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;
45 #define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
48 =head1 Unicode Support
49 These are various utility functions for manipulating UTF8-encoded
50 strings. For the uninitiated, this is a method of representing arbitrary
51 Unicode characters as a variable number of bytes, in such a way that
52 characters in the ASCII range are unmodified, and a zero byte never appears
53 within non-zero characters.
59 Perl__force_out_malformed_utf8_message(pTHX_
60 const U8 *const p, /* First byte in UTF-8 sequence */
61 const U8 * const e, /* Final byte in sequence (may include
63 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
64 usually 0, or some DISALLOW flags */
65 const bool die_here) /* If TRUE, this function does not return */
67 /* This core-only function is to be called when a malformed UTF-8 character
68 * is found, in order to output the detailed information about the
69 * malformation before dieing. The reason it exists is for the occasions
70 * when such a malformation is fatal, but warnings might be turned off, so
71 * that normally they would not be actually output. This ensures that they
72 * do get output. Because a sequence may be malformed in more than one
73 * way, multiple messages may be generated, so we can't make them fatal, as
74 * that would cause the first one to die.
76 * Instead we pretend -W was passed to perl, then die afterwards. The
77 * flexibility is here to return to the caller so they can finish up and
81 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
87 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
89 PL_curcop->cop_warnings = pWARN_ALL;
92 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
97 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
98 " be called only when there are errors found");
102 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
107 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
108 U32 categories, /* Packed warning categories */
109 U32 flag) /* Flag associated with this message */
111 /* Creates, populates, and returns an HV* that describes an error message
112 * for the translators between UTF8 and code point */
114 SV* msg_sv = newSVpv(message, 0);
115 SV* category_sv = newSVuv(categories);
116 SV* flag_bit_sv = newSVuv(flag);
118 HV* msg_hv = newHV();
120 PERL_ARGS_ASSERT_NEW_MSG_HV;
122 (void) hv_stores(msg_hv, "text", msg_sv);
123 (void) hv_stores(msg_hv, "warn_categories", category_sv);
124 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
130 =for apidoc uvoffuni_to_utf8_flags
132 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
133 Instead, B<Almost all code should use L</uvchr_to_utf8> or
134 L</uvchr_to_utf8_flags>>.
136 This function is like them, but the input is a strict Unicode
137 (as opposed to native) code point. Only in very rare circumstances should code
138 not be using the native code point.
140 For details, see the description for L</uvchr_to_utf8_flags>.
146 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
148 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
150 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
153 /* All these formats take a single UV code point argument */
154 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
155 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
156 " is not recommended for open interchange";
157 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
158 " may not be portable";
159 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
160 " Unicode, requires a Perl extension," \
161 " and so is not portable";
163 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
165 if (flags & UNICODE_WARN_SURROGATE) { \
166 U32 category = packWARN(WARN_SURROGATE); \
167 const char * format = surrogate_cp_format; \
169 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
171 UNICODE_GOT_SURROGATE); \
174 Perl_ck_warner_d(aTHX_ category, format, uv); \
177 if (flags & UNICODE_DISALLOW_SURROGATE) { \
182 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
184 if (flags & UNICODE_WARN_NONCHAR) { \
185 U32 category = packWARN(WARN_NONCHAR); \
186 const char * format = nonchar_cp_format; \
188 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
190 UNICODE_GOT_NONCHAR); \
193 Perl_ck_warner_d(aTHX_ category, format, uv); \
196 if (flags & UNICODE_DISALLOW_NONCHAR) { \
201 /* Use shorter names internally in this file */
202 #define SHIFT UTF_ACCUMULATION_SHIFT
204 #define MARK UTF_CONTINUATION_MARK
205 #define MASK UTF_CONTINUATION_MASK
208 =for apidoc uvchr_to_utf8_flags_msgs
210 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
212 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
214 This function is for code that wants any warning and/or error messages to be
215 returned to the caller rather than be displayed. All messages that would have
216 been displayed if all lexcial warnings are enabled will be returned.
218 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
219 placed after all the others, C<msgs>. If this parameter is 0, this function
220 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
221 be a pointer to an C<HV *> variable, in which this function creates a new HV to
222 contain any appropriate messages. The hash has three key-value pairs, as
229 The text of the message as a C<SVpv>.
231 =item C<warn_categories>
233 The warning category (or categories) packed into a C<SVuv>.
237 A single flag bit associated with this message, in a C<SVuv>.
238 The bit corresponds to some bit in the C<*errors> return value,
239 such as C<UNICODE_GOT_SURROGATE>.
243 It's important to note that specifying this parameter as non-null will cause
244 any warnings this function would otherwise generate to be suppressed, and
245 instead be placed in C<*msgs>. The caller can check the lexical warnings state
246 (or not) when choosing what to do with the returned messages.
248 The caller, of course, is responsible for freeing any returned HV.
253 /* Undocumented; we don't want people using this. Instead they should use
254 * uvchr_to_utf8_flags_msgs() */
256 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
258 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
264 if (OFFUNI_IS_INVARIANT(uv)) {
265 *d++ = LATIN1_TO_NATIVE(uv);
269 if (uv <= MAX_UTF8_TWO_BYTE) {
270 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
271 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
275 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
276 * below, the 16 is for start bytes E0-EF (which are all the possible ones
277 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
278 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
279 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
280 * 0x800-0xFFFF on ASCII */
281 if (uv < (16 * (1U << (2 * SHIFT)))) {
282 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
283 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
284 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
286 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
287 aren't tested here */
288 /* The most likely code points in this range are below the surrogates.
289 * Do an extra test to quickly exclude those. */
290 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
291 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
292 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
294 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
296 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
297 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
304 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
305 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
306 * happen starting with 4-byte characters on ASCII platforms. We unify the
307 * code for these with EBCDIC, even though some of them require 5-bytes on
308 * those, because khw believes the code saving is worth the very slight
309 * performance hit on these high EBCDIC code points. */
311 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
312 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
313 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
315 if ( (flags & UNICODE_WARN_SUPER)
316 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
317 && UNICODE_IS_PERL_EXTENDED(uv)))
319 const char * format = super_cp_format;
320 U32 category = packWARN(WARN_NON_UNICODE);
321 U32 flag = UNICODE_GOT_SUPER;
323 /* Choose the more dire applicable warning */
324 if (UNICODE_IS_PERL_EXTENDED(uv)) {
325 format = perl_extended_cp_format;
326 if (flags & (UNICODE_WARN_PERL_EXTENDED
327 |UNICODE_DISALLOW_PERL_EXTENDED))
329 flag = UNICODE_GOT_PERL_EXTENDED;
334 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
338 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
341 if ( (flags & UNICODE_DISALLOW_SUPER)
342 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
343 && UNICODE_IS_PERL_EXTENDED(uv)))
348 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
349 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
352 /* Test for and handle 4-byte result. In the test immediately below, the
353 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
354 * characters). The 3 is for 3 continuation bytes; these each contribute
355 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
356 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
357 * 0x1_0000-0x1F_FFFF on ASCII */
358 if (uv < (8 * (1U << (3 * SHIFT)))) {
359 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
360 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
362 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
364 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
365 characters. The end-plane non-characters for EBCDIC were
366 handled just above */
367 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
368 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
370 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
371 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
378 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
379 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
380 * format. The unrolled version above turns out to not save all that much
381 * time, and at these high code points (well above the legal Unicode range
382 * on ASCII platforms, and well above anything in common use in EBCDIC),
383 * khw believes that less code outweighs slight performance gains. */
386 STRLEN len = OFFUNISKIP(uv);
389 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
390 uv >>= UTF_ACCUMULATION_SHIFT;
392 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
398 =for apidoc uvchr_to_utf8
400 Adds the UTF-8 representation of the native code point C<uv> to the end
401 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
402 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
403 the byte after the end of the new character. In other words,
405 d = uvchr_to_utf8(d, uv);
407 is the recommended wide native character-aware way of saying
411 This function accepts any code point from 0..C<IV_MAX> as input.
412 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
414 It is possible to forbid or warn on non-Unicode code points, or those that may
415 be problematic by using L</uvchr_to_utf8_flags>.
420 /* This is also a macro */
421 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
424 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
426 return uvchr_to_utf8(d, uv);
430 =for apidoc uvchr_to_utf8_flags
432 Adds the UTF-8 representation of the native code point C<uv> to the end
433 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
434 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
435 the byte after the end of the new character. In other words,
437 d = uvchr_to_utf8_flags(d, uv, flags);
441 d = uvchr_to_utf8_flags(d, uv, 0);
443 This is the Unicode-aware way of saying
447 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
448 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
450 Specifying C<flags> can further restrict what is allowed and not warned on, as
453 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
454 the function will raise a warning, provided UTF8 warnings are enabled. If
455 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
456 NULL. If both flags are set, the function will both warn and return NULL.
458 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
459 affect how the function handles a Unicode non-character.
461 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
462 affect the handling of code points that are above the Unicode maximum of
463 0x10FFFF. Languages other than Perl may not be able to accept files that
466 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
467 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
468 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
469 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
470 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
471 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
472 above-Unicode and surrogate flags, but not the non-character ones, as
474 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
475 See L<perlunicode/Noncharacter code points>.
477 Extremely high code points were never specified in any standard, and require an
478 extension to UTF-8 to express, which Perl does. It is likely that programs
479 written in something other than Perl would not be able to read files that
480 contain these; nor would Perl understand files written by something that uses a
481 different extension. For these reasons, there is a separate set of flags that
482 can warn and/or disallow these extremely high code points, even if other
483 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
484 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
485 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
486 treat all above-Unicode code points, including these, as malformations. (Note
487 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
488 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
490 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
491 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
492 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
493 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
494 platforms,these flags can apply to code points that actually do fit in 31 bits.
495 The new names accurately describe the situation in all cases.
500 /* This is also a macro */
501 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
504 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
506 return uvchr_to_utf8_flags(d, uv, flags);
512 S_is_utf8_cp_above_31_bits(const U8 * const s,
514 const bool consider_overlongs)
516 /* Returns TRUE if the first code point represented by the Perl-extended-
517 * UTF-8-encoded string starting at 's', and looking no further than 'e -
518 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
520 * The function handles the case where the input bytes do not include all
521 * the ones necessary to represent a full character. That is, they may be
522 * the intial bytes of the representation of a code point, but possibly
523 * the final ones necessary for the complete representation may be beyond
526 * The function also can handle the case where the input is an overlong
527 * sequence. If 'consider_overlongs' is 0, the function assumes the
528 * input is not overlong, without checking, and will return based on that
529 * assumption. If this parameter is 1, the function will go to the trouble
530 * of figuring out if it actually evaluates to above or below 31 bits.
532 * The sequence is otherwise assumed to be well-formed, without checking.
535 const STRLEN len = e - s;
538 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
540 assert(! UTF8_IS_INVARIANT(*s) && e > s);
544 PERL_UNUSED_ARG(consider_overlongs);
546 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
547 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
548 * also be the start byte for a 31-bit code point; we need at least 2
549 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
550 * the start byte for an overlong sequence, but for 30-bit or smaller code
551 * points, so we don't have to worry about overlongs on EBCDIC.) */
562 /* On ASCII, FE and FF are the only start bytes that can evaluate to
563 * needing more than 31 bits. */
564 if (LIKELY(*s < 0xFE)) {
568 /* What we have left are FE and FF. Both of these require more than 31
569 * bits unless they are for overlongs. */
570 if (! consider_overlongs) {
574 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
575 * above 31 bits. But we need more than one byte to discern this, so if
576 * passed just the start byte, it could be an overlong evaluating to
582 /* Having excluded len==1, and knowing that FE and FF are both valid start
583 * bytes, we can call the function below to see if the sequence is
584 * overlong. (We don't need the full generality of the called function,
585 * but for these huge code points, speed shouldn't be a consideration, and
586 * the compiler does have enough information, since it's static to this
587 * file, to optimize to just the needed parts.) */
588 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
590 /* If it isn't overlong, more than 31 bits are required. */
591 if (is_overlong == 0) {
595 /* If it is indeterminate if it is overlong, return that */
596 if (is_overlong < 0) {
600 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
601 * the max it can be is 2**31 - 1 */
608 /* Here, ASCII and EBCDIC rejoin:
609 * On ASCII: We have an overlong sequence starting with FF
610 * On EBCDIC: We have a sequence starting with FE. */
612 { /* For C89, use a block so the declaration can be close to its use */
616 /* U+7FFFFFFF (2 ** 31 - 1)
617 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
618 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
619 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
620 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
621 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
622 * U+80000000 (2 ** 31):
623 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
624 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
625 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
626 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
628 * and since we know that *s = \xfe, any continuation sequcence
629 * following it that is gt the below is above 31 bits
630 [0] [1] [2] [3] [4] [5] [6] */
631 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
635 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
636 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
637 * FF overlong for U+80000000 (2 ** 31):
638 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
639 * and since we know that *s = \xff, any continuation sequcence
640 * following it that is gt the below is above 30 bits
641 [0] [1] [2] [3] [4] [5] [6] */
642 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
646 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
647 const STRLEN cmp_len = MIN(conts_len, len - 1);
649 /* Now compare the continuation bytes in s with the ones we have
650 * compiled in that are for the largest 30 bit code point. If we have
651 * enough bytes available to determine the answer, or the bytes we do
652 * have differ from them, we can compare the two to get a definitive
653 * answer (Note that in UTF-EBCDIC, the two lowest possible
654 * continuation bytes are \x41 and \x42.) */
655 if (cmp_len >= conts_len || memNE(s + 1,
656 conts_for_highest_30_bit,
659 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
662 /* Here, all the bytes we have are the same as the highest 30-bit code
663 * point, but we are missing so many bytes that we can't make the
671 PERL_STATIC_INLINE int
672 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
674 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
675 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
676 * it isn't, and -1 if there isn't enough information to tell. This last
677 * return value can happen if the sequence is incomplete, missing some
678 * trailing bytes that would form a complete character. If there are
679 * enough bytes to make a definitive decision, this function does so.
680 * Usually 2 bytes sufficient.
682 * Overlongs can occur whenever the number of continuation bytes changes.
683 * That means whenever the number of leading 1 bits in a start byte
684 * increases from the next lower start byte. That happens for start bytes
685 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
686 * start bytes have already been excluded, so don't need to be tested here;
687 * ASCII platforms: C0, C1
688 * EBCDIC platforms C0, C1, C2, C3, C4, E0
691 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
692 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
694 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
695 assert(len > 1 && UTF8_IS_START(*s));
697 /* Each platform has overlongs after the start bytes given above (expressed
698 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
699 * the logic is the same, except the E0 overlong has already been excluded
700 * on EBCDIC platforms. The values below were found by manually
701 * inspecting the UTF-8 patterns. See the tables in utf8.h and
705 # define F0_ABOVE_OVERLONG 0xB0
706 # define F8_ABOVE_OVERLONG 0xA8
707 # define FC_ABOVE_OVERLONG 0xA4
708 # define FE_ABOVE_OVERLONG 0xA2
709 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
713 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
717 # define F0_ABOVE_OVERLONG 0x90
718 # define F8_ABOVE_OVERLONG 0x88
719 # define FC_ABOVE_OVERLONG 0x84
720 # define FE_ABOVE_OVERLONG 0x82
721 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
725 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
726 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
727 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
728 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
733 /* Check for the FF overlong */
734 return isFF_OVERLONG(s, len);
737 PERL_STATIC_INLINE int
738 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
740 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
741 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
742 * it isn't, and -1 if there isn't enough information to tell. This last
743 * return value can happen if the sequence is incomplete, missing some
744 * trailing bytes that would form a complete character. If there are
745 * enough bytes to make a definitive decision, this function does so. */
747 PERL_ARGS_ASSERT_ISFF_OVERLONG;
749 /* To be an FF overlong, all the available bytes must match */
750 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
751 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
756 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
757 * be there; what comes after them doesn't matter. See tables in utf8.h,
759 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
763 /* The missing bytes could cause the result to go one way or the other, so
764 * the result is indeterminate */
768 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
769 # ifdef EBCDIC /* Actually is I8 */
770 # define HIGHEST_REPRESENTABLE_UTF8 \
771 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
773 # define HIGHEST_REPRESENTABLE_UTF8 \
774 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
778 PERL_STATIC_INLINE int
779 S_does_utf8_overflow(const U8 * const s,
781 const bool consider_overlongs)
783 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
784 * 'e' - 1 would overflow an IV on this platform; that is if it represents
785 * a code point larger than the highest representable code point. It
786 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
787 * enough information to tell. This last return value can happen if the
788 * sequence is incomplete, missing some trailing bytes that would form a
789 * complete character. If there are enough bytes to make a definitive
790 * decision, this function does so.
792 * If 'consider_overlongs' is TRUE, the function checks for the possibility
793 * that the sequence is an overlong that doesn't overflow. Otherwise, it
794 * assumes the sequence is not an overlong. This can give different
795 * results only on ASCII 32-bit platforms.
797 * (For ASCII platforms, we could use memcmp() because we don't have to
798 * convert each byte to I8, but it's very rare input indeed that would
799 * approach overflow, so the loop below will likely only get executed once.)
801 * 'e' - 1 must not be beyond a full character. */
804 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
805 assert(s <= e && s + UTF8SKIP(s) >= e);
807 #if ! defined(UV_IS_QUAD)
809 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
813 PERL_UNUSED_ARG(consider_overlongs);
816 const STRLEN len = e - s;
818 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
820 for (x = s; x < e; x++, y++) {
822 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
826 /* If this byte is larger than the corresponding highest UTF-8
827 * byte, the sequence overflow; otherwise the byte is less than,
828 * and so the sequence doesn't overflow */
829 return NATIVE_UTF8_TO_I8(*x) > *y;
833 /* Got to the end and all bytes are the same. If the input is a whole
834 * character, it doesn't overflow. And if it is a partial character,
835 * there's not enough information to tell */
836 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
849 /* This is the portions of the above function that deal with UV_MAX instead of
850 * IV_MAX. They are left here in case we want to combine them so that internal
851 * uses can have larger code points. The only logic difference is that the
852 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
856 /* Anything larger than this will overflow the word if it were converted into a UV */
857 #if defined(UV_IS_QUAD)
858 # ifdef EBCDIC /* Actually is I8 */
859 # define HIGHEST_REPRESENTABLE_UTF8 \
860 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
862 # define HIGHEST_REPRESENTABLE_UTF8 \
863 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
867 # define HIGHEST_REPRESENTABLE_UTF8 \
868 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
870 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
874 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
876 /* On 32 bit ASCII machines, many overlongs that start with FF don't
878 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
880 /* To be such an overlong, the first bytes of 's' must match
881 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
882 * don't have any additional bytes available, the sequence, when
883 * completed might or might not fit in 32 bits. But if we have that
884 * next byte, we can tell for sure. If it is <= 0x83, then it does
886 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
890 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
893 /* Starting with the #else, the rest of the function is identical except
894 * 1. we need to move the 'len' declaration to be global to the function
895 * 2. the endif move to just after the UNUSED_ARG.
896 * An empty endif is given just below to satisfy the preprocessor
902 #undef F0_ABOVE_OVERLONG
903 #undef F8_ABOVE_OVERLONG
904 #undef FC_ABOVE_OVERLONG
905 #undef FE_ABOVE_OVERLONG
906 #undef FF_OVERLONG_PREFIX
909 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
914 /* A helper function that should not be called directly.
916 * This function returns non-zero if the string beginning at 's' and
917 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
918 * code point; otherwise it returns 0. The examination stops after the
919 * first code point in 's' is validated, not looking at the rest of the
920 * input. If 'e' is such that there are not enough bytes to represent a
921 * complete code point, this function will return non-zero anyway, if the
922 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
923 * excluded by 'flags'.
925 * A non-zero return gives the number of bytes required to represent the
926 * code point. Be aware that if the input is for a partial character, the
927 * return will be larger than 'e - s'.
929 * This function assumes that the code point represented is UTF-8 variant.
930 * The caller should have excluded the possibility of it being invariant
931 * before calling this function.
933 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
934 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
935 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
936 * disallowed by the flags. If the input is only for a partial character,
937 * the function will return non-zero if there is any sequence of
938 * well-formed UTF-8 that, when appended to the input sequence, could
939 * result in an allowed code point; otherwise it returns 0. Non characters
940 * cannot be determined based on partial character input. But many of the
941 * other excluded types can be determined with just the first one or two
946 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
948 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
949 |UTF8_DISALLOW_PERL_EXTENDED)));
950 assert(! UTF8_IS_INVARIANT(*s));
952 /* A variant char must begin with a start byte */
953 if (UNLIKELY(! UTF8_IS_START(*s))) {
957 /* Examine a maximum of a single whole code point */
958 if (e - s > UTF8SKIP(s)) {
964 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
965 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
967 /* Here, we are disallowing some set of largish code points, and the
968 * first byte indicates the sequence is for a code point that could be
969 * in the excluded set. We generally don't have to look beyond this or
970 * the second byte to see if the sequence is actually for one of the
971 * excluded classes. The code below is derived from this table:
973 * UTF-8 UTF-EBCDIC I8
974 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
975 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
976 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
978 * Keep in mind that legal continuation bytes range between \x80..\xBF
979 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
980 * continuation bytes. Hence, we don't have to test the upper edge
981 * because if any of those is encountered, the sequence is malformed,
982 * and would fail elsewhere in this function.
984 * The code here likewise assumes that there aren't other
985 * malformations; again the function should fail elsewhere because of
986 * these. For example, an overlong beginning with FC doesn't actually
987 * have to be a super; it could actually represent a small code point,
988 * even U+0000. But, since overlongs (and other malformations) are
989 * illegal, the function should return FALSE in either case.
992 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
993 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
994 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
996 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
998 && ((s1) & 0xFE ) == 0xB6)
999 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1001 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1002 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1003 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1004 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1007 if ( (flags & UTF8_DISALLOW_SUPER)
1008 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1010 return 0; /* Above Unicode */
1013 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1014 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1020 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1022 if ( (flags & UTF8_DISALLOW_SUPER)
1023 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1025 return 0; /* Above Unicode */
1028 if ( (flags & UTF8_DISALLOW_SURROGATE)
1029 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1031 return 0; /* Surrogate */
1034 if ( (flags & UTF8_DISALLOW_NONCHAR)
1035 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1037 return 0; /* Noncharacter code point */
1042 /* Make sure that all that follows are continuation bytes */
1043 for (x = s + 1; x < e; x++) {
1044 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1049 /* Here is syntactically valid. Next, make sure this isn't the start of an
1051 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1055 /* And finally, that the code point represented fits in a word on this
1057 if (0 < does_utf8_overflow(s, e,
1058 0 /* Don't consider overlongs */
1068 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1070 /* Returns a mortalized C string that is a displayable copy of the 'len'
1071 * bytes starting at 'start'. 'format' gives how to display each byte.
1072 * Currently, there are only two formats, so it is currently a bool:
1074 * 1 ab (that is a space between two hex digit bytes)
1077 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1079 const U8 * s = start;
1080 const U8 * const e = start + len;
1084 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1086 Newx(output, output_len, char);
1090 for (s = start; s < e; s++) {
1091 const unsigned high_nibble = (*s & 0xF0) >> 4;
1092 const unsigned low_nibble = (*s & 0x0F);
1104 if (high_nibble < 10) {
1105 *d++ = high_nibble + '0';
1108 *d++ = high_nibble - 10 + 'a';
1111 if (low_nibble < 10) {
1112 *d++ = low_nibble + '0';
1115 *d++ = low_nibble - 10 + 'a';
1123 PERL_STATIC_INLINE char *
1124 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1126 /* How many bytes to print */
1129 /* Which one is the non-continuation */
1130 const STRLEN non_cont_byte_pos,
1132 /* How many bytes should there be? */
1133 const STRLEN expect_len)
1135 /* Return the malformation warning text for an unexpected continuation
1138 const char * const where = (non_cont_byte_pos == 1)
1140 : Perl_form(aTHX_ "%d bytes",
1141 (int) non_cont_byte_pos);
1143 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1145 /* We don't need to pass this parameter, but since it has already been
1146 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1147 assert(expect_len == UTF8SKIP(s));
1149 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1150 " %s after start byte 0x%02x; need %d bytes, got %d)",
1152 _byte_dump_string(s, print_len, 0),
1153 *(s + non_cont_byte_pos),
1157 (int) non_cont_byte_pos);
1162 =for apidoc utf8n_to_uvchr
1164 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1165 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1167 Bottom level UTF-8 decode routine.
1168 Returns the native code point value of the first character in the string C<s>,
1169 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1170 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1171 the length, in bytes, of that character.
1173 The value of C<flags> determines the behavior when C<s> does not point to a
1174 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1175 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1176 is the next possible position in C<s> that could begin a non-malformed
1177 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1178 is raised. Some UTF-8 input sequences may contain multiple malformations.
1179 This function tries to find every possible one in each call, so multiple
1180 warnings can be raised for the same sequence.
1182 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1183 individual types of malformations, such as the sequence being overlong (that
1184 is, when there is a shorter sequence that can express the same code point;
1185 overlong sequences are expressly forbidden in the UTF-8 standard due to
1186 potential security issues). Another malformation example is the first byte of
1187 a character not being a legal first byte. See F<utf8.h> for the list of such
1188 flags. Even if allowed, this function generally returns the Unicode
1189 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1190 F<utf8.h> to override this behavior for the overlong malformations, but don't
1191 do that except for very specialized purposes.
1193 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1194 flags) malformation is found. If this flag is set, the routine assumes that
1195 the caller will raise a warning, and this function will silently just set
1196 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1198 Note that this API requires disambiguation between successful decoding a C<NUL>
1199 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1200 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1201 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1202 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1203 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1205 Certain code points are considered problematic. These are Unicode surrogates,
1206 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1207 By default these are considered regular code points, but certain situations
1208 warrant special handling for them, which can be specified using the C<flags>
1209 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1210 three classes are treated as malformations and handled as such. The flags
1211 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1212 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1213 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1214 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1215 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1217 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1218 The difference between traditional strictness and C9 strictness is that the
1219 latter does not forbid non-character code points. (They are still discouraged,
1220 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1222 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1223 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1224 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1225 raised for their respective categories, but otherwise the code points are
1226 considered valid (not malformations). To get a category to both be treated as
1227 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1228 (But note that warnings are not raised if lexically disabled nor if
1229 C<UTF8_CHECK_ONLY> is also specified.)
1231 Extremely high code points were never specified in any standard, and require an
1232 extension to UTF-8 to express, which Perl does. It is likely that programs
1233 written in something other than Perl would not be able to read files that
1234 contain these; nor would Perl understand files written by something that uses a
1235 different extension. For these reasons, there is a separate set of flags that
1236 can warn and/or disallow these extremely high code points, even if other
1237 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1238 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1239 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1240 above-Unicode code points, including these, as malformations.
1241 (Note that the Unicode standard considers anything above 0x10FFFF to be
1242 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1245 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1246 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1247 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1248 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1249 can apply to code points that actually do fit in 31 bits. This happens on
1250 EBCDIC platforms, and sometimes when the L<overlong
1251 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1252 describe the situation in all cases.
1255 All other code points corresponding to Unicode characters, including private
1256 use and those yet to be assigned, are never considered malformed and never
1261 Also implemented as a macro in utf8.h
1265 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1270 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1272 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1275 /* The tables below come from http://bjoern.hoehrmann.de/utf-8/decoder/dfa/,
1276 * which requires this copyright notice */
1278 /* Copyright (c) 2008-2009 Bjoern Hoehrmann <bjoern@hoehrmann.de>
1280 Permission is hereby granted, free of charge, to any person obtaining a copy of
1281 this software and associated documentation files (the "Software"), to deal in
1282 the Software without restriction, including without limitation the rights to
1283 use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies
1284 of the Software, and to permit persons to whom the Software is furnished to do
1285 so, subject to the following conditions:
1287 The above copyright notice and this permission notice shall be included in all
1288 copies or substantial portions of the Software.
1290 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
1291 IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
1292 FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
1293 AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
1294 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
1295 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
1301 static U8 utf8d_C9[] = {
1302 /* The first part of the table maps bytes to character classes that
1303 * to reduce the size of the transition table and create bitmasks. */
1304 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1305 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1306 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1307 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1308 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1309 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1310 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1311 10,3,3,3,3,3,3,3,3,3,3,3,3,4,3,3, 11,6,6,6,5,8,8,8,8,8,8,8,8,8,8,8, /*-FF*/
1313 /* The second part is a transition table that maps a combination
1314 * of a state of the automaton and a character class to a state. */
1315 0,12,24,36,60,96,84,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,
1316 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,
1317 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,
1318 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,
1319 12,36,12,12,12,12,12,12,12,12,12,12
1326 /* This is a version of the above table customized for Perl that doesn't
1327 * exclude surrogates and accepts start bytes up through F7 (representing
1329 static U8 dfa_tab_for_perl[] = {
1330 /* The first part of the table maps bytes to character classes to reduce
1331 * the size of the transition table and create bitmasks. */
1332 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-1F*/
1333 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-3F*/
1334 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-5F*/
1335 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0, /*-7F*/
1336 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, 9,9,9,9,9,9,9,9,9,9,9,9,9,9,9,9, /*-9F*/
1337 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7, /*-BF*/
1338 8,8,2,2,2,2,2,2,2,2,2,2,2,2,2,2, 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /*-DF*/
1339 10,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, 11,4,4,4,4,4,4,4,8,8,8,8,8,8,8,8, /*-FF*/
1341 /* The second part is a transition table that maps a combination
1342 * of a state of the automaton and a character class to a state. */
1343 0,12,24,36,96,12,12,12,12,12,48,72, 12,12,12,12,12,12,12,12,12,12,12,12,/*23*/
1344 12, 0,12,12,12,12,12, 0,12, 0,12,12, 12,24,12,12,12,12,12,24,12,24,12,12,/*47*/
1345 12,12,12,12,12,12,12,24,12,12,12,12, 12,24,12,12,12,12,12,12,12,24,12,12,/*71*/
1346 12,12,12,12,12,12,12,36,12,36,12,12, 12,36,12,12,12,12,12,36,12,36,12,12,/*95*/
1347 12,36,12,12,12,12,12,36,12,36,12,12 /* 96- 107 */
1349 /* The customization was to repurpose the surrogates type '4' to instead be
1350 * for start bytes F1-F7. Types 5 and 6 are now unused, and their entries in
1351 * the transition part of the table are set to 12, so are illegal.
1353 * To do higher code points would require expansion and some rearrangement of
1354 * the table. The type '1' entries for continuation bytes 80-8f would have to
1355 * be split into several types, because they aren't treated uniformly for
1356 * higher start bytes, since overlongs for F8 are 80-87; FC: 80-83; and FE:
1357 * 80-81. We start needing to worry about overflow if FE is included.
1358 * Ignoring, FE and FF, we could use type 5 for F9-FB, and 6 for FD (remember
1359 * from the web site that these are used to right shift). FE would
1360 * necessarily be type 7; and FF, type 8. And new states would have to be
1361 * created for F8 and FC (and FE and FF if used), so quite a bit of work would
1364 * XXX Better would be to customize the table so that the noncharacters are
1365 * excluded. This again is non trivial, but doing so would simplify the code
1366 * that uses this, and might make it small enough to make it inlinable */
1373 =for apidoc utf8n_to_uvchr_error
1375 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1376 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1378 This function is for code that needs to know what the precise malformation(s)
1379 are when an error is found. If you also need to know the generated warning
1380 messages, use L</utf8n_to_uvchr_msgs>() instead.
1382 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1383 all the others, C<errors>. If this parameter is 0, this function behaves
1384 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1385 to a C<U32> variable, which this function sets to indicate any errors found.
1386 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1387 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1388 of these bits will be set if a malformation is found, even if the input
1389 C<flags> parameter indicates that the given malformation is allowed; those
1390 exceptions are noted:
1394 =item C<UTF8_GOT_PERL_EXTENDED>
1396 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1397 set only if the input C<flags> parameter contains either the
1398 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1400 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1401 and so some extension must be used to express them. Perl uses a natural
1402 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1403 extension to represent even higher ones, so that any code point that fits in a
1404 64-bit word can be represented. Text using these extensions is not likely to
1405 be portable to non-Perl code. We lump both of these extensions together and
1406 refer to them as Perl extended UTF-8. There exist other extensions that people
1407 have invented, incompatible with Perl's.
1409 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1410 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1411 than on ASCII. Prior to that, code points 2**31 and higher were simply
1412 unrepresentable, and a different, incompatible method was used to represent
1413 code points between 2**30 and 2**31 - 1.
1415 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1416 Perl extended UTF-8 is used.
1418 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1419 may use for backward compatibility. That name is misleading, as this flag may
1420 be set when the code point actually does fit in 31 bits. This happens on
1421 EBCDIC platforms, and sometimes when the L<overlong
1422 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1423 describes the situation in all cases.
1425 =item C<UTF8_GOT_CONTINUATION>
1427 The input sequence was malformed in that the first byte was a a UTF-8
1430 =item C<UTF8_GOT_EMPTY>
1432 The input C<curlen> parameter was 0.
1434 =item C<UTF8_GOT_LONG>
1436 The input sequence was malformed in that there is some other sequence that
1437 evaluates to the same code point, but that sequence is shorter than this one.
1439 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1440 it was discovered that this created security issues.
1442 =item C<UTF8_GOT_NONCHAR>
1444 The code point represented by the input UTF-8 sequence is for a Unicode
1445 non-character code point.
1446 This bit is set only if the input C<flags> parameter contains either the
1447 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1449 =item C<UTF8_GOT_NON_CONTINUATION>
1451 The input sequence was malformed in that a non-continuation type byte was found
1452 in a position where only a continuation type one should be.
1454 =item C<UTF8_GOT_OVERFLOW>
1456 The input sequence was malformed in that it is for a code point that is not
1457 representable in the number of bits available in an IV on the current platform.
1459 =item C<UTF8_GOT_SHORT>
1461 The input sequence was malformed in that C<curlen> is smaller than required for
1462 a complete sequence. In other words, the input is for a partial character
1465 =item C<UTF8_GOT_SUPER>
1467 The input sequence was malformed in that it is for a non-Unicode code point;
1468 that is, one above the legal Unicode maximum.
1469 This bit is set only if the input C<flags> parameter contains either the
1470 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1472 =item C<UTF8_GOT_SURROGATE>
1474 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1476 This bit is set only if the input C<flags> parameter contains either the
1477 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1481 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1482 flag to suppress any warnings, and then examine the C<*errors> return.
1486 Also implemented as a macro in utf8.h
1490 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1496 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1498 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1503 =for apidoc utf8n_to_uvchr_msgs
1505 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1506 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1508 This function is for code that needs to know what the precise malformation(s)
1509 are when an error is found, and wants the corresponding warning and/or error
1510 messages to be returned to the caller rather than be displayed. All messages
1511 that would have been displayed if all lexcial warnings are enabled will be
1514 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1515 placed after all the others, C<msgs>. If this parameter is 0, this function
1516 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1517 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1518 contain any appropriate messages. The elements of the array are ordered so
1519 that the first message that would have been displayed is in the 0th element,
1520 and so on. Each element is a hash with three key-value pairs, as follows:
1526 The text of the message as a C<SVpv>.
1528 =item C<warn_categories>
1530 The warning category (or categories) packed into a C<SVuv>.
1534 A single flag bit associated with this message, in a C<SVuv>.
1535 The bit corresponds to some bit in the C<*errors> return value,
1536 such as C<UTF8_GOT_LONG>.
1540 It's important to note that specifying this parameter as non-null will cause
1541 any warnings this function would otherwise generate to be suppressed, and
1542 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1543 (or not) when choosing what to do with the returned messages.
1545 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1548 The caller, of course, is responsible for freeing any returned AV.
1554 Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1561 const U8 * const s0 = s;
1562 const U8 * send = s0 + curlen;
1563 U32 possible_problems = 0; /* A bit is set here for each potential problem
1564 found as we go along */
1566 STRLEN expectlen = 0; /* How long should this sequence be?
1567 (initialized to silence compilers' wrong
1569 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1570 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1571 this gets set and discarded */
1573 /* The below are used only if there is both an overlong malformation and a
1574 * too short one. Otherwise the first two are set to 's0' and 'send', and
1575 * the third not used at all */
1576 U8 * adjusted_s0 = (U8 *) s0;
1577 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1578 routine; see [perl #130921] */
1579 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1583 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
1589 errors = &discard_errors;
1592 /* The order of malformation tests here is important. We should consume as
1593 * few bytes as possible in order to not skip any valid character. This is
1594 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1595 * http://unicode.org/reports/tr36 for more discussion as to why. For
1596 * example, once we've done a UTF8SKIP, we can tell the expected number of
1597 * bytes, and could fail right off the bat if the input parameters indicate
1598 * that there are too few available. But it could be that just that first
1599 * byte is garbled, and the intended character occupies fewer bytes. If we
1600 * blindly assumed that the first byte is correct, and skipped based on
1601 * that number, we could skip over a valid input character. So instead, we
1602 * always examine the sequence byte-by-byte.
1604 * We also should not consume too few bytes, otherwise someone could inject
1605 * things. For example, an input could be deliberately designed to
1606 * overflow, and if this code bailed out immediately upon discovering that,
1607 * returning to the caller C<*retlen> pointing to the very next byte (one
1608 * which is actually part of of the overflowing sequence), that could look
1609 * legitimate to the caller, which could discard the initial partial
1610 * sequence and process the rest, inappropriately.
1612 * Some possible input sequences are malformed in more than one way. This
1613 * function goes to lengths to try to find all of them. This is necessary
1614 * for correctness, as the inputs may allow one malformation but not
1615 * another, and if we abandon searching for others after finding the
1616 * allowed one, we could allow in something that shouldn't have been.
1619 if (UNLIKELY(curlen == 0)) {
1620 possible_problems |= UTF8_GOT_EMPTY;
1622 uv = UNICODE_REPLACEMENT;
1623 goto ready_to_handle_errors;
1626 expectlen = UTF8SKIP(s);
1628 /* A well-formed UTF-8 character, as the vast majority of calls to this
1629 * function will be for, has this expected length. For efficiency, set
1630 * things up here to return it. It will be overriden only in those rare
1631 * cases where a malformation is found */
1633 *retlen = expectlen;
1636 /* An invariant is trivially well-formed */
1637 if (UTF8_IS_INVARIANT(*s0)) {
1643 /* Measurements show that this dfa is somewhat faster than the regular code
1644 * below, so use it first, dropping down for the non-normal cases. */
1646 # define PERL_UTF8_DECODE_REJECT 12
1648 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
1649 UV type = dfa_tab_for_perl[*s];
1652 uv = (*s & 0x3fu) | (uv << UTF_ACCUMULATION_SHIFT);
1653 state = dfa_tab_for_perl[256 + state + type];
1656 uv = (0xff >> type) & (*s);
1657 state = dfa_tab_for_perl[256 + type];
1662 /* If this could be a code point that the flags don't allow (the first
1663 * surrogate is the first such possible one), delve further, but we already
1664 * have calculated 'uv' */
1665 if ( (flags & (UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1666 |UTF8_WARN_ILLEGAL_INTERCHANGE))
1667 && uv >= UNICODE_SURROGATE_FIRST)
1669 curlen = s + 1 - s0;
1679 /* Here, is some sort of failure. Use the full mechanism */
1685 /* A continuation character can't start a valid sequence */
1686 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1687 possible_problems |= UTF8_GOT_CONTINUATION;
1689 uv = UNICODE_REPLACEMENT;
1690 goto ready_to_handle_errors;
1693 /* Here is not a continuation byte, nor an invariant. The only thing left
1694 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1695 * because it excludes start bytes like \xC0 that always lead to
1698 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1699 * that indicate the number of bytes in the character's whole UTF-8
1700 * sequence, leaving just the bits that are part of the value. */
1701 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1703 /* Setup the loop end point, making sure to not look past the end of the
1704 * input string, and flag it as too short if the size isn't big enough. */
1705 if (UNLIKELY(curlen < expectlen)) {
1706 possible_problems |= UTF8_GOT_SHORT;
1710 send = (U8*) s0 + expectlen;
1713 /* Now, loop through the remaining bytes in the character's sequence,
1714 * accumulating each into the working value as we go. */
1715 for (s = s0 + 1; s < send; s++) {
1716 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1717 uv = UTF8_ACCUMULATE(uv, *s);
1721 /* Here, found a non-continuation before processing all expected bytes.
1722 * This byte indicates the beginning of a new character, so quit, even
1723 * if allowing this malformation. */
1724 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1726 } /* End of loop through the character's bytes */
1728 /* Save how many bytes were actually in the character */
1731 /* Note that there are two types of too-short malformation. One is when
1732 * there is actual wrong data before the normal termination of the
1733 * sequence. The other is that the sequence wasn't complete before the end
1734 * of the data we are allowed to look at, based on the input 'curlen'.
1735 * This means that we were passed data for a partial character, but it is
1736 * valid as far as we saw. The other is definitely invalid. This
1737 * distinction could be important to a caller, so the two types are kept
1740 * A convenience macro that matches either of the too-short conditions. */
1741 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1743 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1745 uv = UNICODE_REPLACEMENT;
1748 /* Check for overflow. The algorithm requires us to not look past the end
1749 * of the current character, even if partial, so the upper limit is 's' */
1750 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1751 1 /* Do consider overlongs */
1754 possible_problems |= UTF8_GOT_OVERFLOW;
1755 uv = UNICODE_REPLACEMENT;
1758 /* Check for overlong. If no problems so far, 'uv' is the correct code
1759 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1760 * we must look at the UTF-8 byte sequence itself to see if it is for an
1762 if ( ( LIKELY(! possible_problems)
1763 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1764 || ( UNLIKELY(possible_problems)
1765 && ( UNLIKELY(! UTF8_IS_START(*s0))
1767 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1770 possible_problems |= UTF8_GOT_LONG;
1772 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1774 /* The calculation in the 'true' branch of this 'if'
1775 * below won't work if overflows, and isn't needed
1776 * anyway. Further below we handle all overflow
1778 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1780 UV min_uv = uv_so_far;
1783 /* Here, the input is both overlong and is missing some trailing
1784 * bytes. There is no single code point it could be for, but there
1785 * may be enough information present to determine if what we have
1786 * so far is for an unallowed code point, such as for a surrogate.
1787 * The code further below has the intelligence to determine this,
1788 * but just for non-overlong UTF-8 sequences. What we do here is
1789 * calculate the smallest code point the input could represent if
1790 * there were no too short malformation. Then we compute and save
1791 * the UTF-8 for that, which is what the code below looks at
1792 * instead of the raw input. It turns out that the smallest such
1793 * code point is all we need. */
1794 for (i = curlen; i < expectlen; i++) {
1795 min_uv = UTF8_ACCUMULATE(min_uv,
1796 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1799 adjusted_s0 = temp_char_buf;
1800 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1806 /* Here, we have found all the possible problems, except for when the input
1807 * is for a problematic code point not allowed by the input parameters. */
1809 /* uv is valid for overlongs */
1810 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1812 /* isn't problematic if < this */
1813 && uv >= UNICODE_SURROGATE_FIRST)
1814 || ( UNLIKELY(possible_problems)
1816 /* if overflow, we know without looking further
1817 * precisely which of the problematic types it is,
1818 * and we deal with those in the overflow handling
1820 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1821 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1822 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1823 && ((flags & ( UTF8_DISALLOW_NONCHAR
1824 |UTF8_DISALLOW_SURROGATE
1825 |UTF8_DISALLOW_SUPER
1826 |UTF8_DISALLOW_PERL_EXTENDED
1828 |UTF8_WARN_SURROGATE
1830 |UTF8_WARN_PERL_EXTENDED))))
1832 /* If there were no malformations, or the only malformation is an
1833 * overlong, 'uv' is valid */
1834 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1835 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1836 possible_problems |= UTF8_GOT_SURROGATE;
1838 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1839 possible_problems |= UTF8_GOT_SUPER;
1841 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1842 possible_problems |= UTF8_GOT_NONCHAR;
1845 else { /* Otherwise, need to look at the source UTF-8, possibly
1846 adjusted to be non-overlong */
1848 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1849 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1851 possible_problems |= UTF8_GOT_SUPER;
1853 else if (curlen > 1) {
1854 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1855 NATIVE_UTF8_TO_I8(*adjusted_s0),
1856 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1858 possible_problems |= UTF8_GOT_SUPER;
1860 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1861 NATIVE_UTF8_TO_I8(*adjusted_s0),
1862 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1864 possible_problems |= UTF8_GOT_SURROGATE;
1868 /* We need a complete well-formed UTF-8 character to discern
1869 * non-characters, so can't look for them here */
1873 ready_to_handle_errors:
1876 * curlen contains the number of bytes in the sequence that
1877 * this call should advance the input by.
1878 * avail_len gives the available number of bytes passed in, but
1879 * only if this is less than the expected number of
1880 * bytes, based on the code point's start byte.
1881 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1882 * is set in it for each potential problem found.
1883 * uv contains the code point the input sequence
1884 * represents; or if there is a problem that prevents
1885 * a well-defined value from being computed, it is
1886 * some subsitute value, typically the REPLACEMENT
1888 * s0 points to the first byte of the character
1889 * s points to just after were we left off processing
1891 * send points to just after where that character should
1892 * end, based on how many bytes the start byte tells
1893 * us should be in it, but no further than s0 +
1897 if (UNLIKELY(possible_problems)) {
1898 bool disallowed = FALSE;
1899 const U32 orig_problems = possible_problems;
1905 while (possible_problems) { /* Handle each possible problem */
1907 char * message = NULL;
1908 U32 this_flag_bit = 0;
1910 /* Each 'if' clause handles one problem. They are ordered so that
1911 * the first ones' messages will be displayed before the later
1912 * ones; this is kinda in decreasing severity order. But the
1913 * overlong must come last, as it changes 'uv' looked at by the
1915 if (possible_problems & UTF8_GOT_OVERFLOW) {
1917 /* Overflow means also got a super and are using Perl's
1918 * extended UTF-8, but we handle all three cases here */
1920 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1921 *errors |= UTF8_GOT_OVERFLOW;
1923 /* But the API says we flag all errors found */
1924 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1925 *errors |= UTF8_GOT_SUPER;
1928 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1930 *errors |= UTF8_GOT_PERL_EXTENDED;
1933 /* Disallow if any of the three categories say to */
1934 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1935 || (flags & ( UTF8_DISALLOW_SUPER
1936 |UTF8_DISALLOW_PERL_EXTENDED)))
1941 /* Likewise, warn if any say to */
1942 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1943 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1946 /* The warnings code explicitly says it doesn't handle the
1947 * case of packWARN2 and two categories which have
1948 * parent-child relationship. Even if it works now to
1949 * raise the warning if either is enabled, it wouldn't
1950 * necessarily do so in the future. We output (only) the
1951 * most dire warning */
1952 if (! (flags & UTF8_CHECK_ONLY)) {
1953 if (msgs || ckWARN_d(WARN_UTF8)) {
1954 pack_warn = packWARN(WARN_UTF8);
1956 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1957 pack_warn = packWARN(WARN_NON_UNICODE);
1960 message = Perl_form(aTHX_ "%s: %s (overflows)",
1962 _byte_dump_string(s0, curlen, 0));
1963 this_flag_bit = UTF8_GOT_OVERFLOW;
1968 else if (possible_problems & UTF8_GOT_EMPTY) {
1969 possible_problems &= ~UTF8_GOT_EMPTY;
1970 *errors |= UTF8_GOT_EMPTY;
1972 if (! (flags & UTF8_ALLOW_EMPTY)) {
1974 /* This so-called malformation is now treated as a bug in
1975 * the caller. If you have nothing to decode, skip calling
1981 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1983 pack_warn = packWARN(WARN_UTF8);
1984 message = Perl_form(aTHX_ "%s (empty string)",
1986 this_flag_bit = UTF8_GOT_EMPTY;
1990 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1991 possible_problems &= ~UTF8_GOT_CONTINUATION;
1992 *errors |= UTF8_GOT_CONTINUATION;
1994 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1997 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1999 pack_warn = packWARN(WARN_UTF8);
2000 message = Perl_form(aTHX_
2001 "%s: %s (unexpected continuation byte 0x%02x,"
2002 " with no preceding start byte)",
2004 _byte_dump_string(s0, 1, 0), *s0);
2005 this_flag_bit = UTF8_GOT_CONTINUATION;
2009 else if (possible_problems & UTF8_GOT_SHORT) {
2010 possible_problems &= ~UTF8_GOT_SHORT;
2011 *errors |= UTF8_GOT_SHORT;
2013 if (! (flags & UTF8_ALLOW_SHORT)) {
2016 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2018 pack_warn = packWARN(WARN_UTF8);
2019 message = Perl_form(aTHX_
2020 "%s: %s (too short; %d byte%s available, need %d)",
2022 _byte_dump_string(s0, send - s0, 0),
2024 avail_len == 1 ? "" : "s",
2026 this_flag_bit = UTF8_GOT_SHORT;
2031 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
2032 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2033 *errors |= UTF8_GOT_NON_CONTINUATION;
2035 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2038 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2041 /* If we don't know for sure that the input length is
2042 * valid, avoid as much as possible reading past the
2043 * end of the buffer */
2044 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2047 pack_warn = packWARN(WARN_UTF8);
2048 message = Perl_form(aTHX_ "%s",
2049 unexpected_non_continuation_text(s0,
2053 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2057 else if (possible_problems & UTF8_GOT_SURROGATE) {
2058 possible_problems &= ~UTF8_GOT_SURROGATE;
2060 if (flags & UTF8_WARN_SURROGATE) {
2061 *errors |= UTF8_GOT_SURROGATE;
2063 if ( ! (flags & UTF8_CHECK_ONLY)
2064 && (msgs || ckWARN_d(WARN_SURROGATE)))
2066 pack_warn = packWARN(WARN_SURROGATE);
2068 /* These are the only errors that can occur with a
2069 * surrogate when the 'uv' isn't valid */
2070 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2071 message = Perl_form(aTHX_
2072 "UTF-16 surrogate (any UTF-8 sequence that"
2073 " starts with \"%s\" is for a surrogate)",
2074 _byte_dump_string(s0, curlen, 0));
2077 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2079 this_flag_bit = UTF8_GOT_SURROGATE;
2083 if (flags & UTF8_DISALLOW_SURROGATE) {
2085 *errors |= UTF8_GOT_SURROGATE;
2088 else if (possible_problems & UTF8_GOT_SUPER) {
2089 possible_problems &= ~UTF8_GOT_SUPER;
2091 if (flags & UTF8_WARN_SUPER) {
2092 *errors |= UTF8_GOT_SUPER;
2094 if ( ! (flags & UTF8_CHECK_ONLY)
2095 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2097 pack_warn = packWARN(WARN_NON_UNICODE);
2099 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2100 message = Perl_form(aTHX_
2101 "Any UTF-8 sequence that starts with"
2102 " \"%s\" is for a non-Unicode code point,"
2103 " may not be portable",
2104 _byte_dump_string(s0, curlen, 0));
2107 message = Perl_form(aTHX_ super_cp_format, uv);
2109 this_flag_bit = UTF8_GOT_SUPER;
2113 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2114 * and before possibly bailing out, so that the more dire
2115 * warning will override the regular one. */
2116 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2117 if ( ! (flags & UTF8_CHECK_ONLY)
2118 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2119 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2121 pack_warn = packWARN(WARN_NON_UNICODE);
2123 /* If it is an overlong that evaluates to a code point
2124 * that doesn't have to use the Perl extended UTF-8, it
2125 * still used it, and so we output a message that
2126 * doesn't refer to the code point. The same is true
2127 * if there was a SHORT malformation where the code
2128 * point is not valid. In that case, 'uv' will have
2129 * been set to the REPLACEMENT CHAR, and the message
2130 * below without the code point in it will be selected
2132 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2133 message = Perl_form(aTHX_
2134 perl_extended_cp_format, uv);
2137 message = Perl_form(aTHX_
2138 "Any UTF-8 sequence that starts with"
2139 " \"%s\" is a Perl extension, and"
2140 " so is not portable",
2141 _byte_dump_string(s0, curlen, 0));
2143 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2146 if (flags & ( UTF8_WARN_PERL_EXTENDED
2147 |UTF8_DISALLOW_PERL_EXTENDED))
2149 *errors |= UTF8_GOT_PERL_EXTENDED;
2151 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2157 if (flags & UTF8_DISALLOW_SUPER) {
2158 *errors |= UTF8_GOT_SUPER;
2162 else if (possible_problems & UTF8_GOT_NONCHAR) {
2163 possible_problems &= ~UTF8_GOT_NONCHAR;
2165 if (flags & UTF8_WARN_NONCHAR) {
2166 *errors |= UTF8_GOT_NONCHAR;
2168 if ( ! (flags & UTF8_CHECK_ONLY)
2169 && (msgs || ckWARN_d(WARN_NONCHAR)))
2171 /* The code above should have guaranteed that we don't
2172 * get here with errors other than overlong */
2173 assert (! (orig_problems
2174 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2176 pack_warn = packWARN(WARN_NONCHAR);
2177 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2178 this_flag_bit = UTF8_GOT_NONCHAR;
2182 if (flags & UTF8_DISALLOW_NONCHAR) {
2184 *errors |= UTF8_GOT_NONCHAR;
2187 else if (possible_problems & UTF8_GOT_LONG) {
2188 possible_problems &= ~UTF8_GOT_LONG;
2189 *errors |= UTF8_GOT_LONG;
2191 if (flags & UTF8_ALLOW_LONG) {
2193 /* We don't allow the actual overlong value, unless the
2194 * special extra bit is also set */
2195 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2196 & ~UTF8_ALLOW_LONG)))
2198 uv = UNICODE_REPLACEMENT;
2205 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2207 pack_warn = packWARN(WARN_UTF8);
2209 /* These error types cause 'uv' to be something that
2210 * isn't what was intended, so can't use it in the
2211 * message. The other error types either can't
2212 * generate an overlong, or else the 'uv' is valid */
2214 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2216 message = Perl_form(aTHX_
2217 "%s: %s (any UTF-8 sequence that starts"
2218 " with \"%s\" is overlong which can and"
2219 " should be represented with a"
2220 " different, shorter sequence)",
2222 _byte_dump_string(s0, send - s0, 0),
2223 _byte_dump_string(s0, curlen, 0));
2226 U8 tmpbuf[UTF8_MAXBYTES+1];
2227 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2229 /* Don't use U+ for non-Unicode code points, which
2230 * includes those in the Latin1 range */
2231 const char * preface = ( uv > PERL_UNICODE_MAX
2238 message = Perl_form(aTHX_
2239 "%s: %s (overlong; instead use %s to represent"
2242 _byte_dump_string(s0, send - s0, 0),
2243 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2245 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2246 small code points */
2249 this_flag_bit = UTF8_GOT_LONG;
2252 } /* End of looking through the possible flags */
2254 /* Display the message (if any) for the problem being handled in
2255 * this iteration of the loop */
2258 assert(this_flag_bit);
2260 if (*msgs == NULL) {
2264 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2269 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2272 Perl_warner(aTHX_ pack_warn, "%s", message);
2274 } /* End of 'while (possible_problems)' */
2276 /* Since there was a possible problem, the returned length may need to
2277 * be changed from the one stored at the beginning of this function.
2278 * Instead of trying to figure out if that's needed, just do it. */
2284 if (flags & UTF8_CHECK_ONLY && retlen) {
2285 *retlen = ((STRLEN) -1);
2291 return UNI_TO_NATIVE(uv);
2295 =for apidoc utf8_to_uvchr_buf
2297 Returns the native code point of the first character in the string C<s> which
2298 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2299 C<*retlen> will be set to the length, in bytes, of that character.
2301 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2302 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2303 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2304 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2305 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2306 the next possible position in C<s> that could begin a non-malformed character.
2307 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2312 Also implemented as a macro in utf8.h
2318 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2320 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2324 return utf8n_to_uvchr(s, send - s, retlen,
2325 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
2328 /* This is marked as deprecated
2330 =for apidoc utf8_to_uvuni_buf
2332 Only in very rare circumstances should code need to be dealing in Unicode
2333 (as opposed to native) code points. In those few cases, use
2334 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2335 are not absolutely sure this is one of those cases, then assume it isn't and
2336 use plain C<utf8_to_uvchr_buf> instead.
2338 Returns the Unicode (not-native) code point of the first character in the
2340 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2341 C<retlen> will be set to the length, in bytes, of that character.
2343 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2344 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2345 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2346 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2347 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2348 next possible position in C<s> that could begin a non-malformed character.
2349 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
2355 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2357 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2361 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2365 =for apidoc utf8_length
2367 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2368 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2369 same place, it returns 0 with no warning raised.
2371 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2372 and returns the number of valid characters.
2378 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2382 PERL_ARGS_ASSERT_UTF8_LENGTH;
2384 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2385 * the bitops (especially ~) can create illegal UTF-8.
2386 * In other words: in Perl UTF-8 is not just for Unicode. */
2389 goto warn_and_return;
2399 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2400 "%s in %s", unees, OP_DESC(PL_op));
2402 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2409 =for apidoc bytes_cmp_utf8
2411 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2412 sequence of characters (stored as UTF-8)
2413 in C<u>, C<ulen>. Returns 0 if they are
2414 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2415 if the first string is greater than the second string.
2417 -1 or +1 is returned if the shorter string was identical to the start of the
2418 longer string. -2 or +2 is returned if
2419 there was a difference between characters
2426 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2428 const U8 *const bend = b + blen;
2429 const U8 *const uend = u + ulen;
2431 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2433 while (b < bend && u < uend) {
2435 if (!UTF8_IS_INVARIANT(c)) {
2436 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2439 if (UTF8_IS_CONTINUATION(c1)) {
2440 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2442 /* diag_listed_as: Malformed UTF-8 character%s */
2443 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2445 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2446 PL_op ? " in " : "",
2447 PL_op ? OP_DESC(PL_op) : "");
2452 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2453 "%s in %s", unees, OP_DESC(PL_op));
2455 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2456 return -2; /* Really want to return undef :-) */
2463 return *b < c ? -2 : +2;
2468 if (b == bend && u == uend)
2471 return b < bend ? +1 : -1;
2475 =for apidoc utf8_to_bytes
2477 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2478 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2479 updates C<*lenp> to contain the new length.
2480 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2482 Upon successful return, the number of variants in the string can be computed by
2483 having saved the value of C<*lenp> before the call, and subtracting the
2484 after-call value of C<*lenp> from it.
2486 If you need a copy of the string, see L</bytes_from_utf8>.
2492 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2496 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2497 PERL_UNUSED_CONTEXT;
2499 /* This is a no-op if no variants at all in the input */
2500 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2505 U8 * const save = s;
2506 U8 * const send = s + *lenp;
2509 /* Nothing before the first variant needs to be changed, so start the real
2513 if (! UTF8_IS_INVARIANT(*s)) {
2514 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2515 *lenp = ((STRLEN) -1);
2523 /* Is downgradable, so do it */
2524 d = s = first_variant;
2527 if (! UVCHR_IS_INVARIANT(c)) {
2528 /* Then it is two-byte encoded */
2529 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2542 =for apidoc bytes_from_utf8
2544 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2545 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2546 actually encoded in UTF-8.
2548 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2551 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2552 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2553 C<*lenp> are unchanged, and the return value is the original C<s>.
2555 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2556 newly created string containing a downgraded copy of C<s>, and whose length is
2557 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2558 caller is responsible for arranging for the memory used by this string to get
2561 Upon successful return, the number of variants in the string can be computed by
2562 having saved the value of C<*lenp> before the call, and subtracting the
2563 after-call value of C<*lenp> from it.
2567 There is a macro that avoids this function call, but this is retained for
2568 anyone who calls it with the Perl_ prefix */
2571 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2573 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2574 PERL_UNUSED_CONTEXT;
2576 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2580 No = here because currently externally undocumented
2581 for apidoc bytes_from_utf8_loc
2583 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2584 to store the location of the first character in C<"s"> that cannot be
2585 converted to non-UTF8.
2587 If that parameter is C<NULL>, this function behaves identically to
2590 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2591 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2593 Otherwise, the function returns a newly created C<NUL>-terminated string
2594 containing the non-UTF8 equivalent of the convertible first portion of
2595 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2596 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2597 and C<*first_non_downgradable> is set to C<NULL>.
2599 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2600 first character in the original string that wasn't converted. C<*is_utf8p> is
2601 unchanged. Note that the new string may have length 0.
2603 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2604 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2605 converts as many characters in it as possible stopping at the first one it
2606 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2607 set to point to that. The function returns the portion that could be converted
2608 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2609 not including the terminating C<NUL>. If the very first character in the
2610 original could not be converted, C<*lenp> will be 0, and the new string will
2611 contain just a single C<NUL>. If the entire input string was converted,
2612 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2614 Upon successful return, the number of variants in the converted portion of the
2615 string can be computed by having saved the value of C<*lenp> before the call,
2616 and subtracting the after-call value of C<*lenp> from it.
2624 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2627 const U8 *original = s;
2628 U8 *converted_start;
2629 const U8 *send = s + *lenp;
2631 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2634 if (first_unconverted) {
2635 *first_unconverted = NULL;
2638 return (U8 *) original;
2641 Newx(d, (*lenp) + 1, U8);
2643 converted_start = d;
2646 if (! UTF8_IS_INVARIANT(c)) {
2648 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2649 * have to stop now */
2650 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2651 if (first_unconverted) {
2652 *first_unconverted = s - 1;
2653 goto finish_and_return;
2656 Safefree(converted_start);
2657 return (U8 *) original;
2661 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2667 /* Here, converted the whole of the input */
2669 if (first_unconverted) {
2670 *first_unconverted = NULL;
2675 *lenp = d - converted_start;
2677 /* Trim unused space */
2678 Renew(converted_start, *lenp + 1, U8);
2680 return converted_start;
2684 =for apidoc bytes_to_utf8
2686 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2688 Returns a pointer to the newly-created string, and sets C<*lenp> to
2689 reflect the new length in bytes. The caller is responsible for arranging for
2690 the memory used by this string to get freed.
2692 Upon successful return, the number of variants in the string can be computed by
2693 having saved the value of C<*lenp> before the call, and subtracting it from the
2694 after-call value of C<*lenp>.
2696 A C<NUL> character will be written after the end of the string.
2698 If you want to convert to UTF-8 from encodings other than
2699 the native (Latin1 or EBCDIC),
2700 see L</sv_recode_to_utf8>().
2706 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2708 const U8 * const send = s + (*lenp);
2712 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2713 PERL_UNUSED_CONTEXT;
2715 Newx(d, (*lenp) * 2 + 1, U8);
2719 append_utf8_from_native_byte(*s, &d);
2726 /* Trim unused space */
2727 Renew(dst, *lenp + 1, U8);
2733 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2734 * use utf16_to_utf8_reversed().
2736 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2737 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2738 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2740 * These functions don't check for overflow. The worst case is every code
2741 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2742 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2743 * destination must be pre-extended to 2 times the source length.
2745 * Do not use in-place. We optimize for native, for obvious reasons. */
2748 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2753 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2756 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2762 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2764 if (OFFUNI_IS_INVARIANT(uv)) {
2765 *d++ = LATIN1_TO_NATIVE((U8) uv);
2768 if (uv <= MAX_UTF8_TWO_BYTE) {
2769 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2770 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2774 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2775 #define LAST_HIGH_SURROGATE 0xDBFF
2776 #define FIRST_LOW_SURROGATE 0xDC00
2777 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2778 #define FIRST_IN_PLANE1 0x10000
2780 /* This assumes that most uses will be in the first Unicode plane, not
2781 * needing surrogates */
2782 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2783 && uv <= UNICODE_SURROGATE_LAST))
2785 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2786 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2789 UV low = (p[0] << 8) + p[1];
2790 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2791 || UNLIKELY(low > LAST_LOW_SURROGATE))
2793 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2796 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2797 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2801 d = uvoffuni_to_utf8_flags(d, uv, 0);
2803 if (uv < FIRST_IN_PLANE1) {
2804 *d++ = (U8)(( uv >> 12) | 0xe0);
2805 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2806 *d++ = (U8)(( uv & 0x3f) | 0x80);
2810 *d++ = (U8)(( uv >> 18) | 0xf0);
2811 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2812 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2813 *d++ = (U8)(( uv & 0x3f) | 0x80);
2818 *newlen = d - dstart;
2822 /* Note: this one is slightly destructive of the source. */
2825 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2828 U8* const send = s + bytelen;
2830 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2833 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2837 const U8 tmp = s[0];
2842 return utf16_to_utf8(p, d, bytelen, newlen);
2846 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2848 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2851 /* Internal function so we can deprecate the external one, and call
2852 this one from other deprecated functions in this file */
2855 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2857 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2861 return is_utf8_common(p, NULL,
2862 "This is buggy if this gets used",
2867 Perl__is_uni_perl_idcont(pTHX_ UV c)
2869 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2873 Perl__is_uni_perl_idstart(pTHX_ UV c)
2875 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2879 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2882 /* We have the latin1-range values compiled into the core, so just use
2883 * those, converting the result to UTF-8. The only difference between upper
2884 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2885 * either "SS" or "Ss". Which one to use is passed into the routine in
2886 * 'S_or_s' to avoid a test */
2888 UV converted = toUPPER_LATIN1_MOD(c);
2890 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2892 assert(S_or_s == 'S' || S_or_s == 's');
2894 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2895 characters in this range */
2896 *p = (U8) converted;
2901 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2902 * which it maps to one of them, so as to only have to have one check for
2903 * it in the main case */
2904 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2906 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2907 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2910 converted = GREEK_CAPITAL_LETTER_MU;
2912 #if UNICODE_MAJOR_VERSION > 2 \
2913 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2914 && UNICODE_DOT_DOT_VERSION >= 8)
2915 case LATIN_SMALL_LETTER_SHARP_S:
2922 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2923 " '%c' to map to '%c'",
2924 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2925 NOT_REACHED; /* NOTREACHED */
2929 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2930 *p = UTF8_TWO_BYTE_LO(converted);
2936 /* If compiled on an early Unicode version, there may not be auxiliary tables
2938 #ifndef HAS_UC_AUX_TABLES
2939 # define UC_AUX_TABLE_ptrs NULL
2940 # define UC_AUX_TABLE_lengths NULL
2942 #ifndef HAS_TC_AUX_TABLES
2943 # define TC_AUX_TABLE_ptrs NULL
2944 # define TC_AUX_TABLE_lengths NULL
2946 #ifndef HAS_LC_AUX_TABLES
2947 # define LC_AUX_TABLE_ptrs NULL
2948 # define LC_AUX_TABLE_lengths NULL
2950 #ifndef HAS_CF_AUX_TABLES
2951 # define CF_AUX_TABLE_ptrs NULL
2952 # define CF_AUX_TABLE_lengths NULL
2954 #ifndef HAS_UC_AUX_TABLES
2955 # define UC_AUX_TABLE_ptrs NULL
2956 # define UC_AUX_TABLE_lengths NULL
2959 /* Call the function to convert a UTF-8 encoded character to the specified case.
2960 * Note that there may be more than one character in the result.
2961 * 's' is a pointer to the first byte of the input character
2962 * 'd' will be set to the first byte of the string of changed characters. It
2963 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2964 * 'lenp' will be set to the length in bytes of the string of changed characters
2966 * The functions return the ordinal of the first character in the string of
2968 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2969 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2970 Uppercase_Mapping_invmap, \
2971 UC_AUX_TABLE_ptrs, \
2972 UC_AUX_TABLE_lengths, \
2974 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2975 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2976 Titlecase_Mapping_invmap, \
2977 TC_AUX_TABLE_ptrs, \
2978 TC_AUX_TABLE_lengths, \
2980 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2981 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2982 Lowercase_Mapping_invmap, \
2983 LC_AUX_TABLE_ptrs, \
2984 LC_AUX_TABLE_lengths, \
2988 /* This additionally has the input parameter 'specials', which if non-zero will
2989 * cause this to use the specials hash for folding (meaning get full case
2990 * folding); otherwise, when zero, this implies a simple case fold */
2991 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2993 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2994 Case_Folding_invmap, \
2995 CF_AUX_TABLE_ptrs, \
2996 CF_AUX_TABLE_lengths, \
2998 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2999 Simple_Case_Folding_invmap, \
3004 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
3006 /* Convert the Unicode character whose ordinal is <c> to its uppercase
3007 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
3008 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
3009 * the changed version may be longer than the original character.
3011 * The ordinal of the first character of the changed version is returned
3012 * (but note, as explained above, that there may be more.) */
3014 PERL_ARGS_ASSERT_TO_UNI_UPPER;
3017 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
3020 uvchr_to_utf8(p, c);
3021 return CALL_UPPER_CASE(c, p, p, lenp);
3025 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
3027 PERL_ARGS_ASSERT_TO_UNI_TITLE;
3030 return _to_upper_title_latin1((U8) c, p, lenp, 's');
3033 uvchr_to_utf8(p, c);
3034 return CALL_TITLE_CASE(c, p, p, lenp);
3038 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
3040 /* We have the latin1-range values compiled into the core, so just use
3041 * those, converting the result to UTF-8. Since the result is always just
3042 * one character, we allow <p> to be NULL */
3044 U8 converted = toLOWER_LATIN1(c);
3046 PERL_UNUSED_ARG(dummy);
3049 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
3054 /* Result is known to always be < 256, so can use the EIGHT_BIT
3056 *p = UTF8_EIGHT_BIT_HI(converted);
3057 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3065 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3067 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3070 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3073 uvchr_to_utf8(p, c);
3074 return CALL_LOWER_CASE(c, p, p, lenp);
3078 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3080 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3081 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3082 * FOLD_FLAGS_FULL iff full folding is to be used;
3084 * Not to be used for locale folds
3089 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3091 assert (! (flags & FOLD_FLAGS_LOCALE));
3093 if (UNLIKELY(c == MICRO_SIGN)) {
3094 converted = GREEK_SMALL_LETTER_MU;
3096 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3097 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3098 || UNICODE_DOT_DOT_VERSION > 0)
3099 else if ( (flags & FOLD_FLAGS_FULL)
3100 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3102 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3103 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3104 * under those circumstances. */
3105 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3106 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3107 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3109 return LATIN_SMALL_LETTER_LONG_S;
3119 else { /* In this range the fold of all other characters is their lower
3121 converted = toLOWER_LATIN1(c);
3124 if (UVCHR_IS_INVARIANT(converted)) {
3125 *p = (U8) converted;
3129 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3130 *p = UTF8_TWO_BYTE_LO(converted);
3138 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3141 /* Not currently externally documented, and subject to change
3142 * <flags> bits meanings:
3143 * FOLD_FLAGS_FULL iff full folding is to be used;
3144 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3145 * locale are to be used.
3146 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3149 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3151 if (flags & FOLD_FLAGS_LOCALE) {
3152 /* Treat a UTF-8 locale as not being in locale at all, except for
3153 * potentially warning */
3154 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3155 if (IN_UTF8_CTYPE_LOCALE) {
3156 flags &= ~FOLD_FLAGS_LOCALE;
3159 goto needs_full_generality;
3164 return _to_fold_latin1((U8) c, p, lenp,
3165 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3168 /* Here, above 255. If no special needs, just use the macro */
3169 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3170 uvchr_to_utf8(p, c);
3171 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
3173 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3174 the special flags. */
3175 U8 utf8_c[UTF8_MAXBYTES + 1];
3177 needs_full_generality:
3178 uvchr_to_utf8(utf8_c, c);
3179 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3184 PERL_STATIC_INLINE bool
3185 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
3186 const char *const swashname, SV* const invlist)
3188 /* returns a boolean giving whether or not the UTF8-encoded character that
3189 * starts at <p> is in the swash indicated by <swashname>. <swash>
3190 * contains a pointer to where the swash indicated by <swashname>
3191 * is to be stored; which this routine will do, so that future calls will
3192 * look at <*swash> and only generate a swash if it is not null. <invlist>
3193 * is NULL or an inversion list that defines the swash. If not null, it
3194 * saves time during initialization of the swash.
3196 * Note that it is assumed that the buffer length of <p> is enough to
3197 * contain all the bytes that comprise the character. Thus, <*p> should
3198 * have been checked before this call for mal-formedness enough to assure
3201 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3203 /* The API should have included a length for the UTF-8 character in <p>,
3204 * but it doesn't. We therefore assume that p has been validated at least
3205 * as far as there being enough bytes available in it to accommodate the
3206 * character without reading beyond the end, and pass that number on to the
3207 * validating routine */
3208 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
3209 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
3210 _UTF8_NO_CONFIDENCE_IN_CURLEN,
3212 NOT_REACHED; /* NOTREACHED */
3216 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3222 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3223 *swash = _core_swash_init("utf8",
3225 /* Only use the name if there is no inversion
3226 * list; otherwise will go out to disk */
3227 (invlist) ? "" : swashname,
3229 &PL_sv_undef, 1, 0, invlist, &flags);
3232 return swash_fetch(*swash, p, TRUE) != 0;
3235 PERL_STATIC_INLINE bool
3236 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3237 SV **swash, const char *const swashname,
3240 /* returns a boolean giving whether or not the UTF8-encoded character that
3241 * starts at <p>, and extending no further than <e - 1> is in the swash
3242 * indicated by <swashname>. <swash> contains a pointer to where the swash
3243 * indicated by <swashname> is to be stored; which this routine will do, so
3244 * that future calls will look at <*swash> and only generate a swash if it
3245 * is not null. <invlist> is NULL or an inversion list that defines the
3246 * swash. If not null, it saves time during initialization of the swash.
3249 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3251 if (! isUTF8_CHAR(p, e)) {
3252 _force_out_malformed_utf8_message(p, e, 0, 1);
3253 NOT_REACHED; /* NOTREACHED */
3257 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3263 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3264 *swash = _core_swash_init("utf8",
3266 /* Only use the name if there is no inversion
3267 * list; otherwise will go out to disk */
3268 (invlist) ? "" : swashname,
3270 &PL_sv_undef, 1, 0, invlist, &flags);
3273 return swash_fetch(*swash, p, TRUE) != 0;
3277 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3278 const char * const alternative,
3279 const bool use_locale,
3280 const char * const file,
3281 const unsigned line)
3285 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3287 if (ckWARN_d(WARN_DEPRECATED)) {
3289 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3290 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3291 if (! PL_seen_deprecated_macro) {
3292 PL_seen_deprecated_macro = newHV();
3294 if (! hv_store(PL_seen_deprecated_macro, key,
3295 strlen(key), &PL_sv_undef, 0))
3297 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3300 if (instr(file, "mathoms.c")) {
3301 Perl_warner(aTHX_ WARN_DEPRECATED,
3302 "In %s, line %d, starting in Perl v5.30, %s()"
3303 " will be removed. Avoid this message by"
3304 " converting to use %s().\n",
3305 file, line, name, alternative);
3308 Perl_warner(aTHX_ WARN_DEPRECATED,
3309 "In %s, line %d, starting in Perl v5.30, %s() will"
3310 " require an additional parameter. Avoid this"
3311 " message by converting to use %s().\n",
3312 file, line, name, alternative);
3319 Perl__is_utf8_FOO(pTHX_ U8 classnum,
3321 const char * const name,
3322 const char * const alternative,
3323 const bool use_utf8,
3324 const bool use_locale,
3325 const char * const file,
3326 const unsigned line)
3328 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3330 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3332 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
3342 case _CC_ALPHANUMERIC:
3346 return is_utf8_common(p,
3348 "This is buggy if this gets used",
3349 PL_XPosix_ptrs[classnum]);
3352 return is_XPERLSPACE_high(p);
3354 return is_HORIZWS_high(p);
3356 return is_XDIGIT_high(p);
3362 return is_VERTWS_high(p);
3364 return is_utf8_common(p, NULL,
3365 "This is buggy if this gets used",
3366 PL_utf8_perl_idstart);
3368 return is_utf8_common(p, NULL,
3369 "This is buggy if this gets used",
3370 PL_utf8_perl_idcont);
3374 /* idcont is the same as wordchar below 256 */
3375 if (classnum == _CC_IDCONT) {
3376 classnum = _CC_WORDCHAR;
3378 else if (classnum == _CC_IDFIRST) {
3382 classnum = _CC_ALPHA;
3386 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3387 return _generic_isCC(*p, classnum);
3390 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3393 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3394 return isFOO_lc(classnum, *p);
3397 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3400 NOT_REACHED; /* NOTREACHED */
3404 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3407 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3409 return is_utf8_common_with_len(p, e, NULL,
3410 "This is buggy if this gets used",
3411 PL_XPosix_ptrs[classnum]);
3415 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3417 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3419 return is_utf8_common_with_len(p, e, NULL,
3420 "This is buggy if this gets used",
3421 PL_utf8_perl_idstart);
3425 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3427 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3431 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
3435 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3437 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3439 return is_utf8_common_with_len(p, e, NULL,
3440 "This is buggy if this gets used",
3441 PL_utf8_perl_idcont);
3445 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3447 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3449 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
3453 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3455 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3457 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
3461 Perl__is_utf8_mark(pTHX_ const U8 *p)
3463 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3465 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
3469 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3470 U8* ustrp, STRLEN *lenp,
3471 SV *invlist, const int * const invmap,
3472 const unsigned int * const * const aux_tables,
3473 const U8 * const aux_table_lengths,
3474 const char * const normal)
3478 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3479 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3480 * to name the new case in any generated messages, as a fallback if the
3481 * operation being used is not available. The new case is given by the
3482 * data structures in the remaining arguments.
3484 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3485 * entire changed case string, and the return value is the first code point
3488 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3490 /* For code points that don't change case, we already know that the output
3491 * of this function is the unchanged input, so we can skip doing look-ups
3492 * for them. Unfortunately the case-changing code points are scattered
3493 * around. But there are some long consecutive ranges where there are no
3494 * case changing code points. By adding tests, we can eliminate the lookup
3495 * for all the ones in such ranges. This is currently done here only for
3496 * just a few cases where the scripts are in common use in modern commerce
3497 * (and scripts adjacent to those which can be included without additional
3500 if (uv1 >= 0x0590) {
3501 /* This keeps from needing further processing the code points most
3502 * likely to be used in the following non-cased scripts: Hebrew,
3503 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3504 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3505 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3510 /* The following largish code point ranges also don't have case
3511 * changes, but khw didn't think they warranted extra tests to speed
3512 * them up (which would slightly slow down everything else above them):
3513 * 1100..139F Hangul Jamo, Ethiopic
3514 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3515 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3516 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3517 * Combining Diacritical Marks Extended, Balinese,
3518 * Sundanese, Batak, Lepcha, Ol Chiki
3519 * 2000..206F General Punctuation
3522 if (uv1 >= 0x2D30) {
3524 /* This keeps the from needing further processing the code points
3525 * most likely to be used in the following non-cased major scripts:
3526 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3528 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3529 * event that Unicode eventually allocates the unused block as of
3530 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3531 * that the test suite will start having failures to alert you
3532 * should that happen) */
3537 if (uv1 >= 0xAC00) {
3538 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3539 if (ckWARN_d(WARN_SURROGATE)) {
3540 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3541 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3542 "Operation \"%s\" returns its argument for"
3543 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3548 /* AC00..FAFF Catches Hangul syllables and private use, plus
3554 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3555 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3556 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3557 MAX_EXTERNALLY_LEGAL_CP);
3559 if (ckWARN_d(WARN_NON_UNICODE)) {
3560 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3561 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3562 "Operation \"%s\" returns its argument for"
3563 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3567 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3569 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3572 /* As of Unicode 10.0, this means we avoid swash creation
3573 * for anything beyond high Plane 1 (below emojis) */
3580 /* Note that non-characters are perfectly legal, so no warning should
3586 const unsigned int * cp_list;
3588 SSize_t index = _invlist_search(invlist, uv1);
3589 IV base = invmap[index];
3591 /* The data structures are set up so that if 'base' is non-negative,
3592 * the case change is 1-to-1; and if 0, the change is to itself */
3600 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3601 lc = base + uv1 - invlist_array(invlist)[index];
3602 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3606 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3607 * requires an auxiliary table look up. abs(base) gives the index into
3608 * a list of such tables which points to the proper aux table. And a
3609 * parallel list gives the length of each corresponding aux table. */
3610 cp_list = aux_tables[-base];
3612 /* Create the string of UTF-8 from the mapped-to code points */
3614 for (i = 0; i < aux_table_lengths[-base]; i++) {
3615 d = uvchr_to_utf8(d, cp_list[i]);
3623 /* Here, there was no mapping defined, which means that the code point maps
3624 * to itself. Return the inputs */
3627 if (p != ustrp) { /* Don't copy onto itself */
3628 Copy(p, ustrp, len, U8);
3639 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3640 const unsigned int ** remaining_folds_to)
3642 /* Returns the count of the number of code points that fold to the input
3643 * 'cp' (besides itself).
3645 * If the return is 0, there is nothing else that folds to it, and
3646 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3648 * If the return is 1, '*first_folds_to' is set to the single code point,
3649 * and '*remaining_folds_to' is set to NULL.
3651 * Otherwise, '*first_folds_to' is set to a code point, and
3652 * '*remaining_fold_to' is set to an array that contains the others. The
3653 * length of this array is the returned count minus 1.
3655 * The reason for this convolution is to avoid having to deal with
3656 * allocating and freeing memory. The lists are already constructed, so
3657 * the return can point to them, but single code points aren't, so would
3658 * need to be constructed if we didn't employ something like this API */
3660 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3661 int base = _Perl_IVCF_invmap[index];
3663 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3665 if (base == 0) { /* No fold */
3666 *first_folds_to = 0;
3667 *remaining_folds_to = NULL;
3671 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3677 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3679 /* The data structure is set up so that the absolute value of 'base' is
3680 * an index into a table of pointers to arrays, with the array
3681 * corresponding to the index being the list of code points that fold
3682 * to 'cp', and the parallel array containing the length of the list
3684 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3685 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3688 return IVCF_AUX_TABLE_lengths[-base];
3693 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3694 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3695 *remaining_folds_to = NULL;
3700 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3701 U8* const ustrp, STRLEN *lenp)
3703 /* This is called when changing the case of a UTF-8-encoded character above
3704 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3705 * result contains a character that crosses the 255/256 boundary, disallow
3706 * the change, and return the original code point. See L<perlfunc/lc> for
3709 * p points to the original string whose case was changed; assumed
3710 * by this routine to be well-formed
3711 * result the code point of the first character in the changed-case string
3712 * ustrp points to the changed-case string (<result> represents its
3714 * lenp points to the length of <ustrp> */
3716 UV original; /* To store the first code point of <p> */
3718 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3720 assert(UTF8_IS_ABOVE_LATIN1(*p));
3722 /* We know immediately if the first character in the string crosses the
3723 * boundary, so can skip testing */
3726 /* Look at every character in the result; if any cross the
3727 * boundary, the whole thing is disallowed */
3728 U8* s = ustrp + UTF8SKIP(ustrp);
3729 U8* e = ustrp + *lenp;
3731 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3737 /* Here, no characters crossed, result is ok as-is, but we warn. */
3738 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3744 /* Failed, have to return the original */
3745 original = valid_utf8_to_uvchr(p, lenp);
3747 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3748 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3749 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3750 " locale; resolved to \"\\x{%" UVXf "}\".",
3754 Copy(p, ustrp, *lenp, char);
3759 S_check_and_deprecate(pTHX_ const U8 *p,
3761 const unsigned int type, /* See below */
3762 const bool use_locale, /* Is this a 'LC_'
3764 const char * const file,
3765 const unsigned line)
3767 /* This is a temporary function to deprecate the unsafe calls to the case
3768 * changing macros and functions. It keeps all the special stuff in just
3771 * It updates *e with the pointer to the end of the input string. If using
3772 * the old-style macros, *e is NULL on input, and so this function assumes
3773 * the input string is long enough to hold the entire UTF-8 sequence, and
3774 * sets *e accordingly, but it then returns a flag to pass the
3775 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3776 * using the full length if possible.
3778 * It also does the assert that *e > p when *e is not NULL. This should be
3779 * migrated to the callers when this function gets deleted.
3781 * The 'type' parameter is used for the caller to specify which case
3782 * changing function this is called from: */
3784 # define DEPRECATE_TO_UPPER 0
3785 # define DEPRECATE_TO_TITLE 1
3786 # define DEPRECATE_TO_LOWER 2
3787 # define DEPRECATE_TO_FOLD 3
3789 U32 utf8n_flags = 0;
3791 const char * alternative;
3793 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3796 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3797 *e = p + UTF8SKIP(p);
3799 /* For mathoms.c calls, we use the function name we know is stored
3800 * there. It could be part of a larger path */
3801 if (type == DEPRECATE_TO_UPPER) {
3802 name = instr(file, "mathoms.c")
3805 alternative = "toUPPER_utf8_safe";
3807 else if (type == DEPRECATE_TO_TITLE) {
3808 name = instr(file, "mathoms.c")
3811 alternative = "toTITLE_utf8_safe";
3813 else if (type == DEPRECATE_TO_LOWER) {
3814 name = instr(file, "mathoms.c")
3817 alternative = "toLOWER_utf8_safe";
3819 else if (type == DEPRECATE_TO_FOLD) {
3820 name = instr(file, "mathoms.c")
3823 alternative = "toFOLD_utf8_safe";
3825 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3827 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3836 /* The process for changing the case is essentially the same for the four case
3837 * change types, except there are complications for folding. Otherwise the
3838 * difference is only which case to change to. To make sure that they all do
3839 * the same thing, the bodies of the functions are extracted out into the
3840 * following two macros. The functions are written with the same variable
3841 * names, and these are known and used inside these macros. It would be
3842 * better, of course, to have inline functions to do it, but since different
3843 * macros are called, depending on which case is being changed to, this is not
3844 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3845 * function can start with the common start macro, then finish with its special
3846 * handling; while the other three cases can just use the common end macro.
3848 * The algorithm is to use the proper (passed in) macro or function to change
3849 * the case for code points that are below 256. The macro is used if using
3850 * locale rules for the case change; the function if not. If the code point is
3851 * above 255, it is computed from the input UTF-8, and another macro is called
3852 * to do the conversion. If necessary, the output is converted to UTF-8. If
3853 * using a locale, we have to check that the change did not cross the 255/256
3854 * boundary, see check_locale_boundary_crossing() for further details.
3856 * The macros are split with the correct case change for the below-256 case
3857 * stored into 'result', and in the middle of an else clause for the above-255
3858 * case. At that point in the 'else', 'result' is not the final result, but is
3859 * the input code point calculated from the UTF-8. The fold code needs to
3860 * realize all this and take it from there.
3862 * If you read the two macros as sequential, it's easier to understand what's
3864 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3865 L1_func_extra_param) \
3867 if (flags & (locale_flags)) { \
3868 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3869 /* Treat a UTF-8 locale as not being in locale at all */ \
3870 if (IN_UTF8_CTYPE_LOCALE) { \
3871 flags &= ~(locale_flags); \
3875 if (UTF8_IS_INVARIANT(*p)) { \
3876 if (flags & (locale_flags)) { \
3877 result = LC_L1_change_macro(*p); \
3880 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3883 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3884 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3885 if (flags & (locale_flags)) { \
3886 result = LC_L1_change_macro(c); \
3889 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3892 else { /* malformed UTF-8 or ord above 255 */ \
3893 STRLEN len_result; \
3894 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3895 if (len_result == (STRLEN) -1) { \
3896 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3900 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3901 result = change_macro(result, p, ustrp, lenp); \
3903 if (flags & (locale_flags)) { \
3904 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3909 /* Here, used locale rules. Convert back to UTF-8 */ \
3910 if (UTF8_IS_INVARIANT(result)) { \
3911 *ustrp = (U8) result; \
3915 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3916 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3923 =for apidoc to_utf8_upper
3925 Instead use L</toUPPER_utf8_safe>.
3929 /* Not currently externally documented, and subject to change:
3930 * <flags> is set iff iff the rules from the current underlying locale are to
3934 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3939 const char * const file,
3943 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3944 cBOOL(flags), file, line);
3946 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3948 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3949 /* 2nd char of uc(U+DF) is 'S' */
3950 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3951 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3955 =for apidoc to_utf8_title
3957 Instead use L</toTITLE_utf8_safe>.
3961 /* Not currently externally documented, and subject to change:
3962 * <flags> is set iff the rules from the current underlying locale are to be
3963 * used. Since titlecase is not defined in POSIX, for other than a
3964 * UTF-8 locale, uppercase is used instead for code points < 256.
3968 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3973 const char * const file,
3977 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3978 cBOOL(flags), file, line);
3980 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3982 /* 2nd char of ucfirst(U+DF) is 's' */
3983 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3984 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3988 =for apidoc to_utf8_lower
3990 Instead use L</toLOWER_utf8_safe>.
3994 /* Not currently externally documented, and subject to change:
3995 * <flags> is set iff iff the rules from the current underlying locale are to
4000 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
4005 const char * const file,
4009 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
4010 cBOOL(flags), file, line);
4012 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
4014 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
4015 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
4019 =for apidoc to_utf8_fold
4021 Instead use L</toFOLD_utf8_safe>.
4025 /* Not currently externally documented, and subject to change,
4027 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
4028 * locale are to be used.
4029 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
4030 * otherwise simple folds
4031 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
4036 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
4041 const char * const file,
4045 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
4046 cBOOL(flags), file, line);
4048 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
4050 /* These are mutually exclusive */
4051 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
4053 assert(p != ustrp); /* Otherwise overwrites */
4055 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
4056 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
4058 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
4060 if (flags & FOLD_FLAGS_LOCALE) {
4062 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
4063 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4064 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4066 /* Special case these two characters, as what normally gets
4067 * returned under locale doesn't work */
4068 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S))
4070 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4071 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4072 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
4073 "resolved to \"\\x{17F}\\x{17F}\".");
4078 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T))
4080 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4081 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4082 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
4083 "resolved to \"\\x{FB06}\".");
4084 goto return_ligature_st;
4087 #if UNICODE_MAJOR_VERSION == 3 \
4088 && UNICODE_DOT_VERSION == 0 \
4089 && UNICODE_DOT_DOT_VERSION == 1
4090 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
4092 /* And special case this on this Unicode version only, for the same
4093 * reaons the other two are special cased. They would cross the
4094 * 255/256 boundary which is forbidden under /l, and so the code
4095 * wouldn't catch that they are equivalent (which they are only in
4097 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) {
4098 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4099 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4100 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
4101 "resolved to \"\\x{0131}\".");
4102 goto return_dotless_i;
4106 return check_locale_boundary_crossing(p, result, ustrp, lenp);
4108 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
4112 /* This is called when changing the case of a UTF-8-encoded
4113 * character above the ASCII range, and the result should not
4114 * contain an ASCII character. */
4116 UV original; /* To store the first code point of <p> */
4118 /* Look at every character in the result; if any cross the
4119 * boundary, the whole thing is disallowed */
4121 U8* e = ustrp + *lenp;
4124 /* Crossed, have to return the original */
4125 original = valid_utf8_to_uvchr(p, lenp);
4127 /* But in these instances, there is an alternative we can
4128 * return that is valid */
4129 if (original == LATIN_SMALL_LETTER_SHARP_S
4130 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
4131 || original == LATIN_CAPITAL_LETTER_SHARP_S
4136 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
4137 goto return_ligature_st;
4139 #if UNICODE_MAJOR_VERSION == 3 \
4140 && UNICODE_DOT_VERSION == 0 \
4141 && UNICODE_DOT_DOT_VERSION == 1
4143 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4144 goto return_dotless_i;
4147 Copy(p, ustrp, *lenp, char);
4153 /* Here, no characters crossed, result is ok as-is */
4158 /* Here, used locale rules. Convert back to UTF-8 */
4159 if (UTF8_IS_INVARIANT(result)) {
4160 *ustrp = (U8) result;
4164 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
4165 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
4172 /* Certain folds to 'ss' are prohibited by the options, but they do allow
4173 * folds to a string of two of these characters. By returning this
4174 * instead, then, e.g.,
4175 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
4178 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
4179 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
4181 return LATIN_SMALL_LETTER_LONG_S;
4184 /* Two folds to 'st' are prohibited by the options; instead we pick one and
4185 * have the other one fold to it */
4187 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
4188 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
4189 return LATIN_SMALL_LIGATURE_ST;
4191 #if UNICODE_MAJOR_VERSION == 3 \
4192 && UNICODE_DOT_VERSION == 0 \
4193 && UNICODE_DOT_DOT_VERSION == 1
4196 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
4197 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
4198 return LATIN_SMALL_LETTER_DOTLESS_I;
4205 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
4206 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
4207 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
4211 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4212 I32 minbits, I32 none)
4214 PERL_ARGS_ASSERT_SWASH_INIT;
4216 /* Returns a copy of a swash initiated by the called function. This is the
4217 * public interface, and returning a copy prevents others from doing
4218 * mischief on the original */
4220 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
4225 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4226 I32 minbits, I32 none, SV* invlist,
4230 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
4231 * use the following define */
4233 #define CORE_SWASH_INIT_RETURN(x) \
4234 PL_curpm= old_PL_curpm; \
4237 /* Initialize and return a swash, creating it if necessary. It does this
4238 * by calling utf8_heavy.pl in the general case. The returned value may be
4239 * the swash's inversion list instead if the input parameters allow it.
4240 * Which is returned should be immaterial to callers, as the only
4241 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
4242 * and swash_to_invlist() handle both these transparently.
4244 * This interface should only be used by functions that won't destroy or
4245 * adversely change the swash, as doing so affects all other uses of the
4246 * swash in the program; the general public should use 'Perl_swash_init'
4249 * pkg is the name of the package that <name> should be in.
4250 * name is the name of the swash to find. Typically it is a Unicode
4251 * property name, including user-defined ones
4252 * listsv is a string to initialize the swash with. It must be of the form
4253 * documented as the subroutine return value in
4254 * L<perlunicode/User-Defined Character Properties>
4255 * minbits is the number of bits required to represent each data element.
4256 * It is '1' for binary properties.
4257 * none I (khw) do not understand this one, but it is used only in tr///.
4258 * invlist is an inversion list to initialize the swash with (or NULL)
4259 * flags_p if non-NULL is the address of various input and output flag bits
4260 * to the routine, as follows: ('I' means is input to the routine;
4261 * 'O' means output from the routine. Only flags marked O are
4262 * meaningful on return.)
4263 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
4264 * came from a user-defined property. (I O)
4265 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
4266 * when the swash cannot be located, to simply return NULL. (I)
4267 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
4268 * return of an inversion list instead of a swash hash if this routine
4269 * thinks that would result in faster execution of swash_fetch() later
4272 * Thus there are three possible inputs to find the swash: <name>,
4273 * <listsv>, and <invlist>. At least one must be specified. The result
4274 * will be the union of the specified ones, although <listsv>'s various
4275 * actions can intersect, etc. what <name> gives. To avoid going out to
4276 * disk at all, <invlist> should specify completely what the swash should
4277 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
4279 * <invlist> is only valid for binary properties */
4281 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
4283 SV* retval = &PL_sv_undef;
4284 HV* swash_hv = NULL;
4285 const bool use_invlist= (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST);
4287 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
4288 assert(! invlist || minbits == 1);
4290 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
4291 regex that triggered the swash init and the swash init
4292 perl logic itself. See perl #122747 */
4294 /* If data was passed in to go out to utf8_heavy to find the swash of, do
4296 if (listsv != &PL_sv_undef || strNE(name, "")) {
4298 const size_t pkg_len = strlen(pkg);
4299 const size_t name_len = strlen(name);
4300 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
4304 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
4306 PUSHSTACKi(PERLSI_MAGIC);
4310 /* We might get here via a subroutine signature which uses a utf8
4311 * parameter name, at which point PL_subname will have been set
4312 * but not yet used. */
4313 save_item(PL_subname);
4314 if (PL_parser && PL_parser->error_count)
4315 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
4316 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4317 if (!method) { /* demand load UTF-8 */
4319 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4320 GvSV(PL_errgv) = NULL;
4321 #ifndef NO_TAINT_SUPPORT
4322 /* It is assumed that callers of this routine are not passing in
4323 * any user derived data. */
4324 /* Need to do this after save_re_context() as it will set
4325 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
4326 * in Perl_magic_get). Even line to create errsv_save can turn on
4328 SAVEBOOL(TAINT_get);
4331 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
4334 /* Not ERRSV, as there is no need to vivify a scalar we are
4335 about to discard. */
4336 SV * const errsv = GvSV(PL_errgv);
4337 if (!SvTRUE(errsv)) {
4338 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4339 SvREFCNT_dec(errsv);
4347 mPUSHp(pkg, pkg_len);
4348 mPUSHp(name, name_len);
4353 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4354 GvSV(PL_errgv) = NULL;
4355 /* If we already have a pointer to the method, no need to use
4356 * call_method() to repeat the lookup. */
4358 ? call_sv(MUTABLE_SV(method), G_SCALAR)
4359 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
4361 retval = *PL_stack_sp--;
4362 SvREFCNT_inc(retval);
4365 /* Not ERRSV. See above. */
4366 SV * const errsv = GvSV(PL_errgv);
4367 if (!SvTRUE(errsv)) {
4368 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4369 SvREFCNT_dec(errsv);
4374 if (IN_PERL_COMPILETIME) {
4375 CopHINTS_set(PL_curcop, PL_hints);
4377 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
4378 if (SvPOK(retval)) {
4380 /* If caller wants to handle missing properties, let them */
4381 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
4382 CORE_SWASH_INIT_RETURN(NULL);
4385 "Can't find Unicode property definition \"%" SVf "\"",
4387 NOT_REACHED; /* NOTREACHED */
4390 } /* End of calling the module to find the swash */
4392 /* If this operation fetched a swash, and we will need it later, get it */
4393 if (retval != &PL_sv_undef
4394 && (minbits == 1 || (flags_p
4396 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
4398 swash_hv = MUTABLE_HV(SvRV(retval));
4400 /* If we don't already know that there is a user-defined component to
4401 * this swash, and the user has indicated they wish to know if there is
4402 * one (by passing <flags_p>), find out */
4403 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
4404 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
4405 if (user_defined && SvUV(*user_defined)) {
4406 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
4411 /* Make sure there is an inversion list for binary properties */
4413 SV** swash_invlistsvp = NULL;
4414 SV* swash_invlist = NULL;
4415 bool invlist_in_swash_is_valid = FALSE;
4416 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
4417 an unclaimed reference count */
4419 /* If this operation fetched a swash, get its already existing
4420 * inversion list, or create one for it */
4423 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
4424 if (swash_invlistsvp) {
4425 swash_invlist = *swash_invlistsvp;
4426 invlist_in_swash_is_valid = TRUE;
4429 swash_invlist = _swash_to_invlist(retval);
4430 swash_invlist_unclaimed = TRUE;
4434 /* If an inversion list was passed in, have to include it */
4437 /* Any fetched swash will by now have an inversion list in it;
4438 * otherwise <swash_invlist> will be NULL, indicating that we
4439 * didn't fetch a swash */
4440 if (swash_invlist) {
4442 /* Add the passed-in inversion list, which invalidates the one
4443 * already stored in the swash */
4444 invlist_in_swash_is_valid = FALSE;
4445 SvREADONLY_off(swash_invlist); /* Turned on again below */
4446 _invlist_union(invlist, swash_invlist, &swash_invlist);
4450 /* Here, there is no swash already. Set up a minimal one, if
4451 * we are going to return a swash */
4452 if (! use_invlist) {
4454 retval = newRV_noinc(MUTABLE_SV(swash_hv));
4456 swash_invlist = invlist;
4460 /* Here, we have computed the union of all the passed-in data. It may
4461 * be that there was an inversion list in the swash which didn't get
4462 * touched; otherwise save the computed one */
4463 if (! invlist_in_swash_is_valid && ! use_invlist) {
4464 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
4466 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4468 /* We just stole a reference count. */
4469 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
4470 else SvREFCNT_inc_simple_void_NN(swash_invlist);
4473 /* The result is immutable. Forbid attempts to change it. */
4474 SvREADONLY_on(swash_invlist);
4477 SvREFCNT_dec(retval);
4478 if (!swash_invlist_unclaimed)
4479 SvREFCNT_inc_simple_void_NN(swash_invlist);
4480 retval = newRV_noinc(swash_invlist);
4484 CORE_SWASH_INIT_RETURN(retval);
4485 #undef CORE_SWASH_INIT_RETURN
4489 /* This API is wrong for special case conversions since we may need to
4490 * return several Unicode characters for a single Unicode character
4491 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
4492 * the lower-level routine, and it is similarly broken for returning
4493 * multiple values. --jhi
4494 * For those, you should use S__to_utf8_case() instead */
4495 /* Now SWASHGET is recasted into S_swatch_get in this file. */
4498 * Returns the value of property/mapping C<swash> for the first character
4499 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4500 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
4501 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
4503 * A "swash" is a hash which contains initially the keys/values set up by
4504 * SWASHNEW. The purpose is to be able to completely represent a Unicode
4505 * property for all possible code points. Things are stored in a compact form
4506 * (see utf8_heavy.pl) so that calculation is required to find the actual
4507 * property value for a given code point. As code points are looked up, new
4508 * key/value pairs are added to the hash, so that the calculation doesn't have
4509 * to ever be re-done. Further, each calculation is done, not just for the
4510 * desired one, but for a whole block of code points adjacent to that one.
4511 * For binary properties on ASCII machines, the block is usually for 64 code
4512 * points, starting with a code point evenly divisible by 64. Thus if the
4513 * property value for code point 257 is requested, the code goes out and
4514 * calculates the property values for all 64 code points between 256 and 319,
4515 * and stores these as a single 64-bit long bit vector, called a "swatch",
4516 * under the key for code point 256. The key is the UTF-8 encoding for code
4517 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
4518 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
4519 * for code point 258 is then requested, this code realizes that it would be
4520 * stored under the key for 256, and would find that value and extract the
4521 * relevant bit, offset from 256.
4523 * Non-binary properties are stored in as many bits as necessary to represent
4524 * their values (32 currently, though the code is more general than that), not
4525 * as single bits, but the principle is the same: the value for each key is a
4526 * vector that encompasses the property values for all code points whose UTF-8
4527 * representations are represented by the key. That is, for all code points
4528 * whose UTF-8 representations are length N bytes, and the key is the first N-1
4532 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
4534 HV *const hv = MUTABLE_HV(SvRV(swash));
4539 const U8 *tmps = NULL;
4543 PERL_ARGS_ASSERT_SWASH_FETCH;
4545 /* If it really isn't a hash, it isn't really swash; must be an inversion
4547 if (SvTYPE(hv) != SVt_PVHV) {
4548 return _invlist_contains_cp((SV*)hv,
4550 ? valid_utf8_to_uvchr(ptr, NULL)
4554 /* We store the values in a "swatch" which is a vec() value in a swash
4555 * hash. Code points 0-255 are a single vec() stored with key length
4556 * (klen) 0. All other code points have a UTF-8 representation
4557 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
4558 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
4559 * length for them is the length of the encoded char - 1. ptr[klen] is the
4560 * final byte in the sequence representing the character */
4561 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
4566 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
4569 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
4572 klen = UTF8SKIP(ptr) - 1;
4574 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
4575 * the vec is the final byte in the sequence. (In EBCDIC this is
4576 * converted to I8 to get consecutive values.) To help you visualize
4578 * Straight 1047 After final byte
4579 * UTF-8 UTF-EBCDIC I8 transform
4580 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
4581 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
4583 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
4584 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
4586 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
4587 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
4589 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
4590 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
4592 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
4593 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
4595 * (There are no discontinuities in the elided (...) entries.)
4596 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
4597 * key for the next 31, up through U+043F, whose UTF-8 final byte is
4598 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
4599 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
4600 * index into the vec() swatch (after subtracting 0x80, which we
4601 * actually do with an '&').
4602 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
4603 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4604 * dicontinuities which go away by transforming it into I8, and we
4605 * effectively subtract 0xA0 to get the index. */
4606 needents = (1 << UTF_ACCUMULATION_SHIFT);
4607 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4611 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4612 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4613 * it's nothing to sniff at.) Pity we usually come through at least
4614 * two function calls to get here...
4616 * NB: this code assumes that swatches are never modified, once generated!
4619 if (hv == PL_last_swash_hv &&
4620 klen == PL_last_swash_klen &&
4621 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4623 tmps = PL_last_swash_tmps;
4624 slen = PL_last_swash_slen;
4627 /* Try our second-level swatch cache, kept in a hash. */
4628 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4630 /* If not cached, generate it via swatch_get */
4631 if (!svp || !SvPOK(*svp)
4632 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4635 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4636 swatch = swatch_get(swash,
4637 code_point & ~((UV)needents - 1),
4640 else { /* For the first 256 code points, the swatch has a key of
4642 swatch = swatch_get(swash, 0, needents);
4645 if (IN_PERL_COMPILETIME)
4646 CopHINTS_set(PL_curcop, PL_hints);
4648 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4650 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4651 || (slen << 3) < needents)
4652 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4653 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4654 svp, tmps, (UV)slen, (UV)needents);
4657 PL_last_swash_hv = hv;
4658 assert(klen <= sizeof(PL_last_swash_key));
4659 PL_last_swash_klen = (U8)klen;
4660 /* FIXME change interpvar.h? */
4661 PL_last_swash_tmps = (U8 *) tmps;
4662 PL_last_swash_slen = slen;
4664 Copy(ptr, PL_last_swash_key, klen, U8);
4667 switch ((int)((slen << 3) / needents)) {
4669 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4671 return ((UV) tmps[off]);
4675 ((UV) tmps[off ] << 8) +
4676 ((UV) tmps[off + 1]);
4680 ((UV) tmps[off ] << 24) +
4681 ((UV) tmps[off + 1] << 16) +
4682 ((UV) tmps[off + 2] << 8) +
4683 ((UV) tmps[off + 3]);
4685 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4686 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4687 NORETURN_FUNCTION_END;
4690 /* Read a single line of the main body of the swash input text. These are of
4693 * where each number is hex. The first two numbers form the minimum and
4694 * maximum of a range, and the third is the value associated with the range.
4695 * Not all swashes should have a third number
4697 * On input: l points to the beginning of the line to be examined; it points
4698 * to somewhere in the string of the whole input text, and is
4699 * terminated by a \n or the null string terminator.
4700 * lend points to the null terminator of that string
4701 * wants_value is non-zero if the swash expects a third number
4702 * typestr is the name of the swash's mapping, like 'ToLower'
4703 * On output: *min, *max, and *val are set to the values read from the line.
4704 * returns a pointer just beyond the line examined. If there was no
4705 * valid min number on the line, returns lend+1
4709 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4710 const bool wants_value, const U8* const typestr)
4712 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4713 STRLEN numlen; /* Length of the number */
4714 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4715 | PERL_SCAN_DISALLOW_PREFIX
4716 | PERL_SCAN_SILENT_NON_PORTABLE;
4718 /* nl points to the next \n in the scan */
4719 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4721 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4723 /* Get the first number on the line: the range minimum */
4725 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4726 *max = *min; /* So can never return without setting max */
4727 if (numlen) /* If found a hex number, position past it */
4729 else if (nl) { /* Else, go handle next line, if any */
4730 return nl + 1; /* 1 is length of "\n" */
4732 else { /* Else, no next line */
4733 return lend + 1; /* to LIST's end at which \n is not found */
4736 /* The max range value follows, separated by a BLANK */
4739 flags = PERL_SCAN_SILENT_ILLDIGIT
4740 | PERL_SCAN_DISALLOW_PREFIX
4741 | PERL_SCAN_SILENT_NON_PORTABLE;
4743 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4746 else /* If no value here, it is a single element range */
4749 /* Non-binary tables have a third entry: what the first element of the
4750 * range maps to. The map for those currently read here is in hex */
4754 flags = PERL_SCAN_SILENT_ILLDIGIT
4755 | PERL_SCAN_DISALLOW_PREFIX
4756 | PERL_SCAN_SILENT_NON_PORTABLE;
4758 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4767 /* diag_listed_as: To%s: illegal mapping '%s' */
4768 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4774 *val = 0; /* bits == 1, then any val should be ignored */
4776 else { /* Nothing following range min, should be single element with no
4781 /* diag_listed_as: To%s: illegal mapping '%s' */
4782 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4786 *val = 0; /* bits == 1, then val should be ignored */
4789 /* Position to next line if any, or EOF */
4799 * Returns a swatch (a bit vector string) for a code point sequence
4800 * that starts from the value C<start> and comprises the number C<span>.
4801 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4802 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4805 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4808 U8 *l, *lend, *x, *xend, *s, *send;
4809 STRLEN lcur, xcur, scur;
4810 HV *const hv = MUTABLE_HV(SvRV(swash));
4811 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4813 SV** listsvp = NULL; /* The string containing the main body of the table */
4814 SV** extssvp = NULL;
4815 SV** invert_it_svp = NULL;
4818 STRLEN octets; /* if bits == 1, then octets == 0 */
4820 UV end = start + span;
4822 if (invlistsvp == NULL) {
4823 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4824 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4825 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4826 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4827 listsvp = hv_fetchs(hv, "LIST", FALSE);
4828 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4830 bits = SvUV(*bitssvp);
4831 none = SvUV(*nonesvp);
4832 typestr = (U8*)SvPV_nolen(*typesvp);
4838 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4840 PERL_ARGS_ASSERT_SWATCH_GET;
4842 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4843 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4847 /* If overflowed, use the max possible */
4853 /* create and initialize $swatch */
4854 scur = octets ? (span * octets) : (span + 7) / 8;
4855 swatch = newSV(scur);
4857 s = (U8*)SvPVX(swatch);
4858 if (octets && none) {
4859 const U8* const e = s + scur;
4862 *s++ = (U8)(none & 0xff);
4863 else if (bits == 16) {
4864 *s++ = (U8)((none >> 8) & 0xff);
4865 *s++ = (U8)( none & 0xff);
4867 else if (bits == 32) {
4868 *s++ = (U8)((none >> 24) & 0xff);
4869 *s++ = (U8)((none >> 16) & 0xff);
4870 *s++ = (U8)((none >> 8) & 0xff);
4871 *s++ = (U8)( none & 0xff);
4877 (void)memzero((U8*)s, scur + 1);
4879 SvCUR_set(swatch, scur);
4880 s = (U8*)SvPVX(swatch);
4882 if (invlistsvp) { /* If has an inversion list set up use that */
4883 _invlist_populate_swatch(*invlistsvp, start, end, s);
4887 /* read $swash->{LIST} */
4888 l = (U8*)SvPV(*listsvp, lcur);
4891 UV min, max, val, upper;
4892 l = swash_scan_list_line(l, lend, &min, &max, &val,
4893 cBOOL(octets), typestr);
4898 /* If looking for something beyond this range, go try the next one */
4902 /* <end> is generally 1 beyond where we want to set things, but at the
4903 * platform's infinity, where we can't go any higher, we want to
4904 * include the code point at <end> */
4907 : (max != UV_MAX || end != UV_MAX)
4914 if (!none || val < none) {
4919 for (key = min; key <= upper; key++) {
4921 /* offset must be non-negative (start <= min <= key < end) */
4922 offset = octets * (key - start);
4924 s[offset] = (U8)(val & 0xff);
4925 else if (bits == 16) {
4926 s[offset ] = (U8)((val >> 8) & 0xff);
4927 s[offset + 1] = (U8)( val & 0xff);
4929 else if (bits == 32) {
4930 s[offset ] = (U8)((val >> 24) & 0xff);
4931 s[offset + 1] = (U8)((val >> 16) & 0xff);
4932 s[offset + 2] = (U8)((val >> 8) & 0xff);
4933 s[offset + 3] = (U8)( val & 0xff);
4936 if (!none || val < none)
4940 else { /* bits == 1, then val should be ignored */
4945 for (key = min; key <= upper; key++) {
4946 const STRLEN offset = (STRLEN)(key - start);
4947 s[offset >> 3] |= 1 << (offset & 7);
4952 /* Invert if the data says it should be. Assumes that bits == 1 */
4953 if (invert_it_svp && SvUV(*invert_it_svp)) {
4955 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4956 * be 0, and their inversion should also be 0, as we don't succeed any
4957 * Unicode property matches for non-Unicode code points */
4958 if (start <= PERL_UNICODE_MAX) {
4960 /* The code below assumes that we never cross the
4961 * Unicode/above-Unicode boundary in a range, as otherwise we would
4962 * have to figure out where to stop flipping the bits. Since this
4963 * boundary is divisible by a large power of 2, and swatches comes
4964 * in small powers of 2, this should be a valid assumption */
4965 assert(start + span - 1 <= PERL_UNICODE_MAX);
4975 /* read $swash->{EXTRAS}
4976 * This code also copied to swash_to_invlist() below */
4977 x = (U8*)SvPV(*extssvp, xcur);
4985 SV **otherbitssvp, *other;
4989 const U8 opc = *x++;
4993 nl = (U8*)memchr(x, '\n', xend - x);
4995 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4997 x = nl + 1; /* 1 is length of "\n" */
5001 x = xend; /* to EXTRAS' end at which \n is not found */
5008 namelen = nl - namestr;
5012 namelen = xend - namestr;
5016 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
5017 otherhv = MUTABLE_HV(SvRV(*othersvp));
5018 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
5019 otherbits = (STRLEN)SvUV(*otherbitssvp);
5020 if (bits < otherbits)
5021 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
5022 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
5024 /* The "other" swatch must be destroyed after. */
5025 other = swatch_get(*othersvp, start, span);
5026 o = (U8*)SvPV(other, olen);
5029 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
5031 s = (U8*)SvPV(swatch, slen);
5032 if (bits == 1 && otherbits == 1) {
5034 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
5035 "mismatch, slen=%" UVuf ", olen=%" UVuf,
5036 (UV)slen, (UV)olen);
5060 STRLEN otheroctets = otherbits >> 3;
5062 U8* const send = s + slen;
5067 if (otherbits == 1) {
5068 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
5072 STRLEN vlen = otheroctets;
5080 if (opc == '+' && otherval)
5081 NOOP; /* replace with otherval */
5082 else if (opc == '!' && !otherval)
5084 else if (opc == '-' && otherval)
5086 else if (opc == '&' && !otherval)
5089 s += octets; /* no replacement */
5094 *s++ = (U8)( otherval & 0xff);
5095 else if (bits == 16) {
5096 *s++ = (U8)((otherval >> 8) & 0xff);
5097 *s++ = (U8)( otherval & 0xff);
5099 else if (bits == 32) {
5100 *s++ = (U8)((otherval >> 24) & 0xff);
5101 *s++ = (U8)((otherval >> 16) & 0xff);
5102 *s++ = (U8)((otherval >> 8) & 0xff);
5103 *s++ = (U8)( otherval & 0xff);
5107 sv_free(other); /* through with it! */
5113 Perl__swash_to_invlist(pTHX_ SV* const swash)
5116 /* Subject to change or removal. For use only in one place in regcomp.c.
5117 * Ownership is given to one reference count in the returned SV* */
5122 HV *const hv = MUTABLE_HV(SvRV(swash));
5123 UV elements = 0; /* Number of elements in the inversion list */
5133 STRLEN octets; /* if bits == 1, then octets == 0 */
5139 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
5141 /* If not a hash, it must be the swash's inversion list instead */
5142 if (SvTYPE(hv) != SVt_PVHV) {
5143 return SvREFCNT_inc_simple_NN((SV*) hv);
5146 /* The string containing the main body of the table */
5147 listsvp = hv_fetchs(hv, "LIST", FALSE);
5148 typesvp = hv_fetchs(hv, "TYPE", FALSE);
5149 bitssvp = hv_fetchs(hv, "BITS", FALSE);
5150 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
5151 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
5153 typestr = (U8*)SvPV_nolen(*typesvp);
5154 bits = SvUV(*bitssvp);
5155 octets = bits >> 3; /* if bits == 1, then octets == 0 */
5157 /* read $swash->{LIST} */
5158 if (SvPOK(*listsvp)) {
5159 l = (U8*)SvPV(*listsvp, lcur);
5162 /* LIST legitimately doesn't contain a string during compilation phases
5163 * of Perl itself, before the Unicode tables are generated. In this
5164 * case, just fake things up by creating an empty list */
5171 if (*l == 'V') { /* Inversion list format */
5172 const char *after_atou = (char *) lend;
5174 UV* other_elements_ptr;
5176 /* The first number is a count of the rest */
5178 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
5179 Perl_croak(aTHX_ "panic: Expecting a valid count of elements"
5180 " at start of inversion list");
5182 if (elements == 0) {
5183 invlist = _new_invlist(0);
5186 l = (U8 *) after_atou;
5188 /* Get the 0th element, which is needed to setup the inversion list
5190 while (isSPACE(*l)) l++;
5191 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
5192 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for"
5195 l = (U8 *) after_atou;
5196 invlist = _setup_canned_invlist(elements, element0,
5197 &other_elements_ptr);
5200 /* Then just populate the rest of the input */
5201 while (elements-- > 0) {
5203 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more"
5204 " elements than available", elements);
5206 while (isSPACE(*l)) l++;
5207 if (!grok_atoUV((const char *)l, other_elements_ptr++,
5210 Perl_croak(aTHX_ "panic: Expecting a valid element"
5211 " in inversion list");
5213 l = (U8 *) after_atou;
5219 /* Scan the input to count the number of lines to preallocate array
5220 * size based on worst possible case, which is each line in the input
5221 * creates 2 elements in the inversion list: 1) the beginning of a
5222 * range in the list; 2) the beginning of a range not in the list. */
5223 while ((loc = (char *) memchr(loc, '\n', lend - (U8 *) loc)) != NULL) {
5228 /* If the ending is somehow corrupt and isn't a new line, add another
5229 * element for the final range that isn't in the inversion list */
5230 if (! (*lend == '\n'
5231 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
5236 invlist = _new_invlist(elements);
5238 /* Now go through the input again, adding each range to the list */
5241 UV val; /* Not used by this function */
5243 l = swash_scan_list_line(l, lend, &start, &end, &val,
5244 cBOOL(octets), typestr);
5250 invlist = _add_range_to_invlist(invlist, start, end);
5254 /* Invert if the data says it should be */
5255 if (invert_it_svp && SvUV(*invert_it_svp)) {
5256 _invlist_invert(invlist);
5259 /* This code is copied from swatch_get()
5260 * read $swash->{EXTRAS} */
5261 x = (U8*)SvPV(*extssvp, xcur);
5269 SV **otherbitssvp, *other;
5272 const U8 opc = *x++;
5276 nl = (U8*)memchr(x, '\n', xend - x);
5278 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
5280 x = nl + 1; /* 1 is length of "\n" */
5284 x = xend; /* to EXTRAS' end at which \n is not found */
5291 namelen = nl - namestr;
5295 namelen = xend - namestr;
5299 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
5300 otherhv = MUTABLE_HV(SvRV(*othersvp));
5301 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
5302 otherbits = (STRLEN)SvUV(*otherbitssvp);
5304 if (bits != otherbits || bits != 1) {
5305 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
5306 "properties, bits=%" UVuf ", otherbits=%" UVuf,
5307 (UV)bits, (UV)otherbits);
5310 /* The "other" swatch must be destroyed after. */
5311 other = _swash_to_invlist((SV *)*othersvp);
5313 /* End of code copied from swatch_get() */
5316 _invlist_union(invlist, other, &invlist);
5319 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
5322 _invlist_subtract(invlist, other, &invlist);
5325 _invlist_intersection(invlist, other, &invlist);
5330 sv_free(other); /* through with it! */
5333 SvREADONLY_on(invlist);
5338 Perl__get_swash_invlist(pTHX_ SV* const swash)
5342 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
5344 if (! SvROK(swash)) {
5348 /* If it really isn't a hash, it isn't really swash; must be an inversion
5350 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
5354 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
5363 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
5365 /* May change: warns if surrogates, non-character code points, or
5366 * non-Unicode code points are in 's' which has length 'len' bytes.
5367 * Returns TRUE if none found; FALSE otherwise. The only other validity
5368 * check is to make sure that this won't exceed the string's length nor
5371 const U8* const e = s + len;
5374 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
5377 if (UTF8SKIP(s) > len) {
5378 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
5379 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
5382 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
5383 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
5384 if ( ckWARN_d(WARN_NON_UNICODE)
5385 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
5386 0 /* Don't consider overlongs */
5389 /* A side effect of this function will be to warn */
5390 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
5394 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
5395 if (ckWARN_d(WARN_SURROGATE)) {
5396 /* This has a different warning than the one the called
5397 * function would output, so can't just call it, unlike we
5398 * do for the non-chars and above-unicodes */
5399 UV uv = utf8_to_uvchr_buf(s, e, NULL);
5400 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
5401 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
5406 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
5407 && (ckWARN_d(WARN_NONCHAR)))
5409 /* A side effect of this function will be to warn */
5410 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5421 =for apidoc pv_uni_display
5423 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5424 length C<len>, the displayable version being at most C<pvlim> bytes long
5425 (if longer, the rest is truncated and C<"..."> will be appended).
5427 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5428 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5429 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5430 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5431 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5432 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5434 The pointer to the PV of the C<dsv> is returned.
5436 See also L</sv_uni_display>.
5440 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
5446 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5450 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5452 /* This serves double duty as a flag and a character to print after
5453 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5457 if (pvlim && SvCUR(dsv) >= pvlim) {
5461 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5463 const unsigned char c = (unsigned char)u & 0xFF;
5464 if (flags & UNI_DISPLAY_BACKSLASH) {
5481 const char string = ok;
5482 sv_catpvs(dsv, "\\");
5483 sv_catpvn(dsv, &string, 1);
5486 /* isPRINT() is the locale-blind version. */
5487 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5488 const char string = c;
5489 sv_catpvn(dsv, &string, 1);
5494 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5497 sv_catpvs(dsv, "...");
5503 =for apidoc sv_uni_display
5505 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5506 the displayable version being at most C<pvlim> bytes long
5507 (if longer, the rest is truncated and "..." will be appended).
5509 The C<flags> argument is as in L</pv_uni_display>().
5511 The pointer to the PV of the C<dsv> is returned.
5516 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5518 const char * const ptr =
5519 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5521 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5523 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5524 SvCUR(ssv), pvlim, flags);
5528 =for apidoc foldEQ_utf8
5530 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
5531 both of which may be in UTF-8) are the same case-insensitively; false
5532 otherwise. How far into the strings to compare is determined by other input
5535 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5536 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
5537 C<u2> with respect to C<s2>.
5539 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
5540 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
5541 The scan will not be considered to be a match unless the goal is reached, and
5542 scanning won't continue past that goal. Correspondingly for C<l2> with respect
5545 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
5546 pointer is considered an end pointer to the position 1 byte past the maximum
5547 point in C<s1> beyond which scanning will not continue under any circumstances.
5548 (This routine assumes that UTF-8 encoded input strings are not malformed;
5549 malformed input can cause it to read past C<pe1>). This means that if both
5550 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
5551 will never be successful because it can never
5552 get as far as its goal (and in fact is asserted against). Correspondingly for
5553 C<pe2> with respect to C<s2>.
5555 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5556 C<l2> must be non-zero), and if both do, both have to be
5557 reached for a successful match. Also, if the fold of a character is multiple
5558 characters, all of them must be matched (see tr21 reference below for
5561 Upon a successful match, if C<pe1> is non-C<NULL>,
5562 it will be set to point to the beginning of the I<next> character of C<s1>
5563 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5565 For case-insensitiveness, the "casefolding" of Unicode is used
5566 instead of upper/lowercasing both the characters, see
5567 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5571 /* A flags parameter has been added which may change, and hence isn't
5572 * externally documented. Currently it is:
5573 * 0 for as-documented above
5574 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5575 ASCII one, to not match
5576 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5577 * locale are to be used.
5578 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5579 * routine. This allows that step to be skipped.
5580 * Currently, this requires s1 to be encoded as UTF-8
5581 * (u1 must be true), which is asserted for.
5582 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5583 * cross certain boundaries. Hence, the caller should
5584 * let this function do the folding instead of
5585 * pre-folding. This code contains an assertion to
5586 * that effect. However, if the caller knows what
5587 * it's doing, it can pass this flag to indicate that,
5588 * and the assertion is skipped.
5589 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5590 * FOLDEQ_S2_FOLDS_SANE
5593 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
5594 const char *s2, char **pe2, UV l2, bool u2,
5597 const U8 *p1 = (const U8*)s1; /* Point to current char */
5598 const U8 *p2 = (const U8*)s2;
5599 const U8 *g1 = NULL; /* goal for s1 */
5600 const U8 *g2 = NULL;
5601 const U8 *e1 = NULL; /* Don't scan s1 past this */
5602 U8 *f1 = NULL; /* Point to current folded */
5603 const U8 *e2 = NULL;
5605 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5606 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5607 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5608 U8 flags_for_folder = FOLD_FLAGS_FULL;
5610 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5612 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5613 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5614 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5615 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5616 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5617 /* The algorithm is to trial the folds without regard to the flags on
5618 * the first line of the above assert(), and then see if the result
5619 * violates them. This means that the inputs can't be pre-folded to a
5620 * violating result, hence the assert. This could be changed, with the
5621 * addition of extra tests here for the already-folded case, which would
5622 * slow it down. That cost is more than any possible gain for when these
5623 * flags are specified, as the flags indicate /il or /iaa matching which
5624 * is less common than /iu, and I (khw) also believe that real-world /il
5625 * and /iaa matches are most likely to involve code points 0-255, and this
5626 * function only under rare conditions gets called for 0-255. */
5628 if (flags & FOLDEQ_LOCALE) {
5629 if (IN_UTF8_CTYPE_LOCALE) {
5630 flags &= ~FOLDEQ_LOCALE;
5633 flags_for_folder |= FOLD_FLAGS_LOCALE;
5642 g1 = (const U8*)s1 + l1;
5650 g2 = (const U8*)s2 + l2;
5653 /* Must have at least one goal */
5658 /* Will never match if goal is out-of-bounds */
5659 assert(! e1 || e1 >= g1);
5661 /* Here, there isn't an end pointer, or it is beyond the goal. We
5662 * only go as far as the goal */
5666 assert(e1); /* Must have an end for looking at s1 */
5669 /* Same for goal for s2 */
5671 assert(! e2 || e2 >= g2);
5678 /* If both operands are already folded, we could just do a memEQ on the
5679 * whole strings at once, but it would be better if the caller realized
5680 * this and didn't even call us */
5682 /* Look through both strings, a character at a time */
5683 while (p1 < e1 && p2 < e2) {
5685 /* If at the beginning of a new character in s1, get its fold to use
5686 * and the length of the fold. */
5688 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5694 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5696 /* We have to forbid mixing ASCII with non-ASCII if the
5697 * flags so indicate. And, we can short circuit having to
5698 * call the general functions for this common ASCII case,
5699 * all of whose non-locale folds are also ASCII, and hence
5700 * UTF-8 invariants, so the UTF8ness of the strings is not
5702 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5706 *foldbuf1 = toFOLD(*p1);
5709 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5711 else { /* Not UTF-8, get UTF-8 fold */
5712 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5718 if (n2 == 0) { /* Same for s2 */
5719 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5725 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5726 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5730 *foldbuf2 = toFOLD(*p2);
5733 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5736 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5742 /* Here f1 and f2 point to the beginning of the strings to compare.
5743 * These strings are the folds of the next character from each input
5744 * string, stored in UTF-8. */
5746 /* While there is more to look for in both folds, see if they
5747 * continue to match */
5749 U8 fold_length = UTF8SKIP(f1);
5750 if (fold_length != UTF8SKIP(f2)
5751 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5752 function call for single
5754 || memNE((char*)f1, (char*)f2, fold_length))
5756 return 0; /* mismatch */
5759 /* Here, they matched, advance past them */
5766 /* When reach the end of any fold, advance the input past it */
5768 p1 += u1 ? UTF8SKIP(p1) : 1;
5771 p2 += u2 ? UTF8SKIP(p2) : 1;
5773 } /* End of loop through both strings */
5775 /* A match is defined by each scan that specified an explicit length
5776 * reaching its final goal, and the other not having matched a partial
5777 * character (which can happen when the fold of a character is more than one
5779 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5783 /* Successful match. Set output pointers */
5793 /* XXX The next two functions should likely be moved to mathoms.c once all
5794 * occurrences of them are removed from the core; some cpan-upstream modules
5798 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5800 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5802 return uvoffuni_to_utf8_flags(d, uv, 0);
5806 =for apidoc utf8n_to_uvuni
5808 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5810 This function was useful for code that wanted to handle both EBCDIC and
5811 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5812 distinctions between the platforms have mostly been made invisible to most
5813 code, so this function is quite unlikely to be what you want. If you do need
5814 this precise functionality, use instead
5815 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5816 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5822 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5824 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5826 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5830 =for apidoc uvuni_to_utf8_flags
5832 Instead you almost certainly want to use L</uvchr_to_utf8> or
5833 L</uvchr_to_utf8_flags>.
5835 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5836 which itself, while not deprecated, should be used only in isolated
5837 circumstances. These functions were useful for code that wanted to handle
5838 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5839 v5.20, the distinctions between the platforms have mostly been made invisible
5840 to most code, so this function is quite unlikely to be what you want.
5846 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5848 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5850 return uvoffuni_to_utf8_flags(d, uv, flags);
5854 * ex: set ts=8 sts=4 sw=4 et: