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)";
40 /* Be sure to synchronize this message with the similar one in regcomp.c */
41 static const char cp_above_legal_max[] =
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
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 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
106 U32 categories, /* Packed warning categories */
107 U32 flag) /* Flag associated with this message */
109 /* Creates, populates, and returns an HV* that describes an error message
110 * for the translators between UTF8 and code point */
112 SV* msg_sv = newSVpv(message, 0);
113 SV* category_sv = newSVuv(categories);
114 SV* flag_bit_sv = newSVuv(flag);
116 HV* msg_hv = newHV();
118 PERL_ARGS_ASSERT_NEW_MSG_HV;
120 (void) hv_stores(msg_hv, "text", msg_sv);
121 (void) hv_stores(msg_hv, "warn_categories", category_sv);
122 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
128 =for apidoc uvoffuni_to_utf8_flags
130 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
131 Instead, B<Almost all code should use L</uvchr_to_utf8> or
132 L</uvchr_to_utf8_flags>>.
134 This function is like them, but the input is a strict Unicode
135 (as opposed to native) code point. Only in very rare circumstances should code
136 not be using the native code point.
138 For details, see the description for L</uvchr_to_utf8_flags>.
144 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
146 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
148 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
151 /* All these formats take a single UV code point argument */
152 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
153 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
154 " is not recommended for open interchange";
155 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
156 " may not be portable";
157 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
158 " Unicode, requires a Perl extension," \
159 " and so is not portable";
161 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
163 if (flags & UNICODE_WARN_SURROGATE) { \
164 U32 category = packWARN(WARN_SURROGATE); \
165 const char * format = surrogate_cp_format; \
167 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
169 UNICODE_GOT_SURROGATE); \
172 Perl_ck_warner_d(aTHX_ category, format, uv); \
175 if (flags & UNICODE_DISALLOW_SURROGATE) { \
180 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
182 if (flags & UNICODE_WARN_NONCHAR) { \
183 U32 category = packWARN(WARN_NONCHAR); \
184 const char * format = nonchar_cp_format; \
186 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
188 UNICODE_GOT_NONCHAR); \
191 Perl_ck_warner_d(aTHX_ category, format, uv); \
194 if (flags & UNICODE_DISALLOW_NONCHAR) { \
199 /* Use shorter names internally in this file */
200 #define SHIFT UTF_ACCUMULATION_SHIFT
202 #define MARK UTF_CONTINUATION_MARK
203 #define MASK UTF_CONTINUATION_MASK
206 =for apidoc uvchr_to_utf8_flags_msgs
208 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
210 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
212 This function is for code that wants any warning and/or error messages to be
213 returned to the caller rather than be displayed. All messages that would have
214 been displayed if all lexical warnings are enabled will be returned.
216 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
217 placed after all the others, C<msgs>. If this parameter is 0, this function
218 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
219 be a pointer to an C<HV *> variable, in which this function creates a new HV to
220 contain any appropriate messages. The hash has three key-value pairs, as
227 The text of the message as a C<SVpv>.
229 =item C<warn_categories>
231 The warning category (or categories) packed into a C<SVuv>.
235 A single flag bit associated with this message, in a C<SVuv>.
236 The bit corresponds to some bit in the C<*errors> return value,
237 such as C<UNICODE_GOT_SURROGATE>.
241 It's important to note that specifying this parameter as non-null will cause
242 any warnings this function would otherwise generate to be suppressed, and
243 instead be placed in C<*msgs>. The caller can check the lexical warnings state
244 (or not) when choosing what to do with the returned messages.
246 The caller, of course, is responsible for freeing any returned HV.
251 /* Undocumented; we don't want people using this. Instead they should use
252 * uvchr_to_utf8_flags_msgs() */
254 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
256 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
262 if (OFFUNI_IS_INVARIANT(uv)) {
263 *d++ = LATIN1_TO_NATIVE(uv);
267 if (uv <= MAX_UTF8_TWO_BYTE) {
268 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
269 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
273 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
274 * below, the 16 is for start bytes E0-EF (which are all the possible ones
275 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
276 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
277 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
278 * 0x800-0xFFFF on ASCII */
279 if (uv < (16 * (1U << (2 * SHIFT)))) {
280 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
281 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
282 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
284 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
285 aren't tested here */
286 /* The most likely code points in this range are below the surrogates.
287 * Do an extra test to quickly exclude those. */
288 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
289 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
290 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
292 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
294 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
295 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
302 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
303 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
304 * happen starting with 4-byte characters on ASCII platforms. We unify the
305 * code for these with EBCDIC, even though some of them require 5-bytes on
306 * those, because khw believes the code saving is worth the very slight
307 * performance hit on these high EBCDIC code points. */
309 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
310 if (UNLIKELY(uv > MAX_LEGAL_CP)) {
311 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP);
313 if ( (flags & UNICODE_WARN_SUPER)
314 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
315 && UNICODE_IS_PERL_EXTENDED(uv)))
317 const char * format = super_cp_format;
318 U32 category = packWARN(WARN_NON_UNICODE);
319 U32 flag = UNICODE_GOT_SUPER;
321 /* Choose the more dire applicable warning */
322 if (UNICODE_IS_PERL_EXTENDED(uv)) {
323 format = perl_extended_cp_format;
324 if (flags & (UNICODE_WARN_PERL_EXTENDED
325 |UNICODE_DISALLOW_PERL_EXTENDED))
327 flag = UNICODE_GOT_PERL_EXTENDED;
332 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
336 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
339 if ( (flags & UNICODE_DISALLOW_SUPER)
340 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
341 && UNICODE_IS_PERL_EXTENDED(uv)))
346 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
347 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
350 /* Test for and handle 4-byte result. In the test immediately below, the
351 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
352 * characters). The 3 is for 3 continuation bytes; these each contribute
353 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
354 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
355 * 0x1_0000-0x1F_FFFF on ASCII */
356 if (uv < (8 * (1U << (3 * SHIFT)))) {
357 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
358 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
359 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
360 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
362 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
363 characters. The end-plane non-characters for EBCDIC were
364 handled just above */
365 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
366 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
368 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
369 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
376 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
377 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
378 * format. The unrolled version above turns out to not save all that much
379 * time, and at these high code points (well above the legal Unicode range
380 * on ASCII platforms, and well above anything in common use in EBCDIC),
381 * khw believes that less code outweighs slight performance gains. */
384 STRLEN len = OFFUNISKIP(uv);
387 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
390 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
396 =for apidoc uvchr_to_utf8
398 Adds the UTF-8 representation of the native code point C<uv> to the end
399 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
400 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
401 the byte after the end of the new character. In other words,
403 d = uvchr_to_utf8(d, uv);
405 is the recommended wide native character-aware way of saying
409 This function accepts any code point from 0..C<IV_MAX> as input.
410 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
412 It is possible to forbid or warn on non-Unicode code points, or those that may
413 be problematic by using L</uvchr_to_utf8_flags>.
418 /* This is also a macro */
419 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
422 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
424 return uvchr_to_utf8(d, uv);
428 =for apidoc uvchr_to_utf8_flags
430 Adds the UTF-8 representation of the native code point C<uv> to the end
431 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
432 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
433 the byte after the end of the new character. In other words,
435 d = uvchr_to_utf8_flags(d, uv, flags);
439 d = uvchr_to_utf8_flags(d, uv, 0);
441 This is the Unicode-aware way of saying
445 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
446 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
448 Specifying C<flags> can further restrict what is allowed and not warned on, as
451 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
452 the function will raise a warning, provided UTF8 warnings are enabled. If
453 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
454 NULL. If both flags are set, the function will both warn and return NULL.
456 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
457 affect how the function handles a Unicode non-character.
459 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
460 affect the handling of code points that are above the Unicode maximum of
461 0x10FFFF. Languages other than Perl may not be able to accept files that
464 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
465 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
466 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
467 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
468 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
469 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
470 above-Unicode and surrogate flags, but not the non-character ones, as
472 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
473 See L<perlunicode/Noncharacter code points>.
475 Extremely high code points were never specified in any standard, and require an
476 extension to UTF-8 to express, which Perl does. It is likely that programs
477 written in something other than Perl would not be able to read files that
478 contain these; nor would Perl understand files written by something that uses a
479 different extension. For these reasons, there is a separate set of flags that
480 can warn and/or disallow these extremely high code points, even if other
481 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
482 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
483 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
484 treat all above-Unicode code points, including these, as malformations. (Note
485 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
486 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
488 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
489 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
490 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
491 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
492 platforms,these flags can apply to code points that actually do fit in 31 bits.
493 The new names accurately describe the situation in all cases.
498 /* This is also a macro */
499 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
502 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
504 return uvchr_to_utf8_flags(d, uv, flags);
510 S_is_utf8_cp_above_31_bits(const U8 * const s,
512 const bool consider_overlongs)
514 /* Returns TRUE if the first code point represented by the Perl-extended-
515 * UTF-8-encoded string starting at 's', and looking no further than 'e -
516 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
518 * The function handles the case where the input bytes do not include all
519 * the ones necessary to represent a full character. That is, they may be
520 * the intial bytes of the representation of a code point, but possibly
521 * the final ones necessary for the complete representation may be beyond
524 * The function also can handle the case where the input is an overlong
525 * sequence. If 'consider_overlongs' is 0, the function assumes the
526 * input is not overlong, without checking, and will return based on that
527 * assumption. If this parameter is 1, the function will go to the trouble
528 * of figuring out if it actually evaluates to above or below 31 bits.
530 * The sequence is otherwise assumed to be well-formed, without checking.
533 const STRLEN len = e - s;
536 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
538 assert(! UTF8_IS_INVARIANT(*s) && e > s);
542 PERL_UNUSED_ARG(consider_overlongs);
544 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
545 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
546 * also be the start byte for a 31-bit code point; we need at least 2
547 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
548 * the start byte for an overlong sequence, but for 30-bit or smaller code
549 * points, so we don't have to worry about overlongs on EBCDIC.) */
560 /* On ASCII, FE and FF are the only start bytes that can evaluate to
561 * needing more than 31 bits. */
562 if (LIKELY(*s < 0xFE)) {
566 /* What we have left are FE and FF. Both of these require more than 31
567 * bits unless they are for overlongs. */
568 if (! consider_overlongs) {
572 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
573 * above 31 bits. But we need more than one byte to discern this, so if
574 * passed just the start byte, it could be an overlong evaluating to
580 /* Having excluded len==1, and knowing that FE and FF are both valid start
581 * bytes, we can call the function below to see if the sequence is
582 * overlong. (We don't need the full generality of the called function,
583 * but for these huge code points, speed shouldn't be a consideration, and
584 * the compiler does have enough information, since it's static to this
585 * file, to optimize to just the needed parts.) */
586 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
588 /* If it isn't overlong, more than 31 bits are required. */
589 if (is_overlong == 0) {
593 /* If it is indeterminate if it is overlong, return that */
594 if (is_overlong < 0) {
598 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
599 * the max it can be is 2**31 - 1 */
606 /* Here, ASCII and EBCDIC rejoin:
607 * On ASCII: We have an overlong sequence starting with FF
608 * On EBCDIC: We have a sequence starting with FE. */
610 { /* For C89, use a block so the declaration can be close to its use */
614 /* U+7FFFFFFF (2 ** 31 - 1)
615 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
616 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
617 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
618 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
619 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
620 * U+80000000 (2 ** 31):
621 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
622 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
623 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
624 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
626 * and since we know that *s = \xfe, any continuation sequcence
627 * following it that is gt the below is above 31 bits
628 [0] [1] [2] [3] [4] [5] [6] */
629 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
633 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
634 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
635 * FF overlong for U+80000000 (2 ** 31):
636 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
637 * and since we know that *s = \xff, any continuation sequcence
638 * following it that is gt the below is above 30 bits
639 [0] [1] [2] [3] [4] [5] [6] */
640 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
644 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
645 const STRLEN cmp_len = MIN(conts_len, len - 1);
647 /* Now compare the continuation bytes in s with the ones we have
648 * compiled in that are for the largest 30 bit code point. If we have
649 * enough bytes available to determine the answer, or the bytes we do
650 * have differ from them, we can compare the two to get a definitive
651 * answer (Note that in UTF-EBCDIC, the two lowest possible
652 * continuation bytes are \x41 and \x42.) */
653 if (cmp_len >= conts_len || memNE(s + 1,
654 conts_for_highest_30_bit,
657 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
660 /* Here, all the bytes we have are the same as the highest 30-bit code
661 * point, but we are missing so many bytes that we can't make the
669 PERL_STATIC_INLINE int
670 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
672 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
673 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
674 * it isn't, and -1 if there isn't enough information to tell. This last
675 * return value can happen if the sequence is incomplete, missing some
676 * trailing bytes that would form a complete character. If there are
677 * enough bytes to make a definitive decision, this function does so.
678 * Usually 2 bytes sufficient.
680 * Overlongs can occur whenever the number of continuation bytes changes.
681 * That means whenever the number of leading 1 bits in a start byte
682 * increases from the next lower start byte. That happens for start bytes
683 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
684 * start bytes have already been excluded, so don't need to be tested here;
685 * ASCII platforms: C0, C1
686 * EBCDIC platforms C0, C1, C2, C3, C4, E0
689 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
690 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
692 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
693 assert(len > 1 && UTF8_IS_START(*s));
695 /* Each platform has overlongs after the start bytes given above (expressed
696 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
697 * the logic is the same, except the E0 overlong has already been excluded
698 * on EBCDIC platforms. The values below were found by manually
699 * inspecting the UTF-8 patterns. See the tables in utf8.h and
703 # define F0_ABOVE_OVERLONG 0xB0
704 # define F8_ABOVE_OVERLONG 0xA8
705 # define FC_ABOVE_OVERLONG 0xA4
706 # define FE_ABOVE_OVERLONG 0xA2
707 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
711 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
715 # define F0_ABOVE_OVERLONG 0x90
716 # define F8_ABOVE_OVERLONG 0x88
717 # define FC_ABOVE_OVERLONG 0x84
718 # define FE_ABOVE_OVERLONG 0x82
719 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
723 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
724 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
725 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
726 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
731 /* Check for the FF overlong */
732 return isFF_OVERLONG(s, len);
735 PERL_STATIC_INLINE int
736 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
738 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
739 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
740 * it isn't, and -1 if there isn't enough information to tell. This last
741 * return value can happen if the sequence is incomplete, missing some
742 * trailing bytes that would form a complete character. If there are
743 * enough bytes to make a definitive decision, this function does so. */
745 PERL_ARGS_ASSERT_ISFF_OVERLONG;
747 /* To be an FF overlong, all the available bytes must match */
748 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
749 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
754 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
755 * be there; what comes after them doesn't matter. See tables in utf8.h,
757 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
761 /* The missing bytes could cause the result to go one way or the other, so
762 * the result is indeterminate */
766 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
767 # ifdef EBCDIC /* Actually is I8 */
768 # define HIGHEST_REPRESENTABLE_UTF8 \
769 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
771 # define HIGHEST_REPRESENTABLE_UTF8 \
772 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
776 PERL_STATIC_INLINE int
777 S_does_utf8_overflow(const U8 * const s,
779 const bool consider_overlongs)
781 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
782 * 'e' - 1 would overflow an IV on this platform; that is if it represents
783 * a code point larger than the highest representable code point. It
784 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
785 * enough information to tell. This last return value can happen if the
786 * sequence is incomplete, missing some trailing bytes that would form a
787 * complete character. If there are enough bytes to make a definitive
788 * decision, this function does so.
790 * If 'consider_overlongs' is TRUE, the function checks for the possibility
791 * that the sequence is an overlong that doesn't overflow. Otherwise, it
792 * assumes the sequence is not an overlong. This can give different
793 * results only on ASCII 32-bit platforms.
795 * (For ASCII platforms, we could use memcmp() because we don't have to
796 * convert each byte to I8, but it's very rare input indeed that would
797 * approach overflow, so the loop below will likely only get executed once.)
799 * 'e' - 1 must not be beyond a full character. */
802 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
803 assert(s <= e && s + UTF8SKIP(s) >= e);
805 #if ! defined(UV_IS_QUAD)
807 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
811 PERL_UNUSED_ARG(consider_overlongs);
814 const STRLEN len = e - s;
816 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
818 for (x = s; x < e; x++, y++) {
820 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
824 /* If this byte is larger than the corresponding highest UTF-8
825 * byte, the sequence overflow; otherwise the byte is less than,
826 * and so the sequence doesn't overflow */
827 return NATIVE_UTF8_TO_I8(*x) > *y;
831 /* Got to the end and all bytes are the same. If the input is a whole
832 * character, it doesn't overflow. And if it is a partial character,
833 * there's not enough information to tell */
834 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
847 /* This is the portions of the above function that deal with UV_MAX instead of
848 * IV_MAX. They are left here in case we want to combine them so that internal
849 * uses can have larger code points. The only logic difference is that the
850 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
854 /* Anything larger than this will overflow the word if it were converted into a UV */
855 #if defined(UV_IS_QUAD)
856 # ifdef EBCDIC /* Actually is I8 */
857 # define HIGHEST_REPRESENTABLE_UTF8 \
858 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
860 # define HIGHEST_REPRESENTABLE_UTF8 \
861 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
865 # define HIGHEST_REPRESENTABLE_UTF8 \
866 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
868 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
872 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
874 /* On 32 bit ASCII machines, many overlongs that start with FF don't
876 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
878 /* To be such an overlong, the first bytes of 's' must match
879 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
880 * don't have any additional bytes available, the sequence, when
881 * completed might or might not fit in 32 bits. But if we have that
882 * next byte, we can tell for sure. If it is <= 0x83, then it does
884 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
888 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
891 /* Starting with the #else, the rest of the function is identical except
892 * 1. we need to move the 'len' declaration to be global to the function
893 * 2. the endif move to just after the UNUSED_ARG.
894 * An empty endif is given just below to satisfy the preprocessor
900 #undef F0_ABOVE_OVERLONG
901 #undef F8_ABOVE_OVERLONG
902 #undef FC_ABOVE_OVERLONG
903 #undef FE_ABOVE_OVERLONG
904 #undef FF_OVERLONG_PREFIX
907 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
912 /* A helper function that should not be called directly.
914 * This function returns non-zero if the string beginning at 's' and
915 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
916 * code point; otherwise it returns 0. The examination stops after the
917 * first code point in 's' is validated, not looking at the rest of the
918 * input. If 'e' is such that there are not enough bytes to represent a
919 * complete code point, this function will return non-zero anyway, if the
920 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
921 * excluded by 'flags'.
923 * A non-zero return gives the number of bytes required to represent the
924 * code point. Be aware that if the input is for a partial character, the
925 * return will be larger than 'e - s'.
927 * This function assumes that the code point represented is UTF-8 variant.
928 * The caller should have excluded the possibility of it being invariant
929 * before calling this function.
931 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
932 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
933 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
934 * disallowed by the flags. If the input is only for a partial character,
935 * the function will return non-zero if there is any sequence of
936 * well-formed UTF-8 that, when appended to the input sequence, could
937 * result in an allowed code point; otherwise it returns 0. Non characters
938 * cannot be determined based on partial character input. But many of the
939 * other excluded types can be determined with just the first one or two
944 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
946 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
947 |UTF8_DISALLOW_PERL_EXTENDED)));
948 assert(! UTF8_IS_INVARIANT(*s));
950 /* A variant char must begin with a start byte */
951 if (UNLIKELY(! UTF8_IS_START(*s))) {
955 /* Examine a maximum of a single whole code point */
956 if (e - s > UTF8SKIP(s)) {
962 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
963 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
965 /* Here, we are disallowing some set of largish code points, and the
966 * first byte indicates the sequence is for a code point that could be
967 * in the excluded set. We generally don't have to look beyond this or
968 * the second byte to see if the sequence is actually for one of the
969 * excluded classes. The code below is derived from this table:
971 * UTF-8 UTF-EBCDIC I8
972 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
973 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
974 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
976 * Keep in mind that legal continuation bytes range between \x80..\xBF
977 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
978 * continuation bytes. Hence, we don't have to test the upper edge
979 * because if any of those is encountered, the sequence is malformed,
980 * and would fail elsewhere in this function.
982 * The code here likewise assumes that there aren't other
983 * malformations; again the function should fail elsewhere because of
984 * these. For example, an overlong beginning with FC doesn't actually
985 * have to be a super; it could actually represent a small code point,
986 * even U+0000. But, since overlongs (and other malformations) are
987 * illegal, the function should return FALSE in either case.
990 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
991 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
992 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
994 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
996 && ((s1) & 0xFE ) == 0xB6)
997 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
999 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1000 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1001 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1002 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1005 if ( (flags & UTF8_DISALLOW_SUPER)
1006 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1008 return 0; /* Above Unicode */
1011 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1012 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1018 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1020 if ( (flags & UTF8_DISALLOW_SUPER)
1021 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1023 return 0; /* Above Unicode */
1026 if ( (flags & UTF8_DISALLOW_SURROGATE)
1027 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1029 return 0; /* Surrogate */
1032 if ( (flags & UTF8_DISALLOW_NONCHAR)
1033 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1035 return 0; /* Noncharacter code point */
1040 /* Make sure that all that follows are continuation bytes */
1041 for (x = s + 1; x < e; x++) {
1042 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1047 /* Here is syntactically valid. Next, make sure this isn't the start of an
1049 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1053 /* And finally, that the code point represented fits in a word on this
1055 if (0 < does_utf8_overflow(s, e,
1056 0 /* Don't consider overlongs */
1066 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1068 /* Returns a mortalized C string that is a displayable copy of the 'len'
1069 * bytes starting at 'start'. 'format' gives how to display each byte.
1070 * Currently, there are only two formats, so it is currently a bool:
1072 * 1 ab (that is a space between two hex digit bytes)
1075 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1077 const U8 * s = start;
1078 const U8 * const e = start + len;
1082 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1084 Newx(output, output_len, char);
1088 for (s = start; s < e; s++) {
1089 const unsigned high_nibble = (*s & 0xF0) >> 4;
1090 const unsigned low_nibble = (*s & 0x0F);
1102 if (high_nibble < 10) {
1103 *d++ = high_nibble + '0';
1106 *d++ = high_nibble - 10 + 'a';
1109 if (low_nibble < 10) {
1110 *d++ = low_nibble + '0';
1113 *d++ = low_nibble - 10 + 'a';
1121 PERL_STATIC_INLINE char *
1122 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1124 /* Max number of bytes to print */
1127 /* Which one is the non-continuation */
1128 const STRLEN non_cont_byte_pos,
1130 /* How many bytes should there be? */
1131 const STRLEN expect_len)
1133 /* Return the malformation warning text for an unexpected continuation
1136 const char * const where = (non_cont_byte_pos == 1)
1138 : Perl_form(aTHX_ "%d bytes",
1139 (int) non_cont_byte_pos);
1140 const U8 * x = s + non_cont_byte_pos;
1141 const U8 * e = s + print_len;
1143 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1145 /* We don't need to pass this parameter, but since it has already been
1146 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1147 assert(expect_len == UTF8SKIP(s));
1149 /* As a defensive coding measure, don't output anything past a NUL. Such
1150 * bytes shouldn't be in the middle of a malformation, and could mark the
1151 * end of the allocated string, and what comes after is undefined */
1152 for (; x < e; x++) {
1154 x++; /* Output this particular NUL */
1159 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1160 " %s after start byte 0x%02x; need %d bytes, got %d)",
1162 _byte_dump_string(s, x - s, 0),
1163 *(s + non_cont_byte_pos),
1167 (int) non_cont_byte_pos);
1172 =for apidoc utf8n_to_uvchr
1174 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1175 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1177 Bottom level UTF-8 decode routine.
1178 Returns the native code point value of the first character in the string C<s>,
1179 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1180 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1181 the length, in bytes, of that character.
1183 The value of C<flags> determines the behavior when C<s> does not point to a
1184 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1185 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1186 is the next possible position in C<s> that could begin a non-malformed
1187 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1188 is raised. Some UTF-8 input sequences may contain multiple malformations.
1189 This function tries to find every possible one in each call, so multiple
1190 warnings can be raised for the same sequence.
1192 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1193 individual types of malformations, such as the sequence being overlong (that
1194 is, when there is a shorter sequence that can express the same code point;
1195 overlong sequences are expressly forbidden in the UTF-8 standard due to
1196 potential security issues). Another malformation example is the first byte of
1197 a character not being a legal first byte. See F<utf8.h> for the list of such
1198 flags. Even if allowed, this function generally returns the Unicode
1199 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1200 F<utf8.h> to override this behavior for the overlong malformations, but don't
1201 do that except for very specialized purposes.
1203 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1204 flags) malformation is found. If this flag is set, the routine assumes that
1205 the caller will raise a warning, and this function will silently just set
1206 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1208 Note that this API requires disambiguation between successful decoding a C<NUL>
1209 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1210 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1211 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1212 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1213 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1215 Certain code points are considered problematic. These are Unicode surrogates,
1216 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1217 By default these are considered regular code points, but certain situations
1218 warrant special handling for them, which can be specified using the C<flags>
1219 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1220 three classes are treated as malformations and handled as such. The flags
1221 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1222 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1223 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1224 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1225 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1227 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1228 The difference between traditional strictness and C9 strictness is that the
1229 latter does not forbid non-character code points. (They are still discouraged,
1230 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1232 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1233 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1234 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1235 raised for their respective categories, but otherwise the code points are
1236 considered valid (not malformations). To get a category to both be treated as
1237 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1238 (But note that warnings are not raised if lexically disabled nor if
1239 C<UTF8_CHECK_ONLY> is also specified.)
1241 Extremely high code points were never specified in any standard, and require an
1242 extension to UTF-8 to express, which Perl does. It is likely that programs
1243 written in something other than Perl would not be able to read files that
1244 contain these; nor would Perl understand files written by something that uses a
1245 different extension. For these reasons, there is a separate set of flags that
1246 can warn and/or disallow these extremely high code points, even if other
1247 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1248 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1249 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1250 above-Unicode code points, including these, as malformations.
1251 (Note that the Unicode standard considers anything above 0x10FFFF to be
1252 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1255 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1256 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1257 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1258 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1259 can apply to code points that actually do fit in 31 bits. This happens on
1260 EBCDIC platforms, and sometimes when the L<overlong
1261 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1262 describe the situation in all cases.
1265 All other code points corresponding to Unicode characters, including private
1266 use and those yet to be assigned, are never considered malformed and never
1271 Also implemented as a macro in utf8.h
1275 Perl_utf8n_to_uvchr(const U8 *s,
1280 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1282 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1287 =for apidoc utf8n_to_uvchr_error
1289 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1290 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1292 This function is for code that needs to know what the precise malformation(s)
1293 are when an error is found. If you also need to know the generated warning
1294 messages, use L</utf8n_to_uvchr_msgs>() instead.
1296 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1297 all the others, C<errors>. If this parameter is 0, this function behaves
1298 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1299 to a C<U32> variable, which this function sets to indicate any errors found.
1300 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1301 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1302 of these bits will be set if a malformation is found, even if the input
1303 C<flags> parameter indicates that the given malformation is allowed; those
1304 exceptions are noted:
1308 =item C<UTF8_GOT_PERL_EXTENDED>
1310 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1311 set only if the input C<flags> parameter contains either the
1312 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1314 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1315 and so some extension must be used to express them. Perl uses a natural
1316 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1317 extension to represent even higher ones, so that any code point that fits in a
1318 64-bit word can be represented. Text using these extensions is not likely to
1319 be portable to non-Perl code. We lump both of these extensions together and
1320 refer to them as Perl extended UTF-8. There exist other extensions that people
1321 have invented, incompatible with Perl's.
1323 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1324 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1325 than on ASCII. Prior to that, code points 2**31 and higher were simply
1326 unrepresentable, and a different, incompatible method was used to represent
1327 code points between 2**30 and 2**31 - 1.
1329 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1330 Perl extended UTF-8 is used.
1332 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1333 may use for backward compatibility. That name is misleading, as this flag may
1334 be set when the code point actually does fit in 31 bits. This happens on
1335 EBCDIC platforms, and sometimes when the L<overlong
1336 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1337 describes the situation in all cases.
1339 =item C<UTF8_GOT_CONTINUATION>
1341 The input sequence was malformed in that the first byte was a a UTF-8
1344 =item C<UTF8_GOT_EMPTY>
1346 The input C<curlen> parameter was 0.
1348 =item C<UTF8_GOT_LONG>
1350 The input sequence was malformed in that there is some other sequence that
1351 evaluates to the same code point, but that sequence is shorter than this one.
1353 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1354 it was discovered that this created security issues.
1356 =item C<UTF8_GOT_NONCHAR>
1358 The code point represented by the input UTF-8 sequence is for a Unicode
1359 non-character code point.
1360 This bit is set only if the input C<flags> parameter contains either the
1361 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1363 =item C<UTF8_GOT_NON_CONTINUATION>
1365 The input sequence was malformed in that a non-continuation type byte was found
1366 in a position where only a continuation type one should be. See also
1367 L</C<UTF8_GOT_SHORT>>.
1369 =item C<UTF8_GOT_OVERFLOW>
1371 The input sequence was malformed in that it is for a code point that is not
1372 representable in the number of bits available in an IV on the current platform.
1374 =item C<UTF8_GOT_SHORT>
1376 The input sequence was malformed in that C<curlen> is smaller than required for
1377 a complete sequence. In other words, the input is for a partial character
1381 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1382 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1383 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1384 sequence was looked at. If no other flags are present, it means that the
1385 sequence was valid as far as it went. Depending on the application, this could
1386 mean one of three things:
1392 The C<curlen> length parameter passed in was too small, and the function was
1393 prevented from examining all the necessary bytes.
1397 The buffer being looked at is based on reading data, and the data received so
1398 far stopped in the middle of a character, so that the next read will
1399 read the remainder of this character. (It is up to the caller to deal with the
1400 split bytes somehow.)
1404 This is a real error, and the partial sequence is all we're going to get.
1408 =item C<UTF8_GOT_SUPER>
1410 The input sequence was malformed in that it is for a non-Unicode code point;
1411 that is, one above the legal Unicode maximum.
1412 This bit is set only if the input C<flags> parameter contains either the
1413 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1415 =item C<UTF8_GOT_SURROGATE>
1417 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1419 This bit is set only if the input C<flags> parameter contains either the
1420 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1424 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1425 flag to suppress any warnings, and then examine the C<*errors> return.
1429 Also implemented as a macro in utf8.h
1433 Perl_utf8n_to_uvchr_error(const U8 *s,
1439 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1441 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1446 =for apidoc utf8n_to_uvchr_msgs
1448 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1449 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1451 This function is for code that needs to know what the precise malformation(s)
1452 are when an error is found, and wants the corresponding warning and/or error
1453 messages to be returned to the caller rather than be displayed. All messages
1454 that would have been displayed if all lexcial warnings are enabled will be
1457 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1458 placed after all the others, C<msgs>. If this parameter is 0, this function
1459 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1460 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1461 contain any appropriate messages. The elements of the array are ordered so
1462 that the first message that would have been displayed is in the 0th element,
1463 and so on. Each element is a hash with three key-value pairs, as follows:
1469 The text of the message as a C<SVpv>.
1471 =item C<warn_categories>
1473 The warning category (or categories) packed into a C<SVuv>.
1477 A single flag bit associated with this message, in a C<SVuv>.
1478 The bit corresponds to some bit in the C<*errors> return value,
1479 such as C<UTF8_GOT_LONG>.
1483 It's important to note that specifying this parameter as non-null will cause
1484 any warnings this function would otherwise generate to be suppressed, and
1485 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1486 (or not) when choosing what to do with the returned messages.
1488 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1491 The caller, of course, is responsible for freeing any returned AV.
1497 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1504 const U8 * const s0 = s;
1505 const U8 * send = s0 + curlen;
1506 U32 possible_problems; /* A bit is set here for each potential problem
1507 found as we go along */
1509 STRLEN expectlen; /* How long should this sequence be? */
1510 STRLEN avail_len; /* When input is too short, gives what that is */
1511 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1512 gets set and discarded */
1514 /* The below are used only if there is both an overlong malformation and a
1515 * too short one. Otherwise the first two are set to 's0' and 'send', and
1516 * the third not used at all */
1518 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1519 routine; see [perl #130921] */
1523 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1525 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1526 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1527 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1532 /* Each of the affected Hanguls starts with \xED */
1534 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1545 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1546 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1547 | (s0[2] & UTF_CONTINUATION_MASK);
1552 /* In conjunction with the exhaustive tests that can be enabled in
1553 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1554 * what it is intended to do, and that no flaws in it are masked by
1555 * dropping down and executing the code below
1556 assert(! isUTF8_CHAR(s0, send)
1557 || UTF8_IS_SURROGATE(s0, send)
1558 || UTF8_IS_SUPER(s0, send)
1559 || UTF8_IS_NONCHAR(s0,send));
1564 possible_problems = 0;
1568 adjusted_s0 = (U8 *) s0;
1575 errors = &discard_errors;
1578 /* The order of malformation tests here is important. We should consume as
1579 * few bytes as possible in order to not skip any valid character. This is
1580 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1581 * http://unicode.org/reports/tr36 for more discussion as to why. For
1582 * example, once we've done a UTF8SKIP, we can tell the expected number of
1583 * bytes, and could fail right off the bat if the input parameters indicate
1584 * that there are too few available. But it could be that just that first
1585 * byte is garbled, and the intended character occupies fewer bytes. If we
1586 * blindly assumed that the first byte is correct, and skipped based on
1587 * that number, we could skip over a valid input character. So instead, we
1588 * always examine the sequence byte-by-byte.
1590 * We also should not consume too few bytes, otherwise someone could inject
1591 * things. For example, an input could be deliberately designed to
1592 * overflow, and if this code bailed out immediately upon discovering that,
1593 * returning to the caller C<*retlen> pointing to the very next byte (one
1594 * which is actually part of of the overflowing sequence), that could look
1595 * legitimate to the caller, which could discard the initial partial
1596 * sequence and process the rest, inappropriately.
1598 * Some possible input sequences are malformed in more than one way. This
1599 * function goes to lengths to try to find all of them. This is necessary
1600 * for correctness, as the inputs may allow one malformation but not
1601 * another, and if we abandon searching for others after finding the
1602 * allowed one, we could allow in something that shouldn't have been.
1605 if (UNLIKELY(curlen == 0)) {
1606 possible_problems |= UTF8_GOT_EMPTY;
1608 uv = UNICODE_REPLACEMENT;
1609 goto ready_to_handle_errors;
1612 expectlen = UTF8SKIP(s);
1614 /* A well-formed UTF-8 character, as the vast majority of calls to this
1615 * function will be for, has this expected length. For efficiency, set
1616 * things up here to return it. It will be overriden only in those rare
1617 * cases where a malformation is found */
1619 *retlen = expectlen;
1622 /* A continuation character can't start a valid sequence */
1623 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1624 possible_problems |= UTF8_GOT_CONTINUATION;
1626 uv = UNICODE_REPLACEMENT;
1627 goto ready_to_handle_errors;
1630 /* Here is not a continuation byte, nor an invariant. The only thing left
1631 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1632 * because it excludes start bytes like \xC0 that always lead to
1635 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1636 * that indicate the number of bytes in the character's whole UTF-8
1637 * sequence, leaving just the bits that are part of the value. */
1638 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1640 /* Setup the loop end point, making sure to not look past the end of the
1641 * input string, and flag it as too short if the size isn't big enough. */
1642 if (UNLIKELY(curlen < expectlen)) {
1643 possible_problems |= UTF8_GOT_SHORT;
1647 send = (U8*) s0 + expectlen;
1650 /* Now, loop through the remaining bytes in the character's sequence,
1651 * accumulating each into the working value as we go. */
1652 for (s = s0 + 1; s < send; s++) {
1653 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1654 uv = UTF8_ACCUMULATE(uv, *s);
1658 /* Here, found a non-continuation before processing all expected bytes.
1659 * This byte indicates the beginning of a new character, so quit, even
1660 * if allowing this malformation. */
1661 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1663 } /* End of loop through the character's bytes */
1665 /* Save how many bytes were actually in the character */
1668 /* Note that there are two types of too-short malformation. One is when
1669 * there is actual wrong data before the normal termination of the
1670 * sequence. The other is that the sequence wasn't complete before the end
1671 * of the data we are allowed to look at, based on the input 'curlen'.
1672 * This means that we were passed data for a partial character, but it is
1673 * valid as far as we saw. The other is definitely invalid. This
1674 * distinction could be important to a caller, so the two types are kept
1677 * A convenience macro that matches either of the too-short conditions. */
1678 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1680 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1682 uv = UNICODE_REPLACEMENT;
1685 /* Check for overflow. The algorithm requires us to not look past the end
1686 * of the current character, even if partial, so the upper limit is 's' */
1687 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1688 1 /* Do consider overlongs */
1691 possible_problems |= UTF8_GOT_OVERFLOW;
1692 uv = UNICODE_REPLACEMENT;
1695 /* Check for overlong. If no problems so far, 'uv' is the correct code
1696 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1697 * we must look at the UTF-8 byte sequence itself to see if it is for an
1699 if ( ( LIKELY(! possible_problems)
1700 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1701 || ( UNLIKELY(possible_problems)
1702 && ( UNLIKELY(! UTF8_IS_START(*s0))
1704 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1707 possible_problems |= UTF8_GOT_LONG;
1709 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1711 /* The calculation in the 'true' branch of this 'if'
1712 * below won't work if overflows, and isn't needed
1713 * anyway. Further below we handle all overflow
1715 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1717 UV min_uv = uv_so_far;
1720 /* Here, the input is both overlong and is missing some trailing
1721 * bytes. There is no single code point it could be for, but there
1722 * may be enough information present to determine if what we have
1723 * so far is for an unallowed code point, such as for a surrogate.
1724 * The code further below has the intelligence to determine this,
1725 * but just for non-overlong UTF-8 sequences. What we do here is
1726 * calculate the smallest code point the input could represent if
1727 * there were no too short malformation. Then we compute and save
1728 * the UTF-8 for that, which is what the code below looks at
1729 * instead of the raw input. It turns out that the smallest such
1730 * code point is all we need. */
1731 for (i = curlen; i < expectlen; i++) {
1732 min_uv = UTF8_ACCUMULATE(min_uv,
1733 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1736 adjusted_s0 = temp_char_buf;
1737 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1741 /* Here, we have found all the possible problems, except for when the input
1742 * is for a problematic code point not allowed by the input parameters. */
1744 /* uv is valid for overlongs */
1745 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1747 /* isn't problematic if < this */
1748 && uv >= UNICODE_SURROGATE_FIRST)
1749 || ( UNLIKELY(possible_problems)
1751 /* if overflow, we know without looking further
1752 * precisely which of the problematic types it is,
1753 * and we deal with those in the overflow handling
1755 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1756 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1757 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1758 && ((flags & ( UTF8_DISALLOW_NONCHAR
1759 |UTF8_DISALLOW_SURROGATE
1760 |UTF8_DISALLOW_SUPER
1761 |UTF8_DISALLOW_PERL_EXTENDED
1763 |UTF8_WARN_SURROGATE
1765 |UTF8_WARN_PERL_EXTENDED))))
1767 /* If there were no malformations, or the only malformation is an
1768 * overlong, 'uv' is valid */
1769 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1770 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1771 possible_problems |= UTF8_GOT_SURROGATE;
1773 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1774 possible_problems |= UTF8_GOT_SUPER;
1776 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1777 possible_problems |= UTF8_GOT_NONCHAR;
1780 else { /* Otherwise, need to look at the source UTF-8, possibly
1781 adjusted to be non-overlong */
1783 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1784 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1786 possible_problems |= UTF8_GOT_SUPER;
1788 else if (curlen > 1) {
1789 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1790 NATIVE_UTF8_TO_I8(*adjusted_s0),
1791 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1793 possible_problems |= UTF8_GOT_SUPER;
1795 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1796 NATIVE_UTF8_TO_I8(*adjusted_s0),
1797 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1799 possible_problems |= UTF8_GOT_SURROGATE;
1803 /* We need a complete well-formed UTF-8 character to discern
1804 * non-characters, so can't look for them here */
1808 ready_to_handle_errors:
1811 * curlen contains the number of bytes in the sequence that
1812 * this call should advance the input by.
1813 * avail_len gives the available number of bytes passed in, but
1814 * only if this is less than the expected number of
1815 * bytes, based on the code point's start byte.
1816 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1817 * is set in it for each potential problem found.
1818 * uv contains the code point the input sequence
1819 * represents; or if there is a problem that prevents
1820 * a well-defined value from being computed, it is
1821 * some subsitute value, typically the REPLACEMENT
1823 * s0 points to the first byte of the character
1824 * s points to just after were we left off processing
1826 * send points to just after where that character should
1827 * end, based on how many bytes the start byte tells
1828 * us should be in it, but no further than s0 +
1832 if (UNLIKELY(possible_problems)) {
1833 bool disallowed = FALSE;
1834 const U32 orig_problems = possible_problems;
1840 while (possible_problems) { /* Handle each possible problem */
1842 char * message = NULL;
1843 U32 this_flag_bit = 0;
1845 /* Each 'if' clause handles one problem. They are ordered so that
1846 * the first ones' messages will be displayed before the later
1847 * ones; this is kinda in decreasing severity order. But the
1848 * overlong must come last, as it changes 'uv' looked at by the
1850 if (possible_problems & UTF8_GOT_OVERFLOW) {
1852 /* Overflow means also got a super and are using Perl's
1853 * extended UTF-8, but we handle all three cases here */
1855 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1856 *errors |= UTF8_GOT_OVERFLOW;
1858 /* But the API says we flag all errors found */
1859 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1860 *errors |= UTF8_GOT_SUPER;
1863 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1865 *errors |= UTF8_GOT_PERL_EXTENDED;
1868 /* Disallow if any of the three categories say to */
1869 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1870 || (flags & ( UTF8_DISALLOW_SUPER
1871 |UTF8_DISALLOW_PERL_EXTENDED)))
1876 /* Likewise, warn if any say to */
1877 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1878 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1881 /* The warnings code explicitly says it doesn't handle the
1882 * case of packWARN2 and two categories which have
1883 * parent-child relationship. Even if it works now to
1884 * raise the warning if either is enabled, it wouldn't
1885 * necessarily do so in the future. We output (only) the
1886 * most dire warning */
1887 if (! (flags & UTF8_CHECK_ONLY)) {
1888 if (msgs || ckWARN_d(WARN_UTF8)) {
1889 pack_warn = packWARN(WARN_UTF8);
1891 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1892 pack_warn = packWARN(WARN_NON_UNICODE);
1895 message = Perl_form(aTHX_ "%s: %s (overflows)",
1897 _byte_dump_string(s0, curlen, 0));
1898 this_flag_bit = UTF8_GOT_OVERFLOW;
1903 else if (possible_problems & UTF8_GOT_EMPTY) {
1904 possible_problems &= ~UTF8_GOT_EMPTY;
1905 *errors |= UTF8_GOT_EMPTY;
1907 if (! (flags & UTF8_ALLOW_EMPTY)) {
1909 /* This so-called malformation is now treated as a bug in
1910 * the caller. If you have nothing to decode, skip calling
1916 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1918 pack_warn = packWARN(WARN_UTF8);
1919 message = Perl_form(aTHX_ "%s (empty string)",
1921 this_flag_bit = UTF8_GOT_EMPTY;
1925 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1926 possible_problems &= ~UTF8_GOT_CONTINUATION;
1927 *errors |= UTF8_GOT_CONTINUATION;
1929 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1932 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1934 pack_warn = packWARN(WARN_UTF8);
1935 message = Perl_form(aTHX_
1936 "%s: %s (unexpected continuation byte 0x%02x,"
1937 " with no preceding start byte)",
1939 _byte_dump_string(s0, 1, 0), *s0);
1940 this_flag_bit = UTF8_GOT_CONTINUATION;
1944 else if (possible_problems & UTF8_GOT_SHORT) {
1945 possible_problems &= ~UTF8_GOT_SHORT;
1946 *errors |= UTF8_GOT_SHORT;
1948 if (! (flags & UTF8_ALLOW_SHORT)) {
1951 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1953 pack_warn = packWARN(WARN_UTF8);
1954 message = Perl_form(aTHX_
1955 "%s: %s (too short; %d byte%s available, need %d)",
1957 _byte_dump_string(s0, send - s0, 0),
1959 avail_len == 1 ? "" : "s",
1961 this_flag_bit = UTF8_GOT_SHORT;
1966 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1967 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1968 *errors |= UTF8_GOT_NON_CONTINUATION;
1970 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1973 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1976 /* If we don't know for sure that the input length is
1977 * valid, avoid as much as possible reading past the
1978 * end of the buffer */
1979 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1982 pack_warn = packWARN(WARN_UTF8);
1983 message = Perl_form(aTHX_ "%s",
1984 unexpected_non_continuation_text(s0,
1988 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
1992 else if (possible_problems & UTF8_GOT_SURROGATE) {
1993 possible_problems &= ~UTF8_GOT_SURROGATE;
1995 if (flags & UTF8_WARN_SURROGATE) {
1996 *errors |= UTF8_GOT_SURROGATE;
1998 if ( ! (flags & UTF8_CHECK_ONLY)
1999 && (msgs || ckWARN_d(WARN_SURROGATE)))
2001 pack_warn = packWARN(WARN_SURROGATE);
2003 /* These are the only errors that can occur with a
2004 * surrogate when the 'uv' isn't valid */
2005 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2006 message = Perl_form(aTHX_
2007 "UTF-16 surrogate (any UTF-8 sequence that"
2008 " starts with \"%s\" is for a surrogate)",
2009 _byte_dump_string(s0, curlen, 0));
2012 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2014 this_flag_bit = UTF8_GOT_SURROGATE;
2018 if (flags & UTF8_DISALLOW_SURROGATE) {
2020 *errors |= UTF8_GOT_SURROGATE;
2023 else if (possible_problems & UTF8_GOT_SUPER) {
2024 possible_problems &= ~UTF8_GOT_SUPER;
2026 if (flags & UTF8_WARN_SUPER) {
2027 *errors |= UTF8_GOT_SUPER;
2029 if ( ! (flags & UTF8_CHECK_ONLY)
2030 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2032 pack_warn = packWARN(WARN_NON_UNICODE);
2034 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2035 message = Perl_form(aTHX_
2036 "Any UTF-8 sequence that starts with"
2037 " \"%s\" is for a non-Unicode code point,"
2038 " may not be portable",
2039 _byte_dump_string(s0, curlen, 0));
2042 message = Perl_form(aTHX_ super_cp_format, uv);
2044 this_flag_bit = UTF8_GOT_SUPER;
2048 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2049 * and before possibly bailing out, so that the more dire
2050 * warning will override the regular one. */
2051 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2052 if ( ! (flags & UTF8_CHECK_ONLY)
2053 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2054 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2056 pack_warn = packWARN(WARN_NON_UNICODE);
2058 /* If it is an overlong that evaluates to a code point
2059 * that doesn't have to use the Perl extended UTF-8, it
2060 * still used it, and so we output a message that
2061 * doesn't refer to the code point. The same is true
2062 * if there was a SHORT malformation where the code
2063 * point is not valid. In that case, 'uv' will have
2064 * been set to the REPLACEMENT CHAR, and the message
2065 * below without the code point in it will be selected
2067 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2068 message = Perl_form(aTHX_
2069 perl_extended_cp_format, uv);
2072 message = Perl_form(aTHX_
2073 "Any UTF-8 sequence that starts with"
2074 " \"%s\" is a Perl extension, and"
2075 " so is not portable",
2076 _byte_dump_string(s0, curlen, 0));
2078 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2081 if (flags & ( UTF8_WARN_PERL_EXTENDED
2082 |UTF8_DISALLOW_PERL_EXTENDED))
2084 *errors |= UTF8_GOT_PERL_EXTENDED;
2086 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2092 if (flags & UTF8_DISALLOW_SUPER) {
2093 *errors |= UTF8_GOT_SUPER;
2097 else if (possible_problems & UTF8_GOT_NONCHAR) {
2098 possible_problems &= ~UTF8_GOT_NONCHAR;
2100 if (flags & UTF8_WARN_NONCHAR) {
2101 *errors |= UTF8_GOT_NONCHAR;
2103 if ( ! (flags & UTF8_CHECK_ONLY)
2104 && (msgs || ckWARN_d(WARN_NONCHAR)))
2106 /* The code above should have guaranteed that we don't
2107 * get here with errors other than overlong */
2108 assert (! (orig_problems
2109 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2111 pack_warn = packWARN(WARN_NONCHAR);
2112 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2113 this_flag_bit = UTF8_GOT_NONCHAR;
2117 if (flags & UTF8_DISALLOW_NONCHAR) {
2119 *errors |= UTF8_GOT_NONCHAR;
2122 else if (possible_problems & UTF8_GOT_LONG) {
2123 possible_problems &= ~UTF8_GOT_LONG;
2124 *errors |= UTF8_GOT_LONG;
2126 if (flags & UTF8_ALLOW_LONG) {
2128 /* We don't allow the actual overlong value, unless the
2129 * special extra bit is also set */
2130 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2131 & ~UTF8_ALLOW_LONG)))
2133 uv = UNICODE_REPLACEMENT;
2140 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2142 pack_warn = packWARN(WARN_UTF8);
2144 /* These error types cause 'uv' to be something that
2145 * isn't what was intended, so can't use it in the
2146 * message. The other error types either can't
2147 * generate an overlong, or else the 'uv' is valid */
2149 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2151 message = Perl_form(aTHX_
2152 "%s: %s (any UTF-8 sequence that starts"
2153 " with \"%s\" is overlong which can and"
2154 " should be represented with a"
2155 " different, shorter sequence)",
2157 _byte_dump_string(s0, send - s0, 0),
2158 _byte_dump_string(s0, curlen, 0));
2161 U8 tmpbuf[UTF8_MAXBYTES+1];
2162 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2164 /* Don't use U+ for non-Unicode code points, which
2165 * includes those in the Latin1 range */
2166 const char * preface = ( uv > PERL_UNICODE_MAX
2173 message = Perl_form(aTHX_
2174 "%s: %s (overlong; instead use %s to represent"
2177 _byte_dump_string(s0, send - s0, 0),
2178 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2180 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2181 small code points */
2184 this_flag_bit = UTF8_GOT_LONG;
2187 } /* End of looking through the possible flags */
2189 /* Display the message (if any) for the problem being handled in
2190 * this iteration of the loop */
2193 assert(this_flag_bit);
2195 if (*msgs == NULL) {
2199 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2204 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2207 Perl_warner(aTHX_ pack_warn, "%s", message);
2209 } /* End of 'while (possible_problems)' */
2211 /* Since there was a possible problem, the returned length may need to
2212 * be changed from the one stored at the beginning of this function.
2213 * Instead of trying to figure out if that's needed, just do it. */
2219 if (flags & UTF8_CHECK_ONLY && retlen) {
2220 *retlen = ((STRLEN) -1);
2226 return UNI_TO_NATIVE(uv);
2230 =for apidoc utf8_to_uvchr_buf
2232 Returns the native code point of the first character in the string C<s> which
2233 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2234 C<*retlen> will be set to the length, in bytes, of that character.
2236 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2237 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2238 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2239 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2240 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2241 the next possible position in C<s> that could begin a non-malformed character.
2242 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2247 Also implemented as a macro in utf8.h
2253 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2255 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2259 return utf8n_to_uvchr(s, send - s, retlen,
2260 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
2263 /* This is marked as deprecated
2265 =for apidoc utf8_to_uvuni_buf
2267 Only in very rare circumstances should code need to be dealing in Unicode
2268 (as opposed to native) code points. In those few cases, use
2269 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2270 are not absolutely sure this is one of those cases, then assume it isn't and
2271 use plain C<utf8_to_uvchr_buf> instead.
2273 Returns the Unicode (not-native) code point of the first character in the
2275 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2276 C<retlen> will be set to the length, in bytes, of that character.
2278 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2279 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2280 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2281 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2282 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2283 next possible position in C<s> that could begin a non-malformed character.
2284 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
2290 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2292 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2296 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2300 =for apidoc utf8_length
2302 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2303 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2304 same place, it returns 0 with no warning raised.
2306 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2307 and returns the number of valid characters.
2313 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2317 PERL_ARGS_ASSERT_UTF8_LENGTH;
2319 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2320 * the bitops (especially ~) can create illegal UTF-8.
2321 * In other words: in Perl UTF-8 is not just for Unicode. */
2323 if (UNLIKELY(e < s))
2324 goto warn_and_return;
2330 if (UNLIKELY(e != s)) {
2334 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2335 "%s in %s", unees, OP_DESC(PL_op));
2337 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2344 =for apidoc bytes_cmp_utf8
2346 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2347 sequence of characters (stored as UTF-8)
2348 in C<u>, C<ulen>. Returns 0 if they are
2349 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2350 if the first string is greater than the second string.
2352 -1 or +1 is returned if the shorter string was identical to the start of the
2353 longer string. -2 or +2 is returned if
2354 there was a difference between characters
2361 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2363 const U8 *const bend = b + blen;
2364 const U8 *const uend = u + ulen;
2366 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2368 while (b < bend && u < uend) {
2370 if (!UTF8_IS_INVARIANT(c)) {
2371 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2374 if (UTF8_IS_CONTINUATION(c1)) {
2375 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2377 /* diag_listed_as: Malformed UTF-8 character%s */
2378 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2380 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2381 PL_op ? " in " : "",
2382 PL_op ? OP_DESC(PL_op) : "");
2387 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2388 "%s in %s", unees, OP_DESC(PL_op));
2390 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2391 return -2; /* Really want to return undef :-) */
2398 return *b < c ? -2 : +2;
2403 if (b == bend && u == uend)
2406 return b < bend ? +1 : -1;
2410 =for apidoc utf8_to_bytes
2412 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2413 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2414 updates C<*lenp> to contain the new length.
2415 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2417 Upon successful return, the number of variants in the string can be computed by
2418 having saved the value of C<*lenp> before the call, and subtracting the
2419 after-call value of C<*lenp> from it.
2421 If you need a copy of the string, see L</bytes_from_utf8>.
2427 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2431 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2432 PERL_UNUSED_CONTEXT;
2434 /* This is a no-op if no variants at all in the input */
2435 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2440 U8 * const save = s;
2441 U8 * const send = s + *lenp;
2444 /* Nothing before the first variant needs to be changed, so start the real
2448 if (! UTF8_IS_INVARIANT(*s)) {
2449 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2450 *lenp = ((STRLEN) -1);
2458 /* Is downgradable, so do it */
2459 d = s = first_variant;
2462 if (! UVCHR_IS_INVARIANT(c)) {
2463 /* Then it is two-byte encoded */
2464 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2477 =for apidoc bytes_from_utf8
2479 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2480 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2481 actually encoded in UTF-8.
2483 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2486 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2487 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2488 C<*lenp> are unchanged, and the return value is the original C<s>.
2490 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2491 newly created string containing a downgraded copy of C<s>, and whose length is
2492 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2493 caller is responsible for arranging for the memory used by this string to get
2496 Upon successful return, the number of variants in the string can be computed by
2497 having saved the value of C<*lenp> before the call, and subtracting the
2498 after-call value of C<*lenp> from it.
2502 There is a macro that avoids this function call, but this is retained for
2503 anyone who calls it with the Perl_ prefix */
2506 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2508 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2509 PERL_UNUSED_CONTEXT;
2511 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2515 No = here because currently externally undocumented
2516 for apidoc bytes_from_utf8_loc
2518 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2519 to store the location of the first character in C<"s"> that cannot be
2520 converted to non-UTF8.
2522 If that parameter is C<NULL>, this function behaves identically to
2525 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2526 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2528 Otherwise, the function returns a newly created C<NUL>-terminated string
2529 containing the non-UTF8 equivalent of the convertible first portion of
2530 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2531 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2532 and C<*first_non_downgradable> is set to C<NULL>.
2534 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2535 first character in the original string that wasn't converted. C<*is_utf8p> is
2536 unchanged. Note that the new string may have length 0.
2538 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2539 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2540 converts as many characters in it as possible stopping at the first one it
2541 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2542 set to point to that. The function returns the portion that could be converted
2543 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2544 not including the terminating C<NUL>. If the very first character in the
2545 original could not be converted, C<*lenp> will be 0, and the new string will
2546 contain just a single C<NUL>. If the entire input string was converted,
2547 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2549 Upon successful return, the number of variants in the converted portion of the
2550 string can be computed by having saved the value of C<*lenp> before the call,
2551 and subtracting the after-call value of C<*lenp> from it.
2559 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2562 const U8 *original = s;
2563 U8 *converted_start;
2564 const U8 *send = s + *lenp;
2566 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2569 if (first_unconverted) {
2570 *first_unconverted = NULL;
2573 return (U8 *) original;
2576 Newx(d, (*lenp) + 1, U8);
2578 converted_start = d;
2581 if (! UTF8_IS_INVARIANT(c)) {
2583 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2584 * have to stop now */
2585 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2586 if (first_unconverted) {
2587 *first_unconverted = s - 1;
2588 goto finish_and_return;
2591 Safefree(converted_start);
2592 return (U8 *) original;
2596 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2602 /* Here, converted the whole of the input */
2604 if (first_unconverted) {
2605 *first_unconverted = NULL;
2610 *lenp = d - converted_start;
2612 /* Trim unused space */
2613 Renew(converted_start, *lenp + 1, U8);
2615 return converted_start;
2619 =for apidoc bytes_to_utf8
2621 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2623 Returns a pointer to the newly-created string, and sets C<*lenp> to
2624 reflect the new length in bytes. The caller is responsible for arranging for
2625 the memory used by this string to get freed.
2627 Upon successful return, the number of variants in the string can be computed by
2628 having saved the value of C<*lenp> before the call, and subtracting it from the
2629 after-call value of C<*lenp>.
2631 A C<NUL> character will be written after the end of the string.
2633 If you want to convert to UTF-8 from encodings other than
2634 the native (Latin1 or EBCDIC),
2635 see L</sv_recode_to_utf8>().
2641 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2643 const U8 * const send = s + (*lenp);
2647 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2648 PERL_UNUSED_CONTEXT;
2650 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2651 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2655 append_utf8_from_native_byte(*s, &d);
2666 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2667 * use utf16_to_utf8_reversed().
2669 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2670 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2671 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2673 * These functions don't check for overflow. The worst case is every code
2674 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2675 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2676 * destination must be pre-extended to 2 times the source length.
2678 * Do not use in-place. We optimize for native, for obvious reasons. */
2681 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2686 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2689 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2695 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2697 if (OFFUNI_IS_INVARIANT(uv)) {
2698 *d++ = LATIN1_TO_NATIVE((U8) uv);
2701 if (uv <= MAX_UTF8_TWO_BYTE) {
2702 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2703 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2707 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2708 #define LAST_HIGH_SURROGATE 0xDBFF
2709 #define FIRST_LOW_SURROGATE 0xDC00
2710 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2711 #define FIRST_IN_PLANE1 0x10000
2713 /* This assumes that most uses will be in the first Unicode plane, not
2714 * needing surrogates */
2715 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2716 && uv <= UNICODE_SURROGATE_LAST))
2718 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2719 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2722 UV low = (p[0] << 8) + p[1];
2723 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2724 || UNLIKELY(low > LAST_LOW_SURROGATE))
2726 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2729 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2730 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2734 d = uvoffuni_to_utf8_flags(d, uv, 0);
2736 if (uv < FIRST_IN_PLANE1) {
2737 *d++ = (U8)(( uv >> 12) | 0xe0);
2738 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2739 *d++ = (U8)(( uv & 0x3f) | 0x80);
2743 *d++ = (U8)(( uv >> 18) | 0xf0);
2744 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2745 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2746 *d++ = (U8)(( uv & 0x3f) | 0x80);
2751 *newlen = d - dstart;
2755 /* Note: this one is slightly destructive of the source. */
2758 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2761 U8* const send = s + bytelen;
2763 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2766 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2770 const U8 tmp = s[0];
2775 return utf16_to_utf8(p, d, bytelen, newlen);
2779 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2782 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2785 /* Internal function so we can deprecate the external one, and call
2786 this one from other deprecated functions in this file */
2789 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2793 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2797 return is_utf8_common(p, PL_utf8_idstart);
2801 Perl__is_uni_perl_idcont(pTHX_ UV c)
2804 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2808 Perl__is_uni_perl_idstart(pTHX_ UV c)
2811 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2815 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2818 /* We have the latin1-range values compiled into the core, so just use
2819 * those, converting the result to UTF-8. The only difference between upper
2820 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2821 * either "SS" or "Ss". Which one to use is passed into the routine in
2822 * 'S_or_s' to avoid a test */
2824 UV converted = toUPPER_LATIN1_MOD(c);
2826 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2828 assert(S_or_s == 'S' || S_or_s == 's');
2830 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2831 characters in this range */
2832 *p = (U8) converted;
2837 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2838 * which it maps to one of them, so as to only have to have one check for
2839 * it in the main case */
2840 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2842 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2843 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2846 converted = GREEK_CAPITAL_LETTER_MU;
2848 #if UNICODE_MAJOR_VERSION > 2 \
2849 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2850 && UNICODE_DOT_DOT_VERSION >= 8)
2851 case LATIN_SMALL_LETTER_SHARP_S:
2858 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2859 " '%c' to map to '%c'",
2860 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2861 NOT_REACHED; /* NOTREACHED */
2865 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2866 *p = UTF8_TWO_BYTE_LO(converted);
2872 /* If compiled on an early Unicode version, there may not be auxiliary tables
2874 #ifndef HAS_UC_AUX_TABLES
2875 # define UC_AUX_TABLE_ptrs NULL
2876 # define UC_AUX_TABLE_lengths NULL
2878 #ifndef HAS_TC_AUX_TABLES
2879 # define TC_AUX_TABLE_ptrs NULL
2880 # define TC_AUX_TABLE_lengths NULL
2882 #ifndef HAS_LC_AUX_TABLES
2883 # define LC_AUX_TABLE_ptrs NULL
2884 # define LC_AUX_TABLE_lengths NULL
2886 #ifndef HAS_CF_AUX_TABLES
2887 # define CF_AUX_TABLE_ptrs NULL
2888 # define CF_AUX_TABLE_lengths NULL
2890 #ifndef HAS_UC_AUX_TABLES
2891 # define UC_AUX_TABLE_ptrs NULL
2892 # define UC_AUX_TABLE_lengths NULL
2895 /* Call the function to convert a UTF-8 encoded character to the specified case.
2896 * Note that there may be more than one character in the result.
2897 * 's' is a pointer to the first byte of the input character
2898 * 'd' will be set to the first byte of the string of changed characters. It
2899 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2900 * 'lenp' will be set to the length in bytes of the string of changed characters
2902 * The functions return the ordinal of the first character in the string of
2904 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2905 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2906 Uppercase_Mapping_invmap, \
2907 UC_AUX_TABLE_ptrs, \
2908 UC_AUX_TABLE_lengths, \
2910 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2911 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2912 Titlecase_Mapping_invmap, \
2913 TC_AUX_TABLE_ptrs, \
2914 TC_AUX_TABLE_lengths, \
2916 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2917 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2918 Lowercase_Mapping_invmap, \
2919 LC_AUX_TABLE_ptrs, \
2920 LC_AUX_TABLE_lengths, \
2924 /* This additionally has the input parameter 'specials', which if non-zero will
2925 * cause this to use the specials hash for folding (meaning get full case
2926 * folding); otherwise, when zero, this implies a simple case fold */
2927 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2929 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2930 Case_Folding_invmap, \
2931 CF_AUX_TABLE_ptrs, \
2932 CF_AUX_TABLE_lengths, \
2934 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2935 Simple_Case_Folding_invmap, \
2940 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2942 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2943 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2944 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2945 * the changed version may be longer than the original character.
2947 * The ordinal of the first character of the changed version is returned
2948 * (but note, as explained above, that there may be more.) */
2951 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2954 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2957 return CALL_UPPER_CASE(c, NULL, p, lenp);
2961 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2964 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2967 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2970 return CALL_TITLE_CASE(c, NULL, p, lenp);
2974 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2976 /* We have the latin1-range values compiled into the core, so just use
2977 * those, converting the result to UTF-8. Since the result is always just
2978 * one character, we allow <p> to be NULL */
2980 U8 converted = toLOWER_LATIN1(c);
2982 PERL_UNUSED_ARG(dummy);
2985 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2990 /* Result is known to always be < 256, so can use the EIGHT_BIT
2992 *p = UTF8_EIGHT_BIT_HI(converted);
2993 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3001 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3004 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3007 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3010 return CALL_LOWER_CASE(c, NULL, p, lenp);
3014 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3016 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3017 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3018 * FOLD_FLAGS_FULL iff full folding is to be used;
3020 * Not to be used for locale folds
3025 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3027 assert (! (flags & FOLD_FLAGS_LOCALE));
3029 if (UNLIKELY(c == MICRO_SIGN)) {
3030 converted = GREEK_SMALL_LETTER_MU;
3032 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3033 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3034 || UNICODE_DOT_DOT_VERSION > 0)
3035 else if ( (flags & FOLD_FLAGS_FULL)
3036 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3038 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3039 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3040 * under those circumstances. */
3041 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3042 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3043 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3045 return LATIN_SMALL_LETTER_LONG_S;
3055 else { /* In this range the fold of all other characters is their lower
3057 converted = toLOWER_LATIN1(c);
3060 if (UVCHR_IS_INVARIANT(converted)) {
3061 *p = (U8) converted;
3065 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3066 *p = UTF8_TWO_BYTE_LO(converted);
3074 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3077 /* Not currently externally documented, and subject to change
3078 * <flags> bits meanings:
3079 * FOLD_FLAGS_FULL iff full folding is to be used;
3080 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3081 * locale are to be used.
3082 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3086 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3088 if (flags & FOLD_FLAGS_LOCALE) {
3089 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3090 * except for potentially warning */
3091 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3092 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3093 flags &= ~FOLD_FLAGS_LOCALE;
3096 goto needs_full_generality;
3101 return _to_fold_latin1((U8) c, p, lenp,
3102 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3105 /* Here, above 255. If no special needs, just use the macro */
3106 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3107 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3109 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3110 the special flags. */
3111 U8 utf8_c[UTF8_MAXBYTES + 1];
3113 needs_full_generality:
3114 uvchr_to_utf8(utf8_c, c);
3115 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3120 PERL_STATIC_INLINE bool
3121 S_is_utf8_common(pTHX_ const U8 *const p, SV* const invlist)
3123 /* returns a boolean giving whether or not the UTF8-encoded character that
3124 * starts at <p> is in the inversion list indicated by <invlist>.
3126 * Note that it is assumed that the buffer length of <p> is enough to
3127 * contain all the bytes that comprise the character. Thus, <*p> should
3128 * have been checked before this call for mal-formedness enough to assure
3129 * that. This function, does make sure to not look past any NUL, so it is
3130 * safe to use on C, NUL-terminated, strings */
3131 STRLEN len = my_strnlen((char *) p, UTF8SKIP(p));
3133 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3135 /* The API should have included a length for the UTF-8 character in <p>,
3136 * but it doesn't. We therefore assume that p has been validated at least
3137 * as far as there being enough bytes available in it to accommodate the
3138 * character without reading beyond the end, and pass that number on to the
3139 * validating routine */
3140 if (! isUTF8_CHAR(p, p + len)) {
3141 _force_out_malformed_utf8_message(p, p + len, _UTF8_NO_CONFIDENCE_IN_CURLEN,
3143 NOT_REACHED; /* NOTREACHED */
3146 return is_utf8_common_with_len(p, p + len, invlist);
3149 PERL_STATIC_INLINE bool
3150 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3153 /* returns a boolean giving whether or not the UTF8-encoded character that
3154 * starts at <p>, and extending no further than <e - 1> is in the inversion
3155 * list <invlist>. */
3157 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3159 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3161 if (cp == 0 && (p >= e || *p != '\0')) {
3162 _force_out_malformed_utf8_message(p, e, 0, 1);
3163 NOT_REACHED; /* NOTREACHED */
3167 return _invlist_contains_cp(invlist, cp);
3171 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3172 const char * const alternative,
3173 const bool use_locale,
3174 const char * const file,
3175 const unsigned line)
3179 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3181 if (ckWARN_d(WARN_DEPRECATED)) {
3183 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3184 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3185 if (! PL_seen_deprecated_macro) {
3186 PL_seen_deprecated_macro = newHV();
3188 if (! hv_store(PL_seen_deprecated_macro, key,
3189 strlen(key), &PL_sv_undef, 0))
3191 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3194 if (instr(file, "mathoms.c")) {
3195 Perl_warner(aTHX_ WARN_DEPRECATED,
3196 "In %s, line %d, starting in Perl v5.32, %s()"
3197 " will be removed. Avoid this message by"
3198 " converting to use %s().\n",
3199 file, line, name, alternative);
3202 Perl_warner(aTHX_ WARN_DEPRECATED,
3203 "In %s, line %d, starting in Perl v5.32, %s() will"
3204 " require an additional parameter. Avoid this"
3205 " message by converting to use %s().\n",
3206 file, line, name, alternative);
3213 Perl__is_utf8_FOO(pTHX_ U8 classnum,
3215 const char * const name,
3216 const char * const alternative,
3217 const bool use_utf8,
3218 const bool use_locale,
3219 const char * const file,
3220 const unsigned line)
3223 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3225 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3227 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
3237 case _CC_ALPHANUMERIC:
3241 return is_utf8_common(p, PL_XPosix_ptrs[classnum]);
3244 return is_XPERLSPACE_high(p);
3246 return is_HORIZWS_high(p);
3248 return is_XDIGIT_high(p);
3254 return is_VERTWS_high(p);
3256 return is_utf8_common(p, PL_utf8_perl_idstart);
3258 return is_utf8_common(p, PL_utf8_perl_idcont);
3262 /* idcont is the same as wordchar below 256 */
3263 if (classnum == _CC_IDCONT) {
3264 classnum = _CC_WORDCHAR;
3266 else if (classnum == _CC_IDFIRST) {
3270 classnum = _CC_ALPHA;
3274 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3275 return _generic_isCC(*p, classnum);
3278 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3281 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3282 return isFOO_lc(classnum, *p);
3285 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3288 NOT_REACHED; /* NOTREACHED */
3292 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3296 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3298 return is_utf8_common_with_len(p, e, PL_XPosix_ptrs[classnum]);
3302 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3305 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3307 return is_utf8_common_with_len(p, e, PL_utf8_perl_idstart);
3311 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3314 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3318 return is_utf8_common(p, PL_utf8_xidstart);
3322 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3325 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3327 return is_utf8_common_with_len(p, e, PL_utf8_perl_idcont);
3331 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3334 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3336 return is_utf8_common(p, PL_utf8_idcont);
3340 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3343 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3345 return is_utf8_common(p, PL_utf8_xidcont);
3349 Perl__is_utf8_mark(pTHX_ const U8 *p)
3352 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3354 return is_utf8_common(p, PL_utf8_mark);
3358 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3359 U8* ustrp, STRLEN *lenp,
3360 SV *invlist, const int * const invmap,
3361 const unsigned int * const * const aux_tables,
3362 const U8 * const aux_table_lengths,
3363 const char * const normal)
3367 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3368 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3369 * to name the new case in any generated messages, as a fallback if the
3370 * operation being used is not available. The new case is given by the
3371 * data structures in the remaining arguments.
3373 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3374 * entire changed case string, and the return value is the first code point
3377 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3379 /* For code points that don't change case, we already know that the output
3380 * of this function is the unchanged input, so we can skip doing look-ups
3381 * for them. Unfortunately the case-changing code points are scattered
3382 * around. But there are some long consecutive ranges where there are no
3383 * case changing code points. By adding tests, we can eliminate the lookup
3384 * for all the ones in such ranges. This is currently done here only for
3385 * just a few cases where the scripts are in common use in modern commerce
3386 * (and scripts adjacent to those which can be included without additional
3389 if (uv1 >= 0x0590) {
3390 /* This keeps from needing further processing the code points most
3391 * likely to be used in the following non-cased scripts: Hebrew,
3392 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3393 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3394 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3399 /* The following largish code point ranges also don't have case
3400 * changes, but khw didn't think they warranted extra tests to speed
3401 * them up (which would slightly slow down everything else above them):
3402 * 1100..139F Hangul Jamo, Ethiopic
3403 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3404 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3405 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3406 * Combining Diacritical Marks Extended, Balinese,
3407 * Sundanese, Batak, Lepcha, Ol Chiki
3408 * 2000..206F General Punctuation
3411 if (uv1 >= 0x2D30) {
3413 /* This keeps the from needing further processing the code points
3414 * most likely to be used in the following non-cased major scripts:
3415 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3417 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3418 * event that Unicode eventually allocates the unused block as of
3419 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3420 * that the test suite will start having failures to alert you
3421 * should that happen) */
3426 if (uv1 >= 0xAC00) {
3427 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3428 if (ckWARN_d(WARN_SURROGATE)) {
3429 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3430 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3431 "Operation \"%s\" returns its argument for"
3432 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3437 /* AC00..FAFF Catches Hangul syllables and private use, plus
3443 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3444 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3445 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3448 if (ckWARN_d(WARN_NON_UNICODE)) {
3449 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3450 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3451 "Operation \"%s\" returns its argument for"
3452 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3456 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3458 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3461 /* As of Unicode 10.0, this means we avoid swash creation
3462 * for anything beyond high Plane 1 (below emojis) */
3469 /* Note that non-characters are perfectly legal, so no warning should
3475 const unsigned int * cp_list;
3478 /* 'index' is guaranteed to be non-negative, as this is an inversion
3479 * map that covers all possible inputs. See [perl #133365] */
3480 SSize_t index = _invlist_search(invlist, uv1);
3481 IV base = invmap[index];
3483 /* The data structures are set up so that if 'base' is non-negative,
3484 * the case change is 1-to-1; and if 0, the change is to itself */
3492 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3493 lc = base + uv1 - invlist_array(invlist)[index];
3494 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3498 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3499 * requires an auxiliary table look up. abs(base) gives the index into
3500 * a list of such tables which points to the proper aux table. And a
3501 * parallel list gives the length of each corresponding aux table. */
3502 cp_list = aux_tables[-base];
3504 /* Create the string of UTF-8 from the mapped-to code points */
3506 for (i = 0; i < aux_table_lengths[-base]; i++) {
3507 d = uvchr_to_utf8(d, cp_list[i]);
3515 /* Here, there was no mapping defined, which means that the code point maps
3516 * to itself. Return the inputs */
3520 if (p != ustrp) { /* Don't copy onto itself */
3521 Copy(p, ustrp, len, U8);
3526 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3534 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3535 const unsigned int ** remaining_folds_to)
3537 /* Returns the count of the number of code points that fold to the input
3538 * 'cp' (besides itself).
3540 * If the return is 0, there is nothing else that folds to it, and
3541 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3543 * If the return is 1, '*first_folds_to' is set to the single code point,
3544 * and '*remaining_folds_to' is set to NULL.
3546 * Otherwise, '*first_folds_to' is set to a code point, and
3547 * '*remaining_fold_to' is set to an array that contains the others. The
3548 * length of this array is the returned count minus 1.
3550 * The reason for this convolution is to avoid having to deal with
3551 * allocating and freeing memory. The lists are already constructed, so
3552 * the return can point to them, but single code points aren't, so would
3553 * need to be constructed if we didn't employ something like this API */
3556 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3557 * that covers all possible inputs. See [perl #133365] */
3558 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3559 int base = _Perl_IVCF_invmap[index];
3561 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3563 if (base == 0) { /* No fold */
3564 *first_folds_to = 0;
3565 *remaining_folds_to = NULL;
3569 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3575 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3577 /* The data structure is set up so that the absolute value of 'base' is
3578 * an index into a table of pointers to arrays, with the array
3579 * corresponding to the index being the list of code points that fold
3580 * to 'cp', and the parallel array containing the length of the list
3582 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3583 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3586 return IVCF_AUX_TABLE_lengths[-base];
3591 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3592 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3593 *remaining_folds_to = NULL;
3598 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3599 U8* const ustrp, STRLEN *lenp)
3601 /* This is called when changing the case of a UTF-8-encoded character above
3602 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3603 * result contains a character that crosses the 255/256 boundary, disallow
3604 * the change, and return the original code point. See L<perlfunc/lc> for
3607 * p points to the original string whose case was changed; assumed
3608 * by this routine to be well-formed
3609 * result the code point of the first character in the changed-case string
3610 * ustrp points to the changed-case string (<result> represents its
3612 * lenp points to the length of <ustrp> */
3614 UV original; /* To store the first code point of <p> */
3616 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3618 assert(UTF8_IS_ABOVE_LATIN1(*p));
3620 /* We know immediately if the first character in the string crosses the
3621 * boundary, so can skip testing */
3624 /* Look at every character in the result; if any cross the
3625 * boundary, the whole thing is disallowed */
3626 U8* s = ustrp + UTF8SKIP(ustrp);
3627 U8* e = ustrp + *lenp;
3629 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3635 /* Here, no characters crossed, result is ok as-is, but we warn. */
3636 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3642 /* Failed, have to return the original */
3643 original = valid_utf8_to_uvchr(p, lenp);
3645 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3646 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3647 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3648 " locale; resolved to \"\\x{%" UVXf "}\".",
3652 Copy(p, ustrp, *lenp, char);
3657 S_check_and_deprecate(pTHX_ const U8 *p,
3659 const unsigned int type, /* See below */
3660 const bool use_locale, /* Is this a 'LC_'
3662 const char * const file,
3663 const unsigned line)
3665 /* This is a temporary function to deprecate the unsafe calls to the case
3666 * changing macros and functions. It keeps all the special stuff in just
3669 * It updates *e with the pointer to the end of the input string. If using
3670 * the old-style macros, *e is NULL on input, and so this function assumes
3671 * the input string is long enough to hold the entire UTF-8 sequence, and
3672 * sets *e accordingly, but it then returns a flag to pass the
3673 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3674 * using the full length if possible.
3676 * It also does the assert that *e > p when *e is not NULL. This should be
3677 * migrated to the callers when this function gets deleted.
3679 * The 'type' parameter is used for the caller to specify which case
3680 * changing function this is called from: */
3682 # define DEPRECATE_TO_UPPER 0
3683 # define DEPRECATE_TO_TITLE 1
3684 # define DEPRECATE_TO_LOWER 2
3685 # define DEPRECATE_TO_FOLD 3
3687 U32 utf8n_flags = 0;
3689 const char * alternative;
3691 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3694 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3696 /* strnlen() makes this function safe for the common case of
3697 * NUL-terminated strings */
3698 *e = p + my_strnlen((char *) p, UTF8SKIP(p));
3700 /* For mathoms.c calls, we use the function name we know is stored
3701 * there. It could be part of a larger path */
3702 if (type == DEPRECATE_TO_UPPER) {
3703 name = instr(file, "mathoms.c")
3706 alternative = "toUPPER_utf8_safe";
3708 else if (type == DEPRECATE_TO_TITLE) {
3709 name = instr(file, "mathoms.c")
3712 alternative = "toTITLE_utf8_safe";
3714 else if (type == DEPRECATE_TO_LOWER) {
3715 name = instr(file, "mathoms.c")
3718 alternative = "toLOWER_utf8_safe";
3720 else if (type == DEPRECATE_TO_FOLD) {
3721 name = instr(file, "mathoms.c")
3724 alternative = "toFOLD_utf8_safe";
3726 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3728 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3738 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3739 U8 * ustrp, STRLEN *lenp)
3741 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3742 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3743 * Otherwise, it returns the first code point of the Turkic foldcased
3744 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3745 * contain *lenp bytes
3747 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3748 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3751 PERL_ARGS_ASSERT_TURKIC_FC;
3754 if (UNLIKELY(*p == 'I')) {
3756 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3757 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3758 return LATIN_SMALL_LETTER_DOTLESS_I;
3761 if (UNLIKELY(memBEGINs(p, e - p,
3762 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3773 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3774 U8 * ustrp, STRLEN *lenp)
3776 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3777 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3778 * Otherwise, it returns the first code point of the Turkic lowercased
3779 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3780 * contain *lenp bytes */
3783 PERL_ARGS_ASSERT_TURKIC_LC;
3786 /* A 'I' requires context as to what to do */
3787 if (UNLIKELY(*p0 == 'I')) {
3788 const U8 * p = p0 + 1;
3790 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3791 * modified by a dot above lowercases to 'i' even in turkic locales. */
3795 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3801 /* For the dot above to modify the 'I', it must be part of a
3802 * combining sequence immediately following the 'I', and no other
3803 * modifier with a ccc of 230 may intervene */
3804 cp = utf8_to_uvchr_buf(p, e, NULL);
3805 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3809 /* Here the combining sequence continues */
3814 /* In all other cases the lc is the same as the fold */
3815 return turkic_fc(p0, e, ustrp, lenp);
3819 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3820 U8 * ustrp, STRLEN *lenp)
3822 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3823 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3824 * Otherwise, it returns the first code point of the Turkic upper or
3825 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3826 * ustrp will contain *lenp bytes
3828 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3829 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3832 PERL_ARGS_ASSERT_TURKIC_UC;
3837 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3838 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3839 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3842 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3851 /* The process for changing the case is essentially the same for the four case
3852 * change types, except there are complications for folding. Otherwise the
3853 * difference is only which case to change to. To make sure that they all do
3854 * the same thing, the bodies of the functions are extracted out into the
3855 * following two macros. The functions are written with the same variable
3856 * names, and these are known and used inside these macros. It would be
3857 * better, of course, to have inline functions to do it, but since different
3858 * macros are called, depending on which case is being changed to, this is not
3859 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3860 * function can start with the common start macro, then finish with its special
3861 * handling; while the other three cases can just use the common end macro.
3863 * The algorithm is to use the proper (passed in) macro or function to change
3864 * the case for code points that are below 256. The macro is used if using
3865 * locale rules for the case change; the function if not. If the code point is
3866 * above 255, it is computed from the input UTF-8, and another macro is called
3867 * to do the conversion. If necessary, the output is converted to UTF-8. If
3868 * using a locale, we have to check that the change did not cross the 255/256
3869 * boundary, see check_locale_boundary_crossing() for further details.
3871 * The macros are split with the correct case change for the below-256 case
3872 * stored into 'result', and in the middle of an else clause for the above-255
3873 * case. At that point in the 'else', 'result' is not the final result, but is
3874 * the input code point calculated from the UTF-8. The fold code needs to
3875 * realize all this and take it from there.
3877 * To deal with Turkic locales, the function specified by the parameter
3878 * 'turkic' is called when appropriate.
3880 * If you read the two macros as sequential, it's easier to understand what's
3882 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3883 L1_func_extra_param, turkic) \
3885 if (flags & (locale_flags)) { \
3886 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3887 if (IN_UTF8_CTYPE_LOCALE) { \
3888 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3889 UV ret = turkic(p, e, ustrp, lenp); \
3890 if (ret) return ret; \
3893 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3895 flags &= ~(locale_flags); \
3899 if (UTF8_IS_INVARIANT(*p)) { \
3900 if (flags & (locale_flags)) { \
3901 result = LC_L1_change_macro(*p); \
3904 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3907 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3908 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3909 if (flags & (locale_flags)) { \
3910 result = LC_L1_change_macro(c); \
3913 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3916 else { /* malformed UTF-8 or ord above 255 */ \
3917 STRLEN len_result; \
3918 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3919 if (len_result == (STRLEN) -1) { \
3920 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3924 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3925 result = change_macro(result, p, ustrp, lenp); \
3927 if (flags & (locale_flags)) { \
3928 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3933 /* Here, used locale rules. Convert back to UTF-8 */ \
3934 if (UTF8_IS_INVARIANT(result)) { \
3935 *ustrp = (U8) result; \
3939 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3940 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3947 =for apidoc to_utf8_upper
3949 Instead use L</toUPPER_utf8_safe>.
3953 /* Not currently externally documented, and subject to change:
3954 * <flags> is set iff iff the rules from the current underlying locale are to
3958 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3963 const char * const file,
3968 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3969 cBOOL(flags), file, line);
3971 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3973 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3974 /* 2nd char of uc(U+DF) is 'S' */
3975 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3977 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3981 =for apidoc to_utf8_title
3983 Instead use L</toTITLE_utf8_safe>.
3987 /* Not currently externally documented, and subject to change:
3988 * <flags> is set iff the rules from the current underlying locale are to be
3989 * used. Since titlecase is not defined in POSIX, for other than a
3990 * UTF-8 locale, uppercase is used instead for code points < 256.
3994 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3999 const char * const file,
4004 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
4005 cBOOL(flags), file, line);
4007 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
4009 /* 2nd char of ucfirst(U+DF) is 's' */
4010 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
4012 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
4016 =for apidoc to_utf8_lower
4018 Instead use L</toLOWER_utf8_safe>.
4022 /* Not currently externally documented, and subject to change:
4023 * <flags> is set iff iff the rules from the current underlying locale are to
4028 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
4033 const char * const file,
4038 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
4039 cBOOL(flags), file, line);
4041 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
4043 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
4045 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
4049 =for apidoc to_utf8_fold
4051 Instead use L</toFOLD_utf8_safe>.
4055 /* Not currently externally documented, and subject to change,
4057 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
4058 * locale are to be used.
4059 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
4060 * otherwise simple folds
4061 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
4066 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
4071 const char * const file,
4076 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
4077 cBOOL(flags), file, line);
4079 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
4081 /* These are mutually exclusive */
4082 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
4084 assert(p != ustrp); /* Otherwise overwrites */
4086 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
4087 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
4090 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
4092 if (flags & FOLD_FLAGS_LOCALE) {
4094 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
4095 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4096 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4098 /* Special case these two characters, as what normally gets
4099 * returned under locale doesn't work */
4100 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
4102 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4103 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4104 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
4105 "resolved to \"\\x{17F}\\x{17F}\".");
4110 if (memBEGINs((char *) p, e - p, LONG_S_T))
4112 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4113 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4114 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
4115 "resolved to \"\\x{FB06}\".");
4116 goto return_ligature_st;
4119 #if UNICODE_MAJOR_VERSION == 3 \
4120 && UNICODE_DOT_VERSION == 0 \
4121 && UNICODE_DOT_DOT_VERSION == 1
4122 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
4124 /* And special case this on this Unicode version only, for the same
4125 * reaons the other two are special cased. They would cross the
4126 * 255/256 boundary which is forbidden under /l, and so the code
4127 * wouldn't catch that they are equivalent (which they are only in
4129 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
4130 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4131 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4132 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
4133 "resolved to \"\\x{0131}\".");
4134 goto return_dotless_i;
4138 return check_locale_boundary_crossing(p, result, ustrp, lenp);
4140 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
4144 /* This is called when changing the case of a UTF-8-encoded
4145 * character above the ASCII range, and the result should not
4146 * contain an ASCII character. */
4148 UV original; /* To store the first code point of <p> */
4150 /* Look at every character in the result; if any cross the
4151 * boundary, the whole thing is disallowed */
4153 U8* e = ustrp + *lenp;
4156 /* Crossed, have to return the original */
4157 original = valid_utf8_to_uvchr(p, lenp);
4159 /* But in these instances, there is an alternative we can
4160 * return that is valid */
4161 if (original == LATIN_SMALL_LETTER_SHARP_S
4162 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
4163 || original == LATIN_CAPITAL_LETTER_SHARP_S
4168 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
4169 goto return_ligature_st;
4171 #if UNICODE_MAJOR_VERSION == 3 \
4172 && UNICODE_DOT_VERSION == 0 \
4173 && UNICODE_DOT_DOT_VERSION == 1
4175 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4176 goto return_dotless_i;
4179 Copy(p, ustrp, *lenp, char);
4185 /* Here, no characters crossed, result is ok as-is */
4190 /* Here, used locale rules. Convert back to UTF-8 */
4191 if (UTF8_IS_INVARIANT(result)) {
4192 *ustrp = (U8) result;
4196 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
4197 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
4204 /* Certain folds to 'ss' are prohibited by the options, but they do allow
4205 * folds to a string of two of these characters. By returning this
4206 * instead, then, e.g.,
4207 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
4210 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
4211 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
4213 return LATIN_SMALL_LETTER_LONG_S;
4216 /* Two folds to 'st' are prohibited by the options; instead we pick one and
4217 * have the other one fold to it */
4219 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
4220 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
4221 return LATIN_SMALL_LIGATURE_ST;
4223 #if UNICODE_MAJOR_VERSION == 3 \
4224 && UNICODE_DOT_VERSION == 0 \
4225 && UNICODE_DOT_DOT_VERSION == 1
4228 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
4229 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
4230 return LATIN_SMALL_LETTER_DOTLESS_I;
4237 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
4238 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
4239 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
4243 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4244 I32 minbits, I32 none)
4246 /* Returns a copy of a swash initiated by the called function. This is the
4247 * public interface, and returning a copy prevents others from doing
4248 * mischief on the original. The only remaining use of this is in tr/// */
4250 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
4251 * use the following define */
4253 #define SWASH_INIT_RETURN(x) \
4254 PL_curpm= old_PL_curpm; \
4257 /* Initialize and return a swash, creating it if necessary. It does this
4258 * by calling utf8_heavy.pl in the general case.
4260 * pkg is the name of the package that <name> should be in.
4261 * name is the name of the swash to find.
4262 * listsv is a string to initialize the swash with. It must be of the form
4263 * documented as the subroutine return value in
4264 * L<perlunicode/User-Defined Character Properties>
4265 * minbits is the number of bits required to represent each data element.
4266 * none I (khw) do not understand this one, but it is used only in tr///.
4268 * Thus there are two possible inputs to find the swash: <name> and
4269 * <listsv>. At least one must be specified. The result
4270 * will be the union of the specified ones, although <listsv>'s various
4271 * actions can intersect, etc. what <name> gives. To avoid going out to
4272 * disk at all, <invlist> should specify completely what the swash should
4273 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
4276 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
4278 SV* retval = &PL_sv_undef;
4280 PERL_ARGS_ASSERT_SWASH_INIT;
4282 assert(listsv != &PL_sv_undef || strNE(name, ""));
4284 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
4285 regex that triggered the swash init and the swash init
4286 perl logic itself. See perl #122747 */
4288 /* If data was passed in to go out to utf8_heavy to find the swash of, do
4290 if (listsv != &PL_sv_undef || strNE(name, "")) {
4292 const size_t pkg_len = strlen(pkg);
4293 const size_t name_len = strlen(name);
4294 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
4299 PUSHSTACKi(PERLSI_MAGIC);
4303 /* We might get here via a subroutine signature which uses a utf8
4304 * parameter name, at which point PL_subname will have been set
4305 * but not yet used. */
4306 save_item(PL_subname);
4307 if (PL_parser && PL_parser->error_count)
4308 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
4309 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4310 if (!method) { /* demand load UTF-8 */
4312 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4313 GvSV(PL_errgv) = NULL;
4314 #ifndef NO_TAINT_SUPPORT
4315 /* It is assumed that callers of this routine are not passing in
4316 * any user derived data. */
4317 /* Need to do this after save_re_context() as it will set
4318 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
4319 * in Perl_magic_get). Even line to create errsv_save can turn on
4321 SAVEBOOL(TAINT_get);
4324 require_pv("utf8_heavy.pl");
4326 /* Not ERRSV, as there is no need to vivify a scalar we are
4327 about to discard. */
4328 SV * const errsv = GvSV(PL_errgv);
4329 if (!SvTRUE(errsv)) {
4330 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4331 SvREFCNT_dec(errsv);
4339 mPUSHp(pkg, pkg_len);
4340 mPUSHp(name, name_len);
4345 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4346 GvSV(PL_errgv) = NULL;
4347 /* If we already have a pointer to the method, no need to use
4348 * call_method() to repeat the lookup. */
4350 ? call_sv(MUTABLE_SV(method), G_SCALAR)
4351 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
4353 retval = *PL_stack_sp--;
4354 SvREFCNT_inc(retval);
4357 /* Not ERRSV. See above. */
4358 SV * const errsv = GvSV(PL_errgv);
4359 if (!SvTRUE(errsv)) {
4360 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4361 SvREFCNT_dec(errsv);
4366 if (IN_PERL_COMPILETIME) {
4367 CopHINTS_set(PL_curcop, PL_hints);
4369 } /* End of calling the module to find the swash */
4371 SWASH_INIT_RETURN(retval);
4372 #undef SWASH_INIT_RETURN
4376 /* This API is wrong for special case conversions since we may need to
4377 * return several Unicode characters for a single Unicode character
4378 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
4379 * the lower-level routine, and it is similarly broken for returning
4380 * multiple values. --jhi
4381 * For those, you should use S__to_utf8_case() instead */
4382 /* Now SWASHGET is recasted into S_swatch_get in this file. */
4385 * Returns the value of property/mapping C<swash> for the first character
4386 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4387 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
4388 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
4390 * A "swash" is a hash which contains initially the keys/values set up by
4391 * SWASHNEW. The purpose is to be able to completely represent a Unicode
4392 * property for all possible code points. Things are stored in a compact form
4393 * (see utf8_heavy.pl) so that calculation is required to find the actual
4394 * property value for a given code point. As code points are looked up, new
4395 * key/value pairs are added to the hash, so that the calculation doesn't have
4396 * to ever be re-done. Further, each calculation is done, not just for the
4397 * desired one, but for a whole block of code points adjacent to that one.
4398 * For binary properties on ASCII machines, the block is usually for 64 code
4399 * points, starting with a code point evenly divisible by 64. Thus if the
4400 * property value for code point 257 is requested, the code goes out and
4401 * calculates the property values for all 64 code points between 256 and 319,
4402 * and stores these as a single 64-bit long bit vector, called a "swatch",
4403 * under the key for code point 256. The key is the UTF-8 encoding for code
4404 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
4405 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
4406 * for code point 258 is then requested, this code realizes that it would be
4407 * stored under the key for 256, and would find that value and extract the
4408 * relevant bit, offset from 256.
4410 * Non-binary properties are stored in as many bits as necessary to represent
4411 * their values (32 currently, though the code is more general than that), not
4412 * as single bits, but the principle is the same: the value for each key is a
4413 * vector that encompasses the property values for all code points whose UTF-8
4414 * representations are represented by the key. That is, for all code points
4415 * whose UTF-8 representations are length N bytes, and the key is the first N-1
4419 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
4421 HV *const hv = MUTABLE_HV(SvRV(swash));
4426 const U8 *tmps = NULL;
4430 PERL_ARGS_ASSERT_SWASH_FETCH;
4432 /* If it really isn't a hash, it isn't really swash; must be an inversion
4434 if (SvTYPE(hv) != SVt_PVHV) {
4435 return _invlist_contains_cp((SV*)hv,
4437 ? valid_utf8_to_uvchr(ptr, NULL)
4441 /* We store the values in a "swatch" which is a vec() value in a swash
4442 * hash. Code points 0-255 are a single vec() stored with key length
4443 * (klen) 0. All other code points have a UTF-8 representation
4444 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
4445 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
4446 * length for them is the length of the encoded char - 1. ptr[klen] is the
4447 * final byte in the sequence representing the character */
4448 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
4453 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
4456 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
4459 klen = UTF8SKIP(ptr) - 1;
4461 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
4462 * the vec is the final byte in the sequence. (In EBCDIC this is
4463 * converted to I8 to get consecutive values.) To help you visualize
4465 * Straight 1047 After final byte
4466 * UTF-8 UTF-EBCDIC I8 transform
4467 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
4468 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
4470 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
4471 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
4473 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
4474 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
4476 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
4477 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
4479 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
4480 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
4482 * (There are no discontinuities in the elided (...) entries.)
4483 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
4484 * key for the next 31, up through U+043F, whose UTF-8 final byte is
4485 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
4486 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
4487 * index into the vec() swatch (after subtracting 0x80, which we
4488 * actually do with an '&').
4489 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
4490 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4491 * dicontinuities which go away by transforming it into I8, and we
4492 * effectively subtract 0xA0 to get the index. */
4493 needents = (1 << UTF_ACCUMULATION_SHIFT);
4494 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4498 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4499 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4500 * it's nothing to sniff at.) Pity we usually come through at least
4501 * two function calls to get here...
4503 * NB: this code assumes that swatches are never modified, once generated!
4506 if (hv == PL_last_swash_hv &&
4507 klen == PL_last_swash_klen &&
4508 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4510 tmps = PL_last_swash_tmps;
4511 slen = PL_last_swash_slen;
4514 /* Try our second-level swatch cache, kept in a hash. */
4515 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4517 /* If not cached, generate it via swatch_get */
4518 if (!svp || !SvPOK(*svp)
4519 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4522 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4523 swatch = swatch_get(swash,
4524 code_point & ~((UV)needents - 1),
4527 else { /* For the first 256 code points, the swatch has a key of
4529 swatch = swatch_get(swash, 0, needents);
4532 if (IN_PERL_COMPILETIME)
4533 CopHINTS_set(PL_curcop, PL_hints);
4535 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4537 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4538 || (slen << 3) < needents)
4539 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4540 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4541 svp, tmps, (UV)slen, (UV)needents);
4544 PL_last_swash_hv = hv;
4545 assert(klen <= sizeof(PL_last_swash_key));
4546 PL_last_swash_klen = (U8)klen;
4547 /* FIXME change interpvar.h? */
4548 PL_last_swash_tmps = (U8 *) tmps;
4549 PL_last_swash_slen = slen;
4551 Copy(ptr, PL_last_swash_key, klen, U8);
4554 switch ((int)((slen << 3) / needents)) {
4556 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4558 return ((UV) tmps[off]);
4562 ((UV) tmps[off ] << 8) +
4563 ((UV) tmps[off + 1]);
4567 ((UV) tmps[off ] << 24) +
4568 ((UV) tmps[off + 1] << 16) +
4569 ((UV) tmps[off + 2] << 8) +
4570 ((UV) tmps[off + 3]);
4572 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4573 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4574 NORETURN_FUNCTION_END;
4577 /* Read a single line of the main body of the swash input text. These are of
4580 * where each number is hex. The first two numbers form the minimum and
4581 * maximum of a range, and the third is the value associated with the range.
4582 * Not all swashes should have a third number
4584 * On input: l points to the beginning of the line to be examined; it points
4585 * to somewhere in the string of the whole input text, and is
4586 * terminated by a \n or the null string terminator.
4587 * lend points to the null terminator of that string
4588 * wants_value is non-zero if the swash expects a third number
4589 * typestr is the name of the swash's mapping, like 'ToLower'
4590 * On output: *min, *max, and *val are set to the values read from the line.
4591 * returns a pointer just beyond the line examined. If there was no
4592 * valid min number on the line, returns lend+1
4596 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4597 const bool wants_value, const U8* const typestr)
4599 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4600 STRLEN numlen; /* Length of the number */
4601 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4602 | PERL_SCAN_DISALLOW_PREFIX
4603 | PERL_SCAN_SILENT_NON_PORTABLE;
4605 /* nl points to the next \n in the scan */
4606 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4608 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4610 /* Get the first number on the line: the range minimum */
4612 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4613 *max = *min; /* So can never return without setting max */
4614 if (numlen) /* If found a hex number, position past it */
4616 else if (nl) { /* Else, go handle next line, if any */
4617 return nl + 1; /* 1 is length of "\n" */
4619 else { /* Else, no next line */
4620 return lend + 1; /* to LIST's end at which \n is not found */
4623 /* The max range value follows, separated by a BLANK */
4626 flags = PERL_SCAN_SILENT_ILLDIGIT
4627 | PERL_SCAN_DISALLOW_PREFIX
4628 | PERL_SCAN_SILENT_NON_PORTABLE;
4630 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4633 else /* If no value here, it is a single element range */
4636 /* Non-binary tables have a third entry: what the first element of the
4637 * range maps to. The map for those currently read here is in hex */
4641 flags = PERL_SCAN_SILENT_ILLDIGIT
4642 | PERL_SCAN_DISALLOW_PREFIX
4643 | PERL_SCAN_SILENT_NON_PORTABLE;
4645 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4654 /* diag_listed_as: To%s: illegal mapping '%s' */
4655 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4661 *val = 0; /* bits == 1, then any val should be ignored */
4663 else { /* Nothing following range min, should be single element with no
4668 /* diag_listed_as: To%s: illegal mapping '%s' */
4669 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4673 *val = 0; /* bits == 1, then val should be ignored */
4676 /* Position to next line if any, or EOF */
4686 * Returns a swatch (a bit vector string) for a code point sequence
4687 * that starts from the value C<start> and comprises the number C<span>.
4688 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4689 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4692 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4695 U8 *l, *lend, *x, *xend, *s;
4696 STRLEN lcur, xcur, scur;
4697 HV *const hv = MUTABLE_HV(SvRV(swash));
4699 SV** listsvp = NULL; /* The string containing the main body of the table */
4700 SV** extssvp = NULL;
4703 STRLEN octets; /* if bits == 1, then octets == 0 */
4705 UV end = start + span;
4707 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4708 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4709 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4710 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4711 listsvp = hv_fetchs(hv, "LIST", FALSE);
4713 bits = SvUV(*bitssvp);
4714 none = SvUV(*nonesvp);
4715 typestr = (U8*)SvPV_nolen(*typesvp);
4716 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4718 PERL_ARGS_ASSERT_SWATCH_GET;
4720 if (bits != 8 && bits != 16 && bits != 32) {
4721 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4725 /* If overflowed, use the max possible */
4731 /* create and initialize $swatch */
4732 scur = octets ? (span * octets) : (span + 7) / 8;
4733 swatch = newSV(scur);
4735 s = (U8*)SvPVX(swatch);
4736 if (octets && none) {
4737 const U8* const e = s + scur;
4740 *s++ = (U8)(none & 0xff);
4741 else if (bits == 16) {
4742 *s++ = (U8)((none >> 8) & 0xff);
4743 *s++ = (U8)( none & 0xff);
4745 else if (bits == 32) {
4746 *s++ = (U8)((none >> 24) & 0xff);
4747 *s++ = (U8)((none >> 16) & 0xff);
4748 *s++ = (U8)((none >> 8) & 0xff);
4749 *s++ = (U8)( none & 0xff);
4755 (void)memzero((U8*)s, scur + 1);
4757 SvCUR_set(swatch, scur);
4758 s = (U8*)SvPVX(swatch);
4760 /* read $swash->{LIST} */
4761 l = (U8*)SvPV(*listsvp, lcur);
4764 UV min = 0, max = 0, val = 0, upper;
4765 l = swash_scan_list_line(l, lend, &min, &max, &val,
4766 cBOOL(octets), typestr);
4771 /* If looking for something beyond this range, go try the next one */
4775 /* <end> is generally 1 beyond where we want to set things, but at the
4776 * platform's infinity, where we can't go any higher, we want to
4777 * include the code point at <end> */
4780 : (max != UV_MAX || end != UV_MAX)
4787 if (!none || val < none) {
4792 for (key = min; key <= upper; key++) {
4794 /* offset must be non-negative (start <= min <= key < end) */
4795 offset = octets * (key - start);
4797 s[offset] = (U8)(val & 0xff);
4798 else if (bits == 16) {
4799 s[offset ] = (U8)((val >> 8) & 0xff);
4800 s[offset + 1] = (U8)( val & 0xff);
4802 else if (bits == 32) {
4803 s[offset ] = (U8)((val >> 24) & 0xff);
4804 s[offset + 1] = (U8)((val >> 16) & 0xff);
4805 s[offset + 2] = (U8)((val >> 8) & 0xff);
4806 s[offset + 3] = (U8)( val & 0xff);
4809 if (!none || val < none)
4815 /* read $swash->{EXTRAS} */
4816 x = (U8*)SvPV(*extssvp, xcur);
4824 SV **otherbitssvp, *other;
4828 const U8 opc = *x++;
4832 nl = (U8*)memchr(x, '\n', xend - x);
4834 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4836 x = nl + 1; /* 1 is length of "\n" */
4840 x = xend; /* to EXTRAS' end at which \n is not found */
4847 namelen = nl - namestr;
4851 namelen = xend - namestr;
4855 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4856 otherhv = MUTABLE_HV(SvRV(*othersvp));
4857 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4858 otherbits = (STRLEN)SvUV(*otherbitssvp);
4859 if (bits < otherbits)
4860 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4861 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4863 /* The "other" swatch must be destroyed after. */
4864 other = swatch_get(*othersvp, start, span);
4865 o = (U8*)SvPV(other, olen);
4868 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4870 s = (U8*)SvPV(swatch, slen);
4872 STRLEN otheroctets = otherbits >> 3;
4874 U8* const send = s + slen;
4879 if (otherbits == 1) {
4880 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4884 STRLEN vlen = otheroctets;
4892 if (opc == '+' && otherval)
4893 NOOP; /* replace with otherval */
4894 else if (opc == '!' && !otherval)
4896 else if (opc == '-' && otherval)
4898 else if (opc == '&' && !otherval)
4901 s += octets; /* no replacement */
4906 *s++ = (U8)( otherval & 0xff);
4907 else if (bits == 16) {
4908 *s++ = (U8)((otherval >> 8) & 0xff);
4909 *s++ = (U8)( otherval & 0xff);
4911 else if (bits == 32) {
4912 *s++ = (U8)((otherval >> 24) & 0xff);
4913 *s++ = (U8)((otherval >> 16) & 0xff);
4914 *s++ = (U8)((otherval >> 8) & 0xff);
4915 *s++ = (U8)( otherval & 0xff);
4919 sv_free(other); /* through with it! */
4925 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4927 /* May change: warns if surrogates, non-character code points, or
4928 * non-Unicode code points are in 's' which has length 'len' bytes.
4929 * Returns TRUE if none found; FALSE otherwise. The only other validity
4930 * check is to make sure that this won't exceed the string's length nor
4933 const U8* const e = s + len;
4936 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4939 if (UTF8SKIP(s) > len) {
4940 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4941 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4944 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4945 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4946 if ( ckWARN_d(WARN_NON_UNICODE)
4947 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
4948 0 /* Don't consider overlongs */
4951 /* A side effect of this function will be to warn */
4952 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4956 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4957 if (ckWARN_d(WARN_SURROGATE)) {
4958 /* This has a different warning than the one the called
4959 * function would output, so can't just call it, unlike we
4960 * do for the non-chars and above-unicodes */
4961 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4962 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4963 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4968 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4969 && (ckWARN_d(WARN_NONCHAR)))
4971 /* A side effect of this function will be to warn */
4972 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4983 =for apidoc pv_uni_display
4985 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4986 length C<len>, the displayable version being at most C<pvlim> bytes long
4987 (if longer, the rest is truncated and C<"..."> will be appended).
4989 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4990 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4991 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4992 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4993 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4994 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4996 The pointer to the PV of the C<dsv> is returned.
4998 See also L</sv_uni_display>.
5002 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
5008 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5012 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5014 /* This serves double duty as a flag and a character to print after
5015 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5019 if (pvlim && SvCUR(dsv) >= pvlim) {
5023 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5025 const unsigned char c = (unsigned char)u & 0xFF;
5026 if (flags & UNI_DISPLAY_BACKSLASH) {
5043 const char string = ok;
5044 sv_catpvs(dsv, "\\");
5045 sv_catpvn(dsv, &string, 1);
5048 /* isPRINT() is the locale-blind version. */
5049 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5050 const char string = c;
5051 sv_catpvn(dsv, &string, 1);
5056 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5059 sv_catpvs(dsv, "...");
5065 =for apidoc sv_uni_display
5067 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5068 the displayable version being at most C<pvlim> bytes long
5069 (if longer, the rest is truncated and "..." will be appended).
5071 The C<flags> argument is as in L</pv_uni_display>().
5073 The pointer to the PV of the C<dsv> is returned.
5078 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5080 const char * const ptr =
5081 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5083 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5085 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5086 SvCUR(ssv), pvlim, flags);
5090 =for apidoc foldEQ_utf8
5092 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
5093 both of which may be in UTF-8) are the same case-insensitively; false
5094 otherwise. How far into the strings to compare is determined by other input
5097 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5098 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
5099 C<u2> with respect to C<s2>.
5101 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
5102 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
5103 The scan will not be considered to be a match unless the goal is reached, and
5104 scanning won't continue past that goal. Correspondingly for C<l2> with respect
5107 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
5108 pointer is considered an end pointer to the position 1 byte past the maximum
5109 point in C<s1> beyond which scanning will not continue under any circumstances.
5110 (This routine assumes that UTF-8 encoded input strings are not malformed;
5111 malformed input can cause it to read past C<pe1>). This means that if both
5112 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
5113 will never be successful because it can never
5114 get as far as its goal (and in fact is asserted against). Correspondingly for
5115 C<pe2> with respect to C<s2>.
5117 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5118 C<l2> must be non-zero), and if both do, both have to be
5119 reached for a successful match. Also, if the fold of a character is multiple
5120 characters, all of them must be matched (see tr21 reference below for
5123 Upon a successful match, if C<pe1> is non-C<NULL>,
5124 it will be set to point to the beginning of the I<next> character of C<s1>
5125 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5127 For case-insensitiveness, the "casefolding" of Unicode is used
5128 instead of upper/lowercasing both the characters, see
5129 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5133 /* A flags parameter has been added which may change, and hence isn't
5134 * externally documented. Currently it is:
5135 * 0 for as-documented above
5136 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5137 ASCII one, to not match
5138 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5139 * locale are to be used.
5140 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5141 * routine. This allows that step to be skipped.
5142 * Currently, this requires s1 to be encoded as UTF-8
5143 * (u1 must be true), which is asserted for.
5144 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5145 * cross certain boundaries. Hence, the caller should
5146 * let this function do the folding instead of
5147 * pre-folding. This code contains an assertion to
5148 * that effect. However, if the caller knows what
5149 * it's doing, it can pass this flag to indicate that,
5150 * and the assertion is skipped.
5151 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
5152 * to s2, and s2 doesn't have to be UTF-8 encoded.
5153 * This introduces an asymmetry to save a few branches
5154 * in a loop. Currently, this is not a problem, as
5155 * never are both inputs pre-folded. Simply call this
5156 * function with the pre-folded one as the second
5158 * FOLDEQ_S2_FOLDS_SANE
5161 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
5162 const char *s2, char **pe2, UV l2, bool u2,
5165 const U8 *p1 = (const U8*)s1; /* Point to current char */
5166 const U8 *p2 = (const U8*)s2;
5167 const U8 *g1 = NULL; /* goal for s1 */
5168 const U8 *g2 = NULL;
5169 const U8 *e1 = NULL; /* Don't scan s1 past this */
5170 U8 *f1 = NULL; /* Point to current folded */
5171 const U8 *e2 = NULL;
5173 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5174 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5175 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5176 U8 flags_for_folder = FOLD_FLAGS_FULL;
5178 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5180 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5181 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
5182 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5183 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
5184 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5185 /* The algorithm is to trial the folds without regard to the flags on
5186 * the first line of the above assert(), and then see if the result
5187 * violates them. This means that the inputs can't be pre-folded to a
5188 * violating result, hence the assert. This could be changed, with the
5189 * addition of extra tests here for the already-folded case, which would
5190 * slow it down. That cost is more than any possible gain for when these
5191 * flags are specified, as the flags indicate /il or /iaa matching which
5192 * is less common than /iu, and I (khw) also believe that real-world /il
5193 * and /iaa matches are most likely to involve code points 0-255, and this
5194 * function only under rare conditions gets called for 0-255. */
5196 if (flags & FOLDEQ_LOCALE) {
5197 if (IN_UTF8_CTYPE_LOCALE) {
5198 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
5199 flags_for_folder |= FOLD_FLAGS_LOCALE;
5202 flags &= ~FOLDEQ_LOCALE;
5206 flags_for_folder |= FOLD_FLAGS_LOCALE;
5209 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
5210 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
5218 g1 = (const U8*)s1 + l1;
5226 g2 = (const U8*)s2 + l2;
5229 /* Must have at least one goal */
5234 /* Will never match if goal is out-of-bounds */
5235 assert(! e1 || e1 >= g1);
5237 /* Here, there isn't an end pointer, or it is beyond the goal. We
5238 * only go as far as the goal */
5242 assert(e1); /* Must have an end for looking at s1 */
5245 /* Same for goal for s2 */
5247 assert(! e2 || e2 >= g2);
5254 /* If both operands are already folded, we could just do a memEQ on the
5255 * whole strings at once, but it would be better if the caller realized
5256 * this and didn't even call us */
5258 /* Look through both strings, a character at a time */
5259 while (p1 < e1 && p2 < e2) {
5261 /* If at the beginning of a new character in s1, get its fold to use
5262 * and the length of the fold. */
5264 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5270 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5272 /* We have to forbid mixing ASCII with non-ASCII if the
5273 * flags so indicate. And, we can short circuit having to
5274 * call the general functions for this common ASCII case,
5275 * all of whose non-locale folds are also ASCII, and hence
5276 * UTF-8 invariants, so the UTF8ness of the strings is not
5278 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5282 *foldbuf1 = toFOLD(*p1);
5285 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5287 else { /* Not UTF-8, get UTF-8 fold */
5288 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5294 if (n2 == 0) { /* Same for s2 */
5295 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5297 /* Point to the already-folded character. But for non-UTF-8
5298 * variants, convert to UTF-8 for the algorithm below */
5299 if (UTF8_IS_INVARIANT(*p2)) {
5308 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
5309 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
5315 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5316 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5320 *foldbuf2 = toFOLD(*p2);
5323 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5326 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5332 /* Here f1 and f2 point to the beginning of the strings to compare.
5333 * These strings are the folds of the next character from each input
5334 * string, stored in UTF-8. */
5336 /* While there is more to look for in both folds, see if they
5337 * continue to match */
5339 U8 fold_length = UTF8SKIP(f1);
5340 if (fold_length != UTF8SKIP(f2)
5341 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5342 function call for single
5344 || memNE((char*)f1, (char*)f2, fold_length))
5346 return 0; /* mismatch */
5349 /* Here, they matched, advance past them */
5356 /* When reach the end of any fold, advance the input past it */
5358 p1 += u1 ? UTF8SKIP(p1) : 1;
5361 p2 += u2 ? UTF8SKIP(p2) : 1;
5363 } /* End of loop through both strings */
5365 /* A match is defined by each scan that specified an explicit length
5366 * reaching its final goal, and the other not having matched a partial
5367 * character (which can happen when the fold of a character is more than one
5369 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5373 /* Successful match. Set output pointers */
5383 /* XXX The next two functions should likely be moved to mathoms.c once all
5384 * occurrences of them are removed from the core; some cpan-upstream modules
5388 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5390 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5392 return uvoffuni_to_utf8_flags(d, uv, 0);
5396 =for apidoc utf8n_to_uvuni
5398 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5400 This function was useful for code that wanted to handle both EBCDIC and
5401 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5402 distinctions between the platforms have mostly been made invisible to most
5403 code, so this function is quite unlikely to be what you want. If you do need
5404 this precise functionality, use instead
5405 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5406 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5412 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5414 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5416 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5420 =for apidoc uvuni_to_utf8_flags
5422 Instead you almost certainly want to use L</uvchr_to_utf8> or
5423 L</uvchr_to_utf8_flags>.
5425 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5426 which itself, while not deprecated, should be used only in isolated
5427 circumstances. These functions were useful for code that wanted to handle
5428 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5429 v5.20, the distinctions between the platforms have mostly been made invisible
5430 to most code, so this function is quite unlikely to be what you want.
5436 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5438 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5440 return uvoffuni_to_utf8_flags(d, uv, flags);
5444 =for apidoc utf8_to_uvchr
5446 Returns the native code point of the first character in the string C<s>
5447 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
5448 length, in bytes, of that character.
5450 Some, but not all, UTF-8 malformations are detected, and in fact, some
5451 malformed input could cause reading beyond the end of the input buffer, which
5452 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
5454 If C<s> points to one of the detected malformations, and UTF8 warnings are
5455 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
5456 C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or
5457 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
5458 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
5459 next possible position in C<s> that could begin a non-malformed character.
5460 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
5466 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
5468 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
5470 /* This function is unsafe if malformed UTF-8 input is given it, which is
5471 * why the function is deprecated. If the first byte of the input
5472 * indicates that there are more bytes remaining in the sequence that forms
5473 * the character than there are in the input buffer, it can read past the
5474 * end. But we can make it safe if the input string happens to be
5475 * NUL-terminated, as many strings in Perl are, by refusing to read past a
5476 * NUL. A NUL indicates the start of the next character anyway. If the
5477 * input isn't NUL-terminated, the function remains unsafe, as it always
5480 * An initial NUL has to be handled separately, but all ASCIIs can be
5481 * handled the same way, speeding up this common case */
5483 if (UTF8_IS_INVARIANT(*s)) { /* Assumes 's' contains at least 1 byte */
5487 return utf8_to_uvchr_buf(s,
5488 s + my_strnlen((char *) s, UTF8SKIP(s)),
5493 * ex: set ts=8 sts=4 sw=4 et: