3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
40 /* Be sure to synchronize this message with the similar one in regcomp.c */
41 static const char cp_above_legal_max[] =
42 "Use of code point 0x%" UVXf " is not allowed; the"
43 " permissible max is 0x%" UVXf;
46 =head1 Unicode Support
47 These are various utility functions for manipulating UTF8-encoded
48 strings. For the uninitiated, this is a method of representing arbitrary
49 Unicode characters as a variable number of bytes, in such a way that
50 characters in the ASCII range are unmodified, and a zero byte never appears
51 within non-zero characters.
56 /* helper for Perl__force_out_malformed_utf8_message(). Like
57 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
61 S_restore_cop_warnings(pTHX_ void *p)
63 if (!specialWARN(PL_curcop->cop_warnings))
64 PerlMemShared_free(PL_curcop->cop_warnings);
65 PL_curcop->cop_warnings = (STRLEN*)p;
70 Perl__force_out_malformed_utf8_message(pTHX_
71 const U8 *const p, /* First byte in UTF-8 sequence */
72 const U8 * const e, /* Final byte in sequence (may include
74 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
75 usually 0, or some DISALLOW flags */
76 const bool die_here) /* If TRUE, this function does not return */
78 /* This core-only function is to be called when a malformed UTF-8 character
79 * is found, in order to output the detailed information about the
80 * malformation before dieing. The reason it exists is for the occasions
81 * when such a malformation is fatal, but warnings might be turned off, so
82 * that normally they would not be actually output. This ensures that they
83 * do get output. Because a sequence may be malformed in more than one
84 * way, multiple messages may be generated, so we can't make them fatal, as
85 * that would cause the first one to die.
87 * Instead we pretend -W was passed to perl, then die afterwards. The
88 * flexibility is here to return to the caller so they can finish up and
92 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
98 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
100 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
101 * than PL_compiling */
102 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
103 (void*)PL_curcop->cop_warnings);
104 PL_curcop->cop_warnings = pWARN_ALL;
107 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
112 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
113 " be called only when there are errors found");
117 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
122 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
123 U32 categories, /* Packed warning categories */
124 U32 flag) /* Flag associated with this message */
126 /* Creates, populates, and returns an HV* that describes an error message
127 * for the translators between UTF8 and code point */
129 SV* msg_sv = newSVpv(message, 0);
130 SV* category_sv = newSVuv(categories);
131 SV* flag_bit_sv = newSVuv(flag);
133 HV* msg_hv = newHV();
135 PERL_ARGS_ASSERT_NEW_MSG_HV;
137 (void) hv_stores(msg_hv, "text", msg_sv);
138 (void) hv_stores(msg_hv, "warn_categories", category_sv);
139 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
145 =for apidoc uvoffuni_to_utf8_flags
147 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
148 Instead, B<Almost all code should use L</uvchr_to_utf8> or
149 L</uvchr_to_utf8_flags>>.
151 This function is like them, but the input is a strict Unicode
152 (as opposed to native) code point. Only in very rare circumstances should code
153 not be using the native code point.
155 For details, see the description for L</uvchr_to_utf8_flags>.
161 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
163 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
165 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
168 /* All these formats take a single UV code point argument */
169 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
170 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
171 " is not recommended for open interchange";
172 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
173 " may not be portable";
174 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
175 " Unicode, requires a Perl extension," \
176 " and so is not portable";
178 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
180 if (flags & UNICODE_WARN_SURROGATE) { \
181 U32 category = packWARN(WARN_SURROGATE); \
182 const char * format = surrogate_cp_format; \
184 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
186 UNICODE_GOT_SURROGATE); \
189 Perl_ck_warner_d(aTHX_ category, format, uv); \
192 if (flags & UNICODE_DISALLOW_SURROGATE) { \
197 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
199 if (flags & UNICODE_WARN_NONCHAR) { \
200 U32 category = packWARN(WARN_NONCHAR); \
201 const char * format = nonchar_cp_format; \
203 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
205 UNICODE_GOT_NONCHAR); \
208 Perl_ck_warner_d(aTHX_ category, format, uv); \
211 if (flags & UNICODE_DISALLOW_NONCHAR) { \
216 /* Use shorter names internally in this file */
217 #define SHIFT UTF_ACCUMULATION_SHIFT
219 #define MARK UTF_CONTINUATION_MARK
220 #define MASK UTF_CONTINUATION_MASK
223 =for apidoc uvchr_to_utf8_flags_msgs
225 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
227 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
229 This function is for code that wants any warning and/or error messages to be
230 returned to the caller rather than be displayed. All messages that would have
231 been displayed if all lexical warnings are enabled will be returned.
233 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
234 placed after all the others, C<msgs>. If this parameter is 0, this function
235 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
236 be a pointer to an C<HV *> variable, in which this function creates a new HV to
237 contain any appropriate messages. The hash has three key-value pairs, as
244 The text of the message as a C<SVpv>.
246 =item C<warn_categories>
248 The warning category (or categories) packed into a C<SVuv>.
252 A single flag bit associated with this message, in a C<SVuv>.
253 The bit corresponds to some bit in the C<*errors> return value,
254 such as C<UNICODE_GOT_SURROGATE>.
258 It's important to note that specifying this parameter as non-null will cause
259 any warnings this function would otherwise generate to be suppressed, and
260 instead be placed in C<*msgs>. The caller can check the lexical warnings state
261 (or not) when choosing what to do with the returned messages.
263 The caller, of course, is responsible for freeing any returned HV.
268 /* Undocumented; we don't want people using this. Instead they should use
269 * uvchr_to_utf8_flags_msgs() */
271 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
273 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
279 if (OFFUNI_IS_INVARIANT(uv)) {
280 *d++ = LATIN1_TO_NATIVE(uv);
284 if (uv <= MAX_UTF8_TWO_BYTE) {
285 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
286 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
290 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
291 * below, the 16 is for start bytes E0-EF (which are all the possible ones
292 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
293 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
294 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
295 * 0x800-0xFFFF on ASCII */
296 if (uv < (16 * (1U << (2 * SHIFT)))) {
297 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
298 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
299 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
301 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
302 aren't tested here */
303 /* The most likely code points in this range are below the surrogates.
304 * Do an extra test to quickly exclude those. */
305 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
306 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
307 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
309 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
311 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
312 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
319 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
320 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
321 * happen starting with 4-byte characters on ASCII platforms. We unify the
322 * code for these with EBCDIC, even though some of them require 5-bytes on
323 * those, because khw believes the code saving is worth the very slight
324 * performance hit on these high EBCDIC code points. */
326 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
327 if (UNLIKELY(uv > MAX_LEGAL_CP)) {
328 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_LEGAL_CP);
330 if ( (flags & UNICODE_WARN_SUPER)
331 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
332 && UNICODE_IS_PERL_EXTENDED(uv)))
334 const char * format = super_cp_format;
335 U32 category = packWARN(WARN_NON_UNICODE);
336 U32 flag = UNICODE_GOT_SUPER;
338 /* Choose the more dire applicable warning */
339 if (UNICODE_IS_PERL_EXTENDED(uv)) {
340 format = perl_extended_cp_format;
341 if (flags & (UNICODE_WARN_PERL_EXTENDED
342 |UNICODE_DISALLOW_PERL_EXTENDED))
344 flag = UNICODE_GOT_PERL_EXTENDED;
349 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
353 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
356 if ( (flags & UNICODE_DISALLOW_SUPER)
357 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
358 && UNICODE_IS_PERL_EXTENDED(uv)))
363 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
364 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
367 /* Test for and handle 4-byte result. In the test immediately below, the
368 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
369 * characters). The 3 is for 3 continuation bytes; these each contribute
370 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
371 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
372 * 0x1_0000-0x1F_FFFF on ASCII */
373 if (uv < (8 * (1U << (3 * SHIFT)))) {
374 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
375 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
376 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
377 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
379 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
380 characters. The end-plane non-characters for EBCDIC were
381 handled just above */
382 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
383 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
385 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
386 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
393 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
394 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
395 * format. The unrolled version above turns out to not save all that much
396 * time, and at these high code points (well above the legal Unicode range
397 * on ASCII platforms, and well above anything in common use in EBCDIC),
398 * khw believes that less code outweighs slight performance gains. */
401 STRLEN len = OFFUNISKIP(uv);
404 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
407 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
413 =for apidoc uvchr_to_utf8
415 Adds the UTF-8 representation of the native code point C<uv> to the end
416 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
417 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
418 the byte after the end of the new character. In other words,
420 d = uvchr_to_utf8(d, uv);
422 is the recommended wide native character-aware way of saying
426 This function accepts any code point from 0..C<IV_MAX> as input.
427 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
429 It is possible to forbid or warn on non-Unicode code points, or those that may
430 be problematic by using L</uvchr_to_utf8_flags>.
435 /* This is also a macro */
436 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
439 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
441 return uvchr_to_utf8(d, uv);
445 =for apidoc uvchr_to_utf8_flags
447 Adds the UTF-8 representation of the native code point C<uv> to the end
448 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
449 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
450 the byte after the end of the new character. In other words,
452 d = uvchr_to_utf8_flags(d, uv, flags);
456 d = uvchr_to_utf8_flags(d, uv, 0);
458 This is the Unicode-aware way of saying
462 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
463 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
465 Specifying C<flags> can further restrict what is allowed and not warned on, as
468 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
469 the function will raise a warning, provided UTF8 warnings are enabled. If
470 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
471 NULL. If both flags are set, the function will both warn and return NULL.
473 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
474 affect how the function handles a Unicode non-character.
476 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
477 affect the handling of code points that are above the Unicode maximum of
478 0x10FFFF. Languages other than Perl may not be able to accept files that
481 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
482 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
483 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
484 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
485 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
486 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
487 above-Unicode and surrogate flags, but not the non-character ones, as
489 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
490 See L<perlunicode/Noncharacter code points>.
492 Extremely high code points were never specified in any standard, and require an
493 extension to UTF-8 to express, which Perl does. It is likely that programs
494 written in something other than Perl would not be able to read files that
495 contain these; nor would Perl understand files written by something that uses a
496 different extension. For these reasons, there is a separate set of flags that
497 can warn and/or disallow these extremely high code points, even if other
498 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
499 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
500 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
501 treat all above-Unicode code points, including these, as malformations. (Note
502 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
503 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
505 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
506 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
507 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
508 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
509 platforms,these flags can apply to code points that actually do fit in 31 bits.
510 The new names accurately describe the situation in all cases.
515 /* This is also a macro */
516 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
519 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
521 return uvchr_to_utf8_flags(d, uv, flags);
527 S_is_utf8_cp_above_31_bits(const U8 * const s,
529 const bool consider_overlongs)
531 /* Returns TRUE if the first code point represented by the Perl-extended-
532 * UTF-8-encoded string starting at 's', and looking no further than 'e -
533 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
535 * The function handles the case where the input bytes do not include all
536 * the ones necessary to represent a full character. That is, they may be
537 * the intial bytes of the representation of a code point, but possibly
538 * the final ones necessary for the complete representation may be beyond
541 * The function also can handle the case where the input is an overlong
542 * sequence. If 'consider_overlongs' is 0, the function assumes the
543 * input is not overlong, without checking, and will return based on that
544 * assumption. If this parameter is 1, the function will go to the trouble
545 * of figuring out if it actually evaluates to above or below 31 bits.
547 * The sequence is otherwise assumed to be well-formed, without checking.
550 const STRLEN len = e - s;
553 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
555 assert(! UTF8_IS_INVARIANT(*s) && e > s);
559 PERL_UNUSED_ARG(consider_overlongs);
561 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
562 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
563 * also be the start byte for a 31-bit code point; we need at least 2
564 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
565 * the start byte for an overlong sequence, but for 30-bit or smaller code
566 * points, so we don't have to worry about overlongs on EBCDIC.) */
577 /* On ASCII, FE and FF are the only start bytes that can evaluate to
578 * needing more than 31 bits. */
579 if (LIKELY(*s < 0xFE)) {
583 /* What we have left are FE and FF. Both of these require more than 31
584 * bits unless they are for overlongs. */
585 if (! consider_overlongs) {
589 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
590 * above 31 bits. But we need more than one byte to discern this, so if
591 * passed just the start byte, it could be an overlong evaluating to
597 /* Having excluded len==1, and knowing that FE and FF are both valid start
598 * bytes, we can call the function below to see if the sequence is
599 * overlong. (We don't need the full generality of the called function,
600 * but for these huge code points, speed shouldn't be a consideration, and
601 * the compiler does have enough information, since it's static to this
602 * file, to optimize to just the needed parts.) */
603 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
605 /* If it isn't overlong, more than 31 bits are required. */
606 if (is_overlong == 0) {
610 /* If it is indeterminate if it is overlong, return that */
611 if (is_overlong < 0) {
615 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
616 * the max it can be is 2**31 - 1 */
623 /* Here, ASCII and EBCDIC rejoin:
624 * On ASCII: We have an overlong sequence starting with FF
625 * On EBCDIC: We have a sequence starting with FE. */
627 { /* For C89, use a block so the declaration can be close to its use */
631 /* U+7FFFFFFF (2 ** 31 - 1)
632 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
633 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
634 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
635 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
636 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
637 * U+80000000 (2 ** 31):
638 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
639 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
640 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
641 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
643 * and since we know that *s = \xfe, any continuation sequcence
644 * following it that is gt the below is above 31 bits
645 [0] [1] [2] [3] [4] [5] [6] */
646 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
650 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
651 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
652 * FF overlong for U+80000000 (2 ** 31):
653 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
654 * and since we know that *s = \xff, any continuation sequcence
655 * following it that is gt the below is above 30 bits
656 [0] [1] [2] [3] [4] [5] [6] */
657 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
661 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
662 const STRLEN cmp_len = MIN(conts_len, len - 1);
664 /* Now compare the continuation bytes in s with the ones we have
665 * compiled in that are for the largest 30 bit code point. If we have
666 * enough bytes available to determine the answer, or the bytes we do
667 * have differ from them, we can compare the two to get a definitive
668 * answer (Note that in UTF-EBCDIC, the two lowest possible
669 * continuation bytes are \x41 and \x42.) */
670 if (cmp_len >= conts_len || memNE(s + 1,
671 conts_for_highest_30_bit,
674 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
677 /* Here, all the bytes we have are the same as the highest 30-bit code
678 * point, but we are missing so many bytes that we can't make the
686 PERL_STATIC_INLINE int
687 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
689 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
690 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
691 * it isn't, and -1 if there isn't enough information to tell. This last
692 * return value can happen if the sequence is incomplete, missing some
693 * trailing bytes that would form a complete character. If there are
694 * enough bytes to make a definitive decision, this function does so.
695 * Usually 2 bytes sufficient.
697 * Overlongs can occur whenever the number of continuation bytes changes.
698 * That means whenever the number of leading 1 bits in a start byte
699 * increases from the next lower start byte. That happens for start bytes
700 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
701 * start bytes have already been excluded, so don't need to be tested here;
702 * ASCII platforms: C0, C1
703 * EBCDIC platforms C0, C1, C2, C3, C4, E0
706 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
707 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
709 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
710 assert(len > 1 && UTF8_IS_START(*s));
712 /* Each platform has overlongs after the start bytes given above (expressed
713 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
714 * the logic is the same, except the E0 overlong has already been excluded
715 * on EBCDIC platforms. The values below were found by manually
716 * inspecting the UTF-8 patterns. See the tables in utf8.h and
720 # define F0_ABOVE_OVERLONG 0xB0
721 # define F8_ABOVE_OVERLONG 0xA8
722 # define FC_ABOVE_OVERLONG 0xA4
723 # define FE_ABOVE_OVERLONG 0xA2
724 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
728 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
732 # define F0_ABOVE_OVERLONG 0x90
733 # define F8_ABOVE_OVERLONG 0x88
734 # define FC_ABOVE_OVERLONG 0x84
735 # define FE_ABOVE_OVERLONG 0x82
736 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
740 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
741 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
742 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
743 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
748 /* Check for the FF overlong */
749 return isFF_OVERLONG(s, len);
752 PERL_STATIC_INLINE int
753 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
755 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
756 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
757 * it isn't, and -1 if there isn't enough information to tell. This last
758 * return value can happen if the sequence is incomplete, missing some
759 * trailing bytes that would form a complete character. If there are
760 * enough bytes to make a definitive decision, this function does so. */
762 PERL_ARGS_ASSERT_ISFF_OVERLONG;
764 /* To be an FF overlong, all the available bytes must match */
765 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
766 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
771 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
772 * be there; what comes after them doesn't matter. See tables in utf8.h,
774 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
778 /* The missing bytes could cause the result to go one way or the other, so
779 * the result is indeterminate */
783 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
784 # ifdef EBCDIC /* Actually is I8 */
785 # define HIGHEST_REPRESENTABLE_UTF8 \
786 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
788 # define HIGHEST_REPRESENTABLE_UTF8 \
789 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
793 PERL_STATIC_INLINE int
794 S_does_utf8_overflow(const U8 * const s,
796 const bool consider_overlongs)
798 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
799 * 'e' - 1 would overflow an IV on this platform; that is if it represents
800 * a code point larger than the highest representable code point. It
801 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
802 * enough information to tell. This last return value can happen if the
803 * sequence is incomplete, missing some trailing bytes that would form a
804 * complete character. If there are enough bytes to make a definitive
805 * decision, this function does so.
807 * If 'consider_overlongs' is TRUE, the function checks for the possibility
808 * that the sequence is an overlong that doesn't overflow. Otherwise, it
809 * assumes the sequence is not an overlong. This can give different
810 * results only on ASCII 32-bit platforms.
812 * (For ASCII platforms, we could use memcmp() because we don't have to
813 * convert each byte to I8, but it's very rare input indeed that would
814 * approach overflow, so the loop below will likely only get executed once.)
816 * 'e' - 1 must not be beyond a full character. */
819 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
820 assert(s <= e && s + UTF8SKIP(s) >= e);
822 #if ! defined(UV_IS_QUAD)
824 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
828 PERL_UNUSED_ARG(consider_overlongs);
831 const STRLEN len = e - s;
833 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
835 for (x = s; x < e; x++, y++) {
837 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
841 /* If this byte is larger than the corresponding highest UTF-8
842 * byte, the sequence overflow; otherwise the byte is less than,
843 * and so the sequence doesn't overflow */
844 return NATIVE_UTF8_TO_I8(*x) > *y;
848 /* Got to the end and all bytes are the same. If the input is a whole
849 * character, it doesn't overflow. And if it is a partial character,
850 * there's not enough information to tell */
851 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
864 /* This is the portions of the above function that deal with UV_MAX instead of
865 * IV_MAX. They are left here in case we want to combine them so that internal
866 * uses can have larger code points. The only logic difference is that the
867 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
871 /* Anything larger than this will overflow the word if it were converted into a UV */
872 #if defined(UV_IS_QUAD)
873 # ifdef EBCDIC /* Actually is I8 */
874 # define HIGHEST_REPRESENTABLE_UTF8 \
875 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
877 # define HIGHEST_REPRESENTABLE_UTF8 \
878 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
882 # define HIGHEST_REPRESENTABLE_UTF8 \
883 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
885 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
889 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
891 /* On 32 bit ASCII machines, many overlongs that start with FF don't
893 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
895 /* To be such an overlong, the first bytes of 's' must match
896 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
897 * don't have any additional bytes available, the sequence, when
898 * completed might or might not fit in 32 bits. But if we have that
899 * next byte, we can tell for sure. If it is <= 0x83, then it does
901 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
905 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
908 /* Starting with the #else, the rest of the function is identical except
909 * 1. we need to move the 'len' declaration to be global to the function
910 * 2. the endif move to just after the UNUSED_ARG.
911 * An empty endif is given just below to satisfy the preprocessor
917 #undef F0_ABOVE_OVERLONG
918 #undef F8_ABOVE_OVERLONG
919 #undef FC_ABOVE_OVERLONG
920 #undef FE_ABOVE_OVERLONG
921 #undef FF_OVERLONG_PREFIX
924 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
929 /* A helper function that should not be called directly.
931 * This function returns non-zero if the string beginning at 's' and
932 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
933 * code point; otherwise it returns 0. The examination stops after the
934 * first code point in 's' is validated, not looking at the rest of the
935 * input. If 'e' is such that there are not enough bytes to represent a
936 * complete code point, this function will return non-zero anyway, if the
937 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
938 * excluded by 'flags'.
940 * A non-zero return gives the number of bytes required to represent the
941 * code point. Be aware that if the input is for a partial character, the
942 * return will be larger than 'e - s'.
944 * This function assumes that the code point represented is UTF-8 variant.
945 * The caller should have excluded the possibility of it being invariant
946 * before calling this function.
948 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
949 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
950 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
951 * disallowed by the flags. If the input is only for a partial character,
952 * the function will return non-zero if there is any sequence of
953 * well-formed UTF-8 that, when appended to the input sequence, could
954 * result in an allowed code point; otherwise it returns 0. Non characters
955 * cannot be determined based on partial character input. But many of the
956 * other excluded types can be determined with just the first one or two
961 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
963 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
964 |UTF8_DISALLOW_PERL_EXTENDED)));
965 assert(! UTF8_IS_INVARIANT(*s));
967 /* A variant char must begin with a start byte */
968 if (UNLIKELY(! UTF8_IS_START(*s))) {
972 /* Examine a maximum of a single whole code point */
973 if (e - s > UTF8SKIP(s)) {
979 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
980 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
982 /* Here, we are disallowing some set of largish code points, and the
983 * first byte indicates the sequence is for a code point that could be
984 * in the excluded set. We generally don't have to look beyond this or
985 * the second byte to see if the sequence is actually for one of the
986 * excluded classes. The code below is derived from this table:
988 * UTF-8 UTF-EBCDIC I8
989 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
990 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
991 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
993 * Keep in mind that legal continuation bytes range between \x80..\xBF
994 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
995 * continuation bytes. Hence, we don't have to test the upper edge
996 * because if any of those is encountered, the sequence is malformed,
997 * and would fail elsewhere in this function.
999 * The code here likewise assumes that there aren't other
1000 * malformations; again the function should fail elsewhere because of
1001 * these. For example, an overlong beginning with FC doesn't actually
1002 * have to be a super; it could actually represent a small code point,
1003 * even U+0000. But, since overlongs (and other malformations) are
1004 * illegal, the function should return FALSE in either case.
1007 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1008 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
1009 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
1011 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1013 && ((s1) & 0xFE ) == 0xB6)
1014 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1016 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1017 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1018 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1019 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1022 if ( (flags & UTF8_DISALLOW_SUPER)
1023 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1025 return 0; /* Above Unicode */
1028 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1029 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1035 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1037 if ( (flags & UTF8_DISALLOW_SUPER)
1038 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1040 return 0; /* Above Unicode */
1043 if ( (flags & UTF8_DISALLOW_SURROGATE)
1044 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1046 return 0; /* Surrogate */
1049 if ( (flags & UTF8_DISALLOW_NONCHAR)
1050 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1052 return 0; /* Noncharacter code point */
1057 /* Make sure that all that follows are continuation bytes */
1058 for (x = s + 1; x < e; x++) {
1059 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1064 /* Here is syntactically valid. Next, make sure this isn't the start of an
1066 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1070 /* And finally, that the code point represented fits in a word on this
1072 if (0 < does_utf8_overflow(s, e,
1073 0 /* Don't consider overlongs */
1083 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1085 /* Returns a mortalized C string that is a displayable copy of the 'len'
1086 * bytes starting at 'start'. 'format' gives how to display each byte.
1087 * Currently, there are only two formats, so it is currently a bool:
1089 * 1 ab (that is a space between two hex digit bytes)
1092 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1094 const U8 * s = start;
1095 const U8 * const e = start + len;
1099 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1101 Newx(output, output_len, char);
1105 for (s = start; s < e; s++) {
1106 const unsigned high_nibble = (*s & 0xF0) >> 4;
1107 const unsigned low_nibble = (*s & 0x0F);
1119 if (high_nibble < 10) {
1120 *d++ = high_nibble + '0';
1123 *d++ = high_nibble - 10 + 'a';
1126 if (low_nibble < 10) {
1127 *d++ = low_nibble + '0';
1130 *d++ = low_nibble - 10 + 'a';
1138 PERL_STATIC_INLINE char *
1139 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1141 /* Max number of bytes to print */
1144 /* Which one is the non-continuation */
1145 const STRLEN non_cont_byte_pos,
1147 /* How many bytes should there be? */
1148 const STRLEN expect_len)
1150 /* Return the malformation warning text for an unexpected continuation
1153 const char * const where = (non_cont_byte_pos == 1)
1155 : Perl_form(aTHX_ "%d bytes",
1156 (int) non_cont_byte_pos);
1157 const U8 * x = s + non_cont_byte_pos;
1158 const U8 * e = s + print_len;
1160 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1162 /* We don't need to pass this parameter, but since it has already been
1163 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1164 assert(expect_len == UTF8SKIP(s));
1166 /* As a defensive coding measure, don't output anything past a NUL. Such
1167 * bytes shouldn't be in the middle of a malformation, and could mark the
1168 * end of the allocated string, and what comes after is undefined */
1169 for (; x < e; x++) {
1171 x++; /* Output this particular NUL */
1176 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1177 " %s after start byte 0x%02x; need %d bytes, got %d)",
1179 _byte_dump_string(s, x - s, 0),
1180 *(s + non_cont_byte_pos),
1184 (int) non_cont_byte_pos);
1189 =for apidoc utf8n_to_uvchr
1191 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1192 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1194 Bottom level UTF-8 decode routine.
1195 Returns the native code point value of the first character in the string C<s>,
1196 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1197 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1198 the length, in bytes, of that character.
1200 The value of C<flags> determines the behavior when C<s> does not point to a
1201 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1202 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1203 is the next possible position in C<s> that could begin a non-malformed
1204 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1205 is raised. Some UTF-8 input sequences may contain multiple malformations.
1206 This function tries to find every possible one in each call, so multiple
1207 warnings can be raised for the same sequence.
1209 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1210 individual types of malformations, such as the sequence being overlong (that
1211 is, when there is a shorter sequence that can express the same code point;
1212 overlong sequences are expressly forbidden in the UTF-8 standard due to
1213 potential security issues). Another malformation example is the first byte of
1214 a character not being a legal first byte. See F<utf8.h> for the list of such
1215 flags. Even if allowed, this function generally returns the Unicode
1216 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1217 F<utf8.h> to override this behavior for the overlong malformations, but don't
1218 do that except for very specialized purposes.
1220 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1221 flags) malformation is found. If this flag is set, the routine assumes that
1222 the caller will raise a warning, and this function will silently just set
1223 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1225 Note that this API requires disambiguation between successful decoding a C<NUL>
1226 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1227 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1228 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1229 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1230 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1232 Certain code points are considered problematic. These are Unicode surrogates,
1233 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1234 By default these are considered regular code points, but certain situations
1235 warrant special handling for them, which can be specified using the C<flags>
1236 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1237 three classes are treated as malformations and handled as such. The flags
1238 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1239 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1240 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1241 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1242 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1244 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1245 The difference between traditional strictness and C9 strictness is that the
1246 latter does not forbid non-character code points. (They are still discouraged,
1247 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1249 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1250 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1251 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1252 raised for their respective categories, but otherwise the code points are
1253 considered valid (not malformations). To get a category to both be treated as
1254 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1255 (But note that warnings are not raised if lexically disabled nor if
1256 C<UTF8_CHECK_ONLY> is also specified.)
1258 Extremely high code points were never specified in any standard, and require an
1259 extension to UTF-8 to express, which Perl does. It is likely that programs
1260 written in something other than Perl would not be able to read files that
1261 contain these; nor would Perl understand files written by something that uses a
1262 different extension. For these reasons, there is a separate set of flags that
1263 can warn and/or disallow these extremely high code points, even if other
1264 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1265 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1266 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1267 above-Unicode code points, including these, as malformations.
1268 (Note that the Unicode standard considers anything above 0x10FFFF to be
1269 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1272 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1273 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1274 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1275 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1276 can apply to code points that actually do fit in 31 bits. This happens on
1277 EBCDIC platforms, and sometimes when the L<overlong
1278 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1279 describe the situation in all cases.
1282 All other code points corresponding to Unicode characters, including private
1283 use and those yet to be assigned, are never considered malformed and never
1288 Also implemented as a macro in utf8.h
1292 Perl_utf8n_to_uvchr(const U8 *s,
1297 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1299 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1304 =for apidoc utf8n_to_uvchr_error
1306 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1307 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1309 This function is for code that needs to know what the precise malformation(s)
1310 are when an error is found. If you also need to know the generated warning
1311 messages, use L</utf8n_to_uvchr_msgs>() instead.
1313 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1314 all the others, C<errors>. If this parameter is 0, this function behaves
1315 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1316 to a C<U32> variable, which this function sets to indicate any errors found.
1317 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1318 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1319 of these bits will be set if a malformation is found, even if the input
1320 C<flags> parameter indicates that the given malformation is allowed; those
1321 exceptions are noted:
1325 =item C<UTF8_GOT_PERL_EXTENDED>
1327 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1328 set only if the input C<flags> parameter contains either the
1329 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1331 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1332 and so some extension must be used to express them. Perl uses a natural
1333 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1334 extension to represent even higher ones, so that any code point that fits in a
1335 64-bit word can be represented. Text using these extensions is not likely to
1336 be portable to non-Perl code. We lump both of these extensions together and
1337 refer to them as Perl extended UTF-8. There exist other extensions that people
1338 have invented, incompatible with Perl's.
1340 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1341 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1342 than on ASCII. Prior to that, code points 2**31 and higher were simply
1343 unrepresentable, and a different, incompatible method was used to represent
1344 code points between 2**30 and 2**31 - 1.
1346 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1347 Perl extended UTF-8 is used.
1349 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1350 may use for backward compatibility. That name is misleading, as this flag may
1351 be set when the code point actually does fit in 31 bits. This happens on
1352 EBCDIC platforms, and sometimes when the L<overlong
1353 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1354 describes the situation in all cases.
1356 =item C<UTF8_GOT_CONTINUATION>
1358 The input sequence was malformed in that the first byte was a a UTF-8
1361 =item C<UTF8_GOT_EMPTY>
1363 The input C<curlen> parameter was 0.
1365 =item C<UTF8_GOT_LONG>
1367 The input sequence was malformed in that there is some other sequence that
1368 evaluates to the same code point, but that sequence is shorter than this one.
1370 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1371 it was discovered that this created security issues.
1373 =item C<UTF8_GOT_NONCHAR>
1375 The code point represented by the input UTF-8 sequence is for a Unicode
1376 non-character code point.
1377 This bit is set only if the input C<flags> parameter contains either the
1378 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1380 =item C<UTF8_GOT_NON_CONTINUATION>
1382 The input sequence was malformed in that a non-continuation type byte was found
1383 in a position where only a continuation type one should be. See also
1384 L</C<UTF8_GOT_SHORT>>.
1386 =item C<UTF8_GOT_OVERFLOW>
1388 The input sequence was malformed in that it is for a code point that is not
1389 representable in the number of bits available in an IV on the current platform.
1391 =item C<UTF8_GOT_SHORT>
1393 The input sequence was malformed in that C<curlen> is smaller than required for
1394 a complete sequence. In other words, the input is for a partial character
1398 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1399 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1400 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1401 sequence was looked at. If no other flags are present, it means that the
1402 sequence was valid as far as it went. Depending on the application, this could
1403 mean one of three things:
1409 The C<curlen> length parameter passed in was too small, and the function was
1410 prevented from examining all the necessary bytes.
1414 The buffer being looked at is based on reading data, and the data received so
1415 far stopped in the middle of a character, so that the next read will
1416 read the remainder of this character. (It is up to the caller to deal with the
1417 split bytes somehow.)
1421 This is a real error, and the partial sequence is all we're going to get.
1425 =item C<UTF8_GOT_SUPER>
1427 The input sequence was malformed in that it is for a non-Unicode code point;
1428 that is, one above the legal Unicode maximum.
1429 This bit is set only if the input C<flags> parameter contains either the
1430 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1432 =item C<UTF8_GOT_SURROGATE>
1434 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1436 This bit is set only if the input C<flags> parameter contains either the
1437 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1441 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1442 flag to suppress any warnings, and then examine the C<*errors> return.
1446 Also implemented as a macro in utf8.h
1450 Perl_utf8n_to_uvchr_error(const U8 *s,
1456 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1458 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1463 =for apidoc utf8n_to_uvchr_msgs
1465 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1466 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1468 This function is for code that needs to know what the precise malformation(s)
1469 are when an error is found, and wants the corresponding warning and/or error
1470 messages to be returned to the caller rather than be displayed. All messages
1471 that would have been displayed if all lexcial warnings are enabled will be
1474 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1475 placed after all the others, C<msgs>. If this parameter is 0, this function
1476 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1477 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1478 contain any appropriate messages. The elements of the array are ordered so
1479 that the first message that would have been displayed is in the 0th element,
1480 and so on. Each element is a hash with three key-value pairs, as follows:
1486 The text of the message as a C<SVpv>.
1488 =item C<warn_categories>
1490 The warning category (or categories) packed into a C<SVuv>.
1494 A single flag bit associated with this message, in a C<SVuv>.
1495 The bit corresponds to some bit in the C<*errors> return value,
1496 such as C<UTF8_GOT_LONG>.
1500 It's important to note that specifying this parameter as non-null will cause
1501 any warnings this function would otherwise generate to be suppressed, and
1502 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1503 (or not) when choosing what to do with the returned messages.
1505 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1508 The caller, of course, is responsible for freeing any returned AV.
1514 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1521 const U8 * const s0 = s;
1522 const U8 * send = s0 + curlen;
1523 U32 possible_problems; /* A bit is set here for each potential problem
1524 found as we go along */
1526 STRLEN expectlen; /* How long should this sequence be? */
1527 STRLEN avail_len; /* When input is too short, gives what that is */
1528 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1529 gets set and discarded */
1531 /* The below are used only if there is both an overlong malformation and a
1532 * too short one. Otherwise the first two are set to 's0' and 'send', and
1533 * the third not used at all */
1535 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1536 routine; see [perl #130921] */
1540 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1542 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1543 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1544 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1549 /* Each of the affected Hanguls starts with \xED */
1551 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1562 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1563 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1564 | (s0[2] & UTF_CONTINUATION_MASK);
1569 /* In conjunction with the exhaustive tests that can be enabled in
1570 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1571 * what it is intended to do, and that no flaws in it are masked by
1572 * dropping down and executing the code below
1573 assert(! isUTF8_CHAR(s0, send)
1574 || UTF8_IS_SURROGATE(s0, send)
1575 || UTF8_IS_SUPER(s0, send)
1576 || UTF8_IS_NONCHAR(s0,send));
1581 possible_problems = 0;
1585 adjusted_s0 = (U8 *) s0;
1592 errors = &discard_errors;
1595 /* The order of malformation tests here is important. We should consume as
1596 * few bytes as possible in order to not skip any valid character. This is
1597 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1598 * http://unicode.org/reports/tr36 for more discussion as to why. For
1599 * example, once we've done a UTF8SKIP, we can tell the expected number of
1600 * bytes, and could fail right off the bat if the input parameters indicate
1601 * that there are too few available. But it could be that just that first
1602 * byte is garbled, and the intended character occupies fewer bytes. If we
1603 * blindly assumed that the first byte is correct, and skipped based on
1604 * that number, we could skip over a valid input character. So instead, we
1605 * always examine the sequence byte-by-byte.
1607 * We also should not consume too few bytes, otherwise someone could inject
1608 * things. For example, an input could be deliberately designed to
1609 * overflow, and if this code bailed out immediately upon discovering that,
1610 * returning to the caller C<*retlen> pointing to the very next byte (one
1611 * which is actually part of of the overflowing sequence), that could look
1612 * legitimate to the caller, which could discard the initial partial
1613 * sequence and process the rest, inappropriately.
1615 * Some possible input sequences are malformed in more than one way. This
1616 * function goes to lengths to try to find all of them. This is necessary
1617 * for correctness, as the inputs may allow one malformation but not
1618 * another, and if we abandon searching for others after finding the
1619 * allowed one, we could allow in something that shouldn't have been.
1622 if (UNLIKELY(curlen == 0)) {
1623 possible_problems |= UTF8_GOT_EMPTY;
1625 uv = UNICODE_REPLACEMENT;
1626 goto ready_to_handle_errors;
1629 expectlen = UTF8SKIP(s);
1631 /* A well-formed UTF-8 character, as the vast majority of calls to this
1632 * function will be for, has this expected length. For efficiency, set
1633 * things up here to return it. It will be overriden only in those rare
1634 * cases where a malformation is found */
1636 *retlen = expectlen;
1639 /* A continuation character can't start a valid sequence */
1640 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1641 possible_problems |= UTF8_GOT_CONTINUATION;
1643 uv = UNICODE_REPLACEMENT;
1644 goto ready_to_handle_errors;
1647 /* Here is not a continuation byte, nor an invariant. The only thing left
1648 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1649 * because it excludes start bytes like \xC0 that always lead to
1652 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1653 * that indicate the number of bytes in the character's whole UTF-8
1654 * sequence, leaving just the bits that are part of the value. */
1655 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1657 /* Setup the loop end point, making sure to not look past the end of the
1658 * input string, and flag it as too short if the size isn't big enough. */
1659 if (UNLIKELY(curlen < expectlen)) {
1660 possible_problems |= UTF8_GOT_SHORT;
1664 send = (U8*) s0 + expectlen;
1667 /* Now, loop through the remaining bytes in the character's sequence,
1668 * accumulating each into the working value as we go. */
1669 for (s = s0 + 1; s < send; s++) {
1670 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1671 uv = UTF8_ACCUMULATE(uv, *s);
1675 /* Here, found a non-continuation before processing all expected bytes.
1676 * This byte indicates the beginning of a new character, so quit, even
1677 * if allowing this malformation. */
1678 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1680 } /* End of loop through the character's bytes */
1682 /* Save how many bytes were actually in the character */
1685 /* Note that there are two types of too-short malformation. One is when
1686 * there is actual wrong data before the normal termination of the
1687 * sequence. The other is that the sequence wasn't complete before the end
1688 * of the data we are allowed to look at, based on the input 'curlen'.
1689 * This means that we were passed data for a partial character, but it is
1690 * valid as far as we saw. The other is definitely invalid. This
1691 * distinction could be important to a caller, so the two types are kept
1694 * A convenience macro that matches either of the too-short conditions. */
1695 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1697 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1699 uv = UNICODE_REPLACEMENT;
1702 /* Check for overflow. The algorithm requires us to not look past the end
1703 * of the current character, even if partial, so the upper limit is 's' */
1704 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1705 1 /* Do consider overlongs */
1708 possible_problems |= UTF8_GOT_OVERFLOW;
1709 uv = UNICODE_REPLACEMENT;
1712 /* Check for overlong. If no problems so far, 'uv' is the correct code
1713 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1714 * we must look at the UTF-8 byte sequence itself to see if it is for an
1716 if ( ( LIKELY(! possible_problems)
1717 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1718 || ( UNLIKELY(possible_problems)
1719 && ( UNLIKELY(! UTF8_IS_START(*s0))
1721 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1724 possible_problems |= UTF8_GOT_LONG;
1726 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1728 /* The calculation in the 'true' branch of this 'if'
1729 * below won't work if overflows, and isn't needed
1730 * anyway. Further below we handle all overflow
1732 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1734 UV min_uv = uv_so_far;
1737 /* Here, the input is both overlong and is missing some trailing
1738 * bytes. There is no single code point it could be for, but there
1739 * may be enough information present to determine if what we have
1740 * so far is for an unallowed code point, such as for a surrogate.
1741 * The code further below has the intelligence to determine this,
1742 * but just for non-overlong UTF-8 sequences. What we do here is
1743 * calculate the smallest code point the input could represent if
1744 * there were no too short malformation. Then we compute and save
1745 * the UTF-8 for that, which is what the code below looks at
1746 * instead of the raw input. It turns out that the smallest such
1747 * code point is all we need. */
1748 for (i = curlen; i < expectlen; i++) {
1749 min_uv = UTF8_ACCUMULATE(min_uv,
1750 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1753 adjusted_s0 = temp_char_buf;
1754 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1758 /* Here, we have found all the possible problems, except for when the input
1759 * is for a problematic code point not allowed by the input parameters. */
1761 /* uv is valid for overlongs */
1762 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1764 /* isn't problematic if < this */
1765 && uv >= UNICODE_SURROGATE_FIRST)
1766 || ( UNLIKELY(possible_problems)
1768 /* if overflow, we know without looking further
1769 * precisely which of the problematic types it is,
1770 * and we deal with those in the overflow handling
1772 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1773 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1774 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1775 && ((flags & ( UTF8_DISALLOW_NONCHAR
1776 |UTF8_DISALLOW_SURROGATE
1777 |UTF8_DISALLOW_SUPER
1778 |UTF8_DISALLOW_PERL_EXTENDED
1780 |UTF8_WARN_SURROGATE
1782 |UTF8_WARN_PERL_EXTENDED))))
1784 /* If there were no malformations, or the only malformation is an
1785 * overlong, 'uv' is valid */
1786 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1787 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1788 possible_problems |= UTF8_GOT_SURROGATE;
1790 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1791 possible_problems |= UTF8_GOT_SUPER;
1793 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1794 possible_problems |= UTF8_GOT_NONCHAR;
1797 else { /* Otherwise, need to look at the source UTF-8, possibly
1798 adjusted to be non-overlong */
1800 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1801 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1803 possible_problems |= UTF8_GOT_SUPER;
1805 else if (curlen > 1) {
1806 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1807 NATIVE_UTF8_TO_I8(*adjusted_s0),
1808 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1810 possible_problems |= UTF8_GOT_SUPER;
1812 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1813 NATIVE_UTF8_TO_I8(*adjusted_s0),
1814 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1816 possible_problems |= UTF8_GOT_SURROGATE;
1820 /* We need a complete well-formed UTF-8 character to discern
1821 * non-characters, so can't look for them here */
1825 ready_to_handle_errors:
1828 * curlen contains the number of bytes in the sequence that
1829 * this call should advance the input by.
1830 * avail_len gives the available number of bytes passed in, but
1831 * only if this is less than the expected number of
1832 * bytes, based on the code point's start byte.
1833 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1834 * is set in it for each potential problem found.
1835 * uv contains the code point the input sequence
1836 * represents; or if there is a problem that prevents
1837 * a well-defined value from being computed, it is
1838 * some subsitute value, typically the REPLACEMENT
1840 * s0 points to the first byte of the character
1841 * s points to just after were we left off processing
1843 * send points to just after where that character should
1844 * end, based on how many bytes the start byte tells
1845 * us should be in it, but no further than s0 +
1849 if (UNLIKELY(possible_problems)) {
1850 bool disallowed = FALSE;
1851 const U32 orig_problems = possible_problems;
1857 while (possible_problems) { /* Handle each possible problem */
1859 char * message = NULL;
1860 U32 this_flag_bit = 0;
1862 /* Each 'if' clause handles one problem. They are ordered so that
1863 * the first ones' messages will be displayed before the later
1864 * ones; this is kinda in decreasing severity order. But the
1865 * overlong must come last, as it changes 'uv' looked at by the
1867 if (possible_problems & UTF8_GOT_OVERFLOW) {
1869 /* Overflow means also got a super and are using Perl's
1870 * extended UTF-8, but we handle all three cases here */
1872 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1873 *errors |= UTF8_GOT_OVERFLOW;
1875 /* But the API says we flag all errors found */
1876 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1877 *errors |= UTF8_GOT_SUPER;
1880 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1882 *errors |= UTF8_GOT_PERL_EXTENDED;
1885 /* Disallow if any of the three categories say to */
1886 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1887 || (flags & ( UTF8_DISALLOW_SUPER
1888 |UTF8_DISALLOW_PERL_EXTENDED)))
1893 /* Likewise, warn if any say to */
1894 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1895 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1898 /* The warnings code explicitly says it doesn't handle the
1899 * case of packWARN2 and two categories which have
1900 * parent-child relationship. Even if it works now to
1901 * raise the warning if either is enabled, it wouldn't
1902 * necessarily do so in the future. We output (only) the
1903 * most dire warning */
1904 if (! (flags & UTF8_CHECK_ONLY)) {
1905 if (msgs || ckWARN_d(WARN_UTF8)) {
1906 pack_warn = packWARN(WARN_UTF8);
1908 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1909 pack_warn = packWARN(WARN_NON_UNICODE);
1912 message = Perl_form(aTHX_ "%s: %s (overflows)",
1914 _byte_dump_string(s0, curlen, 0));
1915 this_flag_bit = UTF8_GOT_OVERFLOW;
1920 else if (possible_problems & UTF8_GOT_EMPTY) {
1921 possible_problems &= ~UTF8_GOT_EMPTY;
1922 *errors |= UTF8_GOT_EMPTY;
1924 if (! (flags & UTF8_ALLOW_EMPTY)) {
1926 /* This so-called malformation is now treated as a bug in
1927 * the caller. If you have nothing to decode, skip calling
1933 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1935 pack_warn = packWARN(WARN_UTF8);
1936 message = Perl_form(aTHX_ "%s (empty string)",
1938 this_flag_bit = UTF8_GOT_EMPTY;
1942 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1943 possible_problems &= ~UTF8_GOT_CONTINUATION;
1944 *errors |= UTF8_GOT_CONTINUATION;
1946 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1949 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1951 pack_warn = packWARN(WARN_UTF8);
1952 message = Perl_form(aTHX_
1953 "%s: %s (unexpected continuation byte 0x%02x,"
1954 " with no preceding start byte)",
1956 _byte_dump_string(s0, 1, 0), *s0);
1957 this_flag_bit = UTF8_GOT_CONTINUATION;
1961 else if (possible_problems & UTF8_GOT_SHORT) {
1962 possible_problems &= ~UTF8_GOT_SHORT;
1963 *errors |= UTF8_GOT_SHORT;
1965 if (! (flags & UTF8_ALLOW_SHORT)) {
1968 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1970 pack_warn = packWARN(WARN_UTF8);
1971 message = Perl_form(aTHX_
1972 "%s: %s (too short; %d byte%s available, need %d)",
1974 _byte_dump_string(s0, send - s0, 0),
1976 avail_len == 1 ? "" : "s",
1978 this_flag_bit = UTF8_GOT_SHORT;
1983 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1984 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1985 *errors |= UTF8_GOT_NON_CONTINUATION;
1987 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1990 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1993 /* If we don't know for sure that the input length is
1994 * valid, avoid as much as possible reading past the
1995 * end of the buffer */
1996 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1999 pack_warn = packWARN(WARN_UTF8);
2000 message = Perl_form(aTHX_ "%s",
2001 unexpected_non_continuation_text(s0,
2005 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2009 else if (possible_problems & UTF8_GOT_SURROGATE) {
2010 possible_problems &= ~UTF8_GOT_SURROGATE;
2012 if (flags & UTF8_WARN_SURROGATE) {
2013 *errors |= UTF8_GOT_SURROGATE;
2015 if ( ! (flags & UTF8_CHECK_ONLY)
2016 && (msgs || ckWARN_d(WARN_SURROGATE)))
2018 pack_warn = packWARN(WARN_SURROGATE);
2020 /* These are the only errors that can occur with a
2021 * surrogate when the 'uv' isn't valid */
2022 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2023 message = Perl_form(aTHX_
2024 "UTF-16 surrogate (any UTF-8 sequence that"
2025 " starts with \"%s\" is for a surrogate)",
2026 _byte_dump_string(s0, curlen, 0));
2029 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2031 this_flag_bit = UTF8_GOT_SURROGATE;
2035 if (flags & UTF8_DISALLOW_SURROGATE) {
2037 *errors |= UTF8_GOT_SURROGATE;
2040 else if (possible_problems & UTF8_GOT_SUPER) {
2041 possible_problems &= ~UTF8_GOT_SUPER;
2043 if (flags & UTF8_WARN_SUPER) {
2044 *errors |= UTF8_GOT_SUPER;
2046 if ( ! (flags & UTF8_CHECK_ONLY)
2047 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2049 pack_warn = packWARN(WARN_NON_UNICODE);
2051 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2052 message = Perl_form(aTHX_
2053 "Any UTF-8 sequence that starts with"
2054 " \"%s\" is for a non-Unicode code point,"
2055 " may not be portable",
2056 _byte_dump_string(s0, curlen, 0));
2059 message = Perl_form(aTHX_ super_cp_format, uv);
2061 this_flag_bit = UTF8_GOT_SUPER;
2065 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2066 * and before possibly bailing out, so that the more dire
2067 * warning will override the regular one. */
2068 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2069 if ( ! (flags & UTF8_CHECK_ONLY)
2070 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2071 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2073 pack_warn = packWARN(WARN_NON_UNICODE);
2075 /* If it is an overlong that evaluates to a code point
2076 * that doesn't have to use the Perl extended UTF-8, it
2077 * still used it, and so we output a message that
2078 * doesn't refer to the code point. The same is true
2079 * if there was a SHORT malformation where the code
2080 * point is not valid. In that case, 'uv' will have
2081 * been set to the REPLACEMENT CHAR, and the message
2082 * below without the code point in it will be selected
2084 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2085 message = Perl_form(aTHX_
2086 perl_extended_cp_format, uv);
2089 message = Perl_form(aTHX_
2090 "Any UTF-8 sequence that starts with"
2091 " \"%s\" is a Perl extension, and"
2092 " so is not portable",
2093 _byte_dump_string(s0, curlen, 0));
2095 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2098 if (flags & ( UTF8_WARN_PERL_EXTENDED
2099 |UTF8_DISALLOW_PERL_EXTENDED))
2101 *errors |= UTF8_GOT_PERL_EXTENDED;
2103 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2109 if (flags & UTF8_DISALLOW_SUPER) {
2110 *errors |= UTF8_GOT_SUPER;
2114 else if (possible_problems & UTF8_GOT_NONCHAR) {
2115 possible_problems &= ~UTF8_GOT_NONCHAR;
2117 if (flags & UTF8_WARN_NONCHAR) {
2118 *errors |= UTF8_GOT_NONCHAR;
2120 if ( ! (flags & UTF8_CHECK_ONLY)
2121 && (msgs || ckWARN_d(WARN_NONCHAR)))
2123 /* The code above should have guaranteed that we don't
2124 * get here with errors other than overlong */
2125 assert (! (orig_problems
2126 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2128 pack_warn = packWARN(WARN_NONCHAR);
2129 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2130 this_flag_bit = UTF8_GOT_NONCHAR;
2134 if (flags & UTF8_DISALLOW_NONCHAR) {
2136 *errors |= UTF8_GOT_NONCHAR;
2139 else if (possible_problems & UTF8_GOT_LONG) {
2140 possible_problems &= ~UTF8_GOT_LONG;
2141 *errors |= UTF8_GOT_LONG;
2143 if (flags & UTF8_ALLOW_LONG) {
2145 /* We don't allow the actual overlong value, unless the
2146 * special extra bit is also set */
2147 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2148 & ~UTF8_ALLOW_LONG)))
2150 uv = UNICODE_REPLACEMENT;
2157 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2159 pack_warn = packWARN(WARN_UTF8);
2161 /* These error types cause 'uv' to be something that
2162 * isn't what was intended, so can't use it in the
2163 * message. The other error types either can't
2164 * generate an overlong, or else the 'uv' is valid */
2166 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2168 message = Perl_form(aTHX_
2169 "%s: %s (any UTF-8 sequence that starts"
2170 " with \"%s\" is overlong which can and"
2171 " should be represented with a"
2172 " different, shorter sequence)",
2174 _byte_dump_string(s0, send - s0, 0),
2175 _byte_dump_string(s0, curlen, 0));
2178 U8 tmpbuf[UTF8_MAXBYTES+1];
2179 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2181 /* Don't use U+ for non-Unicode code points, which
2182 * includes those in the Latin1 range */
2183 const char * preface = ( uv > PERL_UNICODE_MAX
2190 message = Perl_form(aTHX_
2191 "%s: %s (overlong; instead use %s to represent"
2194 _byte_dump_string(s0, send - s0, 0),
2195 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2197 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2198 small code points */
2201 this_flag_bit = UTF8_GOT_LONG;
2204 } /* End of looking through the possible flags */
2206 /* Display the message (if any) for the problem being handled in
2207 * this iteration of the loop */
2210 assert(this_flag_bit);
2212 if (*msgs == NULL) {
2216 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2221 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2224 Perl_warner(aTHX_ pack_warn, "%s", message);
2226 } /* End of 'while (possible_problems)' */
2228 /* Since there was a possible problem, the returned length may need to
2229 * be changed from the one stored at the beginning of this function.
2230 * Instead of trying to figure out if that's needed, just do it. */
2236 if (flags & UTF8_CHECK_ONLY && retlen) {
2237 *retlen = ((STRLEN) -1);
2243 return UNI_TO_NATIVE(uv);
2247 =for apidoc utf8_to_uvchr_buf
2249 Returns the native code point of the first character in the string C<s> which
2250 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2251 C<*retlen> will be set to the length, in bytes, of that character.
2253 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2254 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2255 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2256 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2257 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2258 the next possible position in C<s> that could begin a non-malformed character.
2259 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2264 Also implemented as a macro in utf8.h
2270 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2272 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2274 return _utf8_to_uvchr_buf(s, send, retlen);
2277 /* This is marked as deprecated
2279 =for apidoc utf8_to_uvuni_buf
2281 Only in very rare circumstances should code need to be dealing in Unicode
2282 (as opposed to native) code points. In those few cases, use
2283 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2284 are not absolutely sure this is one of those cases, then assume it isn't and
2285 use plain C<utf8_to_uvchr_buf> instead.
2287 Returns the Unicode (not-native) code point of the first character in the
2289 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2290 C<retlen> will be set to the length, in bytes, of that character.
2292 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2293 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2294 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2295 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2296 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2297 next possible position in C<s> that could begin a non-malformed character.
2298 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
2304 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2306 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2310 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2314 =for apidoc utf8_length
2316 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2317 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2318 same place, it returns 0 with no warning raised.
2320 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2321 and returns the number of valid characters.
2327 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2331 PERL_ARGS_ASSERT_UTF8_LENGTH;
2333 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2334 * the bitops (especially ~) can create illegal UTF-8.
2335 * In other words: in Perl UTF-8 is not just for Unicode. */
2337 if (UNLIKELY(e < s))
2338 goto warn_and_return;
2344 if (UNLIKELY(e != s)) {
2348 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2349 "%s in %s", unees, OP_DESC(PL_op));
2351 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2358 =for apidoc bytes_cmp_utf8
2360 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2361 sequence of characters (stored as UTF-8)
2362 in C<u>, C<ulen>. Returns 0 if they are
2363 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2364 if the first string is greater than the second string.
2366 -1 or +1 is returned if the shorter string was identical to the start of the
2367 longer string. -2 or +2 is returned if
2368 there was a difference between characters
2375 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2377 const U8 *const bend = b + blen;
2378 const U8 *const uend = u + ulen;
2380 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2382 while (b < bend && u < uend) {
2384 if (!UTF8_IS_INVARIANT(c)) {
2385 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2388 if (UTF8_IS_CONTINUATION(c1)) {
2389 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2391 /* diag_listed_as: Malformed UTF-8 character%s */
2392 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2394 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2395 PL_op ? " in " : "",
2396 PL_op ? OP_DESC(PL_op) : "");
2401 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2402 "%s in %s", unees, OP_DESC(PL_op));
2404 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2405 return -2; /* Really want to return undef :-) */
2412 return *b < c ? -2 : +2;
2417 if (b == bend && u == uend)
2420 return b < bend ? +1 : -1;
2424 =for apidoc utf8_to_bytes
2426 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2427 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2428 updates C<*lenp> to contain the new length.
2429 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2431 Upon successful return, the number of variants in the string can be computed by
2432 having saved the value of C<*lenp> before the call, and subtracting the
2433 after-call value of C<*lenp> from it.
2435 If you need a copy of the string, see L</bytes_from_utf8>.
2441 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2445 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2446 PERL_UNUSED_CONTEXT;
2448 /* This is a no-op if no variants at all in the input */
2449 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2454 U8 * const save = s;
2455 U8 * const send = s + *lenp;
2458 /* Nothing before the first variant needs to be changed, so start the real
2462 if (! UTF8_IS_INVARIANT(*s)) {
2463 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2464 *lenp = ((STRLEN) -1);
2472 /* Is downgradable, so do it */
2473 d = s = first_variant;
2476 if (! UVCHR_IS_INVARIANT(c)) {
2477 /* Then it is two-byte encoded */
2478 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2491 =for apidoc bytes_from_utf8
2493 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2494 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2495 actually encoded in UTF-8.
2497 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2500 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2501 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2502 C<*lenp> are unchanged, and the return value is the original C<s>.
2504 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2505 newly created string containing a downgraded copy of C<s>, and whose length is
2506 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2507 caller is responsible for arranging for the memory used by this string to get
2510 Upon successful return, the number of variants in the string can be computed by
2511 having saved the value of C<*lenp> before the call, and subtracting the
2512 after-call value of C<*lenp> from it.
2516 There is a macro that avoids this function call, but this is retained for
2517 anyone who calls it with the Perl_ prefix */
2520 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2522 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2523 PERL_UNUSED_CONTEXT;
2525 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2529 No = here because currently externally undocumented
2530 for apidoc bytes_from_utf8_loc
2532 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2533 to store the location of the first character in C<"s"> that cannot be
2534 converted to non-UTF8.
2536 If that parameter is C<NULL>, this function behaves identically to
2539 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2540 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2542 Otherwise, the function returns a newly created C<NUL>-terminated string
2543 containing the non-UTF8 equivalent of the convertible first portion of
2544 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2545 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2546 and C<*first_non_downgradable> is set to C<NULL>.
2548 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2549 first character in the original string that wasn't converted. C<*is_utf8p> is
2550 unchanged. Note that the new string may have length 0.
2552 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2553 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2554 converts as many characters in it as possible stopping at the first one it
2555 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2556 set to point to that. The function returns the portion that could be converted
2557 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2558 not including the terminating C<NUL>. If the very first character in the
2559 original could not be converted, C<*lenp> will be 0, and the new string will
2560 contain just a single C<NUL>. If the entire input string was converted,
2561 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2563 Upon successful return, the number of variants in the converted portion of the
2564 string can be computed by having saved the value of C<*lenp> before the call,
2565 and subtracting the after-call value of C<*lenp> from it.
2573 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2576 const U8 *original = s;
2577 U8 *converted_start;
2578 const U8 *send = s + *lenp;
2580 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2583 if (first_unconverted) {
2584 *first_unconverted = NULL;
2587 return (U8 *) original;
2590 Newx(d, (*lenp) + 1, U8);
2592 converted_start = d;
2595 if (! UTF8_IS_INVARIANT(c)) {
2597 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2598 * have to stop now */
2599 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2600 if (first_unconverted) {
2601 *first_unconverted = s - 1;
2602 goto finish_and_return;
2605 Safefree(converted_start);
2606 return (U8 *) original;
2610 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2616 /* Here, converted the whole of the input */
2618 if (first_unconverted) {
2619 *first_unconverted = NULL;
2624 *lenp = d - converted_start;
2626 /* Trim unused space */
2627 Renew(converted_start, *lenp + 1, U8);
2629 return converted_start;
2633 =for apidoc bytes_to_utf8
2635 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2637 Returns a pointer to the newly-created string, and sets C<*lenp> to
2638 reflect the new length in bytes. The caller is responsible for arranging for
2639 the memory used by this string to get freed.
2641 Upon successful return, the number of variants in the string can be computed by
2642 having saved the value of C<*lenp> before the call, and subtracting it from the
2643 after-call value of C<*lenp>.
2645 A C<NUL> character will be written after the end of the string.
2647 If you want to convert to UTF-8 from encodings other than
2648 the native (Latin1 or EBCDIC),
2649 see L</sv_recode_to_utf8>().
2655 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2657 const U8 * const send = s + (*lenp);
2661 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2662 PERL_UNUSED_CONTEXT;
2664 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2665 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2669 append_utf8_from_native_byte(*s, &d);
2680 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2681 * use utf16_to_utf8_reversed().
2683 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2684 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2685 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2687 * These functions don't check for overflow. The worst case is every code
2688 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2689 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2690 * destination must be pre-extended to 2 times the source length.
2692 * Do not use in-place. We optimize for native, for obvious reasons. */
2695 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2700 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2703 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2709 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2711 if (OFFUNI_IS_INVARIANT(uv)) {
2712 *d++ = LATIN1_TO_NATIVE((U8) uv);
2715 if (uv <= MAX_UTF8_TWO_BYTE) {
2716 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2717 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2721 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2722 #define LAST_HIGH_SURROGATE 0xDBFF
2723 #define FIRST_LOW_SURROGATE 0xDC00
2724 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2725 #define FIRST_IN_PLANE1 0x10000
2727 /* This assumes that most uses will be in the first Unicode plane, not
2728 * needing surrogates */
2729 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2730 && uv <= UNICODE_SURROGATE_LAST))
2732 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2733 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2736 UV low = (p[0] << 8) + p[1];
2737 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2738 || UNLIKELY(low > LAST_LOW_SURROGATE))
2740 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2743 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2744 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2748 d = uvoffuni_to_utf8_flags(d, uv, 0);
2750 if (uv < FIRST_IN_PLANE1) {
2751 *d++ = (U8)(( uv >> 12) | 0xe0);
2752 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2753 *d++ = (U8)(( uv & 0x3f) | 0x80);
2757 *d++ = (U8)(( uv >> 18) | 0xf0);
2758 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2759 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2760 *d++ = (U8)(( uv & 0x3f) | 0x80);
2765 *newlen = d - dstart;
2769 /* Note: this one is slightly destructive of the source. */
2772 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2775 U8* const send = s + bytelen;
2777 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2780 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2784 const U8 tmp = s[0];
2789 return utf16_to_utf8(p, d, bytelen, newlen);
2793 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2796 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2799 /* Internal function so we can deprecate the external one, and call
2800 this one from other deprecated functions in this file */
2803 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2807 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2811 return is_utf8_common(p, PL_utf8_idstart);
2815 Perl__is_uni_perl_idcont(pTHX_ UV c)
2818 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2822 Perl__is_uni_perl_idstart(pTHX_ UV c)
2825 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2829 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2832 /* We have the latin1-range values compiled into the core, so just use
2833 * those, converting the result to UTF-8. The only difference between upper
2834 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2835 * either "SS" or "Ss". Which one to use is passed into the routine in
2836 * 'S_or_s' to avoid a test */
2838 UV converted = toUPPER_LATIN1_MOD(c);
2840 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2842 assert(S_or_s == 'S' || S_or_s == 's');
2844 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2845 characters in this range */
2846 *p = (U8) converted;
2851 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2852 * which it maps to one of them, so as to only have to have one check for
2853 * it in the main case */
2854 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2856 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2857 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2860 converted = GREEK_CAPITAL_LETTER_MU;
2862 #if UNICODE_MAJOR_VERSION > 2 \
2863 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2864 && UNICODE_DOT_DOT_VERSION >= 8)
2865 case LATIN_SMALL_LETTER_SHARP_S:
2872 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2873 " '%c' to map to '%c'",
2874 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2875 NOT_REACHED; /* NOTREACHED */
2879 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2880 *p = UTF8_TWO_BYTE_LO(converted);
2886 /* If compiled on an early Unicode version, there may not be auxiliary tables
2888 #ifndef HAS_UC_AUX_TABLES
2889 # define UC_AUX_TABLE_ptrs NULL
2890 # define UC_AUX_TABLE_lengths NULL
2892 #ifndef HAS_TC_AUX_TABLES
2893 # define TC_AUX_TABLE_ptrs NULL
2894 # define TC_AUX_TABLE_lengths NULL
2896 #ifndef HAS_LC_AUX_TABLES
2897 # define LC_AUX_TABLE_ptrs NULL
2898 # define LC_AUX_TABLE_lengths NULL
2900 #ifndef HAS_CF_AUX_TABLES
2901 # define CF_AUX_TABLE_ptrs NULL
2902 # define CF_AUX_TABLE_lengths NULL
2904 #ifndef HAS_UC_AUX_TABLES
2905 # define UC_AUX_TABLE_ptrs NULL
2906 # define UC_AUX_TABLE_lengths NULL
2909 /* Call the function to convert a UTF-8 encoded character to the specified case.
2910 * Note that there may be more than one character in the result.
2911 * 's' is a pointer to the first byte of the input character
2912 * 'd' will be set to the first byte of the string of changed characters. It
2913 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2914 * 'lenp' will be set to the length in bytes of the string of changed characters
2916 * The functions return the ordinal of the first character in the string of
2918 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2919 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2920 Uppercase_Mapping_invmap, \
2921 UC_AUX_TABLE_ptrs, \
2922 UC_AUX_TABLE_lengths, \
2924 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2925 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2926 Titlecase_Mapping_invmap, \
2927 TC_AUX_TABLE_ptrs, \
2928 TC_AUX_TABLE_lengths, \
2930 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2931 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2932 Lowercase_Mapping_invmap, \
2933 LC_AUX_TABLE_ptrs, \
2934 LC_AUX_TABLE_lengths, \
2938 /* This additionally has the input parameter 'specials', which if non-zero will
2939 * cause this to use the specials hash for folding (meaning get full case
2940 * folding); otherwise, when zero, this implies a simple case fold */
2941 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2943 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2944 Case_Folding_invmap, \
2945 CF_AUX_TABLE_ptrs, \
2946 CF_AUX_TABLE_lengths, \
2948 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2949 Simple_Case_Folding_invmap, \
2954 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2956 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2957 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2958 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2959 * the changed version may be longer than the original character.
2961 * The ordinal of the first character of the changed version is returned
2962 * (but note, as explained above, that there may be more.) */
2965 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2968 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2971 return CALL_UPPER_CASE(c, NULL, p, lenp);
2975 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2978 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2981 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2984 return CALL_TITLE_CASE(c, NULL, p, lenp);
2988 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2990 /* We have the latin1-range values compiled into the core, so just use
2991 * those, converting the result to UTF-8. Since the result is always just
2992 * one character, we allow <p> to be NULL */
2994 U8 converted = toLOWER_LATIN1(c);
2996 PERL_UNUSED_ARG(dummy);
2999 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
3004 /* Result is known to always be < 256, so can use the EIGHT_BIT
3006 *p = UTF8_EIGHT_BIT_HI(converted);
3007 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3015 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3018 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3021 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3024 return CALL_LOWER_CASE(c, NULL, p, lenp);
3028 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3030 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3031 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3032 * FOLD_FLAGS_FULL iff full folding is to be used;
3034 * Not to be used for locale folds
3039 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3041 assert (! (flags & FOLD_FLAGS_LOCALE));
3043 if (UNLIKELY(c == MICRO_SIGN)) {
3044 converted = GREEK_SMALL_LETTER_MU;
3046 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3047 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3048 || UNICODE_DOT_DOT_VERSION > 0)
3049 else if ( (flags & FOLD_FLAGS_FULL)
3050 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3052 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3053 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3054 * under those circumstances. */
3055 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3056 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3057 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3059 return LATIN_SMALL_LETTER_LONG_S;
3069 else { /* In this range the fold of all other characters is their lower
3071 converted = toLOWER_LATIN1(c);
3074 if (UVCHR_IS_INVARIANT(converted)) {
3075 *p = (U8) converted;
3079 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3080 *p = UTF8_TWO_BYTE_LO(converted);
3088 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3091 /* Not currently externally documented, and subject to change
3092 * <flags> bits meanings:
3093 * FOLD_FLAGS_FULL iff full folding is to be used;
3094 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3095 * locale are to be used.
3096 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3100 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3102 if (flags & FOLD_FLAGS_LOCALE) {
3103 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3104 * except for potentially warning */
3105 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3106 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3107 flags &= ~FOLD_FLAGS_LOCALE;
3110 goto needs_full_generality;
3115 return _to_fold_latin1((U8) c, p, lenp,
3116 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3119 /* Here, above 255. If no special needs, just use the macro */
3120 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3121 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3123 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3124 the special flags. */
3125 U8 utf8_c[UTF8_MAXBYTES + 1];
3127 needs_full_generality:
3128 uvchr_to_utf8(utf8_c, c);
3129 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3134 PERL_STATIC_INLINE bool
3135 S_is_utf8_common(pTHX_ const U8 *const p, SV* const invlist)
3137 /* returns a boolean giving whether or not the UTF8-encoded character that
3138 * starts at <p> is in the inversion list indicated by <invlist>.
3140 * Note that it is assumed that the buffer length of <p> is enough to
3141 * contain all the bytes that comprise the character. Thus, <*p> should
3142 * have been checked before this call for mal-formedness enough to assure
3143 * that. This function, does make sure to not look past any NUL, so it is
3144 * safe to use on C, NUL-terminated, strings */
3145 STRLEN len = my_strnlen((char *) p, UTF8SKIP(p));
3147 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3149 /* The API should have included a length for the UTF-8 character in <p>,
3150 * but it doesn't. We therefore assume that p has been validated at least
3151 * as far as there being enough bytes available in it to accommodate the
3152 * character without reading beyond the end, and pass that number on to the
3153 * validating routine */
3154 if (! isUTF8_CHAR(p, p + len)) {
3155 _force_out_malformed_utf8_message(p, p + len, _UTF8_NO_CONFIDENCE_IN_CURLEN,
3157 NOT_REACHED; /* NOTREACHED */
3160 return is_utf8_common_with_len(p, p + len, invlist);
3163 PERL_STATIC_INLINE bool
3164 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3167 /* returns a boolean giving whether or not the UTF8-encoded character that
3168 * starts at <p>, and extending no further than <e - 1> is in the inversion
3169 * list <invlist>. */
3171 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3173 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3175 if (cp == 0 && (p >= e || *p != '\0')) {
3176 _force_out_malformed_utf8_message(p, e, 0, 1);
3177 NOT_REACHED; /* NOTREACHED */
3181 return _invlist_contains_cp(invlist, cp);
3185 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3186 const char * const alternative,
3187 const bool use_locale,
3188 const char * const file,
3189 const unsigned line)
3193 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3195 if (ckWARN_d(WARN_DEPRECATED)) {
3197 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3198 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3199 if (! PL_seen_deprecated_macro) {
3200 PL_seen_deprecated_macro = newHV();
3202 if (! hv_store(PL_seen_deprecated_macro, key,
3203 strlen(key), &PL_sv_undef, 0))
3205 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3208 if (instr(file, "mathoms.c")) {
3209 Perl_warner(aTHX_ WARN_DEPRECATED,
3210 "In %s, line %d, starting in Perl v5.32, %s()"
3211 " will be removed. Avoid this message by"
3212 " converting to use %s().\n",
3213 file, line, name, alternative);
3216 Perl_warner(aTHX_ WARN_DEPRECATED,
3217 "In %s, line %d, starting in Perl v5.32, %s() will"
3218 " require an additional parameter. Avoid this"
3219 " message by converting to use %s().\n",
3220 file, line, name, alternative);
3227 Perl__is_utf8_FOO(pTHX_ U8 classnum,
3229 const char * const name,
3230 const char * const alternative,
3231 const bool use_utf8,
3232 const bool use_locale,
3233 const char * const file,
3234 const unsigned line)
3237 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3239 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3241 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
3251 case _CC_ALPHANUMERIC:
3255 return is_utf8_common(p, PL_XPosix_ptrs[classnum]);
3258 return is_XPERLSPACE_high(p);
3260 return is_HORIZWS_high(p);
3262 return is_XDIGIT_high(p);
3268 return is_VERTWS_high(p);
3270 return is_utf8_common(p, PL_utf8_perl_idstart);
3272 return is_utf8_common(p, PL_utf8_perl_idcont);
3276 /* idcont is the same as wordchar below 256 */
3277 if (classnum == _CC_IDCONT) {
3278 classnum = _CC_WORDCHAR;
3280 else if (classnum == _CC_IDFIRST) {
3284 classnum = _CC_ALPHA;
3288 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3289 return _generic_isCC(*p, classnum);
3292 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3295 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3296 return isFOO_lc(classnum, *p);
3299 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3302 NOT_REACHED; /* NOTREACHED */
3306 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3310 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3312 return is_utf8_common_with_len(p, e, PL_XPosix_ptrs[classnum]);
3316 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3319 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3321 return is_utf8_common_with_len(p, e, PL_utf8_perl_idstart);
3325 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3328 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3332 return is_utf8_common(p, PL_utf8_xidstart);
3336 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3339 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3341 return is_utf8_common_with_len(p, e, PL_utf8_perl_idcont);
3345 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3348 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3350 return is_utf8_common(p, PL_utf8_idcont);
3354 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3357 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3359 return is_utf8_common(p, PL_utf8_xidcont);
3363 Perl__is_utf8_mark(pTHX_ const U8 *p)
3366 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3368 return is_utf8_common(p, PL_utf8_mark);
3372 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3373 U8* ustrp, STRLEN *lenp,
3374 SV *invlist, const int * const invmap,
3375 const unsigned int * const * const aux_tables,
3376 const U8 * const aux_table_lengths,
3377 const char * const normal)
3381 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3382 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3383 * to name the new case in any generated messages, as a fallback if the
3384 * operation being used is not available. The new case is given by the
3385 * data structures in the remaining arguments.
3387 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3388 * entire changed case string, and the return value is the first code point
3391 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3393 /* For code points that don't change case, we already know that the output
3394 * of this function is the unchanged input, so we can skip doing look-ups
3395 * for them. Unfortunately the case-changing code points are scattered
3396 * around. But there are some long consecutive ranges where there are no
3397 * case changing code points. By adding tests, we can eliminate the lookup
3398 * for all the ones in such ranges. This is currently done here only for
3399 * just a few cases where the scripts are in common use in modern commerce
3400 * (and scripts adjacent to those which can be included without additional
3403 if (uv1 >= 0x0590) {
3404 /* This keeps from needing further processing the code points most
3405 * likely to be used in the following non-cased scripts: Hebrew,
3406 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3407 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3408 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3413 /* The following largish code point ranges also don't have case
3414 * changes, but khw didn't think they warranted extra tests to speed
3415 * them up (which would slightly slow down everything else above them):
3416 * 1100..139F Hangul Jamo, Ethiopic
3417 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3418 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3419 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3420 * Combining Diacritical Marks Extended, Balinese,
3421 * Sundanese, Batak, Lepcha, Ol Chiki
3422 * 2000..206F General Punctuation
3425 if (uv1 >= 0x2D30) {
3427 /* This keeps the from needing further processing the code points
3428 * most likely to be used in the following non-cased major scripts:
3429 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3431 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3432 * event that Unicode eventually allocates the unused block as of
3433 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3434 * that the test suite will start having failures to alert you
3435 * should that happen) */
3440 if (uv1 >= 0xAC00) {
3441 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3442 if (ckWARN_d(WARN_SURROGATE)) {
3443 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3444 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3445 "Operation \"%s\" returns its argument for"
3446 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3451 /* AC00..FAFF Catches Hangul syllables and private use, plus
3457 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3458 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3459 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3462 if (ckWARN_d(WARN_NON_UNICODE)) {
3463 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3464 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3465 "Operation \"%s\" returns its argument for"
3466 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3470 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3472 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3475 /* As of Unicode 10.0, this means we avoid swash creation
3476 * for anything beyond high Plane 1 (below emojis) */
3483 /* Note that non-characters are perfectly legal, so no warning should
3489 const unsigned int * cp_list;
3492 /* 'index' is guaranteed to be non-negative, as this is an inversion
3493 * map that covers all possible inputs. See [perl #133365] */
3494 SSize_t index = _invlist_search(invlist, uv1);
3495 IV base = invmap[index];
3497 /* The data structures are set up so that if 'base' is non-negative,
3498 * the case change is 1-to-1; and if 0, the change is to itself */
3506 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3507 lc = base + uv1 - invlist_array(invlist)[index];
3508 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3512 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3513 * requires an auxiliary table look up. abs(base) gives the index into
3514 * a list of such tables which points to the proper aux table. And a
3515 * parallel list gives the length of each corresponding aux table. */
3516 cp_list = aux_tables[-base];
3518 /* Create the string of UTF-8 from the mapped-to code points */
3520 for (i = 0; i < aux_table_lengths[-base]; i++) {
3521 d = uvchr_to_utf8(d, cp_list[i]);
3529 /* Here, there was no mapping defined, which means that the code point maps
3530 * to itself. Return the inputs */
3534 if (p != ustrp) { /* Don't copy onto itself */
3535 Copy(p, ustrp, len, U8);
3540 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3548 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3549 const unsigned int ** remaining_folds_to)
3551 /* Returns the count of the number of code points that fold to the input
3552 * 'cp' (besides itself).
3554 * If the return is 0, there is nothing else that folds to it, and
3555 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3557 * If the return is 1, '*first_folds_to' is set to the single code point,
3558 * and '*remaining_folds_to' is set to NULL.
3560 * Otherwise, '*first_folds_to' is set to a code point, and
3561 * '*remaining_fold_to' is set to an array that contains the others. The
3562 * length of this array is the returned count minus 1.
3564 * The reason for this convolution is to avoid having to deal with
3565 * allocating and freeing memory. The lists are already constructed, so
3566 * the return can point to them, but single code points aren't, so would
3567 * need to be constructed if we didn't employ something like this API */
3570 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3571 * that covers all possible inputs. See [perl #133365] */
3572 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3573 int base = _Perl_IVCF_invmap[index];
3575 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3577 if (base == 0) { /* No fold */
3578 *first_folds_to = 0;
3579 *remaining_folds_to = NULL;
3583 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3589 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3591 /* The data structure is set up so that the absolute value of 'base' is
3592 * an index into a table of pointers to arrays, with the array
3593 * corresponding to the index being the list of code points that fold
3594 * to 'cp', and the parallel array containing the length of the list
3596 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3597 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3600 return IVCF_AUX_TABLE_lengths[-base];
3605 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3606 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3607 *remaining_folds_to = NULL;
3612 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3613 U8* const ustrp, STRLEN *lenp)
3615 /* This is called when changing the case of a UTF-8-encoded character above
3616 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3617 * result contains a character that crosses the 255/256 boundary, disallow
3618 * the change, and return the original code point. See L<perlfunc/lc> for
3621 * p points to the original string whose case was changed; assumed
3622 * by this routine to be well-formed
3623 * result the code point of the first character in the changed-case string
3624 * ustrp points to the changed-case string (<result> represents its
3626 * lenp points to the length of <ustrp> */
3628 UV original; /* To store the first code point of <p> */
3630 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3632 assert(UTF8_IS_ABOVE_LATIN1(*p));
3634 /* We know immediately if the first character in the string crosses the
3635 * boundary, so can skip testing */
3638 /* Look at every character in the result; if any cross the
3639 * boundary, the whole thing is disallowed */
3640 U8* s = ustrp + UTF8SKIP(ustrp);
3641 U8* e = ustrp + *lenp;
3643 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3649 /* Here, no characters crossed, result is ok as-is, but we warn. */
3650 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3656 /* Failed, have to return the original */
3657 original = valid_utf8_to_uvchr(p, lenp);
3659 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3660 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3661 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3662 " locale; resolved to \"\\x{%" UVXf "}\".",
3666 Copy(p, ustrp, *lenp, char);
3671 S_check_and_deprecate(pTHX_ const U8 *p,
3673 const unsigned int type, /* See below */
3674 const bool use_locale, /* Is this a 'LC_'
3676 const char * const file,
3677 const unsigned line)
3679 /* This is a temporary function to deprecate the unsafe calls to the case
3680 * changing macros and functions. It keeps all the special stuff in just
3683 * It updates *e with the pointer to the end of the input string. If using
3684 * the old-style macros, *e is NULL on input, and so this function assumes
3685 * the input string is long enough to hold the entire UTF-8 sequence, and
3686 * sets *e accordingly, but it then returns a flag to pass the
3687 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3688 * using the full length if possible.
3690 * It also does the assert that *e > p when *e is not NULL. This should be
3691 * migrated to the callers when this function gets deleted.
3693 * The 'type' parameter is used for the caller to specify which case
3694 * changing function this is called from: */
3696 # define DEPRECATE_TO_UPPER 0
3697 # define DEPRECATE_TO_TITLE 1
3698 # define DEPRECATE_TO_LOWER 2
3699 # define DEPRECATE_TO_FOLD 3
3701 U32 utf8n_flags = 0;
3703 const char * alternative;
3705 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3708 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3710 /* strnlen() makes this function safe for the common case of
3711 * NUL-terminated strings */
3712 *e = p + my_strnlen((char *) p, UTF8SKIP(p));
3714 /* For mathoms.c calls, we use the function name we know is stored
3715 * there. It could be part of a larger path */
3716 if (type == DEPRECATE_TO_UPPER) {
3717 name = instr(file, "mathoms.c")
3720 alternative = "toUPPER_utf8_safe";
3722 else if (type == DEPRECATE_TO_TITLE) {
3723 name = instr(file, "mathoms.c")
3726 alternative = "toTITLE_utf8_safe";
3728 else if (type == DEPRECATE_TO_LOWER) {
3729 name = instr(file, "mathoms.c")
3732 alternative = "toLOWER_utf8_safe";
3734 else if (type == DEPRECATE_TO_FOLD) {
3735 name = instr(file, "mathoms.c")
3738 alternative = "toFOLD_utf8_safe";
3740 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3742 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3752 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3753 U8 * ustrp, STRLEN *lenp)
3755 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3756 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3757 * Otherwise, it returns the first code point of the Turkic foldcased
3758 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3759 * contain *lenp bytes
3761 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3762 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3765 PERL_ARGS_ASSERT_TURKIC_FC;
3768 if (UNLIKELY(*p == 'I')) {
3770 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3771 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3772 return LATIN_SMALL_LETTER_DOTLESS_I;
3775 if (UNLIKELY(memBEGINs(p, e - p,
3776 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3787 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3788 U8 * ustrp, STRLEN *lenp)
3790 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3791 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3792 * Otherwise, it returns the first code point of the Turkic lowercased
3793 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3794 * contain *lenp bytes */
3797 PERL_ARGS_ASSERT_TURKIC_LC;
3800 /* A 'I' requires context as to what to do */
3801 if (UNLIKELY(*p0 == 'I')) {
3802 const U8 * p = p0 + 1;
3804 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3805 * modified by a dot above lowercases to 'i' even in turkic locales. */
3809 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3815 /* For the dot above to modify the 'I', it must be part of a
3816 * combining sequence immediately following the 'I', and no other
3817 * modifier with a ccc of 230 may intervene */
3818 cp = utf8_to_uvchr_buf(p, e, NULL);
3819 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3823 /* Here the combining sequence continues */
3828 /* In all other cases the lc is the same as the fold */
3829 return turkic_fc(p0, e, ustrp, lenp);
3833 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3834 U8 * ustrp, STRLEN *lenp)
3836 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3837 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3838 * Otherwise, it returns the first code point of the Turkic upper or
3839 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3840 * ustrp will contain *lenp bytes
3842 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3843 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3846 PERL_ARGS_ASSERT_TURKIC_UC;
3851 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3852 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3853 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3856 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3865 /* The process for changing the case is essentially the same for the four case
3866 * change types, except there are complications for folding. Otherwise the
3867 * difference is only which case to change to. To make sure that they all do
3868 * the same thing, the bodies of the functions are extracted out into the
3869 * following two macros. The functions are written with the same variable
3870 * names, and these are known and used inside these macros. It would be
3871 * better, of course, to have inline functions to do it, but since different
3872 * macros are called, depending on which case is being changed to, this is not
3873 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3874 * function can start with the common start macro, then finish with its special
3875 * handling; while the other three cases can just use the common end macro.
3877 * The algorithm is to use the proper (passed in) macro or function to change
3878 * the case for code points that are below 256. The macro is used if using
3879 * locale rules for the case change; the function if not. If the code point is
3880 * above 255, it is computed from the input UTF-8, and another macro is called
3881 * to do the conversion. If necessary, the output is converted to UTF-8. If
3882 * using a locale, we have to check that the change did not cross the 255/256
3883 * boundary, see check_locale_boundary_crossing() for further details.
3885 * The macros are split with the correct case change for the below-256 case
3886 * stored into 'result', and in the middle of an else clause for the above-255
3887 * case. At that point in the 'else', 'result' is not the final result, but is
3888 * the input code point calculated from the UTF-8. The fold code needs to
3889 * realize all this and take it from there.
3891 * To deal with Turkic locales, the function specified by the parameter
3892 * 'turkic' is called when appropriate.
3894 * If you read the two macros as sequential, it's easier to understand what's
3896 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3897 L1_func_extra_param, turkic) \
3899 if (flags & (locale_flags)) { \
3900 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3901 if (IN_UTF8_CTYPE_LOCALE) { \
3902 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3903 UV ret = turkic(p, e, ustrp, lenp); \
3904 if (ret) return ret; \
3907 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3909 flags &= ~(locale_flags); \
3913 if (UTF8_IS_INVARIANT(*p)) { \
3914 if (flags & (locale_flags)) { \
3915 result = LC_L1_change_macro(*p); \
3918 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3921 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3922 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3923 if (flags & (locale_flags)) { \
3924 result = LC_L1_change_macro(c); \
3927 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3930 else { /* malformed UTF-8 or ord above 255 */ \
3931 STRLEN len_result; \
3932 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3933 if (len_result == (STRLEN) -1) { \
3934 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3938 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3939 result = change_macro(result, p, ustrp, lenp); \
3941 if (flags & (locale_flags)) { \
3942 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3947 /* Here, used locale rules. Convert back to UTF-8 */ \
3948 if (UTF8_IS_INVARIANT(result)) { \
3949 *ustrp = (U8) result; \
3953 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3954 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3961 =for apidoc to_utf8_upper
3963 Instead use L</toUPPER_utf8_safe>.
3967 /* Not currently externally documented, and subject to change:
3968 * <flags> is set iff iff the rules from the current underlying locale are to
3972 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3977 const char * const file,
3982 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3983 cBOOL(flags), file, line);
3985 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3987 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3988 /* 2nd char of uc(U+DF) is 'S' */
3989 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3991 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3995 =for apidoc to_utf8_title
3997 Instead use L</toTITLE_utf8_safe>.
4001 /* Not currently externally documented, and subject to change:
4002 * <flags> is set iff the rules from the current underlying locale are to be
4003 * used. Since titlecase is not defined in POSIX, for other than a
4004 * UTF-8 locale, uppercase is used instead for code points < 256.
4008 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
4013 const char * const file,
4018 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
4019 cBOOL(flags), file, line);
4021 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
4023 /* 2nd char of ucfirst(U+DF) is 's' */
4024 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
4026 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
4030 =for apidoc to_utf8_lower
4032 Instead use L</toLOWER_utf8_safe>.
4036 /* Not currently externally documented, and subject to change:
4037 * <flags> is set iff iff the rules from the current underlying locale are to
4042 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
4047 const char * const file,
4052 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
4053 cBOOL(flags), file, line);
4055 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
4057 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
4059 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
4063 =for apidoc to_utf8_fold
4065 Instead use L</toFOLD_utf8_safe>.
4069 /* Not currently externally documented, and subject to change,
4071 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
4072 * locale are to be used.
4073 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
4074 * otherwise simple folds
4075 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
4080 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
4085 const char * const file,
4090 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
4091 cBOOL(flags), file, line);
4093 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
4095 /* These are mutually exclusive */
4096 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
4098 assert(p != ustrp); /* Otherwise overwrites */
4100 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
4101 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
4104 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
4106 if (flags & FOLD_FLAGS_LOCALE) {
4108 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
4109 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4110 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4112 /* Special case these two characters, as what normally gets
4113 * returned under locale doesn't work */
4114 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
4116 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4117 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4118 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
4119 "resolved to \"\\x{17F}\\x{17F}\".");
4124 if (memBEGINs((char *) p, e - p, LONG_S_T))
4126 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4127 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4128 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
4129 "resolved to \"\\x{FB06}\".");
4130 goto return_ligature_st;
4133 #if UNICODE_MAJOR_VERSION == 3 \
4134 && UNICODE_DOT_VERSION == 0 \
4135 && UNICODE_DOT_DOT_VERSION == 1
4136 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
4138 /* And special case this on this Unicode version only, for the same
4139 * reaons the other two are special cased. They would cross the
4140 * 255/256 boundary which is forbidden under /l, and so the code
4141 * wouldn't catch that they are equivalent (which they are only in
4143 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
4144 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4145 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4146 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
4147 "resolved to \"\\x{0131}\".");
4148 goto return_dotless_i;
4152 return check_locale_boundary_crossing(p, result, ustrp, lenp);
4154 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
4158 /* This is called when changing the case of a UTF-8-encoded
4159 * character above the ASCII range, and the result should not
4160 * contain an ASCII character. */
4162 UV original; /* To store the first code point of <p> */
4164 /* Look at every character in the result; if any cross the
4165 * boundary, the whole thing is disallowed */
4167 U8* e = ustrp + *lenp;
4170 /* Crossed, have to return the original */
4171 original = valid_utf8_to_uvchr(p, lenp);
4173 /* But in these instances, there is an alternative we can
4174 * return that is valid */
4175 if (original == LATIN_SMALL_LETTER_SHARP_S
4176 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
4177 || original == LATIN_CAPITAL_LETTER_SHARP_S
4182 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
4183 goto return_ligature_st;
4185 #if UNICODE_MAJOR_VERSION == 3 \
4186 && UNICODE_DOT_VERSION == 0 \
4187 && UNICODE_DOT_DOT_VERSION == 1
4189 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4190 goto return_dotless_i;
4193 Copy(p, ustrp, *lenp, char);
4199 /* Here, no characters crossed, result is ok as-is */
4204 /* Here, used locale rules. Convert back to UTF-8 */
4205 if (UTF8_IS_INVARIANT(result)) {
4206 *ustrp = (U8) result;
4210 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
4211 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
4218 /* Certain folds to 'ss' are prohibited by the options, but they do allow
4219 * folds to a string of two of these characters. By returning this
4220 * instead, then, e.g.,
4221 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
4224 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
4225 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
4227 return LATIN_SMALL_LETTER_LONG_S;
4230 /* Two folds to 'st' are prohibited by the options; instead we pick one and
4231 * have the other one fold to it */
4233 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
4234 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
4235 return LATIN_SMALL_LIGATURE_ST;
4237 #if UNICODE_MAJOR_VERSION == 3 \
4238 && UNICODE_DOT_VERSION == 0 \
4239 && UNICODE_DOT_DOT_VERSION == 1
4242 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
4243 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
4244 return LATIN_SMALL_LETTER_DOTLESS_I;
4251 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
4252 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
4253 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
4257 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4258 I32 minbits, I32 none)
4260 /* Returns a copy of a swash initiated by the called function. This is the
4261 * public interface, and returning a copy prevents others from doing
4262 * mischief on the original. The only remaining use of this is in tr/// */
4264 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
4265 * use the following define */
4267 #define SWASH_INIT_RETURN(x) \
4268 PL_curpm= old_PL_curpm; \
4271 /* Initialize and return a swash, creating it if necessary. It does this
4272 * by calling utf8_heavy.pl in the general case.
4274 * pkg is the name of the package that <name> should be in.
4275 * name is the name of the swash to find.
4276 * listsv is a string to initialize the swash with. It must be of the form
4277 * documented as the subroutine return value in
4278 * L<perlunicode/User-Defined Character Properties>
4279 * minbits is the number of bits required to represent each data element.
4280 * none I (khw) do not understand this one, but it is used only in tr///.
4282 * Thus there are two possible inputs to find the swash: <name> and
4283 * <listsv>. At least one must be specified. The result
4284 * will be the union of the specified ones, although <listsv>'s various
4285 * actions can intersect, etc. what <name> gives. To avoid going out to
4286 * disk at all, <invlist> should specify completely what the swash should
4287 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
4290 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
4292 SV* retval = &PL_sv_undef;
4294 PERL_ARGS_ASSERT_SWASH_INIT;
4296 assert(listsv != &PL_sv_undef || strNE(name, ""));
4298 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
4299 regex that triggered the swash init and the swash init
4300 perl logic itself. See perl #122747 */
4302 /* If data was passed in to go out to utf8_heavy to find the swash of, do
4304 if (listsv != &PL_sv_undef || strNE(name, "")) {
4306 const size_t pkg_len = strlen(pkg);
4307 const size_t name_len = strlen(name);
4308 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
4313 PUSHSTACKi(PERLSI_MAGIC);
4317 /* We might get here via a subroutine signature which uses a utf8
4318 * parameter name, at which point PL_subname will have been set
4319 * but not yet used. */
4320 save_item(PL_subname);
4321 if (PL_parser && PL_parser->error_count)
4322 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
4323 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4324 if (!method) { /* demand load UTF-8 */
4326 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4327 GvSV(PL_errgv) = NULL;
4328 #ifndef NO_TAINT_SUPPORT
4329 /* It is assumed that callers of this routine are not passing in
4330 * any user derived data. */
4331 /* Need to do this after save_re_context() as it will set
4332 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
4333 * in Perl_magic_get). Even line to create errsv_save can turn on
4335 SAVEBOOL(TAINT_get);
4338 require_pv("utf8_heavy.pl");
4340 /* Not ERRSV, as there is no need to vivify a scalar we are
4341 about to discard. */
4342 SV * const errsv = GvSV(PL_errgv);
4343 if (!SvTRUE(errsv)) {
4344 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4345 SvREFCNT_dec(errsv);
4353 mPUSHp(pkg, pkg_len);
4354 mPUSHp(name, name_len);
4359 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4360 GvSV(PL_errgv) = NULL;
4361 /* If we already have a pointer to the method, no need to use
4362 * call_method() to repeat the lookup. */
4364 ? call_sv(MUTABLE_SV(method), G_SCALAR)
4365 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
4367 retval = *PL_stack_sp--;
4368 SvREFCNT_inc(retval);
4371 /* Not ERRSV. See above. */
4372 SV * const errsv = GvSV(PL_errgv);
4373 if (!SvTRUE(errsv)) {
4374 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4375 SvREFCNT_dec(errsv);
4380 if (IN_PERL_COMPILETIME) {
4381 CopHINTS_set(PL_curcop, PL_hints);
4383 } /* End of calling the module to find the swash */
4385 SWASH_INIT_RETURN(retval);
4386 #undef SWASH_INIT_RETURN
4390 /* This API is wrong for special case conversions since we may need to
4391 * return several Unicode characters for a single Unicode character
4392 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
4393 * the lower-level routine, and it is similarly broken for returning
4394 * multiple values. --jhi
4395 * For those, you should use S__to_utf8_case() instead */
4396 /* Now SWASHGET is recasted into S_swatch_get in this file. */
4399 * Returns the value of property/mapping C<swash> for the first character
4400 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4401 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
4402 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
4404 * A "swash" is a hash which contains initially the keys/values set up by
4405 * SWASHNEW. The purpose is to be able to completely represent a Unicode
4406 * property for all possible code points. Things are stored in a compact form
4407 * (see utf8_heavy.pl) so that calculation is required to find the actual
4408 * property value for a given code point. As code points are looked up, new
4409 * key/value pairs are added to the hash, so that the calculation doesn't have
4410 * to ever be re-done. Further, each calculation is done, not just for the
4411 * desired one, but for a whole block of code points adjacent to that one.
4412 * For binary properties on ASCII machines, the block is usually for 64 code
4413 * points, starting with a code point evenly divisible by 64. Thus if the
4414 * property value for code point 257 is requested, the code goes out and
4415 * calculates the property values for all 64 code points between 256 and 319,
4416 * and stores these as a single 64-bit long bit vector, called a "swatch",
4417 * under the key for code point 256. The key is the UTF-8 encoding for code
4418 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
4419 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
4420 * for code point 258 is then requested, this code realizes that it would be
4421 * stored under the key for 256, and would find that value and extract the
4422 * relevant bit, offset from 256.
4424 * Non-binary properties are stored in as many bits as necessary to represent
4425 * their values (32 currently, though the code is more general than that), not
4426 * as single bits, but the principle is the same: the value for each key is a
4427 * vector that encompasses the property values for all code points whose UTF-8
4428 * representations are represented by the key. That is, for all code points
4429 * whose UTF-8 representations are length N bytes, and the key is the first N-1
4433 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
4435 HV *const hv = MUTABLE_HV(SvRV(swash));
4440 const U8 *tmps = NULL;
4444 PERL_ARGS_ASSERT_SWASH_FETCH;
4446 /* If it really isn't a hash, it isn't really swash; must be an inversion
4448 if (SvTYPE(hv) != SVt_PVHV) {
4449 return _invlist_contains_cp((SV*)hv,
4451 ? valid_utf8_to_uvchr(ptr, NULL)
4455 /* We store the values in a "swatch" which is a vec() value in a swash
4456 * hash. Code points 0-255 are a single vec() stored with key length
4457 * (klen) 0. All other code points have a UTF-8 representation
4458 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
4459 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
4460 * length for them is the length of the encoded char - 1. ptr[klen] is the
4461 * final byte in the sequence representing the character */
4462 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
4467 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
4470 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
4473 klen = UTF8SKIP(ptr) - 1;
4475 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
4476 * the vec is the final byte in the sequence. (In EBCDIC this is
4477 * converted to I8 to get consecutive values.) To help you visualize
4479 * Straight 1047 After final byte
4480 * UTF-8 UTF-EBCDIC I8 transform
4481 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
4482 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
4484 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
4485 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
4487 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
4488 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
4490 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
4491 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
4493 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
4494 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
4496 * (There are no discontinuities in the elided (...) entries.)
4497 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
4498 * key for the next 31, up through U+043F, whose UTF-8 final byte is
4499 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
4500 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
4501 * index into the vec() swatch (after subtracting 0x80, which we
4502 * actually do with an '&').
4503 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
4504 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4505 * dicontinuities which go away by transforming it into I8, and we
4506 * effectively subtract 0xA0 to get the index. */
4507 needents = (1 << UTF_ACCUMULATION_SHIFT);
4508 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4512 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4513 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4514 * it's nothing to sniff at.) Pity we usually come through at least
4515 * two function calls to get here...
4517 * NB: this code assumes that swatches are never modified, once generated!
4520 if (hv == PL_last_swash_hv &&
4521 klen == PL_last_swash_klen &&
4522 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4524 tmps = PL_last_swash_tmps;
4525 slen = PL_last_swash_slen;
4528 /* Try our second-level swatch cache, kept in a hash. */
4529 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4531 /* If not cached, generate it via swatch_get */
4532 if (!svp || !SvPOK(*svp)
4533 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4536 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4537 swatch = swatch_get(swash,
4538 code_point & ~((UV)needents - 1),
4541 else { /* For the first 256 code points, the swatch has a key of
4543 swatch = swatch_get(swash, 0, needents);
4546 if (IN_PERL_COMPILETIME)
4547 CopHINTS_set(PL_curcop, PL_hints);
4549 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4551 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4552 || (slen << 3) < needents)
4553 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4554 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4555 svp, tmps, (UV)slen, (UV)needents);
4558 PL_last_swash_hv = hv;
4559 assert(klen <= sizeof(PL_last_swash_key));
4560 PL_last_swash_klen = (U8)klen;
4561 /* FIXME change interpvar.h? */
4562 PL_last_swash_tmps = (U8 *) tmps;
4563 PL_last_swash_slen = slen;
4565 Copy(ptr, PL_last_swash_key, klen, U8);
4568 switch ((int)((slen << 3) / needents)) {
4570 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4572 return ((UV) tmps[off]);
4576 ((UV) tmps[off ] << 8) +
4577 ((UV) tmps[off + 1]);
4581 ((UV) tmps[off ] << 24) +
4582 ((UV) tmps[off + 1] << 16) +
4583 ((UV) tmps[off + 2] << 8) +
4584 ((UV) tmps[off + 3]);
4586 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4587 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4588 NORETURN_FUNCTION_END;
4591 /* Read a single line of the main body of the swash input text. These are of
4594 * where each number is hex. The first two numbers form the minimum and
4595 * maximum of a range, and the third is the value associated with the range.
4596 * Not all swashes should have a third number
4598 * On input: l points to the beginning of the line to be examined; it points
4599 * to somewhere in the string of the whole input text, and is
4600 * terminated by a \n or the null string terminator.
4601 * lend points to the null terminator of that string
4602 * wants_value is non-zero if the swash expects a third number
4603 * typestr is the name of the swash's mapping, like 'ToLower'
4604 * On output: *min, *max, and *val are set to the values read from the line.
4605 * returns a pointer just beyond the line examined. If there was no
4606 * valid min number on the line, returns lend+1
4610 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4611 const bool wants_value, const U8* const typestr)
4613 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4614 STRLEN numlen; /* Length of the number */
4615 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4616 | PERL_SCAN_DISALLOW_PREFIX
4617 | PERL_SCAN_SILENT_NON_PORTABLE;
4619 /* nl points to the next \n in the scan */
4620 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4622 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4624 /* Get the first number on the line: the range minimum */
4626 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4627 *max = *min; /* So can never return without setting max */
4628 if (numlen) /* If found a hex number, position past it */
4630 else if (nl) { /* Else, go handle next line, if any */
4631 return nl + 1; /* 1 is length of "\n" */
4633 else { /* Else, no next line */
4634 return lend + 1; /* to LIST's end at which \n is not found */
4637 /* The max range value follows, separated by a BLANK */
4640 flags = PERL_SCAN_SILENT_ILLDIGIT
4641 | PERL_SCAN_DISALLOW_PREFIX
4642 | PERL_SCAN_SILENT_NON_PORTABLE;
4644 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4647 else /* If no value here, it is a single element range */
4650 /* Non-binary tables have a third entry: what the first element of the
4651 * range maps to. The map for those currently read here is in hex */
4655 flags = PERL_SCAN_SILENT_ILLDIGIT
4656 | PERL_SCAN_DISALLOW_PREFIX
4657 | PERL_SCAN_SILENT_NON_PORTABLE;
4659 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4668 /* diag_listed_as: To%s: illegal mapping '%s' */
4669 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4675 *val = 0; /* bits == 1, then any val should be ignored */
4677 else { /* Nothing following range min, should be single element with no
4682 /* diag_listed_as: To%s: illegal mapping '%s' */
4683 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4687 *val = 0; /* bits == 1, then val should be ignored */
4690 /* Position to next line if any, or EOF */
4700 * Returns a swatch (a bit vector string) for a code point sequence
4701 * that starts from the value C<start> and comprises the number C<span>.
4702 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4703 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4706 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4709 U8 *l, *lend, *x, *xend, *s;
4710 STRLEN lcur, xcur, scur;
4711 HV *const hv = MUTABLE_HV(SvRV(swash));
4713 SV** listsvp = NULL; /* The string containing the main body of the table */
4714 SV** extssvp = NULL;
4717 STRLEN octets; /* if bits == 1, then octets == 0 */
4719 UV end = start + span;
4721 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4722 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4723 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4724 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4725 listsvp = hv_fetchs(hv, "LIST", FALSE);
4727 bits = SvUV(*bitssvp);
4728 none = SvUV(*nonesvp);
4729 typestr = (U8*)SvPV_nolen(*typesvp);
4730 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4732 PERL_ARGS_ASSERT_SWATCH_GET;
4734 if (bits != 8 && bits != 16 && bits != 32) {
4735 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4739 /* If overflowed, use the max possible */
4745 /* create and initialize $swatch */
4746 scur = octets ? (span * octets) : (span + 7) / 8;
4747 swatch = newSV(scur);
4749 s = (U8*)SvPVX(swatch);
4750 if (octets && none) {
4751 const U8* const e = s + scur;
4754 *s++ = (U8)(none & 0xff);
4755 else if (bits == 16) {
4756 *s++ = (U8)((none >> 8) & 0xff);
4757 *s++ = (U8)( none & 0xff);
4759 else if (bits == 32) {
4760 *s++ = (U8)((none >> 24) & 0xff);
4761 *s++ = (U8)((none >> 16) & 0xff);
4762 *s++ = (U8)((none >> 8) & 0xff);
4763 *s++ = (U8)( none & 0xff);
4769 (void)memzero((U8*)s, scur + 1);
4771 SvCUR_set(swatch, scur);
4772 s = (U8*)SvPVX(swatch);
4774 /* read $swash->{LIST} */
4775 l = (U8*)SvPV(*listsvp, lcur);
4778 UV min = 0, max = 0, val = 0, upper;
4779 l = swash_scan_list_line(l, lend, &min, &max, &val,
4780 cBOOL(octets), typestr);
4785 /* If looking for something beyond this range, go try the next one */
4789 /* <end> is generally 1 beyond where we want to set things, but at the
4790 * platform's infinity, where we can't go any higher, we want to
4791 * include the code point at <end> */
4794 : (max != UV_MAX || end != UV_MAX)
4801 if (!none || val < none) {
4806 for (key = min; key <= upper; key++) {
4808 /* offset must be non-negative (start <= min <= key < end) */
4809 offset = octets * (key - start);
4811 s[offset] = (U8)(val & 0xff);
4812 else if (bits == 16) {
4813 s[offset ] = (U8)((val >> 8) & 0xff);
4814 s[offset + 1] = (U8)( val & 0xff);
4816 else if (bits == 32) {
4817 s[offset ] = (U8)((val >> 24) & 0xff);
4818 s[offset + 1] = (U8)((val >> 16) & 0xff);
4819 s[offset + 2] = (U8)((val >> 8) & 0xff);
4820 s[offset + 3] = (U8)( val & 0xff);
4823 if (!none || val < none)
4829 /* read $swash->{EXTRAS} */
4830 x = (U8*)SvPV(*extssvp, xcur);
4838 SV **otherbitssvp, *other;
4842 const U8 opc = *x++;
4846 nl = (U8*)memchr(x, '\n', xend - x);
4848 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4850 x = nl + 1; /* 1 is length of "\n" */
4854 x = xend; /* to EXTRAS' end at which \n is not found */
4861 namelen = nl - namestr;
4865 namelen = xend - namestr;
4869 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4870 otherhv = MUTABLE_HV(SvRV(*othersvp));
4871 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4872 otherbits = (STRLEN)SvUV(*otherbitssvp);
4873 if (bits < otherbits)
4874 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4875 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4877 /* The "other" swatch must be destroyed after. */
4878 other = swatch_get(*othersvp, start, span);
4879 o = (U8*)SvPV(other, olen);
4882 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4884 s = (U8*)SvPV(swatch, slen);
4886 STRLEN otheroctets = otherbits >> 3;
4888 U8* const send = s + slen;
4893 if (otherbits == 1) {
4894 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4898 STRLEN vlen = otheroctets;
4906 if (opc == '+' && otherval)
4907 NOOP; /* replace with otherval */
4908 else if (opc == '!' && !otherval)
4910 else if (opc == '-' && otherval)
4912 else if (opc == '&' && !otherval)
4915 s += octets; /* no replacement */
4920 *s++ = (U8)( otherval & 0xff);
4921 else if (bits == 16) {
4922 *s++ = (U8)((otherval >> 8) & 0xff);
4923 *s++ = (U8)( otherval & 0xff);
4925 else if (bits == 32) {
4926 *s++ = (U8)((otherval >> 24) & 0xff);
4927 *s++ = (U8)((otherval >> 16) & 0xff);
4928 *s++ = (U8)((otherval >> 8) & 0xff);
4929 *s++ = (U8)( otherval & 0xff);
4933 sv_free(other); /* through with it! */
4939 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4941 /* May change: warns if surrogates, non-character code points, or
4942 * non-Unicode code points are in 's' which has length 'len' bytes.
4943 * Returns TRUE if none found; FALSE otherwise. The only other validity
4944 * check is to make sure that this won't exceed the string's length nor
4947 const U8* const e = s + len;
4950 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4953 if (UTF8SKIP(s) > len) {
4954 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4955 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4958 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4959 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4960 if ( ckWARN_d(WARN_NON_UNICODE)
4961 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
4962 0 /* Don't consider overlongs */
4965 /* A side effect of this function will be to warn */
4966 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4970 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4971 if (ckWARN_d(WARN_SURROGATE)) {
4972 /* This has a different warning than the one the called
4973 * function would output, so can't just call it, unlike we
4974 * do for the non-chars and above-unicodes */
4975 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4976 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4977 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4982 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4983 && (ckWARN_d(WARN_NONCHAR)))
4985 /* A side effect of this function will be to warn */
4986 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4997 =for apidoc pv_uni_display
4999 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5000 length C<len>, the displayable version being at most C<pvlim> bytes long
5001 (if longer, the rest is truncated and C<"..."> will be appended).
5003 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5004 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5005 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5006 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5007 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5008 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5010 The pointer to the PV of the C<dsv> is returned.
5012 See also L</sv_uni_display>.
5016 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
5022 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5026 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5028 /* This serves double duty as a flag and a character to print after
5029 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5033 if (pvlim && SvCUR(dsv) >= pvlim) {
5037 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5039 const unsigned char c = (unsigned char)u & 0xFF;
5040 if (flags & UNI_DISPLAY_BACKSLASH) {
5057 const char string = ok;
5058 sv_catpvs(dsv, "\\");
5059 sv_catpvn(dsv, &string, 1);
5062 /* isPRINT() is the locale-blind version. */
5063 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5064 const char string = c;
5065 sv_catpvn(dsv, &string, 1);
5070 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5073 sv_catpvs(dsv, "...");
5079 =for apidoc sv_uni_display
5081 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5082 the displayable version being at most C<pvlim> bytes long
5083 (if longer, the rest is truncated and "..." will be appended).
5085 The C<flags> argument is as in L</pv_uni_display>().
5087 The pointer to the PV of the C<dsv> is returned.
5092 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5094 const char * const ptr =
5095 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5097 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5099 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5100 SvCUR(ssv), pvlim, flags);
5104 =for apidoc foldEQ_utf8
5106 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
5107 both of which may be in UTF-8) are the same case-insensitively; false
5108 otherwise. How far into the strings to compare is determined by other input
5111 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5112 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
5113 C<u2> with respect to C<s2>.
5115 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
5116 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
5117 The scan will not be considered to be a match unless the goal is reached, and
5118 scanning won't continue past that goal. Correspondingly for C<l2> with respect
5121 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
5122 pointer is considered an end pointer to the position 1 byte past the maximum
5123 point in C<s1> beyond which scanning will not continue under any circumstances.
5124 (This routine assumes that UTF-8 encoded input strings are not malformed;
5125 malformed input can cause it to read past C<pe1>). This means that if both
5126 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
5127 will never be successful because it can never
5128 get as far as its goal (and in fact is asserted against). Correspondingly for
5129 C<pe2> with respect to C<s2>.
5131 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5132 C<l2> must be non-zero), and if both do, both have to be
5133 reached for a successful match. Also, if the fold of a character is multiple
5134 characters, all of them must be matched (see tr21 reference below for
5137 Upon a successful match, if C<pe1> is non-C<NULL>,
5138 it will be set to point to the beginning of the I<next> character of C<s1>
5139 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5141 For case-insensitiveness, the "casefolding" of Unicode is used
5142 instead of upper/lowercasing both the characters, see
5143 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5147 /* A flags parameter has been added which may change, and hence isn't
5148 * externally documented. Currently it is:
5149 * 0 for as-documented above
5150 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5151 ASCII one, to not match
5152 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5153 * locale are to be used.
5154 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5155 * routine. This allows that step to be skipped.
5156 * Currently, this requires s1 to be encoded as UTF-8
5157 * (u1 must be true), which is asserted for.
5158 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5159 * cross certain boundaries. Hence, the caller should
5160 * let this function do the folding instead of
5161 * pre-folding. This code contains an assertion to
5162 * that effect. However, if the caller knows what
5163 * it's doing, it can pass this flag to indicate that,
5164 * and the assertion is skipped.
5165 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
5166 * to s2, and s2 doesn't have to be UTF-8 encoded.
5167 * This introduces an asymmetry to save a few branches
5168 * in a loop. Currently, this is not a problem, as
5169 * never are both inputs pre-folded. Simply call this
5170 * function with the pre-folded one as the second
5172 * FOLDEQ_S2_FOLDS_SANE
5175 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
5176 const char *s2, char **pe2, UV l2, bool u2,
5179 const U8 *p1 = (const U8*)s1; /* Point to current char */
5180 const U8 *p2 = (const U8*)s2;
5181 const U8 *g1 = NULL; /* goal for s1 */
5182 const U8 *g2 = NULL;
5183 const U8 *e1 = NULL; /* Don't scan s1 past this */
5184 U8 *f1 = NULL; /* Point to current folded */
5185 const U8 *e2 = NULL;
5187 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5188 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5189 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5190 U8 flags_for_folder = FOLD_FLAGS_FULL;
5192 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5194 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5195 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
5196 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5197 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
5198 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5199 /* The algorithm is to trial the folds without regard to the flags on
5200 * the first line of the above assert(), and then see if the result
5201 * violates them. This means that the inputs can't be pre-folded to a
5202 * violating result, hence the assert. This could be changed, with the
5203 * addition of extra tests here for the already-folded case, which would
5204 * slow it down. That cost is more than any possible gain for when these
5205 * flags are specified, as the flags indicate /il or /iaa matching which
5206 * is less common than /iu, and I (khw) also believe that real-world /il
5207 * and /iaa matches are most likely to involve code points 0-255, and this
5208 * function only under rare conditions gets called for 0-255. */
5210 if (flags & FOLDEQ_LOCALE) {
5211 if (IN_UTF8_CTYPE_LOCALE) {
5212 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
5213 flags_for_folder |= FOLD_FLAGS_LOCALE;
5216 flags &= ~FOLDEQ_LOCALE;
5220 flags_for_folder |= FOLD_FLAGS_LOCALE;
5223 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
5224 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
5232 g1 = (const U8*)s1 + l1;
5240 g2 = (const U8*)s2 + l2;
5243 /* Must have at least one goal */
5248 /* Will never match if goal is out-of-bounds */
5249 assert(! e1 || e1 >= g1);
5251 /* Here, there isn't an end pointer, or it is beyond the goal. We
5252 * only go as far as the goal */
5256 assert(e1); /* Must have an end for looking at s1 */
5259 /* Same for goal for s2 */
5261 assert(! e2 || e2 >= g2);
5268 /* If both operands are already folded, we could just do a memEQ on the
5269 * whole strings at once, but it would be better if the caller realized
5270 * this and didn't even call us */
5272 /* Look through both strings, a character at a time */
5273 while (p1 < e1 && p2 < e2) {
5275 /* If at the beginning of a new character in s1, get its fold to use
5276 * and the length of the fold. */
5278 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5284 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5286 /* We have to forbid mixing ASCII with non-ASCII if the
5287 * flags so indicate. And, we can short circuit having to
5288 * call the general functions for this common ASCII case,
5289 * all of whose non-locale folds are also ASCII, and hence
5290 * UTF-8 invariants, so the UTF8ness of the strings is not
5292 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5296 *foldbuf1 = toFOLD(*p1);
5299 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5301 else { /* Not UTF-8, get UTF-8 fold */
5302 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5308 if (n2 == 0) { /* Same for s2 */
5309 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5311 /* Point to the already-folded character. But for non-UTF-8
5312 * variants, convert to UTF-8 for the algorithm below */
5313 if (UTF8_IS_INVARIANT(*p2)) {
5322 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
5323 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
5329 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5330 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5334 *foldbuf2 = toFOLD(*p2);
5337 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5340 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5346 /* Here f1 and f2 point to the beginning of the strings to compare.
5347 * These strings are the folds of the next character from each input
5348 * string, stored in UTF-8. */
5350 /* While there is more to look for in both folds, see if they
5351 * continue to match */
5353 U8 fold_length = UTF8SKIP(f1);
5354 if (fold_length != UTF8SKIP(f2)
5355 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5356 function call for single
5358 || memNE((char*)f1, (char*)f2, fold_length))
5360 return 0; /* mismatch */
5363 /* Here, they matched, advance past them */
5370 /* When reach the end of any fold, advance the input past it */
5372 p1 += u1 ? UTF8SKIP(p1) : 1;
5375 p2 += u2 ? UTF8SKIP(p2) : 1;
5377 } /* End of loop through both strings */
5379 /* A match is defined by each scan that specified an explicit length
5380 * reaching its final goal, and the other not having matched a partial
5381 * character (which can happen when the fold of a character is more than one
5383 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5387 /* Successful match. Set output pointers */
5397 /* XXX The next two functions should likely be moved to mathoms.c once all
5398 * occurrences of them are removed from the core; some cpan-upstream modules
5402 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5404 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5406 return uvoffuni_to_utf8_flags(d, uv, 0);
5410 =for apidoc utf8n_to_uvuni
5412 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5414 This function was useful for code that wanted to handle both EBCDIC and
5415 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5416 distinctions between the platforms have mostly been made invisible to most
5417 code, so this function is quite unlikely to be what you want. If you do need
5418 this precise functionality, use instead
5419 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5420 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5426 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5428 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5430 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5434 =for apidoc uvuni_to_utf8_flags
5436 Instead you almost certainly want to use L</uvchr_to_utf8> or
5437 L</uvchr_to_utf8_flags>.
5439 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5440 which itself, while not deprecated, should be used only in isolated
5441 circumstances. These functions were useful for code that wanted to handle
5442 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5443 v5.20, the distinctions between the platforms have mostly been made invisible
5444 to most code, so this function is quite unlikely to be what you want.
5450 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5452 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5454 return uvoffuni_to_utf8_flags(d, uv, flags);
5458 =for apidoc utf8_to_uvchr
5460 Returns the native code point of the first character in the string C<s>
5461 which is assumed to be in UTF-8 encoding; C<retlen> will be set to the
5462 length, in bytes, of that character.
5464 Some, but not all, UTF-8 malformations are detected, and in fact, some
5465 malformed input could cause reading beyond the end of the input buffer, which
5466 is why this function is deprecated. Use L</utf8_to_uvchr_buf> instead.
5468 If C<s> points to one of the detected malformations, and UTF8 warnings are
5469 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
5470 C<NULL>) to -1. If those warnings are off, the computed value if well-defined (or
5471 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
5472 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
5473 next possible position in C<s> that could begin a non-malformed character.
5474 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
5480 Perl_utf8_to_uvchr(pTHX_ const U8 *s, STRLEN *retlen)
5482 PERL_ARGS_ASSERT_UTF8_TO_UVCHR;
5484 /* This function is unsafe if malformed UTF-8 input is given it, which is
5485 * why the function is deprecated. If the first byte of the input
5486 * indicates that there are more bytes remaining in the sequence that forms
5487 * the character than there are in the input buffer, it can read past the
5488 * end. But we can make it safe if the input string happens to be
5489 * NUL-terminated, as many strings in Perl are, by refusing to read past a
5490 * NUL. A NUL indicates the start of the next character anyway. If the
5491 * input isn't NUL-terminated, the function remains unsafe, as it always
5494 * An initial NUL has to be handled separately, but all ASCIIs can be
5495 * handled the same way, speeding up this common case */
5497 if (UTF8_IS_INVARIANT(*s)) { /* Assumes 's' contains at least 1 byte */
5504 return utf8_to_uvchr_buf(s,
5505 s + my_strnlen((char *) s, UTF8SKIP(s)),
5510 * ex: set ts=8 sts=4 sw=4 et: