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)";
41 =head1 Unicode Support
42 These are various utility functions for manipulating UTF8-encoded
43 strings. For the uninitiated, this is a method of representing arbitrary
44 Unicode characters as a variable number of bytes, in such a way that
45 characters in the ASCII range are unmodified, and a zero byte never appears
46 within non-zero characters.
51 /* helper for Perl__force_out_malformed_utf8_message(). Like
52 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
56 S_restore_cop_warnings(pTHX_ void *p)
58 free_and_set_cop_warnings(PL_curcop, (STRLEN*) p);
63 Perl__force_out_malformed_utf8_message(pTHX_
64 const U8 *const p, /* First byte in UTF-8 sequence */
65 const U8 * const e, /* Final byte in sequence (may include
67 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
68 usually 0, or some DISALLOW flags */
69 const bool die_here) /* If TRUE, this function does not return */
71 /* This core-only function is to be called when a malformed UTF-8 character
72 * is found, in order to output the detailed information about the
73 * malformation before dieing. The reason it exists is for the occasions
74 * when such a malformation is fatal, but warnings might be turned off, so
75 * that normally they would not be actually output. This ensures that they
76 * do get output. Because a sequence may be malformed in more than one
77 * way, multiple messages may be generated, so we can't make them fatal, as
78 * that would cause the first one to die.
80 * Instead we pretend -W was passed to perl, then die afterwards. The
81 * flexibility is here to return to the caller so they can finish up and
85 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
91 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
93 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
94 * than PL_compiling */
95 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
96 (void*)PL_curcop->cop_warnings);
97 PL_curcop->cop_warnings = pWARN_ALL;
100 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
105 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
106 " be called only when there are errors found");
110 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
115 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
116 U32 categories, /* Packed warning categories */
117 U32 flag) /* Flag associated with this message */
119 /* Creates, populates, and returns an HV* that describes an error message
120 * for the translators between UTF8 and code point */
122 SV* msg_sv = newSVpv(message, 0);
123 SV* category_sv = newSVuv(categories);
124 SV* flag_bit_sv = newSVuv(flag);
126 HV* msg_hv = newHV();
128 PERL_ARGS_ASSERT_NEW_MSG_HV;
130 (void) hv_stores(msg_hv, "text", msg_sv);
131 (void) hv_stores(msg_hv, "warn_categories", category_sv);
132 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
138 =for apidoc uvoffuni_to_utf8_flags
140 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
141 Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
142 L<perlapi/uvchr_to_utf8_flags>>.
144 This function is like them, but the input is a strict Unicode
145 (as opposed to native) code point. Only in very rare circumstances should code
146 not be using the native code point.
148 For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
154 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
156 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
158 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
161 /* All these formats take a single UV code point argument */
162 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
163 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
164 " is not recommended for open interchange";
165 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
166 " may not be portable";
168 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
170 if (flags & UNICODE_WARN_SURROGATE) { \
171 U32 category = packWARN(WARN_SURROGATE); \
172 const char * format = surrogate_cp_format; \
174 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
176 UNICODE_GOT_SURROGATE); \
179 Perl_ck_warner_d(aTHX_ category, format, uv); \
182 if (flags & UNICODE_DISALLOW_SURROGATE) { \
187 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
189 if (flags & UNICODE_WARN_NONCHAR) { \
190 U32 category = packWARN(WARN_NONCHAR); \
191 const char * format = nonchar_cp_format; \
193 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
195 UNICODE_GOT_NONCHAR); \
198 Perl_ck_warner_d(aTHX_ category, format, uv); \
201 if (flags & UNICODE_DISALLOW_NONCHAR) { \
206 /* Use shorter names internally in this file */
207 #define SHIFT UTF_ACCUMULATION_SHIFT
209 #define MARK UTF_CONTINUATION_MARK
210 #define MASK UTF_CONTINUATION_MASK
213 =for apidoc uvchr_to_utf8_flags_msgs
215 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
217 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
219 This function is for code that wants any warning and/or error messages to be
220 returned to the caller rather than be displayed. All messages that would have
221 been displayed if all lexical warnings are enabled will be returned.
223 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
224 placed after all the others, C<msgs>. If this parameter is 0, this function
225 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
226 be a pointer to an C<HV *> variable, in which this function creates a new HV to
227 contain any appropriate messages. The hash has three key-value pairs, as
234 The text of the message as a C<SVpv>.
236 =item C<warn_categories>
238 The warning category (or categories) packed into a C<SVuv>.
242 A single flag bit associated with this message, in a C<SVuv>.
243 The bit corresponds to some bit in the C<*errors> return value,
244 such as C<UNICODE_GOT_SURROGATE>.
248 It's important to note that specifying this parameter as non-null will cause
249 any warnings this function would otherwise generate to be suppressed, and
250 instead be placed in C<*msgs>. The caller can check the lexical warnings state
251 (or not) when choosing what to do with the returned messages.
253 The caller, of course, is responsible for freeing any returned HV.
258 /* Undocumented; we don't want people using this. Instead they should use
259 * uvchr_to_utf8_flags_msgs() */
261 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
263 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
269 if (OFFUNI_IS_INVARIANT(uv)) {
270 *d++ = LATIN1_TO_NATIVE(uv);
274 if (uv <= MAX_UTF8_TWO_BYTE) {
275 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
276 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
280 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
281 * below, the 16 is for start bytes E0-EF (which are all the possible ones
282 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
283 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
284 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
285 * 0x800-0xFFFF on ASCII */
286 if (uv < (16 * (1U << (2 * SHIFT)))) {
287 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
288 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
289 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
291 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
292 aren't tested here */
293 /* The most likely code points in this range are below the surrogates.
294 * Do an extra test to quickly exclude those. */
295 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
296 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
297 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
299 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
301 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
302 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
309 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
310 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
311 * happen starting with 4-byte characters on ASCII platforms. We unify the
312 * code for these with EBCDIC, even though some of them require 5-bytes on
313 * those, because khw believes the code saving is worth the very slight
314 * performance hit on these high EBCDIC code points. */
316 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
317 if (UNLIKELY( uv > MAX_LEGAL_CP
318 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
320 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv));
322 if ( (flags & UNICODE_WARN_SUPER)
323 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
324 && UNICODE_IS_PERL_EXTENDED(uv)))
326 const char * format = super_cp_format;
327 U32 category = packWARN(WARN_NON_UNICODE);
328 U32 flag = UNICODE_GOT_SUPER;
330 /* Choose the more dire applicable warning */
331 if (UNICODE_IS_PERL_EXTENDED(uv)) {
332 format = PL_extended_cp_format;
333 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
334 if (flags & (UNICODE_WARN_PERL_EXTENDED
335 |UNICODE_DISALLOW_PERL_EXTENDED))
337 flag = UNICODE_GOT_PERL_EXTENDED;
342 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
345 else if ( ckWARN_d(WARN_NON_UNICODE)
346 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
347 && ckWARN(WARN_PORTABLE)))
349 Perl_warner(aTHX_ category, format, uv);
352 if ( (flags & UNICODE_DISALLOW_SUPER)
353 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
354 && UNICODE_IS_PERL_EXTENDED(uv)))
359 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
360 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
363 /* Test for and handle 4-byte result. In the test immediately below, the
364 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
365 * characters). The 3 is for 3 continuation bytes; these each contribute
366 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
367 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
368 * 0x1_0000-0x1F_FFFF on ASCII */
369 if (uv < (8 * (1U << (3 * SHIFT)))) {
370 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
371 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
372 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
373 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
375 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
376 characters. The end-plane non-characters for EBCDIC were
377 handled just above */
378 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
379 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
381 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
382 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
389 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
390 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
391 * format. The unrolled version above turns out to not save all that much
392 * time, and at these high code points (well above the legal Unicode range
393 * on ASCII platforms, and well above anything in common use in EBCDIC),
394 * khw believes that less code outweighs slight performance gains. */
397 STRLEN len = OFFUNISKIP(uv);
400 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
403 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
409 =for apidoc uvchr_to_utf8
411 Adds the UTF-8 representation of the native code point C<uv> to the end
412 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
413 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
414 the byte after the end of the new character. In other words,
416 d = uvchr_to_utf8(d, uv);
418 is the recommended wide native character-aware way of saying
422 This function accepts any code point from 0..C<IV_MAX> as input.
423 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
425 It is possible to forbid or warn on non-Unicode code points, or those that may
426 be problematic by using L</uvchr_to_utf8_flags>.
431 /* This is also a macro */
432 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
435 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
437 return uvchr_to_utf8(d, uv);
441 =for apidoc uvchr_to_utf8_flags
443 Adds the UTF-8 representation of the native code point C<uv> to the end
444 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
445 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
446 the byte after the end of the new character. In other words,
448 d = uvchr_to_utf8_flags(d, uv, flags);
452 d = uvchr_to_utf8_flags(d, uv, 0);
454 This is the Unicode-aware way of saying
458 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
459 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
461 Specifying C<flags> can further restrict what is allowed and not warned on, as
464 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
465 the function will raise a warning, provided UTF8 warnings are enabled. If
466 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
467 NULL. If both flags are set, the function will both warn and return NULL.
469 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
470 affect how the function handles a Unicode non-character.
472 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
473 affect the handling of code points that are above the Unicode maximum of
474 0x10FFFF. Languages other than Perl may not be able to accept files that
477 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
478 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
479 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
480 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
481 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
482 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
483 above-Unicode and surrogate flags, but not the non-character ones, as
485 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
486 See L<perlunicode/Noncharacter code points>.
488 Extremely high code points were never specified in any standard, and require an
489 extension to UTF-8 to express, which Perl does. It is likely that programs
490 written in something other than Perl would not be able to read files that
491 contain these; nor would Perl understand files written by something that uses a
492 different extension. For these reasons, there is a separate set of flags that
493 can warn and/or disallow these extremely high code points, even if other
494 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
495 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
496 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
497 treat all above-Unicode code points, including these, as malformations. (Note
498 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
499 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
501 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
502 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
503 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
504 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
505 platforms,these flags can apply to code points that actually do fit in 31 bits.
506 The new names accurately describe the situation in all cases.
511 /* This is also a macro */
512 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
515 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
517 return uvchr_to_utf8_flags(d, uv, flags);
523 S_is_utf8_cp_above_31_bits(const U8 * const s,
525 const bool consider_overlongs)
527 /* Returns TRUE if the first code point represented by the Perl-extended-
528 * UTF-8-encoded string starting at 's', and looking no further than 'e -
529 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
531 * The function handles the case where the input bytes do not include all
532 * the ones necessary to represent a full character. That is, they may be
533 * the intial bytes of the representation of a code point, but possibly
534 * the final ones necessary for the complete representation may be beyond
537 * The function also can handle the case where the input is an overlong
538 * sequence. If 'consider_overlongs' is 0, the function assumes the
539 * input is not overlong, without checking, and will return based on that
540 * assumption. If this parameter is 1, the function will go to the trouble
541 * of figuring out if it actually evaluates to above or below 31 bits.
543 * The sequence is otherwise assumed to be well-formed, without checking.
546 const STRLEN len = e - s;
549 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
551 assert(! UTF8_IS_INVARIANT(*s) && e > s);
555 PERL_UNUSED_ARG(consider_overlongs);
557 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
558 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
559 * also be the start byte for a 31-bit code point; we need at least 2
560 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
561 * the start byte for an overlong sequence, but for 30-bit or smaller code
562 * points, so we don't have to worry about overlongs on EBCDIC.) */
573 /* On ASCII, FE and FF are the only start bytes that can evaluate to
574 * needing more than 31 bits. */
575 if (LIKELY(*s < 0xFE)) {
579 /* What we have left are FE and FF. Both of these require more than 31
580 * bits unless they are for overlongs. */
581 if (! consider_overlongs) {
585 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
586 * above 31 bits. But we need more than one byte to discern this, so if
587 * passed just the start byte, it could be an overlong evaluating to
593 /* Having excluded len==1, and knowing that FE and FF are both valid start
594 * bytes, we can call the function below to see if the sequence is
595 * overlong. (We don't need the full generality of the called function,
596 * but for these huge code points, speed shouldn't be a consideration, and
597 * the compiler does have enough information, since it's static to this
598 * file, to optimize to just the needed parts.) */
599 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
601 /* If it isn't overlong, more than 31 bits are required. */
602 if (is_overlong == 0) {
606 /* If it is indeterminate if it is overlong, return that */
607 if (is_overlong < 0) {
611 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
612 * the max it can be is 2**31 - 1 */
619 /* Here, ASCII and EBCDIC rejoin:
620 * On ASCII: We have an overlong sequence starting with FF
621 * On EBCDIC: We have a sequence starting with FE. */
623 { /* For C89, use a block so the declaration can be close to its use */
627 /* U+7FFFFFFF (2 ** 31 - 1)
628 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
629 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
630 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
631 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
632 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
633 * U+80000000 (2 ** 31):
634 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
635 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
636 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
637 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
639 * and since we know that *s = \xfe, any continuation sequcence
640 * following it that is gt the below is above 31 bits
641 [0] [1] [2] [3] [4] [5] [6] */
642 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
646 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
647 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
648 * FF overlong for U+80000000 (2 ** 31):
649 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
650 * and since we know that *s = \xff, any continuation sequcence
651 * following it that is gt the below is above 30 bits
652 [0] [1] [2] [3] [4] [5] [6] */
653 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
657 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
658 const STRLEN cmp_len = MIN(conts_len, len - 1);
660 /* Now compare the continuation bytes in s with the ones we have
661 * compiled in that are for the largest 30 bit code point. If we have
662 * enough bytes available to determine the answer, or the bytes we do
663 * have differ from them, we can compare the two to get a definitive
664 * answer (Note that in UTF-EBCDIC, the two lowest possible
665 * continuation bytes are \x41 and \x42.) */
666 if (cmp_len >= conts_len || memNE(s + 1,
667 conts_for_highest_30_bit,
670 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
673 /* Here, all the bytes we have are the same as the highest 30-bit code
674 * point, but we are missing so many bytes that we can't make the
682 PERL_STATIC_INLINE int
683 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
685 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
686 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
687 * it isn't, and -1 if there isn't enough information to tell. This last
688 * return value can happen if the sequence is incomplete, missing some
689 * trailing bytes that would form a complete character. If there are
690 * enough bytes to make a definitive decision, this function does so.
691 * Usually 2 bytes sufficient.
693 * Overlongs can occur whenever the number of continuation bytes changes.
694 * That means whenever the number of leading 1 bits in a start byte
695 * increases from the next lower start byte. That happens for start bytes
696 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
697 * start bytes have already been excluded, so don't need to be tested here;
698 * ASCII platforms: C0, C1
699 * EBCDIC platforms C0, C1, C2, C3, C4, E0
702 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
703 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
705 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
706 assert(len > 1 && UTF8_IS_START(*s));
708 /* Each platform has overlongs after the start bytes given above (expressed
709 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
710 * the logic is the same, except the E0 overlong has already been excluded
711 * on EBCDIC platforms. The values below were found by manually
712 * inspecting the UTF-8 patterns. See the tables in utf8.h and
716 # define F0_ABOVE_OVERLONG 0xB0
717 # define F8_ABOVE_OVERLONG 0xA8
718 # define FC_ABOVE_OVERLONG 0xA4
719 # define FE_ABOVE_OVERLONG 0xA2
720 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
724 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
728 # define F0_ABOVE_OVERLONG 0x90
729 # define F8_ABOVE_OVERLONG 0x88
730 # define FC_ABOVE_OVERLONG 0x84
731 # define FE_ABOVE_OVERLONG 0x82
732 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
736 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
737 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
738 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
739 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
744 /* Check for the FF overlong */
745 return isFF_OVERLONG(s, len);
748 PERL_STATIC_INLINE int
749 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
751 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
752 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
753 * it isn't, and -1 if there isn't enough information to tell. This last
754 * return value can happen if the sequence is incomplete, missing some
755 * trailing bytes that would form a complete character. If there are
756 * enough bytes to make a definitive decision, this function does so. */
758 PERL_ARGS_ASSERT_ISFF_OVERLONG;
760 /* To be an FF overlong, all the available bytes must match */
761 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
762 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
767 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
768 * be there; what comes after them doesn't matter. See tables in utf8.h,
770 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
774 /* The missing bytes could cause the result to go one way or the other, so
775 * the result is indeterminate */
779 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
780 # ifdef EBCDIC /* Actually is I8 */
781 # define HIGHEST_REPRESENTABLE_UTF8 \
782 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
784 # define HIGHEST_REPRESENTABLE_UTF8 \
785 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
789 PERL_STATIC_INLINE int
790 S_does_utf8_overflow(const U8 * const s,
792 const bool consider_overlongs)
794 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
795 * 'e' - 1 would overflow an IV on this platform; that is if it represents
796 * a code point larger than the highest representable code point. It
797 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
798 * enough information to tell. This last return value can happen if the
799 * sequence is incomplete, missing some trailing bytes that would form a
800 * complete character. If there are enough bytes to make a definitive
801 * decision, this function does so.
803 * If 'consider_overlongs' is TRUE, the function checks for the possibility
804 * that the sequence is an overlong that doesn't overflow. Otherwise, it
805 * assumes the sequence is not an overlong. This can give different
806 * results only on ASCII 32-bit platforms.
808 * (For ASCII platforms, we could use memcmp() because we don't have to
809 * convert each byte to I8, but it's very rare input indeed that would
810 * approach overflow, so the loop below will likely only get executed once.)
812 * 'e' - 1 must not be beyond a full character. */
815 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
816 assert(s <= e && s + UTF8SKIP(s) >= e);
818 #if ! defined(UV_IS_QUAD)
820 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
824 PERL_UNUSED_ARG(consider_overlongs);
827 const STRLEN len = e - s;
829 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
831 for (x = s; x < e; x++, y++) {
833 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
837 /* If this byte is larger than the corresponding highest UTF-8
838 * byte, the sequence overflow; otherwise the byte is less than,
839 * and so the sequence doesn't overflow */
840 return NATIVE_UTF8_TO_I8(*x) > *y;
844 /* Got to the end and all bytes are the same. If the input is a whole
845 * character, it doesn't overflow. And if it is a partial character,
846 * there's not enough information to tell */
847 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
860 /* This is the portions of the above function that deal with UV_MAX instead of
861 * IV_MAX. They are left here in case we want to combine them so that internal
862 * uses can have larger code points. The only logic difference is that the
863 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
867 /* Anything larger than this will overflow the word if it were converted into a UV */
868 #if defined(UV_IS_QUAD)
869 # ifdef EBCDIC /* Actually is I8 */
870 # define HIGHEST_REPRESENTABLE_UTF8 \
871 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
873 # define HIGHEST_REPRESENTABLE_UTF8 \
874 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
878 # define HIGHEST_REPRESENTABLE_UTF8 \
879 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
881 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
885 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
887 /* On 32 bit ASCII machines, many overlongs that start with FF don't
889 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
891 /* To be such an overlong, the first bytes of 's' must match
892 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
893 * don't have any additional bytes available, the sequence, when
894 * completed might or might not fit in 32 bits. But if we have that
895 * next byte, we can tell for sure. If it is <= 0x83, then it does
897 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
901 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
904 /* Starting with the #else, the rest of the function is identical except
905 * 1. we need to move the 'len' declaration to be global to the function
906 * 2. the endif move to just after the UNUSED_ARG.
907 * An empty endif is given just below to satisfy the preprocessor
913 #undef F0_ABOVE_OVERLONG
914 #undef F8_ABOVE_OVERLONG
915 #undef FC_ABOVE_OVERLONG
916 #undef FE_ABOVE_OVERLONG
917 #undef FF_OVERLONG_PREFIX
920 Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
925 /* A helper function that should not be called directly.
927 * This function returns non-zero if the string beginning at 's' and
928 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
929 * code point; otherwise it returns 0. The examination stops after the
930 * first code point in 's' is validated, not looking at the rest of the
931 * input. If 'e' is such that there are not enough bytes to represent a
932 * complete code point, this function will return non-zero anyway, if the
933 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
934 * excluded by 'flags'.
936 * A non-zero return gives the number of bytes required to represent the
937 * code point. Be aware that if the input is for a partial character, the
938 * return will be larger than 'e - s'.
940 * This function assumes that the code point represented is UTF-8 variant.
941 * The caller should have excluded the possibility of it being invariant
942 * before calling this function.
944 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
945 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
946 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
947 * disallowed by the flags. If the input is only for a partial character,
948 * the function will return non-zero if there is any sequence of
949 * well-formed UTF-8 that, when appended to the input sequence, could
950 * result in an allowed code point; otherwise it returns 0. Non characters
951 * cannot be determined based on partial character input. But many of the
952 * other excluded types can be determined with just the first one or two
957 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
959 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
960 |UTF8_DISALLOW_PERL_EXTENDED)));
961 assert(! UTF8_IS_INVARIANT(*s));
963 /* A variant char must begin with a start byte */
964 if (UNLIKELY(! UTF8_IS_START(*s))) {
968 /* Examine a maximum of a single whole code point */
969 if (e - s > UTF8SKIP(s)) {
975 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
976 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
978 /* Here, we are disallowing some set of largish code points, and the
979 * first byte indicates the sequence is for a code point that could be
980 * in the excluded set. We generally don't have to look beyond this or
981 * the second byte to see if the sequence is actually for one of the
982 * excluded classes. The code below is derived from this table:
984 * UTF-8 UTF-EBCDIC I8
985 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
986 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
987 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
989 * Keep in mind that legal continuation bytes range between \x80..\xBF
990 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
991 * continuation bytes. Hence, we don't have to test the upper edge
992 * because if any of those is encountered, the sequence is malformed,
993 * and would fail elsewhere in this function.
995 * The code here likewise assumes that there aren't other
996 * malformations; again the function should fail elsewhere because of
997 * these. For example, an overlong beginning with FC doesn't actually
998 * have to be a super; it could actually represent a small code point,
999 * even U+0000. But, since overlongs (and other malformations) are
1000 * illegal, the function should return FALSE in either case.
1003 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1004 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
1005 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
1007 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1009 && ((s1) & 0xFE ) == 0xB6)
1010 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1012 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1013 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1014 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1015 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1018 if ( (flags & UTF8_DISALLOW_SUPER)
1019 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1021 return 0; /* Above Unicode */
1024 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1025 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1031 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1033 if ( (flags & UTF8_DISALLOW_SUPER)
1034 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1036 return 0; /* Above Unicode */
1039 if ( (flags & UTF8_DISALLOW_SURROGATE)
1040 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1042 return 0; /* Surrogate */
1045 if ( (flags & UTF8_DISALLOW_NONCHAR)
1046 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1048 return 0; /* Noncharacter code point */
1053 /* Make sure that all that follows are continuation bytes */
1054 for (x = s + 1; x < e; x++) {
1055 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1060 /* Here is syntactically valid. Next, make sure this isn't the start of an
1062 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1066 /* And finally, that the code point represented fits in a word on this
1068 if (0 < does_utf8_overflow(s, e,
1069 0 /* Don't consider overlongs */
1079 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1081 /* Returns a mortalized C string that is a displayable copy of the 'len'
1082 * bytes starting at 'start'. 'format' gives how to display each byte.
1083 * Currently, there are only two formats, so it is currently a bool:
1085 * 1 ab (that is a space between two hex digit bytes)
1088 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1090 const U8 * s = start;
1091 const U8 * const e = start + len;
1095 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1097 Newx(output, output_len, char);
1101 for (s = start; s < e; s++) {
1102 const unsigned high_nibble = (*s & 0xF0) >> 4;
1103 const unsigned low_nibble = (*s & 0x0F);
1115 if (high_nibble < 10) {
1116 *d++ = high_nibble + '0';
1119 *d++ = high_nibble - 10 + 'a';
1122 if (low_nibble < 10) {
1123 *d++ = low_nibble + '0';
1126 *d++ = low_nibble - 10 + 'a';
1134 PERL_STATIC_INLINE char *
1135 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1137 /* Max number of bytes to print */
1140 /* Which one is the non-continuation */
1141 const STRLEN non_cont_byte_pos,
1143 /* How many bytes should there be? */
1144 const STRLEN expect_len)
1146 /* Return the malformation warning text for an unexpected continuation
1149 const char * const where = (non_cont_byte_pos == 1)
1151 : Perl_form(aTHX_ "%d bytes",
1152 (int) non_cont_byte_pos);
1153 const U8 * x = s + non_cont_byte_pos;
1154 const U8 * e = s + print_len;
1156 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1158 /* We don't need to pass this parameter, but since it has already been
1159 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1160 assert(expect_len == UTF8SKIP(s));
1162 /* As a defensive coding measure, don't output anything past a NUL. Such
1163 * bytes shouldn't be in the middle of a malformation, and could mark the
1164 * end of the allocated string, and what comes after is undefined */
1165 for (; x < e; x++) {
1167 x++; /* Output this particular NUL */
1172 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1173 " %s after start byte 0x%02x; need %d bytes, got %d)",
1175 _byte_dump_string(s, x - s, 0),
1176 *(s + non_cont_byte_pos),
1180 (int) non_cont_byte_pos);
1185 =for apidoc utf8n_to_uvchr
1187 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1188 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1191 Bottom level UTF-8 decode routine.
1192 Returns the native code point value of the first character in the string C<s>,
1193 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1194 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1195 the length, in bytes, of that character.
1197 The value of C<flags> determines the behavior when C<s> does not point to a
1198 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1199 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1200 is the next possible position in C<s> that could begin a non-malformed
1201 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1202 is raised. Some UTF-8 input sequences may contain multiple malformations.
1203 This function tries to find every possible one in each call, so multiple
1204 warnings can be raised for the same sequence.
1206 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1207 individual types of malformations, such as the sequence being overlong (that
1208 is, when there is a shorter sequence that can express the same code point;
1209 overlong sequences are expressly forbidden in the UTF-8 standard due to
1210 potential security issues). Another malformation example is the first byte of
1211 a character not being a legal first byte. See F<utf8.h> for the list of such
1212 flags. Even if allowed, this function generally returns the Unicode
1213 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1214 F<utf8.h> to override this behavior for the overlong malformations, but don't
1215 do that except for very specialized purposes.
1217 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1218 flags) malformation is found. If this flag is set, the routine assumes that
1219 the caller will raise a warning, and this function will silently just set
1220 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1222 Note that this API requires disambiguation between successful decoding a C<NUL>
1223 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1224 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1225 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1226 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1227 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1229 Certain code points are considered problematic. These are Unicode surrogates,
1230 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1231 By default these are considered regular code points, but certain situations
1232 warrant special handling for them, which can be specified using the C<flags>
1233 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1234 three classes are treated as malformations and handled as such. The flags
1235 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1236 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1237 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1238 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1239 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1241 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
1242 The difference between traditional strictness and C9 strictness is that the
1243 latter does not forbid non-character code points. (They are still discouraged,
1244 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1246 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1247 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1248 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1249 raised for their respective categories, but otherwise the code points are
1250 considered valid (not malformations). To get a category to both be treated as
1251 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1252 (But note that warnings are not raised if lexically disabled nor if
1253 C<UTF8_CHECK_ONLY> is also specified.)
1255 Extremely high code points were never specified in any standard, and require an
1256 extension to UTF-8 to express, which Perl does. It is likely that programs
1257 written in something other than Perl would not be able to read files that
1258 contain these; nor would Perl understand files written by something that uses a
1259 different extension. For these reasons, there is a separate set of flags that
1260 can warn and/or disallow these extremely high code points, even if other
1261 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1262 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1263 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1264 above-Unicode code points, including these, as malformations.
1265 (Note that the Unicode standard considers anything above 0x10FFFF to be
1266 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1269 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1270 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1271 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1272 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1273 can apply to code points that actually do fit in 31 bits. This happens on
1274 EBCDIC platforms, and sometimes when the L<overlong
1275 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1276 describe the situation in all cases.
1279 All other code points corresponding to Unicode characters, including private
1280 use and those yet to be assigned, are never considered malformed and never
1283 =for apidoc Amnh||UTF8_CHECK_ONLY
1284 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1285 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1286 =for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1287 =for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1288 =for apidoc Amnh||UTF8_DISALLOW_SUPER
1289 =for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1290 =for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1291 =for apidoc Amnh||UTF8_WARN_SURROGATE
1292 =for apidoc Amnh||UTF8_WARN_NONCHAR
1293 =for apidoc Amnh||UTF8_WARN_SUPER
1294 =for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1295 =for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1299 Also implemented as a macro in utf8.h
1303 Perl_utf8n_to_uvchr(const U8 *s,
1308 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1310 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1315 =for apidoc utf8n_to_uvchr_error
1317 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1318 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1321 This function is for code that needs to know what the precise malformation(s)
1322 are when an error is found. If you also need to know the generated warning
1323 messages, use L</utf8n_to_uvchr_msgs>() instead.
1325 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1326 all the others, C<errors>. If this parameter is 0, this function behaves
1327 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1328 to a C<U32> variable, which this function sets to indicate any errors found.
1329 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1330 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1331 of these bits will be set if a malformation is found, even if the input
1332 C<flags> parameter indicates that the given malformation is allowed; those
1333 exceptions are noted:
1337 =item C<UTF8_GOT_PERL_EXTENDED>
1339 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1340 set only if the input C<flags> parameter contains either the
1341 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1343 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1344 and so some extension must be used to express them. Perl uses a natural
1345 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1346 extension to represent even higher ones, so that any code point that fits in a
1347 64-bit word can be represented. Text using these extensions is not likely to
1348 be portable to non-Perl code. We lump both of these extensions together and
1349 refer to them as Perl extended UTF-8. There exist other extensions that people
1350 have invented, incompatible with Perl's.
1352 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1353 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1354 than on ASCII. Prior to that, code points 2**31 and higher were simply
1355 unrepresentable, and a different, incompatible method was used to represent
1356 code points between 2**30 and 2**31 - 1.
1358 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1359 Perl extended UTF-8 is used.
1361 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1362 may use for backward compatibility. That name is misleading, as this flag may
1363 be set when the code point actually does fit in 31 bits. This happens on
1364 EBCDIC platforms, and sometimes when the L<overlong
1365 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1366 describes the situation in all cases.
1368 =item C<UTF8_GOT_CONTINUATION>
1370 The input sequence was malformed in that the first byte was a a UTF-8
1373 =item C<UTF8_GOT_EMPTY>
1375 The input C<curlen> parameter was 0.
1377 =item C<UTF8_GOT_LONG>
1379 The input sequence was malformed in that there is some other sequence that
1380 evaluates to the same code point, but that sequence is shorter than this one.
1382 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1383 it was discovered that this created security issues.
1385 =item C<UTF8_GOT_NONCHAR>
1387 The code point represented by the input UTF-8 sequence is for a Unicode
1388 non-character code point.
1389 This bit is set only if the input C<flags> parameter contains either the
1390 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1392 =item C<UTF8_GOT_NON_CONTINUATION>
1394 The input sequence was malformed in that a non-continuation type byte was found
1395 in a position where only a continuation type one should be. See also
1396 L</C<UTF8_GOT_SHORT>>.
1398 =item C<UTF8_GOT_OVERFLOW>
1400 The input sequence was malformed in that it is for a code point that is not
1401 representable in the number of bits available in an IV on the current platform.
1403 =item C<UTF8_GOT_SHORT>
1405 The input sequence was malformed in that C<curlen> is smaller than required for
1406 a complete sequence. In other words, the input is for a partial character
1410 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1411 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1412 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1413 sequence was looked at. If no other flags are present, it means that the
1414 sequence was valid as far as it went. Depending on the application, this could
1415 mean one of three things:
1421 The C<curlen> length parameter passed in was too small, and the function was
1422 prevented from examining all the necessary bytes.
1426 The buffer being looked at is based on reading data, and the data received so
1427 far stopped in the middle of a character, so that the next read will
1428 read the remainder of this character. (It is up to the caller to deal with the
1429 split bytes somehow.)
1433 This is a real error, and the partial sequence is all we're going to get.
1437 =item C<UTF8_GOT_SUPER>
1439 The input sequence was malformed in that it is for a non-Unicode code point;
1440 that is, one above the legal Unicode maximum.
1441 This bit is set only if the input C<flags> parameter contains either the
1442 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1444 =item C<UTF8_GOT_SURROGATE>
1446 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1448 This bit is set only if the input C<flags> parameter contains either the
1449 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1453 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1454 flag to suppress any warnings, and then examine the C<*errors> return.
1458 Also implemented as a macro in utf8.h
1462 Perl_utf8n_to_uvchr_error(const U8 *s,
1468 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1470 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1475 =for apidoc utf8n_to_uvchr_msgs
1477 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1478 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1481 This function is for code that needs to know what the precise malformation(s)
1482 are when an error is found, and wants the corresponding warning and/or error
1483 messages to be returned to the caller rather than be displayed. All messages
1484 that would have been displayed if all lexcial warnings are enabled will be
1487 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1488 placed after all the others, C<msgs>. If this parameter is 0, this function
1489 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1490 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1491 contain any appropriate messages. The elements of the array are ordered so
1492 that the first message that would have been displayed is in the 0th element,
1493 and so on. Each element is a hash with three key-value pairs, as follows:
1499 The text of the message as a C<SVpv>.
1501 =item C<warn_categories>
1503 The warning category (or categories) packed into a C<SVuv>.
1507 A single flag bit associated with this message, in a C<SVuv>.
1508 The bit corresponds to some bit in the C<*errors> return value,
1509 such as C<UTF8_GOT_LONG>.
1513 It's important to note that specifying this parameter as non-null will cause
1514 any warnings this function would otherwise generate to be suppressed, and
1515 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1516 (or not) when choosing what to do with the returned messages.
1518 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1521 The caller, of course, is responsible for freeing any returned AV.
1527 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1534 const U8 * const s0 = s;
1535 const U8 * send = s0 + curlen;
1536 U32 possible_problems; /* A bit is set here for each potential problem
1537 found as we go along */
1539 STRLEN expectlen; /* How long should this sequence be? */
1540 STRLEN avail_len; /* When input is too short, gives what that is */
1541 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1542 gets set and discarded */
1544 /* The below are used only if there is both an overlong malformation and a
1545 * too short one. Otherwise the first two are set to 's0' and 'send', and
1546 * the third not used at all */
1548 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1549 routine; see [perl #130921] */
1553 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1555 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1556 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1557 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1562 /* Each of the affected Hanguls starts with \xED */
1564 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1575 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1576 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1577 | (s0[2] & UTF_CONTINUATION_MASK);
1582 /* In conjunction with the exhaustive tests that can be enabled in
1583 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1584 * what it is intended to do, and that no flaws in it are masked by
1585 * dropping down and executing the code below
1586 assert(! isUTF8_CHAR(s0, send)
1587 || UTF8_IS_SURROGATE(s0, send)
1588 || UTF8_IS_SUPER(s0, send)
1589 || UTF8_IS_NONCHAR(s0,send));
1594 possible_problems = 0;
1598 adjusted_s0 = (U8 *) s0;
1605 errors = &discard_errors;
1608 /* The order of malformation tests here is important. We should consume as
1609 * few bytes as possible in order to not skip any valid character. This is
1610 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1611 * https://unicode.org/reports/tr36 for more discussion as to why. For
1612 * example, once we've done a UTF8SKIP, we can tell the expected number of
1613 * bytes, and could fail right off the bat if the input parameters indicate
1614 * that there are too few available. But it could be that just that first
1615 * byte is garbled, and the intended character occupies fewer bytes. If we
1616 * blindly assumed that the first byte is correct, and skipped based on
1617 * that number, we could skip over a valid input character. So instead, we
1618 * always examine the sequence byte-by-byte.
1620 * We also should not consume too few bytes, otherwise someone could inject
1621 * things. For example, an input could be deliberately designed to
1622 * overflow, and if this code bailed out immediately upon discovering that,
1623 * returning to the caller C<*retlen> pointing to the very next byte (one
1624 * which is actually part of of the overflowing sequence), that could look
1625 * legitimate to the caller, which could discard the initial partial
1626 * sequence and process the rest, inappropriately.
1628 * Some possible input sequences are malformed in more than one way. This
1629 * function goes to lengths to try to find all of them. This is necessary
1630 * for correctness, as the inputs may allow one malformation but not
1631 * another, and if we abandon searching for others after finding the
1632 * allowed one, we could allow in something that shouldn't have been.
1635 if (UNLIKELY(curlen == 0)) {
1636 possible_problems |= UTF8_GOT_EMPTY;
1638 uv = UNICODE_REPLACEMENT;
1639 goto ready_to_handle_errors;
1642 expectlen = UTF8SKIP(s);
1644 /* A well-formed UTF-8 character, as the vast majority of calls to this
1645 * function will be for, has this expected length. For efficiency, set
1646 * things up here to return it. It will be overriden only in those rare
1647 * cases where a malformation is found */
1649 *retlen = expectlen;
1652 /* A continuation character can't start a valid sequence */
1653 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1654 possible_problems |= UTF8_GOT_CONTINUATION;
1656 uv = UNICODE_REPLACEMENT;
1657 goto ready_to_handle_errors;
1660 /* Here is not a continuation byte, nor an invariant. The only thing left
1661 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1662 * because it excludes start bytes like \xC0 that always lead to
1665 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1666 * that indicate the number of bytes in the character's whole UTF-8
1667 * sequence, leaving just the bits that are part of the value. */
1668 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1670 /* Setup the loop end point, making sure to not look past the end of the
1671 * input string, and flag it as too short if the size isn't big enough. */
1672 if (UNLIKELY(curlen < expectlen)) {
1673 possible_problems |= UTF8_GOT_SHORT;
1677 send = (U8*) s0 + expectlen;
1680 /* Now, loop through the remaining bytes in the character's sequence,
1681 * accumulating each into the working value as we go. */
1682 for (s = s0 + 1; s < send; s++) {
1683 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1684 uv = UTF8_ACCUMULATE(uv, *s);
1688 /* Here, found a non-continuation before processing all expected bytes.
1689 * This byte indicates the beginning of a new character, so quit, even
1690 * if allowing this malformation. */
1691 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1693 } /* End of loop through the character's bytes */
1695 /* Save how many bytes were actually in the character */
1698 /* Note that there are two types of too-short malformation. One is when
1699 * there is actual wrong data before the normal termination of the
1700 * sequence. The other is that the sequence wasn't complete before the end
1701 * of the data we are allowed to look at, based on the input 'curlen'.
1702 * This means that we were passed data for a partial character, but it is
1703 * valid as far as we saw. The other is definitely invalid. This
1704 * distinction could be important to a caller, so the two types are kept
1707 * A convenience macro that matches either of the too-short conditions. */
1708 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1710 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1712 uv = UNICODE_REPLACEMENT;
1715 /* Check for overflow. The algorithm requires us to not look past the end
1716 * of the current character, even if partial, so the upper limit is 's' */
1717 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1718 1 /* Do consider overlongs */
1721 possible_problems |= UTF8_GOT_OVERFLOW;
1722 uv = UNICODE_REPLACEMENT;
1725 /* Check for overlong. If no problems so far, 'uv' is the correct code
1726 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1727 * we must look at the UTF-8 byte sequence itself to see if it is for an
1729 if ( ( LIKELY(! possible_problems)
1730 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1731 || ( UNLIKELY(possible_problems)
1732 && ( UNLIKELY(! UTF8_IS_START(*s0))
1734 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1737 possible_problems |= UTF8_GOT_LONG;
1739 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1741 /* The calculation in the 'true' branch of this 'if'
1742 * below won't work if overflows, and isn't needed
1743 * anyway. Further below we handle all overflow
1745 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1747 UV min_uv = uv_so_far;
1750 /* Here, the input is both overlong and is missing some trailing
1751 * bytes. There is no single code point it could be for, but there
1752 * may be enough information present to determine if what we have
1753 * so far is for an unallowed code point, such as for a surrogate.
1754 * The code further below has the intelligence to determine this,
1755 * but just for non-overlong UTF-8 sequences. What we do here is
1756 * calculate the smallest code point the input could represent if
1757 * there were no too short malformation. Then we compute and save
1758 * the UTF-8 for that, which is what the code below looks at
1759 * instead of the raw input. It turns out that the smallest such
1760 * code point is all we need. */
1761 for (i = curlen; i < expectlen; i++) {
1762 min_uv = UTF8_ACCUMULATE(min_uv,
1763 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1766 adjusted_s0 = temp_char_buf;
1767 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1771 /* Here, we have found all the possible problems, except for when the input
1772 * is for a problematic code point not allowed by the input parameters. */
1774 /* uv is valid for overlongs */
1775 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1777 /* isn't problematic if < this */
1778 && uv >= UNICODE_SURROGATE_FIRST)
1779 || ( UNLIKELY(possible_problems)
1781 /* if overflow, we know without looking further
1782 * precisely which of the problematic types it is,
1783 * and we deal with those in the overflow handling
1785 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1786 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1787 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1788 && ((flags & ( UTF8_DISALLOW_NONCHAR
1789 |UTF8_DISALLOW_SURROGATE
1790 |UTF8_DISALLOW_SUPER
1791 |UTF8_DISALLOW_PERL_EXTENDED
1793 |UTF8_WARN_SURROGATE
1795 |UTF8_WARN_PERL_EXTENDED))))
1797 /* If there were no malformations, or the only malformation is an
1798 * overlong, 'uv' is valid */
1799 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1800 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1801 possible_problems |= UTF8_GOT_SURROGATE;
1803 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1804 possible_problems |= UTF8_GOT_SUPER;
1806 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1807 possible_problems |= UTF8_GOT_NONCHAR;
1810 else { /* Otherwise, need to look at the source UTF-8, possibly
1811 adjusted to be non-overlong */
1813 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1814 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1816 possible_problems |= UTF8_GOT_SUPER;
1818 else if (curlen > 1) {
1819 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1820 NATIVE_UTF8_TO_I8(*adjusted_s0),
1821 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1823 possible_problems |= UTF8_GOT_SUPER;
1825 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1826 NATIVE_UTF8_TO_I8(*adjusted_s0),
1827 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1829 possible_problems |= UTF8_GOT_SURROGATE;
1833 /* We need a complete well-formed UTF-8 character to discern
1834 * non-characters, so can't look for them here */
1838 ready_to_handle_errors:
1841 * curlen contains the number of bytes in the sequence that
1842 * this call should advance the input by.
1843 * avail_len gives the available number of bytes passed in, but
1844 * only if this is less than the expected number of
1845 * bytes, based on the code point's start byte.
1846 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1847 * is set in it for each potential problem found.
1848 * uv contains the code point the input sequence
1849 * represents; or if there is a problem that prevents
1850 * a well-defined value from being computed, it is
1851 * some subsitute value, typically the REPLACEMENT
1853 * s0 points to the first byte of the character
1854 * s points to just after were we left off processing
1856 * send points to just after where that character should
1857 * end, based on how many bytes the start byte tells
1858 * us should be in it, but no further than s0 +
1862 if (UNLIKELY(possible_problems)) {
1863 bool disallowed = FALSE;
1864 const U32 orig_problems = possible_problems;
1870 while (possible_problems) { /* Handle each possible problem */
1872 char * message = NULL;
1873 U32 this_flag_bit = 0;
1875 /* Each 'if' clause handles one problem. They are ordered so that
1876 * the first ones' messages will be displayed before the later
1877 * ones; this is kinda in decreasing severity order. But the
1878 * overlong must come last, as it changes 'uv' looked at by the
1880 if (possible_problems & UTF8_GOT_OVERFLOW) {
1882 /* Overflow means also got a super and are using Perl's
1883 * extended UTF-8, but we handle all three cases here */
1885 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1886 *errors |= UTF8_GOT_OVERFLOW;
1888 /* But the API says we flag all errors found */
1889 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1890 *errors |= UTF8_GOT_SUPER;
1893 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1895 *errors |= UTF8_GOT_PERL_EXTENDED;
1898 /* Disallow if any of the three categories say to */
1899 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1900 || (flags & ( UTF8_DISALLOW_SUPER
1901 |UTF8_DISALLOW_PERL_EXTENDED)))
1906 /* Likewise, warn if any say to */
1907 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1908 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1911 /* The warnings code explicitly says it doesn't handle the
1912 * case of packWARN2 and two categories which have
1913 * parent-child relationship. Even if it works now to
1914 * raise the warning if either is enabled, it wouldn't
1915 * necessarily do so in the future. We output (only) the
1916 * most dire warning */
1917 if (! (flags & UTF8_CHECK_ONLY)) {
1918 if (msgs || ckWARN_d(WARN_UTF8)) {
1919 pack_warn = packWARN(WARN_UTF8);
1921 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1922 pack_warn = packWARN(WARN_NON_UNICODE);
1925 message = Perl_form(aTHX_ "%s: %s (overflows)",
1927 _byte_dump_string(s0, curlen, 0));
1928 this_flag_bit = UTF8_GOT_OVERFLOW;
1933 else if (possible_problems & UTF8_GOT_EMPTY) {
1934 possible_problems &= ~UTF8_GOT_EMPTY;
1935 *errors |= UTF8_GOT_EMPTY;
1937 if (! (flags & UTF8_ALLOW_EMPTY)) {
1939 /* This so-called malformation is now treated as a bug in
1940 * the caller. If you have nothing to decode, skip calling
1946 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1948 pack_warn = packWARN(WARN_UTF8);
1949 message = Perl_form(aTHX_ "%s (empty string)",
1951 this_flag_bit = UTF8_GOT_EMPTY;
1955 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1956 possible_problems &= ~UTF8_GOT_CONTINUATION;
1957 *errors |= UTF8_GOT_CONTINUATION;
1959 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1962 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1964 pack_warn = packWARN(WARN_UTF8);
1965 message = Perl_form(aTHX_
1966 "%s: %s (unexpected continuation byte 0x%02x,"
1967 " with no preceding start byte)",
1969 _byte_dump_string(s0, 1, 0), *s0);
1970 this_flag_bit = UTF8_GOT_CONTINUATION;
1974 else if (possible_problems & UTF8_GOT_SHORT) {
1975 possible_problems &= ~UTF8_GOT_SHORT;
1976 *errors |= UTF8_GOT_SHORT;
1978 if (! (flags & UTF8_ALLOW_SHORT)) {
1981 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1983 pack_warn = packWARN(WARN_UTF8);
1984 message = Perl_form(aTHX_
1985 "%s: %s (too short; %d byte%s available, need %d)",
1987 _byte_dump_string(s0, send - s0, 0),
1989 avail_len == 1 ? "" : "s",
1991 this_flag_bit = UTF8_GOT_SHORT;
1996 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1997 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1998 *errors |= UTF8_GOT_NON_CONTINUATION;
2000 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2003 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2006 /* If we don't know for sure that the input length is
2007 * valid, avoid as much as possible reading past the
2008 * end of the buffer */
2009 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2011 : (int) (send - s0);
2012 pack_warn = packWARN(WARN_UTF8);
2013 message = Perl_form(aTHX_ "%s",
2014 unexpected_non_continuation_text(s0,
2018 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2022 else if (possible_problems & UTF8_GOT_SURROGATE) {
2023 possible_problems &= ~UTF8_GOT_SURROGATE;
2025 if (flags & UTF8_WARN_SURROGATE) {
2026 *errors |= UTF8_GOT_SURROGATE;
2028 if ( ! (flags & UTF8_CHECK_ONLY)
2029 && (msgs || ckWARN_d(WARN_SURROGATE)))
2031 pack_warn = packWARN(WARN_SURROGATE);
2033 /* These are the only errors that can occur with a
2034 * surrogate when the 'uv' isn't valid */
2035 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2036 message = Perl_form(aTHX_
2037 "UTF-16 surrogate (any UTF-8 sequence that"
2038 " starts with \"%s\" is for a surrogate)",
2039 _byte_dump_string(s0, curlen, 0));
2042 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2044 this_flag_bit = UTF8_GOT_SURROGATE;
2048 if (flags & UTF8_DISALLOW_SURROGATE) {
2050 *errors |= UTF8_GOT_SURROGATE;
2053 else if (possible_problems & UTF8_GOT_SUPER) {
2054 possible_problems &= ~UTF8_GOT_SUPER;
2056 if (flags & UTF8_WARN_SUPER) {
2057 *errors |= UTF8_GOT_SUPER;
2059 if ( ! (flags & UTF8_CHECK_ONLY)
2060 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2062 pack_warn = packWARN(WARN_NON_UNICODE);
2064 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2065 message = Perl_form(aTHX_
2066 "Any UTF-8 sequence that starts with"
2067 " \"%s\" is for a non-Unicode code point,"
2068 " may not be portable",
2069 _byte_dump_string(s0, curlen, 0));
2072 message = Perl_form(aTHX_ super_cp_format, uv);
2074 this_flag_bit = UTF8_GOT_SUPER;
2078 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2079 * and before possibly bailing out, so that the more dire
2080 * warning will override the regular one. */
2081 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2082 if ( ! (flags & UTF8_CHECK_ONLY)
2083 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2084 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2085 || ckWARN(WARN_PORTABLE))))
2087 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2089 /* If it is an overlong that evaluates to a code point
2090 * that doesn't have to use the Perl extended UTF-8, it
2091 * still used it, and so we output a message that
2092 * doesn't refer to the code point. The same is true
2093 * if there was a SHORT malformation where the code
2094 * point is not valid. In that case, 'uv' will have
2095 * been set to the REPLACEMENT CHAR, and the message
2096 * below without the code point in it will be selected
2098 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2099 message = Perl_form(aTHX_
2100 PL_extended_cp_format, uv);
2103 message = Perl_form(aTHX_
2104 "Any UTF-8 sequence that starts with"
2105 " \"%s\" is a Perl extension, and"
2106 " so is not portable",
2107 _byte_dump_string(s0, curlen, 0));
2109 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2112 if (flags & ( UTF8_WARN_PERL_EXTENDED
2113 |UTF8_DISALLOW_PERL_EXTENDED))
2115 *errors |= UTF8_GOT_PERL_EXTENDED;
2117 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2123 if (flags & UTF8_DISALLOW_SUPER) {
2124 *errors |= UTF8_GOT_SUPER;
2128 else if (possible_problems & UTF8_GOT_NONCHAR) {
2129 possible_problems &= ~UTF8_GOT_NONCHAR;
2131 if (flags & UTF8_WARN_NONCHAR) {
2132 *errors |= UTF8_GOT_NONCHAR;
2134 if ( ! (flags & UTF8_CHECK_ONLY)
2135 && (msgs || ckWARN_d(WARN_NONCHAR)))
2137 /* The code above should have guaranteed that we don't
2138 * get here with errors other than overlong */
2139 assert (! (orig_problems
2140 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2142 pack_warn = packWARN(WARN_NONCHAR);
2143 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2144 this_flag_bit = UTF8_GOT_NONCHAR;
2148 if (flags & UTF8_DISALLOW_NONCHAR) {
2150 *errors |= UTF8_GOT_NONCHAR;
2153 else if (possible_problems & UTF8_GOT_LONG) {
2154 possible_problems &= ~UTF8_GOT_LONG;
2155 *errors |= UTF8_GOT_LONG;
2157 if (flags & UTF8_ALLOW_LONG) {
2159 /* We don't allow the actual overlong value, unless the
2160 * special extra bit is also set */
2161 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2162 & ~UTF8_ALLOW_LONG)))
2164 uv = UNICODE_REPLACEMENT;
2171 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2173 pack_warn = packWARN(WARN_UTF8);
2175 /* These error types cause 'uv' to be something that
2176 * isn't what was intended, so can't use it in the
2177 * message. The other error types either can't
2178 * generate an overlong, or else the 'uv' is valid */
2180 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2182 message = Perl_form(aTHX_
2183 "%s: %s (any UTF-8 sequence that starts"
2184 " with \"%s\" is overlong which can and"
2185 " should be represented with a"
2186 " different, shorter sequence)",
2188 _byte_dump_string(s0, send - s0, 0),
2189 _byte_dump_string(s0, curlen, 0));
2192 U8 tmpbuf[UTF8_MAXBYTES+1];
2193 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2195 /* Don't use U+ for non-Unicode code points, which
2196 * includes those in the Latin1 range */
2197 const char * preface = ( uv > PERL_UNICODE_MAX
2204 message = Perl_form(aTHX_
2205 "%s: %s (overlong; instead use %s to represent"
2208 _byte_dump_string(s0, send - s0, 0),
2209 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2211 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2212 small code points */
2215 this_flag_bit = UTF8_GOT_LONG;
2218 } /* End of looking through the possible flags */
2220 /* Display the message (if any) for the problem being handled in
2221 * this iteration of the loop */
2224 assert(this_flag_bit);
2226 if (*msgs == NULL) {
2230 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2235 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2238 Perl_warner(aTHX_ pack_warn, "%s", message);
2240 } /* End of 'while (possible_problems)' */
2242 /* Since there was a possible problem, the returned length may need to
2243 * be changed from the one stored at the beginning of this function.
2244 * Instead of trying to figure out if that's needed, just do it. */
2250 if (flags & UTF8_CHECK_ONLY && retlen) {
2251 *retlen = ((STRLEN) -1);
2257 return UNI_TO_NATIVE(uv);
2261 =for apidoc utf8_to_uvchr_buf
2263 Returns the native code point of the first character in the string C<s> which
2264 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2265 C<*retlen> will be set to the length, in bytes, of that character.
2267 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2268 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2269 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2270 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2271 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2272 the next possible position in C<s> that could begin a non-malformed character.
2273 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2278 Also implemented as a macro in utf8.h
2284 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2286 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2288 return utf8_to_uvchr_buf_helper(s, send, retlen);
2291 /* This is marked as deprecated
2293 =for apidoc utf8_to_uvuni_buf
2295 Only in very rare circumstances should code need to be dealing in Unicode
2296 (as opposed to native) code points. In those few cases, use
2297 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2298 If you are not absolutely sure this is one of those cases, then assume it isn't
2299 and use plain C<utf8_to_uvchr_buf> instead.
2301 Returns the Unicode (not-native) code point of the first character in the
2303 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2304 C<retlen> will be set to the length, in bytes, of that character.
2306 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2307 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2308 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2309 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2310 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2311 next possible position in C<s> that could begin a non-malformed character.
2312 See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2319 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2321 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2325 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2329 =for apidoc utf8_length
2331 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2332 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2333 same place, it returns 0 with no warning raised.
2335 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2336 and returns the number of valid characters.
2342 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2346 PERL_ARGS_ASSERT_UTF8_LENGTH;
2348 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2349 * the bitops (especially ~) can create illegal UTF-8.
2350 * In other words: in Perl UTF-8 is not just for Unicode. */
2352 if (UNLIKELY(e < s))
2353 goto warn_and_return;
2359 if (UNLIKELY(e != s)) {
2363 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2364 "%s in %s", unees, OP_DESC(PL_op));
2366 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2373 =for apidoc bytes_cmp_utf8
2375 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2376 sequence of characters (stored as UTF-8)
2377 in C<u>, C<ulen>. Returns 0 if they are
2378 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2379 if the first string is greater than the second string.
2381 -1 or +1 is returned if the shorter string was identical to the start of the
2382 longer string. -2 or +2 is returned if
2383 there was a difference between characters
2390 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2392 const U8 *const bend = b + blen;
2393 const U8 *const uend = u + ulen;
2395 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2397 while (b < bend && u < uend) {
2399 if (!UTF8_IS_INVARIANT(c)) {
2400 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2403 if (UTF8_IS_CONTINUATION(c1)) {
2404 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2406 /* diag_listed_as: Malformed UTF-8 character%s */
2407 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2409 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2410 PL_op ? " in " : "",
2411 PL_op ? OP_DESC(PL_op) : "");
2416 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2417 "%s in %s", unees, OP_DESC(PL_op));
2419 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2420 return -2; /* Really want to return undef :-) */
2427 return *b < c ? -2 : +2;
2432 if (b == bend && u == uend)
2435 return b < bend ? +1 : -1;
2439 =for apidoc utf8_to_bytes
2441 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2442 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2443 updates C<*lenp> to contain the new length.
2444 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2446 Upon successful return, the number of variants in the string can be computed by
2447 having saved the value of C<*lenp> before the call, and subtracting the
2448 after-call value of C<*lenp> from it.
2450 If you need a copy of the string, see L</bytes_from_utf8>.
2456 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2460 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2461 PERL_UNUSED_CONTEXT;
2463 /* This is a no-op if no variants at all in the input */
2464 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2469 U8 * const save = s;
2470 U8 * const send = s + *lenp;
2473 /* Nothing before the first variant needs to be changed, so start the real
2477 if (! UTF8_IS_INVARIANT(*s)) {
2478 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2479 *lenp = ((STRLEN) -1);
2487 /* Is downgradable, so do it */
2488 d = s = first_variant;
2491 if (! UVCHR_IS_INVARIANT(c)) {
2492 /* Then it is two-byte encoded */
2493 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2506 =for apidoc bytes_from_utf8
2508 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2509 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2510 actually encoded in UTF-8.
2512 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2515 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2516 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2517 C<*lenp> are unchanged, and the return value is the original C<s>.
2519 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2520 newly created string containing a downgraded copy of C<s>, and whose length is
2521 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2522 caller is responsible for arranging for the memory used by this string to get
2525 Upon successful return, the number of variants in the string can be computed by
2526 having saved the value of C<*lenp> before the call, and subtracting the
2527 after-call value of C<*lenp> from it.
2531 There is a macro that avoids this function call, but this is retained for
2532 anyone who calls it with the Perl_ prefix */
2535 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2537 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2538 PERL_UNUSED_CONTEXT;
2540 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2546 This is not currently externally documented because we don't want people to use
2547 it for now. XXX Perhaps that is too paranoid, and it should be documented?
2549 =for apidoc bytes_from_utf8_loc
2551 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2552 to store the location of the first character in C<"s"> that cannot be
2553 converted to non-UTF8.
2555 If that parameter is C<NULL>, this function behaves identically to
2558 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2559 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2561 Otherwise, the function returns a newly created C<NUL>-terminated string
2562 containing the non-UTF8 equivalent of the convertible first portion of
2563 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2564 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2565 and C<*first_non_downgradable> is set to C<NULL>.
2567 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2568 first character in the original string that wasn't converted. C<*is_utf8p> is
2569 unchanged. Note that the new string may have length 0.
2571 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2572 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2573 converts as many characters in it as possible stopping at the first one it
2574 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2575 set to point to that. The function returns the portion that could be converted
2576 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2577 not including the terminating C<NUL>. If the very first character in the
2578 original could not be converted, C<*lenp> will be 0, and the new string will
2579 contain just a single C<NUL>. If the entire input string was converted,
2580 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2582 Upon successful return, the number of variants in the converted portion of the
2583 string can be computed by having saved the value of C<*lenp> before the call,
2584 and subtracting the after-call value of C<*lenp> from it.
2592 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2595 const U8 *original = s;
2596 U8 *converted_start;
2597 const U8 *send = s + *lenp;
2599 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2602 if (first_unconverted) {
2603 *first_unconverted = NULL;
2606 return (U8 *) original;
2609 Newx(d, (*lenp) + 1, U8);
2611 converted_start = d;
2614 if (! UTF8_IS_INVARIANT(c)) {
2616 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2617 * have to stop now */
2618 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2619 if (first_unconverted) {
2620 *first_unconverted = s - 1;
2621 goto finish_and_return;
2624 Safefree(converted_start);
2625 return (U8 *) original;
2629 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2635 /* Here, converted the whole of the input */
2637 if (first_unconverted) {
2638 *first_unconverted = NULL;
2643 *lenp = d - converted_start;
2645 /* Trim unused space */
2646 Renew(converted_start, *lenp + 1, U8);
2648 return converted_start;
2652 =for apidoc bytes_to_utf8
2654 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2656 Returns a pointer to the newly-created string, and sets C<*lenp> to
2657 reflect the new length in bytes. The caller is responsible for arranging for
2658 the memory used by this string to get freed.
2660 Upon successful return, the number of variants in the string can be computed by
2661 having saved the value of C<*lenp> before the call, and subtracting it from the
2662 after-call value of C<*lenp>.
2664 A C<NUL> character will be written after the end of the string.
2666 If you want to convert to UTF-8 from encodings other than
2667 the native (Latin1 or EBCDIC),
2668 see L</sv_recode_to_utf8>().
2674 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2676 const U8 * const send = s + (*lenp);
2680 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2681 PERL_UNUSED_CONTEXT;
2683 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2684 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2688 append_utf8_from_native_byte(*s, &d);
2699 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2700 * use utf16_to_utf8_reversed().
2702 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2703 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2704 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2706 * These functions don't check for overflow. The worst case is every code
2707 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2708 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2709 * destination must be pre-extended to 2 times the source length.
2711 * Do not use in-place. We optimize for native, for obvious reasons. */
2714 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2719 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2722 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2728 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2730 if (OFFUNI_IS_INVARIANT(uv)) {
2731 *d++ = LATIN1_TO_NATIVE((U8) uv);
2734 if (uv <= MAX_UTF8_TWO_BYTE) {
2735 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2736 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2740 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2741 #define LAST_HIGH_SURROGATE 0xDBFF
2742 #define FIRST_LOW_SURROGATE 0xDC00
2743 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2744 #define FIRST_IN_PLANE1 0x10000
2746 /* This assumes that most uses will be in the first Unicode plane, not
2747 * needing surrogates */
2748 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2749 UNICODE_SURROGATE_LAST)))
2751 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2752 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2755 UV low = (p[0] << 8) + p[1];
2756 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2757 LAST_LOW_SURROGATE)))
2759 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2762 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2763 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2767 d = uvoffuni_to_utf8_flags(d, uv, 0);
2769 if (uv < FIRST_IN_PLANE1) {
2770 *d++ = (U8)(( uv >> 12) | 0xe0);
2771 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2772 *d++ = (U8)(( uv & 0x3f) | 0x80);
2776 *d++ = (U8)(( uv >> 18) | 0xf0);
2777 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2778 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2779 *d++ = (U8)(( uv & 0x3f) | 0x80);
2784 *newlen = d - dstart;
2788 /* Note: this one is slightly destructive of the source. */
2791 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2794 U8* const send = s + bytelen;
2796 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2799 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2803 const U8 tmp = s[0];
2808 return utf16_to_utf8(p, d, bytelen, newlen);
2812 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2815 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2819 Perl__is_uni_perl_idcont(pTHX_ UV c)
2822 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2826 Perl__is_uni_perl_idstart(pTHX_ UV c)
2829 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2833 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2836 /* We have the latin1-range values compiled into the core, so just use
2837 * those, converting the result to UTF-8. The only difference between upper
2838 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2839 * either "SS" or "Ss". Which one to use is passed into the routine in
2840 * 'S_or_s' to avoid a test */
2842 UV converted = toUPPER_LATIN1_MOD(c);
2844 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2846 assert(S_or_s == 'S' || S_or_s == 's');
2848 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2849 characters in this range */
2850 *p = (U8) converted;
2855 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2856 * which it maps to one of them, so as to only have to have one check for
2857 * it in the main case */
2858 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2860 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2861 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2864 converted = GREEK_CAPITAL_LETTER_MU;
2866 #if UNICODE_MAJOR_VERSION > 2 \
2867 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2868 && UNICODE_DOT_DOT_VERSION >= 8)
2869 case LATIN_SMALL_LETTER_SHARP_S:
2876 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2877 " '%c' to map to '%c'",
2878 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2879 NOT_REACHED; /* NOTREACHED */
2883 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2884 *p = UTF8_TWO_BYTE_LO(converted);
2890 /* If compiled on an early Unicode version, there may not be auxiliary tables
2892 #ifndef HAS_UC_AUX_TABLES
2893 # define UC_AUX_TABLE_ptrs NULL
2894 # define UC_AUX_TABLE_lengths NULL
2896 #ifndef HAS_TC_AUX_TABLES
2897 # define TC_AUX_TABLE_ptrs NULL
2898 # define TC_AUX_TABLE_lengths NULL
2900 #ifndef HAS_LC_AUX_TABLES
2901 # define LC_AUX_TABLE_ptrs NULL
2902 # define LC_AUX_TABLE_lengths NULL
2904 #ifndef HAS_CF_AUX_TABLES
2905 # define CF_AUX_TABLE_ptrs NULL
2906 # define CF_AUX_TABLE_lengths NULL
2908 #ifndef HAS_UC_AUX_TABLES
2909 # define UC_AUX_TABLE_ptrs NULL
2910 # define UC_AUX_TABLE_lengths NULL
2913 /* Call the function to convert a UTF-8 encoded character to the specified case.
2914 * Note that there may be more than one character in the result.
2915 * 's' is a pointer to the first byte of the input character
2916 * 'd' will be set to the first byte of the string of changed characters. It
2917 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2918 * 'lenp' will be set to the length in bytes of the string of changed characters
2920 * The functions return the ordinal of the first character in the string of
2922 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2923 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2924 Uppercase_Mapping_invmap, \
2925 UC_AUX_TABLE_ptrs, \
2926 UC_AUX_TABLE_lengths, \
2928 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2929 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2930 Titlecase_Mapping_invmap, \
2931 TC_AUX_TABLE_ptrs, \
2932 TC_AUX_TABLE_lengths, \
2934 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2935 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2936 Lowercase_Mapping_invmap, \
2937 LC_AUX_TABLE_ptrs, \
2938 LC_AUX_TABLE_lengths, \
2942 /* This additionally has the input parameter 'specials', which if non-zero will
2943 * cause this to use the specials hash for folding (meaning get full case
2944 * folding); otherwise, when zero, this implies a simple case fold */
2945 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2947 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2948 Case_Folding_invmap, \
2949 CF_AUX_TABLE_ptrs, \
2950 CF_AUX_TABLE_lengths, \
2952 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2953 Simple_Case_Folding_invmap, \
2958 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2960 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2961 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2962 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2963 * the changed version may be longer than the original character.
2965 * The ordinal of the first character of the changed version is returned
2966 * (but note, as explained above, that there may be more.) */
2969 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2972 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2975 return CALL_UPPER_CASE(c, NULL, p, lenp);
2979 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2982 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2985 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2988 return CALL_TITLE_CASE(c, NULL, p, lenp);
2992 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2994 /* We have the latin1-range values compiled into the core, so just use
2995 * those, converting the result to UTF-8. Since the result is always just
2996 * one character, we allow <p> to be NULL */
2998 U8 converted = toLOWER_LATIN1(c);
3000 PERL_UNUSED_ARG(dummy);
3003 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
3008 /* Result is known to always be < 256, so can use the EIGHT_BIT
3010 *p = UTF8_EIGHT_BIT_HI(converted);
3011 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3019 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3022 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3025 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3028 return CALL_LOWER_CASE(c, NULL, p, lenp);
3032 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3034 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3035 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3036 * FOLD_FLAGS_FULL iff full folding is to be used;
3038 * Not to be used for locale folds
3043 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3045 assert (! (flags & FOLD_FLAGS_LOCALE));
3047 if (UNLIKELY(c == MICRO_SIGN)) {
3048 converted = GREEK_SMALL_LETTER_MU;
3050 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3051 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3052 || UNICODE_DOT_DOT_VERSION > 0)
3053 else if ( (flags & FOLD_FLAGS_FULL)
3054 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3056 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3057 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3058 * under those circumstances. */
3059 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3060 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3061 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3063 return LATIN_SMALL_LETTER_LONG_S;
3073 else { /* In this range the fold of all other characters is their lower
3075 converted = toLOWER_LATIN1(c);
3078 if (UVCHR_IS_INVARIANT(converted)) {
3079 *p = (U8) converted;
3083 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3084 *p = UTF8_TWO_BYTE_LO(converted);
3092 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3095 /* Not currently externally documented, and subject to change
3096 * <flags> bits meanings:
3097 * FOLD_FLAGS_FULL iff full folding is to be used;
3098 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3099 * locale are to be used.
3100 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3104 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3106 if (flags & FOLD_FLAGS_LOCALE) {
3107 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3108 * except for potentially warning */
3109 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3110 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3111 flags &= ~FOLD_FLAGS_LOCALE;
3114 goto needs_full_generality;
3119 return _to_fold_latin1((U8) c, p, lenp,
3120 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3123 /* Here, above 255. If no special needs, just use the macro */
3124 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3125 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3127 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3128 the special flags. */
3129 U8 utf8_c[UTF8_MAXBYTES + 1];
3131 needs_full_generality:
3132 uvchr_to_utf8(utf8_c, c);
3133 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3138 PERL_STATIC_INLINE bool
3139 S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3142 /* returns a boolean giving whether or not the UTF8-encoded character that
3143 * starts at <p>, and extending no further than <e - 1> is in the inversion
3144 * list <invlist>. */
3146 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3148 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3150 if (cp == 0 && (p >= e || *p != '\0')) {
3151 _force_out_malformed_utf8_message(p, e, 0, 1);
3152 NOT_REACHED; /* NOTREACHED */
3156 return _invlist_contains_cp(invlist, cp);
3159 #if 0 /* Not currently used, but may be needed in the future */
3160 PERLVAR(I, seen_deprecated_macro, HV *)
3163 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3164 const char * const alternative,
3165 const bool use_locale,
3166 const char * const file,
3167 const unsigned line)
3171 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3173 if (ckWARN_d(WARN_DEPRECATED)) {
3175 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3176 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3177 if (! PL_seen_deprecated_macro) {
3178 PL_seen_deprecated_macro = newHV();
3180 if (! hv_store(PL_seen_deprecated_macro, key,
3181 strlen(key), &PL_sv_undef, 0))
3183 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3186 if (instr(file, "mathoms.c")) {
3187 Perl_warner(aTHX_ WARN_DEPRECATED,
3188 "In %s, line %d, starting in Perl v5.32, %s()"
3189 " will be removed. Avoid this message by"
3190 " converting to use %s().\n",
3191 file, line, name, alternative);
3194 Perl_warner(aTHX_ WARN_DEPRECATED,
3195 "In %s, line %d, starting in Perl v5.32, %s() will"
3196 " require an additional parameter. Avoid this"
3197 " message by converting to use %s().\n",
3198 file, line, name, alternative);
3206 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
3209 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3211 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
3215 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
3218 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
3220 return is_utf8_common(p, e, PL_utf8_perl_idstart);
3224 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
3227 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
3229 return is_utf8_common(p, e, PL_utf8_perl_idcont);
3233 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3234 U8* ustrp, STRLEN *lenp,
3235 SV *invlist, const I32 * const invmap,
3236 const U32 * const * const aux_tables,
3237 const U8 * const aux_table_lengths,
3238 const char * const normal)
3242 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3243 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3244 * to name the new case in any generated messages, as a fallback if the
3245 * operation being used is not available. The new case is given by the
3246 * data structures in the remaining arguments.
3248 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3249 * entire changed case string, and the return value is the first code point
3252 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3254 /* For code points that don't change case, we already know that the output
3255 * of this function is the unchanged input, so we can skip doing look-ups
3256 * for them. Unfortunately the case-changing code points are scattered
3257 * around. But there are some long consecutive ranges where there are no
3258 * case changing code points. By adding tests, we can eliminate the lookup
3259 * for all the ones in such ranges. This is currently done here only for
3260 * just a few cases where the scripts are in common use in modern commerce
3261 * (and scripts adjacent to those which can be included without additional
3264 if (uv1 >= 0x0590) {
3265 /* This keeps from needing further processing the code points most
3266 * likely to be used in the following non-cased scripts: Hebrew,
3267 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3268 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3269 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3274 /* The following largish code point ranges also don't have case
3275 * changes, but khw didn't think they warranted extra tests to speed
3276 * them up (which would slightly slow down everything else above them):
3277 * 1100..139F Hangul Jamo, Ethiopic
3278 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3279 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3280 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3281 * Combining Diacritical Marks Extended, Balinese,
3282 * Sundanese, Batak, Lepcha, Ol Chiki
3283 * 2000..206F General Punctuation
3286 if (uv1 >= 0x2D30) {
3288 /* This keeps the from needing further processing the code points
3289 * most likely to be used in the following non-cased major scripts:
3290 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3292 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3293 * event that Unicode eventually allocates the unused block as of
3294 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3295 * that the test suite will start having failures to alert you
3296 * should that happen) */
3301 if (uv1 >= 0xAC00) {
3302 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3303 if (ckWARN_d(WARN_SURROGATE)) {
3304 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3305 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3306 "Operation \"%s\" returns its argument for"
3307 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3312 /* AC00..FAFF Catches Hangul syllables and private use, plus
3318 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3319 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3320 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1));
3322 if (ckWARN_d(WARN_NON_UNICODE)) {
3323 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3324 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3325 "Operation \"%s\" returns its argument for"
3326 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3330 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3332 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3341 /* Note that non-characters are perfectly legal, so no warning should
3347 const U32 * cp_list;
3350 /* 'index' is guaranteed to be non-negative, as this is an inversion
3351 * map that covers all possible inputs. See [perl #133365] */
3352 SSize_t index = _invlist_search(invlist, uv1);
3353 I32 base = invmap[index];
3355 /* The data structures are set up so that if 'base' is non-negative,
3356 * the case change is 1-to-1; and if 0, the change is to itself */
3364 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3365 lc = base + uv1 - invlist_array(invlist)[index];
3366 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3370 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3371 * requires an auxiliary table look up. abs(base) gives the index into
3372 * a list of such tables which points to the proper aux table. And a
3373 * parallel list gives the length of each corresponding aux table. */
3374 cp_list = aux_tables[-base];
3376 /* Create the string of UTF-8 from the mapped-to code points */
3378 for (i = 0; i < aux_table_lengths[-base]; i++) {
3379 d = uvchr_to_utf8(d, cp_list[i]);
3387 /* Here, there was no mapping defined, which means that the code point maps
3388 * to itself. Return the inputs */
3392 if (p != ustrp) { /* Don't copy onto itself */
3393 Copy(p, ustrp, len, U8);
3398 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3406 Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
3407 const U32 ** remaining_folds_to)
3409 /* Returns the count of the number of code points that fold to the input
3410 * 'cp' (besides itself).
3412 * If the return is 0, there is nothing else that folds to it, and
3413 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3415 * If the return is 1, '*first_folds_to' is set to the single code point,
3416 * and '*remaining_folds_to' is set to NULL.
3418 * Otherwise, '*first_folds_to' is set to a code point, and
3419 * '*remaining_fold_to' is set to an array that contains the others. The
3420 * length of this array is the returned count minus 1.
3422 * The reason for this convolution is to avoid having to deal with
3423 * allocating and freeing memory. The lists are already constructed, so
3424 * the return can point to them, but single code points aren't, so would
3425 * need to be constructed if we didn't employ something like this API
3427 * The code points returned by this function are all legal Unicode, which
3428 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3429 * constructed with this size (to save space and memory), and we return
3430 * pointers, so they must be this size */
3433 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3434 * that covers all possible inputs. See [perl #133365] */
3435 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3436 I32 base = _Perl_IVCF_invmap[index];
3438 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3440 if (base == 0) { /* No fold */
3441 *first_folds_to = 0;
3442 *remaining_folds_to = NULL;
3446 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3452 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3454 /* The data structure is set up so that the absolute value of 'base' is
3455 * an index into a table of pointers to arrays, with the array
3456 * corresponding to the index being the list of code points that fold
3457 * to 'cp', and the parallel array containing the length of the list
3459 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3460 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3461 /* +1 excludes first_folds_to */
3462 return IVCF_AUX_TABLE_lengths[-base];
3467 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3468 *first_folds_to = (U32) (base + cp
3469 - invlist_array(PL_utf8_foldclosures)[index]);
3470 *remaining_folds_to = NULL;
3475 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3476 U8* const ustrp, STRLEN *lenp)
3478 /* This is called when changing the case of a UTF-8-encoded character above
3479 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3480 * result contains a character that crosses the 255/256 boundary, disallow
3481 * the change, and return the original code point. See L<perlfunc/lc> for
3484 * p points to the original string whose case was changed; assumed
3485 * by this routine to be well-formed
3486 * result the code point of the first character in the changed-case string
3487 * ustrp points to the changed-case string (<result> represents its
3489 * lenp points to the length of <ustrp> */
3491 UV original; /* To store the first code point of <p> */
3493 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3495 assert(UTF8_IS_ABOVE_LATIN1(*p));
3497 /* We know immediately if the first character in the string crosses the
3498 * boundary, so can skip testing */
3501 /* Look at every character in the result; if any cross the
3502 * boundary, the whole thing is disallowed */
3503 U8* s = ustrp + UTF8SKIP(ustrp);
3504 U8* e = ustrp + *lenp;
3506 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3512 /* Here, no characters crossed, result is ok as-is, but we warn. */
3513 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3519 /* Failed, have to return the original */
3520 original = valid_utf8_to_uvchr(p, lenp);
3522 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3523 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3524 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3525 " locale; resolved to \"\\x{%" UVXf "}\".",
3529 Copy(p, ustrp, *lenp, char);
3534 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3535 U8 * ustrp, STRLEN *lenp)
3537 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3538 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3539 * Otherwise, it returns the first code point of the Turkic foldcased
3540 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3541 * contain *lenp bytes
3543 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3544 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3547 PERL_ARGS_ASSERT_TURKIC_FC;
3550 if (UNLIKELY(*p == 'I')) {
3552 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3553 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3554 return LATIN_SMALL_LETTER_DOTLESS_I;
3557 if (UNLIKELY(memBEGINs(p, e - p,
3558 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3569 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3570 U8 * ustrp, STRLEN *lenp)
3572 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3573 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3574 * Otherwise, it returns the first code point of the Turkic lowercased
3575 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3576 * contain *lenp bytes */
3579 PERL_ARGS_ASSERT_TURKIC_LC;
3582 /* A 'I' requires context as to what to do */
3583 if (UNLIKELY(*p0 == 'I')) {
3584 const U8 * p = p0 + 1;
3586 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3587 * modified by a dot above lowercases to 'i' even in turkic locales. */
3591 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3597 /* For the dot above to modify the 'I', it must be part of a
3598 * combining sequence immediately following the 'I', and no other
3599 * modifier with a ccc of 230 may intervene */
3600 cp = utf8_to_uvchr_buf(p, e, NULL);
3601 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3605 /* Here the combining sequence continues */
3610 /* In all other cases the lc is the same as the fold */
3611 return turkic_fc(p0, e, ustrp, lenp);
3615 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3616 U8 * ustrp, STRLEN *lenp)
3618 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3619 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3620 * Otherwise, it returns the first code point of the Turkic upper or
3621 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3622 * ustrp will contain *lenp bytes
3624 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3625 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3628 PERL_ARGS_ASSERT_TURKIC_UC;
3633 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3634 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3635 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3638 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3647 /* The process for changing the case is essentially the same for the four case
3648 * change types, except there are complications for folding. Otherwise the
3649 * difference is only which case to change to. To make sure that they all do
3650 * the same thing, the bodies of the functions are extracted out into the
3651 * following two macros. The functions are written with the same variable
3652 * names, and these are known and used inside these macros. It would be
3653 * better, of course, to have inline functions to do it, but since different
3654 * macros are called, depending on which case is being changed to, this is not
3655 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3656 * function can start with the common start macro, then finish with its special
3657 * handling; while the other three cases can just use the common end macro.
3659 * The algorithm is to use the proper (passed in) macro or function to change
3660 * the case for code points that are below 256. The macro is used if using
3661 * locale rules for the case change; the function if not. If the code point is
3662 * above 255, it is computed from the input UTF-8, and another macro is called
3663 * to do the conversion. If necessary, the output is converted to UTF-8. If
3664 * using a locale, we have to check that the change did not cross the 255/256
3665 * boundary, see check_locale_boundary_crossing() for further details.
3667 * The macros are split with the correct case change for the below-256 case
3668 * stored into 'result', and in the middle of an else clause for the above-255
3669 * case. At that point in the 'else', 'result' is not the final result, but is
3670 * the input code point calculated from the UTF-8. The fold code needs to
3671 * realize all this and take it from there.
3673 * To deal with Turkic locales, the function specified by the parameter
3674 * 'turkic' is called when appropriate.
3676 * If you read the two macros as sequential, it's easier to understand what's
3678 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3679 L1_func_extra_param, turkic) \
3681 if (flags & (locale_flags)) { \
3682 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3683 if (IN_UTF8_CTYPE_LOCALE) { \
3684 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3685 UV ret = turkic(p, e, ustrp, lenp); \
3686 if (ret) return ret; \
3689 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3691 flags &= ~(locale_flags); \
3695 if (UTF8_IS_INVARIANT(*p)) { \
3696 if (flags & (locale_flags)) { \
3697 result = LC_L1_change_macro(*p); \
3700 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3703 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3704 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3705 if (flags & (locale_flags)) { \
3706 result = LC_L1_change_macro(c); \
3709 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3712 else { /* malformed UTF-8 or ord above 255 */ \
3713 STRLEN len_result; \
3714 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3715 if (len_result == (STRLEN) -1) { \
3716 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
3719 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3720 result = change_macro(result, p, ustrp, lenp); \
3722 if (flags & (locale_flags)) { \
3723 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3728 /* Here, used locale rules. Convert back to UTF-8 */ \
3729 if (UTF8_IS_INVARIANT(result)) { \
3730 *ustrp = (U8) result; \
3734 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3735 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3741 /* Not currently externally documented, and subject to change:
3742 * <flags> is set iff iff the rules from the current underlying locale are to
3746 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3755 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3757 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3758 /* 2nd char of uc(U+DF) is 'S' */
3759 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3761 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3764 /* Not currently externally documented, and subject to change:
3765 * <flags> is set iff the rules from the current underlying locale are to be
3766 * used. Since titlecase is not defined in POSIX, for other than a
3767 * UTF-8 locale, uppercase is used instead for code points < 256.
3771 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3780 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3782 /* 2nd char of ucfirst(U+DF) is 's' */
3783 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
3785 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3788 /* Not currently externally documented, and subject to change:
3789 * <flags> is set iff iff the rules from the current underlying locale are to
3794 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3803 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3805 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
3807 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3810 /* Not currently externally documented, and subject to change,
3812 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3813 * locale are to be used.
3814 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3815 * otherwise simple folds
3816 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3821 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3830 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3832 /* These are mutually exclusive */
3833 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3835 assert(p != ustrp); /* Otherwise overwrites */
3837 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3838 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
3841 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3843 if (flags & FOLD_FLAGS_LOCALE) {
3845 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3846 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3847 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3849 /* Special case these two characters, as what normally gets
3850 * returned under locale doesn't work */
3851 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
3853 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3854 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3855 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3856 "resolved to \"\\x{17F}\\x{17F}\".");
3861 if (memBEGINs((char *) p, e - p, LONG_S_T))
3863 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3864 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3865 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3866 "resolved to \"\\x{FB06}\".");
3867 goto return_ligature_st;
3870 #if UNICODE_MAJOR_VERSION == 3 \
3871 && UNICODE_DOT_VERSION == 0 \
3872 && UNICODE_DOT_DOT_VERSION == 1
3873 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3875 /* And special case this on this Unicode version only, for the same
3876 * reaons the other two are special cased. They would cross the
3877 * 255/256 boundary which is forbidden under /l, and so the code
3878 * wouldn't catch that they are equivalent (which they are only in
3880 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
3881 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3882 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3883 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3884 "resolved to \"\\x{0131}\".");
3885 goto return_dotless_i;
3889 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3891 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3895 /* This is called when changing the case of a UTF-8-encoded
3896 * character above the ASCII range, and the result should not
3897 * contain an ASCII character. */
3899 UV original; /* To store the first code point of <p> */
3901 /* Look at every character in the result; if any cross the
3902 * boundary, the whole thing is disallowed */
3904 U8* send = ustrp + *lenp;
3907 /* Crossed, have to return the original */
3908 original = valid_utf8_to_uvchr(p, lenp);
3910 /* But in these instances, there is an alternative we can
3911 * return that is valid */
3912 if (original == LATIN_SMALL_LETTER_SHARP_S
3913 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3914 || original == LATIN_CAPITAL_LETTER_SHARP_S
3919 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3920 goto return_ligature_st;
3922 #if UNICODE_MAJOR_VERSION == 3 \
3923 && UNICODE_DOT_VERSION == 0 \
3924 && UNICODE_DOT_DOT_VERSION == 1
3926 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3927 goto return_dotless_i;
3930 Copy(p, ustrp, *lenp, char);
3936 /* Here, no characters crossed, result is ok as-is */
3941 /* Here, used locale rules. Convert back to UTF-8 */
3942 if (UTF8_IS_INVARIANT(result)) {
3943 *ustrp = (U8) result;
3947 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3948 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3955 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3956 * folds to a string of two of these characters. By returning this
3957 * instead, then, e.g.,
3958 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3961 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3962 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3964 return LATIN_SMALL_LETTER_LONG_S;
3967 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3968 * have the other one fold to it */
3970 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3971 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3972 return LATIN_SMALL_LIGATURE_ST;
3974 #if UNICODE_MAJOR_VERSION == 3 \
3975 && UNICODE_DOT_VERSION == 0 \
3976 && UNICODE_DOT_DOT_VERSION == 1
3979 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3980 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3981 return LATIN_SMALL_LETTER_DOTLESS_I;
3988 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3990 /* May change: warns if surrogates, non-character code points, or
3991 * non-Unicode code points are in 's' which has length 'len' bytes.
3992 * Returns TRUE if none found; FALSE otherwise. The only other validity
3993 * check is to make sure that this won't exceed the string's length nor
3996 const U8* const e = s + len;
3999 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4002 if (UTF8SKIP(s) > len) {
4003 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4004 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4007 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4008 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4009 if ( ckWARN_d(WARN_NON_UNICODE)
4010 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
4011 0 /* Don't consider overlongs */
4014 /* A side effect of this function will be to warn */
4015 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4019 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4020 if (ckWARN_d(WARN_SURROGATE)) {
4021 /* This has a different warning than the one the called
4022 * function would output, so can't just call it, unlike we
4023 * do for the non-chars and above-unicodes */
4024 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4025 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4026 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4031 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4032 && (ckWARN_d(WARN_NONCHAR)))
4034 /* A side effect of this function will be to warn */
4035 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4046 =for apidoc pv_uni_display
4048 Build to the scalar C<dsv> a displayable version of the UTF-8 encoded string
4049 C<spv>, length C<len>, the displayable version being at most C<pvlim> bytes
4050 long (if longer, the rest is truncated and C<"..."> will be appended).
4052 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4053 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4054 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4055 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4056 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4057 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4059 Additionally, there is now C<UNI_DISPLAY_BACKSPACE> which allows C<\b> for a
4060 backspace, but only when C<UNI_DISPLAY_BACKSLASH> also is set.
4062 The pointer to the PV of the C<dsv> is returned.
4064 See also L</sv_uni_display>.
4068 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
4074 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4078 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4082 if (pvlim && SvCUR(dsv) >= pvlim) {
4086 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4088 const unsigned char c = (unsigned char)u & 0xFF;
4089 if (flags & UNI_DISPLAY_BACKSLASH) {
4090 if ( isMNEMONIC_CNTRL(c)
4092 || (flags & UNI_DISPLAY_BACKSPACE)))
4094 const char * mnemonic = cntrl_to_mnemonic(c);
4095 sv_catpvn(dsv, mnemonic, strlen(mnemonic));
4098 else if (c == '\\') {
4099 sv_catpvs(dsv, "\\\\");
4103 /* isPRINT() is the locale-blind version. */
4104 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4105 const char string = c;
4106 sv_catpvn(dsv, &string, 1);
4111 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
4114 sv_catpvs(dsv, "...");
4120 =for apidoc sv_uni_display
4122 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4123 the displayable version being at most C<pvlim> bytes long
4124 (if longer, the rest is truncated and "..." will be appended).
4126 The C<flags> argument is as in L</pv_uni_display>().
4128 The pointer to the PV of the C<dsv> is returned.
4133 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4135 const char * const ptr =
4136 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4138 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4140 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4141 SvCUR(ssv), pvlim, flags);
4145 =for apidoc foldEQ_utf8
4147 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
4148 both of which may be in UTF-8) are the same case-insensitively; false
4149 otherwise. How far into the strings to compare is determined by other input
4152 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4153 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
4154 C<u2> with respect to C<s2>.
4156 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
4157 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
4158 The scan will not be considered to be a match unless the goal is reached, and
4159 scanning won't continue past that goal. Correspondingly for C<l2> with respect
4162 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
4163 pointer is considered an end pointer to the position 1 byte past the maximum
4164 point in C<s1> beyond which scanning will not continue under any circumstances.
4165 (This routine assumes that UTF-8 encoded input strings are not malformed;
4166 malformed input can cause it to read past C<pe1>). This means that if both
4167 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
4168 will never be successful because it can never
4169 get as far as its goal (and in fact is asserted against). Correspondingly for
4170 C<pe2> with respect to C<s2>.
4172 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4173 C<l2> must be non-zero), and if both do, both have to be
4174 reached for a successful match. Also, if the fold of a character is multiple
4175 characters, all of them must be matched (see tr21 reference below for
4178 Upon a successful match, if C<pe1> is non-C<NULL>,
4179 it will be set to point to the beginning of the I<next> character of C<s1>
4180 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4182 For case-insensitiveness, the "casefolding" of Unicode is used
4183 instead of upper/lowercasing both the characters, see
4184 L<https://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4188 /* A flags parameter has been added which may change, and hence isn't
4189 * externally documented. Currently it is:
4190 * 0 for as-documented above
4191 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4192 ASCII one, to not match
4193 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4194 * locale are to be used.
4195 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4196 * routine. This allows that step to be skipped.
4197 * Currently, this requires s1 to be encoded as UTF-8
4198 * (u1 must be true), which is asserted for.
4199 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4200 * cross certain boundaries. Hence, the caller should
4201 * let this function do the folding instead of
4202 * pre-folding. This code contains an assertion to
4203 * that effect. However, if the caller knows what
4204 * it's doing, it can pass this flag to indicate that,
4205 * and the assertion is skipped.
4206 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
4207 * to s2, and s2 doesn't have to be UTF-8 encoded.
4208 * This introduces an asymmetry to save a few branches
4209 * in a loop. Currently, this is not a problem, as
4210 * never are both inputs pre-folded. Simply call this
4211 * function with the pre-folded one as the second
4213 * FOLDEQ_S2_FOLDS_SANE
4216 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
4217 const char *s2, char **pe2, UV l2, bool u2,
4220 const U8 *p1 = (const U8*)s1; /* Point to current char */
4221 const U8 *p2 = (const U8*)s2;
4222 const U8 *g1 = NULL; /* goal for s1 */
4223 const U8 *g2 = NULL;
4224 const U8 *e1 = NULL; /* Don't scan s1 past this */
4225 U8 *f1 = NULL; /* Point to current folded */
4226 const U8 *e2 = NULL;
4228 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4229 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4230 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4231 U8 flags_for_folder = FOLD_FLAGS_FULL;
4233 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4235 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4236 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
4237 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4238 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
4239 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4240 /* The algorithm is to trial the folds without regard to the flags on
4241 * the first line of the above assert(), and then see if the result
4242 * violates them. This means that the inputs can't be pre-folded to a
4243 * violating result, hence the assert. This could be changed, with the
4244 * addition of extra tests here for the already-folded case, which would
4245 * slow it down. That cost is more than any possible gain for when these
4246 * flags are specified, as the flags indicate /il or /iaa matching which
4247 * is less common than /iu, and I (khw) also believe that real-world /il
4248 * and /iaa matches are most likely to involve code points 0-255, and this
4249 * function only under rare conditions gets called for 0-255. */
4251 if (flags & FOLDEQ_LOCALE) {
4252 if (IN_UTF8_CTYPE_LOCALE) {
4253 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
4254 flags_for_folder |= FOLD_FLAGS_LOCALE;
4257 flags &= ~FOLDEQ_LOCALE;
4261 flags_for_folder |= FOLD_FLAGS_LOCALE;
4264 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
4265 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
4273 g1 = (const U8*)s1 + l1;
4281 g2 = (const U8*)s2 + l2;
4284 /* Must have at least one goal */
4289 /* Will never match if goal is out-of-bounds */
4290 assert(! e1 || e1 >= g1);
4292 /* Here, there isn't an end pointer, or it is beyond the goal. We
4293 * only go as far as the goal */
4297 assert(e1); /* Must have an end for looking at s1 */
4300 /* Same for goal for s2 */
4302 assert(! e2 || e2 >= g2);
4309 /* If both operands are already folded, we could just do a memEQ on the
4310 * whole strings at once, but it would be better if the caller realized
4311 * this and didn't even call us */
4313 /* Look through both strings, a character at a time */
4314 while (p1 < e1 && p2 < e2) {
4316 /* If at the beginning of a new character in s1, get its fold to use
4317 * and the length of the fold. */
4319 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4325 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4327 /* We have to forbid mixing ASCII with non-ASCII if the
4328 * flags so indicate. And, we can short circuit having to
4329 * call the general functions for this common ASCII case,
4330 * all of whose non-locale folds are also ASCII, and hence
4331 * UTF-8 invariants, so the UTF8ness of the strings is not
4333 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4337 *foldbuf1 = toFOLD(*p1);
4340 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
4342 else { /* Not UTF-8, get UTF-8 fold */
4343 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4349 if (n2 == 0) { /* Same for s2 */
4350 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4352 /* Point to the already-folded character. But for non-UTF-8
4353 * variants, convert to UTF-8 for the algorithm below */
4354 if (UTF8_IS_INVARIANT(*p2)) {
4363 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
4364 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
4370 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4371 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4375 *foldbuf2 = toFOLD(*p2);
4378 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
4381 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4387 /* Here f1 and f2 point to the beginning of the strings to compare.
4388 * These strings are the folds of the next character from each input
4389 * string, stored in UTF-8. */
4391 /* While there is more to look for in both folds, see if they
4392 * continue to match */
4394 U8 fold_length = UTF8SKIP(f1);
4395 if (fold_length != UTF8SKIP(f2)
4396 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4397 function call for single
4399 || memNE((char*)f1, (char*)f2, fold_length))
4401 return 0; /* mismatch */
4404 /* Here, they matched, advance past them */
4411 /* When reach the end of any fold, advance the input past it */
4413 p1 += u1 ? UTF8SKIP(p1) : 1;
4416 p2 += u2 ? UTF8SKIP(p2) : 1;
4418 } /* End of loop through both strings */
4420 /* A match is defined by each scan that specified an explicit length
4421 * reaching its final goal, and the other not having matched a partial
4422 * character (which can happen when the fold of a character is more than one
4424 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4428 /* Successful match. Set output pointers */
4439 * ex: set ts=8 sts=4 sw=4 et: