This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Allow allowing UTF-8 overflow malformation
[perl5.git] / utf8.c
CommitLineData
a0ed51b3
LW
1/* utf8.c
2 *
1129b882 3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
b94e2f88 4 * by Larry Wall and others
a0ed51b3
LW
5 *
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
8 *
9 */
10
11/*
4ac71550
TC
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
15 *
cdad3b53 16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
a0ed51b3
LW
17 *
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
4ac71550 20 * as is the custom in the West, if you wish to be answered?'
cdad3b53 21 * --Gandalf, addressing Théoden's door wardens
4ac71550
TC
22 *
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
24 *
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
4ac71550
TC
27 *
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
a0ed51b3
LW
29 */
30
31#include "EXTERN.h"
864dbfa3 32#define PERL_IN_UTF8_C
a0ed51b3 33#include "perl.h"
b992490d 34#include "invlist_inline.h"
a0ed51b3 35
806547a7 36static const char malformed_text[] = "Malformed UTF-8 character";
27da23d5 37static const char unees[] =
806547a7 38 "Malformed UTF-8 character (unexpected end of string)";
760c7c2f 39static const char cp_above_legal_max[] =
147e3846 40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf;
760c7c2f 41
114d9c4d 42#define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
901b21bf 43
48ef279e 44/*
ccfc67b7 45=head1 Unicode Support
7fefc6c1 46These are various utility functions for manipulating UTF8-encoded
72d33970 47strings. For the uninitiated, this is a method of representing arbitrary
61296642 48Unicode characters as a variable number of bytes, in such a way that
56da48f7
DM
49characters in the ASCII range are unmodified, and a zero byte never appears
50within non-zero characters.
166f8a29 51
eaf7a4d2
CS
52=cut
53*/
54
9cbfb8ab
KW
55void
56Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
59 multiple chars */
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
63{
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
72 *
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
75 * die themselves */
76 U32 errors;
77
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
79
80 ENTER;
81 SAVESPTR(PL_dowarn);
82 SAVESPTR(PL_curcop);
83
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
85 if (PL_curcop) {
86 PL_curcop->cop_warnings = pWARN_ALL;
87 }
88
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
90
91 LEAVE;
92
93 if (! errors) {
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
96 }
97
98 if (die_here) {
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
100 }
101}
102
eaf7a4d2 103/*
378516de 104=for apidoc uvoffuni_to_utf8_flags
eebe1485 105
a27992cc 106THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af
KW
107Instead, B<Almost all code should use L</uvchr_to_utf8> or
108L</uvchr_to_utf8_flags>>.
a27992cc 109
de69f3af
KW
110This function is like them, but the input is a strict Unicode
111(as opposed to native) code point. Only in very rare circumstances should code
112not be using the native code point.
949cf498 113
efa9cd84 114For details, see the description for L</uvchr_to_utf8_flags>.
949cf498 115
eebe1485
SC
116=cut
117*/
118
8ee1cdcb
KW
119#define HANDLE_UNICODE_SURROGATE(uv, flags) \
120 STMT_START { \
121 if (flags & UNICODE_WARN_SURROGATE) { \
122 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
147e3846 123 "UTF-16 surrogate U+%04" UVXf, uv); \
8ee1cdcb
KW
124 } \
125 if (flags & UNICODE_DISALLOW_SURROGATE) { \
126 return NULL; \
127 } \
128 } STMT_END;
129
130#define HANDLE_UNICODE_NONCHAR(uv, flags) \
131 STMT_START { \
132 if (flags & UNICODE_WARN_NONCHAR) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
147e3846 134 "Unicode non-character U+%04" UVXf " is not " \
8ee1cdcb
KW
135 "recommended for open interchange", uv); \
136 } \
137 if (flags & UNICODE_DISALLOW_NONCHAR) { \
138 return NULL; \
139 } \
140 } STMT_END;
141
ba6ed43c
KW
142/* Use shorter names internally in this file */
143#define SHIFT UTF_ACCUMULATION_SHIFT
144#undef MARK
145#define MARK UTF_CONTINUATION_MARK
146#define MASK UTF_CONTINUATION_MASK
147
dfe13c55 148U8 *
378516de 149Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
a0ed51b3 150{
378516de 151 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
7918f24d 152
2d1545e5 153 if (OFFUNI_IS_INVARIANT(uv)) {
4c8cd605 154 *d++ = LATIN1_TO_NATIVE(uv);
d9432125
KW
155 return d;
156 }
facc1dc2 157
3ea68d71 158 if (uv <= MAX_UTF8_TWO_BYTE) {
facc1dc2
KW
159 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
160 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
3ea68d71
KW
161 return d;
162 }
d9432125 163
ba6ed43c
KW
164 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
165 * below, the 16 is for start bytes E0-EF (which are all the possible ones
166 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
167 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
168 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
169 * 0x800-0xFFFF on ASCII */
170 if (uv < (16 * (1U << (2 * SHIFT)))) {
171 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
172 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
173 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
174
175#ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
176 aren't tested here */
177 /* The most likely code points in this range are below the surrogates.
178 * Do an extra test to quickly exclude those. */
179 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
180 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
181 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
182 {
8ee1cdcb
KW
183 HANDLE_UNICODE_NONCHAR(uv, flags);
184 }
185 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
186 HANDLE_UNICODE_SURROGATE(uv, flags);
760c7c2f 187 }
ba6ed43c
KW
188 }
189#endif
190 return d;
191 }
192
193 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
194 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
195 * happen starting with 4-byte characters on ASCII platforms. We unify the
196 * code for these with EBCDIC, even though some of them require 5-bytes on
197 * those, because khw believes the code saving is worth the very slight
198 * performance hit on these high EBCDIC code points. */
199
200 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
a5bf80e0
KW
201 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
202 && ckWARN_d(WARN_DEPRECATED))
203 {
204 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
205 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
206 }
207 if ( (flags & UNICODE_WARN_SUPER)
208 || ( UNICODE_IS_ABOVE_31_BIT(uv)
209 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
210 {
211 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
212
213 /* Choose the more dire applicable warning */
214 (UNICODE_IS_ABOVE_31_BIT(uv))
147e3846
KW
215 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
216 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
a5bf80e0
KW
217 uv);
218 }
219 if (flags & UNICODE_DISALLOW_SUPER
220 || ( UNICODE_IS_ABOVE_31_BIT(uv)
221 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
222 {
223 return NULL;
224 }
225 }
ba6ed43c
KW
226 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
227 HANDLE_UNICODE_NONCHAR(uv, flags);
507b9800 228 }
d9432125 229
ba6ed43c
KW
230 /* Test for and handle 4-byte result. In the test immediately below, the
231 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
232 * characters). The 3 is for 3 continuation bytes; these each contribute
233 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
234 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
235 * 0x1_0000-0x1F_FFFF on ASCII */
236 if (uv < (8 * (1U << (3 * SHIFT)))) {
237 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
238 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
239 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
240 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
241
242#ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
243 characters. The end-plane non-characters for EBCDIC were
244 handled just above */
245 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
246 HANDLE_UNICODE_NONCHAR(uv, flags);
d528804a 247 }
ba6ed43c
KW
248 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
249 HANDLE_UNICODE_SURROGATE(uv, flags);
250 }
251#endif
252
253 return d;
254 }
255
256 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
257 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
258 * format. The unrolled version above turns out to not save all that much
259 * time, and at these high code points (well above the legal Unicode range
260 * on ASCII platforms, and well above anything in common use in EBCDIC),
261 * khw believes that less code outweighs slight performance gains. */
262
d9432125 263 {
5aaebcb3 264 STRLEN len = OFFUNISKIP(uv);
1d72bdf6
NIS
265 U8 *p = d+len-1;
266 while (p > d) {
4c8cd605 267 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
1d72bdf6
NIS
268 uv >>= UTF_ACCUMULATION_SHIFT;
269 }
4c8cd605 270 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
1d72bdf6
NIS
271 return d+len;
272 }
a0ed51b3 273}
a5bf80e0 274
646ca15d 275/*
07693fe6
KW
276=for apidoc uvchr_to_utf8
277
bcb1a2d4 278Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 279of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
280C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
281the byte after the end of the new character. In other words,
07693fe6
KW
282
283 d = uvchr_to_utf8(d, uv);
284
285is the recommended wide native character-aware way of saying
286
287 *(d++) = uv;
288
760c7c2f
KW
289This function accepts any UV as input, but very high code points (above
290C<IV_MAX> on the platform) will raise a deprecation warning. This is
291typically 0x7FFF_FFFF in a 32-bit word.
292
293It is possible to forbid or warn on non-Unicode code points, or those that may
294be problematic by using L</uvchr_to_utf8_flags>.
de69f3af 295
07693fe6
KW
296=cut
297*/
298
de69f3af
KW
299/* This is also a macro */
300PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
301
07693fe6
KW
302U8 *
303Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
304{
de69f3af 305 return uvchr_to_utf8(d, uv);
07693fe6
KW
306}
307
de69f3af
KW
308/*
309=for apidoc uvchr_to_utf8_flags
310
311Adds the UTF-8 representation of the native code point C<uv> to the end
f2fc1b45 312of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
c749c9fd
KW
313C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
314the byte after the end of the new character. In other words,
de69f3af
KW
315
316 d = uvchr_to_utf8_flags(d, uv, flags);
317
318or, in most cases,
319
320 d = uvchr_to_utf8_flags(d, uv, 0);
321
322This is the Unicode-aware way of saying
323
324 *(d++) = uv;
325
760c7c2f
KW
326If C<flags> is 0, this function accepts any UV as input, but very high code
327points (above C<IV_MAX> for the platform) will raise a deprecation warning.
328This is typically 0x7FFF_FFFF in a 32-bit word.
329
330Specifying C<flags> can further restrict what is allowed and not warned on, as
331follows:
de69f3af 332
796b6530 333If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
7ee537e6
KW
334the function will raise a warning, provided UTF8 warnings are enabled. If
335instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
336NULL. If both flags are set, the function will both warn and return NULL.
de69f3af 337
760c7c2f
KW
338Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
339affect how the function handles a Unicode non-character.
93e6dbd6 340
760c7c2f
KW
341And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
342affect the handling of code points that are above the Unicode maximum of
3430x10FFFF. Languages other than Perl may not be able to accept files that
344contain these.
93e6dbd6
KW
345
346The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
347the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
ecc1615f
KW
348three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
349allowed inputs to the strict UTF-8 traditionally defined by Unicode.
350Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
351C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
352above-Unicode and surrogate flags, but not the non-character ones, as
353defined in
354L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
355See L<perlunicode/Noncharacter code points>.
93e6dbd6 356
ab8e6d41
KW
357Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
358so using them is more problematic than other above-Unicode code points. Perl
359invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
360likely that non-Perl languages will not be able to read files that contain
361these that written by the perl interpreter; nor would Perl understand files
362written by something that uses a different extension. For these reasons, there
363is a separate set of flags that can warn and/or disallow these extremely high
364code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
365C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
366are entirely independent from the deprecation warning for code points above
367C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
368code point that needs more than 31 bits to represent. When that happens,
369effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
37032-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
ab8e6d41
KW
371above-Unicode code points, including these, as malformations; and
372C<UNICODE_WARN_SUPER> warns on these.)
373
374On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
375extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
376than on ASCII. Prior to that, code points 2**31 and higher were simply
377unrepresentable, and a different, incompatible method was used to represent
378code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
379and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
380platforms, warning and disallowing 2**31 and higher.
de69f3af 381
de69f3af
KW
382=cut
383*/
384
385/* This is also a macro */
386PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
387
07693fe6
KW
388U8 *
389Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
390{
de69f3af 391 return uvchr_to_utf8_flags(d, uv, flags);
07693fe6
KW
392}
393
83dc0f42
KW
394PERL_STATIC_INLINE bool
395S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
396{
397 /* Returns TRUE if the first code point represented by the Perl-extended-
398 * UTF-8-encoded string starting at 's', and looking no further than 'e -
399 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
400 *
401 * The function handles the case where the input bytes do not include all
402 * the ones necessary to represent a full character. That is, they may be
403 * the intial bytes of the representation of a code point, but possibly
404 * the final ones necessary for the complete representation may be beyond
405 * 'e - 1'.
406 *
407 * The function assumes that the sequence is well-formed UTF-8 as far as it
408 * goes, and is for a UTF-8 variant code point. If the sequence is
409 * incomplete, the function returns FALSE if there is any well-formed
410 * UTF-8 byte sequence that can complete it in such a way that a code point
411 * < 2**31 is produced; otherwise it returns TRUE.
412 *
413 * Getting this exactly right is slightly tricky, and has to be done in
414 * several places in this file, so is centralized here. It is based on the
415 * following table:
416 *
417 * U+7FFFFFFF (2 ** 31 - 1)
418 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
419 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
420 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
421 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
422 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
423 * U+80000000 (2 ** 31):
424 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
425 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
426 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
428 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
429 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
430 */
431
432#ifdef EBCDIC
433
37086697
KW
434 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
435 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
83dc0f42
KW
436 const STRLEN prefix_len = sizeof(prefix) - 1;
437 const STRLEN len = e - s;
f880f78a 438 const STRLEN cmp_len = MIN(prefix_len, len - 1);
83dc0f42
KW
439
440#else
441
442 PERL_UNUSED_ARG(e);
443
444#endif
445
446 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
447
448 assert(! UTF8_IS_INVARIANT(*s));
449
450#ifndef EBCDIC
451
452 /* Technically, a start byte of FE can be for a code point that fits into
453 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
454 * malformation. */
455 return (*s >= 0xFE);
456
457#else
458
459 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
460 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
461 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
462 * bits. */
463 if (*s != 0xFE || len == 1) {
464 return FALSE;
465 }
466
467 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
468 * \x41 and \x42. */
469 return cBOOL(memGT(s + 1, prefix, cmp_len));
470
471#endif
472
473}
474
12a4bed3
KW
475PERL_STATIC_INLINE bool
476S_does_utf8_overflow(const U8 * const s, const U8 * e)
477{
478 const U8 *x;
479 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
480
c551bb01
KW
481#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
482
483 const STRLEN len = e - s;
484
485#endif
486
12a4bed3
KW
487 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
488 * platform, that is if it represents a code point larger than the highest
489 * representable code point. (For ASCII platforms, we could use memcmp()
490 * because we don't have to convert each byte to I8, but it's very rare
491 * input indeed that would approach overflow, so the loop below will likely
492 * only get executed once.
493 *
494 * 'e' must not be beyond a full character. If it is less than a full
495 * character, the function returns FALSE if there is any input beyond 'e'
496 * that could result in a non-overflowing code point */
497
498 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
b0b342d4
KW
499 assert(s <= e && s + UTF8SKIP(s) >= e);
500
501#if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
502
503 /* On 32 bit ASCII machines, many overlongs that start with FF don't
504 * overflow */
505
c551bb01 506 if (isFF_OVERLONG(s, len)) {
b0b342d4
KW
507 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
508 return memGE(s, max_32_bit_overlong,
c551bb01 509 MIN(len, sizeof(max_32_bit_overlong) - 1));
b0b342d4
KW
510 }
511
512#endif
12a4bed3
KW
513
514 for (x = s; x < e; x++, y++) {
515
516 /* If this byte is larger than the corresponding highest UTF-8 byte, it
517 * overflows */
518 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
519 return TRUE;
520 }
521
522 /* If not the same as this byte, it must be smaller, doesn't overflow */
523 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
524 return FALSE;
525 }
526 }
527
528 /* Got to the end and all bytes are the same. If the input is a whole
529 * character, it doesn't overflow. And if it is a partial character,
530 * there's not enough information to tell, so assume doesn't overflow */
531 return FALSE;
532}
533
534PERL_STATIC_INLINE bool
535S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
536{
537 /* Overlongs can occur whenever the number of continuation bytes
538 * changes. That means whenever the number of leading 1 bits in a start
539 * byte increases from the next lower start byte. That happens for start
540 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
541 * illegal start bytes have already been excluded, so don't need to be
542 * tested here;
543 * ASCII platforms: C0, C1
544 * EBCDIC platforms C0, C1, C2, C3, C4, E0
545 *
546 * At least a second byte is required to determine if other sequences will
547 * be an overlong. */
548
549 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
550 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
551
552 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
553 assert(len > 1 && UTF8_IS_START(*s));
554
555 /* Each platform has overlongs after the start bytes given above (expressed
556 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
557 * the logic is the same, except the E0 overlong has already been excluded
558 * on EBCDIC platforms. The values below were found by manually
559 * inspecting the UTF-8 patterns. See the tables in utf8.h and
560 * utfebcdic.h. */
561
562# ifdef EBCDIC
563# define F0_ABOVE_OVERLONG 0xB0
564# define F8_ABOVE_OVERLONG 0xA8
565# define FC_ABOVE_OVERLONG 0xA4
566# define FE_ABOVE_OVERLONG 0xA2
567# define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
568 /* I8(0xfe) is FF */
569# else
570
571 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
572 return TRUE;
573 }
574
575# define F0_ABOVE_OVERLONG 0x90
576# define F8_ABOVE_OVERLONG 0x88
577# define FC_ABOVE_OVERLONG 0x84
578# define FE_ABOVE_OVERLONG 0x82
579# define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
580# endif
581
582
583 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
584 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
585 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
586 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
587 {
588 return TRUE;
589 }
590
b0b342d4
KW
591 /* Check for the FF overlong */
592 return isFF_OVERLONG(s, len);
593}
594
595PERL_STATIC_INLINE bool
596S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
597{
598 PERL_ARGS_ASSERT_ISFF_OVERLONG;
12a4bed3
KW
599
600 /* Check for the FF overlong. This happens only if all these bytes match;
601 * what comes after them doesn't matter. See tables in utf8.h,
b0b342d4 602 * utfebcdic.h. */
12a4bed3 603
b0b342d4
KW
604 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
605 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
606 sizeof(FF_OVERLONG_PREFIX) - 1));
12a4bed3
KW
607}
608
609#undef F0_ABOVE_OVERLONG
610#undef F8_ABOVE_OVERLONG
611#undef FC_ABOVE_OVERLONG
612#undef FE_ABOVE_OVERLONG
613#undef FF_OVERLONG_PREFIX
614
35f8c9bd 615STRLEN
edc2c47a 616Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
35f8c9bd 617{
2b479609 618 STRLEN len;
12a4bed3 619 const U8 *x;
35f8c9bd 620
2b479609
KW
621 /* A helper function that should not be called directly.
622 *
623 * This function returns non-zero if the string beginning at 's' and
624 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
625 * code point; otherwise it returns 0. The examination stops after the
626 * first code point in 's' is validated, not looking at the rest of the
627 * input. If 'e' is such that there are not enough bytes to represent a
628 * complete code point, this function will return non-zero anyway, if the
629 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
630 * excluded by 'flags'.
631 *
632 * A non-zero return gives the number of bytes required to represent the
633 * code point. Be aware that if the input is for a partial character, the
634 * return will be larger than 'e - s'.
635 *
636 * This function assumes that the code point represented is UTF-8 variant.
637 * The caller should have excluded this possibility before calling this
638 * function.
639 *
640 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
641 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
642 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
643 * disallowed by the flags. If the input is only for a partial character,
644 * the function will return non-zero if there is any sequence of
645 * well-formed UTF-8 that, when appended to the input sequence, could
646 * result in an allowed code point; otherwise it returns 0. Non characters
647 * cannot be determined based on partial character input. But many of the
648 * other excluded types can be determined with just the first one or two
649 * bytes.
650 *
651 */
652
653 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
654
655 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
656 |UTF8_DISALLOW_ABOVE_31_BIT)));
657 assert(! UTF8_IS_INVARIANT(*s));
35f8c9bd 658
2b479609 659 /* A variant char must begin with a start byte */
35f8c9bd
KW
660 if (UNLIKELY(! UTF8_IS_START(*s))) {
661 return 0;
662 }
663
edc2c47a
KW
664 /* Examine a maximum of a single whole code point */
665 if (e - s > UTF8SKIP(s)) {
666 e = s + UTF8SKIP(s);
667 }
668
2b479609
KW
669 len = e - s;
670
671 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
672 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
35f8c9bd 673
2b479609
KW
674 /* The code below is derived from this table. Keep in mind that legal
675 * continuation bytes range between \x80..\xBF for UTF-8, and
676 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
677 * Hence, we don't have to test the upper edge because if any of those
678 * are encountered, the sequence is malformed, and will fail elsewhere
679 * in this function.
680 * UTF-8 UTF-EBCDIC I8
681 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
682 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
683 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
684 *
685 */
686
687#ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
688# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
19794540 689# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
2b479609 690
19794540
KW
691# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
692 /* B6 and B7 */ \
693 && ((s1) & 0xFE ) == 0xB6)
2b479609
KW
694#else
695# define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
19794540
KW
696# define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
697# define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
2b479609
KW
698#endif
699
700 if ( (flags & UTF8_DISALLOW_SUPER)
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) {
702 return 0; /* Above Unicode */
703 }
704
705 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
706 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
707 {
708 return 0; /* Above 31 bits */
709 }
710
711 if (len > 1) {
712 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
713
714 if ( (flags & UTF8_DISALLOW_SUPER)
19794540 715 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
2b479609
KW
716 {
717 return 0; /* Above Unicode */
718 }
719
720 if ( (flags & UTF8_DISALLOW_SURROGATE)
19794540 721 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
2b479609
KW
722 {
723 return 0; /* Surrogate */
724 }
725
726 if ( (flags & UTF8_DISALLOW_NONCHAR)
727 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
728 {
729 return 0; /* Noncharacter code point */
730 }
731 }
732 }
733
734 /* Make sure that all that follows are continuation bytes */
35f8c9bd
KW
735 for (x = s + 1; x < e; x++) {
736 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
737 return 0;
738 }
739 }
740
af13dd8a 741 /* Here is syntactically valid. Next, make sure this isn't the start of an
12a4bed3
KW
742 * overlong. */
743 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
744 return 0;
af13dd8a
KW
745 }
746
12a4bed3
KW
747 /* And finally, that the code point represented fits in a word on this
748 * platform */
749 if (does_utf8_overflow(s, e)) {
750 return 0;
35f8c9bd
KW
751 }
752
2b479609 753 return UTF8SKIP(s);
35f8c9bd
KW
754}
755
7cf8d05d
KW
756STATIC char *
757S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
758{
759 /* Returns a mortalized C string that is a displayable copy of the 'len'
760 * bytes starting at 's', each in a \xXY format. */
761
762 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
763 trailing NUL */
764 const U8 * const e = s + len;
765 char * output;
766 char * d;
767
768 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
769
770 Newx(output, output_len, char);
771 SAVEFREEPV(output);
772
773 d = output;
774 for (; s < e; s++) {
775 const unsigned high_nibble = (*s & 0xF0) >> 4;
776 const unsigned low_nibble = (*s & 0x0F);
777
778 *d++ = '\\';
779 *d++ = 'x';
780
781 if (high_nibble < 10) {
782 *d++ = high_nibble + '0';
783 }
784 else {
785 *d++ = high_nibble - 10 + 'a';
786 }
787
788 if (low_nibble < 10) {
789 *d++ = low_nibble + '0';
790 }
791 else {
792 *d++ = low_nibble - 10 + 'a';
793 }
794 }
795
796 *d = '\0';
797 return output;
798}
799
806547a7 800PERL_STATIC_INLINE char *
7cf8d05d
KW
801S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
802
803 /* How many bytes to print */
3cc6a05e 804 STRLEN print_len,
7cf8d05d
KW
805
806 /* Which one is the non-continuation */
807 const STRLEN non_cont_byte_pos,
808
809 /* How many bytes should there be? */
810 const STRLEN expect_len)
806547a7
KW
811{
812 /* Return the malformation warning text for an unexpected continuation
813 * byte. */
814
7cf8d05d 815 const char * const where = (non_cont_byte_pos == 1)
806547a7 816 ? "immediately"
7cf8d05d
KW
817 : Perl_form(aTHX_ "%d bytes",
818 (int) non_cont_byte_pos);
3cc6a05e 819 unsigned int i;
806547a7
KW
820
821 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
822
7cf8d05d
KW
823 /* We don't need to pass this parameter, but since it has already been
824 * calculated, it's likely faster to pass it; verify under DEBUGGING */
825 assert(expect_len == UTF8SKIP(s));
826
3cc6a05e
KW
827 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
828 * length that is larger than is actually available in the buffer. If we
829 * print all the bytes based on that length, we will read past the buffer
830 * end. Often, the strings are NUL terminated, so to lower the chances of
831 * this happening, print the malformed bytes only up through any NUL. */
832 for (i = 1; i < print_len; i++) {
833 if (*(s + i) == '\0') {
834 print_len = i + 1; /* +1 gets the NUL printed */
835 break;
836 }
837 }
838
7cf8d05d
KW
839 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
840 " %s after start byte 0x%02x; need %d bytes, got %d)",
841 malformed_text,
842 _byte_dump_string(s, print_len),
843 *(s + non_cont_byte_pos),
844 where,
845 *s,
846 (int) expect_len,
847 (int) non_cont_byte_pos);
806547a7
KW
848}
849
35f8c9bd
KW
850/*
851
de69f3af 852=for apidoc utf8n_to_uvchr
378516de
KW
853
854THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
de69f3af 855Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
67e989fb 856
9041c2e3 857Bottom level UTF-8 decode routine.
de69f3af 858Returns the native code point value of the first character in the string C<s>,
746afd53
KW
859which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
860C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
861the length, in bytes, of that character.
949cf498
KW
862
863The value of C<flags> determines the behavior when C<s> does not point to a
2b5e7bc2
KW
864well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
865causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
866is the next possible position in C<s> that could begin a non-malformed
867character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
868is raised. Some UTF-8 input sequences may contain multiple malformations.
869This function tries to find every possible one in each call, so multiple
870warnings can be raised for each sequence.
949cf498
KW
871
872Various ALLOW flags can be set in C<flags> to allow (and not warn on)
873individual types of malformations, such as the sequence being overlong (that
874is, when there is a shorter sequence that can express the same code point;
875overlong sequences are expressly forbidden in the UTF-8 standard due to
876potential security issues). Another malformation example is the first byte of
877a character not being a legal first byte. See F<utf8.h> for the list of such
94953955
KW
878flags. Even if allowed, this function generally returns the Unicode
879REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
880F<utf8.h> to override this behavior for the overlong malformations, but don't
881do that except for very specialized purposes.
949cf498 882
796b6530 883The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
949cf498
KW
884flags) malformation is found. If this flag is set, the routine assumes that
885the caller will raise a warning, and this function will silently just set
d088425d
KW
886C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
887
75200dff 888Note that this API requires disambiguation between successful decoding a C<NUL>
796b6530 889character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
111fa700
KW
890in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
891be set to 1. To disambiguate, upon a zero return, see if the first byte of
892C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
f9380377 893error. Or you can use C<L</utf8n_to_uvchr_error>>.
949cf498
KW
894
895Certain code points are considered problematic. These are Unicode surrogates,
746afd53 896Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
949cf498 897By default these are considered regular code points, but certain situations
ecc1615f
KW
898warrant special handling for them, which can be specified using the C<flags>
899parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
900three classes are treated as malformations and handled as such. The flags
901C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
902C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
903disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
904restricts the allowed inputs to the strict UTF-8 traditionally defined by
905Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
906definition given by
907L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
908The difference between traditional strictness and C9 strictness is that the
909latter does not forbid non-character code points. (They are still discouraged,
910however.) For more discussion see L<perlunicode/Noncharacter code points>.
911
912The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
913C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
796b6530
KW
914C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
915raised for their respective categories, but otherwise the code points are
916considered valid (not malformations). To get a category to both be treated as
917a malformation and raise a warning, specify both the WARN and DISALLOW flags.
949cf498 918(But note that warnings are not raised if lexically disabled nor if
796b6530 919C<UTF8_CHECK_ONLY> is also specified.)
949cf498 920
760c7c2f
KW
921It is now deprecated to have very high code points (above C<IV_MAX> on the
922platforms) and this function will raise a deprecation warning for these (unless
d5944cab 923such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
760c7c2f 924in a 32-bit word.
ab8e6d41
KW
925
926Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
927so using them is more problematic than other above-Unicode code points. Perl
928invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
929likely that non-Perl languages will not be able to read files that contain
aff2be59 930these; nor would Perl understand files
ab8e6d41
KW
931written by something that uses a different extension. For these reasons, there
932is a separate set of flags that can warn and/or disallow these extremely high
933code points, even if other above-Unicode ones are accepted. These are the
760c7c2f
KW
934C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
935are entirely independent from the deprecation warning for code points above
936C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
937code point that needs more than 31 bits to represent. When that happens,
938effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
93932-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
ab8e6d41
KW
940above-Unicode code points, including these, as malformations; and
941C<UTF8_WARN_SUPER> warns on these.)
942
943On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
944extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
945than on ASCII. Prior to that, code points 2**31 and higher were simply
946unrepresentable, and a different, incompatible method was used to represent
947code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
948and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
949platforms, warning and disallowing 2**31 and higher.
949cf498
KW
950
951All other code points corresponding to Unicode characters, including private
952use and those yet to be assigned, are never considered malformed and never
953warn.
67e989fb 954
37607a96 955=cut
f9380377
KW
956
957Also implemented as a macro in utf8.h
958*/
959
960UV
961Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
962 STRLEN curlen,
963 STRLEN *retlen,
964 const U32 flags)
965{
966 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
967
968 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
969}
970
971/*
972
973=for apidoc utf8n_to_uvchr_error
974
975THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
976Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
977
978This function is for code that needs to know what the precise malformation(s)
979are when an error is found.
980
981It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
982all the others, C<errors>. If this parameter is 0, this function behaves
983identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
984to a C<U32> variable, which this function sets to indicate any errors found.
985Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
986C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
987of these bits will be set if a malformation is found, even if the input
7a65503b 988C<flags> parameter indicates that the given malformation is allowed; those
f9380377
KW
989exceptions are noted:
990
991=over 4
992
993=item C<UTF8_GOT_ABOVE_31_BIT>
994
995The code point represented by the input UTF-8 sequence occupies more than 31
996bits.
997This bit is set only if the input C<flags> parameter contains either the
998C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
999
1000=item C<UTF8_GOT_CONTINUATION>
1001
1002The input sequence was malformed in that the first byte was a a UTF-8
1003continuation byte.
1004
1005=item C<UTF8_GOT_EMPTY>
1006
1007The input C<curlen> parameter was 0.
1008
1009=item C<UTF8_GOT_LONG>
1010
1011The input sequence was malformed in that there is some other sequence that
1012evaluates to the same code point, but that sequence is shorter than this one.
1013
1014=item C<UTF8_GOT_NONCHAR>
1015
1016The code point represented by the input UTF-8 sequence is for a Unicode
1017non-character code point.
1018This bit is set only if the input C<flags> parameter contains either the
1019C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1020
1021=item C<UTF8_GOT_NON_CONTINUATION>
1022
1023The input sequence was malformed in that a non-continuation type byte was found
1024in a position where only a continuation type one should be.
1025
1026=item C<UTF8_GOT_OVERFLOW>
1027
1028The input sequence was malformed in that it is for a code point that is not
1029representable in the number of bits available in a UV on the current platform.
1030
1031=item C<UTF8_GOT_SHORT>
1032
1033The input sequence was malformed in that C<curlen> is smaller than required for
1034a complete sequence. In other words, the input is for a partial character
1035sequence.
1036
1037=item C<UTF8_GOT_SUPER>
1038
1039The input sequence was malformed in that it is for a non-Unicode code point;
1040that is, one above the legal Unicode maximum.
1041This bit is set only if the input C<flags> parameter contains either the
1042C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1043
1044=item C<UTF8_GOT_SURROGATE>
1045
1046The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1047code point.
1048This bit is set only if the input C<flags> parameter contains either the
1049C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1050
1051=back
1052
133551d8
KW
1053To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1054flag to suppress any warnings, and then examine the C<*errors> return.
1055
f9380377 1056=cut
37607a96 1057*/
67e989fb 1058
a0ed51b3 1059UV
f9380377
KW
1060Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1061 STRLEN curlen,
1062 STRLEN *retlen,
1063 const U32 flags,
1064 U32 * errors)
a0ed51b3 1065{
d4c19fe8 1066 const U8 * const s0 = s;
2b5e7bc2
KW
1067 U8 * send = NULL; /* (initialized to silence compilers' wrong
1068 warning) */
1069 U32 possible_problems = 0; /* A bit is set here for each potential problem
1070 found as we go along */
eb83ed87 1071 UV uv = *s;
2b5e7bc2
KW
1072 STRLEN expectlen = 0; /* How long should this sequence be?
1073 (initialized to silence compilers' wrong
1074 warning) */
e308b348 1075 STRLEN avail_len = 0; /* When input is too short, gives what that is */
f9380377
KW
1076 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1077 this gets set and discarded */
a0dbb045 1078
2b5e7bc2
KW
1079 /* The below are used only if there is both an overlong malformation and a
1080 * too short one. Otherwise the first two are set to 's0' and 'send', and
1081 * the third not used at all */
1082 U8 * adjusted_s0 = (U8 *) s0;
5ec712b1
KW
1083 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1084 warning) */
2b5e7bc2 1085 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
7918f24d 1086
f9380377
KW
1087 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1088
1089 if (errors) {
1090 *errors = 0;
1091 }
1092 else {
1093 errors = &discard_errors;
1094 }
a0dbb045 1095
eb83ed87
KW
1096 /* The order of malformation tests here is important. We should consume as
1097 * few bytes as possible in order to not skip any valid character. This is
1098 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1099 * http://unicode.org/reports/tr36 for more discussion as to why. For
1100 * example, once we've done a UTF8SKIP, we can tell the expected number of
1101 * bytes, and could fail right off the bat if the input parameters indicate
1102 * that there are too few available. But it could be that just that first
1103 * byte is garbled, and the intended character occupies fewer bytes. If we
1104 * blindly assumed that the first byte is correct, and skipped based on
1105 * that number, we could skip over a valid input character. So instead, we
1106 * always examine the sequence byte-by-byte.
1107 *
1108 * We also should not consume too few bytes, otherwise someone could inject
1109 * things. For example, an input could be deliberately designed to
1110 * overflow, and if this code bailed out immediately upon discovering that,
e2660c54 1111 * returning to the caller C<*retlen> pointing to the very next byte (one
eb83ed87
KW
1112 * which is actually part of of the overflowing sequence), that could look
1113 * legitimate to the caller, which could discard the initial partial
2b5e7bc2
KW
1114 * sequence and process the rest, inappropriately.
1115 *
1116 * Some possible input sequences are malformed in more than one way. This
1117 * function goes to lengths to try to find all of them. This is necessary
1118 * for correctness, as the inputs may allow one malformation but not
1119 * another, and if we abandon searching for others after finding the
1120 * allowed one, we could allow in something that shouldn't have been.
1121 */
eb83ed87 1122
b5b9af04 1123 if (UNLIKELY(curlen == 0)) {
2b5e7bc2
KW
1124 possible_problems |= UTF8_GOT_EMPTY;
1125 curlen = 0;
5a48568d 1126 uv = UNICODE_REPLACEMENT;
2b5e7bc2 1127 goto ready_to_handle_errors;
0c443dc2
JH
1128 }
1129
eb83ed87
KW
1130 expectlen = UTF8SKIP(s);
1131
1132 /* A well-formed UTF-8 character, as the vast majority of calls to this
1133 * function will be for, has this expected length. For efficiency, set
1134 * things up here to return it. It will be overriden only in those rare
1135 * cases where a malformation is found */
1136 if (retlen) {
1137 *retlen = expectlen;
1138 }
1139
1140 /* An invariant is trivially well-formed */
1d72bdf6 1141 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1142 return uv;
a0ed51b3 1143 }
67e989fb 1144
eb83ed87 1145 /* A continuation character can't start a valid sequence */
b5b9af04 1146 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1147 possible_problems |= UTF8_GOT_CONTINUATION;
1148 curlen = 1;
1149 uv = UNICODE_REPLACEMENT;
1150 goto ready_to_handle_errors;
ba210ebe 1151 }
9041c2e3 1152
dcd27b3c
KW
1153 /* Here is not a continuation byte, nor an invariant. The only thing left
1154 * is a start byte (possibly for an overlong) */
1155
534752c1
KW
1156 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1157 * that indicate the number of bytes in the character's whole UTF-8
1158 * sequence, leaving just the bits that are part of the value. */
1159 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1160
e308b348
KW
1161 /* Setup the loop end point, making sure to not look past the end of the
1162 * input string, and flag it as too short if the size isn't big enough. */
1163 send = (U8*) s0;
1164 if (UNLIKELY(curlen < expectlen)) {
1165 possible_problems |= UTF8_GOT_SHORT;
1166 avail_len = curlen;
1167 send += curlen;
1168 }
1169 else {
1170 send += expectlen;
1171 }
1172 adjusted_send = send;
1173
eb83ed87 1174 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1175 * accumulating each into the working value as we go. */
eb83ed87 1176 for (s = s0 + 1; s < send; s++) {
b5b9af04 1177 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1178 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1179 continue;
1180 }
1181
1182 /* Here, found a non-continuation before processing all expected bytes.
1183 * This byte indicates the beginning of a new character, so quit, even
1184 * if allowing this malformation. */
2b5e7bc2 1185 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1186 break;
eb83ed87
KW
1187 } /* End of loop through the character's bytes */
1188
1189 /* Save how many bytes were actually in the character */
1190 curlen = s - s0;
1191
e308b348
KW
1192 /* A convenience macro that matches either of the too-short conditions. */
1193# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1194
e308b348 1195 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1196 uv_so_far = uv;
1197 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1198 }
1199
2b5e7bc2
KW
1200 /* Note that there are two types of too-short malformation. One is when
1201 * there is actual wrong data before the normal termination of the
1202 * sequence. The other is that the sequence wasn't complete before the end
1203 * of the data we are allowed to look at, based on the input 'curlen'.
1204 * This means that we were passed data for a partial character, but it is
1205 * valid as far as we saw. The other is definitely invalid. This
1206 * distinction could be important to a caller, so the two types are kept
1207 * separate. */
1208
1209 /* Check for overflow */
1210 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1211 possible_problems |= UTF8_GOT_OVERFLOW;
1212 uv = UNICODE_REPLACEMENT;
eb83ed87 1213 }
eb83ed87 1214
2b5e7bc2
KW
1215 /* Check for overlong. If no problems so far, 'uv' is the correct code
1216 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1217 * we must look at the UTF-8 byte sequence itself to see if it is for an
1218 * overlong */
1219 if ( ( LIKELY(! possible_problems)
1220 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1221 || ( UNLIKELY( possible_problems)
1222 && ( UNLIKELY(! UTF8_IS_START(*s0))
1223 || ( curlen > 1
1224 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1225 send - s0))))))
2f8f112e 1226 {
2b5e7bc2
KW
1227 possible_problems |= UTF8_GOT_LONG;
1228
2b5e7bc2
KW
1229 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1230 UV min_uv = uv_so_far;
1231 STRLEN i;
1232
1233 /* Here, the input is both overlong and is missing some trailing
1234 * bytes. There is no single code point it could be for, but there
1235 * may be enough information present to determine if what we have
1236 * so far is for an unallowed code point, such as for a surrogate.
1237 * The code below has the intelligence to determine this, but just
1238 * for non-overlong UTF-8 sequences. What we do here is calculate
1239 * the smallest code point the input could represent if there were
1240 * no too short malformation. Then we compute and save the UTF-8
1241 * for that, which is what the code below looks at instead of the
1242 * raw input. It turns out that the smallest such code point is
1243 * all we need. */
1244 for (i = curlen; i < expectlen; i++) {
1245 min_uv = UTF8_ACCUMULATE(min_uv,
1246 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1247 }
1248
1249 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1250 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1251 to free it ourselves if
1252 warnings are made fatal */
1253 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1254 }
eb83ed87
KW
1255 }
1256
2b5e7bc2
KW
1257 /* Now check that the input isn't for a problematic code point not allowed
1258 * by the input parameters. */
1259 /* isn't problematic if < this */
1260 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1261 || ( UNLIKELY(possible_problems)
d60baaa7
KW
1262
1263 /* if overflow, we know without looking further
1264 * precisely which of the problematic types it is,
1265 * and we deal with those in the overflow handling
1266 * code */
1267 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
2b5e7bc2 1268 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1269 && ((flags & ( UTF8_DISALLOW_NONCHAR
1270 |UTF8_DISALLOW_SURROGATE
1271 |UTF8_DISALLOW_SUPER
1272 |UTF8_DISALLOW_ABOVE_31_BIT
1273 |UTF8_WARN_NONCHAR
1274 |UTF8_WARN_SURROGATE
1275 |UTF8_WARN_SUPER
1276 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1277 /* In case of a malformation, 'uv' is not valid, and has
1278 * been changed to something in the Unicode range.
1279 * Currently we don't output a deprecation message if there
1280 * is already a malformation, so we don't have to special
1281 * case the test immediately below */
760c7c2f
KW
1282 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1283 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1284 {
2b5e7bc2
KW
1285 /* If there were no malformations, or the only malformation is an
1286 * overlong, 'uv' is valid */
1287 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1288 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1289 possible_problems |= UTF8_GOT_SURROGATE;
1290 }
1291 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1292 possible_problems |= UTF8_GOT_SUPER;
1293 }
1294 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1295 possible_problems |= UTF8_GOT_NONCHAR;
1296 }
1297 }
1298 else { /* Otherwise, need to look at the source UTF-8, possibly
1299 adjusted to be non-overlong */
1300
1301 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1302 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1303 {
2b5e7bc2
KW
1304 possible_problems |= UTF8_GOT_SUPER;
1305 }
1306 else if (curlen > 1) {
1307 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1308 NATIVE_UTF8_TO_I8(*adjusted_s0),
1309 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1310 {
2b5e7bc2 1311 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1312 }
2b5e7bc2
KW
1313 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1314 NATIVE_UTF8_TO_I8(*adjusted_s0),
1315 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1316 {
1317 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1318 }
1319 }
c0236afe 1320
2b5e7bc2
KW
1321 /* We need a complete well-formed UTF-8 character to discern
1322 * non-characters, so can't look for them here */
1323 }
1324 }
949cf498 1325
2b5e7bc2
KW
1326 ready_to_handle_errors:
1327
1328 /* At this point:
1329 * curlen contains the number of bytes in the sequence that
1330 * this call should advance the input by.
e308b348
KW
1331 * avail_len gives the available number of bytes passed in, but
1332 * only if this is less than the expected number of
1333 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1334 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1335 * is set in it for each potential problem found.
1336 * uv contains the code point the input sequence
1337 * represents; or if there is a problem that prevents
1338 * a well-defined value from being computed, it is
1339 * some subsitute value, typically the REPLACEMENT
1340 * CHARACTER.
1341 * s0 points to the first byte of the character
1342 * send points to just after where that (potentially
1343 * partial) character ends
1344 * adjusted_s0 normally is the same as s0, but in case of an
1345 * overlong for which the UTF-8 matters below, it is
1346 * the first byte of the shortest form representation
1347 * of the input.
1348 * adjusted_send normally is the same as 'send', but if adjusted_s0
1349 * is set to something other than s0, this points one
1350 * beyond its end
1351 */
eb83ed87 1352
2b5e7bc2
KW
1353 if (UNLIKELY(possible_problems)) {
1354 bool disallowed = FALSE;
1355 const U32 orig_problems = possible_problems;
1356
1357 while (possible_problems) { /* Handle each possible problem */
1358 UV pack_warn = 0;
1359 char * message = NULL;
1360
1361 /* Each 'if' clause handles one problem. They are ordered so that
1362 * the first ones' messages will be displayed before the later
1363 * ones; this is kinda in decreasing severity order */
1364 if (possible_problems & UTF8_GOT_OVERFLOW) {
1365
1366 /* Overflow means also got a super and above 31 bits, but we
1367 * handle all three cases here */
1368 possible_problems
1369 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1370 *errors |= UTF8_GOT_OVERFLOW;
1371
1372 /* But the API says we flag all errors found */
1373 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1374 *errors |= UTF8_GOT_SUPER;
1375 }
1376 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1377 *errors |= UTF8_GOT_ABOVE_31_BIT;
1378 }
2b5e7bc2 1379
d60baaa7
KW
1380 /* Disallow if any of the three categories say to */
1381 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1382 || (flags & ( UTF8_DISALLOW_SUPER
1383 |UTF8_DISALLOW_ABOVE_31_BIT)))
1384 {
1385 disallowed = TRUE;
1386 }
1387
1388
1389 /* Likewise, warn if any say to, plus if deprecation warnings
1390 * are on, because this code point is above IV_MAX */
1391 if ( ckWARN_d(WARN_DEPRECATED)
1392 || ! (flags & UTF8_ALLOW_OVERFLOW)
1393 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1394 {
2b5e7bc2
KW
1395
1396 /* The warnings code explicitly says it doesn't handle the case
1397 * of packWARN2 and two categories which have parent-child
1398 * relationship. Even if it works now to raise the warning if
1399 * either is enabled, it wouldn't necessarily do so in the
1400 * future. We output (only) the most dire warning*/
1401 if (! (flags & UTF8_CHECK_ONLY)) {
1402 if (ckWARN_d(WARN_UTF8)) {
1403 pack_warn = packWARN(WARN_UTF8);
1404 }
1405 else if (ckWARN_d(WARN_NON_UNICODE)) {
1406 pack_warn = packWARN(WARN_NON_UNICODE);
1407 }
1408 if (pack_warn) {
1409 message = Perl_form(aTHX_ "%s: %s (overflows)",
1410 malformed_text,
1411 _byte_dump_string(s0, send - s0));
1412 }
1413 }
d60baaa7 1414 }
2b5e7bc2
KW
1415 }
1416 else if (possible_problems & UTF8_GOT_EMPTY) {
1417 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1418 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1419
1420 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1421
1422 /* This so-called malformation is now treated as a bug in
1423 * the caller. If you have nothing to decode, skip calling
1424 * this function */
1425 assert(0);
1426
2b5e7bc2
KW
1427 disallowed = TRUE;
1428 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1429 pack_warn = packWARN(WARN_UTF8);
1430 message = Perl_form(aTHX_ "%s (empty string)",
1431 malformed_text);
1432 }
1433 }
1434 }
1435 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1436 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1437 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1438
1439 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1440 disallowed = TRUE;
1441 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1442 pack_warn = packWARN(WARN_UTF8);
1443 message = Perl_form(aTHX_
1444 "%s: %s (unexpected continuation byte 0x%02x,"
1445 " with no preceding start byte)",
1446 malformed_text,
1447 _byte_dump_string(s0, 1), *s0);
1448 }
1449 }
1450 }
2b5e7bc2
KW
1451 else if (possible_problems & UTF8_GOT_SHORT) {
1452 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1453 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1454
1455 if (! (flags & UTF8_ALLOW_SHORT)) {
1456 disallowed = TRUE;
1457 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1458 pack_warn = packWARN(WARN_UTF8);
1459 message = Perl_form(aTHX_
9a6c9c81 1460 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1461 malformed_text,
1462 _byte_dump_string(s0, send - s0),
e308b348
KW
1463 (int)avail_len,
1464 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1465 (int)expectlen);
1466 }
1467 }
ba210ebe 1468
2b5e7bc2 1469 }
e308b348
KW
1470 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1471 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1472 *errors |= UTF8_GOT_NON_CONTINUATION;
1473
1474 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1475 disallowed = TRUE;
1476 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1477 pack_warn = packWARN(WARN_UTF8);
1478 message = Perl_form(aTHX_ "%s",
1479 unexpected_non_continuation_text(s0,
1480 send - s0,
1481 s - s0,
1482 (int) expectlen));
1483 }
1484 }
1485 }
2b5e7bc2
KW
1486 else if (possible_problems & UTF8_GOT_LONG) {
1487 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1488 *errors |= UTF8_GOT_LONG;
2b5e7bc2 1489
94953955
KW
1490 if (flags & UTF8_ALLOW_LONG) {
1491
1492 /* We don't allow the actual overlong value, unless the
1493 * special extra bit is also set */
1494 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1495 & ~UTF8_ALLOW_LONG)))
1496 {
1497 uv = UNICODE_REPLACEMENT;
1498 }
1499 }
1500 else {
2b5e7bc2
KW
1501 disallowed = TRUE;
1502
1503 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1504 pack_warn = packWARN(WARN_UTF8);
1505
1506 /* These error types cause 'uv' to be something that
1507 * isn't what was intended, so can't use it in the
1508 * message. The other error types either can't
1509 * generate an overlong, or else the 'uv' is valid */
1510 if (orig_problems &
1511 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1512 {
1513 message = Perl_form(aTHX_
1514 "%s: %s (any UTF-8 sequence that starts"
1515 " with \"%s\" is overlong which can and"
1516 " should be represented with a"
1517 " different, shorter sequence)",
1518 malformed_text,
1519 _byte_dump_string(s0, send - s0),
1520 _byte_dump_string(s0, curlen));
1521 }
1522 else {
1523 U8 tmpbuf[UTF8_MAXBYTES+1];
1524 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1525 uv, 0);
1526 message = Perl_form(aTHX_
1527 "%s: %s (overlong; instead use %s to represent"
147e3846 1528 " U+%0*" UVXf ")",
2b5e7bc2
KW
1529 malformed_text,
1530 _byte_dump_string(s0, send - s0),
1531 _byte_dump_string(tmpbuf, e - tmpbuf),
1532 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1533 small code points */
1534 uv);
1535 }
1536 }
1537 }
1538 }
1539 else if (possible_problems & UTF8_GOT_SURROGATE) {
1540 possible_problems &= ~UTF8_GOT_SURROGATE;
1541
f9380377
KW
1542 if (flags & UTF8_WARN_SURROGATE) {
1543 *errors |= UTF8_GOT_SURROGATE;
1544
1545 if ( ! (flags & UTF8_CHECK_ONLY)
1546 && ckWARN_d(WARN_SURROGATE))
1547 {
2b5e7bc2
KW
1548 pack_warn = packWARN(WARN_SURROGATE);
1549
1550 /* These are the only errors that can occur with a
1551 * surrogate when the 'uv' isn't valid */
1552 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1553 message = Perl_form(aTHX_
1554 "UTF-16 surrogate (any UTF-8 sequence that"
1555 " starts with \"%s\" is for a surrogate)",
1556 _byte_dump_string(s0, curlen));
1557 }
1558 else {
1559 message = Perl_form(aTHX_
147e3846 1560 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1561 }
f9380377 1562 }
2b5e7bc2 1563 }
ba210ebe 1564
2b5e7bc2
KW
1565 if (flags & UTF8_DISALLOW_SURROGATE) {
1566 disallowed = TRUE;
f9380377 1567 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1568 }
1569 }
1570 else if (possible_problems & UTF8_GOT_SUPER) {
1571 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1572
f9380377
KW
1573 if (flags & UTF8_WARN_SUPER) {
1574 *errors |= UTF8_GOT_SUPER;
1575
1576 if ( ! (flags & UTF8_CHECK_ONLY)
1577 && ckWARN_d(WARN_NON_UNICODE))
1578 {
2b5e7bc2
KW
1579 pack_warn = packWARN(WARN_NON_UNICODE);
1580
1581 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1582 message = Perl_form(aTHX_
1583 "Any UTF-8 sequence that starts with"
1584 " \"%s\" is for a non-Unicode code point,"
1585 " may not be portable",
1586 _byte_dump_string(s0, curlen));
1587 }
1588 else {
1589 message = Perl_form(aTHX_
147e3846 1590 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1591 " Unicode, may not be portable",
1592 uv);
1593 }
f9380377 1594 }
2b5e7bc2 1595 }
ba210ebe 1596
2b5e7bc2
KW
1597 /* The maximum code point ever specified by a standard was
1598 * 2**31 - 1. Anything larger than that is a Perl extension
1599 * that very well may not be understood by other applications
1600 * (including earlier perl versions on EBCDIC platforms). We
1601 * test for these after the regular SUPER ones, and before
1602 * possibly bailing out, so that the slightly more dire warning
1603 * will override the regular one. */
1604 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1605 |UTF8_WARN_SUPER
1606 |UTF8_DISALLOW_ABOVE_31_BIT))
1607 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1608 && UNLIKELY(is_utf8_cp_above_31_bits(
1609 adjusted_s0,
1610 adjusted_send)))
1611 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1612 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1613 {
1614 if ( ! (flags & UTF8_CHECK_ONLY)
1615 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1616 && ckWARN_d(WARN_UTF8))
1617 {
1618 pack_warn = packWARN(WARN_UTF8);
1619
1620 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1621 message = Perl_form(aTHX_
1622 "Any UTF-8 sequence that starts with"
1623 " \"%s\" is for a non-Unicode code"
1624 " point, and is not portable",
1625 _byte_dump_string(s0, curlen));
1626 }
1627 else {
1628 message = Perl_form(aTHX_
147e3846 1629 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1630 " and not portable",
1631 uv);
1632 }
1633 }
1634
f9380377
KW
1635 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1636 *errors |= UTF8_GOT_ABOVE_31_BIT;
1637
1638 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1639 disallowed = TRUE;
1640 }
2b5e7bc2
KW
1641 }
1642 }
eb83ed87 1643
2b5e7bc2 1644 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1645 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1646 disallowed = TRUE;
1647 }
eb83ed87 1648
2b5e7bc2
KW
1649 /* The deprecated warning overrides any non-deprecated one. If
1650 * there are other problems, a deprecation message is not
1651 * really helpful, so don't bother to raise it in that case.
1652 * This also keeps the code from having to handle the case
1653 * where 'uv' is not valid. */
1654 if ( ! (orig_problems
1655 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1656 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1657 && ckWARN_d(WARN_DEPRECATED))
1658 {
1659 message = Perl_form(aTHX_ cp_above_legal_max,
1660 uv, MAX_NON_DEPRECATED_CP);
1661 pack_warn = packWARN(WARN_DEPRECATED);
1662 }
1663 }
1664 else if (possible_problems & UTF8_GOT_NONCHAR) {
1665 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1666
f9380377
KW
1667 if (flags & UTF8_WARN_NONCHAR) {
1668 *errors |= UTF8_GOT_NONCHAR;
1669
1670 if ( ! (flags & UTF8_CHECK_ONLY)
1671 && ckWARN_d(WARN_NONCHAR))
1672 {
2b5e7bc2
KW
1673 /* The code above should have guaranteed that we don't
1674 * get here with errors other than overlong */
1675 assert (! (orig_problems
1676 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1677
1678 pack_warn = packWARN(WARN_NONCHAR);
1679 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1680 " U+%04" UVXf " is not recommended"
2b5e7bc2 1681 " for open interchange", uv);
f9380377 1682 }
2b5e7bc2 1683 }
5b311467 1684
2b5e7bc2
KW
1685 if (flags & UTF8_DISALLOW_NONCHAR) {
1686 disallowed = TRUE;
f9380377 1687 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1688 }
1689 } /* End of looking through the possible flags */
1690
1691 /* Display the message (if any) for the problem being handled in
1692 * this iteration of the loop */
1693 if (message) {
1694 if (PL_op)
1695 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1696 OP_DESC(PL_op));
1697 else
1698 Perl_warner(aTHX_ pack_warn, "%s", message);
1699 }
1700 } /* End of 'while (possible_problems) {' */
a0dbb045 1701
2b5e7bc2
KW
1702 /* Since there was a possible problem, the returned length may need to
1703 * be changed from the one stored at the beginning of this function.
1704 * Instead of trying to figure out if that's needed, just do it. */
1705 if (retlen) {
1706 *retlen = curlen;
1707 }
a0dbb045 1708
2b5e7bc2
KW
1709 if (disallowed) {
1710 if (flags & UTF8_CHECK_ONLY && retlen) {
1711 *retlen = ((STRLEN) -1);
1712 }
1713 return 0;
1714 }
eb83ed87 1715 }
ba210ebe 1716
2b5e7bc2 1717 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1718}
1719
8e84507e 1720/*
ec5f19d0
KW
1721=for apidoc utf8_to_uvchr_buf
1722
1723Returns the native code point of the first character in the string C<s> which
1724is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1725C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1726
524080c4
KW
1727If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1728enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1729C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1730(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1731C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1732the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1733See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1734returned.
ec5f19d0 1735
760c7c2f
KW
1736Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1737unless those are turned off.
1738
ec5f19d0 1739=cut
52be2536
KW
1740
1741Also implemented as a macro in utf8.h
1742
ec5f19d0
KW
1743*/
1744
1745
1746UV
1747Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1748{
ec5f19d0
KW
1749 assert(s < send);
1750
1751 return utf8n_to_uvchr(s, send - s, retlen,
1752 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1753}
1754
52be2536
KW
1755/* This is marked as deprecated
1756 *
ec5f19d0
KW
1757=for apidoc utf8_to_uvuni_buf
1758
de69f3af
KW
1759Only in very rare circumstances should code need to be dealing in Unicode
1760(as opposed to native) code points. In those few cases, use
1761C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1762
1763Returns the Unicode (not-native) code point of the first character in the
1764string C<s> which
ec5f19d0
KW
1765is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1766C<retlen> will be set to the length, in bytes, of that character.
1767
524080c4
KW
1768If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1769enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1770NULL) to -1. If those warnings are off, the computed value if well-defined (or
1771the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1772is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1773next possible position in C<s> that could begin a non-malformed character.
de69f3af 1774See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1775
760c7c2f
KW
1776Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1777unless those are turned off.
1778
ec5f19d0
KW
1779=cut
1780*/
1781
1782UV
1783Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1784{
1785 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1786
1787 assert(send > s);
1788
5962d97e
KW
1789 /* Call the low level routine, asking for checks */
1790 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1791}
1792
b76347f2 1793/*
87cea99e 1794=for apidoc utf8_length
b76347f2
JH
1795
1796Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1797Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1798up past C<e>, croaks.
b76347f2
JH
1799
1800=cut
1801*/
1802
1803STRLEN
35a4481c 1804Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1805{
1806 STRLEN len = 0;
1807
7918f24d
NC
1808 PERL_ARGS_ASSERT_UTF8_LENGTH;
1809
8850bf83
JH
1810 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1811 * the bitops (especially ~) can create illegal UTF-8.
1812 * In other words: in Perl UTF-8 is not just for Unicode. */
1813
a3b680e6
AL
1814 if (e < s)
1815 goto warn_and_return;
b76347f2 1816 while (s < e) {
4cbf4130 1817 s += UTF8SKIP(s);
8e91ec7f
AV
1818 len++;
1819 }
1820
1821 if (e != s) {
1822 len--;
1823 warn_and_return:
9b387841
NC
1824 if (PL_op)
1825 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1826 "%s in %s", unees, OP_DESC(PL_op));
1827 else
61a12c31 1828 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1829 }
1830
1831 return len;
1832}
1833
b06226ff 1834/*
fed3ba5d
NC
1835=for apidoc bytes_cmp_utf8
1836
a1433954 1837Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1838sequence of characters (stored as UTF-8)
1839in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1840equal, -1 or -2 if the first string is less than the second string, +1 or +2
1841if the first string is greater than the second string.
1842
1843-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1844longer string. -2 or +2 is returned if
1845there was a difference between characters
fed3ba5d
NC
1846within the strings.
1847
1848=cut
1849*/
1850
1851int
1852Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1853{
1854 const U8 *const bend = b + blen;
1855 const U8 *const uend = u + ulen;
1856
1857 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1858
1859 while (b < bend && u < uend) {
1860 U8 c = *u++;
1861 if (!UTF8_IS_INVARIANT(c)) {
1862 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1863 if (u < uend) {
1864 U8 c1 = *u++;
1865 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1866 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1867 } else {
2b5e7bc2 1868 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1869 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1870 "%s %s%s",
7cf8d05d 1871 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1872 PL_op ? " in " : "",
1873 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1874 return -2;
1875 }
1876 } else {
1877 if (PL_op)
1878 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1879 "%s in %s", unees, OP_DESC(PL_op));
1880 else
61a12c31 1881 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1882 return -2; /* Really want to return undef :-) */
1883 }
1884 } else {
1885 return -2;
1886 }
1887 }
1888 if (*b != c) {
1889 return *b < c ? -2 : +2;
1890 }
1891 ++b;
1892 }
1893
1894 if (b == bend && u == uend)
1895 return 0;
1896
1897 return b < bend ? +1 : -1;
1898}
1899
1900/*
87cea99e 1901=for apidoc utf8_to_bytes
6940069f 1902
2bbc8d55 1903Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1904Unlike L</bytes_to_utf8>, this over-writes the original string, and
1905updates C<len> to contain the new length.
67e989fb 1906Returns zero on failure, setting C<len> to -1.
6940069f 1907
a1433954 1908If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1909
6940069f
GS
1910=cut
1911*/
1912
1913U8 *
37607a96 1914Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1915{
d4c19fe8
AL
1916 U8 * const save = s;
1917 U8 * const send = s + *len;
6940069f 1918 U8 *d;
246fae53 1919
7918f24d 1920 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1921 PERL_UNUSED_CONTEXT;
7918f24d 1922
1e54db1a 1923 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1924 while (s < send) {
d59937ca
KW
1925 if (! UTF8_IS_INVARIANT(*s)) {
1926 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1927 *len = ((STRLEN) -1);
1928 return 0;
1929 }
1930 s++;
dcad2880 1931 }
d59937ca 1932 s++;
246fae53 1933 }
dcad2880
JH
1934
1935 d = s = save;
6940069f 1936 while (s < send) {
80e0b38f
KW
1937 U8 c = *s++;
1938 if (! UTF8_IS_INVARIANT(c)) {
1939 /* Then it is two-byte encoded */
a62b247b 1940 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1941 s++;
1942 }
1943 *d++ = c;
6940069f
GS
1944 }
1945 *d = '\0';
246fae53 1946 *len = d - save;
6940069f
GS
1947 return save;
1948}
1949
1950/*
87cea99e 1951=for apidoc bytes_from_utf8
f9a63242 1952
2bbc8d55 1953Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1954Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1955the newly-created string, and updates C<len> to contain the new
1956length. Returns the original string if no conversion occurs, C<len>
72d33970 1957is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19580 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1959in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1960
37607a96
PK
1961=cut
1962*/
f9a63242
JH
1963
1964U8 *
e1ec3a88 1965Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1966{
f9a63242 1967 U8 *d;
e1ec3a88
AL
1968 const U8 *start = s;
1969 const U8 *send;
f9a63242
JH
1970 I32 count = 0;
1971
7918f24d 1972 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1973 PERL_UNUSED_CONTEXT;
f9a63242 1974 if (!*is_utf8)
73d840c0 1975 return (U8 *)start;
f9a63242 1976
1e54db1a 1977 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1978 for (send = s + *len; s < send;) {
d59937ca
KW
1979 if (! UTF8_IS_INVARIANT(*s)) {
1980 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1981 return (U8 *)start;
d59937ca
KW
1982 }
1983 count++;
1984 s++;
db42d148 1985 }
d59937ca 1986 s++;
f9a63242
JH
1987 }
1988
35da51f7 1989 *is_utf8 = FALSE;
f9a63242 1990
212542aa 1991 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1992 s = start; start = d;
f9a63242
JH
1993 while (s < send) {
1994 U8 c = *s++;
1a91c45d 1995 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1996 /* Then it is two-byte encoded */
a62b247b 1997 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1998 s++;
c4d5f83a
NIS
1999 }
2000 *d++ = c;
f9a63242
JH
2001 }
2002 *d = '\0';
2003 *len = d - start;
73d840c0 2004 return (U8 *)start;
f9a63242
JH
2005}
2006
2007/*
87cea99e 2008=for apidoc bytes_to_utf8
6940069f 2009
ff97e5cf
KW
2010Converts a string C<s> of length C<len> bytes from the native encoding into
2011UTF-8.
6662521e 2012Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 2013reflect the new length in bytes.
6940069f 2014
75200dff 2015A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
2016
2017If you want to convert to UTF-8 from encodings other than
2018the native (Latin1 or EBCDIC),
a1433954 2019see L</sv_recode_to_utf8>().
c9ada85f 2020
497711e7 2021=cut
6940069f
GS
2022*/
2023
c682ebef
FC
2024/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2025 likewise need duplication. */
2026
6940069f 2027U8*
35a4481c 2028Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 2029{
35a4481c 2030 const U8 * const send = s + (*len);
6940069f
GS
2031 U8 *d;
2032 U8 *dst;
7918f24d
NC
2033
2034 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2035 PERL_UNUSED_CONTEXT;
6940069f 2036
212542aa 2037 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2038 dst = d;
2039
2040 while (s < send) {
55d09dc8
KW
2041 append_utf8_from_native_byte(*s, &d);
2042 s++;
6940069f
GS
2043 }
2044 *d = '\0';
6662521e 2045 *len = d-dst;
6940069f
GS
2046 return dst;
2047}
2048
a0ed51b3 2049/*
dea0fc0b 2050 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2051 *
2052 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2053 * We optimize for native, for obvious reasons. */
2054
2055U8*
dea0fc0b 2056Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2057{
dea0fc0b
JH
2058 U8* pend;
2059 U8* dstart = d;
2060
7918f24d
NC
2061 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2062
dea0fc0b 2063 if (bytelen & 1)
147e3846 2064 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2065
2066 pend = p + bytelen;
2067
a0ed51b3 2068 while (p < pend) {
dea0fc0b
JH
2069 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2070 p += 2;
2d1545e5 2071 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2072 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2073 continue;
2074 }
56d37426
KW
2075 if (uv <= MAX_UTF8_TWO_BYTE) {
2076 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2077 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2078 continue;
2079 }
46956fad
KW
2080#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2081#define LAST_HIGH_SURROGATE 0xDBFF
2082#define FIRST_LOW_SURROGATE 0xDC00
2083#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2084
2085 /* This assumes that most uses will be in the first Unicode plane, not
2086 * needing surrogates */
2087 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2088 && uv <= UNICODE_SURROGATE_LAST))
2089 {
2090 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2091 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2092 }
2093 else {
01ea242b 2094 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2095 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2096 || UNLIKELY(low > LAST_LOW_SURROGATE))
2097 {
01ea242b 2098 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2099 }
2100 p += 2;
46956fad
KW
2101 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2102 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2103 }
a0ed51b3 2104 }
56d37426
KW
2105#ifdef EBCDIC
2106 d = uvoffuni_to_utf8_flags(d, uv, 0);
2107#else
a0ed51b3 2108 if (uv < 0x10000) {
eb160463
GS
2109 *d++ = (U8)(( uv >> 12) | 0xe0);
2110 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2111 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2112 continue;
2113 }
2114 else {
eb160463
GS
2115 *d++ = (U8)(( uv >> 18) | 0xf0);
2116 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2117 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2118 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2119 continue;
2120 }
56d37426 2121#endif
a0ed51b3 2122 }
dea0fc0b 2123 *newlen = d - dstart;
a0ed51b3
LW
2124 return d;
2125}
2126
2127/* Note: this one is slightly destructive of the source. */
2128
2129U8*
dea0fc0b 2130Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2131{
2132 U8* s = (U8*)p;
d4c19fe8 2133 U8* const send = s + bytelen;
7918f24d
NC
2134
2135 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2136
e0ea5e2d 2137 if (bytelen & 1)
147e3846 2138 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2139 (UV)bytelen);
2140
a0ed51b3 2141 while (s < send) {
d4c19fe8 2142 const U8 tmp = s[0];
a0ed51b3
LW
2143 s[0] = s[1];
2144 s[1] = tmp;
2145 s += 2;
2146 }
dea0fc0b 2147 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2148}
2149
922e8cb4
KW
2150bool
2151Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2152{
2153 U8 tmpbuf[UTF8_MAXBYTES+1];
2154 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2155 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2156}
2157
f9ae8fb6
JD
2158/* Internal function so we can deprecate the external one, and call
2159 this one from other deprecated functions in this file */
2160
f2645549
KW
2161bool
2162Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2163{
f2645549 2164 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2165
2166 if (*p == '_')
2167 return TRUE;
f25ce844 2168 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2169}
2170
5092f92a 2171bool
eba68aa0
KW
2172Perl__is_uni_perl_idcont(pTHX_ UV c)
2173{
2174 U8 tmpbuf[UTF8_MAXBYTES+1];
2175 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2176 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2177}
2178
2179bool
f91dcd13
KW
2180Perl__is_uni_perl_idstart(pTHX_ UV c)
2181{
2182 U8 tmpbuf[UTF8_MAXBYTES+1];
2183 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2184 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2185}
2186
3a4c58c9
KW
2187UV
2188Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2189{
2190 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2191 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2192 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2193 * either "SS" or "Ss". Which one to use is passed into the routine in
2194 * 'S_or_s' to avoid a test */
2195
2196 UV converted = toUPPER_LATIN1_MOD(c);
2197
2198 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2199
2200 assert(S_or_s == 'S' || S_or_s == 's');
2201
6f2d5cbc 2202 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2203 characters in this range */
3a4c58c9
KW
2204 *p = (U8) converted;
2205 *lenp = 1;
2206 return converted;
2207 }
2208
2209 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2210 * which it maps to one of them, so as to only have to have one check for
2211 * it in the main case */
2212 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2213 switch (c) {
2214 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2215 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2216 break;
2217 case MICRO_SIGN:
2218 converted = GREEK_CAPITAL_LETTER_MU;
2219 break;
79e064b9
KW
2220#if UNICODE_MAJOR_VERSION > 2 \
2221 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2222 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2223 case LATIN_SMALL_LETTER_SHARP_S:
2224 *(p)++ = 'S';
2225 *p = S_or_s;
2226 *lenp = 2;
2227 return 'S';
79e064b9 2228#endif
3a4c58c9
KW
2229 default:
2230 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2231 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2232 }
2233 }
2234
2235 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2236 *p = UTF8_TWO_BYTE_LO(converted);
2237 *lenp = 2;
2238
2239 return converted;
2240}
2241
50bda2c3
KW
2242/* Call the function to convert a UTF-8 encoded character to the specified case.
2243 * Note that there may be more than one character in the result.
2244 * INP is a pointer to the first byte of the input character
2245 * OUTP will be set to the first byte of the string of changed characters. It
2246 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2247 * LENP will be set to the length in bytes of the string of changed characters
2248 *
2249 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2250#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2251#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2252#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2253
b9992569
KW
2254/* This additionally has the input parameter 'specials', which if non-zero will
2255 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2256 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2257#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2258
84afefe6
JH
2259UV
2260Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2261{
a1433954
KW
2262 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2263 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2264 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2265 * the changed version may be longer than the original character.
2266 *
2267 * The ordinal of the first character of the changed version is returned
2268 * (but note, as explained above, that there may be more.) */
2269
7918f24d
NC
2270 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2271
3a4c58c9
KW
2272 if (c < 256) {
2273 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2274 }
2275
0ebc6274 2276 uvchr_to_utf8(p, c);
b9992569 2277 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2278}
2279
84afefe6
JH
2280UV
2281Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2282{
7918f24d
NC
2283 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2284
3a4c58c9
KW
2285 if (c < 256) {
2286 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2287 }
2288
0ebc6274 2289 uvchr_to_utf8(p, c);
b9992569 2290 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2291}
2292
afc16117 2293STATIC U8
eaf412bf 2294S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2295{
2296 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2297 * those, converting the result to UTF-8. Since the result is always just
a1433954 2298 * one character, we allow <p> to be NULL */
afc16117
KW
2299
2300 U8 converted = toLOWER_LATIN1(c);
2301
eaf412bf
KW
2302 PERL_UNUSED_ARG(dummy);
2303
afc16117 2304 if (p != NULL) {
6f2d5cbc 2305 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2306 *p = converted;
2307 *lenp = 1;
2308 }
2309 else {
430c9760
KW
2310 /* Result is known to always be < 256, so can use the EIGHT_BIT
2311 * macros */
2312 *p = UTF8_EIGHT_BIT_HI(converted);
2313 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2314 *lenp = 2;
2315 }
2316 }
2317 return converted;
2318}
2319
84afefe6
JH
2320UV
2321Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2322{
7918f24d
NC
2323 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2324
afc16117 2325 if (c < 256) {
eaf412bf 2326 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2327 }
2328
afc16117 2329 uvchr_to_utf8(p, c);
b9992569 2330 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2331}
2332
84afefe6 2333UV
51910141 2334Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2335{
51910141 2336 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2337 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2338 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2339 *
2340 * Not to be used for locale folds
51910141 2341 */
f673fad4 2342
a1dde8de
KW
2343 UV converted;
2344
2345 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2346 PERL_UNUSED_CONTEXT;
a1dde8de 2347
1ca267a5
KW
2348 assert (! (flags & FOLD_FLAGS_LOCALE));
2349
659a7c2d 2350 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2351 converted = GREEK_SMALL_LETTER_MU;
2352 }
9b63e895
KW
2353#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2354 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2355 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2356 else if ( (flags & FOLD_FLAGS_FULL)
2357 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2358 {
1ca267a5
KW
2359 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2360 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2361 * under those circumstances. */
2362 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2363 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2364 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2365 p, *lenp, U8);
2366 return LATIN_SMALL_LETTER_LONG_S;
2367 }
2368 else {
4f489194
KW
2369 *(p)++ = 's';
2370 *p = 's';
2371 *lenp = 2;
2372 return 's';
1ca267a5 2373 }
a1dde8de 2374 }
9b63e895 2375#endif
a1dde8de
KW
2376 else { /* In this range the fold of all other characters is their lower
2377 case */
2378 converted = toLOWER_LATIN1(c);
2379 }
2380
6f2d5cbc 2381 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2382 *p = (U8) converted;
2383 *lenp = 1;
2384 }
2385 else {
2386 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2387 *p = UTF8_TWO_BYTE_LO(converted);
2388 *lenp = 2;
2389 }
2390
2391 return converted;
2392}
2393
2394UV
31f05a37 2395Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2396{
4b593389 2397
a0270393
KW
2398 /* Not currently externally documented, and subject to change
2399 * <flags> bits meanings:
2400 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2401 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2402 * locale are to be used.
a0270393
KW
2403 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2404 */
4b593389 2405
36bb2ab6 2406 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2407
780fcc9f
KW
2408 if (flags & FOLD_FLAGS_LOCALE) {
2409 /* Treat a UTF-8 locale as not being in locale at all */
2410 if (IN_UTF8_CTYPE_LOCALE) {
2411 flags &= ~FOLD_FLAGS_LOCALE;
2412 }
2413 else {
2414 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2415 goto needs_full_generality;
780fcc9f 2416 }
31f05a37
KW
2417 }
2418
a1dde8de 2419 if (c < 256) {
e7b7ac46 2420 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2421 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2422 }
2423
2f306ab9 2424 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2425 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2426 uvchr_to_utf8(p, c);
b9992569 2427 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
2428 }
2429 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2430 the special flags. */
2431 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2432
2433 needs_full_generality:
a0270393 2434 uvchr_to_utf8(utf8_c, c);
445bf929 2435 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 2436 }
84afefe6
JH
2437}
2438
26483009 2439PERL_STATIC_INLINE bool
5141f98e 2440S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2441 const char *const swashname, SV* const invlist)
bde6a22d 2442{
ea317ccb
KW
2443 /* returns a boolean giving whether or not the UTF8-encoded character that
2444 * starts at <p> is in the swash indicated by <swashname>. <swash>
2445 * contains a pointer to where the swash indicated by <swashname>
2446 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2447 * look at <*swash> and only generate a swash if it is not null. <invlist>
2448 * is NULL or an inversion list that defines the swash. If not null, it
2449 * saves time during initialization of the swash.
ea317ccb
KW
2450 *
2451 * Note that it is assumed that the buffer length of <p> is enough to
2452 * contain all the bytes that comprise the character. Thus, <*p> should
2453 * have been checked before this call for mal-formedness enough to assure
2454 * that. */
2455
7918f24d
NC
2456 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2457
492a624f 2458 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2459 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2460 * as far as there being enough bytes available in it to accommodate the
2461 * character without reading beyond the end, and pass that number on to the
2462 * validating routine */
6302f837 2463 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94
KW
2464 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2465 0,
2466 1 /* Die */ );
2467 NOT_REACHED; /* NOTREACHED */
28123549 2468 }
86ae6e94 2469
87367d5f
KW
2470 if (!*swash) {
2471 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2472 *swash = _core_swash_init("utf8",
2473
2474 /* Only use the name if there is no inversion
2475 * list; otherwise will go out to disk */
2476 (invlist) ? "" : swashname,
2477
2478 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2479 }
28123549 2480
bde6a22d
NC
2481 return swash_fetch(*swash, p, TRUE) != 0;
2482}
2483
da8c1a98
KW
2484PERL_STATIC_INLINE bool
2485S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2486 const char *const swashname, SV* const invlist)
2487{
2488 /* returns a boolean giving whether or not the UTF8-encoded character that
2489 * starts at <p>, and extending no further than <e - 1> is in the swash
2490 * indicated by <swashname>. <swash> contains a pointer to where the swash
2491 * indicated by <swashname> is to be stored; which this routine will do, so
2492 * that future calls will look at <*swash> and only generate a swash if it
2493 * is not null. <invlist> is NULL or an inversion list that defines the
2494 * swash. If not null, it saves time during initialization of the swash.
2495 */
2496
2497 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2498
2499 if (! isUTF8_CHAR(p, e)) {
2500 _force_out_malformed_utf8_message(p, e, 0, 1);
2501 NOT_REACHED; /* NOTREACHED */
2502 }
2503
2504 if (!*swash) {
2505 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2506 *swash = _core_swash_init("utf8",
2507
2508 /* Only use the name if there is no inversion
2509 * list; otherwise will go out to disk */
2510 (invlist) ? "" : swashname,
2511
2512 &PL_sv_undef, 1, 0, invlist, &flags);
2513 }
2514
2515 return swash_fetch(*swash, p, TRUE) != 0;
2516}
2517
bde6a22d 2518bool
922e8cb4
KW
2519Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2520{
922e8cb4
KW
2521 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2522
2523 assert(classnum < _FIRST_NON_SWASH_CC);
2524
f25ce844
KW
2525 return is_utf8_common(p,
2526 &PL_utf8_swash_ptrs[classnum],
2527 swash_property_names[classnum],
2528 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
2529}
2530
2531bool
da8c1a98
KW
2532Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2533 const U8 * const e)
2534{
2535 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2536
2537 assert(classnum < _FIRST_NON_SWASH_CC);
2538
2539 return is_utf8_common_with_len(p,
2540 e,
2541 &PL_utf8_swash_ptrs[classnum],
2542 swash_property_names[classnum],
2543 PL_XPosix_ptrs[classnum]);
2544}
2545
2546bool
f2645549 2547Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 2548{
f2645549 2549 SV* invlist = NULL;
7918f24d 2550
f2645549 2551 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 2552
f2645549
KW
2553 if (! PL_utf8_perl_idstart) {
2554 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2555 }
60071a22 2556 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
2557}
2558
2559bool
da8c1a98
KW
2560Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2561{
2562 SV* invlist = NULL;
2563
2564 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2565
2566 if (! PL_utf8_perl_idstart) {
2567 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2568 }
2569 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2570 "_Perl_IDStart", invlist);
2571}
2572
2573bool
f2645549 2574Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2575{
f2645549 2576 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2577
2578 if (*p == '_')
2579 return TRUE;
f25ce844 2580 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2581}
2582
2583bool
eba68aa0
KW
2584Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2585{
b24b43f7 2586 SV* invlist = NULL;
eba68aa0
KW
2587
2588 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2589
b24b43f7
KW
2590 if (! PL_utf8_perl_idcont) {
2591 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2592 }
60071a22 2593 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
2594}
2595
eba68aa0 2596bool
da8c1a98
KW
2597Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2598{
2599 SV* invlist = NULL;
2600
2601 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2602
2603 if (! PL_utf8_perl_idcont) {
2604 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2605 }
2606 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2607 "_Perl_IDCont", invlist);
2608}
2609
2610bool
f2645549 2611Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2612{
f2645549 2613 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2614
f25ce844 2615 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2616}
2617
2618bool
f2645549 2619Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2620{
f2645549 2621 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2622
f25ce844 2623 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2624}
2625
2626bool
7dbf68d2
KW
2627Perl__is_utf8_mark(pTHX_ const U8 *p)
2628{
7dbf68d2
KW
2629 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2630
f25ce844 2631 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2632}
2633
6b5c0936 2634/*
87cea99e 2635=for apidoc to_utf8_case
6b5c0936 2636
9da1e7cb
KW
2637Instead use the appropriate one of L</toUPPER_utf8>,
2638L</toTITLE_utf8>,
2639L</toLOWER_utf8>,
2640or L</toFOLD_utf8>.
2641
6fae5207 2642C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2643the character that is being converted. This routine assumes that the character
2644at C<p> is well-formed.
6b5c0936 2645
6fae5207
KW
2646C<ustrp> is a pointer to the character buffer to put the
2647conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2648of the result.
2649
6fae5207 2650C<swashp> is a pointer to the swash to use.
6b5c0936 2651
a1433954 2652Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2653and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2654but not always, a multicharacter mapping), is tried first.
6b5c0936 2655
4a8240a3
KW
2656C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2657any special mappings; C<""> means to use the special mappings. Values other
2658than these two are treated as the name of the hash containing the special
2659mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2660
796b6530
KW
2661C<normal> is a string like C<"ToLower"> which means the swash
2662C<%utf8::ToLower>.
0134edef 2663
760c7c2f
KW
2664Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2665unless those are turned off.
2666
0134edef 2667=cut */
6b5c0936 2668
2104c8d9 2669UV
9a957fbc
AL
2670Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2671 SV **swashp, const char *normal, const char *special)
a0ed51b3 2672{
b9992569
KW
2673 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2674
2675 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2676}
2677
2678 /* change namve uv1 to 'from' */
6a4a25f4 2679STATIC UV
b9992569
KW
2680S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2681 SV **swashp, const char *normal, const char *special)
2682{
0134edef 2683 STRLEN len = 0;
7918f24d 2684
b9992569 2685 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2686
36eaa811
KW
2687 /* For code points that don't change case, we already know that the output
2688 * of this function is the unchanged input, so we can skip doing look-ups
2689 * for them. Unfortunately the case-changing code points are scattered
2690 * around. But there are some long consecutive ranges where there are no
2691 * case changing code points. By adding tests, we can eliminate the lookup
2692 * for all the ones in such ranges. This is currently done here only for
2693 * just a few cases where the scripts are in common use in modern commerce
2694 * (and scripts adjacent to those which can be included without additional
2695 * tests). */
2696
2697 if (uv1 >= 0x0590) {
2698 /* This keeps from needing further processing the code points most
2699 * likely to be used in the following non-cased scripts: Hebrew,
2700 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2701 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2702 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2703 if (uv1 < 0x10A0) {
2704 goto cases_to_self;
2705 }
2706
2707 /* The following largish code point ranges also don't have case
2708 * changes, but khw didn't think they warranted extra tests to speed
2709 * them up (which would slightly slow down everything else above them):
2710 * 1100..139F Hangul Jamo, Ethiopic
2711 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2712 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2713 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2714 * Combining Diacritical Marks Extended, Balinese,
2715 * Sundanese, Batak, Lepcha, Ol Chiki
2716 * 2000..206F General Punctuation
2717 */
2718
2719 if (uv1 >= 0x2D30) {
2720
2721 /* This keeps the from needing further processing the code points
2722 * most likely to be used in the following non-cased major scripts:
2723 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2724 *
2725 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2726 * event that Unicode eventually allocates the unused block as of
2727 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2728 * that the test suite will start having failures to alert you
2729 * should that happen) */
2730 if (uv1 < 0xA640) {
2731 goto cases_to_self;
2732 }
2733
2734 if (uv1 >= 0xAC00) {
2735 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2736 if (ckWARN_d(WARN_SURROGATE)) {
2737 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2738 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2739 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2740 }
2741 goto cases_to_self;
2742 }
36eaa811
KW
2743
2744 /* AC00..FAFF Catches Hangul syllables and private use, plus
2745 * some others */
2746 if (uv1 < 0xFB00) {
2747 goto cases_to_self;
2748
2749 }
2750
5af9bc97
KW
2751 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2752 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2753 && ckWARN_d(WARN_DEPRECATED))
2754 {
2755 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2756 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2757 }
2758 if (ckWARN_d(WARN_NON_UNICODE)) {
2759 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2760 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2761 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2762 }
2763 goto cases_to_self;
2764 }
3bfc1e70
KW
2765#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2766 if (UNLIKELY(uv1
2767 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2768 {
2769
2770 /* As of this writing, this means we avoid swash creation
2771 * for anything beyond low Plane 1 */
2772 goto cases_to_self;
2773 }
2774#endif
36eaa811
KW
2775 }
2776 }
9ae3ac1a 2777
36eaa811
KW
2778 /* Note that non-characters are perfectly legal, so no warning should
2779 * be given. There are so few of them, that it isn't worth the extra
2780 * tests to avoid swash creation */
9ae3ac1a
KW
2781 }
2782
0134edef 2783 if (!*swashp) /* load on-demand */
5ab9d2ef 2784 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2785
a6f87d8c 2786 if (special) {
0134edef 2787 /* It might be "special" (sometimes, but not always,
2a37f04d 2788 * a multicharacter mapping) */
4a8240a3 2789 HV *hv = NULL;
b08cf34e
JH
2790 SV **svp;
2791
4a8240a3
KW
2792 /* If passed in the specials name, use that; otherwise use any
2793 * given in the swash */
2794 if (*special != '\0') {
2795 hv = get_hv(special, 0);
2796 }
2797 else {
2798 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2799 if (svp) {
2800 hv = MUTABLE_HV(SvRV(*svp));
2801 }
2802 }
2803
176fe009 2804 if (hv
5f560d8a 2805 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2806 && (*svp))
2807 {
cfd0369c 2808 const char *s;
47654450 2809
cfd0369c 2810 s = SvPV_const(*svp, len);
47654450 2811 if (len == 1)
f4cd282c 2812 /* EIGHTBIT */
c80e42f3 2813 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2814 else {
d2dcd0fb 2815 Copy(s, ustrp, len, U8);
29e98929 2816 }
983ffd37 2817 }
0134edef
JH
2818 }
2819
2820 if (!len && *swashp) {
4a4088c4 2821 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2822
0134edef
JH
2823 if (uv2) {
2824 /* It was "normal" (a single character mapping). */
f4cd282c 2825 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2826 }
2827 }
1feea2c7 2828
cbe07460
KW
2829 if (len) {
2830 if (lenp) {
2831 *lenp = len;
2832 }
2833 return valid_utf8_to_uvchr(ustrp, 0);
2834 }
2835
2836 /* Here, there was no mapping defined, which means that the code point maps
2837 * to itself. Return the inputs */
e24dfe9c 2838 cases_to_self:
bfdf22ec 2839 len = UTF8SKIP(p);
ca9fab46
KW
2840 if (p != ustrp) { /* Don't copy onto itself */
2841 Copy(p, ustrp, len, U8);
2842 }
0134edef 2843
2a37f04d
JH
2844 if (lenp)
2845 *lenp = len;
2846
f4cd282c 2847 return uv1;
cbe07460 2848
a0ed51b3
LW
2849}
2850
051a06d4 2851STATIC UV
357aadde 2852S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2853{
4a4088c4 2854 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2855 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2856 * result contains a character that crosses the 255/256 boundary, disallow
2857 * the change, and return the original code point. See L<perlfunc/lc> for
2858 * why;
051a06d4 2859 *
a1433954
KW
2860 * p points to the original string whose case was changed; assumed
2861 * by this routine to be well-formed
051a06d4
KW
2862 * result the code point of the first character in the changed-case string
2863 * ustrp points to the changed-case string (<result> represents its first char)
2864 * lenp points to the length of <ustrp> */
2865
2866 UV original; /* To store the first code point of <p> */
2867
2868 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2869
a4f12ed7 2870 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2871
2872 /* We know immediately if the first character in the string crosses the
2873 * boundary, so can skip */
2874 if (result > 255) {
2875
2876 /* Look at every character in the result; if any cross the
2877 * boundary, the whole thing is disallowed */
2878 U8* s = ustrp + UTF8SKIP(ustrp);
2879 U8* e = ustrp + *lenp;
2880 while (s < e) {
a4f12ed7 2881 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2882 goto bad_crossing;
2883 }
2884 s += UTF8SKIP(s);
2885 }
2886
613abc6d
KW
2887 /* Here, no characters crossed, result is ok as-is, but we warn. */
2888 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2889 return result;
2890 }
2891
7b52d656 2892 bad_crossing:
051a06d4
KW
2893
2894 /* Failed, have to return the original */
4b88fb76 2895 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2896
2897 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2898 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2899 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2900 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2901 OP_DESC(PL_op),
ab0b796c
KW
2902 original,
2903 original);
051a06d4
KW
2904 Copy(p, ustrp, *lenp, char);
2905 return original;
2906}
2907
eaf412bf
KW
2908/* The process for changing the case is essentially the same for the four case
2909 * change types, except there are complications for folding. Otherwise the
2910 * difference is only which case to change to. To make sure that they all do
2911 * the same thing, the bodies of the functions are extracted out into the
2912 * following two macros. The functions are written with the same variable
2913 * names, and these are known and used inside these macros. It would be
2914 * better, of course, to have inline functions to do it, but since different
2915 * macros are called, depending on which case is being changed to, this is not
2916 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2917 * function can start with the common start macro, then finish with its special
2918 * handling; while the other three cases can just use the common end macro.
2919 *
2920 * The algorithm is to use the proper (passed in) macro or function to change
2921 * the case for code points that are below 256. The macro is used if using
2922 * locale rules for the case change; the function if not. If the code point is
2923 * above 255, it is computed from the input UTF-8, and another macro is called
2924 * to do the conversion. If necessary, the output is converted to UTF-8. If
2925 * using a locale, we have to check that the change did not cross the 255/256
2926 * boundary, see check_locale_boundary_crossing() for further details.
2927 *
2928 * The macros are split with the correct case change for the below-256 case
2929 * stored into 'result', and in the middle of an else clause for the above-255
2930 * case. At that point in the 'else', 'result' is not the final result, but is
2931 * the input code point calculated from the UTF-8. The fold code needs to
2932 * realize all this and take it from there.
2933 *
2934 * If you read the two macros as sequential, it's easier to understand what's
2935 * going on. */
2936#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2937 L1_func_extra_param) \
2938 if (flags & (locale_flags)) { \
2939 /* Treat a UTF-8 locale as not being in locale at all */ \
2940 if (IN_UTF8_CTYPE_LOCALE) { \
2941 flags &= ~(locale_flags); \
2942 } \
2943 else { \
2944 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2945 } \
2946 } \
2947 \
2948 if (UTF8_IS_INVARIANT(*p)) { \
2949 if (flags & (locale_flags)) { \
2950 result = LC_L1_change_macro(*p); \
2951 } \
2952 else { \
2953 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2954 } \
2955 } \
2956 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2957 if (flags & (locale_flags)) { \
2958 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2959 *(p+1))); \
2960 } \
2961 else { \
2962 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2963 ustrp, lenp, L1_func_extra_param); \
2964 } \
2965 } \
2966 else { /* malformed UTF-8 */ \
2967 result = valid_utf8_to_uvchr(p, NULL); \
2968
2969#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2970 result = change_macro(result, p, ustrp, lenp); \
2971 \
2972 if (flags & (locale_flags)) { \
2973 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2974 } \
2975 return result; \
2976 } \
2977 \
2978 /* Here, used locale rules. Convert back to UTF-8 */ \
2979 if (UTF8_IS_INVARIANT(result)) { \
2980 *ustrp = (U8) result; \
2981 *lenp = 1; \
2982 } \
2983 else { \
2984 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2985 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2986 *lenp = 2; \
2987 } \
2988 \
2989 return result;
2990
d3e79532 2991/*
87cea99e 2992=for apidoc to_utf8_upper
d3e79532 2993
1f607577 2994Instead use L</toUPPER_utf8>.
a1433954 2995
d3e79532
JH
2996=cut */
2997
051a06d4 2998/* Not currently externally documented, and subject to change:
31f05a37
KW
2999 * <flags> is set iff iff the rules from the current underlying locale are to
3000 * be used. */
051a06d4 3001
2104c8d9 3002UV
31f05a37 3003Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3004{
051a06d4
KW
3005 UV result;
3006
3007 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 3008
eaf412bf
KW
3009 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3010 /* 2nd char of uc(U+DF) is 'S' */
3011 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3012 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 3013}
a0ed51b3 3014
d3e79532 3015/*
87cea99e 3016=for apidoc to_utf8_title
d3e79532 3017
1f607577 3018Instead use L</toTITLE_utf8>.
a1433954 3019
d3e79532
JH
3020=cut */
3021
051a06d4 3022/* Not currently externally documented, and subject to change:
31f05a37
KW
3023 * <flags> is set iff the rules from the current underlying locale are to be
3024 * used. Since titlecase is not defined in POSIX, for other than a
3025 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 3026 */
051a06d4 3027
983ffd37 3028UV
31f05a37 3029Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 3030{
051a06d4
KW
3031 UV result;
3032
3033 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3034
eaf412bf
KW
3035 /* 2nd char of ucfirst(U+DF) is 's' */
3036 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3037 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3038}
3039
d3e79532 3040/*
87cea99e 3041=for apidoc to_utf8_lower
d3e79532 3042
1f607577 3043Instead use L</toLOWER_utf8>.
a1433954 3044
d3e79532
JH
3045=cut */
3046
051a06d4 3047/* Not currently externally documented, and subject to change:
31f05a37
KW
3048 * <flags> is set iff iff the rules from the current underlying locale are to
3049 * be used.
3050 */
051a06d4 3051
2104c8d9 3052UV
31f05a37 3053Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3054{
051a06d4
KW
3055 UV result;
3056
051a06d4 3057 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3058
eaf412bf
KW
3059 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3060 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3061}
3062
d3e79532 3063/*
87cea99e 3064=for apidoc to_utf8_fold
d3e79532 3065
1f607577 3066Instead use L</toFOLD_utf8>.
a1433954 3067
d3e79532
JH
3068=cut */
3069
051a06d4
KW
3070/* Not currently externally documented, and subject to change,
3071 * in <flags>
31f05a37
KW
3072 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3073 * locale are to be used.
051a06d4
KW
3074 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3075 * otherwise simple folds
a0270393
KW
3076 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3077 * prohibited
445bf929 3078 */
36bb2ab6 3079
b4e400f9 3080UV
445bf929 3081Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3082{
051a06d4
KW
3083 UV result;
3084
36bb2ab6 3085 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3086
a0270393
KW
3087 /* These are mutually exclusive */
3088 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3089
50ba90ff
KW
3090 assert(p != ustrp); /* Otherwise overwrites */
3091
eaf412bf
KW
3092 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3093 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3094
eaf412bf 3095 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3096
1ca267a5
KW
3097 if (flags & FOLD_FLAGS_LOCALE) {
3098
76f2ffcd 3099# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3100 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3101
3102# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3103# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3104
3105 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3106
538e84ed
KW
3107 /* Special case these two characters, as what normally gets
3108 * returned under locale doesn't work */
76f2ffcd
KW
3109 if (UTF8SKIP(p) == cap_sharp_s_len
3110 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3111 {
ab0b796c
KW
3112 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3113 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3114 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3115 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3116 goto return_long_s;
3117 }
0766489e
KW
3118 else
3119#endif
3120 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3121 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3122 {
ab0b796c
KW
3123 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3124 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3125 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3126 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3127 goto return_ligature_st;
3128 }
74894415
KW
3129
3130#if UNICODE_MAJOR_VERSION == 3 \
3131 && UNICODE_DOT_VERSION == 0 \
3132 && UNICODE_DOT_DOT_VERSION == 1
3133# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3134
3135 /* And special case this on this Unicode version only, for the same
3136 * reaons the other two are special cased. They would cross the
3137 * 255/256 boundary which is forbidden under /l, and so the code
3138 * wouldn't catch that they are equivalent (which they are only in
3139 * this release) */
3140 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3141 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3142 {
3143 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3144 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3145 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3146 "resolved to \"\\x{0131}\".");
3147 goto return_dotless_i;
3148 }
3149#endif
3150
357aadde 3151 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3152 }
a0270393
KW
3153 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3154 return result;
3155 }
3156 else {
4a4088c4 3157 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3158 * character above the ASCII range, and the result should not
3159 * contain an ASCII character. */
a0270393
KW
3160
3161 UV original; /* To store the first code point of <p> */
3162
3163 /* Look at every character in the result; if any cross the
3164 * boundary, the whole thing is disallowed */
3165 U8* s = ustrp;
3166 U8* e = ustrp + *lenp;
3167 while (s < e) {
3168 if (isASCII(*s)) {
3169 /* Crossed, have to return the original */
3170 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3171
9fc2026f 3172 /* But in these instances, there is an alternative we can
1ca267a5 3173 * return that is valid */
0766489e
KW
3174 if (original == LATIN_SMALL_LETTER_SHARP_S
3175#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3176 || original == LATIN_CAPITAL_LETTER_SHARP_S
3177#endif
3178 ) {
1ca267a5
KW
3179 goto return_long_s;
3180 }
9fc2026f
KW
3181 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3182 goto return_ligature_st;
3183 }
74894415
KW
3184#if UNICODE_MAJOR_VERSION == 3 \
3185 && UNICODE_DOT_VERSION == 0 \
3186 && UNICODE_DOT_DOT_VERSION == 1
3187
3188 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3189 goto return_dotless_i;
3190 }
3191#endif
a0270393
KW
3192 Copy(p, ustrp, *lenp, char);
3193 return original;
3194 }
3195 s += UTF8SKIP(s);
3196 }
051a06d4 3197
a0270393
KW
3198 /* Here, no characters crossed, result is ok as-is */
3199 return result;
3200 }
051a06d4
KW
3201 }
3202
4a4088c4 3203 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3204 if (UTF8_IS_INVARIANT(result)) {
3205 *ustrp = (U8) result;
3206 *lenp = 1;
3207 }
3208 else {
62cb07ea
KW
3209 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3210 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3211 *lenp = 2;
3212 }
3213
051a06d4 3214 return result;
1ca267a5
KW
3215
3216 return_long_s:
3217 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3218 * folds to a string of two of these characters. By returning this
3219 * instead, then, e.g.,
3220 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3221 * works. */
3222
3223 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3224 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3225 ustrp, *lenp, U8);
3226 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3227
3228 return_ligature_st:
3229 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3230 * have the other one fold to it */
3231
3232 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3233 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3234 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3235
3236#if UNICODE_MAJOR_VERSION == 3 \
3237 && UNICODE_DOT_VERSION == 0 \
3238 && UNICODE_DOT_DOT_VERSION == 1
3239
3240 return_dotless_i:
3241 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3242 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3243 return LATIN_SMALL_LETTER_DOTLESS_I;
3244
3245#endif
3246
a0ed51b3
LW
3247}
3248
711a919c 3249/* Note:
f90a9a02 3250 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3251 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3252 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3253 */
c4a5db0c 3254
a0ed51b3 3255SV*
7fc63493 3256Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3257{
c4a5db0c
KW
3258 PERL_ARGS_ASSERT_SWASH_INIT;
3259
3260 /* Returns a copy of a swash initiated by the called function. This is the
3261 * public interface, and returning a copy prevents others from doing
3262 * mischief on the original */
3263
5d3d13d1 3264 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3265}
3266
3267SV*
5d3d13d1 3268Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3269{
2c1f00b9
YO
3270
3271 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3272 * use the following define */
3273
3274#define CORE_SWASH_INIT_RETURN(x) \
3275 PL_curpm= old_PL_curpm; \
3276 return x
3277
c4a5db0c 3278 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3279 * by calling utf8_heavy.pl in the general case. The returned value may be
3280 * the swash's inversion list instead if the input parameters allow it.
3281 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3282 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3283 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3284 *
3285 * This interface should only be used by functions that won't destroy or
3286 * adversely change the swash, as doing so affects all other uses of the
3287 * swash in the program; the general public should use 'Perl_swash_init'
3288 * instead.
3289 *
3290 * pkg is the name of the package that <name> should be in.
3291 * name is the name of the swash to find. Typically it is a Unicode
3292 * property name, including user-defined ones
3293 * listsv is a string to initialize the swash with. It must be of the form
3294 * documented as the subroutine return value in
3295 * L<perlunicode/User-Defined Character Properties>
3296 * minbits is the number of bits required to represent each data element.
3297 * It is '1' for binary properties.
3298 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3299 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3300 * flags_p if non-NULL is the address of various input and output flag bits
3301 * to the routine, as follows: ('I' means is input to the routine;
3302 * 'O' means output from the routine. Only flags marked O are
3303 * meaningful on return.)
3304 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3305 * came from a user-defined property. (I O)
5d3d13d1
KW
3306 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3307 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3308 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3309 * return of an inversion list instead of a swash hash if this routine
3310 * thinks that would result in faster execution of swash_fetch() later
3311 * on. (I)
9a53f6cf
KW
3312 *
3313 * Thus there are three possible inputs to find the swash: <name>,
3314 * <listsv>, and <invlist>. At least one must be specified. The result
3315 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3316 * actions can intersect, etc. what <name> gives. To avoid going out to
3317 * disk at all, <invlist> should specify completely what the swash should
3318 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3319 *
3320 * <invlist> is only valid for binary properties */
c4a5db0c 3321
2c1f00b9
YO
3322 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3323
c4a5db0c 3324 SV* retval = &PL_sv_undef;
83199d38 3325 HV* swash_hv = NULL;
87367d5f
KW
3326 const int invlist_swash_boundary =
3327 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3328 ? 512 /* Based on some benchmarking, but not extensive, see commit
3329 message */
3330 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3331
3332 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3333 assert(! invlist || minbits == 1);
3334
2c1f00b9
YO
3335 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3336 that triggered the swash init and the swash init perl logic itself.
3337 See perl #122747 */
3338
9a53f6cf
KW
3339 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3340 * so */
3341 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3342 dSP;
3343 const size_t pkg_len = strlen(pkg);
3344 const size_t name_len = strlen(name);
3345 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3346 SV* errsv_save;
3347 GV *method;
3348
3349 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3350
3351 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3352 ENTER;
69794297 3353 SAVEHINTS();
2782061f 3354 save_re_context();
650f067c
JL
3355 /* We might get here via a subroutine signature which uses a utf8
3356 * parameter name, at which point PL_subname will have been set
3357 * but not yet used. */
3358 save_item(PL_subname);
69794297
KW
3359 if (PL_parser && PL_parser->error_count)
3360 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3361 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3362 if (!method) { /* demand load UTF-8 */
69794297 3363 ENTER;
db2c6cb3
FC
3364 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3365 GvSV(PL_errgv) = NULL;
1a419e6b 3366#ifndef NO_TAINT_SUPPORT
69794297
KW
3367 /* It is assumed that callers of this routine are not passing in
3368 * any user derived data. */
2782061f
DM
3369 /* Need to do this after save_re_context() as it will set
3370 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3371 * in Perl_magic_get). Even line to create errsv_save can turn on
3372 * PL_tainted. */
284167a5
SM
3373 SAVEBOOL(TAINT_get);
3374 TAINT_NOT;
3375#endif
69794297
KW
3376 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3377 NULL);
eed484f9 3378 {
db2c6cb3
FC
3379 /* Not ERRSV, as there is no need to vivify a scalar we are
3380 about to discard. */
3381 SV * const errsv = GvSV(PL_errgv);
3382 if (!SvTRUE(errsv)) {
3383 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3384 SvREFCNT_dec(errsv);
3385 }
eed484f9 3386 }
69794297
KW
3387 LEAVE;
3388 }
3389 SPAGAIN;
3390 PUSHMARK(SP);
3391 EXTEND(SP,5);
3392 mPUSHp(pkg, pkg_len);
3393 mPUSHp(name, name_len);
3394 PUSHs(listsv);
3395 mPUSHi(minbits);
3396 mPUSHi(none);
3397 PUTBACK;
db2c6cb3
FC
3398 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3399 GvSV(PL_errgv) = NULL;
69794297
KW
3400 /* If we already have a pointer to the method, no need to use
3401 * call_method() to repeat the lookup. */
c41800a8
KW
3402 if (method
3403 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3404 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3405 {
3406 retval = *PL_stack_sp--;
3407 SvREFCNT_inc(retval);
3408 }
eed484f9 3409 {
db2c6cb3
FC
3410 /* Not ERRSV. See above. */
3411 SV * const errsv = GvSV(PL_errgv);
3412 if (!SvTRUE(errsv)) {
3413 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3414 SvREFCNT_dec(errsv);
3415 }
eed484f9 3416 }
ce3b816e 3417 LEAVE;
69794297
KW
3418 POPSTACK;
3419 if (IN_PERL_COMPILETIME) {
3420 CopHINTS_set(PL_curcop, PL_hints);
3421 }
3422 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3423 if (SvPOK(retval)) {
69794297
KW
3424
3425 /* If caller wants to handle missing properties, let them */
5d3d13d1 3426 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3427 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3428 }
3429 Perl_croak(aTHX_
147e3846 3430 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3431 SVfARG(retval));
a25b5927 3432 NOT_REACHED; /* NOTREACHED */
d95e4a00 3433 }
69794297 3434 }
9a53f6cf 3435 } /* End of calling the module to find the swash */
36eb48b4 3436
83199d38
KW
3437 /* If this operation fetched a swash, and we will need it later, get it */
3438 if (retval != &PL_sv_undef
3439 && (minbits == 1 || (flags_p
3440 && ! (*flags_p
3441 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3442 {
3443 swash_hv = MUTABLE_HV(SvRV(retval));
3444
3445 /* If we don't already know that there is a user-defined component to
3446 * this swash, and the user has indicated they wish to know if there is
3447 * one (by passing <flags_p>), find out */
3448 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3449 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3450 if (user_defined && SvUV(*user_defined)) {
3451 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3452 }
3453 }
3454 }
3455
36eb48b4
KW
3456 /* Make sure there is an inversion list for binary properties */
3457 if (minbits == 1) {
3458 SV** swash_invlistsvp = NULL;
3459 SV* swash_invlist = NULL;
9a53f6cf 3460 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3461 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3462 an unclaimed reference count */
36eb48b4 3463
9a53f6cf 3464 /* If this operation fetched a swash, get its already existing
83199d38 3465 * inversion list, or create one for it */
36eb48b4 3466
83199d38 3467 if (swash_hv) {
5c9f4bd2 3468 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3469 if (swash_invlistsvp) {
3470 swash_invlist = *swash_invlistsvp;
3471 invlist_in_swash_is_valid = TRUE;
3472 }
3473 else {
36eb48b4 3474 swash_invlist = _swash_to_invlist(retval);
02c85471 3475 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3476 }
3477 }
3478
3479 /* If an inversion list was passed in, have to include it */
3480 if (invlist) {
3481
3482 /* Any fetched swash will by now have an inversion list in it;
3483 * otherwise <swash_invlist> will be NULL, indicating that we
3484 * didn't fetch a swash */
3485 if (swash_invlist) {
3486
3487 /* Add the passed-in inversion list, which invalidates the one
3488 * already stored in the swash */
3489 invlist_in_swash_is_valid = FALSE;
eee4c920 3490 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3491 _invlist_union(invlist, swash_invlist, &swash_invlist);
3492 }
3493 else {
3494
87367d5f
KW
3495 /* Here, there is no swash already. Set up a minimal one, if
3496 * we are going to return a swash */
3497 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3498 swash_hv = newHV();
4aca0fe6 3499 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3500 }
9a53f6cf
KW
3501 swash_invlist = invlist;
3502 }
9a53f6cf
KW
3503 }
3504
3505 /* Here, we have computed the union of all the passed-in data. It may
3506 * be that there was an inversion list in the swash which didn't get
538e84ed 3507 * touched; otherwise save the computed one */
87367d5f
KW
3508 if (! invlist_in_swash_is_valid
3509 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3510 {
5c9f4bd2 3511 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3512 {
3513 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3514 }
cc34d8c5
FC
3515 /* We just stole a reference count. */
3516 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3517 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3518 }
87367d5f 3519
ee3222e3 3520 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3521 SvREADONLY_on(swash_invlist);
3522
c41800a8 3523 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3524 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3525 SvREFCNT_dec(retval);
02c85471
FC
3526 if (!swash_invlist_unclaimed)
3527 SvREFCNT_inc_simple_void_NN(swash_invlist);
3528 retval = newRV_noinc(swash_invlist);
87367d5f 3529 }
36eb48b4
KW
3530 }
3531
2c1f00b9
YO
3532 CORE_SWASH_INIT_RETURN(retval);
3533#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3534}
3535
035d37be
JH
3536
3537/* This API is wrong for special case conversions since we may need to
3538 * return several Unicode characters for a single Unicode character
3539 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3540 * the lower-level routine, and it is similarly broken for returning
38684baa 3541 * multiple values. --jhi
b9992569 3542 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3543/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3544
3545/* Note:
3546 * Returns the value of property/mapping C<swash> for the first character
3547 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3548 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3549 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3550 *
3551 * A "swash" is a hash which contains initially the keys/values set up by
3552 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3553 * property for all possible code points. Things are stored in a compact form
3554 * (see utf8_heavy.pl) so that calculation is required to find the actual
3555 * property value for a given code point. As code points are looked up, new
3556 * key/value pairs are added to the hash, so that the calculation doesn't have
3557 * to ever be re-done. Further, each calculation is done, not just for the
3558 * desired one, but for a whole block of code points adjacent to that one.
3559 * For binary properties on ASCII machines, the block is usually for 64 code
3560 * points, starting with a code point evenly divisible by 64. Thus if the
3561 * property value for code point 257 is requested, the code goes out and
3562 * calculates the property values for all 64 code points between 256 and 319,
3563 * and stores these as a single 64-bit long bit vector, called a "swatch",
3564 * under the key for code point 256. The key is the UTF-8 encoding for code
3565 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3566 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3567 * for code point 258 is then requested, this code realizes that it would be
3568 * stored under the key for 256, and would find that value and extract the
3569 * relevant bit, offset from 256.
3570 *
3571 * Non-binary properties are stored in as many bits as necessary to represent
3572 * their values (32 currently, though the code is more general than that), not
fc273927 3573 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3574 * vector that encompasses the property values for all code points whose UTF-8
3575 * representations are represented by the key. That is, for all code points
3576 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3577 * bytes of that.
680c470c 3578 */
a0ed51b3 3579UV
680c470c 3580Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3581{
ef8f7699 3582 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3583 U32 klen;
3584 U32 off;
9b56a019 3585 STRLEN slen = 0;
7d85a32c 3586 STRLEN needents;
cfd0369c 3587 const U8 *tmps = NULL;
979f2922 3588 SV *swatch;
08fb1ac5 3589 const U8 c = *ptr;
3568d838 3590
7918f24d
NC
3591 PERL_ARGS_ASSERT_SWASH_FETCH;
3592
87367d5f
KW
3593 /* If it really isn't a hash, it isn't really swash; must be an inversion
3594 * list */
3595 if (SvTYPE(hv) != SVt_PVHV) {
3596 return _invlist_contains_cp((SV*)hv,
3597 (do_utf8)
3598 ? valid_utf8_to_uvchr(ptr, NULL)
3599 : c);
3600 }
3601
08fb1ac5
KW
3602 /* We store the values in a "swatch" which is a vec() value in a swash
3603 * hash. Code points 0-255 are a single vec() stored with key length
3604 * (klen) 0. All other code points have a UTF-8 representation
3605 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3606 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3607 * length for them is the length of the encoded char - 1. ptr[klen] is the
3608 * final byte in the sequence representing the character */
3609 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3610 klen = 0;
3611 needents = 256;
3612 off = c;
3568d838 3613 }
08fb1ac5
KW
3614 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3615 klen = 0;
3616 needents = 256;
a62b247b 3617 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3618 }
3619 else {
08fb1ac5
KW
3620 klen = UTF8SKIP(ptr) - 1;
3621
3622 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3623 * the vec is the final byte in the sequence. (In EBCDIC this is
3624 * converted to I8 to get consecutive values.) To help you visualize
3625 * all this:
3626 * Straight 1047 After final byte
3627 * UTF-8 UTF-EBCDIC I8 transform
3628 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3629 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3630 * ...
3631 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3632 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3633 * ...
3634 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3635 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3636 * ...
3637 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3638 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3639 * ...
3640 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3641 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3642 *
3643 * (There are no discontinuities in the elided (...) entries.)
3644 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3645 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3646 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3647 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3648 * index into the vec() swatch (after subtracting 0x80, which we
3649 * actually do with an '&').
3650 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3651 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3652 * dicontinuities which go away by transforming it into I8, and we
3653 * effectively subtract 0xA0 to get the index. */
979f2922 3654 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3655 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3656 }
7d85a32c 3657
a0ed51b3 3658 /*
4a4088c4 3659 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
3660 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3661 * it's nothing to sniff at.) Pity we usually come through at least
3662 * two function calls to get here...
3663 *
3664 * NB: this code assumes that swatches are never modified, once generated!
3665 */
3666
3568d838 3667 if (hv == PL_last_swash_hv &&
a0ed51b3 3668 klen == PL_last_swash_klen &&
27da23d5 3669 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3670 {
3671 tmps = PL_last_swash_tmps;
3672 slen = PL_last_swash_slen;
3673 }
3674 else {
3675 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3676 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3677
b0e3252e 3678 /* If not cached, generate it via swatch_get */
979f2922 3679 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3680 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3681 {
3682 if (klen) {
3683 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3684 swatch = swatch_get(swash,
3685 code_point & ~((UV)needents - 1),
3686 needents);
3687 }
3688 else { /* For the first 256 code points, the swatch has a key of
3689 length 0 */
3690 swatch = swatch_get(swash, 0, needents);
3691 }
979f2922 3692
923e4eb5 3693 if (IN_PERL_COMPILETIME)
623e6609 3694 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3695
979f2922 3696 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3697
979f2922
ST
3698 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3699 || (slen << 3) < needents)
5637ef5b 3700 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
147e3846 3701 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
5637ef5b 3702 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3703 }
3704
3705 PL_last_swash_hv = hv;
16d8f38a 3706 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3707 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3708 /* FIXME change interpvar.h? */
3709 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3710 PL_last_swash_slen = slen;
3711 if (klen)
3712 Copy(ptr, PL_last_swash_key, klen, U8);
3713 }
3714
9faf8d75 3715 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3716 case 1:
e7aca353 3717 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3718 case 8:
e7aca353 3719 return ((UV) tmps[off]);
a0ed51b3
LW
3720 case 16:
3721 off <<= 1;
e7aca353
JH
3722 return
3723 ((UV) tmps[off ] << 8) +
3724 ((UV) tmps[off + 1]);
a0ed51b3
LW
3725 case 32:
3726 off <<= 2;