sequence that starts with"
1602 " \"%s\" is for a non-Unicode code"
1603 " point, and is not portable",
1604 _byte_dump_string(s0, curlen));
1605 }
1606 else {
1607 message = Perl_form(aTHX_
147e3846 1608 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1609 " and not portable",
1610 uv);
1611 }
1612 }
1613
f9380377
KW
1614 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1615 *errors |= UTF8_GOT_ABOVE_31_BIT;
1616
1617 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1618 disallowed = TRUE;
1619 }
2b5e7bc2
KW
1620 }
1621 }
eb83ed87 1622
2b5e7bc2 1623 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1624 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1625 disallowed = TRUE;
1626 }
eb83ed87 1627
2b5e7bc2
KW
1628 /* The deprecated warning overrides any non-deprecated one. If
1629 * there are other problems, a deprecation message is not
1630 * really helpful, so don't bother to raise it in that case.
1631 * This also keeps the code from having to handle the case
1632 * where 'uv' is not valid. */
1633 if ( ! (orig_problems
1634 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1635 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1636 && ckWARN_d(WARN_DEPRECATED))
1637 {
1638 message = Perl_form(aTHX_ cp_above_legal_max,
1639 uv, MAX_NON_DEPRECATED_CP);
1640 pack_warn = packWARN(WARN_DEPRECATED);
1641 }
1642 }
1643 else if (possible_problems & UTF8_GOT_NONCHAR) {
1644 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1645
f9380377
KW
1646 if (flags & UTF8_WARN_NONCHAR) {
1647 *errors |= UTF8_GOT_NONCHAR;
1648
1649 if ( ! (flags & UTF8_CHECK_ONLY)
1650 && ckWARN_d(WARN_NONCHAR))
1651 {
2b5e7bc2
KW
1652 /* The code above should have guaranteed that we don't
1653 * get here with errors other than overlong */
1654 assert (! (orig_problems
1655 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1656
1657 pack_warn = packWARN(WARN_NONCHAR);
1658 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1659 " U+%04" UVXf " is not recommended"
2b5e7bc2 1660 " for open interchange", uv);
f9380377 1661 }
2b5e7bc2 1662 }
5b311467 1663
2b5e7bc2
KW
1664 if (flags & UTF8_DISALLOW_NONCHAR) {
1665 disallowed = TRUE;
f9380377 1666 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1667 }
1668 } /* End of looking through the possible flags */
1669
1670 /* Display the message (if any) for the problem being handled in
1671 * this iteration of the loop */
1672 if (message) {
1673 if (PL_op)
1674 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1675 OP_DESC(PL_op));
1676 else
1677 Perl_warner(aTHX_ pack_warn, "%s", message);
1678 }
1679 } /* End of 'while (possible_problems) {' */
a0dbb045 1680
2b5e7bc2
KW
1681 /* Since there was a possible problem, the returned length may need to
1682 * be changed from the one stored at the beginning of this function.
1683 * Instead of trying to figure out if that's needed, just do it. */
1684 if (retlen) {
1685 *retlen = curlen;
1686 }
a0dbb045 1687
2b5e7bc2
KW
1688 if (disallowed) {
1689 if (flags & UTF8_CHECK_ONLY && retlen) {
1690 *retlen = ((STRLEN) -1);
1691 }
1692 return 0;
1693 }
eb83ed87 1694 }
ba210ebe 1695
2b5e7bc2 1696 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1697}
1698
8e84507e 1699/*
ec5f19d0
KW
1700=for apidoc utf8_to_uvchr_buf
1701
1702Returns the native code point of the first character in the string C<s> which
1703is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1704C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1705
524080c4
KW
1706If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1707enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1708C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1709(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1710C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1711the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1712See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1713returned.
ec5f19d0 1714
760c7c2f
KW
1715Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1716unless those are turned off.
1717
ec5f19d0 1718=cut
52be2536
KW
1719
1720Also implemented as a macro in utf8.h
1721
ec5f19d0
KW
1722*/
1723
1724
1725UV
1726Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1727{
ec5f19d0
KW
1728 assert(s < send);
1729
1730 return utf8n_to_uvchr(s, send - s, retlen,
1731 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1732}
1733
52be2536
KW
1734/* This is marked as deprecated
1735 *
ec5f19d0
KW
1736=for apidoc utf8_to_uvuni_buf
1737
de69f3af
KW
1738Only in very rare circumstances should code need to be dealing in Unicode
1739(as opposed to native) code points. In those few cases, use
1740C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1741
1742Returns the Unicode (not-native) code point of the first character in the
1743string C<s> which
ec5f19d0
KW
1744is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1745C<retlen> will be set to the length, in bytes, of that character.
1746
524080c4
KW
1747If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1748enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1749NULL) to -1. If those warnings are off, the computed value if well-defined (or
1750the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1751is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1752next possible position in C<s> that could begin a non-malformed character.
de69f3af 1753See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1754
760c7c2f
KW
1755Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1756unless those are turned off.
1757
ec5f19d0
KW
1758=cut
1759*/
1760
1761UV
1762Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1763{
1764 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1765
1766 assert(send > s);
1767
5962d97e
KW
1768 /* Call the low level routine, asking for checks */
1769 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1770}
1771
b76347f2 1772/*
87cea99e 1773=for apidoc utf8_length
b76347f2
JH
1774
1775Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1776Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1777up past C<e>, croaks.
b76347f2
JH
1778
1779=cut
1780*/
1781
1782STRLEN
35a4481c 1783Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1784{
1785 STRLEN len = 0;
1786
7918f24d
NC
1787 PERL_ARGS_ASSERT_UTF8_LENGTH;
1788
8850bf83
JH
1789 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1790 * the bitops (especially ~) can create illegal UTF-8.
1791 * In other words: in Perl UTF-8 is not just for Unicode. */
1792
a3b680e6
AL
1793 if (e < s)
1794 goto warn_and_return;
b76347f2 1795 while (s < e) {
4cbf4130 1796 s += UTF8SKIP(s);
8e91ec7f
AV
1797 len++;
1798 }
1799
1800 if (e != s) {
1801 len--;
1802 warn_and_return:
9b387841
NC
1803 if (PL_op)
1804 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1805 "%s in %s", unees, OP_DESC(PL_op));
1806 else
61a12c31 1807 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1808 }
1809
1810 return len;
1811}
1812
b06226ff 1813/*
fed3ba5d
NC
1814=for apidoc bytes_cmp_utf8
1815
a1433954 1816Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1817sequence of characters (stored as UTF-8)
1818in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1819equal, -1 or -2 if the first string is less than the second string, +1 or +2
1820if the first string is greater than the second string.
1821
1822-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1823longer string. -2 or +2 is returned if
1824there was a difference between characters
fed3ba5d
NC
1825within the strings.
1826
1827=cut
1828*/
1829
1830int
1831Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1832{
1833 const U8 *const bend = b + blen;
1834 const U8 *const uend = u + ulen;
1835
1836 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1837
1838 while (b < bend && u < uend) {
1839 U8 c = *u++;
1840 if (!UTF8_IS_INVARIANT(c)) {
1841 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1842 if (u < uend) {
1843 U8 c1 = *u++;
1844 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1845 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1846 } else {
2b5e7bc2 1847 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1848 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1849 "%s %s%s",
7cf8d05d 1850 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1851 PL_op ? " in " : "",
1852 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1853 return -2;
1854 }
1855 } else {
1856 if (PL_op)
1857 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1858 "%s in %s", unees, OP_DESC(PL_op));
1859 else
61a12c31 1860 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1861 return -2; /* Really want to return undef :-) */
1862 }
1863 } else {
1864 return -2;
1865 }
1866 }
1867 if (*b != c) {
1868 return *b < c ? -2 : +2;
1869 }
1870 ++b;
1871 }
1872
1873 if (b == bend && u == uend)
1874 return 0;
1875
1876 return b < bend ? +1 : -1;
1877}
1878
1879/*
87cea99e 1880=for apidoc utf8_to_bytes
6940069f 1881
2bbc8d55 1882Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1883Unlike L</bytes_to_utf8>, this over-writes the original string, and
1884updates C<len> to contain the new length.
67e989fb 1885Returns zero on failure, setting C<len> to -1.
6940069f 1886
a1433954 1887If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1888
6940069f
GS
1889=cut
1890*/
1891
1892U8 *
37607a96 1893Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1894{
d4c19fe8
AL
1895 U8 * const save = s;
1896 U8 * const send = s + *len;
6940069f 1897 U8 *d;
246fae53 1898
7918f24d 1899 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1900 PERL_UNUSED_CONTEXT;
7918f24d 1901
1e54db1a 1902 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1903 while (s < send) {
d59937ca
KW
1904 if (! UTF8_IS_INVARIANT(*s)) {
1905 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1906 *len = ((STRLEN) -1);
1907 return 0;
1908 }
1909 s++;
dcad2880 1910 }
d59937ca 1911 s++;
246fae53 1912 }
dcad2880
JH
1913
1914 d = s = save;
6940069f 1915 while (s < send) {
80e0b38f
KW
1916 U8 c = *s++;
1917 if (! UTF8_IS_INVARIANT(c)) {
1918 /* Then it is two-byte encoded */
a62b247b 1919 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1920 s++;
1921 }
1922 *d++ = c;
6940069f
GS
1923 }
1924 *d = '\0';
246fae53 1925 *len = d - save;
6940069f
GS
1926 return save;
1927}
1928
1929/*
87cea99e 1930=for apidoc bytes_from_utf8
f9a63242 1931
2bbc8d55 1932Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1933Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1934the newly-created string, and updates C<len> to contain the new
1935length. Returns the original string if no conversion occurs, C<len>
72d33970 1936is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19370 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1938in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1939
37607a96
PK
1940=cut
1941*/
f9a63242
JH
1942
1943U8 *
e1ec3a88 1944Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1945{
f9a63242 1946 U8 *d;
e1ec3a88
AL
1947 const U8 *start = s;
1948 const U8 *send;
f9a63242
JH
1949 I32 count = 0;
1950
7918f24d 1951 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1952 PERL_UNUSED_CONTEXT;
f9a63242 1953 if (!*is_utf8)
73d840c0 1954 return (U8 *)start;
f9a63242 1955
1e54db1a 1956 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1957 for (send = s + *len; s < send;) {
d59937ca
KW
1958 if (! UTF8_IS_INVARIANT(*s)) {
1959 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1960 return (U8 *)start;
d59937ca
KW
1961 }
1962 count++;
1963 s++;
db42d148 1964 }
d59937ca 1965 s++;
f9a63242
JH
1966 }
1967
35da51f7 1968 *is_utf8 = FALSE;
f9a63242 1969
212542aa 1970 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1971 s = start; start = d;
f9a63242
JH
1972 while (s < send) {
1973 U8 c = *s++;
1a91c45d 1974 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1975 /* Then it is two-byte encoded */
a62b247b 1976 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1977 s++;
c4d5f83a
NIS
1978 }
1979 *d++ = c;
f9a63242
JH
1980 }
1981 *d = '\0';
1982 *len = d - start;
73d840c0 1983 return (U8 *)start;
f9a63242
JH
1984}
1985
1986/*
87cea99e 1987=for apidoc bytes_to_utf8
6940069f 1988
ff97e5cf
KW
1989Converts a string C<s> of length C<len> bytes from the native encoding into
1990UTF-8.
6662521e 1991Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1992reflect the new length in bytes.
6940069f 1993
75200dff 1994A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1995
1996If you want to convert to UTF-8 from encodings other than
1997the native (Latin1 or EBCDIC),
a1433954 1998see L</sv_recode_to_utf8>().
c9ada85f 1999
497711e7 2000=cut
6940069f
GS
2001*/
2002
c682ebef
FC
2003/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2004 likewise need duplication. */
2005
6940069f 2006U8*
35a4481c 2007Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 2008{
35a4481c 2009 const U8 * const send = s + (*len);
6940069f
GS
2010 U8 *d;
2011 U8 *dst;
7918f24d
NC
2012
2013 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2014 PERL_UNUSED_CONTEXT;
6940069f 2015
212542aa 2016 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2017 dst = d;
2018
2019 while (s < send) {
55d09dc8
KW
2020 append_utf8_from_native_byte(*s, &d);
2021 s++;
6940069f
GS
2022 }
2023 *d = '\0';
6662521e 2024 *len = d-dst;
6940069f
GS
2025 return dst;
2026}
2027
a0ed51b3 2028/*
dea0fc0b 2029 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2030 *
2031 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2032 * We optimize for native, for obvious reasons. */
2033
2034U8*
dea0fc0b 2035Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2036{
dea0fc0b
JH
2037 U8* pend;
2038 U8* dstart = d;
2039
7918f24d
NC
2040 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2041
dea0fc0b 2042 if (bytelen & 1)
147e3846 2043 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2044
2045 pend = p + bytelen;
2046
a0ed51b3 2047 while (p < pend) {
dea0fc0b
JH
2048 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2049 p += 2;
2d1545e5 2050 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2051 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2052 continue;
2053 }
56d37426
KW
2054 if (uv <= MAX_UTF8_TWO_BYTE) {
2055 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2056 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2057 continue;
2058 }
46956fad
KW
2059#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2060#define LAST_HIGH_SURROGATE 0xDBFF
2061#define FIRST_LOW_SURROGATE 0xDC00
2062#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2063
2064 /* This assumes that most uses will be in the first Unicode plane, not
2065 * needing surrogates */
2066 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2067 && uv <= UNICODE_SURROGATE_LAST))
2068 {
2069 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2070 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2071 }
2072 else {
01ea242b 2073 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2074 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2075 || UNLIKELY(low > LAST_LOW_SURROGATE))
2076 {
01ea242b 2077 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2078 }
2079 p += 2;
46956fad
KW
2080 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2081 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2082 }
a0ed51b3 2083 }
56d37426
KW
2084#ifdef EBCDIC
2085 d = uvoffuni_to_utf8_flags(d, uv, 0);
2086#else
a0ed51b3 2087 if (uv < 0x10000) {
eb160463
GS
2088 *d++ = (U8)(( uv >> 12) | 0xe0);
2089 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2090 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2091 continue;
2092 }
2093 else {
eb160463
GS
2094 *d++ = (U8)(( uv >> 18) | 0xf0);
2095 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2096 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2097 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2098 continue;
2099 }
56d37426 2100#endif
a0ed51b3 2101 }
dea0fc0b 2102 *newlen = d - dstart;
a0ed51b3
LW
2103 return d;
2104}
2105
2106/* Note: this one is slightly destructive of the source. */
2107
2108U8*
dea0fc0b 2109Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2110{
2111 U8* s = (U8*)p;
d4c19fe8 2112 U8* const send = s + bytelen;
7918f24d
NC
2113
2114 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2115
e0ea5e2d 2116 if (bytelen & 1)
147e3846 2117 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2118 (UV)bytelen);
2119
a0ed51b3 2120 while (s < send) {
d4c19fe8 2121 const U8 tmp = s[0];
a0ed51b3
LW
2122 s[0] = s[1];
2123 s[1] = tmp;
2124 s += 2;
2125 }
dea0fc0b 2126 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2127}
2128
922e8cb4
KW
2129bool
2130Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2131{
2132 U8 tmpbuf[UTF8_MAXBYTES+1];
2133 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2134 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2135}
2136
f9ae8fb6
JD
2137/* Internal function so we can deprecate the external one, and call
2138 this one from other deprecated functions in this file */
2139
f2645549
KW
2140bool
2141Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2142{
f2645549 2143 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2144
2145 if (*p == '_')
2146 return TRUE;
f25ce844 2147 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2148}
2149
5092f92a 2150bool
eba68aa0
KW
2151Perl__is_uni_perl_idcont(pTHX_ UV c)
2152{
2153 U8 tmpbuf[UTF8_MAXBYTES+1];
2154 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2155 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2156}
2157
2158bool
f91dcd13
KW
2159Perl__is_uni_perl_idstart(pTHX_ UV c)
2160{
2161 U8 tmpbuf[UTF8_MAXBYTES+1];
2162 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2163 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2164}
2165
3a4c58c9
KW
2166UV
2167Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2168{
2169 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2170 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2171 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2172 * either "SS" or "Ss". Which one to use is passed into the routine in
2173 * 'S_or_s' to avoid a test */
2174
2175 UV converted = toUPPER_LATIN1_MOD(c);
2176
2177 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2178
2179 assert(S_or_s == 'S' || S_or_s == 's');
2180
6f2d5cbc 2181 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2182 characters in this range */
3a4c58c9
KW
2183 *p = (U8) converted;
2184 *lenp = 1;
2185 return converted;
2186 }
2187
2188 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2189 * which it maps to one of them, so as to only have to have one check for
2190 * it in the main case */
2191 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2192 switch (c) {
2193 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2194 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2195 break;
2196 case MICRO_SIGN:
2197 converted = GREEK_CAPITAL_LETTER_MU;
2198 break;
79e064b9
KW
2199#if UNICODE_MAJOR_VERSION > 2 \
2200 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2201 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2202 case LATIN_SMALL_LETTER_SHARP_S:
2203 *(p)++ = 'S';
2204 *p = S_or_s;
2205 *lenp = 2;
2206 return 'S';
79e064b9 2207#endif
3a4c58c9
KW
2208 default:
2209 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2210 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2211 }
2212 }
2213
2214 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2215 *p = UTF8_TWO_BYTE_LO(converted);
2216 *lenp = 2;
2217
2218 return converted;
2219}
2220
50bda2c3
KW
2221/* Call the function to convert a UTF-8 encoded character to the specified case.
2222 * Note that there may be more than one character in the result.
2223 * INP is a pointer to the first byte of the input character
2224 * OUTP will be set to the first byte of the string of changed characters. It
2225 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2226 * LENP will be set to the length in bytes of the string of changed characters
2227 *
2228 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2229#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2230#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2231#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2232
b9992569
KW
2233/* This additionally has the input parameter 'specials', which if non-zero will
2234 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2235 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2236#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2237
84afefe6
JH
2238UV
2239Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2240{
a1433954
KW
2241 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2242 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2243 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2244 * the changed version may be longer than the original character.
2245 *
2246 * The ordinal of the first character of the changed version is returned
2247 * (but note, as explained above, that there may be more.) */
2248
7918f24d
NC
2249 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2250
3a4c58c9
KW
2251 if (c < 256) {
2252 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2253 }
2254
0ebc6274 2255 uvchr_to_utf8(p, c);
b9992569 2256 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2257}
2258
84afefe6
JH
2259UV
2260Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2261{
7918f24d
NC
2262 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2263
3a4c58c9
KW
2264 if (c < 256) {
2265 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2266 }
2267
0ebc6274 2268 uvchr_to_utf8(p, c);
b9992569 2269 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2270}
2271
afc16117 2272STATIC U8
eaf412bf 2273S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2274{
2275 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2276 * those, converting the result to UTF-8. Since the result is always just
a1433954 2277 * one character, we allow <p> to be NULL */
afc16117
KW
2278
2279 U8 converted = toLOWER_LATIN1(c);
2280
eaf412bf
KW
2281 PERL_UNUSED_ARG(dummy);
2282
afc16117 2283 if (p != NULL) {
6f2d5cbc 2284 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2285 *p = converted;
2286 *lenp = 1;
2287 }
2288 else {
430c9760
KW
2289 /* Result is known to always be < 256, so can use the EIGHT_BIT
2290 * macros */
2291 *p = UTF8_EIGHT_BIT_HI(converted);
2292 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2293 *lenp = 2;
2294 }
2295 }
2296 return converted;
2297}
2298
84afefe6
JH
2299UV
2300Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2301{
7918f24d
NC
2302 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2303
afc16117 2304 if (c < 256) {
eaf412bf 2305 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2306 }
2307
afc16117 2308 uvchr_to_utf8(p, c);
b9992569 2309 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2310}
2311
84afefe6 2312UV
51910141 2313Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2314{
51910141 2315 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2316 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2317 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2318 *
2319 * Not to be used for locale folds
51910141 2320 */
f673fad4 2321
a1dde8de
KW
2322 UV converted;
2323
2324 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2325 PERL_UNUSED_CONTEXT;
a1dde8de 2326
1ca267a5
KW
2327 assert (! (flags & FOLD_FLAGS_LOCALE));
2328
659a7c2d 2329 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2330 converted = GREEK_SMALL_LETTER_MU;
2331 }
9b63e895
KW
2332#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2333 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2334 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2335 else if ( (flags & FOLD_FLAGS_FULL)
2336 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2337 {
1ca267a5
KW
2338 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2339 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2340 * under those circumstances. */
2341 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2342 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2343 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2344 p, *lenp, U8);
2345 return LATIN_SMALL_LETTER_LONG_S;
2346 }
2347 else {
4f489194
KW
2348 *(p)++ = 's';
2349 *p = 's';
2350 *lenp = 2;
2351 return 's';
1ca267a5 2352 }
a1dde8de 2353 }
9b63e895 2354#endif
a1dde8de
KW
2355 else { /* In this range the fold of all other characters is their lower
2356 case */
2357 converted = toLOWER_LATIN1(c);
2358 }
2359
6f2d5cbc 2360 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2361 *p = (U8) converted;
2362 *lenp = 1;
2363 }
2364 else {
2365 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2366 *p = UTF8_TWO_BYTE_LO(converted);
2367 *lenp = 2;
2368 }
2369
2370 return converted;
2371}
2372
2373UV
31f05a37 2374Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2375{
4b593389 2376
a0270393
KW
2377 /* Not currently externally documented, and subject to change
2378 * <flags> bits meanings:
2379 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2380 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2381 * locale are to be used.
a0270393
KW
2382 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2383 */
4b593389 2384
36bb2ab6 2385 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2386
780fcc9f
KW
2387 if (flags & FOLD_FLAGS_LOCALE) {
2388 /* Treat a UTF-8 locale as not being in locale at all */
2389 if (IN_UTF8_CTYPE_LOCALE) {
2390 flags &= ~FOLD_FLAGS_LOCALE;
2391 }
2392 else {
2393 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2394 goto needs_full_generality;
780fcc9f 2395 }
31f05a37
KW
2396 }
2397
a1dde8de 2398 if (c < 256) {
e7b7ac46 2399 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2400 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2401 }
2402
2f306ab9 2403 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2404 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2405 uvchr_to_utf8(p, c);
b9992569 2406 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
2407 }
2408 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2409 the special flags. */
2410 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2411
2412 needs_full_generality:
a0270393 2413 uvchr_to_utf8(utf8_c, c);
445bf929 2414 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 2415 }
84afefe6
JH
2416}
2417
26483009 2418PERL_STATIC_INLINE bool
5141f98e 2419S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2420 const char *const swashname, SV* const invlist)
bde6a22d 2421{
ea317ccb
KW
2422 /* returns a boolean giving whether or not the UTF8-encoded character that
2423 * starts at <p> is in the swash indicated by <swashname>. <swash>
2424 * contains a pointer to where the swash indicated by <swashname>
2425 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2426 * look at <*swash> and only generate a swash if it is not null. <invlist>
2427 * is NULL or an inversion list that defines the swash. If not null, it
2428 * saves time during initialization of the swash.
ea317ccb
KW
2429 *
2430 * Note that it is assumed that the buffer length of <p> is enough to
2431 * contain all the bytes that comprise the character. Thus, <*p> should
2432 * have been checked before this call for mal-formedness enough to assure
2433 * that. */
2434
7918f24d
NC
2435 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2436
492a624f 2437 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2438 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2439 * as far as there being enough bytes available in it to accommodate the
2440 * character without reading beyond the end, and pass that number on to the
2441 * validating routine */
6302f837 2442 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94
KW
2443 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2444 0,
2445 1 /* Die */ );
2446 NOT_REACHED; /* NOTREACHED */
28123549 2447 }
86ae6e94 2448
87367d5f
KW
2449 if (!*swash) {
2450 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2451 *swash = _core_swash_init("utf8",
2452
2453 /* Only use the name if there is no inversion
2454 * list; otherwise will go out to disk */
2455 (invlist) ? "" : swashname,
2456
2457 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2458 }
28123549 2459
bde6a22d
NC
2460 return swash_fetch(*swash, p, TRUE) != 0;
2461}
2462
da8c1a98
KW
2463PERL_STATIC_INLINE bool
2464S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2465 const char *const swashname, SV* const invlist)
2466{
2467 /* returns a boolean giving whether or not the UTF8-encoded character that
2468 * starts at <p>, and extending no further than <e - 1> is in the swash
2469 * indicated by <swashname>. <swash> contains a pointer to where the swash
2470 * indicated by <swashname> is to be stored; which this routine will do, so
2471 * that future calls will look at <*swash> and only generate a swash if it
2472 * is not null. <invlist> is NULL or an inversion list that defines the
2473 * swash. If not null, it saves time during initialization of the swash.
2474 */
2475
2476 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2477
2478 if (! isUTF8_CHAR(p, e)) {
2479 _force_out_malformed_utf8_message(p, e, 0, 1);
2480 NOT_REACHED; /* NOTREACHED */
2481 }
2482
2483 if (!*swash) {
2484 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2485 *swash = _core_swash_init("utf8",
2486
2487 /* Only use the name if there is no inversion
2488 * list; otherwise will go out to disk */
2489 (invlist) ? "" : swashname,
2490
2491 &PL_sv_undef, 1, 0, invlist, &flags);
2492 }
2493
2494 return swash_fetch(*swash, p, TRUE) != 0;
2495}
2496
bde6a22d 2497bool
922e8cb4
KW
2498Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2499{
922e8cb4
KW
2500 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2501
2502 assert(classnum < _FIRST_NON_SWASH_CC);
2503
f25ce844
KW
2504 return is_utf8_common(p,
2505 &PL_utf8_swash_ptrs[classnum],
2506 swash_property_names[classnum],
2507 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
2508}
2509
2510bool
da8c1a98
KW
2511Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2512 const U8 * const e)
2513{
2514 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2515
2516 assert(classnum < _FIRST_NON_SWASH_CC);
2517
2518 return is_utf8_common_with_len(p,
2519 e,
2520 &PL_utf8_swash_ptrs[classnum],
2521 swash_property_names[classnum],
2522 PL_XPosix_ptrs[classnum]);
2523}
2524
2525bool
f2645549 2526Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 2527{
f2645549 2528 SV* invlist = NULL;
7918f24d 2529
f2645549 2530 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 2531
f2645549
KW
2532 if (! PL_utf8_perl_idstart) {
2533 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2534 }
60071a22 2535 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
2536}
2537
2538bool
da8c1a98
KW
2539Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2540{
2541 SV* invlist = NULL;
2542
2543 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2544
2545 if (! PL_utf8_perl_idstart) {
2546 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2547 }
2548 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2549 "_Perl_IDStart", invlist);
2550}
2551
2552bool
f2645549 2553Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2554{
f2645549 2555 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2556
2557 if (*p == '_')
2558 return TRUE;
f25ce844 2559 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2560}
2561
2562bool
eba68aa0
KW
2563Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2564{
b24b43f7 2565 SV* invlist = NULL;
eba68aa0
KW
2566
2567 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2568
b24b43f7
KW
2569 if (! PL_utf8_perl_idcont) {
2570 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2571 }
60071a22 2572 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
2573}
2574
eba68aa0 2575bool
da8c1a98
KW
2576Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2577{
2578 SV* invlist = NULL;
2579
2580 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2581
2582 if (! PL_utf8_perl_idcont) {
2583 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2584 }
2585 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2586 "_Perl_IDCont", invlist);
2587}
2588
2589bool
f2645549 2590Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2591{
f2645549 2592 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2593
f25ce844 2594 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2595}
2596
2597bool
f2645549 2598Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2599{
f2645549 2600 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2601
f25ce844 2602 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2603}
2604
2605bool
7dbf68d2
KW
2606Perl__is_utf8_mark(pTHX_ const U8 *p)
2607{
7dbf68d2
KW
2608 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2609
f25ce844 2610 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2611}
2612
6b5c0936 2613/*
87cea99e 2614=for apidoc to_utf8_case
6b5c0936 2615
9da1e7cb
KW
2616Instead use the appropriate one of L</toUPPER_utf8>,
2617L</toTITLE_utf8>,
2618L</toLOWER_utf8>,
2619or L</toFOLD_utf8>.
2620
6fae5207 2621C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2622the character that is being converted. This routine assumes that the character
2623at C<p> is well-formed.
6b5c0936 2624
6fae5207
KW
2625C<ustrp> is a pointer to the character buffer to put the
2626conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2627of the result.
2628
6fae5207 2629C<swashp> is a pointer to the swash to use.
6b5c0936 2630
a1433954 2631Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2632and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2633but not always, a multicharacter mapping), is tried first.
6b5c0936 2634
4a8240a3
KW
2635C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2636any special mappings; C<""> means to use the special mappings. Values other
2637than these two are treated as the name of the hash containing the special
2638mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2639
796b6530
KW
2640C<normal> is a string like C<"ToLower"> which means the swash
2641C<%utf8::ToLower>.
0134edef 2642
760c7c2f
KW
2643Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2644unless those are turned off.
2645
0134edef 2646=cut */
6b5c0936 2647
2104c8d9 2648UV
9a957fbc
AL
2649Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2650 SV **swashp, const char *normal, const char *special)
a0ed51b3 2651{
b9992569
KW
2652 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2653
2654 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2655}
2656
2657 /* change namve uv1 to 'from' */
6a4a25f4 2658STATIC UV
b9992569
KW
2659S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2660 SV **swashp, const char *normal, const char *special)
2661{
0134edef 2662 STRLEN len = 0;
7918f24d 2663
b9992569 2664 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2665
36eaa811
KW
2666 /* For code points that don't change case, we already know that the output
2667 * of this function is the unchanged input, so we can skip doing look-ups
2668 * for them. Unfortunately the case-changing code points are scattered
2669 * around. But there are some long consecutive ranges where there are no
2670 * case changing code points. By adding tests, we can eliminate the lookup
2671 * for all the ones in such ranges. This is currently done here only for
2672 * just a few cases where the scripts are in common use in modern commerce
2673 * (and scripts adjacent to those which can be included without additional
2674 * tests). */
2675
2676 if (uv1 >= 0x0590) {
2677 /* This keeps from needing further processing the code points most
2678 * likely to be used in the following non-cased scripts: Hebrew,
2679 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2680 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2681 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2682 if (uv1 < 0x10A0) {
2683 goto cases_to_self;
2684 }
2685
2686 /* The following largish code point ranges also don't have case
2687 * changes, but khw didn't think they warranted extra tests to speed
2688 * them up (which would slightly slow down everything else above them):
2689 * 1100..139F Hangul Jamo, Ethiopic
2690 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2691 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2692 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2693 * Combining Diacritical Marks Extended, Balinese,
2694 * Sundanese, Batak, Lepcha, Ol Chiki
2695 * 2000..206F General Punctuation
2696 */
2697
2698 if (uv1 >= 0x2D30) {
2699
2700 /* This keeps the from needing further processing the code points
2701 * most likely to be used in the following non-cased major scripts:
2702 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2703 *
2704 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2705 * event that Unicode eventually allocates the unused block as of
2706 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2707 * that the test suite will start having failures to alert you
2708 * should that happen) */
2709 if (uv1 < 0xA640) {
2710 goto cases_to_self;
2711 }
2712
2713 if (uv1 >= 0xAC00) {
2714 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2715 if (ckWARN_d(WARN_SURROGATE)) {
2716 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2717 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2718 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2719 }
2720 goto cases_to_self;
2721 }
36eaa811
KW
2722
2723 /* AC00..FAFF Catches Hangul syllables and private use, plus
2724 * some others */
2725 if (uv1 < 0xFB00) {
2726 goto cases_to_self;
2727
2728 }
2729
5af9bc97
KW
2730 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2731 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2732 && ckWARN_d(WARN_DEPRECATED))
2733 {
2734 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2735 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2736 }
2737 if (ckWARN_d(WARN_NON_UNICODE)) {
2738 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2739 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2740 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2741 }
2742 goto cases_to_self;
2743 }
3bfc1e70
KW
2744#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2745 if (UNLIKELY(uv1
2746 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2747 {
2748
2749 /* As of this writing, this means we avoid swash creation
2750 * for anything beyond low Plane 1 */
2751 goto cases_to_self;
2752 }
2753#endif
36eaa811
KW
2754 }
2755 }
9ae3ac1a 2756
36eaa811
KW
2757 /* Note that non-characters are perfectly legal, so no warning should
2758 * be given. There are so few of them, that it isn't worth the extra
2759 * tests to avoid swash creation */
9ae3ac1a
KW
2760 }
2761
0134edef 2762 if (!*swashp) /* load on-demand */
5ab9d2ef 2763 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2764
a6f87d8c 2765 if (special) {
0134edef 2766 /* It might be "special" (sometimes, but not always,
2a37f04d 2767 * a multicharacter mapping) */
4a8240a3 2768 HV *hv = NULL;
b08cf34e
JH
2769 SV **svp;
2770
4a8240a3
KW
2771 /* If passed in the specials name, use that; otherwise use any
2772 * given in the swash */
2773 if (*special != '\0') {
2774 hv = get_hv(special, 0);
2775 }
2776 else {
2777 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2778 if (svp) {
2779 hv = MUTABLE_HV(SvRV(*svp));
2780 }
2781 }
2782
176fe009 2783 if (hv
5f560d8a 2784 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2785 && (*svp))
2786 {
cfd0369c 2787 const char *s;
47654450 2788
cfd0369c 2789 s = SvPV_const(*svp, len);
47654450 2790 if (len == 1)
f4cd282c 2791 /* EIGHTBIT */
c80e42f3 2792 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2793 else {
d2dcd0fb 2794 Copy(s, ustrp, len, U8);
29e98929 2795 }
983ffd37 2796 }
0134edef
JH
2797 }
2798
2799 if (!len && *swashp) {
4a4088c4 2800 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2801
0134edef
JH
2802 if (uv2) {
2803 /* It was "normal" (a single character mapping). */
f4cd282c 2804 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2805 }
2806 }
1feea2c7 2807
cbe07460
KW
2808 if (len) {
2809 if (lenp) {
2810 *lenp = len;
2811 }
2812 return valid_utf8_to_uvchr(ustrp, 0);
2813 }
2814
2815 /* Here, there was no mapping defined, which means that the code point maps
2816 * to itself. Return the inputs */
e24dfe9c 2817 cases_to_self:
bfdf22ec 2818 len = UTF8SKIP(p);
ca9fab46
KW
2819 if (p != ustrp) { /* Don't copy onto itself */
2820 Copy(p, ustrp, len, U8);
2821 }
0134edef 2822
2a37f04d
JH
2823 if (lenp)
2824 *lenp = len;
2825
f4cd282c 2826 return uv1;
cbe07460 2827
a0ed51b3
LW
2828}
2829
051a06d4 2830STATIC UV
357aadde 2831S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2832{
4a4088c4 2833 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2834 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2835 * result contains a character that crosses the 255/256 boundary, disallow
2836 * the change, and return the original code point. See L<perlfunc/lc> for
2837 * why;
051a06d4 2838 *
a1433954
KW
2839 * p points to the original string whose case was changed; assumed
2840 * by this routine to be well-formed
051a06d4
KW
2841 * result the code point of the first character in the changed-case string
2842 * ustrp points to the changed-case string (<result> represents its first char)
2843 * lenp points to the length of <ustrp> */
2844
2845 UV original; /* To store the first code point of <p> */
2846
2847 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2848
a4f12ed7 2849 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2850
2851 /* We know immediately if the first character in the string crosses the
2852 * boundary, so can skip */
2853 if (result > 255) {
2854
2855 /* Look at every character in the result; if any cross the
2856 * boundary, the whole thing is disallowed */
2857 U8* s = ustrp + UTF8SKIP(ustrp);
2858 U8* e = ustrp + *lenp;
2859 while (s < e) {
a4f12ed7 2860 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2861 goto bad_crossing;
2862 }
2863 s += UTF8SKIP(s);
2864 }
2865
613abc6d
KW
2866 /* Here, no characters crossed, result is ok as-is, but we warn. */
2867 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2868 return result;
2869 }
2870
7b52d656 2871 bad_crossing:
051a06d4
KW
2872
2873 /* Failed, have to return the original */
4b88fb76 2874 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2875
2876 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2877 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2878 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2879 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2880 OP_DESC(PL_op),
ab0b796c
KW
2881 original,
2882 original);
051a06d4
KW
2883 Copy(p, ustrp, *lenp, char);
2884 return original;
2885}
2886
eaf412bf
KW
2887/* The process for changing the case is essentially the same for the four case
2888 * change types, except there are complications for folding. Otherwise the
2889 * difference is only which case to change to. To make sure that they all do
2890 * the same thing, the bodies of the functions are extracted out into the
2891 * following two macros. The functions are written with the same variable
2892 * names, and these are known and used inside these macros. It would be
2893 * better, of course, to have inline functions to do it, but since different
2894 * macros are called, depending on which case is being changed to, this is not
2895 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2896 * function can start with the common start macro, then finish with its special
2897 * handling; while the other three cases can just use the common end macro.
2898 *
2899 * The algorithm is to use the proper (passed in) macro or function to change
2900 * the case for code points that are below 256. The macro is used if using
2901 * locale rules for the case change; the function if not. If the code point is
2902 * above 255, it is computed from the input UTF-8, and another macro is called
2903 * to do the conversion. If necessary, the output is converted to UTF-8. If
2904 * using a locale, we have to check that the change did not cross the 255/256
2905 * boundary, see check_locale_boundary_crossing() for further details.
2906 *
2907 * The macros are split with the correct case change for the below-256 case
2908 * stored into 'result', and in the middle of an else clause for the above-255
2909 * case. At that point in the 'else', 'result' is not the final result, but is
2910 * the input code point calculated from the UTF-8. The fold code needs to
2911 * realize all this and take it from there.
2912 *
2913 * If you read the two macros as sequential, it's easier to understand what's
2914 * going on. */
2915#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2916 L1_func_extra_param) \
2917 if (flags & (locale_flags)) { \
2918 /* Treat a UTF-8 locale as not being in locale at all */ \
2919 if (IN_UTF8_CTYPE_LOCALE) { \
2920 flags &= ~(locale_flags); \
2921 } \
2922 else { \
2923 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2924 } \
2925 } \
2926 \
2927 if (UTF8_IS_INVARIANT(*p)) { \
2928 if (flags & (locale_flags)) { \
2929 result = LC_L1_change_macro(*p); \
2930 } \
2931 else { \
2932 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2933 } \
2934 } \
2935 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2936 if (flags & (locale_flags)) { \
2937 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2938 *(p+1))); \
2939 } \
2940 else { \
2941 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2942 ustrp, lenp, L1_func_extra_param); \
2943 } \
2944 } \
2945 else { /* malformed UTF-8 */ \
2946 result = valid_utf8_to_uvchr(p, NULL); \
2947
2948#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2949 result = change_macro(result, p, ustrp, lenp); \
2950 \
2951 if (flags & (locale_flags)) { \
2952 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2953 } \
2954 return result; \
2955 } \
2956 \
2957 /* Here, used locale rules. Convert back to UTF-8 */ \
2958 if (UTF8_IS_INVARIANT(result)) { \
2959 *ustrp = (U8) result; \
2960 *lenp = 1; \
2961 } \
2962 else { \
2963 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2964 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2965 *lenp = 2; \
2966 } \
2967 \
2968 return result;
2969
d3e79532 2970/*
87cea99e 2971=for apidoc to_utf8_upper
d3e79532 2972
1f607577 2973Instead use L</toUPPER_utf8>.
a1433954 2974
d3e79532
JH
2975=cut */
2976
051a06d4 2977/* Not currently externally documented, and subject to change:
31f05a37
KW
2978 * <flags> is set iff iff the rules from the current underlying locale are to
2979 * be used. */
051a06d4 2980
2104c8d9 2981UV
31f05a37 2982Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2983{
051a06d4
KW
2984 UV result;
2985
2986 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2987
eaf412bf
KW
2988 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
2989 /* 2nd char of uc(U+DF) is 'S' */
2990 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
2991 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 2992}
a0ed51b3 2993
d3e79532 2994/*
87cea99e 2995=for apidoc to_utf8_title
d3e79532 2996
1f607577 2997Instead use L</toTITLE_utf8>.
a1433954 2998
d3e79532
JH
2999=cut */
3000
051a06d4 3001/* Not currently externally documented, and subject to change:
31f05a37
KW
3002 * <flags> is set iff the rules from the current underlying locale are to be
3003 * used. Since titlecase is not defined in POSIX, for other than a
3004 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3005 */
051a06d4 3006
983ffd37 3007UV
31f05a37 3008Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 3009{
051a06d4
KW
3010 UV result;
3011
3012 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3013
eaf412bf
KW
3014 /* 2nd char of ucfirst(U+DF) is 's' */
3015 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3016 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3017}
3018
d3e79532 3019/*
87cea99e 3020=for apidoc to_utf8_lower
d3e79532 3021
1f607577 3022Instead use L</toLOWER_utf8>.
a1433954 3023
d3e79532
JH
3024=cut */
3025
051a06d4 3026/* Not currently externally documented, and subject to change:
31f05a37
KW
3027 * <flags> is set iff iff the rules from the current underlying locale are to
3028 * be used.
3029 */
051a06d4 3030
2104c8d9 3031UV
31f05a37 3032Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3033{
051a06d4
KW
3034 UV result;
3035
051a06d4 3036 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3037
eaf412bf
KW
3038 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3039 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3040}
3041
d3e79532 3042/*
87cea99e 3043=for apidoc to_utf8_fold
d3e79532 3044
1f607577 3045Instead use L</toFOLD_utf8>.
a1433954 3046
d3e79532
JH
3047=cut */
3048
051a06d4
KW
3049/* Not currently externally documented, and subject to change,
3050 * in <flags>
31f05a37
KW
3051 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3052 * locale are to be used.
051a06d4
KW
3053 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3054 * otherwise simple folds
a0270393
KW
3055 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3056 * prohibited
445bf929 3057 */
36bb2ab6 3058
b4e400f9 3059UV
445bf929 3060Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3061{
051a06d4
KW
3062 UV result;
3063
36bb2ab6 3064 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3065
a0270393
KW
3066 /* These are mutually exclusive */
3067 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3068
50ba90ff
KW
3069 assert(p != ustrp); /* Otherwise overwrites */
3070
eaf412bf
KW
3071 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3072 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3073
eaf412bf 3074 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3075
1ca267a5
KW
3076 if (flags & FOLD_FLAGS_LOCALE) {
3077
76f2ffcd 3078# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3079 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3080
3081# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3082# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3083
3084 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3085
538e84ed
KW
3086 /* Special case these two characters, as what normally gets
3087 * returned under locale doesn't work */
76f2ffcd
KW
3088 if (UTF8SKIP(p) == cap_sharp_s_len
3089 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3090 {
ab0b796c
KW
3091 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3092 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3093 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3094 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3095 goto return_long_s;
3096 }
0766489e
KW
3097 else
3098#endif
3099 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3100 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3101 {
ab0b796c
KW
3102 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3103 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3104 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3105 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3106 goto return_ligature_st;
3107 }
74894415
KW
3108
3109#if UNICODE_MAJOR_VERSION == 3 \
3110 && UNICODE_DOT_VERSION == 0 \
3111 && UNICODE_DOT_DOT_VERSION == 1
3112# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3113
3114 /* And special case this on this Unicode version only, for the same
3115 * reaons the other two are special cased. They would cross the
3116 * 255/256 boundary which is forbidden under /l, and so the code
3117 * wouldn't catch that they are equivalent (which they are only in
3118 * this release) */
3119 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3120 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3121 {
3122 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3123 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3124 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3125 "resolved to \"\\x{0131}\".");
3126 goto return_dotless_i;
3127 }
3128#endif
3129
357aadde 3130 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3131 }
a0270393
KW
3132 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3133 return result;
3134 }
3135 else {
4a4088c4 3136 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3137 * character above the ASCII range, and the result should not
3138 * contain an ASCII character. */
a0270393
KW
3139
3140 UV original; /* To store the first code point of <p> */
3141
3142 /* Look at every character in the result; if any cross the
3143 * boundary, the whole thing is disallowed */
3144 U8* s = ustrp;
3145 U8* e = ustrp + *lenp;
3146 while (s < e) {
3147 if (isASCII(*s)) {
3148 /* Crossed, have to return the original */
3149 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3150
9fc2026f 3151 /* But in these instances, there is an alternative we can
1ca267a5 3152 * return that is valid */
0766489e
KW
3153 if (original == LATIN_SMALL_LETTER_SHARP_S
3154#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3155 || original == LATIN_CAPITAL_LETTER_SHARP_S
3156#endif
3157 ) {
1ca267a5
KW
3158 goto return_long_s;
3159 }
9fc2026f
KW
3160 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3161 goto return_ligature_st;
3162 }
74894415
KW
3163#if UNICODE_MAJOR_VERSION == 3 \
3164 && UNICODE_DOT_VERSION == 0 \
3165 && UNICODE_DOT_DOT_VERSION == 1
3166
3167 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3168 goto return_dotless_i;
3169 }
3170#endif
a0270393
KW
3171 Copy(p, ustrp, *lenp, char);
3172 return original;
3173 }
3174 s += UTF8SKIP(s);
3175 }
051a06d4 3176
a0270393
KW
3177 /* Here, no characters crossed, result is ok as-is */
3178 return result;
3179 }
051a06d4
KW
3180 }
3181
4a4088c4 3182 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3183 if (UTF8_IS_INVARIANT(result)) {
3184 *ustrp = (U8) result;
3185 *lenp = 1;
3186 }
3187 else {
62cb07ea
KW
3188 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3189 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3190 *lenp = 2;
3191 }
3192
051a06d4 3193 return result;
1ca267a5
KW
3194
3195 return_long_s:
3196 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3197 * folds to a string of two of these characters. By returning this
3198 * instead, then, e.g.,
3199 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3200 * works. */
3201
3202 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3203 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3204 ustrp, *lenp, U8);
3205 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3206
3207 return_ligature_st:
3208 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3209 * have the other one fold to it */
3210
3211 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3212 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3213 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3214
3215#if UNICODE_MAJOR_VERSION == 3 \
3216 && UNICODE_DOT_VERSION == 0 \
3217 && UNICODE_DOT_DOT_VERSION == 1
3218
3219 return_dotless_i:
3220 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3221 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3222 return LATIN_SMALL_LETTER_DOTLESS_I;
3223
3224#endif
3225
a0ed51b3
LW
3226}
3227
711a919c 3228/* Note:
f90a9a02 3229 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3230 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3231 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3232 */
c4a5db0c 3233
a0ed51b3 3234SV*
7fc63493 3235Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3236{
c4a5db0c
KW
3237 PERL_ARGS_ASSERT_SWASH_INIT;
3238
3239 /* Returns a copy of a swash initiated by the called function. This is the
3240 * public interface, and returning a copy prevents others from doing
3241 * mischief on the original */
3242
5d3d13d1 3243 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3244}
3245
3246SV*
5d3d13d1 3247Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3248{
2c1f00b9
YO
3249
3250 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3251 * use the following define */
3252
3253#define CORE_SWASH_INIT_RETURN(x) \
3254 PL_curpm= old_PL_curpm; \
3255 return x
3256
c4a5db0c 3257 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3258 * by calling utf8_heavy.pl in the general case. The returned value may be
3259 * the swash's inversion list instead if the input parameters allow it.
