3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
41 =head1 Unicode Support
42 These are various utility functions for manipulating UTF8-encoded
43 strings. For the uninitiated, this is a method of representing arbitrary
44 Unicode characters as a variable number of bytes, in such a way that
45 characters in the ASCII range are unmodified, and a zero byte never appears
46 within non-zero characters.
51 /* helper for Perl__force_out_malformed_utf8_message(). Like
52 * SAVECOMPILEWARNINGS(), but works with PL_curcop rather than
56 S_restore_cop_warnings(pTHX_ void *p)
58 if (!specialWARN(PL_curcop->cop_warnings))
59 PerlMemShared_free(PL_curcop->cop_warnings);
60 PL_curcop->cop_warnings = (STRLEN*)p;
65 Perl__force_out_malformed_utf8_message(pTHX_
66 const U8 *const p, /* First byte in UTF-8 sequence */
67 const U8 * const e, /* Final byte in sequence (may include
69 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
70 usually 0, or some DISALLOW flags */
71 const bool die_here) /* If TRUE, this function does not return */
73 /* This core-only function is to be called when a malformed UTF-8 character
74 * is found, in order to output the detailed information about the
75 * malformation before dieing. The reason it exists is for the occasions
76 * when such a malformation is fatal, but warnings might be turned off, so
77 * that normally they would not be actually output. This ensures that they
78 * do get output. Because a sequence may be malformed in more than one
79 * way, multiple messages may be generated, so we can't make them fatal, as
80 * that would cause the first one to die.
82 * Instead we pretend -W was passed to perl, then die afterwards. The
83 * flexibility is here to return to the caller so they can finish up and
87 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
93 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
95 /* this is like SAVECOMPILEWARNINGS() except with PL_curcop rather
96 * than PL_compiling */
97 SAVEDESTRUCTOR_X(S_restore_cop_warnings,
98 (void*)PL_curcop->cop_warnings);
99 PL_curcop->cop_warnings = pWARN_ALL;
102 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
107 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
108 " be called only when there are errors found");
112 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
117 S_new_msg_hv(pTHX_ const char * const message, /* The message text */
118 U32 categories, /* Packed warning categories */
119 U32 flag) /* Flag associated with this message */
121 /* Creates, populates, and returns an HV* that describes an error message
122 * for the translators between UTF8 and code point */
124 SV* msg_sv = newSVpv(message, 0);
125 SV* category_sv = newSVuv(categories);
126 SV* flag_bit_sv = newSVuv(flag);
128 HV* msg_hv = newHV();
130 PERL_ARGS_ASSERT_NEW_MSG_HV;
132 (void) hv_stores(msg_hv, "text", msg_sv);
133 (void) hv_stores(msg_hv, "warn_categories", category_sv);
134 (void) hv_stores(msg_hv, "flag_bit", flag_bit_sv);
140 =for apidoc uvoffuni_to_utf8_flags
142 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
143 Instead, B<Almost all code should use L<perlapi/uvchr_to_utf8> or
144 L<perlapi/uvchr_to_utf8_flags>>.
146 This function is like them, but the input is a strict Unicode
147 (as opposed to native) code point. Only in very rare circumstances should code
148 not be using the native code point.
150 For details, see the description for L<perlapi/uvchr_to_utf8_flags>.
156 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
158 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
160 return uvoffuni_to_utf8_flags_msgs(d, uv, flags, NULL);
163 /* All these formats take a single UV code point argument */
164 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
165 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
166 " is not recommended for open interchange";
167 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
168 " may not be portable";
170 #define HANDLE_UNICODE_SURROGATE(uv, flags, msgs) \
172 if (flags & UNICODE_WARN_SURROGATE) { \
173 U32 category = packWARN(WARN_SURROGATE); \
174 const char * format = surrogate_cp_format; \
176 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
178 UNICODE_GOT_SURROGATE); \
181 Perl_ck_warner_d(aTHX_ category, format, uv); \
184 if (flags & UNICODE_DISALLOW_SURROGATE) { \
189 #define HANDLE_UNICODE_NONCHAR(uv, flags, msgs) \
191 if (flags & UNICODE_WARN_NONCHAR) { \
192 U32 category = packWARN(WARN_NONCHAR); \
193 const char * format = nonchar_cp_format; \
195 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv), \
197 UNICODE_GOT_NONCHAR); \
200 Perl_ck_warner_d(aTHX_ category, format, uv); \
203 if (flags & UNICODE_DISALLOW_NONCHAR) { \
208 /* Use shorter names internally in this file */
209 #define SHIFT UTF_ACCUMULATION_SHIFT
211 #define MARK UTF_CONTINUATION_MARK
212 #define MASK UTF_CONTINUATION_MASK
215 =for apidoc uvchr_to_utf8_flags_msgs
217 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
219 Most code should use C<L</uvchr_to_utf8_flags>()> rather than call this directly.
221 This function is for code that wants any warning and/or error messages to be
222 returned to the caller rather than be displayed. All messages that would have
223 been displayed if all lexical warnings are enabled will be returned.
225 It is just like C<L</uvchr_to_utf8_flags>> but it takes an extra parameter
226 placed after all the others, C<msgs>. If this parameter is 0, this function
227 behaves identically to C<L</uvchr_to_utf8_flags>>. Otherwise, C<msgs> should
228 be a pointer to an C<HV *> variable, in which this function creates a new HV to
229 contain any appropriate messages. The hash has three key-value pairs, as
236 The text of the message as a C<SVpv>.
238 =item C<warn_categories>
240 The warning category (or categories) packed into a C<SVuv>.
244 A single flag bit associated with this message, in a C<SVuv>.
245 The bit corresponds to some bit in the C<*errors> return value,
246 such as C<UNICODE_GOT_SURROGATE>.
250 It's important to note that specifying this parameter as non-null will cause
251 any warnings this function would otherwise generate to be suppressed, and
252 instead be placed in C<*msgs>. The caller can check the lexical warnings state
253 (or not) when choosing what to do with the returned messages.
255 The caller, of course, is responsible for freeing any returned HV.
260 /* Undocumented; we don't want people using this. Instead they should use
261 * uvchr_to_utf8_flags_msgs() */
263 Perl_uvoffuni_to_utf8_flags_msgs(pTHX_ U8 *d, UV uv, const UV flags, HV** msgs)
265 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS_MSGS;
271 if (OFFUNI_IS_INVARIANT(uv)) {
272 *d++ = LATIN1_TO_NATIVE(uv);
276 if (uv <= MAX_UTF8_TWO_BYTE) {
277 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
278 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
282 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
283 * below, the 16 is for start bytes E0-EF (which are all the possible ones
284 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
285 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
286 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
287 * 0x800-0xFFFF on ASCII */
288 if (uv < (16 * (1U << (2 * SHIFT)))) {
289 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
290 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
291 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
293 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
294 aren't tested here */
295 /* The most likely code points in this range are below the surrogates.
296 * Do an extra test to quickly exclude those. */
297 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
298 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
299 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
301 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
303 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
304 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
311 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
312 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
313 * happen starting with 4-byte characters on ASCII platforms. We unify the
314 * code for these with EBCDIC, even though some of them require 5-bytes on
315 * those, because khw believes the code saving is worth the very slight
316 * performance hit on these high EBCDIC code points. */
318 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
319 if (UNLIKELY( uv > MAX_LEGAL_CP
320 && ! (flags & UNICODE_ALLOW_ABOVE_IV_MAX)))
322 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv));
324 if ( (flags & UNICODE_WARN_SUPER)
325 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
326 && UNICODE_IS_PERL_EXTENDED(uv)))
328 const char * format = super_cp_format;
329 U32 category = packWARN(WARN_NON_UNICODE);
330 U32 flag = UNICODE_GOT_SUPER;
332 /* Choose the more dire applicable warning */
333 if (UNICODE_IS_PERL_EXTENDED(uv)) {
334 format = PL_extended_cp_format;
335 category = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
336 if (flags & (UNICODE_WARN_PERL_EXTENDED
337 |UNICODE_DISALLOW_PERL_EXTENDED))
339 flag = UNICODE_GOT_PERL_EXTENDED;
344 *msgs = new_msg_hv(Perl_form(aTHX_ format, uv),
347 else if ( ckWARN_d(WARN_NON_UNICODE)
348 || ( (flag & UNICODE_GOT_PERL_EXTENDED)
349 && ckWARN(WARN_PORTABLE)))
351 Perl_warner(aTHX_ category, format, uv);
354 if ( (flags & UNICODE_DISALLOW_SUPER)
355 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
356 && UNICODE_IS_PERL_EXTENDED(uv)))
361 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
362 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
365 /* Test for and handle 4-byte result. In the test immediately below, the
366 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
367 * characters). The 3 is for 3 continuation bytes; these each contribute
368 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
369 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
370 * 0x1_0000-0x1F_FFFF on ASCII */
371 if (uv < (8 * (1U << (3 * SHIFT)))) {
372 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
373 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
374 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
375 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
377 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
378 characters. The end-plane non-characters for EBCDIC were
379 handled just above */
380 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
381 HANDLE_UNICODE_NONCHAR(uv, flags, msgs);
383 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
384 HANDLE_UNICODE_SURROGATE(uv, flags, msgs);
391 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
392 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
393 * format. The unrolled version above turns out to not save all that much
394 * time, and at these high code points (well above the legal Unicode range
395 * on ASCII platforms, and well above anything in common use in EBCDIC),
396 * khw believes that less code outweighs slight performance gains. */
399 STRLEN len = OFFUNISKIP(uv);
402 *p-- = I8_TO_NATIVE_UTF8((uv & MASK) | MARK);
405 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
411 =for apidoc uvchr_to_utf8
413 Adds the UTF-8 representation of the native code point C<uv> to the end
414 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
415 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
416 the byte after the end of the new character. In other words,
418 d = uvchr_to_utf8(d, uv);
420 is the recommended wide native character-aware way of saying
424 This function accepts any code point from 0..C<IV_MAX> as input.
425 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
427 It is possible to forbid or warn on non-Unicode code points, or those that may
428 be problematic by using L</uvchr_to_utf8_flags>.
433 /* This is also a macro */
434 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
437 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
439 return uvchr_to_utf8(d, uv);
443 =for apidoc uvchr_to_utf8_flags
445 Adds the UTF-8 representation of the native code point C<uv> to the end
446 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
447 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
448 the byte after the end of the new character. In other words,
450 d = uvchr_to_utf8_flags(d, uv, flags);
454 d = uvchr_to_utf8_flags(d, uv, 0);
456 This is the Unicode-aware way of saying
460 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
461 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
463 Specifying C<flags> can further restrict what is allowed and not warned on, as
466 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
467 the function will raise a warning, provided UTF8 warnings are enabled. If
468 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
469 NULL. If both flags are set, the function will both warn and return NULL.
471 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
472 affect how the function handles a Unicode non-character.
474 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
475 affect the handling of code points that are above the Unicode maximum of
476 0x10FFFF. Languages other than Perl may not be able to accept files that
479 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
480 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
481 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
482 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
483 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
484 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
485 above-Unicode and surrogate flags, but not the non-character ones, as
487 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
488 See L<perlunicode/Noncharacter code points>.
490 Extremely high code points were never specified in any standard, and require an
491 extension to UTF-8 to express, which Perl does. It is likely that programs
492 written in something other than Perl would not be able to read files that
493 contain these; nor would Perl understand files written by something that uses a
494 different extension. For these reasons, there is a separate set of flags that
495 can warn and/or disallow these extremely high code points, even if other
496 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
497 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
498 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
499 treat all above-Unicode code points, including these, as malformations. (Note
500 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
501 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
503 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
504 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
505 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
506 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because on EBCDIC
507 platforms,these flags can apply to code points that actually do fit in 31 bits.
508 The new names accurately describe the situation in all cases.
513 /* This is also a macro */
514 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
517 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
519 return uvchr_to_utf8_flags(d, uv, flags);
525 S_is_utf8_cp_above_31_bits(const U8 * const s,
527 const bool consider_overlongs)
529 /* Returns TRUE if the first code point represented by the Perl-extended-
530 * UTF-8-encoded string starting at 's', and looking no further than 'e -
531 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
533 * The function handles the case where the input bytes do not include all
534 * the ones necessary to represent a full character. That is, they may be
535 * the intial bytes of the representation of a code point, but possibly
536 * the final ones necessary for the complete representation may be beyond
539 * The function also can handle the case where the input is an overlong
540 * sequence. If 'consider_overlongs' is 0, the function assumes the
541 * input is not overlong, without checking, and will return based on that
542 * assumption. If this parameter is 1, the function will go to the trouble
543 * of figuring out if it actually evaluates to above or below 31 bits.
545 * The sequence is otherwise assumed to be well-formed, without checking.
548 const STRLEN len = e - s;
551 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
553 assert(! UTF8_IS_INVARIANT(*s) && e > s);
557 PERL_UNUSED_ARG(consider_overlongs);
559 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
560 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
561 * also be the start byte for a 31-bit code point; we need at least 2
562 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
563 * the start byte for an overlong sequence, but for 30-bit or smaller code
564 * points, so we don't have to worry about overlongs on EBCDIC.) */
575 /* On ASCII, FE and FF are the only start bytes that can evaluate to
576 * needing more than 31 bits. */
577 if (LIKELY(*s < 0xFE)) {
581 /* What we have left are FE and FF. Both of these require more than 31
582 * bits unless they are for overlongs. */
583 if (! consider_overlongs) {
587 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
588 * above 31 bits. But we need more than one byte to discern this, so if
589 * passed just the start byte, it could be an overlong evaluating to
595 /* Having excluded len==1, and knowing that FE and FF are both valid start
596 * bytes, we can call the function below to see if the sequence is
597 * overlong. (We don't need the full generality of the called function,
598 * but for these huge code points, speed shouldn't be a consideration, and
599 * the compiler does have enough information, since it's static to this
600 * file, to optimize to just the needed parts.) */
601 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
603 /* If it isn't overlong, more than 31 bits are required. */
604 if (is_overlong == 0) {
608 /* If it is indeterminate if it is overlong, return that */
609 if (is_overlong < 0) {
613 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
614 * the max it can be is 2**31 - 1 */
621 /* Here, ASCII and EBCDIC rejoin:
622 * On ASCII: We have an overlong sequence starting with FF
623 * On EBCDIC: We have a sequence starting with FE. */
625 { /* For C89, use a block so the declaration can be close to its use */
629 /* U+7FFFFFFF (2 ** 31 - 1)
630 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
631 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
632 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
633 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
634 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
635 * U+80000000 (2 ** 31):
636 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
637 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
638 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
639 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
641 * and since we know that *s = \xfe, any continuation sequcence
642 * following it that is gt the below is above 31 bits
643 [0] [1] [2] [3] [4] [5] [6] */
644 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
648 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
649 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
650 * FF overlong for U+80000000 (2 ** 31):
651 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
652 * and since we know that *s = \xff, any continuation sequcence
653 * following it that is gt the below is above 30 bits
654 [0] [1] [2] [3] [4] [5] [6] */
655 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
659 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
660 const STRLEN cmp_len = MIN(conts_len, len - 1);
662 /* Now compare the continuation bytes in s with the ones we have
663 * compiled in that are for the largest 30 bit code point. If we have
664 * enough bytes available to determine the answer, or the bytes we do
665 * have differ from them, we can compare the two to get a definitive
666 * answer (Note that in UTF-EBCDIC, the two lowest possible
667 * continuation bytes are \x41 and \x42.) */
668 if (cmp_len >= conts_len || memNE(s + 1,
669 conts_for_highest_30_bit,
672 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
675 /* Here, all the bytes we have are the same as the highest 30-bit code
676 * point, but we are missing so many bytes that we can't make the
684 PERL_STATIC_INLINE int
685 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
687 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
688 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
689 * it isn't, and -1 if there isn't enough information to tell. This last
690 * return value can happen if the sequence is incomplete, missing some
691 * trailing bytes that would form a complete character. If there are
692 * enough bytes to make a definitive decision, this function does so.
693 * Usually 2 bytes sufficient.
695 * Overlongs can occur whenever the number of continuation bytes changes.
696 * That means whenever the number of leading 1 bits in a start byte
697 * increases from the next lower start byte. That happens for start bytes
698 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
699 * start bytes have already been excluded, so don't need to be tested here;
700 * ASCII platforms: C0, C1
701 * EBCDIC platforms C0, C1, C2, C3, C4, E0
704 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
705 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
707 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
708 assert(len > 1 && UTF8_IS_START(*s));
710 /* Each platform has overlongs after the start bytes given above (expressed
711 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
712 * the logic is the same, except the E0 overlong has already been excluded
713 * on EBCDIC platforms. The values below were found by manually
714 * inspecting the UTF-8 patterns. See the tables in utf8.h and
718 # define F0_ABOVE_OVERLONG 0xB0
719 # define F8_ABOVE_OVERLONG 0xA8
720 # define FC_ABOVE_OVERLONG 0xA4
721 # define FE_ABOVE_OVERLONG 0xA2
722 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
726 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
730 # define F0_ABOVE_OVERLONG 0x90
731 # define F8_ABOVE_OVERLONG 0x88
732 # define FC_ABOVE_OVERLONG 0x84
733 # define FE_ABOVE_OVERLONG 0x82
734 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
738 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
739 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
740 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
741 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
746 /* Check for the FF overlong */
747 return isFF_OVERLONG(s, len);
750 PERL_STATIC_INLINE int
751 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
753 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
754 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
755 * it isn't, and -1 if there isn't enough information to tell. This last
756 * return value can happen if the sequence is incomplete, missing some
757 * trailing bytes that would form a complete character. If there are
758 * enough bytes to make a definitive decision, this function does so. */
760 PERL_ARGS_ASSERT_ISFF_OVERLONG;
762 /* To be an FF overlong, all the available bytes must match */
763 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
764 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
769 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
770 * be there; what comes after them doesn't matter. See tables in utf8.h,
772 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
776 /* The missing bytes could cause the result to go one way or the other, so
777 * the result is indeterminate */
781 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
782 # ifdef EBCDIC /* Actually is I8 */
783 # define HIGHEST_REPRESENTABLE_UTF8 \
784 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
786 # define HIGHEST_REPRESENTABLE_UTF8 \
787 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
791 PERL_STATIC_INLINE int
792 S_does_utf8_overflow(const U8 * const s,
794 const bool consider_overlongs)
796 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
797 * 'e' - 1 would overflow an IV on this platform; that is if it represents
798 * a code point larger than the highest representable code point. It
799 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
800 * enough information to tell. This last return value can happen if the
801 * sequence is incomplete, missing some trailing bytes that would form a
802 * complete character. If there are enough bytes to make a definitive
803 * decision, this function does so.
805 * If 'consider_overlongs' is TRUE, the function checks for the possibility
806 * that the sequence is an overlong that doesn't overflow. Otherwise, it
807 * assumes the sequence is not an overlong. This can give different
808 * results only on ASCII 32-bit platforms.
810 * (For ASCII platforms, we could use memcmp() because we don't have to
811 * convert each byte to I8, but it's very rare input indeed that would
812 * approach overflow, so the loop below will likely only get executed once.)
814 * 'e' - 1 must not be beyond a full character. */
817 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
818 assert(s <= e && s + UTF8SKIP(s) >= e);
820 #if ! defined(UV_IS_QUAD)
822 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
826 PERL_UNUSED_ARG(consider_overlongs);
829 const STRLEN len = e - s;
831 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
833 for (x = s; x < e; x++, y++) {
835 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
839 /* If this byte is larger than the corresponding highest UTF-8
840 * byte, the sequence overflow; otherwise the byte is less than,
841 * and so the sequence doesn't overflow */
842 return NATIVE_UTF8_TO_I8(*x) > *y;
846 /* Got to the end and all bytes are the same. If the input is a whole
847 * character, it doesn't overflow. And if it is a partial character,
848 * there's not enough information to tell */
849 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
862 /* This is the portions of the above function that deal with UV_MAX instead of
863 * IV_MAX. They are left here in case we want to combine them so that internal
864 * uses can have larger code points. The only logic difference is that the
865 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
869 /* Anything larger than this will overflow the word if it were converted into a UV */
870 #if defined(UV_IS_QUAD)
871 # ifdef EBCDIC /* Actually is I8 */
872 # define HIGHEST_REPRESENTABLE_UTF8 \
873 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
875 # define HIGHEST_REPRESENTABLE_UTF8 \
876 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
880 # define HIGHEST_REPRESENTABLE_UTF8 \
881 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
883 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
887 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
889 /* On 32 bit ASCII machines, many overlongs that start with FF don't
891 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
893 /* To be such an overlong, the first bytes of 's' must match
894 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
895 * don't have any additional bytes available, the sequence, when
896 * completed might or might not fit in 32 bits. But if we have that
897 * next byte, we can tell for sure. If it is <= 0x83, then it does
899 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
903 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
906 /* Starting with the #else, the rest of the function is identical except
907 * 1. we need to move the 'len' declaration to be global to the function
908 * 2. the endif move to just after the UNUSED_ARG.
909 * An empty endif is given just below to satisfy the preprocessor
915 #undef F0_ABOVE_OVERLONG
916 #undef F8_ABOVE_OVERLONG
917 #undef FC_ABOVE_OVERLONG
918 #undef FE_ABOVE_OVERLONG
919 #undef FF_OVERLONG_PREFIX
922 Perl_is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
927 /* A helper function that should not be called directly.
929 * This function returns non-zero if the string beginning at 's' and
930 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
931 * code point; otherwise it returns 0. The examination stops after the
932 * first code point in 's' is validated, not looking at the rest of the
933 * input. If 'e' is such that there are not enough bytes to represent a
934 * complete code point, this function will return non-zero anyway, if the
935 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
936 * excluded by 'flags'.
938 * A non-zero return gives the number of bytes required to represent the
939 * code point. Be aware that if the input is for a partial character, the
940 * return will be larger than 'e - s'.
942 * This function assumes that the code point represented is UTF-8 variant.
943 * The caller should have excluded the possibility of it being invariant
944 * before calling this function.
946 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
947 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
948 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
949 * disallowed by the flags. If the input is only for a partial character,
950 * the function will return non-zero if there is any sequence of
951 * well-formed UTF-8 that, when appended to the input sequence, could
952 * result in an allowed code point; otherwise it returns 0. Non characters
953 * cannot be determined based on partial character input. But many of the
954 * other excluded types can be determined with just the first one or two
959 PERL_ARGS_ASSERT_IS_UTF8_CHAR_HELPER;
961 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
962 |UTF8_DISALLOW_PERL_EXTENDED)));
963 assert(! UTF8_IS_INVARIANT(*s));
965 /* A variant char must begin with a start byte */
966 if (UNLIKELY(! UTF8_IS_START(*s))) {
970 /* Examine a maximum of a single whole code point */
971 if (e - s > UTF8SKIP(s)) {
977 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
978 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
980 /* Here, we are disallowing some set of largish code points, and the
981 * first byte indicates the sequence is for a code point that could be
982 * in the excluded set. We generally don't have to look beyond this or
983 * the second byte to see if the sequence is actually for one of the
984 * excluded classes. The code below is derived from this table:
986 * UTF-8 UTF-EBCDIC I8
987 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
988 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
989 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
991 * Keep in mind that legal continuation bytes range between \x80..\xBF
992 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
993 * continuation bytes. Hence, we don't have to test the upper edge
994 * because if any of those is encountered, the sequence is malformed,
995 * and would fail elsewhere in this function.
997 * The code here likewise assumes that there aren't other
998 * malformations; again the function should fail elsewhere because of
999 * these. For example, an overlong beginning with FC doesn't actually
1000 * have to be a super; it could actually represent a small code point,
1001 * even U+0000. But, since overlongs (and other malformations) are
1002 * illegal, the function should return FALSE in either case.
1005 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
1006 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
1007 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
1009 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
1011 && ((s1) & 0xFE ) == 0xB6)
1012 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
1014 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
1015 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
1016 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
1017 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
1020 if ( (flags & UTF8_DISALLOW_SUPER)
1021 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1023 return 0; /* Above Unicode */
1026 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
1027 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
1033 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
1035 if ( (flags & UTF8_DISALLOW_SUPER)
1036 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
1038 return 0; /* Above Unicode */
1041 if ( (flags & UTF8_DISALLOW_SURROGATE)
1042 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
1044 return 0; /* Surrogate */
1047 if ( (flags & UTF8_DISALLOW_NONCHAR)
1048 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
1050 return 0; /* Noncharacter code point */
1055 /* Make sure that all that follows are continuation bytes */
1056 for (x = s + 1; x < e; x++) {
1057 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
1062 /* Here is syntactically valid. Next, make sure this isn't the start of an
1064 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
1068 /* And finally, that the code point represented fits in a word on this
1070 if (0 < does_utf8_overflow(s, e,
1071 0 /* Don't consider overlongs */
1081 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
1083 /* Returns a mortalized C string that is a displayable copy of the 'len'
1084 * bytes starting at 'start'. 'format' gives how to display each byte.
1085 * Currently, there are only two formats, so it is currently a bool:
1087 * 1 ab (that is a space between two hex digit bytes)
1090 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
1092 const U8 * s = start;
1093 const U8 * const e = start + len;
1097 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
1099 Newx(output, output_len, char);
1103 for (s = start; s < e; s++) {
1104 const unsigned high_nibble = (*s & 0xF0) >> 4;
1105 const unsigned low_nibble = (*s & 0x0F);
1117 if (high_nibble < 10) {
1118 *d++ = high_nibble + '0';
1121 *d++ = high_nibble - 10 + 'a';
1124 if (low_nibble < 10) {
1125 *d++ = low_nibble + '0';
1128 *d++ = low_nibble - 10 + 'a';
1136 PERL_STATIC_INLINE char *
1137 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1139 /* Max number of bytes to print */
1142 /* Which one is the non-continuation */
1143 const STRLEN non_cont_byte_pos,
1145 /* How many bytes should there be? */
1146 const STRLEN expect_len)
1148 /* Return the malformation warning text for an unexpected continuation
1151 const char * const where = (non_cont_byte_pos == 1)
1153 : Perl_form(aTHX_ "%d bytes",
1154 (int) non_cont_byte_pos);
1155 const U8 * x = s + non_cont_byte_pos;
1156 const U8 * e = s + print_len;
1158 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1160 /* We don't need to pass this parameter, but since it has already been
1161 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1162 assert(expect_len == UTF8SKIP(s));
1164 /* As a defensive coding measure, don't output anything past a NUL. Such
1165 * bytes shouldn't be in the middle of a malformation, and could mark the
1166 * end of the allocated string, and what comes after is undefined */
1167 for (; x < e; x++) {
1169 x++; /* Output this particular NUL */
1174 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1175 " %s after start byte 0x%02x; need %d bytes, got %d)",
1177 _byte_dump_string(s, x - s, 0),
1178 *(s + non_cont_byte_pos),
1182 (int) non_cont_byte_pos);
1187 =for apidoc utf8n_to_uvchr
1189 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1190 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1193 Bottom level UTF-8 decode routine.
1194 Returns the native code point value of the first character in the string C<s>,
1195 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1196 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1197 the length, in bytes, of that character.
1199 The value of C<flags> determines the behavior when C<s> does not point to a
1200 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1201 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1202 is the next possible position in C<s> that could begin a non-malformed
1203 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1204 is raised. Some UTF-8 input sequences may contain multiple malformations.
1205 This function tries to find every possible one in each call, so multiple
1206 warnings can be raised for the same sequence.
1208 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1209 individual types of malformations, such as the sequence being overlong (that
1210 is, when there is a shorter sequence that can express the same code point;
1211 overlong sequences are expressly forbidden in the UTF-8 standard due to
1212 potential security issues). Another malformation example is the first byte of
1213 a character not being a legal first byte. See F<utf8.h> for the list of such
1214 flags. Even if allowed, this function generally returns the Unicode
1215 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1216 F<utf8.h> to override this behavior for the overlong malformations, but don't
1217 do that except for very specialized purposes.
1219 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1220 flags) malformation is found. If this flag is set, the routine assumes that
1221 the caller will raise a warning, and this function will silently just set
1222 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1224 Note that this API requires disambiguation between successful decoding a C<NUL>
1225 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1226 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1227 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1228 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1229 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1231 Certain code points are considered problematic. These are Unicode surrogates,
1232 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1233 By default these are considered regular code points, but certain situations
1234 warrant special handling for them, which can be specified using the C<flags>
1235 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1236 three classes are treated as malformations and handled as such. The flags
1237 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1238 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1239 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1240 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1241 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1243 L<Unicode Corrigendum #9|https://www.unicode.org/versions/corrigendum9.html>.
1244 The difference between traditional strictness and C9 strictness is that the
1245 latter does not forbid non-character code points. (They are still discouraged,
1246 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1248 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1249 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1250 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1251 raised for their respective categories, but otherwise the code points are
1252 considered valid (not malformations). To get a category to both be treated as
1253 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1254 (But note that warnings are not raised if lexically disabled nor if
1255 C<UTF8_CHECK_ONLY> is also specified.)
1257 Extremely high code points were never specified in any standard, and require an
1258 extension to UTF-8 to express, which Perl does. It is likely that programs
1259 written in something other than Perl would not be able to read files that
1260 contain these; nor would Perl understand files written by something that uses a
1261 different extension. For these reasons, there is a separate set of flags that
1262 can warn and/or disallow these extremely high code points, even if other
1263 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1264 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1265 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1266 above-Unicode code points, including these, as malformations.
1267 (Note that the Unicode standard considers anything above 0x10FFFF to be
1268 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1271 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1272 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1273 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1274 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1275 can apply to code points that actually do fit in 31 bits. This happens on
1276 EBCDIC platforms, and sometimes when the L<overlong
1277 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1278 describe the situation in all cases.
1281 All other code points corresponding to Unicode characters, including private
1282 use and those yet to be assigned, are never considered malformed and never
1285 =for apidoc Amnh||UTF8_CHECK_ONLY
1286 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_INTERCHANGE
1287 =for apidoc Amnh||UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE
1288 =for apidoc Amnh||UTF8_DISALLOW_SURROGATE
1289 =for apidoc Amnh||UTF8_DISALLOW_NONCHAR
1290 =for apidoc Amnh||UTF8_DISALLOW_SUPER
1291 =for apidoc Amnh||UTF8_WARN_ILLEGAL_INTERCHANGE
1292 =for apidoc Amnh||UTF8_WARN_ILLEGAL_C9_INTERCHANGE
1293 =for apidoc Amnh||UTF8_WARN_SURROGATE
1294 =for apidoc Amnh||UTF8_WARN_NONCHAR
1295 =for apidoc Amnh||UTF8_WARN_SUPER
1296 =for apidoc Amnh||UTF8_WARN_PERL_EXTENDED
1297 =for apidoc Amnh||UTF8_DISALLOW_PERL_EXTENDED
1301 Also implemented as a macro in utf8.h
1305 Perl_utf8n_to_uvchr(const U8 *s,
1310 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1312 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1317 =for apidoc utf8n_to_uvchr_error
1319 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1320 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1323 This function is for code that needs to know what the precise malformation(s)
1324 are when an error is found. If you also need to know the generated warning
1325 messages, use L</utf8n_to_uvchr_msgs>() instead.
1327 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1328 all the others, C<errors>. If this parameter is 0, this function behaves
1329 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1330 to a C<U32> variable, which this function sets to indicate any errors found.
1331 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1332 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1333 of these bits will be set if a malformation is found, even if the input
1334 C<flags> parameter indicates that the given malformation is allowed; those
1335 exceptions are noted:
1339 =item C<UTF8_GOT_PERL_EXTENDED>
1341 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1342 set only if the input C<flags> parameter contains either the
1343 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1345 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1346 and so some extension must be used to express them. Perl uses a natural
1347 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1348 extension to represent even higher ones, so that any code point that fits in a
1349 64-bit word can be represented. Text using these extensions is not likely to
1350 be portable to non-Perl code. We lump both of these extensions together and
1351 refer to them as Perl extended UTF-8. There exist other extensions that people
1352 have invented, incompatible with Perl's.
1354 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1355 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1356 than on ASCII. Prior to that, code points 2**31 and higher were simply
1357 unrepresentable, and a different, incompatible method was used to represent
1358 code points between 2**30 and 2**31 - 1.
1360 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1361 Perl extended UTF-8 is used.
1363 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1364 may use for backward compatibility. That name is misleading, as this flag may
1365 be set when the code point actually does fit in 31 bits. This happens on
1366 EBCDIC platforms, and sometimes when the L<overlong
1367 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1368 describes the situation in all cases.
1370 =item C<UTF8_GOT_CONTINUATION>
1372 The input sequence was malformed in that the first byte was a a UTF-8
1375 =item C<UTF8_GOT_EMPTY>
1377 The input C<curlen> parameter was 0.
1379 =item C<UTF8_GOT_LONG>
1381 The input sequence was malformed in that there is some other sequence that
1382 evaluates to the same code point, but that sequence is shorter than this one.
1384 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1385 it was discovered that this created security issues.
1387 =item C<UTF8_GOT_NONCHAR>
1389 The code point represented by the input UTF-8 sequence is for a Unicode
1390 non-character code point.
1391 This bit is set only if the input C<flags> parameter contains either the
1392 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1394 =item C<UTF8_GOT_NON_CONTINUATION>
1396 The input sequence was malformed in that a non-continuation type byte was found
1397 in a position where only a continuation type one should be. See also
1398 L</C<UTF8_GOT_SHORT>>.
1400 =item C<UTF8_GOT_OVERFLOW>
1402 The input sequence was malformed in that it is for a code point that is not
1403 representable in the number of bits available in an IV on the current platform.
1405 =item C<UTF8_GOT_SHORT>
1407 The input sequence was malformed in that C<curlen> is smaller than required for
1408 a complete sequence. In other words, the input is for a partial character
1412 C<UTF8_GOT_SHORT> and C<UTF8_GOT_NON_CONTINUATION> both indicate a too short
1413 sequence. The difference is that C<UTF8_GOT_NON_CONTINUATION> indicates always
1414 that there is an error, while C<UTF8_GOT_SHORT> means that an incomplete
1415 sequence was looked at. If no other flags are present, it means that the
1416 sequence was valid as far as it went. Depending on the application, this could
1417 mean one of three things:
1423 The C<curlen> length parameter passed in was too small, and the function was
1424 prevented from examining all the necessary bytes.
1428 The buffer being looked at is based on reading data, and the data received so
1429 far stopped in the middle of a character, so that the next read will
1430 read the remainder of this character. (It is up to the caller to deal with the
1431 split bytes somehow.)
1435 This is a real error, and the partial sequence is all we're going to get.
1439 =item C<UTF8_GOT_SUPER>
1441 The input sequence was malformed in that it is for a non-Unicode code point;
1442 that is, one above the legal Unicode maximum.
1443 This bit is set only if the input C<flags> parameter contains either the
1444 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1446 =item C<UTF8_GOT_SURROGATE>
1448 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1450 This bit is set only if the input C<flags> parameter contains either the
1451 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1455 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1456 flag to suppress any warnings, and then examine the C<*errors> return.
1460 Also implemented as a macro in utf8.h
1464 Perl_utf8n_to_uvchr_error(const U8 *s,
1470 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1472 return utf8n_to_uvchr_msgs(s, curlen, retlen, flags, errors, NULL);
1477 =for apidoc utf8n_to_uvchr_msgs
1479 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1480 Most code should use L</utf8_to_uvchr_buf>() rather than call this
1483 This function is for code that needs to know what the precise malformation(s)
1484 are when an error is found, and wants the corresponding warning and/or error
1485 messages to be returned to the caller rather than be displayed. All messages
1486 that would have been displayed if all lexcial warnings are enabled will be
1489 It is just like C<L</utf8n_to_uvchr_error>> but it takes an extra parameter
1490 placed after all the others, C<msgs>. If this parameter is 0, this function
1491 behaves identically to C<L</utf8n_to_uvchr_error>>. Otherwise, C<msgs> should
1492 be a pointer to an C<AV *> variable, in which this function creates a new AV to
1493 contain any appropriate messages. The elements of the array are ordered so
1494 that the first message that would have been displayed is in the 0th element,
1495 and so on. Each element is a hash with three key-value pairs, as follows:
1501 The text of the message as a C<SVpv>.
1503 =item C<warn_categories>
1505 The warning category (or categories) packed into a C<SVuv>.
1509 A single flag bit associated with this message, in a C<SVuv>.
1510 The bit corresponds to some bit in the C<*errors> return value,
1511 such as C<UTF8_GOT_LONG>.
1515 It's important to note that specifying this parameter as non-null will cause
1516 any warnings this function would otherwise generate to be suppressed, and
1517 instead be placed in C<*msgs>. The caller can check the lexical warnings state
1518 (or not) when choosing what to do with the returned messages.
1520 If the flag C<UTF8_CHECK_ONLY> is passed, no warnings are generated, and hence
1523 The caller, of course, is responsible for freeing any returned AV.
1529 Perl__utf8n_to_uvchr_msgs_helper(const U8 *s,
1536 const U8 * const s0 = s;
1537 const U8 * send = s0 + curlen;
1538 U32 possible_problems; /* A bit is set here for each potential problem
1539 found as we go along */
1541 STRLEN expectlen; /* How long should this sequence be? */
1542 STRLEN avail_len; /* When input is too short, gives what that is */
1543 U32 discard_errors; /* Used to save branches when 'errors' is NULL; this
1544 gets set and discarded */
1546 /* The below are used only if there is both an overlong malformation and a
1547 * too short one. Otherwise the first two are set to 's0' and 'send', and
1548 * the third not used at all */
1550 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1551 routine; see [perl #130921] */
1555 PERL_ARGS_ASSERT__UTF8N_TO_UVCHR_MSGS_HELPER;
1557 /* Here, is one of: a) malformed; b) a problematic code point (surrogate,
1558 * non-unicode, or nonchar); or c) on ASCII platforms, one of the Hangul
1559 * syllables that the dfa doesn't properly handle. Quickly dispose of the
1564 /* Each of the affected Hanguls starts with \xED */
1566 if (is_HANGUL_ED_utf8_safe(s0, send)) {
1577 return ((0xED & UTF_START_MASK(3)) << (2 * UTF_ACCUMULATION_SHIFT))
1578 | ((s0[1] & UTF_CONTINUATION_MASK) << UTF_ACCUMULATION_SHIFT)
1579 | (s0[2] & UTF_CONTINUATION_MASK);
1584 /* In conjunction with the exhaustive tests that can be enabled in
1585 * APItest/t/utf8_warn_base.pl, this can make sure the dfa does precisely
1586 * what it is intended to do, and that no flaws in it are masked by
1587 * dropping down and executing the code below
1588 assert(! isUTF8_CHAR(s0, send)
1589 || UTF8_IS_SURROGATE(s0, send)
1590 || UTF8_IS_SUPER(s0, send)
1591 || UTF8_IS_NONCHAR(s0,send));
1596 possible_problems = 0;
1600 adjusted_s0 = (U8 *) s0;
1607 errors = &discard_errors;
1610 /* The order of malformation tests here is important. We should consume as
1611 * few bytes as possible in order to not skip any valid character. This is
1612 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1613 * https://unicode.org/reports/tr36 for more discussion as to why. For
1614 * example, once we've done a UTF8SKIP, we can tell the expected number of
1615 * bytes, and could fail right off the bat if the input parameters indicate
1616 * that there are too few available. But it could be that just that first
1617 * byte is garbled, and the intended character occupies fewer bytes. If we
1618 * blindly assumed that the first byte is correct, and skipped based on
1619 * that number, we could skip over a valid input character. So instead, we
1620 * always examine the sequence byte-by-byte.
1622 * We also should not consume too few bytes, otherwise someone could inject
1623 * things. For example, an input could be deliberately designed to
1624 * overflow, and if this code bailed out immediately upon discovering that,
1625 * returning to the caller C<*retlen> pointing to the very next byte (one
1626 * which is actually part of of the overflowing sequence), that could look
1627 * legitimate to the caller, which could discard the initial partial
1628 * sequence and process the rest, inappropriately.
1630 * Some possible input sequences are malformed in more than one way. This
1631 * function goes to lengths to try to find all of them. This is necessary
1632 * for correctness, as the inputs may allow one malformation but not
1633 * another, and if we abandon searching for others after finding the
1634 * allowed one, we could allow in something that shouldn't have been.
1637 if (UNLIKELY(curlen == 0)) {
1638 possible_problems |= UTF8_GOT_EMPTY;
1640 uv = UNICODE_REPLACEMENT;
1641 goto ready_to_handle_errors;
1644 expectlen = UTF8SKIP(s);
1646 /* A well-formed UTF-8 character, as the vast majority of calls to this
1647 * function will be for, has this expected length. For efficiency, set
1648 * things up here to return it. It will be overriden only in those rare
1649 * cases where a malformation is found */
1651 *retlen = expectlen;
1654 /* A continuation character can't start a valid sequence */
1655 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1656 possible_problems |= UTF8_GOT_CONTINUATION;
1658 uv = UNICODE_REPLACEMENT;
1659 goto ready_to_handle_errors;
1662 /* Here is not a continuation byte, nor an invariant. The only thing left
1663 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1664 * because it excludes start bytes like \xC0 that always lead to
1667 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1668 * that indicate the number of bytes in the character's whole UTF-8
1669 * sequence, leaving just the bits that are part of the value. */
1670 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1672 /* Setup the loop end point, making sure to not look past the end of the
1673 * input string, and flag it as too short if the size isn't big enough. */
1674 if (UNLIKELY(curlen < expectlen)) {
1675 possible_problems |= UTF8_GOT_SHORT;
1679 send = (U8*) s0 + expectlen;
1682 /* Now, loop through the remaining bytes in the character's sequence,
1683 * accumulating each into the working value as we go. */
1684 for (s = s0 + 1; s < send; s++) {
1685 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1686 uv = UTF8_ACCUMULATE(uv, *s);
1690 /* Here, found a non-continuation before processing all expected bytes.
1691 * This byte indicates the beginning of a new character, so quit, even
1692 * if allowing this malformation. */
1693 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1695 } /* End of loop through the character's bytes */
1697 /* Save how many bytes were actually in the character */
1700 /* Note that there are two types of too-short malformation. One is when
1701 * there is actual wrong data before the normal termination of the
1702 * sequence. The other is that the sequence wasn't complete before the end
1703 * of the data we are allowed to look at, based on the input 'curlen'.
1704 * This means that we were passed data for a partial character, but it is
1705 * valid as far as we saw. The other is definitely invalid. This
1706 * distinction could be important to a caller, so the two types are kept
1709 * A convenience macro that matches either of the too-short conditions. */
1710 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1712 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1714 uv = UNICODE_REPLACEMENT;
1717 /* Check for overflow. The algorithm requires us to not look past the end
1718 * of the current character, even if partial, so the upper limit is 's' */
1719 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1720 1 /* Do consider overlongs */
1723 possible_problems |= UTF8_GOT_OVERFLOW;
1724 uv = UNICODE_REPLACEMENT;
1727 /* Check for overlong. If no problems so far, 'uv' is the correct code
1728 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1729 * we must look at the UTF-8 byte sequence itself to see if it is for an
1731 if ( ( LIKELY(! possible_problems)
1732 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1733 || ( UNLIKELY(possible_problems)
1734 && ( UNLIKELY(! UTF8_IS_START(*s0))
1736 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1739 possible_problems |= UTF8_GOT_LONG;
1741 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1743 /* The calculation in the 'true' branch of this 'if'
1744 * below won't work if overflows, and isn't needed
1745 * anyway. Further below we handle all overflow
1747 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1749 UV min_uv = uv_so_far;
1752 /* Here, the input is both overlong and is missing some trailing
1753 * bytes. There is no single code point it could be for, but there
1754 * may be enough information present to determine if what we have
1755 * so far is for an unallowed code point, such as for a surrogate.
1756 * The code further below has the intelligence to determine this,
1757 * but just for non-overlong UTF-8 sequences. What we do here is
1758 * calculate the smallest code point the input could represent if
1759 * there were no too short malformation. Then we compute and save
1760 * the UTF-8 for that, which is what the code below looks at
1761 * instead of the raw input. It turns out that the smallest such
1762 * code point is all we need. */
1763 for (i = curlen; i < expectlen; i++) {
1764 min_uv = UTF8_ACCUMULATE(min_uv,
1765 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1768 adjusted_s0 = temp_char_buf;
1769 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1773 /* Here, we have found all the possible problems, except for when the input
1774 * is for a problematic code point not allowed by the input parameters. */
1776 /* uv is valid for overlongs */
1777 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1779 /* isn't problematic if < this */
1780 && uv >= UNICODE_SURROGATE_FIRST)
1781 || ( UNLIKELY(possible_problems)
1783 /* if overflow, we know without looking further
1784 * precisely which of the problematic types it is,
1785 * and we deal with those in the overflow handling
1787 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1788 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1789 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1790 && ((flags & ( UTF8_DISALLOW_NONCHAR
1791 |UTF8_DISALLOW_SURROGATE
1792 |UTF8_DISALLOW_SUPER
1793 |UTF8_DISALLOW_PERL_EXTENDED
1795 |UTF8_WARN_SURROGATE
1797 |UTF8_WARN_PERL_EXTENDED))))
1799 /* If there were no malformations, or the only malformation is an
1800 * overlong, 'uv' is valid */
1801 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1802 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1803 possible_problems |= UTF8_GOT_SURROGATE;
1805 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1806 possible_problems |= UTF8_GOT_SUPER;
1808 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1809 possible_problems |= UTF8_GOT_NONCHAR;
1812 else { /* Otherwise, need to look at the source UTF-8, possibly
1813 adjusted to be non-overlong */
1815 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1816 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1818 possible_problems |= UTF8_GOT_SUPER;
1820 else if (curlen > 1) {
1821 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1822 NATIVE_UTF8_TO_I8(*adjusted_s0),
1823 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1825 possible_problems |= UTF8_GOT_SUPER;
1827 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1828 NATIVE_UTF8_TO_I8(*adjusted_s0),
1829 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1831 possible_problems |= UTF8_GOT_SURROGATE;
1835 /* We need a complete well-formed UTF-8 character to discern
1836 * non-characters, so can't look for them here */
1840 ready_to_handle_errors:
1843 * curlen contains the number of bytes in the sequence that
1844 * this call should advance the input by.
1845 * avail_len gives the available number of bytes passed in, but
1846 * only if this is less than the expected number of
1847 * bytes, based on the code point's start byte.
1848 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1849 * is set in it for each potential problem found.
1850 * uv contains the code point the input sequence
1851 * represents; or if there is a problem that prevents
1852 * a well-defined value from being computed, it is
1853 * some subsitute value, typically the REPLACEMENT
1855 * s0 points to the first byte of the character
1856 * s points to just after were we left off processing
1858 * send points to just after where that character should
1859 * end, based on how many bytes the start byte tells
1860 * us should be in it, but no further than s0 +
1864 if (UNLIKELY(possible_problems)) {
1865 bool disallowed = FALSE;
1866 const U32 orig_problems = possible_problems;
1872 while (possible_problems) { /* Handle each possible problem */
1874 char * message = NULL;
1875 U32 this_flag_bit = 0;
1877 /* Each 'if' clause handles one problem. They are ordered so that
1878 * the first ones' messages will be displayed before the later
1879 * ones; this is kinda in decreasing severity order. But the
1880 * overlong must come last, as it changes 'uv' looked at by the
1882 if (possible_problems & UTF8_GOT_OVERFLOW) {
1884 /* Overflow means also got a super and are using Perl's
1885 * extended UTF-8, but we handle all three cases here */
1887 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1888 *errors |= UTF8_GOT_OVERFLOW;
1890 /* But the API says we flag all errors found */
1891 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1892 *errors |= UTF8_GOT_SUPER;
1895 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1897 *errors |= UTF8_GOT_PERL_EXTENDED;
1900 /* Disallow if any of the three categories say to */
1901 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1902 || (flags & ( UTF8_DISALLOW_SUPER
1903 |UTF8_DISALLOW_PERL_EXTENDED)))
1908 /* Likewise, warn if any say to */
1909 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1910 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1913 /* The warnings code explicitly says it doesn't handle the
1914 * case of packWARN2 and two categories which have
1915 * parent-child relationship. Even if it works now to
1916 * raise the warning if either is enabled, it wouldn't
1917 * necessarily do so in the future. We output (only) the
1918 * most dire warning */
1919 if (! (flags & UTF8_CHECK_ONLY)) {
1920 if (msgs || ckWARN_d(WARN_UTF8)) {
1921 pack_warn = packWARN(WARN_UTF8);
1923 else if (msgs || ckWARN_d(WARN_NON_UNICODE)) {
1924 pack_warn = packWARN(WARN_NON_UNICODE);
1927 message = Perl_form(aTHX_ "%s: %s (overflows)",
1929 _byte_dump_string(s0, curlen, 0));
1930 this_flag_bit = UTF8_GOT_OVERFLOW;
1935 else if (possible_problems & UTF8_GOT_EMPTY) {
1936 possible_problems &= ~UTF8_GOT_EMPTY;
1937 *errors |= UTF8_GOT_EMPTY;
1939 if (! (flags & UTF8_ALLOW_EMPTY)) {
1941 /* This so-called malformation is now treated as a bug in
1942 * the caller. If you have nothing to decode, skip calling
1948 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1950 pack_warn = packWARN(WARN_UTF8);
1951 message = Perl_form(aTHX_ "%s (empty string)",
1953 this_flag_bit = UTF8_GOT_EMPTY;
1957 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1958 possible_problems &= ~UTF8_GOT_CONTINUATION;
1959 *errors |= UTF8_GOT_CONTINUATION;
1961 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1964 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1966 pack_warn = packWARN(WARN_UTF8);
1967 message = Perl_form(aTHX_
1968 "%s: %s (unexpected continuation byte 0x%02x,"
1969 " with no preceding start byte)",
1971 _byte_dump_string(s0, 1, 0), *s0);
1972 this_flag_bit = UTF8_GOT_CONTINUATION;
1976 else if (possible_problems & UTF8_GOT_SHORT) {
1977 possible_problems &= ~UTF8_GOT_SHORT;
1978 *errors |= UTF8_GOT_SHORT;
1980 if (! (flags & UTF8_ALLOW_SHORT)) {
1983 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
1985 pack_warn = packWARN(WARN_UTF8);
1986 message = Perl_form(aTHX_
1987 "%s: %s (too short; %d byte%s available, need %d)",
1989 _byte_dump_string(s0, send - s0, 0),
1991 avail_len == 1 ? "" : "s",
1993 this_flag_bit = UTF8_GOT_SHORT;
1998 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1999 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
2000 *errors |= UTF8_GOT_NON_CONTINUATION;
2002 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
2005 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2008 /* If we don't know for sure that the input length is
2009 * valid, avoid as much as possible reading past the
2010 * end of the buffer */
2011 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
2013 : (int) (send - s0);
2014 pack_warn = packWARN(WARN_UTF8);
2015 message = Perl_form(aTHX_ "%s",
2016 unexpected_non_continuation_text(s0,
2020 this_flag_bit = UTF8_GOT_NON_CONTINUATION;
2024 else if (possible_problems & UTF8_GOT_SURROGATE) {
2025 possible_problems &= ~UTF8_GOT_SURROGATE;
2027 if (flags & UTF8_WARN_SURROGATE) {
2028 *errors |= UTF8_GOT_SURROGATE;
2030 if ( ! (flags & UTF8_CHECK_ONLY)
2031 && (msgs || ckWARN_d(WARN_SURROGATE)))
2033 pack_warn = packWARN(WARN_SURROGATE);
2035 /* These are the only errors that can occur with a
2036 * surrogate when the 'uv' isn't valid */
2037 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2038 message = Perl_form(aTHX_
2039 "UTF-16 surrogate (any UTF-8 sequence that"
2040 " starts with \"%s\" is for a surrogate)",
2041 _byte_dump_string(s0, curlen, 0));
2044 message = Perl_form(aTHX_ surrogate_cp_format, uv);
2046 this_flag_bit = UTF8_GOT_SURROGATE;
2050 if (flags & UTF8_DISALLOW_SURROGATE) {
2052 *errors |= UTF8_GOT_SURROGATE;
2055 else if (possible_problems & UTF8_GOT_SUPER) {
2056 possible_problems &= ~UTF8_GOT_SUPER;
2058 if (flags & UTF8_WARN_SUPER) {
2059 *errors |= UTF8_GOT_SUPER;
2061 if ( ! (flags & UTF8_CHECK_ONLY)
2062 && (msgs || ckWARN_d(WARN_NON_UNICODE)))
2064 pack_warn = packWARN(WARN_NON_UNICODE);
2066 if (orig_problems & UTF8_GOT_TOO_SHORT) {
2067 message = Perl_form(aTHX_
2068 "Any UTF-8 sequence that starts with"
2069 " \"%s\" is for a non-Unicode code point,"
2070 " may not be portable",
2071 _byte_dump_string(s0, curlen, 0));
2074 message = Perl_form(aTHX_ super_cp_format, uv);
2076 this_flag_bit = UTF8_GOT_SUPER;
2080 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
2081 * and before possibly bailing out, so that the more dire
2082 * warning will override the regular one. */
2083 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
2084 if ( ! (flags & UTF8_CHECK_ONLY)
2085 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
2086 && (msgs || ( ckWARN_d(WARN_NON_UNICODE)
2087 || ckWARN(WARN_PORTABLE))))
2089 pack_warn = packWARN2(WARN_NON_UNICODE, WARN_PORTABLE);
2091 /* If it is an overlong that evaluates to a code point
2092 * that doesn't have to use the Perl extended UTF-8, it
2093 * still used it, and so we output a message that
2094 * doesn't refer to the code point. The same is true
2095 * if there was a SHORT malformation where the code
2096 * point is not valid. In that case, 'uv' will have
2097 * been set to the REPLACEMENT CHAR, and the message
2098 * below without the code point in it will be selected
2100 if (UNICODE_IS_PERL_EXTENDED(uv)) {
2101 message = Perl_form(aTHX_
2102 PL_extended_cp_format, uv);
2105 message = Perl_form(aTHX_
2106 "Any UTF-8 sequence that starts with"
2107 " \"%s\" is a Perl extension, and"
2108 " so is not portable",
2109 _byte_dump_string(s0, curlen, 0));
2111 this_flag_bit = UTF8_GOT_PERL_EXTENDED;
2114 if (flags & ( UTF8_WARN_PERL_EXTENDED
2115 |UTF8_DISALLOW_PERL_EXTENDED))
2117 *errors |= UTF8_GOT_PERL_EXTENDED;
2119 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
2125 if (flags & UTF8_DISALLOW_SUPER) {
2126 *errors |= UTF8_GOT_SUPER;
2130 else if (possible_problems & UTF8_GOT_NONCHAR) {
2131 possible_problems &= ~UTF8_GOT_NONCHAR;
2133 if (flags & UTF8_WARN_NONCHAR) {
2134 *errors |= UTF8_GOT_NONCHAR;
2136 if ( ! (flags & UTF8_CHECK_ONLY)
2137 && (msgs || ckWARN_d(WARN_NONCHAR)))
2139 /* The code above should have guaranteed that we don't
2140 * get here with errors other than overlong */
2141 assert (! (orig_problems
2142 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
2144 pack_warn = packWARN(WARN_NONCHAR);
2145 message = Perl_form(aTHX_ nonchar_cp_format, uv);
2146 this_flag_bit = UTF8_GOT_NONCHAR;
2150 if (flags & UTF8_DISALLOW_NONCHAR) {
2152 *errors |= UTF8_GOT_NONCHAR;
2155 else if (possible_problems & UTF8_GOT_LONG) {
2156 possible_problems &= ~UTF8_GOT_LONG;
2157 *errors |= UTF8_GOT_LONG;
2159 if (flags & UTF8_ALLOW_LONG) {
2161 /* We don't allow the actual overlong value, unless the
2162 * special extra bit is also set */
2163 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
2164 & ~UTF8_ALLOW_LONG)))
2166 uv = UNICODE_REPLACEMENT;
2173 || ckWARN_d(WARN_UTF8)) && ! (flags & UTF8_CHECK_ONLY))
2175 pack_warn = packWARN(WARN_UTF8);
2177 /* These error types cause 'uv' to be something that
2178 * isn't what was intended, so can't use it in the
2179 * message. The other error types either can't
2180 * generate an overlong, or else the 'uv' is valid */
2182 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
2184 message = Perl_form(aTHX_
2185 "%s: %s (any UTF-8 sequence that starts"
2186 " with \"%s\" is overlong which can and"
2187 " should be represented with a"
2188 " different, shorter sequence)",
2190 _byte_dump_string(s0, send - s0, 0),
2191 _byte_dump_string(s0, curlen, 0));
2194 U8 tmpbuf[UTF8_MAXBYTES+1];
2195 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
2197 /* Don't use U+ for non-Unicode code points, which
2198 * includes those in the Latin1 range */
2199 const char * preface = ( uv > PERL_UNICODE_MAX
2206 message = Perl_form(aTHX_
2207 "%s: %s (overlong; instead use %s to represent"
2210 _byte_dump_string(s0, send - s0, 0),
2211 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
2213 ((uv < 256) ? 2 : 4), /* Field width of 2 for
2214 small code points */
2217 this_flag_bit = UTF8_GOT_LONG;
2220 } /* End of looking through the possible flags */
2222 /* Display the message (if any) for the problem being handled in
2223 * this iteration of the loop */
2226 assert(this_flag_bit);
2228 if (*msgs == NULL) {
2232 av_push(*msgs, newRV_noinc((SV*) new_msg_hv(message,
2237 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
2240 Perl_warner(aTHX_ pack_warn, "%s", message);
2242 } /* End of 'while (possible_problems)' */
2244 /* Since there was a possible problem, the returned length may need to
2245 * be changed from the one stored at the beginning of this function.
2246 * Instead of trying to figure out if that's needed, just do it. */
2252 if (flags & UTF8_CHECK_ONLY && retlen) {
2253 *retlen = ((STRLEN) -1);
2259 return UNI_TO_NATIVE(uv);
2263 =for apidoc utf8_to_uvchr_buf
2265 Returns the native code point of the first character in the string C<s> which
2266 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2267 C<*retlen> will be set to the length, in bytes, of that character.
2269 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2270 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2271 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
2272 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
2273 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
2274 the next possible position in C<s> that could begin a non-malformed character.
2275 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2280 Also implemented as a macro in utf8.h
2286 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2288 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
2290 return utf8_to_uvchr_buf_helper(s, send, retlen);
2293 /* This is marked as deprecated
2295 =for apidoc utf8_to_uvuni_buf
2297 Only in very rare circumstances should code need to be dealing in Unicode
2298 (as opposed to native) code points. In those few cases, use
2299 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|perlapi/utf8_to_uvchr_buf>> instead.
2300 If you are not absolutely sure this is one of those cases, then assume it isn't
2301 and use plain C<utf8_to_uvchr_buf> instead.
2303 Returns the Unicode (not-native) code point of the first character in the
2305 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
2306 C<retlen> will be set to the length, in bytes, of that character.
2308 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
2309 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
2310 NULL) to -1. If those warnings are off, the computed value if well-defined (or
2311 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
2312 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
2313 next possible position in C<s> that could begin a non-malformed character.
2314 See L<perlapi/utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
2321 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
2323 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2327 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2331 =for apidoc utf8_length
2333 Returns the number of characters in the sequence of UTF-8-encoded bytes starting
2334 at C<s> and ending at the byte just before C<e>. If <s> and <e> point to the
2335 same place, it returns 0 with no warning raised.
2337 If C<e E<lt> s> or if the scan would end up past C<e>, it raises a UTF8 warning
2338 and returns the number of valid characters.
2344 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2348 PERL_ARGS_ASSERT_UTF8_LENGTH;
2350 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2351 * the bitops (especially ~) can create illegal UTF-8.
2352 * In other words: in Perl UTF-8 is not just for Unicode. */
2354 if (UNLIKELY(e < s))
2355 goto warn_and_return;
2361 if (UNLIKELY(e != s)) {
2365 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2366 "%s in %s", unees, OP_DESC(PL_op));
2368 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2375 =for apidoc bytes_cmp_utf8
2377 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2378 sequence of characters (stored as UTF-8)
2379 in C<u>, C<ulen>. Returns 0 if they are
2380 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2381 if the first string is greater than the second string.
2383 -1 or +1 is returned if the shorter string was identical to the start of the
2384 longer string. -2 or +2 is returned if
2385 there was a difference between characters
2392 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2394 const U8 *const bend = b + blen;
2395 const U8 *const uend = u + ulen;
2397 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2399 while (b < bend && u < uend) {
2401 if (!UTF8_IS_INVARIANT(c)) {
2402 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2405 if (UTF8_IS_CONTINUATION(c1)) {
2406 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2408 /* diag_listed_as: Malformed UTF-8 character%s */
2409 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2411 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2412 PL_op ? " in " : "",
2413 PL_op ? OP_DESC(PL_op) : "");
2418 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2419 "%s in %s", unees, OP_DESC(PL_op));
2421 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2422 return -2; /* Really want to return undef :-) */
2429 return *b < c ? -2 : +2;
2434 if (b == bend && u == uend)
2437 return b < bend ? +1 : -1;
2441 =for apidoc utf8_to_bytes
2443 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2444 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2445 updates C<*lenp> to contain the new length.
2446 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2448 Upon successful return, the number of variants in the string can be computed by
2449 having saved the value of C<*lenp> before the call, and subtracting the
2450 after-call value of C<*lenp> from it.
2452 If you need a copy of the string, see L</bytes_from_utf8>.
2458 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2462 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2463 PERL_UNUSED_CONTEXT;
2465 /* This is a no-op if no variants at all in the input */
2466 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2471 U8 * const save = s;
2472 U8 * const send = s + *lenp;
2475 /* Nothing before the first variant needs to be changed, so start the real
2479 if (! UTF8_IS_INVARIANT(*s)) {
2480 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2481 *lenp = ((STRLEN) -1);
2489 /* Is downgradable, so do it */
2490 d = s = first_variant;
2493 if (! UVCHR_IS_INVARIANT(c)) {
2494 /* Then it is two-byte encoded */
2495 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2508 =for apidoc bytes_from_utf8
2510 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2511 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2512 actually encoded in UTF-8.
2514 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2517 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2518 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2519 C<*lenp> are unchanged, and the return value is the original C<s>.
2521 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2522 newly created string containing a downgraded copy of C<s>, and whose length is
2523 returned in C<*lenp>, updated. The new string is C<NUL>-terminated. The
2524 caller is responsible for arranging for the memory used by this string to get
2527 Upon successful return, the number of variants in the string can be computed by
2528 having saved the value of C<*lenp> before the call, and subtracting the
2529 after-call value of C<*lenp> from it.
2533 There is a macro that avoids this function call, but this is retained for
2534 anyone who calls it with the Perl_ prefix */
2537 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2539 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2540 PERL_UNUSED_CONTEXT;
2542 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2548 This is not currently externally documented because we don't want people to use
2549 it for now. XXX Perhaps that is too paranoid, and it should be documented?
2551 =for apidoc bytes_from_utf8_loc
2553 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2554 to store the location of the first character in C<"s"> that cannot be
2555 converted to non-UTF8.
2557 If that parameter is C<NULL>, this function behaves identically to
2560 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2561 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2563 Otherwise, the function returns a newly created C<NUL>-terminated string
2564 containing the non-UTF8 equivalent of the convertible first portion of
2565 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2566 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2567 and C<*first_non_downgradable> is set to C<NULL>.
2569 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2570 first character in the original string that wasn't converted. C<*is_utf8p> is
2571 unchanged. Note that the new string may have length 0.
2573 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2574 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2575 converts as many characters in it as possible stopping at the first one it
2576 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2577 set to point to that. The function returns the portion that could be converted
2578 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2579 not including the terminating C<NUL>. If the very first character in the
2580 original could not be converted, C<*lenp> will be 0, and the new string will
2581 contain just a single C<NUL>. If the entire input string was converted,
2582 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2584 Upon successful return, the number of variants in the converted portion of the
2585 string can be computed by having saved the value of C<*lenp> before the call,
2586 and subtracting the after-call value of C<*lenp> from it.
2594 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2597 const U8 *original = s;
2598 U8 *converted_start;
2599 const U8 *send = s + *lenp;
2601 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2604 if (first_unconverted) {
2605 *first_unconverted = NULL;
2608 return (U8 *) original;
2611 Newx(d, (*lenp) + 1, U8);
2613 converted_start = d;
2616 if (! UTF8_IS_INVARIANT(c)) {
2618 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2619 * have to stop now */
2620 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2621 if (first_unconverted) {
2622 *first_unconverted = s - 1;
2623 goto finish_and_return;
2626 Safefree(converted_start);
2627 return (U8 *) original;
2631 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2637 /* Here, converted the whole of the input */
2639 if (first_unconverted) {
2640 *first_unconverted = NULL;
2645 *lenp = d - converted_start;
2647 /* Trim unused space */
2648 Renew(converted_start, *lenp + 1, U8);
2650 return converted_start;
2654 =for apidoc bytes_to_utf8
2656 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2658 Returns a pointer to the newly-created string, and sets C<*lenp> to
2659 reflect the new length in bytes. The caller is responsible for arranging for
2660 the memory used by this string to get freed.
2662 Upon successful return, the number of variants in the string can be computed by
2663 having saved the value of C<*lenp> before the call, and subtracting it from the
2664 after-call value of C<*lenp>.
2666 A C<NUL> character will be written after the end of the string.
2668 If you want to convert to UTF-8 from encodings other than
2669 the native (Latin1 or EBCDIC),
2670 see L</sv_recode_to_utf8>().
2676 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2678 const U8 * const send = s + (*lenp);
2682 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2683 PERL_UNUSED_CONTEXT;
2685 /* 1 for each byte + 1 for each byte that expands to two, + trailing NUL */
2686 Newx(d, (*lenp) + variant_under_utf8_count(s, send) + 1, U8);
2690 append_utf8_from_native_byte(*s, &d);
2701 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2702 * use utf16_to_utf8_reversed().
2704 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2705 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2706 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2708 * These functions don't check for overflow. The worst case is every code
2709 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2710 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2711 * destination must be pre-extended to 2 times the source length.
2713 * Do not use in-place. We optimize for native, for obvious reasons. */
2716 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2721 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2724 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2730 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2732 if (OFFUNI_IS_INVARIANT(uv)) {
2733 *d++ = LATIN1_TO_NATIVE((U8) uv);
2736 if (uv <= MAX_UTF8_TWO_BYTE) {
2737 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2738 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2742 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2743 #define LAST_HIGH_SURROGATE 0xDBFF
2744 #define FIRST_LOW_SURROGATE 0xDC00
2745 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2746 #define FIRST_IN_PLANE1 0x10000
2748 /* This assumes that most uses will be in the first Unicode plane, not
2749 * needing surrogates */
2750 if (UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST,
2751 UNICODE_SURROGATE_LAST)))
2753 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2754 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2757 UV low = (p[0] << 8) + p[1];
2758 if (UNLIKELY(! inRANGE(low, FIRST_LOW_SURROGATE,
2759 LAST_LOW_SURROGATE)))
2761 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2764 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2765 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2769 d = uvoffuni_to_utf8_flags(d, uv, 0);
2771 if (uv < FIRST_IN_PLANE1) {
2772 *d++ = (U8)(( uv >> 12) | 0xe0);
2773 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2774 *d++ = (U8)(( uv & 0x3f) | 0x80);
2778 *d++ = (U8)(( uv >> 18) | 0xf0);
2779 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2780 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2781 *d++ = (U8)(( uv & 0x3f) | 0x80);
2786 *newlen = d - dstart;
2790 /* Note: this one is slightly destructive of the source. */
2793 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, Size_t bytelen, Size_t *newlen)
2796 U8* const send = s + bytelen;
2798 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2801 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2805 const U8 tmp = s[0];
2810 return utf16_to_utf8(p, d, bytelen, newlen);
2814 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2817 return _invlist_contains_cp(PL_XPosix_ptrs[classnum], c);
2821 Perl__is_uni_perl_idcont(pTHX_ UV c)
2824 return _invlist_contains_cp(PL_utf8_perl_idcont, c);
2828 Perl__is_uni_perl_idstart(pTHX_ UV c)
2831 return _invlist_contains_cp(PL_utf8_perl_idstart, c);
2835 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2838 /* We have the latin1-range values compiled into the core, so just use
2839 * those, converting the result to UTF-8. The only difference between upper
2840 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2841 * either "SS" or "Ss". Which one to use is passed into the routine in
2842 * 'S_or_s' to avoid a test */
2844 UV converted = toUPPER_LATIN1_MOD(c);
2846 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2848 assert(S_or_s == 'S' || S_or_s == 's');
2850 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2851 characters in this range */
2852 *p = (U8) converted;
2857 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2858 * which it maps to one of them, so as to only have to have one check for
2859 * it in the main case */
2860 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2862 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2863 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2866 converted = GREEK_CAPITAL_LETTER_MU;
2868 #if UNICODE_MAJOR_VERSION > 2 \
2869 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2870 && UNICODE_DOT_DOT_VERSION >= 8)
2871 case LATIN_SMALL_LETTER_SHARP_S:
2878 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2879 " '%c' to map to '%c'",
2880 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2881 NOT_REACHED; /* NOTREACHED */
2885 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2886 *p = UTF8_TWO_BYTE_LO(converted);
2892 /* If compiled on an early Unicode version, there may not be auxiliary tables
2894 #ifndef HAS_UC_AUX_TABLES
2895 # define UC_AUX_TABLE_ptrs NULL
2896 # define UC_AUX_TABLE_lengths NULL
2898 #ifndef HAS_TC_AUX_TABLES
2899 # define TC_AUX_TABLE_ptrs NULL
2900 # define TC_AUX_TABLE_lengths NULL
2902 #ifndef HAS_LC_AUX_TABLES
2903 # define LC_AUX_TABLE_ptrs NULL
2904 # define LC_AUX_TABLE_lengths NULL
2906 #ifndef HAS_CF_AUX_TABLES
2907 # define CF_AUX_TABLE_ptrs NULL
2908 # define CF_AUX_TABLE_lengths NULL
2910 #ifndef HAS_UC_AUX_TABLES
2911 # define UC_AUX_TABLE_ptrs NULL
2912 # define UC_AUX_TABLE_lengths NULL
2915 /* Call the function to convert a UTF-8 encoded character to the specified case.
2916 * Note that there may be more than one character in the result.
2917 * 's' is a pointer to the first byte of the input character
2918 * 'd' will be set to the first byte of the string of changed characters. It
2919 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2920 * 'lenp' will be set to the length in bytes of the string of changed characters
2922 * The functions return the ordinal of the first character in the string of
2924 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2925 _to_utf8_case(uv, s, d, lenp, PL_utf8_toupper, \
2926 Uppercase_Mapping_invmap, \
2927 UC_AUX_TABLE_ptrs, \
2928 UC_AUX_TABLE_lengths, \
2930 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2931 _to_utf8_case(uv, s, d, lenp, PL_utf8_totitle, \
2932 Titlecase_Mapping_invmap, \
2933 TC_AUX_TABLE_ptrs, \
2934 TC_AUX_TABLE_lengths, \
2936 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2937 _to_utf8_case(uv, s, d, lenp, PL_utf8_tolower, \
2938 Lowercase_Mapping_invmap, \
2939 LC_AUX_TABLE_ptrs, \
2940 LC_AUX_TABLE_lengths, \
2944 /* This additionally has the input parameter 'specials', which if non-zero will
2945 * cause this to use the specials hash for folding (meaning get full case
2946 * folding); otherwise, when zero, this implies a simple case fold */
2947 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2949 ? _to_utf8_case(uv, s, d, lenp, PL_utf8_tofold, \
2950 Case_Folding_invmap, \
2951 CF_AUX_TABLE_ptrs, \
2952 CF_AUX_TABLE_lengths, \
2954 : _to_utf8_case(uv, s, d, lenp, PL_utf8_tosimplefold, \
2955 Simple_Case_Folding_invmap, \
2960 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2962 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2963 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2964 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2965 * the changed version may be longer than the original character.
2967 * The ordinal of the first character of the changed version is returned
2968 * (but note, as explained above, that there may be more.) */
2971 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2974 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2977 return CALL_UPPER_CASE(c, NULL, p, lenp);
2981 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2984 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2987 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2990 return CALL_TITLE_CASE(c, NULL, p, lenp);
2994 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2996 /* We have the latin1-range values compiled into the core, so just use
2997 * those, converting the result to UTF-8. Since the result is always just
2998 * one character, we allow <p> to be NULL */
3000 U8 converted = toLOWER_LATIN1(c);
3002 PERL_UNUSED_ARG(dummy);
3005 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
3010 /* Result is known to always be < 256, so can use the EIGHT_BIT
3012 *p = UTF8_EIGHT_BIT_HI(converted);
3013 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
3021 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
3024 PERL_ARGS_ASSERT_TO_UNI_LOWER;
3027 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
3030 return CALL_LOWER_CASE(c, NULL, p, lenp);
3034 Perl__to_fold_latin1(const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
3036 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
3037 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3038 * FOLD_FLAGS_FULL iff full folding is to be used;
3040 * Not to be used for locale folds
3045 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
3047 assert (! (flags & FOLD_FLAGS_LOCALE));
3049 if (UNLIKELY(c == MICRO_SIGN)) {
3050 converted = GREEK_SMALL_LETTER_MU;
3052 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
3053 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
3054 || UNICODE_DOT_DOT_VERSION > 0)
3055 else if ( (flags & FOLD_FLAGS_FULL)
3056 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
3058 /* If can't cross 127/128 boundary, can't return "ss"; instead return
3059 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
3060 * under those circumstances. */
3061 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
3062 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3063 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3065 return LATIN_SMALL_LETTER_LONG_S;
3075 else { /* In this range the fold of all other characters is their lower
3077 converted = toLOWER_LATIN1(c);
3080 if (UVCHR_IS_INVARIANT(converted)) {
3081 *p = (U8) converted;
3085 *(p)++ = UTF8_TWO_BYTE_HI(converted);
3086 *p = UTF8_TWO_BYTE_LO(converted);
3094 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
3097 /* Not currently externally documented, and subject to change
3098 * <flags> bits meanings:
3099 * FOLD_FLAGS_FULL iff full folding is to be used;
3100 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3101 * locale are to be used.
3102 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
3106 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
3108 if (flags & FOLD_FLAGS_LOCALE) {
3109 /* Treat a non-Turkic UTF-8 locale as not being in locale at all,
3110 * except for potentially warning */
3111 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
3112 if (IN_UTF8_CTYPE_LOCALE && ! PL_in_utf8_turkic_locale) {
3113 flags &= ~FOLD_FLAGS_LOCALE;
3116 goto needs_full_generality;
3121 return _to_fold_latin1((U8) c, p, lenp,
3122 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3125 /* Here, above 255. If no special needs, just use the macro */
3126 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
3127 return CALL_FOLD_CASE(c, NULL, p, lenp, flags & FOLD_FLAGS_FULL);
3129 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
3130 the special flags. */
3131 U8 utf8_c[UTF8_MAXBYTES + 1];
3133 needs_full_generality:
3134 uvchr_to_utf8(utf8_c, c);
3135 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
3140 PERL_STATIC_INLINE bool
3141 S_is_utf8_common(pTHX_ const U8 *const p, const U8 * const e,
3144 /* returns a boolean giving whether or not the UTF8-encoded character that
3145 * starts at <p>, and extending no further than <e - 1> is in the inversion
3146 * list <invlist>. */
3148 UV cp = utf8n_to_uvchr(p, e - p, NULL, 0);
3150 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
3152 if (cp == 0 && (p >= e || *p != '\0')) {
3153 _force_out_malformed_utf8_message(p, e, 0, 1);
3154 NOT_REACHED; /* NOTREACHED */
3158 return _invlist_contains_cp(invlist, cp);
3161 #if 0 /* Not currently used, but may be needed in the future */
3162 PERLVAR(I, seen_deprecated_macro, HV *)
3165 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
3166 const char * const alternative,
3167 const bool use_locale,
3168 const char * const file,
3169 const unsigned line)
3173 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
3175 if (ckWARN_d(WARN_DEPRECATED)) {
3177 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
3178 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
3179 if (! PL_seen_deprecated_macro) {
3180 PL_seen_deprecated_macro = newHV();
3182 if (! hv_store(PL_seen_deprecated_macro, key,
3183 strlen(key), &PL_sv_undef, 0))
3185 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3188 if (instr(file, "mathoms.c")) {
3189 Perl_warner(aTHX_ WARN_DEPRECATED,
3190 "In %s, line %d, starting in Perl v5.32, %s()"
3191 " will be removed. Avoid this message by"
3192 " converting to use %s().\n",
3193 file, line, name, alternative);
3196 Perl_warner(aTHX_ WARN_DEPRECATED,
3197 "In %s, line %d, starting in Perl v5.32, %s() will"
3198 " require an additional parameter. Avoid this"
3199 " message by converting to use %s().\n",
3200 file, line, name, alternative);
3208 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p, const U8 * const e)
3211 PERL_ARGS_ASSERT__IS_UTF8_FOO;
3213 return is_utf8_common(p, e, PL_XPosix_ptrs[classnum]);
3217 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p, const U8 * const e)
3220 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
3222 return is_utf8_common(p, e, PL_utf8_perl_idstart);
3226 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p, const U8 * const e)
3229 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
3231 return is_utf8_common(p, e, PL_utf8_perl_idcont);
3235 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p,
3236 U8* ustrp, STRLEN *lenp,
3237 SV *invlist, const I32 * const invmap,
3238 const U32 * const * const aux_tables,
3239 const U8 * const aux_table_lengths,
3240 const char * const normal)
3244 /* Change the case of code point 'uv1' whose UTF-8 representation (assumed
3245 * by this routine to be valid) begins at 'p'. 'normal' is a string to use
3246 * to name the new case in any generated messages, as a fallback if the
3247 * operation being used is not available. The new case is given by the
3248 * data structures in the remaining arguments.
3250 * On return 'ustrp' points to '*lenp' UTF-8 encoded bytes representing the
3251 * entire changed case string, and the return value is the first code point
3254 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3256 /* For code points that don't change case, we already know that the output
3257 * of this function is the unchanged input, so we can skip doing look-ups
3258 * for them. Unfortunately the case-changing code points are scattered
3259 * around. But there are some long consecutive ranges where there are no
3260 * case changing code points. By adding tests, we can eliminate the lookup
3261 * for all the ones in such ranges. This is currently done here only for
3262 * just a few cases where the scripts are in common use in modern commerce
3263 * (and scripts adjacent to those which can be included without additional
3266 if (uv1 >= 0x0590) {
3267 /* This keeps from needing further processing the code points most
3268 * likely to be used in the following non-cased scripts: Hebrew,
3269 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3270 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3271 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3276 /* The following largish code point ranges also don't have case
3277 * changes, but khw didn't think they warranted extra tests to speed
3278 * them up (which would slightly slow down everything else above them):
3279 * 1100..139F Hangul Jamo, Ethiopic
3280 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3281 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3282 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3283 * Combining Diacritical Marks Extended, Balinese,
3284 * Sundanese, Batak, Lepcha, Ol Chiki
3285 * 2000..206F General Punctuation
3288 if (uv1 >= 0x2D30) {
3290 /* This keeps the from needing further processing the code points
3291 * most likely to be used in the following non-cased major scripts:
3292 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3294 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3295 * event that Unicode eventually allocates the unused block as of
3296 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3297 * that the test suite will start having failures to alert you
3298 * should that happen) */
3303 if (uv1 >= 0xAC00) {
3304 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3305 if (ckWARN_d(WARN_SURROGATE)) {
3306 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3307 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3308 "Operation \"%s\" returns its argument for"
3309 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3314 /* AC00..FAFF Catches Hangul syllables and private use, plus
3320 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3321 if (UNLIKELY(uv1 > MAX_LEGAL_CP)) {
3322 Perl_croak(aTHX_ "%s", form_cp_too_large_msg(16, NULL, 0, uv1));
3324 if (ckWARN_d(WARN_NON_UNICODE)) {
3325 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3326 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3327 "Operation \"%s\" returns its argument for"
3328 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3332 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3334 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3343 /* Note that non-characters are perfectly legal, so no warning should
3349 const U32 * cp_list;
3352 /* 'index' is guaranteed to be non-negative, as this is an inversion
3353 * map that covers all possible inputs. See [perl #133365] */
3354 SSize_t index = _invlist_search(invlist, uv1);
3355 I32 base = invmap[index];
3357 /* The data structures are set up so that if 'base' is non-negative,
3358 * the case change is 1-to-1; and if 0, the change is to itself */
3366 /* This computes, e.g. lc(H) as 'H - A + a', using the lc table */
3367 lc = base + uv1 - invlist_array(invlist)[index];
3368 *lenp = uvchr_to_utf8(ustrp, lc) - ustrp;
3372 /* Here 'base' is negative. That means the mapping is 1-to-many, and
3373 * requires an auxiliary table look up. abs(base) gives the index into
3374 * a list of such tables which points to the proper aux table. And a
3375 * parallel list gives the length of each corresponding aux table. */
3376 cp_list = aux_tables[-base];
3378 /* Create the string of UTF-8 from the mapped-to code points */
3380 for (i = 0; i < aux_table_lengths[-base]; i++) {
3381 d = uvchr_to_utf8(d, cp_list[i]);
3389 /* Here, there was no mapping defined, which means that the code point maps
3390 * to itself. Return the inputs */
3394 if (p != ustrp) { /* Don't copy onto itself */
3395 Copy(p, ustrp, len, U8);
3400 *lenp = uvchr_to_utf8(ustrp, uv1) - ustrp;
3408 Perl__inverse_folds(pTHX_ const UV cp, U32 * first_folds_to,
3409 const U32 ** remaining_folds_to)
3411 /* Returns the count of the number of code points that fold to the input
3412 * 'cp' (besides itself).
3414 * If the return is 0, there is nothing else that folds to it, and
3415 * '*first_folds_to' is set to 0, and '*remaining_folds_to' is set to NULL.
3417 * If the return is 1, '*first_folds_to' is set to the single code point,
3418 * and '*remaining_folds_to' is set to NULL.
3420 * Otherwise, '*first_folds_to' is set to a code point, and
3421 * '*remaining_fold_to' is set to an array that contains the others. The
3422 * length of this array is the returned count minus 1.
3424 * The reason for this convolution is to avoid having to deal with
3425 * allocating and freeing memory. The lists are already constructed, so
3426 * the return can point to them, but single code points aren't, so would
3427 * need to be constructed if we didn't employ something like this API
3429 * The code points returned by this function are all legal Unicode, which
3430 * occupy at most 21 bits, and so a U32 is sufficient, and the lists are
3431 * constructed with this size (to save space and memory), and we return
3432 * pointers, so they must be this size */
3435 /* 'index' is guaranteed to be non-negative, as this is an inversion map
3436 * that covers all possible inputs. See [perl #133365] */
3437 SSize_t index = _invlist_search(PL_utf8_foldclosures, cp);
3438 I32 base = _Perl_IVCF_invmap[index];
3440 PERL_ARGS_ASSERT__INVERSE_FOLDS;
3442 if (base == 0) { /* No fold */
3443 *first_folds_to = 0;
3444 *remaining_folds_to = NULL;
3448 #ifndef HAS_IVCF_AUX_TABLES /* This Unicode version only has 1-1 folds */
3454 if (UNLIKELY(base < 0)) { /* Folds to more than one character */
3456 /* The data structure is set up so that the absolute value of 'base' is
3457 * an index into a table of pointers to arrays, with the array
3458 * corresponding to the index being the list of code points that fold
3459 * to 'cp', and the parallel array containing the length of the list
3461 *first_folds_to = IVCF_AUX_TABLE_ptrs[-base][0];
3462 *remaining_folds_to = IVCF_AUX_TABLE_ptrs[-base] + 1;
3463 /* +1 excludes first_folds_to */
3464 return IVCF_AUX_TABLE_lengths[-base];
3469 /* Only the single code point. This works like 'fc(G) = G - A + a' */
3470 *first_folds_to = (U32) (base + cp
3471 - invlist_array(PL_utf8_foldclosures)[index]);
3472 *remaining_folds_to = NULL;
3477 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3478 U8* const ustrp, STRLEN *lenp)
3480 /* This is called when changing the case of a UTF-8-encoded character above
3481 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3482 * result contains a character that crosses the 255/256 boundary, disallow
3483 * the change, and return the original code point. See L<perlfunc/lc> for
3486 * p points to the original string whose case was changed; assumed
3487 * by this routine to be well-formed
3488 * result the code point of the first character in the changed-case string
3489 * ustrp points to the changed-case string (<result> represents its
3491 * lenp points to the length of <ustrp> */
3493 UV original; /* To store the first code point of <p> */
3495 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3497 assert(UTF8_IS_ABOVE_LATIN1(*p));
3499 /* We know immediately if the first character in the string crosses the
3500 * boundary, so can skip testing */
3503 /* Look at every character in the result; if any cross the
3504 * boundary, the whole thing is disallowed */
3505 U8* s = ustrp + UTF8SKIP(ustrp);
3506 U8* e = ustrp + *lenp;
3508 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3514 /* Here, no characters crossed, result is ok as-is, but we warn. */
3515 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3521 /* Failed, have to return the original */
3522 original = valid_utf8_to_uvchr(p, lenp);
3524 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3525 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3526 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3527 " locale; resolved to \"\\x{%" UVXf "}\".",
3531 Copy(p, ustrp, *lenp, char);
3536 S_turkic_fc(pTHX_ const U8 * const p, const U8 * const e,
3537 U8 * ustrp, STRLEN *lenp)
3539 /* Returns 0 if the foldcase of the input UTF-8 encoded sequence from
3540 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3541 * Otherwise, it returns the first code point of the Turkic foldcased
3542 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3543 * contain *lenp bytes
3545 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3546 * I WITH DOT ABOVE form a case pair, as do 'I' and LATIN SMALL LETTER
3549 PERL_ARGS_ASSERT_TURKIC_FC;
3552 if (UNLIKELY(*p == 'I')) {
3554 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_SMALL_LETTER_DOTLESS_I);
3555 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_SMALL_LETTER_DOTLESS_I);
3556 return LATIN_SMALL_LETTER_DOTLESS_I;
3559 if (UNLIKELY(memBEGINs(p, e - p,
3560 LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8)))
3571 S_turkic_lc(pTHX_ const U8 * const p0, const U8 * const e,
3572 U8 * ustrp, STRLEN *lenp)
3574 /* Returns 0 if the lowercase of the input UTF-8 encoded sequence from
3575 * p0..e-1 according to Turkic rules is the same as for non-Turkic.
3576 * Otherwise, it returns the first code point of the Turkic lowercased
3577 * sequence, and the entire sequence will be stored in *ustrp. ustrp will
3578 * contain *lenp bytes */
3581 PERL_ARGS_ASSERT_TURKIC_LC;
3584 /* A 'I' requires context as to what to do */
3585 if (UNLIKELY(*p0 == 'I')) {
3586 const U8 * p = p0 + 1;
3588 /* According to the Unicode SpecialCasing.txt file, a capital 'I'
3589 * modified by a dot above lowercases to 'i' even in turkic locales. */
3593 if (memBEGINs(p, e - p, COMBINING_DOT_ABOVE_UTF8)) {
3599 /* For the dot above to modify the 'I', it must be part of a
3600 * combining sequence immediately following the 'I', and no other
3601 * modifier with a ccc of 230 may intervene */
3602 cp = utf8_to_uvchr_buf(p, e, NULL);
3603 if (! _invlist_contains_cp(PL_CCC_non0_non230, cp)) {
3607 /* Here the combining sequence continues */
3612 /* In all other cases the lc is the same as the fold */
3613 return turkic_fc(p0, e, ustrp, lenp);
3617 S_turkic_uc(pTHX_ const U8 * const p, const U8 * const e,
3618 U8 * ustrp, STRLEN *lenp)
3620 /* Returns 0 if the upper or title-case of the input UTF-8 encoded sequence
3621 * from p0..e-1 according to Turkic rules is the same as for non-Turkic.
3622 * Otherwise, it returns the first code point of the Turkic upper or
3623 * title-cased sequence, and the entire sequence will be stored in *ustrp.
3624 * ustrp will contain *lenp bytes
3626 * Turkic differs only from non-Turkic in that 'i' and LATIN CAPITAL LETTER
3627 * I WITH DOT ABOVE form a case pair, as do 'I' and and LATIN SMALL LETTER
3630 PERL_ARGS_ASSERT_TURKIC_UC;
3635 ustrp[0] = UTF8_TWO_BYTE_HI(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3636 ustrp[1] = UTF8_TWO_BYTE_LO(LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE);
3637 return LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE;
3640 if (memBEGINs(p, e - p, LATIN_SMALL_LETTER_DOTLESS_I_UTF8)) {
3649 /* The process for changing the case is essentially the same for the four case
3650 * change types, except there are complications for folding. Otherwise the
3651 * difference is only which case to change to. To make sure that they all do
3652 * the same thing, the bodies of the functions are extracted out into the
3653 * following two macros. The functions are written with the same variable
3654 * names, and these are known and used inside these macros. It would be
3655 * better, of course, to have inline functions to do it, but since different
3656 * macros are called, depending on which case is being changed to, this is not
3657 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3658 * function can start with the common start macro, then finish with its special
3659 * handling; while the other three cases can just use the common end macro.
3661 * The algorithm is to use the proper (passed in) macro or function to change
3662 * the case for code points that are below 256. The macro is used if using
3663 * locale rules for the case change; the function if not. If the code point is
3664 * above 255, it is computed from the input UTF-8, and another macro is called
3665 * to do the conversion. If necessary, the output is converted to UTF-8. If
3666 * using a locale, we have to check that the change did not cross the 255/256
3667 * boundary, see check_locale_boundary_crossing() for further details.
3669 * The macros are split with the correct case change for the below-256 case
3670 * stored into 'result', and in the middle of an else clause for the above-255
3671 * case. At that point in the 'else', 'result' is not the final result, but is
3672 * the input code point calculated from the UTF-8. The fold code needs to
3673 * realize all this and take it from there.
3675 * To deal with Turkic locales, the function specified by the parameter
3676 * 'turkic' is called when appropriate.
3678 * If you read the two macros as sequential, it's easier to understand what's
3680 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3681 L1_func_extra_param, turkic) \
3683 if (flags & (locale_flags)) { \
3684 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3685 if (IN_UTF8_CTYPE_LOCALE) { \
3686 if (UNLIKELY(PL_in_utf8_turkic_locale)) { \
3687 UV ret = turkic(p, e, ustrp, lenp); \
3688 if (ret) return ret; \
3691 /* Otherwise, treat a UTF-8 locale as not being in locale at \
3693 flags &= ~(locale_flags); \
3697 if (UTF8_IS_INVARIANT(*p)) { \
3698 if (flags & (locale_flags)) { \
3699 result = LC_L1_change_macro(*p); \
3702 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3705 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3706 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)); \
3707 if (flags & (locale_flags)) { \
3708 result = LC_L1_change_macro(c); \
3711 return L1_func(c, ustrp, lenp, L1_func_extra_param); \
3714 else { /* malformed UTF-8 or ord above 255 */ \
3715 STRLEN len_result; \
3716 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3717 if (len_result == (STRLEN) -1) { \
3718 _force_out_malformed_utf8_message(p, e, 0, 1 /* Die */ ); \
3721 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3722 result = change_macro(result, p, ustrp, lenp); \
3724 if (flags & (locale_flags)) { \
3725 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3730 /* Here, used locale rules. Convert back to UTF-8 */ \
3731 if (UTF8_IS_INVARIANT(result)) { \
3732 *ustrp = (U8) result; \
3736 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3737 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3743 /* Not currently externally documented, and subject to change:
3744 * <flags> is set iff iff the rules from the current underlying locale are to
3748 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3757 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3759 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3760 /* 2nd char of uc(U+DF) is 'S' */
3761 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S',
3763 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3766 /* Not currently externally documented, and subject to change:
3767 * <flags> is set iff the rules from the current underlying locale are to be
3768 * used. Since titlecase is not defined in POSIX, for other than a
3769 * UTF-8 locale, uppercase is used instead for code points < 256.
3773 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3782 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3784 /* 2nd char of ucfirst(U+DF) is 's' */
3785 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's',
3787 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3790 /* Not currently externally documented, and subject to change:
3791 * <flags> is set iff iff the rules from the current underlying locale are to
3796 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3805 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3807 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */,
3809 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3812 /* Not currently externally documented, and subject to change,
3814 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3815 * locale are to be used.
3816 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3817 * otherwise simple folds
3818 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3823 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3832 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3834 /* These are mutually exclusive */
3835 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3837 assert(p != ustrp); /* Otherwise overwrites */
3839 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3840 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)),
3843 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3845 if (flags & FOLD_FLAGS_LOCALE) {
3847 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3848 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3849 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3851 /* Special case these two characters, as what normally gets
3852 * returned under locale doesn't work */
3853 if (memBEGINs((char *) p, e - p, CAP_SHARP_S))
3855 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3856 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3857 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3858 "resolved to \"\\x{17F}\\x{17F}\".");
3863 if (memBEGINs((char *) p, e - p, LONG_S_T))
3865 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3866 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3867 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3868 "resolved to \"\\x{FB06}\".");
3869 goto return_ligature_st;
3872 #if UNICODE_MAJOR_VERSION == 3 \
3873 && UNICODE_DOT_VERSION == 0 \
3874 && UNICODE_DOT_DOT_VERSION == 1
3875 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3877 /* And special case this on this Unicode version only, for the same
3878 * reaons the other two are special cased. They would cross the
3879 * 255/256 boundary which is forbidden under /l, and so the code
3880 * wouldn't catch that they are equivalent (which they are only in
3882 else if (memBEGINs((char *) p, e - p, DOTTED_I)) {
3883 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3884 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3885 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3886 "resolved to \"\\x{0131}\".");
3887 goto return_dotless_i;
3891 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3893 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3897 /* This is called when changing the case of a UTF-8-encoded
3898 * character above the ASCII range, and the result should not
3899 * contain an ASCII character. */
3901 UV original; /* To store the first code point of <p> */
3903 /* Look at every character in the result; if any cross the
3904 * boundary, the whole thing is disallowed */
3906 U8* send = ustrp + *lenp;
3909 /* Crossed, have to return the original */
3910 original = valid_utf8_to_uvchr(p, lenp);
3912 /* But in these instances, there is an alternative we can
3913 * return that is valid */
3914 if (original == LATIN_SMALL_LETTER_SHARP_S
3915 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3916 || original == LATIN_CAPITAL_LETTER_SHARP_S
3921 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3922 goto return_ligature_st;
3924 #if UNICODE_MAJOR_VERSION == 3 \
3925 && UNICODE_DOT_VERSION == 0 \
3926 && UNICODE_DOT_DOT_VERSION == 1
3928 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3929 goto return_dotless_i;
3932 Copy(p, ustrp, *lenp, char);
3938 /* Here, no characters crossed, result is ok as-is */
3943 /* Here, used locale rules. Convert back to UTF-8 */
3944 if (UTF8_IS_INVARIANT(result)) {
3945 *ustrp = (U8) result;
3949 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3950 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3957 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3958 * folds to a string of two of these characters. By returning this
3959 * instead, then, e.g.,
3960 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3963 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3964 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3966 return LATIN_SMALL_LETTER_LONG_S;
3969 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3970 * have the other one fold to it */
3972 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3973 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3974 return LATIN_SMALL_LIGATURE_ST;
3976 #if UNICODE_MAJOR_VERSION == 3 \
3977 && UNICODE_DOT_VERSION == 0 \
3978 && UNICODE_DOT_DOT_VERSION == 1
3981 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3982 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3983 return LATIN_SMALL_LETTER_DOTLESS_I;
3990 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
3992 /* May change: warns if surrogates, non-character code points, or
3993 * non-Unicode code points are in 's' which has length 'len' bytes.
3994 * Returns TRUE if none found; FALSE otherwise. The only other validity
3995 * check is to make sure that this won't exceed the string's length nor
3998 const U8* const e = s + len;
4001 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4004 if (UTF8SKIP(s) > len) {
4005 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4006 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4009 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4010 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4011 if ( ckWARN_d(WARN_NON_UNICODE)
4012 || UNLIKELY(0 < does_utf8_overflow(s, s + len,
4013 0 /* Don't consider overlongs */
4016 /* A side effect of this function will be to warn */
4017 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4021 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4022 if (ckWARN_d(WARN_SURROGATE)) {
4023 /* This has a different warning than the one the called
4024 * function would output, so can't just call it, unlike we
4025 * do for the non-chars and above-unicodes */
4026 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4027 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4028 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
4033 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
4034 && (ckWARN_d(WARN_NONCHAR)))
4036 /* A side effect of this function will be to warn */
4037 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4048 =for apidoc pv_uni_display
4050 Build to the scalar C<dsv> a displayable version of the UTF-8 encoded string
4051 C<spv>, length C<len>, the displayable version being at most C<pvlim> bytes
4052 long (if longer, the rest is truncated and C<"..."> will be appended).
4054 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4055 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4056 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4057 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4058 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4059 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4061 Additionally, there is now C<UNI_DISPLAY_BACKSPACE> which allows C<\b> for a
4062 backspace, but only when C<UNI_DISPLAY_BACKSLASH> also is set.
4064 The pointer to the PV of the C<dsv> is returned.
4066 See also L</sv_uni_display>.
4070 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
4076 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4080 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4084 if (pvlim && SvCUR(dsv) >= pvlim) {
4088 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4090 const unsigned char c = (unsigned char)u & 0xFF;
4091 if (flags & UNI_DISPLAY_BACKSLASH) {
4092 if ( isMNEMONIC_CNTRL(c)
4094 || (flags & UNI_DISPLAY_BACKSPACE)))
4096 const char * mnemonic = cntrl_to_mnemonic(c);
4097 sv_catpvn(dsv, mnemonic, strlen(mnemonic));
4100 else if (c == '\\') {
4101 sv_catpvs(dsv, "\\\\");
4105 /* isPRINT() is the locale-blind version. */
4106 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4107 const char string = c;
4108 sv_catpvn(dsv, &string, 1);
4113 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
4116 sv_catpvs(dsv, "...");
4122 =for apidoc sv_uni_display
4124 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4125 the displayable version being at most C<pvlim> bytes long
4126 (if longer, the rest is truncated and "..." will be appended).
4128 The C<flags> argument is as in L</pv_uni_display>().
4130 The pointer to the PV of the C<dsv> is returned.
4135 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4137 const char * const ptr =
4138 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4140 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4142 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4143 SvCUR(ssv), pvlim, flags);
4147 =for apidoc foldEQ_utf8
4149 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
4150 both of which may be in UTF-8) are the same case-insensitively; false
4151 otherwise. How far into the strings to compare is determined by other input
4154 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4155 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
4156 C<u2> with respect to C<s2>.
4158 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
4159 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
4160 The scan will not be considered to be a match unless the goal is reached, and
4161 scanning won't continue past that goal. Correspondingly for C<l2> with respect
4164 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
4165 pointer is considered an end pointer to the position 1 byte past the maximum
4166 point in C<s1> beyond which scanning will not continue under any circumstances.
4167 (This routine assumes that UTF-8 encoded input strings are not malformed;
4168 malformed input can cause it to read past C<pe1>). This means that if both
4169 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
4170 will never be successful because it can never
4171 get as far as its goal (and in fact is asserted against). Correspondingly for
4172 C<pe2> with respect to C<s2>.
4174 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4175 C<l2> must be non-zero), and if both do, both have to be
4176 reached for a successful match. Also, if the fold of a character is multiple
4177 characters, all of them must be matched (see tr21 reference below for
4180 Upon a successful match, if C<pe1> is non-C<NULL>,
4181 it will be set to point to the beginning of the I<next> character of C<s1>
4182 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4184 For case-insensitiveness, the "casefolding" of Unicode is used
4185 instead of upper/lowercasing both the characters, see
4186 L<https://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4190 /* A flags parameter has been added which may change, and hence isn't
4191 * externally documented. Currently it is:
4192 * 0 for as-documented above
4193 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4194 ASCII one, to not match
4195 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4196 * locale are to be used.
4197 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4198 * routine. This allows that step to be skipped.
4199 * Currently, this requires s1 to be encoded as UTF-8
4200 * (u1 must be true), which is asserted for.
4201 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4202 * cross certain boundaries. Hence, the caller should
4203 * let this function do the folding instead of
4204 * pre-folding. This code contains an assertion to
4205 * that effect. However, if the caller knows what
4206 * it's doing, it can pass this flag to indicate that,
4207 * and the assertion is skipped.
4208 * FOLDEQ_S2_ALREADY_FOLDED Similar to FOLDEQ_S1_ALREADY_FOLDED, but applies
4209 * to s2, and s2 doesn't have to be UTF-8 encoded.
4210 * This introduces an asymmetry to save a few branches
4211 * in a loop. Currently, this is not a problem, as
4212 * never are both inputs pre-folded. Simply call this
4213 * function with the pre-folded one as the second
4215 * FOLDEQ_S2_FOLDS_SANE
4218 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
4219 const char *s2, char **pe2, UV l2, bool u2,
4222 const U8 *p1 = (const U8*)s1; /* Point to current char */
4223 const U8 *p2 = (const U8*)s2;
4224 const U8 *g1 = NULL; /* goal for s1 */
4225 const U8 *g2 = NULL;
4226 const U8 *e1 = NULL; /* Don't scan s1 past this */
4227 U8 *f1 = NULL; /* Point to current folded */
4228 const U8 *e2 = NULL;
4230 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4231 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4232 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4233 U8 flags_for_folder = FOLD_FLAGS_FULL;
4235 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4237 assert( ! ( (flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4238 && (( (flags & FOLDEQ_S1_ALREADY_FOLDED)
4239 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4240 || ( (flags & FOLDEQ_S2_ALREADY_FOLDED)
4241 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4242 /* The algorithm is to trial the folds without regard to the flags on
4243 * the first line of the above assert(), and then see if the result
4244 * violates them. This means that the inputs can't be pre-folded to a
4245 * violating result, hence the assert. This could be changed, with the
4246 * addition of extra tests here for the already-folded case, which would
4247 * slow it down. That cost is more than any possible gain for when these
4248 * flags are specified, as the flags indicate /il or /iaa matching which
4249 * is less common than /iu, and I (khw) also believe that real-world /il
4250 * and /iaa matches are most likely to involve code points 0-255, and this
4251 * function only under rare conditions gets called for 0-255. */
4253 if (flags & FOLDEQ_LOCALE) {
4254 if (IN_UTF8_CTYPE_LOCALE) {
4255 if (UNLIKELY(PL_in_utf8_turkic_locale)) {
4256 flags_for_folder |= FOLD_FLAGS_LOCALE;
4259 flags &= ~FOLDEQ_LOCALE;
4263 flags_for_folder |= FOLD_FLAGS_LOCALE;
4266 if (flags & FOLDEQ_UTF8_NOMIX_ASCII) {
4267 flags_for_folder |= FOLD_FLAGS_NOMIX_ASCII;
4275 g1 = (const U8*)s1 + l1;
4283 g2 = (const U8*)s2 + l2;
4286 /* Must have at least one goal */
4291 /* Will never match if goal is out-of-bounds */
4292 assert(! e1 || e1 >= g1);
4294 /* Here, there isn't an end pointer, or it is beyond the goal. We
4295 * only go as far as the goal */
4299 assert(e1); /* Must have an end for looking at s1 */
4302 /* Same for goal for s2 */
4304 assert(! e2 || e2 >= g2);
4311 /* If both operands are already folded, we could just do a memEQ on the
4312 * whole strings at once, but it would be better if the caller realized
4313 * this and didn't even call us */
4315 /* Look through both strings, a character at a time */
4316 while (p1 < e1 && p2 < e2) {
4318 /* If at the beginning of a new character in s1, get its fold to use
4319 * and the length of the fold. */
4321 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4327 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4329 /* We have to forbid mixing ASCII with non-ASCII if the
4330 * flags so indicate. And, we can short circuit having to
4331 * call the general functions for this common ASCII case,
4332 * all of whose non-locale folds are also ASCII, and hence
4333 * UTF-8 invariants, so the UTF8ness of the strings is not
4335 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4339 *foldbuf1 = toFOLD(*p1);
4342 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
4344 else { /* Not UTF-8, get UTF-8 fold */
4345 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
4351 if (n2 == 0) { /* Same for s2 */
4352 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
4354 /* Point to the already-folded character. But for non-UTF-8
4355 * variants, convert to UTF-8 for the algorithm below */
4356 if (UTF8_IS_INVARIANT(*p2)) {
4365 foldbuf2[0] = UTF8_EIGHT_BIT_HI(*p2);
4366 foldbuf2[1] = UTF8_EIGHT_BIT_LO(*p2);
4372 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
4373 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
4377 *foldbuf2 = toFOLD(*p2);
4380 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
4383 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
4389 /* Here f1 and f2 point to the beginning of the strings to compare.
4390 * These strings are the folds of the next character from each input
4391 * string, stored in UTF-8. */
4393 /* While there is more to look for in both folds, see if they
4394 * continue to match */
4396 U8 fold_length = UTF8SKIP(f1);
4397 if (fold_length != UTF8SKIP(f2)
4398 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
4399 function call for single
4401 || memNE((char*)f1, (char*)f2, fold_length))
4403 return 0; /* mismatch */
4406 /* Here, they matched, advance past them */
4413 /* When reach the end of any fold, advance the input past it */
4415 p1 += u1 ? UTF8SKIP(p1) : 1;
4418 p2 += u2 ? UTF8SKIP(p2) : 1;
4420 } /* End of loop through both strings */
4422 /* A match is defined by each scan that specified an explicit length
4423 * reaching its final goal, and the other not having matched a partial
4424 * character (which can happen when the fold of a character is more than one
4426 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
4430 /* Successful match. Set output pointers */
4441 * ex: set ts=8 sts=4 sw=4 et: