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"
35 #include "uni_keywords.h"
37 static const char malformed_text[] = "Malformed UTF-8 character";
38 static const char unees[] =
39 "Malformed UTF-8 character (unexpected end of string)";
41 /* Be sure to synchronize this message with the similar one in regcomp.c */
42 static const char cp_above_legal_max[] =
43 "Use of code point 0x%" UVXf " is not allowed; the"
44 " permissible max is 0x%" UVXf;
46 #define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
49 =head1 Unicode Support
50 These are various utility functions for manipulating UTF8-encoded
51 strings. For the uninitiated, this is a method of representing arbitrary
52 Unicode characters as a variable number of bytes, in such a way that
53 characters in the ASCII range are unmodified, and a zero byte never appears
54 within non-zero characters.
60 Perl__force_out_malformed_utf8_message(pTHX_
61 const U8 *const p, /* First byte in UTF-8 sequence */
62 const U8 * const e, /* Final byte in sequence (may include
64 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
65 usually 0, or some DISALLOW flags */
66 const bool die_here) /* If TRUE, this function does not return */
68 /* This core-only function is to be called when a malformed UTF-8 character
69 * is found, in order to output the detailed information about the
70 * malformation before dieing. The reason it exists is for the occasions
71 * when such a malformation is fatal, but warnings might be turned off, so
72 * that normally they would not be actually output. This ensures that they
73 * do get output. Because a sequence may be malformed in more than one
74 * way, multiple messages may be generated, so we can't make them fatal, as
75 * that would cause the first one to die.
77 * Instead we pretend -W was passed to perl, then die afterwards. The
78 * flexibility is here to return to the caller so they can finish up and
82 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
88 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
90 PL_curcop->cop_warnings = pWARN_ALL;
93 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
98 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
99 " be called only when there are errors found");
103 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
108 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
109 U32 categories, /* Packed warning categories */
110 U32 flag) /* Flag associated with this message */
112 /* Creates, populates, and returns an HV* that describes an error message
113 * for the translators between UTF8 and code point */
115 SV* msg_sv = newSVpv(message, 0);
116 SV* category_sv = newSVuv(categories);
117 SV* flag_bit_sv = newSVuv(flag);
119 HV* msg_hv = newHV();
121 PERL_ARGS_ASSERT_NEW_MSG_HV;
123 (void) hv_stores(msg_hv, "text", msg_sv);
124 (void) hv_stores(msg_hv, "warn_categories", category_sv);
125 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
131 =for apidoc uvoffuni_to_utf8_flags
133 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
134 Instead, B<Almost all code should use L</uvchr_to_utf8> or
135 L</uvchr_to_utf8_flags>>.
137 This function is like them, but the input is a strict Unicode
138 (as opposed to native) code point. Only in very rare circumstances should code
139 not be using the native code point.
141 For details, see the description for L</uvchr_to_utf8_flags>.
147 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
149 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
151 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
154 /* All these formats take a single UV code point argument */
155 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
156 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
157 " is not recommended for open interchange";
158 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
159 " may not be portable";
160 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
161 " Unicode, requires a Perl extension," \
162 " and so is not portable";
164 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
166 if (flags & UNICODE_WARN_SURROGATE) { \
167 U32 category = packWARN(WARN_SURROGATE); \
168 const char * format = surrogate_cp_format; \
170 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
172 UNICODE_GOT_SURROGATE); \
175 Perl_ck_warner_d(aTHX_ category, format, uv); \
178 if (flags & UNICODE_DISALLOW_SURROGATE) { \
183 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
185 if (flags & UNICODE_WARN_NONCHAR) { \
186 U32 category = packWARN(WARN_NONCHAR); \
187 const char * format = nonchar_cp_format; \
189 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
191 UNICODE_GOT_NONCHAR); \
194 Perl_ck_warner_d(aTHX_ category, format, uv); \
197 if (flags & UNICODE_DISALLOW_NONCHAR) { \
202 /* Use shorter names internally in this file */
203 #define SHIFT UTF_ACCUMULATION_SHIFT
205 #define MARK UTF_CONTINUATION_MARK
206 #define MASK UTF_CONTINUATION_MASK
209 =for apidoc uvchr_to_utf8_flags_msgs
211 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
213 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
215 This function is for code that wants any warning and/or error messages to be
216 returned to the caller rather than be displayed. All messages that would have
217 been displayed if all lexical warnings are enabled will be returned.
219 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
220 placed after all the others, C<msgs>. If this parameter is 0, this function
221 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
222 be a pointer to an C<HV *> variable, in which this function creates a new HV to
223 contain any appropriate messages. The hash has three key-value pairs, as
230 The text of the message as a C<SVpv>.
232 =item C<warn_categories>
234 The warning category (or categories) packed into a C<SVuv>.
238 A single flag bit associated with this message, in a C<SVuv>.
239 The bit corresponds to some bit in the C<*errors> return value,
240 such as C<UNICODE_GOT_SURROGATE>.
244 It's important to note that specifying this parameter as non-null will cause
245 any warnings this function would otherwise generate to be suppressed, and
246 instead be placed in C<*msgs>. The caller can check the lexical warnings state
247 (or not) when choosing what to do with the returned messages.
249 The caller, of course, is responsible for freeing any returned HV.
254 /* Undocumented; we don't want people using this. Instead they should use
255 * uvchr_to_utf8_flags_msgs() */
257 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
259 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
265 if (OFFUNI_IS_INVARIANT(uv)) {
266 *d++ = LATIN1_TO_NATIVE(uv);
270 if (uv <= MAX_UTF8_TWO_BYTE) {
271 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
272 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
276 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
277 * below, the 16 is for start bytes E0-EF (which are all the possible ones
278 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
279 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
280 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
281 * 0x800-0xFFFF on ASCII */
282 if (uv < (16 * (1U << (2 * SHIFT)))) {
283 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
284 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
285 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
287 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
288 aren't tested here */
289 /* The most likely code points in this range are below the surrogates.
290 * Do an extra test to quickly exclude those. */
291 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
292 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
293 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
295 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
297 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
298 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
305 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
306 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
307 * happen starting with 4-byte characters on ASCII platforms. We unify the
308 * code for these with EBCDIC, even though some of them require 5-bytes on
309 * those, because khw believes the code saving is worth the very slight
310 * performance hit on these high EBCDIC code points. */
312 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
313 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
314 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
316 if ( (flags & UNICODE_WARN_SUPER)
317 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
318 && UNICODE_IS_PERL_EXTENDED(uv)))
320 const char * format = super_cp_format;
321 U32 category = packWARN(WARN_NON_UNICODE);
322 U32 flag = UNICODE_GOT_SUPER;
324 /* Choose the more dire applicable warning */
325 if (UNICODE_IS_PERL_EXTENDED(uv)) {
326 format = perl_extended_cp_format;
327 if (flags & (UNICODE_WARN_PERL_EXTENDED
328 |UNICODE_DISALLOW_PERL_EXTENDED))
330 flag = UNICODE_GOT_PERL_EXTENDED;
335 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
339 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE), format, uv);
342 if ( (flags & UNICODE_DISALLOW_SUPER)
343 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
344 && UNICODE_IS_PERL_EXTENDED(uv)))
349 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
350 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
353 /* Test for and handle 4-byte result. In the test immediately below, the
354 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
355 * characters). The 3 is for 3 continuation bytes; these each contribute
356 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
357 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
358 * 0x1_0000-0x1F_FFFF on ASCII */
359 if (uv < (8 * (1U << (3 * SHIFT)))) {
360 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
361 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
362 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
363 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
365 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
366 characters. The end-plane non-characters for EBCDIC were
367 handled just above */
368 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
369 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
371 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
372 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
379 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
380 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
381 * format. The unrolled version above turns out to not save all that much
382 * time, and at these high code points (well above the legal Unicode range
383 * on ASCII platforms, and well above anything in common use in EBCDIC),
384 * khw believes that less code outweighs slight performance gains. */
387 STRLEN len = OFFUNISKIP(uv);
390 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
393 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
399 =for apidoc uvchr_to_utf8
401 Adds the UTF-8 representation of the native code point C<uv> to the end
402 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
403 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
404 the byte after the end of the new character. In other words,
406 d = uvchr_to_utf8(d, uv);
408 is the recommended wide native character-aware way of saying
412 This function accepts any code point from 0..C<IV_MAX> as input.
413 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
415 It is possible to forbid or warn on non-Unicode code points, or those that may
416 be problematic by using L</uvchr_to_utf8_flags>.
421 /* This is also a macro */
422 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
425 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
427 return uvchr_to_utf8(d, uv);
431 =for apidoc uvchr_to_utf8_flags
433 Adds the UTF-8 representation of the native code point C<uv> to the end
434 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
435 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
436 the byte after the end of the new character. In other words,
438 d = uvchr_to_utf8_flags(d, uv, flags);
442 d = uvchr_to_utf8_flags(d, uv, 0);
444 This is the Unicode-aware way of saying
448 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
449 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
451 Specifying C<flags> can further restrict what is allowed and not warned on, as
454 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
455 the function will raise a warning, provided UTF8 warnings are enabled. If
456 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
457 NULL. If both flags are set, the function will both warn and return NULL.
459 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
460 affect how the function handles a Unicode non-character.
462 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
463 affect the handling of code points that are above the Unicode maximum of
464 0x10FFFF. Languages other than Perl may not be able to accept files that
467 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
468 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
469 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
470 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
471 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
472 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
473 above-Unicode and surrogate flags, but not the non-character ones, as
475 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
476 See L<perlunicode/Noncharacter code points>.
478 Extremely high code points were never specified in any standard, and require an
479 extension to UTF-8 to express, which Perl does. It is likely that programs
480 written in something other than Perl would not be able to read files that
481 contain these; nor would Perl understand files written by something that uses a
482 different extension. For these reasons, there is a separate set of flags that
483 can warn and/or disallow these extremely high code points, even if other
484 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
485 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
486 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
487 treat all above-Unicode code points, including these, as malformations. (Note
488 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
489 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
491 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
492 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
493 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
494 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
495 platforms,these flags can apply to code points that actually do fit in 31 bits.
496 The new names accurately describe the situation in all cases.
501 /* This is also a macro */
502 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
505 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
507 return uvchr_to_utf8_flags(d, uv, flags);
513 S_is_utf8_cp_above_31_bits(const U8 * const s,
515 const bool consider_overlongs)
517 /* Returns TRUE if the first code point represented by the Perl-extended-
518 * UTF-8-encoded string starting at 's', and looking no further than 'e -
519 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
521 * The function handles the case where the input bytes do not include all
522 * the ones necessary to represent a full character. That is, they may be
523 * the intial bytes of the representation of a code point, but possibly
524 * the final ones necessary for the complete representation may be beyond
527 * The function also can handle the case where the input is an overlong
528 * sequence. If 'consider_overlongs' is 0, the function assumes the
529 * input is not overlong, without checking, and will return based on that
530 * assumption. If this parameter is 1, the function will go to the trouble
531 * of figuring out if it actually evaluates to above or below 31 bits.
533 * The sequence is otherwise assumed to be well-formed, without checking.
536 const STRLEN len = e - s;
539 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
541 assert(! UTF8_IS_INVARIANT(*s) && e > s);
545 PERL_UNUSED_ARG(consider_overlongs);
547 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
548 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
549 * also be the start byte for a 31-bit code point; we need at least 2
550 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
551 * the start byte for an overlong sequence, but for 30-bit or smaller code
552 * points, so we don't have to worry about overlongs on EBCDIC.) */
563 /* On ASCII, FE and FF are the only start bytes that can evaluate to
564 * needing more than 31 bits. */
565 if (LIKELY(*s < 0xFE)) {
569 /* What we have left are FE and FF. Both of these require more than 31
570 * bits unless they are for overlongs. */
571 if (! consider_overlongs) {
575 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
576 * above 31 bits. But we need more than one byte to discern this, so if
577 * passed just the start byte, it could be an overlong evaluating to
583 /* Having excluded len==1, and knowing that FE and FF are both valid start
584 * bytes, we can call the function below to see if the sequence is
585 * overlong. (We don't need the full generality of the called function,
586 * but for these huge code points, speed shouldn't be a consideration, and
587 * the compiler does have enough information, since it's static to this
588 * file, to optimize to just the needed parts.) */
589 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
591 /* If it isn't overlong, more than 31 bits are required. */
592 if (is_overlong == 0) {
596 /* If it is indeterminate if it is overlong, return that */
597 if (is_overlong < 0) {
601 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
602 * the max it can be is 2**31 - 1 */
609 /* Here, ASCII and EBCDIC rejoin:
610 * On ASCII: We have an overlong sequence starting with FF
611 * On EBCDIC: We have a sequence starting with FE. */
613 { /* For C89, use a block so the declaration can be close to its use */
617 /* U+7FFFFFFF (2 ** 31 - 1)
618 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
619 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
620 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
621 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
622 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
623 * U+80000000 (2 ** 31):
624 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
625 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
626 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
627 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
629 * and since we know that *s = \xfe, any continuation sequcence
630 * following it that is gt the below is above 31 bits
631 [0] [1] [2] [3] [4] [5] [6] */
632 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
636 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
637 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
638 * FF overlong for U+80000000 (2 ** 31):
639 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
640 * and since we know that *s = \xff, any continuation sequcence
641 * following it that is gt the below is above 30 bits
642 [0] [1] [2] [3] [4] [5] [6] */
643 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
647 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
648 const STRLEN cmp_len = MIN(conts_len, len - 1);
650 /* Now compare the continuation bytes in s with the ones we have
651 * compiled in that are for the largest 30 bit code point. If we have
652 * enough bytes available to determine the answer, or the bytes we do
653 * have differ from them, we can compare the two to get a definitive
654 * answer (Note that in UTF-EBCDIC, the two lowest possible
655 * continuation bytes are \x41 and \x42.) */
656 if (cmp_len >= conts_len || memNE(s + 1,
657 conts_for_highest_30_bit,
660 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
663 /* Here, all the bytes we have are the same as the highest 30-bit code
664 * point, but we are missing so many bytes that we can't make the
672 PERL_STATIC_INLINE int
673 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
675 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
676 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
677 * it isn't, and -1 if there isn't enough information to tell. This last
678 * return value can happen if the sequence is incomplete, missing some
679 * trailing bytes that would form a complete character. If there are
680 * enough bytes to make a definitive decision, this function does so.
681 * Usually 2 bytes sufficient.
683 * Overlongs can occur whenever the number of continuation bytes changes.
684 * That means whenever the number of leading 1 bits in a start byte
685 * increases from the next lower start byte. That happens for start bytes
686 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
687 * start bytes have already been excluded, so don't need to be tested here;
688 * ASCII platforms: C0, C1
689 * EBCDIC platforms C0, C1, C2, C3, C4, E0
692 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
693 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
695 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
696 assert(len > 1 && UTF8_IS_START(*s));
698 /* Each platform has overlongs after the start bytes given above (expressed
699 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
700 * the logic is the same, except the E0 overlong has already been excluded
701 * on EBCDIC platforms. The values below were found by manually
702 * inspecting the UTF-8 patterns. See the tables in utf8.h and
706 # define F0_ABOVE_OVERLONG 0xB0
707 # define F8_ABOVE_OVERLONG 0xA8
708 # define FC_ABOVE_OVERLONG 0xA4
709 # define FE_ABOVE_OVERLONG 0xA2
710 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
714 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
718 # define F0_ABOVE_OVERLONG 0x90
719 # define F8_ABOVE_OVERLONG 0x88
720 # define FC_ABOVE_OVERLONG 0x84
721 # define FE_ABOVE_OVERLONG 0x82
722 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
726 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
727 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
728 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
729 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
734 /* Check for the FF overlong */
735 return isFF_OVERLONG(s, len);
738 PERL_STATIC_INLINE int
739 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
741 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
742 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
743 * it isn't, and -1 if there isn't enough information to tell. This last
744 * return value can happen if the sequence is incomplete, missing some
745 * trailing bytes that would form a complete character. If there are
746 * enough bytes to make a definitive decision, this function does so. */
748 PERL_ARGS_ASSERT_ISFF_OVERLONG;
750 /* To be an FF overlong, all the available bytes must match */
751 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
752 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
757 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
758 * be there; what comes after them doesn't matter. See tables in utf8.h,
760 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
764 /* The missing bytes could cause the result to go one way or the other, so
765 * the result is indeterminate */
769 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
770 # ifdef EBCDIC /* Actually is I8 */
771 # define HIGHEST_REPRESENTABLE_UTF8 \
772 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
774 # define HIGHEST_REPRESENTABLE_UTF8 \
775 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
779 PERL_STATIC_INLINE int
780 S_does_utf8_overflow(const U8 * const s,
782 const bool consider_overlongs)
784 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
785 * 'e' - 1 would overflow an IV on this platform; that is if it represents
786 * a code point larger than the highest representable code point. It
787 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
788 * enough information to tell. This last return value can happen if the
789 * sequence is incomplete, missing some trailing bytes that would form a
790 * complete character. If there are enough bytes to make a definitive
791 * decision, this function does so.
793 * If 'consider_overlongs' is TRUE, the function checks for the possibility
794 * that the sequence is an overlong that doesn't overflow. Otherwise, it
795 * assumes the sequence is not an overlong. This can give different
796 * results only on ASCII 32-bit platforms.
798 * (For ASCII platforms, we could use memcmp() because we don't have to
799 * convert each byte to I8, but it's very rare input indeed that would
800 * approach overflow, so the loop below will likely only get executed once.)
802 * 'e' - 1 must not be beyond a full character. */
805 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
806 assert(s <= e && s + UTF8SKIP(s) >= e);
808 #if ! defined(UV_IS_QUAD)
810 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
814 PERL_UNUSED_ARG(consider_overlongs);
817 const STRLEN len = e - s;
819 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
821 for (x = s; x < e; x++, y++) {
823 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
827 /* If this byte is larger than the corresponding highest UTF-8
828 * byte, the sequence overflow; otherwise the byte is less than,
829 * and so the sequence doesn't overflow */
830 return NATIVE_UTF8_TO_I8(*x) > *y;
834 /* Got to the end and all bytes are the same. If the input is a whole
835 * character, it doesn't overflow. And if it is a partial character,
836 * there's not enough information to tell */
837 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
850 /* This is the portions of the above function that deal with UV_MAX instead of
851 * IV_MAX. They are left here in case we want to combine them so that internal
852 * uses can have larger code points. The only logic difference is that the
853 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
857 /* Anything larger than this will overflow the word if it were converted into a UV */
858 #if defined(UV_IS_QUAD)
859 # ifdef EBCDIC /* Actually is I8 */
860 # define HIGHEST_REPRESENTABLE_UTF8 \
861 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
863 # define HIGHEST_REPRESENTABLE_UTF8 \
864 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
868 # define HIGHEST_REPRESENTABLE_UTF8 \
869 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
871 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
875 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
877 /* On 32 bit ASCII machines, many overlongs that start with FF don't
879 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
881 /* To be such an overlong, the first bytes of 's' must match
882 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
883 * don't have any additional bytes available, the sequence, when
884 * completed might or might not fit in 32 bits. But if we have that
885 * next byte, we can tell for sure. If it is <= 0x83, then it does
887 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
891 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
894 /* Starting with the #else, the rest of the function is identical except
895 * 1. we need to move the 'len' declaration to be global to the function
896 * 2. the endif move to just after the UNUSED_ARG.
897 * An empty endif is given just below to satisfy the preprocessor
903 #undef F0_ABOVE_OVERLONG
904 #undef F8_ABOVE_OVERLONG
905 #undef FC_ABOVE_OVERLONG
906 #undef FE_ABOVE_OVERLONG
907 #undef FF_OVERLONG_PREFIX
910 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
915 /* A helper function that should not be called directly.
917 * This function returns non-zero if the string beginning at 's' and
918 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
919 * code point; otherwise it returns 0. The examination stops after the
920 * first code point in 's' is validated, not looking at the rest of the
921 * input. If 'e' is such that there are not enough bytes to represent a
922 * complete code point, this function will return non-zero anyway, if the
923 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
924 * excluded by 'flags'.
926 * A non-zero return gives the number of bytes required to represent the
927 * code point. Be aware that if the input is for a partial character, the
928 * return will be larger than 'e - s'.
930 * This function assumes that the code point represented is UTF-8 variant.
931 * The caller should have excluded the possibility of it being invariant
932 * before calling this function.
934 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
935 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
936 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
937 * disallowed by the flags. If the input is only for a partial character,
938 * the function will return non-zero if there is any sequence of
939 * well-formed UTF-8 that, when appended to the input sequence, could
940 * result in an allowed code point; otherwise it returns 0. Non characters
941 * cannot be determined based on partial character input. But many of the
942 * other excluded types can be determined with just the first one or two
947 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
949 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
950 |UTF8_DISALLOW_PERL_EXTENDED)));
951 assert(! UTF8_IS_INVARIANT(*s));
953 /* A variant char must begin with a start byte */
954 if (UNLIKELY(! UTF8_IS_START(*s))) {
958 /* Examine a maximum of a single whole code point */
959 if (e - s > UTF8SKIP(s)) {
965 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
966 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
968 /* Here, we are disallowing some set of largish code points, and the
969 * first byte indicates the sequence is for a code point that could be
970 * in the excluded set. We generally don't have to look beyond this or
971 * the second byte to see if the sequence is actually for one of the
972 * excluded classes. The code below is derived from this table:
974 * UTF-8 UTF-EBCDIC I8
975 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
976 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
977 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
979 * Keep in mind that legal continuation bytes range between \x80..\xBF
980 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
981 * continuation bytes. Hence, we don't have to test the upper edge
982 * because if any of those is encountered, the sequence is malformed,
983 * and would fail elsewhere in this function.
985 * The code here likewise assumes that there aren't other
986 * malformations; again the function should fail elsewhere because of
987 * these. For example, an overlong beginning with FC doesn't actually
988 * have to be a super; it could actually represent a small code point,
989 * even U+0000. But, since overlongs (and other malformations) are
990 * illegal, the function should return FALSE in either case.
993 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
994 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
995 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
997 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
999 && ((s1) & 0xFE ) == 0xB6)
1000 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1002 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1003 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1004 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1005 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1008 if ( (flags & UTF8_DISALLOW_SUPER)
1009 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1011 return 0; /* Above Unicode */
1014 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1015 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1021 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1023 if ( (flags & UTF8_DISALLOW_SUPER)
1024 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1026 return 0; /* Above Unicode */
1029 if ( (flags & UTF8_DISALLOW_SURROGATE)
1030 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1032 return 0; /* Surrogate */
1035 if ( (flags & UTF8_DISALLOW_NONCHAR)
1036 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1038 return 0; /* Noncharacter code point */
1043 /* Make sure that all that follows are continuation bytes */
1044 for (x = s + 1; x < e; x++) {
1045 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1050 /* Here is syntactically valid. Next, make sure this isn't the start of an
1052 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1056 /* And finally, that the code point represented fits in a word on this
1058 if (0 < does_utf8_overflow(s, e,
1059 0 /* Don't consider overlongs */
1069 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1071 /* Returns a mortalized C string that is a displayable copy of the 'len'
1072 * bytes starting at 'start'. 'format' gives how to display each byte.
1073 * Currently, there are only two formats, so it is currently a bool:
1075 * 1 ab (that is a space between two hex digit bytes)
1078 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1080 const U8 * s = start;
1081 const U8 * const e = start + len;
1085 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1087 Newx(output, output_len, char);
1091 for (s = start; s < e; s++) {
1092 const unsigned high_nibble = (*s & 0xF0) >> 4;
1093 const unsigned low_nibble = (*s & 0x0F);
1105 if (high_nibble < 10) {
1106 *d++ = high_nibble + '0';
1109 *d++ = high_nibble - 10 + 'a';
1112 if (low_nibble < 10) {
1113 *d++ = low_nibble + '0';
1116 *d++ = low_nibble - 10 + 'a';
1124 PERL_STATIC_INLINE char *
1125 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1127 /* Max number of bytes to print */
1130 /* Which one is the non-continuation */
1131 const STRLEN non_cont_byte_pos,
1133 /* How many bytes should there be? */
1134 const STRLEN expect_len)
1136 /* Return the malformation warning text for an unexpected continuation
1139 const char * const where = (non_cont_byte_pos == 1)
1141 : Perl_form(aTHX_ "%d bytes",
1142 (int) non_cont_byte_pos);
1143 const U8 * x = s + non_cont_byte_pos;
1144 const U8 * e = s + print_len;
1146 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1148 /* We don't need to pass this parameter, but since it has already been
1149 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1150 assert(expect_len == UTF8SKIP(s));
1152 /* As a defensive coding measure, don't output anything past a NUL. Such
1153 * bytes shouldn't be in the middle of a malformation, and could mark the
1154 * end of the allocated string, and what comes after is undefined */
1155 for (; x < e; x++) {
1157 x++; /* Output this particular NUL */
1162 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1163 " %s after start byte 0x%02x; need %d bytes, got %d)",
1165 _byte_dump_string(s, x - s, 0),
1166 *(s + non_cont_byte_pos),
1170 (int) non_cont_byte_pos);
1175 =for apidoc utf8n_to_uvchr
1177 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1178 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1180 Bottom level UTF-8 decode routine.
1181 Returns the native code point value of the first character in the string C<s>,
1182 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1183 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1184 the length, in bytes, of that character.
1186 The value of C<flags> determines the behavior when C<s> does not point to a
1187 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1188 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1189 is the next possible position in C<s> that could begin a non-malformed
1190 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1191 is raised. Some UTF-8 input sequences may contain multiple malformations.
1192 This function tries to find every possible one in each call, so multiple
1193 warnings can be raised for the same sequence.
1195 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1196 individual types of malformations, such as the sequence being overlong (that
1197 is, when there is a shorter sequence that can express the same code point;
1198 overlong sequences are expressly forbidden in the UTF-8 standard due to
1199 potential security issues). Another malformation example is the first byte of
1200 a character not being a legal first byte. See F<utf8.h> for the list of such
1201 flags. Even if allowed, this function generally returns the Unicode
1202 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1203 F<utf8.h> to override this behavior for the overlong malformations, but don't
1204 do that except for very specialized purposes.
1206 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1207 flags) malformation is found. If this flag is set, the routine assumes that
1208 the caller will raise a warning, and this function will silently just set
1209 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1211 Note that this API requires disambiguation between successful decoding a C<NUL>
1212 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1213 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1214 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1215 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1216 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1218 Certain code points are considered problematic. These are Unicode surrogates,
1219 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1220 By default these are considered regular code points, but certain situations
1221 warrant special handling for them, which can be specified using the C<flags>
1222 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1223 three classes are treated as malformations and handled as such. The flags
1224 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1225 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1226 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1227 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1228 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1230 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1231 The difference between traditional strictness and C9 strictness is that the
1232 latter does not forbid non-character code points. (They are still discouraged,
1233 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1235 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1236 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1237 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1238 raised for their respective categories, but otherwise the code points are
1239 considered valid (not malformations). To get a category to both be treated as
1240 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1241 (But note that warnings are not raised if lexically disabled nor if
1242 C<UTF8_CHECK_ONLY> is also specified.)
1244 Extremely high code points were never specified in any standard, and require an
1245 extension to UTF-8 to express, which Perl does. It is likely that programs
1246 written in something other than Perl would not be able to read files that
1247 contain these; nor would Perl understand files written by something that uses a
1248 different extension. For these reasons, there is a separate set of flags that
1249 can warn and/or disallow these extremely high code points, even if other
1250 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1251 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1252 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1253 above-Unicode code points, including these, as malformations.
1254 (Note that the Unicode standard considers anything above 0x10FFFF to be
1255 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1258 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1259 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1260 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1261 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1262 can apply to code points that actually do fit in 31 bits. This happens on
1263 EBCDIC platforms, and sometimes when the L<overlong
1264 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1265 describe the situation in all cases.
1268 All other code points corresponding to Unicode characters, including private
1269 use and those yet to be assigned, are never considered malformed and never
1274 Also implemented as a macro in utf8.h
1278 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1283 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1285 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1290 =for apidoc utf8n_to_uvchr_error
1292 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1293 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1295 This function is for code that needs to know what the precise malformation(s)
1296 are when an error is found. If you also need to know the generated warning
1297 messages, use L</utf8n_to_uvchr_msgs>() instead.
1299 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1300 all the others, C<errors>. If this parameter is 0, this function behaves
1301 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1302 to a C<U32> variable, which this function sets to indicate any errors found.
1303 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1304 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1305 of these bits will be set if a malformation is found, even if the input
1306 C<flags> parameter indicates that the given malformation is allowed; those
1307 exceptions are noted:
1311 =item C<UTF8_GOT_PERL_EXTENDED>
1313 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1314 set only if the input C<flags> parameter contains either the
1315 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1317 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1318 and so some extension must be used to express them. Perl uses a natural
1319 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1320 extension to represent even higher ones, so that any code point that fits in a
1321 64-bit word can be represented. Text using these extensions is not likely to
1322 be portable to non-Perl code. We lump both of these extensions together and
1323 refer to them as Perl extended UTF-8. There exist other extensions that people
1324 have invented, incompatible with Perl's.
1326 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1327 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1328 than on ASCII. Prior to that, code points 2**31 and higher were simply
1329 unrepresentable, and a different, incompatible method was used to represent
1330 code points between 2**30 and 2**31 - 1.
1332 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1333 Perl extended UTF-8 is used.
1335 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1336 may use for backward compatibility. That name is misleading, as this flag may
1337 be set when the code point actually does fit in 31 bits. This happens on
1338 EBCDIC platforms, and sometimes when the L<overlong
1339 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1340 describes the situation in all cases.
1342 =item C<UTF8_GOT_CONTINUATION>
1344 The input sequence was malformed in that the first byte was a a UTF-8
1347 =item C<UTF8_GOT_EMPTY>
1349 The input C<curlen> parameter was 0.
1351 =item C<UTF8_GOT_LONG>
1353 The input sequence was malformed in that there is some other sequence that
1354 evaluates to the same code point, but that sequence is shorter than this one.
1356 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1357 it was discovered that this created security issues.
1359 =item C<UTF8_GOT_NONCHAR>
1361 The code point represented by the input UTF-8 sequence is for a Unicode
1362 non-character code point.
1363 This bit is set only if the input C<flags> parameter contains either the
1364 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1366 =item C<UTF8_GOT_NON_CONTINUATION>
1368 The input sequence was malformed in that a non-continuation type byte was found
1369 in a position where only a continuation type one should be.
1371 =item C<UTF8_GOT_OVERFLOW>
1373 The input sequence was malformed in that it is for a code point that is not
1374 representable in the number of bits available in an IV on the current platform.
1376 =item C<UTF8_GOT_SHORT>
1378 The input sequence was malformed in that C<curlen> is smaller than required for
1379 a complete sequence. In other words, the input is for a partial character
1382 =item C<UTF8_GOT_SUPER>
1384 The input sequence was malformed in that it is for a non-Unicode code point;
1385 that is, one above the legal Unicode maximum.
1386 This bit is set only if the input C<flags> parameter contains either the
1387 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1389 =item C<UTF8_GOT_SURROGATE>
1391 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1393 This bit is set only if the input C<flags> parameter contains either the
1394 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1398 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1399 flag to suppress any warnings, and then examine the C<*errors> return.
1403 Also implemented as a macro in utf8.h
1407 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1413 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1415 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1420 =for apidoc utf8n_to_uvchr_msgs
1422 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1423 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1425 This function is for code that needs to know what the precise malformation(s)
1426 are when an error is found, and wants the corresponding warning and/or error
1427 messages to be returned to the caller rather than be displayed. All messages
1428 that would have been displayed if all lexcial warnings are enabled will be
1431 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1432 placed after all the others, C<msgs>. If this parameter is 0, this function
1433 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1434 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1435 contain any appropriate messages. The elements of the array are ordered so
1436 that the first message that would have been displayed is in the 0th element,
1437 and so on. Each element is a hash with three key-value pairs, as follows:
1443 The text of the message as a C<SVpv>.
1445 =item C<warn_categories>
1447 The warning category (or categories) packed into a C<SVuv>.
1451 A single flag bit associated with this message, in a C<SVuv>.
1452 The bit corresponds to some bit in the C<*errors> return value,
1453 such as C<UTF8_GOT_LONG>.
1457 It's important to note that specifying this parameter as non-null will cause
1458 any warnings this function would otherwise generate to be suppressed, and
1459 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1460 (or not) when choosing what to do with the returned messages.
1462 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1465 The caller, of course, is responsible for freeing any returned AV.
1471 Perl_utf8n_to_uvchr_msgs(pTHX_ const U8 *s,
1478 const U8 * const s0 = s;
1479 const U8 * send = s0 + curlen;
1480 U32 possible_problems; /* A bit is set here for each potential problem
1481 found as we go along */
1483 STRLEN expectlen; /* How long should this sequence be? */
1484 STRLEN avail_len; /* When input is too short, gives what that is */
1485 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1486 gets set and discarded */
1488 /* The below are used only if there is both an overlong malformation and a
1489 * too short one. Otherwise the first two are set to 's0' and 'send', and
1490 * the third not used at all */
1492 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1493 routine; see [perl #130921] */
1497 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_MSGS;
1499 /* Measurements show that this dfa is somewhat faster than the regular code
1500 * below, so use it first, dropping down for the non-normal cases. */
1502 #define PERL_UTF8_DECODE_REJECT 1
1504 while (s < send && LIKELY(state != PERL_UTF8_DECODE_REJECT)) {
1505 UV type = strict_utf8_dfa_tab[*s];
1508 ? ((0xff >> type) & NATIVE_UTF8_TO_I8(*s))
1509 : UTF8_ACCUMULATE(uv, *s);
1510 state = strict_utf8_dfa_tab[256 + state + type];
1514 *retlen = s - s0 + 1;
1529 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1530 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1531 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1536 /* Each of the affected Hanguls starts with \xED */
1538 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1549 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1550 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1551 | (s0[2] & UTF_CONTINUATION_MASK);
1556 /* In conjunction with the exhaustive tests that can be enabled in
1557 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1558 * what it is intended to do, and that no flaws in it are masked by
1559 * dropping down and executing the code below
1560 assert(! isUTF8_CHAR(s0, send)
1561 || UTF8_IS_SURROGATE(s0, send)
1562 || UTF8_IS_SUPER(s0, send)
1563 || UTF8_IS_NONCHAR(s0,send));
1568 possible_problems = 0;
1572 adjusted_s0 = (U8 *) s0;
1579 errors = &discard_errors;
1582 /* The order of malformation tests here is important. We should consume as
1583 * few bytes as possible in order to not skip any valid character. This is
1584 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1585 * http://unicode.org/reports/tr36 for more discussion as to why. For
1586 * example, once we've done a UTF8SKIP, we can tell the expected number of
1587 * bytes, and could fail right off the bat if the input parameters indicate
1588 * that there are too few available. But it could be that just that first
1589 * byte is garbled, and the intended character occupies fewer bytes. If we
1590 * blindly assumed that the first byte is correct, and skipped based on
1591 * that number, we could skip over a valid input character. So instead, we
1592 * always examine the sequence byte-by-byte.
1594 * We also should not consume too few bytes, otherwise someone could inject
1595 * things. For example, an input could be deliberately designed to
1596 * overflow, and if this code bailed out immediately upon discovering that,
1597 * returning to the caller C<*retlen> pointing to the very next byte (one
1598 * which is actually part of of the overflowing sequence), that could look
1599 * legitimate to the caller, which could discard the initial partial
1600 * sequence and process the rest, inappropriately.
1602 * Some possible input sequences are malformed in more than one way. This
1603 * function goes to lengths to try to find all of them. This is necessary
1604 * for correctness, as the inputs may allow one malformation but not
1605 * another, and if we abandon searching for others after finding the
1606 * allowed one, we could allow in something that shouldn't have been.
1609 if (UNLIKELY(curlen == 0)) {
1610 possible_problems |= UTF8_GOT_EMPTY;
1612 uv = UNICODE_REPLACEMENT;
1613 goto ready_to_handle_errors;
1616 expectlen = UTF8SKIP(s);
1618 /* A well-formed UTF-8 character, as the vast majority of calls to this
1619 * function will be for, has this expected length. For efficiency, set
1620 * things up here to return it. It will be overriden only in those rare
1621 * cases where a malformation is found */
1623 *retlen = expectlen;
1626 /* A continuation character can't start a valid sequence */
1627 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1628 possible_problems |= UTF8_GOT_CONTINUATION;
1630 uv = UNICODE_REPLACEMENT;
1631 goto ready_to_handle_errors;
1634 /* Here is not a continuation byte, nor an invariant. The only thing left
1635 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1636 * because it excludes start bytes like \xC0 that always lead to
1639 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1640 * that indicate the number of bytes in the character's whole UTF-8
1641 * sequence, leaving just the bits that are part of the value. */
1642 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1644 /* Setup the loop end point, making sure to not look past the end of the
1645 * input string, and flag it as too short if the size isn't big enough. */
1646 if (UNLIKELY(curlen < expectlen)) {
1647 possible_problems |= UTF8_GOT_SHORT;
1651 send = (U8*) s0 + expectlen;
1654 /* Now, loop through the remaining bytes in the character's sequence,
1655 * accumulating each into the working value as we go. */
1656 for (s = s0 + 1; s < send; s++) {
1657 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1658 uv = UTF8_ACCUMULATE(uv, *s);
1662 /* Here, found a non-continuation before processing all expected bytes.
1663 * This byte indicates the beginning of a new character, so quit, even
1664 * if allowing this malformation. */
1665 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1667 } /* End of loop through the character's bytes */
1669 /* Save how many bytes were actually in the character */
1672 /* Note that there are two types of too-short malformation. One is when
1673 * there is actual wrong data before the normal termination of the
1674 * sequence. The other is that the sequence wasn't complete before the end
1675 * of the data we are allowed to look at, based on the input 'curlen'.
1676 * This means that we were passed data for a partial character, but it is
1677 * valid as far as we saw. The other is definitely invalid. This
1678 * distinction could be important to a caller, so the two types are kept
1681 * A convenience macro that matches either of the too-short conditions. */
1682 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1684 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1686 uv = UNICODE_REPLACEMENT;
1689 /* Check for overflow. The algorithm requires us to not look past the end
1690 * of the current character, even if partial, so the upper limit is 's' */
1691 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1692 1 /* Do consider overlongs */
1695 possible_problems |= UTF8_GOT_OVERFLOW;
1696 uv = UNICODE_REPLACEMENT;
1699 /* Check for overlong. If no problems so far, 'uv' is the correct code
1700 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1701 * we must look at the UTF-8 byte sequence itself to see if it is for an
1703 if ( ( LIKELY(! possible_problems)
1704 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1705 || ( UNLIKELY(possible_problems)
1706 && ( UNLIKELY(! UTF8_IS_START(*s0))
1708 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1711 possible_problems |= UTF8_GOT_LONG;
1713 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1715 /* The calculation in the 'true' branch of this 'if'
1716 * below won't work if overflows, and isn't needed
1717 * anyway. Further below we handle all overflow
1719 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1721 UV min_uv = uv_so_far;
1724 /* Here, the input is both overlong and is missing some trailing
1725 * bytes. There is no single code point it could be for, but there
1726 * may be enough information present to determine if what we have
1727 * so far is for an unallowed code point, such as for a surrogate.
1728 * The code further below has the intelligence to determine this,
1729 * but just for non-overlong UTF-8 sequences. What we do here is
1730 * calculate the smallest code point the input could represent if
1731 * there were no too short malformation. Then we compute and save
1732 * the UTF-8 for that, which is what the code below looks at
1733 * instead of the raw input. It turns out that the smallest such
1734 * code point is all we need. */
1735 for (i = curlen; i < expectlen; i++) {
1736 min_uv = UTF8_ACCUMULATE(min_uv,
1737 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1740 adjusted_s0 = temp_char_buf;
1741 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1745 /* Here, we have found all the possible problems, except for when the input
1746 * is for a problematic code point not allowed by the input parameters. */
1748 /* uv is valid for overlongs */
1749 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1751 /* isn't problematic if < this */
1752 && uv >= UNICODE_SURROGATE_FIRST)
1753 || ( UNLIKELY(possible_problems)
1755 /* if overflow, we know without looking further
1756 * precisely which of the problematic types it is,
1757 * and we deal with those in the overflow handling
1759 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1760 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1761 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1762 && ((flags & ( UTF8_DISALLOW_NONCHAR
1763 |UTF8_DISALLOW_SURROGATE
1764 |UTF8_DISALLOW_SUPER
1765 |UTF8_DISALLOW_PERL_EXTENDED
1767 |UTF8_WARN_SURROGATE
1769 |UTF8_WARN_PERL_EXTENDED))))
1771 /* If there were no malformations, or the only malformation is an
1772 * overlong, 'uv' is valid */
1773 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1774 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1775 possible_problems |= UTF8_GOT_SURROGATE;
1777 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1778 possible_problems |= UTF8_GOT_SUPER;
1780 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1781 possible_problems |= UTF8_GOT_NONCHAR;
1784 else { /* Otherwise, need to look at the source UTF-8, possibly
1785 adjusted to be non-overlong */
1787 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1788 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1790 possible_problems |= UTF8_GOT_SUPER;
1792 else if (curlen > 1) {
1793 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1794 NATIVE_UTF8_TO_I8(*adjusted_s0),
1795 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1797 possible_problems |= UTF8_GOT_SUPER;
1799 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1800 NATIVE_UTF8_TO_I8(*adjusted_s0),
1801 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1803 possible_problems |= UTF8_GOT_SURROGATE;
1807 /* We need a complete well-formed UTF-8 character to discern
1808 * non-characters, so can't look for them here */
1812 ready_to_handle_errors:
1815 * curlen contains the number of bytes in the sequence that
1816 * this call should advance the input by.
1817 * avail_len gives the available number of bytes passed in, but
1818 * only if this is less than the expected number of
1819 * bytes, based on the code point's start byte.
1820 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1821 * is set in it for each potential problem found.
1822 * uv contains the code point the input sequence
1823 * represents; or if there is a problem that prevents
1824 * a well-defined value from being computed, it is
1825 * some subsitute value, typically the REPLACEMENT
1827 * s0 points to the first byte of the character
1828 * s points to just after were we left off processing
1830 * send points to just after where that character should
1831 * end, based on how many bytes the start byte tells
1832 * us should be in it, but no further than s0 +
1836 if (UNLIKELY(possible_problems)) {
1837 bool disallowed = FALSE;
1838 const U32 orig_problems = possible_problems;
1844 while (possible_problems) { /* Handle each possible problem */
1846 char * message = NULL;
1847 U32 this_flag_bit = 0;
1849 /* Each 'if' clause handles one problem. They are ordered so that
1850 * the first ones' messages will be displayed before the later
1851 * ones; this is kinda in decreasing severity order. But the
1852 * overlong must come last, as it changes 'uv' looked at by the
1854 if (possible_problems & UTF8_GOT_OVERFLOW) {
1856 /* Overflow means also got a super and are using Perl's
1857 * extended UTF-8, but we handle all three cases here */
1859 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1860 *errors |= UTF8_GOT_OVERFLOW;
1862 /* But the API says we flag all errors found */
1863 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1864 *errors |= UTF8_GOT_SUPER;
1867 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1869 *errors |= UTF8_GOT_PERL_EXTENDED;
1872 /* Disallow if any of the three categories say to */
1873 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1874 || (flags & ( UTF8_DISALLOW_SUPER
1875 |UTF8_DISALLOW_PERL_EXTENDED)))
1880 /* Likewise, warn if any say to */
1881 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1882 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1885 /* The warnings code explicitly says it doesn't handle the
1886 * case of packWARN2 and two categories which have
1887 * parent-child relationship. Even if it works now to
1888 * raise the warning if either is enabled, it wouldn't
1889 * necessarily do so in the future. We output (only) the
1890 * most dire warning */
1891 if (! (flags & UTF8_CHECK_ONLY)) {
1892 if (msgs || ckWARN_d(WARN_UTF8)) {
1893 pack_warn = packWARN(WARN_UTF8);
1895 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1896 pack_warn = packWARN(WARN_NON_UNICODE);
1899 message = Perl_form(aTHX_ "%s: %s (overflows)",
1901 _byte_dump_string(s0, curlen, 0));
1902 this_flag_bit = UTF8_GOT_OVERFLOW;
1907 else if (possible_problems & UTF8_GOT_EMPTY) {
1908 possible_problems &= ~UTF8_GOT_EMPTY;
1909 *errors |= UTF8_GOT_EMPTY;
1911 if (! (flags & UTF8_ALLOW_EMPTY)) {
1913 /* This so-called malformation is now treated as a bug in
1914 * the caller. If you have nothing to decode, skip calling
1920 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1922 pack_warn = packWARN(WARN_UTF8);
1923 message = Perl_form(aTHX_ "%s (empty string)",
1925 this_flag_bit = UTF8_GOT_EMPTY;
1929 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1930 possible_problems &= ~UTF8_GOT_CONTINUATION;
1931 *errors |= UTF8_GOT_CONTINUATION;
1933 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1936 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1938 pack_warn = packWARN(WARN_UTF8);
1939 message = Perl_form(aTHX_
1940 "%s: %s (unexpected continuation byte 0x%02x,"
1941 " with no preceding start byte)",
1943 _byte_dump_string(s0, 1, 0), *s0);
1944 this_flag_bit = UTF8_GOT_CONTINUATION;
1948 else if (possible_problems & UTF8_GOT_SHORT) {
1949 possible_problems &= ~UTF8_GOT_SHORT;
1950 *errors |= UTF8_GOT_SHORT;
1952 if (! (flags & UTF8_ALLOW_SHORT)) {
1955 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1957 pack_warn = packWARN(WARN_UTF8);
1958 message = Perl_form(aTHX_
1959 "%s: %s (too short; %d byte%s available, need %d)",
1961 _byte_dump_string(s0, send - s0, 0),
1963 avail_len == 1 ? "" : "s",
1965 this_flag_bit = UTF8_GOT_SHORT;
1970 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1971 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1972 *errors |= UTF8_GOT_NON_CONTINUATION;
1974 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1977 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1980 /* If we don't know for sure that the input length is
1981 * valid, avoid as much as possible reading past the
1982 * end of the buffer */
1983 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1986 pack_warn = packWARN(WARN_UTF8);
1987 message = Perl_form(aTHX_ "%s",
1988 unexpected_non_continuation_text(s0,
1992 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
1996 else if (possible_problems & UTF8_GOT_SURROGATE) {
1997 possible_problems &= ~UTF8_GOT_SURROGATE;
1999 if (flags & UTF8_WARN_SURROGATE) {
2000 *errors |= UTF8_GOT_SURROGATE;
2002 if ( ! (flags & UTF8_CHECK_ONLY)
2003 && (msgs || ckWARN_d(WARN_SURROGATE)))
2005 pack_warn = packWARN(WARN_SURROGATE);
2007 /* These are the only errors that can occur with a
2008 * surrogate when the 'uv' isn't valid */
2009 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2010 message = Perl_form(aTHX_
2011 "UTF-16 surrogate (any UTF-8 sequence that"
2012 " starts with \"%s\" is for a surrogate)",
2013 _byte_dump_string(s0, curlen, 0));
2016 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2018 this_flag_bit = UTF8_GOT_SURROGATE;
2022 if (flags & UTF8_DISALLOW_SURROGATE) {
2024 *errors |= UTF8_GOT_SURROGATE;
2027 else if (possible_problems & UTF8_GOT_SUPER) {
2028 possible_problems &= ~UTF8_GOT_SUPER;
2030 if (flags & UTF8_WARN_SUPER) {
2031 *errors |= UTF8_GOT_SUPER;
2033 if ( ! (flags & UTF8_CHECK_ONLY)
2034 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2036 pack_warn = packWARN(WARN_NON_UNICODE);
2038 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2039 message = Perl_form(aTHX_
2040 "Any UTF-8 sequence that starts with"
2041 " \"%s\" is for a non-Unicode code point,"
2042 " may not be portable",
2043 _byte_dump_string(s0, curlen, 0));
2046 message = Perl_form(aTHX_ super_cp_format, uv);
2048 this_flag_bit = UTF8_GOT_SUPER;
2052 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2053 * and before possibly bailing out, so that the more dire
2054 * warning will override the regular one. */
2055 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2056 if ( ! (flags & UTF8_CHECK_ONLY)
2057 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2058 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2060 pack_warn = packWARN(WARN_NON_UNICODE);
2062 /* If it is an overlong that evaluates to a code point
2063 * that doesn't have to use the Perl extended UTF-8, it
2064 * still used it, and so we output a message that
2065 * doesn't refer to the code point. The same is true
2066 * if there was a SHORT malformation where the code
2067 * point is not valid. In that case, 'uv' will have
2068 * been set to the REPLACEMENT CHAR, and the message
2069 * below without the code point in it will be selected
2071 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2072 message = Perl_form(aTHX_
2073 perl_extended_cp_format, uv);
2076 message = Perl_form(aTHX_
2077 "Any UTF-8 sequence that starts with"
2078 " \"%s\" is a Perl extension, and"
2079 " so is not portable",
2080 _byte_dump_string(s0, curlen, 0));
2082 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2085 if (flags & ( UTF8_WARN_PERL_EXTENDED
2086 |UTF8_DISALLOW_PERL_EXTENDED))
2088 *errors |= UTF8_GOT_PERL_EXTENDED;
2090 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2096 if (flags & UTF8_DISALLOW_SUPER) {
2097 *errors |= UTF8_GOT_SUPER;
2101 else if (possible_problems & UTF8_GOT_NONCHAR) {
2102 possible_problems &= ~UTF8_GOT_NONCHAR;
2104 if (flags & UTF8_WARN_NONCHAR) {
2105 *errors |= UTF8_GOT_NONCHAR;
2107 if ( ! (flags & UTF8_CHECK_ONLY)
2108 && (msgs || ckWARN_d(WARN_NONCHAR)))
2110 /* The code above should have guaranteed that we don't
2111 * get here with errors other than overlong */
2112 assert (! (orig_problems
2113 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2115 pack_warn = packWARN(WARN_NONCHAR);
2116 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2117 this_flag_bit = UTF8_GOT_NONCHAR;
2121 if (flags & UTF8_DISALLOW_NONCHAR) {
2123 *errors |= UTF8_GOT_NONCHAR;
2126 else if (possible_problems & UTF8_GOT_LONG) {
2127 possible_problems &= ~UTF8_GOT_LONG;
2128 *errors |= UTF8_GOT_LONG;
2130 if (flags & UTF8_ALLOW_LONG) {
2132 /* We don't allow the actual overlong value, unless the
2133 * special extra bit is also set */
2134 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2135 & ~UTF8_ALLOW_LONG)))
2137 uv = UNICODE_REPLACEMENT;
2144 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2146 pack_warn = packWARN(WARN_UTF8);
2148 /* These error types cause 'uv' to be something that
2149 * isn't what was intended, so can't use it in the
2150 * message. The other error types either can't
2151 * generate an overlong, or else the 'uv' is valid */
2153 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2155 message = Perl_form(aTHX_
2156 "%s: %s (any UTF-8 sequence that starts"
2157 " with \"%s\" is overlong which can and"
2158 " should be represented with a"
2159 " different, shorter sequence)",
2161 _byte_dump_string(s0, send - s0, 0),
2162 _byte_dump_string(s0, curlen, 0));
2165 U8 tmpbuf[UTF8_MAXBYTES+1];
2166 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2168 /* Don't use U+ for non-Unicode code points, which
2169 * includes those in the Latin1 range */
2170 const char * preface = ( uv > PERL_UNICODE_MAX
2177 message = Perl_form(aTHX_
2178 "%s: %s (overlong; instead use %s to represent"
2181 _byte_dump_string(s0, send - s0, 0),
2182 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2184 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2185 small code points */
2188 this_flag_bit = UTF8_GOT_LONG;
2191 } /* End of looking through the possible flags */
2193 /* Display the message (if any) for the problem being handled in
2194 * this iteration of the loop */
2197 assert(this_flag_bit);
2199 if (*msgs == NULL) {
2203 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2208 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2211 Perl_warner(aTHX_ pack_warn, "%s", message);
2213 } /* End of 'while (possible_problems)' */
2215 /* Since there was a possible problem, the returned length may need to
2216 * be changed from the one stored at the beginning of this function.
2217 * Instead of trying to figure out if that's needed, just do it. */
2223 if (flags & UTF8_CHECK_ONLY && retlen) {
2224 *retlen = ((STRLEN) -1);
2230 return UNI_TO_NATIVE(uv);
2234 =for apidoc utf8_to_uvchr_buf
2236 Returns the native code point of the first character in the string C<s> which
2237 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2238 C<*retlen> will be set to the length, in bytes, of that character.
2240 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2241 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2242 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2243 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2244 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2245 the next possible position in C<s> that could begin a non-malformed character.
2246 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2251 Also implemented as a macro in utf8.h
2257 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2259 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2263 return utf8n_to_uvchr(s, send - s, retlen,
2264 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
2267 /* This is marked as deprecated
2269 =for apidoc utf8_to_uvuni_buf
2271 Only in very rare circumstances should code need to be dealing in Unicode
2272 (as opposed to native) code points. In those few cases, use
2273 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2274 are not absolutely sure this is one of those cases, then assume it isn't and
2275 use plain C<utf8_to_uvchr_buf> instead.
2277 Returns the Unicode (not-native) code point of the first character in the
2279 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2280 C<retlen> will be set to the length, in bytes, of that character.
2282 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2283 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2284 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2285 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2286 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2287 next possible position in C<s> that could begin a non-malformed character.
2288 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
2294 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2296 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2300 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2304 =for apidoc utf8_length
2306 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2307 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2308 same place, it returns 0 with no warning raised.
2310 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2311 and returns the number of valid characters.
2317 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2321 PERL_ARGS_ASSERT_UTF8_LENGTH;
2323 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2324 * the bitops (especially ~) can create illegal UTF-8.
2325 * In other words: in Perl UTF-8 is not just for Unicode. */
2328 goto warn_and_return;
2338 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2339 "%s in %s", unees, OP_DESC(PL_op));
2341 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2348 =for apidoc bytes_cmp_utf8
2350 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2351 sequence of characters (stored as UTF-8)
2352 in C<u>, C<ulen>. Returns 0 if they are
2353 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2354 if the first string is greater than the second string.
2356 -1 or +1 is returned if the shorter string was identical to the start of the
2357 longer string. -2 or +2 is returned if
2358 there was a difference between characters
2365 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2367 const U8 *const bend = b + blen;
2368 const U8 *const uend = u + ulen;
2370 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2372 while (b < bend && u < uend) {
2374 if (!UTF8_IS_INVARIANT(c)) {
2375 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2378 if (UTF8_IS_CONTINUATION(c1)) {
2379 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2381 /* diag_listed_as: Malformed UTF-8 character%s */
2382 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2384 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2385 PL_op ? " in " : "",
2386 PL_op ? OP_DESC(PL_op) : "");
2391 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2392 "%s in %s", unees, OP_DESC(PL_op));
2394 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2395 return -2; /* Really want to return undef :-) */
2402 return *b < c ? -2 : +2;
2407 if (b == bend && u == uend)
2410 return b < bend ? +1 : -1;
2414 =for apidoc utf8_to_bytes
2416 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2417 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2418 updates C<*lenp> to contain the new length.
2419 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2421 Upon successful return, the number of variants in the string can be computed by
2422 having saved the value of C<*lenp> before the call, and subtracting the
2423 after-call value of C<*lenp> from it.
2425 If you need a copy of the string, see L</bytes_from_utf8>.
2431 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2435 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2436 PERL_UNUSED_CONTEXT;
2438 /* This is a no-op if no variants at all in the input */
2439 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2444 U8 * const save = s;
2445 U8 * const send = s + *lenp;
2448 /* Nothing before the first variant needs to be changed, so start the real
2452 if (! UTF8_IS_INVARIANT(*s)) {
2453 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2454 *lenp = ((STRLEN) -1);
2462 /* Is downgradable, so do it */
2463 d = s = first_variant;
2466 if (! UVCHR_IS_INVARIANT(c)) {
2467 /* Then it is two-byte encoded */
2468 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2481 =for apidoc bytes_from_utf8
2483 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2484 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2485 actually encoded in UTF-8.
2487 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2490 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2491 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2492 C<*lenp> are unchanged, and the return value is the original C<s>.
2494 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2495 newly created string containing a downgraded copy of C<s>, and whose length is
2496 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2497 caller is responsible for arranging for the memory used by this string to get
2500 Upon successful return, the number of variants in the string can be computed by
2501 having saved the value of C<*lenp> before the call, and subtracting the
2502 after-call value of C<*lenp> from it.
2506 There is a macro that avoids this function call, but this is retained for
2507 anyone who calls it with the Perl_ prefix */
2510 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2512 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2513 PERL_UNUSED_CONTEXT;
2515 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2519 No = here because currently externally undocumented
2520 for apidoc bytes_from_utf8_loc
2522 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2523 to store the location of the first character in C<"s"> that cannot be
2524 converted to non-UTF8.
2526 If that parameter is C<NULL>, this function behaves identically to
2529 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2530 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2532 Otherwise, the function returns a newly created C<NUL>-terminated string
2533 containing the non-UTF8 equivalent of the convertible first portion of
2534 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2535 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2536 and C<*first_non_downgradable> is set to C<NULL>.
2538 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2539 first character in the original string that wasn't converted. C<*is_utf8p> is
2540 unchanged. Note that the new string may have length 0.
2542 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2543 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2544 converts as many characters in it as possible stopping at the first one it
2545 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2546 set to point to that. The function returns the portion that could be converted
2547 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2548 not including the terminating C<NUL>. If the very first character in the
2549 original could not be converted, C<*lenp> will be 0, and the new string will
2550 contain just a single C<NUL>. If the entire input string was converted,
2551 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2553 Upon successful return, the number of variants in the converted portion of the
2554 string can be computed by having saved the value of C<*lenp> before the call,
2555 and subtracting the after-call value of C<*lenp> from it.
2563 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2566 const U8 *original = s;
2567 U8 *converted_start;
2568 const U8 *send = s + *lenp;
2570 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2573 if (first_unconverted) {
2574 *first_unconverted = NULL;
2577 return (U8 *) original;
2580 Newx(d, (*lenp) + 1, U8);
2582 converted_start = d;
2585 if (! UTF8_IS_INVARIANT(c)) {
2587 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2588 * have to stop now */
2589 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2590 if (first_unconverted) {
2591 *first_unconverted = s - 1;
2592 goto finish_and_return;
2595 Safefree(converted_start);
2596 return (U8 *) original;
2600 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2606 /* Here, converted the whole of the input */
2608 if (first_unconverted) {
2609 *first_unconverted = NULL;
2614 *lenp = d - converted_start;
2616 /* Trim unused space */
2617 Renew(converted_start, *lenp + 1, U8);
2619 return converted_start;
2623 =for apidoc bytes_to_utf8
2625 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2627 Returns a pointer to the newly-created string, and sets C<*lenp> to
2628 reflect the new length in bytes. The caller is responsible for arranging for
2629 the memory used by this string to get freed.
2631 Upon successful return, the number of variants in the string can be computed by
2632 having saved the value of C<*lenp> before the call, and subtracting it from the
2633 after-call value of C<*lenp>.
2635 A C<NUL> character will be written after the end of the string.
2637 If you want to convert to UTF-8 from encodings other than
2638 the native (Latin1 or EBCDIC),
2639 see L</sv_recode_to_utf8>().
2645 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2647 const U8 * const send = s + (*lenp);
2651 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2652 PERL_UNUSED_CONTEXT;
2654 Newx(d, (*lenp) * 2 + 1, U8);
2658 append_utf8_from_native_byte(*s, &d);
2665 /* Trim unused space */
2666 Renew(dst, *lenp + 1, U8);
2672 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2673 * use utf16_to_utf8_reversed().
2675 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2676 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2677 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2679 * These functions don't check for overflow. The worst case is every code
2680 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2681 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2682 * destination must be pre-extended to 2 times the source length.
2684 * Do not use in-place. We optimize for native, for obvious reasons. */
2687 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2692 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2695 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2701 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2703 if (OFFUNI_IS_INVARIANT(uv)) {
2704 *d++ = LATIN1_TO_NATIVE((U8) uv);
2707 if (uv <= MAX_UTF8_TWO_BYTE) {
2708 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2709 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2713 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2714 #define LAST_HIGH_SURROGATE 0xDBFF
2715 #define FIRST_LOW_SURROGATE 0xDC00
2716 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2717 #define FIRST_IN_PLANE1 0x10000
2719 /* This assumes that most uses will be in the first Unicode plane, not
2720 * needing surrogates */
2721 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2722 && uv <= UNICODE_SURROGATE_LAST))
2724 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2725 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2728 UV low = (p[0] << 8) + p[1];
2729 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2730 || UNLIKELY(low > LAST_LOW_SURROGATE))
2732 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2735 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2736 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2740 d = uvoffuni_to_utf8_flags(d, uv, 0);
2742 if (uv < FIRST_IN_PLANE1) {
2743 *d++ = (U8)(( uv >> 12) | 0xe0);
2744 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2745 *d++ = (U8)(( uv & 0x3f) | 0x80);
2749 *d++ = (U8)(( uv >> 18) | 0xf0);
2750 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2751 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2752 *d++ = (U8)(( uv & 0x3f) | 0x80);
2757 *newlen = d - dstart;
2761 /* Note: this one is slightly destructive of the source. */
2764 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2767 U8* const send = s + bytelen;
2769 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2772 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2776 const U8 tmp = s[0];
2781 return utf16_to_utf8(p, d, bytelen, newlen);
2785 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2787 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2790 /* Internal function so we can deprecate the external one, and call
2791 this one from other deprecated functions in this file */
2794 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2796 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2800 return is_utf8_common(p, NULL,
2801 "This is buggy if this gets used",
2806 Perl__is_uni_perl_idcont(pTHX_ UV c)
2808 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2812 Perl__is_uni_perl_idstart(pTHX_ UV c)
2814 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2818 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2821 /* We have the latin1-range values compiled into the core, so just use
2822 * those, converting the result to UTF-8. The only difference between upper
2823 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2824 * either "SS" or "Ss". Which one to use is passed into the routine in
2825 * 'S_or_s' to avoid a test */
2827 UV converted = toUPPER_LATIN1_MOD(c);
2829 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2831 assert(S_or_s == 'S' || S_or_s == 's');
2833 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2834 characters in this range */
2835 *p = (U8) converted;
2840 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2841 * which it maps to one of them, so as to only have to have one check for
2842 * it in the main case */
2843 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2845 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2846 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2849 converted = GREEK_CAPITAL_LETTER_MU;
2851 #if UNICODE_MAJOR_VERSION > 2 \
2852 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2853 && UNICODE_DOT_DOT_VERSION >= 8)
2854 case LATIN_SMALL_LETTER_SHARP_S:
2861 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2862 " '%c' to map to '%c'",
2863 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2864 NOT_REACHED; /* NOTREACHED */
2868 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2869 *p = UTF8_TWO_BYTE_LO(converted);
2875 /* If compiled on an early Unicode version, there may not be auxiliary tables
2877 #ifndef HAS_UC_AUX_TABLES
2878 # define UC_AUX_TABLE_ptrs NULL
2879 # define UC_AUX_TABLE_lengths NULL
2881 #ifndef HAS_TC_AUX_TABLES
2882 # define TC_AUX_TABLE_ptrs NULL
2883 # define TC_AUX_TABLE_lengths NULL
2885 #ifndef HAS_LC_AUX_TABLES
2886 # define LC_AUX_TABLE_ptrs NULL
2887 # define LC_AUX_TABLE_lengths NULL
2889 #ifndef HAS_CF_AUX_TABLES
2890 # define CF_AUX_TABLE_ptrs NULL
2891 # define CF_AUX_TABLE_lengths NULL
2893 #ifndef HAS_UC_AUX_TABLES
2894 # define UC_AUX_TABLE_ptrs NULL
2895 # define UC_AUX_TABLE_lengths NULL
2898 /* Call the function to convert a UTF-8 encoded character to the specified case.
2899 * Note that there may be more than one character in the result.
2900 * 's' is a pointer to the first byte of the input character
2901 * 'd' will be set to the first byte of the string of changed characters. It
2902 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2903 * 'lenp' will be set to the length in bytes of the string of changed characters
2905 * The functions return the ordinal of the first character in the string of
2907 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2908 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2909 Uppercase_Mapping_invmap, \
2910 UC_AUX_TABLE_ptrs, \
2911 UC_AUX_TABLE_lengths, \
2913 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2914 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2915 Titlecase_Mapping_invmap, \
2916 TC_AUX_TABLE_ptrs, \
2917 TC_AUX_TABLE_lengths, \
2919 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2920 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2921 Lowercase_Mapping_invmap, \
2922 LC_AUX_TABLE_ptrs, \
2923 LC_AUX_TABLE_lengths, \
2927 /* This additionally has the input parameter 'specials', which if non-zero will
2928 * cause this to use the specials hash for folding (meaning get full case
2929 * folding); otherwise, when zero, this implies a simple case fold */
2930 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2932 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2933 Case_Folding_invmap, \
2934 CF_AUX_TABLE_ptrs, \
2935 CF_AUX_TABLE_lengths, \
2937 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2938 Simple_Case_Folding_invmap, \
2943 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2945 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2946 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2947 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2948 * the changed version may be longer than the original character.
2950 * The ordinal of the first character of the changed version is returned
2951 * (but note, as explained above, that there may be more.) */
2953 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2956 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2959 uvchr_to_utf8(p, c);
2960 return CALL_UPPER_CASE(c, p, p, lenp);
2964 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2966 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2969 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2972 uvchr_to_utf8(p, c);
2973 return CALL_TITLE_CASE(c, p, p, lenp);
2977 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2979 /* We have the latin1-range values compiled into the core, so just use
2980 * those, converting the result to UTF-8. Since the result is always just
2981 * one character, we allow <p> to be NULL */
2983 U8 converted = toLOWER_LATIN1(c);
2985 PERL_UNUSED_ARG(dummy);
2988 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2993 /* Result is known to always be < 256, so can use the EIGHT_BIT
2995 *p = UTF8_EIGHT_BIT_HI(converted);
2996 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3004 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3006 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3009 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3012 uvchr_to_utf8(p, c);
3013 return CALL_LOWER_CASE(c, p, p, lenp);
3017 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3019 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3020 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3021 * FOLD_FLAGS_FULL iff full folding is to be used;
3023 * Not to be used for locale folds
3028 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3030 assert (! (flags & FOLD_FLAGS_LOCALE));
3032 if (UNLIKELY(c == MICRO_SIGN)) {
3033 converted = GREEK_SMALL_LETTER_MU;
3035 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3036 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3037 || UNICODE_DOT_DOT_VERSION > 0)
3038 else if ( (flags & FOLD_FLAGS_FULL)
3039 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3041 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3042 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3043 * under those circumstances. */
3044 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3045 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3046 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3048 return LATIN_SMALL_LETTER_LONG_S;
3058 else { /* In this range the fold of all other characters is their lower
3060 converted = toLOWER_LATIN1(c);
3063 if (UVCHR_IS_INVARIANT(converted)) {
3064 *p = (U8) converted;
3068 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3069 *p = UTF8_TWO_BYTE_LO(converted);
3077 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3080 /* Not currently externally documented, and subject to change
3081 * <flags> bits meanings:
3082 * FOLD_FLAGS_FULL iff full folding is to be used;
3083 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3084 * locale are to be used.
3085 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3088 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3090 if (flags & FOLD_FLAGS_LOCALE) {
3091 /* Treat a UTF-8 locale as not being in locale at all, except for
3092 * potentially warning */
3093 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3094 if (IN_UTF8_CTYPE_LOCALE) {
3095 flags &= ~FOLD_FLAGS_LOCALE;
3098 goto needs_full_generality;
3103 return _to_fold_latin1((U8) c, p, lenp,
3104 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3107 /* Here, above 255. If no special needs, just use the macro */
3108 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3109 uvchr_to_utf8(p, c);
3110 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
3112 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3113 the special flags. */
3114 U8 utf8_c[UTF8_MAXBYTES + 1];
3116 needs_full_generality:
3117 uvchr_to_utf8(utf8_c, c);
3118 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3123 PERL_STATIC_INLINE bool
3124 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
3125 const char *const swashname, SV* const invlist)
3127 /* returns a boolean giving whether or not the UTF8-encoded character that
3128 * starts at <p> is in the swash indicated by <swashname>. <swash>
3129 * contains a pointer to where the swash indicated by <swashname>
3130 * is to be stored; which this routine will do, so that future calls will
3131 * look at <*swash> and only generate a swash if it is not null. <invlist>
3132 * is NULL or an inversion list that defines the swash. If not null, it
3133 * saves time during initialization of the swash.
3135 * Note that it is assumed that the buffer length of <p> is enough to
3136 * contain all the bytes that comprise the character. Thus, <*p> should
3137 * have been checked before this call for mal-formedness enough to assure
3140 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3142 /* The API should have included a length for the UTF-8 character in <p>,
3143 * but it doesn't. We therefore assume that p has been validated at least
3144 * as far as there being enough bytes available in it to accommodate the
3145 * character without reading beyond the end, and pass that number on to the
3146 * validating routine */
3147 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
3148 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
3149 _UTF8_NO_CONFIDENCE_IN_CURLEN,
3151 NOT_REACHED; /* NOTREACHED */
3155 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3161 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3162 *swash = _core_swash_init("utf8",
3164 /* Only use the name if there is no inversion
3165 * list; otherwise will go out to disk */
3166 (invlist) ? "" : swashname,
3168 &PL_sv_undef, 1, 0, invlist, &flags);
3171 return swash_fetch(*swash, p, TRUE) != 0;
3174 PERL_STATIC_INLINE bool
3175 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3176 SV **swash, const char *const swashname,
3179 /* returns a boolean giving whether or not the UTF8-encoded character that
3180 * starts at <p>, and extending no further than <e - 1> is in the swash
3181 * indicated by <swashname>. <swash> contains a pointer to where the swash
3182 * indicated by <swashname> is to be stored; which this routine will do, so
3183 * that future calls will look at <*swash> and only generate a swash if it
3184 * is not null. <invlist> is NULL or an inversion list that defines the
3185 * swash. If not null, it saves time during initialization of the swash.
3188 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3190 if (! isUTF8_CHAR(p, e)) {
3191 _force_out_malformed_utf8_message(p, e, 0, 1);
3192 NOT_REACHED; /* NOTREACHED */
3196 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3202 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3203 *swash = _core_swash_init("utf8",
3205 /* Only use the name if there is no inversion
3206 * list; otherwise will go out to disk */
3207 (invlist) ? "" : swashname,
3209 &PL_sv_undef, 1, 0, invlist, &flags);
3212 return swash_fetch(*swash, p, TRUE) != 0;
3216 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3217 const char * const alternative,
3218 const bool use_locale,
3219 const char * const file,
3220 const unsigned line)
3224 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3226 if (ckWARN_d(WARN_DEPRECATED)) {
3228 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3229 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3230 if (! PL_seen_deprecated_macro) {
3231 PL_seen_deprecated_macro = newHV();
3233 if (! hv_store(PL_seen_deprecated_macro, key,
3234 strlen(key), &PL_sv_undef, 0))
3236 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3239 if (instr(file, "mathoms.c")) {
3240 Perl_warner(aTHX_ WARN_DEPRECATED,
3241 "In %s, line %d, starting in Perl v5.30, %s()"
3242 " will be removed. Avoid this message by"
3243 " converting to use %s().\n",
3244 file, line, name, alternative);
3247 Perl_warner(aTHX_ WARN_DEPRECATED,
3248 "In %s, line %d, starting in Perl v5.30, %s() will"
3249 " require an additional parameter. Avoid this"
3250 " message by converting to use %s().\n",
3251 file, line, name, alternative);
3258 Perl__is_utf8_FOO(pTHX_ U8 classnum,
3260 const char * const name,
3261 const char * const alternative,
3262 const bool use_utf8,
3263 const bool use_locale,
3264 const char * const file,
3265 const unsigned line)
3267 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3269 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3271 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
3281 case _CC_ALPHANUMERIC:
3285 return is_utf8_common(p,
3287 "This is buggy if this gets used",
3288 PL_XPosix_ptrs[classnum]);
3291 return is_XPERLSPACE_high(p);
3293 return is_HORIZWS_high(p);
3295 return is_XDIGIT_high(p);
3301 return is_VERTWS_high(p);
3303 return is_utf8_common(p, NULL,
3304 "This is buggy if this gets used",
3305 PL_utf8_perl_idstart);
3307 return is_utf8_common(p, NULL,
3308 "This is buggy if this gets used",
3309 PL_utf8_perl_idcont);
3313 /* idcont is the same as wordchar below 256 */
3314 if (classnum == _CC_IDCONT) {
3315 classnum = _CC_WORDCHAR;
3317 else if (classnum == _CC_IDFIRST) {
3321 classnum = _CC_ALPHA;
3325 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3326 return _generic_isCC(*p, classnum);
3329 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3332 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3333 return isFOO_lc(classnum, *p);
3336 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3339 NOT_REACHED; /* NOTREACHED */
3343 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3346 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3348 return is_utf8_common_with_len(p, e, NULL,
3349 "This is buggy if this gets used",
3350 PL_XPosix_ptrs[classnum]);
3354 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3356 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3358 return is_utf8_common_with_len(p, e, NULL,
3359 "This is buggy if this gets used",
3360 PL_utf8_perl_idstart);
3364 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3366 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3370 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
3374 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3376 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3378 return is_utf8_common_with_len(p, e, NULL,
3379 "This is buggy if this gets used",
3380 PL_utf8_perl_idcont);
3384 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3386 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3388 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
3392 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3394 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3396 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
3400 Perl__is_utf8_mark(pTHX_ const U8 *p)
3402 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3404 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
3408 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3409 U8* ustrp, STRLEN *lenp,
3410 SV *invlist, const int * const invmap,
3411 const unsigned int * const * const aux_tables,
3412 const U8 * const aux_table_lengths,
3413 const char * const normal)
3417 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3418 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3419 * to name the new case in any generated messages, as a fallback if the
3420 * operation being used is not available. The new case is given by the
3421 * data structures in the remaining arguments.
3423 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3424 * entire changed case string, and the return value is the first code point
3427 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3429 /* For code points that don't change case, we already know that the output
3430 * of this function is the unchanged input, so we can skip doing look-ups
3431 * for them. Unfortunately the case-changing code points are scattered
3432 * around. But there are some long consecutive ranges where there are no
3433 * case changing code points. By adding tests, we can eliminate the lookup
3434 * for all the ones in such ranges. This is currently done here only for
3435 * just a few cases where the scripts are in common use in modern commerce
3436 * (and scripts adjacent to those which can be included without additional
3439 if (uv1 >= 0x0590) {
3440 /* This keeps from needing further processing the code points most
3441 * likely to be used in the following non-cased scripts: Hebrew,
3442 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3443 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3444 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3449 /* The following largish code point ranges also don't have case
3450 * changes, but khw didn't think they warranted extra tests to speed
3451 * them up (which would slightly slow down everything else above them):
3452 * 1100..139F Hangul Jamo, Ethiopic
3453 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3454 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3455 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3456 * Combining Diacritical Marks Extended, Balinese,
3457 * Sundanese, Batak, Lepcha, Ol Chiki
3458 * 2000..206F General Punctuation
3461 if (uv1 >= 0x2D30) {
3463 /* This keeps the from needing further processing the code points
3464 * most likely to be used in the following non-cased major scripts:
3465 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3467 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3468 * event that Unicode eventually allocates the unused block as of
3469 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3470 * that the test suite will start having failures to alert you
3471 * should that happen) */
3476 if (uv1 >= 0xAC00) {
3477 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3478 if (ckWARN_d(WARN_SURROGATE)) {
3479 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3480 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3481 "Operation \"%s\" returns its argument for"
3482 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3487 /* AC00..FAFF Catches Hangul syllables and private use, plus
3493 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3494 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3495 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3496 MAX_EXTERNALLY_LEGAL_CP);
3498 if (ckWARN_d(WARN_NON_UNICODE)) {
3499 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3500 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3501 "Operation \"%s\" returns its argument for"
3502 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3506 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3508 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3511 /* As of Unicode 10.0, this means we avoid swash creation
3512 * for anything beyond high Plane 1 (below emojis) */
3519 /* Note that non-characters are perfectly legal, so no warning should
3525 const unsigned int * cp_list;
3527 SSize_t index = _invlist_search(invlist, uv1);
3528 IV base = invmap[index];
3530 /* The data structures are set up so that if 'base' is non-negative,
3531 * the case change is 1-to-1; and if 0, the change is to itself */
3539 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3540 lc = base + uv1 - invlist_array(invlist)[index];
3541 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3545 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3546 * requires an auxiliary table look up. abs(base) gives the index into
3547 * a list of such tables which points to the proper aux table. And a
3548 * parallel list gives the length of each corresponding aux table. */
3549 cp_list = aux_tables[-base];
3551 /* Create the string of UTF-8 from the mapped-to code points */
3553 for (i = 0; i < aux_table_lengths[-base]; i++) {
3554 d = uvchr_to_utf8(d, cp_list[i]);
3562 /* Here, there was no mapping defined, which means that the code point maps
3563 * to itself. Return the inputs */
3566 if (p != ustrp) { /* Don't copy onto itself */
3567 Copy(p, ustrp, len, U8);
3578 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3579 const unsigned int ** remaining_folds_to)
3581 /* Returns the count of the number of code points that fold to the input
3582 * 'cp' (besides itself).
3584 * If the return is 0, there is nothing else that folds to it, and
3585 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3587 * If the return is 1, '*first_folds_to' is set to the single code point,
3588 * and '*remaining_folds_to' is set to NULL.
3590 * Otherwise, '*first_folds_to' is set to a code point, and
3591 * '*remaining_fold_to' is set to an array that contains the others. The
3592 * length of this array is the returned count minus 1.
3594 * The reason for this convolution is to avoid having to deal with
3595 * allocating and freeing memory. The lists are already constructed, so
3596 * the return can point to them, but single code points aren't, so would
3597 * need to be constructed if we didn't employ something like this API */
3599 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3600 int base = _Perl_IVCF_invmap[index];
3602 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3604 if (base == 0) { /* No fold */
3605 *first_folds_to = 0;
3606 *remaining_folds_to = NULL;
3610 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3616 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3618 /* The data structure is set up so that the absolute value of 'base' is
3619 * an index into a table of pointers to arrays, with the array
3620 * corresponding to the index being the list of code points that fold
3621 * to 'cp', and the parallel array containing the length of the list
3623 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3624 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3627 return IVCF_AUX_TABLE_lengths[-base];
3632 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3633 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3634 *remaining_folds_to = NULL;
3639 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3640 U8* const ustrp, STRLEN *lenp)
3642 /* This is called when changing the case of a UTF-8-encoded character above
3643 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3644 * result contains a character that crosses the 255/256 boundary, disallow
3645 * the change, and return the original code point. See L<perlfunc/lc> for
3648 * p points to the original string whose case was changed; assumed
3649 * by this routine to be well-formed
3650 * result the code point of the first character in the changed-case string
3651 * ustrp points to the changed-case string (<result> represents its
3653 * lenp points to the length of <ustrp> */
3655 UV original; /* To store the first code point of <p> */
3657 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3659 assert(UTF8_IS_ABOVE_LATIN1(*p));
3661 /* We know immediately if the first character in the string crosses the
3662 * boundary, so can skip testing */
3665 /* Look at every character in the result; if any cross the
3666 * boundary, the whole thing is disallowed */
3667 U8* s = ustrp + UTF8SKIP(ustrp);
3668 U8* e = ustrp + *lenp;
3670 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3676 /* Here, no characters crossed, result is ok as-is, but we warn. */
3677 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3683 /* Failed, have to return the original */
3684 original = valid_utf8_to_uvchr(p, lenp);
3686 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3687 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3688 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3689 " locale; resolved to \"\\x{%" UVXf "}\".",
3693 Copy(p, ustrp, *lenp, char);
3698 S_check_and_deprecate(pTHX_ const U8 *p,
3700 const unsigned int type, /* See below */
3701 const bool use_locale, /* Is this a 'LC_'
3703 const char * const file,
3704 const unsigned line)
3706 /* This is a temporary function to deprecate the unsafe calls to the case
3707 * changing macros and functions. It keeps all the special stuff in just
3710 * It updates *e with the pointer to the end of the input string. If using
3711 * the old-style macros, *e is NULL on input, and so this function assumes
3712 * the input string is long enough to hold the entire UTF-8 sequence, and
3713 * sets *e accordingly, but it then returns a flag to pass the
3714 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3715 * using the full length if possible.
3717 * It also does the assert that *e > p when *e is not NULL. This should be
3718 * migrated to the callers when this function gets deleted.
3720 * The 'type' parameter is used for the caller to specify which case
3721 * changing function this is called from: */
3723 # define DEPRECATE_TO_UPPER 0
3724 # define DEPRECATE_TO_TITLE 1
3725 # define DEPRECATE_TO_LOWER 2
3726 # define DEPRECATE_TO_FOLD 3
3728 U32 utf8n_flags = 0;
3730 const char * alternative;
3732 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3735 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3736 *e = p + UTF8SKIP(p);
3738 /* For mathoms.c calls, we use the function name we know is stored
3739 * there. It could be part of a larger path */
3740 if (type == DEPRECATE_TO_UPPER) {
3741 name = instr(file, "mathoms.c")
3744 alternative = "toUPPER_utf8_safe";
3746 else if (type == DEPRECATE_TO_TITLE) {
3747 name = instr(file, "mathoms.c")
3750 alternative = "toTITLE_utf8_safe";
3752 else if (type == DEPRECATE_TO_LOWER) {
3753 name = instr(file, "mathoms.c")
3756 alternative = "toLOWER_utf8_safe";
3758 else if (type == DEPRECATE_TO_FOLD) {
3759 name = instr(file, "mathoms.c")
3762 alternative = "toFOLD_utf8_safe";
3764 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3766 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3775 /* The process for changing the case is essentially the same for the four case
3776 * change types, except there are complications for folding. Otherwise the
3777 * difference is only which case to change to. To make sure that they all do
3778 * the same thing, the bodies of the functions are extracted out into the
3779 * following two macros. The functions are written with the same variable
3780 * names, and these are known and used inside these macros. It would be
3781 * better, of course, to have inline functions to do it, but since different
3782 * macros are called, depending on which case is being changed to, this is not
3783 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3784 * function can start with the common start macro, then finish with its special
3785 * handling; while the other three cases can just use the common end macro.
3787 * The algorithm is to use the proper (passed in) macro or function to change
3788 * the case for code points that are below 256. The macro is used if using
3789 * locale rules for the case change; the function if not. If the code point is
3790 * above 255, it is computed from the input UTF-8, and another macro is called
3791 * to do the conversion. If necessary, the output is converted to UTF-8. If
3792 * using a locale, we have to check that the change did not cross the 255/256
3793 * boundary, see check_locale_boundary_crossing() for further details.
3795 * The macros are split with the correct case change for the below-256 case
3796 * stored into 'result', and in the middle of an else clause for the above-255
3797 * case. At that point in the 'else', 'result' is not the final result, but is
3798 * the input code point calculated from the UTF-8. The fold code needs to
3799 * realize all this and take it from there.
3801 * If you read the two macros as sequential, it's easier to understand what's
3803 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3804 L1_func_extra_param) \
3806 if (flags & (locale_flags)) { \
3807 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3808 /* Treat a UTF-8 locale as not being in locale at all */ \
3809 if (IN_UTF8_CTYPE_LOCALE) { \
3810 flags &= ~(locale_flags); \
3814 if (UTF8_IS_INVARIANT(*p)) { \
3815 if (flags & (locale_flags)) { \
3816 result = LC_L1_change_macro(*p); \
3819 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3822 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3823 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3824 if (flags & (locale_flags)) { \
3825 result = LC_L1_change_macro(c); \
3828 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3831 else { /* malformed UTF-8 or ord above 255 */ \
3832 STRLEN len_result; \
3833 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3834 if (len_result == (STRLEN) -1) { \
3835 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3839 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3840 result = change_macro(result, p, ustrp, lenp); \
3842 if (flags & (locale_flags)) { \
3843 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3848 /* Here, used locale rules. Convert back to UTF-8 */ \
3849 if (UTF8_IS_INVARIANT(result)) { \
3850 *ustrp = (U8) result; \
3854 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3855 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3862 =for apidoc to_utf8_upper
3864 Instead use L</toUPPER_utf8_safe>.
3868 /* Not currently externally documented, and subject to change:
3869 * <flags> is set iff iff the rules from the current underlying locale are to
3873 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3878 const char * const file,
3882 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3883 cBOOL(flags), file, line);
3885 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3887 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3888 /* 2nd char of uc(U+DF) is 'S' */
3889 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3890 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3894 =for apidoc to_utf8_title
3896 Instead use L</toTITLE_utf8_safe>.
3900 /* Not currently externally documented, and subject to change:
3901 * <flags> is set iff the rules from the current underlying locale are to be
3902 * used. Since titlecase is not defined in POSIX, for other than a
3903 * UTF-8 locale, uppercase is used instead for code points < 256.
3907 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3912 const char * const file,
3916 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3917 cBOOL(flags), file, line);
3919 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3921 /* 2nd char of ucfirst(U+DF) is 's' */
3922 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3923 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3927 =for apidoc to_utf8_lower
3929 Instead use L</toLOWER_utf8_safe>.
3933 /* Not currently externally documented, and subject to change:
3934 * <flags> is set iff iff the rules from the current underlying locale are to
3939 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3944 const char * const file,
3948 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3949 cBOOL(flags), file, line);
3951 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3953 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3954 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3958 =for apidoc to_utf8_fold
3960 Instead use L</toFOLD_utf8_safe>.
3964 /* Not currently externally documented, and subject to change,
3966 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3967 * locale are to be used.
3968 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3969 * otherwise simple folds
3970 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3975 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3980 const char * const file,
3984 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3985 cBOOL(flags), file, line);
3987 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3989 /* These are mutually exclusive */
3990 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3992 assert(p != ustrp); /* Otherwise overwrites */
3994 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3995 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3997 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3999 if (flags & FOLD_FLAGS_LOCALE) {
4001 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
4002 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4003 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
4005 /* Special case these two characters, as what normally gets
4006 * returned under locale doesn't work */
4007 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S))
4009 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4010 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4011 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
4012 "resolved to \"\\x{17F}\\x{17F}\".");
4017 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T))
4019 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4020 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4021 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
4022 "resolved to \"\\x{FB06}\".");
4023 goto return_ligature_st;
4026 #if UNICODE_MAJOR_VERSION == 3 \
4027 && UNICODE_DOT_VERSION == 0 \
4028 && UNICODE_DOT_DOT_VERSION == 1
4029 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
4031 /* And special case this on this Unicode version only, for the same
4032 * reaons the other two are special cased. They would cross the
4033 * 255/256 boundary which is forbidden under /l, and so the code
4034 * wouldn't catch that they are equivalent (which they are only in
4036 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) {
4037 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4038 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4039 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
4040 "resolved to \"\\x{0131}\".");
4041 goto return_dotless_i;
4045 return check_locale_boundary_crossing(p, result, ustrp, lenp);
4047 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
4051 /* This is called when changing the case of a UTF-8-encoded
4052 * character above the ASCII range, and the result should not
4053 * contain an ASCII character. */
4055 UV original; /* To store the first code point of <p> */
4057 /* Look at every character in the result; if any cross the
4058 * boundary, the whole thing is disallowed */
4060 U8* e = ustrp + *lenp;
4063 /* Crossed, have to return the original */
4064 original = valid_utf8_to_uvchr(p, lenp);
4066 /* But in these instances, there is an alternative we can
4067 * return that is valid */
4068 if (original == LATIN_SMALL_LETTER_SHARP_S
4069 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
4070 || original == LATIN_CAPITAL_LETTER_SHARP_S
4075 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
4076 goto return_ligature_st;
4078 #if UNICODE_MAJOR_VERSION == 3 \
4079 && UNICODE_DOT_VERSION == 0 \
4080 && UNICODE_DOT_DOT_VERSION == 1
4082 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4083 goto return_dotless_i;
4086 Copy(p, ustrp, *lenp, char);
4092 /* Here, no characters crossed, result is ok as-is */
4097 /* Here, used locale rules. Convert back to UTF-8 */
4098 if (UTF8_IS_INVARIANT(result)) {
4099 *ustrp = (U8) result;
4103 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
4104 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
4111 /* Certain folds to 'ss' are prohibited by the options, but they do allow
4112 * folds to a string of two of these characters. By returning this
4113 * instead, then, e.g.,
4114 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
4117 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
4118 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
4120 return LATIN_SMALL_LETTER_LONG_S;
4123 /* Two folds to 'st' are prohibited by the options; instead we pick one and
4124 * have the other one fold to it */
4126 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
4127 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
4128 return LATIN_SMALL_LIGATURE_ST;
4130 #if UNICODE_MAJOR_VERSION == 3 \
4131 && UNICODE_DOT_VERSION == 0 \
4132 && UNICODE_DOT_DOT_VERSION == 1
4135 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
4136 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
4137 return LATIN_SMALL_LETTER_DOTLESS_I;
4144 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
4145 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
4146 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
4150 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4151 I32 minbits, I32 none)
4153 PERL_ARGS_ASSERT_SWASH_INIT;
4155 /* Returns a copy of a swash initiated by the called function. This is the
4156 * public interface, and returning a copy prevents others from doing
4157 * mischief on the original */
4159 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
4164 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4165 I32 minbits, I32 none, SV* invlist,
4169 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
4170 * use the following define */
4172 #define CORE_SWASH_INIT_RETURN(x) \
4173 PL_curpm= old_PL_curpm; \
4176 /* Initialize and return a swash, creating it if necessary. It does this
4177 * by calling utf8_heavy.pl in the general case. The returned value may be
4178 * the swash's inversion list instead if the input parameters allow it.
4179 * Which is returned should be immaterial to callers, as the only
4180 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
4181 * and swash_to_invlist() handle both these transparently.
4183 * This interface should only be used by functions that won't destroy or
4184 * adversely change the swash, as doing so affects all other uses of the
4185 * swash in the program; the general public should use 'Perl_swash_init'
4188 * pkg is the name of the package that <name> should be in.
4189 * name is the name of the swash to find. Typically it is a Unicode
4190 * property name, including user-defined ones
4191 * listsv is a string to initialize the swash with. It must be of the form
4192 * documented as the subroutine return value in
4193 * L<perlunicode/User-Defined Character Properties>
4194 * minbits is the number of bits required to represent each data element.
4195 * It is '1' for binary properties.
4196 * none I (khw) do not understand this one, but it is used only in tr///.
4197 * invlist is an inversion list to initialize the swash with (or NULL)
4198 * flags_p if non-NULL is the address of various input and output flag bits
4199 * to the routine, as follows: ('I' means is input to the routine;
4200 * 'O' means output from the routine. Only flags marked O are
4201 * meaningful on return.)
4202 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
4203 * came from a user-defined property. (I O)
4204 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
4205 * when the swash cannot be located, to simply return NULL. (I)
4206 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
4207 * return of an inversion list instead of a swash hash if this routine
4208 * thinks that would result in faster execution of swash_fetch() later
4211 * Thus there are three possible inputs to find the swash: <name>,
4212 * <listsv>, and <invlist>. At least one must be specified. The result
4213 * will be the union of the specified ones, although <listsv>'s various
4214 * actions can intersect, etc. what <name> gives. To avoid going out to
4215 * disk at all, <invlist> should specify completely what the swash should
4216 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
4218 * <invlist> is only valid for binary properties */
4220 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
4222 SV* retval = &PL_sv_undef;
4223 HV* swash_hv = NULL;
4224 const bool use_invlist= (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST);
4226 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
4227 assert(! invlist || minbits == 1);
4229 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
4230 regex that triggered the swash init and the swash init
4231 perl logic itself. See perl #122747 */
4233 /* If data was passed in to go out to utf8_heavy to find the swash of, do
4235 if (listsv != &PL_sv_undef || strNE(name, "")) {
4237 const size_t pkg_len = strlen(pkg);
4238 const size_t name_len = strlen(name);
4239 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
4243 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
4245 PUSHSTACKi(PERLSI_MAGIC);
4249 /* We might get here via a subroutine signature which uses a utf8
4250 * parameter name, at which point PL_subname will have been set
4251 * but not yet used. */
4252 save_item(PL_subname);
4253 if (PL_parser && PL_parser->error_count)
4254 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
4255 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4256 if (!method) { /* demand load UTF-8 */
4258 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4259 GvSV(PL_errgv) = NULL;
4260 #ifndef NO_TAINT_SUPPORT
4261 /* It is assumed that callers of this routine are not passing in
4262 * any user derived data. */
4263 /* Need to do this after save_re_context() as it will set
4264 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
4265 * in Perl_magic_get). Even line to create errsv_save can turn on
4267 SAVEBOOL(TAINT_get);
4270 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
4273 /* Not ERRSV, as there is no need to vivify a scalar we are
4274 about to discard. */
4275 SV * const errsv = GvSV(PL_errgv);
4276 if (!SvTRUE(errsv)) {
4277 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4278 SvREFCNT_dec(errsv);
4286 mPUSHp(pkg, pkg_len);
4287 mPUSHp(name, name_len);
4292 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4293 GvSV(PL_errgv) = NULL;
4294 /* If we already have a pointer to the method, no need to use
4295 * call_method() to repeat the lookup. */
4297 ? call_sv(MUTABLE_SV(method), G_SCALAR)
4298 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
4300 retval = *PL_stack_sp--;
4301 SvREFCNT_inc(retval);