This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Return REPLACEMENT for UTF-8 empty malformation
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
147e3846 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf;
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
81 SAVESPTR(PL_dowarn);
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
8ee1cdcb
KW
119#define HANDLE_UNICODE_SURROGATE(uv, flags) \
120 STMT_START { \
121 if (flags & UNICODE_WARN_SURROGATE) { \
122 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
147e3846 123 "UTF-16 surrogate U+%04" UVXf, uv); \
8ee1cdcb
KW
124 } \
125 if (flags & UNICODE_DISALLOW_SURROGATE) { \
126 return NULL; \
127 } \
128 } STMT_END;
129
130#define HANDLE_UNICODE_NONCHAR(uv, flags) \
131 STMT_START { \
132 if (flags & UNICODE_WARN_NONCHAR) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
147e3846 134 "Unicode non-character U+%04" UVXf " is not " \
8ee1cdcb
KW
135 "recommended for open interchange", uv); \
136 } \
137 if (flags & UNICODE_DISALLOW_NONCHAR) { \
138 return NULL; \
139 } \
140 } STMT_END;
141
ba6ed43c
KW
142/* Use shorter names internally in this file */
143#define SHIFT UTF_ACCUMULATION_SHIFT
144#undef MARK
145#define MARK UTF_CONTINUATION_MARK
146#define MASK UTF_CONTINUATION_MASK
147
dfe13c55 148U8 *
378516de 149Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 150{
378516de 151 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 152
2d1545e5 153 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 154 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
155 return d;
156 }
facc1dc2 157
3ea68d71 158 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
159 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
160 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
161 return d;
162 }
d9432125 163
ba6ed43c
KW
164 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
165 * below, the 16 is for start bytes E0-EF (which are all the possible ones
166 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
167 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
168 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
169 * 0x800-0xFFFF on ASCII */
170 if (uv < (16 * (1U << (2 * SHIFT)))) {
171 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
172 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
173 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
174
175#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
176 aren't tested here */
177 /* The most likely code points in this range are below the surrogates.
178 * Do an extra test to quickly exclude those. */
179 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
180 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
181 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
182 {
8ee1cdcb
KW
183 HANDLE_UNICODE_NONCHAR(uv, flags);
184 }
185 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
186 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 187 }
ba6ed43c
KW
188 }
189#endif
190 return d;
191 }
192
193 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
194 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
195 * happen starting with 4-byte characters on ASCII platforms. We unify the
196 * code for these with EBCDIC, even though some of them require 5-bytes on
197 * those, because khw believes the code saving is worth the very slight
198 * performance hit on these high EBCDIC code points. */
199
200 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
201 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
202 && ckWARN_d(WARN_DEPRECATED))
203 {
204 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
205 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
206 }
207 if ( (flags & UNICODE_WARN_SUPER)
208 || ( UNICODE_IS_ABOVE_31_BIT(uv)
209 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
210 {
211 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
212
213 /* Choose the more dire applicable warning */
214 (UNICODE_IS_ABOVE_31_BIT(uv))
147e3846
KW
215 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
216 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
a5bf80e0
KW
217 uv);
218 }
219 if (flags & UNICODE_DISALLOW_SUPER
220 || ( UNICODE_IS_ABOVE_31_BIT(uv)
221 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
222 {
223 return NULL;
224 }
225 }
ba6ed43c
KW
226 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
227 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 228 }
d9432125 229
ba6ed43c
KW
230 /* Test for and handle 4-byte result. In the test immediately below, the
231 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
232 * characters). The 3 is for 3 continuation bytes; these each contribute
233 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
234 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
235 * 0x1_0000-0x1F_FFFF on ASCII */
236 if (uv < (8 * (1U << (3 * SHIFT)))) {
237 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
238 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
239 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
240 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
241
242#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
243 characters. The end-plane non-characters for EBCDIC were
244 handled just above */
245 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
246 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 247 }
ba6ed43c
KW
248 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
249 HANDLE_UNICODE_SURROGATE(uv, flags);
250 }
251#endif
252
253 return d;
254 }
255
256 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
257 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
258 * format. The unrolled version above turns out to not save all that much
259 * time, and at these high code points (well above the legal Unicode range
260 * on ASCII platforms, and well above anything in common use in EBCDIC),
261 * khw believes that less code outweighs slight performance gains. */
262
d9432125 263 {
5aaebcb3 264 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
265 U8 *p = d+len-1;
266 while (p > d) {
4c8cd605 267 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
268 uv >>= UTF_ACCUMULATION_SHIFT;
269 }
4c8cd605 270 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
271 return d+len;
272 }
a0ed51b3 273}
a5bf80e0 274
646ca15d 275/*
07693fe6
KW
276=for apidoc uvchr_to_utf8
277
bcb1a2d4 278Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 279of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
280C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
281the byte after the end of the new character. In other words,
07693fe6
KW
282
283 d = uvchr_to_utf8(d, uv);
284
285is the recommended wide native character-aware way of saying
286
287 *(d++) = uv;
288
760c7c2f
KW
289This function accepts any UV as input, but very high code points (above
290C<IV_MAX> on the platform) will raise a deprecation warning. This is
291typically 0x7FFF_FFFF in a 32-bit word.
292
293It is possible to forbid or warn on non-Unicode code points, or those that may
294be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 295
07693fe6
KW
296=cut
297*/
298
de69f3af
KW
299/* This is also a macro */
300PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
301
07693fe6
KW
302U8 *
303Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
304{
de69f3af 305 return uvchr_to_utf8(d, uv);
07693fe6
KW
306}
307
de69f3af
KW
308/*
309=for apidoc uvchr_to_utf8_flags
310
311Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 312of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
313C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
314the byte after the end of the new character. In other words,
de69f3af
KW
315
316 d = uvchr_to_utf8_flags(d, uv, flags);
317
318or, in most cases,
319
320 d = uvchr_to_utf8_flags(d, uv, 0);
321
322This is the Unicode-aware way of saying
323
324 *(d++) = uv;
325
760c7c2f
KW
326If C<flags> is 0, this function accepts any UV as input, but very high code
327points (above C<IV_MAX> for the platform) will raise a deprecation warning.
328This is typically 0x7FFF_FFFF in a 32-bit word.
329
330Specifying C<flags> can further restrict what is allowed and not warned on, as
331follows:
de69f3af 332
796b6530 333If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
334the function will raise a warning, provided UTF8 warnings are enabled. If
335instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
336NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 337
760c7c2f
KW
338Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
339affect how the function handles a Unicode non-character.
93e6dbd6 340
760c7c2f
KW
341And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
342affect the handling of code points that are above the Unicode maximum of
3430x10FFFF. Languages other than Perl may not be able to accept files that
344contain these.
93e6dbd6
KW
345
346The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
347the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
348three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
349allowed inputs to the strict UTF-8 traditionally defined by Unicode.
350Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
351C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
352above-Unicode and surrogate flags, but not the non-character ones, as
353defined in
354L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
355See L<perlunicode/Noncharacter code points>.
93e6dbd6 356
ab8e6d41
KW
357Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
358so using them is more problematic than other above-Unicode code points. Perl
359invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
360likely that non-Perl languages will not be able to read files that contain
361these that written by the perl interpreter; nor would Perl understand files
362written by something that uses a different extension. For these reasons, there
363is a separate set of flags that can warn and/or disallow these extremely high
364code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
365C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
366are entirely independent from the deprecation warning for code points above
367C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
368code point that needs more than 31 bits to represent. When that happens,
369effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
37032-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
371above-Unicode code points, including these, as malformations; and
372C<UNICODE_WARN_SUPER> warns on these.)
373
374On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
375extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
376than on ASCII. Prior to that, code points 2**31 and higher were simply
377unrepresentable, and a different, incompatible method was used to represent
378code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
379and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
380platforms, warning and disallowing 2**31 and higher.
de69f3af 381
de69f3af
KW
382=cut
383*/
384
385/* This is also a macro */
386PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
387
07693fe6
KW
388U8 *
389Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
390{
de69f3af 391 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
392}
393
83dc0f42
KW
394PERL_STATIC_INLINE bool
395S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
396{
397 /* Returns TRUE if the first code point represented by the Perl-extended-
398 * UTF-8-encoded string starting at 's', and looking no further than 'e -
399 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
400 *
401 * The function handles the case where the input bytes do not include all
402 * the ones necessary to represent a full character. That is, they may be
403 * the intial bytes of the representation of a code point, but possibly
404 * the final ones necessary for the complete representation may be beyond
405 * 'e - 1'.
406 *
407 * The function assumes that the sequence is well-formed UTF-8 as far as it
408 * goes, and is for a UTF-8 variant code point. If the sequence is
409 * incomplete, the function returns FALSE if there is any well-formed
410 * UTF-8 byte sequence that can complete it in such a way that a code point
411 * < 2**31 is produced; otherwise it returns TRUE.
412 *
413 * Getting this exactly right is slightly tricky, and has to be done in
414 * several places in this file, so is centralized here. It is based on the
415 * following table:
416 *
417 * U+7FFFFFFF (2 ** 31 - 1)
418 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
419 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
420 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
421 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
422 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
423 * U+80000000 (2 ** 31):
424 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
425 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
426 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
428 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
429 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
430 */
431
432#ifdef EBCDIC
433
37086697
KW
434 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
435 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
436 const STRLEN prefix_len = sizeof(prefix) - 1;
437 const STRLEN len = e - s;
f880f78a 438 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
439
440#else
441
442 PERL_UNUSED_ARG(e);
443
444#endif
445
446 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
447
448 assert(! UTF8_IS_INVARIANT(*s));
449
450#ifndef EBCDIC
451
452 /* Technically, a start byte of FE can be for a code point that fits into
453 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
454 * malformation. */
455 return (*s >= 0xFE);
456
457#else
458
459 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
460 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
461 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
462 * bits. */
463 if (*s != 0xFE || len == 1) {
464 return FALSE;
465 }
466
467 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
468 * \x41 and \x42. */
469 return cBOOL(memGT(s + 1, prefix, cmp_len));
470
471#endif
472
473}
474
12a4bed3
KW
475PERL_STATIC_INLINE bool
476S_does_utf8_overflow(const U8 * const s, const U8 * e)
477{
478 const U8 *x;
479 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
480
c551bb01
KW
481#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
482
483 const STRLEN len = e - s;
484
485#endif
486
12a4bed3
KW
487 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
488 * platform, that is if it represents a code point larger than the highest
489 * representable code point. (For ASCII platforms, we could use memcmp()
490 * because we don't have to convert each byte to I8, but it's very rare
491 * input indeed that would approach overflow, so the loop below will likely
492 * only get executed once.
493 *
494 * 'e' must not be beyond a full character. If it is less than a full
495 * character, the function returns FALSE if there is any input beyond 'e'
496 * that could result in a non-overflowing code point */
497
498 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
499 assert(s <= e && s + UTF8SKIP(s) >= e);
500
501#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
502
503 /* On 32 bit ASCII machines, many overlongs that start with FF don't
504 * overflow */
505
c551bb01 506 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
507 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
508 return memGE(s, max_32_bit_overlong,
c551bb01 509 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
510 }
511
512#endif
12a4bed3
KW
513
514 for (x = s; x < e; x++, y++) {
515
516 /* If this byte is larger than the corresponding highest UTF-8 byte, it
517 * overflows */
518 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
519 return TRUE;
520 }
521
522 /* If not the same as this byte, it must be smaller, doesn't overflow */
523 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
524 return FALSE;
525 }
526 }
527
528 /* Got to the end and all bytes are the same. If the input is a whole
529 * character, it doesn't overflow. And if it is a partial character,
530 * there's not enough information to tell, so assume doesn't overflow */
531 return FALSE;
532}
533
534PERL_STATIC_INLINE bool
535S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
536{
537 /* Overlongs can occur whenever the number of continuation bytes
538 * changes. That means whenever the number of leading 1 bits in a start
539 * byte increases from the next lower start byte. That happens for start
540 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
541 * illegal start bytes have already been excluded, so don't need to be
542 * tested here;
543 * ASCII platforms: C0, C1
544 * EBCDIC platforms C0, C1, C2, C3, C4, E0
545 *
546 * At least a second byte is required to determine if other sequences will
547 * be an overlong. */
548
549 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
550 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
551
552 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
553 assert(len > 1 && UTF8_IS_START(*s));
554
555 /* Each platform has overlongs after the start bytes given above (expressed
556 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
557 * the logic is the same, except the E0 overlong has already been excluded
558 * on EBCDIC platforms. The values below were found by manually
559 * inspecting the UTF-8 patterns. See the tables in utf8.h and
560 * utfebcdic.h. */
561
562# ifdef EBCDIC
563# define F0_ABOVE_OVERLONG 0xB0
564# define F8_ABOVE_OVERLONG 0xA8
565# define FC_ABOVE_OVERLONG 0xA4
566# define FE_ABOVE_OVERLONG 0xA2
567# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
568 /* I8(0xfe) is FF */
569# else
570
571 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
572 return TRUE;
573 }
574
575# define F0_ABOVE_OVERLONG 0x90
576# define F8_ABOVE_OVERLONG 0x88
577# define FC_ABOVE_OVERLONG 0x84
578# define FE_ABOVE_OVERLONG 0x82
579# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
580# endif
581
582
583 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
584 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
585 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
586 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
587 {
588 return TRUE;
589 }
590
b0b342d4
KW
591 /* Check for the FF overlong */
592 return isFF_OVERLONG(s, len);
593}
594
595PERL_STATIC_INLINE bool
596S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
597{
598 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
599
600 /* Check for the FF overlong. This happens only if all these bytes match;
601 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 602 * utfebcdic.h. */
12a4bed3 603
b0b342d4
KW
604 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
605 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
606 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
607}
608
609#undef F0_ABOVE_OVERLONG
610#undef F8_ABOVE_OVERLONG
611#undef FC_ABOVE_OVERLONG
612#undef FE_ABOVE_OVERLONG
613#undef FF_OVERLONG_PREFIX
614
35f8c9bd 615STRLEN
edc2c47a 616Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 617{
2b479609 618 STRLEN len;
12a4bed3 619 const U8 *x;
35f8c9bd 620
2b479609
KW
621 /* A helper function that should not be called directly.
