3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
40 /* Be sure to synchronize this message with the similar one in regcomp.c */
41 static const char cp_above_legal_max[] =
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
46 =head1 Unicode Support
47 These are various utility functions for manipulating UTF8-encoded
48 strings. For the uninitiated, this is a method of representing arbitrary
49 Unicode characters as a variable number of bytes, in such a way that
50 characters in the ASCII range are unmodified, and a zero byte never appears
51 within non-zero characters.
56 /* helper for Perl__force_out_malformed_utf8_message(). Like
57 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
61 S_restore_cop_warnings(pTHX_ void *p)
63 if (!specialWARN(PL_curcop->cop_warnings))
64 PerlMemShared_free(PL_curcop->cop_warnings);
65 PL_curcop->cop_warnings = (STRLEN*)p;
70 Perl__force_out_malformed_utf8_message(pTHX_
71 const U8 *const p, /* First byte in UTF-8 sequence */
72 const U8 * const e, /* Final byte in sequence (may include
74 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
75 usually 0, or some DISALLOW flags */
76 const bool die_here) /* If TRUE, this function does not return */
78 /* This core-only function is to be called when a malformed UTF-8 character
79 * is found, in order to output the detailed information about the
80 * malformation before dieing. The reason it exists is for the occasions
81 * when such a malformation is fatal, but warnings might be turned off, so
82 * that normally they would not be actually output. This ensures that they
83 * do get output. Because a sequence may be malformed in more than one
84 * way, multiple messages may be generated, so we can't make them fatal, as
85 * that would cause the first one to die.
87 * Instead we pretend -W was passed to perl, then die afterwards. The
88 * flexibility is here to return to the caller so they can finish up and
92 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
98 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
100 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
101 * than PL_compiling */
102 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
103 (void*)PL_curcop->cop_warnings);
104 PL_curcop->cop_warnings = pWARN_ALL;
107 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
112 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
113 " be called only when there are errors found");
117 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
122 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
123 U32 categories, /* Packed warning categories */
124 U32 flag) /* Flag associated with this message */
126 /* Creates, populates, and returns an HV* that describes an error message
127 * for the translators between UTF8 and code point */
129 SV* msg_sv = newSVpv(message, 0);
130 SV* category_sv = newSVuv(categories);
131 SV* flag_bit_sv = newSVuv(flag);
133 HV* msg_hv = newHV();
135 PERL_ARGS_ASSERT_NEW_MSG_HV;
137 (void) hv_stores(msg_hv, "text", msg_sv);
138 (void) hv_stores(msg_hv, "warn_categories", category_sv);
139 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
145 =for apidoc uvoffuni_to_utf8_flags
147 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
148 Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
149 L<perlapi/uvchr_to_utf8_flags>>.
151 This function is like them, but the input is a strict Unicode
152 (as opposed to native) code point. Only in very rare circumstances should code
153 not be using the native code point.
155 For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
161 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
163 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
165 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
168 /* All these formats take a single UV code point argument */
169 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
170 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
171 " is not recommended for open interchange";
172 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
173 " may not be portable";
174 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
175 " Unicode, requires a Perl extension," \
176 " and so is not portable";
178 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
180 if (flags & UNICODE_WARN_SURROGATE) { \
181 U32 category = packWARN(WARN_SURROGATE); \
182 const char * format = surrogate_cp_format; \
184 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
186 UNICODE_GOT_SURROGATE); \
189 Perl_ck_warner_d(aTHX_ category, format, uv); \
192 if (flags & UNICODE_DISALLOW_SURROGATE) { \
197 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
199 if (flags & UNICODE_WARN_NONCHAR) { \
200 U32 category = packWARN(WARN_NONCHAR); \
201 const char * format = nonchar_cp_format; \
203 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
205 UNICODE_GOT_NONCHAR); \
208 Perl_ck_warner_d(aTHX_ category, format, uv); \
211 if (flags & UNICODE_DISALLOW_NONCHAR) { \
216 /* Use shorter names internally in this file */
217 #define SHIFT UTF_ACCUMULATION_SHIFT
219 #define MARK UTF_CONTINUATION_MARK
220 #define MASK UTF_CONTINUATION_MASK
223 =for apidoc uvchr_to_utf8_flags_msgs
225 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
227 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
229 This function is for code that wants any warning and/or error messages to be
230 returned to the caller rather than be displayed. All messages that would have
231 been displayed if all lexical warnings are enabled will be returned.
233 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
234 placed after all the others, C<msgs>. If this parameter is 0, this function
235 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
236 be a pointer to an C<HV *> variable, in which this function creates a new HV to
237 contain any appropriate messages. The hash has three key-value pairs, as
244 The text of the message as a C<SVpv>.
246 =item C<warn_categories>
248 The warning category (or categories) packed into a C<SVuv>.
252 A single flag bit associated with this message, in a C<SVuv>.
253 The bit corresponds to some bit in the C<*errors> return value,
254 such as C<UNICODE_GOT_SURROGATE>.
258 It's important to note that specifying this parameter as non-null will cause
259 any warnings this function would otherwise generate to be suppressed, and
260 instead be placed in C<*msgs>. The caller can check the lexical warnings state
261 (or not) when choosing what to do with the returned messages.
263 The caller, of course, is responsible for freeing any returned HV.
268 /* Undocumented; we don't want people using this. Instead they should use
269 * uvchr_to_utf8_flags_msgs() */
271 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
273 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
279 if (OFFUNI_IS_INVARIANT(uv)) {
280 *d++ = LATIN1_TO_NATIVE(uv);
284 if (uv <= MAX_UTF8_TWO_BYTE) {
285 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
286 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
290 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
291 * below, the 16 is for start bytes E0-EF (which are all the possible ones
292 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
293 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
294 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
295 * 0x800-0xFFFF on ASCII */
296 if (uv < (16 * (1U << (2 * SHIFT)))) {
297 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
298 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
299 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
301 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
302 aren't tested here */
303 /* The most likely code points in this range are below the surrogates.
304 * Do an extra test to quickly exclude those. */
305 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
306 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
307 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
309 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
311 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
312 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
319 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
320 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
321 * happen starting with 4-byte characters on ASCII platforms. We unify the
322 * code for these with EBCDIC, even though some of them require 5-bytes on
323 * those, because khw believes the code saving is worth the very slight
324 * performance hit on these high EBCDIC code points. */
326 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
327 if (UNLIKELY( uv > MAX_LEGAL_CP
328 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
330 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP);
332 if ( (flags & UNICODE_WARN_SUPER)
333 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
334 && UNICODE_IS_PERL_EXTENDED(uv)))
336 const char * format = super_cp_format;
337 U32 category = packWARN(WARN_NON_UNICODE);
338 U32 flag = UNICODE_GOT_SUPER;
340 /* Choose the more dire applicable warning */
341 if (UNICODE_IS_PERL_EXTENDED(uv)) {
342 format = perl_extended_cp_format;
343 if (flags & (UNICODE_WARN_PERL_EXTENDED
344 |UNICODE_DISALLOW_PERL_EXTENDED))
346 flag = UNICODE_GOT_PERL_EXTENDED;
351 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
355 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
358 if ( (flags & UNICODE_DISALLOW_SUPER)
359 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
360 && UNICODE_IS_PERL_EXTENDED(uv)))
365 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
366 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
369 /* Test for and handle 4-byte result. In the test immediately below, the
370 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
371 * characters). The 3 is for 3 continuation bytes; these each contribute
372 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
373 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
374 * 0x1_0000-0x1F_FFFF on ASCII */
375 if (uv < (8 * (1U << (3 * SHIFT)))) {
376 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
377 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
378 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
379 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
381 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
382 characters. The end-plane non-characters for EBCDIC were
383 handled just above */
384 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
385 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
387 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
388 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
395 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
396 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
397 * format. The unrolled version above turns out to not save all that much
398 * time, and at these high code points (well above the legal Unicode range
399 * on ASCII platforms, and well above anything in common use in EBCDIC),
400 * khw believes that less code outweighs slight performance gains. */
403 STRLEN len = OFFUNISKIP(uv);
406 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
409 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
415 =for apidoc uvchr_to_utf8
417 Adds the UTF-8 representation of the native code point C<uv> to the end
418 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
419 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
420 the byte after the end of the new character. In other words,
422 d = uvchr_to_utf8(d, uv);
424 is the recommended wide native character-aware way of saying
428 This function accepts any code point from 0..C<IV_MAX> as input.
429 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
431 It is possible to forbid or warn on non-Unicode code points, or those that may
432 be problematic by using L</uvchr_to_utf8_flags>.
437 /* This is also a macro */
438 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
441 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
443 return uvchr_to_utf8(d, uv);
447 =for apidoc uvchr_to_utf8_flags
449 Adds the UTF-8 representation of the native code point C<uv> to the end
450 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
451 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
452 the byte after the end of the new character. In other words,
454 d = uvchr_to_utf8_flags(d, uv, flags);
458 d = uvchr_to_utf8_flags(d, uv, 0);
460 This is the Unicode-aware way of saying
464 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
465 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
467 Specifying C<flags> can further restrict what is allowed and not warned on, as
470 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
471 the function will raise a warning, provided UTF8 warnings are enabled. If
472 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
473 NULL. If both flags are set, the function will both warn and return NULL.
475 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
476 affect how the function handles a Unicode non-character.
478 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
479 affect the handling of code points that are above the Unicode maximum of
480 0x10FFFF. Languages other than Perl may not be able to accept files that
483 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
484 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
485 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
486 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
487 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
488 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
489 above-Unicode and surrogate flags, but not the non-character ones, as
491 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
492 See L<perlunicode/Noncharacter code points>.
494 Extremely high code points were never specified in any standard, and require an
495 extension to UTF-8 to express, which Perl does. It is likely that programs
496 written in something other than Perl would not be able to read files that
497 contain these; nor would Perl understand files written by something that uses a
498 different extension. For these reasons, there is a separate set of flags that
499 can warn and/or disallow these extremely high code points, even if other
500 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
501 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
502 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
503 treat all above-Unicode code points, including these, as malformations. (Note
504 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
505 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
507 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
508 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
509 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
510 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
511 platforms,these flags can apply to code points that actually do fit in 31 bits.
512 The new names accurately describe the situation in all cases.
517 /* This is also a macro */
518 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
521 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
523 return uvchr_to_utf8_flags(d, uv, flags);
529 S_is_utf8_cp_above_31_bits(const U8 * const s,
531 const bool consider_overlongs)
533 /* Returns TRUE if the first code point represented by the Perl-extended-
534 * UTF-8-encoded string starting at 's', and looking no further than 'e -
535 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
537 * The function handles the case where the input bytes do not include all
538 * the ones necessary to represent a full character. That is, they may be
539 * the intial bytes of the representation of a code point, but possibly
540 * the final ones necessary for the complete representation may be beyond
543 * The function also can handle the case where the input is an overlong
544 * sequence. If 'consider_overlongs' is 0, the function assumes the
545 * input is not overlong, without checking, and will return based on that
546 * assumption. If this parameter is 1, the function will go to the trouble
547 * of figuring out if it actually evaluates to above or below 31 bits.
549 * The sequence is otherwise assumed to be well-formed, without checking.
552 const STRLEN len = e - s;
555 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
557 assert(! UTF8_IS_INVARIANT(*s) && e > s);
561 PERL_UNUSED_ARG(consider_overlongs);
563 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
564 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
565 * also be the start byte for a 31-bit code point; we need at least 2
566 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
567 * the start byte for an overlong sequence, but for 30-bit or smaller code
568 * points, so we don't have to worry about overlongs on EBCDIC.) */
579 /* On ASCII, FE and FF are the only start bytes that can evaluate to
580 * needing more than 31 bits. */
581 if (LIKELY(*s < 0xFE)) {
585 /* What we have left are FE and FF. Both of these require more than 31
586 * bits unless they are for overlongs. */
587 if (! consider_overlongs) {
591 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
592 * above 31 bits. But we need more than one byte to discern this, so if
593 * passed just the start byte, it could be an overlong evaluating to
599 /* Having excluded len==1, and knowing that FE and FF are both valid start
600 * bytes, we can call the function below to see if the sequence is
601 * overlong. (We don't need the full generality of the called function,
602 * but for these huge code points, speed shouldn't be a consideration, and
603 * the compiler does have enough information, since it's static to this
604 * file, to optimize to just the needed parts.) */
605 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
607 /* If it isn't overlong, more than 31 bits are required. */
608 if (is_overlong == 0) {
612 /* If it is indeterminate if it is overlong, return that */
613 if (is_overlong < 0) {
617 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
618 * the max it can be is 2**31 - 1 */
625 /* Here, ASCII and EBCDIC rejoin:
626 * On ASCII: We have an overlong sequence starting with FF
627 * On EBCDIC: We have a sequence starting with FE. */
629 { /* For C89, use a block so the declaration can be close to its use */
633 /* U+7FFFFFFF (2 ** 31 - 1)
634 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
635 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
636 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
637 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
638 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
639 * U+80000000 (2 ** 31):
640 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
641 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
642 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
643 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
645 * and since we know that *s = \xfe, any continuation sequcence
646 * following it that is gt the below is above 31 bits
647 [0] [1] [2] [3] [4] [5] [6] */
648 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
652 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
653 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
654 * FF overlong for U+80000000 (2 ** 31):
655 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
656 * and since we know that *s = \xff, any continuation sequcence
657 * following it that is gt the below is above 30 bits
658 [0] [1] [2] [3] [4] [5] [6] */
659 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
663 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
664 const STRLEN cmp_len = MIN(conts_len, len - 1);
666 /* Now compare the continuation bytes in s with the ones we have
667 * compiled in that are for the largest 30 bit code point. If we have
668 * enough bytes available to determine the answer, or the bytes we do
669 * have differ from them, we can compare the two to get a definitive
670 * answer (Note that in UTF-EBCDIC, the two lowest possible
671 * continuation bytes are \x41 and \x42.) */
672 if (cmp_len >= conts_len || memNE(s + 1,
673 conts_for_highest_30_bit,
676 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
679 /* Here, all the bytes we have are the same as the highest 30-bit code
680 * point, but we are missing so many bytes that we can't make the
688 PERL_STATIC_INLINE int
689 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
691 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
692 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
693 * it isn't, and -1 if there isn't enough information to tell. This last
694 * return value can happen if the sequence is incomplete, missing some
695 * trailing bytes that would form a complete character. If there are
696 * enough bytes to make a definitive decision, this function does so.
697 * Usually 2 bytes sufficient.
699 * Overlongs can occur whenever the number of continuation bytes changes.
700 * That means whenever the number of leading 1 bits in a start byte
701 * increases from the next lower start byte. That happens for start bytes
702 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
703 * start bytes have already been excluded, so don't need to be tested here;
704 * ASCII platforms: C0, C1
705 * EBCDIC platforms C0, C1, C2, C3, C4, E0
708 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
709 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
711 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
712 assert(len > 1 && UTF8_IS_START(*s));
714 /* Each platform has overlongs after the start bytes given above (expressed
715 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
716 * the logic is the same, except the E0 overlong has already been excluded
717 * on EBCDIC platforms. The values below were found by manually
718 * inspecting the UTF-8 patterns. See the tables in utf8.h and
722 # define F0_ABOVE_OVERLONG 0xB0
723 # define F8_ABOVE_OVERLONG 0xA8
724 # define FC_ABOVE_OVERLONG 0xA4
725 # define FE_ABOVE_OVERLONG 0xA2
726 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
730 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
734 # define F0_ABOVE_OVERLONG 0x90
735 # define F8_ABOVE_OVERLONG 0x88
736 # define FC_ABOVE_OVERLONG 0x84
737 # define FE_ABOVE_OVERLONG 0x82
738 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
742 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
743 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
744 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
745 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
750 /* Check for the FF overlong */
751 return isFF_OVERLONG(s, len);
754 PERL_STATIC_INLINE int
755 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
757 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
758 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
759 * it isn't, and -1 if there isn't enough information to tell. This last
760 * return value can happen if the sequence is incomplete, missing some
761 * trailing bytes that would form a complete character. If there are
762 * enough bytes to make a definitive decision, this function does so. */
764 PERL_ARGS_ASSERT_ISFF_OVERLONG;
766 /* To be an FF overlong, all the available bytes must match */
767 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
768 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
773 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
774 * be there; what comes after them doesn't matter. See tables in utf8.h,
776 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
780 /* The missing bytes could cause the result to go one way or the other, so
781 * the result is indeterminate */
785 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
786 # ifdef EBCDIC /* Actually is I8 */
787 # define HIGHEST_REPRESENTABLE_UTF8 \
788 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
790 # define HIGHEST_REPRESENTABLE_UTF8 \
791 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
795 PERL_STATIC_INLINE int
796 S_does_utf8_overflow(const U8 * const s,
798 const bool consider_overlongs)
800 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
801 * 'e' - 1 would overflow an IV on this platform; that is if it represents
802 * a code point larger than the highest representable code point. It
803 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
804 * enough information to tell. This last return value can happen if the
805 * sequence is incomplete, missing some trailing bytes that would form a
806 * complete character. If there are enough bytes to make a definitive
807 * decision, this function does so.
809 * If 'consider_overlongs' is TRUE, the function checks for the possibility
810 * that the sequence is an overlong that doesn't overflow. Otherwise, it
811 * assumes the sequence is not an overlong. This can give different
812 * results only on ASCII 32-bit platforms.
814 * (For ASCII platforms, we could use memcmp() because we don't have to
815 * convert each byte to I8, but it's very rare input indeed that would
816 * approach overflow, so the loop below will likely only get executed once.)
818 * 'e' - 1 must not be beyond a full character. */
821 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
822 assert(s <= e && s + UTF8SKIP(s) >= e);
824 #if ! defined(UV_IS_QUAD)
826 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
830 PERL_UNUSED_ARG(consider_overlongs);
833 const STRLEN len = e - s;
835 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
837 for (x = s; x < e; x++, y++) {
839 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
843 /* If this byte is larger than the corresponding highest UTF-8
844 * byte, the sequence overflow; otherwise the byte is less than,
845 * and so the sequence doesn't overflow */
846 return NATIVE_UTF8_TO_I8(*x) > *y;
850 /* Got to the end and all bytes are the same. If the input is a whole
851 * character, it doesn't overflow. And if it is a partial character,
852 * there's not enough information to tell */
853 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
866 /* This is the portions of the above function that deal with UV_MAX instead of
867 * IV_MAX. They are left here in case we want to combine them so that internal
868 * uses can have larger code points. The only logic difference is that the
869 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
873 /* Anything larger than this will overflow the word if it were converted into a UV */
874 #if defined(UV_IS_QUAD)
875 # ifdef EBCDIC /* Actually is I8 */
876 # define HIGHEST_REPRESENTABLE_UTF8 \
877 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
879 # define HIGHEST_REPRESENTABLE_UTF8 \
880 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
884 # define HIGHEST_REPRESENTABLE_UTF8 \
885 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
887 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
891 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
893 /* On 32 bit ASCII machines, many overlongs that start with FF don't
895 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
897 /* To be such an overlong, the first bytes of 's' must match
898 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
899 * don't have any additional bytes available, the sequence, when
900 * completed might or might not fit in 32 bits. But if we have that
901 * next byte, we can tell for sure. If it is <= 0x83, then it does
903 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
907 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
910 /* Starting with the #else, the rest of the function is identical except
911 * 1. we need to move the 'len' declaration to be global to the function
912 * 2. the endif move to just after the UNUSED_ARG.
913 * An empty endif is given just below to satisfy the preprocessor
919 #undef F0_ABOVE_OVERLONG
920 #undef F8_ABOVE_OVERLONG
921 #undef FC_ABOVE_OVERLONG
922 #undef FE_ABOVE_OVERLONG
923 #undef FF_OVERLONG_PREFIX
926 Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
931 /* A helper function that should not be called directly.
933 * This function returns non-zero if the string beginning at 's' and
934 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
935 * code point; otherwise it returns 0. The examination stops after the
936 * first code point in 's' is validated, not looking at the rest of the
937 * input. If 'e' is such that there are not enough bytes to represent a
938 * complete code point, this function will return non-zero anyway, if the
939 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
940 * excluded by 'flags'.
942 * A non-zero return gives the number of bytes required to represent the
943 * code point. Be aware that if the input is for a partial character, the
944 * return will be larger than 'e - s'.
946 * This function assumes that the code point represented is UTF-8 variant.
947 * The caller should have excluded the possibility of it being invariant
948 * before calling this function.
950 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
951 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
952 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
953 * disallowed by the flags. If the input is only for a partial character,
954 * the function will return non-zero if there is any sequence of
955 * well-formed UTF-8 that, when appended to the input sequence, could
956 * result in an allowed code point; otherwise it returns 0. Non characters
957 * cannot be determined based on partial character input. But many of the
958 * other excluded types can be determined with just the first one or two
963 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
965 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
966 |UTF8_DISALLOW_PERL_EXTENDED)));
967 assert(! UTF8_IS_INVARIANT(*s));
969 /* A variant char must begin with a start byte */
970 if (UNLIKELY(! UTF8_IS_START(*s))) {
974 /* Examine a maximum of a single whole code point */
975 if (e - s > UTF8SKIP(s)) {
981 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
982 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
984 /* Here, we are disallowing some set of largish code points, and the
985 * first byte indicates the sequence is for a code point that could be
986 * in the excluded set. We generally don't have to look beyond this or
987 * the second byte to see if the sequence is actually for one of the
988 * excluded classes. The code below is derived from this table:
990 * UTF-8 UTF-EBCDIC I8
991 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
992 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
993 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
995 * Keep in mind that legal continuation bytes range between \x80..\xBF
996 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
997 * continuation bytes. Hence, we don't have to test the upper edge
998 * because if any of those is encountered, the sequence is malformed,
999 * and would fail elsewhere in this function.
1001 * The code here likewise assumes that there aren't other
1002 * malformations; again the function should fail elsewhere because of
1003 * these. For example, an overlong beginning with FC doesn't actually
1004 * have to be a super; it could actually represent a small code point,
1005 * even U+0000. But, since overlongs (and other malformations) are
1006 * illegal, the function should return FALSE in either case.
1009 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1010 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
1011 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
1013 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1015 && ((s1) & 0xFE ) == 0xB6)
1016 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1018 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1019 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1020 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1021 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1024 if ( (flags & UTF8_DISALLOW_SUPER)
1025 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1027 return 0; /* Above Unicode */
1030 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1031 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1037 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1039 if ( (flags & UTF8_DISALLOW_SUPER)
1040 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1042 return 0; /* Above Unicode */
1045 if ( (flags & UTF8_DISALLOW_SURROGATE)
1046 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1048 return 0; /* Surrogate */
1051 if ( (flags & UTF8_DISALLOW_NONCHAR)
1052 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1054 return 0; /* Noncharacter code point */
1059 /* Make sure that all that follows are continuation bytes */
1060 for (x = s + 1; x < e; x++) {
1061 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1066 /* Here is syntactically valid. Next, make sure this isn't the start of an
1068 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1072 /* And finally, that the code point represented fits in a word on this
1074 if (0 < does_utf8_overflow(s, e,
1075 0 /* Don't consider overlongs */
1085 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1087 /* Returns a mortalized C string that is a displayable copy of the 'len'
1088 * bytes starting at 'start'. 'format' gives how to display each byte.
1089 * Currently, there are only two formats, so it is currently a bool:
1091 * 1 ab (that is a space between two hex digit bytes)
1094 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1096 const U8 * s = start;
1097 const U8 * const e = start + len;
1101 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1103 Newx(output, output_len, char);
1107 for (s = start; s < e; s++) {
1108 const unsigned high_nibble = (*s & 0xF0) >> 4;
1109 const unsigned low_nibble = (*s & 0x0F);
1121 if (high_nibble < 10) {
1122 *d++ = high_nibble + '0';
1125 *d++ = high_nibble - 10 + 'a';
1128 if (low_nibble < 10) {
1129 *d++ = low_nibble + '0';
1132 *d++ = low_nibble - 10 + 'a';
1140 PERL_STATIC_INLINE char *
1141 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1143 /* Max number of bytes to print */
1146 /* Which one is the non-continuation */
1147 const STRLEN non_cont_byte_pos,
1149 /* How many bytes should there be? */
1150 const STRLEN expect_len)
1152 /* Return the malformation warning text for an unexpected continuation
1155 const char * const where = (non_cont_byte_pos == 1)
1157 : Perl_form(aTHX_ "%d bytes",
1158 (int) non_cont_byte_pos);
1159 const U8 * x = s + non_cont_byte_pos;
1160 const U8 * e = s + print_len;
1162 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1164 /* We don't need to pass this parameter, but since it has already been
1165 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1166 assert(expect_len == UTF8SKIP(s));
1168 /* As a defensive coding measure, don't output anything past a NUL. Such
1169 * bytes shouldn't be in the middle of a malformation, and could mark the
1170 * end of the allocated string, and what comes after is undefined */
1171 for (; x < e; x++) {
1173 x++; /* Output this particular NUL */
1178 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1179 " %s after start byte 0x%02x; need %d bytes, got %d)",
1181 _byte_dump_string(s, x - s, 0),
1182 *(s + non_cont_byte_pos),
1186 (int) non_cont_byte_pos);
1191 =for apidoc utf8n_to_uvchr
1193 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1194 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1197 Bottom level UTF-8 decode routine.
1198 Returns the native code point value of the first character in the string C<s>,
1199 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1200 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1201 the length, in bytes, of that character.
1203 The value of C<flags> determines the behavior when C<s> does not point to a
1204 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1205 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1206 is the next possible position in C<s> that could begin a non-malformed
1207 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1208 is raised. Some UTF-8 input sequences may contain multiple malformations.
1209 This function tries to find every possible one in each call, so multiple
1210 warnings can be raised for the same sequence.
1212 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1213 individual types of malformations, such as the sequence being overlong (that
1214 is, when there is a shorter sequence that can express the same code point;
1215 overlong sequences are expressly forbidden in the UTF-8 standard due to
1216 potential security issues). Another malformation example is the first byte of
1217 a character not being a legal first byte. See F<utf8.h> for the list of such
1218 flags. Even if allowed, this function generally returns the Unicode
1219 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1220 F<utf8.h> to override this behavior for the overlong malformations, but don't
1221 do that except for very specialized purposes.
1223 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1224 flags) malformation is found. If this flag is set, the routine assumes that
1225 the caller will raise a warning, and this function will silently just set
1226 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1228 Note that this API requires disambiguation between successful decoding a C<NUL>
1229 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1230 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1231 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1232 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1233 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1235 Certain code points are considered problematic. These are Unicode surrogates,
1236 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1237 By default these are considered regular code points, but certain situations
1238 warrant special handling for them, which can be specified using the C<flags>
1239 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1240 three classes are treated as malformations and handled as such. The flags
1241 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1242 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1243 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1244 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1245 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1247 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
1248 The difference between traditional strictness and C9 strictness is that the
1249 latter does not forbid non-character code points. (They are still discouraged,
1250 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1252 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1253 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1254 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1255 raised for their respective categories, but otherwise the code points are
1256 considered valid (not malformations). To get a category to both be treated as
1257 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1258 (But note that warnings are not raised if lexically disabled nor if
1259 C<UTF8_CHECK_ONLY> is also specified.)
1261 Extremely high code points were never specified in any standard, and require an
1262 extension to UTF-8 to express, which Perl does. It is likely that programs
1263 written in something other than Perl would not be able to read files that
1264 contain these; nor would Perl understand files written by something that uses a
1265 different extension. For these reasons, there is a separate set of flags that
1266 can warn and/or disallow these extremely high code points, even if other
1267 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1268 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1269 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1270 above-Unicode code points, including these, as malformations.
1271 (Note that the Unicode standard considers anything above 0x10FFFF to be
1272 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1275 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1276 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1277 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1278 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1279 can apply to code points that actually do fit in 31 bits. This happens on
1280 EBCDIC platforms, and sometimes when the L<overlong
1281 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1282 describe the situation in all cases.
1285 All other code points corresponding to Unicode characters, including private
1286 use and those yet to be assigned, are never considered malformed and never
1291 Also implemented as a macro in utf8.h
1295 Perl_utf8n_to_uvchr(const U8 *s,
1300 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1302 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1307 =for apidoc utf8n_to_uvchr_error
1309 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1310 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1313 This function is for code that needs to know what the precise malformation(s)
1314 are when an error is found. If you also need to know the generated warning
1315 messages, use L</utf8n_to_uvchr_msgs>() instead.
1317 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1318 all the others, C<errors>. If this parameter is 0, this function behaves
1319 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1320 to a C<U32> variable, which this function sets to indicate any errors found.
1321 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1322 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1323 of these bits will be set if a malformation is found, even if the input
1324 C<flags> parameter indicates that the given malformation is allowed; those
1325 exceptions are noted:
1329 =item C<UTF8_GOT_PERL_EXTENDED>
1331 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1332 set only if the input C<flags> parameter contains either the
1333 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1335 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1336 and so some extension must be used to express them. Perl uses a natural
1337 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1338 extension to represent even higher ones, so that any code point that fits in a
1339 64-bit word can be represented. Text using these extensions is not likely to
1340 be portable to non-Perl code. We lump both of these extensions together and
1341 refer to them as Perl extended UTF-8. There exist other extensions that people
1342 have invented, incompatible with Perl's.
1344 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1345 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1346 than on ASCII. Prior to that, code points 2**31 and higher were simply
1347 unrepresentable, and a different, incompatible method was used to represent
1348 code points between 2**30 and 2**31 - 1.
1350 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1351 Perl extended UTF-8 is used.
1353 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1354 may use for backward compatibility. That name is misleading, as this flag may
1355 be set when the code point actually does fit in 31 bits. This happens on
1356 EBCDIC platforms, and sometimes when the L<overlong
1357 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1358 describes the situation in all cases.
1360 =item C<UTF8_GOT_CONTINUATION>
1362 The input sequence was malformed in that the first byte was a a UTF-8
1365 =item C<UTF8_GOT_EMPTY>
1367 The input C<curlen> parameter was 0.
1369 =item C<UTF8_GOT_LONG>
1371 The input sequence was malformed in that there is some other sequence that
1372 evaluates to the same code point, but that sequence is shorter than this one.
1374 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1375 it was discovered that this created security issues.
1377 =item C<UTF8_GOT_NONCHAR>
1379 The code point represented by the input UTF-8 sequence is for a Unicode
1380 non-character code point.
1381 This bit is set only if the input C<flags> parameter contains either the
1382 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1384 =item C<UTF8_GOT_NON_CONTINUATION>
1386 The input sequence was malformed in that a non-continuation type byte was found
1387 in a position where only a continuation type one should be. See also
1388 L</C<UTF8_GOT_SHORT>>.
1390 =item C<UTF8_GOT_OVERFLOW>
1392 The input sequence was malformed in that it is for a code point that is not
1393 representable in the number of bits available in an IV on the current platform.
1395 =item C<UTF8_GOT_SHORT>
1397 The input sequence was malformed in that C<curlen> is smaller than required for
1398 a complete sequence. In other words, the input is for a partial character
1402 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1403 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1404 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1405 sequence was looked at. If no other flags are present, it means that the
1406 sequence was valid as far as it went. Depending on the application, this could
1407 mean one of three things:
1413 The C<curlen> length parameter passed in was too small, and the function was
1414 prevented from examining all the necessary bytes.
1418 The buffer being looked at is based on reading data, and the data received so
1419 far stopped in the middle of a character, so that the next read will
1420 read the remainder of this character. (It is up to the caller to deal with the
1421 split bytes somehow.)
1425 This is a real error, and the partial sequence is all we're going to get.
1429 =item C<UTF8_GOT_SUPER>
1431 The input sequence was malformed in that it is for a non-Unicode code point;
1432 that is, one above the legal Unicode maximum.
1433 This bit is set only if the input C<flags> parameter contains either the
1434 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1436 =item C<UTF8_GOT_SURROGATE>
1438 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1440 This bit is set only if the input C<flags> parameter contains either the
1441 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1445 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1446 flag to suppress any warnings, and then examine the C<*errors> return.
1450 Also implemented as a macro in utf8.h
1454 Perl_utf8n_to_uvchr_error(const U8 *s,
1460 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1462 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1467 =for apidoc utf8n_to_uvchr_msgs
1469 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1470 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1473 This function is for code that needs to know what the precise malformation(s)
1474 are when an error is found, and wants the corresponding warning and/or error
1475 messages to be returned to the caller rather than be displayed. All messages
1476 that would have been displayed if all lexcial warnings are enabled will be
1479 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1480 placed after all the others, C<msgs>. If this parameter is 0, this function
1481 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1482 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1483 contain any appropriate messages. The elements of the array are ordered so
1484 that the first message that would have been displayed is in the 0th element,
1485 and so on. Each element is a hash with three key-value pairs, as follows:
1491 The text of the message as a C<SVpv>.
1493 =item C<warn_categories>
1495 The warning category (or categories) packed into a C<SVuv>.
1499 A single flag bit associated with this message, in a C<SVuv>.
1500 The bit corresponds to some bit in the C<*errors> return value,
1501 such as C<UTF8_GOT_LONG>.
1505 It's important to note that specifying this parameter as non-null will cause
1506 any warnings this function would otherwise generate to be suppressed, and
1507 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1508 (or not) when choosing what to do with the returned messages.
1510 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1513 The caller, of course, is responsible for freeing any returned AV.
1519 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1526 const U8 * const s0 = s;
1527 const U8 * send = s0 + curlen;
1528 U32 possible_problems; /* A bit is set here for each potential problem
1529 found as we go along */
1531 STRLEN expectlen; /* How long should this sequence be? */
1532 STRLEN avail_len; /* When input is too short, gives what that is */
1533 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1534 gets set and discarded */
1536 /* The below are used only if there is both an overlong malformation and a
1537 * too short one. Otherwise the first two are set to 's0' and 'send', and
1538 * the third not used at all */
1540 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1541 routine; see [perl #130921] */
1545 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1547 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1548 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1549 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1554 /* Each of the affected Hanguls starts with \xED */
1556 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1567 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1568 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1569 | (s0[2] & UTF_CONTINUATION_MASK);
1574 /* In conjunction with the exhaustive tests that can be enabled in
1575 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1576 * what it is intended to do, and that no flaws in it are masked by
1577 * dropping down and executing the code below
1578 assert(! isUTF8_CHAR(s0, send)
1579 || UTF8_IS_SURROGATE(s0, send)
1580 || UTF8_IS_SUPER(s0, send)
1581 || UTF8_IS_NONCHAR(s0,send));
1586 possible_problems = 0;
1590 adjusted_s0 = (U8 *) s0;
1597 errors = &discard_errors;
1600 /* The order of malformation tests here is important. We should consume as
1601 * few bytes as possible in order to not skip any valid character. This is
1602 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1603 * https://unicode.org/reports/tr36 for more discussion as to why. For
1604 * example, once we've done a UTF8SKIP, we can tell the expected number of
1605 * bytes, and could fail right off the bat if the input parameters indicate
1606 * that there are too few available. But it could be that just that first
1607 * byte is garbled, and the intended character occupies fewer bytes. If we
1608 * blindly assumed that the first byte is correct, and skipped based on
1609 * that number, we could skip over a valid input character. So instead, we
1610 * always examine the sequence byte-by-byte.
1612 * We also should not consume too few bytes, otherwise someone could inject
1613 * things. For example, an input could be deliberately designed to
1614 * overflow, and if this code bailed out immediately upon discovering that,
1615 * returning to the caller C<*retlen> pointing to the very next byte (one
1616 * which is actually part of of the overflowing sequence), that could look
1617 * legitimate to the caller, which could discard the initial partial
1618 * sequence and process the rest, inappropriately.
1620 * Some possible input sequences are malformed in more than one way. This
1621 * function goes to lengths to try to find all of them. This is necessary
1622 * for correctness, as the inputs may allow one malformation but not
1623 * another, and if we abandon searching for others after finding the
1624 * allowed one, we could allow in something that shouldn't have been.
1627 if (UNLIKELY(curlen == 0)) {
1628 possible_problems |= UTF8_GOT_EMPTY;
1630 uv = UNICODE_REPLACEMENT;
1631 goto ready_to_handle_errors;
1634 expectlen = UTF8SKIP(s);
1636 /* A well-formed UTF-8 character, as the vast majority of calls to this
1637 * function will be for, has this expected length. For efficiency, set
1638 * things up here to return it. It will be overriden only in those rare
1639 * cases where a malformation is found */
1641 *retlen = expectlen;
1644 /* A continuation character can't start a valid sequence */
1645 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1646 possible_problems |= UTF8_GOT_CONTINUATION;
1648 uv = UNICODE_REPLACEMENT;
1649 goto ready_to_handle_errors;
1652 /* Here is not a continuation byte, nor an invariant. The only thing left
1653 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1654 * because it excludes start bytes like \xC0 that always lead to
1657 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1658 * that indicate the number of bytes in the character's whole UTF-8
1659 * sequence, leaving just the bits that are part of the value. */
1660 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1662 /* Setup the loop end point, making sure to not look past the end of the
1663 * input string, and flag it as too short if the size isn't big enough. */
1664 if (UNLIKELY(curlen < expectlen)) {
1665 possible_problems |= UTF8_GOT_SHORT;
1669 send = (U8*) s0 + expectlen;
1672 /* Now, loop through the remaining bytes in the character's sequence,
1673 * accumulating each into the working value as we go. */
1674 for (s = s0 + 1; s < send; s++) {
1675 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1676 uv = UTF8_ACCUMULATE(uv, *s);
1680 /* Here, found a non-continuation before processing all expected bytes.
1681 * This byte indicates the beginning of a new character, so quit, even
1682 * if allowing this malformation. */
1683 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1685 } /* End of loop through the character's bytes */
1687 /* Save how many bytes were actually in the character */
1690 /* Note that there are two types of too-short malformation. One is when
1691 * there is actual wrong data before the normal termination of the
1692 * sequence. The other is that the sequence wasn't complete before the end
1693 * of the data we are allowed to look at, based on the input 'curlen'.
1694 * This means that we were passed data for a partial character, but it is
1695 * valid as far as we saw. The other is definitely invalid. This
1696 * distinction could be important to a caller, so the two types are kept
1699 * A convenience macro that matches either of the too-short conditions. */
1700 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1702 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1704 uv = UNICODE_REPLACEMENT;
1707 /* Check for overflow. The algorithm requires us to not look past the end
1708 * of the current character, even if partial, so the upper limit is 's' */
1709 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1710 1 /* Do consider overlongs */
1713 possible_problems |= UTF8_GOT_OVERFLOW;
1714 uv = UNICODE_REPLACEMENT;
1717 /* Check for overlong. If no problems so far, 'uv' is the correct code
1718 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1719 * we must look at the UTF-8 byte sequence itself to see if it is for an
1721 if ( ( LIKELY(! possible_problems)
1722 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1723 || ( UNLIKELY(possible_problems)
1724 && ( UNLIKELY(! UTF8_IS_START(*s0))
1726 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1729 possible_problems |= UTF8_GOT_LONG;
1731 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1733 /* The calculation in the 'true' branch of this 'if'
1734 * below won't work if overflows, and isn't needed
1735 * anyway. Further below we handle all overflow
1737 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1739 UV min_uv = uv_so_far;
1742 /* Here, the input is both overlong and is missing some trailing
1743 * bytes. There is no single code point it could be for, but there
1744 * may be enough information present to determine if what we have
1745 * so far is for an unallowed code point, such as for a surrogate.
1746 * The code further below has the intelligence to determine this,
1747 * but just for non-overlong UTF-8 sequences. What we do here is
1748 * calculate the smallest code point the input could represent if
1749 * there were no too short malformation. Then we compute and save
1750 * the UTF-8 for that, which is what the code below looks at
1751 * instead of the raw input. It turns out that the smallest such
1752 * code point is all we need. */
1753 for (i = curlen; i < expectlen; i++) {
1754 min_uv = UTF8_ACCUMULATE(min_uv,
1755 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1758 adjusted_s0 = temp_char_buf;
1759 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1763 /* Here, we have found all the possible problems, except for when the input
1764 * is for a problematic code point not allowed by the input parameters. */
1766 /* uv is valid for overlongs */
1767 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1769 /* isn't problematic if < this */
1770 && uv >= UNICODE_SURROGATE_FIRST)
1771 || ( UNLIKELY(possible_problems)
1773 /* if overflow, we know without looking further
1774 * precisely which of the problematic types it is,
1775 * and we deal with those in the overflow handling
1777 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1778 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1779 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1780 && ((flags & ( UTF8_DISALLOW_NONCHAR
1781 |UTF8_DISALLOW_SURROGATE
1782 |UTF8_DISALLOW_SUPER
1783 |UTF8_DISALLOW_PERL_EXTENDED
1785 |UTF8_WARN_SURROGATE
1787 |UTF8_WARN_PERL_EXTENDED))))
1789 /* If there were no malformations, or the only malformation is an
1790 * overlong, 'uv' is valid */
1791 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1792 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1793 possible_problems |= UTF8_GOT_SURROGATE;
1795 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1796 possible_problems |= UTF8_GOT_SUPER;
1798 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1799 possible_problems |= UTF8_GOT_NONCHAR;
1802 else { /* Otherwise, need to look at the source UTF-8, possibly
1803 adjusted to be non-overlong */
1805 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1806 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1808 possible_problems |= UTF8_GOT_SUPER;
1810 else if (curlen > 1) {
1811 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1812 NATIVE_UTF8_TO_I8(*adjusted_s0),
1813 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1815 possible_problems |= UTF8_GOT_SUPER;
1817 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1818 NATIVE_UTF8_TO_I8(*adjusted_s0),
1819 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1821 possible_problems |= UTF8_GOT_SURROGATE;
1825 /* We need a complete well-formed UTF-8 character to discern
1826 * non-characters, so can't look for them here */
1830 ready_to_handle_errors:
1833 * curlen contains the number of bytes in the sequence that
1834 * this call should advance the input by.
1835 * avail_len gives the available number of bytes passed in, but
1836 * only if this is less than the expected number of
1837 * bytes, based on the code point's start byte.
1838 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1839 * is set in it for each potential problem found.
1840 * uv contains the code point the input sequence
1841 * represents; or if there is a problem that prevents
1842 * a well-defined value from being computed, it is
1843 * some subsitute value, typically the REPLACEMENT
1845 * s0 points to the first byte of the character
1846 * s points to just after were we left off processing
1848 * send points to just after where that character should
1849 * end, based on how many bytes the start byte tells
1850 * us should be in it, but no further than s0 +
1854 if (UNLIKELY(possible_problems)) {
1855 bool disallowed = FALSE;
1856 const U32 orig_problems = possible_problems;
1862 while (possible_problems) { /* Handle each possible problem */
1864 char * message = NULL;
1865 U32 this_flag_bit = 0;
1867 /* Each 'if' clause handles one problem. They are ordered so that
1868 * the first ones' messages will be displayed before the later
1869 * ones; this is kinda in decreasing severity order. But the
1870 * overlong must come last, as it changes 'uv' looked at by the
1872 if (possible_problems & UTF8_GOT_OVERFLOW) {
1874 /* Overflow means also got a super and are using Perl's
1875 * extended UTF-8, but we handle all three cases here */
1877 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1878 *errors |= UTF8_GOT_OVERFLOW;
1880 /* But the API says we flag all errors found */
1881 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1882 *errors |= UTF8_GOT_SUPER;
1885 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1887 *errors |= UTF8_GOT_PERL_EXTENDED;
1890 /* Disallow if any of the three categories say to */
1891 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1892 || (flags & ( UTF8_DISALLOW_SUPER
1893 |UTF8_DISALLOW_PERL_EXTENDED)))
1898 /* Likewise, warn if any say to */
1899 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1900 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1903 /* The warnings code explicitly says it doesn't handle the
1904 * case of packWARN2 and two categories which have
1905 * parent-child relationship. Even if it works now to
1906 * raise the warning if either is enabled, it wouldn't
1907 * necessarily do so in the future. We output (only) the
1908 * most dire warning */
1909 if (! (flags & UTF8_CHECK_ONLY)) {
1910 if (msgs || ckWARN_d(WARN_UTF8)) {
1911 pack_warn = packWARN(WARN_UTF8);
1913 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1914 pack_warn = packWARN(WARN_NON_UNICODE);
1917 message = Perl_form(aTHX_ "%s: %s (overflows)",
1919 _byte_dump_string(s0, curlen, 0));
1920 this_flag_bit = UTF8_GOT_OVERFLOW;
1925 else if (possible_problems & UTF8_GOT_EMPTY) {
1926 possible_problems &= ~UTF8_GOT_EMPTY;
1927 *errors |= UTF8_GOT_EMPTY;
1929 if (! (flags & UTF8_ALLOW_EMPTY)) {
1931 /* This so-called malformation is now treated as a bug in
1932 * the caller. If you have nothing to decode, skip calling
1938 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1940 pack_warn = packWARN(WARN_UTF8);
1941 message = Perl_form(aTHX_ "%s (empty string)",
1943 this_flag_bit = UTF8_GOT_EMPTY;
1947 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1948 possible_problems &= ~UTF8_GOT_CONTINUATION;
1949 *errors |= UTF8_GOT_CONTINUATION;
1951 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1954 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1956 pack_warn = packWARN(WARN_UTF8);
1957 message = Perl_form(aTHX_
1958 "%s: %s (unexpected continuation byte 0x%02x,"
1959 " with no preceding start byte)",
1961 _byte_dump_string(s0, 1, 0), *s0);
1962 this_flag_bit = UTF8_GOT_CONTINUATION;
1966 else if (possible_problems & UTF8_GOT_SHORT) {
1967 possible_problems &= ~UTF8_GOT_SHORT;
1968 *errors |= UTF8_GOT_SHORT;
1970 if (! (flags & UTF8_ALLOW_SHORT)) {
1973 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1975 pack_warn = packWARN(WARN_UTF8);
1976 message = Perl_form(aTHX_
1977 "%s: %s (too short; %d byte%s available, need %d)",
1979 _byte_dump_string(s0, send - s0, 0),
1981 avail_len == 1 ? "" : "s",
1983 this_flag_bit = UTF8_GOT_SHORT;
1988 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1989 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1990 *errors |= UTF8_GOT_NON_CONTINUATION;
1992 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1995 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1998 /* If we don't know for sure that the input length is
1999 * valid, avoid as much as possible reading past the
2000 * end of the buffer */
2001 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2004 pack_warn = packWARN(WARN_UTF8);
2005 message = Perl_form(aTHX_ "%s",
2006 unexpected_non_continuation_text(s0,
2010 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2014 else if (possible_problems & UTF8_GOT_SURROGATE) {
2015 possible_problems &= ~UTF8_GOT_SURROGATE;
2017 if (flags & UTF8_WARN_SURROGATE) {
2018 *errors |= UTF8_GOT_SURROGATE;
2020 if ( ! (flags & UTF8_CHECK_ONLY)
2021 && (msgs || ckWARN_d(WARN_SURROGATE)))
2023 pack_warn = packWARN(WARN_SURROGATE);
2025 /* These are the only errors that can occur with a
2026 * surrogate when the 'uv' isn't valid */
2027 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2028 message = Perl_form(aTHX_
2029 "UTF-16 surrogate (any UTF-8 sequence that"
2030 " starts with \"%s\" is for a surrogate)",
2031 _byte_dump_string(s0, curlen, 0));
2034 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2036 this_flag_bit = UTF8_GOT_SURROGATE;
2040 if (flags & UTF8_DISALLOW_SURROGATE) {
2042 *errors |= UTF8_GOT_SURROGATE;
2045 else if (possible_problems & UTF8_GOT_SUPER) {
2046 possible_problems &= ~UTF8_GOT_SUPER;
2048 if (flags & UTF8_WARN_SUPER) {
2049 *errors |= UTF8_GOT_SUPER;
2051 if ( ! (flags & UTF8_CHECK_ONLY)
2052 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2054 pack_warn = packWARN(WARN_NON_UNICODE);
2056 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2057 message = Perl_form(aTHX_
2058 "Any UTF-8 sequence that starts with"
2059 " \"%s\" is for a non-Unicode code point,"
2060 " may not be portable",
2061 _byte_dump_string(s0, curlen, 0));
2064 message = Perl_form(aTHX_ super_cp_format, uv);
2066 this_flag_bit = UTF8_GOT_SUPER;
2070 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2071 * and before possibly bailing out, so that the more dire
2072 * warning will override the regular one. */
2073 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2074 if ( ! (flags & UTF8_CHECK_ONLY)
2075 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2076 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2078 pack_warn = packWARN(WARN_NON_UNICODE);
2080 /* If it is an overlong that evaluates to a code point
2081 * that doesn't have to use the Perl extended UTF-8, it
2082 * still used it, and so we output a message that
2083 * doesn't refer to the code point. The same is true
2084 * if there was a SHORT malformation where the code
2085 * point is not valid. In that case, 'uv' will have
2086 * been set to the REPLACEMENT CHAR, and the message
2087 * below without the code point in it will be selected
2089 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2090 message = Perl_form(aTHX_
2091 perl_extended_cp_format, uv);
2094 message = Perl_form(aTHX_
2095 "Any UTF-8 sequence that starts with"
2096 " \"%s\" is a Perl extension, and"
2097 " so is not portable",
2098 _byte_dump_string(s0, curlen, 0));
2100 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2103 if (flags & ( UTF8_WARN_PERL_EXTENDED
2104 |UTF8_DISALLOW_PERL_EXTENDED))
2106 *errors |= UTF8_GOT_PERL_EXTENDED;
2108 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2114 if (flags & UTF8_DISALLOW_SUPER) {
2115 *errors |= UTF8_GOT_SUPER;
2119 else if (possible_problems & UTF8_GOT_NONCHAR) {
2120 possible_problems &= ~UTF8_GOT_NONCHAR;
2122 if (flags & UTF8_WARN_NONCHAR) {
2123 *errors |= UTF8_GOT_NONCHAR;
2125 if ( ! (flags & UTF8_CHECK_ONLY)
2126 && (msgs || ckWARN_d(WARN_NONCHAR)))
2128 /* The code above should have guaranteed that we don't
2129 * get here with errors other than overlong */
2130 assert (! (orig_problems
2131 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2133 pack_warn = packWARN(WARN_NONCHAR);
2134 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2135 this_flag_bit = UTF8_GOT_NONCHAR;
2139 if (flags & UTF8_DISALLOW_NONCHAR) {
2141 *errors |= UTF8_GOT_NONCHAR;
2144 else if (possible_problems & UTF8_GOT_LONG) {
2145 possible_problems &= ~UTF8_GOT_LONG;
2146 *errors |= UTF8_GOT_LONG;
2148 if (flags & UTF8_ALLOW_LONG) {
2150 /* We don't allow the actual overlong value, unless the
2151 * special extra bit is also set */
2152 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2153 & ~UTF8_ALLOW_LONG)))
2155 uv = UNICODE_REPLACEMENT;
2162 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2164 pack_warn = packWARN(WARN_UTF8);
2166 /* These error types cause 'uv' to be something that
2167 * isn't what was intended, so can't use it in the
2168 * message. The other error types either can't
2169 * generate an overlong, or else the 'uv' is valid */
2171 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2173 message = Perl_form(aTHX_
2174 "%s: %s (any UTF-8 sequence that starts"
2175 " with \"%s\" is overlong which can and"
2176 " should be represented with a"
2177 " different, shorter sequence)",
2179 _byte_dump_string(s0, send - s0, 0),
2180 _byte_dump_string(s0, curlen, 0));
2183 U8 tmpbuf[UTF8_MAXBYTES+1];
2184 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2186 /* Don't use U+ for non-Unicode code points, which
2187 * includes those in the Latin1 range */
2188 const char * preface = ( uv > PERL_UNICODE_MAX
2195 message = Perl_form(aTHX_
2196 "%s: %s (overlong; instead use %s to represent"
2199 _byte_dump_string(s0, send - s0, 0),
2200 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2202 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2203 small code points */
2206 this_flag_bit = UTF8_GOT_LONG;
2209 } /* End of looking through the possible flags */
2211 /* Display the message (if any) for the problem being handled in
2212 * this iteration of the loop */
2215 assert(this_flag_bit);
2217 if (*msgs == NULL) {
2221 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2226 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2229 Perl_warner(aTHX_ pack_warn, "%s", message);
2231 } /* End of 'while (possible_problems)' */
2233 /* Since there was a possible problem, the returned length may need to
2234 * be changed from the one stored at the beginning of this function.
2235 * Instead of trying to figure out if that's needed, just do it. */
2241 if (flags & UTF8_CHECK_ONLY && retlen) {
2242 *retlen = ((STRLEN) -1);
2248 return UNI_TO_NATIVE(uv);
2252 =for apidoc utf8_to_uvchr_buf
2254 Returns the native code point of the first character in the string C<s> which
2255 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2256 C<*retlen> will be set to the length, in bytes, of that character.
2258 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2259 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2260 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2261 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2262 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2263 the next possible position in C<s> that could begin a non-malformed character.
2264 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2269 Also implemented as a macro in utf8.h
2275 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2277 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2279 return utf8_to_uvchr_buf_helper(s, send, retlen);
2282 /* This is marked as deprecated
2284 =for apidoc utf8_to_uvuni_buf
2286 Only in very rare circumstances should code need to be dealing in Unicode
2287 (as opposed to native) code points. In those few cases, use
2288 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2289 If you are not absolutely sure this is one of those cases, then assume it isn't
2290 and use plain C<utf8_to_uvchr_buf> instead.
2292 Returns the Unicode (not-native) code point of the first character in the
2294 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2295 C<retlen> will be set to the length, in bytes, of that character.
2297 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2298 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2299 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2300 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2301 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2302 next possible position in C<s> that could begin a non-malformed character.
2303 See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2310 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2312 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2316 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2320 =for apidoc utf8_length
2322 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2323 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2324 same place, it returns 0 with no warning raised.
2326 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2327 and returns the number of valid characters.
2333 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2337 PERL_ARGS_ASSERT_UTF8_LENGTH;
2339 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2340 * the bitops (especially ~) can create illegal UTF-8.
2341 * In other words: in Perl UTF-8 is not just for Unicode. */
2343 if (UNLIKELY(e < s))
2344 goto warn_and_return;
2350 if (UNLIKELY(e != s)) {
2354 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2355 "%s in %s", unees, OP_DESC(PL_op));
2357 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2364 =for apidoc bytes_cmp_utf8
2366 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2367 sequence of characters (stored as UTF-8)
2368 in C<u>, C<ulen>. Returns 0 if they are
2369 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2370 if the first string is greater than the second string.
2372 -1 or +1 is returned if the shorter string was identical to the start of the
2373 longer string. -2 or +2 is returned if
2374 there was a difference between characters
2381 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2383 const U8 *const bend = b + blen;
2384 const U8 *const uend = u + ulen;
2386 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2388 while (b < bend && u < uend) {
2390 if (!UTF8_IS_INVARIANT(c)) {
2391 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2394 if (UTF8_IS_CONTINUATION(c1)) {
2395 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2397 /* diag_listed_as: Malformed UTF-8 character%s */
2398 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2400 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2401 PL_op ? " in " : "",
2402 PL_op ? OP_DESC(PL_op) : "");
2407 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2408 "%s in %s", unees, OP_DESC(PL_op));
2410 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2411 return -2; /* Really want to return undef :-) */
2418 return *b < c ? -2 : +2;
2423 if (b == bend && u == uend)
2426 return b < bend ? +1 : -1;
2430 =for apidoc utf8_to_bytes
2432 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2433 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2434 updates C<*lenp> to contain the new length.
2435 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2437 Upon successful return, the number of variants in the string can be computed by
2438 having saved the value of C<*lenp> before the call, and subtracting the
2439 after-call value of C<*lenp> from it.
2441 If you need a copy of the string, see L</bytes_from_utf8>.
2447 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2451 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2452 PERL_UNUSED_CONTEXT;
2454 /* This is a no-op if no variants at all in the input */
2455 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2460 U8 * const save = s;
2461 U8 * const send = s + *lenp;
2464 /* Nothing before the first variant needs to be changed, so start the real
2468 if (! UTF8_IS_INVARIANT(*s)) {
2469 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2470 *lenp = ((STRLEN) -1);
2478 /* Is downgradable, so do it */
2479 d = s = first_variant;
2482 if (! UVCHR_IS_INVARIANT(c)) {
2483 /* Then it is two-byte encoded */
2484 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2497 =for apidoc bytes_from_utf8
2499 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2500 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2501 actually encoded in UTF-8.
2503 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2506 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2507 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2508 C<*lenp> are unchanged, and the return value is the original C<s>.
2510 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2511 newly created string containing a downgraded copy of C<s>, and whose length is
2512 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2513 caller is responsible for arranging for the memory used by this string to get
2516 Upon successful return, the number of variants in the string can be computed by
2517 having saved the value of C<*lenp> before the call, and subtracting the
2518 after-call value of C<*lenp> from it.
2522 There is a macro that avoids this function call, but this is retained for
2523 anyone who calls it with the Perl_ prefix */
2526 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2528 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2529 PERL_UNUSED_CONTEXT;
2531 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2535 No = here because currently externally undocumented
2536 for apidoc bytes_from_utf8_loc
2538 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2539 to store the location of the first character in C<"s"> that cannot be
2540 converted to non-UTF8.
2542 If that parameter is C<NULL>, this function behaves identically to
2545 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2546 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2548 Otherwise, the function returns a newly created C<NUL>-terminated string
2549 containing the non-UTF8 equivalent of the convertible first portion of
2550 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2551 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2552 and C<*first_non_downgradable> is set to C<NULL>.
2554 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2555 first character in the original string that wasn't converted. C<*is_utf8p> is
2556 unchanged. Note that the new string may have length 0.
2558 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2559 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2560 converts as many characters in it as possible stopping at the first one it
2561 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2562 set to point to that. The function returns the portion that could be converted
2563 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2564 not including the terminating C<NUL>. If the very first character in the
2565 original could not be converted, C<*lenp> will be 0, and the new string will
2566 contain just a single C<NUL>. If the entire input string was converted,
2567 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2569 Upon successful return, the number of variants in the converted portion of the
2570 string can be computed by having saved the value of C<*lenp> before the call,
2571 and subtracting the after-call value of C<*lenp> from it.
2579 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2582 const U8 *original = s;
2583 U8 *converted_start;
2584 const U8 *send = s + *lenp;
2586 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2589 if (first_unconverted) {
2590 *first_unconverted = NULL;
2593 return (U8 *) original;
2596 Newx(d, (*lenp) + 1, U8);
2598 converted_start = d;
2601 if (! UTF8_IS_INVARIANT(c)) {
2603 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2604 * have to stop now */
2605 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2606 if (first_unconverted) {
2607 *first_unconverted = s - 1;
2608 goto finish_and_return;
2611 Safefree(converted_start);
2612 return (U8 *) original;
2616 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2622 /* Here, converted the whole of the input */
2624 if (first_unconverted) {
2625 *first_unconverted = NULL;
2630 *lenp = d - converted_start;
2632 /* Trim unused space */
2633 Renew(converted_start, *lenp + 1, U8);
2635 return converted_start;
2639 =for apidoc bytes_to_utf8
2641 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2643 Returns a pointer to the newly-created string, and sets C<*lenp> to
2644 reflect the new length in bytes. The caller is responsible for arranging for
2645 the memory used by this string to get freed.
2647 Upon successful return, the number of variants in the string can be computed by
2648 having saved the value of C<*lenp> before the call, and subtracting it from the
2649 after-call value of C<*lenp>.
2651 A C<NUL> character will be written after the end of the string.
2653 If you want to convert to UTF-8 from encodings other than
2654 the native (Latin1 or EBCDIC),
2655 see L</sv_recode_to_utf8>().
2661 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2663 const U8 * const send = s + (*lenp);
2667 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2668 PERL_UNUSED_CONTEXT;
2670 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2671 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2675 append_utf8_from_native_byte(*s, &d);
2686 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2687 * use utf16_to_utf8_reversed().
2689 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2690 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2691 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2693 * These functions don't check for overflow. The worst case is every code
2694 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2695 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2696 * destination must be pre-extended to 2 times the source length.
2698 * Do not use in-place. We optimize for native, for obvious reasons. */
2701 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2706 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2709 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2715 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2717 if (OFFUNI_IS_INVARIANT(uv)) {
2718 *d++ = LATIN1_TO_NATIVE((U8) uv);
2721 if (uv <= MAX_UTF8_TWO_BYTE) {
2722 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2723 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2727 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2728 #define LAST_HIGH_SURROGATE 0xDBFF
2729 #define FIRST_LOW_SURROGATE 0xDC00
2730 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2731 #define FIRST_IN_PLANE1 0x10000
2733 /* This assumes that most uses will be in the first Unicode plane, not
2734 * needing surrogates */
2735 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2736 && uv <= UNICODE_SURROGATE_LAST))
2738 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2739 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2742 UV low = (p[0] << 8) + p[1];
2743 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2744 || UNLIKELY(low > LAST_LOW_SURROGATE))
2746 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2749 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2750 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2754 d = uvoffuni_to_utf8_flags(d, uv, 0);
2756 if (uv < FIRST_IN_PLANE1) {
2757 *d++ = (U8)(( uv >> 12) | 0xe0);
2758 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2759 *d++ = (U8)(( uv & 0x3f) | 0x80);
2763 *d++ = (U8)(( uv >> 18) | 0xf0);
2764 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2765 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2766 *d++ = (U8)(( uv & 0x3f) | 0x80);
2771 *newlen = d - dstart;
2775 /* Note: this one is slightly destructive of the source. */
2778 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2781 U8* const send = s + bytelen;
2783 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2786 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2790 const U8 tmp = s[0];
2795 return utf16_to_utf8(p, d, bytelen, newlen);
2799 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2802 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2806 Perl__is_uni_perl_idcont(pTHX_ UV c)
2809 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2813 Perl__is_uni_perl_idstart(pTHX_ UV c)
2816 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2820 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2823 /* We have the latin1-range values compiled into the core, so just use
2824 * those, converting the result to UTF-8. The only difference between upper
2825 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2826 * either "SS" or "Ss". Which one to use is passed into the routine in
2827 * 'S_or_s' to avoid a test */
2829 UV converted = toUPPER_LATIN1_MOD(c);
2831 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2833 assert(S_or_s == 'S' || S_or_s == 's');
2835 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2836 characters in this range */
2837 *p = (U8) converted;
2842 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2843 * which it maps to one of them, so as to only have to have one check for
2844 * it in the main case */
2845 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2847 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2848 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2851 converted = GREEK_CAPITAL_LETTER_MU;
2853 #if UNICODE_MAJOR_VERSION > 2 \
2854 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2855 && UNICODE_DOT_DOT_VERSION >= 8)
2856 case LATIN_SMALL_LETTER_SHARP_S:
2863 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2864 " '%c' to map to '%c'",
2865 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2866 NOT_REACHED; /* NOTREACHED */
2870 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2871 *p = UTF8_TWO_BYTE_LO(converted);
2877 /* If compiled on an early Unicode version, there may not be auxiliary tables
2879 #ifndef HAS_UC_AUX_TABLES
2880 # define UC_AUX_TABLE_ptrs NULL
2881 # define UC_AUX_TABLE_lengths NULL
2883 #ifndef HAS_TC_AUX_TABLES
2884 # define TC_AUX_TABLE_ptrs NULL
2885 # define TC_AUX_TABLE_lengths NULL
2887 #ifndef HAS_LC_AUX_TABLES
2888 # define LC_AUX_TABLE_ptrs NULL
2889 # define LC_AUX_TABLE_lengths NULL
2891 #ifndef HAS_CF_AUX_TABLES
2892 # define CF_AUX_TABLE_ptrs NULL
2893 # define CF_AUX_TABLE_lengths NULL
2895 #ifndef HAS_UC_AUX_TABLES
2896 # define UC_AUX_TABLE_ptrs NULL
2897 # define UC_AUX_TABLE_lengths NULL
2900 /* Call the function to convert a UTF-8 encoded character to the specified case.
2901 * Note that there may be more than one character in the result.
2902 * 's' is a pointer to the first byte of the input character
2903 * 'd' will be set to the first byte of the string of changed characters. It
2904 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2905 * 'lenp' will be set to the length in bytes of the string of changed characters
2907 * The functions return the ordinal of the first character in the string of
2909 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2910 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2911 Uppercase_Mapping_invmap, \
2912 UC_AUX_TABLE_ptrs, \
2913 UC_AUX_TABLE_lengths, \
2915 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2916 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2917 Titlecase_Mapping_invmap, \
2918 TC_AUX_TABLE_ptrs, \
2919 TC_AUX_TABLE_lengths, \
2921 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2922 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2923 Lowercase_Mapping_invmap, \
2924 LC_AUX_TABLE_ptrs, \
2925 LC_AUX_TABLE_lengths, \
2929 /* This additionally has the input parameter 'specials', which if non-zero will
2930 * cause this to use the specials hash for folding (meaning get full case
2931 * folding); otherwise, when zero, this implies a simple case fold */
2932 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2934 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2935 Case_Folding_invmap, \
2936 CF_AUX_TABLE_ptrs, \
2937 CF_AUX_TABLE_lengths, \
2939 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2940 Simple_Case_Folding_invmap, \
2945 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2947 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2948 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2949 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2950 * the changed version may be longer than the original character.
2952 * The ordinal of the first character of the changed version is returned
2953 * (but note, as explained above, that there may be more.) */
2956 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2959 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2962 return CALL_UPPER_CASE(c, NULL, p, lenp);
2966 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2969 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2972 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2975 return CALL_TITLE_CASE(c, NULL, p, lenp);
2979 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2981 /* We have the latin1-range values compiled into the core, so just use
2982 * those, converting the result to UTF-8. Since the result is always just
2983 * one character, we allow <p> to be NULL */
2985 U8 converted = toLOWER_LATIN1(c);
2987 PERL_UNUSED_ARG(dummy);
2990 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2995 /* Result is known to always be < 256, so can use the EIGHT_BIT
2997 *p = UTF8_EIGHT_BIT_HI(converted);
2998 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3006 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3009 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3012 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3015 return CALL_LOWER_CASE(c, NULL, p, lenp);
3019 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3021 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3022 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3023 * FOLD_FLAGS_FULL iff full folding is to be used;
3025 * Not to be used for locale folds
3030 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3032 assert (! (flags & FOLD_FLAGS_LOCALE));
3034 if (UNLIKELY(c == MICRO_SIGN)) {
3035 converted = GREEK_SMALL_LETTER_MU;
3037 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3038 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3039 || UNICODE_DOT_DOT_VERSION > 0)
3040 else if ( (flags & FOLD_FLAGS_FULL)
3041 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3043 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3044 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3045 * under those circumstances. */
3046 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3047 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3048 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3050 return LATIN_SMALL_LETTER_LONG_S;
3060 else { /* In this range the fold of all other characters is their lower
3062 converted = toLOWER_LATIN1(c);
3065 if (UVCHR_IS_INVARIANT(converted)) {
3066 *p = (U8) converted;
3070 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3071 *p = UTF8_TWO_BYTE_LO(converted);
3079 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3082 /* Not currently externally documented, and subject to change
3083 * <flags> bits meanings:
3084 * FOLD_FLAGS_FULL iff full folding is to be used;
3085 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3086 * locale are to be used.
3087 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3091 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3093 if (flags & FOLD_FLAGS_LOCALE) {
3094 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3095 * except for potentially warning */
3096 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3097 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3098 flags &= ~FOLD_FLAGS_LOCALE;
3101 goto needs_full_generality;
3106 return _to_fold_latin1((U8) c, p, lenp,
3107 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3110 /* Here, above 255. If no special needs, just use the macro */
3111 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3112 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3114 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3115 the special flags. */
3116 U8 utf8_c[UTF8_MAXBYTES + 1];
3118 needs_full_generality:
3119 uvchr_to_utf8(utf8_c, c);
3120 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3125 PERL_STATIC_INLINE bool
3126 S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3129 /* returns a boolean giving whether or not the UTF8-encoded character that
3130 * starts at <p>, and extending no further than <e - 1> is in the inversion
3131 * list <invlist>. */
3133 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3135 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3137 if (cp == 0 && (p >= e || *p != '\0')) {
3138 _force_out_malformed_utf8_message(p, e, 0, 1);
3139 NOT_REACHED; /* NOTREACHED */
3143 return _invlist_contains_cp(invlist, cp);
3146 #if 0 /* Not currently used, but may be needed in the future */
3147 PERLVAR(I, seen_deprecated_macro, HV *)
3150 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3151 const char * const alternative,
3152 const bool use_locale,
3153 const char * const file,
3154 const unsigned line)
3158 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3160 if (ckWARN_d(WARN_DEPRECATED)) {
3162 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3163 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3164 if (! PL_seen_deprecated_macro) {
3165 PL_seen_deprecated_macro = newHV();
3167 if (! hv_store(PL_seen_deprecated_macro, key,
3168 strlen(key), &PL_sv_undef, 0))
3170 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3173 if (instr(file, "mathoms.c")) {
3174 Perl_warner(aTHX_ WARN_DEPRECATED,
3175 "In %s, line %d, starting in Perl v5.32, %s()"
3176 " will be removed. Avoid this message by"
3177 " converting to use %s().\n",
3178 file, line, name, alternative);
3181 Perl_warner(aTHX_ WARN_DEPRECATED,
3182 "In %s, line %d, starting in Perl v5.32, %s() will"
3183 " require an additional parameter. Avoid this"
3184 " message by converting to use %s().\n",
3185 file, line, name, alternative);
3193 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
3196 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3198 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
3202 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
3205 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
3207 return is_utf8_common(p, e, PL_utf8_perl_idstart);
3211 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
3214 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
3216 return is_utf8_common(p, e, PL_utf8_perl_idcont);
3220 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3221 U8* ustrp, STRLEN *lenp,
3222 SV *invlist, const int * const invmap,
3223 const unsigned int * const * const aux_tables,
3224 const U8 * const aux_table_lengths,
3225 const char * const normal)
3229 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3230 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3231 * to name the new case in any generated messages, as a fallback if the
3232 * operation being used is not available. The new case is given by the
3233 * data structures in the remaining arguments.
3235 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3236 * entire changed case string, and the return value is the first code point
3239 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3241 /* For code points that don't change case, we already know that the output
3242 * of this function is the unchanged input, so we can skip doing look-ups
3243 * for them. Unfortunately the case-changing code points are scattered
3244 * around. But there are some long consecutive ranges where there are no
3245 * case changing code points. By adding tests, we can eliminate the lookup
3246 * for all the ones in such ranges. This is currently done here only for
3247 * just a few cases where the scripts are in common use in modern commerce
3248 * (and scripts adjacent to those which can be included without additional
3251 if (uv1 >= 0x0590) {
3252 /* This keeps from needing further processing the code points most
3253 * likely to be used in the following non-cased scripts: Hebrew,
3254 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3255 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3256 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3261 /* The following largish code point ranges also don't have case
3262 * changes, but khw didn't think they warranted extra tests to speed
3263 * them up (which would slightly slow down everything else above them):
3264 * 1100..139F Hangul Jamo, Ethiopic
3265 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3266 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3267 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3268 * Combining Diacritical Marks Extended, Balinese,
3269 * Sundanese, Batak, Lepcha, Ol Chiki
3270 * 2000..206F General Punctuation
3273 if (uv1 >= 0x2D30) {
3275 /* This keeps the from needing further processing the code points
3276 * most likely to be used in the following non-cased major scripts:
3277 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3279 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3280 * event that Unicode eventually allocates the unused block as of
3281 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3282 * that the test suite will start having failures to alert you
3283 * should that happen) */
3288 if (uv1 >= 0xAC00) {
3289 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3290 if (ckWARN_d(WARN_SURROGATE)) {
3291 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3292 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3293 "Operation \"%s\" returns its argument for"
3294 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3299 /* AC00..FAFF Catches Hangul syllables and private use, plus
3305 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3306 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3307 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3310 if (ckWARN_d(WARN_NON_UNICODE)) {
3311 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3312 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3313 "Operation \"%s\" returns its argument for"
3314 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3318 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3320 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3329 /* Note that non-characters are perfectly legal, so no warning should
3335 const unsigned int * cp_list;
3338 /* 'index' is guaranteed to be non-negative, as this is an inversion
3339 * map that covers all possible inputs. See [perl #133365] */
3340 SSize_t index = _invlist_search(invlist, uv1);
3341 IV base = invmap[index];
3343 /* The data structures are set up so that if 'base' is non-negative,
3344 * the case change is 1-to-1; and if 0, the change is to itself */
3352 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3353 lc = base + uv1 - invlist_array(invlist)[index];
3354 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3358 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3359 * requires an auxiliary table look up. abs(base) gives the index into
3360 * a list of such tables which points to the proper aux table. And a
3361 * parallel list gives the length of each corresponding aux table. */
3362 cp_list = aux_tables[-base];
3364 /* Create the string of UTF-8 from the mapped-to code points */
3366 for (i = 0; i < aux_table_lengths[-base]; i++) {
3367 d = uvchr_to_utf8(d, cp_list[i]);
3375 /* Here, there was no mapping defined, which means that the code point maps
3376 * to itself. Return the inputs */
3380 if (p != ustrp) { /* Don't copy onto itself */
3381 Copy(p, ustrp, len, U8);
3386 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3394 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3395 const unsigned int ** remaining_folds_to)
3397 /* Returns the count of the number of code points that fold to the input
3398 * 'cp' (besides itself).
3400 * If the return is 0, there is nothing else that folds to it, and
3401 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3403 * If the return is 1, '*first_folds_to' is set to the single code point,
3404 * and '*remaining_folds_to' is set to NULL.
3406 * Otherwise, '*first_folds_to' is set to a code point, and
3407 * '*remaining_fold_to' is set to an array that contains the others. The
3408 * length of this array is the returned count minus 1.
3410 * The reason for this convolution is to avoid having to deal with
3411 * allocating and freeing memory. The lists are already constructed, so
3412 * the return can point to them, but single code points aren't, so would
3413 * need to be constructed if we didn't employ something like this API */
3416 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3417 * that covers all possible inputs. See [perl #133365] */
3418 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3419 int base = _Perl_IVCF_invmap[index];
3421 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3423 if (base == 0) { /* No fold */
3424 *first_folds_to = 0;
3425 *remaining_folds_to = NULL;
3429 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3435 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3437 /* The data structure is set up so that the absolute value of 'base' is
3438 * an index into a table of pointers to arrays, with the array
3439 * corresponding to the index being the list of code points that fold
3440 * to 'cp', and the parallel array containing the length of the list
3442 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3443 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3446 return IVCF_AUX_TABLE_lengths[-base];
3451 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3452 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3453 *remaining_folds_to = NULL;
3458 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3459 U8* const ustrp, STRLEN *lenp)
3461 /* This is called when changing the case of a UTF-8-encoded character above
3462 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3463 * result contains a character that crosses the 255/256 boundary, disallow
3464 * the change, and return the original code point. See L<perlfunc/lc> for
3467 * p points to the original string whose case was changed; assumed
3468 * by this routine to be well-formed
3469 * result the code point of the first character in the changed-case string
3470 * ustrp points to the changed-case string (<result> represents its
3472 * lenp points to the length of <ustrp> */
3474 UV original; /* To store the first code point of <p> */
3476 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3478 assert(UTF8_IS_ABOVE_LATIN1(*p));
3480 /* We know immediately if the first character in the string crosses the
3481 * boundary, so can skip testing */
3484 /* Look at every character in the result; if any cross the
3485 * boundary, the whole thing is disallowed */
3486 U8* s = ustrp + UTF8SKIP(ustrp);
3487 U8* e = ustrp + *lenp;
3489 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3495 /* Here, no characters crossed, result is ok as-is, but we warn. */
3496 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3502 /* Failed, have to return the original */
3503 original = valid_utf8_to_uvchr(p, lenp);
3505 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3506 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3507 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3508 " locale; resolved to \"\\x{%" UVXf "}\".",
3512 Copy(p, ustrp, *lenp, char);
3517 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3518 U8 * ustrp, STRLEN *lenp)
3520 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3521 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3522 * Otherwise, it returns the first code point of the Turkic foldcased
3523 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3524 * contain *lenp bytes
3526 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3527 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3530 PERL_ARGS_ASSERT_TURKIC_FC;
3533 if (UNLIKELY(*p == 'I')) {
3535 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3536 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3537 return LATIN_SMALL_LETTER_DOTLESS_I;
3540 if (UNLIKELY(memBEGINs(p, e - p,
3541 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3552 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3553 U8 * ustrp, STRLEN *lenp)
3555 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3556 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3557 * Otherwise, it returns the first code point of the Turkic lowercased
3558 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3559 * contain *lenp bytes */
3562 PERL_ARGS_ASSERT_TURKIC_LC;
3565 /* A 'I' requires context as to what to do */
3566 if (UNLIKELY(*p0 == 'I')) {
3567 const U8 * p = p0 + 1;
3569 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3570 * modified by a dot above lowercases to 'i' even in turkic locales. */
3574 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3580 /* For the dot above to modify the 'I', it must be part of a
3581 * combining sequence immediately following the 'I', and no other
3582 * modifier with a ccc of 230 may intervene */
3583 cp = utf8_to_uvchr_buf(p, e, NULL);
3584 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3588 /* Here the combining sequence continues */
3593 /* In all other cases the lc is the same as the fold */
3594 return turkic_fc(p0, e, ustrp, lenp);
3598 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3599 U8 * ustrp, STRLEN *lenp)
3601 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3602 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3603 * Otherwise, it returns the first code point of the Turkic upper or
3604 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3605 * ustrp will contain *lenp bytes
3607 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3608 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3611 PERL_ARGS_ASSERT_TURKIC_UC;
3616 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3617 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3618 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3621 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3630 /* The process for changing the case is essentially the same for the four case
3631 * change types, except there are complications for folding. Otherwise the
3632 * difference is only which case to change to. To make sure that they all do
3633 * the same thing, the bodies of the functions are extracted out into the
3634 * following two macros. The functions are written with the same variable
3635 * names, and these are known and used inside these macros. It would be
3636 * better, of course, to have inline functions to do it, but since different
3637 * macros are called, depending on which case is being changed to, this is not
3638 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3639 * function can start with the common start macro, then finish with its special
3640 * handling; while the other three cases can just use the common end macro.
3642 * The algorithm is to use the proper (passed in) macro or function to change
3643 * the case for code points that are below 256. The macro is used if using
3644 * locale rules for the case change; the function if not. If the code point is
3645 * above 255, it is computed from the input UTF-8, and another macro is called
3646 * to do the conversion. If necessary, the output is converted to UTF-8. If
3647 * using a locale, we have to check that the change did not cross the 255/256
3648 * boundary, see check_locale_boundary_crossing() for further details.
3650 * The macros are split with the correct case change for the below-256 case
3651 * stored into 'result', and in the middle of an else clause for the above-255
3652 * case. At that point in the 'else', 'result' is not the final result, but is
3653 * the input code point calculated from the UTF-8. The fold code needs to
3654 * realize all this and take it from there.
3656 * To deal with Turkic locales, the function specified by the parameter
3657 * 'turkic' is called when appropriate.
3659 * If you read the two macros as sequential, it's easier to understand what's
3661 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3662 L1_func_extra_param, turkic) \
3664 if (flags & (locale_flags)) { \
3665 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3666 if (IN_UTF8_CTYPE_LOCALE) { \
3667 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3668 UV ret = turkic(p, e, ustrp, lenp); \
3669 if (ret) return ret; \
3672 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3674 flags &= ~(locale_flags); \
3678 if (UTF8_IS_INVARIANT(*p)) { \
3679 if (flags & (locale_flags)) { \
3680 result = LC_L1_change_macro(*p); \
3683 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3686 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3687 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3688 if (flags & (locale_flags)) { \
3689 result = LC_L1_change_macro(c); \
3692 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3695 else { /* malformed UTF-8 or ord above 255 */ \
3696 STRLEN len_result; \
3697 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3698 if (len_result == (STRLEN) -1) { \
3699 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
3702 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3703 result = change_macro(result, p, ustrp, lenp); \
3705 if (flags & (locale_flags)) { \
3706 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3711 /* Here, used locale rules. Convert back to UTF-8 */ \
3712 if (UTF8_IS_INVARIANT(result)) { \
3713 *ustrp = (U8) result; \
3717 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3718 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3724 /* Not currently externally documented, and subject to change:
3725 * <flags> is set iff iff the rules from the current underlying locale are to
3729 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3738 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3740 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3741 /* 2nd char of uc(U+DF) is 'S' */
3742 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3744 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3747 /* Not currently externally documented, and subject to change:
3748 * <flags> is set iff the rules from the current underlying locale are to be
3749 * used. Since titlecase is not defined in POSIX, for other than a
3750 * UTF-8 locale, uppercase is used instead for code points < 256.
3754 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3763 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3765 /* 2nd char of ucfirst(U+DF) is 's' */
3766 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
3768 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3771 /* Not currently externally documented, and subject to change:
3772 * <flags> is set iff iff the rules from the current underlying locale are to
3777 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3786 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3788 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
3790 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3793 /* Not currently externally documented, and subject to change,
3795 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3796 * locale are to be used.
3797 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3798 * otherwise simple folds
3799 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3804 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3813 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3815 /* These are mutually exclusive */
3816 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3818 assert(p != ustrp); /* Otherwise overwrites */
3820 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3821 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
3824 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3826 if (flags & FOLD_FLAGS_LOCALE) {
3828 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3829 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3830 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3832 /* Special case these two characters, as what normally gets
3833 * returned under locale doesn't work */
3834 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
3836 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3837 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3838 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3839 "resolved to \"\\x{17F}\\x{17F}\".");
3844 if (memBEGINs((char *) p, e - p, LONG_S_T))
3846 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3847 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3848 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3849 "resolved to \"\\x{FB06}\".");
3850 goto return_ligature_st;
3853 #if UNICODE_MAJOR_VERSION == 3 \
3854 && UNICODE_DOT_VERSION == 0 \
3855 && UNICODE_DOT_DOT_VERSION == 1
3856 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3858 /* And special case this on this Unicode version only, for the same
3859 * reaons the other two are special cased. They would cross the
3860 * 255/256 boundary which is forbidden under /l, and so the code
3861 * wouldn't catch that they are equivalent (which they are only in
3863 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
3864 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3865 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3866 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3867 "resolved to \"\\x{0131}\".");
3868 goto return_dotless_i;
3872 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3874 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3878 /* This is called when changing the case of a UTF-8-encoded
3879 * character above the ASCII range, and the result should not
3880 * contain an ASCII character. */
3882 UV original; /* To store the first code point of <p> */
3884 /* Look at every character in the result; if any cross the
3885 * boundary, the whole thing is disallowed */
3887 U8* send = ustrp + *lenp;
3890 /* Crossed, have to return the original */
3891 original = valid_utf8_to_uvchr(p, lenp);
3893 /* But in these instances, there is an alternative we can
3894 * return that is valid */
3895 if (original == LATIN_SMALL_LETTER_SHARP_S
3896 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3897 || original == LATIN_CAPITAL_LETTER_SHARP_S
3902 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3903 goto return_ligature_st;
3905 #if UNICODE_MAJOR_VERSION == 3 \
3906 && UNICODE_DOT_VERSION == 0 \
3907 && UNICODE_DOT_DOT_VERSION == 1
3909 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3910 goto return_dotless_i;
3913 Copy(p, ustrp, *lenp, char);
3919 /* Here, no characters crossed, result is ok as-is */
3924 /* Here, used locale rules. Convert back to UTF-8 */
3925 if (UTF8_IS_INVARIANT(result)) {
3926 *ustrp = (U8) result;
3930 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3931 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3938 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3939 * folds to a string of two of these characters. By returning this
3940 * instead, then, e.g.,
3941 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3944 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3945 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3947 return LATIN_SMALL_LETTER_LONG_S;
3950 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3951 * have the other one fold to it */
3953 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3954 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3955 return LATIN_SMALL_LIGATURE_ST;
3957 #if UNICODE_MAJOR_VERSION == 3 \
3958 && UNICODE_DOT_VERSION == 0 \
3959 && UNICODE_DOT_DOT_VERSION == 1
3962 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3963 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3964 return LATIN_SMALL_LETTER_DOTLESS_I;
3971 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3973 /* May change: warns if surrogates, non-character code points, or
3974 * non-Unicode code points are in 's' which has length 'len' bytes.
3975 * Returns TRUE if none found; FALSE otherwise. The only other validity
3976 * check is to make sure that this won't exceed the string's length nor
3979 const U8* const e = s + len;
3982 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
3985 if (UTF8SKIP(s) > len) {
3986 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
3987 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
3990 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
3991 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
3992 if ( ckWARN_d(WARN_NON_UNICODE)
3993 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
3994 0 /* Don't consider overlongs */
3997 /* A side effect of this function will be to warn */
3998 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4002 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4003 if (ckWARN_d(WARN_SURROGATE)) {
4004 /* This has a different warning than the one the called
4005 * function would output, so can't just call it, unlike we
4006 * do for the non-chars and above-unicodes */
4007 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4008 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4009 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4014 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4015 && (ckWARN_d(WARN_NONCHAR)))
4017 /* A side effect of this function will be to warn */
4018 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4029 =for apidoc pv_uni_display
4031 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4032 length C<len>, the displayable version being at most C<pvlim> bytes long
4033 (if longer, the rest is truncated and C<"..."> will be appended).
4035 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4036 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4037 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4038 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4039 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4040 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4042 The pointer to the PV of the C<dsv> is returned.
4044 See also L</sv_uni_display>.
4048 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
4054 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4058 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4060 /* This serves double duty as a flag and a character to print after
4061 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4065 if (pvlim && SvCUR(dsv) >= pvlim) {
4069 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4071 const unsigned char c = (unsigned char)u & 0xFF;
4072 if (flags & UNI_DISPLAY_BACKSLASH) {
4089 const char string = ok;
4090 sv_catpvs(dsv, "\\");
4091 sv_catpvn(dsv, &string, 1);
4094 /* isPRINT() is the locale-blind version. */
4095 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4096 const char string = c;
4097 sv_catpvn(dsv, &string, 1);
4102 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
4105 sv_catpvs(dsv, "...");
4111 =for apidoc sv_uni_display
4113 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4114 the displayable version being at most C<pvlim> bytes long
4115 (if longer, the rest is truncated and "..." will be appended).
4117 The C<flags> argument is as in L</pv_uni_display>().
4119 The pointer to the PV of the C<dsv> is returned.
4124 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4126 const char * const ptr =
4127 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4129 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4131 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4132 SvCUR(ssv), pvlim, flags);
4136 =for apidoc foldEQ_utf8
4138 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
4139 both of which may be in UTF-8) are the same case-insensitively; false
4140 otherwise. How far into the strings to compare is determined by other input
4143 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4144 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
4145 C<u2> with respect to C<s2>.
4147 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
4148 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
4149 The scan will not be considered to be a match unless the goal is reached, and
4150 scanning won't continue past that goal. Correspondingly for C<l2> with respect
4153 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
4154 pointer is considered an end pointer to the position 1 byte past the maximum
4155 point in C<s1> beyond which scanning will not continue under any circumstances.
4156 (This routine assumes that UTF-8 encoded input strings are not malformed;
4157 malformed input can cause it to read past C<pe1>). This means that if both
4158 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
4159 will never be successful because it can never
4160 get as far as its goal (and in fact is asserted against). Correspondingly for
4161 C<pe2> with respect to C<s2>.
4163 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4164 C<l2> must be non-zero), and if both do, both have to be
4165 reached for a successful match. Also, if the fold of a character is multiple
4166 characters, all of them must be matched (see tr21 reference below for
4169 Upon a successful match, if C<pe1> is non-C<NULL>,
4170 it will be set to point to the beginning of the I<next> character of C<s1>
4171 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4173 For case-insensitiveness, the "casefolding" of Unicode is used
4174 instead of upper/lowercasing both the characters, see
4175 L<https://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4179 /* A flags parameter has been added which may change, and hence isn't
4180 * externally documented. Currently it is:
4181 * 0 for as-documented above
4182 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4183 ASCII one, to not match
4184 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4185 * locale are to be used.
4186 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4187 * routine. This allows that step to be skipped.
4188 * Currently, this requires s1 to be encoded as UTF-8
4189 * (u1 must be true), which is asserted for.
4190 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4191 * cross certain boundaries. Hence, the caller should
4192 * let this function do the folding instead of
4193 * pre-folding. This code contains an assertion to
4194 * that effect. However, if the caller knows what
4195 * it's doing, it can pass this flag to indicate that,
4196 * and the assertion is skipped.
4197 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
4198 * to s2, and s2 doesn't have to be UTF-8 encoded.
4199 * This introduces an asymmetry to save a few branches
4200 * in a loop. Currently, this is not a problem, as
4201 * never are both inputs pre-folded. Simply call this
4202 * function with the pre-folded one as the second
4204 * FOLDEQ_S2_FOLDS_SANE
4207 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
4208 const char *s2, char **pe2, UV l2, bool u2,
4211 const U8 *p1 = (const U8*)s1; /* Point to current char */
4212 const U8 *p2 = (const U8*)s2;
4213 const U8 *g1 = NULL; /* goal for s1 */
4214 const U8 *g2 = NULL;
4215 const U8 *e1 = NULL; /* Don't scan s1 past this */
4216 U8 *f1 = NULL; /* Point to current folded */
4217 const U8 *e2 = NULL;
4219 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4220 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4221 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4222 U8 flags_for_folder = FOLD_FLAGS_FULL;
4224 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4226 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4227 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
4228 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4229 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
4230 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4231 /* The algorithm is to trial the folds without regard to the flags on
4232 * the first line of the above assert(), and then see if the result
4233 * violates them. This means that the inputs can't be pre-folded to a
4234 * violating result, hence the assert. This could be changed, with the
4235 * addition of extra tests here for the already-folded case, which would
4236 * slow it down. That cost is more than any possible gain for when these
4237 * flags are specified, as the flags indicate /il or /iaa matching which
4238 * is less common than /iu, and I (khw) also believe that real-world /il
4239 * and /iaa matches are most likely to involve code points 0-255, and this
4240 * function only under rare conditions gets called for 0-255. */
4242 if (flags & FOLDEQ_LOCALE) {
4243 if (IN_UTF8_CTYPE_LOCALE) {
4244 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
4245 flags_for_folder |= FOLD_FLAGS_LOCALE;
4248 flags &= ~FOLDEQ_LOCALE;
4252 flags_for_folder |= FOLD_FLAGS_LOCALE;
4255 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
4256 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
4264 g1 = (const U8*)s1 + l1;
4272 g2 = (const U8*)s2 + l2;
4275 /* Must have at least one goal */
4280 /* Will never match if goal is out-of-bounds */
4281 assert(! e1 || e1 >= g1);
4283 /* Here, there isn't an end pointer, or it is beyond the goal. We
4284 * only go as far as the goal */
4288 assert(e1); /* Must have an end for looking at s1 */
4291 /* Same for goal for s2 */
4293 assert(! e2 || e2 >= g2);
4300 /* If both operands are already folded, we could just do a memEQ on the
4301 * whole strings at once, but it would be better if the caller realized
4302 * this and didn't even call us */
4304 /* Look through both strings, a character at a time */
4305 while (p1 < e1 && p2 < e2) {
4307 /* If at the beginning of a new character in s1, get its fold to use
4308 * and the length of the fold. */
4310 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4316 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4318 /* We have to forbid mixing ASCII with non-ASCII if the
4319 * flags so indicate. And, we can short circuit having to
4320 * call the general functions for this common ASCII case,
4321 * all of whose non-locale folds are also ASCII, and hence
4322 * UTF-8 invariants, so the UTF8ness of the strings is not
4324 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4328 *foldbuf1 = toFOLD(*p1);
4331 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
4333 else { /* Not UTF-8, get UTF-8 fold */
4334 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4340 if (n2 == 0) { /* Same for s2 */
4341 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4343 /* Point to the already-folded character. But for non-UTF-8
4344 * variants, convert to UTF-8 for the algorithm below */
4345 if (UTF8_IS_INVARIANT(*p2)) {
4354 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
4355 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
4361 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4362 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4366 *foldbuf2 = toFOLD(*p2);
4369 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
4372 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4378 /* Here f1 and f2 point to the beginning of the strings to compare.
4379 * These strings are the folds of the next character from each input
4380 * string, stored in UTF-8. */
4382 /* While there is more to look for in both folds, see if they
4383 * continue to match */
4385 U8 fold_length = UTF8SKIP(f1);
4386 if (fold_length != UTF8SKIP(f2)
4387 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4388 function call for single
4390 || memNE((char*)f1, (char*)f2, fold_length))
4392 return 0; /* mismatch */
4395 /* Here, they matched, advance past them */
4402 /* When reach the end of any fold, advance the input past it */
4404 p1 += u1 ? UTF8SKIP(p1) : 1;
4407 p2 += u2 ? UTF8SKIP(p2) : 1;
4409 } /* End of loop through both strings */
4411 /* A match is defined by each scan that specified an explicit length
4412 * reaching its final goal, and the other not having matched a partial
4413 * character (which can happen when the fold of a character is more than one
4415 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4419 /* Successful match. Set output pointers */
4429 /* XXX The next two functions should likely be moved to mathoms.c once all
4430 * occurrences of them are removed from the core; some cpan-upstream modules
4434 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
4436 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
4438 return uvoffuni_to_utf8_flags(d, uv, 0);
4442 =for apidoc utf8n_to_uvuni
4444 Instead use L<perlapi/utf8_to_uvchr_buf>, or rarely, L<perlapi/utf8n_to_uvchr>.
4446 This function was useful for code that wanted to handle both EBCDIC and
4447 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
4448 distinctions between the platforms have mostly been made invisible to most
4449 code, so this function is quite unlikely to be what you want. If you do need
4450 this precise functionality, use instead
4451 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>>
4452 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|perlapi/utf8n_to_uvchr>>.
4458 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
4460 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
4462 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
4466 =for apidoc uvuni_to_utf8_flags
4468 Instead you almost certainly want to use L<perlapi/uvchr_to_utf8> or
4469 L<perlapi/uvchr_to_utf8_flags>.
4471 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
4472 which itself, while not deprecated, should be used only in isolated
4473 circumstances. These functions were useful for code that wanted to handle
4474 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
4475 v5.20, the distinctions between the platforms have mostly been made invisible
4476 to most code, so this function is quite unlikely to be what you want.
4482 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
4484 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
4486 return uvoffuni_to_utf8_flags(d, uv, flags);
4490 =for apidoc utf8_to_uvchr
4492 Returns the native code point of the first character in the string C<s>
4493 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
4494 length, in bytes, of that character.
4496 Some, but not all, UTF-8 malformations are detected, and in fact, some
4497 malformed input could cause reading beyond the end of the input buffer, which
4498 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
4500 If C<s> points to one of the detected malformations, and UTF8 warnings are
4501 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
4502 C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or
4503 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
4504 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
4505 next possible position in C<s> that could begin a non-malformed character.
4506 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
4512 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
4514 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
4516 /* This function is unsafe if malformed UTF-8 input is given it, which is
4517 * why the function is deprecated. If the first byte of the input
4518 * indicates that there are more bytes remaining in the sequence that forms
4519 * the character than there are in the input buffer, it can read past the
4520 * end. But we can make it safe if the input string happens to be
4521 * NUL-terminated, as many strings in Perl are, by refusing to read past a
4522 * NUL, which is what UTF8_CHK_SKIP() does. A NUL indicates the start of
4523 * the next character anyway. If the input isn't NUL-terminated, the
4524 * function remains unsafe, as it always has been. */
4526 return utf8_to_uvchr_buf(s, s + UTF8_CHK_SKIP(s), retlen);
4530 * ex: set ts=8 sts=4 sw=4 et: