This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
regexec.c: Make isFOO_lc() non-static
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
147e3846 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf;
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
81 SAVESPTR(PL_dowarn);
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
8ee1cdcb
KW
119#define HANDLE_UNICODE_SURROGATE(uv, flags) \
120 STMT_START { \
121 if (flags & UNICODE_WARN_SURROGATE) { \
122 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
147e3846 123 "UTF-16 surrogate U+%04" UVXf, uv); \
8ee1cdcb
KW
124 } \
125 if (flags & UNICODE_DISALLOW_SURROGATE) { \
126 return NULL; \
127 } \
128 } STMT_END;
129
130#define HANDLE_UNICODE_NONCHAR(uv, flags) \
131 STMT_START { \
132 if (flags & UNICODE_WARN_NONCHAR) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
147e3846 134 "Unicode non-character U+%04" UVXf " is not " \
8ee1cdcb
KW
135 "recommended for open interchange", uv); \
136 } \
137 if (flags & UNICODE_DISALLOW_NONCHAR) { \
138 return NULL; \
139 } \
140 } STMT_END;
141
ba6ed43c
KW
142/* Use shorter names internally in this file */
143#define SHIFT UTF_ACCUMULATION_SHIFT
144#undef MARK
145#define MARK UTF_CONTINUATION_MARK
146#define MASK UTF_CONTINUATION_MASK
147
dfe13c55 148U8 *
378516de 149Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 150{
378516de 151 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 152
2d1545e5 153 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 154 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
155 return d;
156 }
facc1dc2 157
3ea68d71 158 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
159 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
160 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
161 return d;
162 }
d9432125 163
ba6ed43c
KW
164 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
165 * below, the 16 is for start bytes E0-EF (which are all the possible ones
166 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
167 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
168 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
169 * 0x800-0xFFFF on ASCII */
170 if (uv < (16 * (1U << (2 * SHIFT)))) {
171 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
172 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
173 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
174
175#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
176 aren't tested here */
177 /* The most likely code points in this range are below the surrogates.
178 * Do an extra test to quickly exclude those. */
179 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
180 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
181 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
182 {
8ee1cdcb
KW
183 HANDLE_UNICODE_NONCHAR(uv, flags);
184 }
185 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
186 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 187 }
ba6ed43c
KW
188 }
189#endif
190 return d;
191 }
192
193 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
194 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
195 * happen starting with 4-byte characters on ASCII platforms. We unify the
196 * code for these with EBCDIC, even though some of them require 5-bytes on
197 * those, because khw believes the code saving is worth the very slight
198 * performance hit on these high EBCDIC code points. */
199
200 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
201 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
202 && ckWARN_d(WARN_DEPRECATED))
203 {
204 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
205 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
206 }
207 if ( (flags & UNICODE_WARN_SUPER)
208 || ( UNICODE_IS_ABOVE_31_BIT(uv)
209 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
210 {
211 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
212
213 /* Choose the more dire applicable warning */
214 (UNICODE_IS_ABOVE_31_BIT(uv))
147e3846
KW
215 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
216 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
a5bf80e0
KW
217 uv);
218 }
219 if (flags & UNICODE_DISALLOW_SUPER
220 || ( UNICODE_IS_ABOVE_31_BIT(uv)
221 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
222 {
223 return NULL;
224 }
225 }
ba6ed43c
KW
226 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
227 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 228 }
d9432125 229
ba6ed43c
KW
230 /* Test for and handle 4-byte result. In the test immediately below, the
231 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
232 * characters). The 3 is for 3 continuation bytes; these each contribute
233 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
234 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
235 * 0x1_0000-0x1F_FFFF on ASCII */
236 if (uv < (8 * (1U << (3 * SHIFT)))) {
237 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
238 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
239 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
240 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
241
242#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
243 characters. The end-plane non-characters for EBCDIC were
244 handled just above */
245 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
246 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 247 }
ba6ed43c
KW
248 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
249 HANDLE_UNICODE_SURROGATE(uv, flags);
250 }
251#endif
252
253 return d;
254 }
255
256 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
257 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
258 * format. The unrolled version above turns out to not save all that much
259 * time, and at these high code points (well above the legal Unicode range
260 * on ASCII platforms, and well above anything in common use in EBCDIC),
261 * khw believes that less code outweighs slight performance gains. */
262
d9432125 263 {
5aaebcb3 264 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
265 U8 *p = d+len-1;
266 while (p > d) {
4c8cd605 267 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
268 uv >>= UTF_ACCUMULATION_SHIFT;
269 }
4c8cd605 270 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
271 return d+len;
272 }
a0ed51b3 273}
a5bf80e0 274
646ca15d 275/*
07693fe6
KW
276=for apidoc uvchr_to_utf8
277
bcb1a2d4 278Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 279of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
280C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
281the byte after the end of the new character. In other words,
07693fe6
KW
282
283 d = uvchr_to_utf8(d, uv);
284
285is the recommended wide native character-aware way of saying
286
287 *(d++) = uv;
288
760c7c2f
KW
289This function accepts any UV as input, but very high code points (above
290C<IV_MAX> on the platform) will raise a deprecation warning. This is
291typically 0x7FFF_FFFF in a 32-bit word.
292
293It is possible to forbid or warn on non-Unicode code points, or those that may
294be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 295
07693fe6
KW
296=cut
297*/
298
de69f3af
KW
299/* This is also a macro */
300PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
301
07693fe6
KW
302U8 *
303Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
304{
de69f3af 305 return uvchr_to_utf8(d, uv);
07693fe6
KW
306}
307
de69f3af
KW
308/*
309=for apidoc uvchr_to_utf8_flags
310
311Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 312of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
313C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
314the byte after the end of the new character. In other words,
de69f3af
KW
315
316 d = uvchr_to_utf8_flags(d, uv, flags);
317
318or, in most cases,
319
320 d = uvchr_to_utf8_flags(d, uv, 0);
321
322This is the Unicode-aware way of saying
323
324 *(d++) = uv;
325
760c7c2f
KW
326If C<flags> is 0, this function accepts any UV as input, but very high code
327points (above C<IV_MAX> for the platform) will raise a deprecation warning.
328This is typically 0x7FFF_FFFF in a 32-bit word.
329
330Specifying C<flags> can further restrict what is allowed and not warned on, as
331follows:
de69f3af 332
796b6530 333If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
334the function will raise a warning, provided UTF8 warnings are enabled. If
335instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
336NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 337
760c7c2f
KW
338Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
339affect how the function handles a Unicode non-character.
93e6dbd6 340
760c7c2f
KW
341And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
342affect the handling of code points that are above the Unicode maximum of
3430x10FFFF. Languages other than Perl may not be able to accept files that
344contain these.
93e6dbd6
KW
345
346The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
347the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
348three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
349allowed inputs to the strict UTF-8 traditionally defined by Unicode.
350Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
351C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
352above-Unicode and surrogate flags, but not the non-character ones, as
353defined in
354L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
355See L<perlunicode/Noncharacter code points>.
93e6dbd6 356
ab8e6d41
KW
357Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
358so using them is more problematic than other above-Unicode code points. Perl
359invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
360likely that non-Perl languages will not be able to read files that contain
361these that written by the perl interpreter; nor would Perl understand files
362written by something that uses a different extension. For these reasons, there
363is a separate set of flags that can warn and/or disallow these extremely high
364code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
365C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
366are entirely independent from the deprecation warning for code points above
367C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
368code point that needs more than 31 bits to represent. When that happens,
369effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
37032-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
371above-Unicode code points, including these, as malformations; and
372C<UNICODE_WARN_SUPER> warns on these.)
373
374On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
375extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
376than on ASCII. Prior to that, code points 2**31 and higher were simply
377unrepresentable, and a different, incompatible method was used to represent
378code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
379and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
380platforms, warning and disallowing 2**31 and higher.
de69f3af 381
de69f3af
KW
382=cut
383*/
384
385/* This is also a macro */
386PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
387
07693fe6
KW
388U8 *
389Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
390{
de69f3af 391 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
392}
393
83dc0f42
KW
394PERL_STATIC_INLINE bool
395S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
396{
397 /* Returns TRUE if the first code point represented by the Perl-extended-
398 * UTF-8-encoded string starting at 's', and looking no further than 'e -
399 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
400 *
401 * The function handles the case where the input bytes do not include all
402 * the ones necessary to represent a full character. That is, they may be
403 * the intial bytes of the representation of a code point, but possibly
404 * the final ones necessary for the complete representation may be beyond
405 * 'e - 1'.
406 *
407 * The function assumes that the sequence is well-formed UTF-8 as far as it
408 * goes, and is for a UTF-8 variant code point. If the sequence is
409 * incomplete, the function returns FALSE if there is any well-formed
410 * UTF-8 byte sequence that can complete it in such a way that a code point
411 * < 2**31 is produced; otherwise it returns TRUE.
412 *
413 * Getting this exactly right is slightly tricky, and has to be done in
414 * several places in this file, so is centralized here. It is based on the
415 * following table:
416 *
417 * U+7FFFFFFF (2 ** 31 - 1)
418 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
419 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
420 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
421 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
422 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
423 * U+80000000 (2 ** 31):
424 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
425 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
426 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
428 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
429 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
430 */
431
432#ifdef EBCDIC
433
37086697
KW
434 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
435 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
436 const STRLEN prefix_len = sizeof(prefix) - 1;
437 const STRLEN len = e - s;
f880f78a 438 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
439
440#else
441
442 PERL_UNUSED_ARG(e);
443
444#endif
445
446 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
447
448 assert(! UTF8_IS_INVARIANT(*s));
449
450#ifndef EBCDIC
451
452 /* Technically, a start byte of FE can be for a code point that fits into
453 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
454 * malformation. */
455 return (*s >= 0xFE);
456
457#else
458
459 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
460 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
461 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
462 * bits. */
463 if (*s != 0xFE || len == 1) {
464 return FALSE;
465 }
466
467 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
468 * \x41 and \x42. */
469 return cBOOL(memGT(s + 1, prefix, cmp_len));
470
471#endif
472
473}
474
12a4bed3
KW
475PERL_STATIC_INLINE bool
476S_does_utf8_overflow(const U8 * const s, const U8 * e)
477{
478 const U8 *x;
479 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
480
c551bb01
KW
481#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
482
483 const STRLEN len = e - s;
484
485#endif
486
12a4bed3
KW
487 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
488 * platform, that is if it represents a code point larger than the highest
489 * representable code point. (For ASCII platforms, we could use memcmp()
490 * because we don't have to convert each byte to I8, but it's very rare
491 * input indeed that would approach overflow, so the loop below will likely
492 * only get executed once.
493 *
494 * 'e' must not be beyond a full character. If it is less than a full
495 * character, the function returns FALSE if there is any input beyond 'e'
496 * that could result in a non-overflowing code point */
497
498 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
499 assert(s <= e && s + UTF8SKIP(s) >= e);
500
501#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
502
503 /* On 32 bit ASCII machines, many overlongs that start with FF don't
504 * overflow */
505
c551bb01 506 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
507 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
508 return memGE(s, max_32_bit_overlong,
c551bb01 509 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
510 }
511
512#endif
12a4bed3
KW
513
514 for (x = s; x < e; x++, y++) {
515
516 /* If this byte is larger than the corresponding highest UTF-8 byte, it
517 * overflows */
518 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
519 return TRUE;
520 }
521
522 /* If not the same as this byte, it must be smaller, doesn't overflow */
523 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
524 return FALSE;
525 }
526 }
527
528 /* Got to the end and all bytes are the same. If the input is a whole
529 * character, it doesn't overflow. And if it is a partial character,
530 * there's not enough information to tell, so assume doesn't overflow */
531 return FALSE;
532}
533
534PERL_STATIC_INLINE bool
535S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
536{
537 /* Overlongs can occur whenever the number of continuation bytes
538 * changes. That means whenever the number of leading 1 bits in a start
539 * byte increases from the next lower start byte. That happens for start
540 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
541 * illegal start bytes have already been excluded, so don't need to be
542 * tested here;
543 * ASCII platforms: C0, C1
544 * EBCDIC platforms C0, C1, C2, C3, C4, E0
545 *
546 * At least a second byte is required to determine if other sequences will
547 * be an overlong. */
548
549 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
550 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
551
552 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
553 assert(len > 1 && UTF8_IS_START(*s));
554
555 /* Each platform has overlongs after the start bytes given above (expressed
556 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
557 * the logic is the same, except the E0 overlong has already been excluded
558 * on EBCDIC platforms. The values below were found by manually
559 * inspecting the UTF-8 patterns. See the tables in utf8.h and
560 * utfebcdic.h. */
561
562# ifdef EBCDIC
563# define F0_ABOVE_OVERLONG 0xB0
564# define F8_ABOVE_OVERLONG 0xA8
565# define FC_ABOVE_OVERLONG 0xA4
566# define FE_ABOVE_OVERLONG 0xA2
567# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
568 /* I8(0xfe) is FF */
569# else
570
571 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
572 return TRUE;
573 }
574
575# define F0_ABOVE_OVERLONG 0x90
576# define F8_ABOVE_OVERLONG 0x88
577# define FC_ABOVE_OVERLONG 0x84
578# define FE_ABOVE_OVERLONG 0x82
579# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
580# endif
581
582
583 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
584 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
585 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
586 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
587 {
588 return TRUE;
589 }
590
b0b342d4
KW
591 /* Check for the FF overlong */
592 return isFF_OVERLONG(s, len);
593}
594
595PERL_STATIC_INLINE bool
596S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
597{
598 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
599
600 /* Check for the FF overlong. This happens only if all these bytes match;
601 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 602 * utfebcdic.h. */
12a4bed3 603
b0b342d4
KW
604 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
605 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
606 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
607}
608
609#undef F0_ABOVE_OVERLONG
610#undef F8_ABOVE_OVERLONG
611#undef FC_ABOVE_OVERLONG
612#undef FE_ABOVE_OVERLONG
613#undef FF_OVERLONG_PREFIX
614
35f8c9bd 615STRLEN
edc2c47a 616Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 617{
2b479609 618 STRLEN len;
12a4bed3 619 const U8 *x;
35f8c9bd 620
2b479609
KW
621 /* A helper function that should not be called directly.
622 *
623 * This function returns non-zero if the string beginning at 's' and
624 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
625 * code point; otherwise it returns 0. The examination stops after the
626 * first code point in 's' is validated, not looking at the rest of the
627 * input. If 'e' is such that there are not enough bytes to represent a
628 * complete code point, this function will return non-zero anyway, if the
629 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
630 * excluded by 'flags'.
631 *
632 * A non-zero return gives the number of bytes required to represent the
633 * code point. Be aware that if the input is for a partial character, the
634 * return will be larger than 'e - s'.
635 *
636 * This function assumes that the code point represented is UTF-8 variant.
637 * The caller should have excluded this possibility before calling this
638 * function.
639 *
640 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
641 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
642 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
643 * disallowed by the flags. If the input is only for a partial character,
644 * the function will return non-zero if there is any sequence of
645 * well-formed UTF-8 that, when appended to the input sequence, could
646 * result in an allowed code point; otherwise it returns 0. Non characters
647 * cannot be determined based on partial character input. But many of the
648 * other excluded types can be determined with just the first one or two
649 * bytes.
650 *
651 */
652
653 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
654
655 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
656 |UTF8_DISALLOW_ABOVE_31_BIT)));
657 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 658
2b479609 659 /* A variant char must begin with a start byte */
35f8c9bd
KW
660 if (UNLIKELY(! UTF8_IS_START(*s))) {
661 return 0;
662 }
663
edc2c47a
KW
664 /* Examine a maximum of a single whole code point */
665 if (e - s > UTF8SKIP(s)) {
666 e = s + UTF8SKIP(s);
667 }
668
2b479609
KW
669 len = e - s;
670
671 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
672 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 673
2b479609
KW
674 /* The code below is derived from this table. Keep in mind that legal
675 * continuation bytes range between \x80..\xBF for UTF-8, and
676 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
677 * Hence, we don't have to test the upper edge because if any of those
678 * are encountered, the sequence is malformed, and will fail elsewhere
679 * in this function.
680 * UTF-8 UTF-EBCDIC I8
681 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
682 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
683 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
684 *
685 */
686
687#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
688# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 689# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 690
19794540
KW
691# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
692 /* B6 and B7 */ \
693 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
694#else
695# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
696# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
697# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
698#endif
699
700 if ( (flags & UTF8_DISALLOW_SUPER)
ddb65933
KW
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
702 {
2b479609
KW
703 return 0; /* Above Unicode */
704 }
705
706 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
707 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
708 {
709 return 0; /* Above 31 bits */
710 }
711
712 if (len > 1) {
713 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
714
715 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 716 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
717 {
718 return 0; /* Above Unicode */
719 }
720
721 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 722 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
723 {
724 return 0; /* Surrogate */
725 }
726
727 if ( (flags & UTF8_DISALLOW_NONCHAR)
728 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
729 {
730 return 0; /* Noncharacter code point */
731 }
732 }
733 }
734
735 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
736 for (x = s + 1; x < e; x++) {
737 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
738 return 0;
739 }
740 }
741
af13dd8a 742 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
743 * overlong. */
744 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
745 return 0;
af13dd8a
KW
746 }
747
12a4bed3
KW
748 /* And finally, that the code point represented fits in a word on this
749 * platform */
750 if (does_utf8_overflow(s, e)) {
751 return 0;
35f8c9bd
KW
752 }
753
2b479609 754 return UTF8SKIP(s);
35f8c9bd
KW
755}
756
7cf8d05d
KW
757STATIC char *
758S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
759{
760 /* Returns a mortalized C string that is a displayable copy of the 'len'
761 * bytes starting at 's', each in a \xXY format. */
762
763 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
764 trailing NUL */
765 const U8 * const e = s + len;
766 char * output;
767 char * d;
768
769 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
770
771 Newx(output, output_len, char);
772 SAVEFREEPV(output);
773
774 d = output;
775 for (; s < e; s++) {
776 const unsigned high_nibble = (*s & 0xF0) >> 4;
777 const unsigned low_nibble = (*s & 0x0F);
778
779 *d++ = '\\';
780 *d++ = 'x';
781
782 if (high_nibble < 10) {
783 *d++ = high_nibble + '0';
784 }
785 else {
786 *d++ = high_nibble - 10 + 'a';
787 }
788
789 if (low_nibble < 10) {
790 *d++ = low_nibble + '0';
791 }
792 else {
793 *d++ = low_nibble - 10 + 'a';
794 }
795 }
796
797 *d = '\0';
798 return output;
799}
800
806547a7 801PERL_STATIC_INLINE char *
7cf8d05d
KW
802S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
803
804 /* How many bytes to print */
3cc6a05e 805 STRLEN print_len,
7cf8d05d
KW
806
807 /* Which one is the non-continuation */
808 const STRLEN non_cont_byte_pos,
809
810 /* How many bytes should there be? */
811 const STRLEN expect_len)
806547a7
KW
812{
813 /* Return the malformation warning text for an unexpected continuation
814 * byte. */
815
7cf8d05d 816 const char * const where = (non_cont_byte_pos == 1)
806547a7 817 ? "immediately"
7cf8d05d
KW
818 : Perl_form(aTHX_ "%d bytes",
819 (int) non_cont_byte_pos);
3cc6a05e 820 unsigned int i;
806547a7
KW
821
822 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
823
7cf8d05d
KW
824 /* We don't need to pass this parameter, but since it has already been
825 * calculated, it's likely faster to pass it; verify under DEBUGGING */
826 assert(expect_len == UTF8SKIP(s));
827
3cc6a05e
KW
828 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
829 * length that is larger than is actually available in the buffer. If we
830 * print all the bytes based on that length, we will read past the buffer
831 * end. Often, the strings are NUL terminated, so to lower the chances of
832 * this happening, print the malformed bytes only up through any NUL. */
833 for (i = 1; i < print_len; i++) {
834 if (*(s + i) == '\0') {
835 print_len = i + 1; /* +1 gets the NUL printed */
836 break;
837 }
838 }
839
7cf8d05d
KW
840 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
841 " %s after start byte 0x%02x; need %d bytes, got %d)",
842 malformed_text,
843 _byte_dump_string(s, print_len),
844 *(s + non_cont_byte_pos),
845 where,
846 *s,
847 (int) expect_len,
848 (int) non_cont_byte_pos);
806547a7
KW
849}
850
35f8c9bd
KW
851/*
852
de69f3af 853=for apidoc utf8n_to_uvchr
378516de
KW
854
855THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 856Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 857
9041c2e3 858Bottom level UTF-8 decode routine.
de69f3af 859Returns the native code point value of the first character in the string C<s>,
746afd53
KW
860which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
861C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
862the length, in bytes, of that character.
949cf498
KW
863
864The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
865well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
866causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
867is the next possible position in C<s> that could begin a non-malformed
868character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
869is raised. Some UTF-8 input sequences may contain multiple malformations.
870This function tries to find every possible one in each call, so multiple
871warnings can be raised for each sequence.
949cf498
KW
872
873Various ALLOW flags can be set in C<flags> to allow (and not warn on)
874individual types of malformations, such as the sequence being overlong (that
875is, when there is a shorter sequence that can express the same code point;
876overlong sequences are expressly forbidden in the UTF-8 standard due to
877potential security issues). Another malformation example is the first byte of
878a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
879flags. Even if allowed, this function generally returns the Unicode
880REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
881F<utf8.h> to override this behavior for the overlong malformations, but don't
882do that except for very specialized purposes.
949cf498 883
796b6530 884The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
885flags) malformation is found. If this flag is set, the routine assumes that
886the caller will raise a warning, and this function will silently just set
d088425d
KW
887C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
888
75200dff 889Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 890character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
891in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
892be set to 1. To disambiguate, upon a zero return, see if the first byte of
893C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 894error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
895
896Certain code points are considered problematic. These are Unicode surrogates,
746afd53 897Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 898By default these are considered regular code points, but certain situations
ecc1615f
KW
899warrant special handling for them, which can be specified using the C<flags>
900parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
901three classes are treated as malformations and handled as such. The flags
902C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
903C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
904disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
905restricts the allowed inputs to the strict UTF-8 traditionally defined by
906Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
907definition given by
908L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
909The difference between traditional strictness and C9 strictness is that the
910latter does not forbid non-character code points. (They are still discouraged,
911however.) For more discussion see L<perlunicode/Noncharacter code points>.
912
913The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
914C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
915C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
916raised for their respective categories, but otherwise the code points are
917considered valid (not malformations). To get a category to both be treated as
918a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 919(But note that warnings are not raised if lexically disabled nor if
796b6530 920C<UTF8_CHECK_ONLY> is also specified.)
949cf498 921
760c7c2f
KW
922It is now deprecated to have very high code points (above C<IV_MAX> on the
923platforms) and this function will raise a deprecation warning for these (unless
d5944cab 924such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 925in a 32-bit word.
ab8e6d41
KW
926
927Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
928so using them is more problematic than other above-Unicode code points. Perl
929invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
930likely that non-Perl languages will not be able to read files that contain
aff2be59 931these; nor would Perl understand files
ab8e6d41
KW
932written by something that uses a different extension. For these reasons, there
933is a separate set of flags that can warn and/or disallow these extremely high
934code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
935C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
936are entirely independent from the deprecation warning for code points above
937C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
938code point that needs more than 31 bits to represent. When that happens,
939effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
94032-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
941above-Unicode code points, including these, as malformations; and
942C<UTF8_WARN_SUPER> warns on these.)
943
944On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
945extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
946than on ASCII. Prior to that, code points 2**31 and higher were simply
947unrepresentable, and a different, incompatible method was used to represent
948code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
949and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
950platforms, warning and disallowing 2**31 and higher.
949cf498
KW
951
952All other code points corresponding to Unicode characters, including private
953use and those yet to be assigned, are never considered malformed and never
954warn.
67e989fb 955
37607a96 956=cut
f9380377
KW
957
958Also implemented as a macro in utf8.h
959*/
960
961UV
962Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
963 STRLEN curlen,
964 STRLEN *retlen,
965 const U32 flags)
966{
967 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
968
969 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
970}
971
972/*
973
974=for apidoc utf8n_to_uvchr_error
975
976THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
977Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
978
979This function is for code that needs to know what the precise malformation(s)
980are when an error is found.
981
982It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
983all the others, C<errors>. If this parameter is 0, this function behaves
984identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
985to a C<U32> variable, which this function sets to indicate any errors found.
986Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
987C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
988of these bits will be set if a malformation is found, even if the input
7a65503b 989C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
990exceptions are noted:
991
992=over 4
993
994=item C<UTF8_GOT_ABOVE_31_BIT>
995
996The code point represented by the input UTF-8 sequence occupies more than 31
997bits.
998This bit is set only if the input C<flags> parameter contains either the
999C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
1000
1001=item C<UTF8_GOT_CONTINUATION>
1002
1003The input sequence was malformed in that the first byte was a a UTF-8
1004continuation byte.
1005
1006=item C<UTF8_GOT_EMPTY>
1007
1008The input C<curlen> parameter was 0.
1009
1010=item C<UTF8_GOT_LONG>
1011
1012The input sequence was malformed in that there is some other sequence that
1013evaluates to the same code point, but that sequence is shorter than this one.
1014
1015=item C<UTF8_GOT_NONCHAR>
1016
1017The code point represented by the input UTF-8 sequence is for a Unicode
1018non-character code point.
1019This bit is set only if the input C<flags> parameter contains either the
1020C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1021
1022=item C<UTF8_GOT_NON_CONTINUATION>
1023
1024The input sequence was malformed in that a non-continuation type byte was found
1025in a position where only a continuation type one should be.
1026
1027=item C<UTF8_GOT_OVERFLOW>
1028
1029The input sequence was malformed in that it is for a code point that is not
1030representable in the number of bits available in a UV on the current platform.
1031
1032=item C<UTF8_GOT_SHORT>
1033
1034The input sequence was malformed in that C<curlen> is smaller than required for
1035a complete sequence. In other words, the input is for a partial character
1036sequence.
1037
1038=item C<UTF8_GOT_SUPER>
1039
1040The input sequence was malformed in that it is for a non-Unicode code point;
1041that is, one above the legal Unicode maximum.
1042This bit is set only if the input C<flags> parameter contains either the
1043C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1044
1045=item C<UTF8_GOT_SURROGATE>
1046
1047The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1048code point.
1049This bit is set only if the input C<flags> parameter contains either the
1050C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1051
1052=back
1053
133551d8
KW
1054To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1055flag to suppress any warnings, and then examine the C<*errors> return.
1056
f9380377 1057=cut
37607a96 1058*/
67e989fb 1059
a0ed51b3 1060UV
f9380377
KW
1061Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1062 STRLEN curlen,
1063 STRLEN *retlen,
1064 const U32 flags,
1065 U32 * errors)
a0ed51b3 1066{
d4c19fe8 1067 const U8 * const s0 = s;
2b5e7bc2
KW
1068 U8 * send = NULL; /* (initialized to silence compilers' wrong
1069 warning) */
1070 U32 possible_problems = 0; /* A bit is set here for each potential problem
1071 found as we go along */
eb83ed87 1072 UV uv = *s;
2b5e7bc2
KW
1073 STRLEN expectlen = 0; /* How long should this sequence be?
1074 (initialized to silence compilers' wrong
1075 warning) */
e308b348 1076 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1077 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1078 this gets set and discarded */
a0dbb045 1079
2b5e7bc2
KW
1080 /* The below are used only if there is both an overlong malformation and a
1081 * too short one. Otherwise the first two are set to 's0' and 'send', and
1082 * the third not used at all */
1083 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1084 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1085 warning) */
2b5e7bc2 1086 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1087
f9380377
KW
1088 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1089
1090 if (errors) {
1091 *errors = 0;
1092 }
1093 else {
1094 errors = &discard_errors;
1095 }
a0dbb045 1096
eb83ed87
KW
1097 /* The order of malformation tests here is important. We should consume as
1098 * few bytes as possible in order to not skip any valid character. This is
1099 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1100 * http://unicode.org/reports/tr36 for more discussion as to why. For
1101 * example, once we've done a UTF8SKIP, we can tell the expected number of
1102 * bytes, and could fail right off the bat if the input parameters indicate
1103 * that there are too few available. But it could be that just that first
1104 * byte is garbled, and the intended character occupies fewer bytes. If we
1105 * blindly assumed that the first byte is correct, and skipped based on
1106 * that number, we could skip over a valid input character. So instead, we
1107 * always examine the sequence byte-by-byte.
1108 *
1109 * We also should not consume too few bytes, otherwise someone could inject
1110 * things. For example, an input could be deliberately designed to
1111 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1112 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1113 * which is actually part of of the overflowing sequence), that could look
1114 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1115 * sequence and process the rest, inappropriately.
1116 *
1117 * Some possible input sequences are malformed in more than one way. This
1118 * function goes to lengths to try to find all of them. This is necessary
1119 * for correctness, as the inputs may allow one malformation but not
1120 * another, and if we abandon searching for others after finding the
1121 * allowed one, we could allow in something that shouldn't have been.
1122 */
eb83ed87 1123
b5b9af04 1124 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1125 possible_problems |= UTF8_GOT_EMPTY;
1126 curlen = 0;
5a48568d 1127 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1128 goto ready_to_handle_errors;
0c443dc2
JH
1129 }
1130
eb83ed87
KW
1131 expectlen = UTF8SKIP(s);
1132
1133 /* A well-formed UTF-8 character, as the vast majority of calls to this
1134 * function will be for, has this expected length. For efficiency, set
1135 * things up here to return it. It will be overriden only in those rare
1136 * cases where a malformation is found */
1137 if (retlen) {
1138 *retlen = expectlen;
1139 }
1140
1141 /* An invariant is trivially well-formed */
1d72bdf6 1142 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1143 return uv;
a0ed51b3 1144 }
67e989fb 1145
eb83ed87 1146 /* A continuation character can't start a valid sequence */
b5b9af04 1147 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1148 possible_problems |= UTF8_GOT_CONTINUATION;
1149 curlen = 1;
1150 uv = UNICODE_REPLACEMENT;
1151 goto ready_to_handle_errors;
ba210ebe 1152 }
9041c2e3 1153
dcd27b3c 1154 /* Here is not a continuation byte, nor an invariant. The only thing left
ddb65933
KW
1155 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1156 * because it excludes start bytes like \xC0 that always lead to
1157 * overlongs.) */
dcd27b3c 1158
534752c1
KW
1159 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1160 * that indicate the number of bytes in the character's whole UTF-8
1161 * sequence, leaving just the bits that are part of the value. */
1162 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1163
e308b348
KW
1164 /* Setup the loop end point, making sure to not look past the end of the
1165 * input string, and flag it as too short if the size isn't big enough. */
1166 send = (U8*) s0;
1167 if (UNLIKELY(curlen < expectlen)) {
1168 possible_problems |= UTF8_GOT_SHORT;
1169 avail_len = curlen;
1170 send += curlen;
1171 }
1172 else {
1173 send += expectlen;
1174 }
1175 adjusted_send = send;
1176
eb83ed87 1177 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1178 * accumulating each into the working value as we go. */
eb83ed87 1179 for (s = s0 + 1; s < send; s++) {
b5b9af04 1180 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1181 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1182 continue;
1183 }
1184
1185 /* Here, found a non-continuation before processing all expected bytes.
1186 * This byte indicates the beginning of a new character, so quit, even
1187 * if allowing this malformation. */
2b5e7bc2 1188 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1189 break;
eb83ed87
KW
1190 } /* End of loop through the character's bytes */
1191
1192 /* Save how many bytes were actually in the character */
1193 curlen = s - s0;
1194
e308b348
KW
1195 /* A convenience macro that matches either of the too-short conditions. */
1196# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1197
e308b348 1198 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1199 uv_so_far = uv;
1200 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1201 }
1202
2b5e7bc2
KW
1203 /* Note that there are two types of too-short malformation. One is when
1204 * there is actual wrong data before the normal termination of the
1205 * sequence. The other is that the sequence wasn't complete before the end
1206 * of the data we are allowed to look at, based on the input 'curlen'.
1207 * This means that we were passed data for a partial character, but it is
1208 * valid as far as we saw. The other is definitely invalid. This
1209 * distinction could be important to a caller, so the two types are kept
1210 * separate. */
1211
1212 /* Check for overflow */
1213 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1214 possible_problems |= UTF8_GOT_OVERFLOW;
1215 uv = UNICODE_REPLACEMENT;
eb83ed87 1216 }
eb83ed87 1217
2b5e7bc2
KW
1218 /* Check for overlong. If no problems so far, 'uv' is the correct code
1219 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1220 * we must look at the UTF-8 byte sequence itself to see if it is for an
1221 * overlong */
1222 if ( ( LIKELY(! possible_problems)
1223 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1224 || ( UNLIKELY( possible_problems)
1225 && ( UNLIKELY(! UTF8_IS_START(*s0))
1226 || ( curlen > 1
1227 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1228 send - s0))))))
2f8f112e 1229 {
2b5e7bc2
KW
1230 possible_problems |= UTF8_GOT_LONG;
1231
2b5e7bc2
KW
1232 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1233 UV min_uv = uv_so_far;
1234 STRLEN i;
1235
1236 /* Here, the input is both overlong and is missing some trailing
1237 * bytes. There is no single code point it could be for, but there
1238 * may be enough information present to determine if what we have
1239 * so far is for an unallowed code point, such as for a surrogate.
1240 * The code below has the intelligence to determine this, but just
1241 * for non-overlong UTF-8 sequences. What we do here is calculate
1242 * the smallest code point the input could represent if there were
1243 * no too short malformation. Then we compute and save the UTF-8
1244 * for that, which is what the code below looks at instead of the
1245 * raw input. It turns out that the smallest such code point is
1246 * all we need. */
1247 for (i = curlen; i < expectlen; i++) {
1248 min_uv = UTF8_ACCUMULATE(min_uv,
1249 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1250 }
1251
1252 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1253 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1254 to free it ourselves if
1255 warnings are made fatal */
1256 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1257 }
eb83ed87
KW
1258 }
1259
2b5e7bc2
KW
1260 /* Now check that the input isn't for a problematic code point not allowed
1261 * by the input parameters. */
1262 /* isn't problematic if < this */
1263 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1264 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1265
1266 /* if overflow, we know without looking further
1267 * precisely which of the problematic types it is,
1268 * and we deal with those in the overflow handling
1269 * code */
1270 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1271 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1272 && ((flags & ( UTF8_DISALLOW_NONCHAR
1273 |UTF8_DISALLOW_SURROGATE
1274 |UTF8_DISALLOW_SUPER
1275 |UTF8_DISALLOW_ABOVE_31_BIT
1276 |UTF8_WARN_NONCHAR
1277 |UTF8_WARN_SURROGATE
1278 |UTF8_WARN_SUPER
1279 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1280 /* In case of a malformation, 'uv' is not valid, and has
1281 * been changed to something in the Unicode range.
1282 * Currently we don't output a deprecation message if there
1283 * is already a malformation, so we don't have to special
1284 * case the test immediately below */
760c7c2f
KW
1285 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1286 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1287 {
2b5e7bc2
KW
1288 /* If there were no malformations, or the only malformation is an
1289 * overlong, 'uv' is valid */
1290 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1291 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1292 possible_problems |= UTF8_GOT_SURROGATE;
1293 }
1294 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1295 possible_problems |= UTF8_GOT_SUPER;
1296 }
1297 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1298 possible_problems |= UTF8_GOT_NONCHAR;
1299 }
1300 }
1301 else { /* Otherwise, need to look at the source UTF-8, possibly
1302 adjusted to be non-overlong */
1303
1304 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1305 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1306 {
2b5e7bc2
KW
1307 possible_problems |= UTF8_GOT_SUPER;
1308 }
1309 else if (curlen > 1) {
1310 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1311 NATIVE_UTF8_TO_I8(*adjusted_s0),
1312 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1313 {
2b5e7bc2 1314 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1315 }
2b5e7bc2
KW
1316 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1317 NATIVE_UTF8_TO_I8(*adjusted_s0),
1318 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1319 {
1320 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1321 }
1322 }
c0236afe 1323
2b5e7bc2
KW
1324 /* We need a complete well-formed UTF-8 character to discern
1325 * non-characters, so can't look for them here */
1326 }
1327 }
949cf498 1328
2b5e7bc2
KW
1329 ready_to_handle_errors:
1330
1331 /* At this point:
1332 * curlen contains the number of bytes in the sequence that
1333 * this call should advance the input by.
e308b348
KW
1334 * avail_len gives the available number of bytes passed in, but
1335 * only if this is less than the expected number of
1336 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1337 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1338 * is set in it for each potential problem found.
1339 * uv contains the code point the input sequence
1340 * represents; or if there is a problem that prevents
1341 * a well-defined value from being computed, it is
1342 * some subsitute value, typically the REPLACEMENT
1343 * CHARACTER.
1344 * s0 points to the first byte of the character
1345 * send points to just after where that (potentially
1346 * partial) character ends
1347 * adjusted_s0 normally is the same as s0, but in case of an
1348 * overlong for which the UTF-8 matters below, it is
1349 * the first byte of the shortest form representation
1350 * of the input.
1351 * adjusted_send normally is the same as 'send', but if adjusted_s0
1352 * is set to something other than s0, this points one
1353 * beyond its end
1354 */
eb83ed87 1355
2b5e7bc2
KW
1356 if (UNLIKELY(possible_problems)) {
1357 bool disallowed = FALSE;
1358 const U32 orig_problems = possible_problems;
1359
1360 while (possible_problems) { /* Handle each possible problem */
1361 UV pack_warn = 0;
1362 char * message = NULL;
1363
1364 /* Each 'if' clause handles one problem. They are ordered so that
1365 * the first ones' messages will be displayed before the later
1366 * ones; this is kinda in decreasing severity order */
1367 if (possible_problems & UTF8_GOT_OVERFLOW) {
1368
1369 /* Overflow means also got a super and above 31 bits, but we
1370 * handle all three cases here */
1371 possible_problems
1372 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1373 *errors |= UTF8_GOT_OVERFLOW;
1374
1375 /* But the API says we flag all errors found */
1376 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1377 *errors |= UTF8_GOT_SUPER;
1378 }
ddb65933
KW
1379 if (flags
1380 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1381 {
f9380377
KW
1382 *errors |= UTF8_GOT_ABOVE_31_BIT;
1383 }
2b5e7bc2 1384
d60baaa7
KW
1385 /* Disallow if any of the three categories say to */
1386 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1387 || (flags & ( UTF8_DISALLOW_SUPER
1388 |UTF8_DISALLOW_ABOVE_31_BIT)))
1389 {
1390 disallowed = TRUE;
1391 }
1392
1393
1394 /* Likewise, warn if any say to, plus if deprecation warnings
1395 * are on, because this code point is above IV_MAX */
1396 if ( ckWARN_d(WARN_DEPRECATED)
1397 || ! (flags & UTF8_ALLOW_OVERFLOW)
1398 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1399 {
2b5e7bc2 1400
ddb65933
KW
1401 /* The warnings code explicitly says it doesn't handle the
1402 * case of packWARN2 and two categories which have
1403 * parent-child relationship. Even if it works now to
1404 * raise the warning if either is enabled, it wouldn't
1405 * necessarily do so in the future. We output (only) the
1406 * most dire warning*/
1407 if (! (flags & UTF8_CHECK_ONLY)) {
1408 if (ckWARN_d(WARN_UTF8)) {
1409 pack_warn = packWARN(WARN_UTF8);
1410 }
1411 else if (ckWARN_d(WARN_NON_UNICODE)) {
1412 pack_warn = packWARN(WARN_NON_UNICODE);
1413 }
1414 if (pack_warn) {
1415 message = Perl_form(aTHX_ "%s: %s (overflows)",
1416 malformed_text,
1417 _byte_dump_string(s0, send - s0));
1418 }
2b5e7bc2
KW
1419 }
1420 }
1421 }
1422 else if (possible_problems & UTF8_GOT_EMPTY) {
1423 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1424 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1425
1426 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1427
1428 /* This so-called malformation is now treated as a bug in
1429 * the caller. If you have nothing to decode, skip calling
1430 * this function */
1431 assert(0);
1432
2b5e7bc2
KW
1433 disallowed = TRUE;
1434 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1435 pack_warn = packWARN(WARN_UTF8);
1436 message = Perl_form(aTHX_ "%s (empty string)",
1437 malformed_text);
1438 }
1439 }
1440 }
1441 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1442 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1443 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1444
1445 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1446 disallowed = TRUE;
1447 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1448 pack_warn = packWARN(WARN_UTF8);
1449 message = Perl_form(aTHX_
1450 "%s: %s (unexpected continuation byte 0x%02x,"
1451 " with no preceding start byte)",
1452 malformed_text,
1453 _byte_dump_string(s0, 1), *s0);
1454 }
1455 }
1456 }
2b5e7bc2
KW
1457 else if (possible_problems & UTF8_GOT_SHORT) {
1458 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1459 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1460
1461 if (! (flags & UTF8_ALLOW_SHORT)) {
1462 disallowed = TRUE;
1463 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1464 pack_warn = packWARN(WARN_UTF8);
1465 message = Perl_form(aTHX_
9a6c9c81 1466 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1467 malformed_text,
1468 _byte_dump_string(s0, send - s0),
e308b348
KW
1469 (int)avail_len,
1470 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1471 (int)expectlen);
1472 }
1473 }
ba210ebe 1474
2b5e7bc2 1475 }
e308b348
KW
1476 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1477 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1478 *errors |= UTF8_GOT_NON_CONTINUATION;
1479
1480 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1481 disallowed = TRUE;
1482 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1483 pack_warn = packWARN(WARN_UTF8);
1484 message = Perl_form(aTHX_ "%s",
1485 unexpected_non_continuation_text(s0,
1486 send - s0,
1487 s - s0,
1488 (int) expectlen));
1489 }
1490 }
1491 }
2b5e7bc2
KW
1492 else if (possible_problems & UTF8_GOT_LONG) {
1493 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1494 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1495
94953955
KW
1496 if (flags & UTF8_ALLOW_LONG) {
1497
1498 /* We don't allow the actual overlong value, unless the
1499 * special extra bit is also set */
1500 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1501 & ~UTF8_ALLOW_LONG)))
1502 {
1503 uv = UNICODE_REPLACEMENT;
1504 }
1505 }
1506 else {
2b5e7bc2
KW
1507 disallowed = TRUE;
1508
1509 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1510 pack_warn = packWARN(WARN_UTF8);
1511
1512 /* These error types cause 'uv' to be something that
1513 * isn't what was intended, so can't use it in the
1514 * message. The other error types either can't
1515 * generate an overlong, or else the 'uv' is valid */
1516 if (orig_problems &
1517 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1518 {
1519 message = Perl_form(aTHX_
1520 "%s: %s (any UTF-8 sequence that starts"
1521 " with \"%s\" is overlong which can and"
1522 " should be represented with a"
1523 " different, shorter sequence)",
1524 malformed_text,
1525 _byte_dump_string(s0, send - s0),
1526 _byte_dump_string(s0, curlen));
1527 }
1528 else {
1529 U8 tmpbuf[UTF8_MAXBYTES+1];
1530 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1531 uv, 0);
1532 message = Perl_form(aTHX_
1533 "%s: %s (overlong; instead use %s to represent"
147e3846 1534 " U+%0*" UVXf ")",
2b5e7bc2
KW
1535 malformed_text,
1536 _byte_dump_string(s0, send - s0),
1537 _byte_dump_string(tmpbuf, e - tmpbuf),
1538 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1539 small code points */
1540 uv);
1541 }
1542 }
1543 }
1544 }
1545 else if (possible_problems & UTF8_GOT_SURROGATE) {
1546 possible_problems &= ~UTF8_GOT_SURROGATE;
1547
f9380377
KW
1548 if (flags & UTF8_WARN_SURROGATE) {
1549 *errors |= UTF8_GOT_SURROGATE;
1550
1551 if ( ! (flags & UTF8_CHECK_ONLY)
1552 && ckWARN_d(WARN_SURROGATE))
1553 {
2b5e7bc2
KW
1554 pack_warn = packWARN(WARN_SURROGATE);
1555
1556 /* These are the only errors that can occur with a
1557 * surrogate when the 'uv' isn't valid */
1558 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1559 message = Perl_form(aTHX_
1560 "UTF-16 surrogate (any UTF-8 sequence that"
1561 " starts with \"%s\" is for a surrogate)",
1562 _byte_dump_string(s0, curlen));
1563 }
1564 else {
1565 message = Perl_form(aTHX_
147e3846 1566 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1567 }
f9380377 1568 }
2b5e7bc2 1569 }
ba210ebe 1570
2b5e7bc2
KW
1571 if (flags & UTF8_DISALLOW_SURROGATE) {
1572 disallowed = TRUE;
f9380377 1573 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1574 }
1575 }
1576 else if (possible_problems & UTF8_GOT_SUPER) {
1577 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1578
f9380377
KW
1579 if (flags & UTF8_WARN_SUPER) {
1580 *errors |= UTF8_GOT_SUPER;
1581
1582 if ( ! (flags & UTF8_CHECK_ONLY)
1583 && ckWARN_d(WARN_NON_UNICODE))
1584 {
2b5e7bc2
KW
1585 pack_warn = packWARN(WARN_NON_UNICODE);
1586
1587 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1588 message = Perl_form(aTHX_
1589 "Any UTF-8 sequence that starts with"
1590 " \"%s\" is for a non-Unicode code point,"
1591 " may not be portable",
1592 _byte_dump_string(s0, curlen));
1593 }
1594 else {
1595 message = Perl_form(aTHX_
147e3846 1596 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1597 " Unicode, may not be portable",
1598 uv);
1599 }
f9380377 1600 }
2b5e7bc2 1601 }
ba210ebe 1602
2b5e7bc2
KW
1603 /* The maximum code point ever specified by a standard was
1604 * 2**31 - 1. Anything larger than that is a Perl extension
1605 * that very well may not be understood by other applications
1606 * (including earlier perl versions on EBCDIC platforms). We
1607 * test for these after the regular SUPER ones, and before
1608 * possibly bailing out, so that the slightly more dire warning
1609 * will override the regular one. */
1610 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1611 |UTF8_WARN_SUPER
1612 |UTF8_DISALLOW_ABOVE_31_BIT))
1613 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1614 && UNLIKELY(is_utf8_cp_above_31_bits(
1615 adjusted_s0,
1616 adjusted_send)))
1617 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1618 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1619 {
1620 if ( ! (flags & UTF8_CHECK_ONLY)
1621 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1622 && ckWARN_d(WARN_UTF8))
1623 {
1624 pack_warn = packWARN(WARN_UTF8);
1625
1626 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1627 message = Perl_form(aTHX_
1628 "Any UTF-8 sequence that starts with"
1629 " \"%s\" is for a non-Unicode code"
1630 " point, and is not portable",
1631 _byte_dump_string(s0, curlen));
1632 }
1633 else {
1634 message = Perl_form(aTHX_
147e3846 1635 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1636 " and not portable",
1637 uv);
1638 }
1639 }
1640
ddb65933
KW
1641 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1642 |UTF8_DISALLOW_ABOVE_31_BIT))
1643 {
f9380377
KW
1644 *errors |= UTF8_GOT_ABOVE_31_BIT;
1645
1646 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1647 disallowed = TRUE;
1648 }
2b5e7bc2
KW
1649 }
1650 }
eb83ed87 1651
2b5e7bc2 1652 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1653 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1654 disallowed = TRUE;
1655 }
eb83ed87 1656
2b5e7bc2
KW
1657 /* The deprecated warning overrides any non-deprecated one. If
1658 * there are other problems, a deprecation message is not
1659 * really helpful, so don't bother to raise it in that case.
1660 * This also keeps the code from having to handle the case
1661 * where 'uv' is not valid. */
1662 if ( ! (orig_problems
1663 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1664 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1665 && ckWARN_d(WARN_DEPRECATED))
1666 {
1667 message = Perl_form(aTHX_ cp_above_legal_max,
1668 uv, MAX_NON_DEPRECATED_CP);
1669 pack_warn = packWARN(WARN_DEPRECATED);
1670 }
1671 }
1672 else if (possible_problems & UTF8_GOT_NONCHAR) {
1673 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1674
f9380377
KW
1675 if (flags & UTF8_WARN_NONCHAR) {
1676 *errors |= UTF8_GOT_NONCHAR;
1677
1678 if ( ! (flags & UTF8_CHECK_ONLY)
1679 && ckWARN_d(WARN_NONCHAR))
1680 {
2b5e7bc2
KW
1681 /* The code above should have guaranteed that we don't
1682 * get here with errors other than overlong */
1683 assert (! (orig_problems
1684 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1685
1686 pack_warn = packWARN(WARN_NONCHAR);
1687 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1688 " U+%04" UVXf " is not recommended"
2b5e7bc2 1689 " for open interchange", uv);
f9380377 1690 }
2b5e7bc2 1691 }
5b311467 1692
2b5e7bc2
KW
1693 if (flags & UTF8_DISALLOW_NONCHAR) {
1694 disallowed = TRUE;
f9380377 1695 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1696 }
1697 } /* End of looking through the possible flags */
1698
1699 /* Display the message (if any) for the problem being handled in
1700 * this iteration of the loop */
1701 if (message) {
1702 if (PL_op)
1703 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1704 OP_DESC(PL_op));
1705 else
1706 Perl_warner(aTHX_ pack_warn, "%s", message);
1707 }
ddb65933 1708 } /* End of 'while (possible_problems)' */
a0dbb045 1709
2b5e7bc2
KW
1710 /* Since there was a possible problem, the returned length may need to
1711 * be changed from the one stored at the beginning of this function.
1712 * Instead of trying to figure out if that's needed, just do it. */
1713 if (retlen) {
1714 *retlen = curlen;
1715 }
a0dbb045 1716
2b5e7bc2
KW
1717 if (disallowed) {
1718 if (flags & UTF8_CHECK_ONLY && retlen) {
1719 *retlen = ((STRLEN) -1);
1720 }
1721 return 0;
1722 }
eb83ed87 1723 }
ba210ebe 1724
2b5e7bc2 1725 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1726}
1727
8e84507e 1728/*
ec5f19d0
KW
1729=for apidoc utf8_to_uvchr_buf
1730
1731Returns the native code point of the first character in the string C<s> which
1732is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1733C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1734
524080c4
KW
1735If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1736enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1737C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1738(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1739C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1740the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1741See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1742returned.
ec5f19d0 1743
760c7c2f
KW
1744Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1745unless those are turned off.
1746
ec5f19d0 1747=cut
52be2536
KW
1748
1749Also implemented as a macro in utf8.h
1750
ec5f19d0
KW
1751*/
1752
1753
1754UV
1755Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1756{
ec5f19d0
KW
1757 assert(s < send);
1758
1759 return utf8n_to_uvchr(s, send - s, retlen,
ddb65933 1760 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
ec5f19d0
KW
1761}
1762
52be2536
KW
1763/* This is marked as deprecated
1764 *
ec5f19d0
KW
1765=for apidoc utf8_to_uvuni_buf
1766
de69f3af
KW
1767Only in very rare circumstances should code need to be dealing in Unicode
1768(as opposed to native) code points. In those few cases, use
1769C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1770
1771Returns the Unicode (not-native) code point of the first character in the
1772string C<s> which
ec5f19d0
KW
1773is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1774C<retlen> will be set to the length, in bytes, of that character.
1775
524080c4
KW
1776If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1777enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1778NULL) to -1. If those warnings are off, the computed value if well-defined (or
1779the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1780is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1781next possible position in C<s> that could begin a non-malformed character.
de69f3af 1782See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1783
760c7c2f
KW
1784Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1785unless those are turned off.
1786
ec5f19d0
KW
1787=cut
1788*/
1789
1790UV
1791Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1792{
1793 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1794
1795 assert(send > s);
1796
5962d97e
KW
1797 /* Call the low level routine, asking for checks */
1798 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1799}
1800
b76347f2 1801/*
87cea99e 1802=for apidoc utf8_length
b76347f2
JH
1803
1804Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1805Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1806up past C<e>, croaks.
b76347f2
JH
1807
1808=cut
1809*/
1810
1811STRLEN
35a4481c 1812Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1813{
1814 STRLEN len = 0;
1815
7918f24d
NC
1816 PERL_ARGS_ASSERT_UTF8_LENGTH;
1817
8850bf83
JH
1818 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1819 * the bitops (especially ~) can create illegal UTF-8.
1820 * In other words: in Perl UTF-8 is not just for Unicode. */
1821
a3b680e6
AL
1822 if (e < s)
1823 goto warn_and_return;
b76347f2 1824 while (s < e) {
4cbf4130 1825 s += UTF8SKIP(s);
8e91ec7f
AV
1826 len++;
1827 }
1828
1829 if (e != s) {
1830 len--;
1831 warn_and_return:
9b387841
NC
1832 if (PL_op)
1833 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1834 "%s in %s", unees, OP_DESC(PL_op));
1835 else
61a12c31 1836 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1837 }
1838
1839 return len;
1840}
1841
b06226ff 1842/*
fed3ba5d
NC
1843=for apidoc bytes_cmp_utf8
1844
a1433954 1845Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1846sequence of characters (stored as UTF-8)
1847in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1848equal, -1 or -2 if the first string is less than the second string, +1 or +2
1849if the first string is greater than the second string.
1850
1851-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1852longer string. -2 or +2 is returned if
1853there was a difference between characters
fed3ba5d
NC
1854within the strings.
1855
1856=cut
1857*/
1858
1859int
1860Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1861{
1862 const U8 *const bend = b + blen;
1863 const U8 *const uend = u + ulen;
1864
1865 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1866
1867 while (b < bend && u < uend) {
1868 U8 c = *u++;
1869 if (!UTF8_IS_INVARIANT(c)) {
1870 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1871 if (u < uend) {
1872 U8 c1 = *u++;
1873 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1874 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1875 } else {
2b5e7bc2 1876 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1877 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1878 "%s %s%s",
7cf8d05d 1879 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1880 PL_op ? " in " : "",
1881 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1882 return -2;
1883 }
1884 } else {
1885 if (PL_op)
1886 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1887 "%s in %s", unees, OP_DESC(PL_op));
1888 else
61a12c31 1889 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1890 return -2; /* Really want to return undef :-) */
1891 }
1892 } else {
1893 return -2;
1894 }
1895 }
1896 if (*b != c) {
1897 return *b < c ? -2 : +2;
1898 }
1899 ++b;
1900 }
1901
1902 if (b == bend && u == uend)
1903 return 0;
1904
1905 return b < bend ? +1 : -1;
1906}
1907
1908/*
87cea99e 1909=for apidoc utf8_to_bytes
6940069f 1910
2bbc8d55 1911Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1912Unlike L</bytes_to_utf8>, this over-writes the original string, and
1913updates C<len> to contain the new length.
67e989fb 1914Returns zero on failure, setting C<len> to -1.
6940069f 1915
a1433954 1916If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1917
6940069f
GS
1918=cut
1919*/
1920
1921U8 *
37607a96 1922Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1923{
d4c19fe8
AL
1924 U8 * const save = s;
1925 U8 * const send = s + *len;
6940069f 1926 U8 *d;
246fae53 1927
7918f24d 1928 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1929 PERL_UNUSED_CONTEXT;
7918f24d 1930
1e54db1a 1931 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1932 while (s < send) {
d59937ca
KW
1933 if (! UTF8_IS_INVARIANT(*s)) {
1934 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1935 *len = ((STRLEN) -1);
1936 return 0;
1937 }
1938 s++;
dcad2880 1939 }
d59937ca 1940 s++;
246fae53 1941 }
dcad2880
JH
1942
1943 d = s = save;
6940069f 1944 while (s < send) {
80e0b38f
KW
1945 U8 c = *s++;
1946 if (! UTF8_IS_INVARIANT(c)) {
1947 /* Then it is two-byte encoded */
a62b247b 1948 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1949 s++;
1950 }
1951 *d++ = c;
6940069f
GS
1952 }
1953 *d = '\0';
246fae53 1954 *len = d - save;
6940069f
GS
1955 return save;
1956}
1957
1958/*
87cea99e 1959=for apidoc bytes_from_utf8
f9a63242 1960
2bbc8d55 1961Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1962Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1963the newly-created string, and updates C<len> to contain the new
1964length. Returns the original string if no conversion occurs, C<len>
72d33970 1965is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19660 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1967in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1968
37607a96
PK
1969=cut
1970*/
f9a63242
JH
1971
1972U8 *
e1ec3a88 1973Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1974{
f9a63242 1975 U8 *d;
e1ec3a88
AL
1976 const U8 *start = s;
1977 const U8 *send;
f9a63242
JH
1978 I32 count = 0;
1979
7918f24d 1980 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1981 PERL_UNUSED_CONTEXT;
f9a63242 1982 if (!*is_utf8)
73d840c0 1983 return (U8 *)start;
f9a63242 1984
1e54db1a 1985 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1986 for (send = s + *len; s < send;) {
d59937ca
KW
1987 if (! UTF8_IS_INVARIANT(*s)) {
1988 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1989 return (U8 *)start;
d59937ca
KW
1990 }
1991 count++;
1992 s++;
db42d148 1993 }
d59937ca 1994 s++;
f9a63242
JH
1995 }
1996
35da51f7 1997 *is_utf8 = FALSE;
f9a63242 1998
212542aa 1999 Newx(d, (*len) - count + 1, U8);
ef9edfd0 2000 s = start; start = d;
f9a63242
JH
2001 while (s < send) {
2002 U8 c = *s++;
1a91c45d 2003 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 2004 /* Then it is two-byte encoded */
a62b247b 2005 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 2006 s++;
c4d5f83a
NIS
2007 }
2008 *d++ = c;
f9a63242
JH
2009 }
2010 *d = '\0';
2011 *len = d - start;
73d840c0 2012 return (U8 *)start;
f9a63242
JH
2013}
2014
2015/*
87cea99e 2016=for apidoc bytes_to_utf8
6940069f 2017
ff97e5cf
KW
2018Converts a string C<s> of length C<len> bytes from the native encoding into
2019UTF-8.
6662521e 2020Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 2021reflect the new length in bytes.
6940069f 2022
75200dff 2023A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2024
2025If you want to convert to UTF-8 from encodings other than
2026the native (Latin1 or EBCDIC),
a1433954 2027see L</sv_recode_to_utf8>().
c9ada85f 2028
497711e7 2029=cut
6940069f
GS
2030*/
2031
c682ebef
FC
2032/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2033 likewise need duplication. */
2034
6940069f 2035U8*
35a4481c 2036Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 2037{
35a4481c 2038 const U8 * const send = s + (*len);
6940069f
GS
2039 U8 *d;
2040 U8 *dst;
7918f24d
NC
2041
2042 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2043 PERL_UNUSED_CONTEXT;
6940069f 2044
212542aa 2045 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2046 dst = d;
2047
2048 while (s < send) {
55d09dc8
KW
2049 append_utf8_from_native_byte(*s, &d);
2050 s++;
6940069f
GS
2051 }
2052 *d = '\0';
6662521e 2053 *len = d-dst;
6940069f
GS
2054 return dst;
2055}
2056
a0ed51b3 2057/*
dea0fc0b 2058 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2059 *
2060 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2061 * We optimize for native, for obvious reasons. */
2062
2063U8*
dea0fc0b 2064Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2065{
dea0fc0b
JH
2066 U8* pend;
2067 U8* dstart = d;
2068
7918f24d
NC
2069 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2070
dea0fc0b 2071 if (bytelen & 1)
147e3846 2072 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2073
2074 pend = p + bytelen;
2075
a0ed51b3 2076 while (p < pend) {
dea0fc0b
JH
2077 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2078 p += 2;
2d1545e5 2079 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2080 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2081 continue;
2082 }
56d37426
KW
2083 if (uv <= MAX_UTF8_TWO_BYTE) {
2084 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2085 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2086 continue;
2087 }
46956fad
KW
2088#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2089#define LAST_HIGH_SURROGATE 0xDBFF
2090#define FIRST_LOW_SURROGATE 0xDC00
2091#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2092
2093 /* This assumes that most uses will be in the first Unicode plane, not
2094 * needing surrogates */
2095 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2096 && uv <= UNICODE_SURROGATE_LAST))
2097 {
2098 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2099 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2100 }
2101 else {
01ea242b 2102 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2103 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2104 || UNLIKELY(low > LAST_LOW_SURROGATE))
2105 {
01ea242b 2106 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2107 }
2108 p += 2;
46956fad
KW
2109 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2110 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2111 }
a0ed51b3 2112 }
56d37426
KW
2113#ifdef EBCDIC
2114 d = uvoffuni_to_utf8_flags(d, uv, 0);
2115#else
a0ed51b3 2116 if (uv < 0x10000) {
eb160463
GS
2117 *d++ = (U8)(( uv >> 12) | 0xe0);
2118 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2119 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2120 continue;
2121 }
2122 else {
eb160463
GS
2123 *d++ = (U8)(( uv >> 18) | 0xf0);
2124 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2125 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2126 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2127 continue;
2128 }
56d37426 2129#endif
a0ed51b3 2130 }
dea0fc0b 2131 *newlen = d - dstart;
a0ed51b3
LW
2132 return d;
2133}
2134
2135/* Note: this one is slightly destructive of the source. */
2136
2137U8*
dea0fc0b 2138Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2139{
2140 U8* s = (U8*)p;
d4c19fe8 2141 U8* const send = s + bytelen;
7918f24d
NC
2142
2143 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2144
e0ea5e2d 2145 if (bytelen & 1)
147e3846 2146 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2147 (UV)bytelen);
2148
a0ed51b3 2149 while (s < send) {
d4c19fe8 2150 const U8 tmp = s[0];
a0ed51b3
LW
2151 s[0] = s[1];
2152 s[1] = tmp;
2153 s += 2;
2154 }
dea0fc0b 2155 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2156}
2157
922e8cb4
KW
2158bool
2159Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2160{
2161 U8 tmpbuf[UTF8_MAXBYTES+1];
2162 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2163 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2164}
2165
f9ae8fb6
JD
2166/* Internal function so we can deprecate the external one, and call
2167 this one from other deprecated functions in this file */
2168
f2645549
KW
2169bool
2170Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2171{
f2645549 2172 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2173
2174 if (*p == '_')
2175 return TRUE;
f25ce844 2176 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2177}
2178
5092f92a 2179bool
eba68aa0
KW
2180Perl__is_uni_perl_idcont(pTHX_ UV c)
2181{
2182 U8 tmpbuf[UTF8_MAXBYTES+1];
2183 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2184 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2185}
2186
2187bool
f91dcd13
KW
2188Perl__is_uni_perl_idstart(pTHX_ UV c)
2189{
2190 U8 tmpbuf[UTF8_MAXBYTES+1];
2191 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2192 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2193}
2194
3a4c58c9
KW
2195UV
2196Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2197{
2198 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2199 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2200 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2201 * either "SS" or "Ss". Which one to use is passed into the routine in
2202 * 'S_or_s' to avoid a test */
2203
2204 UV converted = toUPPER_LATIN1_MOD(c);
2205
2206 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2207
2208 assert(S_or_s == 'S' || S_or_s == 's');
2209
6f2d5cbc 2210 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2211 characters in this range */
3a4c58c9
KW
2212 *p = (U8) converted;
2213 *lenp = 1;
2214 return converted;
2215 }
2216
2217 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2218 * which it maps to one of them, so as to only have to have one check for
2219 * it in the main case */
2220 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2221 switch (c) {
2222 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2223 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2224 break;
2225 case MICRO_SIGN:
2226 converted = GREEK_CAPITAL_LETTER_MU;
2227 break;
79e064b9
KW
2228#if UNICODE_MAJOR_VERSION > 2 \
2229 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2230 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2231 case LATIN_SMALL_LETTER_SHARP_S:
2232 *(p)++ = 'S';
2233 *p = S_or_s;
2234 *lenp = 2;
2235 return 'S';
79e064b9 2236#endif
3a4c58c9
KW
2237 default:
2238 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2239 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2240 }
2241 }
2242
2243 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2244 *p = UTF8_TWO_BYTE_LO(converted);
2245 *lenp = 2;
2246
2247 return converted;
2248}
2249
50bda2c3
KW
2250/* Call the function to convert a UTF-8 encoded character to the specified case.
2251 * Note that there may be more than one character in the result.
2252 * INP is a pointer to the first byte of the input character
2253 * OUTP will be set to the first byte of the string of changed characters. It
2254 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2255 * LENP will be set to the length in bytes of the string of changed characters
2256 *
2257 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2258#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2259#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2260#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2261
b9992569
KW
2262/* This additionally has the input parameter 'specials', which if non-zero will
2263 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2264 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2265#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2266
84afefe6
JH
2267UV
2268Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2269{
a1433954
KW
2270 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2271 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2272 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2273 * the changed version may be longer than the original character.
2274 *
2275 * The ordinal of the first character of the changed version is returned
2276 * (but note, as explained above, that there may be more.) */
2277
7918f24d
NC
2278 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2279
3a4c58c9
KW
2280 if (c < 256) {
2281 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2282 }
2283
0ebc6274 2284 uvchr_to_utf8(p, c);
b9992569 2285 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2286}
2287
84afefe6
JH
2288UV
2289Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2290{
7918f24d
NC
2291 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2292
3a4c58c9
KW
2293 if (c < 256) {
2294 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2295 }
2296
0ebc6274 2297 uvchr_to_utf8(p, c);
b9992569 2298 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2299}
2300
afc16117 2301STATIC U8
eaf412bf 2302S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2303{
2304 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2305 * those, converting the result to UTF-8. Since the result is always just
a1433954 2306 * one character, we allow <p> to be NULL */
afc16117
KW
2307
2308 U8 converted = toLOWER_LATIN1(c);
2309
eaf412bf
KW
2310 PERL_UNUSED_ARG(dummy);
2311
afc16117 2312 if (p != NULL) {
6f2d5cbc 2313 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2314 *p = converted;
2315 *lenp = 1;
2316 }
2317 else {
430c9760
KW
2318 /* Result is known to always be < 256, so can use the EIGHT_BIT
2319 * macros */
2320 *p = UTF8_EIGHT_BIT_HI(converted);
2321 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2322 *lenp = 2;
2323 }
2324 }
2325 return converted;
2326}
2327
84afefe6
JH
2328UV
2329Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2330{
7918f24d
NC
2331 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2332
afc16117 2333 if (c < 256) {
eaf412bf 2334 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2335 }
2336
afc16117 2337 uvchr_to_utf8(p, c);
b9992569 2338 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2339}
2340
84afefe6 2341UV
51910141 2342Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2343{
51910141 2344 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2345 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2346 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2347 *
2348 * Not to be used for locale folds
51910141 2349 */
f673fad4 2350
a1dde8de
KW
2351 UV converted;
2352
2353 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2354 PERL_UNUSED_CONTEXT;
a1dde8de 2355
1ca267a5
KW
2356 assert (! (flags & FOLD_FLAGS_LOCALE));
2357
659a7c2d 2358 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2359 converted = GREEK_SMALL_LETTER_MU;
2360 }
9b63e895
KW
2361#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2362 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2363 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2364 else if ( (flags & FOLD_FLAGS_FULL)
2365 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2366 {
1ca267a5
KW
2367 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2368 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2369 * under those circumstances. */
2370 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2371 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2372 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2373 p, *lenp, U8);
2374 return LATIN_SMALL_LETTER_LONG_S;
2375 }
2376 else {
4f489194
KW
2377 *(p)++ = 's';
2378 *p = 's';
2379 *lenp = 2;
2380 return 's';
1ca267a5 2381 }
a1dde8de 2382 }
9b63e895 2383#endif
a1dde8de
KW
2384 else { /* In this range the fold of all other characters is their lower
2385 case */
2386 converted = toLOWER_LATIN1(c);
2387 }
2388
6f2d5cbc 2389 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2390 *p = (U8) converted;
2391 *lenp = 1;
2392 }
2393 else {
2394 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2395 *p = UTF8_TWO_BYTE_LO(converted);
2396 *lenp = 2;
2397 }
2398
2399 return converted;
2400}
2401
2402UV
31f05a37 2403Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2404{
4b593389 2405
a0270393
KW
2406 /* Not currently externally documented, and subject to change
2407 * <flags> bits meanings:
2408 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2409 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2410 * locale are to be used.
a0270393
KW
2411 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2412 */
4b593389 2413
36bb2ab6 2414 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2415
780fcc9f
KW
2416 if (flags & FOLD_FLAGS_LOCALE) {
2417 /* Treat a UTF-8 locale as not being in locale at all */
2418 if (IN_UTF8_CTYPE_LOCALE) {
2419 flags &= ~FOLD_FLAGS_LOCALE;
2420 }
2421 else {
2422 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2423 goto needs_full_generality;
780fcc9f 2424 }
31f05a37
KW
2425 }
2426
a1dde8de 2427 if (c < 256) {
e7b7ac46 2428 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2429 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2430 }
2431
2f306ab9 2432 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2433 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2434 uvchr_to_utf8(p, c);
b9992569 2435 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
2436 }
2437 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2438 the special flags. */
2439 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2440
2441 needs_full_generality:
a0270393 2442 uvchr_to_utf8(utf8_c, c);
445bf929 2443 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 2444 }
84afefe6
JH
2445}
2446
26483009 2447PERL_STATIC_INLINE bool
5141f98e 2448S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2449 const char *const swashname, SV* const invlist)
bde6a22d 2450{
ea317ccb
KW
2451 /* returns a boolean giving whether or not the UTF8-encoded character that
2452 * starts at <p> is in the swash indicated by <swashname>. <swash>
2453 * contains a pointer to where the swash indicated by <swashname>
2454 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2455 * look at <*swash> and only generate a swash if it is not null. <invlist>
2456 * is NULL or an inversion list that defines the swash. If not null, it
2457 * saves time during initialization of the swash.
ea317ccb
KW
2458 *
2459 * Note that it is assumed that the buffer length of <p> is enough to
2460 * contain all the bytes that comprise the character. Thus, <*p> should
2461 * have been checked before this call for mal-formedness enough to assure
2462 * that. */
2463
7918f24d
NC
2464 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2465
492a624f 2466 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2467 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2468 * as far as there being enough bytes available in it to accommodate the
2469 * character without reading beyond the end, and pass that number on to the
2470 * validating routine */
6302f837 2471 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94
KW
2472 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2473 0,
2474 1 /* Die */ );
2475 NOT_REACHED; /* NOTREACHED */
28123549 2476 }
86ae6e94 2477
87367d5f
KW
2478 if (!*swash) {
2479 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2480 *swash = _core_swash_init("utf8",
2481
2482 /* Only use the name if there is no inversion
2483 * list; otherwise will go out to disk */
2484 (invlist) ? "" : swashname,
2485
2486 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2487 }
28123549 2488
bde6a22d
NC
2489 return swash_fetch(*swash, p, TRUE) != 0;
2490}
2491
da8c1a98
KW
2492PERL_STATIC_INLINE bool
2493S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2494 const char *const swashname, SV* const invlist)
2495{
2496 /* returns a boolean giving whether or not the UTF8-encoded character that
2497 * starts at <p>, and extending no further than <e - 1> is in the swash
2498 * indicated by <swashname>. <swash> contains a pointer to where the swash
2499 * indicated by <swashname> is to be stored; which this routine will do, so
2500 * that future calls will look at <*swash> and only generate a swash if it
2501 * is not null. <invlist> is NULL or an inversion list that defines the
2502 * swash. If not null, it saves time during initialization of the swash.
2503 */
2504
2505 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2506
2507 if (! isUTF8_CHAR(p, e)) {
2508 _force_out_malformed_utf8_message(p, e, 0, 1);
2509 NOT_REACHED; /* NOTREACHED */
2510 }
2511
2512 if (!*swash) {
2513 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2514 *swash = _core_swash_init("utf8",
2515
2516 /* Only use the name if there is no inversion
2517 * list; otherwise will go out to disk */
2518 (invlist) ? "" : swashname,
2519
2520 &PL_sv_undef, 1, 0, invlist, &flags);
2521 }
2522
2523 return swash_fetch(*swash, p, TRUE) != 0;
2524}
2525
bde6a22d 2526bool
922e8cb4
KW
2527Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2528{
922e8cb4
KW
2529 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2530
2531 assert(classnum < _FIRST_NON_SWASH_CC);
2532
f25ce844
KW
2533 return is_utf8_common(p,
2534 &PL_utf8_swash_ptrs[classnum],
2535 swash_property_names[classnum],
2536 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
2537}
2538
2539bool
da8c1a98
KW
2540Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2541 const U8 * const e)
2542{
2543 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2544
2545 assert(classnum < _FIRST_NON_SWASH_CC);
2546
2547 return is_utf8_common_with_len(p,
2548 e,
2549 &PL_utf8_swash_ptrs[classnum],
2550 swash_property_names[classnum],
2551 PL_XPosix_ptrs[classnum]);
2552}
2553
2554bool
f2645549 2555Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 2556{
f2645549 2557 SV* invlist = NULL;
7918f24d 2558
f2645549 2559 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 2560
f2645549
KW
2561 if (! PL_utf8_perl_idstart) {
2562 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2563 }
60071a22 2564 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
2565}
2566
2567bool
da8c1a98
KW
2568Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2569{
2570 SV* invlist = NULL;
2571
2572 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2573
2574 if (! PL_utf8_perl_idstart) {
2575 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2576 }
2577 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2578 "_Perl_IDStart", invlist);
2579}
2580
2581bool
f2645549 2582Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2583{
f2645549 2584 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2585
2586 if (*p == '_')
2587 return TRUE;
f25ce844 2588 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2589}
2590
2591bool
eba68aa0
KW
2592Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2593{
b24b43f7 2594 SV* invlist = NULL;
eba68aa0
KW
2595
2596 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2597
b24b43f7
KW
2598 if (! PL_utf8_perl_idcont) {
2599 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2600 }
60071a22 2601 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
2602}
2603
eba68aa0 2604bool
da8c1a98
KW
2605Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2606{
2607 SV* invlist = NULL;
2608
2609 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2610
2611 if (! PL_utf8_perl_idcont) {
2612 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2613 }
2614 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2615 "_Perl_IDCont", invlist);
2616}
2617
2618bool
f2645549 2619Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2620{
f2645549 2621 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2622
f25ce844 2623 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2624}
2625
2626bool
f2645549 2627Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2628{
f2645549 2629 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2630
f25ce844 2631 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2632}
2633
2634bool
7dbf68d2
KW
2635Perl__is_utf8_mark(pTHX_ const U8 *p)
2636{
7dbf68d2
KW
2637 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2638
f25ce844 2639 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2640}
2641
6b5c0936 2642/*
87cea99e 2643=for apidoc to_utf8_case
6b5c0936 2644
9da1e7cb
KW
2645Instead use the appropriate one of L</toUPPER_utf8>,
2646L</toTITLE_utf8>,
2647L</toLOWER_utf8>,
2648or L</toFOLD_utf8>.
2649
6fae5207 2650C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2651the character that is being converted. This routine assumes that the character
2652at C<p> is well-formed.
6b5c0936 2653
6fae5207
KW
2654C<ustrp> is a pointer to the character buffer to put the
2655conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2656of the result.
2657
6fae5207 2658C<swashp> is a pointer to the swash to use.
6b5c0936 2659
a1433954 2660Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2661and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2662but not always, a multicharacter mapping), is tried first.
6b5c0936 2663
4a8240a3
KW
2664C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2665any special mappings; C<""> means to use the special mappings. Values other
2666than these two are treated as the name of the hash containing the special
2667mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2668
796b6530
KW
2669C<normal> is a string like C<"ToLower"> which means the swash
2670C<%utf8::ToLower>.
0134edef 2671
760c7c2f
KW
2672Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2673unless those are turned off.
2674
0134edef 2675=cut */
6b5c0936 2676
2104c8d9 2677UV
9a957fbc
AL
2678Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2679 SV **swashp, const char *normal, const char *special)
a0ed51b3 2680{
b9992569
KW
2681 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2682
2683 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2684}
2685
2686 /* change namve uv1 to 'from' */
6a4a25f4 2687STATIC UV
b9992569
KW
2688S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2689 SV **swashp, const char *normal, const char *special)
2690{
0134edef 2691 STRLEN len = 0;
7918f24d 2692
b9992569 2693 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2694
36eaa811
KW
2695 /* For code points that don't change case, we already know that the output
2696 * of this function is the unchanged input, so we can skip doing look-ups
2697 * for them. Unfortunately the case-changing code points are scattered
2698 * around. But there are some long consecutive ranges where there are no
2699 * case changing code points. By adding tests, we can eliminate the lookup
2700 * for all the ones in such ranges. This is currently done here only for
2701 * just a few cases where the scripts are in common use in modern commerce
2702 * (and scripts adjacent to those which can be included without additional
2703 * tests). */
2704
2705 if (uv1 >= 0x0590) {
2706 /* This keeps from needing further processing the code points most
2707 * likely to be used in the following non-cased scripts: Hebrew,
2708 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2709 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2710 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2711 if (uv1 < 0x10A0) {
2712 goto cases_to_self;
2713 }
2714
2715 /* The following largish code point ranges also don't have case
2716 * changes, but khw didn't think they warranted extra tests to speed
2717 * them up (which would slightly slow down everything else above them):
2718 * 1100..139F Hangul Jamo, Ethiopic
2719 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2720 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2721 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2722 * Combining Diacritical Marks Extended, Balinese,
2723 * Sundanese, Batak, Lepcha, Ol Chiki
2724 * 2000..206F General Punctuation
2725 */
2726
2727 if (uv1 >= 0x2D30) {
2728
2729 /* This keeps the from needing further processing the code points
2730 * most likely to be used in the following non-cased major scripts:
2731 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2732 *
2733 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2734 * event that Unicode eventually allocates the unused block as of
2735 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2736 * that the test suite will start having failures to alert you
2737 * should that happen) */
2738 if (uv1 < 0xA640) {
2739 goto cases_to_self;
2740 }
2741
2742 if (uv1 >= 0xAC00) {
2743 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2744 if (ckWARN_d(WARN_SURROGATE)) {
2745 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2746 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2747 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2748 }
2749 goto cases_to_self;
2750 }
36eaa811
KW
2751
2752 /* AC00..FAFF Catches Hangul syllables and private use, plus
2753 * some others */
2754 if (uv1 < 0xFB00) {
2755 goto cases_to_self;
2756
2757 }
2758
5af9bc97
KW
2759 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2760 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2761 && ckWARN_d(WARN_DEPRECATED))
2762 {
2763 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2764 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2765 }
2766 if (ckWARN_d(WARN_NON_UNICODE)) {
2767 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2768 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2769 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2770 }
2771 goto cases_to_self;
2772 }
3bfc1e70
KW
2773#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2774 if (UNLIKELY(uv1
2775 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2776 {
2777
2778 /* As of this writing, this means we avoid swash creation
2779 * for anything beyond low Plane 1 */
2780 goto cases_to_self;
2781 }
2782#endif
36eaa811
KW
2783 }
2784 }
9ae3ac1a 2785
36eaa811
KW
2786 /* Note that non-characters are perfectly legal, so no warning should
2787 * be given. There are so few of them, that it isn't worth the extra
2788 * tests to avoid swash creation */
9ae3ac1a
KW
2789 }
2790
0134edef 2791 if (!*swashp) /* load on-demand */
5ab9d2ef 2792 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2793
a6f87d8c 2794 if (special) {
0134edef 2795 /* It might be "special" (sometimes, but not always,
2a37f04d 2796 * a multicharacter mapping) */
4a8240a3 2797 HV *hv = NULL;
b08cf34e
JH
2798 SV **svp;
2799
4a8240a3
KW
2800 /* If passed in the specials name, use that; otherwise use any
2801 * given in the swash */
2802 if (*special != '\0') {
2803 hv = get_hv(special, 0);
2804 }
2805 else {
2806 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2807 if (svp) {
2808 hv = MUTABLE_HV(SvRV(*svp));
2809 }
2810 }
2811
176fe009 2812 if (hv
5f560d8a 2813 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2814 && (*svp))
2815 {
cfd0369c 2816 const char *s;
47654450 2817
cfd0369c 2818 s = SvPV_const(*svp, len);
47654450 2819 if (len == 1)
f4cd282c 2820 /* EIGHTBIT */
c80e42f3 2821 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2822 else {
d2dcd0fb 2823 Copy(s, ustrp, len, U8);
29e98929 2824 }
983ffd37 2825 }
0134edef
JH
2826 }
2827
2828 if (!len && *swashp) {
4a4088c4 2829 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2830
0134edef
JH
2831 if (uv2) {
2832 /* It was "normal" (a single character mapping). */
f4cd282c 2833 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2834 }
2835 }
1feea2c7 2836
cbe07460
KW
2837 if (len) {
2838 if (lenp) {
2839 *lenp = len;
2840 }
2841 return valid_utf8_to_uvchr(ustrp, 0);
2842 }
2843
2844 /* Here, there was no mapping defined, which means that the code point maps
2845 * to itself. Return the inputs */
e24dfe9c 2846 cases_to_self:
bfdf22ec 2847 len = UTF8SKIP(p);
ca9fab46
KW
2848 if (p != ustrp) { /* Don't copy onto itself */
2849 Copy(p, ustrp, len, U8);
2850 }
0134edef 2851
2a37f04d
JH
2852 if (lenp)
2853 *lenp = len;
2854
f4cd282c 2855 return uv1;
cbe07460 2856
a0ed51b3
LW
2857}
2858
051a06d4 2859STATIC UV
357aadde 2860S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2861{
4a4088c4 2862 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2863 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2864 * result contains a character that crosses the 255/256 boundary, disallow
2865 * the change, and return the original code point. See L<perlfunc/lc> for
2866 * why;
051a06d4 2867 *
a1433954
KW
2868 * p points to the original string whose case was changed; assumed
2869 * by this routine to be well-formed
051a06d4
KW
2870 * result the code point of the first character in the changed-case string
2871 * ustrp points to the changed-case string (<result> represents its first char)
2872 * lenp points to the length of <ustrp> */
2873
2874 UV original; /* To store the first code point of <p> */
2875
2876 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2877
a4f12ed7 2878 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2879
2880 /* We know immediately if the first character in the string crosses the
2881 * boundary, so can skip */
2882 if (result > 255) {
2883
2884 /* Look at every character in the result; if any cross the
2885 * boundary, the whole thing is disallowed */
2886 U8* s = ustrp + UTF8SKIP(ustrp);
2887 U8* e = ustrp + *lenp;
2888 while (s < e) {
a4f12ed7 2889 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2890 goto bad_crossing;
2891 }
2892 s += UTF8SKIP(s);
2893 }
2894
613abc6d
KW
2895 /* Here, no characters crossed, result is ok as-is, but we warn. */
2896 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2897 return result;
2898 }
2899
7b52d656 2900 bad_crossing:
051a06d4
KW
2901
2902 /* Failed, have to return the original */
4b88fb76 2903 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2904
2905 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2906 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2907 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2908 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2909 OP_DESC(PL_op),
ab0b796c
KW
2910 original,
2911 original);
051a06d4
KW
2912 Copy(p, ustrp, *lenp, char);
2913 return original;
2914}
2915
eaf412bf
KW
2916/* The process for changing the case is essentially the same for the four case
2917 * change types, except there are complications for folding. Otherwise the
2918 * difference is only which case to change to. To make sure that they all do
2919 * the same thing, the bodies of the functions are extracted out into the
2920 * following two macros. The functions are written with the same variable
2921 * names, and these are known and used inside these macros. It would be
2922 * better, of course, to have inline functions to do it, but since different
2923 * macros are called, depending on which case is being changed to, this is not
2924 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2925 * function can start with the common start macro, then finish with its special
2926 * handling; while the other three cases can just use the common end macro.
2927 *
2928 * The algorithm is to use the proper (passed in) macro or function to change
2929 * the case for code points that are below 256. The macro is used if using
2930 * locale rules for the case change; the function if not. If the code point is
2931 * above 255, it is computed from the input UTF-8, and another macro is called
2932 * to do the conversion. If necessary, the output is converted to UTF-8. If
2933 * using a locale, we have to check that the change did not cross the 255/256
2934 * boundary, see check_locale_boundary_crossing() for further details.
2935 *
2936 * The macros are split with the correct case change for the below-256 case
2937 * stored into 'result', and in the middle of an else clause for the above-255
2938 * case. At that point in the 'else', 'result' is not the final result, but is
2939 * the input code point calculated from the UTF-8. The fold code needs to
2940 * realize all this and take it from there.
2941 *
2942 * If you read the two macros as sequential, it's easier to understand what's
2943 * going on. */
2944#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2945 L1_func_extra_param) \
2946 if (flags & (locale_flags)) { \
2947 /* Treat a UTF-8 locale as not being in locale at all */ \
2948 if (IN_UTF8_CTYPE_LOCALE) { \
2949 flags &= ~(locale_flags); \
2950 } \
2951 else { \
2952 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2953 } \
2954 } \
2955 \
2956 if (UTF8_IS_INVARIANT(*p)) { \
2957 if (flags & (locale_flags)) { \
2958 result = LC_L1_change_macro(*p); \
2959 } \
2960 else { \
2961 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2962 } \
2963 } \
2964 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2965 if (flags & (locale_flags)) { \
2966 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2967 *(p+1))); \
2968 } \
2969 else { \
2970 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2971 ustrp, lenp, L1_func_extra_param); \
2972 } \
2973 } \
2974 else { /* malformed UTF-8 */ \
2975 result = valid_utf8_to_uvchr(p, NULL); \
2976
2977#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2978 result = change_macro(result, p, ustrp, lenp); \
2979 \
2980 if (flags & (locale_flags)) { \
2981 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2982 } \
2983 return result; \
2984 } \
2985 \
2986 /* Here, used locale rules. Convert back to UTF-8 */ \
2987 if (UTF8_IS_INVARIANT(result)) { \
2988 *ustrp = (U8) result; \
2989 *lenp = 1; \
2990 } \
2991 else { \
2992 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2993 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2994 *lenp = 2; \
2995 } \
2996 \
2997 return result;
2998
d3e79532 2999/*
87cea99e 3000=for apidoc to_utf8_upper
d3e79532 3001
1f607577 3002Instead use L</toUPPER_utf8>.
a1433954 3003
d3e79532
JH
3004=cut */
3005
051a06d4 3006/* Not currently externally documented, and subject to change:
31f05a37
KW
3007 * <flags> is set iff iff the rules from the current underlying locale are to
3008 * be used. */
051a06d4 3009
2104c8d9 3010UV
31f05a37 3011Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3012{
051a06d4
KW
3013 UV result;
3014
3015 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3016
eaf412bf
KW
3017 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3018 /* 2nd char of uc(U+DF) is 'S' */
3019 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3020 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3021}
a0ed51b3 3022
d3e79532 3023/*
87cea99e 3024=for apidoc to_utf8_title
d3e79532 3025
1f607577 3026Instead use L</toTITLE_utf8>.
a1433954 3027
d3e79532
JH
3028=cut */
3029
051a06d4 3030/* Not currently externally documented, and subject to change:
31f05a37
KW
3031 * <flags> is set iff the rules from the current underlying locale are to be
3032 * used. Since titlecase is not defined in POSIX, for other than a
3033 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3034 */
051a06d4 3035
983ffd37 3036UV
31f05a37 3037Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 3038{
051a06d4
KW
3039 UV result;
3040
3041 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3042
eaf412bf
KW
3043 /* 2nd char of ucfirst(U+DF) is 's' */
3044 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3045 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3046}
3047
d3e79532 3048/*
87cea99e 3049=for apidoc to_utf8_lower
d3e79532 3050
1f607577 3051Instead use L</toLOWER_utf8>.
a1433954 3052
d3e79532
JH
3053=cut */
3054
051a06d4 3055/* Not currently externally documented, and subject to change:
31f05a37
KW
3056 * <flags> is set iff iff the rules from the current underlying locale are to
3057 * be used.
3058 */
051a06d4 3059
2104c8d9 3060UV
31f05a37 3061Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3062{
051a06d4
KW
3063 UV result;
3064
051a06d4 3065 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3066
eaf412bf
KW
3067 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3068 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3069}
3070
d3e79532 3071/*
87cea99e 3072=for apidoc to_utf8_fold
d3e79532 3073
1f607577 3074Instead use L</toFOLD_utf8>.
a1433954 3075
d3e79532
JH
3076=cut */
3077
051a06d4
KW
3078/* Not currently externally documented, and subject to change,
3079 * in <flags>
31f05a37
KW
3080 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3081 * locale are to be used.
051a06d4
KW
3082 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3083 * otherwise simple folds
a0270393
KW
3084 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3085 * prohibited
445bf929 3086 */
36bb2ab6 3087
b4e400f9 3088UV
445bf929 3089Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3090{
051a06d4
KW
3091 UV result;
3092
36bb2ab6 3093 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3094
a0270393
KW
3095 /* These are mutually exclusive */
3096 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3097
50ba90ff
KW
3098 assert(p != ustrp); /* Otherwise overwrites */
3099
eaf412bf
KW
3100 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3101 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3102
eaf412bf 3103 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3104
1ca267a5
KW
3105 if (flags & FOLD_FLAGS_LOCALE) {
3106
76f2ffcd 3107# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3108 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3109
3110# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3111# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3112
3113 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3114
538e84ed
KW
3115 /* Special case these two characters, as what normally gets
3116 * returned under locale doesn't work */
76f2ffcd
KW
3117 if (UTF8SKIP(p) == cap_sharp_s_len
3118 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3119 {
ab0b796c
KW
3120 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3121 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3122 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3123 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3124 goto return_long_s;
3125 }
0766489e
KW
3126 else
3127#endif
3128 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3129 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3130 {
ab0b796c
KW
3131 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3132 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3133 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3134 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3135 goto return_ligature_st;
3136 }
74894415
KW
3137
3138#if UNICODE_MAJOR_VERSION == 3 \
3139 && UNICODE_DOT_VERSION == 0 \
3140 && UNICODE_DOT_DOT_VERSION == 1
3141# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3142
3143 /* And special case this on this Unicode version only, for the same
3144 * reaons the other two are special cased. They would cross the
3145 * 255/256 boundary which is forbidden under /l, and so the code
3146 * wouldn't catch that they are equivalent (which they are only in
3147 * this release) */
3148 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3149 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3150 {
3151 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3152 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3153 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3154 "resolved to \"\\x{0131}\".");
3155 goto return_dotless_i;
3156 }
3157#endif
3158
357aadde 3159 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3160 }
a0270393
KW
3161 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3162 return result;
3163 }
3164 else {
4a4088c4 3165 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3166 * character above the ASCII range, and the result should not
3167 * contain an ASCII character. */
a0270393
KW
3168
3169 UV original; /* To store the first code point of <p> */
3170
3171 /* Look at every character in the result; if any cross the
3172 * boundary, the whole thing is disallowed */
3173 U8* s = ustrp;
3174 U8* e = ustrp + *lenp;
3175 while (s < e) {
3176 if (isASCII(*s)) {
3177 /* Crossed, have to return the original */
3178 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3179
9fc2026f 3180 /* But in these instances, there is an alternative we can
1ca267a5 3181 * return that is valid */
0766489e
KW
3182 if (original == LATIN_SMALL_LETTER_SHARP_S
3183#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3184 || original == LATIN_CAPITAL_LETTER_SHARP_S
3185#endif
3186 ) {
1ca267a5
KW
3187 goto return_long_s;
3188 }
9fc2026f
KW
3189 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3190 goto return_ligature_st;
3191 }
74894415
KW
3192#if UNICODE_MAJOR_VERSION == 3 \
3193 && UNICODE_DOT_VERSION == 0 \
3194 && UNICODE_DOT_DOT_VERSION == 1
3195
3196 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3197 goto return_dotless_i;
3198 }
3199#endif
a0270393
KW
3200 Copy(p, ustrp, *lenp, char);
3201 return original;
3202 }
3203 s += UTF8SKIP(s);
3204 }
051a06d4 3205
a0270393
KW
3206 /* Here, no characters crossed, result is ok as-is */
3207 return result;
3208 }
051a06d4
KW
3209 }
3210
4a4088c4 3211 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3212 if (UTF8_IS_INVARIANT(result)) {
3213 *ustrp = (U8) result;
3214 *lenp = 1;
3215 }
3216 else {
62cb07ea
KW
3217 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3218 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3219 *lenp = 2;
3220 }
3221
051a06d4 3222 return result;
1ca267a5
KW
3223
3224 return_long_s:
3225 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3226 * folds to a string of two of these characters. By returning this
3227 * instead, then, e.g.,
3228 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3229 * works. */
3230
3231 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3232 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3233 ustrp, *lenp, U8);
3234 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3235
3236 return_ligature_st:
3237 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3238 * have the other one fold to it */
3239
3240 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3241 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3242 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3243
3244#if UNICODE_MAJOR_VERSION == 3 \
3245 && UNICODE_DOT_VERSION == 0 \
3246 && UNICODE_DOT_DOT_VERSION == 1
3247
3248 return_dotless_i:
3249 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3250 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3251 return LATIN_SMALL_LETTER_DOTLESS_I;
3252
3253#endif
3254
a0ed51b3
LW
3255}
3256
711a919c 3257/* Note:
f90a9a02 3258 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3259 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3260 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3261 */
c4a5db0c 3262
a0ed51b3 3263SV*
7fc63493 3264Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3265{
c4a5db0c
KW
3266 PERL_ARGS_ASSERT_SWASH_INIT;
3267
3268 /* Returns a copy of a swash initiated by the called function. This is the
3269 * public interface, and returning a copy prevents others from doing
3270 * mischief on the original */
3271
5d3d13d1 3272 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3273}
3274
3275SV*
5d3d13d1 3276Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3277{
2c1f00b9
YO
3278
3279 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3280 * use the following define */
3281
3282#define CORE_SWASH_INIT_RETURN(x) \
3283 PL_curpm= old_PL_curpm; \
3284 return x
3285
c4a5db0c 3286 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3287 * by calling utf8_heavy.pl in the general case. The returned value may be
3288 * the swash's inversion list instead if the input parameters allow it.
3289 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3290 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3291 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3292 *
3293 * This interface should only be used by functions that won't destroy or
3294 * adversely change the swash, as doing so affects all other uses of the
3295 * swash in the program; the general public should use 'Perl_swash_init'
3296 * instead.
3297 *
3298 * pkg is the name of the package that <name> should be in.
3299 * name is the name of the swash to find. Typically it is a Unicode
3300 * property name, including user-defined ones
3301 * listsv is a string to initialize the swash with. It must be of the form
3302 * documented as the subroutine return value in
3303 * L<perlunicode/User-Defined Character Properties>
3304 * minbits is the number of bits required to represent each data element.
3305 * It is '1' for binary properties.
3306 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3307 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3308 * flags_p if non-NULL is the address of various input and output flag bits
3309 * to the routine, as follows: ('I' means is input to the routine;
3310 * 'O' means output from the routine. Only flags marked O are
3311 * meaningful on return.)
3312 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3313 * came from a user-defined property. (I O)
5d3d13d1
KW
3314 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3315 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3316 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3317 * return of an inversion list instead of a swash hash if this routine
3318 * thinks that would result in faster execution of swash_fetch() later
3319 * on. (I)
9a53f6cf
KW
3320 *
3321 * Thus there are three possible inputs to find the swash: <name>,
3322 * <listsv>, and <invlist>. At least one must be specified. The result
3323 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3324 * actions can intersect, etc. what <name> gives. To avoid going out to
3325 * disk at all, <invlist> should specify completely what the swash should
3326 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3327 *
3328 * <invlist> is only valid for binary properties */
c4a5db0c 3329
2c1f00b9
YO
3330 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3331
c4a5db0c 3332 SV* retval = &PL_sv_undef;
83199d38 3333 HV* swash_hv = NULL;
87367d5f
KW
3334 const int invlist_swash_boundary =
3335 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3336 ? 512 /* Based on some benchmarking, but not extensive, see commit
3337 message */
3338 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3339
3340 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3341 assert(! invlist || minbits == 1);
3342
2c1f00b9
YO
3343 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3344 that triggered the swash init and the swash init perl logic itself.
3345 See perl #122747 */
3346
9a53f6cf
KW
3347 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3348 * so */
3349 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3350 dSP;
3351 const size_t pkg_len = strlen(pkg);
3352 const size_t name_len = strlen(name);
3353 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3354 SV* errsv_save;
3355 GV *method;
3356
3357 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3358
3359 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3360 ENTER;
69794297 3361 SAVEHINTS();
2782061f 3362 save_re_context();
650f067c
JL
3363 /* We might get here via a subroutine signature which uses a utf8
3364 * parameter name, at which point PL_subname will have been set
3365 * but not yet used. */
3366 save_item(PL_subname);
69794297
KW
3367 if (PL_parser && PL_parser->error_count)
3368 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3369 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3370 if (!method) { /* demand load UTF-8 */
69794297 3371 ENTER;
db2c6cb3
FC
3372 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3373 GvSV(PL_errgv) = NULL;
1a419e6b 3374#ifndef NO_TAINT_SUPPORT
69794297
KW
3375 /* It is assumed that callers of this routine are not passing in
3376 * any user derived data. */
2782061f
DM
3377 /* Need to do this after save_re_context() as it will set
3378 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3379 * in Perl_magic_get). Even line to create errsv_save can turn on
3380 * PL_tainted. */
284167a5
SM
3381 SAVEBOOL(TAINT_get);
3382 TAINT_NOT;
3383#endif
69794297
KW
3384 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3385 NULL);
eed484f9 3386 {
db2c6cb3
FC
3387 /* Not ERRSV, as there is no need to vivify a scalar we are
3388 about to discard. */
3389 SV * const errsv = GvSV(PL_errgv);
3390 if (!SvTRUE(errsv)) {
3391 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3392 SvREFCNT_dec(errsv);
3393 }
eed484f9 3394 }
69794297
KW
3395 LEAVE;
3396 }
3397 SPAGAIN;
3398 PUSHMARK(SP);
3399 EXTEND(SP,5);
3400 mPUSHp(pkg, pkg_len);
3401 mPUSHp(name, name_len);
3402 PUSHs(listsv);
3403 mPUSHi(minbits);
3404 mPUSHi(none);
3405 PUTBACK;
db2c6cb3
FC
3406 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3407 GvSV(PL_errgv) = NULL;
69794297
KW
3408 /* If we already have a pointer to the method, no need to use
3409 * call_method() to repeat the lookup. */
c41800a8
KW
3410 if (method
3411 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3412 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3413 {
3414 retval = *PL_stack_sp--;
3415 SvREFCNT_inc(retval);
3416 }
eed484f9 3417 {
db2c6cb3
FC
3418 /* Not ERRSV. See above. */
3419 SV * const errsv = GvSV(PL_errgv);
3420 if (!SvTRUE(errsv)) {
3421 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3422 SvREFCNT_dec(errsv);
3423 }
eed484f9 3424 }
ce3b816e 3425 LEAVE;
69794297
KW
3426 POPSTACK;
3427 if (IN_PERL_COMPILETIME) {
3428 CopHINTS_set(PL_curcop, PL_hints);
3429 }
3430 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3431 if (SvPOK(retval)) {
69794297
KW
3432
3433 /* If caller wants to handle missing properties, let them */
5d3d13d1 3434 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3435 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3436 }
3437 Perl_croak(aTHX_
147e3846 3438 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3439 SVfARG(retval));
a25b5927 3440 NOT_REACHED; /* NOTREACHED */
d95e4a00 3441 }
69794297 3442 }
9a53f6cf 3443 } /* End of calling the module to find the swash */
36eb48b4 3444
83199d38
KW
3445 /* If this operation fetched a swash, and we will need it later, get it */
3446 if (retval != &PL_sv_undef
3447 && (minbits == 1 || (flags_p
3448 && ! (*flags_p
3449 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3450 {
3451 swash_hv = MUTABLE_HV(SvRV(retval));
3452
3453 /* If we don't already know that there is a user-defined component to
3454 * this swash, and the user has indicated they wish to know if there is
3455 * one (by passing <flags_p>), find out */
3456 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3457 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3458 if (user_defined && SvUV(*user_defined)) {
3459 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3460 }
3461 }
3462 }
3463
36eb48b4
KW
3464 /* Make sure there is an inversion list for binary properties */
3465 if (minbits == 1) {
3466 SV** swash_invlistsvp = NULL;
3467 SV* swash_invlist = NULL;
9a53f6cf 3468 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3469 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3470 an unclaimed reference count */
36eb48b4 3471
9a53f6cf 3472 /* If this operation fetched a swash, get its already existing
83199d38 3473 * inversion list, or create one for it */
36eb48b4 3474
83199d38 3475 if (swash_hv) {
5c9f4bd2 3476 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3477 if (swash_invlistsvp) {
3478 swash_invlist = *swash_invlistsvp;
3479 invlist_in_swash_is_valid = TRUE;
3480 }
3481 else {
36eb48b4 3482 swash_invlist = _swash_to_invlist(retval);
02c85471 3483 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3484 }
3485 }
3486
3487 /* If an inversion list was passed in, have to include it */
3488 if (invlist) {
3489
3490 /* Any fetched swash will by now have an inversion list in it;
3491 * otherwise <swash_invlist> will be NULL, indicating that we
3492 * didn't fetch a swash */
3493 if (swash_invlist) {
3494
3495 /* Add the passed-in inversion list, which invalidates the one
3496 * already stored in the swash */
3497 invlist_in_swash_is_valid = FALSE;
eee4c920 3498 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3499 _invlist_union(invlist, swash_invlist, &swash_invlist);
3500 }
3501 else {
3502
87367d5f
KW
3503 /* Here, there is no swash already. Set up a minimal one, if
3504 * we are going to return a swash */
3505 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3506 swash_hv = newHV();
4aca0fe6 3507 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3508 }
9a53f6cf
KW
3509 swash_invlist = invlist;
3510 }
9a53f6cf
KW
3511 }
3512
3513 /* Here, we have computed the union of all the passed-in data. It may
3514 * be that there was an inversion list in the swash which didn't get
538e84ed 3515 * touched; otherwise save the computed one */
87367d5f
KW
3516 if (! invlist_in_swash_is_valid
3517 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3518 {
5c9f4bd2 3519 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3520 {
3521 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3522 }
cc34d8c5
FC
3523 /* We just stole a reference count. */
3524 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3525 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3526 }
87367d5f 3527
ee3222e3 3528 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3529 SvREADONLY_on(swash_invlist);
3530
c41800a8 3531 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3532 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3533 SvREFCNT_dec(retval);
02c85471
FC
3534 if (!swash_invlist_unclaimed)
3535 SvREFCNT_inc_simple_void_NN(swash_invlist);
3536 retval = newRV_noinc(swash_invlist);
87367d5f 3537 }
36eb48b4
KW
3538 }
3539
2c1f00b9
YO
3540 CORE_SWASH_INIT_RETURN(retval);
3541#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3542}
3543
035d37be
JH
3544
3545/* This API is wrong for special case conversions since we may need to
3546 * return several Unicode characters for a single Unicode character
3547 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3548 * the lower-level routine, and it is similarly broken for returning
38684baa 3549 * multiple values. --jhi
b9992569 3550 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3551/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3552
3553/* Note:
3554 * Returns the value of property/mapping C<swash> for the first character
3555 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3556 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3557 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3558 *
3559 * A "swash" is a hash which contains initially the keys/values set up by
3560 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3561 * property for all possible code points. Things are stored in a compact form
3562 * (see utf8_heavy.pl) so that calculation is required to find the actual
3563 * property value for a given code point. As code points are looked up, new
3564 * key/value pairs are added to the hash, so that the calculation doesn't have
3565 * to ever be re-done. Further, each calculation is done, not just for the
3566 * desired one, but for a whole block of code points adjacent to that one.
3567 * For binary properties on ASCII machines, the block is usually for 64 code
3568 * points, starting with a code point evenly divisible by 64. Thus if the
3569 * property value for code point 257 is requested, the code goes out and
3570 * calculates the property values for all 64 code points between 256 and 319,
3571 * and stores these as a single 64-bit long bit vector, called a "swatch",
3572 * under the key for code point 256. The key is the UTF-8 encoding for code
3573 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3574 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3575 * for code point 258 is then requested, this code realizes that it would be
3576 * stored under the key for 256, and would find that value and extract the
3577 * relevant bit, offset from 256.
3578 *
3579 * Non-binary properties are stored in as many bits as necessary to represent
3580 * their values (32 currently, though the code is more general than that), not
fc273927 3581 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3582 * vector that encompasses the property values for all code points whose UTF-8
3583 * representations are represented by the key. That is, for all code points
3584 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3585 * bytes of that.
680c470c 3586 */
a0ed51b3 3587UV
680c470c 3588Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3589{
ef8f7699 3590 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3591 U32 klen;
3592 U32 off;
9b56a019 3593 STRLEN slen = 0;
7d85a32c 3594 STRLEN needents;
cfd0369c 3595 const U8 *tmps = NULL;
979f2922 3596 SV *swatch;
08fb1ac5 3597 const U8 c = *ptr;
3568d838 3598
7918f24d
NC
3599 PERL_ARGS_ASSERT_SWASH_FETCH;
3600
87367d5f
KW
3601 /* If it really isn't a hash, it isn't really swash; must be an inversion
3602 * list */
3603 if (SvTYPE(hv) != SVt_PVHV) {
3604 return _invlist_contains_cp((SV*)hv,
3605 (do_utf8)
3606 ? valid_utf8_to_uvchr(ptr, NULL)
3607 : c);
3608 }
3609
08fb1ac5
KW
3610 /* We store the values in a "swatch" which is a vec() value in a swash
3611 * hash. Code points 0-255 are a single vec() stored with key length
3612 * (klen) 0. All other code points have a UTF-8 representation
3613 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3614 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3615 * length for them is the length of the encoded char - 1. ptr[klen] is the
3616 * final byte in the sequence representing the character */
3617 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3618 klen = 0;
3619 needents = 256;
3620 off = c;
3568d838 3621 }
08fb1ac5
KW
3622 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3623 klen = 0;
3624 needents = 256;
a62b247b 3625 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3626 }
3627 else {
08fb1ac5
KW
3628 klen = UTF8SKIP(ptr) - 1;
3629
3630 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3631 * the vec is the final byte in the sequence. (In EBCDIC this is
3632 * converted to I8 to get consecutive values.) To help you visualize
3633 * all this:
3634 * Straight 1047 After final byte
3635 * UTF-8 UTF-EBCDIC I8 transform
3636 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3637 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3638 * ...
3639 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3640 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3641 * ...
3642 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3643 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3644 * ...
3645 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3646 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3647 * ...
3648 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3649 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3650 *
3651 * (There are no discontinuities in the elided (...) entries.)
3652 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3653 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3654 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3655 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3656 * index into the vec() swatch (after subtracting 0x80, which we
3657 * actually do with an '&').
3658 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3659 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3660 * dicontinuities which go away by transforming it into I8, and we
3661 * effectively subtract 0xA0 to get the index. */
979f2922 3662 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3663 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3664 }
7d85a32c 3665
a0ed51b3 3666 /*
4a4088c4 3667 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
3668 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3669 * it's nothing to sniff at.) Pity we usually come through at least
3670 * two function calls to get here...
3671 *
3672 * NB: this code assumes that swatches are never modified, once generated!
3673 */
3674
3568d838 3675 if (hv == PL_last_swash_hv &&
a0ed51b3 3676 klen == PL_last_swash_klen &&
27da23d5 3677 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3678 {
3679 tmps = PL_last_swash_tmps;
3680 slen = PL_last_swash_slen;
3681 }
3682 else {
3683 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3684 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3685
b0e3252e 3686 /* If not cached, generate it via swatch_get */
979f2922 3687 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3688 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3689 {
3690 if (klen) {
3691 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3692 swatch = swatch_get(swash,
3693 code_point & ~((UV)needents - 1),
3694 needents);
3695 }
3696 else { /* For the first 256 code points, the swatch has a key of
3697 length 0 */
3698 swatch = swatch_get(swash, 0, needents);
3699 }
979f2922 3700
923e4eb5 3701 if (IN_PERL_COMPILETIME)
623e6609 3702 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3703
979f2922 3704 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3705
979f2922
ST
3706 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3707 || (slen << 3) < needents)
5637ef5b 3708 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
147e3846 3709 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
5637ef5b 3710 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3711 }
3712
3713 PL_last_swash_hv = hv;
16d8f38a 3714 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3715 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3716 /* FIXME change interpvar.h? */
3717 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3718 PL_last_swash_slen = slen;
3719 if (klen)
3720 Copy(ptr, PL_last_swash_key, klen, U8);
3721 }
3722
9faf8d75 3723 switch ((int)((slen << 3) / needents)) {