622 *
623 * This function returns non-zero if the string beginning at 's' and
624 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
625 * code point; otherwise it returns 0. The examination stops after the
626 * first code point in 's' is validated, not looking at the rest of the
627 * input. If 'e' is such that there are not enough bytes to represent a
628 * complete code point, this function will return non-zero anyway, if the
629 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
630 * excluded by 'flags'.
631 *
632 * A non-zero return gives the number of bytes required to represent the
633 * code point. Be aware that if the input is for a partial character, the
634 * return will be larger than 'e - s'.
635 *
636 * This function assumes that the code point represented is UTF-8 variant.
637 * The caller should have excluded this possibility before calling this
638 * function.
639 *
640 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
641 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
642 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
643 * disallowed by the flags. If the input is only for a partial character,
644 * the function will return non-zero if there is any sequence of
645 * well-formed UTF-8 that, when appended to the input sequence, could
646 * result in an allowed code point; otherwise it returns 0. Non characters
647 * cannot be determined based on partial character input. But many of the
648 * other excluded types can be determined with just the first one or two
649 * bytes.
650 *
651 */
652
653 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
654
655 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
656 |UTF8_DISALLOW_ABOVE_31_BIT)));
657 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 658
2b479609 659 /* A variant char must begin with a start byte */
35f8c9bd
KW
660 if (UNLIKELY(! UTF8_IS_START(*s))) {
661 return 0;
662 }
663
edc2c47a
KW
664 /* Examine a maximum of a single whole code point */
665 if (e - s > UTF8SKIP(s)) {
666 e = s + UTF8SKIP(s);
667 }
668
2b479609
KW
669 len = e - s;
670
671 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
672 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 673
2b479609
KW
674 /* The code below is derived from this table. Keep in mind that legal
675 * continuation bytes range between \x80..\xBF for UTF-8, and
676 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
677 * Hence, we don't have to test the upper edge because if any of those
678 * are encountered, the sequence is malformed, and will fail elsewhere
679 * in this function.
680 * UTF-8 UTF-EBCDIC I8
681 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
682 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
683 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
684 *
685 */
686
687#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
688# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 689# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 690
19794540
KW
691# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
692 /* B6 and B7 */ \
693 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
694#else
695# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
696# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
697# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
698#endif
699
700 if ( (flags & UTF8_DISALLOW_SUPER)
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) {
702 return 0; /* Above Unicode */
703 }
704
705 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
706 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
707 {
708 return 0; /* Above 31 bits */
709 }
710
711 if (len > 1) {
712 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
713
714 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 715 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
716 {
717 return 0; /* Above Unicode */
718 }
719
720 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 721 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
722 {
723 return 0; /* Surrogate */
724 }
725
726 if ( (flags & UTF8_DISALLOW_NONCHAR)
727 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
728 {
729 return 0; /* Noncharacter code point */
730 }
731 }
732 }
733
734 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
735 for (x = s + 1; x < e; x++) {
736 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
737 return 0;
738 }
739 }
740
af13dd8a 741 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
742 * overlong. */
743 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
744 return 0;
af13dd8a
KW
745 }
746
12a4bed3
KW
747 /* And finally, that the code point represented fits in a word on this
748 * platform */
749 if (does_utf8_overflow(s, e)) {
750 return 0;
35f8c9bd
KW
751 }
752
2b479609 753 return UTF8SKIP(s);
35f8c9bd
KW
754}
755
7cf8d05d
KW
756STATIC char *
757S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
758{
759 /* Returns a mortalized C string that is a displayable copy of the 'len'
760 * bytes starting at 's', each in a \xXY format. */
761
762 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
763 trailing NUL */
764 const U8 * const e = s + len;
765 char * output;
766 char * d;
767
768 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
769
770 Newx(output, output_len, char);
771 SAVEFREEPV(output);
772
773 d = output;
774 for (; s < e; s++) {
775 const unsigned high_nibble = (*s & 0xF0) >> 4;
776 const unsigned low_nibble = (*s & 0x0F);
777
778 *d++ = '\\';
779 *d++ = 'x';
780
781 if (high_nibble < 10) {
782 *d++ = high_nibble + '0';
783 }
784 else {
785 *d++ = high_nibble - 10 + 'a';
786 }
787
788 if (low_nibble < 10) {
789 *d++ = low_nibble + '0';
790 }
791 else {
792 *d++ = low_nibble - 10 + 'a';
793 }
794 }
795
796 *d = '\0';
797 return output;
798}
799
806547a7 800PERL_STATIC_INLINE char *
7cf8d05d
KW
801S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
802
803 /* How many bytes to print */
3cc6a05e 804 STRLEN print_len,
7cf8d05d
KW
805
806 /* Which one is the non-continuation */
807 const STRLEN non_cont_byte_pos,
808
809 /* How many bytes should there be? */
810 const STRLEN expect_len)
806547a7
KW
811{
812 /* Return the malformation warning text for an unexpected continuation
813 * byte. */
814
7cf8d05d 815 const char * const where = (non_cont_byte_pos == 1)
806547a7 816 ? "immediately"
7cf8d05d
KW
817 : Perl_form(aTHX_ "%d bytes",
818 (int) non_cont_byte_pos);
3cc6a05e 819 unsigned int i;
806547a7
KW
820
821 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
822
7cf8d05d
KW
823 /* We don't need to pass this parameter, but since it has already been
824 * calculated, it's likely faster to pass it; verify under DEBUGGING */
825 assert(expect_len == UTF8SKIP(s));
826
3cc6a05e
KW
827 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
828 * length that is larger than is actually available in the buffer. If we
829 * print all the bytes based on that length, we will read past the buffer
830 * end. Often, the strings are NUL terminated, so to lower the chances of
831 * this happening, print the malformed bytes only up through any NUL. */
832 for (i = 1; i < print_len; i++) {
833 if (*(s + i) == '\0') {
834 print_len = i + 1; /* +1 gets the NUL printed */
835 break;
836 }
837 }
838
7cf8d05d
KW
839 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
840 " %s after start byte 0x%02x; need %d bytes, got %d)",
841 malformed_text,
842 _byte_dump_string(s, print_len),
843 *(s + non_cont_byte_pos),
844 where,
845 *s,
846 (int) expect_len,
847 (int) non_cont_byte_pos);
806547a7
KW
848}
849
35f8c9bd
KW
850/*
851
de69f3af 852=for apidoc utf8n_to_uvchr
378516de
KW
853
854THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 855Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 856
9041c2e3 857Bottom level UTF-8 decode routine.
de69f3af 858Returns the native code point value of the first character in the string C<s>,
746afd53
KW
859which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
860C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
861the length, in bytes, of that character.
949cf498
KW
862
863The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
864well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
865causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
866is the next possible position in C<s> that could begin a non-malformed
867character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
868is raised. Some UTF-8 input sequences may contain multiple malformations.
869This function tries to find every possible one in each call, so multiple
870warnings can be raised for each sequence.
949cf498
KW
871
872Various ALLOW flags can be set in C<flags> to allow (and not warn on)
873individual types of malformations, such as the sequence being overlong (that
874is, when there is a shorter sequence that can express the same code point;
875overlong sequences are expressly forbidden in the UTF-8 standard due to
876potential security issues). Another malformation example is the first byte of
877a character not being a legal first byte. See F<utf8.h> for the list of such
5a48568d
KW
878flags. For allowed overlong sequences, the computed code point is returned;
879for all other allowed malformations, the Unicode REPLACEMENT CHARACTER is
880returned.
949cf498 881
796b6530 882The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
883flags) malformation is found. If this flag is set, the routine assumes that
884the caller will raise a warning, and this function will silently just set
d088425d
KW
885C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
886
75200dff 887Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 888character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
889in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
890be set to 1. To disambiguate, upon a zero return, see if the first byte of
891C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 892error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
893
894Certain code points are considered problematic. These are Unicode surrogates,
746afd53 895Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 896By default these are considered regular code points, but certain situations
ecc1615f
KW
897warrant special handling for them, which can be specified using the C<flags>
898parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
899three classes are treated as malformations and handled as such. The flags
900C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
901C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
902disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
903restricts the allowed inputs to the strict UTF-8 traditionally defined by
904Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
905definition given by
906L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
907The difference between traditional strictness and C9 strictness is that the
908latter does not forbid non-character code points. (They are still discouraged,
909however.) For more discussion see L<perlunicode/Noncharacter code points>.
910
911The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
912C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
913C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
914raised for their respective categories, but otherwise the code points are
915considered valid (not malformations). To get a category to both be treated as
916a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 917(But note that warnings are not raised if lexically disabled nor if
796b6530 918C<UTF8_CHECK_ONLY> is also specified.)
949cf498 919
760c7c2f
KW
920It is now deprecated to have very high code points (above C<IV_MAX> on the
921platforms) and this function will raise a deprecation warning for these (unless
d5944cab 922such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 923in a 32-bit word.
ab8e6d41
KW
924
925Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
926so using them is more problematic than other above-Unicode code points. Perl
927invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
928likely that non-Perl languages will not be able to read files that contain
aff2be59 929these; nor would Perl understand files
ab8e6d41
KW
930written by something that uses a different extension. For these reasons, there
931is a separate set of flags that can warn and/or disallow these extremely high
932code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
933C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
934are entirely independent from the deprecation warning for code points above
935C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
936code point that needs more than 31 bits to represent. When that happens,
937effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
93832-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
939above-Unicode code points, including these, as malformations; and
940C<UTF8_WARN_SUPER> warns on these.)
941
942On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
943extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
944than on ASCII. Prior to that, code points 2**31 and higher were simply
945unrepresentable, and a different, incompatible method was used to represent
946code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
947and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
948platforms, warning and disallowing 2**31 and higher.
949cf498
KW
949
950All other code points corresponding to Unicode characters, including private
951use and those yet to be assigned, are never considered malformed and never
952warn.
67e989fb 953
37607a96 954=cut
f9380377
KW
955
956Also implemented as a macro in utf8.h
957*/
958
959UV
960Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
961 STRLEN curlen,
962 STRLEN *retlen,
963 const U32 flags)
964{
965 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
966
967 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
968}
969
970/*
971
972=for apidoc utf8n_to_uvchr_error
973
974THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
975Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
976
977This function is for code that needs to know what the precise malformation(s)
978are when an error is found.
979
980It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
981all the others, C<errors>. If this parameter is 0, this function behaves
982identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
983to a C<U32> variable, which this function sets to indicate any errors found.
984Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
985C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
986of these bits will be set if a malformation is found, even if the input
7a65503b 987C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
988exceptions are noted:
989
990=over 4
991
992=item C<UTF8_GOT_ABOVE_31_BIT>
993
994The code point represented by the input UTF-8 sequence occupies more than 31
995bits.
996This bit is set only if the input C<flags> parameter contains either the
997C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
998
999=item C<UTF8_GOT_CONTINUATION>
1000
1001The input sequence was malformed in that the first byte was a a UTF-8
1002continuation byte.
1003
1004=item C<UTF8_GOT_EMPTY>
1005
1006The input C<curlen> parameter was 0.
1007
1008=item C<UTF8_GOT_LONG>
1009
1010The input sequence was malformed in that there is some other sequence that
1011evaluates to the same code point, but that sequence is shorter than this one.
1012
1013=item C<UTF8_GOT_NONCHAR>
1014
1015The code point represented by the input UTF-8 sequence is for a Unicode
1016non-character code point.
1017This bit is set only if the input C<flags> parameter contains either the
1018C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1019
1020=item C<UTF8_GOT_NON_CONTINUATION>
1021
1022The input sequence was malformed in that a non-continuation type byte was found
1023in a position where only a continuation type one should be.
1024
1025=item C<UTF8_GOT_OVERFLOW>
1026
1027The input sequence was malformed in that it is for a code point that is not
1028representable in the number of bits available in a UV on the current platform.
1029
1030=item C<UTF8_GOT_SHORT>
1031
1032The input sequence was malformed in that C<curlen> is smaller than required for
1033a complete sequence. In other words, the input is for a partial character
1034sequence.
1035
1036=item C<UTF8_GOT_SUPER>
1037
1038The input sequence was malformed in that it is for a non-Unicode code point;
1039that is, one above the legal Unicode maximum.
1040This bit is set only if the input C<flags> parameter contains either the
1041C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1042
1043=item C<UTF8_GOT_SURROGATE>
1044
1045The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1046code point.
1047This bit is set only if the input C<flags> parameter contains either the
1048C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1049
1050=back
1051
133551d8
KW
1052To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1053flag to suppress any warnings, and then examine the C<*errors> return.
1054
f9380377 1055=cut
37607a96 1056*/
67e989fb 1057
a0ed51b3 1058UV
f9380377
KW
1059Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1060 STRLEN curlen,
1061 STRLEN *retlen,
1062 const U32 flags,
1063 U32 * errors)
a0ed51b3 1064{
d4c19fe8 1065 const U8 * const s0 = s;
2b5e7bc2
KW
1066 U8 * send = NULL; /* (initialized to silence compilers' wrong
1067 warning) */
1068 U32 possible_problems = 0; /* A bit is set here for each potential problem
1069 found as we go along */
eb83ed87 1070 UV uv = *s;
2b5e7bc2
KW
1071 STRLEN expectlen = 0; /* How long should this sequence be?
1072 (initialized to silence compilers' wrong
1073 warning) */
e308b348 1074 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1075 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1076 this gets set and discarded */
a0dbb045 1077
2b5e7bc2
KW
1078 /* The below are used only if there is both an overlong malformation and a
1079 * too short one. Otherwise the first two are set to 's0' and 'send', and
1080 * the third not used at all */
1081 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1082 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1083 warning) */
2b5e7bc2 1084 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1085
f9380377
KW
1086 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1087
1088 if (errors) {
1089 *errors = 0;
1090 }
1091 else {
1092 errors = &discard_errors;
1093 }
a0dbb045 1094
eb83ed87
KW
1095 /* The order of malformation tests here is important. We should consume as
1096 * few bytes as possible in order to not skip any valid character. This is
1097 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1098 * http://unicode.org/reports/tr36 for more discussion as to why. For
1099 * example, once we've done a UTF8SKIP, we can tell the expected number of
1100 * bytes, and could fail right off the bat if the input parameters indicate
1101 * that there are too few available. But it could be that just that first
1102 * byte is garbled, and the intended character occupies fewer bytes. If we
1103 * blindly assumed that the first byte is correct, and skipped based on
1104 * that number, we could skip over a valid input character. So instead, we
1105 * always examine the sequence byte-by-byte.
1106 *
1107 * We also should not consume too few bytes, otherwise someone could inject
1108 * things. For example, an input could be deliberately designed to
1109 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1110 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1111 * which is actually part of of the overflowing sequence), that could look
1112 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1113 * sequence and process the rest, inappropriately.
1114 *
1115 * Some possible input sequences are malformed in more than one way. This
1116 * function goes to lengths to try to find all of them. This is necessary
1117 * for correctness, as the inputs may allow one malformation but not
1118 * another, and if we abandon searching for others after finding the
1119 * allowed one, we could allow in something that shouldn't have been.
1120 */
eb83ed87 1121
b5b9af04 1122 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1123 possible_problems |= UTF8_GOT_EMPTY;
1124 curlen = 0;
5a48568d 1125 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1126 goto ready_to_handle_errors;
0c443dc2
JH
1127 }
1128
eb83ed87
KW
1129 expectlen = UTF8SKIP(s);
1130
1131 /* A well-formed UTF-8 character, as the vast majority of calls to this
1132 * function will be for, has this expected length. For efficiency, set
1133 * things up here to return it. It will be overriden only in those rare
1134 * cases where a malformation is found */
1135 if (retlen) {
1136 *retlen = expectlen;
1137 }
1138
1139 /* An invariant is trivially well-formed */
1d72bdf6 1140 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1141 return uv;
a0ed51b3 1142 }
67e989fb 1143
eb83ed87 1144 /* A continuation character can't start a valid sequence */
b5b9af04 1145 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1146 possible_problems |= UTF8_GOT_CONTINUATION;
1147 curlen = 1;
1148 uv = UNICODE_REPLACEMENT;
1149 goto ready_to_handle_errors;
ba210ebe 1150 }
9041c2e3 1151
dcd27b3c
KW
1152 /* Here is not a continuation byte, nor an invariant. The only thing left
1153 * is a start byte (possibly for an overlong) */
1154
534752c1
KW
1155 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1156 * that indicate the number of bytes in the character's whole UTF-8
1157 * sequence, leaving just the bits that are part of the value. */
1158 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1159
e308b348
KW
1160 /* Setup the loop end point, making sure to not look past the end of the
1161 * input string, and flag it as too short if the size isn't big enough. */
1162 send = (U8*) s0;
1163 if (UNLIKELY(curlen < expectlen)) {
1164 possible_problems |= UTF8_GOT_SHORT;
1165 avail_len = curlen;
1166 send += curlen;
1167 }
1168 else {
1169 send += expectlen;
1170 }
1171 adjusted_send = send;
1172
eb83ed87 1173 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1174 * accumulating each into the working value as we go. */
eb83ed87 1175 for (s = s0 + 1; s < send; s++) {
b5b9af04 1176 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1177 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1178 continue;
1179 }
1180
1181 /* Here, found a non-continuation before processing all expected bytes.
1182 * This byte indicates the beginning of a new character, so quit, even
1183 * if allowing this malformation. */
2b5e7bc2 1184 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1185 break;
eb83ed87
KW
1186 } /* End of loop through the character's bytes */
1187
1188 /* Save how many bytes were actually in the character */
1189 curlen = s - s0;
1190
e308b348
KW
1191 /* A convenience macro that matches either of the too-short conditions. */
1192# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1193
e308b348 1194 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1195 uv_so_far = uv;
1196 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1197 }
1198
2b5e7bc2
KW
1199 /* Note that there are two types of too-short malformation. One is when
1200 * there is actual wrong data before the normal termination of the
1201 * sequence. The other is that the sequence wasn't complete before the end
1202 * of the data we are allowed to look at, based on the input 'curlen'.
1203 * This means that we were passed data for a partial character, but it is
1204 * valid as far as we saw. The other is definitely invalid. This
1205 * distinction could be important to a caller, so the two types are kept
1206 * separate. */
1207
1208 /* Check for overflow */
1209 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1210 possible_problems |= UTF8_GOT_OVERFLOW;
1211 uv = UNICODE_REPLACEMENT;
eb83ed87 1212 }
eb83ed87 1213
2b5e7bc2
KW
1214 /* Check for overlong. If no problems so far, 'uv' is the correct code
1215 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1216 * we must look at the UTF-8 byte sequence itself to see if it is for an
1217 * overlong */
1218 if ( ( LIKELY(! possible_problems)
1219 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1220 || ( UNLIKELY( possible_problems)
1221 && ( UNLIKELY(! UTF8_IS_START(*s0))
1222 || ( curlen > 1
1223 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1224 send - s0))))))
2f8f112e 1225 {
2b5e7bc2
KW
1226 possible_problems |= UTF8_GOT_LONG;
1227
2b5e7bc2
KW
1228 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1229 UV min_uv = uv_so_far;
1230 STRLEN i;
1231
1232 /* Here, the input is both overlong and is missing some trailing
1233 * bytes. There is no single code point it could be for, but there
1234 * may be enough information present to determine if what we have
1235 * so far is for an unallowed code point, such as for a surrogate.
1236 * The code below has the intelligence to determine this, but just
1237 * for non-overlong UTF-8 sequences. What we do here is calculate
1238 * the smallest code point the input could represent if there were
1239 * no too short malformation. Then we compute and save the UTF-8
1240 * for that, which is what the code below looks at instead of the
1241 * raw input. It turns out that the smallest such code point is
1242 * all we need. */
1243 for (i = curlen; i < expectlen; i++) {
1244 min_uv = UTF8_ACCUMULATE(min_uv,
1245 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1246 }
1247
1248 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1249 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1250 to free it ourselves if
1251 warnings are made fatal */
1252 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1253 }
eb83ed87
KW
1254 }
1255
2b5e7bc2
KW
1256 /* Now check that the input isn't for a problematic code point not allowed
1257 * by the input parameters. */
1258 /* isn't problematic if < this */
1259 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1260 || ( UNLIKELY(possible_problems)
1261 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1262 && ((flags & ( UTF8_DISALLOW_NONCHAR
1263 |UTF8_DISALLOW_SURROGATE
1264 |UTF8_DISALLOW_SUPER
1265 |UTF8_DISALLOW_ABOVE_31_BIT
1266 |UTF8_WARN_NONCHAR
1267 |UTF8_WARN_SURROGATE
1268 |UTF8_WARN_SUPER
1269 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1270 /* In case of a malformation, 'uv' is not valid, and has
1271 * been changed to something in the Unicode range.
1272 * Currently we don't output a deprecation message if there
1273 * is already a malformation, so we don't have to special
1274 * case the test immediately below */
760c7c2f
KW
1275 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1276 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1277 {
2b5e7bc2
KW
1278 /* If there were no malformations, or the only malformation is an
1279 * overlong, 'uv' is valid */
1280 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1281 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1282 possible_problems |= UTF8_GOT_SURROGATE;
1283 }
1284 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1285 possible_problems |= UTF8_GOT_SUPER;
1286 }
1287 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1288 possible_problems |= UTF8_GOT_NONCHAR;
1289 }
1290 }
1291 else { /* Otherwise, need to look at the source UTF-8, possibly
1292 adjusted to be non-overlong */
1293
1294 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1295 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1296 {
2b5e7bc2
KW
1297 possible_problems |= UTF8_GOT_SUPER;
1298 }
1299 else if (curlen > 1) {
1300 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1301 NATIVE_UTF8_TO_I8(*adjusted_s0),
1302 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1303 {
2b5e7bc2 1304 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1305 }
2b5e7bc2
KW
1306 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1307 NATIVE_UTF8_TO_I8(*adjusted_s0),
1308 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1309 {
1310 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1311 }
1312 }
c0236afe 1313
2b5e7bc2
KW
1314 /* We need a complete well-formed UTF-8 character to discern
1315 * non-characters, so can't look for them here */
1316 }
1317 }
949cf498 1318
2b5e7bc2
KW
1319 ready_to_handle_errors:
1320
1321 /* At this point:
1322 * curlen contains the number of bytes in the sequence that
1323 * this call should advance the input by.
e308b348
KW
1324 * avail_len gives the available number of bytes passed in, but
1325 * only if this is less than the expected number of
1326 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1327 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1328 * is set in it for each potential problem found.
1329 * uv contains the code point the input sequence
1330 * represents; or if there is a problem that prevents
1331 * a well-defined value from being computed, it is
1332 * some subsitute value, typically the REPLACEMENT
1333 * CHARACTER.
1334 * s0 points to the first byte of the character
1335 * send points to just after where that (potentially
1336 * partial) character ends
1337 * adjusted_s0 normally is the same as s0, but in case of an
1338 * overlong for which the UTF-8 matters below, it is
1339 * the first byte of the shortest form representation
1340 * of the input.
1341 * adjusted_send normally is the same as 'send', but if adjusted_s0
1342 * is set to something other than s0, this points one
1343 * beyond its end
1344 */
eb83ed87 1345
2b5e7bc2
KW
1346 if (UNLIKELY(possible_problems)) {
1347 bool disallowed = FALSE;
1348 const U32 orig_problems = possible_problems;
1349
1350 while (possible_problems) { /* Handle each possible problem */
1351 UV pack_warn = 0;
1352 char * message = NULL;
1353
1354 /* Each 'if' clause handles one problem. They are ordered so that
1355 * the first ones' messages will be displayed before the later
1356 * ones; this is kinda in decreasing severity order */
1357 if (possible_problems & UTF8_GOT_OVERFLOW) {
1358
1359 /* Overflow means also got a super and above 31 bits, but we
1360 * handle all three cases here */
1361 possible_problems
1362 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1363 *errors |= UTF8_GOT_OVERFLOW;
1364
1365 /* But the API says we flag all errors found */
1366 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1367 *errors |= UTF8_GOT_SUPER;
1368 }
1369 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1370 *errors |= UTF8_GOT_ABOVE_31_BIT;
1371 }
2b5e7bc2
KW
1372
1373 disallowed = TRUE;
1374
1375 /* The warnings code explicitly says it doesn't handle the case
1376 * of packWARN2 and two categories which have parent-child
1377 * relationship. Even if it works now to raise the warning if
1378 * either is enabled, it wouldn't necessarily do so in the
1379 * future. We output (only) the most dire warning*/
1380 if (! (flags & UTF8_CHECK_ONLY)) {
1381 if (ckWARN_d(WARN_UTF8)) {
1382 pack_warn = packWARN(WARN_UTF8);
1383 }
1384 else if (ckWARN_d(WARN_NON_UNICODE)) {
1385 pack_warn = packWARN(WARN_NON_UNICODE);
1386 }
1387 if (pack_warn) {
1388 message = Perl_form(aTHX_ "%s: %s (overflows)",
1389 malformed_text,
1390 _byte_dump_string(s0, send - s0));
1391 }
1392 }
1393 }
1394 else if (possible_problems & UTF8_GOT_EMPTY) {
1395 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1396 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1397
1398 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1399
1400 /* This so-called malformation is now treated as a bug in
1401 * the caller. If you have nothing to decode, skip calling
1402 * this function */
1403 assert(0);
1404
2b5e7bc2
KW
1405 disallowed = TRUE;
1406 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1407 pack_warn = packWARN(WARN_UTF8);
1408 message = Perl_form(aTHX_ "%s (empty string)",
1409 malformed_text);
1410 }
1411 }
1412 }
1413 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1414 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1415 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1416
1417 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1418 disallowed = TRUE;
1419 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1420 pack_warn = packWARN(WARN_UTF8);
1421 message = Perl_form(aTHX_
1422 "%s: %s (unexpected continuation byte 0x%02x,"
1423 " with no preceding start byte)",
1424 malformed_text,
1425 _byte_dump_string(s0, 1), *s0);
1426 }
1427 }
1428 }
2b5e7bc2
KW
1429 else if (possible_problems & UTF8_GOT_SHORT) {
1430 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1431 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1432
1433 if (! (flags & UTF8_ALLOW_SHORT)) {
1434 disallowed = TRUE;
1435 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1436 pack_warn = packWARN(WARN_UTF8);
1437 message = Perl_form(aTHX_
9a6c9c81 1438 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1439 malformed_text,
1440 _byte_dump_string(s0, send - s0),
e308b348
KW
1441 (int)avail_len,
1442 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1443 (int)expectlen);
1444 }
1445 }
ba210ebe 1446
2b5e7bc2 1447 }
e308b348
KW
1448 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1449 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1450 *errors |= UTF8_GOT_NON_CONTINUATION;
1451
1452 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1453 disallowed = TRUE;
1454 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1455 pack_warn = packWARN(WARN_UTF8);
1456 message = Perl_form(aTHX_ "%s",
1457 unexpected_non_continuation_text(s0,
1458 send - s0,
1459 s - s0,
1460 (int) expectlen));
1461 }
1462 }
1463 }
2b5e7bc2
KW
1464 else if (possible_problems & UTF8_GOT_LONG) {
1465 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1466 *errors |= UTF8_GOT_LONG;
2b5e7bc2
KW
1467
1468 if (! (flags & UTF8_ALLOW_LONG)) {
1469 disallowed = TRUE;
1470
1471 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1472 pack_warn = packWARN(WARN_UTF8);
1473
1474 /* These error types cause 'uv' to be something that
1475 * isn't what was intended, so can't use it in the
1476 * message. The other error types either can't
1477 * generate an overlong, or else the 'uv' is valid */
1478 if (orig_problems &
1479 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1480 {
1481 message = Perl_form(aTHX_
1482 "%s: %s (any UTF-8 sequence that starts"
1483 " with \"%s\" is overlong which can and"
1484 " should be represented with a"
1485 " different, shorter sequence)",
1486 malformed_text,
1487 _byte_dump_string(s0, send - s0),
1488 _byte_dump_string(s0, curlen));
1489 }
1490 else {
1491 U8 tmpbuf[UTF8_MAXBYTES+1];
1492 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1493 uv, 0);
1494 message = Perl_form(aTHX_
1495 "%s: %s (overlong; instead use %s to represent"
147e3846 1496 " U+%0*" UVXf ")",
2b5e7bc2
KW
1497 malformed_text,
1498 _byte_dump_string(s0, send - s0),
1499 _byte_dump_string(tmpbuf, e - tmpbuf),
1500 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1501 small code points */
1502 uv);
1503 }
1504 }
1505 }
1506 }
1507 else if (possible_problems & UTF8_GOT_SURROGATE) {
1508 possible_problems &= ~UTF8_GOT_SURROGATE;
1509
f9380377
KW
1510 if (flags & UTF8_WARN_SURROGATE) {
1511 *errors |= UTF8_GOT_SURROGATE;
1512
1513 if ( ! (flags & UTF8_CHECK_ONLY)
1514 && ckWARN_d(WARN_SURROGATE))
1515 {
2b5e7bc2
KW
1516 pack_warn = packWARN(WARN_SURROGATE);
1517
1518 /* These are the only errors that can occur with a
1519 * surrogate when the 'uv' isn't valid */
1520 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1521 message = Perl_form(aTHX_
1522 "UTF-16 surrogate (any UTF-8 sequence that"
1523 " starts with \"%s\" is for a surrogate)",
1524 _byte_dump_string(s0, curlen));
1525 }
1526 else {
1527 message = Perl_form(aTHX_
147e3846 1528 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1529 }
f9380377 1530 }
2b5e7bc2 1531 }
ba210ebe 1532
2b5e7bc2
KW
1533 if (flags & UTF8_DISALLOW_SURROGATE) {
1534 disallowed = TRUE;
f9380377 1535 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1536 }
1537 }
1538 else if (possible_problems & UTF8_GOT_SUPER) {
1539 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1540
f9380377
KW
1541 if (flags & UTF8_WARN_SUPER) {
1542 *errors |= UTF8_GOT_SUPER;
1543
1544 if ( ! (flags & UTF8_CHECK_ONLY)
1545 && ckWARN_d(WARN_NON_UNICODE))
1546 {
2b5e7bc2
KW
1547 pack_warn = packWARN(WARN_NON_UNICODE);
1548
1549 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1550 message = Perl_form(aTHX_
1551 "Any UTF-8 sequence that starts with"
1552 " \"%s\" is for a non-Unicode code point,"
1553 " may not be portable",
1554 _byte_dump_string(s0, curlen));
1555 }
1556 else {
1557 message = Perl_form(aTHX_
147e3846 1558 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1559 " Unicode, may not be portable",
1560 uv);
1561 }
f9380377 1562 }
2b5e7bc2 1563 }
ba210ebe 1564
2b5e7bc2
KW
1565 /* The maximum code point ever specified by a standard was
1566 * 2**31 - 1. Anything larger than that is a Perl extension
1567 * that very well may not be understood by other applications
1568 * (including earlier perl versions on EBCDIC platforms). We
1569 * test for these after the regular SUPER ones, and before
1570 * possibly bailing out, so that the slightly more dire warning
1571 * will override the regular one. */
1572 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1573 |UTF8_WARN_SUPER
1574 |UTF8_DISALLOW_ABOVE_31_BIT))
1575 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1576 && UNLIKELY(is_utf8_cp_above_31_bits(
1577 adjusted_s0,
1578 adjusted_send)))
1579 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1580 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1581 {
1582 if ( ! (flags & UTF8_CHECK_ONLY)
1583 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1584 && ckWARN_d(WARN_UTF8))
1585 {
1586 pack_warn = packWARN(WARN_UTF8);
1587
1588 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1589 message = Perl_form(aTHX_
1590 "Any UTF-8 sequence that starts with"
1591 " \"%s\" is for a non-Unicode code"
1592 " point, and is not portable",
1593 _byte_dump_string(s0, curlen));
1594 }
1595 else {
1596 message = Perl_form(aTHX_
147e3846 1597 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1598 " and not portable",
1599 uv);
1600 }
1601 }
1602
f9380377
KW
1603 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1604 *errors |= UTF8_GOT_ABOVE_31_BIT;
1605
1606 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1607 disallowed = TRUE;
1608 }
2b5e7bc2
KW
1609 }
1610 }
eb83ed87 1611
2b5e7bc2 1612 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1613 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1614 disallowed = TRUE;
1615 }
eb83ed87 1616
2b5e7bc2
KW
1617 /* The deprecated warning overrides any non-deprecated one. If
1618 * there are other problems, a deprecation message is not
1619 * really helpful, so don't bother to raise it in that case.
1620 * This also keeps the code from having to handle the case
1621 * where 'uv' is not valid. */
1622 if ( ! (orig_problems
1623 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1624 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1625 && ckWARN_d(WARN_DEPRECATED))
1626 {
1627 message = Perl_form(aTHX_ cp_above_legal_max,
1628 uv, MAX_NON_DEPRECATED_CP);
1629 pack_warn = packWARN(WARN_DEPRECATED);
1630 }
1631 }
1632 else if (possible_problems & UTF8_GOT_NONCHAR) {
1633 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1634
f9380377
KW
1635 if (flags & UTF8_WARN_NONCHAR) {
1636 *errors |= UTF8_GOT_NONCHAR;
1637
1638 if ( ! (flags & UTF8_CHECK_ONLY)
1639 && ckWARN_d(WARN_NONCHAR))
1640 {
2b5e7bc2
KW
1641 /* The code above should have guaranteed that we don't
1642 * get here with errors other than overlong */
1643 assert (! (orig_problems
1644 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1645
1646 pack_warn = packWARN(WARN_NONCHAR);
1647 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1648 " U+%04" UVXf " is not recommended"
2b5e7bc2 1649 " for open interchange", uv);
f9380377 1650 }
2b5e7bc2 1651 }
5b311467 1652
2b5e7bc2
KW
1653 if (flags & UTF8_DISALLOW_NONCHAR) {
1654 disallowed = TRUE;
f9380377 1655 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1656 }
1657 } /* End of looking through the possible flags */
1658
1659 /* Display the message (if any) for the problem being handled in
1660 * this iteration of the loop */
1661 if (message) {
1662 if (PL_op)
1663 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1664 OP_DESC(PL_op));
1665 else
1666 Perl_warner(aTHX_ pack_warn, "%s", message);
1667 }
1668 } /* End of 'while (possible_problems) {' */
a0dbb045 1669
2b5e7bc2
KW
1670 /* Since there was a possible problem, the returned length may need to
1671 * be changed from the one stored at the beginning of this function.
1672 * Instead of trying to figure out if that's needed, just do it. */
1673 if (retlen) {
1674 *retlen = curlen;
1675 }
a0dbb045 1676
2b5e7bc2
KW
1677 if (disallowed) {
1678 if (flags & UTF8_CHECK_ONLY && retlen) {
1679 *retlen = ((STRLEN) -1);
1680 }
1681 return 0;
1682 }
eb83ed87 1683 }
ba210ebe 1684
2b5e7bc2 1685 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1686}
1687
8e84507e 1688/*
ec5f19d0
KW
1689=for apidoc utf8_to_uvchr_buf
1690
1691Returns the native code point of the first character in the string C<s> which
1692is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1693C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1694
524080c4
KW
1695If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1696enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1697C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1698(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1699C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1700the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1701See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1702returned.
ec5f19d0 1703
760c7c2f
KW
1704Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1705unless those are turned off.
1706
ec5f19d0 1707=cut
52be2536
KW
1708
1709Also implemented as a macro in utf8.h
1710
ec5f19d0
KW
1711*/
1712
1713
1714UV
1715Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1716{
ec5f19d0
KW
1717 assert(s < send);
1718
1719 return utf8n_to_uvchr(s, send - s, retlen,
1720 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1721}
1722
52be2536
KW
1723/* This is marked as deprecated
1724 *
ec5f19d0
KW
1725=for apidoc utf8_to_uvuni_buf
1726
de69f3af
KW
1727Only in very rare circumstances should code need to be dealing in Unicode
1728(as opposed to native) code points. In those few cases, use
1729C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1730
1731Returns the Unicode (not-native) code point of the first character in the
1732string C<s> which
ec5f19d0
KW
1733is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1734C<retlen> will be set to the length, in bytes, of that character.
1735
524080c4
KW
1736If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1737enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1738NULL) to -1. If those warnings are off, the computed value if well-defined (or
1739the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1740is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1741next possible position in C<s> that could begin a non-malformed character.
de69f3af 1742See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
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
KW
1747=cut
1748*/
1749
1750UV
1751Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1752{
1753 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1754
1755 assert(send > s);
1756
5962d97e
KW
1757 /* Call the low level routine, asking for checks */
1758 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1759}
1760
b76347f2 1761/*
87cea99e 1762=for apidoc utf8_length
b76347f2
JH
1763
1764Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1765Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1766up past C<e>, croaks.
b76347f2
JH
1767
1768=cut
1769*/
1770
1771STRLEN
35a4481c 1772Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1773{
1774 STRLEN len = 0;
1775
7918f24d
NC
1776 PERL_ARGS_ASSERT_UTF8_LENGTH;
1777
8850bf83
JH
1778 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1779 * the bitops (especially ~) can create illegal UTF-8.
1780 * In other words: in Perl UTF-8 is not just for Unicode. */
1781
a3b680e6
AL
1782 if (e < s)
1783 goto warn_and_return;
b76347f2 1784 while (s < e) {
4cbf4130 1785 s += UTF8SKIP(s);
8e91ec7f
AV
1786 len++;
1787 }
1788
1789 if (e != s) {
1790 len--;
1791 warn_and_return:
9b387841
NC
1792 if (PL_op)
1793 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1794 "%s in %s", unees, OP_DESC(PL_op));
1795 else
61a12c31 1796 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1797 }
1798
1799 return len;
1800}
1801
b06226ff 1802/*
fed3ba5d
NC
1803=for apidoc bytes_cmp_utf8
1804
a1433954 1805Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1806sequence of characters (stored as UTF-8)
1807in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1808equal, -1 or -2 if the first string is less than the second string, +1 or +2
1809if the first string is greater than the second string.
1810
1811-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1812longer string. -2 or +2 is returned if
1813there was a difference between characters
fed3ba5d
NC
1814within the strings.
1815
1816=cut
1817*/
1818
1819int
1820Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1821{
1822 const U8 *const bend = b + blen;
1823 const U8 *const uend = u + ulen;
1824
1825 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1826
1827 while (b < bend && u < uend) {
1828 U8 c = *u++;
1829 if (!UTF8_IS_INVARIANT(c)) {
1830 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1831 if (u < uend) {
1832 U8 c1 = *u++;
1833 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1834 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1835 } else {
2b5e7bc2 1836 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1837 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1838 "%s %s%s",
7cf8d05d 1839 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1840 PL_op ? " in " : "",
1841 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1842 return -2;
1843 }
1844 } else {
1845 if (PL_op)
1846 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1847 "%s in %s", unees, OP_DESC(PL_op));
1848 else
61a12c31 1849 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1850 return -2; /* Really want to return undef :-) */
1851 }
1852 } else {
1853 return -2;
1854 }
1855 }
1856 if (*b != c) {
1857 return *b < c ? -2 : +2;
1858 }
1859 ++b;
1860 }
1861
1862 if (b == bend && u == uend)
1863 return 0;
1864
1865 return b < bend ? +1 : -1;
1866}
1867
1868/*
87cea99e 1869=for apidoc utf8_to_bytes
6940069f 1870
2bbc8d55 1871Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1872Unlike L</bytes_to_utf8>, this over-writes the original string, and
1873updates C<len> to contain the new length.
67e989fb 1874Returns zero on failure, setting C<len> to -1.
6940069f 1875
a1433954 1876If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1877
6940069f
GS
1878=cut
1879*/
1880
1881U8 *
37607a96 1882Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1883{
d4c19fe8
AL
1884 U8 * const save = s;
1885 U8 * const send = s + *len;
6940069f 1886 U8 *d;
246fae53 1887
7918f24d 1888 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1889 PERL_UNUSED_CONTEXT;
7918f24d 1890
1e54db1a 1891 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1892 while (s < send) {
d59937ca
KW
1893 if (! UTF8_IS_INVARIANT(*s)) {
1894 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1895 *len = ((STRLEN) -1);
1896 return 0;
1897 }
1898 s++;
dcad2880 1899 }
d59937ca 1900 s++;
246fae53 1901 }
dcad2880
JH
1902
1903 d = s = save;
6940069f 1904 while (s < send) {
80e0b38f
KW
1905 U8 c = *s++;
1906 if (! UTF8_IS_INVARIANT(c)) {
1907 /* Then it is two-byte encoded */
a62b247b 1908 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1909 s++;
1910 }
1911 *d++ = c;
6940069f
GS
1912 }
1913 *d = '\0';
246fae53 1914 *len = d - save;
6940069f
GS
1915 return save;
1916}
1917
1918/*
87cea99e 1919=for apidoc bytes_from_utf8
f9a63242 1920
2bbc8d55 1921Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1922Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1923the newly-created string, and updates C<len> to contain the new
1924length. Returns the original string if no conversion occurs, C<len>
72d33970 1925is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19260 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1927in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1928
37607a96
PK
1929=cut
1930*/
f9a63242
JH
1931
1932U8 *
e1ec3a88 1933Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1934{
f9a63242 1935 U8 *d;
e1ec3a88
AL
1936 const U8 *start = s;
1937 const U8 *send;
f9a63242
JH
1938 I32 count = 0;
1939
7918f24d 1940 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1941 PERL_UNUSED_CONTEXT;
f9a63242 1942 if (!*is_utf8)
73d840c0 1943 return (U8 *)start;
f9a63242 1944
1e54db1a 1945 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1946 for (send = s + *len; s < send;) {
d59937ca
KW
1947 if (! UTF8_IS_INVARIANT(*s)) {
1948 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1949 return (U8 *)start;
d59937ca
KW
1950 }
1951 count++;
1952 s++;
db42d148 1953 }
d59937ca 1954 s++;
f9a63242
JH
1955 }
1956
35da51f7 1957 *is_utf8 = FALSE;
f9a63242 1958
212542aa 1959 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1960 s = start; start = d;
f9a63242
JH
1961 while (s < send) {
1962 U8 c = *s++;
1a91c45d 1963 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1964 /* Then it is two-byte encoded */
a62b247b 1965 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1966 s++;
c4d5f83a
NIS
1967 }
1968 *d++ = c;
f9a63242
JH
1969 }
1970 *d = '\0';
1971 *len = d - start;
73d840c0 1972 return (U8 *)start;
f9a63242
JH
1973}
1974
1975/*
87cea99e 1976=for apidoc bytes_to_utf8
6940069f 1977
ff97e5cf
KW
1978Converts a string C<s> of length C<len> bytes from the native encoding into
1979UTF-8.
6662521e 1980Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1981reflect the new length in bytes.
6940069f 1982
75200dff 1983A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1984
1985If you want to convert to UTF-8 from encodings other than
1986the native (Latin1 or EBCDIC),
a1433954 1987see L</sv_recode_to_utf8>().
c9ada85f 1988
497711e7 1989=cut
6940069f
GS
1990*/
1991
c682ebef
FC
1992/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1993 likewise need duplication. */
1994
6940069f 1995U8*
35a4481c 1996Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1997{
35a4481c 1998 const U8 * const send = s + (*len);
6940069f
GS
1999 U8 *d;
2000 U8 *dst;
7918f24d
NC
2001
2002 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2003 PERL_UNUSED_CONTEXT;
6940069f 2004
212542aa 2005 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2006 dst = d;
2007
2008 while (s < send) {
55d09dc8
KW
2009 append_utf8_from_native_byte(*s, &d);
2010 s++;
6940069f
GS
2011 }
2012 *d = '\0';
6662521e 2013 *len = d-dst;
6940069f
GS
2014 return dst;
2015}
2016
a0ed51b3 2017/*
dea0fc0b 2018 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2019 *
2020 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2021 * We optimize for native, for obvious reasons. */
2022
2023U8*
dea0fc0b 2024Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2025{
dea0fc0b
JH
2026 U8* pend;
2027 U8* dstart = d;
2028
7918f24d
NC
2029 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2030
dea0fc0b 2031 if (bytelen & 1)
147e3846 2032 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2033
2034 pend = p + bytelen;
2035
a0ed51b3 2036 while (p < pend) {
dea0fc0b
JH
2037 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2038 p += 2;
2d1545e5 2039 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2040 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2041 continue;
2042 }
56d37426
KW
2043 if (uv <= MAX_UTF8_TWO_BYTE) {
2044 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2045 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2046 continue;
2047 }
46956fad
KW
2048#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2049#define LAST_HIGH_SURROGATE 0xDBFF
2050#define FIRST_LOW_SURROGATE 0xDC00
2051#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2052
2053 /* This assumes that most uses will be in the first Unicode plane, not
2054 * needing surrogates */
2055 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2056 && uv <= UNICODE_SURROGATE_LAST))
2057 {
2058 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2059 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2060 }
2061 else {
01ea242b 2062 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2063 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2064 || UNLIKELY(low > LAST_LOW_SURROGATE))
2065 {
01ea242b 2066 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2067 }
2068 p += 2;
46956fad
KW
2069 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2070 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2071 }
a0ed51b3 2072 }
56d37426
KW
2073#ifdef EBCDIC
2074 d = uvoffuni_to_utf8_flags(d, uv, 0);
2075#else
a0ed51b3 2076 if (uv < 0x10000) {
eb160463
GS
2077 *d++ = (U8)(( uv >> 12) | 0xe0);
2078 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2079 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2080 continue;
2081 }
2082 else {
eb160463
GS
2083 *d++ = (U8)(( uv >> 18) | 0xf0);
2084 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2085 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2086 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2087 continue;
2088 }
56d37426 2089#endif
a0ed51b3 2090 }
dea0fc0b 2091 *newlen = d - dstart;
a0ed51b3
LW
2092 return d;
2093}
2094
2095/* Note: this one is slightly destructive of the source. */
2096
2097U8*
dea0fc0b 2098Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2099{
2100 U8* s = (U8*)p;
d4c19fe8 2101 U8* const send = s + bytelen;
7918f24d
NC
2102
2103 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2104
e0ea5e2d 2105 if (bytelen & 1)
147e3846 2106 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2107 (UV)bytelen);
2108
a0ed51b3 2109 while (s < send) {
d4c19fe8 2110 const U8 tmp = s[0];
a0ed51b3
LW
2111 s[0] = s[1];
2112 s[1] = tmp;
2113 s += 2;
2114 }
dea0fc0b 2115 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2116}
2117
922e8cb4
KW
2118bool
2119Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2120{
2121 U8 tmpbuf[UTF8_MAXBYTES+1];
2122 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2123 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2124}
2125
f9ae8fb6
JD
2126/* Internal function so we can deprecate the external one, and call
2127 this one from other deprecated functions in this file */
2128
f2645549
KW
2129bool
2130Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2131{
f2645549 2132 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2133
2134 if (*p == '_')
2135 return TRUE;
f25ce844 2136 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2137}
2138
5092f92a 2139bool
eba68aa0
KW
2140Perl__is_uni_perl_idcont(pTHX_ UV c)
2141{
2142 U8 tmpbuf[UTF8_MAXBYTES+1];
2143 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2144 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2145}
2146
2147bool
f91dcd13
KW
2148Perl__is_uni_perl_idstart(pTHX_ UV c)
2149{
2150 U8 tmpbuf[UTF8_MAXBYTES+1];
2151 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2152 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2153}
2154
3a4c58c9
KW
2155UV
2156Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2157{
2158 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2159 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2160 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2161 * either "SS" or "Ss". Which one to use is passed into the routine in
2162 * 'S_or_s' to avoid a test */
2163
2164 UV converted = toUPPER_LATIN1_MOD(c);
2165
2166 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2167
2168 assert(S_or_s == 'S' || S_or_s == 's');
2169
6f2d5cbc 2170 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2171 characters in this range */
3a4c58c9
KW
2172 *p = (U8) converted;
2173 *lenp = 1;
2174 return converted;
2175 }
2176
2177 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2178 * which it maps to one of them, so as to only have to have one check for
2179 * it in the main case */
2180 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2181 switch (c) {
2182 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2183 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2184 break;
2185 case MICRO_SIGN:
2186 converted = GREEK_CAPITAL_LETTER_MU;
2187 break;
79e064b9
KW
2188#if UNICODE_MAJOR_VERSION > 2 \
2189 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2190 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2191 case LATIN_SMALL_LETTER_SHARP_S:
2192 *(p)++ = 'S';
2193 *p = S_or_s;
2194 *lenp = 2;
2195 return 'S';
79e064b9 2196#endif
3a4c58c9
KW
2197 default:
2198 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2199 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2200 }
2201 }
2202
2203 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2204 *p = UTF8_TWO_BYTE_LO(converted);
2205 *lenp = 2;
2206
2207 return converted;
2208}
2209
50bda2c3
KW
2210/* Call the function to convert a UTF-8 encoded character to the specified case.
2211 * Note that there may be more than one character in the result.
2212 * INP is a pointer to the first byte of the input character
2213 * OUTP will be set to the first byte of the string of changed characters. It
2214 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2215 * LENP will be set to the length in bytes of the string of changed characters
2216 *
2217 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2218#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2219#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2220#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2221
b9992569
KW
2222/* This additionally has the input parameter 'specials', which if non-zero will
2223 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2224 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2225#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2226
84afefe6
JH
2227UV
2228Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2229{
a1433954
KW
2230 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2231 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2232 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2233 * the changed version may be longer than the original character.
2234 *
2235 * The ordinal of the first character of the changed version is returned
2236 * (but note, as explained above, that there may be more.) */
2237
7918f24d
NC
2238 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2239
3a4c58c9
KW
2240 if (c < 256) {
2241 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2242 }
2243
0ebc6274 2244 uvchr_to_utf8(p, c);
b9992569 2245 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2246}
2247
84afefe6
JH
2248UV
2249Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2250{
7918f24d
NC
2251 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2252
3a4c58c9
KW
2253 if (c < 256) {
2254 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2255 }
2256
0ebc6274 2257 uvchr_to_utf8(p, c);
b9992569 2258 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2259}
2260
afc16117 2261STATIC U8
eaf412bf 2262S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2263{
2264 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2265 * those, converting the result to UTF-8. Since the result is always just
a1433954 2266 * one character, we allow <p> to be NULL */
afc16117
KW
2267
2268 U8 converted = toLOWER_LATIN1(c);
2269
eaf412bf
KW
2270 PERL_UNUSED_ARG(dummy);
2271
afc16117 2272 if (p != NULL) {
6f2d5cbc 2273 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2274 *p = converted;
2275 *lenp = 1;
2276 }
2277 else {
430c9760
KW
2278 /* Result is known to always be < 256, so can use the EIGHT_BIT
2279 * macros */
2280 *p = UTF8_EIGHT_BIT_HI(converted);
2281 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2282 *lenp = 2;
2283 }
2284 }
2285 return converted;
2286}
2287
84afefe6
JH
2288UV
2289Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2290{
7918f24d
NC
2291 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2292
afc16117 2293 if (c < 256) {
eaf412bf 2294 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2295 }
2296
afc16117 2297 uvchr_to_utf8(p, c);
b9992569 2298 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2299}
2300
84afefe6 2301UV
51910141 2302Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2303{
51910141 2304 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2305 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2306 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2307 *
2308 * Not to be used for locale folds
51910141 2309 */
f673fad4 2310
a1dde8de
KW
2311 UV converted;
2312
2313 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2314 PERL_UNUSED_CONTEXT;
a1dde8de 2315
1ca267a5
KW
2316 assert (! (flags & FOLD_FLAGS_LOCALE));
2317
659a7c2d 2318 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2319 converted = GREEK_SMALL_LETTER_MU;
2320 }
9b63e895
KW
2321#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2322 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2323 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2324 else if ( (flags & FOLD_FLAGS_FULL)
2325 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2326 {
1ca267a5
KW
2327 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2328 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2329 * under those circumstances. */
2330 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2331 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2332 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2333 p, *lenp, U8);
2334 return LATIN_SMALL_LETTER_LONG_S;
2335 }
2336 else {
4f489194
KW
2337 *(p)++ = 's';
2338 *p = 's';
2339 *lenp = 2;
2340 return 's';
1ca267a5 2341 }
a1dde8de 2342 }
9b63e895 2343#endif
a1dde8de
KW
2344 else { /* In this range the fold of all other characters is their lower
2345 case */
2346 converted = toLOWER_LATIN1(c);
2347 }
2348
6f2d5cbc 2349 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2350 *p = (U8) converted;
2351 *lenp = 1;
2352 }
2353 else {
2354 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2355 *p = UTF8_TWO_BYTE_LO(converted);
2356 *lenp = 2;
2357 }
2358
2359 return converted;
2360}
2361
2362UV
31f05a37 2363Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2364{
4b593389 2365
a0270393
KW
2366 /* Not currently externally documented, and subject to change
2367 * <flags> bits meanings:
2368 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2369 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2370 * locale are to be used.
a0270393
KW
2371 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2372 */
4b593389 2373
36bb2ab6 2374 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2375
780fcc9f
KW
2376 if (flags & FOLD_FLAGS_LOCALE) {
2377 /* Treat a UTF-8 locale as not being in locale at all */
2378 if (IN_UTF8_CTYPE_LOCALE) {
2379 flags &= ~FOLD_FLAGS_LOCALE;
2380 }
2381 else {
2382 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2383 goto needs_full_generality;
780fcc9f 2384 }
31f05a37
KW
2385 }
2386
a1dde8de 2387 if (c < 256) {
e7b7ac46 2388 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2389 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2390 }
2391
2f306ab9 2392 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2393 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2394 uvchr_to_utf8(p, c);
b9992569 2395 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
2396 }
2397 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2398 the special flags. */
2399 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2400
2401 needs_full_generality:
a0270393 2402 uvchr_to_utf8(utf8_c, c);
445bf929 2403 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 2404 }
84afefe6
JH
2405}
2406
26483009 2407PERL_STATIC_INLINE bool
5141f98e 2408S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2409 const char *const swashname, SV* const invlist)
bde6a22d 2410{
ea317ccb
KW
2411 /* returns a boolean giving whether or not the UTF8-encoded character that
2412 * starts at <p> is in the swash indicated by <swashname>. <swash>
2413 * contains a pointer to where the swash indicated by <swashname>
2414 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2415 * look at <*swash> and only generate a swash if it is not null. <invlist>
2416 * is NULL or an inversion list that defines the swash. If not null, it
2417 * saves time during initialization of the swash.
ea317ccb
KW
2418 *
2419 * Note that it is assumed that the buffer length of <p> is enough to
2420 * contain all the bytes that comprise the character. Thus, <*p> should
2421 * have been checked before this call for mal-formedness enough to assure
2422 * that. */
2423
7918f24d
NC
2424 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2425
492a624f 2426 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2427 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2428 * as far as there being enough bytes available in it to accommodate the
2429 * character without reading beyond the end, and pass that number on to the
2430 * validating routine */
6302f837 2431 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94
KW
2432 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2433 0,
2434 1 /* Die */ );
2435 NOT_REACHED; /* NOTREACHED */
28123549 2436 }
86ae6e94 2437
87367d5f
KW
2438 if (!*swash) {
2439 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2440 *swash = _core_swash_init("utf8",
2441
2442 /* Only use the name if there is no inversion
2443 * list; otherwise will go out to disk */
2444 (invlist) ? "" : swashname,
2445
2446 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2447 }
28123549 2448
bde6a22d
NC
2449 return swash_fetch(*swash, p, TRUE) != 0;
2450}
2451
da8c1a98
KW
2452PERL_STATIC_INLINE bool
2453S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2454 const char *const swashname, SV* const invlist)
2455{
2456 /* returns a boolean giving whether or not the UTF8-encoded character that
2457 * starts at <p>, and extending no further than <e - 1> is in the swash
2458 * indicated by <swashname>. <swash> contains a pointer to where the swash
2459 * indicated by <swashname> is to be stored; which this routine will do, so
2460 * that future calls will look at <*swash> and only generate a swash if it
2461 * is not null. <invlist> is NULL or an inversion list that defines the
2462 * swash. If not null, it saves time during initialization of the swash.
2463 */
2464
2465 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2466
2467 if (! isUTF8_CHAR(p, e)) {
2468 _force_out_malformed_utf8_message(p, e, 0, 1);
2469 NOT_REACHED; /* NOTREACHED */
2470 }
2471
2472 if (!*swash) {
2473 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2474 *swash = _core_swash_init("utf8",
2475
2476 /* Only use the name if there is no inversion
2477 * list; otherwise will go out to disk */
2478 (invlist) ? "" : swashname,
2479
2480 &PL_sv_undef, 1, 0, invlist, &flags);
2481 }
2482
2483 return swash_fetch(*swash, p, TRUE) != 0;
2484}
2485
bde6a22d 2486bool
922e8cb4
KW
2487Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2488{
922e8cb4
KW
2489 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2490
2491 assert(classnum < _FIRST_NON_SWASH_CC);
2492
f25ce844
KW
2493 return is_utf8_common(p,
2494 &PL_utf8_swash_ptrs[classnum],
2495 swash_property_names[classnum],
2496 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
2497}
2498
2499bool
da8c1a98
KW
2500Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2501 const U8 * const e)
2502{
2503 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2504
2505 assert(classnum < _FIRST_NON_SWASH_CC);
2506
2507 return is_utf8_common_with_len(p,
2508 e,
2509 &PL_utf8_swash_ptrs[classnum],
2510 swash_property_names[classnum],
2511 PL_XPosix_ptrs[classnum]);
2512}
2513
2514bool
f2645549 2515Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 2516{
f2645549 2517 SV* invlist = NULL;
7918f24d 2518
f2645549 2519 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 2520
f2645549
KW
2521 if (! PL_utf8_perl_idstart) {
2522 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2523 }
60071a22 2524 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
2525}
2526
2527bool
da8c1a98
KW
2528Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2529{
2530 SV* invlist = NULL;
2531
2532 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2533
2534 if (! PL_utf8_perl_idstart) {
2535 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2536 }
2537 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2538 "_Perl_IDStart", invlist);
2539}
2540
2541bool
f2645549 2542Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2543{
f2645549 2544 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2545
2546 if (*p == '_')
2547 return TRUE;
f25ce844 2548 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2549}
2550
2551bool
eba68aa0
KW
2552Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2553{
b24b43f7 2554 SV* invlist = NULL;
eba68aa0
KW
2555
2556 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2557
b24b43f7
KW
2558 if (! PL_utf8_perl_idcont) {
2559 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2560 }
60071a22 2561 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
2562}
2563
eba68aa0 2564bool
da8c1a98
KW
2565Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2566{
2567 SV* invlist = NULL;
2568
2569 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2570
2571 if (! PL_utf8_perl_idcont) {
2572 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2573 }
2574 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2575 "_Perl_IDCont", invlist);
2576}
2577
2578bool
f2645549 2579Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2580{
f2645549 2581 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2582
f25ce844 2583 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2584}
2585
2586bool
f2645549 2587Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2588{
f2645549 2589 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2590
f25ce844 2591 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2592}
2593
2594bool
7dbf68d2
KW
2595Perl__is_utf8_mark(pTHX_ const U8 *p)
2596{
7dbf68d2
KW
2597 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2598
f25ce844 2599 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2600}
2601
6b5c0936 2602/*
87cea99e 2603=for apidoc to_utf8_case
6b5c0936 2604
9da1e7cb
KW
2605Instead use the appropriate one of L</toUPPER_utf8>,
2606L</toTITLE_utf8>,
2607L</toLOWER_utf8>,
2608or L</toFOLD_utf8>.
2609
6fae5207 2610C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2611the character that is being converted. This routine assumes that the character
2612at C<p> is well-formed.
6b5c0936 2613
6fae5207
KW
2614C<ustrp> is a pointer to the character buffer to put the
2615conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2616of the result.
2617
6fae5207 2618C<swashp> is a pointer to the swash to use.
6b5c0936 2619
a1433954 2620Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2621and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2622but not always, a multicharacter mapping), is tried first.
6b5c0936 2623
4a8240a3
KW
2624C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2625any special mappings; C<""> means to use the special mappings. Values other
2626than these two are treated as the name of the hash containing the special
2627mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2628
796b6530
KW
2629C<normal> is a string like C<"ToLower"> which means the swash
2630C<%utf8::ToLower>.
0134edef 2631
760c7c2f
KW
2632Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2633unless those are turned off.
2634
0134edef 2635=cut */
6b5c0936 2636
2104c8d9 2637UV
9a957fbc
AL
2638Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2639 SV **swashp, const char *normal, const char *special)
a0ed51b3 2640{
b9992569
KW
2641 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2642
2643 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2644}
2645
2646 /* change namve uv1 to 'from' */
6a4a25f4 2647STATIC UV
b9992569
KW
2648S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2649 SV **swashp, const char *normal, const char *special)
2650{
0134edef 2651 STRLEN len = 0;
7918f24d 2652
b9992569 2653 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2654
36eaa811
KW
2655 /* For code points that don't change case, we already know that the output
2656 * of this function is the unchanged input, so we can skip doing look-ups
2657 * for them. Unfortunately the case-changing code points are scattered
2658 * around. But there are some long consecutive ranges where there are no
2659 * case changing code points. By adding tests, we can eliminate the lookup
2660 * for all the ones in such ranges. This is currently done here only for
2661 * just a few cases where the scripts are in common use in modern commerce
2662 * (and scripts adjacent to those which can be included without additional
2663 * tests). */
2664
2665 if (uv1 >= 0x0590) {
2666 /* This keeps from needing further processing the code points most
2667 * likely to be used in the following non-cased scripts: Hebrew,
2668 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2669 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2670 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2671 if (uv1 < 0x10A0) {
2672 goto cases_to_self;
2673 }
2674
2675 /* The following largish code point ranges also don't have case
2676 * changes, but khw didn't think they warranted extra tests to speed
2677 * them up (which would slightly slow down everything else above them):
2678 * 1100..139F Hangul Jamo, Ethiopic
2679 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2680 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2681 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2682 * Combining Diacritical Marks Extended, Balinese,
2683 * Sundanese, Batak, Lepcha, Ol Chiki
2684 * 2000..206F General Punctuation
2685 */
2686
2687 if (uv1 >= 0x2D30) {
2688
2689 /* This keeps the from needing further processing the code points
2690 * most likely to be used in the following non-cased major scripts:
2691 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2692 *
2693 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2694 * event that Unicode eventually allocates the unused block as of
2695 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2696 * that the test suite will start having failures to alert you
2697 * should that happen) */
2698 if (uv1 < 0xA640) {
2699 goto cases_to_self;
2700 }
2701
2702 if (uv1 >= 0xAC00) {
2703 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2704 if (ckWARN_d(WARN_SURROGATE)) {
2705 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2706 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2707 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2708 }
2709 goto cases_to_self;
2710 }
36eaa811
KW
2711
2712 /* AC00..FAFF Catches Hangul syllables and private use, plus
2713 * some others */
2714 if (uv1 < 0xFB00) {
2715 goto cases_to_self;
2716
2717 }
2718
5af9bc97
KW
2719 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2720 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2721 && ckWARN_d(WARN_DEPRECATED))
2722 {
2723 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2724 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2725 }
2726 if (ckWARN_d(WARN_NON_UNICODE)) {
2727 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2728 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2729 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2730 }
2731 goto cases_to_self;
2732 }
3bfc1e70
KW
2733#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2734 if (UNLIKELY(uv1
2735 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2736 {
2737
2738 /* As of this writing, this means we avoid swash creation
2739 * for anything beyond low Plane 1 */
2740 goto cases_to_self;
2741 }
2742#endif
36eaa811
KW
2743 }
2744 }
9ae3ac1a 2745
36eaa811
KW
2746 /* Note that non-characters are perfectly legal, so no warning should
2747 * be given. There are so few of them, that it isn't worth the extra
2748 * tests to avoid swash creation */
9ae3ac1a
KW
2749 }
2750
0134edef 2751 if (!*swashp) /* load on-demand */
5ab9d2ef 2752 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2753
a6f87d8c 2754 if (special) {
0134edef 2755 /* It might be "special" (sometimes, but not always,
2a37f04d 2756 * a multicharacter mapping) */
4a8240a3 2757 HV *hv = NULL;
b08cf34e
JH
2758 SV **svp;
2759
4a8240a3
KW
2760 /* If passed in the specials name, use that; otherwise use any
2761 * given in the swash */
2762 if (*special != '\0') {
2763 hv = get_hv(special, 0);
2764 }
2765 else {
2766 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2767 if (svp) {
2768 hv = MUTABLE_HV(SvRV(*svp));
2769 }
2770 }
2771
176fe009 2772 if (hv
5f560d8a 2773 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2774 && (*svp))
2775 {
cfd0369c 2776 const char *s;
47654450 2777
cfd0369c 2778 s = SvPV_const(*svp, len);
47654450 2779 if (len == 1)
f4cd282c 2780 /* EIGHTBIT */
c80e42f3 2781 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2782 else {
d2dcd0fb 2783 Copy(s, ustrp, len, U8);
29e98929 2784 }
983ffd37 2785 }
0134edef
JH
2786 }
2787
2788 if (!len && *swashp) {
4a4088c4 2789 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2790
0134edef
JH
2791 if (uv2) {
2792 /* It was "normal" (a single character mapping). */
f4cd282c 2793 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2794 }
2795 }
1feea2c7 2796
cbe07460
KW
2797 if (len) {
2798 if (lenp) {
2799 *lenp = len;
2800 }
2801 return valid_utf8_to_uvchr(ustrp, 0);
2802 }
2803
2804 /* Here, there was no mapping defined, which means that the code point maps
2805 * to itself. Return the inputs */
e24dfe9c 2806 cases_to_self:
bfdf22ec 2807 len = UTF8SKIP(p);
ca9fab46
KW
2808 if (p != ustrp) { /* Don't copy onto itself */
2809 Copy(p, ustrp, len, U8);
2810 }
0134edef 2811
2a37f04d
JH
2812 if (lenp)
2813 *lenp = len;
2814
f4cd282c 2815 return uv1;
cbe07460 2816
a0ed51b3
LW
2817}
2818
051a06d4 2819STATIC UV
357aadde 2820S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2821{
4a4088c4 2822 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2823 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2824 * result contains a character that crosses the 255/256 boundary, disallow
2825 * the change, and return the original code point. See L<perlfunc/lc> for
2826 * why;
051a06d4 2827 *
a1433954
KW
2828 * p points to the original string whose case was changed; assumed
2829 * by this routine to be well-formed
051a06d4
KW
2830 * result the code point of the first character in the changed-case string
2831 * ustrp points to the changed-case string (<result> represents its first char)
2832 * lenp points to the length of <ustrp> */
2833
2834 UV original; /* To store the first code point of <p> */
2835
2836 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2837
a4f12ed7 2838 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2839
2840 /* We know immediately if the first character in the string crosses the
2841 * boundary, so can skip */
2842 if (result > 255) {
2843
2844 /* Look at every character in the result; if any cross the
2845 * boundary, the whole thing is disallowed */
2846 U8* s = ustrp + UTF8SKIP(ustrp);
2847 U8* e = ustrp + *lenp;
2848 while (s < e) {
a4f12ed7 2849 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2850 goto bad_crossing;
2851 }
2852 s += UTF8SKIP(s);
2853 }
2854
613abc6d
KW
2855 /* Here, no characters crossed, result is ok as-is, but we warn. */
2856 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2857 return result;
2858 }
2859
7b52d656 2860 bad_crossing:
051a06d4
KW
2861
2862 /* Failed, have to return the original */
4b88fb76 2863 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2864
2865 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2866 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2867 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2868 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2869 OP_DESC(PL_op),
ab0b796c
KW
2870 original,
2871 original);
051a06d4
KW
2872 Copy(p, ustrp, *lenp, char);
2873 return original;
2874}
2875
eaf412bf
KW
2876/* The process for changing the case is essentially the same for the four case
2877 * change types, except there are complications for folding. Otherwise the
2878 * difference is only which case to change to. To make sure that they all do
2879 * the same thing, the bodies of the functions are extracted out into the
2880 * following two macros. The functions are written with the same variable
2881 * names, and these are known and used inside these macros. It would be
2882 * better, of course, to have inline functions to do it, but since different
2883 * macros are called, depending on which case is being changed to, this is not
2884 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2885 * function can start with the common start macro, then finish with its special
2886 * handling; while the other three cases can just use the common end macro.
2887 *
2888 * The algorithm is to use the proper (passed in) macro or function to change
2889 * the case for code points that are below 256. The macro is used if using
2890 * locale rules for the case change; the function if not. If the code point is
2891 * above 255, it is computed from the input UTF-8, and another macro is called
2892 * to do the conversion. If necessary, the output is converted to UTF-8. If
2893 * using a locale, we have to check that the change did not cross the 255/256
2894 * boundary, see check_locale_boundary_crossing() for further details.
2895 *
2896 * The macros are split with the correct case change for the below-256 case
2897 * stored into 'result', and in the middle of an else clause for the above-255
2898 * case. At that point in the 'else', 'result' is not the final result, but is
2899 * the input code point calculated from the UTF-8. The fold code needs to
2900 * realize all this and take it from there.
2901 *
2902 * If you read the two macros as sequential, it's easier to understand what's
2903 * going on. */
2904#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2905 L1_func_extra_param) \
2906 if (flags & (locale_flags)) { \
2907 /* Treat a UTF-8 locale as not being in locale at all */ \
2908 if (IN_UTF8_CTYPE_LOCALE) { \
2909 flags &= ~(locale_flags); \
2910 } \
2911 else { \
2912 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2913 } \
2914 } \
2915 \
2916 if (UTF8_IS_INVARIANT(*p)) { \
2917 if (flags & (locale_flags)) { \
2918 result = LC_L1_change_macro(*p); \
2919 } \
2920 else { \
2921 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2922 } \
2923 } \
2924 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2925 if (flags & (locale_flags)) { \
2926 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2927 *(p+1))); \
2928 } \
2929 else { \
2930 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2931 ustrp, lenp, L1_func_extra_param); \
2932 } \
2933 } \
2934 else { /* malformed UTF-8 */ \
2935 result = valid_utf8_to_uvchr(p, NULL); \
2936
2937#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2938 result = change_macro(result, p, ustrp, lenp); \
2939 \
2940 if (flags & (locale_flags)) { \
2941 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2942 } \
2943 return result; \
2944 } \
2945 \
2946 /* Here, used locale rules. Convert back to UTF-8 */ \
2947 if (UTF8_IS_INVARIANT(result)) { \
2948 *ustrp = (U8) result; \
2949 *lenp = 1; \
2950 } \
2951 else { \
2952 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2953 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2954 *lenp = 2; \
2955 } \
2956 \
2957 return result;
2958
d3e79532 2959/*
87cea99e 2960=for apidoc to_utf8_upper
d3e79532 2961
1f607577 2962Instead use L</toUPPER_utf8>.
a1433954 2963
d3e79532
JH
2964=cut */
2965
051a06d4 2966/* Not currently externally documented, and subject to change:
31f05a37
KW
2967 * <flags> is set iff iff the rules from the current underlying locale are to
2968 * be used. */
051a06d4 2969
2104c8d9 2970UV
31f05a37 2971Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2972{
051a06d4
KW
2973 UV result;
2974
2975 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2976
eaf412bf
KW
2977 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
2978 /* 2nd char of uc(U+DF) is 'S' */
2979 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
2980 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 2981}
a0ed51b3 2982
d3e79532 2983/*
87cea99e 2984=for apidoc to_utf8_title
d3e79532 2985
1f607577 2986Instead use L</toTITLE_utf8>.
a1433954 2987
d3e79532
JH
2988=cut */
2989
051a06d4 2990/* Not currently externally documented, and subject to change:
31f05a37
KW
2991 * <flags> is set iff the rules from the current underlying locale are to be
2992 * used. Since titlecase is not defined in POSIX, for other than a
2993 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2994 */
051a06d4 2995
983ffd37 2996UV
31f05a37 2997Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 2998{
051a06d4
KW
2999 UV result;
3000
3001 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3002
eaf412bf
KW
3003 /* 2nd char of ucfirst(U+DF) is 's' */
3004 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3005 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3006}
3007
d3e79532 3008/*
87cea99e 3009=for apidoc to_utf8_lower
d3e79532 3010
1f607577 3011Instead use L</toLOWER_utf8>.
a1433954 3012
d3e79532
JH
3013=cut */
3014
051a06d4 3015/* Not currently externally documented, and subject to change:
31f05a37
KW
3016 * <flags> is set iff iff the rules from the current underlying locale are to
3017 * be used.
3018 */
051a06d4 3019
2104c8d9 3020UV
31f05a37 3021Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3022{
051a06d4
KW
3023 UV result;
3024
051a06d4 3025 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3026
eaf412bf
KW
3027 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3028 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3029}
3030
d3e79532 3031/*
87cea99e 3032=for apidoc to_utf8_fold
d3e79532 3033
1f607577 3034Instead use L</toFOLD_utf8>.
a1433954 3035
d3e79532
JH
3036=cut */
3037
051a06d4
KW
3038/* Not currently externally documented, and subject to change,
3039 * in <flags>
31f05a37
KW
3040 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3041 * locale are to be used.
051a06d4
KW
3042 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3043 * otherwise simple folds
a0270393
KW
3044 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3045 * prohibited
445bf929 3046 */
36bb2ab6 3047
b4e400f9 3048UV
445bf929 3049Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3050{
051a06d4
KW
3051 UV result;
3052
36bb2ab6 3053 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3054
a0270393
KW
3055 /* These are mutually exclusive */
3056 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3057
50ba90ff
KW
3058 assert(p != ustrp); /* Otherwise overwrites */
3059
eaf412bf
KW
3060 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3061 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3062
eaf412bf 3063 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3064
1ca267a5
KW
3065 if (flags & FOLD_FLAGS_LOCALE) {
3066
76f2ffcd 3067# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3068 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3069
3070# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3071# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3072
3073 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3074
538e84ed
KW
3075 /* Special case these two characters, as what normally gets
3076 * returned under locale doesn't work */
76f2ffcd
KW
3077 if (UTF8SKIP(p) == cap_sharp_s_len
3078 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3079 {
ab0b796c
KW
3080 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3081 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3082 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3083 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3084 goto return_long_s;
3085 }
0766489e
KW
3086 else
3087#endif
3088 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3089 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3090 {
ab0b796c
KW
3091 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3092 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3093 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3094 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3095 goto return_ligature_st;
3096 }
74894415
KW
3097
3098#if UNICODE_MAJOR_VERSION == 3 \
3099 && UNICODE_DOT_VERSION == 0 \
3100 && UNICODE_DOT_DOT_VERSION == 1
3101# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3102
3103 /* And special case this on this Unicode version only, for the same
3104 * reaons the other two are special cased. They would cross the
3105 * 255/256 boundary which is forbidden under /l, and so the code
3106 * wouldn't catch that they are equivalent (which they are only in
3107 * this release) */
3108 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3109 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3110 {
3111 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3112 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3113 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3114 "resolved to \"\\x{0131}\".");
3115 goto return_dotless_i;
3116 }
3117#endif
3118
357aadde 3119 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3120 }
a0270393
KW
3121 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3122 return result;
3123 }
3124 else {
4a4088c4 3125 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3126 * character above the ASCII range, and the result should not
3127 * contain an ASCII character. */
a0270393
KW
3128
3129 UV original; /* To store the first code point of <p> */
3130
3131 /* Look at every character in the result; if any cross the
3132 * boundary, the whole thing is disallowed */
3133 U8* s = ustrp;
3134 U8* e = ustrp + *lenp;
3135 while (s < e) {
3136 if (isASCII(*s)) {
3137 /* Crossed, have to return the original */
3138 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3139
9fc2026f 3140 /* But in these instances, there is an alternative we can
1ca267a5 3141 * return that is valid */
0766489e
KW
3142 if (original == LATIN_SMALL_LETTER_SHARP_S
3143#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3144 || original == LATIN_CAPITAL_LETTER_SHARP_S
3145#endif
3146 ) {
1ca267a5
KW
3147 goto return_long_s;
3148 }
9fc2026f
KW
3149 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3150 goto return_ligature_st;
3151 }
74894415
KW
3152#if UNICODE_MAJOR_VERSION == 3 \
3153 && UNICODE_DOT_VERSION == 0 \
3154 && UNICODE_DOT_DOT_VERSION == 1
3155
3156 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3157 goto return_dotless_i;
3158 }
3159#endif
a0270393
KW
3160 Copy(p, ustrp, *lenp, char);
3161 return original;
3162 }
3163 s += UTF8SKIP(s);
3164 }
051a06d4 3165
a0270393
KW
3166 /* Here, no characters crossed, result is ok as-is */
3167 return result;
3168 }
051a06d4
KW
3169 }
3170
4a4088c4 3171 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3172 if (UTF8_IS_INVARIANT(result)) {
3173 *ustrp = (U8) result;
3174 *lenp = 1;
3175 }
3176 else {
62cb07ea
KW
3177 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3178 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3179 *lenp = 2;
3180 }
3181
051a06d4 3182 return result;
1ca267a5
KW
3183
3184 return_long_s:
3185 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3186 * folds to a string of two of these characters. By returning this
3187 * instead, then, e.g.,
3188 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3189 * works. */
3190
3191 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3192 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3193 ustrp, *lenp, U8);
3194 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3195
3196 return_ligature_st:
3197 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3198 * have the other one fold to it */
3199
3200 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3201 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3202 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3203
3204#if UNICODE_MAJOR_VERSION == 3 \
3205 && UNICODE_DOT_VERSION == 0 \
3206 && UNICODE_DOT_DOT_VERSION == 1
3207
3208 return_dotless_i:
3209 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3210 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3211 return LATIN_SMALL_LETTER_DOTLESS_I;
3212
3213#endif
3214
a0ed51b3
LW
3215}
3216
711a919c 3217/* Note:
f90a9a02 3218 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3219 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3220 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3221 */
c4a5db0c 3222
a0ed51b3 3223SV*
7fc63493 3224Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3225{
c4a5db0c
KW
3226 PERL_ARGS_ASSERT_SWASH_INIT;
3227
3228 /* Returns a copy of a swash initiated by the called function. This is the
3229 * public interface, and returning a copy prevents others from doing
3230 * mischief on the original */
3231
5d3d13d1 3232 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3233}
3234
3235SV*
5d3d13d1 3236Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3237{
2c1f00b9
YO
3238
3239 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3240 * use the following define */
3241
3242#define CORE_SWASH_INIT_RETURN(x) \
3243 PL_curpm= old_PL_curpm; \
3244 return x
3245
c4a5db0c 3246 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3247 * by calling utf8_heavy.pl in the general case. The returned value may be
3248 * the swash's inversion list instead if the input parameters allow it.
3249 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3250 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3251 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3252 *
3253 * This interface should only be used by functions that won't destroy or
3254 * adversely change the swash, as doing so affects all other uses of the
3255 * swash in the program; the general public should use 'Perl_swash_init'
3256 * instead.
3257 *
3258 * pkg is the name of the package that <name> should be in.
3259 * name is the name of the swash to find. Typically it is a Unicode
3260 * property name, including user-defined ones
3261 * listsv is a string to initialize the swash with. It must be of the form
3262 * documented as the subroutine return value in
3263 * L<perlunicode/User-Defined Character Properties>
3264 * minbits is the number of bits required to represent each data element.
3265 * It is '1' for binary properties.
3266 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3267 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3268 * flags_p if non-NULL is the address of various input and output flag bits
3269 * to the routine, as follows: ('I' means is input to the routine;
3270 * 'O' means output from the routine. Only flags marked O are
3271 * meaningful on return.)
3272 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3273 * came from a user-defined property. (I O)
5d3d13d1
KW
3274 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3275 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3276 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3277 * return of an inversion list instead of a swash hash if this routine
3278 * thinks that would result in faster execution of swash_fetch() later
3279 * on. (I)
9a53f6cf
KW
3280 *
3281 * Thus there are three possible inputs to find the swash: <name>,
3282 * <listsv>, and <invlist>. At least one must be specified. The result
3283 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3284 * actions can intersect, etc. what <name> gives. To avoid going out to
3285 * disk at all, <invlist> should specify completely what the swash should
3286 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3287 *
3288 * <invlist> is only valid for binary properties */
c4a5db0c 3289
2c1f00b9
YO
3290 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3291
c4a5db0c 3292 SV* retval = &PL_sv_undef;
83199d38 3293 HV* swash_hv = NULL;
87367d5f
KW
3294 const int invlist_swash_boundary =
3295 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3296 ? 512 /* Based on some benchmarking, but not extensive, see commit
3297 message */
3298 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3299
3300 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3301 assert(! invlist || minbits == 1);
3302
2c1f00b9
YO
3303 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3304 that triggered the swash init and the swash init perl logic itself.
3305 See perl #122747 */
3306
9a53f6cf
KW
3307 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3308 * so */
3309 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3310 dSP;
3311 const size_t pkg_len = strlen(pkg);
3312 const size_t name_len = strlen(name);
3313 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3314 SV* errsv_save;
3315 GV *method;
3316
3317 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3318
3319 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3320 ENTER;
69794297 3321 SAVEHINTS();
2782061f 3322 save_re_context();
650f067c
JL
3323 /* We might get here via a subroutine signature which uses a utf8
3324 * parameter name, at which point PL_subname will have been set
3325 * but not yet used. */
3326 save_item(PL_subname);
69794297
KW
3327 if (PL_parser && PL_parser->error_count)
3328 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3329 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3330 if (!method) { /* demand load UTF-8 */
69794297 3331 ENTER;
db2c6cb3
FC
3332 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3333 GvSV(PL_errgv) = NULL;
1a419e6b 3334#ifndef NO_TAINT_SUPPORT
69794297
KW
3335 /* It is assumed that callers of this routine are not passing in
3336 * any user derived data. */
2782061f
DM
3337 /* Need to do this after save_re_context() as it will set
3338 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3339 * in Perl_magic_get). Even line to create errsv_save can turn on
3340 * PL_tainted. */
284167a5
SM
3341 SAVEBOOL(TAINT_get);
3342 TAINT_NOT;
3343#endif
69794297
KW
3344 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3345 NULL);
eed484f9 3346 {
db2c6cb3
FC
3347 /* Not ERRSV, as there is no need to vivify a scalar we are
3348 about to discard. */
3349 SV * const errsv = GvSV(PL_errgv);
3350 if (!SvTRUE(errsv)) {
3351 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3352 SvREFCNT_dec(errsv);
3353 }
eed484f9 3354 }
69794297
KW
3355 LEAVE;
3356 }
3357 SPAGAIN;
3358 PUSHMARK(SP);
3359 EXTEND(SP,5);
3360 mPUSHp(pkg, pkg_len);
3361 mPUSHp(name, name_len);
3362 PUSHs(listsv);
3363 mPUSHi(minbits);
3364 mPUSHi(none);
3365 PUTBACK;
db2c6cb3
FC
3366 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3367 GvSV(PL_errgv) = NULL;
69794297
KW
3368 /* If we already have a pointer to the method, no need to use
3369 * call_method() to repeat the lookup. */
c41800a8
KW
3370 if (method
3371 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3372 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3373 {
3374 retval = *PL_stack_sp--;
3375 SvREFCNT_inc(retval);
3376 }
eed484f9 3377 {
db2c6cb3
FC
3378 /* Not ERRSV. See above. */
3379 SV * const errsv = GvSV(PL_errgv);
3380 if (!SvTRUE(errsv)) {
3381 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3382 SvREFCNT_dec(errsv);
3383 }
eed484f9 3384 }
ce3b816e 3385 LEAVE;
69794297
KW
3386 POPSTACK;
3387 if (IN_PERL_COMPILETIME) {
3388 CopHINTS_set(PL_curcop, PL_hints);
3389 }
3390 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3391 if (SvPOK(retval)) {
69794297
KW
3392
3393 /* If caller wants to handle missing properties, let them */
5d3d13d1 3394 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3395 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3396 }
3397 Perl_croak(aTHX_
147e3846 3398 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3399 SVfARG(retval));
a25b5927 3400 NOT_REACHED; /* NOTREACHED */
d95e4a00 3401 }
69794297 3402 }
9a53f6cf 3403 } /* End of calling the module to find the swash */
36eb48b4 3404
83199d38
KW
3405 /* If this operation fetched a swash, and we will need it later, get it */
3406 if (retval != &PL_sv_undef
3407 && (minbits == 1 || (flags_p
3408 && ! (*flags_p
3409 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3410 {
3411 swash_hv = MUTABLE_HV(SvRV(retval));
3412
3413 /* If we don't already know that there is a user-defined component to
3414 * this swash, and the user has indicated they wish to know if there is
3415 * one (by passing <flags_p>), find out */
3416 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3417 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3418 if (user_defined && SvUV(*user_defined)) {
3419 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3420 }
3421 }
3422 }
3423
36eb48b4
KW
3424 /* Make sure there is an inversion list for binary properties */
3425 if (minbits == 1) {
3426 SV** swash_invlistsvp = NULL;
3427 SV* swash_invlist = NULL;
9a53f6cf 3428 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3429 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3430 an unclaimed reference count */
36eb48b4 3431
9a53f6cf 3432 /* If this operation fetched a swash, get its already existing
83199d38 3433 * inversion list, or create one for it */
36eb48b4 3434
83199d38 3435 if (swash_hv) {
5c9f4bd2 3436 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3437 if (swash_invlistsvp) {
3438 swash_invlist = *swash_invlistsvp;
3439 invlist_in_swash_is_valid = TRUE;
3440 }
3441 else {
36eb48b4 3442 swash_invlist = _swash_to_invlist(retval);
02c85471 3443 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3444 }
3445 }
3446
3447 /* If an inversion list was passed in, have to include it */
3448 if (invlist) {
3449
3450 /* Any fetched swash will by now have an inversion list in it;
3451 * otherwise <swash_invlist> will be NULL, indicating that we
3452 * didn't fetch a swash */
3453 if (swash_invlist) {
3454
3455 /* Add the passed-in inversion list, which invalidates the one
3456 * already stored in the swash */
3457 invlist_in_swash_is_valid = FALSE;
eee4c920 3458 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3459 _invlist_union(invlist, swash_invlist, &swash_invlist);
3460 }
3461 else {
3462
87367d5f
KW
3463 /* Here, there is no swash already. Set up a minimal one, if
3464 * we are going to return a swash */
3465 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3466 swash_hv = newHV();
4aca0fe6 3467 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3468 }
9a53f6cf
KW
3469 swash_invlist = invlist;
3470 }
9a53f6cf
KW
3471 }
3472
3473 /* Here, we have computed the union of all the passed-in data. It may
3474 * be that there was an inversion list in the swash which didn't get
538e84ed 3475 * touched; otherwise save the computed one */
87367d5f
KW
3476 if (! invlist_in_swash_is_valid
3477 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3478 {
5c9f4bd2 3479 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3480 {
3481 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3482 }
cc34d8c5
FC
3483 /* We just stole a reference count. */
3484 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3485 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3486 }
87367d5f 3487
ee3222e3 3488 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3489 SvREADONLY_on(swash_invlist);
3490
c41800a8 3491 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3492 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3493 SvREFCNT_dec(retval);
02c85471
FC
3494 if (!swash_invlist_unclaimed)
3495 SvREFCNT_inc_simple_void_NN(swash_invlist);
3496 retval = newRV_noinc(swash_invlist);
87367d5f 3497 }
36eb48b4
KW
3498 }
3499
2c1f00b9
YO
3500 CORE_SWASH_INIT_RETURN(retval);
3501#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3502}
3503
035d37be
JH
3504
3505/* This API is wrong for special case conversions since we may need to
3506 * return several Unicode characters for a single Unicode character
3507 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3508 * the lower-level routine, and it is similarly broken for returning
38684baa 3509 * multiple values. --jhi
b9992569 3510 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3511/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3512
3513/* Note:
3514 * Returns the value of property/mapping C<swash> for the first character
3515 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3516 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3517 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3518 *
3519 * A "swash" is a hash which contains initially the keys/values set up by
3520 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3521 * property for all possible code points. Things are stored in a compact form
3522 * (see utf8_heavy.pl) so that calculation is required to find the actual
3523 * property value for a given code point. As code points are looked up, new
3524 * key/value pairs are added to the hash, so that the calculation doesn't have
3525 * to ever be re-done. Further, each calculation is done, not just for the
3526 * desired one, but for a whole block of code points adjacent to that one.
3527 * For binary properties on ASCII machines, the block is usually for 64 code
3528 * points, starting with a code point evenly divisible by 64. Thus if the
3529 * property value for code point 257 is requested, the code goes out and
3530 * calculates the property values for all 64 code points between 256 and 319,
3531 * and stores these as a single 64-bit long bit vector, called a "swatch",
3532 * under the key for code point 256. The key is the UTF-8 encoding for code
3533 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3534 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3535 * for code point 258 is then requested, this code realizes that it would be
3536 * stored under the key for 256, and would find that value and extract the
3537 * relevant bit, offset from 256.
3538 *
3539 * Non-binary properties are stored in as many bits as necessary to represent
3540 * their values (32 currently, though the code is more general than that), not
fc273927 3541 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3542 * vector that encompasses the property values for all code points whose UTF-8
3543 * representations are represented by the key. That is, for all code points
3544 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3545 * bytes of that.
680c470c 3546 */
a0ed51b3 3547UV
680c470c 3548Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3549{
ef8f7699 3550 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3551 U32 klen;
3552 U32 off;
9b56a019 3553 STRLEN slen = 0;
7d85a32c 3554 STRLEN needents;
cfd0369c 3555 const U8 *tmps = NULL;
979f2922 3556 SV *swatch;
08fb1ac5 3557 const U8 c = *ptr;
3568d838 3558
7918f24d
NC
3559 PERL_ARGS_ASSERT_SWASH_FETCH;
3560
87367d5f
KW
3561 /* If it really isn't a hash, it isn't really swash; must be an inversion
3562 * list */
3563 if (SvTYPE(hv) != SVt_PVHV) {
3564 return _invlist_contains_cp((SV*)hv,
3565 (do_utf8)
3566 ? valid_utf8_to_uvchr(ptr, NULL)
3567 : c);
3568 }
3569
08fb1ac5
KW
3570 /* We store the values in a "swatch" which is a vec() value in a swash
3571 * hash. Code points 0-255 are a single vec() stored with key length
3572 * (klen) 0. All other code points have a UTF-8 representation
3573 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3574 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3575 * length for them is the length of the encoded char - 1. ptr[klen] is the
3576 * final byte in the sequence representing the character */
3577 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3578 klen = 0;
3579 needents = 256;
3580 off = c;
3568d838 3581 }
08fb1ac5
KW
3582 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3583 klen = 0;
3584 needents = 256;
a62b247b 3585 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3586 }
3587 else {
08fb1ac5
KW
3588 klen = UTF8SKIP(ptr) - 1;
3589
3590 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3591 * the vec is the final byte in the sequence. (In EBCDIC this is
3592 * converted to I8 to get consecutive values.) To help you visualize
3593 * all this:
3594 * Straight 1047 After final byte
3595 * UTF-8 UTF-EBCDIC I8 transform
3596 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3597 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3598 * ...
3599 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3600 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3601 * ...
3602 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3603 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3604 * ...
3605 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3606 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3607 * ...
3608 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3609 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3610 *
3611 * (There are no discontinuities in the elided (...) entries.)
3612 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3613 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3614 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3615 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3616 * index into the vec() swatch (after subtracting 0x80, which we
3617 * actually do with an '&').
3618 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3619 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3620 * dicontinuities which go away by transforming it into I8, and we
3621 * effectively subtract 0xA0 to get the index. */
979f2922 3622 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3623 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3624 }
7d85a32c 3625
a0ed51b3 3626 /*
4a4088c4 3627 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
3628 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3629 * it's nothing to sniff at.) Pity we usually come through at least
3630 * two function calls to get here...
3631 *
3632 * NB: this code assumes that swatches are never modified, once generated!
3633 */
3634
3568d838 3635 if (hv == PL_last_swash_hv &&
a0ed51b3 3636 klen == PL_last_swash_klen &&
27da23d5 3637 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3638 {
3639 tmps = PL_last_swash_tmps;
3640 slen = PL_last_swash_slen;
3641 }
3642 else {
3643 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3644 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3645
b0e3252e 3646 /* If not cached, generate it via swatch_get */
979f2922 3647 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3648 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3649 {
3650 if (klen) {
3651 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3652 swatch = swatch_get(swash,
3653 code_point & ~((UV)needents - 1),
3654 needents);
3655 }
3656 else { /* For the first 256 code points, the swatch has a key of
3657 length 0 */
3658 swatch = swatch_get(swash, 0, needents);
3659 }
979f2922 3660
923e4eb5 3661 if (IN_PERL_COMPILETIME)
623e6609 3662 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3663
979f2922 3664 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3665
979f2922
ST
3666 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3667 || (slen << 3) < needents)
5637ef5b 3668 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
147e3846 3669 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
5637ef5b 3670 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3671 }
3672
3673 PL_last_swash_hv = hv;
16d8f38a 3674 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3675 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3676 /* FIXME change interpvar.h? */
3677 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3678 PL_last_swash_slen = slen;
3679 if (klen)
3680 Copy(ptr, PL_last_swash_key, klen, U8);
3681 }
3682
9faf8d75 3683 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3684 case 1:
e7aca353 3685 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3686 case 8:
e7aca353 3687 return ((UV) tmps[off]);
a0ed51b3
LW
3688 case 16:
3689 off <<= 1;
e7aca353
JH
3690 return
3691 ((UV) tmps[off ] << 8) +
3692 ((UV) tmps[off + 1]);
a0ed51b3
LW
3693 case 32:
3694 off <<= 2;
e7aca353
JH
3695 return
3696 ((UV) tmps[off ] << 24) +
3697 ((UV) tmps[off + 1] << 16) +
3698 ((UV) tmps[off + 2] << 8) +
3699 ((UV) tmps[off + 3]);
a0ed51b3 3700 }
5637ef5b 3701 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
147e3846 3702 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
670f1322 3703 NORETURN_FUNCTION_END;
a0ed51b3 3704}
2b9d42f0 3705
319009ee
KW
3706/* Read a single line of the main body of the swash input text. These are of
3707 * the form:
3708 * 0053 0056 0073
3709 * where each number is hex. The first two numbers form the minimum and
3710 * maximum of a range, and the third is the value associated with the range.
3711 * Not all swashes should have a third number
3712 *
3713 * On input: l points to the beginning of the line to be examined; it points
3714 * to somewhere in the string of the whole input text, and is
3715 * terminated by a \n or the null string terminator.
3716 * lend points to the null terminator of that string
3717 * wants_value is non-zero if the swash expects a third number
3718 * typestr is the name of the swash's mapping, like 'ToLower'
3719 * On output: *min, *max, and *val are set to the values read from the line.
3720 * returns a pointer just beyond the line examined. If there was no
3721 * valid min number on the line, returns lend+1
3722 */
3723
3724STATIC U8*
3725S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3726 const bool wants_value, const U8* const typestr)
3727