3260 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3261 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3262 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3263 *
3264 * This interface should only be used by functions that won't destroy or
3265 * adversely change the swash, as doing so affects all other uses of the
3266 * swash in the program; the general public should use 'Perl_swash_init'
3267 * instead.
3268 *
3269 * pkg is the name of the package that <name> should be in.
3270 * name is the name of the swash to find. Typically it is a Unicode
3271 * property name, including user-defined ones
3272 * listsv is a string to initialize the swash with. It must be of the form
3273 * documented as the subroutine return value in
3274 * L<perlunicode/User-Defined Character Properties>
3275 * minbits is the number of bits required to represent each data element.
3276 * It is '1' for binary properties.
3277 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3278 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3279 * flags_p if non-NULL is the address of various input and output flag bits
3280 * to the routine, as follows: ('I' means is input to the routine;
3281 * 'O' means output from the routine. Only flags marked O are
3282 * meaningful on return.)
3283 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3284 * came from a user-defined property. (I O)
5d3d13d1
KW
3285 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3286 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3287 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3288 * return of an inversion list instead of a swash hash if this routine
3289 * thinks that would result in faster execution of swash_fetch() later
3290 * on. (I)
9a53f6cf
KW
3291 *
3292 * Thus there are three possible inputs to find the swash: <name>,
3293 * <listsv>, and <invlist>. At least one must be specified. The result
3294 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3295 * actions can intersect, etc. what <name> gives. To avoid going out to
3296 * disk at all, <invlist> should specify completely what the swash should
3297 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3298 *
3299 * <invlist> is only valid for binary properties */
c4a5db0c 3300
2c1f00b9
YO
3301 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3302
c4a5db0c 3303 SV* retval = &PL_sv_undef;
83199d38 3304 HV* swash_hv = NULL;
87367d5f
KW
3305 const int invlist_swash_boundary =
3306 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3307 ? 512 /* Based on some benchmarking, but not extensive, see commit
3308 message */
3309 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3310
3311 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3312 assert(! invlist || minbits == 1);
3313
2c1f00b9
YO
3314 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3315 that triggered the swash init and the swash init perl logic itself.
3316 See perl #122747 */
3317
9a53f6cf
KW
3318 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3319 * so */
3320 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3321 dSP;
3322 const size_t pkg_len = strlen(pkg);
3323 const size_t name_len = strlen(name);
3324 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3325 SV* errsv_save;
3326 GV *method;
3327
3328 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3329
3330 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3331 ENTER;
69794297 3332 SAVEHINTS();
2782061f 3333 save_re_context();
650f067c
JL
3334 /* We might get here via a subroutine signature which uses a utf8
3335 * parameter name, at which point PL_subname will have been set
3336 * but not yet used. */
3337 save_item(PL_subname);
69794297
KW
3338 if (PL_parser && PL_parser->error_count)
3339 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3340 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3341 if (!method) { /* demand load UTF-8 */
69794297 3342 ENTER;
db2c6cb3
FC
3343 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3344 GvSV(PL_errgv) = NULL;
1a419e6b 3345#ifndef NO_TAINT_SUPPORT
69794297
KW
3346 /* It is assumed that callers of this routine are not passing in
3347 * any user derived data. */
2782061f
DM
3348 /* Need to do this after save_re_context() as it will set
3349 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3350 * in Perl_magic_get). Even line to create errsv_save can turn on
3351 * PL_tainted. */
284167a5
SM
3352 SAVEBOOL(TAINT_get);
3353 TAINT_NOT;
3354#endif
69794297
KW
3355 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3356 NULL);
eed484f9 3357 {
db2c6cb3
FC
3358 /* Not ERRSV, as there is no need to vivify a scalar we are
3359 about to discard. */
3360 SV * const errsv = GvSV(PL_errgv);
3361 if (!SvTRUE(errsv)) {
3362 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3363 SvREFCNT_dec(errsv);
3364 }
eed484f9 3365 }
69794297
KW
3366 LEAVE;
3367 }
3368 SPAGAIN;
3369 PUSHMARK(SP);
3370 EXTEND(SP,5);
3371 mPUSHp(pkg, pkg_len);
3372 mPUSHp(name, name_len);
3373 PUSHs(listsv);
3374 mPUSHi(minbits);
3375 mPUSHi(none);
3376 PUTBACK;
db2c6cb3
FC
3377 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3378 GvSV(PL_errgv) = NULL;
69794297
KW
3379 /* If we already have a pointer to the method, no need to use
3380 * call_method() to repeat the lookup. */
c41800a8
KW
3381 if (method
3382 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3383 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3384 {
3385 retval = *PL_stack_sp--;
3386 SvREFCNT_inc(retval);
3387 }
eed484f9 3388 {
db2c6cb3
FC
3389 /* Not ERRSV. See above. */
3390 SV * const errsv = GvSV(PL_errgv);
3391 if (!SvTRUE(errsv)) {
3392 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3393 SvREFCNT_dec(errsv);
3394 }
eed484f9 3395 }
ce3b816e 3396 LEAVE;
69794297
KW
3397 POPSTACK;
3398 if (IN_PERL_COMPILETIME) {
3399 CopHINTS_set(PL_curcop, PL_hints);
3400 }
3401 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3402 if (SvPOK(retval)) {
69794297
KW
3403
3404 /* If caller wants to handle missing properties, let them */
5d3d13d1 3405 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3406 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3407 }
3408 Perl_croak(aTHX_
147e3846 3409 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3410 SVfARG(retval));
a25b5927 3411 NOT_REACHED; /* NOTREACHED */
d95e4a00 3412 }
69794297 3413 }
9a53f6cf 3414 } /* End of calling the module to find the swash */
36eb48b4 3415
83199d38
KW
3416 /* If this operation fetched a swash, and we will need it later, get it */
3417 if (retval != &PL_sv_undef
3418 && (minbits == 1 || (flags_p
3419 && ! (*flags_p
3420 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3421 {
3422 swash_hv = MUTABLE_HV(SvRV(retval));
3423
3424 /* If we don't already know that there is a user-defined component to
3425 * this swash, and the user has indicated they wish to know if there is
3426 * one (by passing <flags_p>), find out */
3427 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3428 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3429 if (user_defined && SvUV(*user_defined)) {
3430 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3431 }
3432 }
3433 }
3434
36eb48b4
KW
3435 /* Make sure there is an inversion list for binary properties */
3436 if (minbits == 1) {
3437 SV** swash_invlistsvp = NULL;
3438 SV* swash_invlist = NULL;
9a53f6cf 3439 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3440 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3441 an unclaimed reference count */
36eb48b4 3442
9a53f6cf 3443 /* If this operation fetched a swash, get its already existing
83199d38 3444 * inversion list, or create one for it */
36eb48b4 3445
83199d38 3446 if (swash_hv) {
5c9f4bd2 3447 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3448 if (swash_invlistsvp) {
3449 swash_invlist = *swash_invlistsvp;
3450 invlist_in_swash_is_valid = TRUE;
3451 }
3452 else {
36eb48b4 3453 swash_invlist = _swash_to_invlist(retval);
02c85471 3454 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3455 }
3456 }
3457
3458 /* If an inversion list was passed in, have to include it */
3459 if (invlist) {
3460
3461 /* Any fetched swash will by now have an inversion list in it;
3462 * otherwise <swash_invlist> will be NULL, indicating that we
3463 * didn't fetch a swash */
3464 if (swash_invlist) {
3465
3466 /* Add the passed-in inversion list, which invalidates the one
3467 * already stored in the swash */
3468 invlist_in_swash_is_valid = FALSE;
eee4c920 3469 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3470 _invlist_union(invlist, swash_invlist, &swash_invlist);
3471 }
3472 else {
3473
87367d5f
KW
3474 /* Here, there is no swash already. Set up a minimal one, if
3475 * we are going to return a swash */
3476 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3477 swash_hv = newHV();
4aca0fe6 3478 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3479 }
9a53f6cf
KW
3480 swash_invlist = invlist;
3481 }
9a53f6cf
KW
3482 }
3483
3484 /* Here, we have computed the union of all the passed-in data. It may
3485 * be that there was an inversion list in the swash which didn't get
538e84ed 3486 * touched; otherwise save the computed one */
87367d5f
KW
3487 if (! invlist_in_swash_is_valid
3488 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3489 {
5c9f4bd2 3490 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3491 {
3492 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3493 }
cc34d8c5
FC
3494 /* We just stole a reference count. */
3495 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3496 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3497 }
87367d5f 3498
ee3222e3 3499 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3500 SvREADONLY_on(swash_invlist);
3501
c41800a8 3502 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3503 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3504 SvREFCNT_dec(retval);
02c85471
FC
3505 if (!swash_invlist_unclaimed)
3506 SvREFCNT_inc_simple_void_NN(swash_invlist);
3507 retval = newRV_noinc(swash_invlist);
87367d5f 3508 }
36eb48b4
KW
3509 }
3510
2c1f00b9
YO
3511 CORE_SWASH_INIT_RETURN(retval);
3512#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3513}
3514
035d37be
JH
3515
3516/* This API is wrong for special case conversions since we may need to
3517 * return several Unicode characters for a single Unicode character
3518 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3519 * the lower-level routine, and it is similarly broken for returning
38684baa 3520 * multiple values. --jhi
b9992569 3521 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3522/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3523
3524/* Note:
3525 * Returns the value of property/mapping C<swash> for the first character
3526 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3527 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3528 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3529 *
3530 * A "swash" is a hash which contains initially the keys/values set up by
3531 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3532 * property for all possible code points. Things are stored in a compact form
3533 * (see utf8_heavy.pl) so that calculation is required to find the actual
3534 * property value for a given code point. As code points are looked up, new
3535 * key/value pairs are added to the hash, so that the calculation doesn't have
3536 * to ever be re-done. Further, each calculation is done, not just for the
3537 * desired one, but for a whole block of code points adjacent to that one.
3538 * For binary properties on ASCII machines, the block is usually for 64 code
3539 * points, starting with a code point evenly divisible by 64. Thus if the
3540 * property value for code point 257 is requested, the code goes out and
3541 * calculates the property values for all 64 code points between 256 and 319,
3542 * and stores these as a single 64-bit long bit vector, called a "swatch",
3543 * under the key for code point 256. The key is the UTF-8 encoding for code
3544 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3545 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3546 * for code point 258 is then requested, this code realizes that it would be
3547 * stored under the key for 256, and would find that value and extract the
3548 * relevant bit, offset from 256.
3549 *
3550 * Non-binary properties are stored in as many bits as necessary to represent
3551 * their values (32 currently, though the code is more general than that), not
fc273927 3552 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3553 * vector that encompasses the property values for all code points whose UTF-8
3554 * representations are represented by the key. That is, for all code points
3555 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3556 * bytes of that.
680c470c 3557 */
a0ed51b3 3558UV
680c470c 3559Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3560{
ef8f7699 3561 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3562 U32 klen;
3563 U32 off;
9b56a019 3564 STRLEN slen = 0;
7d85a32c 3565 STRLEN needents;
cfd0369c 3566 const U8 *tmps = NULL;
979f2922 3567 SV *swatch;
08fb1ac5 3568 const U8 c = *ptr;
3568d838 3569
7918f24d
NC
3570 PERL_ARGS_ASSERT_SWASH_FETCH;
3571
87367d5f
KW
3572 /* If it really isn't a hash, it isn't really swash; must be an inversion
3573 * list */
3574 if (SvTYPE(hv) != SVt_PVHV) {
3575 return _invlist_contains_cp((SV*)hv,
3576 (do_utf8)
3577 ? valid_utf8_to_uvchr(ptr, NULL)
3578 : c);
3579 }
3580
08fb1ac5
KW
3581 /* We store the values in a "swatch" which is a vec() value in a swash
3582 * hash. Code points 0-255 are a single vec() stored with key length
3583 * (klen) 0. All other code points have a UTF-8 representation
3584 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3585 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3586 * length for them is the length of the encoded char - 1. ptr[klen] is the
3587 * final byte in the sequence representing the character */
3588 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3589 klen = 0;
3590 needents = 256;
3591 off = c;
3568d838 3592 }
08fb1ac5
KW
3593 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3594 klen = 0;
3595 needents = 256;
a62b247b 3596 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3597 }
3598 else {
08fb1ac5
KW
3599 klen = UTF8SKIP(ptr) - 1;
3600
3601 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3602 * the vec is the final byte in the sequence. (In EBCDIC this is
3603 * converted to I8 to get consecutive values.) To help you visualize
3604 * all this:
3605 * Straight 1047 After final byte
3606 * UTF-8 UTF-EBCDIC I8 transform
3607 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3608 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3609 * ...
3610 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3611 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3612 * ...
3613 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3614 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3615 * ...
3616 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3617 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3618 * ...
3619 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3620 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3621 *
3622 * (There are no discontinuities in the elided (...) entries.)
3623 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3624 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3625 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3626 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3627 * index into the vec() swatch (after subtracting 0x80, which we
3628 * actually do with an '&').
3629 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3630 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3631 * dicontinuities which go away by transforming it into I8, and we
3632 * effectively subtract 0xA0 to get the index. */
979f2922 3633 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3634 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3635 }
7d85a32c 3636
a0ed51b3 3637 /*
4a4088c4 3638 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
3639 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3640 * it's nothing to sniff at.) Pity we usually come through at least
3641 * two function calls to get here...
3642 *
3643 * NB: this code assumes that swatches are never modified, once generated!
3644 */
3645
3568d838 3646 if (hv == PL_last_swash_hv &&
a0ed51b3 3647 klen == PL_last_swash_klen &&
27da23d5 3648 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3649 {
3650 tmps = PL_last_swash_tmps;
3651 slen = PL_last_swash_slen;
3652 }
3653 else {
3654 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3655 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3656
b0e3252e 3657 /* If not cached, generate it via swatch_get */
979f2922 3658 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3659 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3660 {
3661 if (klen) {
3662 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3663 swatch = swatch_get(swash,
3664 code_point & ~((UV)needents - 1),
3665 needents);
3666 }
3667 else { /* For the first 256 code points, the swatch has a key of
3668 length 0 */
3669 swatch = swatch_get(swash, 0, needents);
3670 }
979f2922 3671
923e4eb5 3672 if (IN_PERL_COMPILETIME)
623e6609 3673 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3674
979f2922 3675 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3676
979f2922
ST
3677 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3678 || (slen << 3) < needents)
5637ef5b 3679 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
147e3846 3680 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
5637ef5b 3681 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3682 }
3683
3684 PL_last_swash_hv = hv;
16d8f38a 3685 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3686 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3687 /* FIXME change interpvar.h? */
3688 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3689 PL_last_swash_slen = slen;
3690 if (klen)
3691 Copy(ptr, PL_last_swash_key, klen, U8);
3692 }
3693
9faf8d75 3694 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3695 case 1:
e7aca353 3696 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3697 case 8:
e7aca353 3698 return ((UV) tmps[off]);
a0ed51b3
LW
3699 case 16:
3700 off <<= 1;
e7aca353
JH
3701 return
3702 ((UV) tmps[off ] << 8) +
3703 ((UV) tmps[off + 1]);
a0ed51b3
LW
3704 case 32:
3705 off <<= 2;
e7aca353
JH
3706 return
3707 ((UV) tmps[off ] << 24) +
3708 ((UV) tmps[off + 1] << 16) +
3709 ((UV) tmps[off + 2] << 8) +
3710 ((UV) tmps[off + 3]);
a0ed51b3 3711 }
5637ef5b 3712 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
147e3846 3713 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
670f1322 3714 NORETURN_FUNCTION_END;
a0ed51b3 3715}
2b9d42f0 3716
319009ee
KW
3717/* Read a single line of the main body of the swash input text. These are of
3718 * the form:
3719 * 0053 0056 0073
3720 * where each number is hex. The first two numbers form the minimum and
3721 * maximum of a range, and the third is the value associated with the range.
3722 * Not all swashes should have a third number
3723 *
3724 * On input: l points to the beginning of the line to be examined; it points
3725 * to somewhere in the string of the whole input text, and is