3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
41 These are various utility functions for manipulating UTF8-encoded
42 strings. For the uninitiated, this is a method of representing arbitrary
43 Unicode characters as a variable number of bytes, in such a way that
44 characters in the ASCII range are unmodified, and a zero byte never appears
45 within non-zero characters.
48 /* helper for Perl__force_out_malformed_utf8_message(). Like
49 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
53 S_restore_cop_warnings(pTHX_ void *p)
55 free_and_set_cop_warnings(PL_curcop, (STRLEN*) p);
60 Perl__force_out_malformed_utf8_message(pTHX_
61 const U8 *const p, /* First byte in UTF-8 sequence */
62 const U8 * const e, /* Final byte in sequence (may include
64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
65 usually 0, or some DISALLOW flags */
66 const bool die_here) /* If TRUE, this function does not return */
68 /* This core-only function is to be called when a malformed UTF-8 character
69 * is found, in order to output the detailed information about the
70 * malformation before dieing. The reason it exists is for the occasions
71 * when such a malformation is fatal, but warnings might be turned off, so
72 * that normally they would not be actually output. This ensures that they
73 * do get output. Because a sequence may be malformed in more than one
74 * way, multiple messages may be generated, so we can't make them fatal, as
75 * that would cause the first one to die.
77 * Instead we pretend -W was passed to perl, then die afterwards. The
78 * flexibility is here to return to the caller so they can finish up and
82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
90 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
91 * than PL_compiling */
92 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
93 (void*)PL_curcop->cop_warnings);
94 PL_curcop->cop_warnings = pWARN_ALL;
97 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
102 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
103 " be called only when there are errors found");
107 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
112 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
113 U32 categories, /* Packed warning categories */
114 U32 flag) /* Flag associated with this message */
116 /* Creates, populates, and returns an HV* that describes an error message
117 * for the translators between UTF8 and code point */
119 SV* msg_sv = newSVpv(message, 0);
120 SV* category_sv = newSVuv(categories);
121 SV* flag_bit_sv = newSVuv(flag);
123 HV* msg_hv = newHV();
125 PERL_ARGS_ASSERT_NEW_MSG_HV;
127 (void) hv_stores(msg_hv, "text", msg_sv);
128 (void) hv_stores(msg_hv, "warn_categories", category_sv);
129 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
135 =for apidoc uvoffuni_to_utf8_flags
137 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
138 Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
139 L<perlapi/uvchr_to_utf8_flags>>.
141 This function is like them, but the input is a strict Unicode
142 (as opposed to native) code point. Only in very rare circumstances should code
143 not be using the native code point.
145 For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
151 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
153 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
155 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
158 /* All these formats take a single UV code point argument */
159 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
160 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
161 " is not recommended for open interchange";
162 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
163 " may not be portable";
165 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
167 if (flags & UNICODE_WARN_SURROGATE) { \
168 U32 category = packWARN(WARN_SURROGATE); \
169 const char * format = surrogate_cp_format; \
171 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
173 UNICODE_GOT_SURROGATE); \
176 Perl_ck_warner_d(aTHX_ category, format, uv); \
179 if (flags & UNICODE_DISALLOW_SURROGATE) { \
184 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
186 if (flags & UNICODE_WARN_NONCHAR) { \
187 U32 category = packWARN(WARN_NONCHAR); \
188 const char * format = nonchar_cp_format; \
190 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
192 UNICODE_GOT_NONCHAR); \
195 Perl_ck_warner_d(aTHX_ category, format, uv); \
198 if (flags & UNICODE_DISALLOW_NONCHAR) { \
203 /* Use shorter names internally in this file */
204 #define SHIFT UTF_ACCUMULATION_SHIFT
206 #define MARK UTF_CONTINUATION_MARK
207 #define MASK UTF_CONTINUATION_MASK
210 =for apidoc uvchr_to_utf8_flags_msgs
212 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
214 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
216 This function is for code that wants any warning and/or error messages to be
217 returned to the caller rather than be displayed. All messages that would have
218 been displayed if all lexical warnings are enabled will be returned.
220 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
221 placed after all the others, C<msgs>. If this parameter is 0, this function
222 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
223 be a pointer to an C<HV *> variable, in which this function creates a new HV to
224 contain any appropriate messages. The hash has three key-value pairs, as
231 The text of the message as a C<SVpv>.
233 =item C<warn_categories>
235 The warning category (or categories) packed into a C<SVuv>.
239 A single flag bit associated with this message, in a C<SVuv>.
240 The bit corresponds to some bit in the C<*errors> return value,
241 such as C<UNICODE_GOT_SURROGATE>.
245 It's important to note that specifying this parameter as non-null will cause
246 any warnings this function would otherwise generate to be suppressed, and
247 instead be placed in C<*msgs>. The caller can check the lexical warnings state
248 (or not) when choosing what to do with the returned messages.
250 The caller, of course, is responsible for freeing any returned HV.
255 /* Undocumented; we don't want people using this. Instead they should use
256 * uvchr_to_utf8_flags_msgs() */
258 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
260 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
266 if (OFFUNI_IS_INVARIANT(uv)) {
267 *d++ = LATIN1_TO_NATIVE(uv);
271 if (uv <= MAX_UTF8_TWO_BYTE) {
272 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
273 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
277 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
278 * below, the 16 is for start bytes E0-EF (which are all the possible ones
279 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
280 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
281 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
282 * 0x800-0xFFFF on ASCII */
283 if (uv < (16 * (1U << (2 * SHIFT)))) {
284 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
285 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
286 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
288 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
289 aren't tested here */
290 /* The most likely code points in this range are below the surrogates.
291 * Do an extra test to quickly exclude those. */
292 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
293 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
294 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
296 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
298 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
299 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
306 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
307 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
308 * happen starting with 4-byte characters on ASCII platforms. We unify the
309 * code for these with EBCDIC, even though some of them require 5-bytes on
310 * those, because khw believes the code saving is worth the very slight
311 * performance hit on these high EBCDIC code points. */
313 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
314 if (UNLIKELY( uv > MAX_LEGAL_CP
315 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
317 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv));
319 if ( (flags & UNICODE_WARN_SUPER)
320 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
321 && UNICODE_IS_PERL_EXTENDED(uv)))
323 const char * format = super_cp_format;
324 U32 category = packWARN(WARN_NON_UNICODE);
325 U32 flag = UNICODE_GOT_SUPER;
327 /* Choose the more dire applicable warning */
328 if (UNICODE_IS_PERL_EXTENDED(uv)) {
329 format = PL_extended_cp_format;
330 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
331 if (flags & (UNICODE_WARN_PERL_EXTENDED
332 |UNICODE_DISALLOW_PERL_EXTENDED))
334 flag = UNICODE_GOT_PERL_EXTENDED;
339 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
342 else if ( ckWARN_d(WARN_NON_UNICODE)
343 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
344 && ckWARN(WARN_PORTABLE)))
346 Perl_warner(aTHX_ category, format, uv);
349 if ( (flags & UNICODE_DISALLOW_SUPER)
350 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
351 && UNICODE_IS_PERL_EXTENDED(uv)))
356 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
357 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
360 /* Test for and handle 4-byte result. In the test immediately below, the
361 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
362 * characters). The 3 is for 3 continuation bytes; these each contribute
363 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
364 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
365 * 0x1_0000-0x1F_FFFF on ASCII */
366 if (uv < (8 * (1U << (3 * SHIFT)))) {
367 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
368 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
369 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
370 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
372 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
373 characters. The end-plane non-characters for EBCDIC were
374 handled just above */
375 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
376 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
378 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
379 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
386 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
387 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
388 * format. The unrolled version above turns out to not save all that much
389 * time, and at these high code points (well above the legal Unicode range
390 * on ASCII platforms, and well above anything in common use in EBCDIC),
391 * khw believes that less code outweighs slight performance gains. */
394 STRLEN len = OFFUNISKIP(uv);
397 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
400 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
406 =for apidoc uvchr_to_utf8
408 Adds the UTF-8 representation of the native code point C<uv> to the end
409 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
410 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
411 the byte after the end of the new character. In other words,
413 d = uvchr_to_utf8(d, uv);
415 is the recommended wide native character-aware way of saying
419 This function accepts any code point from 0..C<IV_MAX> as input.
420 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
422 It is possible to forbid or warn on non-Unicode code points, or those that may
423 be problematic by using L</uvchr_to_utf8_flags>.
428 /* This is also a macro */
429 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
432 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
434 return uvchr_to_utf8(d, uv);
438 =for apidoc uvchr_to_utf8_flags
440 Adds the UTF-8 representation of the native code point C<uv> to the end
441 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
442 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
443 the byte after the end of the new character. In other words,
445 d = uvchr_to_utf8_flags(d, uv, flags);
449 d = uvchr_to_utf8_flags(d, uv, 0);
451 This is the Unicode-aware way of saying
455 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
456 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
458 Specifying C<flags> can further restrict what is allowed and not warned on, as
461 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
462 the function will raise a warning, provided UTF8 warnings are enabled. If
463 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
464 NULL. If both flags are set, the function will both warn and return NULL.
466 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
467 affect how the function handles a Unicode non-character.
469 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
470 affect the handling of code points that are above the Unicode maximum of
471 0x10FFFF. Languages other than Perl may not be able to accept files that
474 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
475 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
476 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
477 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
478 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
479 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
480 above-Unicode and surrogate flags, but not the non-character ones, as
482 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
483 See L<perlunicode/Noncharacter code points>.
485 Extremely high code points were never specified in any standard, and require an
486 extension to UTF-8 to express, which Perl does. It is likely that programs
487 written in something other than Perl would not be able to read files that
488 contain these; nor would Perl understand files written by something that uses a
489 different extension. For these reasons, there is a separate set of flags that
490 can warn and/or disallow these extremely high code points, even if other
491 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
492 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
493 C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
494 treat all above-Unicode code points, including these, as malformations. (Note
495 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
496 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
498 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
499 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
500 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
501 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
502 platforms,these flags can apply to code points that actually do fit in 31 bits.
503 The new names accurately describe the situation in all cases.
505 =for apidoc Amnh||UNICODE_DISALLOW_ABOVE_31_BIT
506 =for apidoc Amnh||UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE
507 =for apidoc Amnh||UNICODE_DISALLOW_ILLEGAL_INTERCHANGE
508 =for apidoc Amnh||UNICODE_DISALLOW_NONCHAR
509 =for apidoc Amnh||UNICODE_DISALLOW_PERL_EXTENDED
510 =for apidoc Amnh||UNICODE_DISALLOW_SUPER
511 =for apidoc Amnh||UNICODE_DISALLOW_SURROGATE
512 =for apidoc Amnh||UNICODE_WARN_ABOVE_31_BIT
513 =for apidoc Amnh||UNICODE_WARN_ILLEGAL_C9_INTERCHANGE
514 =for apidoc Amnh||UNICODE_WARN_ILLEGAL_INTERCHANGE
515 =for apidoc Amnh||UNICODE_WARN_NONCHAR
516 =for apidoc Amnh||UNICODE_WARN_PERL_EXTENDED
517 =for apidoc Amnh||UNICODE_WARN_SUPER
518 =for apidoc Amnh||UNICODE_WARN_SURROGATE
523 /* This is also a macro */
524 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
527 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
529 return uvchr_to_utf8_flags(d, uv, flags);
535 S_is_utf8_cp_above_31_bits(const U8 * const s,
537 const bool consider_overlongs)
539 /* Returns TRUE if the first code point represented by the Perl-extended-
540 * UTF-8-encoded string starting at 's', and looking no further than 'e -
541 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
543 * The function handles the case where the input bytes do not include all
544 * the ones necessary to represent a full character. That is, they may be
545 * the intial bytes of the representation of a code point, but possibly
546 * the final ones necessary for the complete representation may be beyond
549 * The function also can handle the case where the input is an overlong
550 * sequence. If 'consider_overlongs' is 0, the function assumes the
551 * input is not overlong, without checking, and will return based on that
552 * assumption. If this parameter is 1, the function will go to the trouble
553 * of figuring out if it actually evaluates to above or below 31 bits.
555 * The sequence is otherwise assumed to be well-formed, without checking.
558 const STRLEN len = e - s;
561 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
563 assert(! UTF8_IS_INVARIANT(*s) && e > s);
567 PERL_UNUSED_ARG(consider_overlongs);
569 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
570 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
571 * also be the start byte for a 31-bit code point; we need at least 2
572 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
573 * the start byte for an overlong sequence, but for 30-bit or smaller code
574 * points, so we don't have to worry about overlongs on EBCDIC.) */
585 /* On ASCII, FE and FF are the only start bytes that can evaluate to
586 * needing more than 31 bits. */
587 if (LIKELY(*s < 0xFE)) {
591 /* What we have left are FE and FF. Both of these require more than 31
592 * bits unless they are for overlongs. */
593 if (! consider_overlongs) {
597 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
598 * above 31 bits. But we need more than one byte to discern this, so if
599 * passed just the start byte, it could be an overlong evaluating to
605 /* Having excluded len==1, and knowing that FE and FF are both valid start
606 * bytes, we can call the function below to see if the sequence is
607 * overlong. (We don't need the full generality of the called function,
608 * but for these huge code points, speed shouldn't be a consideration, and
609 * the compiler does have enough information, since it's static to this
610 * file, to optimize to just the needed parts.) */
611 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
613 /* If it isn't overlong, more than 31 bits are required. */
614 if (is_overlong == 0) {
618 /* If it is indeterminate if it is overlong, return that */
619 if (is_overlong < 0) {
623 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
624 * the max it can be is 2**31 - 1 */
631 /* Here, ASCII and EBCDIC rejoin:
632 * On ASCII: We have an overlong sequence starting with FF
633 * On EBCDIC: We have a sequence starting with FE. */
635 { /* For C89, use a block so the declaration can be close to its use */
639 /* U+7FFFFFFF (2 ** 31 - 1)
640 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
641 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
642 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
643 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
644 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
645 * U+80000000 (2 ** 31):
646 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
647 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
648 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
649 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
651 * and since we know that *s = \xfe, any continuation sequcence
652 * following it that is gt the below is above 31 bits
653 [0] [1] [2] [3] [4] [5] [6] */
654 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
658 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
659 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
660 * FF overlong for U+80000000 (2 ** 31):
661 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
662 * and since we know that *s = \xff, any continuation sequcence
663 * following it that is gt the below is above 30 bits
664 [0] [1] [2] [3] [4] [5] [6] */
665 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
669 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
670 const STRLEN cmp_len = MIN(conts_len, len - 1);
672 /* Now compare the continuation bytes in s with the ones we have
673 * compiled in that are for the largest 30 bit code point. If we have
674 * enough bytes available to determine the answer, or the bytes we do
675 * have differ from them, we can compare the two to get a definitive
676 * answer (Note that in UTF-EBCDIC, the two lowest possible
677 * continuation bytes are \x41 and \x42.) */
678 if (cmp_len >= conts_len || memNE(s + 1,
679 conts_for_highest_30_bit,
682 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
685 /* Here, all the bytes we have are the same as the highest 30-bit code
686 * point, but we are missing so many bytes that we can't make the
694 PERL_STATIC_INLINE int
695 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
697 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
698 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
699 * it isn't, and -1 if there isn't enough information to tell. This last
700 * return value can happen if the sequence is incomplete, missing some
701 * trailing bytes that would form a complete character. If there are
702 * enough bytes to make a definitive decision, this function does so.
703 * Usually 2 bytes sufficient.
705 * Overlongs can occur whenever the number of continuation bytes changes.
706 * That means whenever the number of leading 1 bits in a start byte
707 * increases from the next lower start byte. That happens for start bytes
708 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
709 * start bytes have already been excluded, so don't need to be tested here;
710 * ASCII platforms: C0, C1
711 * EBCDIC platforms C0, C1, C2, C3, C4, E0
714 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
715 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
717 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
718 assert(len > 1 && UTF8_IS_START(*s));
720 /* Each platform has overlongs after the start bytes given above (expressed
721 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
722 * the logic is the same, except the E0 overlong has already been excluded
723 * on EBCDIC platforms. The values below were found by manually
724 * inspecting the UTF-8 patterns. See the tables in utf8.h and
728 # define F0_ABOVE_OVERLONG 0xB0
729 # define F8_ABOVE_OVERLONG 0xA8
730 # define FC_ABOVE_OVERLONG 0xA4
731 # define FE_ABOVE_OVERLONG 0xA2
732 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
736 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
740 # define F0_ABOVE_OVERLONG 0x90
741 # define F8_ABOVE_OVERLONG 0x88
742 # define FC_ABOVE_OVERLONG 0x84
743 # define FE_ABOVE_OVERLONG 0x82
744 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
748 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
749 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
750 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
751 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
756 /* Check for the FF overlong */
757 return isFF_OVERLONG(s, len);
760 PERL_STATIC_INLINE int
761 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
763 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
764 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
765 * it isn't, and -1 if there isn't enough information to tell. This last
766 * return value can happen if the sequence is incomplete, missing some
767 * trailing bytes that would form a complete character. If there are
768 * enough bytes to make a definitive decision, this function does so. */
770 PERL_ARGS_ASSERT_ISFF_OVERLONG;
772 /* To be an FF overlong, all the available bytes must match */
773 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
774 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
779 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
780 * be there; what comes after them doesn't matter. See tables in utf8.h,
782 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
786 /* The missing bytes could cause the result to go one way or the other, so
787 * the result is indeterminate */
791 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
792 # ifdef EBCDIC /* Actually is I8 */
793 # define HIGHEST_REPRESENTABLE_UTF8 \
794 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
796 # define HIGHEST_REPRESENTABLE_UTF8 \
797 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
801 PERL_STATIC_INLINE int
802 S_does_utf8_overflow(const U8 * const s,
804 const bool consider_overlongs)
806 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
807 * 'e' - 1 would overflow an IV on this platform; that is if it represents
808 * a code point larger than the highest representable code point. It
809 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
810 * enough information to tell. This last return value can happen if the
811 * sequence is incomplete, missing some trailing bytes that would form a
812 * complete character. If there are enough bytes to make a definitive
813 * decision, this function does so.
815 * If 'consider_overlongs' is TRUE, the function checks for the possibility
816 * that the sequence is an overlong that doesn't overflow. Otherwise, it
817 * assumes the sequence is not an overlong. This can give different
818 * results only on ASCII 32-bit platforms.
820 * (For ASCII platforms, we could use memcmp() because we don't have to
821 * convert each byte to I8, but it's very rare input indeed that would
822 * approach overflow, so the loop below will likely only get executed once.)
824 * 'e' - 1 must not be beyond a full character. */
827 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
828 assert(s <= e && s + UTF8SKIP(s) >= e);
830 #if ! defined(UV_IS_QUAD)
832 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
836 PERL_UNUSED_ARG(consider_overlongs);
839 const STRLEN len = e - s;
841 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
843 for (x = s; x < e; x++, y++) {
845 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
849 /* If this byte is larger than the corresponding highest UTF-8
850 * byte, the sequence overflow; otherwise the byte is less than,
851 * and so the sequence doesn't overflow */
852 return NATIVE_UTF8_TO_I8(*x) > *y;
856 /* Got to the end and all bytes are the same. If the input is a whole
857 * character, it doesn't overflow. And if it is a partial character,
858 * there's not enough information to tell */
859 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
872 /* This is the portions of the above function that deal with UV_MAX instead of
873 * IV_MAX. They are left here in case we want to combine them so that internal
874 * uses can have larger code points. The only logic difference is that the
875 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
879 /* Anything larger than this will overflow the word if it were converted into a UV */
880 #if defined(UV_IS_QUAD)
881 # ifdef EBCDIC /* Actually is I8 */
882 # define HIGHEST_REPRESENTABLE_UTF8 \
883 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
885 # define HIGHEST_REPRESENTABLE_UTF8 \
886 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
890 # define HIGHEST_REPRESENTABLE_UTF8 \
891 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
893 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
897 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
899 /* On 32 bit ASCII machines, many overlongs that start with FF don't
901 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
903 /* To be such an overlong, the first bytes of 's' must match
904 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
905 * don't have any additional bytes available, the sequence, when
906 * completed might or might not fit in 32 bits. But if we have that
907 * next byte, we can tell for sure. If it is <= 0x83, then it does
909 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
913 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
916 /* Starting with the #else, the rest of the function is identical except
917 * 1. we need to move the 'len' declaration to be global to the function
918 * 2. the endif move to just after the UNUSED_ARG.
919 * An empty endif is given just below to satisfy the preprocessor
925 #undef F0_ABOVE_OVERLONG
926 #undef F8_ABOVE_OVERLONG
927 #undef FC_ABOVE_OVERLONG
928 #undef FE_ABOVE_OVERLONG
929 #undef FF_OVERLONG_PREFIX
932 Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
937 /* A helper function that should not be called directly.
939 * This function returns non-zero if the string beginning at 's' and
940 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
941 * code point; otherwise it returns 0. The examination stops after the
942 * first code point in 's' is validated, not looking at the rest of the
943 * input. If 'e' is such that there are not enough bytes to represent a
944 * complete code point, this function will return non-zero anyway, if the
945 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
946 * excluded by 'flags'.
948 * A non-zero return gives the number of bytes required to represent the
949 * code point. Be aware that if the input is for a partial character, the
950 * return will be larger than 'e - s'.
952 * This function assumes that the code point represented is UTF-8 variant.
953 * The caller should have excluded the possibility of it being invariant
954 * before calling this function.
956 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
957 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
958 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
959 * disallowed by the flags. If the input is only for a partial character,
960 * the function will return non-zero if there is any sequence of
961 * well-formed UTF-8 that, when appended to the input sequence, could
962 * result in an allowed code point; otherwise it returns 0. Non characters
963 * cannot be determined based on partial character input. But many of the
964 * other excluded types can be determined with just the first one or two
969 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
971 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
972 |UTF8_DISALLOW_PERL_EXTENDED)));
973 assert(! UTF8_IS_INVARIANT(*s));
975 /* A variant char must begin with a start byte */
976 if (UNLIKELY(! UTF8_IS_START(*s))) {
980 /* Examine a maximum of a single whole code point */
981 if (e - s > UTF8SKIP(s)) {
987 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
988 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
990 /* Here, we are disallowing some set of largish code points, and the
991 * first byte indicates the sequence is for a code point that could be
992 * in the excluded set. We generally don't have to look beyond this or
993 * the second byte to see if the sequence is actually for one of the
994 * excluded classes. The code below is derived from this table:
996 * UTF-8 UTF-EBCDIC I8
997 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
998 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
999 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
1001 * Keep in mind that legal continuation bytes range between \x80..\xBF
1002 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
1003 * continuation bytes. Hence, we don't have to test the upper edge
1004 * because if any of those is encountered, the sequence is malformed,
1005 * and would fail elsewhere in this function.
1007 * The code here likewise assumes that there aren't other
1008 * malformations; again the function should fail elsewhere because of
1009 * these. For example, an overlong beginning with FC doesn't actually
1010 * have to be a super; it could actually represent a small code point,
1011 * even U+0000. But, since overlongs (and other malformations) are
1012 * illegal, the function should return FALSE in either case.
1015 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1016 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
1017 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
1019 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1021 && ((s1) & 0xFE ) == 0xB6)
1022 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1024 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1025 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1026 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1027 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1030 if ( (flags & UTF8_DISALLOW_SUPER)
1031 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1033 return 0; /* Above Unicode */
1036 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1037 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1043 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1045 if ( (flags & UTF8_DISALLOW_SUPER)
1046 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1048 return 0; /* Above Unicode */
1051 if ( (flags & UTF8_DISALLOW_SURROGATE)
1052 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1054 return 0; /* Surrogate */
1057 if ( (flags & UTF8_DISALLOW_NONCHAR)
1058 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1060 return 0; /* Noncharacter code point */
1065 /* Make sure that all that follows are continuation bytes */
1066 for (x = s + 1; x < e; x++) {
1067 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1072 /* Here is syntactically valid. Next, make sure this isn't the start of an
1074 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1078 /* And finally, that the code point represented fits in a word on this
1080 if (0 < does_utf8_overflow(s, e,
1081 0 /* Don't consider overlongs */
1091 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1093 /* Returns a mortalized C string that is a displayable copy of the 'len'
1094 * bytes starting at 'start'. 'format' gives how to display each byte.
1095 * Currently, there are only two formats, so it is currently a bool:
1097 * 1 ab (that is a space between two hex digit bytes)
1100 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1102 const U8 * s = start;
1103 const U8 * const e = start + len;
1107 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1109 Newx(output, output_len, char);
1113 for (s = start; s < e; s++) {
1114 const unsigned high_nibble = (*s & 0xF0) >> 4;
1115 const unsigned low_nibble = (*s & 0x0F);
1127 if (high_nibble < 10) {
1128 *d++ = high_nibble + '0';
1131 *d++ = high_nibble - 10 + 'a';
1134 if (low_nibble < 10) {
1135 *d++ = low_nibble + '0';
1138 *d++ = low_nibble - 10 + 'a';
1146 PERL_STATIC_INLINE char *
1147 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1149 /* Max number of bytes to print */
1152 /* Which one is the non-continuation */
1153 const STRLEN non_cont_byte_pos,
1155 /* How many bytes should there be? */
1156 const STRLEN expect_len)
1158 /* Return the malformation warning text for an unexpected continuation
1161 const char * const where = (non_cont_byte_pos == 1)
1163 : Perl_form(aTHX_ "%d bytes",
1164 (int) non_cont_byte_pos);
1165 const U8 * x = s + non_cont_byte_pos;
1166 const U8 * e = s + print_len;
1168 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1170 /* We don't need to pass this parameter, but since it has already been
1171 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1172 assert(expect_len == UTF8SKIP(s));
1174 /* As a defensive coding measure, don't output anything past a NUL. Such
1175 * bytes shouldn't be in the middle of a malformation, and could mark the
1176 * end of the allocated string, and what comes after is undefined */
1177 for (; x < e; x++) {
1179 x++; /* Output this particular NUL */
1184 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1185 " %s after start byte 0x%02x; need %d bytes, got %d)",
1187 _byte_dump_string(s, x - s, 0),
1188 *(s + non_cont_byte_pos),
1192 (int) non_cont_byte_pos);
1197 =for apidoc utf8n_to_uvchr
1199 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1200 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1203 Bottom level UTF-8 decode routine.
1204 Returns the native code point value of the first character in the string C<s>,
1205 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1206 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1207 the length, in bytes, of that character.
1209 The value of C<flags> determines the behavior when C<s> does not point to a
1210 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1211 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1212 is the next possible position in C<s> that could begin a non-malformed
1213 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1214 is raised. Some UTF-8 input sequences may contain multiple malformations.
1215 This function tries to find every possible one in each call, so multiple
1216 warnings can be raised for the same sequence.
1218 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1219 individual types of malformations, such as the sequence being overlong (that
1220 is, when there is a shorter sequence that can express the same code point;
1221 overlong sequences are expressly forbidden in the UTF-8 standard due to
1222 potential security issues). Another malformation example is the first byte of
1223 a character not being a legal first byte. See F<utf8.h> for the list of such
1224 flags. Even if allowed, this function generally returns the Unicode
1225 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1226 F<utf8.h> to override this behavior for the overlong malformations, but don't
1227 do that except for very specialized purposes.
1229 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1230 flags) malformation is found. If this flag is set, the routine assumes that
1231 the caller will raise a warning, and this function will silently just set
1232 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1234 Note that this API requires disambiguation between successful decoding a C<NUL>
1235 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1236 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1237 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1238 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1239 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1241 Certain code points are considered problematic. These are Unicode surrogates,
1242 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1243 By default these are considered regular code points, but certain situations
1244 warrant special handling for them, which can be specified using the C<flags>
1245 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1246 three classes are treated as malformations and handled as such. The flags
1247 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1248 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1249 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1250 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1251 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1253 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
1254 The difference between traditional strictness and C9 strictness is that the
1255 latter does not forbid non-character code points. (They are still discouraged,
1256 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1258 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1259 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1260 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1261 raised for their respective categories, but otherwise the code points are
1262 considered valid (not malformations). To get a category to both be treated as
1263 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1264 (But note that warnings are not raised if lexically disabled nor if
1265 C<UTF8_CHECK_ONLY> is also specified.)
1267 Extremely high code points were never specified in any standard, and require an
1268 extension to UTF-8 to express, which Perl does. It is likely that programs
1269 written in something other than Perl would not be able to read files that
1270 contain these; nor would Perl understand files written by something that uses a
1271 different extension. For these reasons, there is a separate set of flags that
1272 can warn and/or disallow these extremely high code points, even if other
1273 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1274 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1275 C<L</UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1276 above-Unicode code points, including these, as malformations.
1277 (Note that the Unicode standard considers anything above 0x10FFFF to be
1278 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1281 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1282 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1283 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1284 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1285 can apply to code points that actually do fit in 31 bits. This happens on
1286 EBCDIC platforms, and sometimes when the L<overlong
1287 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1288 describe the situation in all cases.
1291 All other code points corresponding to Unicode characters, including private
1292 use and those yet to be assigned, are never considered malformed and never
1295 =for apidoc Amnh||UTF8_CHECK_ONLY
1296 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1297 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1298 =for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1299 =for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1300 =for apidoc Amnh||UTF8_DISALLOW_SUPER
1301 =for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1302 =for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1303 =for apidoc Amnh||UTF8_WARN_SURROGATE
1304 =for apidoc Amnh||UTF8_WARN_NONCHAR
1305 =for apidoc Amnh||UTF8_WARN_SUPER
1306 =for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1307 =for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1311 Also implemented as a macro in utf8.h
1315 Perl_utf8n_to_uvchr(const U8 *s,
1320 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1322 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1327 =for apidoc utf8n_to_uvchr_error
1329 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1330 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1333 This function is for code that needs to know what the precise malformation(s)
1334 are when an error is found. If you also need to know the generated warning
1335 messages, use L</utf8n_to_uvchr_msgs>() instead.
1337 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1338 all the others, C<errors>. If this parameter is 0, this function behaves
1339 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1340 to a C<U32> variable, which this function sets to indicate any errors found.
1341 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1342 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1343 of these bits will be set if a malformation is found, even if the input
1344 C<flags> parameter indicates that the given malformation is allowed; those
1345 exceptions are noted:
1349 =item C<UTF8_GOT_PERL_EXTENDED>
1351 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1352 set only if the input C<flags> parameter contains either the
1353 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1355 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1356 and so some extension must be used to express them. Perl uses a natural
1357 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1358 extension to represent even higher ones, so that any code point that fits in a
1359 64-bit word can be represented. Text using these extensions is not likely to
1360 be portable to non-Perl code. We lump both of these extensions together and
1361 refer to them as Perl extended UTF-8. There exist other extensions that people
1362 have invented, incompatible with Perl's.
1364 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1365 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1366 than on ASCII. Prior to that, code points 2**31 and higher were simply
1367 unrepresentable, and a different, incompatible method was used to represent
1368 code points between 2**30 and 2**31 - 1.
1370 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1371 Perl extended UTF-8 is used.
1373 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1374 may use for backward compatibility. That name is misleading, as this flag may
1375 be set when the code point actually does fit in 31 bits. This happens on
1376 EBCDIC platforms, and sometimes when the L<overlong
1377 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1378 describes the situation in all cases.
1380 =item C<UTF8_GOT_CONTINUATION>
1382 The input sequence was malformed in that the first byte was a UTF-8
1385 =item C<UTF8_GOT_EMPTY>
1387 The input C<curlen> parameter was 0.
1389 =item C<UTF8_GOT_LONG>
1391 The input sequence was malformed in that there is some other sequence that
1392 evaluates to the same code point, but that sequence is shorter than this one.
1394 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1395 it was discovered that this created security issues.
1397 =item C<UTF8_GOT_NONCHAR>
1399 The code point represented by the input UTF-8 sequence is for a Unicode
1400 non-character code point.
1401 This bit is set only if the input C<flags> parameter contains either the
1402 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1404 =item C<UTF8_GOT_NON_CONTINUATION>
1406 The input sequence was malformed in that a non-continuation type byte was found
1407 in a position where only a continuation type one should be. See also
1408 C<L</UTF8_GOT_SHORT>>.
1410 =item C<UTF8_GOT_OVERFLOW>
1412 The input sequence was malformed in that it is for a code point that is not
1413 representable in the number of bits available in an IV on the current platform.
1415 =item C<UTF8_GOT_SHORT>
1417 The input sequence was malformed in that C<curlen> is smaller than required for
1418 a complete sequence. In other words, the input is for a partial character
1422 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1423 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1424 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1425 sequence was looked at. If no other flags are present, it means that the
1426 sequence was valid as far as it went. Depending on the application, this could
1427 mean one of three things:
1433 The C<curlen> length parameter passed in was too small, and the function was
1434 prevented from examining all the necessary bytes.
1438 The buffer being looked at is based on reading data, and the data received so
1439 far stopped in the middle of a character, so that the next read will
1440 read the remainder of this character. (It is up to the caller to deal with the
1441 split bytes somehow.)
1445 This is a real error, and the partial sequence is all we're going to get.
1449 =item C<UTF8_GOT_SUPER>
1451 The input sequence was malformed in that it is for a non-Unicode code point;
1452 that is, one above the legal Unicode maximum.
1453 This bit is set only if the input C<flags> parameter contains either the
1454 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1456 =item C<UTF8_GOT_SURROGATE>
1458 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1460 This bit is set only if the input C<flags> parameter contains either the
1461 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1465 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1466 flag to suppress any warnings, and then examine the C<*errors> return.
1468 =for apidoc Amnh||UTF8_GOT_PERL_EXTENDED
1469 =for apidoc Amnh||UTF8_GOT_CONTINUATION
1470 =for apidoc Amnh||UTF8_GOT_EMPTY
1471 =for apidoc Amnh||UTF8_GOT_LONG
1472 =for apidoc Amnh||UTF8_GOT_NONCHAR
1473 =for apidoc Amnh||UTF8_GOT_NON_CONTINUATION
1474 =for apidoc Amnh||UTF8_GOT_OVERFLOW
1475 =for apidoc Amnh||UTF8_GOT_SHORT
1476 =for apidoc Amnh||UTF8_GOT_SUPER
1477 =for apidoc Amnh||UTF8_GOT_SURROGATE
1481 Also implemented as a macro in utf8.h
1485 Perl_utf8n_to_uvchr_error(const U8 *s,
1491 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1493 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1498 =for apidoc utf8n_to_uvchr_msgs
1500 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1501 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1504 This function is for code that needs to know what the precise malformation(s)
1505 are when an error is found, and wants the corresponding warning and/or error
1506 messages to be returned to the caller rather than be displayed. All messages
1507 that would have been displayed if all lexical warnings are enabled will be
1510 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1511 placed after all the others, C<msgs>. If this parameter is 0, this function
1512 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1513 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1514 contain any appropriate messages. The elements of the array are ordered so
1515 that the first message that would have been displayed is in the 0th element,
1516 and so on. Each element is a hash with three key-value pairs, as follows:
1522 The text of the message as a C<SVpv>.
1524 =item C<warn_categories>
1526 The warning category (or categories) packed into a C<SVuv>.
1530 A single flag bit associated with this message, in a C<SVuv>.
1531 The bit corresponds to some bit in the C<*errors> return value,
1532 such as C<UTF8_GOT_LONG>.
1536 It's important to note that specifying this parameter as non-null will cause
1537 any warnings this function would otherwise generate to be suppressed, and
1538 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1539 (or not) when choosing what to do with the returned messages.
1541 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1544 The caller, of course, is responsible for freeing any returned AV.
1550 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1557 const U8 * const s0 = s;
1558 const U8 * send = s0 + curlen;
1559 U32 possible_problems; /* A bit is set here for each potential problem
1560 found as we go along */
1562 STRLEN expectlen; /* How long should this sequence be? */
1563 STRLEN avail_len; /* When input is too short, gives what that is */
1564 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1565 gets set and discarded */
1567 /* The below are used only if there is both an overlong malformation and a
1568 * too short one. Otherwise the first two are set to 's0' and 'send', and
1569 * the third not used at all */
1571 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1572 routine; see [perl #130921] */
1576 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1578 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1579 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1580 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1585 /* Each of the affected Hanguls starts with \xED */
1587 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1598 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1599 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1600 | (s0[2] & UTF_CONTINUATION_MASK);
1605 /* In conjunction with the exhaustive tests that can be enabled in
1606 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1607 * what it is intended to do, and that no flaws in it are masked by
1608 * dropping down and executing the code below
1609 assert(! isUTF8_CHAR(s0, send)
1610 || UTF8_IS_SURROGATE(s0, send)
1611 || UTF8_IS_SUPER(s0, send)
1612 || UTF8_IS_NONCHAR(s0,send));
1617 possible_problems = 0;
1621 adjusted_s0 = (U8 *) s0;
1628 errors = &discard_errors;
1631 /* The order of malformation tests here is important. We should consume as
1632 * few bytes as possible in order to not skip any valid character. This is
1633 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1634 * https://unicode.org/reports/tr36 for more discussion as to why. For
1635 * example, once we've done a UTF8SKIP, we can tell the expected number of
1636 * bytes, and could fail right off the bat if the input parameters indicate
1637 * that there are too few available. But it could be that just that first
1638 * byte is garbled, and the intended character occupies fewer bytes. If we
1639 * blindly assumed that the first byte is correct, and skipped based on
1640 * that number, we could skip over a valid input character. So instead, we
1641 * always examine the sequence byte-by-byte.
1643 * We also should not consume too few bytes, otherwise someone could inject
1644 * things. For example, an input could be deliberately designed to
1645 * overflow, and if this code bailed out immediately upon discovering that,
1646 * returning to the caller C<*retlen> pointing to the very next byte (one
1647 * which is actually part of the overflowing sequence), that could look
1648 * legitimate to the caller, which could discard the initial partial
1649 * sequence and process the rest, inappropriately.
1651 * Some possible input sequences are malformed in more than one way. This
1652 * function goes to lengths to try to find all of them. This is necessary
1653 * for correctness, as the inputs may allow one malformation but not
1654 * another, and if we abandon searching for others after finding the
1655 * allowed one, we could allow in something that shouldn't have been.
1658 if (UNLIKELY(curlen == 0)) {
1659 possible_problems |= UTF8_GOT_EMPTY;
1661 uv = UNICODE_REPLACEMENT;
1662 goto ready_to_handle_errors;
1665 expectlen = UTF8SKIP(s);
1667 /* A well-formed UTF-8 character, as the vast majority of calls to this
1668 * function will be for, has this expected length. For efficiency, set
1669 * things up here to return it. It will be overriden only in those rare
1670 * cases where a malformation is found */
1672 *retlen = expectlen;
1675 /* A continuation character can't start a valid sequence */
1676 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1677 possible_problems |= UTF8_GOT_CONTINUATION;
1679 uv = UNICODE_REPLACEMENT;
1680 goto ready_to_handle_errors;
1683 /* Here is not a continuation byte, nor an invariant. The only thing left
1684 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1685 * because it excludes start bytes like \xC0 that always lead to
1688 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1689 * that indicate the number of bytes in the character's whole UTF-8
1690 * sequence, leaving just the bits that are part of the value. */
1691 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1693 /* Setup the loop end point, making sure to not look past the end of the
1694 * input string, and flag it as too short if the size isn't big enough. */
1695 if (UNLIKELY(curlen < expectlen)) {
1696 possible_problems |= UTF8_GOT_SHORT;
1700 send = (U8*) s0 + expectlen;
1703 /* Now, loop through the remaining bytes in the character's sequence,
1704 * accumulating each into the working value as we go. */
1705 for (s = s0 + 1; s < send; s++) {
1706 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1707 uv = UTF8_ACCUMULATE(uv, *s);
1711 /* Here, found a non-continuation before processing all expected bytes.
1712 * This byte indicates the beginning of a new character, so quit, even
1713 * if allowing this malformation. */
1714 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1716 } /* End of loop through the character's bytes */
1718 /* Save how many bytes were actually in the character */
1721 /* Note that there are two types of too-short malformation. One is when
1722 * there is actual wrong data before the normal termination of the
1723 * sequence. The other is that the sequence wasn't complete before the end
1724 * of the data we are allowed to look at, based on the input 'curlen'.
1725 * This means that we were passed data for a partial character, but it is
1726 * valid as far as we saw. The other is definitely invalid. This
1727 * distinction could be important to a caller, so the two types are kept
1730 * A convenience macro that matches either of the too-short conditions. */
1731 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1733 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1735 uv = UNICODE_REPLACEMENT;
1738 /* Check for overflow. The algorithm requires us to not look past the end
1739 * of the current character, even if partial, so the upper limit is 's' */
1740 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1741 1 /* Do consider overlongs */
1744 possible_problems |= UTF8_GOT_OVERFLOW;
1745 uv = UNICODE_REPLACEMENT;
1748 /* Check for overlong. If no problems so far, 'uv' is the correct code
1749 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1750 * we must look at the UTF-8 byte sequence itself to see if it is for an
1752 if ( ( LIKELY(! possible_problems)
1753 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1754 || ( UNLIKELY(possible_problems)
1755 && ( UNLIKELY(! UTF8_IS_START(*s0))
1757 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1760 possible_problems |= UTF8_GOT_LONG;
1762 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1764 /* The calculation in the 'true' branch of this 'if'
1765 * below won't work if overflows, and isn't needed
1766 * anyway. Further below we handle all overflow
1768 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1770 UV min_uv = uv_so_far;
1773 /* Here, the input is both overlong and is missing some trailing
1774 * bytes. There is no single code point it could be for, but there
1775 * may be enough information present to determine if what we have
1776 * so far is for an unallowed code point, such as for a surrogate.
1777 * The code further below has the intelligence to determine this,
1778 * but just for non-overlong UTF-8 sequences. What we do here is
1779 * calculate the smallest code point the input could represent if
1780 * there were no too short malformation. Then we compute and save
1781 * the UTF-8 for that, which is what the code below looks at
1782 * instead of the raw input. It turns out that the smallest such
1783 * code point is all we need. */
1784 for (i = curlen; i < expectlen; i++) {
1785 min_uv = UTF8_ACCUMULATE(min_uv,
1786 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1789 adjusted_s0 = temp_char_buf;
1790 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1794 /* Here, we have found all the possible problems, except for when the input
1795 * is for a problematic code point not allowed by the input parameters. */
1797 /* uv is valid for overlongs */
1798 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1800 /* isn't problematic if < this */
1801 && uv >= UNICODE_SURROGATE_FIRST)
1802 || ( UNLIKELY(possible_problems)
1804 /* if overflow, we know without looking further
1805 * precisely which of the problematic types it is,
1806 * and we deal with those in the overflow handling
1808 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1809 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1810 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1811 && ((flags & ( UTF8_DISALLOW_NONCHAR
1812 |UTF8_DISALLOW_SURROGATE
1813 |UTF8_DISALLOW_SUPER
1814 |UTF8_DISALLOW_PERL_EXTENDED
1816 |UTF8_WARN_SURROGATE
1818 |UTF8_WARN_PERL_EXTENDED))))
1820 /* If there were no malformations, or the only malformation is an
1821 * overlong, 'uv' is valid */
1822 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1823 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1824 possible_problems |= UTF8_GOT_SURROGATE;
1826 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1827 possible_problems |= UTF8_GOT_SUPER;
1829 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1830 possible_problems |= UTF8_GOT_NONCHAR;
1833 else { /* Otherwise, need to look at the source UTF-8, possibly
1834 adjusted to be non-overlong */
1836 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1837 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1839 possible_problems |= UTF8_GOT_SUPER;
1841 else if (curlen > 1) {
1842 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1843 NATIVE_UTF8_TO_I8(*adjusted_s0),
1844 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1846 possible_problems |= UTF8_GOT_SUPER;
1848 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1849 NATIVE_UTF8_TO_I8(*adjusted_s0),
1850 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1852 possible_problems |= UTF8_GOT_SURROGATE;
1856 /* We need a complete well-formed UTF-8 character to discern
1857 * non-characters, so can't look for them here */
1861 ready_to_handle_errors:
1864 * curlen contains the number of bytes in the sequence that
1865 * this call should advance the input by.
1866 * avail_len gives the available number of bytes passed in, but
1867 * only if this is less than the expected number of
1868 * bytes, based on the code point's start byte.
1869 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1870 * is set in it for each potential problem found.
1871 * uv contains the code point the input sequence
1872 * represents; or if there is a problem that prevents
1873 * a well-defined value from being computed, it is
1874 * some subsitute value, typically the REPLACEMENT
1876 * s0 points to the first byte of the character
1877 * s points to just after were we left off processing
1879 * send points to just after where that character should
1880 * end, based on how many bytes the start byte tells
1881 * us should be in it, but no further than s0 +
1885 if (UNLIKELY(possible_problems)) {
1886 bool disallowed = FALSE;
1887 const U32 orig_problems = possible_problems;
1893 while (possible_problems) { /* Handle each possible problem */
1895 char * message = NULL;
1896 U32 this_flag_bit = 0;
1898 /* Each 'if' clause handles one problem. They are ordered so that
1899 * the first ones' messages will be displayed before the later
1900 * ones; this is kinda in decreasing severity order. But the
1901 * overlong must come last, as it changes 'uv' looked at by the
1903 if (possible_problems & UTF8_GOT_OVERFLOW) {
1905 /* Overflow means also got a super and are using Perl's
1906 * extended UTF-8, but we handle all three cases here */
1908 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1909 *errors |= UTF8_GOT_OVERFLOW;
1911 /* But the API says we flag all errors found */
1912 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1913 *errors |= UTF8_GOT_SUPER;
1916 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1918 *errors |= UTF8_GOT_PERL_EXTENDED;
1921 /* Disallow if any of the three categories say to */
1922 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1923 || (flags & ( UTF8_DISALLOW_SUPER
1924 |UTF8_DISALLOW_PERL_EXTENDED)))
1929 /* Likewise, warn if any say to */
1930 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1931 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1934 /* The warnings code explicitly says it doesn't handle the
1935 * case of packWARN2 and two categories which have
1936 * parent-child relationship. Even if it works now to
1937 * raise the warning if either is enabled, it wouldn't
1938 * necessarily do so in the future. We output (only) the
1939 * most dire warning */
1940 if (! (flags & UTF8_CHECK_ONLY)) {
1941 if (msgs || ckWARN_d(WARN_UTF8)) {
1942 pack_warn = packWARN(WARN_UTF8);
1944 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1945 pack_warn = packWARN(WARN_NON_UNICODE);
1948 message = Perl_form(aTHX_ "%s: %s (overflows)",
1950 _byte_dump_string(s0, curlen, 0));
1951 this_flag_bit = UTF8_GOT_OVERFLOW;
1956 else if (possible_problems & UTF8_GOT_EMPTY) {
1957 possible_problems &= ~UTF8_GOT_EMPTY;
1958 *errors |= UTF8_GOT_EMPTY;
1960 if (! (flags & UTF8_ALLOW_EMPTY)) {
1962 /* This so-called malformation is now treated as a bug in
1963 * the caller. If you have nothing to decode, skip calling
1969 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1971 pack_warn = packWARN(WARN_UTF8);
1972 message = Perl_form(aTHX_ "%s (empty string)",
1974 this_flag_bit = UTF8_GOT_EMPTY;
1978 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1979 possible_problems &= ~UTF8_GOT_CONTINUATION;
1980 *errors |= UTF8_GOT_CONTINUATION;
1982 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1985 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1987 pack_warn = packWARN(WARN_UTF8);
1988 message = Perl_form(aTHX_
1989 "%s: %s (unexpected continuation byte 0x%02x,"
1990 " with no preceding start byte)",
1992 _byte_dump_string(s0, 1, 0), *s0);
1993 this_flag_bit = UTF8_GOT_CONTINUATION;
1997 else if (possible_problems & UTF8_GOT_SHORT) {
1998 possible_problems &= ~UTF8_GOT_SHORT;
1999 *errors |= UTF8_GOT_SHORT;
2001 if (! (flags & UTF8_ALLOW_SHORT)) {
2004 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2006 pack_warn = packWARN(WARN_UTF8);
2007 message = Perl_form(aTHX_
2008 "%s: %s (too short; %d byte%s available, need %d)",
2010 _byte_dump_string(s0, send - s0, 0),
2012 avail_len == 1 ? "" : "s",
2014 this_flag_bit = UTF8_GOT_SHORT;
2019 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2020 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2021 *errors |= UTF8_GOT_NON_CONTINUATION;
2023 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2026 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2029 /* If we don't know for sure that the input length is
2030 * valid, avoid as much as possible reading past the
2031 * end of the buffer */
2032 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2034 : (int) (send - s0);
2035 pack_warn = packWARN(WARN_UTF8);
2036 message = Perl_form(aTHX_ "%s",
2037 unexpected_non_continuation_text(s0,
2041 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2045 else if (possible_problems & UTF8_GOT_SURROGATE) {
2046 possible_problems &= ~UTF8_GOT_SURROGATE;
2048 if (flags & UTF8_WARN_SURROGATE) {
2049 *errors |= UTF8_GOT_SURROGATE;
2051 if ( ! (flags & UTF8_CHECK_ONLY)
2052 && (msgs || ckWARN_d(WARN_SURROGATE)))
2054 pack_warn = packWARN(WARN_SURROGATE);
2056 /* These are the only errors that can occur with a
2057 * surrogate when the 'uv' isn't valid */
2058 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2059 message = Perl_form(aTHX_
2060 "UTF-16 surrogate (any UTF-8 sequence that"
2061 " starts with \"%s\" is for a surrogate)",
2062 _byte_dump_string(s0, curlen, 0));
2065 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2067 this_flag_bit = UTF8_GOT_SURROGATE;
2071 if (flags & UTF8_DISALLOW_SURROGATE) {
2073 *errors |= UTF8_GOT_SURROGATE;
2076 else if (possible_problems & UTF8_GOT_SUPER) {
2077 possible_problems &= ~UTF8_GOT_SUPER;
2079 if (flags & UTF8_WARN_SUPER) {
2080 *errors |= UTF8_GOT_SUPER;
2082 if ( ! (flags & UTF8_CHECK_ONLY)
2083 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2085 pack_warn = packWARN(WARN_NON_UNICODE);
2087 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2088 message = Perl_form(aTHX_
2089 "Any UTF-8 sequence that starts with"
2090 " \"%s\" is for a non-Unicode code point,"
2091 " may not be portable",
2092 _byte_dump_string(s0, curlen, 0));
2095 message = Perl_form(aTHX_ super_cp_format, uv);
2097 this_flag_bit = UTF8_GOT_SUPER;
2101 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2102 * and before possibly bailing out, so that the more dire
2103 * warning will override the regular one. */
2104 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2105 if ( ! (flags & UTF8_CHECK_ONLY)
2106 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2107 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2108 || ckWARN(WARN_PORTABLE))))
2110 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2112 /* If it is an overlong that evaluates to a code point
2113 * that doesn't have to use the Perl extended UTF-8, it
2114 * still used it, and so we output a message that
2115 * doesn't refer to the code point. The same is true
2116 * if there was a SHORT malformation where the code
2117 * point is not valid. In that case, 'uv' will have
2118 * been set to the REPLACEMENT CHAR, and the message
2119 * below without the code point in it will be selected
2121 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2122 message = Perl_form(aTHX_
2123 PL_extended_cp_format, uv);
2126 message = Perl_form(aTHX_
2127 "Any UTF-8 sequence that starts with"
2128 " \"%s\" is a Perl extension, and"
2129 " so is not portable",
2130 _byte_dump_string(s0, curlen, 0));
2132 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2135 if (flags & ( UTF8_WARN_PERL_EXTENDED
2136 |UTF8_DISALLOW_PERL_EXTENDED))
2138 *errors |= UTF8_GOT_PERL_EXTENDED;
2140 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2146 if (flags & UTF8_DISALLOW_SUPER) {
2147 *errors |= UTF8_GOT_SUPER;
2151 else if (possible_problems & UTF8_GOT_NONCHAR) {
2152 possible_problems &= ~UTF8_GOT_NONCHAR;
2154 if (flags & UTF8_WARN_NONCHAR) {
2155 *errors |= UTF8_GOT_NONCHAR;
2157 if ( ! (flags & UTF8_CHECK_ONLY)
2158 && (msgs || ckWARN_d(WARN_NONCHAR)))
2160 /* The code above should have guaranteed that we don't
2161 * get here with errors other than overlong */
2162 assert (! (orig_problems
2163 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2165 pack_warn = packWARN(WARN_NONCHAR);
2166 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2167 this_flag_bit = UTF8_GOT_NONCHAR;
2171 if (flags & UTF8_DISALLOW_NONCHAR) {
2173 *errors |= UTF8_GOT_NONCHAR;
2176 else if (possible_problems & UTF8_GOT_LONG) {
2177 possible_problems &= ~UTF8_GOT_LONG;
2178 *errors |= UTF8_GOT_LONG;
2180 if (flags & UTF8_ALLOW_LONG) {
2182 /* We don't allow the actual overlong value, unless the
2183 * special extra bit is also set */
2184 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2185 & ~UTF8_ALLOW_LONG)))
2187 uv = UNICODE_REPLACEMENT;
2194 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2196 pack_warn = packWARN(WARN_UTF8);
2198 /* These error types cause 'uv' to be something that
2199 * isn't what was intended, so can't use it in the
2200 * message. The other error types either can't
2201 * generate an overlong, or else the 'uv' is valid */
2203 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2205 message = Perl_form(aTHX_
2206 "%s: %s (any UTF-8 sequence that starts"
2207 " with \"%s\" is overlong which can and"
2208 " should be represented with a"
2209 " different, shorter sequence)",
2211 _byte_dump_string(s0, send - s0, 0),
2212 _byte_dump_string(s0, curlen, 0));
2215 U8 tmpbuf[UTF8_MAXBYTES+1];
2216 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2218 /* Don't use U+ for non-Unicode code points, which
2219 * includes those in the Latin1 range */
2220 const char * preface = ( uv > PERL_UNICODE_MAX
2227 message = Perl_form(aTHX_
2228 "%s: %s (overlong; instead use %s to represent"
2231 _byte_dump_string(s0, send - s0, 0),
2232 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2234 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2235 small code points */
2238 this_flag_bit = UTF8_GOT_LONG;
2241 } /* End of looking through the possible flags */
2243 /* Display the message (if any) for the problem being handled in
2244 * this iteration of the loop */
2247 assert(this_flag_bit);
2249 if (*msgs == NULL) {
2253 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2258 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2261 Perl_warner(aTHX_ pack_warn, "%s", message);
2263 } /* End of 'while (possible_problems)' */
2265 /* Since there was a possible problem, the returned length may need to
2266 * be changed from the one stored at the beginning of this function.
2267 * Instead of trying to figure out if that's needed, just do it. */
2273 if (flags & UTF8_CHECK_ONLY && retlen) {
2274 *retlen = ((STRLEN) -1);
2280 return UNI_TO_NATIVE(uv);
2284 =for apidoc utf8_to_uvchr_buf
2286 Returns the native code point of the first character in the string C<s> which
2287 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2288 C<*retlen> will be set to the length, in bytes, of that character.
2290 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2291 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2292 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2293 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2294 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2295 the next possible position in C<s> that could begin a non-malformed character.
2296 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2301 Also implemented as a macro in utf8.h
2307 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2309 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2311 return utf8_to_uvchr_buf_helper(s, send, retlen);
2314 /* This is marked as deprecated
2316 =for apidoc utf8_to_uvuni_buf
2318 Only in very rare circumstances should code need to be dealing in Unicode
2319 (as opposed to native) code points. In those few cases, use
2320 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2321 If you are not absolutely sure this is one of those cases, then assume it isn't
2322 and use plain C<utf8_to_uvchr_buf> instead.
2324 Returns the Unicode (not-native) code point of the first character in the
2326 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2327 C<retlen> will be set to the length, in bytes, of that character.
2329 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2330 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2331 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2332 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2333 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2334 next possible position in C<s> that could begin a non-malformed character.
2335 See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2342 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2344 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2348 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2352 =for apidoc utf8_length
2354 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2355 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2356 same place, it returns 0 with no warning raised.
2358 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2359 and returns the number of valid characters.
2365 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2369 PERL_ARGS_ASSERT_UTF8_LENGTH;
2371 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2372 * the bitops (especially ~) can create illegal UTF-8.
2373 * In other words: in Perl UTF-8 is not just for Unicode. */
2375 if (UNLIKELY(e < s))
2376 goto warn_and_return;
2382 if (UNLIKELY(e != s)) {
2386 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2387 "%s in %s", unees, OP_DESC(PL_op));
2389 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2396 =for apidoc bytes_cmp_utf8
2398 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2399 sequence of characters (stored as UTF-8)
2400 in C<u>, C<ulen>. Returns 0 if they are
2401 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2402 if the first string is greater than the second string.
2404 -1 or +1 is returned if the shorter string was identical to the start of the
2405 longer string. -2 or +2 is returned if
2406 there was a difference between characters
2413 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2415 const U8 *const bend = b + blen;
2416 const U8 *const uend = u + ulen;
2418 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2420 while (b < bend && u < uend) {
2422 if (!UTF8_IS_INVARIANT(c)) {
2423 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2426 if (UTF8_IS_CONTINUATION(c1)) {
2427 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2429 /* diag_listed_as: Malformed UTF-8 character%s */
2430 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2432 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2433 PL_op ? " in " : "",
2434 PL_op ? OP_DESC(PL_op) : "");
2439 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2440 "%s in %s", unees, OP_DESC(PL_op));
2442 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2443 return -2; /* Really want to return undef :-) */
2450 return *b < c ? -2 : +2;
2455 if (b == bend && u == uend)
2458 return b < bend ? +1 : -1;
2462 =for apidoc utf8_to_bytes
2464 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2465 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2466 updates C<*lenp> to contain the new length.
2467 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2469 Upon successful return, the number of variants in the string can be computed by
2470 having saved the value of C<*lenp> before the call, and subtracting the
2471 after-call value of C<*lenp> from it.
2473 If you need a copy of the string, see L</bytes_from_utf8>.
2479 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2483 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2484 PERL_UNUSED_CONTEXT;
2486 /* This is a no-op if no variants at all in the input */
2487 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2492 U8 * const save = s;
2493 U8 * const send = s + *lenp;
2496 /* Nothing before the first variant needs to be changed, so start the real
2500 if (! UTF8_IS_INVARIANT(*s)) {
2501 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2502 *lenp = ((STRLEN) -1);
2510 /* Is downgradable, so do it */
2511 d = s = first_variant;
2514 if (! UVCHR_IS_INVARIANT(c)) {
2515 /* Then it is two-byte encoded */
2516 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2529 =for apidoc bytes_from_utf8
2531 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2532 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2533 actually encoded in UTF-8.
2535 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2538 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2539 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2540 C<*lenp> are unchanged, and the return value is the original C<s>.
2542 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2543 newly created string containing a downgraded copy of C<s>, and whose length is
2544 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2545 caller is responsible for arranging for the memory used by this string to get
2548 Upon successful return, the number of variants in the string can be computed by
2549 having saved the value of C<*lenp> before the call, and subtracting the
2550 after-call value of C<*lenp> from it.
2554 There is a macro that avoids this function call, but this is retained for
2555 anyone who calls it with the Perl_ prefix */
2558 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2560 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2561 PERL_UNUSED_CONTEXT;
2563 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2567 =for apidoc bytes_from_utf8_loc
2569 Like C<L<perlapi/bytes_from_utf8>()>, but takes an extra parameter, a pointer
2570 to where to store the location of the first character in C<"s"> that cannot be
2571 converted to non-UTF8.
2573 If that parameter is C<NULL>, this function behaves identically to
2576 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2577 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2579 Otherwise, the function returns a newly created C<NUL>-terminated string
2580 containing the non-UTF8 equivalent of the convertible first portion of
2581 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2582 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2583 and C<*first_non_downgradable> is set to C<NULL>.
2585 Otherwise, C<*first_non_downgradable> is set to point to the first byte of the
2586 first character in the original string that wasn't converted. C<*is_utf8p> is
2587 unchanged. Note that the new string may have length 0.
2589 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2590 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2591 converts as many characters in it as possible stopping at the first one it
2592 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2593 set to point to that. The function returns the portion that could be converted
2594 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2595 not including the terminating C<NUL>. If the very first character in the
2596 original could not be converted, C<*lenp> will be 0, and the new string will
2597 contain just a single C<NUL>. If the entire input string was converted,
2598 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2600 Upon successful return, the number of variants in the converted portion of the
2601 string can be computed by having saved the value of C<*lenp> before the call,
2602 and subtracting the after-call value of C<*lenp> from it.
2610 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2613 const U8 *original = s;
2614 U8 *converted_start;
2615 const U8 *send = s + *lenp;
2617 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2620 if (first_unconverted) {
2621 *first_unconverted = NULL;
2624 return (U8 *) original;
2627 Newx(d, (*lenp) + 1, U8);
2629 converted_start = d;
2632 if (! UTF8_IS_INVARIANT(c)) {
2634 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2635 * have to stop now */
2636 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2637 if (first_unconverted) {
2638 *first_unconverted = s - 1;
2639 goto finish_and_return;
2642 Safefree(converted_start);
2643 return (U8 *) original;
2647 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2653 /* Here, converted the whole of the input */
2655 if (first_unconverted) {
2656 *first_unconverted = NULL;
2661 *lenp = d - converted_start;
2663 /* Trim unused space */
2664 Renew(converted_start, *lenp + 1, U8);
2666 return converted_start;
2670 =for apidoc bytes_to_utf8
2672 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2674 Returns a pointer to the newly-created string, and sets C<*lenp> to
2675 reflect the new length in bytes. The caller is responsible for arranging for
2676 the memory used by this string to get freed.
2678 Upon successful return, the number of variants in the string can be computed by
2679 having saved the value of C<*lenp> before the call, and subtracting it from the
2680 after-call value of C<*lenp>.
2682 A C<NUL> character will be written after the end of the string.
2684 If you want to convert to UTF-8 from encodings other than
2685 the native (Latin1 or EBCDIC),
2686 see L</sv_recode_to_utf8>().
2692 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2694 const U8 * const send = s + (*lenp);
2698 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2699 PERL_UNUSED_CONTEXT;
2701 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2702 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2706 append_utf8_from_native_byte(*s, &d);
2717 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2718 * use utf16_to_utf8_reversed().
2720 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2721 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2722 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2724 * These functions don't check for overflow. The worst case is every code
2725 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2726 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2727 * destination must be pre-extended to 2 times the source length.
2729 * Do not use in-place. We optimize for native, for obvious reasons. */
2732 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2737 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2740 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2746 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2748 if (OFFUNI_IS_INVARIANT(uv)) {
2749 *d++ = LATIN1_TO_NATIVE((U8) uv);
2752 if (uv <= MAX_UTF8_TWO_BYTE) {
2753 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2754 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2758 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2759 #define LAST_HIGH_SURROGATE 0xDBFF
2760 #define FIRST_LOW_SURROGATE 0xDC00
2761 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2762 #define FIRST_IN_PLANE1 0x10000
2764 /* This assumes that most uses will be in the first Unicode plane, not
2765 * needing surrogates */
2766 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2767 UNICODE_SURROGATE_LAST)))
2769 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2770 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2773 UV low = (p[0] << 8) + p[1];
2774 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2775 LAST_LOW_SURROGATE)))
2777 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2780 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2781 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2785 d = uvoffuni_to_utf8_flags(d, uv, 0);
2787 if (uv < FIRST_IN_PLANE1) {
2788 *d++ = (U8)(( uv >> 12) | 0xe0);
2789 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2790 *d++ = (U8)(( uv & 0x3f) | 0x80);
2794 *d++ = (U8)(( uv >> 18) | 0xf0);
2795 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2796 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2797 *d++ = (U8)(( uv & 0x3f) | 0x80);
2802 *newlen = d - dstart;
2806 /* Note: this one is slightly destructive of the source. */
2809 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2812 U8* const send = s + bytelen;
2814 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2817 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2821 const U8 tmp = s[0];
2826 return utf16_to_utf8(p, d, bytelen, newlen);
2830 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2832 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2836 Perl__is_uni_perl_idcont(pTHX_ UV c)
2838 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2842 Perl__is_uni_perl_idstart(pTHX_ UV c)
2844 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2848 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2851 /* We have the latin1-range values compiled into the core, so just use
2852 * those, converting the result to UTF-8. The only difference between upper
2853 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2854 * either "SS" or "Ss". Which one to use is passed into the routine in
2855 * 'S_or_s' to avoid a test */
2857 UV converted = toUPPER_LATIN1_MOD(c);
2859 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2861 assert(S_or_s == 'S' || S_or_s == 's');
2863 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2864 characters in this range */
2865 *p = (U8) converted;
2870 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2871 * which it maps to one of them, so as to only have to have one check for
2872 * it in the main case */
2873 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2875 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2876 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2879 converted = GREEK_CAPITAL_LETTER_MU;
2881 #if UNICODE_MAJOR_VERSION > 2 \
2882 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2883 && UNICODE_DOT_DOT_VERSION >= 8)
2884 case LATIN_SMALL_LETTER_SHARP_S:
2891 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2892 " '%c' to map to '%c'",
2893 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2894 NOT_REACHED; /* NOTREACHED */
2898 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2899 *p = UTF8_TWO_BYTE_LO(converted);
2905 /* If compiled on an early Unicode version, there may not be auxiliary tables
2907 #ifndef HAS_UC_AUX_TABLES
2908 # define UC_AUX_TABLE_ptrs NULL
2909 # define UC_AUX_TABLE_lengths NULL
2911 #ifndef HAS_TC_AUX_TABLES
2912 # define TC_AUX_TABLE_ptrs NULL
2913 # define TC_AUX_TABLE_lengths NULL
2915 #ifndef HAS_LC_AUX_TABLES
2916 # define LC_AUX_TABLE_ptrs NULL
2917 # define LC_AUX_TABLE_lengths NULL
2919 #ifndef HAS_CF_AUX_TABLES
2920 # define CF_AUX_TABLE_ptrs NULL
2921 # define CF_AUX_TABLE_lengths NULL
2923 #ifndef HAS_UC_AUX_TABLES
2924 # define UC_AUX_TABLE_ptrs NULL
2925 # define UC_AUX_TABLE_lengths NULL
2928 /* Call the function to convert a UTF-8 encoded character to the specified case.
2929 * Note that there may be more than one character in the result.
2930 * 's' is a pointer to the first byte of the input character
2931 * 'd' will be set to the first byte of the string of changed characters. It
2932 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2933 * 'lenp' will be set to the length in bytes of the string of changed characters
2935 * The functions return the ordinal of the first character in the string of
2937 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2938 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2939 Uppercase_Mapping_invmap, \
2940 UC_AUX_TABLE_ptrs, \
2941 UC_AUX_TABLE_lengths, \
2943 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2944 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2945 Titlecase_Mapping_invmap, \
2946 TC_AUX_TABLE_ptrs, \
2947 TC_AUX_TABLE_lengths, \
2949 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2950 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2951 Lowercase_Mapping_invmap, \
2952 LC_AUX_TABLE_ptrs, \
2953 LC_AUX_TABLE_lengths, \
2957 /* This additionally has the input parameter 'specials', which if non-zero will
2958 * cause this to use the specials hash for folding (meaning get full case
2959 * folding); otherwise, when zero, this implies a simple case fold */
2960 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2962 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2963 Case_Folding_invmap, \
2964 CF_AUX_TABLE_ptrs, \
2965 CF_AUX_TABLE_lengths, \
2967 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2968 Simple_Case_Folding_invmap, \
2973 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2975 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2976 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2977 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2978 * the changed version may be longer than the original character.
2980 * The ordinal of the first character of the changed version is returned
2981 * (but note, as explained above, that there may be more.) */
2983 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2986 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2989 return CALL_UPPER_CASE(c, NULL, p, lenp);
2993 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2995 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2998 return _to_upper_title_latin1((U8) c, p, lenp, 's');
3001 return CALL_TITLE_CASE(c, NULL, p, lenp);
3005 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
3007 /* We have the latin1-range values compiled into the core, so just use
3008 * those, converting the result to UTF-8. Since the result is always just
3009 * one character, we allow <p> to be NULL */
3011 U8 converted = toLOWER_LATIN1(c);
3013 PERL_UNUSED_ARG(dummy);
3016 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
3021 /* Result is known to always be < 256, so can use the EIGHT_BIT
3023 *p = UTF8_EIGHT_BIT_HI(converted);
3024 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3032 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3034 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3037 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3040 return CALL_LOWER_CASE(c, NULL, p, lenp);
3044 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3046 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3047 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3048 * FOLD_FLAGS_FULL iff full folding is to be used;
3050 * Not to be used for locale folds
3055 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3057 assert (! (flags & FOLD_FLAGS_LOCALE));
3059 if (UNLIKELY(c == MICRO_SIGN)) {
3060 converted = GREEK_SMALL_LETTER_MU;
3062 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3063 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3064 || UNICODE_DOT_DOT_VERSION > 0)
3065 else if ( (flags & FOLD_FLAGS_FULL)
3066 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3068 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3069 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3070 * under those circumstances. */
3071 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3072 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3073 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3075 return LATIN_SMALL_LETTER_LONG_S;
3085 else { /* In this range the fold of all other characters is their lower
3087 converted = toLOWER_LATIN1(c);
3090 if (UVCHR_IS_INVARIANT(converted)) {
3091 *p = (U8) converted;
3095 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3096 *p = UTF8_TWO_BYTE_LO(converted);
3104 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3107 /* Not currently externally documented, and subject to change
3108 * <flags> bits meanings:
3109 * FOLD_FLAGS_FULL iff full folding is to be used;
3110 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3111 * locale are to be used.
3112 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3115 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3117 if (flags & FOLD_FLAGS_LOCALE) {
3118 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3119 * except for potentially warning */
3120 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3121 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3122 flags &= ~FOLD_FLAGS_LOCALE;
3125 goto needs_full_generality;
3130 return _to_fold_latin1((U8) c, p, lenp,
3131 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3134 /* Here, above 255. If no special needs, just use the macro */
3135 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3136 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3138 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3139 the special flags. */
3140 U8 utf8_c[UTF8_MAXBYTES + 1];
3142 needs_full_generality:
3143 uvchr_to_utf8(utf8_c, c);
3144 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3149 PERL_STATIC_INLINE bool
3150 S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3153 /* returns a boolean giving whether or not the UTF8-encoded character that
3154 * starts at <p>, and extending no further than <e - 1> is in the inversion
3155 * list <invlist>. */
3157 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3159 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3161 if (cp == 0 && (p >= e || *p != '\0')) {
3162 _force_out_malformed_utf8_message(p, e, 0, 1);
3163 NOT_REACHED; /* NOTREACHED */
3167 return _invlist_contains_cp(invlist, cp);
3170 #if 0 /* Not currently used, but may be needed in the future */
3171 PERLVAR(I, seen_deprecated_macro, HV *)
3174 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3175 const char * const alternative,
3176 const bool use_locale,
3177 const char * const file,
3178 const unsigned line)
3182 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3184 if (ckWARN_d(WARN_DEPRECATED)) {
3186 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3187 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3188 if (! PL_seen_deprecated_macro) {
3189 PL_seen_deprecated_macro = newHV();
3191 if (! hv_store(PL_seen_deprecated_macro, key,
3192 strlen(key), &PL_sv_undef, 0))
3194 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3197 if (instr(file, "mathoms.c")) {
3198 Perl_warner(aTHX_ WARN_DEPRECATED,
3199 "In %s, line %d, starting in Perl v5.32, %s()"
3200 " will be removed. Avoid this message by"
3201 " converting to use %s().\n",
3202 file, line, name, alternative);
3205 Perl_warner(aTHX_ WARN_DEPRECATED,
3206 "In %s, line %d, starting in Perl v5.32, %s() will"
3207 " require an additional parameter. Avoid this"
3208 " message by converting to use %s().\n",
3209 file, line, name, alternative);
3217 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
3219 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3221 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
3225 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
3227 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
3229 return is_utf8_common(p, e, PL_utf8_perl_idstart);
3233 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
3235 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
3237 return is_utf8_common(p, e, PL_utf8_perl_idcont);
3241 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3242 U8* ustrp, STRLEN *lenp,
3243 SV *invlist, const I32 * const invmap,
3244 const U32 * const * const aux_tables,
3245 const U8 * const aux_table_lengths,
3246 const char * const normal)
3250 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3251 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3252 * to name the new case in any generated messages, as a fallback if the
3253 * operation being used is not available. The new case is given by the
3254 * data structures in the remaining arguments.
3256 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3257 * entire changed case string, and the return value is the first code point
3260 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3262 /* For code points that don't change case, we already know that the output
3263 * of this function is the unchanged input, so we can skip doing look-ups
3264 * for them. Unfortunately the case-changing code points are scattered
3265 * around. But there are some long consecutive ranges where there are no
3266 * case changing code points. By adding tests, we can eliminate the lookup
3267 * for all the ones in such ranges. This is currently done here only for
3268 * just a few cases where the scripts are in common use in modern commerce
3269 * (and scripts adjacent to those which can be included without additional
3272 if (uv1 >= 0x0590) {
3273 /* This keeps from needing further processing the code points most
3274 * likely to be used in the following non-cased scripts: Hebrew,
3275 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3276 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3277 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3282 /* The following largish code point ranges also don't have case
3283 * changes, but khw didn't think they warranted extra tests to speed
3284 * them up (which would slightly slow down everything else above them):
3285 * 1100..139F Hangul Jamo, Ethiopic
3286 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3287 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3288 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3289 * Combining Diacritical Marks Extended, Balinese,
3290 * Sundanese, Batak, Lepcha, Ol Chiki
3291 * 2000..206F General Punctuation
3294 if (uv1 >= 0x2D30) {
3296 /* This keeps the from needing further processing the code points
3297 * most likely to be used in the following non-cased major scripts:
3298 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3300 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3301 * event that Unicode eventually allocates the unused block as of
3302 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3303 * that the test suite will start having failures to alert you
3304 * should that happen) */
3309 if (uv1 >= 0xAC00) {
3310 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3311 if (ckWARN_d(WARN_SURROGATE)) {
3312 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3313 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3314 "Operation \"%s\" returns its argument for"
3315 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3320 /* AC00..FAFF Catches Hangul syllables and private use, plus
3326 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3327 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3328 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1));
3330 if (ckWARN_d(WARN_NON_UNICODE)) {
3331 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3332 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3333 "Operation \"%s\" returns its argument for"
3334 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3338 #ifdef HIGHEST_CASE_CHANGING_CP
3339 if (UNLIKELY(uv1 > HIGHEST_CASE_CHANGING_CP)) {
3347 /* Note that non-characters are perfectly legal, so no warning should
3353 const U32 * cp_list;
3356 /* 'index' is guaranteed to be non-negative, as this is an inversion
3357 * map that covers all possible inputs. See [perl #133365] */
3358 SSize_t index = _invlist_search(invlist, uv1);
3359 I32 base = invmap[index];
3361 /* The data structures are set up so that if 'base' is non-negative,
3362 * the case change is 1-to-1; and if 0, the change is to itself */
3370 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3371 lc = base + uv1 - invlist_array(invlist)[index];
3372 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3376 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3377 * requires an auxiliary table look up. abs(base) gives the index into
3378 * a list of such tables which points to the proper aux table. And a
3379 * parallel list gives the length of each corresponding aux table. */
3380 cp_list = aux_tables[-base];
3382 /* Create the string of UTF-8 from the mapped-to code points */
3384 for (i = 0; i < aux_table_lengths[-base]; i++) {
3385 d = uvchr_to_utf8(d, cp_list[i]);
3393 /* Here, there was no mapping defined, which means that the code point maps
3394 * to itself. Return the inputs */
3398 if (p != ustrp) { /* Don't copy onto itself */
3399 Copy(p, ustrp, len, U8);
3404 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3412 Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
3413 const U32 ** remaining_folds_to)
3415 /* Returns the count of the number of code points that fold to the input
3416 * 'cp' (besides itself).
3418 * If the return is 0, there is nothing else that folds to it, and
3419 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3421 * If the return is 1, '*first_folds_to' is set to the single code point,
3422 * and '*remaining_folds_to' is set to NULL.
3424 * Otherwise, '*first_folds_to' is set to a code point, and
3425 * '*remaining_fold_to' is set to an array that contains the others. The
3426 * length of this array is the returned count minus 1.
3428 * The reason for this convolution is to avoid having to deal with
3429 * allocating and freeing memory. The lists are already constructed, so
3430 * the return can point to them, but single code points aren't, so would
3431 * need to be constructed if we didn't employ something like this API
3433 * The code points returned by this function are all legal Unicode, which
3434 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3435 * constructed with this size (to save space and memory), and we return
3436 * pointers, so they must be this size */
3438 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3439 * that covers all possible inputs. See [perl #133365] */
3440 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3441 I32 base = _Perl_IVCF_invmap[index];
3443 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3445 if (base == 0) { /* No fold */
3446 *first_folds_to = 0;
3447 *remaining_folds_to = NULL;
3451 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3457 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3459 /* The data structure is set up so that the absolute value of 'base' is
3460 * an index into a table of pointers to arrays, with the array
3461 * corresponding to the index being the list of code points that fold
3462 * to 'cp', and the parallel array containing the length of the list
3464 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3465 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3466 /* +1 excludes first_folds_to */
3467 return IVCF_AUX_TABLE_lengths[-base];
3472 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3473 *first_folds_to = (U32) (base + cp
3474 - invlist_array(PL_utf8_foldclosures)[index]);
3475 *remaining_folds_to = NULL;
3480 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3481 U8* const ustrp, STRLEN *lenp)
3483 /* This is called when changing the case of a UTF-8-encoded character above
3484 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3485 * result contains a character that crosses the 255/256 boundary, disallow
3486 * the change, and return the original code point. See L<perlfunc/lc> for
3489 * p points to the original string whose case was changed; assumed
3490 * by this routine to be well-formed
3491 * result the code point of the first character in the changed-case string
3492 * ustrp points to the changed-case string (<result> represents its
3494 * lenp points to the length of <ustrp> */
3496 UV original; /* To store the first code point of <p> */
3498 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3500 assert(UTF8_IS_ABOVE_LATIN1(*p));
3502 /* We know immediately if the first character in the string crosses the
3503 * boundary, so can skip testing */
3506 /* Look at every character in the result; if any cross the
3507 * boundary, the whole thing is disallowed */
3508 U8* s = ustrp + UTF8SKIP(ustrp);
3509 U8* e = ustrp + *lenp;
3511 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3517 /* Here, no characters crossed, result is ok as-is, but we warn. */
3518 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3524 /* Failed, have to return the original */
3525 original = valid_utf8_to_uvchr(p, lenp);
3527 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3528 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3529 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3530 " locale; resolved to \"\\x{%" UVXf "}\".",
3534 Copy(p, ustrp, *lenp, char);
3539 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3540 U8 * ustrp, STRLEN *lenp)
3542 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3543 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3544 * Otherwise, it returns the first code point of the Turkic foldcased
3545 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3546 * contain *lenp bytes
3548 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3549 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3552 PERL_ARGS_ASSERT_TURKIC_FC;
3555 if (UNLIKELY(*p == 'I')) {
3557 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3558 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3559 return LATIN_SMALL_LETTER_DOTLESS_I;
3562 if (UNLIKELY(memBEGINs(p, e - p,
3563 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3574 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3575 U8 * ustrp, STRLEN *lenp)
3577 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3578 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3579 * Otherwise, it returns the first code point of the Turkic lowercased
3580 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3581 * contain *lenp bytes */
3583 PERL_ARGS_ASSERT_TURKIC_LC;
3586 /* A 'I' requires context as to what to do */
3587 if (UNLIKELY(*p0 == 'I')) {
3588 const U8 * p = p0 + 1;
3590 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3591 * modified by a dot above lowercases to 'i' even in turkic locales. */
3595 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3601 /* For the dot above to modify the 'I', it must be part of a
3602 * combining sequence immediately following the 'I', and no other
3603 * modifier with a ccc of 230 may intervene */
3604 cp = utf8_to_uvchr_buf(p, e, NULL);
3605 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3609 /* Here the combining sequence continues */
3614 /* In all other cases the lc is the same as the fold */
3615 return turkic_fc(p0, e, ustrp, lenp);
3619 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3620 U8 * ustrp, STRLEN *lenp)
3622 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3623 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3624 * Otherwise, it returns the first code point of the Turkic upper or
3625 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3626 * ustrp will contain *lenp bytes
3628 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3629 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3632 PERL_ARGS_ASSERT_TURKIC_UC;
3637 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3638 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3639 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3642 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3651 /* The process for changing the case is essentially the same for the four case
3652 * change types, except there are complications for folding. Otherwise the
3653 * difference is only which case to change to. To make sure that they all do
3654 * the same thing, the bodies of the functions are extracted out into the
3655 * following two macros. The functions are written with the same variable
3656 * names, and these are known and used inside these macros. It would be
3657 * better, of course, to have inline functions to do it, but since different
3658 * macros are called, depending on which case is being changed to, this is not
3659 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3660 * function can start with the common start macro, then finish with its special
3661 * handling; while the other three cases can just use the common end macro.
3663 * The algorithm is to use the proper (passed in) macro or function to change
3664 * the case for code points that are below 256. The macro is used if using
3665 * locale rules for the case change; the function if not. If the code point is
3666 * above 255, it is computed from the input UTF-8, and another macro is called
3667 * to do the conversion. If necessary, the output is converted to UTF-8. If
3668 * using a locale, we have to check that the change did not cross the 255/256
3669 * boundary, see check_locale_boundary_crossing() for further details.
3671 * The macros are split with the correct case change for the below-256 case
3672 * stored into 'result', and in the middle of an else clause for the above-255
3673 * case. At that point in the 'else', 'result' is not the final result, but is
3674 * the input code point calculated from the UTF-8. The fold code needs to
3675 * realize all this and take it from there.
3677 * To deal with Turkic locales, the function specified by the parameter
3678 * 'turkic' is called when appropriate.
3680 * If you read the two macros as sequential, it's easier to understand what's
3682 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3683 L1_func_extra_param, turkic) \
3685 if (flags & (locale_flags)) { \
3686 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3687 if (IN_UTF8_CTYPE_LOCALE) { \
3688 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3689 UV ret = turkic(p, e, ustrp, lenp); \
3690 if (ret) return ret; \
3693 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3695 flags &= ~(locale_flags); \
3699 if (UTF8_IS_INVARIANT(*p)) { \
3700 if (flags & (locale_flags)) { \
3701 result = LC_L1_change_macro(*p); \
3704 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3707 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3708 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3709 if (flags & (locale_flags)) { \
3710 result = LC_L1_change_macro(c); \
3713 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3716 else { /* malformed UTF-8 or ord above 255 */ \
3717 STRLEN len_result; \
3718 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3719 if (len_result == (STRLEN) -1) { \
3720 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
3723 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3724 result = change_macro(result, p, ustrp, lenp); \
3726 if (flags & (locale_flags)) { \
3727 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3732 /* Here, used locale rules. Convert back to UTF-8 */ \
3733 if (UTF8_IS_INVARIANT(result)) { \
3734 *ustrp = (U8) result; \
3738 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3739 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3745 /* Not currently externally documented, and subject to change:
3746 * <flags> is set iff the rules from the current underlying locale are to
3750 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3758 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3760 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3761 /* 2nd char of uc(U+DF) is 'S' */
3762 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3764 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3767 /* Not currently externally documented, and subject to change:
3768 * <flags> is set iff the rules from the current underlying locale are to be
3769 * used. Since titlecase is not defined in POSIX, for other than a
3770 * UTF-8 locale, uppercase is used instead for code points < 256.
3774 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3782 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3784 /* 2nd char of ucfirst(U+DF) is 's' */
3785 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
3787 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3790 /* Not currently externally documented, and subject to change:
3791 * <flags> is set iff the rules from the current underlying locale are to
3796 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3804 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3806 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
3808 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3811 /* Not currently externally documented, and subject to change,
3813 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3814 * locale are to be used.
3815 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3816 * otherwise simple folds
3817 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3822 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3830 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3832 /* These are mutually exclusive */
3833 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3835 assert(p != ustrp); /* Otherwise overwrites */
3837 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3838 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
3841 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3843 if (flags & FOLD_FLAGS_LOCALE) {
3845 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3846 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3847 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3849 /* Special case these two characters, as what normally gets
3850 * returned under locale doesn't work */
3851 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
3853 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3854 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3855 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3856 "resolved to \"\\x{17F}\\x{17F}\".");
3861 if (memBEGINs((char *) p, e - p, LONG_S_T))
3863 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3864 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3865 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3866 "resolved to \"\\x{FB06}\".");
3867 goto return_ligature_st;
3870 #if UNICODE_MAJOR_VERSION == 3 \
3871 && UNICODE_DOT_VERSION == 0 \
3872 && UNICODE_DOT_DOT_VERSION == 1
3873 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3875 /* And special case this on this Unicode version only, for the same
3876 * reaons the other two are special cased. They would cross the
3877 * 255/256 boundary which is forbidden under /l, and so the code
3878 * wouldn't catch that they are equivalent (which they are only in
3880 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
3881 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3882 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3883 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3884 "resolved to \"\\x{0131}\".");
3885 goto return_dotless_i;
3889 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3891 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3895 /* This is called when changing the case of a UTF-8-encoded
3896 * character above the ASCII range, and the result should not
3897 * contain an ASCII character. */
3899 UV original; /* To store the first code point of <p> */
3901 /* Look at every character in the result; if any cross the
3902 * boundary, the whole thing is disallowed */
3904 U8* send = ustrp + *lenp;
3907 /* Crossed, have to return the original */
3908 original = valid_utf8_to_uvchr(p, lenp);
3910 /* But in these instances, there is an alternative we can
3911 * return that is valid */
3912 if (original == LATIN_SMALL_LETTER_SHARP_S
3913 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3914 || original == LATIN_CAPITAL_LETTER_SHARP_S
3919 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3920 goto return_ligature_st;
3922 #if UNICODE_MAJOR_VERSION == 3 \
3923 && UNICODE_DOT_VERSION == 0 \
3924 && UNICODE_DOT_DOT_VERSION == 1
3926 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3927 goto return_dotless_i;
3930 Copy(p, ustrp, *lenp, char);
3936 /* Here, no characters crossed, result is ok as-is */
3941 /* Here, used locale rules. Convert back to UTF-8 */
3942 if (UTF8_IS_INVARIANT(result)) {
3943 *ustrp = (U8) result;
3947 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3948 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3955 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3956 * folds to a string of two of these characters. By returning this
3957 * instead, then, e.g.,
3958 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3961 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3962 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3964 return LATIN_SMALL_LETTER_LONG_S;
3967 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3968 * have the other one fold to it */
3970 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3971 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3972 return LATIN_SMALL_LIGATURE_ST;
3974 #if UNICODE_MAJOR_VERSION == 3 \
3975 && UNICODE_DOT_VERSION == 0 \
3976 && UNICODE_DOT_DOT_VERSION == 1
3979 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3980 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3981 return LATIN_SMALL_LETTER_DOTLESS_I;
3988 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3990 /* May change: warns if surrogates, non-character code points, or
3991 * non-Unicode code points are in 's' which has length 'len' bytes.
3992 * Returns TRUE if none found; FALSE otherwise. The only other validity
3993 * check is to make sure that this won't exceed the string's length nor
3996 const U8* const e = s + len;
3999 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4002 if (UTF8SKIP(s) > len) {
4003 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4004 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4007 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4008 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4009 if ( ckWARN_d(WARN_NON_UNICODE)
4010 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
4011 0 /* Don't consider overlongs */
4014 /* A side effect of this function will be to warn */
4015 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4019 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4020 if (ckWARN_d(WARN_SURROGATE)) {
4021 /* This has a different warning than the one the called
4022 * function would output, so can't just call it, unlike we
4023 * do for the non-chars and above-unicodes */
4024 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4025 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4026 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4031 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4032 && (ckWARN_d(WARN_NONCHAR)))
4034 /* A side effect of this function will be to warn */
4035 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4046 =for apidoc pv_uni_display
4048 Build to the scalar C<dsv> a displayable version of the UTF-8 encoded string
4049 C<spv>, length C<len>, the displayable version being at most C<pvlim> bytes
4050 long (if longer, the rest is truncated and C<"..."> will be appended).
4052 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4053 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4054 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4055 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4056 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4057 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4059 Additionally, there is now C<UNI_DISPLAY_BACKSPACE> which allows C<\b> for a
4060 backspace, but only when C<UNI_DISPLAY_BACKSLASH> also is set.
4062 The pointer to the PV of the C<dsv> is returned.
4064 See also L</sv_uni_display>.
4066 =for apidoc Amnh||UNI_DISPLAY_BACKSLASH
4067 =for apidoc Amnh||UNI_DISPLAY_BACKSPACE
4068 =for apidoc Amnh||UNI_DISPLAY_ISPRINT
4069 =for apidoc Amnh||UNI_DISPLAY_QQ
4070 =for apidoc Amnh||UNI_DISPLAY_REGEX
4074 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
4080 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4084 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4088 if (pvlim && SvCUR(dsv) >= pvlim) {
4092 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4094 const unsigned char c = (unsigned char)u & 0xFF;
4095 if (flags & UNI_DISPLAY_BACKSLASH) {
4096 if ( isMNEMONIC_CNTRL(c)
4098 || (flags & UNI_DISPLAY_BACKSPACE)))
4100 const char * mnemonic = cntrl_to_mnemonic(c);
4101 sv_catpvn(dsv, mnemonic, strlen(mnemonic));
4104 else if (c == '\\') {
4105 sv_catpvs(dsv, "\\\\");
4109 /* isPRINT() is the locale-blind version. */
4110 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4111 const char string = c;
4112 sv_catpvn(dsv, &string, 1);
4117 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
4120 sv_catpvs(dsv, "...");
4126 =for apidoc sv_uni_display
4128 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4129 the displayable version being at most C<pvlim> bytes long
4130 (if longer, the rest is truncated and "..." will be appended).
4132 The C<flags> argument is as in L</pv_uni_display>().
4134 The pointer to the PV of the C<dsv> is returned.
4139 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4141 const char * const ptr =
4142 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4144 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4146 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4147 SvCUR(ssv), pvlim, flags);
4151 =for apidoc foldEQ_utf8
4153 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
4154 both of which may be in UTF-8) are the same case-insensitively; false
4155 otherwise. How far into the strings to compare is determined by other input
4158 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4159 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
4160 C<u2> with respect to C<s2>.
4162 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
4163 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
4164 The scan will not be considered to be a match unless the goal is reached, and
4165 scanning won't continue past that goal. Correspondingly for C<l2> with respect
4168 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
4169 pointer is considered an end pointer to the position 1 byte past the maximum
4170 point in C<s1> beyond which scanning will not continue under any circumstances.
4171 (This routine assumes that UTF-8 encoded input strings are not malformed;
4172 malformed input can cause it to read past C<pe1>). This means that if both
4173 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
4174 will never be successful because it can never
4175 get as far as its goal (and in fact is asserted against). Correspondingly for
4176 C<pe2> with respect to C<s2>.
4178 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4179 C<l2> must be non-zero), and if both do, both have to be
4180 reached for a successful match. Also, if the fold of a character is multiple
4181 characters, all of them must be matched (see tr21 reference below for
4184 Upon a successful match, if C<pe1> is non-C<NULL>,
4185 it will be set to point to the beginning of the I<next> character of C<s1>
4186 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4188 For case-insensitiveness, the "casefolding" of Unicode is used
4189 instead of upper/lowercasing both the characters, see
4190 L<https://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4192 =for apidoc Cmnh||FOLDEQ_UTF8_NOMIX_ASCII
4193 =for apidoc Cmnh||FOLDEQ_LOCALE
4194 =for apidoc Cmnh||FOLDEQ_S1_ALREADY_FOLDED
4195 =for apidoc Cmnh||FOLDEQ_S1_FOLDS_SANE
4196 =for apidoc Cmnh||FOLDEQ_S2_ALREADY_FOLDED
4197 =for apidoc Cmnh||FOLDEQ_S2_FOLDS_SANE
4201 /* A flags parameter has been added which may change, and hence isn't
4202 * externally documented. Currently it is:
4203 * 0 for as-documented above
4204 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4205 ASCII one, to not match
4206 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4207 * locale are to be used.
4208 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4209 * routine. This allows that step to be skipped.
4210 * Currently, this requires s1 to be encoded as UTF-8
4211 * (u1 must be true), which is asserted for.
4212 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4213 * cross certain boundaries. Hence, the caller should
4214 * let this function do the folding instead of
4215 * pre-folding. This code contains an assertion to
4216 * that effect. However, if the caller knows what
4217 * it's doing, it can pass this flag to indicate that,
4218 * and the assertion is skipped.
4219 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
4220 * to s2, and s2 doesn't have to be UTF-8 encoded.
4221 * This introduces an asymmetry to save a few branches
4222 * in a loop. Currently, this is not a problem, as
4223 * never are both inputs pre-folded. Simply call this
4224 * function with the pre-folded one as the second
4226 * FOLDEQ_S2_FOLDS_SANE
4230 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
4231 const char *s2, char **pe2, UV l2, bool u2,
4234 const U8 *p1 = (const U8*)s1; /* Point to current char */
4235 const U8 *p2 = (const U8*)s2;
4236 const U8 *g1 = NULL; /* goal for s1 */
4237 const U8 *g2 = NULL;
4238 const U8 *e1 = NULL; /* Don't scan s1 past this */
4239 U8 *f1 = NULL; /* Point to current folded */
4240 const U8 *e2 = NULL;
4242 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4243 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4244 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4245 U8 flags_for_folder = FOLD_FLAGS_FULL;
4247 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4249 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4250 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
4251 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4252 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
4253 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4254 /* The algorithm is to trial the folds without regard to the flags on
4255 * the first line of the above assert(), and then see if the result
4256 * violates them. This means that the inputs can't be pre-folded to a
4257 * violating result, hence the assert. This could be changed, with the
4258 * addition of extra tests here for the already-folded case, which would
4259 * slow it down. That cost is more than any possible gain for when these
4260 * flags are specified, as the flags indicate /il or /iaa matching which
4261 * is less common than /iu, and I (khw) also believe that real-world /il
4262 * and /iaa matches are most likely to involve code points 0-255, and this
4263 * function only under rare conditions gets called for 0-255. */
4265 if (flags & FOLDEQ_LOCALE) {
4266 if (IN_UTF8_CTYPE_LOCALE) {
4267 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
4268 flags_for_folder |= FOLD_FLAGS_LOCALE;
4271 flags &= ~FOLDEQ_LOCALE;
4275 flags_for_folder |= FOLD_FLAGS_LOCALE;
4278 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
4279 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
4287 g1 = (const U8*)s1 + l1;
4295 g2 = (const U8*)s2 + l2;
4298 /* Must have at least one goal */
4303 /* Will never match if goal is out-of-bounds */
4304 assert(! e1 || e1 >= g1);
4306 /* Here, there isn't an end pointer, or it is beyond the goal. We
4307 * only go as far as the goal */
4311 assert(e1); /* Must have an end for looking at s1 */
4314 /* Same for goal for s2 */
4316 assert(! e2 || e2 >= g2);
4323 /* If both operands are already folded, we could just do a memEQ on the
4324 * whole strings at once, but it would be better if the caller realized
4325 * this and didn't even call us */
4327 /* Look through both strings, a character at a time */
4328 while (p1 < e1 && p2 < e2) {
4330 /* If at the beginning of a new character in s1, get its fold to use
4331 * and the length of the fold. */
4333 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4339 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4341 /* We have to forbid mixing ASCII with non-ASCII if the
4342 * flags so indicate. And, we can short circuit having to
4343 * call the general functions for this common ASCII case,
4344 * all of whose non-locale folds are also ASCII, and hence
4345 * UTF-8 invariants, so the UTF8ness of the strings is not
4347 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4351 *foldbuf1 = toFOLD(*p1);
4354 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
4356 else { /* Not UTF-8, get UTF-8 fold */
4357 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4363 if (n2 == 0) { /* Same for s2 */
4364 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4366 /* Point to the already-folded character. But for non-UTF-8
4367 * variants, convert to UTF-8 for the algorithm below */
4368 if (UTF8_IS_INVARIANT(*p2)) {
4377 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
4378 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
4384 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4385 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4389 *foldbuf2 = toFOLD(*p2);
4392 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
4395 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4401 /* Here f1 and f2 point to the beginning of the strings to compare.
4402 * These strings are the folds of the next character from each input
4403 * string, stored in UTF-8. */
4405 /* While there is more to look for in both folds, see if they
4406 * continue to match */
4408 U8 fold_length = UTF8SKIP(f1);
4409 if (fold_length != UTF8SKIP(f2)
4410 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4411 function call for single
4413 || memNE((char*)f1, (char*)f2, fold_length))
4415 return 0; /* mismatch */
4418 /* Here, they matched, advance past them */
4425 /* When reach the end of any fold, advance the input past it */
4427 p1 += u1 ? UTF8SKIP(p1) : 1;
4430 p2 += u2 ? UTF8SKIP(p2) : 1;
4432 } /* End of loop through both strings */
4434 /* A match is defined by each scan that specified an explicit length
4435 * reaching its final goal, and the other not having matched a partial
4436 * character (which can happen when the fold of a character is more than one
4438 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4442 /* Successful match. Set output pointers */
4453 * ex: set ts=8 sts=4 sw=4 et: