This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
utf8.c: Forbid zero-length malformation under DEBUGGING
[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
524080c4
KW
878flags. For allowed 0 length strings, this function returns 0; for allowed
879overlong sequences, the computed code point is returned; for all other allowed
880malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
881determinable reasonable value.
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;
1126 uv = 0; /* XXX It could be argued that this should be
1127 UNICODE_REPLACEMENT? */
1128 goto ready_to_handle_errors;
0c443dc2
JH
1129 }
1130
eb83ed87
KW
1131 expectlen = UTF8SKIP(s);
1132
1133 /* A well-formed UTF-8 character, as the vast majority of calls to this
1134 * function will be for, has this expected length. For efficiency, set
1135 * things up here to return it. It will be overriden only in those rare
1136 * cases where a malformation is found */
1137 if (retlen) {
1138 *retlen = expectlen;
1139 }
1140
1141 /* An invariant is trivially well-formed */
1d72bdf6 1142 if (UTF8_IS_INVARIANT(uv)) {
de69f3af 1143 return uv;
a0ed51b3 1144 }
67e989fb 1145
eb83ed87 1146 /* A continuation character can't start a valid sequence */
b5b9af04 1147 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
2b5e7bc2
KW
1148 possible_problems |= UTF8_GOT_CONTINUATION;
1149 curlen = 1;
1150 uv = UNICODE_REPLACEMENT;
1151 goto ready_to_handle_errors;
ba210ebe 1152 }
9041c2e3 1153
dcd27b3c
KW
1154 /* Here is not a continuation byte, nor an invariant. The only thing left
1155 * is a start byte (possibly for an overlong) */
1156
534752c1
KW
1157 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1158 * that indicate the number of bytes in the character's whole UTF-8
1159 * sequence, leaving just the bits that are part of the value. */
1160 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
ba210ebe 1161
e308b348
KW
1162 /* Setup the loop end point, making sure to not look past the end of the
1163 * input string, and flag it as too short if the size isn't big enough. */
1164 send = (U8*) s0;
1165 if (UNLIKELY(curlen < expectlen)) {
1166 possible_problems |= UTF8_GOT_SHORT;
1167 avail_len = curlen;
1168 send += curlen;
1169 }
1170 else {
1171 send += expectlen;
1172 }
1173 adjusted_send = send;
1174
eb83ed87 1175 /* Now, loop through the remaining bytes in the character's sequence,
e308b348 1176 * accumulating each into the working value as we go. */
eb83ed87 1177 for (s = s0 + 1; s < send; s++) {
b5b9af04 1178 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
8850bf83 1179 uv = UTF8_ACCUMULATE(uv, *s);
2b5e7bc2
KW
1180 continue;
1181 }
1182
1183 /* Here, found a non-continuation before processing all expected bytes.
1184 * This byte indicates the beginning of a new character, so quit, even
1185 * if allowing this malformation. */
2b5e7bc2 1186 possible_problems |= UTF8_GOT_NON_CONTINUATION;
e308b348 1187 break;
eb83ed87
KW
1188 } /* End of loop through the character's bytes */
1189
1190 /* Save how many bytes were actually in the character */
1191 curlen = s - s0;
1192
e308b348
KW
1193 /* A convenience macro that matches either of the too-short conditions. */
1194# define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
2f8f112e 1195
e308b348 1196 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
2b5e7bc2
KW
1197 uv_so_far = uv;
1198 uv = UNICODE_REPLACEMENT;
eb83ed87
KW
1199 }
1200
2b5e7bc2
KW
1201 /* Note that there are two types of too-short malformation. One is when
1202 * there is actual wrong data before the normal termination of the
1203 * sequence. The other is that the sequence wasn't complete before the end
1204 * of the data we are allowed to look at, based on the input 'curlen'.
1205 * This means that we were passed data for a partial character, but it is
1206 * valid as far as we saw. The other is definitely invalid. This
1207 * distinction could be important to a caller, so the two types are kept
1208 * separate. */
1209
1210 /* Check for overflow */
1211 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1212 possible_problems |= UTF8_GOT_OVERFLOW;
1213 uv = UNICODE_REPLACEMENT;
eb83ed87 1214 }
eb83ed87 1215
2b5e7bc2
KW
1216 /* Check for overlong. If no problems so far, 'uv' is the correct code
1217 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1218 * we must look at the UTF-8 byte sequence itself to see if it is for an
1219 * overlong */
1220 if ( ( LIKELY(! possible_problems)
1221 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1222 || ( UNLIKELY( possible_problems)
1223 && ( UNLIKELY(! UTF8_IS_START(*s0))
1224 || ( curlen > 1
1225 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1226 send - s0))))))
2f8f112e 1227 {
2b5e7bc2
KW
1228 possible_problems |= UTF8_GOT_LONG;
1229
2b5e7bc2
KW
1230 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1231 UV min_uv = uv_so_far;
1232 STRLEN i;
1233
1234 /* Here, the input is both overlong and is missing some trailing
1235 * bytes. There is no single code point it could be for, but there
1236 * may be enough information present to determine if what we have
1237 * so far is for an unallowed code point, such as for a surrogate.
1238 * The code below has the intelligence to determine this, but just
1239 * for non-overlong UTF-8 sequences. What we do here is calculate
1240 * the smallest code point the input could represent if there were
1241 * no too short malformation. Then we compute and save the UTF-8
1242 * for that, which is what the code below looks at instead of the
1243 * raw input. It turns out that the smallest such code point is
1244 * all we need. */
1245 for (i = curlen; i < expectlen; i++) {
1246 min_uv = UTF8_ACCUMULATE(min_uv,
1247 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1248 }
1249
1250 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1251 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1252 to free it ourselves if
1253 warnings are made fatal */
1254 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1255 }
eb83ed87
KW
1256 }
1257
2b5e7bc2
KW
1258 /* Now check that the input isn't for a problematic code point not allowed
1259 * by the input parameters. */
1260 /* isn't problematic if < this */
1261 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1262 || ( UNLIKELY(possible_problems)
1263 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
760c7c2f
KW
1264 && ((flags & ( UTF8_DISALLOW_NONCHAR
1265 |UTF8_DISALLOW_SURROGATE
1266 |UTF8_DISALLOW_SUPER
1267 |UTF8_DISALLOW_ABOVE_31_BIT
1268 |UTF8_WARN_NONCHAR
1269 |UTF8_WARN_SURROGATE
1270 |UTF8_WARN_SUPER
1271 |UTF8_WARN_ABOVE_31_BIT))
2b5e7bc2
KW
1272 /* In case of a malformation, 'uv' is not valid, and has
1273 * been changed to something in the Unicode range.
1274 * Currently we don't output a deprecation message if there
1275 * is already a malformation, so we don't have to special
1276 * case the test immediately below */
760c7c2f
KW
1277 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1278 && ckWARN_d(WARN_DEPRECATED))))
eb83ed87 1279 {
2b5e7bc2
KW
1280 /* If there were no malformations, or the only malformation is an
1281 * overlong, 'uv' is valid */
1282 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1283 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1284 possible_problems |= UTF8_GOT_SURROGATE;
1285 }
1286 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1287 possible_problems |= UTF8_GOT_SUPER;
1288 }
1289 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1290 possible_problems |= UTF8_GOT_NONCHAR;
1291 }
1292 }
1293 else { /* Otherwise, need to look at the source UTF-8, possibly
1294 adjusted to be non-overlong */
1295
1296 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1297 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
ea5ced44 1298 {
2b5e7bc2
KW
1299 possible_problems |= UTF8_GOT_SUPER;
1300 }
1301 else if (curlen > 1) {
1302 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1303 NATIVE_UTF8_TO_I8(*adjusted_s0),
1304 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
ea5ced44 1305 {
2b5e7bc2 1306 possible_problems |= UTF8_GOT_SUPER;
ea5ced44 1307 }
2b5e7bc2
KW
1308 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1309 NATIVE_UTF8_TO_I8(*adjusted_s0),
1310 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1311 {
1312 possible_problems |= UTF8_GOT_SURROGATE;
ea5ced44
KW
1313 }
1314 }
c0236afe 1315
2b5e7bc2
KW
1316 /* We need a complete well-formed UTF-8 character to discern
1317 * non-characters, so can't look for them here */
1318 }
1319 }
949cf498 1320
2b5e7bc2
KW
1321 ready_to_handle_errors:
1322
1323 /* At this point:
1324 * curlen contains the number of bytes in the sequence that
1325 * this call should advance the input by.
e308b348
KW
1326 * avail_len gives the available number of bytes passed in, but
1327 * only if this is less than the expected number of
1328 * bytes, based on the code point's start byte.
2b5e7bc2
KW
1329 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1330 * is set in it for each potential problem found.
1331 * uv contains the code point the input sequence
1332 * represents; or if there is a problem that prevents
1333 * a well-defined value from being computed, it is
1334 * some subsitute value, typically the REPLACEMENT
1335 * CHARACTER.
1336 * s0 points to the first byte of the character
1337 * send points to just after where that (potentially
1338 * partial) character ends
1339 * adjusted_s0 normally is the same as s0, but in case of an
1340 * overlong for which the UTF-8 matters below, it is
1341 * the first byte of the shortest form representation
1342 * of the input.
1343 * adjusted_send normally is the same as 'send', but if adjusted_s0
1344 * is set to something other than s0, this points one
1345 * beyond its end
1346 */
eb83ed87 1347
2b5e7bc2
KW
1348 if (UNLIKELY(possible_problems)) {
1349 bool disallowed = FALSE;
1350 const U32 orig_problems = possible_problems;
1351
1352 while (possible_problems) { /* Handle each possible problem */
1353 UV pack_warn = 0;
1354 char * message = NULL;
1355
1356 /* Each 'if' clause handles one problem. They are ordered so that
1357 * the first ones' messages will be displayed before the later
1358 * ones; this is kinda in decreasing severity order */
1359 if (possible_problems & UTF8_GOT_OVERFLOW) {
1360
1361 /* Overflow means also got a super and above 31 bits, but we
1362 * handle all three cases here */
1363 possible_problems
1364 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
f9380377
KW
1365 *errors |= UTF8_GOT_OVERFLOW;
1366
1367 /* But the API says we flag all errors found */
1368 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1369 *errors |= UTF8_GOT_SUPER;
1370 }
1371 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1372 *errors |= UTF8_GOT_ABOVE_31_BIT;
1373 }
2b5e7bc2
KW
1374
1375 disallowed = TRUE;
1376
1377 /* The warnings code explicitly says it doesn't handle the case
1378 * of packWARN2 and two categories which have parent-child
1379 * relationship. Even if it works now to raise the warning if
1380 * either is enabled, it wouldn't necessarily do so in the
1381 * future. We output (only) the most dire warning*/
1382 if (! (flags & UTF8_CHECK_ONLY)) {
1383 if (ckWARN_d(WARN_UTF8)) {
1384 pack_warn = packWARN(WARN_UTF8);
1385 }
1386 else if (ckWARN_d(WARN_NON_UNICODE)) {
1387 pack_warn = packWARN(WARN_NON_UNICODE);
1388 }
1389 if (pack_warn) {
1390 message = Perl_form(aTHX_ "%s: %s (overflows)",
1391 malformed_text,
1392 _byte_dump_string(s0, send - s0));
1393 }
1394 }
1395 }
1396 else if (possible_problems & UTF8_GOT_EMPTY) {
1397 possible_problems &= ~UTF8_GOT_EMPTY;
f9380377 1398 *errors |= UTF8_GOT_EMPTY;
2b5e7bc2
KW
1399
1400 if (! (flags & UTF8_ALLOW_EMPTY)) {
d1f8d421
KW
1401
1402 /* This so-called malformation is now treated as a bug in
1403 * the caller. If you have nothing to decode, skip calling
1404 * this function */
1405 assert(0);
1406
2b5e7bc2
KW
1407 disallowed = TRUE;
1408 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1409 pack_warn = packWARN(WARN_UTF8);
1410 message = Perl_form(aTHX_ "%s (empty string)",
1411 malformed_text);
1412 }
1413 }
1414 }
1415 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1416 possible_problems &= ~UTF8_GOT_CONTINUATION;
f9380377 1417 *errors |= UTF8_GOT_CONTINUATION;
2b5e7bc2
KW
1418
1419 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1420 disallowed = TRUE;
1421 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1422 pack_warn = packWARN(WARN_UTF8);
1423 message = Perl_form(aTHX_
1424 "%s: %s (unexpected continuation byte 0x%02x,"
1425 " with no preceding start byte)",
1426 malformed_text,
1427 _byte_dump_string(s0, 1), *s0);
1428 }
1429 }
1430 }
2b5e7bc2
KW
1431 else if (possible_problems & UTF8_GOT_SHORT) {
1432 possible_problems &= ~UTF8_GOT_SHORT;
f9380377 1433 *errors |= UTF8_GOT_SHORT;
2b5e7bc2
KW
1434
1435 if (! (flags & UTF8_ALLOW_SHORT)) {
1436 disallowed = TRUE;
1437 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1438 pack_warn = packWARN(WARN_UTF8);
1439 message = Perl_form(aTHX_
9a6c9c81 1440 "%s: %s (too short; %d byte%s available, need %d)",
2b5e7bc2
KW
1441 malformed_text,
1442 _byte_dump_string(s0, send - s0),
e308b348
KW
1443 (int)avail_len,
1444 avail_len == 1 ? "" : "s",
2b5e7bc2
KW
1445 (int)expectlen);
1446 }
1447 }
ba210ebe 1448
2b5e7bc2 1449 }
e308b348
KW
1450 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1451 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1452 *errors |= UTF8_GOT_NON_CONTINUATION;
1453
1454 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1455 disallowed = TRUE;
1456 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1457 pack_warn = packWARN(WARN_UTF8);
1458 message = Perl_form(aTHX_ "%s",
1459 unexpected_non_continuation_text(s0,
1460 send - s0,
1461 s - s0,
1462 (int) expectlen));
1463 }
1464 }
1465 }
2b5e7bc2
KW
1466 else if (possible_problems & UTF8_GOT_LONG) {
1467 possible_problems &= ~UTF8_GOT_LONG;
f9380377 1468 *errors |= UTF8_GOT_LONG;
2b5e7bc2
KW
1469
1470 if (! (flags & UTF8_ALLOW_LONG)) {
1471 disallowed = TRUE;
1472
1473 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1474 pack_warn = packWARN(WARN_UTF8);
1475
1476 /* These error types cause 'uv' to be something that
1477 * isn't what was intended, so can't use it in the
1478 * message. The other error types either can't
1479 * generate an overlong, or else the 'uv' is valid */
1480 if (orig_problems &
1481 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1482 {
1483 message = Perl_form(aTHX_
1484 "%s: %s (any UTF-8 sequence that starts"
1485 " with \"%s\" is overlong which can and"
1486 " should be represented with a"
1487 " different, shorter sequence)",
1488 malformed_text,
1489 _byte_dump_string(s0, send - s0),
1490 _byte_dump_string(s0, curlen));
1491 }
1492 else {
1493 U8 tmpbuf[UTF8_MAXBYTES+1];
1494 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1495 uv, 0);
1496 message = Perl_form(aTHX_
1497 "%s: %s (overlong; instead use %s to represent"
147e3846 1498 " U+%0*" UVXf ")",
2b5e7bc2
KW
1499 malformed_text,
1500 _byte_dump_string(s0, send - s0),
1501 _byte_dump_string(tmpbuf, e - tmpbuf),
1502 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1503 small code points */
1504 uv);
1505 }
1506 }
1507 }
1508 }
1509 else if (possible_problems & UTF8_GOT_SURROGATE) {
1510 possible_problems &= ~UTF8_GOT_SURROGATE;
1511
f9380377
KW
1512 if (flags & UTF8_WARN_SURROGATE) {
1513 *errors |= UTF8_GOT_SURROGATE;
1514
1515 if ( ! (flags & UTF8_CHECK_ONLY)
1516 && ckWARN_d(WARN_SURROGATE))
1517 {
2b5e7bc2
KW
1518 pack_warn = packWARN(WARN_SURROGATE);
1519
1520 /* These are the only errors that can occur with a
1521 * surrogate when the 'uv' isn't valid */
1522 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1523 message = Perl_form(aTHX_
1524 "UTF-16 surrogate (any UTF-8 sequence that"
1525 " starts with \"%s\" is for a surrogate)",
1526 _byte_dump_string(s0, curlen));
1527 }
1528 else {
1529 message = Perl_form(aTHX_
147e3846 1530 "UTF-16 surrogate U+%04" UVXf, uv);
2b5e7bc2 1531 }
f9380377 1532 }
2b5e7bc2 1533 }
ba210ebe 1534
2b5e7bc2
KW
1535 if (flags & UTF8_DISALLOW_SURROGATE) {
1536 disallowed = TRUE;
f9380377 1537 *errors |= UTF8_GOT_SURROGATE;
2b5e7bc2
KW
1538 }
1539 }
1540 else if (possible_problems & UTF8_GOT_SUPER) {
1541 possible_problems &= ~UTF8_GOT_SUPER;
949cf498 1542
f9380377
KW
1543 if (flags & UTF8_WARN_SUPER) {
1544 *errors |= UTF8_GOT_SUPER;
1545
1546 if ( ! (flags & UTF8_CHECK_ONLY)
1547 && ckWARN_d(WARN_NON_UNICODE))
1548 {
2b5e7bc2
KW
1549 pack_warn = packWARN(WARN_NON_UNICODE);
1550
1551 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1552 message = Perl_form(aTHX_
1553 "Any UTF-8 sequence that starts with"
1554 " \"%s\" is for a non-Unicode code point,"
1555 " may not be portable",
1556 _byte_dump_string(s0, curlen));
1557 }
1558 else {
1559 message = Perl_form(aTHX_
147e3846 1560 "Code point 0x%04" UVXf " is not"
2b5e7bc2
KW
1561 " Unicode, may not be portable",
1562 uv);
1563 }
f9380377 1564 }
2b5e7bc2 1565 }
ba210ebe 1566
2b5e7bc2
KW
1567 /* The maximum code point ever specified by a standard was
1568 * 2**31 - 1. Anything larger than that is a Perl extension
1569 * that very well may not be understood by other applications
1570 * (including earlier perl versions on EBCDIC platforms). We
1571 * test for these after the regular SUPER ones, and before
1572 * possibly bailing out, so that the slightly more dire warning
1573 * will override the regular one. */
1574 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1575 |UTF8_WARN_SUPER
1576 |UTF8_DISALLOW_ABOVE_31_BIT))
1577 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1578 && UNLIKELY(is_utf8_cp_above_31_bits(
1579 adjusted_s0,
1580 adjusted_send)))
1581 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1582 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1583 {
1584 if ( ! (flags & UTF8_CHECK_ONLY)
1585 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1586 && ckWARN_d(WARN_UTF8))
1587 {
1588 pack_warn = packWARN(WARN_UTF8);
1589
1590 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1591 message = Perl_form(aTHX_
1592 "Any UTF-8 sequence that starts with"
1593 " \"%s\" is for a non-Unicode code"
1594 " point, and is not portable",
1595 _byte_dump_string(s0, curlen));
1596 }
1597 else {
1598 message = Perl_form(aTHX_
147e3846 1599 "Code point 0x%" UVXf " is not Unicode,"
2b5e7bc2
KW
1600 " and not portable",
1601 uv);
1602 }
1603 }
1604
f9380377
KW
1605 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1606 *errors |= UTF8_GOT_ABOVE_31_BIT;
1607
1608 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1609 disallowed = TRUE;
1610 }
2b5e7bc2
KW
1611 }
1612 }
eb83ed87 1613
2b5e7bc2 1614 if (flags & UTF8_DISALLOW_SUPER) {
f9380377 1615 *errors |= UTF8_GOT_SUPER;
2b5e7bc2
KW
1616 disallowed = TRUE;
1617 }
eb83ed87 1618
2b5e7bc2
KW
1619 /* The deprecated warning overrides any non-deprecated one. If
1620 * there are other problems, a deprecation message is not
1621 * really helpful, so don't bother to raise it in that case.
1622 * This also keeps the code from having to handle the case
1623 * where 'uv' is not valid. */
1624 if ( ! (orig_problems
1625 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1626 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1627 && ckWARN_d(WARN_DEPRECATED))
1628 {
1629 message = Perl_form(aTHX_ cp_above_legal_max,
1630 uv, MAX_NON_DEPRECATED_CP);
1631 pack_warn = packWARN(WARN_DEPRECATED);
1632 }
1633 }
1634 else if (possible_problems & UTF8_GOT_NONCHAR) {
1635 possible_problems &= ~UTF8_GOT_NONCHAR;
ba210ebe 1636
f9380377
KW
1637 if (flags & UTF8_WARN_NONCHAR) {
1638 *errors |= UTF8_GOT_NONCHAR;
1639
1640 if ( ! (flags & UTF8_CHECK_ONLY)
1641 && ckWARN_d(WARN_NONCHAR))
1642 {
2b5e7bc2
KW
1643 /* The code above should have guaranteed that we don't
1644 * get here with errors other than overlong */
1645 assert (! (orig_problems
1646 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1647
1648 pack_warn = packWARN(WARN_NONCHAR);
1649 message = Perl_form(aTHX_ "Unicode non-character"
147e3846 1650 " U+%04" UVXf " is not recommended"
2b5e7bc2 1651 " for open interchange", uv);
f9380377 1652 }
2b5e7bc2 1653 }
5b311467 1654
2b5e7bc2
KW
1655 if (flags & UTF8_DISALLOW_NONCHAR) {
1656 disallowed = TRUE;
f9380377 1657 *errors |= UTF8_GOT_NONCHAR;
2b5e7bc2
KW
1658 }
1659 } /* End of looking through the possible flags */
1660
1661 /* Display the message (if any) for the problem being handled in
1662 * this iteration of the loop */
1663 if (message) {
1664 if (PL_op)
1665 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1666 OP_DESC(PL_op));
1667 else
1668 Perl_warner(aTHX_ pack_warn, "%s", message);
1669 }
1670 } /* End of 'while (possible_problems) {' */
a0dbb045 1671
2b5e7bc2
KW
1672 /* Since there was a possible problem, the returned length may need to
1673 * be changed from the one stored at the beginning of this function.
1674 * Instead of trying to figure out if that's needed, just do it. */
1675 if (retlen) {
1676 *retlen = curlen;
1677 }
a0dbb045 1678
2b5e7bc2
KW
1679 if (disallowed) {
1680 if (flags & UTF8_CHECK_ONLY && retlen) {
1681 *retlen = ((STRLEN) -1);
1682 }
1683 return 0;
1684 }
eb83ed87 1685 }
ba210ebe 1686
2b5e7bc2 1687 return UNI_TO_NATIVE(uv);
a0ed51b3
LW
1688}
1689
8e84507e 1690/*
ec5f19d0
KW
1691=for apidoc utf8_to_uvchr_buf
1692
1693Returns the native code point of the first character in the string C<s> which
1694is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
524080c4 1695C<*retlen> will be set to the length, in bytes, of that character.
ec5f19d0 1696
524080c4
KW
1697If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1698enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
796b6530 1699C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
173db420 1700(or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
796b6530 1701C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
173db420 1702the next possible position in C<s> that could begin a non-malformed character.
de69f3af 1703See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
173db420 1704returned.
ec5f19d0 1705
760c7c2f
KW
1706Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1707unless those are turned off.
1708
ec5f19d0 1709=cut
52be2536
KW
1710
1711Also implemented as a macro in utf8.h
1712
ec5f19d0
KW
1713*/
1714
1715
1716UV
1717Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1718{
ec5f19d0
KW
1719 assert(s < send);
1720
1721 return utf8n_to_uvchr(s, send - s, retlen,
1722 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1723}
1724
52be2536
KW
1725/* This is marked as deprecated
1726 *
ec5f19d0
KW
1727=for apidoc utf8_to_uvuni_buf
1728
de69f3af
KW
1729Only in very rare circumstances should code need to be dealing in Unicode
1730(as opposed to native) code points. In those few cases, use
1731C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
4f83cdcd
KW
1732
1733Returns the Unicode (not-native) code point of the first character in the
1734string C<s> which
ec5f19d0
KW
1735is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1736C<retlen> will be set to the length, in bytes, of that character.
1737
524080c4
KW
1738If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1739enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1740NULL) to -1. If those warnings are off, the computed value if well-defined (or
1741the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1742is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1743next possible position in C<s> that could begin a non-malformed character.
de69f3af 1744See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
ec5f19d0 1745
760c7c2f
KW
1746Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1747unless those are turned off.
1748
ec5f19d0
KW
1749=cut
1750*/
1751
1752UV
1753Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1754{
1755 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1756
1757 assert(send > s);
1758
5962d97e
KW
1759 /* Call the low level routine, asking for checks */
1760 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
ec5f19d0
KW
1761}
1762
b76347f2 1763/*
87cea99e 1764=for apidoc utf8_length
b76347f2
JH
1765
1766Return the length of the UTF-8 char encoded string C<s> in characters.
02eb7b47
JH
1767Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1768up past C<e>, croaks.
b76347f2
JH
1769
1770=cut
1771*/
1772
1773STRLEN
35a4481c 1774Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
b76347f2
JH
1775{
1776 STRLEN len = 0;
1777
7918f24d
NC
1778 PERL_ARGS_ASSERT_UTF8_LENGTH;
1779
8850bf83
JH
1780 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1781 * the bitops (especially ~) can create illegal UTF-8.
1782 * In other words: in Perl UTF-8 is not just for Unicode. */
1783
a3b680e6
AL
1784 if (e < s)
1785 goto warn_and_return;
b76347f2 1786 while (s < e) {
4cbf4130 1787 s += UTF8SKIP(s);
8e91ec7f
AV
1788 len++;
1789 }
1790
1791 if (e != s) {
1792 len--;
1793 warn_and_return:
9b387841
NC
1794 if (PL_op)
1795 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1796 "%s in %s", unees, OP_DESC(PL_op));
1797 else
61a12c31 1798 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
b76347f2
JH
1799 }
1800
1801 return len;
1802}
1803
b06226ff 1804/*
fed3ba5d
NC
1805=for apidoc bytes_cmp_utf8
1806
a1433954 1807Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
72d33970
FC
1808sequence of characters (stored as UTF-8)
1809in C<u>, C<ulen>. Returns 0 if they are
fed3ba5d
NC
1810equal, -1 or -2 if the first string is less than the second string, +1 or +2
1811if the first string is greater than the second string.
1812
1813-1 or +1 is returned if the shorter string was identical to the start of the
72d33970
FC
1814longer string. -2 or +2 is returned if
1815there was a difference between characters
fed3ba5d
NC
1816within the strings.
1817
1818=cut
1819*/
1820
1821int
1822Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1823{
1824 const U8 *const bend = b + blen;
1825 const U8 *const uend = u + ulen;
1826
1827 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
fed3ba5d
NC
1828
1829 while (b < bend && u < uend) {
1830 U8 c = *u++;
1831 if (!UTF8_IS_INVARIANT(c)) {
1832 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1833 if (u < uend) {
1834 U8 c1 = *u++;
1835 if (UTF8_IS_CONTINUATION(c1)) {
a62b247b 1836 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
fed3ba5d 1837 } else {
2b5e7bc2 1838 /* diag_listed_as: Malformed UTF-8 character%s */
fed3ba5d 1839 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
806547a7 1840 "%s %s%s",
7cf8d05d 1841 unexpected_non_continuation_text(u - 1, 2, 1, 2),
806547a7
KW
1842 PL_op ? " in " : "",
1843 PL_op ? OP_DESC(PL_op) : "");
fed3ba5d
NC
1844 return -2;
1845 }
1846 } else {
1847 if (PL_op)
1848 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1849 "%s in %s", unees, OP_DESC(PL_op));
1850 else
61a12c31 1851 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
fed3ba5d
NC
1852 return -2; /* Really want to return undef :-) */
1853 }
1854 } else {
1855 return -2;
1856 }
1857 }
1858 if (*b != c) {
1859 return *b < c ? -2 : +2;
1860 }
1861 ++b;
1862 }
1863
1864 if (b == bend && u == uend)
1865 return 0;
1866
1867 return b < bend ? +1 : -1;
1868}
1869
1870/*
87cea99e 1871=for apidoc utf8_to_bytes
6940069f 1872
2bbc8d55 1873Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954
KW
1874Unlike L</bytes_to_utf8>, this over-writes the original string, and
1875updates C<len> to contain the new length.
67e989fb 1876Returns zero on failure, setting C<len> to -1.
6940069f 1877
a1433954 1878If you need a copy of the string, see L</bytes_from_utf8>.
95be277c 1879
6940069f
GS
1880=cut
1881*/
1882
1883U8 *
37607a96 1884Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
6940069f 1885{
d4c19fe8
AL
1886 U8 * const save = s;
1887 U8 * const send = s + *len;
6940069f 1888 U8 *d;
246fae53 1889
7918f24d 1890 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
81611534 1891 PERL_UNUSED_CONTEXT;
7918f24d 1892
1e54db1a 1893 /* ensure valid UTF-8 and chars < 256 before updating string */
d4c19fe8 1894 while (s < send) {
d59937ca
KW
1895 if (! UTF8_IS_INVARIANT(*s)) {
1896 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1897 *len = ((STRLEN) -1);
1898 return 0;
1899 }
1900 s++;
dcad2880 1901 }
d59937ca 1902 s++;
246fae53 1903 }
dcad2880
JH
1904
1905 d = s = save;
6940069f 1906 while (s < send) {
80e0b38f
KW
1907 U8 c = *s++;
1908 if (! UTF8_IS_INVARIANT(c)) {
1909 /* Then it is two-byte encoded */
a62b247b 1910 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
80e0b38f
KW
1911 s++;
1912 }
1913 *d++ = c;
6940069f
GS
1914 }
1915 *d = '\0';
246fae53 1916 *len = d - save;
6940069f
GS
1917 return save;
1918}
1919
1920/*
87cea99e 1921=for apidoc bytes_from_utf8
f9a63242 1922
2bbc8d55 1923Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
a1433954 1924Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
ef9edfd0
JH
1925the newly-created string, and updates C<len> to contain the new
1926length. Returns the original string if no conversion occurs, C<len>
72d33970 1927is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
2bbc8d55 19280 if C<s> is converted or consisted entirely of characters that are invariant
4a4088c4 1929in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
f9a63242 1930
37607a96
PK
1931=cut
1932*/
f9a63242
JH
1933
1934U8 *
e1ec3a88 1935Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
f9a63242 1936{
f9a63242 1937 U8 *d;
e1ec3a88
AL
1938 const U8 *start = s;
1939 const U8 *send;
f9a63242
JH
1940 I32 count = 0;
1941
7918f24d 1942 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
96a5add6 1943 PERL_UNUSED_CONTEXT;
f9a63242 1944 if (!*is_utf8)
73d840c0 1945 return (U8 *)start;
f9a63242 1946
1e54db1a 1947 /* ensure valid UTF-8 and chars < 256 before converting string */
f9a63242 1948 for (send = s + *len; s < send;) {
d59937ca
KW
1949 if (! UTF8_IS_INVARIANT(*s)) {
1950 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
73d840c0 1951 return (U8 *)start;
d59937ca
KW
1952 }
1953 count++;
1954 s++;
db42d148 1955 }
d59937ca 1956 s++;
f9a63242
JH
1957 }
1958
35da51f7 1959 *is_utf8 = FALSE;
f9a63242 1960
212542aa 1961 Newx(d, (*len) - count + 1, U8);
ef9edfd0 1962 s = start; start = d;
f9a63242
JH
1963 while (s < send) {
1964 U8 c = *s++;
1a91c45d 1965 if (! UTF8_IS_INVARIANT(c)) {
c4d5f83a 1966 /* Then it is two-byte encoded */
a62b247b 1967 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1a91c45d 1968 s++;
c4d5f83a
NIS
1969 }
1970 *d++ = c;
f9a63242
JH
1971 }
1972 *d = '\0';
1973 *len = d - start;
73d840c0 1974 return (U8 *)start;
f9a63242
JH
1975}
1976
1977/*
87cea99e 1978=for apidoc bytes_to_utf8
6940069f 1979
ff97e5cf
KW
1980Converts a string C<s> of length C<len> bytes from the native encoding into
1981UTF-8.
6662521e 1982Returns a pointer to the newly-created string, and sets C<len> to
ff97e5cf 1983reflect the new length in bytes.
6940069f 1984
75200dff 1985A C<NUL> character will be written after the end of the string.
2bbc8d55
SP
1986
1987If you want to convert to UTF-8 from encodings other than
1988the native (Latin1 or EBCDIC),
a1433954 1989see L</sv_recode_to_utf8>().
c9ada85f 1990
497711e7 1991=cut
6940069f
GS
1992*/
1993
c682ebef
FC
1994/* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1995 likewise need duplication. */
1996
6940069f 1997U8*
35a4481c 1998Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
6940069f 1999{
35a4481c 2000 const U8 * const send = s + (*len);
6940069f
GS
2001 U8 *d;
2002 U8 *dst;
7918f24d
NC
2003
2004 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
96a5add6 2005 PERL_UNUSED_CONTEXT;
6940069f 2006
212542aa 2007 Newx(d, (*len) * 2 + 1, U8);
6940069f
GS
2008 dst = d;
2009
2010 while (s < send) {
55d09dc8
KW
2011 append_utf8_from_native_byte(*s, &d);
2012 s++;
6940069f
GS
2013 }
2014 *d = '\0';
6662521e 2015 *len = d-dst;
6940069f
GS
2016 return dst;
2017}
2018
a0ed51b3 2019/*
dea0fc0b 2020 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
a0ed51b3
LW
2021 *
2022 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2023 * We optimize for native, for obvious reasons. */
2024
2025U8*
dea0fc0b 2026Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3 2027{
dea0fc0b
JH
2028 U8* pend;
2029 U8* dstart = d;
2030
7918f24d
NC
2031 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2032
dea0fc0b 2033 if (bytelen & 1)
147e3846 2034 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
dea0fc0b
JH
2035
2036 pend = p + bytelen;
2037
a0ed51b3 2038 while (p < pend) {
dea0fc0b
JH
2039 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2040 p += 2;
2d1545e5 2041 if (OFFUNI_IS_INVARIANT(uv)) {
56d37426 2042 *d++ = LATIN1_TO_NATIVE((U8) uv);
a0ed51b3
LW
2043 continue;
2044 }
56d37426
KW
2045 if (uv <= MAX_UTF8_TWO_BYTE) {
2046 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2047 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
a0ed51b3
LW
2048 continue;
2049 }
46956fad
KW
2050#define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2051#define LAST_HIGH_SURROGATE 0xDBFF
2052#define FIRST_LOW_SURROGATE 0xDC00
2053#define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
e23c50db
KW
2054
2055 /* This assumes that most uses will be in the first Unicode plane, not
2056 * needing surrogates */
2057 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2058 && uv <= UNICODE_SURROGATE_LAST))
2059 {
2060 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2061 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2062 }
2063 else {
01ea242b 2064 UV low = (p[0] << 8) + p[1];
e23c50db
KW
2065 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2066 || UNLIKELY(low > LAST_LOW_SURROGATE))
2067 {
01ea242b 2068 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
e23c50db
KW
2069 }
2070 p += 2;
46956fad
KW
2071 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2072 + (low - FIRST_LOW_SURROGATE) + 0x10000;
01ea242b 2073 }
a0ed51b3 2074 }
56d37426
KW
2075#ifdef EBCDIC
2076 d = uvoffuni_to_utf8_flags(d, uv, 0);
2077#else
a0ed51b3 2078 if (uv < 0x10000) {
eb160463
GS
2079 *d++ = (U8)(( uv >> 12) | 0xe0);
2080 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2081 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2082 continue;
2083 }
2084 else {
eb160463
GS
2085 *d++ = (U8)(( uv >> 18) | 0xf0);
2086 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2087 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2088 *d++ = (U8)(( uv & 0x3f) | 0x80);
a0ed51b3
LW
2089 continue;
2090 }
56d37426 2091#endif
a0ed51b3 2092 }
dea0fc0b 2093 *newlen = d - dstart;
a0ed51b3
LW
2094 return d;
2095}
2096
2097/* Note: this one is slightly destructive of the source. */
2098
2099U8*
dea0fc0b 2100Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
a0ed51b3
LW
2101{
2102 U8* s = (U8*)p;
d4c19fe8 2103 U8* const send = s + bytelen;
7918f24d
NC
2104
2105 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2106
e0ea5e2d 2107 if (bytelen & 1)
147e3846 2108 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
e0ea5e2d
NC
2109 (UV)bytelen);
2110
a0ed51b3 2111 while (s < send) {
d4c19fe8 2112 const U8 tmp = s[0];
a0ed51b3
LW
2113 s[0] = s[1];
2114 s[1] = tmp;
2115 s += 2;
2116 }
dea0fc0b 2117 return utf16_to_utf8(p, d, bytelen, newlen);
a0ed51b3
LW
2118}
2119
922e8cb4
KW
2120bool
2121Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2122{
2123 U8 tmpbuf[UTF8_MAXBYTES+1];
2124 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2125 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
922e8cb4
KW
2126}
2127
f9ae8fb6
JD
2128/* Internal function so we can deprecate the external one, and call
2129 this one from other deprecated functions in this file */
2130
f2645549
KW
2131bool
2132Perl__is_utf8_idstart(pTHX_ const U8 *p)
61b19385 2133{
f2645549 2134 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
61b19385
KW
2135
2136 if (*p == '_')
2137 return TRUE;
f25ce844 2138 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
61b19385
KW
2139}
2140
5092f92a 2141bool
eba68aa0
KW
2142Perl__is_uni_perl_idcont(pTHX_ UV c)
2143{
2144 U8 tmpbuf[UTF8_MAXBYTES+1];
2145 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2146 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
eba68aa0
KW
2147}
2148
2149bool
f91dcd13
KW
2150Perl__is_uni_perl_idstart(pTHX_ UV c)
2151{
2152 U8 tmpbuf[UTF8_MAXBYTES+1];
2153 uvchr_to_utf8(tmpbuf, c);
da8c1a98 2154 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
f91dcd13
KW
2155}
2156
3a4c58c9
KW
2157UV
2158Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2159{
2160 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2161 * those, converting the result to UTF-8. The only difference between upper
3a4c58c9
KW
2162 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2163 * either "SS" or "Ss". Which one to use is passed into the routine in
2164 * 'S_or_s' to avoid a test */
2165
2166 UV converted = toUPPER_LATIN1_MOD(c);
2167
2168 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2169
2170 assert(S_or_s == 'S' || S_or_s == 's');
2171
6f2d5cbc 2172 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
f4cd282c 2173 characters in this range */
3a4c58c9
KW
2174 *p = (U8) converted;
2175 *lenp = 1;
2176 return converted;
2177 }
2178
2179 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2180 * which it maps to one of them, so as to only have to have one check for
2181 * it in the main case */
2182 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2183 switch (c) {
2184 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2185 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2186 break;
2187 case MICRO_SIGN:
2188 converted = GREEK_CAPITAL_LETTER_MU;
2189 break;
79e064b9
KW
2190#if UNICODE_MAJOR_VERSION > 2 \
2191 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2192 && UNICODE_DOT_DOT_VERSION >= 8)
3a4c58c9
KW
2193 case LATIN_SMALL_LETTER_SHARP_S:
2194 *(p)++ = 'S';
2195 *p = S_or_s;
2196 *lenp = 2;
2197 return 'S';
79e064b9 2198#endif
3a4c58c9
KW
2199 default:
2200 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
e5964223 2201 NOT_REACHED; /* NOTREACHED */
3a4c58c9
KW
2202 }
2203 }
2204
2205 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2206 *p = UTF8_TWO_BYTE_LO(converted);
2207 *lenp = 2;
2208
2209 return converted;
2210}
2211
50bda2c3
KW
2212/* Call the function to convert a UTF-8 encoded character to the specified case.
2213 * Note that there may be more than one character in the result.
2214 * INP is a pointer to the first byte of the input character
2215 * OUTP will be set to the first byte of the string of changed characters. It
2216 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2217 * LENP will be set to the length in bytes of the string of changed characters
2218 *
2219 * The functions return the ordinal of the first character in the string of OUTP */
b9992569
KW
2220#define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2221#define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2222#define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
50bda2c3 2223
b9992569
KW
2224/* This additionally has the input parameter 'specials', which if non-zero will
2225 * cause this to use the specials hash for folding (meaning get full case
50bda2c3 2226 * folding); otherwise, when zero, this implies a simple case fold */
b9992569 2227#define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
c3fd2246 2228
84afefe6
JH
2229UV
2230Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2231{
a1433954
KW
2232 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2233 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2234 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
c3fd2246
KW
2235 * the changed version may be longer than the original character.
2236 *
2237 * The ordinal of the first character of the changed version is returned
2238 * (but note, as explained above, that there may be more.) */
2239
7918f24d
NC
2240 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2241
3a4c58c9
KW
2242 if (c < 256) {
2243 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2244 }
2245
0ebc6274 2246 uvchr_to_utf8(p, c);
b9992569 2247 return CALL_UPPER_CASE(c, p, p, lenp);
a0ed51b3
LW
2248}
2249
84afefe6
JH
2250UV
2251Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2252{
7918f24d
NC
2253 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2254
3a4c58c9
KW
2255 if (c < 256) {
2256 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2257 }
2258
0ebc6274 2259 uvchr_to_utf8(p, c);
b9992569 2260 return CALL_TITLE_CASE(c, p, p, lenp);
a0ed51b3
LW
2261}
2262
afc16117 2263STATIC U8
eaf412bf 2264S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
afc16117
KW
2265{
2266 /* We have the latin1-range values compiled into the core, so just use
4a4088c4 2267 * those, converting the result to UTF-8. Since the result is always just
a1433954 2268 * one character, we allow <p> to be NULL */
afc16117
KW
2269
2270 U8 converted = toLOWER_LATIN1(c);
2271
eaf412bf
KW
2272 PERL_UNUSED_ARG(dummy);
2273
afc16117 2274 if (p != NULL) {
6f2d5cbc 2275 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
afc16117
KW
2276 *p = converted;
2277 *lenp = 1;
2278 }
2279 else {
430c9760
KW
2280 /* Result is known to always be < 256, so can use the EIGHT_BIT
2281 * macros */
2282 *p = UTF8_EIGHT_BIT_HI(converted);
2283 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
afc16117
KW
2284 *lenp = 2;
2285 }
2286 }
2287 return converted;
2288}
2289
84afefe6
JH
2290UV
2291Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
a0ed51b3 2292{
7918f24d
NC
2293 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2294
afc16117 2295 if (c < 256) {
eaf412bf 2296 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
bca00c02
KW
2297 }
2298
afc16117 2299 uvchr_to_utf8(p, c);
b9992569 2300 return CALL_LOWER_CASE(c, p, p, lenp);
a0ed51b3
LW
2301}
2302
84afefe6 2303UV
51910141 2304Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
a1dde8de 2305{
51910141 2306 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
1ca267a5 2307 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
51910141 2308 * FOLD_FLAGS_FULL iff full folding is to be used;
1ca267a5
KW
2309 *
2310 * Not to be used for locale folds
51910141 2311 */
f673fad4 2312
a1dde8de
KW
2313 UV converted;
2314
2315 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
81611534 2316 PERL_UNUSED_CONTEXT;
a1dde8de 2317
1ca267a5
KW
2318 assert (! (flags & FOLD_FLAGS_LOCALE));
2319
659a7c2d 2320 if (UNLIKELY(c == MICRO_SIGN)) {
a1dde8de
KW
2321 converted = GREEK_SMALL_LETTER_MU;
2322 }
9b63e895
KW
2323#if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2324 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2325 || UNICODE_DOT_DOT_VERSION > 0)
659a7c2d
KW
2326 else if ( (flags & FOLD_FLAGS_FULL)
2327 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2328 {
1ca267a5
KW
2329 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2330 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2331 * under those circumstances. */
2332 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2333 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2334 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2335 p, *lenp, U8);
2336 return LATIN_SMALL_LETTER_LONG_S;
2337 }
2338 else {
4f489194
KW
2339 *(p)++ = 's';
2340 *p = 's';
2341 *lenp = 2;
2342 return 's';
1ca267a5 2343 }
a1dde8de 2344 }
9b63e895 2345#endif
a1dde8de
KW
2346 else { /* In this range the fold of all other characters is their lower
2347 case */
2348 converted = toLOWER_LATIN1(c);
2349 }
2350
6f2d5cbc 2351 if (UVCHR_IS_INVARIANT(converted)) {
a1dde8de
KW
2352 *p = (U8) converted;
2353 *lenp = 1;
2354 }
2355 else {
2356 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2357 *p = UTF8_TWO_BYTE_LO(converted);
2358 *lenp = 2;
2359 }
2360
2361 return converted;
2362}
2363
2364UV
31f05a37 2365Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
84afefe6 2366{
4b593389 2367
a0270393
KW
2368 /* Not currently externally documented, and subject to change
2369 * <flags> bits meanings:
2370 * FOLD_FLAGS_FULL iff full folding is to be used;
31f05a37
KW
2371 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2372 * locale are to be used.
a0270393
KW
2373 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2374 */
4b593389 2375
36bb2ab6 2376 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
7918f24d 2377
780fcc9f
KW
2378 if (flags & FOLD_FLAGS_LOCALE) {
2379 /* Treat a UTF-8 locale as not being in locale at all */
2380 if (IN_UTF8_CTYPE_LOCALE) {
2381 flags &= ~FOLD_FLAGS_LOCALE;
2382 }
2383 else {
2384 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
e7b7ac46 2385 goto needs_full_generality;
780fcc9f 2386 }
31f05a37
KW
2387 }
2388
a1dde8de 2389 if (c < 256) {
e7b7ac46 2390 return _to_fold_latin1((U8) c, p, lenp,
31f05a37 2391 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
a1dde8de
KW
2392 }
2393
2f306ab9 2394 /* Here, above 255. If no special needs, just use the macro */
a0270393
KW
2395 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2396 uvchr_to_utf8(p, c);
b9992569 2397 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
a0270393
KW
2398 }
2399 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2400 the special flags. */
2401 U8 utf8_c[UTF8_MAXBYTES + 1];
e7b7ac46
KW
2402
2403 needs_full_generality:
a0270393 2404 uvchr_to_utf8(utf8_c, c);
445bf929 2405 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
a0270393 2406 }
84afefe6
JH
2407}
2408
26483009 2409PERL_STATIC_INLINE bool
5141f98e 2410S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
f25ce844 2411 const char *const swashname, SV* const invlist)
bde6a22d 2412{
ea317ccb
KW
2413 /* returns a boolean giving whether or not the UTF8-encoded character that
2414 * starts at <p> is in the swash indicated by <swashname>. <swash>
2415 * contains a pointer to where the swash indicated by <swashname>
2416 * is to be stored; which this routine will do, so that future calls will
f25ce844
KW
2417 * look at <*swash> and only generate a swash if it is not null. <invlist>
2418 * is NULL or an inversion list that defines the swash. If not null, it
2419 * saves time during initialization of the swash.
ea317ccb
KW
2420 *
2421 * Note that it is assumed that the buffer length of <p> is enough to
2422 * contain all the bytes that comprise the character. Thus, <*p> should
2423 * have been checked before this call for mal-formedness enough to assure
2424 * that. */
2425
7918f24d
NC
2426 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2427
492a624f 2428 /* The API should have included a length for the UTF-8 character in <p>,
28123549 2429 * but it doesn't. We therefore assume that p has been validated at least
492a624f
KW
2430 * as far as there being enough bytes available in it to accommodate the
2431 * character without reading beyond the end, and pass that number on to the
2432 * validating routine */
6302f837 2433 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
86ae6e94
KW
2434 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2435 0,
2436 1 /* Die */ );
2437 NOT_REACHED; /* NOTREACHED */
28123549 2438 }
86ae6e94 2439
87367d5f
KW
2440 if (!*swash) {
2441 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
f25ce844
KW
2442 *swash = _core_swash_init("utf8",
2443
2444 /* Only use the name if there is no inversion
2445 * list; otherwise will go out to disk */
2446 (invlist) ? "" : swashname,
2447
2448 &PL_sv_undef, 1, 0, invlist, &flags);
87367d5f 2449 }
28123549 2450
bde6a22d
NC
2451 return swash_fetch(*swash, p, TRUE) != 0;
2452}
2453
da8c1a98
KW
2454PERL_STATIC_INLINE bool
2455S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2456 const char *const swashname, SV* const invlist)
2457{
2458 /* returns a boolean giving whether or not the UTF8-encoded character that
2459 * starts at <p>, and extending no further than <e - 1> is in the swash
2460 * indicated by <swashname>. <swash> contains a pointer to where the swash
2461 * indicated by <swashname> is to be stored; which this routine will do, so
2462 * that future calls will look at <*swash> and only generate a swash if it
2463 * is not null. <invlist> is NULL or an inversion list that defines the
2464 * swash. If not null, it saves time during initialization of the swash.
2465 */
2466
2467 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2468
2469 if (! isUTF8_CHAR(p, e)) {
2470 _force_out_malformed_utf8_message(p, e, 0, 1);
2471 NOT_REACHED; /* NOTREACHED */
2472 }
2473
2474 if (!*swash) {
2475 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2476 *swash = _core_swash_init("utf8",
2477
2478 /* Only use the name if there is no inversion
2479 * list; otherwise will go out to disk */
2480 (invlist) ? "" : swashname,
2481
2482 &PL_sv_undef, 1, 0, invlist, &flags);
2483 }
2484
2485 return swash_fetch(*swash, p, TRUE) != 0;
2486}
2487
bde6a22d 2488bool
922e8cb4
KW
2489Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2490{
922e8cb4
KW
2491 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2492
2493 assert(classnum < _FIRST_NON_SWASH_CC);
2494
f25ce844
KW
2495 return is_utf8_common(p,
2496 &PL_utf8_swash_ptrs[classnum],
2497 swash_property_names[classnum],
2498 PL_XPosix_ptrs[classnum]);
922e8cb4
KW
2499}
2500
2501bool
da8c1a98
KW
2502Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2503 const U8 * const e)
2504{
2505 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2506
2507 assert(classnum < _FIRST_NON_SWASH_CC);
2508
2509 return is_utf8_common_with_len(p,
2510 e,
2511 &PL_utf8_swash_ptrs[classnum],
2512 swash_property_names[classnum],
2513 PL_XPosix_ptrs[classnum]);
2514}
2515
2516bool
f2645549 2517Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
a0ed51b3 2518{
f2645549 2519 SV* invlist = NULL;
7918f24d 2520
f2645549 2521 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
7918f24d 2522
f2645549
KW
2523 if (! PL_utf8_perl_idstart) {
2524 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2525 }
60071a22 2526 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
82686b01
JH
2527}
2528
2529bool
da8c1a98
KW
2530Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2531{
2532 SV* invlist = NULL;
2533
2534 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2535
2536 if (! PL_utf8_perl_idstart) {
2537 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2538 }
2539 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2540 "_Perl_IDStart", invlist);
2541}
2542
2543bool
f2645549 2544Perl__is_utf8_xidstart(pTHX_ const U8 *p)
c11ff943 2545{
f2645549 2546 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
c11ff943
KW
2547
2548 if (*p == '_')
2549 return TRUE;
f25ce844 2550 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
c11ff943
KW
2551}
2552
2553bool
eba68aa0
KW
2554Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2555{
b24b43f7 2556 SV* invlist = NULL;
eba68aa0
KW
2557
2558 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2559
b24b43f7
KW
2560 if (! PL_utf8_perl_idcont) {
2561 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2562 }
60071a22 2563 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
eba68aa0
KW
2564}
2565
eba68aa0 2566bool
da8c1a98
KW
2567Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2568{
2569 SV* invlist = NULL;
2570
2571 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2572
2573 if (! PL_utf8_perl_idcont) {
2574 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2575 }
2576 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2577 "_Perl_IDCont", invlist);
2578}
2579
2580bool
f2645549 2581Perl__is_utf8_idcont(pTHX_ const U8 *p)
82686b01 2582{
f2645549 2583 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
7918f24d 2584
f25ce844 2585 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
a0ed51b3
LW
2586}
2587
2588bool
f2645549 2589Perl__is_utf8_xidcont(pTHX_ const U8 *p)
c11ff943 2590{
f2645549 2591 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
c11ff943 2592
f25ce844 2593 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
c11ff943
KW
2594}
2595
2596bool
7dbf68d2
KW
2597Perl__is_utf8_mark(pTHX_ const U8 *p)
2598{
7dbf68d2
KW
2599 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2600
f25ce844 2601 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
7dbf68d2
KW
2602}
2603
6b5c0936 2604/*
87cea99e 2605=for apidoc to_utf8_case
6b5c0936 2606
9da1e7cb
KW
2607Instead use the appropriate one of L</toUPPER_utf8>,
2608L</toTITLE_utf8>,
2609L</toLOWER_utf8>,
2610or L</toFOLD_utf8>.
2611
6fae5207 2612C<p> contains the pointer to the UTF-8 string encoding
a1433954
KW
2613the character that is being converted. This routine assumes that the character
2614at C<p> is well-formed.
6b5c0936 2615
6fae5207
KW
2616C<ustrp> is a pointer to the character buffer to put the
2617conversion result to. C<lenp> is a pointer to the length
6b5c0936
JH
2618of the result.
2619
6fae5207 2620C<swashp> is a pointer to the swash to use.
6b5c0936 2621
a1433954 2622Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
796b6530 2623and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
0134edef 2624but not always, a multicharacter mapping), is tried first.
6b5c0936 2625
4a8240a3
KW
2626C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2627any special mappings; C<""> means to use the special mappings. Values other
2628than these two are treated as the name of the hash containing the special
2629mappings, like C<"utf8::ToSpecLower">.
6b5c0936 2630
796b6530
KW
2631C<normal> is a string like C<"ToLower"> which means the swash
2632C<%utf8::ToLower>.
0134edef 2633
760c7c2f
KW
2634Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2635unless those are turned off.
2636
0134edef 2637=cut */
6b5c0936 2638
2104c8d9 2639UV
9a957fbc
AL
2640Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2641 SV **swashp, const char *normal, const char *special)
a0ed51b3 2642{
b9992569
KW
2643 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2644
2645 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2646}
2647
2648 /* change namve uv1 to 'from' */
6a4a25f4 2649STATIC UV
b9992569
KW
2650S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2651 SV **swashp, const char *normal, const char *special)
2652{
0134edef 2653 STRLEN len = 0;
7918f24d 2654
b9992569 2655 PERL_ARGS_ASSERT__TO_UTF8_CASE;
7918f24d 2656
36eaa811
KW
2657 /* For code points that don't change case, we already know that the output
2658 * of this function is the unchanged input, so we can skip doing look-ups
2659 * for them. Unfortunately the case-changing code points are scattered
2660 * around. But there are some long consecutive ranges where there are no
2661 * case changing code points. By adding tests, we can eliminate the lookup
2662 * for all the ones in such ranges. This is currently done here only for
2663 * just a few cases where the scripts are in common use in modern commerce
2664 * (and scripts adjacent to those which can be included without additional
2665 * tests). */
2666
2667 if (uv1 >= 0x0590) {
2668 /* This keeps from needing further processing the code points most
2669 * likely to be used in the following non-cased scripts: Hebrew,
2670 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2671 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2672 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2673 if (uv1 < 0x10A0) {
2674 goto cases_to_self;
2675 }
2676
2677 /* The following largish code point ranges also don't have case
2678 * changes, but khw didn't think they warranted extra tests to speed
2679 * them up (which would slightly slow down everything else above them):
2680 * 1100..139F Hangul Jamo, Ethiopic
2681 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2682 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2683 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2684 * Combining Diacritical Marks Extended, Balinese,
2685 * Sundanese, Batak, Lepcha, Ol Chiki
2686 * 2000..206F General Punctuation
2687 */
2688
2689 if (uv1 >= 0x2D30) {
2690
2691 /* This keeps the from needing further processing the code points
2692 * most likely to be used in the following non-cased major scripts:
2693 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2694 *
2695 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2696 * event that Unicode eventually allocates the unused block as of
2697 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2698 * that the test suite will start having failures to alert you
2699 * should that happen) */
2700 if (uv1 < 0xA640) {
2701 goto cases_to_self;
2702 }
2703
2704 if (uv1 >= 0xAC00) {
2705 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
5af9bc97
KW
2706 if (ckWARN_d(WARN_SURROGATE)) {
2707 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2708 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
147e3846 2709 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
5af9bc97
KW
2710 }
2711 goto cases_to_self;
2712 }
36eaa811
KW
2713
2714 /* AC00..FAFF Catches Hangul syllables and private use, plus
2715 * some others */
2716 if (uv1 < 0xFB00) {
2717 goto cases_to_self;
2718
2719 }
2720
5af9bc97
KW
2721 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2722 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2723 && ckWARN_d(WARN_DEPRECATED))
2724 {
2725 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2726 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2727 }
2728 if (ckWARN_d(WARN_NON_UNICODE)) {
2729 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2730 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
147e3846 2731 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
5af9bc97
KW
2732 }
2733 goto cases_to_self;
2734 }
3bfc1e70
KW
2735#ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2736 if (UNLIKELY(uv1
2737 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2738 {
2739
2740 /* As of this writing, this means we avoid swash creation
2741 * for anything beyond low Plane 1 */
2742 goto cases_to_self;
2743 }
2744#endif
36eaa811
KW
2745 }
2746 }
9ae3ac1a 2747
36eaa811
KW
2748 /* Note that non-characters are perfectly legal, so no warning should
2749 * be given. There are so few of them, that it isn't worth the extra
2750 * tests to avoid swash creation */
9ae3ac1a
KW
2751 }
2752
0134edef 2753 if (!*swashp) /* load on-demand */
5ab9d2ef 2754 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
0134edef 2755
a6f87d8c 2756 if (special) {
0134edef 2757 /* It might be "special" (sometimes, but not always,
2a37f04d 2758 * a multicharacter mapping) */
4a8240a3 2759 HV *hv = NULL;
b08cf34e
JH
2760 SV **svp;
2761
4a8240a3
KW
2762 /* If passed in the specials name, use that; otherwise use any
2763 * given in the swash */
2764 if (*special != '\0') {
2765 hv = get_hv(special, 0);
2766 }
2767 else {
2768 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2769 if (svp) {
2770 hv = MUTABLE_HV(SvRV(*svp));
2771 }
2772 }
2773
176fe009 2774 if (hv
5f560d8a 2775 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
176fe009
KW
2776 && (*svp))
2777 {
cfd0369c 2778 const char *s;
47654450 2779
cfd0369c 2780 s = SvPV_const(*svp, len);
47654450 2781 if (len == 1)
f4cd282c 2782 /* EIGHTBIT */
c80e42f3 2783 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2a37f04d 2784 else {
d2dcd0fb 2785 Copy(s, ustrp, len, U8);
29e98929 2786 }
983ffd37 2787 }
0134edef
JH
2788 }
2789
2790 if (!len && *swashp) {
4a4088c4 2791 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
d4c19fe8 2792
0134edef
JH
2793 if (uv2) {
2794 /* It was "normal" (a single character mapping). */
f4cd282c 2795 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2a37f04d
JH
2796 }
2797 }
1feea2c7 2798
cbe07460
KW
2799 if (len) {
2800 if (lenp) {
2801 *lenp = len;
2802 }
2803 return valid_utf8_to_uvchr(ustrp, 0);
2804 }
2805
2806 /* Here, there was no mapping defined, which means that the code point maps
2807 * to itself. Return the inputs */
e24dfe9c 2808 cases_to_self:
bfdf22ec 2809 len = UTF8SKIP(p);
ca9fab46
KW
2810 if (p != ustrp) { /* Don't copy onto itself */
2811 Copy(p, ustrp, len, U8);
2812 }
0134edef 2813
2a37f04d
JH
2814 if (lenp)
2815 *lenp = len;
2816
f4cd282c 2817 return uv1;
cbe07460 2818
a0ed51b3
LW
2819}
2820
051a06d4 2821STATIC UV
357aadde 2822S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
051a06d4 2823{
4a4088c4 2824 /* This is called when changing the case of a UTF-8-encoded character above
31f05a37
KW
2825 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2826 * result contains a character that crosses the 255/256 boundary, disallow
2827 * the change, and return the original code point. See L<perlfunc/lc> for
2828 * why;
051a06d4 2829 *
a1433954
KW
2830 * p points to the original string whose case was changed; assumed
2831 * by this routine to be well-formed
051a06d4
KW
2832 * result the code point of the first character in the changed-case string
2833 * ustrp points to the changed-case string (<result> represents its first char)
2834 * lenp points to the length of <ustrp> */
2835
2836 UV original; /* To store the first code point of <p> */
2837
2838 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2839
a4f12ed7 2840 assert(UTF8_IS_ABOVE_LATIN1(*p));
051a06d4
KW
2841
2842 /* We know immediately if the first character in the string crosses the
2843 * boundary, so can skip */
2844 if (result > 255) {
2845
2846 /* Look at every character in the result; if any cross the
2847 * boundary, the whole thing is disallowed */
2848 U8* s = ustrp + UTF8SKIP(ustrp);
2849 U8* e = ustrp + *lenp;
2850 while (s < e) {
a4f12ed7 2851 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
051a06d4
KW
2852 goto bad_crossing;
2853 }
2854 s += UTF8SKIP(s);
2855 }
2856
613abc6d
KW
2857 /* Here, no characters crossed, result is ok as-is, but we warn. */
2858 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
051a06d4
KW
2859 return result;
2860 }
2861
7b52d656 2862 bad_crossing:
051a06d4
KW
2863
2864 /* Failed, have to return the original */
4b88fb76 2865 original = valid_utf8_to_uvchr(p, lenp);
ab0b796c
KW
2866
2867 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2868 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
147e3846
KW
2869 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2870 "resolved to \"\\x{%" UVXf "}\".",
357aadde 2871 OP_DESC(PL_op),
ab0b796c
KW
2872 original,
2873 original);
051a06d4
KW
2874 Copy(p, ustrp, *lenp, char);
2875 return original;
2876}
2877
eaf412bf
KW
2878/* The process for changing the case is essentially the same for the four case
2879 * change types, except there are complications for folding. Otherwise the
2880 * difference is only which case to change to. To make sure that they all do
2881 * the same thing, the bodies of the functions are extracted out into the
2882 * following two macros. The functions are written with the same variable
2883 * names, and these are known and used inside these macros. It would be
2884 * better, of course, to have inline functions to do it, but since different
2885 * macros are called, depending on which case is being changed to, this is not
2886 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2887 * function can start with the common start macro, then finish with its special
2888 * handling; while the other three cases can just use the common end macro.
2889 *
2890 * The algorithm is to use the proper (passed in) macro or function to change
2891 * the case for code points that are below 256. The macro is used if using
2892 * locale rules for the case change; the function if not. If the code point is
2893 * above 255, it is computed from the input UTF-8, and another macro is called
2894 * to do the conversion. If necessary, the output is converted to UTF-8. If
2895 * using a locale, we have to check that the change did not cross the 255/256
2896 * boundary, see check_locale_boundary_crossing() for further details.
2897 *
2898 * The macros are split with the correct case change for the below-256 case
2899 * stored into 'result', and in the middle of an else clause for the above-255
2900 * case. At that point in the 'else', 'result' is not the final result, but is
2901 * the input code point calculated from the UTF-8. The fold code needs to
2902 * realize all this and take it from there.
2903 *
2904 * If you read the two macros as sequential, it's easier to understand what's
2905 * going on. */
2906#define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2907 L1_func_extra_param) \
2908 if (flags & (locale_flags)) { \
2909 /* Treat a UTF-8 locale as not being in locale at all */ \
2910 if (IN_UTF8_CTYPE_LOCALE) { \
2911 flags &= ~(locale_flags); \
2912 } \
2913 else { \
2914 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2915 } \
2916 } \
2917 \
2918 if (UTF8_IS_INVARIANT(*p)) { \
2919 if (flags & (locale_flags)) { \
2920 result = LC_L1_change_macro(*p); \
2921 } \
2922 else { \
2923 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2924 } \
2925 } \
2926 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2927 if (flags & (locale_flags)) { \
2928 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2929 *(p+1))); \
2930 } \
2931 else { \
2932 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2933 ustrp, lenp, L1_func_extra_param); \
2934 } \
2935 } \
2936 else { /* malformed UTF-8 */ \
2937 result = valid_utf8_to_uvchr(p, NULL); \
2938
2939#define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2940 result = change_macro(result, p, ustrp, lenp); \
2941 \
2942 if (flags & (locale_flags)) { \
2943 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2944 } \
2945 return result; \
2946 } \
2947 \
2948 /* Here, used locale rules. Convert back to UTF-8 */ \
2949 if (UTF8_IS_INVARIANT(result)) { \
2950 *ustrp = (U8) result; \
2951 *lenp = 1; \
2952 } \
2953 else { \
2954 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2955 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2956 *lenp = 2; \
2957 } \
2958 \
2959 return result;
2960
d3e79532 2961/*
87cea99e 2962=for apidoc to_utf8_upper
d3e79532 2963
1f607577 2964Instead use L</toUPPER_utf8>.
a1433954 2965
d3e79532
JH
2966=cut */
2967
051a06d4 2968/* Not currently externally documented, and subject to change:
31f05a37
KW
2969 * <flags> is set iff iff the rules from the current underlying locale are to
2970 * be used. */
051a06d4 2971
2104c8d9 2972UV
31f05a37 2973Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 2974{
051a06d4
KW
2975 UV result;
2976
2977 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
7918f24d 2978
eaf412bf
KW
2979 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
2980 /* 2nd char of uc(U+DF) is 'S' */
2981 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
2982 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
983ffd37 2983}
a0ed51b3 2984
d3e79532 2985/*
87cea99e 2986=for apidoc to_utf8_title
d3e79532 2987
1f607577 2988Instead use L</toTITLE_utf8>.
a1433954 2989
d3e79532
JH
2990=cut */
2991
051a06d4 2992/* Not currently externally documented, and subject to change:
31f05a37
KW
2993 * <flags> is set iff the rules from the current underlying locale are to be
2994 * used. Since titlecase is not defined in POSIX, for other than a
2995 * UTF-8 locale, uppercase is used instead for code points < 256.
445bf929 2996 */
051a06d4 2997
983ffd37 2998UV
31f05a37 2999Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
983ffd37 3000{
051a06d4
KW
3001 UV result;
3002
3003 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
7918f24d 3004
eaf412bf
KW
3005 /* 2nd char of ucfirst(U+DF) is 's' */
3006 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3007 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
a0ed51b3
LW
3008}
3009
d3e79532 3010/*
87cea99e 3011=for apidoc to_utf8_lower
d3e79532 3012
1f607577 3013Instead use L</toLOWER_utf8>.
a1433954 3014
d3e79532
JH
3015=cut */
3016
051a06d4 3017/* Not currently externally documented, and subject to change:
31f05a37
KW
3018 * <flags> is set iff iff the rules from the current underlying locale are to
3019 * be used.
3020 */
051a06d4 3021
2104c8d9 3022UV
31f05a37 3023Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
a0ed51b3 3024{
051a06d4
KW
3025 UV result;
3026
051a06d4 3027 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
7918f24d 3028
eaf412bf
KW
3029 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3030 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
b4e400f9
JH
3031}
3032
d3e79532 3033/*
87cea99e 3034=for apidoc to_utf8_fold
d3e79532 3035
1f607577 3036Instead use L</toFOLD_utf8>.
a1433954 3037
d3e79532
JH
3038=cut */
3039
051a06d4
KW
3040/* Not currently externally documented, and subject to change,
3041 * in <flags>
31f05a37
KW
3042 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3043 * locale are to be used.
051a06d4
KW
3044 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3045 * otherwise simple folds
a0270393
KW
3046 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3047 * prohibited
445bf929 3048 */
36bb2ab6 3049
b4e400f9 3050UV
445bf929 3051Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
b4e400f9 3052{
051a06d4
KW
3053 UV result;
3054
36bb2ab6 3055 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
7918f24d 3056
a0270393
KW
3057 /* These are mutually exclusive */
3058 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3059
50ba90ff
KW
3060 assert(p != ustrp); /* Otherwise overwrites */
3061
eaf412bf
KW
3062 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3063 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
31f05a37 3064
eaf412bf 3065 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
a1dde8de 3066
1ca267a5
KW
3067 if (flags & FOLD_FLAGS_LOCALE) {
3068
76f2ffcd 3069# define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
0766489e
KW
3070 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3071
3072# ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3073# define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
76f2ffcd
KW
3074
3075 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
76f2ffcd 3076
538e84ed
KW
3077 /* Special case these two characters, as what normally gets
3078 * returned under locale doesn't work */
76f2ffcd
KW
3079 if (UTF8SKIP(p) == cap_sharp_s_len
3080 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
1ca267a5 3081 {
ab0b796c
KW
3082 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3083 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3084 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3085 "resolved to \"\\x{17F}\\x{17F}\".");
1ca267a5
KW
3086 goto return_long_s;
3087 }
0766489e
KW
3088 else
3089#endif
3090 if (UTF8SKIP(p) == long_s_t_len
76f2ffcd 3091 && memEQ((char *) p, LONG_S_T, long_s_t_len))
9fc2026f 3092 {
ab0b796c
KW
3093 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3094 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3095 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3096 "resolved to \"\\x{FB06}\".");
9fc2026f
KW
3097 goto return_ligature_st;
3098 }
74894415
KW
3099
3100#if UNICODE_MAJOR_VERSION == 3 \
3101 && UNICODE_DOT_VERSION == 0 \
3102 && UNICODE_DOT_DOT_VERSION == 1
3103# define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3104
3105 /* And special case this on this Unicode version only, for the same
3106 * reaons the other two are special cased. They would cross the
3107 * 255/256 boundary which is forbidden under /l, and so the code
3108 * wouldn't catch that they are equivalent (which they are only in
3109 * this release) */
3110 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3111 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3112 {
3113 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3114 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3115 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3116 "resolved to \"\\x{0131}\".");
3117 goto return_dotless_i;
3118 }
3119#endif
3120
357aadde 3121 return check_locale_boundary_crossing(p, result, ustrp, lenp);
051a06d4 3122 }
a0270393
KW
3123 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3124 return result;
3125 }
3126 else {
4a4088c4 3127 /* This is called when changing the case of a UTF-8-encoded
9fc2026f
KW
3128 * character above the ASCII range, and the result should not
3129 * contain an ASCII character. */
a0270393
KW
3130
3131 UV original; /* To store the first code point of <p> */
3132
3133 /* Look at every character in the result; if any cross the
3134 * boundary, the whole thing is disallowed */
3135 U8* s = ustrp;
3136 U8* e = ustrp + *lenp;
3137 while (s < e) {
3138 if (isASCII(*s)) {
3139 /* Crossed, have to return the original */
3140 original = valid_utf8_to_uvchr(p, lenp);
1ca267a5 3141
9fc2026f 3142 /* But in these instances, there is an alternative we can
1ca267a5 3143 * return that is valid */
0766489e
KW
3144 if (original == LATIN_SMALL_LETTER_SHARP_S
3145#ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3146 || original == LATIN_CAPITAL_LETTER_SHARP_S
3147#endif
3148 ) {
1ca267a5
KW
3149 goto return_long_s;
3150 }
9fc2026f
KW
3151 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3152 goto return_ligature_st;
3153 }
74894415
KW
3154#if UNICODE_MAJOR_VERSION == 3 \
3155 && UNICODE_DOT_VERSION == 0 \
3156 && UNICODE_DOT_DOT_VERSION == 1
3157
3158 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3159 goto return_dotless_i;
3160 }
3161#endif
a0270393
KW
3162 Copy(p, ustrp, *lenp, char);
3163 return original;
3164 }
3165 s += UTF8SKIP(s);
3166 }
051a06d4 3167
a0270393
KW
3168 /* Here, no characters crossed, result is ok as-is */
3169 return result;
3170 }
051a06d4
KW
3171 }
3172
4a4088c4 3173 /* Here, used locale rules. Convert back to UTF-8 */
051a06d4
KW
3174 if (UTF8_IS_INVARIANT(result)) {
3175 *ustrp = (U8) result;
3176 *lenp = 1;
3177 }
3178 else {
62cb07ea
KW
3179 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3180 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
051a06d4
KW
3181 *lenp = 2;
3182 }
3183
051a06d4 3184 return result;
1ca267a5
KW
3185
3186 return_long_s:
3187 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3188 * folds to a string of two of these characters. By returning this
3189 * instead, then, e.g.,
3190 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3191 * works. */
3192
3193 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3194 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3195 ustrp, *lenp, U8);
3196 return LATIN_SMALL_LETTER_LONG_S;
9fc2026f
KW
3197
3198 return_ligature_st:
3199 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3200 * have the other one fold to it */
3201
3202 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3203 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3204 return LATIN_SMALL_LIGATURE_ST;
74894415
KW
3205
3206#if UNICODE_MAJOR_VERSION == 3 \
3207 && UNICODE_DOT_VERSION == 0 \
3208 && UNICODE_DOT_DOT_VERSION == 1
3209
3210 return_dotless_i:
3211 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3212 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3213 return LATIN_SMALL_LETTER_DOTLESS_I;
3214
3215#endif
3216
a0ed51b3
LW
3217}
3218
711a919c 3219/* Note:
f90a9a02 3220 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
711a919c
ST
3221 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3222 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3223 */
c4a5db0c 3224
a0ed51b3 3225SV*
7fc63493 3226Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
a0ed51b3 3227{
c4a5db0c
KW
3228 PERL_ARGS_ASSERT_SWASH_INIT;
3229
3230 /* Returns a copy of a swash initiated by the called function. This is the
3231 * public interface, and returning a copy prevents others from doing
3232 * mischief on the original */
3233
5d3d13d1 3234 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
c4a5db0c
KW
3235}
3236
3237SV*
5d3d13d1 3238Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
c4a5db0c 3239{
2c1f00b9
YO
3240
3241 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3242 * use the following define */
3243
3244#define CORE_SWASH_INIT_RETURN(x) \
3245 PL_curpm= old_PL_curpm; \
3246 return x
3247
c4a5db0c 3248 /* Initialize and return a swash, creating it if necessary. It does this
87367d5f
KW
3249 * by calling utf8_heavy.pl in the general case. The returned value may be
3250 * the swash's inversion list instead if the input parameters allow it.
3251 * Which is returned should be immaterial to callers, as the only
923b6d4e
KW
3252 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3253 * and swash_to_invlist() handle both these transparently.
c4a5db0c
KW
3254 *
3255 * This interface should only be used by functions that won't destroy or
3256 * adversely change the swash, as doing so affects all other uses of the
3257 * swash in the program; the general public should use 'Perl_swash_init'
3258 * instead.
3259 *
3260 * pkg is the name of the package that <name> should be in.
3261 * name is the name of the swash to find. Typically it is a Unicode
3262 * property name, including user-defined ones
3263 * listsv is a string to initialize the swash with. It must be of the form
3264 * documented as the subroutine return value in
3265 * L<perlunicode/User-Defined Character Properties>
3266 * minbits is the number of bits required to represent each data element.
3267 * It is '1' for binary properties.
3268 * none I (khw) do not understand this one, but it is used only in tr///.
9a53f6cf 3269 * invlist is an inversion list to initialize the swash with (or NULL)
83199d38
KW
3270 * flags_p if non-NULL is the address of various input and output flag bits
3271 * to the routine, as follows: ('I' means is input to the routine;
3272 * 'O' means output from the routine. Only flags marked O are
3273 * meaningful on return.)
3274 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3275 * came from a user-defined property. (I O)
5d3d13d1
KW
3276 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3277 * when the swash cannot be located, to simply return NULL. (I)
87367d5f
KW
3278 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3279 * return of an inversion list instead of a swash hash if this routine
3280 * thinks that would result in faster execution of swash_fetch() later
3281 * on. (I)
9a53f6cf
KW
3282 *
3283 * Thus there are three possible inputs to find the swash: <name>,
3284 * <listsv>, and <invlist>. At least one must be specified. The result
3285 * will be the union of the specified ones, although <listsv>'s various
aabbdbda
KW
3286 * actions can intersect, etc. what <name> gives. To avoid going out to
3287 * disk at all, <invlist> should specify completely what the swash should
3288 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
9a53f6cf
KW
3289 *
3290 * <invlist> is only valid for binary properties */
c4a5db0c 3291
2c1f00b9
YO
3292 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3293
c4a5db0c 3294 SV* retval = &PL_sv_undef;
83199d38 3295 HV* swash_hv = NULL;
87367d5f
KW
3296 const int invlist_swash_boundary =
3297 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3298 ? 512 /* Based on some benchmarking, but not extensive, see commit
3299 message */
3300 : -1; /* Never return just an inversion list */
9a53f6cf
KW
3301
3302 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3303 assert(! invlist || minbits == 1);
3304
2c1f00b9
YO
3305 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3306 that triggered the swash init and the swash init perl logic itself.
3307 See perl #122747 */
3308
9a53f6cf
KW
3309 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3310 * so */
3311 if (listsv != &PL_sv_undef || strNE(name, "")) {
69794297
KW
3312 dSP;
3313 const size_t pkg_len = strlen(pkg);
3314 const size_t name_len = strlen(name);
3315 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3316 SV* errsv_save;
3317 GV *method;
3318
3319 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3320
3321 PUSHSTACKi(PERLSI_MAGIC);
ce3b816e 3322 ENTER;
69794297 3323 SAVEHINTS();
2782061f 3324 save_re_context();
650f067c
JL
3325 /* We might get here via a subroutine signature which uses a utf8
3326 * parameter name, at which point PL_subname will have been set
3327 * but not yet used. */
3328 save_item(PL_subname);
69794297
KW
3329 if (PL_parser && PL_parser->error_count)
3330 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3331 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4a4088c4 3332 if (!method) { /* demand load UTF-8 */
69794297 3333 ENTER;
db2c6cb3
FC
3334 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3335 GvSV(PL_errgv) = NULL;
1a419e6b 3336#ifndef NO_TAINT_SUPPORT
69794297
KW
3337 /* It is assumed that callers of this routine are not passing in
3338 * any user derived data. */
2782061f
DM
3339 /* Need to do this after save_re_context() as it will set
3340 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3341 * in Perl_magic_get). Even line to create errsv_save can turn on
3342 * PL_tainted. */
284167a5
SM
3343 SAVEBOOL(TAINT_get);
3344 TAINT_NOT;
3345#endif
69794297
KW
3346 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3347 NULL);
eed484f9 3348 {
db2c6cb3
FC
3349 /* Not ERRSV, as there is no need to vivify a scalar we are
3350 about to discard. */
3351 SV * const errsv = GvSV(PL_errgv);
3352 if (!SvTRUE(errsv)) {
3353 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3354 SvREFCNT_dec(errsv);
3355 }
eed484f9 3356 }
69794297
KW
3357 LEAVE;
3358 }
3359 SPAGAIN;
3360 PUSHMARK(SP);
3361 EXTEND(SP,5);
3362 mPUSHp(pkg, pkg_len);
3363 mPUSHp(name, name_len);
3364 PUSHs(listsv);
3365 mPUSHi(minbits);
3366 mPUSHi(none);
3367 PUTBACK;
db2c6cb3
FC
3368 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3369 GvSV(PL_errgv) = NULL;
69794297
KW
3370 /* If we already have a pointer to the method, no need to use
3371 * call_method() to repeat the lookup. */
c41800a8
KW
3372 if (method
3373 ? call_sv(MUTABLE_SV(method), G_SCALAR)
69794297
KW
3374 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3375 {
3376 retval = *PL_stack_sp--;
3377 SvREFCNT_inc(retval);
3378 }
eed484f9 3379 {
db2c6cb3
FC
3380 /* Not ERRSV. See above. */
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 }
ce3b816e 3387 LEAVE;
69794297
KW
3388 POPSTACK;
3389 if (IN_PERL_COMPILETIME) {
3390 CopHINTS_set(PL_curcop, PL_hints);
3391 }
3392 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
d95e4a00 3393 if (SvPOK(retval)) {
69794297
KW
3394
3395 /* If caller wants to handle missing properties, let them */
5d3d13d1 3396 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
2c1f00b9 3397 CORE_SWASH_INIT_RETURN(NULL);
69794297
KW
3398 }
3399 Perl_croak(aTHX_
147e3846 3400 "Can't find Unicode property definition \"%" SVf "\"",
69794297 3401 SVfARG(retval));
a25b5927 3402 NOT_REACHED; /* NOTREACHED */
d95e4a00 3403 }
69794297 3404 }
9a53f6cf 3405 } /* End of calling the module to find the swash */
36eb48b4 3406
83199d38
KW
3407 /* If this operation fetched a swash, and we will need it later, get it */
3408 if (retval != &PL_sv_undef
3409 && (minbits == 1 || (flags_p
3410 && ! (*flags_p
3411 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3412 {
3413 swash_hv = MUTABLE_HV(SvRV(retval));
3414
3415 /* If we don't already know that there is a user-defined component to
3416 * this swash, and the user has indicated they wish to know if there is
3417 * one (by passing <flags_p>), find out */
3418 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3419 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3420 if (user_defined && SvUV(*user_defined)) {
3421 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3422 }
3423 }
3424 }
3425
36eb48b4
KW
3426 /* Make sure there is an inversion list for binary properties */
3427 if (minbits == 1) {
3428 SV** swash_invlistsvp = NULL;
3429 SV* swash_invlist = NULL;
9a53f6cf 3430 bool invlist_in_swash_is_valid = FALSE;
02c85471
FC
3431 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3432 an unclaimed reference count */
36eb48b4 3433
9a53f6cf 3434 /* If this operation fetched a swash, get its already existing
83199d38 3435 * inversion list, or create one for it */
36eb48b4 3436
83199d38 3437 if (swash_hv) {
5c9f4bd2 3438 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
9a53f6cf
KW
3439 if (swash_invlistsvp) {
3440 swash_invlist = *swash_invlistsvp;
3441 invlist_in_swash_is_valid = TRUE;
3442 }
3443 else {
36eb48b4 3444 swash_invlist = _swash_to_invlist(retval);
02c85471 3445 swash_invlist_unclaimed = TRUE;
9a53f6cf
KW
3446 }
3447 }
3448
3449 /* If an inversion list was passed in, have to include it */
3450 if (invlist) {
3451
3452 /* Any fetched swash will by now have an inversion list in it;
3453 * otherwise <swash_invlist> will be NULL, indicating that we
3454 * didn't fetch a swash */
3455 if (swash_invlist) {
3456
3457 /* Add the passed-in inversion list, which invalidates the one
3458 * already stored in the swash */
3459 invlist_in_swash_is_valid = FALSE;
eee4c920 3460 SvREADONLY_off(swash_invlist); /* Turned on again below */
9a53f6cf
KW
3461 _invlist_union(invlist, swash_invlist, &swash_invlist);
3462 }
3463 else {
3464
87367d5f
KW
3465 /* Here, there is no swash already. Set up a minimal one, if
3466 * we are going to return a swash */
3467 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
971d486f 3468 swash_hv = newHV();
4aca0fe6 3469 retval = newRV_noinc(MUTABLE_SV(swash_hv));
87367d5f 3470 }
9a53f6cf
KW
3471 swash_invlist = invlist;
3472 }
9a53f6cf
KW
3473 }
3474
3475 /* Here, we have computed the union of all the passed-in data. It may
3476 * be that there was an inversion list in the swash which didn't get
538e84ed 3477 * touched; otherwise save the computed one */
87367d5f
KW
3478 if (! invlist_in_swash_is_valid
3479 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3480 {
5c9f4bd2 3481 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
69794297
KW
3482 {
3483 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3484 }
cc34d8c5
FC
3485 /* We just stole a reference count. */
3486 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3487 else SvREFCNT_inc_simple_void_NN(swash_invlist);
9a53f6cf 3488 }
87367d5f 3489
ee3222e3 3490 /* The result is immutable. Forbid attempts to change it. */
dbfdbd26
KW
3491 SvREADONLY_on(swash_invlist);
3492
c41800a8 3493 /* Use the inversion list stand-alone if small enough */
87367d5f
KW
3494 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3495 SvREFCNT_dec(retval);
02c85471
FC
3496 if (!swash_invlist_unclaimed)
3497 SvREFCNT_inc_simple_void_NN(swash_invlist);
3498 retval = newRV_noinc(swash_invlist);
87367d5f 3499 }
36eb48b4
KW
3500 }
3501
2c1f00b9
YO
3502 CORE_SWASH_INIT_RETURN(retval);
3503#undef CORE_SWASH_INIT_RETURN
a0ed51b3
LW
3504}
3505
035d37be
JH
3506
3507/* This API is wrong for special case conversions since we may need to
3508 * return several Unicode characters for a single Unicode character
3509 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3510 * the lower-level routine, and it is similarly broken for returning
38684baa 3511 * multiple values. --jhi
b9992569 3512 * For those, you should use S__to_utf8_case() instead */
b0e3252e 3513/* Now SWASHGET is recasted into S_swatch_get in this file. */
680c470c
ST
3514
3515/* Note:
3516 * Returns the value of property/mapping C<swash> for the first character
3517 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4a4088c4 3518 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3d0f8846 3519 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
af2af982
KW
3520 *
3521 * A "swash" is a hash which contains initially the keys/values set up by
3522 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3523 * property for all possible code points. Things are stored in a compact form
3524 * (see utf8_heavy.pl) so that calculation is required to find the actual
3525 * property value for a given code point. As code points are looked up, new
3526 * key/value pairs are added to the hash, so that the calculation doesn't have
3527 * to ever be re-done. Further, each calculation is done, not just for the
3528 * desired one, but for a whole block of code points adjacent to that one.
3529 * For binary properties on ASCII machines, the block is usually for 64 code
3530 * points, starting with a code point evenly divisible by 64. Thus if the
3531 * property value for code point 257 is requested, the code goes out and
3532 * calculates the property values for all 64 code points between 256 and 319,
3533 * and stores these as a single 64-bit long bit vector, called a "swatch",
3534 * under the key for code point 256. The key is the UTF-8 encoding for code
3535 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3536 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3537 * for code point 258 is then requested, this code realizes that it would be
3538 * stored under the key for 256, and would find that value and extract the
3539 * relevant bit, offset from 256.
3540 *
3541 * Non-binary properties are stored in as many bits as necessary to represent
3542 * their values (32 currently, though the code is more general than that), not
fc273927 3543 * as single bits, but the principle is the same: the value for each key is a
af2af982
KW
3544 * vector that encompasses the property values for all code points whose UTF-8
3545 * representations are represented by the key. That is, for all code points
3546 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3547 * bytes of that.
680c470c 3548 */
a0ed51b3 3549UV
680c470c 3550Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
a0ed51b3 3551{
ef8f7699 3552 HV *const hv = MUTABLE_HV(SvRV(swash));
3568d838
JH
3553 U32 klen;
3554 U32 off;
9b56a019 3555 STRLEN slen = 0;
7d85a32c 3556 STRLEN needents;
cfd0369c 3557 const U8 *tmps = NULL;
979f2922 3558 SV *swatch;
08fb1ac5 3559 const U8 c = *ptr;
3568d838 3560
7918f24d
NC
3561 PERL_ARGS_ASSERT_SWASH_FETCH;
3562
87367d5f
KW
3563 /* If it really isn't a hash, it isn't really swash; must be an inversion
3564 * list */
3565 if (SvTYPE(hv) != SVt_PVHV) {
3566 return _invlist_contains_cp((SV*)hv,
3567 (do_utf8)
3568 ? valid_utf8_to_uvchr(ptr, NULL)
3569 : c);
3570 }
3571
08fb1ac5
KW
3572 /* We store the values in a "swatch" which is a vec() value in a swash
3573 * hash. Code points 0-255 are a single vec() stored with key length
3574 * (klen) 0. All other code points have a UTF-8 representation
3575 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3576 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3577 * length for them is the length of the encoded char - 1. ptr[klen] is the
3578 * final byte in the sequence representing the character */
3579 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3580 klen = 0;
3581 needents = 256;
3582 off = c;
3568d838 3583 }
08fb1ac5
KW
3584 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3585 klen = 0;
3586 needents = 256;
a62b247b 3587 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
979f2922
ST
3588 }
3589 else {
08fb1ac5
KW
3590 klen = UTF8SKIP(ptr) - 1;
3591
3592 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3593 * the vec is the final byte in the sequence. (In EBCDIC this is
3594 * converted to I8 to get consecutive values.) To help you visualize
3595 * all this:
3596 * Straight 1047 After final byte
3597 * UTF-8 UTF-EBCDIC I8 transform
3598 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3599 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3600 * ...
3601 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3602 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3603 * ...
3604 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3605 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3606 * ...
3607 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3608 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3609 * ...
3610 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3611 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3612 *
3613 * (There are no discontinuities in the elided (...) entries.)
3614 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3615 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3616 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3617 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3618 * index into the vec() swatch (after subtracting 0x80, which we
3619 * actually do with an '&').
3620 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3621 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3622 * dicontinuities which go away by transforming it into I8, and we
3623 * effectively subtract 0xA0 to get the index. */
979f2922 3624 needents = (1 << UTF_ACCUMULATION_SHIFT);
bc3632a8 3625 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
979f2922 3626 }
7d85a32c 3627
a0ed51b3 3628 /*
4a4088c4 3629 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
a0ed51b3
LW
3630 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3631 * it's nothing to sniff at.) Pity we usually come through at least
3632 * two function calls to get here...
3633 *
3634 * NB: this code assumes that swatches are never modified, once generated!
3635 */
3636
3568d838 3637 if (hv == PL_last_swash_hv &&
a0ed51b3 3638 klen == PL_last_swash_klen &&
27da23d5 3639 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
a0ed51b3
LW
3640 {
3641 tmps = PL_last_swash_tmps;
3642 slen = PL_last_swash_slen;
3643 }
3644 else {
3645 /* Try our second-level swatch cache, kept in a hash. */
e1ec3a88 3646 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
a0ed51b3 3647
b0e3252e 3648 /* If not cached, generate it via swatch_get */
979f2922 3649 if (!svp || !SvPOK(*svp)
08fb1ac5
KW
3650 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3651 {
3652 if (klen) {
3653 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3654 swatch = swatch_get(swash,
3655 code_point & ~((UV)needents - 1),
3656 needents);
3657 }
3658 else { /* For the first 256 code points, the swatch has a key of
3659 length 0 */
3660 swatch = swatch_get(swash, 0, needents);
3661 }
979f2922 3662
923e4eb5 3663 if (IN_PERL_COMPILETIME)
623e6609 3664 CopHINTS_set(PL_curcop, PL_hints);
a0ed51b3 3665
979f2922 3666 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
a0ed51b3 3667
979f2922
ST
3668 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3669 || (slen << 3) < needents)
5637ef5b 3670 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
147e3846 3671 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
5637ef5b 3672 svp, tmps, (UV)slen, (UV)needents);
a0ed51b3
LW
3673 }
3674
3675 PL_last_swash_hv = hv;
16d8f38a 3676 assert(klen <= sizeof(PL_last_swash_key));
eac04b2e 3677 PL_last_swash_klen = (U8)klen;
cfd0369c
NC
3678 /* FIXME change interpvar.h? */
3679 PL_last_swash_tmps = (U8 *) tmps;
a0ed51b3
LW
3680 PL_last_swash_slen = slen;
3681 if (klen)
3682 Copy(ptr, PL_last_swash_key, klen, U8);
3683 }
3684
9faf8d75 3685 switch ((int)((slen << 3) / needents)) {
a0ed51b3 3686 case 1:
e7aca353 3687 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
a0ed51b3 3688 case 8:
e7aca353 3689 return ((UV) tmps[off]);
a0ed51b3
LW
3690 case 16:
3691 off <<= 1;
e7aca353
JH
3692 return
3693 ((UV) tmps[off ] << 8) +
3694 ((UV) tmps[off + 1]);
a0ed51b3
LW
3695 case 32:
3696 off <<= 2;
e7aca353
JH
3697 return
3698 ((UV) tmps[off ] << 24) +
3699 ((UV) tmps[off + 1] << 16) +
3700 ((UV) tmps[off + 2] << 8) +
3701 ((UV) tmps[off + 3]);
a0ed51b3 3702 }
5637ef5b 3703 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
147e3846 3704 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
670f1322 3705 NORETURN_FUNCTION_END;
a0ed51b3 3706}
2b9d42f0 3707
319009ee
KW
3708/* Read a single line of the main body of the swash input text. These are of
3709 * the form:
3710 * 0053 0056 0073
3711 * where each number is hex. The first two numbers form the minimum and
3712 * maximum of a range, and the third is the value associated with the range.
3713 * Not all swashes should have a third number
3714 *
3715 * On input: l points to the beginning of the line to be examined; it points
3716 * to somewhere in the string of the whole input text, and is
3717 * terminated by a \n or the null string terminator.
3718 * lend points to the null terminator of that string
3719 * wants_value is non-zero if the swash expects a third number
3720 * typestr is the name of the swash's mapping, like 'ToLower'
3721 * On output: *min, *max, and *val are set to the values read from the line.
3722 * returns a pointer just beyond the line examined. If there was no
3723 * valid min number on the line, returns lend+1
3724 */
3725
3726STATIC U8*
3727S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3728 co