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(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(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_helper(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_HELPER;
1499 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1500 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1501 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1506 /* Each of the affected Hanguls starts with \xED */
1508 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1519 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1520 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1521 | (s0[2] & UTF_CONTINUATION_MASK);
1526 /* In conjunction with the exhaustive tests that can be enabled in
1527 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1528 * what it is intended to do, and that no flaws in it are masked by
1529 * dropping down and executing the code below
1530 assert(! isUTF8_CHAR(s0, send)
1531 || UTF8_IS_SURROGATE(s0, send)
1532 || UTF8_IS_SUPER(s0, send)
1533 || UTF8_IS_NONCHAR(s0,send));
1538 possible_problems = 0;
1542 adjusted_s0 = (U8 *) s0;
1549 errors = &discard_errors;
1552 /* The order of malformation tests here is important. We should consume as
1553 * few bytes as possible in order to not skip any valid character. This is
1554 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1555 * http://unicode.org/reports/tr36 for more discussion as to why. For
1556 * example, once we've done a UTF8SKIP, we can tell the expected number of
1557 * bytes, and could fail right off the bat if the input parameters indicate
1558 * that there are too few available. But it could be that just that first
1559 * byte is garbled, and the intended character occupies fewer bytes. If we
1560 * blindly assumed that the first byte is correct, and skipped based on
1561 * that number, we could skip over a valid input character. So instead, we
1562 * always examine the sequence byte-by-byte.
1564 * We also should not consume too few bytes, otherwise someone could inject
1565 * things. For example, an input could be deliberately designed to
1566 * overflow, and if this code bailed out immediately upon discovering that,
1567 * returning to the caller C<*retlen> pointing to the very next byte (one
1568 * which is actually part of of the overflowing sequence), that could look
1569 * legitimate to the caller, which could discard the initial partial
1570 * sequence and process the rest, inappropriately.
1572 * Some possible input sequences are malformed in more than one way. This
1573 * function goes to lengths to try to find all of them. This is necessary
1574 * for correctness, as the inputs may allow one malformation but not
1575 * another, and if we abandon searching for others after finding the
1576 * allowed one, we could allow in something that shouldn't have been.
1579 if (UNLIKELY(curlen == 0)) {
1580 possible_problems |= UTF8_GOT_EMPTY;
1582 uv = UNICODE_REPLACEMENT;
1583 goto ready_to_handle_errors;
1586 expectlen = UTF8SKIP(s);
1588 /* A well-formed UTF-8 character, as the vast majority of calls to this
1589 * function will be for, has this expected length. For efficiency, set
1590 * things up here to return it. It will be overriden only in those rare
1591 * cases where a malformation is found */
1593 *retlen = expectlen;
1596 /* A continuation character can't start a valid sequence */
1597 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1598 possible_problems |= UTF8_GOT_CONTINUATION;
1600 uv = UNICODE_REPLACEMENT;
1601 goto ready_to_handle_errors;
1604 /* Here is not a continuation byte, nor an invariant. The only thing left
1605 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1606 * because it excludes start bytes like \xC0 that always lead to
1609 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1610 * that indicate the number of bytes in the character's whole UTF-8
1611 * sequence, leaving just the bits that are part of the value. */
1612 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1614 /* Setup the loop end point, making sure to not look past the end of the
1615 * input string, and flag it as too short if the size isn't big enough. */
1616 if (UNLIKELY(curlen < expectlen)) {
1617 possible_problems |= UTF8_GOT_SHORT;
1621 send = (U8*) s0 + expectlen;
1624 /* Now, loop through the remaining bytes in the character's sequence,
1625 * accumulating each into the working value as we go. */
1626 for (s = s0 + 1; s < send; s++) {
1627 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1628 uv = UTF8_ACCUMULATE(uv, *s);
1632 /* Here, found a non-continuation before processing all expected bytes.
1633 * This byte indicates the beginning of a new character, so quit, even
1634 * if allowing this malformation. */
1635 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1637 } /* End of loop through the character's bytes */
1639 /* Save how many bytes were actually in the character */
1642 /* Note that there are two types of too-short malformation. One is when
1643 * there is actual wrong data before the normal termination of the
1644 * sequence. The other is that the sequence wasn't complete before the end
1645 * of the data we are allowed to look at, based on the input 'curlen'.
1646 * This means that we were passed data for a partial character, but it is
1647 * valid as far as we saw. The other is definitely invalid. This
1648 * distinction could be important to a caller, so the two types are kept
1651 * A convenience macro that matches either of the too-short conditions. */
1652 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1654 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1656 uv = UNICODE_REPLACEMENT;
1659 /* Check for overflow. The algorithm requires us to not look past the end
1660 * of the current character, even if partial, so the upper limit is 's' */
1661 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1662 1 /* Do consider overlongs */
1665 possible_problems |= UTF8_GOT_OVERFLOW;
1666 uv = UNICODE_REPLACEMENT;
1669 /* Check for overlong. If no problems so far, 'uv' is the correct code
1670 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1671 * we must look at the UTF-8 byte sequence itself to see if it is for an
1673 if ( ( LIKELY(! possible_problems)
1674 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1675 || ( UNLIKELY(possible_problems)
1676 && ( UNLIKELY(! UTF8_IS_START(*s0))
1678 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1681 possible_problems |= UTF8_GOT_LONG;
1683 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1685 /* The calculation in the 'true' branch of this 'if'
1686 * below won't work if overflows, and isn't needed
1687 * anyway. Further below we handle all overflow
1689 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1691 UV min_uv = uv_so_far;
1694 /* Here, the input is both overlong and is missing some trailing
1695 * bytes. There is no single code point it could be for, but there
1696 * may be enough information present to determine if what we have
1697 * so far is for an unallowed code point, such as for a surrogate.
1698 * The code further below has the intelligence to determine this,
1699 * but just for non-overlong UTF-8 sequences. What we do here is
1700 * calculate the smallest code point the input could represent if
1701 * there were no too short malformation. Then we compute and save
1702 * the UTF-8 for that, which is what the code below looks at
1703 * instead of the raw input. It turns out that the smallest such
1704 * code point is all we need. */
1705 for (i = curlen; i < expectlen; i++) {
1706 min_uv = UTF8_ACCUMULATE(min_uv,
1707 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1710 adjusted_s0 = temp_char_buf;
1711 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1715 /* Here, we have found all the possible problems, except for when the input
1716 * is for a problematic code point not allowed by the input parameters. */
1718 /* uv is valid for overlongs */
1719 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1721 /* isn't problematic if < this */
1722 && uv >= UNICODE_SURROGATE_FIRST)
1723 || ( UNLIKELY(possible_problems)
1725 /* if overflow, we know without looking further
1726 * precisely which of the problematic types it is,
1727 * and we deal with those in the overflow handling
1729 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1730 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1731 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1732 && ((flags & ( UTF8_DISALLOW_NONCHAR
1733 |UTF8_DISALLOW_SURROGATE
1734 |UTF8_DISALLOW_SUPER
1735 |UTF8_DISALLOW_PERL_EXTENDED
1737 |UTF8_WARN_SURROGATE
1739 |UTF8_WARN_PERL_EXTENDED))))
1741 /* If there were no malformations, or the only malformation is an
1742 * overlong, 'uv' is valid */
1743 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1744 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1745 possible_problems |= UTF8_GOT_SURROGATE;
1747 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1748 possible_problems |= UTF8_GOT_SUPER;
1750 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1751 possible_problems |= UTF8_GOT_NONCHAR;
1754 else { /* Otherwise, need to look at the source UTF-8, possibly
1755 adjusted to be non-overlong */
1757 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1758 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1760 possible_problems |= UTF8_GOT_SUPER;
1762 else if (curlen > 1) {
1763 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1764 NATIVE_UTF8_TO_I8(*adjusted_s0),
1765 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1767 possible_problems |= UTF8_GOT_SUPER;
1769 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1770 NATIVE_UTF8_TO_I8(*adjusted_s0),
1771 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1773 possible_problems |= UTF8_GOT_SURROGATE;
1777 /* We need a complete well-formed UTF-8 character to discern
1778 * non-characters, so can't look for them here */
1782 ready_to_handle_errors:
1785 * curlen contains the number of bytes in the sequence that
1786 * this call should advance the input by.
1787 * avail_len gives the available number of bytes passed in, but
1788 * only if this is less than the expected number of
1789 * bytes, based on the code point's start byte.
1790 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1791 * is set in it for each potential problem found.
1792 * uv contains the code point the input sequence
1793 * represents; or if there is a problem that prevents
1794 * a well-defined value from being computed, it is
1795 * some subsitute value, typically the REPLACEMENT
1797 * s0 points to the first byte of the character
1798 * s points to just after were we left off processing
1800 * send points to just after where that character should
1801 * end, based on how many bytes the start byte tells
1802 * us should be in it, but no further than s0 +
1806 if (UNLIKELY(possible_problems)) {
1807 bool disallowed = FALSE;
1808 const U32 orig_problems = possible_problems;
1814 while (possible_problems) { /* Handle each possible problem */
1816 char * message = NULL;
1817 U32 this_flag_bit = 0;
1819 /* Each 'if' clause handles one problem. They are ordered so that
1820 * the first ones' messages will be displayed before the later
1821 * ones; this is kinda in decreasing severity order. But the
1822 * overlong must come last, as it changes 'uv' looked at by the
1824 if (possible_problems & UTF8_GOT_OVERFLOW) {
1826 /* Overflow means also got a super and are using Perl's
1827 * extended UTF-8, but we handle all three cases here */
1829 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1830 *errors |= UTF8_GOT_OVERFLOW;
1832 /* But the API says we flag all errors found */
1833 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1834 *errors |= UTF8_GOT_SUPER;
1837 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1839 *errors |= UTF8_GOT_PERL_EXTENDED;
1842 /* Disallow if any of the three categories say to */
1843 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1844 || (flags & ( UTF8_DISALLOW_SUPER
1845 |UTF8_DISALLOW_PERL_EXTENDED)))
1850 /* Likewise, warn if any say to */
1851 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1852 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1855 /* The warnings code explicitly says it doesn't handle the
1856 * case of packWARN2 and two categories which have
1857 * parent-child relationship. Even if it works now to
1858 * raise the warning if either is enabled, it wouldn't
1859 * necessarily do so in the future. We output (only) the
1860 * most dire warning */
1861 if (! (flags & UTF8_CHECK_ONLY)) {
1862 if (msgs || ckWARN_d(WARN_UTF8)) {
1863 pack_warn = packWARN(WARN_UTF8);
1865 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1866 pack_warn = packWARN(WARN_NON_UNICODE);
1869 message = Perl_form(aTHX_ "%s: %s (overflows)",
1871 _byte_dump_string(s0, curlen, 0));
1872 this_flag_bit = UTF8_GOT_OVERFLOW;
1877 else if (possible_problems & UTF8_GOT_EMPTY) {
1878 possible_problems &= ~UTF8_GOT_EMPTY;
1879 *errors |= UTF8_GOT_EMPTY;
1881 if (! (flags & UTF8_ALLOW_EMPTY)) {
1883 /* This so-called malformation is now treated as a bug in
1884 * the caller. If you have nothing to decode, skip calling
1890 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1892 pack_warn = packWARN(WARN_UTF8);
1893 message = Perl_form(aTHX_ "%s (empty string)",
1895 this_flag_bit = UTF8_GOT_EMPTY;
1899 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1900 possible_problems &= ~UTF8_GOT_CONTINUATION;
1901 *errors |= UTF8_GOT_CONTINUATION;
1903 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1906 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1908 pack_warn = packWARN(WARN_UTF8);
1909 message = Perl_form(aTHX_
1910 "%s: %s (unexpected continuation byte 0x%02x,"
1911 " with no preceding start byte)",
1913 _byte_dump_string(s0, 1, 0), *s0);
1914 this_flag_bit = UTF8_GOT_CONTINUATION;
1918 else if (possible_problems & UTF8_GOT_SHORT) {
1919 possible_problems &= ~UTF8_GOT_SHORT;
1920 *errors |= UTF8_GOT_SHORT;
1922 if (! (flags & UTF8_ALLOW_SHORT)) {
1925 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1927 pack_warn = packWARN(WARN_UTF8);
1928 message = Perl_form(aTHX_
1929 "%s: %s (too short; %d byte%s available, need %d)",
1931 _byte_dump_string(s0, send - s0, 0),
1933 avail_len == 1 ? "" : "s",
1935 this_flag_bit = UTF8_GOT_SHORT;
1940 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1941 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1942 *errors |= UTF8_GOT_NON_CONTINUATION;
1944 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1947 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1950 /* If we don't know for sure that the input length is
1951 * valid, avoid as much as possible reading past the
1952 * end of the buffer */
1953 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1956 pack_warn = packWARN(WARN_UTF8);
1957 message = Perl_form(aTHX_ "%s",
1958 unexpected_non_continuation_text(s0,
1962 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
1966 else if (possible_problems & UTF8_GOT_SURROGATE) {
1967 possible_problems &= ~UTF8_GOT_SURROGATE;
1969 if (flags & UTF8_WARN_SURROGATE) {
1970 *errors |= UTF8_GOT_SURROGATE;
1972 if ( ! (flags & UTF8_CHECK_ONLY)
1973 && (msgs || ckWARN_d(WARN_SURROGATE)))
1975 pack_warn = packWARN(WARN_SURROGATE);
1977 /* These are the only errors that can occur with a
1978 * surrogate when the 'uv' isn't valid */
1979 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1980 message = Perl_form(aTHX_
1981 "UTF-16 surrogate (any UTF-8 sequence that"
1982 " starts with \"%s\" is for a surrogate)",
1983 _byte_dump_string(s0, curlen, 0));
1986 message = Perl_form(aTHX_ surrogate_cp_format, uv);
1988 this_flag_bit = UTF8_GOT_SURROGATE;
1992 if (flags & UTF8_DISALLOW_SURROGATE) {
1994 *errors |= UTF8_GOT_SURROGATE;
1997 else if (possible_problems & UTF8_GOT_SUPER) {
1998 possible_problems &= ~UTF8_GOT_SUPER;
2000 if (flags & UTF8_WARN_SUPER) {
2001 *errors |= UTF8_GOT_SUPER;
2003 if ( ! (flags & UTF8_CHECK_ONLY)
2004 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2006 pack_warn = packWARN(WARN_NON_UNICODE);
2008 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2009 message = Perl_form(aTHX_
2010 "Any UTF-8 sequence that starts with"
2011 " \"%s\" is for a non-Unicode code point,"
2012 " may not be portable",
2013 _byte_dump_string(s0, curlen, 0));
2016 message = Perl_form(aTHX_ super_cp_format, uv);
2018 this_flag_bit = UTF8_GOT_SUPER;
2022 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2023 * and before possibly bailing out, so that the more dire
2024 * warning will override the regular one. */
2025 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2026 if ( ! (flags & UTF8_CHECK_ONLY)
2027 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2028 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2030 pack_warn = packWARN(WARN_NON_UNICODE);
2032 /* If it is an overlong that evaluates to a code point
2033 * that doesn't have to use the Perl extended UTF-8, it
2034 * still used it, and so we output a message that
2035 * doesn't refer to the code point. The same is true
2036 * if there was a SHORT malformation where the code
2037 * point is not valid. In that case, 'uv' will have
2038 * been set to the REPLACEMENT CHAR, and the message
2039 * below without the code point in it will be selected
2041 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2042 message = Perl_form(aTHX_
2043 perl_extended_cp_format, uv);
2046 message = Perl_form(aTHX_
2047 "Any UTF-8 sequence that starts with"
2048 " \"%s\" is a Perl extension, and"
2049 " so is not portable",
2050 _byte_dump_string(s0, curlen, 0));
2052 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2055 if (flags & ( UTF8_WARN_PERL_EXTENDED
2056 |UTF8_DISALLOW_PERL_EXTENDED))
2058 *errors |= UTF8_GOT_PERL_EXTENDED;
2060 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2066 if (flags & UTF8_DISALLOW_SUPER) {
2067 *errors |= UTF8_GOT_SUPER;
2071 else if (possible_problems & UTF8_GOT_NONCHAR) {
2072 possible_problems &= ~UTF8_GOT_NONCHAR;
2074 if (flags & UTF8_WARN_NONCHAR) {
2075 *errors |= UTF8_GOT_NONCHAR;
2077 if ( ! (flags & UTF8_CHECK_ONLY)
2078 && (msgs || ckWARN_d(WARN_NONCHAR)))
2080 /* The code above should have guaranteed that we don't
2081 * get here with errors other than overlong */
2082 assert (! (orig_problems
2083 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2085 pack_warn = packWARN(WARN_NONCHAR);
2086 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2087 this_flag_bit = UTF8_GOT_NONCHAR;
2091 if (flags & UTF8_DISALLOW_NONCHAR) {
2093 *errors |= UTF8_GOT_NONCHAR;
2096 else if (possible_problems & UTF8_GOT_LONG) {
2097 possible_problems &= ~UTF8_GOT_LONG;
2098 *errors |= UTF8_GOT_LONG;
2100 if (flags & UTF8_ALLOW_LONG) {
2102 /* We don't allow the actual overlong value, unless the
2103 * special extra bit is also set */
2104 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2105 & ~UTF8_ALLOW_LONG)))
2107 uv = UNICODE_REPLACEMENT;
2114 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2116 pack_warn = packWARN(WARN_UTF8);
2118 /* These error types cause 'uv' to be something that
2119 * isn't what was intended, so can't use it in the
2120 * message. The other error types either can't
2121 * generate an overlong, or else the 'uv' is valid */
2123 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2125 message = Perl_form(aTHX_
2126 "%s: %s (any UTF-8 sequence that starts"
2127 " with \"%s\" is overlong which can and"
2128 " should be represented with a"
2129 " different, shorter sequence)",
2131 _byte_dump_string(s0, send - s0, 0),
2132 _byte_dump_string(s0, curlen, 0));
2135 U8 tmpbuf[UTF8_MAXBYTES+1];
2136 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2138 /* Don't use U+ for non-Unicode code points, which
2139 * includes those in the Latin1 range */
2140 const char * preface = ( uv > PERL_UNICODE_MAX
2147 message = Perl_form(aTHX_
2148 "%s: %s (overlong; instead use %s to represent"
2151 _byte_dump_string(s0, send - s0, 0),
2152 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2154 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2155 small code points */
2158 this_flag_bit = UTF8_GOT_LONG;
2161 } /* End of looking through the possible flags */
2163 /* Display the message (if any) for the problem being handled in
2164 * this iteration of the loop */
2167 assert(this_flag_bit);
2169 if (*msgs == NULL) {
2173 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2178 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2181 Perl_warner(aTHX_ pack_warn, "%s", message);
2183 } /* End of 'while (possible_problems)' */
2185 /* Since there was a possible problem, the returned length may need to
2186 * be changed from the one stored at the beginning of this function.
2187 * Instead of trying to figure out if that's needed, just do it. */
2193 if (flags & UTF8_CHECK_ONLY && retlen) {
2194 *retlen = ((STRLEN) -1);
2200 return UNI_TO_NATIVE(uv);
2204 =for apidoc utf8_to_uvchr_buf
2206 Returns the native code point of the first character in the string C<s> which
2207 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2208 C<*retlen> will be set to the length, in bytes, of that character.
2210 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2211 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2212 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2213 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2214 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2215 the next possible position in C<s> that could begin a non-malformed character.
2216 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2221 Also implemented as a macro in utf8.h
2227 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2229 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2233 return utf8n_to_uvchr(s, send - s, retlen,
2234 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
2237 /* This is marked as deprecated
2239 =for apidoc utf8_to_uvuni_buf
2241 Only in very rare circumstances should code need to be dealing in Unicode
2242 (as opposed to native) code points. In those few cases, use
2243 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead. If you
2244 are not absolutely sure this is one of those cases, then assume it isn't and
2245 use plain C<utf8_to_uvchr_buf> instead.
2247 Returns the Unicode (not-native) code point of the first character in the
2249 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2250 C<retlen> will be set to the length, in bytes, of that character.
2252 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2253 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2254 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2255 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2256 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2257 next possible position in C<s> that could begin a non-malformed character.
2258 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
2264 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2266 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2270 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2274 =for apidoc utf8_length
2276 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2277 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2278 same place, it returns 0 with no warning raised.
2280 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2281 and returns the number of valid characters.
2287 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2291 PERL_ARGS_ASSERT_UTF8_LENGTH;
2293 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2294 * the bitops (especially ~) can create illegal UTF-8.
2295 * In other words: in Perl UTF-8 is not just for Unicode. */
2298 goto warn_and_return;
2308 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2309 "%s in %s", unees, OP_DESC(PL_op));
2311 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2318 =for apidoc bytes_cmp_utf8
2320 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2321 sequence of characters (stored as UTF-8)
2322 in C<u>, C<ulen>. Returns 0 if they are
2323 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2324 if the first string is greater than the second string.
2326 -1 or +1 is returned if the shorter string was identical to the start of the
2327 longer string. -2 or +2 is returned if
2328 there was a difference between characters
2335 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2337 const U8 *const bend = b + blen;
2338 const U8 *const uend = u + ulen;
2340 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2342 while (b < bend && u < uend) {
2344 if (!UTF8_IS_INVARIANT(c)) {
2345 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2348 if (UTF8_IS_CONTINUATION(c1)) {
2349 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2351 /* diag_listed_as: Malformed UTF-8 character%s */
2352 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2354 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2355 PL_op ? " in " : "",
2356 PL_op ? OP_DESC(PL_op) : "");
2361 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2362 "%s in %s", unees, OP_DESC(PL_op));
2364 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2365 return -2; /* Really want to return undef :-) */
2372 return *b < c ? -2 : +2;
2377 if (b == bend && u == uend)
2380 return b < bend ? +1 : -1;
2384 =for apidoc utf8_to_bytes
2386 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2387 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2388 updates C<*lenp> to contain the new length.
2389 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2391 Upon successful return, the number of variants in the string can be computed by
2392 having saved the value of C<*lenp> before the call, and subtracting the
2393 after-call value of C<*lenp> from it.
2395 If you need a copy of the string, see L</bytes_from_utf8>.
2401 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2405 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2406 PERL_UNUSED_CONTEXT;
2408 /* This is a no-op if no variants at all in the input */
2409 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2414 U8 * const save = s;
2415 U8 * const send = s + *lenp;
2418 /* Nothing before the first variant needs to be changed, so start the real
2422 if (! UTF8_IS_INVARIANT(*s)) {
2423 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2424 *lenp = ((STRLEN) -1);
2432 /* Is downgradable, so do it */
2433 d = s = first_variant;
2436 if (! UVCHR_IS_INVARIANT(c)) {
2437 /* Then it is two-byte encoded */
2438 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2451 =for apidoc bytes_from_utf8
2453 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2454 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2455 actually encoded in UTF-8.
2457 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2460 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2461 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2462 C<*lenp> are unchanged, and the return value is the original C<s>.
2464 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2465 newly created string containing a downgraded copy of C<s>, and whose length is
2466 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2467 caller is responsible for arranging for the memory used by this string to get
2470 Upon successful return, the number of variants in the string can be computed by
2471 having saved the value of C<*lenp> before the call, and subtracting the
2472 after-call value of C<*lenp> from it.
2476 There is a macro that avoids this function call, but this is retained for
2477 anyone who calls it with the Perl_ prefix */
2480 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2482 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2483 PERL_UNUSED_CONTEXT;
2485 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2489 No = here because currently externally undocumented
2490 for apidoc bytes_from_utf8_loc
2492 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2493 to store the location of the first character in C<"s"> that cannot be
2494 converted to non-UTF8.
2496 If that parameter is C<NULL>, this function behaves identically to
2499 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2500 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2502 Otherwise, the function returns a newly created C<NUL>-terminated string
2503 containing the non-UTF8 equivalent of the convertible first portion of
2504 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2505 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2506 and C<*first_non_downgradable> is set to C<NULL>.
2508 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2509 first character in the original string that wasn't converted. C<*is_utf8p> is
2510 unchanged. Note that the new string may have length 0.
2512 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2513 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2514 converts as many characters in it as possible stopping at the first one it
2515 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2516 set to point to that. The function returns the portion that could be converted
2517 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2518 not including the terminating C<NUL>. If the very first character in the
2519 original could not be converted, C<*lenp> will be 0, and the new string will
2520 contain just a single C<NUL>. If the entire input string was converted,
2521 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2523 Upon successful return, the number of variants in the converted portion of the
2524 string can be computed by having saved the value of C<*lenp> before the call,
2525 and subtracting the after-call value of C<*lenp> from it.
2533 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2536 const U8 *original = s;
2537 U8 *converted_start;
2538 const U8 *send = s + *lenp;
2540 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2543 if (first_unconverted) {
2544 *first_unconverted = NULL;
2547 return (U8 *) original;
2550 Newx(d, (*lenp) + 1, U8);
2552 converted_start = d;
2555 if (! UTF8_IS_INVARIANT(c)) {
2557 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2558 * have to stop now */
2559 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2560 if (first_unconverted) {
2561 *first_unconverted = s - 1;
2562 goto finish_and_return;
2565 Safefree(converted_start);
2566 return (U8 *) original;
2570 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2576 /* Here, converted the whole of the input */
2578 if (first_unconverted) {
2579 *first_unconverted = NULL;
2584 *lenp = d - converted_start;
2586 /* Trim unused space */
2587 Renew(converted_start, *lenp + 1, U8);
2589 return converted_start;
2593 =for apidoc bytes_to_utf8
2595 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2597 Returns a pointer to the newly-created string, and sets C<*lenp> to
2598 reflect the new length in bytes. The caller is responsible for arranging for
2599 the memory used by this string to get freed.
2601 Upon successful return, the number of variants in the string can be computed by
2602 having saved the value of C<*lenp> before the call, and subtracting it from the
2603 after-call value of C<*lenp>.
2605 A C<NUL> character will be written after the end of the string.
2607 If you want to convert to UTF-8 from encodings other than
2608 the native (Latin1 or EBCDIC),
2609 see L</sv_recode_to_utf8>().
2615 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2617 const U8 * const send = s + (*lenp);
2621 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2622 PERL_UNUSED_CONTEXT;
2624 Newx(d, (*lenp) * 2 + 1, U8);
2628 append_utf8_from_native_byte(*s, &d);
2635 /* Trim unused space */
2636 Renew(dst, *lenp + 1, U8);
2642 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2643 * use utf16_to_utf8_reversed().
2645 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2646 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2647 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2649 * These functions don't check for overflow. The worst case is every code
2650 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2651 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2652 * destination must be pre-extended to 2 times the source length.
2654 * Do not use in-place. We optimize for native, for obvious reasons. */
2657 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2662 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2665 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2671 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2673 if (OFFUNI_IS_INVARIANT(uv)) {
2674 *d++ = LATIN1_TO_NATIVE((U8) uv);
2677 if (uv <= MAX_UTF8_TWO_BYTE) {
2678 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2679 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2683 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2684 #define LAST_HIGH_SURROGATE 0xDBFF
2685 #define FIRST_LOW_SURROGATE 0xDC00
2686 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2687 #define FIRST_IN_PLANE1 0x10000
2689 /* This assumes that most uses will be in the first Unicode plane, not
2690 * needing surrogates */
2691 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2692 && uv <= UNICODE_SURROGATE_LAST))
2694 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2695 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2698 UV low = (p[0] << 8) + p[1];
2699 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2700 || UNLIKELY(low > LAST_LOW_SURROGATE))
2702 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2705 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2706 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2710 d = uvoffuni_to_utf8_flags(d, uv, 0);
2712 if (uv < FIRST_IN_PLANE1) {
2713 *d++ = (U8)(( uv >> 12) | 0xe0);
2714 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2715 *d++ = (U8)(( uv & 0x3f) | 0x80);
2719 *d++ = (U8)(( uv >> 18) | 0xf0);
2720 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2721 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2722 *d++ = (U8)(( uv & 0x3f) | 0x80);
2727 *newlen = d - dstart;
2731 /* Note: this one is slightly destructive of the source. */
2734 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2737 U8* const send = s + bytelen;
2739 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2742 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2746 const U8 tmp = s[0];
2751 return utf16_to_utf8(p, d, bytelen, newlen);
2755 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2757 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2760 /* Internal function so we can deprecate the external one, and call
2761 this one from other deprecated functions in this file */
2764 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2766 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2770 return is_utf8_common(p, NULL,
2771 "This is buggy if this gets used",
2776 Perl__is_uni_perl_idcont(pTHX_ UV c)
2778 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2782 Perl__is_uni_perl_idstart(pTHX_ UV c)
2784 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2788 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2791 /* We have the latin1-range values compiled into the core, so just use
2792 * those, converting the result to UTF-8. The only difference between upper
2793 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2794 * either "SS" or "Ss". Which one to use is passed into the routine in
2795 * 'S_or_s' to avoid a test */
2797 UV converted = toUPPER_LATIN1_MOD(c);
2799 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2801 assert(S_or_s == 'S' || S_or_s == 's');
2803 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2804 characters in this range */
2805 *p = (U8) converted;
2810 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2811 * which it maps to one of them, so as to only have to have one check for
2812 * it in the main case */
2813 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2815 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2816 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2819 converted = GREEK_CAPITAL_LETTER_MU;
2821 #if UNICODE_MAJOR_VERSION > 2 \
2822 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2823 && UNICODE_DOT_DOT_VERSION >= 8)
2824 case LATIN_SMALL_LETTER_SHARP_S:
2831 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2832 " '%c' to map to '%c'",
2833 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2834 NOT_REACHED; /* NOTREACHED */
2838 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2839 *p = UTF8_TWO_BYTE_LO(converted);
2845 /* If compiled on an early Unicode version, there may not be auxiliary tables
2847 #ifndef HAS_UC_AUX_TABLES
2848 # define UC_AUX_TABLE_ptrs NULL
2849 # define UC_AUX_TABLE_lengths NULL
2851 #ifndef HAS_TC_AUX_TABLES
2852 # define TC_AUX_TABLE_ptrs NULL
2853 # define TC_AUX_TABLE_lengths NULL
2855 #ifndef HAS_LC_AUX_TABLES
2856 # define LC_AUX_TABLE_ptrs NULL
2857 # define LC_AUX_TABLE_lengths NULL
2859 #ifndef HAS_CF_AUX_TABLES
2860 # define CF_AUX_TABLE_ptrs NULL
2861 # define CF_AUX_TABLE_lengths NULL
2863 #ifndef HAS_UC_AUX_TABLES
2864 # define UC_AUX_TABLE_ptrs NULL
2865 # define UC_AUX_TABLE_lengths NULL
2868 /* Call the function to convert a UTF-8 encoded character to the specified case.
2869 * Note that there may be more than one character in the result.
2870 * 's' is a pointer to the first byte of the input character
2871 * 'd' will be set to the first byte of the string of changed characters. It
2872 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2873 * 'lenp' will be set to the length in bytes of the string of changed characters
2875 * The functions return the ordinal of the first character in the string of
2877 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2878 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2879 Uppercase_Mapping_invmap, \
2880 UC_AUX_TABLE_ptrs, \
2881 UC_AUX_TABLE_lengths, \
2883 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2884 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2885 Titlecase_Mapping_invmap, \
2886 TC_AUX_TABLE_ptrs, \
2887 TC_AUX_TABLE_lengths, \
2889 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2890 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2891 Lowercase_Mapping_invmap, \
2892 LC_AUX_TABLE_ptrs, \
2893 LC_AUX_TABLE_lengths, \
2897 /* This additionally has the input parameter 'specials', which if non-zero will
2898 * cause this to use the specials hash for folding (meaning get full case
2899 * folding); otherwise, when zero, this implies a simple case fold */
2900 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2902 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2903 Case_Folding_invmap, \
2904 CF_AUX_TABLE_ptrs, \
2905 CF_AUX_TABLE_lengths, \
2907 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2908 Simple_Case_Folding_invmap, \
2913 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2915 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2916 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2917 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2918 * the changed version may be longer than the original character.
2920 * The ordinal of the first character of the changed version is returned
2921 * (but note, as explained above, that there may be more.) */
2923 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2926 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2929 uvchr_to_utf8(p, c);
2930 return CALL_UPPER_CASE(c, p, p, lenp);
2934 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2936 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2939 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2942 uvchr_to_utf8(p, c);
2943 return CALL_TITLE_CASE(c, p, p, lenp);
2947 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2949 /* We have the latin1-range values compiled into the core, so just use
2950 * those, converting the result to UTF-8. Since the result is always just
2951 * one character, we allow <p> to be NULL */
2953 U8 converted = toLOWER_LATIN1(c);
2955 PERL_UNUSED_ARG(dummy);
2958 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2963 /* Result is known to always be < 256, so can use the EIGHT_BIT
2965 *p = UTF8_EIGHT_BIT_HI(converted);
2966 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2974 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2976 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2979 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2982 uvchr_to_utf8(p, c);
2983 return CALL_LOWER_CASE(c, p, p, lenp);
2987 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2989 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2990 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2991 * FOLD_FLAGS_FULL iff full folding is to be used;
2993 * Not to be used for locale folds
2998 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3000 assert (! (flags & FOLD_FLAGS_LOCALE));
3002 if (UNLIKELY(c == MICRO_SIGN)) {
3003 converted = GREEK_SMALL_LETTER_MU;
3005 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3006 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3007 || UNICODE_DOT_DOT_VERSION > 0)
3008 else if ( (flags & FOLD_FLAGS_FULL)
3009 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3011 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3012 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3013 * under those circumstances. */
3014 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3015 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3016 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3018 return LATIN_SMALL_LETTER_LONG_S;
3028 else { /* In this range the fold of all other characters is their lower
3030 converted = toLOWER_LATIN1(c);
3033 if (UVCHR_IS_INVARIANT(converted)) {
3034 *p = (U8) converted;
3038 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3039 *p = UTF8_TWO_BYTE_LO(converted);
3047 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3050 /* Not currently externally documented, and subject to change
3051 * <flags> bits meanings:
3052 * FOLD_FLAGS_FULL iff full folding is to be used;
3053 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3054 * locale are to be used.
3055 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3058 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3060 if (flags & FOLD_FLAGS_LOCALE) {
3061 /* Treat a UTF-8 locale as not being in locale at all, except for
3062 * potentially warning */
3063 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3064 if (IN_UTF8_CTYPE_LOCALE) {
3065 flags &= ~FOLD_FLAGS_LOCALE;
3068 goto needs_full_generality;
3073 return _to_fold_latin1((U8) c, p, lenp,
3074 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3077 /* Here, above 255. If no special needs, just use the macro */
3078 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3079 uvchr_to_utf8(p, c);
3080 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
3082 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3083 the special flags. */
3084 U8 utf8_c[UTF8_MAXBYTES + 1];
3086 needs_full_generality:
3087 uvchr_to_utf8(utf8_c, c);
3088 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3093 PERL_STATIC_INLINE bool
3094 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
3095 const char *const swashname, SV* const invlist)
3097 /* returns a boolean giving whether or not the UTF8-encoded character that
3098 * starts at <p> is in the swash indicated by <swashname>. <swash>
3099 * contains a pointer to where the swash indicated by <swashname>
3100 * is to be stored; which this routine will do, so that future calls will
3101 * look at <*swash> and only generate a swash if it is not null. <invlist>
3102 * is NULL or an inversion list that defines the swash. If not null, it
3103 * saves time during initialization of the swash.
3105 * Note that it is assumed that the buffer length of <p> is enough to
3106 * contain all the bytes that comprise the character. Thus, <*p> should
3107 * have been checked before this call for mal-formedness enough to assure
3110 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3112 /* The API should have included a length for the UTF-8 character in <p>,
3113 * but it doesn't. We therefore assume that p has been validated at least
3114 * as far as there being enough bytes available in it to accommodate the
3115 * character without reading beyond the end, and pass that number on to the
3116 * validating routine */
3117 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
3118 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
3119 _UTF8_NO_CONFIDENCE_IN_CURLEN,
3121 NOT_REACHED; /* NOTREACHED */
3125 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3131 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3132 *swash = _core_swash_init("utf8",
3134 /* Only use the name if there is no inversion
3135 * list; otherwise will go out to disk */
3136 (invlist) ? "" : swashname,
3138 &PL_sv_undef, 1, 0, invlist, &flags);
3141 return swash_fetch(*swash, p, TRUE) != 0;
3144 PERL_STATIC_INLINE bool
3145 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
3146 SV **swash, const char *const swashname,
3149 /* returns a boolean giving whether or not the UTF8-encoded character that
3150 * starts at <p>, and extending no further than <e - 1> is in the swash
3151 * indicated by <swashname>. <swash> contains a pointer to where the swash
3152 * indicated by <swashname> is to be stored; which this routine will do, so
3153 * that future calls will look at <*swash> and only generate a swash if it
3154 * is not null. <invlist> is NULL or an inversion list that defines the
3155 * swash. If not null, it saves time during initialization of the swash.
3158 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
3160 if (! isUTF8_CHAR(p, e)) {
3161 _force_out_malformed_utf8_message(p, e, 0, 1);
3162 NOT_REACHED; /* NOTREACHED */
3166 return _invlist_contains_cp(invlist, valid_utf8_to_uvchr(p, NULL));
3172 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
3173 *swash = _core_swash_init("utf8",
3175 /* Only use the name if there is no inversion
3176 * list; otherwise will go out to disk */
3177 (invlist) ? "" : swashname,
3179 &PL_sv_undef, 1, 0, invlist, &flags);
3182 return swash_fetch(*swash, p, TRUE) != 0;
3186 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3187 const char * const alternative,
3188 const bool use_locale,
3189 const char * const file,
3190 const unsigned line)
3194 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3196 if (ckWARN_d(WARN_DEPRECATED)) {
3198 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3199 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3200 if (! PL_seen_deprecated_macro) {
3201 PL_seen_deprecated_macro = newHV();
3203 if (! hv_store(PL_seen_deprecated_macro, key,
3204 strlen(key), &PL_sv_undef, 0))
3206 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3209 if (instr(file, "mathoms.c")) {
3210 Perl_warner(aTHX_ WARN_DEPRECATED,
3211 "In %s, line %d, starting in Perl v5.30, %s()"
3212 " will be removed. Avoid this message by"
3213 " converting to use %s().\n",
3214 file, line, name, alternative);
3217 Perl_warner(aTHX_ WARN_DEPRECATED,
3218 "In %s, line %d, starting in Perl v5.30, %s() will"
3219 " require an additional parameter. Avoid this"
3220 " message by converting to use %s().\n",
3221 file, line, name, alternative);
3228 Perl__is_utf8_FOO(pTHX_ U8 classnum,
3230 const char * const name,
3231 const char * const alternative,
3232 const bool use_utf8,
3233 const bool use_locale,
3234 const char * const file,
3235 const unsigned line)
3237 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3239 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3241 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
3251 case _CC_ALPHANUMERIC:
3255 return is_utf8_common(p,
3257 "This is buggy if this gets used",
3258 PL_XPosix_ptrs[classnum]);
3261 return is_XPERLSPACE_high(p);
3263 return is_HORIZWS_high(p);
3265 return is_XDIGIT_high(p);
3271 return is_VERTWS_high(p);
3273 return is_utf8_common(p, NULL,
3274 "This is buggy if this gets used",
3275 PL_utf8_perl_idstart);
3277 return is_utf8_common(p, NULL,
3278 "This is buggy if this gets used",
3279 PL_utf8_perl_idcont);
3283 /* idcont is the same as wordchar below 256 */
3284 if (classnum == _CC_IDCONT) {
3285 classnum = _CC_WORDCHAR;
3287 else if (classnum == _CC_IDFIRST) {
3291 classnum = _CC_ALPHA;
3295 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3296 return _generic_isCC(*p, classnum);
3299 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
3302 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
3303 return isFOO_lc(classnum, *p);
3306 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
3309 NOT_REACHED; /* NOTREACHED */
3313 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
3316 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
3318 return is_utf8_common_with_len(p, e, NULL,
3319 "This is buggy if this gets used",
3320 PL_XPosix_ptrs[classnum]);
3324 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3326 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3328 return is_utf8_common_with_len(p, e, NULL,
3329 "This is buggy if this gets used",
3330 PL_utf8_perl_idstart);
3334 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3336 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3340 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
3344 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3346 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3348 return is_utf8_common_with_len(p, e, NULL,
3349 "This is buggy if this gets used",
3350 PL_utf8_perl_idcont);
3354 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3356 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3358 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
3362 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3364 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3366 return is_utf8_common(p, &PL_utf8_xidcont, "XIdContinue", NULL);
3370 Perl__is_utf8_mark(pTHX_ const U8 *p)
3372 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3374 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
3378 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3379 U8* ustrp, STRLEN *lenp,
3380 SV *invlist, const int * const invmap,
3381 const unsigned int * const * const aux_tables,
3382 const U8 * const aux_table_lengths,
3383 const char * const normal)
3387 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3388 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3389 * to name the new case in any generated messages, as a fallback if the
3390 * operation being used is not available. The new case is given by the
3391 * data structures in the remaining arguments.
3393 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3394 * entire changed case string, and the return value is the first code point
3397 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3399 /* For code points that don't change case, we already know that the output
3400 * of this function is the unchanged input, so we can skip doing look-ups
3401 * for them. Unfortunately the case-changing code points are scattered
3402 * around. But there are some long consecutive ranges where there are no
3403 * case changing code points. By adding tests, we can eliminate the lookup
3404 * for all the ones in such ranges. This is currently done here only for
3405 * just a few cases where the scripts are in common use in modern commerce
3406 * (and scripts adjacent to those which can be included without additional
3409 if (uv1 >= 0x0590) {
3410 /* This keeps from needing further processing the code points most
3411 * likely to be used in the following non-cased scripts: Hebrew,
3412 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3413 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3414 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3419 /* The following largish code point ranges also don't have case
3420 * changes, but khw didn't think they warranted extra tests to speed
3421 * them up (which would slightly slow down everything else above them):
3422 * 1100..139F Hangul Jamo, Ethiopic
3423 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3424 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3425 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3426 * Combining Diacritical Marks Extended, Balinese,
3427 * Sundanese, Batak, Lepcha, Ol Chiki
3428 * 2000..206F General Punctuation
3431 if (uv1 >= 0x2D30) {
3433 /* This keeps the from needing further processing the code points
3434 * most likely to be used in the following non-cased major scripts:
3435 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3437 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3438 * event that Unicode eventually allocates the unused block as of
3439 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3440 * that the test suite will start having failures to alert you
3441 * should that happen) */
3446 if (uv1 >= 0xAC00) {
3447 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3448 if (ckWARN_d(WARN_SURROGATE)) {
3449 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3450 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3451 "Operation \"%s\" returns its argument for"
3452 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3457 /* AC00..FAFF Catches Hangul syllables and private use, plus
3463 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3464 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3465 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3466 MAX_EXTERNALLY_LEGAL_CP);
3468 if (ckWARN_d(WARN_NON_UNICODE)) {
3469 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3470 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3471 "Operation \"%s\" returns its argument for"
3472 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3476 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3478 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3481 /* As of Unicode 10.0, this means we avoid swash creation
3482 * for anything beyond high Plane 1 (below emojis) */
3489 /* Note that non-characters are perfectly legal, so no warning should
3495 const unsigned int * cp_list;
3497 SSize_t index = _invlist_search(invlist, uv1);
3498 IV base = invmap[index];
3500 /* The data structures are set up so that if 'base' is non-negative,
3501 * the case change is 1-to-1; and if 0, the change is to itself */
3509 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3510 lc = base + uv1 - invlist_array(invlist)[index];
3511 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3515 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3516 * requires an auxiliary table look up. abs(base) gives the index into
3517 * a list of such tables which points to the proper aux table. And a
3518 * parallel list gives the length of each corresponding aux table. */
3519 cp_list = aux_tables[-base];
3521 /* Create the string of UTF-8 from the mapped-to code points */
3523 for (i = 0; i < aux_table_lengths[-base]; i++) {
3524 d = uvchr_to_utf8(d, cp_list[i]);
3532 /* Here, there was no mapping defined, which means that the code point maps
3533 * to itself. Return the inputs */
3536 if (p != ustrp) { /* Don't copy onto itself */
3537 Copy(p, ustrp, len, U8);
3548 Perl__inverse_folds(pTHX_ const UV cp, unsigned int * first_folds_to,
3549 const unsigned int ** remaining_folds_to)
3551 /* Returns the count of the number of code points that fold to the input
3552 * 'cp' (besides itself).
3554 * If the return is 0, there is nothing else that folds to it, and
3555 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3557 * If the return is 1, '*first_folds_to' is set to the single code point,
3558 * and '*remaining_folds_to' is set to NULL.
3560 * Otherwise, '*first_folds_to' is set to a code point, and
3561 * '*remaining_fold_to' is set to an array that contains the others. The
3562 * length of this array is the returned count minus 1.
3564 * The reason for this convolution is to avoid having to deal with
3565 * allocating and freeing memory. The lists are already constructed, so
3566 * the return can point to them, but single code points aren't, so would
3567 * need to be constructed if we didn't employ something like this API */
3569 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3570 int base = _Perl_IVCF_invmap[index];
3572 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3574 if (base == 0) { /* No fold */
3575 *first_folds_to = 0;
3576 *remaining_folds_to = NULL;
3580 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3586 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3588 /* The data structure is set up so that the absolute value of 'base' is
3589 * an index into a table of pointers to arrays, with the array
3590 * corresponding to the index being the list of code points that fold
3591 * to 'cp', and the parallel array containing the length of the list
3593 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3594 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1; /* +1 excludes
3597 return IVCF_AUX_TABLE_lengths[-base];
3602 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3603 *first_folds_to = base + cp - invlist_array(PL_utf8_foldclosures)[index];
3604 *remaining_folds_to = NULL;
3609 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3610 U8* const ustrp, STRLEN *lenp)
3612 /* This is called when changing the case of a UTF-8-encoded character above
3613 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3614 * result contains a character that crosses the 255/256 boundary, disallow
3615 * the change, and return the original code point. See L<perlfunc/lc> for
3618 * p points to the original string whose case was changed; assumed
3619 * by this routine to be well-formed
3620 * result the code point of the first character in the changed-case string
3621 * ustrp points to the changed-case string (<result> represents its
3623 * lenp points to the length of <ustrp> */
3625 UV original; /* To store the first code point of <p> */
3627 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3629 assert(UTF8_IS_ABOVE_LATIN1(*p));
3631 /* We know immediately if the first character in the string crosses the
3632 * boundary, so can skip testing */
3635 /* Look at every character in the result; if any cross the
3636 * boundary, the whole thing is disallowed */
3637 U8* s = ustrp + UTF8SKIP(ustrp);
3638 U8* e = ustrp + *lenp;
3640 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3646 /* Here, no characters crossed, result is ok as-is, but we warn. */
3647 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3653 /* Failed, have to return the original */
3654 original = valid_utf8_to_uvchr(p, lenp);
3656 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3657 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3658 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3659 " locale; resolved to \"\\x{%" UVXf "}\".",
3663 Copy(p, ustrp, *lenp, char);
3668 S_check_and_deprecate(pTHX_ const U8 *p,
3670 const unsigned int type, /* See below */
3671 const bool use_locale, /* Is this a 'LC_'
3673 const char * const file,
3674 const unsigned line)
3676 /* This is a temporary function to deprecate the unsafe calls to the case
3677 * changing macros and functions. It keeps all the special stuff in just
3680 * It updates *e with the pointer to the end of the input string. If using
3681 * the old-style macros, *e is NULL on input, and so this function assumes
3682 * the input string is long enough to hold the entire UTF-8 sequence, and
3683 * sets *e accordingly, but it then returns a flag to pass the
3684 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3685 * using the full length if possible.
3687 * It also does the assert that *e > p when *e is not NULL. This should be
3688 * migrated to the callers when this function gets deleted.
3690 * The 'type' parameter is used for the caller to specify which case
3691 * changing function this is called from: */
3693 # define DEPRECATE_TO_UPPER 0
3694 # define DEPRECATE_TO_TITLE 1
3695 # define DEPRECATE_TO_LOWER 2
3696 # define DEPRECATE_TO_FOLD 3
3698 U32 utf8n_flags = 0;
3700 const char * alternative;
3702 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3705 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3706 *e = p + UTF8SKIP(p);
3708 /* For mathoms.c calls, we use the function name we know is stored
3709 * there. It could be part of a larger path */
3710 if (type == DEPRECATE_TO_UPPER) {
3711 name = instr(file, "mathoms.c")
3714 alternative = "toUPPER_utf8_safe";
3716 else if (type == DEPRECATE_TO_TITLE) {
3717 name = instr(file, "mathoms.c")
3720 alternative = "toTITLE_utf8_safe";
3722 else if (type == DEPRECATE_TO_LOWER) {
3723 name = instr(file, "mathoms.c")
3726 alternative = "toLOWER_utf8_safe";
3728 else if (type == DEPRECATE_TO_FOLD) {
3729 name = instr(file, "mathoms.c")
3732 alternative = "toFOLD_utf8_safe";
3734 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3736 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3745 /* The process for changing the case is essentially the same for the four case
3746 * change types, except there are complications for folding. Otherwise the
3747 * difference is only which case to change to. To make sure that they all do
3748 * the same thing, the bodies of the functions are extracted out into the
3749 * following two macros. The functions are written with the same variable
3750 * names, and these are known and used inside these macros. It would be
3751 * better, of course, to have inline functions to do it, but since different
3752 * macros are called, depending on which case is being changed to, this is not
3753 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3754 * function can start with the common start macro, then finish with its special
3755 * handling; while the other three cases can just use the common end macro.
3757 * The algorithm is to use the proper (passed in) macro or function to change
3758 * the case for code points that are below 256. The macro is used if using
3759 * locale rules for the case change; the function if not. If the code point is
3760 * above 255, it is computed from the input UTF-8, and another macro is called
3761 * to do the conversion. If necessary, the output is converted to UTF-8. If
3762 * using a locale, we have to check that the change did not cross the 255/256
3763 * boundary, see check_locale_boundary_crossing() for further details.
3765 * The macros are split with the correct case change for the below-256 case
3766 * stored into 'result', and in the middle of an else clause for the above-255
3767 * case. At that point in the 'else', 'result' is not the final result, but is
3768 * the input code point calculated from the UTF-8. The fold code needs to
3769 * realize all this and take it from there.
3771 * If you read the two macros as sequential, it's easier to understand what's
3773 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3774 L1_func_extra_param) \
3776 if (flags & (locale_flags)) { \
3777 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3778 /* Treat a UTF-8 locale as not being in locale at all */ \
3779 if (IN_UTF8_CTYPE_LOCALE) { \
3780 flags &= ~(locale_flags); \
3784 if (UTF8_IS_INVARIANT(*p)) { \
3785 if (flags & (locale_flags)) { \
3786 result = LC_L1_change_macro(*p); \
3789 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3792 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3793 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3794 if (flags & (locale_flags)) { \
3795 result = LC_L1_change_macro(c); \
3798 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3801 else { /* malformed UTF-8 or ord above 255 */ \
3802 STRLEN len_result; \
3803 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3804 if (len_result == (STRLEN) -1) { \
3805 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3809 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3810 result = change_macro(result, p, ustrp, lenp); \
3812 if (flags & (locale_flags)) { \
3813 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3818 /* Here, used locale rules. Convert back to UTF-8 */ \
3819 if (UTF8_IS_INVARIANT(result)) { \
3820 *ustrp = (U8) result; \
3824 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3825 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3832 =for apidoc to_utf8_upper
3834 Instead use L</toUPPER_utf8_safe>.
3838 /* Not currently externally documented, and subject to change:
3839 * <flags> is set iff iff the rules from the current underlying locale are to
3843 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3848 const char * const file,
3852 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3853 cBOOL(flags), file, line);
3855 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3857 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3858 /* 2nd char of uc(U+DF) is 'S' */
3859 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3860 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3864 =for apidoc to_utf8_title
3866 Instead use L</toTITLE_utf8_safe>.
3870 /* Not currently externally documented, and subject to change:
3871 * <flags> is set iff the rules from the current underlying locale are to be
3872 * used. Since titlecase is not defined in POSIX, for other than a
3873 * UTF-8 locale, uppercase is used instead for code points < 256.
3877 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3882 const char * const file,
3886 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3887 cBOOL(flags), file, line);
3889 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3891 /* 2nd char of ucfirst(U+DF) is 's' */
3892 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3893 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3897 =for apidoc to_utf8_lower
3899 Instead use L</toLOWER_utf8_safe>.
3903 /* Not currently externally documented, and subject to change:
3904 * <flags> is set iff iff the rules from the current underlying locale are to
3909 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3914 const char * const file,
3918 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3919 cBOOL(flags), file, line);
3921 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3923 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3924 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3928 =for apidoc to_utf8_fold
3930 Instead use L</toFOLD_utf8_safe>.
3934 /* Not currently externally documented, and subject to change,
3936 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3937 * locale are to be used.
3938 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3939 * otherwise simple folds
3940 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3945 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3950 const char * const file,
3954 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3955 cBOOL(flags), file, line);
3957 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3959 /* These are mutually exclusive */
3960 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3962 assert(p != ustrp); /* Otherwise overwrites */
3964 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3965 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3967 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3969 if (flags & FOLD_FLAGS_LOCALE) {
3971 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3972 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3973 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3975 /* Special case these two characters, as what normally gets
3976 * returned under locale doesn't work */
3977 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S))
3979 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3980 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3981 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3982 "resolved to \"\\x{17F}\\x{17F}\".");
3987 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T))
3989 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3990 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3991 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3992 "resolved to \"\\x{FB06}\".");
3993 goto return_ligature_st;
3996 #if UNICODE_MAJOR_VERSION == 3 \
3997 && UNICODE_DOT_VERSION == 0 \
3998 && UNICODE_DOT_DOT_VERSION == 1
3999 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
4001 /* And special case this on this Unicode version only, for the same
4002 * reaons the other two are special cased. They would cross the
4003 * 255/256 boundary which is forbidden under /l, and so the code
4004 * wouldn't catch that they are equivalent (which they are only in
4006 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) {
4007 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
4008 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
4009 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
4010 "resolved to \"\\x{0131}\".");
4011 goto return_dotless_i;
4015 return check_locale_boundary_crossing(p, result, ustrp, lenp);
4017 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
4021 /* This is called when changing the case of a UTF-8-encoded
4022 * character above the ASCII range, and the result should not
4023 * contain an ASCII character. */
4025 UV original; /* To store the first code point of <p> */
4027 /* Look at every character in the result; if any cross the
4028 * boundary, the whole thing is disallowed */
4030 U8* e = ustrp + *lenp;
4033 /* Crossed, have to return the original */
4034 original = valid_utf8_to_uvchr(p, lenp);
4036 /* But in these instances, there is an alternative we can
4037 * return that is valid */
4038 if (original == LATIN_SMALL_LETTER_SHARP_S
4039 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
4040 || original == LATIN_CAPITAL_LETTER_SHARP_S
4045 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
4046 goto return_ligature_st;
4048 #if UNICODE_MAJOR_VERSION == 3 \
4049 && UNICODE_DOT_VERSION == 0 \
4050 && UNICODE_DOT_DOT_VERSION == 1
4052 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
4053 goto return_dotless_i;
4056 Copy(p, ustrp, *lenp, char);
4062 /* Here, no characters crossed, result is ok as-is */
4067 /* Here, used locale rules. Convert back to UTF-8 */
4068 if (UTF8_IS_INVARIANT(result)) {
4069 *ustrp = (U8) result;
4073 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
4074 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
4081 /* Certain folds to 'ss' are prohibited by the options, but they do allow
4082 * folds to a string of two of these characters. By returning this
4083 * instead, then, e.g.,
4084 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
4087 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
4088 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
4090 return LATIN_SMALL_LETTER_LONG_S;
4093 /* Two folds to 'st' are prohibited by the options; instead we pick one and
4094 * have the other one fold to it */
4096 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
4097 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
4098 return LATIN_SMALL_LIGATURE_ST;
4100 #if UNICODE_MAJOR_VERSION == 3 \
4101 && UNICODE_DOT_VERSION == 0 \
4102 && UNICODE_DOT_DOT_VERSION == 1
4105 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
4106 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
4107 return LATIN_SMALL_LETTER_DOTLESS_I;
4114 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
4115 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
4116 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
4120 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4121 I32 minbits, I32 none)
4123 PERL_ARGS_ASSERT_SWASH_INIT;
4125 /* Returns a copy of a swash initiated by the called function. This is the
4126 * public interface, and returning a copy prevents others from doing
4127 * mischief on the original */
4129 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
4134 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
4135 I32 minbits, I32 none, SV* invlist,
4139 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
4140 * use the following define */
4142 #define CORE_SWASH_INIT_RETURN(x) \
4143 PL_curpm= old_PL_curpm; \
4146 /* Initialize and return a swash, creating it if necessary. It does this
4147 * by calling utf8_heavy.pl in the general case. The returned value may be
4148 * the swash's inversion list instead if the input parameters allow it.
4149 * Which is returned should be immaterial to callers, as the only
4150 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
4151 * and swash_to_invlist() handle both these transparently.
4153 * This interface should only be used by functions that won't destroy or
4154 * adversely change the swash, as doing so affects all other uses of the
4155 * swash in the program; the general public should use 'Perl_swash_init'
4158 * pkg is the name of the package that <name> should be in.
4159 * name is the name of the swash to find. Typically it is a Unicode
4160 * property name, including user-defined ones
4161 * listsv is a string to initialize the swash with. It must be of the form
4162 * documented as the subroutine return value in
4163 * L<perlunicode/User-Defined Character Properties>
4164 * minbits is the number of bits required to represent each data element.
4165 * It is '1' for binary properties.
4166 * none I (khw) do not understand this one, but it is used only in tr///.
4167 * invlist is an inversion list to initialize the swash with (or NULL)
4168 * flags_p if non-NULL is the address of various input and output flag bits
4169 * to the routine, as follows: ('I' means is input to the routine;
4170 * 'O' means output from the routine. Only flags marked O are
4171 * meaningful on return.)
4172 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
4173 * came from a user-defined property. (I O)
4174 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
4175 * when the swash cannot be located, to simply return NULL. (I)
4176 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
4177 * return of an inversion list instead of a swash hash if this routine
4178 * thinks that would result in faster execution of swash_fetch() later
4181 * Thus there are three possible inputs to find the swash: <name>,
4182 * <listsv>, and <invlist>. At least one must be specified. The result
4183 * will be the union of the specified ones, although <listsv>'s various
4184 * actions can intersect, etc. what <name> gives. To avoid going out to
4185 * disk at all, <invlist> should specify completely what the swash should
4186 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
4188 * <invlist> is only valid for binary properties */
4190 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
4192 SV* retval = &PL_sv_undef;
4193 HV* swash_hv = NULL;
4194 const bool use_invlist= (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST);
4196 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
4197 assert(! invlist || minbits == 1);
4199 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
4200 regex that triggered the swash init and the swash init
4201 perl logic itself. See perl #122747 */
4203 /* If data was passed in to go out to utf8_heavy to find the swash of, do
4205 if (listsv != &PL_sv_undef || strNE(name, "")) {
4207 const size_t pkg_len = strlen(pkg);
4208 const size_t name_len = strlen(name);
4209 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
4213 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
4215 PUSHSTACKi(PERLSI_MAGIC);
4219 /* We might get here via a subroutine signature which uses a utf8
4220 * parameter name, at which point PL_subname will have been set
4221 * but not yet used. */
4222 save_item(PL_subname);
4223 if (PL_parser && PL_parser->error_count)
4224 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
4225 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
4226 if (!method) { /* demand load UTF-8 */
4228 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4229 GvSV(PL_errgv) = NULL;
4230 #ifndef NO_TAINT_SUPPORT
4231 /* It is assumed that callers of this routine are not passing in
4232 * any user derived data. */
4233 /* Need to do this after save_re_context() as it will set
4234 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
4235 * in Perl_magic_get). Even line to create errsv_save can turn on
4237 SAVEBOOL(TAINT_get);
4240 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
4243 /* Not ERRSV, as there is no need to vivify a scalar we are
4244 about to discard. */
4245 SV * const errsv = GvSV(PL_errgv);
4246 if (!SvTRUE(errsv)) {
4247 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4248 SvREFCNT_dec(errsv);
4256 mPUSHp(pkg, pkg_len);
4257 mPUSHp(name, name_len);
4262 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
4263 GvSV(PL_errgv) = NULL;
4264 /* If we already have a pointer to the method, no need to use
4265 * call_method() to repeat the lookup. */
4267 ? call_sv(MUTABLE_SV(method), G_SCALAR)
4268 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
4270 retval = *PL_stack_sp--;
4271 SvREFCNT_inc(retval);
4274 /* Not ERRSV. See above. */
4275 SV * const errsv = GvSV(PL_errgv);
4276 if (!SvTRUE(errsv)) {
4277 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
4278 SvREFCNT_dec(errsv);
4283 if (IN_PERL_COMPILETIME) {
4284 CopHINTS_set(PL_curcop, PL_hints);
4286 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
4287 if (SvPOK(retval)) {
4289 /* If caller wants to handle missing properties, let them */
4290 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
4291 CORE_SWASH_INIT_RETURN(NULL);
4294 "Can't find Unicode property definition \"%" SVf "\"",
4296 NOT_REACHED; /* NOTREACHED */