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)";
39 static const char cp_above_legal_max[] =
40 "Use of code point 0x%" UVXf " is not allowed; "
41 "the permissible max is 0x%" UVXf;
43 #define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
46 =head1 Unicode Support
47 These are various utility functions for manipulating UTF8-encoded
48 strings. For the uninitiated, this is a method of representing arbitrary
49 Unicode characters as a variable number of bytes, in such a way that
50 characters in the ASCII range are unmodified, and a zero byte never appears
51 within non-zero characters.
57 Perl__force_out_malformed_utf8_message(pTHX_
58 const U8 *const p, /* First byte in UTF-8 sequence */
59 const U8 * const e, /* Final byte in sequence (may include
61 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
62 usually 0, or some DISALLOW flags */
63 const bool die_here) /* If TRUE, this function does not return */
65 /* This core-only function is to be called when a malformed UTF-8 character
66 * is found, in order to output the detailed information about the
67 * malformation before dieing. The reason it exists is for the occasions
68 * when such a malformation is fatal, but warnings might be turned off, so
69 * that normally they would not be actually output. This ensures that they
70 * do get output. Because a sequence may be malformed in more than one
71 * way, multiple messages may be generated, so we can't make them fatal, as
72 * that would cause the first one to die.
74 * Instead we pretend -W was passed to perl, then die afterwards. The
75 * flexibility is here to return to the caller so they can finish up and
79 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
85 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
87 PL_curcop->cop_warnings = pWARN_ALL;
90 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
95 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
96 " be called only when there are errors found");
100 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
105 =for apidoc uvoffuni_to_utf8_flags
107 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
108 Instead, B<Almost all code should use L</uvchr_to_utf8> or
109 L</uvchr_to_utf8_flags>>.
111 This function is like them, but the input is a strict Unicode
112 (as opposed to native) code point. Only in very rare circumstances should code
113 not be using the native code point.
115 For details, see the description for L</uvchr_to_utf8_flags>.
120 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
122 if (flags & UNICODE_WARN_SURROGATE) { \
123 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
124 "UTF-16 surrogate U+%04" UVXf, uv); \
126 if (flags & UNICODE_DISALLOW_SURROGATE) { \
131 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
133 if (flags & UNICODE_WARN_NONCHAR) { \
134 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
135 "Unicode non-character U+%04" UVXf " is not " \
136 "recommended for open interchange", uv); \
138 if (flags & UNICODE_DISALLOW_NONCHAR) { \
143 /* Use shorter names internally in this file */
144 #define SHIFT UTF_ACCUMULATION_SHIFT
146 #define MARK UTF_CONTINUATION_MARK
147 #define MASK UTF_CONTINUATION_MASK
150 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
152 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
154 if (OFFUNI_IS_INVARIANT(uv)) {
155 *d++ = LATIN1_TO_NATIVE(uv);
159 if (uv <= MAX_UTF8_TWO_BYTE) {
160 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
161 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
165 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
166 * below, the 16 is for start bytes E0-EF (which are all the possible ones
167 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
168 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
169 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
170 * 0x800-0xFFFF on ASCII */
171 if (uv < (16 * (1U << (2 * SHIFT)))) {
172 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
173 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
174 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
176 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
177 aren't tested here */
178 /* The most likely code points in this range are below the surrogates.
179 * Do an extra test to quickly exclude those. */
180 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
181 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
182 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
184 HANDLE_UNICODE_NONCHAR(uv, flags);
186 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
187 HANDLE_UNICODE_SURROGATE(uv, flags);
194 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
195 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
196 * happen starting with 4-byte characters on ASCII platforms. We unify the
197 * code for these with EBCDIC, even though some of them require 5-bytes on
198 * those, because khw believes the code saving is worth the very slight
199 * performance hit on these high EBCDIC code points. */
201 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
202 if (UNLIKELY(uv > MAX_NON_DEPRECATED_CP)) {
203 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
205 if ( (flags & UNICODE_WARN_SUPER)
206 || ( UNICODE_IS_ABOVE_31_BIT(uv)
207 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
209 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
211 /* Choose the more dire applicable warning */
212 (UNICODE_IS_ABOVE_31_BIT(uv))
213 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
214 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
217 if (flags & UNICODE_DISALLOW_SUPER
218 || ( UNICODE_IS_ABOVE_31_BIT(uv)
219 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
224 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
225 HANDLE_UNICODE_NONCHAR(uv, flags);
228 /* Test for and handle 4-byte result. In the test immediately below, the
229 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
230 * characters). The 3 is for 3 continuation bytes; these each contribute
231 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
232 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
233 * 0x1_0000-0x1F_FFFF on ASCII */
234 if (uv < (8 * (1U << (3 * SHIFT)))) {
235 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
236 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
237 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
238 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
240 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
241 characters. The end-plane non-characters for EBCDIC were
242 handled just above */
243 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
244 HANDLE_UNICODE_NONCHAR(uv, flags);
246 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
247 HANDLE_UNICODE_SURROGATE(uv, flags);
254 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
255 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
256 * format. The unrolled version above turns out to not save all that much
257 * time, and at these high code points (well above the legal Unicode range
258 * on ASCII platforms, and well above anything in common use in EBCDIC),
259 * khw believes that less code outweighs slight performance gains. */
262 STRLEN len = OFFUNISKIP(uv);
265 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
266 uv >>= UTF_ACCUMULATION_SHIFT;
268 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
274 =for apidoc uvchr_to_utf8
276 Adds the UTF-8 representation of the native code point C<uv> to the end
277 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
278 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
279 the byte after the end of the new character. In other words,
281 d = uvchr_to_utf8(d, uv);
283 is the recommended wide native character-aware way of saying
287 This function accepts any UV as input, but very high code points (above
288 C<IV_MAX> on the platform) will raise a deprecation warning. This is
289 typically 0x7FFF_FFFF in a 32-bit word.
291 It is possible to forbid or warn on non-Unicode code points, or those that may
292 be problematic by using L</uvchr_to_utf8_flags>.
297 /* This is also a macro */
298 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
301 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
303 return uvchr_to_utf8(d, uv);
307 =for apidoc uvchr_to_utf8_flags
309 Adds the UTF-8 representation of the native code point C<uv> to the end
310 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
311 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
312 the byte after the end of the new character. In other words,
314 d = uvchr_to_utf8_flags(d, uv, flags);
318 d = uvchr_to_utf8_flags(d, uv, 0);
320 This is the Unicode-aware way of saying
324 If C<flags> is 0, this function accepts any UV as input, but very high code
325 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
326 This is typically 0x7FFF_FFFF in a 32-bit word.
328 Specifying C<flags> can further restrict what is allowed and not warned on, as
331 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
332 the function will raise a warning, provided UTF8 warnings are enabled. If
333 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
334 NULL. If both flags are set, the function will both warn and return NULL.
336 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
337 affect how the function handles a Unicode non-character.
339 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
340 affect the handling of code points that are above the Unicode maximum of
341 0x10FFFF. Languages other than Perl may not be able to accept files that
344 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
345 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
346 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
347 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
348 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
349 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
350 above-Unicode and surrogate flags, but not the non-character ones, as
352 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
353 See L<perlunicode/Noncharacter code points>.
355 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
356 so using them is more problematic than other above-Unicode code points. Perl
357 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
358 likely that non-Perl languages will not be able to read files that contain
359 these that written by the perl interpreter; nor would Perl understand files
360 written by something that uses a different extension. For these reasons, there
361 is a separate set of flags that can warn and/or disallow these extremely high
362 code points, even if other above-Unicode ones are accepted. These are the
363 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
364 are entirely independent from the deprecation warning for code points above
365 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
366 code point that needs more than 31 bits to represent. When that happens,
367 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
368 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
369 above-Unicode code points, including these, as malformations; and
370 C<UNICODE_WARN_SUPER> warns on these.)
372 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
373 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
374 than on ASCII. Prior to that, code points 2**31 and higher were simply
375 unrepresentable, and a different, incompatible method was used to represent
376 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
377 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
378 platforms, warning and disallowing 2**31 and higher.
383 /* This is also a macro */
384 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
387 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
389 return uvchr_to_utf8_flags(d, uv, flags);
392 PERL_STATIC_INLINE bool
393 S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
395 /* Returns TRUE if the first code point represented by the Perl-extended-
396 * UTF-8-encoded string starting at 's', and looking no further than 'e -
397 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
399 * The function handles the case where the input bytes do not include all
400 * the ones necessary to represent a full character. That is, they may be
401 * the intial bytes of the representation of a code point, but possibly
402 * the final ones necessary for the complete representation may be beyond
405 * The function assumes that the sequence is well-formed UTF-8 as far as it
406 * goes, and is for a UTF-8 variant code point. If the sequence is
407 * incomplete, the function returns FALSE if there is any well-formed
408 * UTF-8 byte sequence that can complete it in such a way that a code point
409 * < 2**31 is produced; otherwise it returns TRUE.
411 * Getting this exactly right is slightly tricky, and has to be done in
412 * several places in this file, so is centralized here. It is based on the
415 * U+7FFFFFFF (2 ** 31 - 1)
416 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
417 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
418 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
419 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
420 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
421 * U+80000000 (2 ** 31):
422 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
423 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
424 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
425 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
426 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
432 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
433 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
434 const STRLEN prefix_len = sizeof(prefix) - 1;
435 const STRLEN len = e - s;
436 const STRLEN cmp_len = MIN(prefix_len, len - 1);
444 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
446 assert(! UTF8_IS_INVARIANT(*s));
450 /* Technically, a start byte of FE can be for a code point that fits into
451 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
457 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
458 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
459 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
461 if (*s != 0xFE || len == 1) {
465 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
467 return cBOOL(memGT(s + 1, prefix, cmp_len));
473 PERL_STATIC_INLINE bool
474 S_does_utf8_overflow(const U8 * const s, const U8 * e)
477 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
479 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
481 const STRLEN len = e - s;
485 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
486 * platform, that is if it represents a code point larger than the highest
487 * representable code point. (For ASCII platforms, we could use memcmp()
488 * because we don't have to convert each byte to I8, but it's very rare
489 * input indeed that would approach overflow, so the loop below will likely
490 * only get executed once.
492 * 'e' must not be beyond a full character. If it is less than a full
493 * character, the function returns FALSE if there is any input beyond 'e'
494 * that could result in a non-overflowing code point */
496 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
497 assert(s <= e && s + UTF8SKIP(s) >= e);
499 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
501 /* On 32 bit ASCII machines, many overlongs that start with FF don't
504 if (isFF_OVERLONG(s, len)) {
505 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
506 return memGE(s, max_32_bit_overlong,
507 MIN(len, sizeof(max_32_bit_overlong) - 1));
512 for (x = s; x < e; x++, y++) {
514 /* If this byte is larger than the corresponding highest UTF-8 byte, it
516 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
520 /* If not the same as this byte, it must be smaller, doesn't overflow */
521 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
526 /* Got to the end and all bytes are the same. If the input is a whole
527 * character, it doesn't overflow. And if it is a partial character,
528 * there's not enough information to tell, so assume doesn't overflow */
532 PERL_STATIC_INLINE bool
533 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
535 /* Overlongs can occur whenever the number of continuation bytes
536 * changes. That means whenever the number of leading 1 bits in a start
537 * byte increases from the next lower start byte. That happens for start
538 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
539 * illegal start bytes have already been excluded, so don't need to be
541 * ASCII platforms: C0, C1
542 * EBCDIC platforms C0, C1, C2, C3, C4, E0
544 * At least a second byte is required to determine if other sequences will
547 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
548 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
550 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
551 assert(len > 1 && UTF8_IS_START(*s));
553 /* Each platform has overlongs after the start bytes given above (expressed
554 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
555 * the logic is the same, except the E0 overlong has already been excluded
556 * on EBCDIC platforms. The values below were found by manually
557 * inspecting the UTF-8 patterns. See the tables in utf8.h and
561 # define F0_ABOVE_OVERLONG 0xB0
562 # define F8_ABOVE_OVERLONG 0xA8
563 # define FC_ABOVE_OVERLONG 0xA4
564 # define FE_ABOVE_OVERLONG 0xA2
565 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
569 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
573 # define F0_ABOVE_OVERLONG 0x90
574 # define F8_ABOVE_OVERLONG 0x88
575 # define FC_ABOVE_OVERLONG 0x84
576 # define FE_ABOVE_OVERLONG 0x82
577 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
581 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
582 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
583 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
584 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
589 /* Check for the FF overlong */
590 return isFF_OVERLONG(s, len);
593 PERL_STATIC_INLINE bool
594 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
596 PERL_ARGS_ASSERT_ISFF_OVERLONG;
598 /* Check for the FF overlong. This happens only if all these bytes match;
599 * what comes after them doesn't matter. See tables in utf8.h,
602 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
603 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
604 sizeof(FF_OVERLONG_PREFIX) - 1));
607 #undef F0_ABOVE_OVERLONG
608 #undef F8_ABOVE_OVERLONG
609 #undef FC_ABOVE_OVERLONG
610 #undef FE_ABOVE_OVERLONG
611 #undef FF_OVERLONG_PREFIX
614 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
619 /* A helper function that should not be called directly.
621 * This function returns non-zero if the string beginning at 's' and
622 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
623 * code point; otherwise it returns 0. The examination stops after the
624 * first code point in 's' is validated, not looking at the rest of the
625 * input. If 'e' is such that there are not enough bytes to represent a
626 * complete code point, this function will return non-zero anyway, if the
627 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
628 * excluded by 'flags'.
630 * A non-zero return gives the number of bytes required to represent the
631 * code point. Be aware that if the input is for a partial character, the
632 * return will be larger than 'e - s'.
634 * This function assumes that the code point represented is UTF-8 variant.
635 * The caller should have excluded this possibility before calling this
638 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
639 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
640 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
641 * disallowed by the flags. If the input is only for a partial character,
642 * the function will return non-zero if there is any sequence of
643 * well-formed UTF-8 that, when appended to the input sequence, could
644 * result in an allowed code point; otherwise it returns 0. Non characters
645 * cannot be determined based on partial character input. But many of the
646 * other excluded types can be determined with just the first one or two
651 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
653 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
654 |UTF8_DISALLOW_ABOVE_31_BIT)));
655 assert(! UTF8_IS_INVARIANT(*s));
657 /* A variant char must begin with a start byte */
658 if (UNLIKELY(! UTF8_IS_START(*s))) {
662 /* Examine a maximum of a single whole code point */
663 if (e - s > UTF8SKIP(s)) {
669 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
670 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
672 /* The code below is derived from this table. Keep in mind that legal
673 * continuation bytes range between \x80..\xBF for UTF-8, and
674 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
675 * Hence, we don't have to test the upper edge because if any of those
676 * are encountered, the sequence is malformed, and will fail elsewhere
678 * UTF-8 UTF-EBCDIC I8
679 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
680 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
681 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
685 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
686 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
687 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
689 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
691 && ((s1) & 0xFE ) == 0xB6)
693 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
694 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
695 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
698 if ( (flags & UTF8_DISALLOW_SUPER)
699 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
701 return 0; /* Above Unicode */
704 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
705 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
707 return 0; /* Above 31 bits */
711 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
713 if ( (flags & UTF8_DISALLOW_SUPER)
714 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
716 return 0; /* Above Unicode */
719 if ( (flags & UTF8_DISALLOW_SURROGATE)
720 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
722 return 0; /* Surrogate */
725 if ( (flags & UTF8_DISALLOW_NONCHAR)
726 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
728 return 0; /* Noncharacter code point */
733 /* Make sure that all that follows are continuation bytes */
734 for (x = s + 1; x < e; x++) {
735 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
740 /* Here is syntactically valid. Next, make sure this isn't the start of an
742 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
746 /* And finally, that the code point represented fits in a word on this
748 if (does_utf8_overflow(s, e)) {
756 Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
758 /* Returns a mortalized C string that is a displayable copy of the 'len'
759 * bytes starting at 's'. 'format' gives how to display each byte.
760 * Currently, there are only two formats, so it is currently a bool:
762 * 1 ab (that is a space between two hex digit bytes)
765 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
767 const U8 * const e = s + len;
771 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
773 Newx(output, output_len, char);
778 const unsigned high_nibble = (*s & 0xF0) >> 4;
779 const unsigned low_nibble = (*s & 0x0F);
789 if (high_nibble < 10) {
790 *d++ = high_nibble + '0';
793 *d++ = high_nibble - 10 + 'a';
796 if (low_nibble < 10) {
797 *d++ = low_nibble + '0';
800 *d++ = low_nibble - 10 + 'a';
808 PERL_STATIC_INLINE char *
809 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
811 /* How many bytes to print */
814 /* Which one is the non-continuation */
815 const STRLEN non_cont_byte_pos,
817 /* How many bytes should there be? */
818 const STRLEN expect_len)
820 /* Return the malformation warning text for an unexpected continuation
823 const char * const where = (non_cont_byte_pos == 1)
825 : Perl_form(aTHX_ "%d bytes",
826 (int) non_cont_byte_pos);
828 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
830 /* We don't need to pass this parameter, but since it has already been
831 * calculated, it's likely faster to pass it; verify under DEBUGGING */
832 assert(expect_len == UTF8SKIP(s));
834 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
835 " %s after start byte 0x%02x; need %d bytes, got %d)",
837 _byte_dump_string(s, print_len, 0),
838 *(s + non_cont_byte_pos),
842 (int) non_cont_byte_pos);
847 =for apidoc utf8n_to_uvchr
849 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
850 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
852 Bottom level UTF-8 decode routine.
853 Returns the native code point value of the first character in the string C<s>,
854 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
855 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
856 the length, in bytes, of that character.
858 The value of C<flags> determines the behavior when C<s> does not point to a
859 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
860 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
861 is the next possible position in C<s> that could begin a non-malformed
862 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
863 is raised. Some UTF-8 input sequences may contain multiple malformations.
864 This function tries to find every possible one in each call, so multiple
865 warnings can be raised for each sequence.
867 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
868 individual types of malformations, such as the sequence being overlong (that
869 is, when there is a shorter sequence that can express the same code point;
870 overlong sequences are expressly forbidden in the UTF-8 standard due to
871 potential security issues). Another malformation example is the first byte of
872 a character not being a legal first byte. See F<utf8.h> for the list of such
873 flags. Even if allowed, this function generally returns the Unicode
874 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
875 F<utf8.h> to override this behavior for the overlong malformations, but don't
876 do that except for very specialized purposes.
878 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
879 flags) malformation is found. If this flag is set, the routine assumes that
880 the caller will raise a warning, and this function will silently just set
881 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
883 Note that this API requires disambiguation between successful decoding a C<NUL>
884 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
885 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
886 be set to 1. To disambiguate, upon a zero return, see if the first byte of
887 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
888 error. Or you can use C<L</utf8n_to_uvchr_error>>.
890 Certain code points are considered problematic. These are Unicode surrogates,
891 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
892 By default these are considered regular code points, but certain situations
893 warrant special handling for them, which can be specified using the C<flags>
894 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
895 three classes are treated as malformations and handled as such. The flags
896 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
897 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
898 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
899 restricts the allowed inputs to the strict UTF-8 traditionally defined by
900 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
902 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
903 The difference between traditional strictness and C9 strictness is that the
904 latter does not forbid non-character code points. (They are still discouraged,
905 however.) For more discussion see L<perlunicode/Noncharacter code points>.
907 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
908 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
909 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
910 raised for their respective categories, but otherwise the code points are
911 considered valid (not malformations). To get a category to both be treated as
912 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
913 (But note that warnings are not raised if lexically disabled nor if
914 C<UTF8_CHECK_ONLY> is also specified.)
916 It is now deprecated to have very high code points (above C<IV_MAX> on the
917 platforms) and this function will raise a deprecation warning for these (unless
918 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
921 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
922 so using them is more problematic than other above-Unicode code points. Perl
923 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
924 likely that non-Perl languages will not be able to read files that contain
925 these; nor would Perl understand files
926 written by something that uses a different extension. For these reasons, there
927 is a separate set of flags that can warn and/or disallow these extremely high
928 code points, even if other above-Unicode ones are accepted. These are the
929 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
930 are entirely independent from the deprecation warning for code points above
931 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
932 code point that needs more than 31 bits to represent. When that happens,
933 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
934 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
935 above-Unicode code points, including these, as malformations; and
936 C<UTF8_WARN_SUPER> warns on these.)
938 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
939 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
940 than on ASCII. Prior to that, code points 2**31 and higher were simply
941 unrepresentable, and a different, incompatible method was used to represent
942 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
943 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
944 platforms, warning and disallowing 2**31 and higher.
946 All other code points corresponding to Unicode characters, including private
947 use and those yet to be assigned, are never considered malformed and never
952 Also implemented as a macro in utf8.h
956 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
961 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
963 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
968 =for apidoc utf8n_to_uvchr_error
970 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
971 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
973 This function is for code that needs to know what the precise malformation(s)
974 are when an error is found.
976 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
977 all the others, C<errors>. If this parameter is 0, this function behaves
978 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
979 to a C<U32> variable, which this function sets to indicate any errors found.
980 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
981 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
982 of these bits will be set if a malformation is found, even if the input
983 C<flags> parameter indicates that the given malformation is allowed; those
984 exceptions are noted:
988 =item C<UTF8_GOT_ABOVE_31_BIT>
990 The code point represented by the input UTF-8 sequence occupies more than 31
992 This bit is set only if the input C<flags> parameter contains either the
993 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
995 =item C<UTF8_GOT_CONTINUATION>
997 The input sequence was malformed in that the first byte was a a UTF-8
1000 =item C<UTF8_GOT_EMPTY>
1002 The input C<curlen> parameter was 0.
1004 =item C<UTF8_GOT_LONG>
1006 The input sequence was malformed in that there is some other sequence that
1007 evaluates to the same code point, but that sequence is shorter than this one.
1009 =item C<UTF8_GOT_NONCHAR>
1011 The code point represented by the input UTF-8 sequence is for a Unicode
1012 non-character code point.
1013 This bit is set only if the input C<flags> parameter contains either the
1014 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1016 =item C<UTF8_GOT_NON_CONTINUATION>
1018 The input sequence was malformed in that a non-continuation type byte was found
1019 in a position where only a continuation type one should be.
1021 =item C<UTF8_GOT_OVERFLOW>
1023 The input sequence was malformed in that it is for a code point that is not
1024 representable in the number of bits available in a UV on the current platform.
1026 =item C<UTF8_GOT_SHORT>
1028 The input sequence was malformed in that C<curlen> is smaller than required for
1029 a complete sequence. In other words, the input is for a partial character
1032 =item C<UTF8_GOT_SUPER>
1034 The input sequence was malformed in that it is for a non-Unicode code point;
1035 that is, one above the legal Unicode maximum.
1036 This bit is set only if the input C<flags> parameter contains either the
1037 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1039 =item C<UTF8_GOT_SURROGATE>
1041 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1043 This bit is set only if the input C<flags> parameter contains either the
1044 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1048 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1049 flag to suppress any warnings, and then examine the C<*errors> return.
1055 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1061 const U8 * const s0 = s;
1062 U8 * send = NULL; /* (initialized to silence compilers' wrong
1064 U32 possible_problems = 0; /* A bit is set here for each potential problem
1065 found as we go along */
1067 STRLEN expectlen = 0; /* How long should this sequence be?
1068 (initialized to silence compilers' wrong
1070 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1071 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1072 this gets set and discarded */
1074 /* The below are used only if there is both an overlong malformation and a
1075 * too short one. Otherwise the first two are set to 's0' and 'send', and
1076 * the third not used at all */
1077 U8 * adjusted_s0 = (U8 *) s0;
1078 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1080 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1081 routine; see [perl #130921] */
1082 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1084 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1090 errors = &discard_errors;
1093 /* The order of malformation tests here is important. We should consume as
1094 * few bytes as possible in order to not skip any valid character. This is
1095 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1096 * http://unicode.org/reports/tr36 for more discussion as to why. For
1097 * example, once we've done a UTF8SKIP, we can tell the expected number of
1098 * bytes, and could fail right off the bat if the input parameters indicate
1099 * that there are too few available. But it could be that just that first
1100 * byte is garbled, and the intended character occupies fewer bytes. If we
1101 * blindly assumed that the first byte is correct, and skipped based on
1102 * that number, we could skip over a valid input character. So instead, we
1103 * always examine the sequence byte-by-byte.
1105 * We also should not consume too few bytes, otherwise someone could inject
1106 * things. For example, an input could be deliberately designed to
1107 * overflow, and if this code bailed out immediately upon discovering that,
1108 * returning to the caller C<*retlen> pointing to the very next byte (one
1109 * which is actually part of of the overflowing sequence), that could look
1110 * legitimate to the caller, which could discard the initial partial
1111 * sequence and process the rest, inappropriately.
1113 * Some possible input sequences are malformed in more than one way. This
1114 * function goes to lengths to try to find all of them. This is necessary
1115 * for correctness, as the inputs may allow one malformation but not
1116 * another, and if we abandon searching for others after finding the
1117 * allowed one, we could allow in something that shouldn't have been.
1120 if (UNLIKELY(curlen == 0)) {
1121 possible_problems |= UTF8_GOT_EMPTY;
1123 uv = UNICODE_REPLACEMENT;
1124 goto ready_to_handle_errors;
1127 expectlen = UTF8SKIP(s);
1129 /* A well-formed UTF-8 character, as the vast majority of calls to this
1130 * function will be for, has this expected length. For efficiency, set
1131 * things up here to return it. It will be overriden only in those rare
1132 * cases where a malformation is found */
1134 *retlen = expectlen;
1137 /* An invariant is trivially well-formed */
1138 if (UTF8_IS_INVARIANT(uv)) {
1142 /* A continuation character can't start a valid sequence */
1143 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1144 possible_problems |= UTF8_GOT_CONTINUATION;
1146 uv = UNICODE_REPLACEMENT;
1147 goto ready_to_handle_errors;
1150 /* Here is not a continuation byte, nor an invariant. The only thing left
1151 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1152 * because it excludes start bytes like \xC0 that always lead to
1155 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1156 * that indicate the number of bytes in the character's whole UTF-8
1157 * sequence, leaving just the bits that are part of the value. */
1158 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1160 /* Setup the loop end point, making sure to not look past the end of the
1161 * input string, and flag it as too short if the size isn't big enough. */
1163 if (UNLIKELY(curlen < expectlen)) {
1164 possible_problems |= UTF8_GOT_SHORT;
1171 adjusted_send = send;
1173 /* Now, loop through the remaining bytes in the character's sequence,
1174 * accumulating each into the working value as we go. */
1175 for (s = s0 + 1; s < send; s++) {
1176 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1177 uv = UTF8_ACCUMULATE(uv, *s);
1181 /* Here, found a non-continuation before processing all expected bytes.
1182 * This byte indicates the beginning of a new character, so quit, even
1183 * if allowing this malformation. */
1184 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1186 } /* End of loop through the character's bytes */
1188 /* Save how many bytes were actually in the character */
1191 /* Note that there are two types of too-short malformation. One is when
1192 * there is actual wrong data before the normal termination of the
1193 * sequence. The other is that the sequence wasn't complete before the end
1194 * of the data we are allowed to look at, based on the input 'curlen'.
1195 * This means that we were passed data for a partial character, but it is
1196 * valid as far as we saw. The other is definitely invalid. This
1197 * distinction could be important to a caller, so the two types are kept
1200 * A convenience macro that matches either of the too-short conditions. */
1201 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1203 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1205 uv = UNICODE_REPLACEMENT;
1208 /* Check for overflow */
1209 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1210 possible_problems |= UTF8_GOT_OVERFLOW;
1211 uv = UNICODE_REPLACEMENT;
1214 /* Check for overlong. If no problems so far, 'uv' is the correct code
1215 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1216 * we must look at the UTF-8 byte sequence itself to see if it is for an
1218 if ( ( LIKELY(! possible_problems)
1219 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1220 || ( UNLIKELY( possible_problems)
1221 && ( UNLIKELY(! UTF8_IS_START(*s0))
1223 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1226 possible_problems |= UTF8_GOT_LONG;
1228 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1229 UV min_uv = uv_so_far;
1232 /* Here, the input is both overlong and is missing some trailing
1233 * bytes. There is no single code point it could be for, but there
1234 * may be enough information present to determine if what we have
1235 * so far is for an unallowed code point, such as for a surrogate.
1236 * The code below has the intelligence to determine this, but just
1237 * for non-overlong UTF-8 sequences. What we do here is calculate
1238 * the smallest code point the input could represent if there were
1239 * no too short malformation. Then we compute and save the UTF-8
1240 * for that, which is what the code below looks at instead of the
1241 * raw input. It turns out that the smallest such code point is
1243 for (i = curlen; i < expectlen; i++) {
1244 min_uv = UTF8_ACCUMULATE(min_uv,
1245 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1248 adjusted_s0 = temp_char_buf;
1249 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1253 /* Now check that the input isn't for a problematic code point not allowed
1254 * by the input parameters. */
1255 /* isn't problematic if < this */
1256 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1257 || ( UNLIKELY(possible_problems)
1259 /* if overflow, we know without looking further
1260 * precisely which of the problematic types it is,
1261 * and we deal with those in the overflow handling
1263 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1264 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1265 && ((flags & ( UTF8_DISALLOW_NONCHAR
1266 |UTF8_DISALLOW_SURROGATE
1267 |UTF8_DISALLOW_SUPER
1268 |UTF8_DISALLOW_ABOVE_31_BIT
1270 |UTF8_WARN_SURROGATE
1272 |UTF8_WARN_ABOVE_31_BIT))
1273 /* In case of a malformation, 'uv' is not valid, and has
1274 * been changed to something in the Unicode range.
1275 * Currently we don't output a deprecation message if there
1276 * is already a malformation, so we don't have to special
1277 * case the test immediately below */
1278 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1279 && ckWARN_d(WARN_DEPRECATED))))
1281 /* If there were no malformations, or the only malformation is an
1282 * overlong, 'uv' is valid */
1283 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1284 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1285 possible_problems |= UTF8_GOT_SURROGATE;
1287 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1288 possible_problems |= UTF8_GOT_SUPER;
1290 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1291 possible_problems |= UTF8_GOT_NONCHAR;
1294 else { /* Otherwise, need to look at the source UTF-8, possibly
1295 adjusted to be non-overlong */
1297 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1298 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1300 possible_problems |= UTF8_GOT_SUPER;
1302 else if (curlen > 1) {
1303 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1304 NATIVE_UTF8_TO_I8(*adjusted_s0),
1305 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1307 possible_problems |= UTF8_GOT_SUPER;
1309 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1310 NATIVE_UTF8_TO_I8(*adjusted_s0),
1311 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1313 possible_problems |= UTF8_GOT_SURROGATE;
1317 /* We need a complete well-formed UTF-8 character to discern
1318 * non-characters, so can't look for them here */
1322 ready_to_handle_errors:
1325 * curlen contains the number of bytes in the sequence that
1326 * this call should advance the input by.
1327 * avail_len gives the available number of bytes passed in, but
1328 * only if this is less than the expected number of
1329 * bytes, based on the code point's start byte.
1330 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1331 * is set in it for each potential problem found.
1332 * uv contains the code point the input sequence
1333 * represents; or if there is a problem that prevents
1334 * a well-defined value from being computed, it is
1335 * some subsitute value, typically the REPLACEMENT
1337 * s0 points to the first byte of the character
1338 * send points to just after where that (potentially
1339 * partial) character ends
1340 * adjusted_s0 normally is the same as s0, but in case of an
1341 * overlong for which the UTF-8 matters below, it is
1342 * the first byte of the shortest form representation
1344 * adjusted_send normally is the same as 'send', but if adjusted_s0
1345 * is set to something other than s0, this points one
1349 if (UNLIKELY(possible_problems)) {
1350 bool disallowed = FALSE;
1351 const U32 orig_problems = possible_problems;
1353 while (possible_problems) { /* Handle each possible problem */
1355 char * message = NULL;
1357 /* Each 'if' clause handles one problem. They are ordered so that
1358 * the first ones' messages will be displayed before the later
1359 * ones; this is kinda in decreasing severity order */
1360 if (possible_problems & UTF8_GOT_OVERFLOW) {
1362 /* Overflow means also got a super and above 31 bits, but we
1363 * handle all three cases here */
1365 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1366 *errors |= UTF8_GOT_OVERFLOW;
1368 /* But the API says we flag all errors found */
1369 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1370 *errors |= UTF8_GOT_SUPER;
1373 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1375 *errors |= UTF8_GOT_ABOVE_31_BIT;
1378 /* Disallow if any of the three categories say to */
1379 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1380 || (flags & ( UTF8_DISALLOW_SUPER
1381 |UTF8_DISALLOW_ABOVE_31_BIT)))
1387 /* Likewise, warn if any say to, plus if deprecation warnings
1388 * are on, because this code point is above IV_MAX */
1389 if ( ckWARN_d(WARN_DEPRECATED)
1390 || ! (flags & UTF8_ALLOW_OVERFLOW)
1391 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1394 /* The warnings code explicitly says it doesn't handle the
1395 * case of packWARN2 and two categories which have
1396 * parent-child relationship. Even if it works now to
1397 * raise the warning if either is enabled, it wouldn't
1398 * necessarily do so in the future. We output (only) the
1399 * most dire warning*/
1400 if (! (flags & UTF8_CHECK_ONLY)) {
1401 if (ckWARN_d(WARN_UTF8)) {
1402 pack_warn = packWARN(WARN_UTF8);
1404 else if (ckWARN_d(WARN_NON_UNICODE)) {
1405 pack_warn = packWARN(WARN_NON_UNICODE);
1408 message = Perl_form(aTHX_ "%s: %s (overflows)",
1410 _byte_dump_string(s0, send - s0, 0));
1415 else if (possible_problems & UTF8_GOT_EMPTY) {
1416 possible_problems &= ~UTF8_GOT_EMPTY;
1417 *errors |= UTF8_GOT_EMPTY;
1419 if (! (flags & UTF8_ALLOW_EMPTY)) {
1421 /* This so-called malformation is now treated as a bug in
1422 * the caller. If you have nothing to decode, skip calling
1427 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1428 pack_warn = packWARN(WARN_UTF8);
1429 message = Perl_form(aTHX_ "%s (empty string)",
1434 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1435 possible_problems &= ~UTF8_GOT_CONTINUATION;
1436 *errors |= UTF8_GOT_CONTINUATION;
1438 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1440 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1441 pack_warn = packWARN(WARN_UTF8);
1442 message = Perl_form(aTHX_
1443 "%s: %s (unexpected continuation byte 0x%02x,"
1444 " with no preceding start byte)",
1446 _byte_dump_string(s0, 1, 0), *s0);
1450 else if (possible_problems & UTF8_GOT_SHORT) {
1451 possible_problems &= ~UTF8_GOT_SHORT;
1452 *errors |= UTF8_GOT_SHORT;
1454 if (! (flags & UTF8_ALLOW_SHORT)) {
1456 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1457 pack_warn = packWARN(WARN_UTF8);
1458 message = Perl_form(aTHX_
1459 "%s: %s (too short; %d byte%s available, need %d)",
1461 _byte_dump_string(s0, send - s0, 0),
1463 avail_len == 1 ? "" : "s",
1469 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1470 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1471 *errors |= UTF8_GOT_NON_CONTINUATION;
1473 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1475 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1477 /* If we don't know for sure that the input length is
1478 * valid, avoid as much as possible reading past the
1479 * end of the buffer */
1480 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1483 pack_warn = packWARN(WARN_UTF8);
1484 message = Perl_form(aTHX_ "%s",
1485 unexpected_non_continuation_text(s0,
1492 else if (possible_problems & UTF8_GOT_LONG) {
1493 possible_problems &= ~UTF8_GOT_LONG;
1494 *errors |= UTF8_GOT_LONG;
1496 if (flags & UTF8_ALLOW_LONG) {
1498 /* We don't allow the actual overlong value, unless the
1499 * special extra bit is also set */
1500 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1501 & ~UTF8_ALLOW_LONG)))
1503 uv = UNICODE_REPLACEMENT;
1509 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1510 pack_warn = packWARN(WARN_UTF8);
1512 /* These error types cause 'uv' to be something that
1513 * isn't what was intended, so can't use it in the
1514 * message. The other error types either can't
1515 * generate an overlong, or else the 'uv' is valid */
1517 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1519 message = Perl_form(aTHX_
1520 "%s: %s (any UTF-8 sequence that starts"
1521 " with \"%s\" is overlong which can and"
1522 " should be represented with a"
1523 " different, shorter sequence)",
1525 _byte_dump_string(s0, send - s0, 0),
1526 _byte_dump_string(s0, curlen, 0));
1529 U8 tmpbuf[UTF8_MAXBYTES+1];
1530 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1532 message = Perl_form(aTHX_
1533 "%s: %s (overlong; instead use %s to represent"
1536 _byte_dump_string(s0, send - s0, 0),
1537 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1538 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1539 small code points */
1545 else if (possible_problems & UTF8_GOT_SURROGATE) {
1546 possible_problems &= ~UTF8_GOT_SURROGATE;
1548 if (flags & UTF8_WARN_SURROGATE) {
1549 *errors |= UTF8_GOT_SURROGATE;
1551 if ( ! (flags & UTF8_CHECK_ONLY)
1552 && ckWARN_d(WARN_SURROGATE))
1554 pack_warn = packWARN(WARN_SURROGATE);
1556 /* These are the only errors that can occur with a
1557 * surrogate when the 'uv' isn't valid */
1558 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1559 message = Perl_form(aTHX_
1560 "UTF-16 surrogate (any UTF-8 sequence that"
1561 " starts with \"%s\" is for a surrogate)",
1562 _byte_dump_string(s0, curlen, 0));
1565 message = Perl_form(aTHX_
1566 "UTF-16 surrogate U+%04" UVXf, uv);
1571 if (flags & UTF8_DISALLOW_SURROGATE) {
1573 *errors |= UTF8_GOT_SURROGATE;
1576 else if (possible_problems & UTF8_GOT_SUPER) {
1577 possible_problems &= ~UTF8_GOT_SUPER;
1579 if (flags & UTF8_WARN_SUPER) {
1580 *errors |= UTF8_GOT_SUPER;
1582 if ( ! (flags & UTF8_CHECK_ONLY)
1583 && ckWARN_d(WARN_NON_UNICODE))
1585 pack_warn = packWARN(WARN_NON_UNICODE);
1587 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1588 message = Perl_form(aTHX_
1589 "Any UTF-8 sequence that starts with"
1590 " \"%s\" is for a non-Unicode code point,"
1591 " may not be portable",
1592 _byte_dump_string(s0, curlen, 0));
1595 message = Perl_form(aTHX_
1596 "Code point 0x%04" UVXf " is not"
1597 " Unicode, may not be portable",
1603 /* The maximum code point ever specified by a standard was
1604 * 2**31 - 1. Anything larger than that is a Perl extension
1605 * that very well may not be understood by other applications
1606 * (including earlier perl versions on EBCDIC platforms). We
1607 * test for these after the regular SUPER ones, and before
1608 * possibly bailing out, so that the slightly more dire warning
1609 * will override the regular one. */
1610 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1612 |UTF8_DISALLOW_ABOVE_31_BIT))
1613 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1614 && UNLIKELY(is_utf8_cp_above_31_bits(
1617 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1618 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1620 if ( ! (flags & UTF8_CHECK_ONLY)
1621 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1622 && ckWARN_d(WARN_UTF8))
1624 pack_warn = packWARN(WARN_UTF8);
1626 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1627 message = Perl_form(aTHX_
1628 "Any UTF-8 sequence that starts with"
1629 " \"%s\" is for a non-Unicode code"
1630 " point, and is not portable",
1631 _byte_dump_string(s0, curlen, 0));
1634 message = Perl_form(aTHX_
1635 "Code point 0x%" UVXf " is not Unicode,"
1636 " and not portable",
1641 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1642 |UTF8_DISALLOW_ABOVE_31_BIT))
1644 *errors |= UTF8_GOT_ABOVE_31_BIT;
1646 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1652 if (flags & UTF8_DISALLOW_SUPER) {
1653 *errors |= UTF8_GOT_SUPER;
1657 /* The deprecated warning overrides any non-deprecated one. If
1658 * there are other problems, a deprecation message is not
1659 * really helpful, so don't bother to raise it in that case.
1660 * This also keeps the code from having to handle the case
1661 * where 'uv' is not valid. */
1662 if ( ! (orig_problems
1663 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1664 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)) {
1665 Perl_croak(aTHX_ cp_above_legal_max, uv,
1666 MAX_NON_DEPRECATED_CP);
1669 else if (possible_problems & UTF8_GOT_NONCHAR) {
1670 possible_problems &= ~UTF8_GOT_NONCHAR;
1672 if (flags & UTF8_WARN_NONCHAR) {
1673 *errors |= UTF8_GOT_NONCHAR;
1675 if ( ! (flags & UTF8_CHECK_ONLY)
1676 && ckWARN_d(WARN_NONCHAR))
1678 /* The code above should have guaranteed that we don't
1679 * get here with errors other than overlong */
1680 assert (! (orig_problems
1681 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1683 pack_warn = packWARN(WARN_NONCHAR);
1684 message = Perl_form(aTHX_ "Unicode non-character"
1685 " U+%04" UVXf " is not recommended"
1686 " for open interchange", uv);
1690 if (flags & UTF8_DISALLOW_NONCHAR) {
1692 *errors |= UTF8_GOT_NONCHAR;
1694 } /* End of looking through the possible flags */
1696 /* Display the message (if any) for the problem being handled in
1697 * this iteration of the loop */
1700 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1703 Perl_warner(aTHX_ pack_warn, "%s", message);
1705 } /* End of 'while (possible_problems)' */
1707 /* Since there was a possible problem, the returned length may need to
1708 * be changed from the one stored at the beginning of this function.
1709 * Instead of trying to figure out if that's needed, just do it. */
1715 if (flags & UTF8_CHECK_ONLY && retlen) {
1716 *retlen = ((STRLEN) -1);
1722 return UNI_TO_NATIVE(uv);
1726 =for apidoc utf8_to_uvchr_buf
1728 Returns the native code point of the first character in the string C<s> which
1729 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1730 C<*retlen> will be set to the length, in bytes, of that character.
1732 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1733 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1734 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1735 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1736 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1737 the next possible position in C<s> that could begin a non-malformed character.
1738 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1741 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1742 unless those are turned off.
1746 Also implemented as a macro in utf8.h
1752 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1754 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1758 return utf8n_to_uvchr(s, send - s, retlen,
1759 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1762 /* This is marked as deprecated
1764 =for apidoc utf8_to_uvuni_buf
1766 Only in very rare circumstances should code need to be dealing in Unicode
1767 (as opposed to native) code points. In those few cases, use
1768 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1770 Returns the Unicode (not-native) code point of the first character in the
1772 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1773 C<retlen> will be set to the length, in bytes, of that character.
1775 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1776 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1777 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1778 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1779 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1780 next possible position in C<s> that could begin a non-malformed character.
1781 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1783 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1784 unless those are turned off.
1790 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1792 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1796 /* Call the low level routine, asking for checks */
1797 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1801 =for apidoc utf8_length
1803 Return the length of the UTF-8 char encoded string C<s> in characters.
1804 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1805 up past C<e>, croaks.
1811 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1815 PERL_ARGS_ASSERT_UTF8_LENGTH;
1817 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1818 * the bitops (especially ~) can create illegal UTF-8.
1819 * In other words: in Perl UTF-8 is not just for Unicode. */
1822 goto warn_and_return;
1832 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1833 "%s in %s", unees, OP_DESC(PL_op));
1835 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1842 =for apidoc bytes_cmp_utf8
1844 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1845 sequence of characters (stored as UTF-8)
1846 in C<u>, C<ulen>. Returns 0 if they are
1847 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1848 if the first string is greater than the second string.
1850 -1 or +1 is returned if the shorter string was identical to the start of the
1851 longer string. -2 or +2 is returned if
1852 there was a difference between characters
1859 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1861 const U8 *const bend = b + blen;
1862 const U8 *const uend = u + ulen;
1864 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1866 while (b < bend && u < uend) {
1868 if (!UTF8_IS_INVARIANT(c)) {
1869 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1872 if (UTF8_IS_CONTINUATION(c1)) {
1873 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1875 /* diag_listed_as: Malformed UTF-8 character%s */
1876 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1878 unexpected_non_continuation_text(u - 1, 2, 1, 2),
1879 PL_op ? " in " : "",
1880 PL_op ? OP_DESC(PL_op) : "");
1885 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1886 "%s in %s", unees, OP_DESC(PL_op));
1888 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1889 return -2; /* Really want to return undef :-) */
1896 return *b < c ? -2 : +2;
1901 if (b == bend && u == uend)
1904 return b < bend ? +1 : -1;
1908 =for apidoc utf8_to_bytes
1910 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1911 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1912 updates C<len> to contain the new length.
1913 Returns zero on failure, setting C<len> to -1.
1915 If you need a copy of the string, see L</bytes_from_utf8>.
1921 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1923 U8 * const save = s;
1924 U8 * const send = s + *len;
1927 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1928 PERL_UNUSED_CONTEXT;
1930 /* ensure valid UTF-8 and chars < 256 before updating string */
1932 if (! UTF8_IS_INVARIANT(*s)) {
1933 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1934 *len = ((STRLEN) -1);
1945 if (! UTF8_IS_INVARIANT(c)) {
1946 /* Then it is two-byte encoded */
1947 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1958 =for apidoc bytes_from_utf8
1960 Converts a potentially UTF-8 encoded string C<s> of length C<len> into native
1961 byte encoding. On input, the boolean C<*is_utf8> gives whether or not C<s> is
1962 actually encoded in UTF-8.
1964 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
1967 Do nothing if C<*is_utf8> is 0, or if there are code points in the string
1968 not expressible in native byte encoding. In these cases, C<*is_utf8> and
1969 C<*len> are unchanged, and the return value is the original C<s>.
1971 Otherwise, C<*is_utf8> is set to 0, and the return value is a pointer to a
1972 newly created string containing a downgraded copy of C<s>, and whose length is
1973 returned in C<*len>, updated.
1979 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1982 const U8 *start = s;
1986 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1987 PERL_UNUSED_CONTEXT;
1991 /* ensure valid UTF-8 and chars < 256 before converting string */
1992 for (send = s + *len; s < send;) {
1993 if (! UTF8_IS_INVARIANT(*s)) {
1994 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2005 Newx(d, (*len) - count + 1, U8);
2006 s = start; start = d;
2009 if (! UTF8_IS_INVARIANT(c)) {
2010 /* Then it is two-byte encoded */
2011 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2022 =for apidoc bytes_to_utf8
2024 Converts a string C<s> of length C<len> bytes from the native encoding into
2026 Returns a pointer to the newly-created string, and sets C<len> to
2027 reflect the new length in bytes.
2029 A C<NUL> character will be written after the end of the string.
2031 If you want to convert to UTF-8 from encodings other than
2032 the native (Latin1 or EBCDIC),
2033 see L</sv_recode_to_utf8>().
2038 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2039 likewise need duplication. */
2042 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
2044 const U8 * const send = s + (*len);
2048 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2049 PERL_UNUSED_CONTEXT;
2051 Newx(d, (*len) * 2 + 1, U8);
2055 append_utf8_from_native_byte(*s, &d);
2064 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2066 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2067 * We optimize for native, for obvious reasons. */
2070 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2075 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2078 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
2083 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2085 if (OFFUNI_IS_INVARIANT(uv)) {
2086 *d++ = LATIN1_TO_NATIVE((U8) uv);
2089 if (uv <= MAX_UTF8_TWO_BYTE) {
2090 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2091 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2094 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2095 #define LAST_HIGH_SURROGATE 0xDBFF
2096 #define FIRST_LOW_SURROGATE 0xDC00
2097 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2099 /* This assumes that most uses will be in the first Unicode plane, not
2100 * needing surrogates */
2101 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2102 && uv <= UNICODE_SURROGATE_LAST))
2104 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2105 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2108 UV low = (p[0] << 8) + p[1];
2109 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2110 || UNLIKELY(low > LAST_LOW_SURROGATE))
2112 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2115 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2116 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2120 d = uvoffuni_to_utf8_flags(d, uv, 0);
2123 *d++ = (U8)(( uv >> 12) | 0xe0);
2124 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2125 *d++ = (U8)(( uv & 0x3f) | 0x80);
2129 *d++ = (U8)(( uv >> 18) | 0xf0);
2130 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2131 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2132 *d++ = (U8)(( uv & 0x3f) | 0x80);
2137 *newlen = d - dstart;
2141 /* Note: this one is slightly destructive of the source. */
2144 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2147 U8* const send = s + bytelen;
2149 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2152 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2156 const U8 tmp = s[0];
2161 return utf16_to_utf8(p, d, bytelen, newlen);
2165 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2167 U8 tmpbuf[UTF8_MAXBYTES+1];
2168 uvchr_to_utf8(tmpbuf, c);
2169 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2172 /* Internal function so we can deprecate the external one, and call
2173 this one from other deprecated functions in this file */
2176 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2178 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2182 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2186 Perl__is_uni_perl_idcont(pTHX_ UV c)
2188 U8 tmpbuf[UTF8_MAXBYTES+1];
2189 uvchr_to_utf8(tmpbuf, c);
2190 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2194 Perl__is_uni_perl_idstart(pTHX_ UV c)
2196 U8 tmpbuf[UTF8_MAXBYTES+1];
2197 uvchr_to_utf8(tmpbuf, c);
2198 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2202 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2204 /* We have the latin1-range values compiled into the core, so just use
2205 * those, converting the result to UTF-8. The only difference between upper
2206 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2207 * either "SS" or "Ss". Which one to use is passed into the routine in
2208 * 'S_or_s' to avoid a test */
2210 UV converted = toUPPER_LATIN1_MOD(c);
2212 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2214 assert(S_or_s == 'S' || S_or_s == 's');
2216 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2217 characters in this range */
2218 *p = (U8) converted;
2223 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2224 * which it maps to one of them, so as to only have to have one check for
2225 * it in the main case */
2226 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2228 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2229 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2232 converted = GREEK_CAPITAL_LETTER_MU;
2234 #if UNICODE_MAJOR_VERSION > 2 \
2235 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2236 && UNICODE_DOT_DOT_VERSION >= 8)
2237 case LATIN_SMALL_LETTER_SHARP_S:
2244 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2245 NOT_REACHED; /* NOTREACHED */
2249 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2250 *p = UTF8_TWO_BYTE_LO(converted);
2256 /* Call the function to convert a UTF-8 encoded character to the specified case.
2257 * Note that there may be more than one character in the result.
2258 * INP is a pointer to the first byte of the input character
2259 * OUTP will be set to the first byte of the string of changed characters. It
2260 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2261 * LENP will be set to the length in bytes of the string of changed characters
2263 * The functions return the ordinal of the first character in the string of OUTP */
2264 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2265 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2266 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2268 /* This additionally has the input parameter 'specials', which if non-zero will
2269 * cause this to use the specials hash for folding (meaning get full case
2270 * folding); otherwise, when zero, this implies a simple case fold */
2271 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2274 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2276 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2277 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2278 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2279 * the changed version may be longer than the original character.
2281 * The ordinal of the first character of the changed version is returned
2282 * (but note, as explained above, that there may be more.) */
2284 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2287 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2290 uvchr_to_utf8(p, c);
2291 return CALL_UPPER_CASE(c, p, p, lenp);
2295 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2297 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2300 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2303 uvchr_to_utf8(p, c);
2304 return CALL_TITLE_CASE(c, p, p, lenp);
2308 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2310 /* We have the latin1-range values compiled into the core, so just use
2311 * those, converting the result to UTF-8. Since the result is always just
2312 * one character, we allow <p> to be NULL */
2314 U8 converted = toLOWER_LATIN1(c);
2316 PERL_UNUSED_ARG(dummy);
2319 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2324 /* Result is known to always be < 256, so can use the EIGHT_BIT
2326 *p = UTF8_EIGHT_BIT_HI(converted);
2327 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2335 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2337 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2340 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2343 uvchr_to_utf8(p, c);
2344 return CALL_LOWER_CASE(c, p, p, lenp);
2348 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2350 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2351 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2352 * FOLD_FLAGS_FULL iff full folding is to be used;
2354 * Not to be used for locale folds
2359 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2360 PERL_UNUSED_CONTEXT;
2362 assert (! (flags & FOLD_FLAGS_LOCALE));
2364 if (UNLIKELY(c == MICRO_SIGN)) {
2365 converted = GREEK_SMALL_LETTER_MU;
2367 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2368 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2369 || UNICODE_DOT_DOT_VERSION > 0)
2370 else if ( (flags & FOLD_FLAGS_FULL)
2371 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2373 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2374 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2375 * under those circumstances. */
2376 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2377 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2378 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2380 return LATIN_SMALL_LETTER_LONG_S;
2390 else { /* In this range the fold of all other characters is their lower
2392 converted = toLOWER_LATIN1(c);
2395 if (UVCHR_IS_INVARIANT(converted)) {
2396 *p = (U8) converted;
2400 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2401 *p = UTF8_TWO_BYTE_LO(converted);
2409 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2412 /* Not currently externally documented, and subject to change
2413 * <flags> bits meanings:
2414 * FOLD_FLAGS_FULL iff full folding is to be used;
2415 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2416 * locale are to be used.
2417 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2420 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2422 if (flags & FOLD_FLAGS_LOCALE) {
2423 /* Treat a UTF-8 locale as not being in locale at all */
2424 if (IN_UTF8_CTYPE_LOCALE) {
2425 flags &= ~FOLD_FLAGS_LOCALE;
2428 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2429 goto needs_full_generality;
2434 return _to_fold_latin1((U8) c, p, lenp,
2435 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2438 /* Here, above 255. If no special needs, just use the macro */
2439 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2440 uvchr_to_utf8(p, c);
2441 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2443 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2444 the special flags. */
2445 U8 utf8_c[UTF8_MAXBYTES + 1];
2447 needs_full_generality:
2448 uvchr_to_utf8(utf8_c, c);
2449 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
2453 PERL_STATIC_INLINE bool
2454 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2455 const char *const swashname, SV* const invlist)
2457 /* returns a boolean giving whether or not the UTF8-encoded character that
2458 * starts at <p> is in the swash indicated by <swashname>. <swash>
2459 * contains a pointer to where the swash indicated by <swashname>
2460 * is to be stored; which this routine will do, so that future calls will
2461 * look at <*swash> and only generate a swash if it is not null. <invlist>
2462 * is NULL or an inversion list that defines the swash. If not null, it
2463 * saves time during initialization of the swash.
2465 * Note that it is assumed that the buffer length of <p> is enough to
2466 * contain all the bytes that comprise the character. Thus, <*p> should
2467 * have been checked before this call for mal-formedness enough to assure
2470 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2472 /* The API should have included a length for the UTF-8 character in <p>,
2473 * but it doesn't. We therefore assume that p has been validated at least
2474 * as far as there being enough bytes available in it to accommodate the
2475 * character without reading beyond the end, and pass that number on to the
2476 * validating routine */
2477 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2478 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2479 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2481 NOT_REACHED; /* NOTREACHED */
2485 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2486 *swash = _core_swash_init("utf8",
2488 /* Only use the name if there is no inversion
2489 * list; otherwise will go out to disk */
2490 (invlist) ? "" : swashname,
2492 &PL_sv_undef, 1, 0, invlist, &flags);
2495 return swash_fetch(*swash, p, TRUE) != 0;
2498 PERL_STATIC_INLINE bool
2499 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2500 const char *const swashname, SV* const invlist)
2502 /* returns a boolean giving whether or not the UTF8-encoded character that
2503 * starts at <p>, and extending no further than <e - 1> is in the swash
2504 * indicated by <swashname>. <swash> contains a pointer to where the swash
2505 * indicated by <swashname> is to be stored; which this routine will do, so
2506 * that future calls will look at <*swash> and only generate a swash if it
2507 * is not null. <invlist> is NULL or an inversion list that defines the
2508 * swash. If not null, it saves time during initialization of the swash.
2511 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2513 if (! isUTF8_CHAR(p, e)) {
2514 _force_out_malformed_utf8_message(p, e, 0, 1);
2515 NOT_REACHED; /* NOTREACHED */
2519 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2520 *swash = _core_swash_init("utf8",
2522 /* Only use the name if there is no inversion
2523 * list; otherwise will go out to disk */
2524 (invlist) ? "" : swashname,
2526 &PL_sv_undef, 1, 0, invlist, &flags);
2529 return swash_fetch(*swash, p, TRUE) != 0;
2533 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2534 const char * const alternative,
2535 const bool use_locale,
2536 const char * const file,
2537 const unsigned line)
2541 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2543 if (ckWARN_d(WARN_DEPRECATED)) {
2545 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2546 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2547 if (! PL_seen_deprecated_macro) {
2548 PL_seen_deprecated_macro = newHV();
2550 if (! hv_store(PL_seen_deprecated_macro, key,
2551 strlen(key), &PL_sv_undef, 0))
2553 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2556 if (instr(file, "mathoms.c")) {
2557 Perl_warner(aTHX_ WARN_DEPRECATED,
2558 "In %s, line %d, starting in Perl v5.30, %s()"
2559 " will be removed. Avoid this message by"
2560 " converting to use %s().\n",
2561 file, line, name, alternative);
2564 Perl_warner(aTHX_ WARN_DEPRECATED,
2565 "In %s, line %d, starting in Perl v5.30, %s() will"
2566 " require an additional parameter. Avoid this"
2567 " message by converting to use %s().\n",
2568 file, line, name, alternative);
2575 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2577 const char * const name,
2578 const char * const alternative,
2579 const bool use_utf8,
2580 const bool use_locale,
2581 const char * const file,
2582 const unsigned line)
2584 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2586 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2588 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2598 case _CC_ALPHANUMERIC:
2602 return is_utf8_common(p,
2603 &PL_utf8_swash_ptrs[classnum],
2604 swash_property_names[classnum],
2605 PL_XPosix_ptrs[classnum]);
2608 return is_XPERLSPACE_high(p);
2610 return is_HORIZWS_high(p);
2612 return is_XDIGIT_high(p);
2618 return is_VERTWS_high(p);
2620 if (! PL_utf8_perl_idstart) {
2621 PL_utf8_perl_idstart
2622 = _new_invlist_C_array(_Perl_IDStart_invlist);
2624 return is_utf8_common(p, &PL_utf8_perl_idstart,
2625 "_Perl_IDStart", NULL);
2627 if (! PL_utf8_perl_idcont) {
2629 = _new_invlist_C_array(_Perl_IDCont_invlist);
2631 return is_utf8_common(p, &PL_utf8_perl_idcont,
2632 "_Perl_IDCont", NULL);
2636 /* idcont is the same as wordchar below 256 */
2637 if (classnum == _CC_IDCONT) {
2638 classnum = _CC_WORDCHAR;
2640 else if (classnum == _CC_IDFIRST) {
2644 classnum = _CC_ALPHA;
2648 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2649 return _generic_isCC(*p, classnum);
2652 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2655 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2656 return isFOO_lc(classnum, *p);
2659 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2662 NOT_REACHED; /* NOTREACHED */
2666 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2669 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2671 assert(classnum < _FIRST_NON_SWASH_CC);
2673 return is_utf8_common_with_len(p,
2675 &PL_utf8_swash_ptrs[classnum],
2676 swash_property_names[classnum],
2677 PL_XPosix_ptrs[classnum]);
2681 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2685 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2687 if (! PL_utf8_perl_idstart) {
2688 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2690 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2691 "_Perl_IDStart", invlist);
2695 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2697 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2701 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2705 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2709 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2711 if (! PL_utf8_perl_idcont) {
2712 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2714 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2715 "_Perl_IDCont", invlist);
2719 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2721 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2723 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2727 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2729 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2731 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2735 Perl__is_utf8_mark(pTHX_ const U8 *p)
2737 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2739 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2742 /* change namve uv1 to 'from' */
2744 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2745 SV **swashp, const char *normal, const char *special)
2749 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2751 /* For code points that don't change case, we already know that the output
2752 * of this function is the unchanged input, so we can skip doing look-ups
2753 * for them. Unfortunately the case-changing code points are scattered
2754 * around. But there are some long consecutive ranges where there are no
2755 * case changing code points. By adding tests, we can eliminate the lookup
2756 * for all the ones in such ranges. This is currently done here only for
2757 * just a few cases where the scripts are in common use in modern commerce
2758 * (and scripts adjacent to those which can be included without additional
2761 if (uv1 >= 0x0590) {
2762 /* This keeps from needing further processing the code points most
2763 * likely to be used in the following non-cased scripts: Hebrew,
2764 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2765 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2766 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2771 /* The following largish code point ranges also don't have case
2772 * changes, but khw didn't think they warranted extra tests to speed
2773 * them up (which would slightly slow down everything else above them):
2774 * 1100..139F Hangul Jamo, Ethiopic
2775 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2776 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2777 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2778 * Combining Diacritical Marks Extended, Balinese,
2779 * Sundanese, Batak, Lepcha, Ol Chiki
2780 * 2000..206F General Punctuation
2783 if (uv1 >= 0x2D30) {
2785 /* This keeps the from needing further processing the code points
2786 * most likely to be used in the following non-cased major scripts:
2787 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2789 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2790 * event that Unicode eventually allocates the unused block as of
2791 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2792 * that the test suite will start having failures to alert you
2793 * should that happen) */
2798 if (uv1 >= 0xAC00) {
2799 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2800 if (ckWARN_d(WARN_SURROGATE)) {
2801 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2802 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2803 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2808 /* AC00..FAFF Catches Hangul syllables and private use, plus
2815 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2816 if (UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)) {
2817 Perl_croak(aTHX_ cp_above_legal_max, uv1,
2818 MAX_NON_DEPRECATED_CP);
2820 if (ckWARN_d(WARN_NON_UNICODE)) {
2821 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2822 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2823 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2827 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2829 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2832 /* As of this writing, this means we avoid swash creation
2833 * for anything beyond low Plane 1 */
2840 /* Note that non-characters are perfectly legal, so no warning should
2841 * be given. There are so few of them, that it isn't worth the extra
2842 * tests to avoid swash creation */
2845 if (!*swashp) /* load on-demand */
2846 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2849 /* It might be "special" (sometimes, but not always,
2850 * a multicharacter mapping) */
2854 /* If passed in the specials name, use that; otherwise use any
2855 * given in the swash */
2856 if (*special != '\0') {
2857 hv = get_hv(special, 0);
2860 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2862 hv = MUTABLE_HV(SvRV(*svp));
2867 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2872 s = SvPV_const(*svp, len);
2875 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2877 Copy(s, ustrp, len, U8);
2882 if (!len && *swashp) {
2883 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2886 /* It was "normal" (a single character mapping). */
2887 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2895 return valid_utf8_to_uvchr(ustrp, 0);
2898 /* Here, there was no mapping defined, which means that the code point maps
2899 * to itself. Return the inputs */
2902 if (p != ustrp) { /* Don't copy onto itself */
2903 Copy(p, ustrp, len, U8);
2914 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2916 /* This is called when changing the case of a UTF-8-encoded character above
2917 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2918 * result contains a character that crosses the 255/256 boundary, disallow
2919 * the change, and return the original code point. See L<perlfunc/lc> for
2922 * p points to the original string whose case was changed; assumed
2923 * by this routine to be well-formed
2924 * result the code point of the first character in the changed-case string
2925 * ustrp points to the changed-case string (<result> represents its first char)
2926 * lenp points to the length of <ustrp> */
2928 UV original; /* To store the first code point of <p> */
2930 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2932 assert(UTF8_IS_ABOVE_LATIN1(*p));
2934 /* We know immediately if the first character in the string crosses the
2935 * boundary, so can skip */
2938 /* Look at every character in the result; if any cross the
2939 * boundary, the whole thing is disallowed */
2940 U8* s = ustrp + UTF8SKIP(ustrp);
2941 U8* e = ustrp + *lenp;
2943 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2949 /* Here, no characters crossed, result is ok as-is, but we warn. */
2950 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
2956 /* Failed, have to return the original */
2957 original = valid_utf8_to_uvchr(p, lenp);
2959 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2960 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2961 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2962 "resolved to \"\\x{%" UVXf "}\".",
2966 Copy(p, ustrp, *lenp, char);
2971 S_check_and_deprecate(pTHX_ const U8 *p,
2973 const unsigned int type, /* See below */
2974 const bool use_locale, /* Is this a 'LC_'
2976 const char * const file,
2977 const unsigned line)
2979 /* This is a temporary function to deprecate the unsafe calls to the case
2980 * changing macros and functions. It keeps all the special stuff in just
2983 * It updates *e with the pointer to the end of the input string. If using
2984 * the old-style macros, *e is NULL on input, and so this function assumes
2985 * the input string is long enough to hold the entire UTF-8 sequence, and
2986 * sets *e accordingly, but it then returns a flag to pass the
2987 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
2988 * using the full length if possible.
2990 * It also does the assert that *e > p when *e is not NULL. This should be
2991 * migrated to the callers when this function gets deleted.
2993 * The 'type' parameter is used for the caller to specify which case
2994 * changing function this is called from: */
2996 # define DEPRECATE_TO_UPPER 0
2997 # define DEPRECATE_TO_TITLE 1
2998 # define DEPRECATE_TO_LOWER 2
2999 # define DEPRECATE_TO_FOLD 3
3001 U32 utf8n_flags = 0;
3003 const char * alternative;
3005 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3008 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3009 *e = p + UTF8SKIP(p);
3011 /* For mathoms.c calls, we use the function name we know is stored
3012 * there. It could be part of a larger path */
3013 if (type == DEPRECATE_TO_UPPER) {
3014 name = instr(file, "mathoms.c")
3017 alternative = "toUPPER_utf8_safe";
3019 else if (type == DEPRECATE_TO_TITLE) {
3020 name = instr(file, "mathoms.c")
3023 alternative = "toTITLE_utf8_safe";
3025 else if (type == DEPRECATE_TO_LOWER) {
3026 name = instr(file, "mathoms.c")
3029 alternative = "toLOWER_utf8_safe";
3031 else if (type == DEPRECATE_TO_FOLD) {
3032 name = instr(file, "mathoms.c")
3035 alternative = "toFOLD_utf8_safe";
3037 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3039 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3048 /* The process for changing the case is essentially the same for the four case
3049 * change types, except there are complications for folding. Otherwise the
3050 * difference is only which case to change to. To make sure that they all do
3051 * the same thing, the bodies of the functions are extracted out into the
3052 * following two macros. The functions are written with the same variable
3053 * names, and these are known and used inside these macros. It would be
3054 * better, of course, to have inline functions to do it, but since different
3055 * macros are called, depending on which case is being changed to, this is not
3056 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3057 * function can start with the common start macro, then finish with its special
3058 * handling; while the other three cases can just use the common end macro.
3060 * The algorithm is to use the proper (passed in) macro or function to change
3061 * the case for code points that are below 256. The macro is used if using
3062 * locale rules for the case change; the function if not. If the code point is
3063 * above 255, it is computed from the input UTF-8, and another macro is called
3064 * to do the conversion. If necessary, the output is converted to UTF-8. If
3065 * using a locale, we have to check that the change did not cross the 255/256
3066 * boundary, see check_locale_boundary_crossing() for further details.
3068 * The macros are split with the correct case change for the below-256 case
3069 * stored into 'result', and in the middle of an else clause for the above-255
3070 * case. At that point in the 'else', 'result' is not the final result, but is
3071 * the input code point calculated from the UTF-8. The fold code needs to
3072 * realize all this and take it from there.
3074 * If you read the two macros as sequential, it's easier to understand what's
3076 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3077 L1_func_extra_param) \
3079 if (flags & (locale_flags)) { \
3080 /* Treat a UTF-8 locale as not being in locale at all */ \
3081 if (IN_UTF8_CTYPE_LOCALE) { \
3082 flags &= ~(locale_flags); \
3085 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3089 if (UTF8_IS_INVARIANT(*p)) { \
3090 if (flags & (locale_flags)) { \
3091 result = LC_L1_change_macro(*p); \
3094 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3097 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3098 if (flags & (locale_flags)) { \
3099 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3103 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3104 ustrp, lenp, L1_func_extra_param); \
3107 else { /* malformed UTF-8 or ord above 255 */ \
3108 STRLEN len_result; \
3109 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3110 if (len_result == (STRLEN) -1) { \
3111 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3115 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3116 result = change_macro(result, p, ustrp, lenp); \
3118 if (flags & (locale_flags)) { \
3119 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3124 /* Here, used locale rules. Convert back to UTF-8 */ \
3125 if (UTF8_IS_INVARIANT(result)) { \
3126 *ustrp = (U8) result; \
3130 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3131 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3138 =for apidoc to_utf8_upper
3140 Instead use L</toUPPER_utf8_safe>.
3144 /* Not currently externally documented, and subject to change:
3145 * <flags> is set iff iff the rules from the current underlying locale are to
3149 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3154 const char * const file,
3158 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3159 cBOOL(flags), file, line);
3161 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3163 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3164 /* 2nd char of uc(U+DF) is 'S' */
3165 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3166 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3170 =for apidoc to_utf8_title
3172 Instead use L</toTITLE_utf8_safe>.
3176 /* Not currently externally documented, and subject to change:
3177 * <flags> is set iff the rules from the current underlying locale are to be
3178 * used. Since titlecase is not defined in POSIX, for other than a
3179 * UTF-8 locale, uppercase is used instead for code points < 256.
3183 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3188 const char * const file,
3192 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3193 cBOOL(flags), file, line);
3195 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3197 /* 2nd char of ucfirst(U+DF) is 's' */
3198 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3199 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3203 =for apidoc to_utf8_lower
3205 Instead use L</toLOWER_utf8_safe>.
3209 /* Not currently externally documented, and subject to change:
3210 * <flags> is set iff iff the rules from the current underlying locale are to
3215 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3220 const char * const file,
3224 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3225 cBOOL(flags), file, line);
3227 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3229 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3230 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3234 =for apidoc to_utf8_fold
3236 Instead use L</toFOLD_utf8_safe>.
3240 /* Not currently externally documented, and subject to change,
3242 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3243 * locale are to be used.
3244 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3245 * otherwise simple folds
3246 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3251 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3256 const char * const file,
3260 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3261 cBOOL(flags), file, line);
3263 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3265 /* These are mutually exclusive */
3266 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3268 assert(p != ustrp); /* Otherwise overwrites */
3270 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3271 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3273 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3275 if (flags & FOLD_FLAGS_LOCALE) {
3277 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3278 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3280 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3281 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3283 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3285 /* Special case these two characters, as what normally gets
3286 * returned under locale doesn't work */
3287 if (UTF8SKIP(p) == cap_sharp_s_len
3288 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3290 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3291 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3292 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3293 "resolved to \"\\x{17F}\\x{17F}\".");
3298 if (UTF8SKIP(p) == long_s_t_len
3299 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3301 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3302 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3303 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3304 "resolved to \"\\x{FB06}\".");
3305 goto return_ligature_st;
3308 #if UNICODE_MAJOR_VERSION == 3 \
3309 && UNICODE_DOT_VERSION == 0 \
3310 && UNICODE_DOT_DOT_VERSION == 1
3311 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3313 /* And special case this on this Unicode version only, for the same
3314 * reaons the other two are special cased. They would cross the
3315 * 255/256 boundary which is forbidden under /l, and so the code
3316 * wouldn't catch that they are equivalent (which they are only in
3318 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3319 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3321 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3322 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3323 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3324 "resolved to \"\\x{0131}\".");
3325 goto return_dotless_i;
3329 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3331 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3335 /* This is called when changing the case of a UTF-8-encoded
3336 * character above the ASCII range, and the result should not
3337 * contain an ASCII character. */
3339 UV original; /* To store the first code point of <p> */
3341 /* Look at every character in the result; if any cross the
3342 * boundary, the whole thing is disallowed */
3344 U8* e = ustrp + *lenp;
3347 /* Crossed, have to return the original */
3348 original = valid_utf8_to_uvchr(p, lenp);
3350 /* But in these instances, there is an alternative we can
3351 * return that is valid */
3352 if (original == LATIN_SMALL_LETTER_SHARP_S
3353 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3354 || original == LATIN_CAPITAL_LETTER_SHARP_S
3359 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3360 goto return_ligature_st;
3362 #if UNICODE_MAJOR_VERSION == 3 \
3363 && UNICODE_DOT_VERSION == 0 \
3364 && UNICODE_DOT_DOT_VERSION == 1
3366 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3367 goto return_dotless_i;
3370 Copy(p, ustrp, *lenp, char);
3376 /* Here, no characters crossed, result is ok as-is */
3381 /* Here, used locale rules. Convert back to UTF-8 */
3382 if (UTF8_IS_INVARIANT(result)) {
3383 *ustrp = (U8) result;
3387 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3388 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3395 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3396 * folds to a string of two of these characters. By returning this
3397 * instead, then, e.g.,
3398 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3401 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3402 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3404 return LATIN_SMALL_LETTER_LONG_S;
3407 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3408 * have the other one fold to it */
3410 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3411 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3412 return LATIN_SMALL_LIGATURE_ST;
3414 #if UNICODE_MAJOR_VERSION == 3 \
3415 && UNICODE_DOT_VERSION == 0 \
3416 && UNICODE_DOT_DOT_VERSION == 1
3419 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3420 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3421 return LATIN_SMALL_LETTER_DOTLESS_I;
3428 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3429 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3430 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3434 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3436 PERL_ARGS_ASSERT_SWASH_INIT;
3438 /* Returns a copy of a swash initiated by the called function. This is the
3439 * public interface, and returning a copy prevents others from doing
3440 * mischief on the original */
3442 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3446 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3449 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3450 * use the following define */
3452 #define CORE_SWASH_INIT_RETURN(x) \
3453 PL_curpm= old_PL_curpm; \
3456 /* Initialize and return a swash, creating it if necessary. It does this
3457 * by calling utf8_heavy.pl in the general case. The returned value may be
3458 * the swash's inversion list instead if the input parameters allow it.
3459 * Which is returned should be immaterial to callers, as the only
3460 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3461 * and swash_to_invlist() handle both these transparently.
3463 * This interface should only be used by functions that won't destroy or
3464 * adversely change the swash, as doing so affects all other uses of the
3465 * swash in the program; the general public should use 'Perl_swash_init'
3468 * pkg is the name of the package that <name> should be in.
3469 * name is the name of the swash to find. Typically it is a Unicode
3470 * property name, including user-defined ones
3471 * listsv is a string to initialize the swash with. It must be of the form
3472 * documented as the subroutine return value in
3473 * L<perlunicode/User-Defined Character Properties>
3474 * minbits is the number of bits required to represent each data element.
3475 * It is '1' for binary properties.
3476 * none I (khw) do not understand this one, but it is used only in tr///.
3477 * invlist is an inversion list to initialize the swash with (or NULL)
3478 * flags_p if non-NULL is the address of various input and output flag bits
3479 * to the routine, as follows: ('I' means is input to the routine;
3480 * 'O' means output from the routine. Only flags marked O are
3481 * meaningful on return.)
3482 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3483 * came from a user-defined property. (I O)
3484 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3485 * when the swash cannot be located, to simply return NULL. (I)
3486 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3487 * return of an inversion list instead of a swash hash if this routine
3488 * thinks that would result in faster execution of swash_fetch() later
3491 * Thus there are three possible inputs to find the swash: <name>,
3492 * <listsv>, and <invlist>. At least one must be specified. The result
3493 * will be the union of the specified ones, although <listsv>'s various
3494 * actions can intersect, etc. what <name> gives. To avoid going out to
3495 * disk at all, <invlist> should specify completely what the swash should
3496 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3498 * <invlist> is only valid for binary properties */
3500 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3502 SV* retval = &PL_sv_undef;
3503 HV* swash_hv = NULL;
3504 const int invlist_swash_boundary =
3505 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3506 ? 512 /* Based on some benchmarking, but not extensive, see commit
3508 : -1; /* Never return just an inversion list */
3510 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3511 assert(! invlist || minbits == 1);
3513 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3514 that triggered the swash init and the swash init perl logic itself.
3517 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3519 if (listsv != &PL_sv_undef || strNE(name, "")) {
3521 const size_t pkg_len = strlen(pkg);
3522 const size_t name_len = strlen(name);
3523 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3527 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3529 PUSHSTACKi(PERLSI_MAGIC);
3533 /* We might get here via a subroutine signature which uses a utf8
3534 * parameter name, at which point PL_subname will have been set
3535 * but not yet used. */
3536 save_item(PL_subname);
3537 if (PL_parser && PL_parser->error_count)
3538 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3539 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3540 if (!method) { /* demand load UTF-8 */
3542 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3543 GvSV(PL_errgv) = NULL;
3544 #ifndef NO_TAINT_SUPPORT
3545 /* It is assumed that callers of this routine are not passing in
3546 * any user derived data. */
3547 /* Need to do this after save_re_context() as it will set
3548 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3549 * in Perl_magic_get). Even line to create errsv_save can turn on
3551 SAVEBOOL(TAINT_get);
3554 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3557 /* Not ERRSV, as there is no need to vivify a scalar we are
3558 about to discard. */
3559 SV * const errsv = GvSV(PL_errgv);
3560 if (!SvTRUE(errsv)) {
3561 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3562 SvREFCNT_dec(errsv);
3570 mPUSHp(pkg, pkg_len);
3571 mPUSHp(name, name_len);
3576 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3577 GvSV(PL_errgv) = NULL;
3578 /* If we already have a pointer to the method, no need to use
3579 * call_method() to repeat the lookup. */
3581 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3582 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3584 retval = *PL_stack_sp--;
3585 SvREFCNT_inc(retval);
3588 /* Not ERRSV. See above. */
3589 SV * const errsv = GvSV(PL_errgv);
3590 if (!SvTRUE(errsv)) {
3591 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3592 SvREFCNT_dec(errsv);
3597 if (IN_PERL_COMPILETIME) {
3598 CopHINTS_set(PL_curcop, PL_hints);
3600 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3601 if (SvPOK(retval)) {
3603 /* If caller wants to handle missing properties, let them */
3604 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3605 CORE_SWASH_INIT_RETURN(NULL);
3608 "Can't find Unicode property definition \"%" SVf "\"",
3610 NOT_REACHED; /* NOTREACHED */
3613 } /* End of calling the module to find the swash */
3615 /* If this operation fetched a swash, and we will need it later, get it */
3616 if (retval != &PL_sv_undef
3617 && (minbits == 1 || (flags_p
3619 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3621 swash_hv = MUTABLE_HV(SvRV(retval));
3623 /* If we don't already know that there is a user-defined component to
3624 * this swash, and the user has indicated they wish to know if there is
3625 * one (by passing <flags_p>), find out */
3626 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3627 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3628 if (user_defined && SvUV(*user_defined)) {
3629 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3634 /* Make sure there is an inversion list for binary properties */
3636 SV** swash_invlistsvp = NULL;
3637 SV* swash_invlist = NULL;
3638 bool invlist_in_swash_is_valid = FALSE;
3639 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3640 an unclaimed reference count */
3642 /* If this operation fetched a swash, get its already existing
3643 * inversion list, or create one for it */
3646 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3647 if (swash_invlistsvp) {
3648 swash_invlist = *swash_invlistsvp;
3649 invlist_in_swash_is_valid = TRUE;
3652 swash_invlist = _swash_to_invlist(retval);
3653 swash_invlist_unclaimed = TRUE;
3657 /* If an inversion list was passed in, have to include it */
3660 /* Any fetched swash will by now have an inversion list in it;
3661 * otherwise <swash_invlist> will be NULL, indicating that we
3662 * didn't fetch a swash */
3663 if (swash_invlist) {
3665 /* Add the passed-in inversion list, which invalidates the one
3666 * already stored in the swash */
3667 invlist_in_swash_is_valid = FALSE;
3668 SvREADONLY_off(swash_invlist); /* Turned on again below */
3669 _invlist_union(invlist, swash_invlist, &swash_invlist);
3673 /* Here, there is no swash already. Set up a minimal one, if
3674 * we are going to return a swash */
3675 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3677 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3679 swash_invlist = invlist;
3683 /* Here, we have computed the union of all the passed-in data. It may
3684 * be that there was an inversion list in the swash which didn't get
3685 * touched; otherwise save the computed one */
3686 if (! invlist_in_swash_is_valid
3687 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3689 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3691 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3693 /* We just stole a reference count. */
3694 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3695 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3698 /* The result is immutable. Forbid attempts to change it. */
3699 SvREADONLY_on(swash_invlist);
3701 /* Use the inversion list stand-alone if small enough */
3702 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3703 SvREFCNT_dec(retval);
3704 if (!swash_invlist_unclaimed)
3705 SvREFCNT_inc_simple_void_NN(swash_invlist);
3706 retval = newRV_noinc(swash_invlist);
3710 CORE_SWASH_INIT_RETURN(retval);
3711 #undef CORE_SWASH_INIT_RETURN
3715 /* This API is wrong for special case conversions since we may need to
3716 * return several Unicode characters for a single Unicode character
3717 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3718 * the lower-level routine, and it is similarly broken for returning
3719 * multiple values. --jhi
3720 * For those, you should use S__to_utf8_case() instead */
3721 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3724 * Returns the value of property/mapping C<swash> for the first character
3725 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3726 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3727 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3729 * A "swash" is a hash which contains initially the keys/values set up by
3730 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3731 * property for all possible code points. Things are stored in a compact form
3732 * (see utf8_heavy.pl) so that calculation is required to find the actual
3733 * property value for a given code point. As code points are looked up, new
3734 * key/value pairs are added to the hash, so that the calculation doesn't have
3735 * to ever be re-done. Further, each calculation is done, not just for the
3736 * desired one, but for a whole block of code points adjacent to that one.
3737 * For binary properties on ASCII machines, the block is usually for 64 code
3738 * points, starting with a code point evenly divisible by 64. Thus if the
3739 * property value for code point 257 is requested, the code goes out and
3740 * calculates the property values for all 64 code points between 256 and 319,
3741 * and stores these as a single 64-bit long bit vector, called a "swatch",
3742 * under the key for code point 256. The key is the UTF-8 encoding for code
3743 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3744 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3745 * for code point 258 is then requested, this code realizes that it would be
3746 * stored under the key for 256, and would find that value and extract the
3747 * relevant bit, offset from 256.
3749 * Non-binary properties are stored in as many bits as necessary to represent
3750 * their values (32 currently, though the code is more general than that), not
3751 * as single bits, but the principle is the same: the value for each key is a
3752 * vector that encompasses the property values for all code points whose UTF-8
3753 * representations are represented by the key. That is, for all code points
3754 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3758 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3760 HV *const hv = MUTABLE_HV(SvRV(swash));
3765 const U8 *tmps = NULL;
3769 PERL_ARGS_ASSERT_SWASH_FETCH;
3771 /* If it really isn't a hash, it isn't really swash; must be an inversion
3773 if (SvTYPE(hv) != SVt_PVHV) {
3774 return _invlist_contains_cp((SV*)hv,
3776 ? valid_utf8_to_uvchr(ptr, NULL)
3780 /* We store the values in a "swatch" which is a vec() value in a swash
3781 * hash. Code points 0-255 are a single vec() stored with key length
3782 * (klen) 0. All other code points have a UTF-8 representation
3783 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3784 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3785 * length for them is the length of the encoded char - 1. ptr[klen] is the
3786 * final byte in the sequence representing the character */
3787 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3792 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3795 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3798 klen = UTF8SKIP(ptr) - 1;
3800 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3801 * the vec is the final byte in the sequence. (In EBCDIC this is
3802 * converted to I8 to get consecutive values.) To help you visualize
3804 * Straight 1047 After final byte
3805 * UTF-8 UTF-EBCDIC I8 transform
3806 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3807 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3809 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3810 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3812 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3813 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3815 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3816 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3818 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3819 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3821 * (There are no discontinuities in the elided (...) entries.)
3822 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3823 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3824 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3825 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3826 * index into the vec() swatch (after subtracting 0x80, which we
3827 * actually do with an '&').
3828 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3829 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3830 * dicontinuities which go away by transforming it into I8, and we
3831 * effectively subtract 0xA0 to get the index. */
3832 needents = (1 << UTF_ACCUMULATION_SHIFT);
3833 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3837 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3838 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3839 * it's nothing to sniff at.) Pity we usually come through at least
3840 * two function calls to get here...
3842 * NB: this code assumes that swatches are never modified, once generated!
3845 if (hv == PL_last_swash_hv &&
3846 klen == PL_last_swash_klen &&
3847 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3849 tmps = PL_last_swash_tmps;
3850 slen = PL_last_swash_slen;
3853 /* Try our second-level swatch cache, kept in a hash. */
3854 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3856 /* If not cached, generate it via swatch_get */
3857 if (!svp || !SvPOK(*svp)
3858 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3861 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3862 swatch = swatch_get(swash,
3863 code_point & ~((UV)needents - 1),
3866 else { /* For the first 256 code points, the swatch has a key of
3868 swatch = swatch_get(swash, 0, needents);
3871 if (IN_PERL_COMPILETIME)
3872 CopHINTS_set(PL_curcop, PL_hints);
3874 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3876 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3877 || (slen << 3) < needents)
3878 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3879 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3880 svp, tmps, (UV)slen, (UV)needents);
3883 PL_last_swash_hv = hv;
3884 assert(klen <= sizeof(PL_last_swash_key));
3885 PL_last_swash_klen = (U8)klen;
3886 /* FIXME change interpvar.h? */
3887 PL_last_swash_tmps = (U8 *) tmps;
3888 PL_last_swash_slen = slen;
3890 Copy(ptr, PL_last_swash_key, klen, U8);
3893 switch ((int)((slen << 3) / needents)) {
3895 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3897 return ((UV) tmps[off]);
3901 ((UV) tmps[off ] << 8) +
3902 ((UV) tmps[off + 1]);
3906 ((UV) tmps[off ] << 24) +
3907 ((UV) tmps[off + 1] << 16) +
3908 ((UV) tmps[off + 2] << 8) +
3909 ((UV) tmps[off + 3]);
3911 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3912 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
3913 NORETURN_FUNCTION_END;
3916 /* Read a single line of the main body of the swash input text. These are of
3919 * where each number is hex. The first two numbers form the minimum and
3920 * maximum of a range, and the third is the value associated with the range.
3921 * Not all swashes should have a third number
3923 * On input: l points to the beginning of the line to be examined; it points
3924 * to somewhere in the string of the whole input text, and is
3925 * terminated by a \n or the null string terminator.
3926 * lend points to the null terminator of that string
3927 * wants_value is non-zero if the swash expects a third number
3928 * typestr is the name of the swash's mapping, like 'ToLower'
3929 * On output: *min, *max, and *val are set to the values read from the line.
3930 * returns a pointer just beyond the line examined. If there was no
3931 * valid min number on the line, returns lend+1
3935 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3936 const bool wants_value, const U8* const typestr)
3938 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3939 STRLEN numlen; /* Length of the number */
3940 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3941 | PERL_SCAN_DISALLOW_PREFIX
3942 | PERL_SCAN_SILENT_NON_PORTABLE;
3944 /* nl points to the next \n in the scan */
3945 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3947 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3949 /* Get the first number on the line: the range minimum */
3951 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3952 *max = *min; /* So can never return without setting max */
3953 if (numlen) /* If found a hex number, position past it */
3955 else if (nl) { /* Else, go handle next line, if any */
3956 return nl + 1; /* 1 is length of "\n" */
3958 else { /* Else, no next line */
3959 return lend + 1; /* to LIST's end at which \n is not found */
3962 /* The max range value follows, separated by a BLANK */
3965 flags = PERL_SCAN_SILENT_ILLDIGIT
3966 | PERL_SCAN_DISALLOW_PREFIX
3967 | PERL_SCAN_SILENT_NON_PORTABLE;
3969 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3972 else /* If no value here, it is a single element range */
3975 /* Non-binary tables have a third entry: what the first element of the
3976 * range maps to. The map for those currently read here is in hex */
3980 flags = PERL_SCAN_SILENT_ILLDIGIT
3981 | PERL_SCAN_DISALLOW_PREFIX
3982 | PERL_SCAN_SILENT_NON_PORTABLE;
3984 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3993 /* diag_listed_as: To%s: illegal mapping '%s' */
3994 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4000 *val = 0; /* bits == 1, then any val should be ignored */
4002 else { /* Nothing following range min, should be single element with no
4007 /* diag_listed_as: To%s: illegal mapping '%s' */
4008 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4012 *val = 0; /* bits == 1, then val should be ignored */
4015 /* Position to next line if any, or EOF */
4025 * Returns a swatch (a bit vector string) for a code point sequence
4026 * that starts from the value C<start> and comprises the number C<span>.
4027 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4028 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4031 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4034 U8 *l, *lend, *x, *xend, *s, *send;
4035 STRLEN lcur, xcur, scur;
4036 HV *const hv = MUTABLE_HV(SvRV(swash));
4037 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4039 SV** listsvp = NULL; /* The string containing the main body of the table */
4040 SV** extssvp = NULL;
4041 SV** invert_it_svp = NULL;
4044 STRLEN octets; /* if bits == 1, then octets == 0 */
4046 UV end = start + span;
4048 if (invlistsvp == NULL) {
4049 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4050 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4051 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4052 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4053 listsvp = hv_fetchs(hv, "LIST", FALSE);
4054 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4056 bits = SvUV(*bitssvp);
4057 none = SvUV(*nonesvp);
4058 typestr = (U8*)SvPV_nolen(*typesvp);
4064 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4066 PERL_ARGS_ASSERT_SWATCH_GET;
4068 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4069 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4073 /* If overflowed, use the max possible */
4079 /* create and initialize $swatch */
4080 scur = octets ? (span * octets) : (span + 7) / 8;
4081 swatch = newSV(scur);
4083 s = (U8*)SvPVX(swatch);
4084 if (octets && none) {
4085 const U8* const e = s + scur;
4088 *s++ = (U8)(none & 0xff);
4089 else if (bits == 16) {
4090 *s++ = (U8)((none >> 8) & 0xff);
4091 *s++ = (U8)( none & 0xff);
4093 else if (bits == 32) {
4094 *s++ = (U8)((none >> 24) & 0xff);
4095 *s++ = (U8)((none >> 16) & 0xff);
4096 *s++ = (U8)((none >> 8) & 0xff);
4097 *s++ = (U8)( none & 0xff);
4103 (void)memzero((U8*)s, scur + 1);
4105 SvCUR_set(swatch, scur);
4106 s = (U8*)SvPVX(swatch);
4108 if (invlistsvp) { /* If has an inversion list set up use that */
4109 _invlist_populate_swatch(*invlistsvp, start, end, s);
4113 /* read $swash->{LIST} */
4114 l = (U8*)SvPV(*listsvp, lcur);
4117 UV min, max, val, upper;
4118 l = swash_scan_list_line(l, lend, &min, &max, &val,
4119 cBOOL(octets), typestr);
4124 /* If looking for something beyond this range, go try the next one */
4128 /* <end> is generally 1 beyond where we want to set things, but at the
4129 * platform's infinity, where we can't go any higher, we want to
4130 * include the code point at <end> */
4133 : (max != UV_MAX || end != UV_MAX)
4140 if (!none || val < none) {
4145 for (key = min; key <= upper; key++) {
4147 /* offset must be non-negative (start <= min <= key < end) */
4148 offset = octets * (key - start);
4150 s[offset] = (U8)(val & 0xff);
4151 else if (bits == 16) {
4152 s[offset ] = (U8)((val >> 8) & 0xff);
4153 s[offset + 1] = (U8)( val & 0xff);
4155 else if (bits == 32) {
4156 s[offset ] = (U8)((val >> 24) & 0xff);
4157 s[offset + 1] = (U8)((val >> 16) & 0xff);
4158 s[offset + 2] = (U8)((val >> 8) & 0xff);
4159 s[offset + 3] = (U8)( val & 0xff);
4162 if (!none || val < none)
4166 else { /* bits == 1, then val should be ignored */
4171 for (key = min; key <= upper; key++) {
4172 const STRLEN offset = (STRLEN)(key - start);
4173 s[offset >> 3] |= 1 << (offset & 7);
4178 /* Invert if the data says it should be. Assumes that bits == 1 */
4179 if (invert_it_svp && SvUV(*invert_it_svp)) {
4181 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4182 * be 0, and their inversion should also be 0, as we don't succeed any
4183 * Unicode property matches for non-Unicode code points */
4184 if (start <= PERL_UNICODE_MAX) {
4186 /* The code below assumes that we never cross the
4187 * Unicode/above-Unicode boundary in a range, as otherwise we would
4188 * have to figure out where to stop flipping the bits. Since this
4189 * boundary is divisible by a large power of 2, and swatches comes
4190 * in small powers of 2, this should be a valid assumption */
4191 assert(start + span - 1 <= PERL_UNICODE_MAX);
4201 /* read $swash->{EXTRAS}
4202 * This code also copied to swash_to_invlist() below */
4203 x = (U8*)SvPV(*extssvp, xcur);
4211 SV **otherbitssvp, *other;
4215 const U8 opc = *x++;
4219 nl = (U8*)memchr(x, '\n', xend - x);
4221 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4223 x = nl + 1; /* 1 is length of "\n" */
4227 x = xend; /* to EXTRAS' end at which \n is not found */
4234 namelen = nl - namestr;
4238 namelen = xend - namestr;
4242 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4243 otherhv = MUTABLE_HV(SvRV(*othersvp));
4244 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4245 otherbits = (STRLEN)SvUV(*otherbitssvp);
4246 if (bits < otherbits)
4247 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4248 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4250 /* The "other" swatch must be destroyed after. */
4251 other = swatch_get(*othersvp, start, span);
4252 o = (U8*)SvPV(other, olen);
4255 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4257 s = (U8*)SvPV(swatch, slen);
4258 if (bits == 1 && otherbits == 1) {
4260 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4261 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4262 (UV)slen, (UV)olen);
4286 STRLEN otheroctets = otherbits >> 3;
4288 U8* const send = s + slen;
4293 if (otherbits == 1) {
4294 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4298 STRLEN vlen = otheroctets;
4306 if (opc == '+' && otherval)
4307 NOOP; /* replace with otherval */
4308 else if (opc == '!' && !otherval)
4310 else if (opc == '-' && otherval)
4312 else if (opc == '&' && !otherval)
4315 s += octets; /* no replacement */
4320 *s++ = (U8)( otherval & 0xff);
4321 else if (bits == 16) {
4322 *s++ = (U8)((otherval >> 8) & 0xff);
4323 *s++ = (U8)( otherval & 0xff);
4325 else if (bits == 32) {
4326 *s++ = (U8)((otherval >> 24) & 0xff);
4327 *s++ = (U8)((otherval >> 16) & 0xff);
4328 *s++ = (U8)((otherval >> 8) & 0xff);
4329 *s++ = (U8)( otherval & 0xff);
4333 sv_free(other); /* through with it! */
4339 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4342 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4343 * Can't be used on a property that is subject to user override, as it
4344 * relies on the value of SPECIALS in the swash which would be set by
4345 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4346 * for overridden properties
4348 * Returns a hash which is the inversion and closure of a swash mapping.
4349 * For example, consider the input lines:
4354 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4355 * 006C. The value for each key is an array. For 006C, the array would
4356 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4357 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4359 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4360 * keys are only code points that are folded-to, so it isn't a full closure.
4362 * Essentially, for any code point, it gives all the code points that map to
4363 * it, or the list of 'froms' for that point.
4365 * Currently it ignores any additions or deletions from other swashes,
4366 * looking at just the main body of the swash, and if there are SPECIALS
4367 * in the swash, at that hash
4369 * The specials hash can be extra code points, and most likely consists of
4370 * maps from single code points to multiple ones (each expressed as a string
4371 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4372 * However consider this possible input in the specials hash:
4373 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4374 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4376 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4377 * currently handle. But it also means that FB05 and FB06 are equivalent in
4378 * a 1-1 mapping which we should handle, and this relationship may not be in
4379 * the main table. Therefore this function examines all the multi-char
4380 * sequences and adds the 1-1 mappings that come out of that.
4382 * XXX This function was originally intended to be multipurpose, but its
4383 * only use is quite likely to remain for constructing the inversion of
4384 * the CaseFolding (//i) property. If it were more general purpose for
4385 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4386 * because certain folds are prohibited under /iaa and /il. As an example,
4387 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4388 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4389 * prohibited, so we would not figure out that they fold to each other.
4390 * Code could be written to automatically figure this out, similar to the
4391 * code that does this for multi-character folds, but this is the only case
4392 * where something like this is ever likely to happen, as all the single
4393 * char folds to the 0-255 range are now quite settled. Instead there is a
4394 * little special code that is compiled only for this Unicode version. This
4395 * is smaller and didn't require much coding time to do. But this makes
4396 * this routine strongly tied to being used just for CaseFolding. If ever
4397 * it should be generalized, this would have to be fixed */
4401 HV *const hv = MUTABLE_HV(SvRV(swash));
4403 /* The string containing the main body of the table. This will have its
4404 * assertion fail if the swash has been converted to its inversion list */
4405 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4407 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4408 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4409 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4410 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4411 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4412 const STRLEN bits = SvUV(*bitssvp);
4413 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4414 const UV none = SvUV(*nonesvp);
4415 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4419 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4421 /* Must have at least 8 bits to get the mappings */
4422 if (bits != 8 && bits != 16 && bits != 32) {
4423 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4427 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4428 mapping to more than one character */
4430 /* Construct an inverse mapping hash for the specials */
4431 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4432 HV * specials_inverse = newHV();
4433 char *char_from; /* the lhs of the map */
4434 I32 from_len; /* its byte length */
4435 char *char_to; /* the rhs of the map */
4436 I32 to_len; /* its byte length */
4437 SV *sv_to; /* and in a sv */
4438 AV* from_list; /* list of things that map to each 'to' */
4440 hv_iterinit(specials_hv);
4442 /* The keys are the characters (in UTF-8) that map to the corresponding
4443 * UTF-8 string value. Iterate through the list creating the inverse
4445 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4447 if (! SvPOK(sv_to)) {
4448 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4449 "unexpectedly is not a string, flags=%lu",
4450 (unsigned long)SvFLAGS(sv_to));
4452 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "Found mapping from %" UVXf ", First char of to is %" UVXf "\n", valid_utf8_to_uvchr((U8*) char_from, 0), valid_utf8_to_uvchr((U8*) SvPVX(sv_to), 0)));*/
4454 /* Each key in the inverse list is a mapped-to value, and the key's
4455 * hash value is a list of the strings (each in UTF-8) that map to
4456 * it. Those strings are all one character long */
4457 if ((listp = hv_fetch(specials_inverse,
4461 from_list = (AV*) *listp;
4463 else { /* No entry yet for it: create one */
4464 from_list = newAV();
4465 if (! hv_store(specials_inverse,
4468 (SV*) from_list, 0))
4470 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4474 /* Here have the list associated with this 'to' (perhaps newly
4475 * created and empty). Just add to it. Note that we ASSUME that
4476 * the input is guaranteed to not have duplications, so we don't
4477 * check for that. Duplications just slow down execution time. */
4478 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4481 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4482 * it looking for cases like the FB05/FB06 examples above. There would
4483 * be an entry in the hash like
4484 * 'st' => [ FB05, FB06 ]
4485 * In this example we will create two lists that get stored in the
4486 * returned hash, 'ret':
4487 * FB05 => [ FB05, FB06 ]
4488 * FB06 => [ FB05, FB06 ]
4490 * Note that there is nothing to do if the array only has one element.
4491 * (In the normal 1-1 case handled below, we don't have to worry about
4492 * two lists, as everything gets tied to the single list that is
4493 * generated for the single character 'to'. But here, we are omitting
4494 * that list, ('st' in the example), so must have multiple lists.) */
4495 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4496 &char_to, &to_len)))
4498 if (av_tindex_skip_len_mg(from_list) > 0) {
4501 /* We iterate over all combinations of i,j to place each code
4502 * point on each list */
4503 for (i = 0; i <= av_tindex_skip_len_mg(from_list); i++) {
4505 AV* i_list = newAV();
4506 SV** entryp = av_fetch(from_list, i, FALSE);
4507 if (entryp == NULL) {
4508 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4510 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4511 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4513 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4514 (SV*) i_list, FALSE))
4516 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4519 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4520 for (j = 0; j <= av_tindex_skip_len_mg(from_list); j++) {
4521 entryp = av_fetch(from_list, j, FALSE);
4522 if (entryp == NULL) {
4523 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4526 /* When i==j this adds itself to the list */
4527 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4528 (U8*) SvPVX(*entryp),
4529 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4531 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
4536 SvREFCNT_dec(specials_inverse); /* done with it */
4537 } /* End of specials */
4539 /* read $swash->{LIST} */
4541 #if UNICODE_MAJOR_VERSION == 3 \
4542 && UNICODE_DOT_VERSION == 0 \
4543 && UNICODE_DOT_DOT_VERSION == 1
4545 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4546 * rule so that things work under /iaa and /il */
4548 SV * mod_listsv = sv_mortalcopy(*listsvp);
4549 sv_catpv(mod_listsv, "130\t130\t131\n");
4550 l = (U8*)SvPV(mod_listsv, lcur);
4554 l = (U8*)SvPV(*listsvp, lcur);
4560 /* Go through each input line */
4564 l = swash_scan_list_line(l, lend, &min, &max, &val,
4565 cBOOL(octets), typestr);
4570 /* Each element in the range is to be inverted */
4571 for (inverse = min; inverse <= max; inverse++) {
4575 bool found_key = FALSE;
4576 bool found_inverse = FALSE;
4578 /* The key is the inverse mapping */
4579 char key[UTF8_MAXBYTES+1];
4580 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4581 STRLEN key_len = key_end - key;
4583 /* Get the list for the map */
4584 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4585 list = (AV*) *listp;
4587 else { /* No entry yet for it: create one */
4589 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4590 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4594 /* Look through list to see if this inverse mapping already is
4595 * listed, or if there is a mapping to itself already */
4596 for (i = 0; i <= av_tindex_skip_len_mg(list); i++) {
4597 SV** entryp = av_fetch(list, i, FALSE);
4600 if (entryp == NULL) {
4601 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4605 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4609 if (uv == inverse) {
4610 found_inverse = TRUE;
4613 /* No need to continue searching if found everything we are
4615 if (found_key && found_inverse) {
4620 /* Make sure there is a mapping to itself on the list */
4622 av_push(list, newSVuv(val));
4623 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4627 /* Simply add the value to the list */
4628 if (! found_inverse) {
4629 av_push(list, newSVuv(inverse));
4630 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4633 /* swatch_get() increments the value of val for each element in the
4634 * range. That makes more compact tables possible. You can
4635 * express the capitalization, for example, of all consecutive
4636 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4637 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4638 * and it's not documented; it appears to be used only in
4639 * implementing tr//; I copied the semantics from swatch_get(), just
4641 if (!none || val < none) {
4651 Perl__swash_to_invlist(pTHX_ SV* const swash)
4654 /* Subject to change or removal. For use only in one place in regcomp.c.
4655 * Ownership is given to one reference count in the returned SV* */
4660 HV *const hv = MUTABLE_HV(SvRV(swash));
4661 UV elements = 0; /* Number of elements in the inversion list */
4671 STRLEN octets; /* if bits == 1, then octets == 0 */
4677 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4679 /* If not a hash, it must be the swash's inversion list instead */
4680 if (SvTYPE(hv) != SVt_PVHV) {
4681 return SvREFCNT_inc_simple_NN((SV*) hv);
4684 /* The string containing the main body of the table */
4685 listsvp = hv_fetchs(hv, "LIST", FALSE);
4686 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4687 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4688 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4689 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4691 typestr = (U8*)SvPV_nolen(*typesvp);
4692 bits = SvUV(*bitssvp);
4693 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4695 /* read $swash->{LIST} */
4696 if (SvPOK(*listsvp)) {
4697 l = (U8*)SvPV(*listsvp, lcur);
4700 /* LIST legitimately doesn't contain a string during compilation phases
4701 * of Perl itself, before the Unicode tables are generated. In this
4702 * case, just fake things up by creating an empty list */
4709 if (*l == 'V') { /* Inversion list format */
4710 const char *after_atou = (char *) lend;
4712 UV* other_elements_ptr;
4714 /* The first number is a count of the rest */
4716 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4717 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
4719 if (elements == 0) {
4720 invlist = _new_invlist(0);
4723 l = (U8 *) after_atou;
4725 /* Get the 0th element, which is needed to setup the inversion list */
4726 while (isSPACE(*l)) l++;
4727 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4728 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
4730 l = (U8 *) after_atou;
4731 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
4734 /* Then just populate the rest of the input */
4735 while (elements-- > 0) {
4737 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more elements than available", elements);
4739 while (isSPACE(*l)) l++;
4740 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
4741 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
4743 l = (U8 *) after_atou;
4749 /* Scan the input to count the number of lines to preallocate array
4750 * size based on worst possible case, which is each line in the input
4751 * creates 2 elements in the inversion list: 1) the beginning of a
4752 * range in the list; 2) the beginning of a range not in the list. */
4753 while ((loc = (strchr(loc, '\n'))) != NULL) {
4758 /* If the ending is somehow corrupt and isn't a new line, add another
4759 * element for the final range that isn't in the inversion list */
4760 if (! (*lend == '\n'
4761 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4766 invlist = _new_invlist(elements);
4768 /* Now go through the input again, adding each range to the list */
4771 UV val; /* Not used by this function */
4773 l = swash_scan_list_line(l, lend, &start, &end, &val,
4774 cBOOL(octets), typestr);
4780 invlist = _add_range_to_invlist(invlist, start, end);
4784 /* Invert if the data says it should be */
4785 if (invert_it_svp && SvUV(*invert_it_svp)) {
4786 _invlist_invert(invlist);
4789 /* This code is copied from swatch_get()
4790 * read $swash->{EXTRAS} */
4791 x = (U8*)SvPV(*extssvp, xcur);
4799 SV **otherbitssvp, *other;
4802 const U8 opc = *x++;
4806 nl = (U8*)memchr(x, '\n', xend - x);
4808 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4810 x = nl + 1; /* 1 is length of "\n" */
4814 x = xend; /* to EXTRAS' end at which \n is not found */
4821 namelen = nl - namestr;
4825 namelen = xend - namestr;
4829 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4830 otherhv = MUTABLE_HV(SvRV(*othersvp));
4831 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4832 otherbits = (STRLEN)SvUV(*otherbitssvp);
4834 if (bits != otherbits || bits != 1) {
4835 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4836 "properties, bits=%" UVuf ", otherbits=%" UVuf,
4837 (UV)bits, (UV)otherbits);
4840 /* The "other" swatch must be destroyed after. */
4841 other = _swash_to_invlist((SV *)*othersvp);
4843 /* End of code copied from swatch_get() */
4846 _invlist_union(invlist, other, &invlist);
4849 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4852 _invlist_subtract(invlist, other, &invlist);
4855 _invlist_intersection(invlist, other, &invlist);
4860 sv_free(other); /* through with it! */
4863 SvREADONLY_on(invlist);
4868 Perl__get_swash_invlist(pTHX_ SV* const swash)
4872 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4874 if (! SvROK(swash)) {
4878 /* If it really isn't a hash, it isn't really swash; must be an inversion
4880 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4884 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4893 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4895 /* May change: warns if surrogates, non-character code points, or
4896 * non-Unicode code points are in s which has length len bytes. Returns
4897 * TRUE if none found; FALSE otherwise. The only other validity check is
4898 * to make sure that this won't exceed the string's length.
4900 * Code points above the platform's C<IV_MAX> will raise a deprecation
4901 * warning, unless those are turned off. */
4903 const U8* const e = s + len;
4906 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4909 if (UTF8SKIP(s) > len) {
4910 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4911 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4914 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4915 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4916 if ( ckWARN_d(WARN_NON_UNICODE)
4917 || ( ckWARN_d(WARN_DEPRECATED)
4919 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
4920 #else /* Below is 64-bit words */
4921 /* 2**63 and up meet these conditions provided we have
4925 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
4928 /* s[1] being above 0x80 overflows */
4933 /* A side effect of this function will be to warn */
4934 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4938 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4939 if (ckWARN_d(WARN_SURROGATE)) {
4940 /* This has a different warning than the one the called
4941 * function would output, so can't just call it, unlike we
4942 * do for the non-chars and above-unicodes */
4943 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4944 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4945 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", uv);
4949 else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
4950 /* A side effect of this function will be to warn */
4951 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4962 =for apidoc pv_uni_display
4964 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4965 length C<len>, the displayable version being at most C<pvlim> bytes long
4966 (if longer, the rest is truncated and C<"..."> will be appended).
4968 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4969 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4970 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4971 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4972 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4973 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4975 The pointer to the PV of the C<dsv> is returned.
4977 See also L</sv_uni_display>.
4981 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4986 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4990 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4992 /* This serves double duty as a flag and a character to print after
4993 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4997 if (pvlim && SvCUR(dsv) >= pvlim) {
5001 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5003 const unsigned char c = (unsigned char)u & 0xFF;
5004 if (flags & UNI_DISPLAY_BACKSLASH) {
5021 const char string = ok;
5022 sv_catpvs(dsv, "\\");
5023 sv_catpvn(dsv, &string, 1);
5026 /* isPRINT() is the locale-blind version. */
5027 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5028 const char string = c;
5029 sv_catpvn(dsv, &string, 1);
5034 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5037 sv_catpvs(dsv, "...");
5043 =for apidoc sv_uni_display
5045 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5046 the displayable version being at most C<pvlim> bytes long
5047 (if longer, the rest is truncated and "..." will be appended).
5049 The C<flags> argument is as in L</pv_uni_display>().
5051 The pointer to the PV of the C<dsv> is returned.
5056 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5058 const char * const ptr =
5059 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5061 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5063 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5064 SvCUR(ssv), pvlim, flags);
5068 =for apidoc foldEQ_utf8
5070 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
5071 of which may be in UTF-8) are the same case-insensitively; false otherwise.
5072 How far into the strings to compare is determined by other input parameters.
5074 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5075 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
5076 with respect to C<s2>.
5078 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
5079 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
5080 scan will not be considered to be a match unless the goal is reached, and
5081 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
5084 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
5085 considered an end pointer to the position 1 byte past the maximum point
5086 in C<s1> beyond which scanning will not continue under any circumstances.
5087 (This routine assumes that UTF-8 encoded input strings are not malformed;
5088 malformed input can cause it to read past C<pe1>).
5089 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
5090 is less than C<s1>+C<l1>, the match will never be successful because it can
5092 get as far as its goal (and in fact is asserted against). Correspondingly for
5093 C<pe2> with respect to C<s2>.
5095 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5096 C<l2> must be non-zero), and if both do, both have to be
5097 reached for a successful match. Also, if the fold of a character is multiple
5098 characters, all of them must be matched (see tr21 reference below for
5101 Upon a successful match, if C<pe1> is non-C<NULL>,
5102 it will be set to point to the beginning of the I<next> character of C<s1>
5103 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5105 For case-insensitiveness, the "casefolding" of Unicode is used
5106 instead of upper/lowercasing both the characters, see
5107 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5111 /* A flags parameter has been added which may change, and hence isn't
5112 * externally documented. Currently it is:
5113 * 0 for as-documented above
5114 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5115 ASCII one, to not match
5116 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5117 * locale are to be used.
5118 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5119 * routine. This allows that step to be skipped.
5120 * Currently, this requires s1 to be encoded as UTF-8
5121 * (u1 must be true), which is asserted for.
5122 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5123 * cross certain boundaries. Hence, the caller should
5124 * let this function do the folding instead of
5125 * pre-folding. This code contains an assertion to
5126 * that effect. However, if the caller knows what
5127 * it's doing, it can pass this flag to indicate that,
5128 * and the assertion is skipped.
5129 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5130 * FOLDEQ_S2_FOLDS_SANE
5133 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, const char *s2, char **pe2, UV l2, bool u2, U32 flags)
5135 const U8 *p1 = (const U8*)s1; /* Point to current char */
5136 const U8 *p2 = (const U8*)s2;
5137 const U8 *g1 = NULL; /* goal for s1 */
5138 const U8 *g2 = NULL;
5139 const U8 *e1 = NULL; /* Don't scan s1 past this */
5140 U8 *f1 = NULL; /* Point to current folded */
5141 const U8 *e2 = NULL;
5143 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5144 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5145 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5146 U8 flags_for_folder = FOLD_FLAGS_FULL;
5148 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5150 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5151 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5152 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5153 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5154 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5155 /* The algorithm is to trial the folds without regard to the flags on
5156 * the first line of the above assert(), and then see if the result
5157 * violates them. This means that the inputs can't be pre-folded to a
5158 * violating result, hence the assert. This could be changed, with the
5159 * addition of extra tests here for the already-folded case, which would
5160 * slow it down. That cost is more than any possible gain for when these
5161 * flags are specified, as the flags indicate /il or /iaa matching which
5162 * is less common than /iu, and I (khw) also believe that real-world /il
5163 * and /iaa matches are most likely to involve code points 0-255, and this
5164 * function only under rare conditions gets called for 0-255. */
5166 if (flags & FOLDEQ_LOCALE) {
5167 if (IN_UTF8_CTYPE_LOCALE) {
5168 flags &= ~FOLDEQ_LOCALE;
5171 flags_for_folder |= FOLD_FLAGS_LOCALE;
5180 g1 = (const U8*)s1 + l1;
5188 g2 = (const U8*)s2 + l2;
5191 /* Must have at least one goal */
5196 /* Will never match if goal is out-of-bounds */
5197 assert(! e1 || e1 >= g1);
5199 /* Here, there isn't an end pointer, or it is beyond the goal. We
5200 * only go as far as the goal */
5204 assert(e1); /* Must have an end for looking at s1 */
5207 /* Same for goal for s2 */
5209 assert(! e2 || e2 >= g2);
5216 /* If both operands are already folded, we could just do a memEQ on the
5217 * whole strings at once, but it would be better if the caller realized
5218 * this and didn't even call us */
5220 /* Look through both strings, a character at a time */
5221 while (p1 < e1 && p2 < e2) {
5223 /* If at the beginning of a new character in s1, get its fold to use
5224 * and the length of the fold. */
5226 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5232 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5234 /* We have to forbid mixing ASCII with non-ASCII if the
5235 * flags so indicate. And, we can short circuit having to
5236 * call the general functions for this common ASCII case,
5237 * all of whose non-locale folds are also ASCII, and hence
5238 * UTF-8 invariants, so the UTF8ness of the strings is not
5240 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5244 *foldbuf1 = toFOLD(*p1);
5247 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5249 else { /* Not UTF-8, get UTF-8 fold */
5250 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5256 if (n2 == 0) { /* Same for s2 */
5257 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5263 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5264 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5268 *foldbuf2 = toFOLD(*p2);
5271 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5274 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5280 /* Here f1 and f2 point to the beginning of the strings to compare.
5281 * These strings are the folds of the next character from each input
5282 * string, stored in UTF-8. */
5284 /* While there is more to look for in both folds, see if they
5285 * continue to match */
5287 U8 fold_length = UTF8SKIP(f1);
5288 if (fold_length != UTF8SKIP(f2)
5289 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5290 function call for single
5292 || memNE((char*)f1, (char*)f2, fold_length))
5294 return 0; /* mismatch */
5297 /* Here, they matched, advance past them */
5304 /* When reach the end of any fold, advance the input past it */
5306 p1 += u1 ? UTF8SKIP(p1) : 1;
5309 p2 += u2 ? UTF8SKIP(p2) : 1;
5311 } /* End of loop through both strings */
5313 /* A match is defined by each scan that specified an explicit length
5314 * reaching its final goal, and the other not having matched a partial
5315 * character (which can happen when the fold of a character is more than one
5317 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5321 /* Successful match. Set output pointers */
5331 /* XXX The next two functions should likely be moved to mathoms.c once all
5332 * occurrences of them are removed from the core; some cpan-upstream modules
5336 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5338 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5340 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5344 =for apidoc utf8n_to_uvuni
5346 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5348 This function was useful for code that wanted to handle both EBCDIC and
5349 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5350 distinctions between the platforms have mostly been made invisible to most
5351 code, so this function is quite unlikely to be what you want. If you do need
5352 this precise functionality, use instead
5353 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5354 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5360 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5362 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5364 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5368 =for apidoc uvuni_to_utf8_flags
5370 Instead you almost certainly want to use L</uvchr_to_utf8> or
5371 L</uvchr_to_utf8_flags>.
5373 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5374 which itself, while not deprecated, should be used only in isolated
5375 circumstances. These functions were useful for code that wanted to handle
5376 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5377 v5.20, the distinctions between the platforms have mostly been made invisible
5378 to most code, so this function is quite unlikely to be what you want.
5384 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5386 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5388 return uvoffuni_to_utf8_flags(d, uv, flags);
5392 * ex: set ts=8 sts=4 sw=4 et: