3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
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.
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.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
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,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
39 static const char cp_above_legal_max[] =
40 "Use of code point 0x%" UVXf " is not allowed; the"
41 " permissible max is 0x%" UVXf;
43 #define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
46 =head1 Unicode Support
47 These are various utility functions for manipulating UTF8-encoded
48 strings. For the uninitiated, this is a method of representing arbitrary
49 Unicode characters as a variable number of bytes, in such a way that
50 characters in the ASCII range are unmodified, and a zero byte never appears
51 within non-zero characters.
57 Perl__force_out_malformed_utf8_message(pTHX_
58 const U8 *const p, /* First byte in UTF-8 sequence */
59 const U8 * const e, /* Final byte in sequence (may include
61 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
62 usually 0, or some DISALLOW flags */
63 const bool die_here) /* If TRUE, this function does not return */
65 /* This core-only function is to be called when a malformed UTF-8 character
66 * is found, in order to output the detailed information about the
67 * malformation before dieing. The reason it exists is for the occasions
68 * when such a malformation is fatal, but warnings might be turned off, so
69 * that normally they would not be actually output. This ensures that they
70 * do get output. Because a sequence may be malformed in more than one
71 * way, multiple messages may be generated, so we can't make them fatal, as
72 * that would cause the first one to die.
74 * Instead we pretend -W was passed to perl, then die afterwards. The
75 * flexibility is here to return to the caller so they can finish up and
79 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
85 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
87 PL_curcop->cop_warnings = pWARN_ALL;
90 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
95 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
96 " be called only when there are errors found");
100 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
105 =for apidoc uvoffuni_to_utf8_flags
107 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
108 Instead, B<Almost all code should use L</uvchr_to_utf8> or
109 L</uvchr_to_utf8_flags>>.
111 This function is like them, but the input is a strict Unicode
112 (as opposed to native) code point. Only in very rare circumstances should code
113 not be using the native code point.
115 For details, see the description for L</uvchr_to_utf8_flags>.
120 /* All these formats take a single UV code point argument */
121 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
122 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
123 " is not recommended for open interchange";
124 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
125 " may not be portable";
126 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
127 " Unicode, requires a Perl extension," \
128 " and so is not portable";
130 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
132 if (flags & UNICODE_WARN_SURROGATE) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
134 surrogate_cp_format, uv); \
136 if (flags & UNICODE_DISALLOW_SURROGATE) { \
141 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
143 if (flags & UNICODE_WARN_NONCHAR) { \
144 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
145 nonchar_cp_format, uv); \
147 if (flags & UNICODE_DISALLOW_NONCHAR) { \
152 /* Use shorter names internally in this file */
153 #define SHIFT UTF_ACCUMULATION_SHIFT
155 #define MARK UTF_CONTINUATION_MARK
156 #define MASK UTF_CONTINUATION_MASK
159 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
161 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
163 if (OFFUNI_IS_INVARIANT(uv)) {
164 *d++ = LATIN1_TO_NATIVE(uv);
168 if (uv <= MAX_UTF8_TWO_BYTE) {
169 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
170 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
174 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
175 * below, the 16 is for start bytes E0-EF (which are all the possible ones
176 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
177 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
178 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
179 * 0x800-0xFFFF on ASCII */
180 if (uv < (16 * (1U << (2 * SHIFT)))) {
181 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
182 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
183 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
185 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
186 aren't tested here */
187 /* The most likely code points in this range are below the surrogates.
188 * Do an extra test to quickly exclude those. */
189 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
190 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
191 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
193 HANDLE_UNICODE_NONCHAR(uv, flags);
195 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
196 HANDLE_UNICODE_SURROGATE(uv, flags);
203 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
204 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
205 * happen starting with 4-byte characters on ASCII platforms. We unify the
206 * code for these with EBCDIC, even though some of them require 5-bytes on
207 * those, because khw believes the code saving is worth the very slight
208 * performance hit on these high EBCDIC code points. */
210 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
211 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
212 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
214 if ( (flags & UNICODE_WARN_SUPER)
215 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
216 && UNICODE_IS_PERL_EXTENDED(uv)))
218 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
220 /* Choose the more dire applicable warning */
221 (UNICODE_IS_PERL_EXTENDED(uv))
222 ? perl_extended_cp_format
226 if ( (flags & UNICODE_DISALLOW_SUPER)
227 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
228 && UNICODE_IS_PERL_EXTENDED(uv)))
233 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
234 HANDLE_UNICODE_NONCHAR(uv, flags);
237 /* Test for and handle 4-byte result. In the test immediately below, the
238 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
239 * characters). The 3 is for 3 continuation bytes; these each contribute
240 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
241 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
242 * 0x1_0000-0x1F_FFFF on ASCII */
243 if (uv < (8 * (1U << (3 * SHIFT)))) {
244 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
245 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
246 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
247 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
249 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
250 characters. The end-plane non-characters for EBCDIC were
251 handled just above */
252 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
253 HANDLE_UNICODE_NONCHAR(uv, flags);
255 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
256 HANDLE_UNICODE_SURROGATE(uv, flags);
263 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
264 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
265 * format. The unrolled version above turns out to not save all that much
266 * time, and at these high code points (well above the legal Unicode range
267 * on ASCII platforms, and well above anything in common use in EBCDIC),
268 * khw believes that less code outweighs slight performance gains. */
271 STRLEN len = OFFUNISKIP(uv);
274 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
275 uv >>= UTF_ACCUMULATION_SHIFT;
277 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
283 =for apidoc uvchr_to_utf8
285 Adds the UTF-8 representation of the native code point C<uv> to the end
286 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
287 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
288 the byte after the end of the new character. In other words,
290 d = uvchr_to_utf8(d, uv);
292 is the recommended wide native character-aware way of saying
296 This function accepts any code point from 0..C<IV_MAX> as input.
297 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
299 It is possible to forbid or warn on non-Unicode code points, or those that may
300 be problematic by using L</uvchr_to_utf8_flags>.
305 /* This is also a macro */
306 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
309 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
311 return uvchr_to_utf8(d, uv);
315 =for apidoc uvchr_to_utf8_flags
317 Adds the UTF-8 representation of the native code point C<uv> to the end
318 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
319 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
320 the byte after the end of the new character. In other words,
322 d = uvchr_to_utf8_flags(d, uv, flags);
326 d = uvchr_to_utf8_flags(d, uv, 0);
328 This is the Unicode-aware way of saying
332 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
333 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
335 Specifying C<flags> can further restrict what is allowed and not warned on, as
338 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
339 the function will raise a warning, provided UTF8 warnings are enabled. If
340 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
341 NULL. If both flags are set, the function will both warn and return NULL.
343 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
344 affect how the function handles a Unicode non-character.
346 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
347 affect the handling of code points that are above the Unicode maximum of
348 0x10FFFF. Languages other than Perl may not be able to accept files that
351 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
352 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
353 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
354 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
355 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
356 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
357 above-Unicode and surrogate flags, but not the non-character ones, as
359 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
360 See L<perlunicode/Noncharacter code points>.
362 Extremely high code points were never specified in any standard, and require an
363 extension to UTF-8 to express, which Perl does. It is likely that programs
364 written in something other than Perl would not be able to read files that
365 contain these; nor would Perl understand files written by something that uses a
366 different extension. For these reasons, there is a separate set of flags that
367 can warn and/or disallow these extremely high code points, even if other
368 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
369 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
370 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
371 treat all above-Unicode code points, including these, as malformations. (Note
372 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
373 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
375 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
376 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
377 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
378 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because these
379 flags can apply to code points that actually do fit in 31 bits. This happens
380 on EBCDIC platforms, and sometimes when the L<overlong
381 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
382 describe the situation in all cases.
387 /* This is also a macro */
388 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
391 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
393 return uvchr_to_utf8_flags(d, uv, flags);
399 S_is_utf8_cp_above_31_bits(const U8 * const s,
401 const bool consider_overlongs)
403 /* Returns TRUE if the first code point represented by the Perl-extended-
404 * UTF-8-encoded string starting at 's', and looking no further than 'e -
405 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
407 * The function handles the case where the input bytes do not include all
408 * the ones necessary to represent a full character. That is, they may be
409 * the intial bytes of the representation of a code point, but possibly
410 * the final ones necessary for the complete representation may be beyond
413 * The function also can handle the case where the input is an overlong
414 * sequence. If 'consider_overlongs' is 0, the function assumes the
415 * input is not overlong, without checking, and will return based on that
416 * assumption. If this parameter is 1, the function will go to the trouble
417 * of figuring out if it actually evaluates to above or below 31 bits.
419 * The sequence is otherwise assumed to be well-formed, without checking.
422 const STRLEN len = e - s;
425 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
427 assert(! UTF8_IS_INVARIANT(*s) && e > s);
431 PERL_UNUSED_ARG(consider_overlongs);
433 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
434 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
435 * also be the start byte for a 31-bit code point; we need at least 2
436 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
437 * the start byte for an overlong sequence, but for 30-bit or smaller code
438 * points, so we don't have to worry about overlongs on EBCDIC.) */
449 /* On ASCII, FE and FF are the only start bytes that can evaluate to
450 * needing more than 31 bits. */
451 if (LIKELY(*s < 0xFE)) {
455 /* What we have left are FE and FF. Both of these require more than 31
456 * bits unless they are for overlongs. */
457 if (! consider_overlongs) {
461 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
462 * above 31 bits. But we need more than one byte to discern this, so if
463 * passed just the start byte, it could be an overlong evaluating to
469 /* Having excluded len==1, and knowing that FE and FF are both valid start
470 * bytes, we can call the function below to see if the sequence is
471 * overlong. (We don't need the full generality of the called function,
472 * but for these huge code points, speed shouldn't be a consideration, and
473 * the compiler does have enough information, since it's static to this
474 * file, to optimize to just the needed parts.) */
475 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
477 /* If it isn't overlong, more than 31 bits are required. */
478 if (is_overlong == 0) {
482 /* If it is indeterminate if it is overlong, return that */
483 if (is_overlong < 0) {
487 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
488 * the max it can be is 2**31 - 1 */
495 /* Here, ASCII and EBCDIC rejoin:
496 * On ASCII: We have an overlong sequence starting with FF
497 * On EBCDIC: We have a sequence starting with FE. */
499 { /* For C89, use a block so the declaration can be close to its use */
503 /* U+7FFFFFFF (2 ** 31 - 1)
504 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
505 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
506 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
507 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
508 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
509 * U+80000000 (2 ** 31):
510 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
511 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
512 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
513 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
515 * and since we know that *s = \xfe, any continuation sequcence
516 * following it that is gt the below is above 31 bits
517 [0] [1] [2] [3] [4] [5] [6] */
518 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
522 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
523 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
524 * FF overlong for U+80000000 (2 ** 31):
525 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
526 * and since we know that *s = \xff, any continuation sequcence
527 * following it that is gt the below is above 30 bits
528 [0] [1] [2] [3] [4] [5] [6] */
529 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
533 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
534 const STRLEN cmp_len = MIN(conts_len, len - 1);
536 /* Now compare the continuation bytes in s with the ones we have
537 * compiled in that are for the largest 30 bit code point. If we have
538 * enough bytes available to determine the answer, or the bytes we do
539 * have differ from them, we can compare the two to get a definitive
540 * answer (Note that in UTF-EBCDIC, the two lowest possible
541 * continuation bytes are \x41 and \x42.) */
542 if (cmp_len >= conts_len || memNE(s + 1,
543 conts_for_highest_30_bit,
546 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
549 /* Here, all the bytes we have are the same as the highest 30-bit code
550 * point, but we are missing so many bytes that we can't make the
558 PERL_STATIC_INLINE int
559 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
561 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
562 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
563 * it isn't, and -1 if there isn't enough information to tell. This last
564 * return value can happen if the sequence is incomplete, missing some
565 * trailing bytes that would form a complete character. If there are
566 * enough bytes to make a definitive decision, this function does so.
567 * Usually 2 bytes sufficient.
569 * Overlongs can occur whenever the number of continuation bytes changes.
570 * That means whenever the number of leading 1 bits in a start byte
571 * increases from the next lower start byte. That happens for start bytes
572 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
573 * start bytes have already been excluded, so don't need to be tested here;
574 * ASCII platforms: C0, C1
575 * EBCDIC platforms C0, C1, C2, C3, C4, E0
578 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
579 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
581 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
582 assert(len > 1 && UTF8_IS_START(*s));
584 /* Each platform has overlongs after the start bytes given above (expressed
585 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
586 * the logic is the same, except the E0 overlong has already been excluded
587 * on EBCDIC platforms. The values below were found by manually
588 * inspecting the UTF-8 patterns. See the tables in utf8.h and
592 # define F0_ABOVE_OVERLONG 0xB0
593 # define F8_ABOVE_OVERLONG 0xA8
594 # define FC_ABOVE_OVERLONG 0xA4
595 # define FE_ABOVE_OVERLONG 0xA2
596 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
600 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
604 # define F0_ABOVE_OVERLONG 0x90
605 # define F8_ABOVE_OVERLONG 0x88
606 # define FC_ABOVE_OVERLONG 0x84
607 # define FE_ABOVE_OVERLONG 0x82
608 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
612 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
613 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
614 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
615 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
620 /* Check for the FF overlong */
621 return isFF_OVERLONG(s, len);
624 PERL_STATIC_INLINE int
625 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
627 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
628 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
629 * it isn't, and -1 if there isn't enough information to tell. This last
630 * return value can happen if the sequence is incomplete, missing some
631 * trailing bytes that would form a complete character. If there are
632 * enough bytes to make a definitive decision, this function does so. */
634 PERL_ARGS_ASSERT_ISFF_OVERLONG;
636 /* To be an FF overlong, all the available bytes must match */
637 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
638 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
643 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
644 * be there; what comes after them doesn't matter. See tables in utf8.h,
646 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
650 /* The missing bytes could cause the result to go one way or the other, so
651 * the result is indeterminate */
655 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
656 # ifdef EBCDIC /* Actually is I8 */
657 # define HIGHEST_REPRESENTABLE_UTF8 \
658 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
660 # define HIGHEST_REPRESENTABLE_UTF8 \
661 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
665 PERL_STATIC_INLINE int
666 S_does_utf8_overflow(const U8 * const s,
668 const bool consider_overlongs)
670 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
671 * 'e' - 1 would overflow an IV on this platform; that is if it represents
672 * a code point larger than the highest representable code point. It
673 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
674 * enough information to tell. This last return value can happen if the
675 * sequence is incomplete, missing some trailing bytes that would form a
676 * complete character. If there are enough bytes to make a definitive
677 * decision, this function does so.
679 * If 'consider_overlongs' is TRUE, the function checks for the possibility
680 * that the sequence is an overlong that doesn't overflow. Otherwise, it
681 * assumes the sequence is not an overlong. This can give different
682 * results only on ASCII 32-bit platforms.
684 * (For ASCII platforms, we could use memcmp() because we don't have to
685 * convert each byte to I8, but it's very rare input indeed that would
686 * approach overflow, so the loop below will likely only get executed once.)
688 * 'e' - 1 must not be beyond a full character. */
691 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
692 assert(s <= e && s + UTF8SKIP(s) >= e);
694 #if ! defined(UV_IS_QUAD)
696 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
700 PERL_UNUSED_ARG(consider_overlongs);
703 const STRLEN len = e - s;
705 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
707 for (x = s; x < e; x++, y++) {
709 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
713 /* If this byte is larger than the corresponding highest UTF-8
714 * byte, the sequence overflow; otherwise the byte is less than,
715 * and so the sequence doesn't overflow */
716 return NATIVE_UTF8_TO_I8(*x) > *y;
720 /* Got to the end and all bytes are the same. If the input is a whole
721 * character, it doesn't overflow. And if it is a partial character,
722 * there's not enough information to tell */
723 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
736 /* This is the portions of the above function that deal with UV_MAX instead of
737 * IV_MAX. They are left here in case we want to combine them so that internal
738 * uses can have larger code points. The only logic difference is that the
739 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
743 /* Anything larger than this will overflow the word if it were converted into a UV */
744 #if defined(UV_IS_QUAD)
745 # ifdef EBCDIC /* Actually is I8 */
746 # define HIGHEST_REPRESENTABLE_UTF8 \
747 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
749 # define HIGHEST_REPRESENTABLE_UTF8 \
750 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
754 # define HIGHEST_REPRESENTABLE_UTF8 \
755 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
757 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
761 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
763 /* On 32 bit ASCII machines, many overlongs that start with FF don't
765 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
767 /* To be such an overlong, the first bytes of 's' must match
768 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
769 * don't have any additional bytes available, the sequence, when
770 * completed might or might not fit in 32 bits. But if we have that
771 * next byte, we can tell for sure. If it is <= 0x83, then it does
773 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
777 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
780 /* Starting with the #else, the rest of the function is identical except
781 * 1. we need to move the 'len' declaration to be global to the function
782 * 2. the endif move to just after the UNUSED_ARG.
783 * An empty endif is given just below to satisfy the preprocessor
789 #undef F0_ABOVE_OVERLONG
790 #undef F8_ABOVE_OVERLONG
791 #undef FC_ABOVE_OVERLONG
792 #undef FE_ABOVE_OVERLONG
793 #undef FF_OVERLONG_PREFIX
796 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
801 /* A helper function that should not be called directly.
803 * This function returns non-zero if the string beginning at 's' and
804 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
805 * code point; otherwise it returns 0. The examination stops after the
806 * first code point in 's' is validated, not looking at the rest of the
807 * input. If 'e' is such that there are not enough bytes to represent a
808 * complete code point, this function will return non-zero anyway, if the
809 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
810 * excluded by 'flags'.
812 * A non-zero return gives the number of bytes required to represent the
813 * code point. Be aware that if the input is for a partial character, the
814 * return will be larger than 'e - s'.
816 * This function assumes that the code point represented is UTF-8 variant.
817 * The caller should have excluded the possibility of it being invariant
818 * before calling this function.
820 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
821 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
822 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
823 * disallowed by the flags. If the input is only for a partial character,
824 * the function will return non-zero if there is any sequence of
825 * well-formed UTF-8 that, when appended to the input sequence, could
826 * result in an allowed code point; otherwise it returns 0. Non characters
827 * cannot be determined based on partial character input. But many of the
828 * other excluded types can be determined with just the first one or two
833 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
835 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
836 |UTF8_DISALLOW_PERL_EXTENDED)));
837 assert(! UTF8_IS_INVARIANT(*s));
839 /* A variant char must begin with a start byte */
840 if (UNLIKELY(! UTF8_IS_START(*s))) {
844 /* Examine a maximum of a single whole code point */
845 if (e - s > UTF8SKIP(s)) {
851 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
852 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
854 /* Here, we are disallowing some set of largish code points, and the
855 * first byte indicates the sequence is for a code point that could be
856 * in the excluded set. We generally don't have to look beyond this or
857 * the second byte to see if the sequence is actually for one of the
858 * excluded classes. The code below is derived from this table:
860 * UTF-8 UTF-EBCDIC I8
861 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
862 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
863 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
865 * Keep in mind that legal continuation bytes range between \x80..\xBF
866 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
867 * continuation bytes. Hence, we don't have to test the upper edge
868 * because if any of those is encountered, the sequence is malformed,
869 * and would fail elsewhere in this function.
871 * The code here likewise assumes that there aren't other
872 * malformations; again the function should fail elsewhere because of
873 * these. For example, an overlong beginning with FC doesn't actually
874 * have to be a super; it could actually represent a small code point,
875 * even U+0000. But, since overlongs (and other malformations) are
876 * illegal, the function should return FALSE in either case.
879 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
880 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
881 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
883 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
885 && ((s1) & 0xFE ) == 0xB6)
886 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
888 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
889 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
890 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
891 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
894 if ( (flags & UTF8_DISALLOW_SUPER)
895 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
897 return 0; /* Above Unicode */
900 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
901 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
907 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
909 if ( (flags & UTF8_DISALLOW_SUPER)
910 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
912 return 0; /* Above Unicode */
915 if ( (flags & UTF8_DISALLOW_SURROGATE)
916 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
918 return 0; /* Surrogate */
921 if ( (flags & UTF8_DISALLOW_NONCHAR)
922 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
924 return 0; /* Noncharacter code point */
929 /* Make sure that all that follows are continuation bytes */
930 for (x = s + 1; x < e; x++) {
931 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
936 /* Here is syntactically valid. Next, make sure this isn't the start of an
938 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
942 /* And finally, that the code point represented fits in a word on this
944 if (0 < does_utf8_overflow(s, e,
945 0 /* Don't consider overlongs */
955 Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
957 /* Returns a mortalized C string that is a displayable copy of the 'len'
958 * bytes starting at 's'. 'format' gives how to display each byte.
959 * Currently, there are only two formats, so it is currently a bool:
961 * 1 ab (that is a space between two hex digit bytes)
964 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
966 const U8 * const e = s + len;
970 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
972 Newx(output, output_len, char);
977 const unsigned high_nibble = (*s & 0xF0) >> 4;
978 const unsigned low_nibble = (*s & 0x0F);
988 if (high_nibble < 10) {
989 *d++ = high_nibble + '0';
992 *d++ = high_nibble - 10 + 'a';
995 if (low_nibble < 10) {
996 *d++ = low_nibble + '0';
999 *d++ = low_nibble - 10 + 'a';
1007 PERL_STATIC_INLINE char *
1008 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1010 /* How many bytes to print */
1013 /* Which one is the non-continuation */
1014 const STRLEN non_cont_byte_pos,
1016 /* How many bytes should there be? */
1017 const STRLEN expect_len)
1019 /* Return the malformation warning text for an unexpected continuation
1022 const char * const where = (non_cont_byte_pos == 1)
1024 : Perl_form(aTHX_ "%d bytes",
1025 (int) non_cont_byte_pos);
1027 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1029 /* We don't need to pass this parameter, but since it has already been
1030 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1031 assert(expect_len == UTF8SKIP(s));
1033 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1034 " %s after start byte 0x%02x; need %d bytes, got %d)",
1036 _byte_dump_string(s, print_len, 0),
1037 *(s + non_cont_byte_pos),
1041 (int) non_cont_byte_pos);
1046 =for apidoc utf8n_to_uvchr
1048 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1049 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1051 Bottom level UTF-8 decode routine.
1052 Returns the native code point value of the first character in the string C<s>,
1053 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1054 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1055 the length, in bytes, of that character.
1057 The value of C<flags> determines the behavior when C<s> does not point to a
1058 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1059 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1060 is the next possible position in C<s> that could begin a non-malformed
1061 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1062 is raised. Some UTF-8 input sequences may contain multiple malformations.
1063 This function tries to find every possible one in each call, so multiple
1064 warnings can be raised for the same sequence.
1066 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1067 individual types of malformations, such as the sequence being overlong (that
1068 is, when there is a shorter sequence that can express the same code point;
1069 overlong sequences are expressly forbidden in the UTF-8 standard due to
1070 potential security issues). Another malformation example is the first byte of
1071 a character not being a legal first byte. See F<utf8.h> for the list of such
1072 flags. Even if allowed, this function generally returns the Unicode
1073 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1074 F<utf8.h> to override this behavior for the overlong malformations, but don't
1075 do that except for very specialized purposes.
1077 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1078 flags) malformation is found. If this flag is set, the routine assumes that
1079 the caller will raise a warning, and this function will silently just set
1080 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1082 Note that this API requires disambiguation between successful decoding a C<NUL>
1083 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1084 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1085 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1086 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1087 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1089 Certain code points are considered problematic. These are Unicode surrogates,
1090 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1091 By default these are considered regular code points, but certain situations
1092 warrant special handling for them, which can be specified using the C<flags>
1093 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1094 three classes are treated as malformations and handled as such. The flags
1095 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1096 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1097 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1098 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1099 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1101 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1102 The difference between traditional strictness and C9 strictness is that the
1103 latter does not forbid non-character code points. (They are still discouraged,
1104 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1106 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1107 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1108 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1109 raised for their respective categories, but otherwise the code points are
1110 considered valid (not malformations). To get a category to both be treated as
1111 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1112 (But note that warnings are not raised if lexically disabled nor if
1113 C<UTF8_CHECK_ONLY> is also specified.)
1115 Extremely high code points were never specified in any standard, and require an
1116 extension to UTF-8 to express, which Perl does. It is likely that programs
1117 written in something other than Perl would not be able to read files that
1118 contain these; nor would Perl understand files written by something that uses a
1119 different extension. For these reasons, there is a separate set of flags that
1120 can warn and/or disallow these extremely high code points, even if other
1121 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1122 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1123 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1124 above-Unicode code points, including these, as malformations.
1125 (Note that the Unicode standard considers anything above 0x10FFFF to be
1126 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1129 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1130 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1131 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1132 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1133 can apply to code points that actually do fit in 31 bits. This happens on
1134 EBCDIC platforms, and sometimes when the L<overlong
1135 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1136 describe the situation in all cases.
1139 All other code points corresponding to Unicode characters, including private
1140 use and those yet to be assigned, are never considered malformed and never
1145 Also implemented as a macro in utf8.h
1149 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1154 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1156 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1161 =for apidoc utf8n_to_uvchr_error
1163 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1164 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1166 This function is for code that needs to know what the precise malformation(s)
1167 are when an error is found.
1169 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1170 all the others, C<errors>. If this parameter is 0, this function behaves
1171 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1172 to a C<U32> variable, which this function sets to indicate any errors found.
1173 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1174 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1175 of these bits will be set if a malformation is found, even if the input
1176 C<flags> parameter indicates that the given malformation is allowed; those
1177 exceptions are noted:
1181 =item C<UTF8_GOT_PERL_EXTENDED>
1183 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1184 set only if the input C<flags> parameter contains either the
1185 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1187 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1188 and so some extension must be used to express them. Perl uses a natural
1189 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1190 extension to represent even higher ones, so that any code point that fits in a
1191 64-bit word can be represented. Text using these extensions is not likely to
1192 be portable to non-Perl code. We lump both of these extensions together and
1193 refer to them as Perl extended UTF-8. There exist other extensions that people
1194 have invented, incompatible with Perl's.
1196 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1197 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1198 than on ASCII. Prior to that, code points 2**31 and higher were simply
1199 unrepresentable, and a different, incompatible method was used to represent
1200 code points between 2**30 and 2**31 - 1.
1202 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1203 Perl extended UTF-8 is used.
1205 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1206 may use for backward compatibility. That name is misleading, as this flag may
1207 be set when the code point actually does fit in 31 bits. This happens on
1208 EBCDIC platforms, and sometimes when the L<overlong
1209 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1210 describes the situation in all cases.
1212 =item C<UTF8_GOT_CONTINUATION>
1214 The input sequence was malformed in that the first byte was a a UTF-8
1217 =item C<UTF8_GOT_EMPTY>
1219 The input C<curlen> parameter was 0.
1221 =item C<UTF8_GOT_LONG>
1223 The input sequence was malformed in that there is some other sequence that
1224 evaluates to the same code point, but that sequence is shorter than this one.
1226 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1227 it was discovered that this created security issues.
1229 =item C<UTF8_GOT_NONCHAR>
1231 The code point represented by the input UTF-8 sequence is for a Unicode
1232 non-character code point.
1233 This bit is set only if the input C<flags> parameter contains either the
1234 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1236 =item C<UTF8_GOT_NON_CONTINUATION>
1238 The input sequence was malformed in that a non-continuation type byte was found
1239 in a position where only a continuation type one should be.
1241 =item C<UTF8_GOT_OVERFLOW>
1243 The input sequence was malformed in that it is for a code point that is not
1244 representable in the number of bits available in an IV on the current platform.
1246 =item C<UTF8_GOT_SHORT>
1248 The input sequence was malformed in that C<curlen> is smaller than required for
1249 a complete sequence. In other words, the input is for a partial character
1252 =item C<UTF8_GOT_SUPER>
1254 The input sequence was malformed in that it is for a non-Unicode code point;
1255 that is, one above the legal Unicode maximum.
1256 This bit is set only if the input C<flags> parameter contains either the
1257 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1259 =item C<UTF8_GOT_SURROGATE>
1261 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1263 This bit is set only if the input C<flags> parameter contains either the
1264 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1268 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1269 flag to suppress any warnings, and then examine the C<*errors> return.
1275 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1281 const U8 * const s0 = s;
1282 U8 * send = NULL; /* (initialized to silence compilers' wrong
1284 U32 possible_problems = 0; /* A bit is set here for each potential problem
1285 found as we go along */
1287 STRLEN expectlen = 0; /* How long should this sequence be?
1288 (initialized to silence compilers' wrong
1290 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1291 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1292 this gets set and discarded */
1294 /* The below are used only if there is both an overlong malformation and a
1295 * too short one. Otherwise the first two are set to 's0' and 'send', and
1296 * the third not used at all */
1297 U8 * adjusted_s0 = (U8 *) s0;
1298 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1299 routine; see [perl #130921] */
1300 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1302 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1308 errors = &discard_errors;
1311 /* The order of malformation tests here is important. We should consume as
1312 * few bytes as possible in order to not skip any valid character. This is
1313 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1314 * http://unicode.org/reports/tr36 for more discussion as to why. For
1315 * example, once we've done a UTF8SKIP, we can tell the expected number of
1316 * bytes, and could fail right off the bat if the input parameters indicate
1317 * that there are too few available. But it could be that just that first
1318 * byte is garbled, and the intended character occupies fewer bytes. If we
1319 * blindly assumed that the first byte is correct, and skipped based on
1320 * that number, we could skip over a valid input character. So instead, we
1321 * always examine the sequence byte-by-byte.
1323 * We also should not consume too few bytes, otherwise someone could inject
1324 * things. For example, an input could be deliberately designed to
1325 * overflow, and if this code bailed out immediately upon discovering that,
1326 * returning to the caller C<*retlen> pointing to the very next byte (one
1327 * which is actually part of of the overflowing sequence), that could look
1328 * legitimate to the caller, which could discard the initial partial
1329 * sequence and process the rest, inappropriately.
1331 * Some possible input sequences are malformed in more than one way. This
1332 * function goes to lengths to try to find all of them. This is necessary
1333 * for correctness, as the inputs may allow one malformation but not
1334 * another, and if we abandon searching for others after finding the
1335 * allowed one, we could allow in something that shouldn't have been.
1338 if (UNLIKELY(curlen == 0)) {
1339 possible_problems |= UTF8_GOT_EMPTY;
1341 uv = UNICODE_REPLACEMENT;
1342 goto ready_to_handle_errors;
1345 expectlen = UTF8SKIP(s);
1347 /* A well-formed UTF-8 character, as the vast majority of calls to this
1348 * function will be for, has this expected length. For efficiency, set
1349 * things up here to return it. It will be overriden only in those rare
1350 * cases where a malformation is found */
1352 *retlen = expectlen;
1355 /* An invariant is trivially well-formed */
1356 if (UTF8_IS_INVARIANT(uv)) {
1360 /* A continuation character can't start a valid sequence */
1361 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1362 possible_problems |= UTF8_GOT_CONTINUATION;
1364 uv = UNICODE_REPLACEMENT;
1365 goto ready_to_handle_errors;
1368 /* Here is not a continuation byte, nor an invariant. The only thing left
1369 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1370 * because it excludes start bytes like \xC0 that always lead to
1373 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1374 * that indicate the number of bytes in the character's whole UTF-8
1375 * sequence, leaving just the bits that are part of the value. */
1376 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1378 /* Setup the loop end point, making sure to not look past the end of the
1379 * input string, and flag it as too short if the size isn't big enough. */
1381 if (UNLIKELY(curlen < expectlen)) {
1382 possible_problems |= UTF8_GOT_SHORT;
1390 /* Now, loop through the remaining bytes in the character's sequence,
1391 * accumulating each into the working value as we go. */
1392 for (s = s0 + 1; s < send; s++) {
1393 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1394 uv = UTF8_ACCUMULATE(uv, *s);
1398 /* Here, found a non-continuation before processing all expected bytes.
1399 * This byte indicates the beginning of a new character, so quit, even
1400 * if allowing this malformation. */
1401 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1403 } /* End of loop through the character's bytes */
1405 /* Save how many bytes were actually in the character */
1408 /* Note that there are two types of too-short malformation. One is when
1409 * there is actual wrong data before the normal termination of the
1410 * sequence. The other is that the sequence wasn't complete before the end
1411 * of the data we are allowed to look at, based on the input 'curlen'.
1412 * This means that we were passed data for a partial character, but it is
1413 * valid as far as we saw. The other is definitely invalid. This
1414 * distinction could be important to a caller, so the two types are kept
1417 * A convenience macro that matches either of the too-short conditions. */
1418 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1420 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1422 uv = UNICODE_REPLACEMENT;
1425 /* Check for overflow. The algorithm requires us to not look past the end
1426 * of the current character, even if partial, so the upper limit is 's' */
1427 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1428 1 /* Do consider overlongs */
1431 possible_problems |= UTF8_GOT_OVERFLOW;
1432 uv = UNICODE_REPLACEMENT;
1435 /* Check for overlong. If no problems so far, 'uv' is the correct code
1436 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1437 * we must look at the UTF-8 byte sequence itself to see if it is for an
1439 if ( ( LIKELY(! possible_problems)
1440 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1441 || ( UNLIKELY(possible_problems)
1442 && ( UNLIKELY(! UTF8_IS_START(*s0))
1444 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1447 possible_problems |= UTF8_GOT_LONG;
1449 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1451 /* The calculation in the 'true' branch of this 'if'
1452 * below won't work if overflows, and isn't needed
1453 * anyway. Further below we handle all overflow
1455 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1457 UV min_uv = uv_so_far;
1460 /* Here, the input is both overlong and is missing some trailing
1461 * bytes. There is no single code point it could be for, but there
1462 * may be enough information present to determine if what we have
1463 * so far is for an unallowed code point, such as for a surrogate.
1464 * The code further below has the intelligence to determine this,
1465 * but just for non-overlong UTF-8 sequences. What we do here is
1466 * calculate the smallest code point the input could represent if
1467 * there were no too short malformation. Then we compute and save
1468 * the UTF-8 for that, which is what the code below looks at
1469 * instead of the raw input. It turns out that the smallest such
1470 * code point is all we need. */
1471 for (i = curlen; i < expectlen; i++) {
1472 min_uv = UTF8_ACCUMULATE(min_uv,
1473 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1476 adjusted_s0 = temp_char_buf;
1477 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1481 /* Here, we have found all the possible problems, except for when the input
1482 * is for a problematic code point not allowed by the input parameters. */
1484 /* uv is valid for overlongs */
1485 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1487 /* isn't problematic if < this */
1488 && uv >= UNICODE_SURROGATE_FIRST)
1489 || ( UNLIKELY(possible_problems)
1491 /* if overflow, we know without looking further
1492 * precisely which of the problematic types it is,
1493 * and we deal with those in the overflow handling
1495 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1496 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1497 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1498 && ((flags & ( UTF8_DISALLOW_NONCHAR
1499 |UTF8_DISALLOW_SURROGATE
1500 |UTF8_DISALLOW_SUPER
1501 |UTF8_DISALLOW_PERL_EXTENDED
1503 |UTF8_WARN_SURROGATE
1505 |UTF8_WARN_PERL_EXTENDED))))
1507 /* If there were no malformations, or the only malformation is an
1508 * overlong, 'uv' is valid */
1509 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1510 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1511 possible_problems |= UTF8_GOT_SURROGATE;
1513 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1514 possible_problems |= UTF8_GOT_SUPER;
1516 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1517 possible_problems |= UTF8_GOT_NONCHAR;
1520 else { /* Otherwise, need to look at the source UTF-8, possibly
1521 adjusted to be non-overlong */
1523 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1524 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1526 possible_problems |= UTF8_GOT_SUPER;
1528 else if (curlen > 1) {
1529 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1530 NATIVE_UTF8_TO_I8(*adjusted_s0),
1531 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1533 possible_problems |= UTF8_GOT_SUPER;
1535 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1536 NATIVE_UTF8_TO_I8(*adjusted_s0),
1537 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1539 possible_problems |= UTF8_GOT_SURROGATE;
1543 /* We need a complete well-formed UTF-8 character to discern
1544 * non-characters, so can't look for them here */
1548 ready_to_handle_errors:
1551 * curlen contains the number of bytes in the sequence that
1552 * this call should advance the input by.
1553 * avail_len gives the available number of bytes passed in, but
1554 * only if this is less than the expected number of
1555 * bytes, based on the code point's start byte.
1556 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1557 * is set in it for each potential problem found.
1558 * uv contains the code point the input sequence
1559 * represents; or if there is a problem that prevents
1560 * a well-defined value from being computed, it is
1561 * some subsitute value, typically the REPLACEMENT
1563 * s0 points to the first byte of the character
1564 * s points to just after were we left off processing
1566 * send points to just after where that character should
1567 * end, based on how many bytes the start byte tells
1568 * us should be in it, but no further than s0 +
1572 if (UNLIKELY(possible_problems)) {
1573 bool disallowed = FALSE;
1574 const U32 orig_problems = possible_problems;
1576 while (possible_problems) { /* Handle each possible problem */
1578 char * message = NULL;
1580 /* Each 'if' clause handles one problem. They are ordered so that
1581 * the first ones' messages will be displayed before the later
1582 * ones; this is kinda in decreasing severity order. But the
1583 * overlong must come last, as it changes 'uv' looked at by the
1585 if (possible_problems & UTF8_GOT_OVERFLOW) {
1587 /* Overflow means also got a super and are using Perl's
1588 * extended UTF-8, but we handle all three cases here */
1590 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1591 *errors |= UTF8_GOT_OVERFLOW;
1593 /* But the API says we flag all errors found */
1594 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1595 *errors |= UTF8_GOT_SUPER;
1598 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1600 *errors |= UTF8_GOT_PERL_EXTENDED;
1603 /* Disallow if any of the three categories say to */
1604 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1605 || (flags & ( UTF8_DISALLOW_SUPER
1606 |UTF8_DISALLOW_PERL_EXTENDED)))
1611 /* Likewise, warn if any say to */
1612 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1613 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1616 /* The warnings code explicitly says it doesn't handle the
1617 * case of packWARN2 and two categories which have
1618 * parent-child relationship. Even if it works now to
1619 * raise the warning if either is enabled, it wouldn't
1620 * necessarily do so in the future. We output (only) the
1621 * most dire warning */
1622 if (! (flags & UTF8_CHECK_ONLY)) {
1623 if (ckWARN_d(WARN_UTF8)) {
1624 pack_warn = packWARN(WARN_UTF8);
1626 else if (ckWARN_d(WARN_NON_UNICODE)) {
1627 pack_warn = packWARN(WARN_NON_UNICODE);
1630 message = Perl_form(aTHX_ "%s: %s (overflows)",
1632 _byte_dump_string(s0, curlen, 0));
1637 else if (possible_problems & UTF8_GOT_EMPTY) {
1638 possible_problems &= ~UTF8_GOT_EMPTY;
1639 *errors |= UTF8_GOT_EMPTY;
1641 if (! (flags & UTF8_ALLOW_EMPTY)) {
1643 /* This so-called malformation is now treated as a bug in
1644 * the caller. If you have nothing to decode, skip calling
1649 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1650 pack_warn = packWARN(WARN_UTF8);
1651 message = Perl_form(aTHX_ "%s (empty string)",
1656 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1657 possible_problems &= ~UTF8_GOT_CONTINUATION;
1658 *errors |= UTF8_GOT_CONTINUATION;
1660 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1662 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1663 pack_warn = packWARN(WARN_UTF8);
1664 message = Perl_form(aTHX_
1665 "%s: %s (unexpected continuation byte 0x%02x,"
1666 " with no preceding start byte)",
1668 _byte_dump_string(s0, 1, 0), *s0);
1672 else if (possible_problems & UTF8_GOT_SHORT) {
1673 possible_problems &= ~UTF8_GOT_SHORT;
1674 *errors |= UTF8_GOT_SHORT;
1676 if (! (flags & UTF8_ALLOW_SHORT)) {
1678 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1679 pack_warn = packWARN(WARN_UTF8);
1680 message = Perl_form(aTHX_
1681 "%s: %s (too short; %d byte%s available, need %d)",
1683 _byte_dump_string(s0, send - s0, 0),
1685 avail_len == 1 ? "" : "s",
1691 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1692 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1693 *errors |= UTF8_GOT_NON_CONTINUATION;
1695 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1697 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1699 /* If we don't know for sure that the input length is
1700 * valid, avoid as much as possible reading past the
1701 * end of the buffer */
1702 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1705 pack_warn = packWARN(WARN_UTF8);
1706 message = Perl_form(aTHX_ "%s",
1707 unexpected_non_continuation_text(s0,
1714 else if (possible_problems & UTF8_GOT_SURROGATE) {
1715 possible_problems &= ~UTF8_GOT_SURROGATE;
1717 if (flags & UTF8_WARN_SURROGATE) {
1718 *errors |= UTF8_GOT_SURROGATE;
1720 if ( ! (flags & UTF8_CHECK_ONLY)
1721 && ckWARN_d(WARN_SURROGATE))
1723 pack_warn = packWARN(WARN_SURROGATE);
1725 /* These are the only errors that can occur with a
1726 * surrogate when the 'uv' isn't valid */
1727 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1728 message = Perl_form(aTHX_
1729 "UTF-16 surrogate (any UTF-8 sequence that"
1730 " starts with \"%s\" is for a surrogate)",
1731 _byte_dump_string(s0, curlen, 0));
1734 message = Perl_form(aTHX_ surrogate_cp_format, uv);
1739 if (flags & UTF8_DISALLOW_SURROGATE) {
1741 *errors |= UTF8_GOT_SURROGATE;
1744 else if (possible_problems & UTF8_GOT_SUPER) {
1745 possible_problems &= ~UTF8_GOT_SUPER;
1747 if (flags & UTF8_WARN_SUPER) {
1748 *errors |= UTF8_GOT_SUPER;
1750 if ( ! (flags & UTF8_CHECK_ONLY)
1751 && ckWARN_d(WARN_NON_UNICODE))
1753 pack_warn = packWARN(WARN_NON_UNICODE);
1755 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1756 message = Perl_form(aTHX_
1757 "Any UTF-8 sequence that starts with"
1758 " \"%s\" is for a non-Unicode code point,"
1759 " may not be portable",
1760 _byte_dump_string(s0, curlen, 0));
1763 message = Perl_form(aTHX_ super_cp_format, uv);
1768 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1769 * and before possibly bailing out, so that the more dire
1770 * warning will override the regular one. */
1771 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
1772 if ( ! (flags & UTF8_CHECK_ONLY)
1773 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
1774 && ckWARN_d(WARN_NON_UNICODE))
1776 pack_warn = packWARN(WARN_NON_UNICODE);
1778 /* If it is an overlong that evaluates to a code point
1779 * that doesn't have to use the Perl extended UTF-8, it
1780 * still used it, and so we output a message that
1781 * doesn't refer to the code point. The same is true
1782 * if there was a SHORT malformation where the code
1783 * point is not valid. In that case, 'uv' will have
1784 * been set to the REPLACEMENT CHAR, and the message
1785 * below without the code point in it will be selected
1787 if (UNICODE_IS_PERL_EXTENDED(uv)) {
1788 message = Perl_form(aTHX_
1789 perl_extended_cp_format, uv);
1792 message = Perl_form(aTHX_
1793 "Any UTF-8 sequence that starts with"
1794 " \"%s\" is a Perl extension, and"
1795 " so is not portable",
1796 _byte_dump_string(s0, curlen, 0));
1800 if (flags & ( UTF8_WARN_PERL_EXTENDED
1801 |UTF8_DISALLOW_PERL_EXTENDED))
1803 *errors |= UTF8_GOT_PERL_EXTENDED;
1805 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
1811 if (flags & UTF8_DISALLOW_SUPER) {
1812 *errors |= UTF8_GOT_SUPER;
1816 else if (possible_problems & UTF8_GOT_NONCHAR) {
1817 possible_problems &= ~UTF8_GOT_NONCHAR;
1819 if (flags & UTF8_WARN_NONCHAR) {
1820 *errors |= UTF8_GOT_NONCHAR;
1822 if ( ! (flags & UTF8_CHECK_ONLY)
1823 && ckWARN_d(WARN_NONCHAR))
1825 /* The code above should have guaranteed that we don't
1826 * get here with errors other than overlong */
1827 assert (! (orig_problems
1828 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1830 pack_warn = packWARN(WARN_NONCHAR);
1831 message = Perl_form(aTHX_ nonchar_cp_format, uv);
1835 if (flags & UTF8_DISALLOW_NONCHAR) {
1837 *errors |= UTF8_GOT_NONCHAR;
1840 else if (possible_problems & UTF8_GOT_LONG) {
1841 possible_problems &= ~UTF8_GOT_LONG;
1842 *errors |= UTF8_GOT_LONG;
1844 if (flags & UTF8_ALLOW_LONG) {
1846 /* We don't allow the actual overlong value, unless the
1847 * special extra bit is also set */
1848 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1849 & ~UTF8_ALLOW_LONG)))
1851 uv = UNICODE_REPLACEMENT;
1857 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1858 pack_warn = packWARN(WARN_UTF8);
1860 /* These error types cause 'uv' to be something that
1861 * isn't what was intended, so can't use it in the
1862 * message. The other error types either can't
1863 * generate an overlong, or else the 'uv' is valid */
1865 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1867 message = Perl_form(aTHX_
1868 "%s: %s (any UTF-8 sequence that starts"
1869 " with \"%s\" is overlong which can and"
1870 " should be represented with a"
1871 " different, shorter sequence)",
1873 _byte_dump_string(s0, send - s0, 0),
1874 _byte_dump_string(s0, curlen, 0));
1877 U8 tmpbuf[UTF8_MAXBYTES+1];
1878 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1880 const char * preface = (uv <= PERL_UNICODE_MAX)
1883 message = Perl_form(aTHX_
1884 "%s: %s (overlong; instead use %s to represent"
1887 _byte_dump_string(s0, send - s0, 0),
1888 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1890 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1891 small code points */
1896 } /* End of looking through the possible flags */
1898 /* Display the message (if any) for the problem being handled in
1899 * this iteration of the loop */
1902 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1905 Perl_warner(aTHX_ pack_warn, "%s", message);
1907 } /* End of 'while (possible_problems)' */
1909 /* Since there was a possible problem, the returned length may need to
1910 * be changed from the one stored at the beginning of this function.
1911 * Instead of trying to figure out if that's needed, just do it. */
1917 if (flags & UTF8_CHECK_ONLY && retlen) {
1918 *retlen = ((STRLEN) -1);
1924 return UNI_TO_NATIVE(uv);
1928 =for apidoc utf8_to_uvchr_buf
1930 Returns the native code point of the first character in the string C<s> which
1931 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1932 C<*retlen> will be set to the length, in bytes, of that character.
1934 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1935 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1936 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1937 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1938 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1939 the next possible position in C<s> that could begin a non-malformed character.
1940 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1945 Also implemented as a macro in utf8.h
1951 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1953 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1957 return utf8n_to_uvchr(s, send - s, retlen,
1958 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1961 /* This is marked as deprecated
1963 =for apidoc utf8_to_uvuni_buf
1965 Only in very rare circumstances should code need to be dealing in Unicode
1966 (as opposed to native) code points. In those few cases, use
1967 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1969 Returns the Unicode (not-native) code point of the first character in the
1971 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1972 C<retlen> will be set to the length, in bytes, of that character.
1974 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1975 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1976 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1977 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1978 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1979 next possible position in C<s> that could begin a non-malformed character.
1980 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1986 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1988 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1992 /* Call the low level routine, asking for checks */
1993 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1997 =for apidoc utf8_length
1999 Return the length of the UTF-8 char encoded string C<s> in characters.
2000 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
2001 up past C<e>, croaks.
2007 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2011 PERL_ARGS_ASSERT_UTF8_LENGTH;
2013 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2014 * the bitops (especially ~) can create illegal UTF-8.
2015 * In other words: in Perl UTF-8 is not just for Unicode. */
2018 goto warn_and_return;
2028 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2029 "%s in %s", unees, OP_DESC(PL_op));
2031 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2038 =for apidoc bytes_cmp_utf8
2040 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2041 sequence of characters (stored as UTF-8)
2042 in C<u>, C<ulen>. Returns 0 if they are
2043 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2044 if the first string is greater than the second string.
2046 -1 or +1 is returned if the shorter string was identical to the start of the
2047 longer string. -2 or +2 is returned if
2048 there was a difference between characters
2055 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2057 const U8 *const bend = b + blen;
2058 const U8 *const uend = u + ulen;
2060 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2062 while (b < bend && u < uend) {
2064 if (!UTF8_IS_INVARIANT(c)) {
2065 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2068 if (UTF8_IS_CONTINUATION(c1)) {
2069 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2071 /* diag_listed_as: Malformed UTF-8 character%s */
2072 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2074 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2075 PL_op ? " in " : "",
2076 PL_op ? OP_DESC(PL_op) : "");
2081 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2082 "%s in %s", unees, OP_DESC(PL_op));
2084 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2085 return -2; /* Really want to return undef :-) */
2092 return *b < c ? -2 : +2;
2097 if (b == bend && u == uend)
2100 return b < bend ? +1 : -1;
2104 =for apidoc utf8_to_bytes
2106 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2107 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2108 updates C<*lenp> to contain the new length.
2109 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2111 Upon successful return, the number of variants in the string can be computed by
2112 having saved the value of C<*lenp> before the call, and subtracting the
2113 after-call value of C<*lenp> from it.
2115 If you need a copy of the string, see L</bytes_from_utf8>.
2121 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2125 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2126 PERL_UNUSED_CONTEXT;
2128 /* This is a no-op if no variants at all in the input */
2129 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2134 U8 * const save = s;
2135 U8 * const send = s + *lenp;
2138 /* Nothing before the first variant needs to be changed, so start the real
2142 if (! UTF8_IS_INVARIANT(*s)) {
2143 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2144 *lenp = ((STRLEN) -1);
2152 /* Is downgradable, so do it */
2153 d = s = first_variant;
2156 if (! UVCHR_IS_INVARIANT(c)) {
2157 /* Then it is two-byte encoded */
2158 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2171 =for apidoc bytes_from_utf8
2173 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2174 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2175 actually encoded in UTF-8.
2177 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2180 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2181 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2182 C<*lenp> are unchanged, and the return value is the original C<s>.
2184 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2185 newly created string containing a downgraded copy of C<s>, and whose length is
2186 returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
2188 Upon successful return, the number of variants in the string can be computed by
2189 having saved the value of C<*lenp> before the call, and subtracting the
2190 after-call value of C<*lenp> from it.
2194 There is a macro that avoids this function call, but this is retained for
2195 anyone who calls it with the Perl_ prefix */
2198 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2200 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2201 PERL_UNUSED_CONTEXT;
2203 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2207 No = here because currently externally undocumented
2208 for apidoc bytes_from_utf8_loc
2210 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2211 to store the location of the first character in C<"s"> that cannot be
2212 converted to non-UTF8.
2214 If that parameter is C<NULL>, this function behaves identically to
2217 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2218 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2220 Otherwise, the function returns a newly created C<NUL>-terminated string
2221 containing the non-UTF8 equivalent of the convertible first portion of
2222 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2223 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2224 and C<*first_non_downgradable> is set to C<NULL>.
2226 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2227 first character in the original string that wasn't converted. C<*is_utf8p> is
2228 unchanged. Note that the new string may have length 0.
2230 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2231 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2232 converts as many characters in it as possible stopping at the first one it
2233 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2234 set to point to that. The function returns the portion that could be converted
2235 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2236 not including the terminating C<NUL>. If the very first character in the
2237 original could not be converted, C<*lenp> will be 0, and the new string will
2238 contain just a single C<NUL>. If the entire input string was converted,
2239 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2241 Upon successful return, the number of variants in the converted portion of the
2242 string can be computed by having saved the value of C<*lenp> before the call,
2243 and subtracting the after-call value of C<*lenp> from it.
2251 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2254 const U8 *original = s;
2255 U8 *converted_start;
2256 const U8 *send = s + *lenp;
2258 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2261 if (first_unconverted) {
2262 *first_unconverted = NULL;
2265 return (U8 *) original;
2268 Newx(d, (*lenp) + 1, U8);
2270 converted_start = d;
2273 if (! UTF8_IS_INVARIANT(c)) {
2275 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2276 * have to stop now */
2277 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2278 if (first_unconverted) {
2279 *first_unconverted = s - 1;
2280 goto finish_and_return;
2283 Safefree(converted_start);
2284 return (U8 *) original;
2288 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2294 /* Here, converted the whole of the input */
2296 if (first_unconverted) {
2297 *first_unconverted = NULL;
2302 *lenp = d - converted_start;
2304 /* Trim unused space */
2305 Renew(converted_start, *lenp + 1, U8);
2307 return converted_start;
2311 =for apidoc bytes_to_utf8
2313 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2315 Returns a pointer to the newly-created string, and sets C<*lenp> to
2316 reflect the new length in bytes.
2318 Upon successful return, the number of variants in the string can be computed by
2319 having saved the value of C<*lenp> before the call, and subtracting it from the
2320 after-call value of C<*lenp>.
2322 A C<NUL> character will be written after the end of the string.
2324 If you want to convert to UTF-8 from encodings other than
2325 the native (Latin1 or EBCDIC),
2326 see L</sv_recode_to_utf8>().
2332 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2334 const U8 * const send = s + (*lenp);
2338 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2339 PERL_UNUSED_CONTEXT;
2341 Newx(d, (*lenp) * 2 + 1, U8);
2345 append_utf8_from_native_byte(*s, &d);
2354 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2356 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2357 * We optimize for native, for obvious reasons. */
2360 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2365 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2368 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2374 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2376 if (OFFUNI_IS_INVARIANT(uv)) {
2377 *d++ = LATIN1_TO_NATIVE((U8) uv);
2380 if (uv <= MAX_UTF8_TWO_BYTE) {
2381 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2382 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2385 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2386 #define LAST_HIGH_SURROGATE 0xDBFF
2387 #define FIRST_LOW_SURROGATE 0xDC00
2388 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2390 /* This assumes that most uses will be in the first Unicode plane, not
2391 * needing surrogates */
2392 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2393 && uv <= UNICODE_SURROGATE_LAST))
2395 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2396 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2399 UV low = (p[0] << 8) + p[1];
2400 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2401 || UNLIKELY(low > LAST_LOW_SURROGATE))
2403 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2406 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2407 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2411 d = uvoffuni_to_utf8_flags(d, uv, 0);
2414 *d++ = (U8)(( uv >> 12) | 0xe0);
2415 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2416 *d++ = (U8)(( uv & 0x3f) | 0x80);
2420 *d++ = (U8)(( uv >> 18) | 0xf0);
2421 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2422 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2423 *d++ = (U8)(( uv & 0x3f) | 0x80);
2428 *newlen = d - dstart;
2432 /* Note: this one is slightly destructive of the source. */
2435 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2438 U8* const send = s + bytelen;
2440 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2443 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2447 const U8 tmp = s[0];
2452 return utf16_to_utf8(p, d, bytelen, newlen);
2456 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2458 U8 tmpbuf[UTF8_MAXBYTES+1];
2459 uvchr_to_utf8(tmpbuf, c);
2460 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2463 /* Internal function so we can deprecate the external one, and call
2464 this one from other deprecated functions in this file */
2467 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2469 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2473 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2477 Perl__is_uni_perl_idcont(pTHX_ UV c)
2479 U8 tmpbuf[UTF8_MAXBYTES+1];
2480 uvchr_to_utf8(tmpbuf, c);
2481 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2485 Perl__is_uni_perl_idstart(pTHX_ UV c)
2487 U8 tmpbuf[UTF8_MAXBYTES+1];
2488 uvchr_to_utf8(tmpbuf, c);
2489 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2493 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2496 /* We have the latin1-range values compiled into the core, so just use
2497 * those, converting the result to UTF-8. The only difference between upper
2498 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2499 * either "SS" or "Ss". Which one to use is passed into the routine in
2500 * 'S_or_s' to avoid a test */
2502 UV converted = toUPPER_LATIN1_MOD(c);
2504 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2506 assert(S_or_s == 'S' || S_or_s == 's');
2508 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2509 characters in this range */
2510 *p = (U8) converted;
2515 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2516 * which it maps to one of them, so as to only have to have one check for
2517 * it in the main case */
2518 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2520 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2521 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2524 converted = GREEK_CAPITAL_LETTER_MU;
2526 #if UNICODE_MAJOR_VERSION > 2 \
2527 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2528 && UNICODE_DOT_DOT_VERSION >= 8)
2529 case LATIN_SMALL_LETTER_SHARP_S:
2536 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2537 " '%c' to map to '%c'",
2538 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2539 NOT_REACHED; /* NOTREACHED */
2543 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2544 *p = UTF8_TWO_BYTE_LO(converted);
2550 /* Call the function to convert a UTF-8 encoded character to the specified case.
2551 * Note that there may be more than one character in the result.
2552 * INP is a pointer to the first byte of the input character
2553 * OUTP will be set to the first byte of the string of changed characters. It
2554 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2555 * LENP will be set to the length in bytes of the string of changed characters
2557 * The functions return the ordinal of the first character in the string of
2559 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2560 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2561 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2562 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2563 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2564 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2566 /* This additionally has the input parameter 'specials', which if non-zero will
2567 * cause this to use the specials hash for folding (meaning get full case
2568 * folding); otherwise, when zero, this implies a simple case fold */
2569 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2570 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2573 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2575 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2576 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2577 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2578 * the changed version may be longer than the original character.
2580 * The ordinal of the first character of the changed version is returned
2581 * (but note, as explained above, that there may be more.) */
2583 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2586 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2589 uvchr_to_utf8(p, c);
2590 return CALL_UPPER_CASE(c, p, p, lenp);
2594 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2596 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2599 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2602 uvchr_to_utf8(p, c);
2603 return CALL_TITLE_CASE(c, p, p, lenp);
2607 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2609 /* We have the latin1-range values compiled into the core, so just use
2610 * those, converting the result to UTF-8. Since the result is always just
2611 * one character, we allow <p> to be NULL */
2613 U8 converted = toLOWER_LATIN1(c);
2615 PERL_UNUSED_ARG(dummy);
2618 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2623 /* Result is known to always be < 256, so can use the EIGHT_BIT
2625 *p = UTF8_EIGHT_BIT_HI(converted);
2626 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2634 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2636 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2639 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2642 uvchr_to_utf8(p, c);
2643 return CALL_LOWER_CASE(c, p, p, lenp);
2647 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2648 const unsigned int flags)
2650 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2651 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2652 * FOLD_FLAGS_FULL iff full folding is to be used;
2654 * Not to be used for locale folds
2659 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2660 PERL_UNUSED_CONTEXT;
2662 assert (! (flags & FOLD_FLAGS_LOCALE));
2664 if (UNLIKELY(c == MICRO_SIGN)) {
2665 converted = GREEK_SMALL_LETTER_MU;
2667 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2668 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2669 || UNICODE_DOT_DOT_VERSION > 0)
2670 else if ( (flags & FOLD_FLAGS_FULL)
2671 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2673 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2674 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2675 * under those circumstances. */
2676 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2677 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2678 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2680 return LATIN_SMALL_LETTER_LONG_S;
2690 else { /* In this range the fold of all other characters is their lower
2692 converted = toLOWER_LATIN1(c);
2695 if (UVCHR_IS_INVARIANT(converted)) {
2696 *p = (U8) converted;
2700 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2701 *p = UTF8_TWO_BYTE_LO(converted);
2709 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2712 /* Not currently externally documented, and subject to change
2713 * <flags> bits meanings:
2714 * FOLD_FLAGS_FULL iff full folding is to be used;
2715 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2716 * locale are to be used.
2717 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2720 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2722 if (flags & FOLD_FLAGS_LOCALE) {
2723 /* Treat a UTF-8 locale as not being in locale at all */
2724 if (IN_UTF8_CTYPE_LOCALE) {
2725 flags &= ~FOLD_FLAGS_LOCALE;
2728 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2729 goto needs_full_generality;
2734 return _to_fold_latin1((U8) c, p, lenp,
2735 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2738 /* Here, above 255. If no special needs, just use the macro */
2739 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2740 uvchr_to_utf8(p, c);
2741 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2743 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2744 the special flags. */
2745 U8 utf8_c[UTF8_MAXBYTES + 1];
2747 needs_full_generality:
2748 uvchr_to_utf8(utf8_c, c);
2749 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2754 PERL_STATIC_INLINE bool
2755 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2756 const char *const swashname, SV* const invlist)
2758 /* returns a boolean giving whether or not the UTF8-encoded character that
2759 * starts at <p> is in the swash indicated by <swashname>. <swash>
2760 * contains a pointer to where the swash indicated by <swashname>
2761 * is to be stored; which this routine will do, so that future calls will
2762 * look at <*swash> and only generate a swash if it is not null. <invlist>
2763 * is NULL or an inversion list that defines the swash. If not null, it
2764 * saves time during initialization of the swash.
2766 * Note that it is assumed that the buffer length of <p> is enough to
2767 * contain all the bytes that comprise the character. Thus, <*p> should
2768 * have been checked before this call for mal-formedness enough to assure
2771 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2773 /* The API should have included a length for the UTF-8 character in <p>,
2774 * but it doesn't. We therefore assume that p has been validated at least
2775 * as far as there being enough bytes available in it to accommodate the
2776 * character without reading beyond the end, and pass that number on to the
2777 * validating routine */
2778 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2779 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2780 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2782 NOT_REACHED; /* NOTREACHED */
2786 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2787 *swash = _core_swash_init("utf8",
2789 /* Only use the name if there is no inversion
2790 * list; otherwise will go out to disk */
2791 (invlist) ? "" : swashname,
2793 &PL_sv_undef, 1, 0, invlist, &flags);
2796 return swash_fetch(*swash, p, TRUE) != 0;
2799 PERL_STATIC_INLINE bool
2800 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2801 SV **swash, const char *const swashname,
2804 /* returns a boolean giving whether or not the UTF8-encoded character that
2805 * starts at <p>, and extending no further than <e - 1> is in the swash
2806 * indicated by <swashname>. <swash> contains a pointer to where the swash
2807 * indicated by <swashname> is to be stored; which this routine will do, so
2808 * that future calls will look at <*swash> and only generate a swash if it
2809 * is not null. <invlist> is NULL or an inversion list that defines the
2810 * swash. If not null, it saves time during initialization of the swash.
2813 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2815 if (! isUTF8_CHAR(p, e)) {
2816 _force_out_malformed_utf8_message(p, e, 0, 1);
2817 NOT_REACHED; /* NOTREACHED */
2821 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2822 *swash = _core_swash_init("utf8",
2824 /* Only use the name if there is no inversion
2825 * list; otherwise will go out to disk */
2826 (invlist) ? "" : swashname,
2828 &PL_sv_undef, 1, 0, invlist, &flags);
2831 return swash_fetch(*swash, p, TRUE) != 0;
2835 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2836 const char * const alternative,
2837 const bool use_locale,
2838 const char * const file,
2839 const unsigned line)
2843 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2845 if (ckWARN_d(WARN_DEPRECATED)) {
2847 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2848 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2849 if (! PL_seen_deprecated_macro) {
2850 PL_seen_deprecated_macro = newHV();
2852 if (! hv_store(PL_seen_deprecated_macro, key,
2853 strlen(key), &PL_sv_undef, 0))
2855 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2858 if (instr(file, "mathoms.c")) {
2859 Perl_warner(aTHX_ WARN_DEPRECATED,
2860 "In %s, line %d, starting in Perl v5.30, %s()"
2861 " will be removed. Avoid this message by"
2862 " converting to use %s().\n",
2863 file, line, name, alternative);
2866 Perl_warner(aTHX_ WARN_DEPRECATED,
2867 "In %s, line %d, starting in Perl v5.30, %s() will"
2868 " require an additional parameter. Avoid this"
2869 " message by converting to use %s().\n",
2870 file, line, name, alternative);
2877 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2879 const char * const name,
2880 const char * const alternative,
2881 const bool use_utf8,
2882 const bool use_locale,
2883 const char * const file,
2884 const unsigned line)
2886 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2888 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2890 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2900 case _CC_ALPHANUMERIC:
2904 return is_utf8_common(p,
2905 &PL_utf8_swash_ptrs[classnum],
2906 swash_property_names[classnum],
2907 PL_XPosix_ptrs[classnum]);
2910 return is_XPERLSPACE_high(p);
2912 return is_HORIZWS_high(p);
2914 return is_XDIGIT_high(p);
2920 return is_VERTWS_high(p);
2922 if (! PL_utf8_perl_idstart) {
2923 PL_utf8_perl_idstart
2924 = _new_invlist_C_array(_Perl_IDStart_invlist);
2926 return is_utf8_common(p, &PL_utf8_perl_idstart,
2927 "_Perl_IDStart", NULL);
2929 if (! PL_utf8_perl_idcont) {
2931 = _new_invlist_C_array(_Perl_IDCont_invlist);
2933 return is_utf8_common(p, &PL_utf8_perl_idcont,
2934 "_Perl_IDCont", NULL);
2938 /* idcont is the same as wordchar below 256 */
2939 if (classnum == _CC_IDCONT) {
2940 classnum = _CC_WORDCHAR;
2942 else if (classnum == _CC_IDFIRST) {
2946 classnum = _CC_ALPHA;
2950 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2951 return _generic_isCC(*p, classnum);
2954 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2957 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2958 return isFOO_lc(classnum, *p);
2961 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2964 NOT_REACHED; /* NOTREACHED */
2968 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2971 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2973 assert(classnum < _FIRST_NON_SWASH_CC);
2975 return is_utf8_common_with_len(p,
2977 &PL_utf8_swash_ptrs[classnum],
2978 swash_property_names[classnum],
2979 PL_XPosix_ptrs[classnum]);
2983 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2987 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2989 if (! PL_utf8_perl_idstart) {
2990 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2992 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2993 "_Perl_IDStart", invlist);
2997 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2999 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3003 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
3007 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3011 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3013 if (! PL_utf8_perl_idcont) {
3014 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3016 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3017 "_Perl_IDCont", invlist);
3021 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3023 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3025 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
3029 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3031 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3033 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
3037 Perl__is_utf8_mark(pTHX_ const U8 *p)
3039 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3041 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
3044 /* change namve uv1 to 'from' */
3046 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
3047 SV **swashp, const char *normal, const char *special)
3051 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3053 /* For code points that don't change case, we already know that the output
3054 * of this function is the unchanged input, so we can skip doing look-ups
3055 * for them. Unfortunately the case-changing code points are scattered
3056 * around. But there are some long consecutive ranges where there are no
3057 * case changing code points. By adding tests, we can eliminate the lookup
3058 * for all the ones in such ranges. This is currently done here only for
3059 * just a few cases where the scripts are in common use in modern commerce
3060 * (and scripts adjacent to those which can be included without additional
3063 if (uv1 >= 0x0590) {
3064 /* This keeps from needing further processing the code points most
3065 * likely to be used in the following non-cased scripts: Hebrew,
3066 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3067 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3068 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3073 /* The following largish code point ranges also don't have case
3074 * changes, but khw didn't think they warranted extra tests to speed
3075 * them up (which would slightly slow down everything else above them):
3076 * 1100..139F Hangul Jamo, Ethiopic
3077 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3078 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3079 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3080 * Combining Diacritical Marks Extended, Balinese,
3081 * Sundanese, Batak, Lepcha, Ol Chiki
3082 * 2000..206F General Punctuation
3085 if (uv1 >= 0x2D30) {
3087 /* This keeps the from needing further processing the code points
3088 * most likely to be used in the following non-cased major scripts:
3089 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3091 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3092 * event that Unicode eventually allocates the unused block as of
3093 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3094 * that the test suite will start having failures to alert you
3095 * should that happen) */
3100 if (uv1 >= 0xAC00) {
3101 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3102 if (ckWARN_d(WARN_SURROGATE)) {
3103 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3104 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3105 "Operation \"%s\" returns its argument for"
3106 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3111 /* AC00..FAFF Catches Hangul syllables and private use, plus
3118 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3119 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3120 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3121 MAX_EXTERNALLY_LEGAL_CP);
3123 if (ckWARN_d(WARN_NON_UNICODE)) {
3124 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3125 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3126 "Operation \"%s\" returns its argument for"
3127 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3131 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3133 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3136 /* As of Unicode 10.0, this means we avoid swash creation
3137 * for anything beyond high Plane 1 (below emojis) */
3144 /* Note that non-characters are perfectly legal, so no warning should
3145 * be given. There are so few of them, that it isn't worth the extra
3146 * tests to avoid swash creation */
3149 if (!*swashp) /* load on-demand */
3150 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3154 /* It might be "special" (sometimes, but not always,
3155 * a multicharacter mapping) */
3159 /* If passed in the specials name, use that; otherwise use any
3160 * given in the swash */
3161 if (*special != '\0') {
3162 hv = get_hv(special, 0);
3165 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3167 hv = MUTABLE_HV(SvRV(*svp));
3172 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
3177 s = SvPV_const(*svp, len);
3180 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
3182 Copy(s, ustrp, len, U8);
3187 if (!len && *swashp) {
3188 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
3191 /* It was "normal" (a single character mapping). */
3192 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
3200 return valid_utf8_to_uvchr(ustrp, 0);
3203 /* Here, there was no mapping defined, which means that the code point maps
3204 * to itself. Return the inputs */
3207 if (p != ustrp) { /* Don't copy onto itself */
3208 Copy(p, ustrp, len, U8);
3219 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3220 U8* const ustrp, STRLEN *lenp)
3222 /* This is called when changing the case of a UTF-8-encoded character above
3223 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3224 * result contains a character that crosses the 255/256 boundary, disallow
3225 * the change, and return the original code point. See L<perlfunc/lc> for
3228 * p points to the original string whose case was changed; assumed
3229 * by this routine to be well-formed
3230 * result the code point of the first character in the changed-case string
3231 * ustrp points to the changed-case string (<result> represents its
3233 * lenp points to the length of <ustrp> */
3235 UV original; /* To store the first code point of <p> */
3237 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3239 assert(UTF8_IS_ABOVE_LATIN1(*p));
3241 /* We know immediately if the first character in the string crosses the
3242 * boundary, so can skip */
3245 /* Look at every character in the result; if any cross the
3246 * boundary, the whole thing is disallowed */
3247 U8* s = ustrp + UTF8SKIP(ustrp);
3248 U8* e = ustrp + *lenp;
3250 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3256 /* Here, no characters crossed, result is ok as-is, but we warn. */
3257 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3263 /* Failed, have to return the original */
3264 original = valid_utf8_to_uvchr(p, lenp);
3266 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3267 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3268 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3269 " locale; resolved to \"\\x{%" UVXf "}\".",
3273 Copy(p, ustrp, *lenp, char);
3278 S_check_and_deprecate(pTHX_ const U8 *p,
3280 const unsigned int type, /* See below */
3281 const bool use_locale, /* Is this a 'LC_'
3283 const char * const file,
3284 const unsigned line)
3286 /* This is a temporary function to deprecate the unsafe calls to the case
3287 * changing macros and functions. It keeps all the special stuff in just
3290 * It updates *e with the pointer to the end of the input string. If using
3291 * the old-style macros, *e is NULL on input, and so this function assumes
3292 * the input string is long enough to hold the entire UTF-8 sequence, and
3293 * sets *e accordingly, but it then returns a flag to pass the
3294 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3295 * using the full length if possible.
3297 * It also does the assert that *e > p when *e is not NULL. This should be
3298 * migrated to the callers when this function gets deleted.
3300 * The 'type' parameter is used for the caller to specify which case
3301 * changing function this is called from: */
3303 # define DEPRECATE_TO_UPPER 0
3304 # define DEPRECATE_TO_TITLE 1
3305 # define DEPRECATE_TO_LOWER 2
3306 # define DEPRECATE_TO_FOLD 3
3308 U32 utf8n_flags = 0;
3310 const char * alternative;
3312 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3315 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3316 *e = p + UTF8SKIP(p);
3318 /* For mathoms.c calls, we use the function name we know is stored
3319 * there. It could be part of a larger path */
3320 if (type == DEPRECATE_TO_UPPER) {
3321 name = instr(file, "mathoms.c")
3324 alternative = "toUPPER_utf8_safe";
3326 else if (type == DEPRECATE_TO_TITLE) {
3327 name = instr(file, "mathoms.c")
3330 alternative = "toTITLE_utf8_safe";
3332 else if (type == DEPRECATE_TO_LOWER) {
3333 name = instr(file, "mathoms.c")
3336 alternative = "toLOWER_utf8_safe";
3338 else if (type == DEPRECATE_TO_FOLD) {
3339 name = instr(file, "mathoms.c")
3342 alternative = "toFOLD_utf8_safe";
3344 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3346 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3355 /* The process for changing the case is essentially the same for the four case
3356 * change types, except there are complications for folding. Otherwise the
3357 * difference is only which case to change to. To make sure that they all do
3358 * the same thing, the bodies of the functions are extracted out into the
3359 * following two macros. The functions are written with the same variable
3360 * names, and these are known and used inside these macros. It would be
3361 * better, of course, to have inline functions to do it, but since different
3362 * macros are called, depending on which case is being changed to, this is not
3363 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3364 * function can start with the common start macro, then finish with its special
3365 * handling; while the other three cases can just use the common end macro.
3367 * The algorithm is to use the proper (passed in) macro or function to change
3368 * the case for code points that are below 256. The macro is used if using
3369 * locale rules for the case change; the function if not. If the code point is
3370 * above 255, it is computed from the input UTF-8, and another macro is called
3371 * to do the conversion. If necessary, the output is converted to UTF-8. If
3372 * using a locale, we have to check that the change did not cross the 255/256
3373 * boundary, see check_locale_boundary_crossing() for further details.
3375 * The macros are split with the correct case change for the below-256 case
3376 * stored into 'result', and in the middle of an else clause for the above-255
3377 * case. At that point in the 'else', 'result' is not the final result, but is
3378 * the input code point calculated from the UTF-8. The fold code needs to
3379 * realize all this and take it from there.
3381 * If you read the two macros as sequential, it's easier to understand what's
3383 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3384 L1_func_extra_param) \
3386 if (flags & (locale_flags)) { \
3387 /* Treat a UTF-8 locale as not being in locale at all */ \
3388 if (IN_UTF8_CTYPE_LOCALE) { \
3389 flags &= ~(locale_flags); \
3392 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3396 if (UTF8_IS_INVARIANT(*p)) { \
3397 if (flags & (locale_flags)) { \
3398 result = LC_L1_change_macro(*p); \
3401 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3404 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3405 if (flags & (locale_flags)) { \
3406 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3410 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3411 ustrp, lenp, L1_func_extra_param); \
3414 else { /* malformed UTF-8 or ord above 255 */ \
3415 STRLEN len_result; \
3416 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3417 if (len_result == (STRLEN) -1) { \
3418 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3422 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3423 result = change_macro(result, p, ustrp, lenp); \
3425 if (flags & (locale_flags)) { \
3426 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3431 /* Here, used locale rules. Convert back to UTF-8 */ \
3432 if (UTF8_IS_INVARIANT(result)) { \
3433 *ustrp = (U8) result; \
3437 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3438 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3445 =for apidoc to_utf8_upper
3447 Instead use L</toUPPER_utf8_safe>.
3451 /* Not currently externally documented, and subject to change:
3452 * <flags> is set iff iff the rules from the current underlying locale are to
3456 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3461 const char * const file,
3465 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3466 cBOOL(flags), file, line);
3468 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3470 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3471 /* 2nd char of uc(U+DF) is 'S' */
3472 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3473 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3477 =for apidoc to_utf8_title
3479 Instead use L</toTITLE_utf8_safe>.
3483 /* Not currently externally documented, and subject to change:
3484 * <flags> is set iff the rules from the current underlying locale are to be
3485 * used. Since titlecase is not defined in POSIX, for other than a
3486 * UTF-8 locale, uppercase is used instead for code points < 256.
3490 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3495 const char * const file,
3499 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3500 cBOOL(flags), file, line);
3502 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3504 /* 2nd char of ucfirst(U+DF) is 's' */
3505 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3506 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3510 =for apidoc to_utf8_lower
3512 Instead use L</toLOWER_utf8_safe>.
3516 /* Not currently externally documented, and subject to change:
3517 * <flags> is set iff iff the rules from the current underlying locale are to
3522 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3527 const char * const file,
3531 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3532 cBOOL(flags), file, line);
3534 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3536 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3537 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3541 =for apidoc to_utf8_fold
3543 Instead use L</toFOLD_utf8_safe>.
3547 /* Not currently externally documented, and subject to change,
3549 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3550 * locale are to be used.
3551 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3552 * otherwise simple folds
3553 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3558 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3563 const char * const file,
3567 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3568 cBOOL(flags), file, line);
3570 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3572 /* These are mutually exclusive */
3573 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3575 assert(p != ustrp); /* Otherwise overwrites */
3577 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3578 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3580 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3582 if (flags & FOLD_FLAGS_LOCALE) {
3584 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3585 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3587 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3588 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3590 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3592 /* Special case these two characters, as what normally gets
3593 * returned under locale doesn't work */
3594 if (UTF8SKIP(p) == cap_sharp_s_len
3595 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3597 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3598 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3599 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3600 "resolved to \"\\x{17F}\\x{17F}\".");
3605 if (UTF8SKIP(p) == long_s_t_len
3606 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3608 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3609 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3610 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3611 "resolved to \"\\x{FB06}\".");
3612 goto return_ligature_st;
3615 #if UNICODE_MAJOR_VERSION == 3 \
3616 && UNICODE_DOT_VERSION == 0 \
3617 && UNICODE_DOT_DOT_VERSION == 1
3618 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3620 /* And special case this on this Unicode version only, for the same
3621 * reaons the other two are special cased. They would cross the
3622 * 255/256 boundary which is forbidden under /l, and so the code
3623 * wouldn't catch that they are equivalent (which they are only in
3625 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3626 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3628 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3629 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3630 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3631 "resolved to \"\\x{0131}\".");
3632 goto return_dotless_i;
3636 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3638 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3642 /* This is called when changing the case of a UTF-8-encoded
3643 * character above the ASCII range, and the result should not
3644 * contain an ASCII character. */
3646 UV original; /* To store the first code point of <p> */
3648 /* Look at every character in the result; if any cross the
3649 * boundary, the whole thing is disallowed */
3651 U8* e = ustrp + *lenp;
3654 /* Crossed, have to return the original */
3655 original = valid_utf8_to_uvchr(p, lenp);
3657 /* But in these instances, there is an alternative we can
3658 * return that is valid */
3659 if (original == LATIN_SMALL_LETTER_SHARP_S
3660 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3661 || original == LATIN_CAPITAL_LETTER_SHARP_S
3666 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3667 goto return_ligature_st;
3669 #if UNICODE_MAJOR_VERSION == 3 \
3670 && UNICODE_DOT_VERSION == 0 \
3671 && UNICODE_DOT_DOT_VERSION == 1
3673 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3674 goto return_dotless_i;
3677 Copy(p, ustrp, *lenp, char);
3683 /* Here, no characters crossed, result is ok as-is */
3688 /* Here, used locale rules. Convert back to UTF-8 */
3689 if (UTF8_IS_INVARIANT(result)) {
3690 *ustrp = (U8) result;
3694 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3695 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3702 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3703 * folds to a string of two of these characters. By returning this
3704 * instead, then, e.g.,
3705 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3708 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3709 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3711 return LATIN_SMALL_LETTER_LONG_S;
3714 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3715 * have the other one fold to it */
3717 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3718 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3719 return LATIN_SMALL_LIGATURE_ST;
3721 #if UNICODE_MAJOR_VERSION == 3 \
3722 && UNICODE_DOT_VERSION == 0 \
3723 && UNICODE_DOT_DOT_VERSION == 1
3726 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3727 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3728 return LATIN_SMALL_LETTER_DOTLESS_I;
3735 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3736 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3737 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3741 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3742 I32 minbits, I32 none)
3744 PERL_ARGS_ASSERT_SWASH_INIT;
3746 /* Returns a copy of a swash initiated by the called function. This is the
3747 * public interface, and returning a copy prevents others from doing
3748 * mischief on the original */
3750 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3755 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3756 I32 minbits, I32 none, SV* invlist,
3760 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3761 * use the following define */
3763 #define CORE_SWASH_INIT_RETURN(x) \
3764 PL_curpm= old_PL_curpm; \
3767 /* Initialize and return a swash, creating it if necessary. It does this
3768 * by calling utf8_heavy.pl in the general case. The returned value may be
3769 * the swash's inversion list instead if the input parameters allow it.
3770 * Which is returned should be immaterial to callers, as the only
3771 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3772 * and swash_to_invlist() handle both these transparently.
3774 * This interface should only be used by functions that won't destroy or
3775 * adversely change the swash, as doing so affects all other uses of the
3776 * swash in the program; the general public should use 'Perl_swash_init'
3779 * pkg is the name of the package that <name> should be in.
3780 * name is the name of the swash to find. Typically it is a Unicode
3781 * property name, including user-defined ones
3782 * listsv is a string to initialize the swash with. It must be of the form
3783 * documented as the subroutine return value in
3784 * L<perlunicode/User-Defined Character Properties>
3785 * minbits is the number of bits required to represent each data element.
3786 * It is '1' for binary properties.
3787 * none I (khw) do not understand this one, but it is used only in tr///.
3788 * invlist is an inversion list to initialize the swash with (or NULL)
3789 * flags_p if non-NULL is the address of various input and output flag bits
3790 * to the routine, as follows: ('I' means is input to the routine;
3791 * 'O' means output from the routine. Only flags marked O are
3792 * meaningful on return.)
3793 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3794 * came from a user-defined property. (I O)
3795 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3796 * when the swash cannot be located, to simply return NULL. (I)
3797 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3798 * return of an inversion list instead of a swash hash if this routine
3799 * thinks that would result in faster execution of swash_fetch() later
3802 * Thus there are three possible inputs to find the swash: <name>,
3803 * <listsv>, and <invlist>. At least one must be specified. The result
3804 * will be the union of the specified ones, although <listsv>'s various
3805 * actions can intersect, etc. what <name> gives. To avoid going out to
3806 * disk at all, <invlist> should specify completely what the swash should
3807 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3809 * <invlist> is only valid for binary properties */
3811 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3813 SV* retval = &PL_sv_undef;
3814 HV* swash_hv = NULL;
3815 const int invlist_swash_boundary =
3816 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3817 ? 512 /* Based on some benchmarking, but not extensive, see commit
3819 : -1; /* Never return just an inversion list */
3821 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3822 assert(! invlist || minbits == 1);
3824 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3825 regex that triggered the swash init and the swash init
3826 perl logic itself. See perl #122747 */
3828 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3830 if (listsv != &PL_sv_undef || strNE(name, "")) {
3832 const size_t pkg_len = strlen(pkg);
3833 const size_t name_len = strlen(name);
3834 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3838 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3840 PUSHSTACKi(PERLSI_MAGIC);
3844 /* We might get here via a subroutine signature which uses a utf8
3845 * parameter name, at which point PL_subname will have been set
3846 * but not yet used. */
3847 save_item(PL_subname);
3848 if (PL_parser && PL_parser->error_count)
3849 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3850 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3851 if (!method) { /* demand load UTF-8 */
3853 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3854 GvSV(PL_errgv) = NULL;
3855 #ifndef NO_TAINT_SUPPORT
3856 /* It is assumed that callers of this routine are not passing in
3857 * any user derived data. */
3858 /* Need to do this after save_re_context() as it will set
3859 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3860 * in Perl_magic_get). Even line to create errsv_save can turn on
3862 SAVEBOOL(TAINT_get);
3865 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3868 /* Not ERRSV, as there is no need to vivify a scalar we are
3869 about to discard. */
3870 SV * const errsv = GvSV(PL_errgv);
3871 if (!SvTRUE(errsv)) {
3872 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3873 SvREFCNT_dec(errsv);
3881 mPUSHp(pkg, pkg_len);
3882 mPUSHp(name, name_len);
3887 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3888 GvSV(PL_errgv) = NULL;
3889 /* If we already have a pointer to the method, no need to use
3890 * call_method() to repeat the lookup. */
3892 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3893 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3895 retval = *PL_stack_sp--;
3896 SvREFCNT_inc(retval);
3899 /* Not ERRSV. See above. */
3900 SV * const errsv = GvSV(PL_errgv);
3901 if (!SvTRUE(errsv)) {
3902 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3903 SvREFCNT_dec(errsv);
3908 if (IN_PERL_COMPILETIME) {
3909 CopHINTS_set(PL_curcop, PL_hints);
3911 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3912 if (SvPOK(retval)) {
3914 /* If caller wants to handle missing properties, let them */
3915 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3916 CORE_SWASH_INIT_RETURN(NULL);
3919 "Can't find Unicode property definition \"%" SVf "\"",
3921 NOT_REACHED; /* NOTREACHED */
3924 } /* End of calling the module to find the swash */
3926 /* If this operation fetched a swash, and we will need it later, get it */
3927 if (retval != &PL_sv_undef
3928 && (minbits == 1 || (flags_p
3930 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3932 swash_hv = MUTABLE_HV(SvRV(retval));
3934 /* If we don't already know that there is a user-defined component to
3935 * this swash, and the user has indicated they wish to know if there is
3936 * one (by passing <flags_p>), find out */
3937 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3938 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3939 if (user_defined && SvUV(*user_defined)) {
3940 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3945 /* Make sure there is an inversion list for binary properties */
3947 SV** swash_invlistsvp = NULL;
3948 SV* swash_invlist = NULL;
3949 bool invlist_in_swash_is_valid = FALSE;
3950 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3951 an unclaimed reference count */
3953 /* If this operation fetched a swash, get its already existing
3954 * inversion list, or create one for it */
3957 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3958 if (swash_invlistsvp) {
3959 swash_invlist = *swash_invlistsvp;
3960 invlist_in_swash_is_valid = TRUE;
3963 swash_invlist = _swash_to_invlist(retval);
3964 swash_invlist_unclaimed = TRUE;
3968 /* If an inversion list was passed in, have to include it */
3971 /* Any fetched swash will by now have an inversion list in it;
3972 * otherwise <swash_invlist> will be NULL, indicating that we
3973 * didn't fetch a swash */
3974 if (swash_invlist) {
3976 /* Add the passed-in inversion list, which invalidates the one
3977 * already stored in the swash */
3978 invlist_in_swash_is_valid = FALSE;
3979 SvREADONLY_off(swash_invlist); /* Turned on again below */
3980 _invlist_union(invlist, swash_invlist, &swash_invlist);
3984 /* Here, there is no swash already. Set up a minimal one, if
3985 * we are going to return a swash */
3986 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3988 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3990 swash_invlist = invlist;
3994 /* Here, we have computed the union of all the passed-in data. It may
3995 * be that there was an inversion list in the swash which didn't get
3996 * touched; otherwise save the computed one */
3997 if (! invlist_in_swash_is_valid
3998 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
4000 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
4002 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4004 /* We just stole a reference count. */
4005 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
4006 else SvREFCNT_inc_simple_void_NN(swash_invlist);
4009 /* The result is immutable. Forbid attempts to change it. */
4010 SvREADONLY_on(swash_invlist);
4012 /* Use the inversion list stand-alone if small enough */
4013 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
4014 SvREFCNT_dec(retval);
4015 if (!swash_invlist_unclaimed)
4016 SvREFCNT_inc_simple_void_NN(swash_invlist);
4017 retval = newRV_noinc(swash_invlist);
4021 CORE_SWASH_INIT_RETURN(retval);
4022 #undef CORE_SWASH_INIT_RETURN
4026 /* This API is wrong for special case conversions since we may need to
4027 * return several Unicode characters for a single Unicode character
4028 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
4029 * the lower-level routine, and it is similarly broken for returning
4030 * multiple values. --jhi
4031 * For those, you should use S__to_utf8_case() instead */
4032 /* Now SWASHGET is recasted into S_swatch_get in this file. */
4035 * Returns the value of property/mapping C<swash> for the first character
4036 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4037 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
4038 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
4040 * A "swash" is a hash which contains initially the keys/values set up by
4041 * SWASHNEW. The purpose is to be able to completely represent a Unicode
4042 * property for all possible code points. Things are stored in a compact form
4043 * (see utf8_heavy.pl) so that calculation is required to find the actual
4044 * property value for a given code point. As code points are looked up, new
4045 * key/value pairs are added to the hash, so that the calculation doesn't have
4046 * to ever be re-done. Further, each calculation is done, not just for the
4047 * desired one, but for a whole block of code points adjacent to that one.
4048 * For binary properties on ASCII machines, the block is usually for 64 code
4049 * points, starting with a code point evenly divisible by 64. Thus if the
4050 * property value for code point 257 is requested, the code goes out and
4051 * calculates the property values for all 64 code points between 256 and 319,
4052 * and stores these as a single 64-bit long bit vector, called a "swatch",
4053 * under the key for code point 256. The key is the UTF-8 encoding for code
4054 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
4055 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
4056 * for code point 258 is then requested, this code realizes that it would be
4057 * stored under the key for 256, and would find that value and extract the
4058 * relevant bit, offset from 256.
4060 * Non-binary properties are stored in as many bits as necessary to represent
4061 * their values (32 currently, though the code is more general than that), not
4062 * as single bits, but the principle is the same: the value for each key is a
4063 * vector that encompasses the property values for all code points whose UTF-8
4064 * representations are represented by the key. That is, for all code points
4065 * whose UTF-8 representations are length N bytes, and the key is the first N-1
4069 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
4071 HV *const hv = MUTABLE_HV(SvRV(swash));
4076 const U8 *tmps = NULL;
4080 PERL_ARGS_ASSERT_SWASH_FETCH;
4082 /* If it really isn't a hash, it isn't really swash; must be an inversion
4084 if (SvTYPE(hv) != SVt_PVHV) {
4085 return _invlist_contains_cp((SV*)hv,
4087 ? valid_utf8_to_uvchr(ptr, NULL)
4091 /* We store the values in a "swatch" which is a vec() value in a swash
4092 * hash. Code points 0-255 are a single vec() stored with key length
4093 * (klen) 0. All other code points have a UTF-8 representation
4094 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
4095 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
4096 * length for them is the length of the encoded char - 1. ptr[klen] is the
4097 * final byte in the sequence representing the character */
4098 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
4103 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
4106 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
4109 klen = UTF8SKIP(ptr) - 1;
4111 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
4112 * the vec is the final byte in the sequence. (In EBCDIC this is
4113 * converted to I8 to get consecutive values.) To help you visualize
4115 * Straight 1047 After final byte
4116 * UTF-8 UTF-EBCDIC I8 transform
4117 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
4118 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
4120 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
4121 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
4123 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
4124 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
4126 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
4127 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
4129 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
4130 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
4132 * (There are no discontinuities in the elided (...) entries.)
4133 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
4134 * key for the next 31, up through U+043F, whose UTF-8 final byte is
4135 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
4136 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
4137 * index into the vec() swatch (after subtracting 0x80, which we
4138 * actually do with an '&').
4139 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
4140 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4141 * dicontinuities which go away by transforming it into I8, and we
4142 * effectively subtract 0xA0 to get the index. */
4143 needents = (1 << UTF_ACCUMULATION_SHIFT);
4144 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4148 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4149 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4150 * it's nothing to sniff at.) Pity we usually come through at least
4151 * two function calls to get here...
4153 * NB: this code assumes that swatches are never modified, once generated!
4156 if (hv == PL_last_swash_hv &&
4157 klen == PL_last_swash_klen &&
4158 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4160 tmps = PL_last_swash_tmps;
4161 slen = PL_last_swash_slen;
4164 /* Try our second-level swatch cache, kept in a hash. */
4165 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4167 /* If not cached, generate it via swatch_get */
4168 if (!svp || !SvPOK(*svp)
4169 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4172 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4173 swatch = swatch_get(swash,
4174 code_point & ~((UV)needents - 1),
4177 else { /* For the first 256 code points, the swatch has a key of
4179 swatch = swatch_get(swash, 0, needents);
4182 if (IN_PERL_COMPILETIME)
4183 CopHINTS_set(PL_curcop, PL_hints);
4185 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4187 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4188 || (slen << 3) < needents)
4189 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4190 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4191 svp, tmps, (UV)slen, (UV)needents);
4194 PL_last_swash_hv = hv;
4195 assert(klen <= sizeof(PL_last_swash_key));
4196 PL_last_swash_klen = (U8)klen;
4197 /* FIXME change interpvar.h? */
4198 PL_last_swash_tmps = (U8 *) tmps;
4199 PL_last_swash_slen = slen;
4201 Copy(ptr, PL_last_swash_key, klen, U8);
4204 switch ((int)((slen << 3) / needents)) {
4206 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4208 return ((UV) tmps[off]);
4212 ((UV) tmps[off ] << 8) +
4213 ((UV) tmps[off + 1]);
4217 ((UV) tmps[off ] << 24) +
4218 ((UV) tmps[off + 1] << 16) +
4219 ((UV) tmps[off + 2] << 8) +
4220 ((UV) tmps[off + 3]);
4222 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4223 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4224 NORETURN_FUNCTION_END;
4227 /* Read a single line of the main body of the swash input text. These are of
4230 * where each number is hex. The first two numbers form the minimum and
4231 * maximum of a range, and the third is the value associated with the range.
4232 * Not all swashes should have a third number
4234 * On input: l points to the beginning of the line to be examined; it points
4235 * to somewhere in the string of the whole input text, and is
4236 * terminated by a \n or the null string terminator.
4237 * lend points to the null terminator of that string
4238 * wants_value is non-zero if the swash expects a third number
4239 * typestr is the name of the swash's mapping, like 'ToLower'
4240 * On output: *min, *max, and *val are set to the values read from the line.
4241 * returns a pointer just beyond the line examined. If there was no
4242 * valid min number on the line, returns lend+1
4246 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4247 const bool wants_value, const U8* const typestr)
4249 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4250 STRLEN numlen; /* Length of the number */
4251 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4252 | PERL_SCAN_DISALLOW_PREFIX
4253 | PERL_SCAN_SILENT_NON_PORTABLE;
4255 /* nl points to the next \n in the scan */
4256 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4258 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4260 /* Get the first number on the line: the range minimum */
4262 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4263 *max = *min; /* So can never return without setting max */
4264 if (numlen) /* If found a hex number, position past it */
4266 else if (nl) { /* Else, go handle next line, if any */
4267 return nl + 1; /* 1 is length of "\n" */
4269 else { /* Else, no next line */
4270 return lend + 1; /* to LIST's end at which \n is not found */
4273 /* The max range value follows, separated by a BLANK */
4276 flags = PERL_SCAN_SILENT_ILLDIGIT
4277 | PERL_SCAN_DISALLOW_PREFIX
4278 | PERL_SCAN_SILENT_NON_PORTABLE;
4280 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4283 else /* If no value here, it is a single element range */
4286 /* Non-binary tables have a third entry: what the first element of the
4287 * range maps to. The map for those currently read here is in hex */
4291 flags = PERL_SCAN_SILENT_ILLDIGIT
4292 | PERL_SCAN_DISALLOW_PREFIX
4293 | PERL_SCAN_SILENT_NON_PORTABLE;
4295 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4304 /* diag_listed_as: To%s: illegal mapping '%s' */
4305 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4311 *val = 0; /* bits == 1, then any val should be ignored */
4313 else { /* Nothing following range min, should be single element with no
4318 /* diag_listed_as: To%s: illegal mapping '%s' */
4319 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4323 *val = 0; /* bits == 1, then val should be ignored */
4326 /* Position to next line if any, or EOF */
4336 * Returns a swatch (a bit vector string) for a code point sequence
4337 * that starts from the value C<start> and comprises the number C<span>.
4338 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4339 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4342 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4345 U8 *l, *lend, *x, *xend, *s, *send;
4346 STRLEN lcur, xcur, scur;
4347 HV *const hv = MUTABLE_HV(SvRV(swash));
4348 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4350 SV** listsvp = NULL; /* The string containing the main body of the table */
4351 SV** extssvp = NULL;
4352 SV** invert_it_svp = NULL;
4355 STRLEN octets; /* if bits == 1, then octets == 0 */
4357 UV end = start + span;
4359 if (invlistsvp == NULL) {
4360 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4361 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4362 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4363 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4364 listsvp = hv_fetchs(hv, "LIST", FALSE);
4365 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4367 bits = SvUV(*bitssvp);
4368 none = SvUV(*nonesvp);
4369 typestr = (U8*)SvPV_nolen(*typesvp);
4375 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4377 PERL_ARGS_ASSERT_SWATCH_GET;
4379 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4380 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4384 /* If overflowed, use the max possible */
4390 /* create and initialize $swatch */
4391 scur = octets ? (span * octets) : (span + 7) / 8;
4392 swatch = newSV(scur);
4394 s = (U8*)SvPVX(swatch);
4395 if (octets && none) {
4396 const U8* const e = s + scur;
4399 *s++ = (U8)(none & 0xff);
4400 else if (bits == 16) {
4401 *s++ = (U8)((none >> 8) & 0xff);
4402 *s++ = (U8)( none & 0xff);
4404 else if (bits == 32) {
4405 *s++ = (U8)((none >> 24) & 0xff);
4406 *s++ = (U8)((none >> 16) & 0xff);
4407 *s++ = (U8)((none >> 8) & 0xff);
4408 *s++ = (U8)( none & 0xff);
4414 (void)memzero((U8*)s, scur + 1);
4416 SvCUR_set(swatch, scur);
4417 s = (U8*)SvPVX(swatch);
4419 if (invlistsvp) { /* If has an inversion list set up use that */
4420 _invlist_populate_swatch(*invlistsvp, start, end, s);
4424 /* read $swash->{LIST} */
4425 l = (U8*)SvPV(*listsvp, lcur);
4428 UV min, max, val, upper;
4429 l = swash_scan_list_line(l, lend, &min, &max, &val,
4430 cBOOL(octets), typestr);
4435 /* If looking for something beyond this range, go try the next one */
4439 /* <end> is generally 1 beyond where we want to set things, but at the
4440 * platform's infinity, where we can't go any higher, we want to
4441 * include the code point at <end> */
4444 : (max != UV_MAX || end != UV_MAX)
4451 if (!none || val < none) {
4456 for (key = min; key <= upper; key++) {
4458 /* offset must be non-negative (start <= min <= key < end) */
4459 offset = octets * (key - start);
4461 s[offset] = (U8)(val & 0xff);
4462 else if (bits == 16) {
4463 s[offset ] = (U8)((val >> 8) & 0xff);
4464 s[offset + 1] = (U8)( val & 0xff);
4466 else if (bits == 32) {
4467 s[offset ] = (U8)((val >> 24) & 0xff);
4468 s[offset + 1] = (U8)((val >> 16) & 0xff);
4469 s[offset + 2] = (U8)((val >> 8) & 0xff);
4470 s[offset + 3] = (U8)( val & 0xff);
4473 if (!none || val < none)
4477 else { /* bits == 1, then val should be ignored */
4482 for (key = min; key <= upper; key++) {
4483 const STRLEN offset = (STRLEN)(key - start);
4484 s[offset >> 3] |= 1 << (offset & 7);
4489 /* Invert if the data says it should be. Assumes that bits == 1 */
4490 if (invert_it_svp && SvUV(*invert_it_svp)) {
4492 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4493 * be 0, and their inversion should also be 0, as we don't succeed any
4494 * Unicode property matches for non-Unicode code points */
4495 if (start <= PERL_UNICODE_MAX) {
4497 /* The code below assumes that we never cross the
4498 * Unicode/above-Unicode boundary in a range, as otherwise we would
4499 * have to figure out where to stop flipping the bits. Since this
4500 * boundary is divisible by a large power of 2, and swatches comes
4501 * in small powers of 2, this should be a valid assumption */
4502 assert(start + span - 1 <= PERL_UNICODE_MAX);
4512 /* read $swash->{EXTRAS}
4513 * This code also copied to swash_to_invlist() below */
4514 x = (U8*)SvPV(*extssvp, xcur);
4522 SV **otherbitssvp, *other;
4526 const U8 opc = *x++;
4530 nl = (U8*)memchr(x, '\n', xend - x);
4532 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4534 x = nl + 1; /* 1 is length of "\n" */
4538 x = xend; /* to EXTRAS' end at which \n is not found */
4545 namelen = nl - namestr;
4549 namelen = xend - namestr;
4553 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4554 otherhv = MUTABLE_HV(SvRV(*othersvp));
4555 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4556 otherbits = (STRLEN)SvUV(*otherbitssvp);
4557 if (bits < otherbits)
4558 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4559 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4561 /* The "other" swatch must be destroyed after. */
4562 other = swatch_get(*othersvp, start, span);
4563 o = (U8*)SvPV(other, olen);
4566 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4568 s = (U8*)SvPV(swatch, slen);
4569 if (bits == 1 && otherbits == 1) {
4571 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4572 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4573 (UV)slen, (UV)olen);
4597 STRLEN otheroctets = otherbits >> 3;
4599 U8* const send = s + slen;
4604 if (otherbits == 1) {
4605 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4609 STRLEN vlen = otheroctets;
4617 if (opc == '+' && otherval)
4618 NOOP; /* replace with otherval */
4619 else if (opc == '!' && !otherval)
4621 else if (opc == '-' && otherval)
4623 else if (opc == '&' && !otherval)
4626 s += octets; /* no replacement */
4631 *s++ = (U8)( otherval & 0xff);
4632 else if (bits == 16) {
4633 *s++ = (U8)((otherval >> 8) & 0xff);
4634 *s++ = (U8)( otherval & 0xff);
4636 else if (bits == 32) {
4637 *s++ = (U8)((otherval >> 24) & 0xff);
4638 *s++ = (U8)((otherval >> 16) & 0xff);
4639 *s++ = (U8)((otherval >> 8) & 0xff);
4640 *s++ = (U8)( otherval & 0xff);
4644 sv_free(other); /* through with it! */
4650 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4653 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4654 * Can't be used on a property that is subject to user override, as it
4655 * relies on the value of SPECIALS in the swash which would be set by
4656 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4657 * for overridden properties
4659 * Returns a hash which is the inversion and closure of a swash mapping.
4660 * For example, consider the input lines:
4665 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4666 * 006C. The value for each key is an array. For 006C, the array would
4667 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4668 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4670 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4671 * keys are only code points that are folded-to, so it isn't a full closure.
4673 * Essentially, for any code point, it gives all the code points that map to
4674 * it, or the list of 'froms' for that point.
4676 * Currently it ignores any additions or deletions from other swashes,
4677 * looking at just the main body of the swash, and if there are SPECIALS
4678 * in the swash, at that hash
4680 * The specials hash can be extra code points, and most likely consists of
4681 * maps from single code points to multiple ones (each expressed as a string
4682 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4683 * However consider this possible input in the specials hash:
4684 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4685 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4687 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4688 * currently handle. But it also means that FB05 and FB06 are equivalent in
4689 * a 1-1 mapping which we should handle, and this relationship may not be in
4690 * the main table. Therefore this function examines all the multi-char
4691 * sequences and adds the 1-1 mappings that come out of that.
4693 * XXX This function was originally intended to be multipurpose, but its
4694 * only use is quite likely to remain for constructing the inversion of
4695 * the CaseFolding (//i) property. If it were more general purpose for
4696 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4697 * because certain folds are prohibited under /iaa and /il. As an example,
4698 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4699 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4700 * prohibited, so we would not figure out that they fold to each other.
4701 * Code could be written to automatically figure this out, similar to the
4702 * code that does this for multi-character folds, but this is the only case
4703 * where something like this is ever likely to happen, as all the single
4704 * char folds to the 0-255 range are now quite settled. Instead there is a
4705 * little special code that is compiled only for this Unicode version. This
4706 * is smaller and didn't require much coding time to do. But this makes
4707 * this routine strongly tied to being used just for CaseFolding. If ever
4708 * it should be generalized, this would have to be fixed */
4712 HV *const hv = MUTABLE_HV(SvRV(swash));
4714 /* The string containing the main body of the table. This will have its
4715 * assertion fail if the swash has been converted to its inversion list */
4716 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4718 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4719 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4720 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4721 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4722 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4723 const STRLEN bits = SvUV(*bitssvp);
4724 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4725 const UV none = SvUV(*nonesvp);
4726 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4730 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4732 /* Must have at least 8 bits to get the mappings */
4733 if (bits != 8 && bits != 16 && bits != 32) {
4734 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"
4738 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4739 mapping to more than one character */
4741 /* Construct an inverse mapping hash for the specials */
4742 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4743 HV * specials_inverse = newHV();
4744 char *char_from; /* the lhs of the map */
4745 I32 from_len; /* its byte length */
4746 char *char_to; /* the rhs of the map */
4747 I32 to_len; /* its byte length */
4748 SV *sv_to; /* and in a sv */
4749 AV* from_list; /* list of things that map to each 'to' */
4751 hv_iterinit(specials_hv);
4753 /* The keys are the characters (in UTF-8) that map to the corresponding
4754 * UTF-8 string value. Iterate through the list creating the inverse
4756 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4758 if (! SvPOK(sv_to)) {
4759 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4760 "unexpectedly is not a string, flags=%lu",
4761 (unsigned long)SvFLAGS(sv_to));
4763 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %" UVXf ", First char of to is %" UVXf "\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
4765 /* Each key in the inverse list is a mapped-to value, and the key's
4766 * hash value is a list of the strings (each in UTF-8) that map to
4767 * it. Those strings are all one character long */
4768 if ((listp = hv_fetch(specials_inverse,
4772 from_list = (AV*) *listp;
4774 else { /* No entry yet for it: create one */
4775 from_list = newAV();
4776 if (! hv_store(specials_inverse,
4779 (SV*) from_list, 0))
4781 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4785 /* Here have the list associated with this 'to' (perhaps newly
4786 * created and empty). Just add to it. Note that we ASSUME that
4787 * the input is guaranteed to not have duplications, so we don't
4788 * check for that. Duplications just slow down execution time. */
4789 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4792 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4793 * it looking for cases like the FB05/FB06 examples above. There would
4794 * be an entry in the hash like
4795 * 'st' => [ FB05, FB06 ]
4796 * In this example we will create two lists that get stored in the
4797 * returned hash, 'ret':
4798 * FB05 => [ FB05, FB06 ]
4799 * FB06 => [ FB05, FB06 ]
4801 * Note that there is nothing to do if the array only has one element.
4802 * (In the normal 1-1 case handled below, we don't have to worry about
4803 * two lists, as everything gets tied to the single list that is
4804 * generated for the single character 'to'. But here, we are omitting
4805 * that list, ('st' in the example), so must have multiple lists.) */
4806 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4807 &char_to, &to_len)))
4809 if (av_tindex_skip_len_mg(from_list) > 0) {
4812 /* We iterate over all combinations of i,j to place each code
4813 * point on each list */
4814 for (i = 0; i <= av_tindex_skip_len_mg(from_list); i++) {
4816 AV* i_list = newAV();
4817 SV** entryp = av_fetch(from_list, i, FALSE);
4818 if (entryp == NULL) {
4819 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly"
4822 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4823 Perl_croak(aTHX_ "panic: unexpected entry for %s",
4826 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4827 (SV*) i_list, FALSE))
4829 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4832 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4833 for (j = 0; j <= av_tindex_skip_len_mg(from_list); j++) {
4834 entryp = av_fetch(from_list, j, FALSE);
4835 if (entryp == NULL) {
4836 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4839 /* When i==j this adds itself to the list */
4840 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4841 (U8*) SvPVX(*entryp),
4842 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4844 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
4849 SvREFCNT_dec(specials_inverse); /* done with it */
4850 } /* End of specials */
4852 /* read $swash->{LIST} */
4854 #if UNICODE_MAJOR_VERSION == 3 \
4855 && UNICODE_DOT_VERSION == 0 \
4856 && UNICODE_DOT_DOT_VERSION == 1
4858 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4859 * rule so that things work under /iaa and /il */
4861 SV * mod_listsv = sv_mortalcopy(*listsvp);
4862 sv_catpv(mod_listsv, "130\t130\t131\n");
4863 l = (U8*)SvPV(mod_listsv, lcur);
4867 l = (U8*)SvPV(*listsvp, lcur);
4873 /* Go through each input line */
4877 l = swash_scan_list_line(l, lend, &min, &max, &val,
4878 cBOOL(octets), typestr);
4883 /* Each element in the range is to be inverted */
4884 for (inverse = min; inverse <= max; inverse++) {
4888 bool found_key = FALSE;
4889 bool found_inverse = FALSE;
4891 /* The key is the inverse mapping */
4892 char key[UTF8_MAXBYTES+1];
4893 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4894 STRLEN key_len = key_end - key;
4896 /* Get the list for the map */
4897 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4898 list = (AV*) *listp;
4900 else { /* No entry yet for it: create one */
4902 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4903 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4907 /* Look through list to see if this inverse mapping already is
4908 * listed, or if there is a mapping to itself already */
4909 for (i = 0; i <= av_tindex_skip_len_mg(list); i++) {
4910 SV** entryp = av_fetch(list, i, FALSE);
4913 if (entryp == NULL) {
4914 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4918 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4922 if (uv == inverse) {
4923 found_inverse = TRUE;
4926 /* No need to continue searching if found everything we are
4928 if (found_key && found_inverse) {
4933 /* Make sure there is a mapping to itself on the list */
4935 av_push(list, newSVuv(val));
4936 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4940 /* Simply add the value to the list */
4941 if (! found_inverse) {
4942 av_push(list, newSVuv(inverse));
4943 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4946 /* swatch_get() increments the value of val for each element in the
4947 * range. That makes more compact tables possible. You can
4948 * express the capitalization, for example, of all consecutive
4949 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4950 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4951 * and it's not documented; it appears to be used only in
4952 * implementing tr//; I copied the semantics from swatch_get(), just
4954 if (!none || val < none) {
4964 Perl__swash_to_invlist(pTHX_ SV* const swash)
4967 /* Subject to change or removal. For use only in one place in regcomp.c.
4968 * Ownership is given to one reference count in the returned SV* */
4973 HV *const hv = MUTABLE_HV(SvRV(swash));
4974 UV elements = 0; /* Number of elements in the inversion list */
4984 STRLEN octets; /* if bits == 1, then octets == 0 */
4990 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4992 /* If not a hash, it must be the swash's inversion list instead */
4993 if (SvTYPE(hv) != SVt_PVHV) {
4994 return SvREFCNT_inc_simple_NN((SV*) hv);
4997 /* The string containing the main body of the table */
4998 listsvp = hv_fetchs(hv, "LIST", FALSE);
4999 typesvp = hv_fetchs(hv, "TYPE", FALSE);
5000 bitssvp = hv_fetchs(hv, "BITS", FALSE);
5001 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
5002 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
5004 typestr = (U8*)SvPV_nolen(*typesvp);
5005 bits = SvUV(*bitssvp);
5006 octets = bits >> 3; /* if bits == 1, then octets == 0 */
5008 /* read $swash->{LIST} */
5009 if (SvPOK(*listsvp)) {
5010 l = (U8*)SvPV(*listsvp, lcur);
5013 /* LIST legitimately doesn't contain a string during compilation phases
5014 * of Perl itself, before the Unicode tables are generated. In this
5015 * case, just fake things up by creating an empty list */
5022 if (*l == 'V') { /* Inversion list format */
5023 const char *after_atou = (char *) lend;
5025 UV* other_elements_ptr;
5027 /* The first number is a count of the rest */
5029 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
5030 Perl_croak(aTHX_ "panic: Expecting a valid count of elements"
5031 " at start of inversion list");
5033 if (elements == 0) {
5034 invlist = _new_invlist(0);
5037 l = (U8 *) after_atou;
5039 /* Get the 0th element, which is needed to setup the inversion list
5041 while (isSPACE(*l)) l++;
5042 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
5043 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for"
5046 l = (U8 *) after_atou;
5047 invlist = _setup_canned_invlist(elements, element0,
5048 &other_elements_ptr);
5051 /* Then just populate the rest of the input */
5052 while (elements-- > 0) {
5054 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more"
5055 " elements than available", elements);
5057 while (isSPACE(*l)) l++;
5058 if (!grok_atoUV((const char *)l, other_elements_ptr++,
5061 Perl_croak(aTHX_ "panic: Expecting a valid element"
5062 " in inversion list");
5064 l = (U8 *) after_atou;
5070 /* Scan the input to count the number of lines to preallocate array
5071 * size based on worst possible case, which is each line in the input
5072 * creates 2 elements in the inversion list: 1) the beginning of a
5073 * range in the list; 2) the beginning of a range not in the list. */
5074 while ((loc = (strchr(loc, '\n'))) != NULL) {
5079 /* If the ending is somehow corrupt and isn't a new line, add another
5080 * element for the final range that isn't in the inversion list */
5081 if (! (*lend == '\n'
5082 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
5087 invlist = _new_invlist(elements);
5089 /* Now go through the input again, adding each range to the list */
5092 UV val; /* Not used by this function */
5094 l = swash_scan_list_line(l, lend, &start, &end, &val,
5095 cBOOL(octets), typestr);
5101 invlist = _add_range_to_invlist(invlist, start, end);
5105 /* Invert if the data says it should be */
5106 if (invert_it_svp && SvUV(*invert_it_svp)) {
5107 _invlist_invert(invlist);
5110 /* This code is copied from swatch_get()
5111 * read $swash->{EXTRAS} */
5112 x = (U8*)SvPV(*extssvp, xcur);
5120 SV **otherbitssvp, *other;
5123 const U8 opc = *x++;
5127 nl = (U8*)memchr(x, '\n', xend - x);
5129 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
5131 x = nl + 1; /* 1 is length of "\n" */
5135 x = xend; /* to EXTRAS' end at which \n is not found */
5142 namelen = nl - namestr;
5146 namelen = xend - namestr;
5150 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
5151 otherhv = MUTABLE_HV(SvRV(*othersvp));
5152 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
5153 otherbits = (STRLEN)SvUV(*otherbitssvp);
5155 if (bits != otherbits || bits != 1) {
5156 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
5157 "properties, bits=%" UVuf ", otherbits=%" UVuf,
5158 (UV)bits, (UV)otherbits);
5161 /* The "other" swatch must be destroyed after. */
5162 other = _swash_to_invlist((SV *)*othersvp);
5164 /* End of code copied from swatch_get() */
5167 _invlist_union(invlist, other, &invlist);
5170 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
5173 _invlist_subtract(invlist, other, &invlist);
5176 _invlist_intersection(invlist, other, &invlist);
5181 sv_free(other); /* through with it! */
5184 SvREADONLY_on(invlist);
5189 Perl__get_swash_invlist(pTHX_ SV* const swash)
5193 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
5195 if (! SvROK(swash)) {
5199 /* If it really isn't a hash, it isn't really swash; must be an inversion
5201 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
5205 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
5214 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
5216 /* May change: warns if surrogates, non-character code points, or
5217 * non-Unicode code points are in 's' which has length 'len' bytes.
5218 * Returns TRUE if none found; FALSE otherwise. The only other validity
5219 * check is to make sure that this won't exceed the string's length nor
5222 const U8* const e = s + len;
5225 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
5228 if (UTF8SKIP(s) > len) {
5229 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
5230 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
5233 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
5234 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
5235 if ( ckWARN_d(WARN_NON_UNICODE)
5236 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
5237 0 /* Don't consider overlongs */
5240 /* A side effect of this function will be to warn */
5241 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
5245 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
5246 if (ckWARN_d(WARN_SURROGATE)) {
5247 /* This has a different warning than the one the called
5248 * function would output, so can't just call it, unlike we
5249 * do for the non-chars and above-unicodes */
5250 UV uv = utf8_to_uvchr_buf(s, e, NULL);
5251 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
5252 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
5257 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
5258 && (ckWARN_d(WARN_NONCHAR)))
5260 /* A side effect of this function will be to warn */
5261 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5272 =for apidoc pv_uni_display
5274 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5275 length C<len>, the displayable version being at most C<pvlim> bytes long
5276 (if longer, the rest is truncated and C<"..."> will be appended).
5278 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5279 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5280 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5281 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5282 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5283 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5285 The pointer to the PV of the C<dsv> is returned.
5287 See also L</sv_uni_display>.
5291 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
5297 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5301 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5303 /* This serves double duty as a flag and a character to print after
5304 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5308 if (pvlim && SvCUR(dsv) >= pvlim) {
5312 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5314 const unsigned char c = (unsigned char)u & 0xFF;
5315 if (flags & UNI_DISPLAY_BACKSLASH) {
5332 const char string = ok;
5333 sv_catpvs(dsv, "\\");
5334 sv_catpvn(dsv, &string, 1);
5337 /* isPRINT() is the locale-blind version. */
5338 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5339 const char string = c;
5340 sv_catpvn(dsv, &string, 1);
5345 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5348 sv_catpvs(dsv, "...");
5354 =for apidoc sv_uni_display
5356 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5357 the displayable version being at most C<pvlim> bytes long
5358 (if longer, the rest is truncated and "..." will be appended).
5360 The C<flags> argument is as in L</pv_uni_display>().
5362 The pointer to the PV of the C<dsv> is returned.
5367 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5369 const char * const ptr =
5370 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5372 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5374 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5375 SvCUR(ssv), pvlim, flags);
5379 =for apidoc foldEQ_utf8
5381 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
5382 both of which may be in UTF-8) are the same case-insensitively; false
5383 otherwise. How far into the strings to compare is determined by other input
5386 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5387 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
5388 C<u2> with respect to C<s2>.
5390 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
5391 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
5392 The scan will not be considered to be a match unless the goal is reached, and
5393 scanning won't continue past that goal. Correspondingly for C<l2> with respect
5396 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
5397 pointer is considered an end pointer to the position 1 byte past the maximum
5398 point in C<s1> beyond which scanning will not continue under any circumstances.
5399 (This routine assumes that UTF-8 encoded input strings are not malformed;
5400 malformed input can cause it to read past C<pe1>). This means that if both
5401 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
5402 will never be successful because it can never
5403 get as far as its goal (and in fact is asserted against). Correspondingly for
5404 C<pe2> with respect to C<s2>.
5406 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5407 C<l2> must be non-zero), and if both do, both have to be
5408 reached for a successful match. Also, if the fold of a character is multiple
5409 characters, all of them must be matched (see tr21 reference below for
5412 Upon a successful match, if C<pe1> is non-C<NULL>,
5413 it will be set to point to the beginning of the I<next> character of C<s1>
5414 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5416 For case-insensitiveness, the "casefolding" of Unicode is used
5417 instead of upper/lowercasing both the characters, see
5418 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5422 /* A flags parameter has been added which may change, and hence isn't
5423 * externally documented. Currently it is:
5424 * 0 for as-documented above
5425 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5426 ASCII one, to not match
5427 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5428 * locale are to be used.
5429 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5430 * routine. This allows that step to be skipped.
5431 * Currently, this requires s1 to be encoded as UTF-8
5432 * (u1 must be true), which is asserted for.
5433 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5434 * cross certain boundaries. Hence, the caller should
5435 * let this function do the folding instead of
5436 * pre-folding. This code contains an assertion to
5437 * that effect. However, if the caller knows what
5438 * it's doing, it can pass this flag to indicate that,
5439 * and the assertion is skipped.
5440 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5441 * FOLDEQ_S2_FOLDS_SANE
5444 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
5445 const char *s2, char **pe2, UV l2, bool u2,
5448 const U8 *p1 = (const U8*)s1; /* Point to current char */
5449 const U8 *p2 = (const U8*)s2;
5450 const U8 *g1 = NULL; /* goal for s1 */
5451 const U8 *g2 = NULL;
5452 const U8 *e1 = NULL; /* Don't scan s1 past this */
5453 U8 *f1 = NULL; /* Point to current folded */
5454 const U8 *e2 = NULL;
5456 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5457 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5458 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5459 U8 flags_for_folder = FOLD_FLAGS_FULL;
5461 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5463 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5464 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5465 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5466 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5467 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5468 /* The algorithm is to trial the folds without regard to the flags on
5469 * the first line of the above assert(), and then see if the result
5470 * violates them. This means that the inputs can't be pre-folded to a
5471 * violating result, hence the assert. This could be changed, with the
5472 * addition of extra tests here for the already-folded case, which would
5473 * slow it down. That cost is more than any possible gain for when these
5474 * flags are specified, as the flags indicate /il or /iaa matching which
5475 * is less common than /iu, and I (khw) also believe that real-world /il
5476 * and /iaa matches are most likely to involve code points 0-255, and this
5477 * function only under rare conditions gets called for 0-255. */
5479 if (flags & FOLDEQ_LOCALE) {
5480 if (IN_UTF8_CTYPE_LOCALE) {
5481 flags &= ~FOLDEQ_LOCALE;
5484 flags_for_folder |= FOLD_FLAGS_LOCALE;
5493 g1 = (const U8*)s1 + l1;
5501 g2 = (const U8*)s2 + l2;
5504 /* Must have at least one goal */
5509 /* Will never match if goal is out-of-bounds */
5510 assert(! e1 || e1 >= g1);
5512 /* Here, there isn't an end pointer, or it is beyond the goal. We
5513 * only go as far as the goal */
5517 assert(e1); /* Must have an end for looking at s1 */
5520 /* Same for goal for s2 */
5522 assert(! e2 || e2 >= g2);
5529 /* If both operands are already folded, we could just do a memEQ on the
5530 * whole strings at once, but it would be better if the caller realized
5531 * this and didn't even call us */
5533 /* Look through both strings, a character at a time */
5534 while (p1 < e1 && p2 < e2) {
5536 /* If at the beginning of a new character in s1, get its fold to use
5537 * and the length of the fold. */
5539 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5545 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5547 /* We have to forbid mixing ASCII with non-ASCII if the
5548 * flags so indicate. And, we can short circuit having to
5549 * call the general functions for this common ASCII case,
5550 * all of whose non-locale folds are also ASCII, and hence
5551 * UTF-8 invariants, so the UTF8ness of the strings is not
5553 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5557 *foldbuf1 = toFOLD(*p1);
5560 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5562 else { /* Not UTF-8, get UTF-8 fold */
5563 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5569 if (n2 == 0) { /* Same for s2 */
5570 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5576 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5577 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5581 *foldbuf2 = toFOLD(*p2);
5584 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5587 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5593 /* Here f1 and f2 point to the beginning of the strings to compare.
5594 * These strings are the folds of the next character from each input
5595 * string, stored in UTF-8. */
5597 /* While there is more to look for in both folds, see if they
5598 * continue to match */
5600 U8 fold_length = UTF8SKIP(f1);
5601 if (fold_length != UTF8SKIP(f2)
5602 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5603 function call for single
5605 || memNE((char*)f1, (char*)f2, fold_length))
5607 return 0; /* mismatch */
5610 /* Here, they matched, advance past them */
5617 /* When reach the end of any fold, advance the input past it */
5619 p1 += u1 ? UTF8SKIP(p1) : 1;
5622 p2 += u2 ? UTF8SKIP(p2) : 1;
5624 } /* End of loop through both strings */
5626 /* A match is defined by each scan that specified an explicit length
5627 * reaching its final goal, and the other not having matched a partial
5628 * character (which can happen when the fold of a character is more than one
5630 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5634 /* Successful match. Set output pointers */
5644 /* XXX The next two functions should likely be moved to mathoms.c once all
5645 * occurrences of them are removed from the core; some cpan-upstream modules
5649 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5651 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5653 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5657 =for apidoc utf8n_to_uvuni
5659 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5661 This function was useful for code that wanted to handle both EBCDIC and
5662 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5663 distinctions between the platforms have mostly been made invisible to most
5664 code, so this function is quite unlikely to be what you want. If you do need
5665 this precise functionality, use instead
5666 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5667 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5673 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5675 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5677 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5681 =for apidoc uvuni_to_utf8_flags
5683 Instead you almost certainly want to use L</uvchr_to_utf8> or
5684 L</uvchr_to_utf8_flags>.
5686 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5687 which itself, while not deprecated, should be used only in isolated
5688 circumstances. These functions were useful for code that wanted to handle
5689 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5690 v5.20, the distinctions between the platforms have mostly been made invisible
5691 to most code, so this function is quite unlikely to be what you want.
5697 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5699 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5701 return uvoffuni_to_utf8_flags(d, uv, flags);
5705 * ex: set ts=8 sts=4 sw=4 et: