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<*lenp> from UTF-8 into native byte encoding.
1911 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1912 updates C<*lenp> to contain the new length.
1913 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
1915 Upon successful return, the number of variants in the string can be computed by
1916 having saved the value of C<*lenp> before the call, and subtracting the
1917 after-call value of C<*lenp> from it.
1919 If you need a copy of the string, see L</bytes_from_utf8>.
1925 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
1929 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1930 PERL_UNUSED_CONTEXT;
1932 /* This is a no-op if no variants at all in the input */
1933 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
1938 U8 * const save = s;
1939 U8 * const send = s + *lenp;
1942 /* Nothing before the first variant needs to be changed, so start the real
1946 if (! UTF8_IS_INVARIANT(*s)) {
1947 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1948 *lenp = ((STRLEN) -1);
1956 /* Is downgradable, so do it */
1957 d = s = first_variant;
1960 if (! UVCHR_IS_INVARIANT(c)) {
1961 /* Then it is two-byte encoded */
1962 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1975 =for apidoc bytes_from_utf8
1977 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
1978 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
1979 actually encoded in UTF-8.
1981 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
1984 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
1985 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
1986 C<*lenp> are unchanged, and the return value is the original C<s>.
1988 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
1989 newly created string containing a downgraded copy of C<s>, and whose length is
1990 returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
1992 Upon successful return, the number of variants in the string can be computed by
1993 having saved the value of C<*lenp> before the call, and subtracting the
1994 after-call value of C<*lenp> from it.
1998 There is a macro that avoids this function call, but this is retained for
1999 anyone who calls it with the Perl_ prefix */
2002 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2004 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2005 PERL_UNUSED_CONTEXT;
2007 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2011 No = here because currently externally undocumented
2012 for apidoc bytes_from_utf8_loc
2014 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2015 to store the location of the first character in C<"s"> that cannot be
2016 converted to non-UTF8.
2018 If that parameter is C<NULL>, this function behaves identically to
2021 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2022 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2024 Otherwise, the function returns a newly created C<NUL>-terminated string
2025 containing the non-UTF8 equivalent of the convertible first portion of
2026 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2027 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2028 and C<*first_non_downgradable> is set to C<NULL>.
2030 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2031 first character in the original string that wasn't converted. C<*is_utf8p> is
2032 unchanged. Note that the new string may have length 0.
2034 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2035 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2036 converts as many characters in it as possible stopping at the first one it
2037 finds one that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2038 set to point to that. The function returns the portion that could be converted
2039 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2040 not including the terminating C<NUL>. If the very first character in the
2041 original could not be converted, C<*lenp> will be 0, and the new string will
2042 contain just a single C<NUL>. If the entire input string was converted,
2043 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2045 Upon successful return, the number of variants in the converted portion of the
2046 string can be computed by having saved the value of C<*lenp> before the call,
2047 and subtracting the after-call value of C<*lenp> from it.
2055 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2058 const U8 *original = s;
2059 U8 *converted_start;
2060 const U8 *send = s + *lenp;
2062 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2065 if (first_unconverted) {
2066 *first_unconverted = NULL;
2069 return (U8 *) original;
2072 Newx(d, (*lenp) + 1, U8);
2074 converted_start = d;
2077 if (! UTF8_IS_INVARIANT(c)) {
2079 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2080 * have to stop now */
2081 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2082 if (first_unconverted) {
2083 *first_unconverted = s - 1;
2084 goto finish_and_return;
2087 Safefree(converted_start);
2088 return (U8 *) original;
2092 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2098 /* Here, converted the whole of the input */
2100 if (first_unconverted) {
2101 *first_unconverted = NULL;
2106 *lenp = d - converted_start;
2108 /* Trim unused space */
2109 Renew(converted_start, *lenp + 1, U8);
2111 return converted_start;
2115 =for apidoc bytes_to_utf8
2117 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2119 Returns a pointer to the newly-created string, and sets C<*lenp> to
2120 reflect the new length in bytes.
2122 Upon successful return, the number of variants in the string can be computed by
2123 having saved the value of C<*lenp> before the call, and subtracting it from the
2124 after-call value of C<*lenp>.
2126 A C<NUL> character will be written after the end of the string.
2128 If you want to convert to UTF-8 from encodings other than
2129 the native (Latin1 or EBCDIC),
2130 see L</sv_recode_to_utf8>().
2136 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2138 const U8 * const send = s + (*lenp);
2142 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2143 PERL_UNUSED_CONTEXT;
2145 Newx(d, (*lenp) * 2 + 1, U8);
2149 append_utf8_from_native_byte(*s, &d);
2158 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2160 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2161 * We optimize for native, for obvious reasons. */
2164 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2169 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2172 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
2177 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2179 if (OFFUNI_IS_INVARIANT(uv)) {
2180 *d++ = LATIN1_TO_NATIVE((U8) uv);
2183 if (uv <= MAX_UTF8_TWO_BYTE) {
2184 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2185 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2188 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2189 #define LAST_HIGH_SURROGATE 0xDBFF
2190 #define FIRST_LOW_SURROGATE 0xDC00
2191 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2193 /* This assumes that most uses will be in the first Unicode plane, not
2194 * needing surrogates */
2195 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2196 && uv <= UNICODE_SURROGATE_LAST))
2198 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2199 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2202 UV low = (p[0] << 8) + p[1];
2203 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2204 || UNLIKELY(low > LAST_LOW_SURROGATE))
2206 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2209 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2210 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2214 d = uvoffuni_to_utf8_flags(d, uv, 0);
2217 *d++ = (U8)(( uv >> 12) | 0xe0);
2218 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2219 *d++ = (U8)(( uv & 0x3f) | 0x80);
2223 *d++ = (U8)(( uv >> 18) | 0xf0);
2224 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2225 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2226 *d++ = (U8)(( uv & 0x3f) | 0x80);
2231 *newlen = d - dstart;
2235 /* Note: this one is slightly destructive of the source. */
2238 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2241 U8* const send = s + bytelen;
2243 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2246 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2250 const U8 tmp = s[0];
2255 return utf16_to_utf8(p, d, bytelen, newlen);
2259 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2261 U8 tmpbuf[UTF8_MAXBYTES+1];
2262 uvchr_to_utf8(tmpbuf, c);
2263 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2266 /* Internal function so we can deprecate the external one, and call
2267 this one from other deprecated functions in this file */
2270 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2272 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2276 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2280 Perl__is_uni_perl_idcont(pTHX_ UV c)
2282 U8 tmpbuf[UTF8_MAXBYTES+1];
2283 uvchr_to_utf8(tmpbuf, c);
2284 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2288 Perl__is_uni_perl_idstart(pTHX_ UV c)
2290 U8 tmpbuf[UTF8_MAXBYTES+1];
2291 uvchr_to_utf8(tmpbuf, c);
2292 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2296 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2298 /* We have the latin1-range values compiled into the core, so just use
2299 * those, converting the result to UTF-8. The only difference between upper
2300 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2301 * either "SS" or "Ss". Which one to use is passed into the routine in
2302 * 'S_or_s' to avoid a test */
2304 UV converted = toUPPER_LATIN1_MOD(c);
2306 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2308 assert(S_or_s == 'S' || S_or_s == 's');
2310 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2311 characters in this range */
2312 *p = (U8) converted;
2317 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2318 * which it maps to one of them, so as to only have to have one check for
2319 * it in the main case */
2320 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2322 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2323 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2326 converted = GREEK_CAPITAL_LETTER_MU;
2328 #if UNICODE_MAJOR_VERSION > 2 \
2329 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2330 && UNICODE_DOT_DOT_VERSION >= 8)
2331 case LATIN_SMALL_LETTER_SHARP_S:
2338 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2339 NOT_REACHED; /* NOTREACHED */
2343 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2344 *p = UTF8_TWO_BYTE_LO(converted);
2350 /* Call the function to convert a UTF-8 encoded character to the specified case.
2351 * Note that there may be more than one character in the result.
2352 * INP is a pointer to the first byte of the input character
2353 * OUTP will be set to the first byte of the string of changed characters. It
2354 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2355 * LENP will be set to the length in bytes of the string of changed characters
2357 * The functions return the ordinal of the first character in the string of OUTP */
2358 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2359 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2360 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2362 /* This additionally has the input parameter 'specials', which if non-zero will
2363 * cause this to use the specials hash for folding (meaning get full case
2364 * folding); otherwise, when zero, this implies a simple case fold */
2365 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2368 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2370 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2371 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2372 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2373 * the changed version may be longer than the original character.
2375 * The ordinal of the first character of the changed version is returned
2376 * (but note, as explained above, that there may be more.) */
2378 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2381 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2384 uvchr_to_utf8(p, c);
2385 return CALL_UPPER_CASE(c, p, p, lenp);
2389 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2391 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2394 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2397 uvchr_to_utf8(p, c);
2398 return CALL_TITLE_CASE(c, p, p, lenp);
2402 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2404 /* We have the latin1-range values compiled into the core, so just use
2405 * those, converting the result to UTF-8. Since the result is always just
2406 * one character, we allow <p> to be NULL */
2408 U8 converted = toLOWER_LATIN1(c);
2410 PERL_UNUSED_ARG(dummy);
2413 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2418 /* Result is known to always be < 256, so can use the EIGHT_BIT
2420 *p = UTF8_EIGHT_BIT_HI(converted);
2421 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2429 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2431 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2434 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2437 uvchr_to_utf8(p, c);
2438 return CALL_LOWER_CASE(c, p, p, lenp);
2442 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2444 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2445 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2446 * FOLD_FLAGS_FULL iff full folding is to be used;
2448 * Not to be used for locale folds
2453 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2454 PERL_UNUSED_CONTEXT;
2456 assert (! (flags & FOLD_FLAGS_LOCALE));
2458 if (UNLIKELY(c == MICRO_SIGN)) {
2459 converted = GREEK_SMALL_LETTER_MU;
2461 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2462 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2463 || UNICODE_DOT_DOT_VERSION > 0)
2464 else if ( (flags & FOLD_FLAGS_FULL)
2465 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2467 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2468 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2469 * under those circumstances. */
2470 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2471 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2472 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2474 return LATIN_SMALL_LETTER_LONG_S;
2484 else { /* In this range the fold of all other characters is their lower
2486 converted = toLOWER_LATIN1(c);
2489 if (UVCHR_IS_INVARIANT(converted)) {
2490 *p = (U8) converted;
2494 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2495 *p = UTF8_TWO_BYTE_LO(converted);
2503 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2506 /* Not currently externally documented, and subject to change
2507 * <flags> bits meanings:
2508 * FOLD_FLAGS_FULL iff full folding is to be used;
2509 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2510 * locale are to be used.
2511 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2514 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2516 if (flags & FOLD_FLAGS_LOCALE) {
2517 /* Treat a UTF-8 locale as not being in locale at all */
2518 if (IN_UTF8_CTYPE_LOCALE) {
2519 flags &= ~FOLD_FLAGS_LOCALE;
2522 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2523 goto needs_full_generality;
2528 return _to_fold_latin1((U8) c, p, lenp,
2529 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2532 /* Here, above 255. If no special needs, just use the macro */
2533 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2534 uvchr_to_utf8(p, c);
2535 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2537 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2538 the special flags. */
2539 U8 utf8_c[UTF8_MAXBYTES + 1];
2541 needs_full_generality:
2542 uvchr_to_utf8(utf8_c, c);
2543 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
2547 PERL_STATIC_INLINE bool
2548 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2549 const char *const swashname, SV* const invlist)
2551 /* returns a boolean giving whether or not the UTF8-encoded character that
2552 * starts at <p> is in the swash indicated by <swashname>. <swash>
2553 * contains a pointer to where the swash indicated by <swashname>
2554 * is to be stored; which this routine will do, so that future calls will
2555 * look at <*swash> and only generate a swash if it is not null. <invlist>
2556 * is NULL or an inversion list that defines the swash. If not null, it
2557 * saves time during initialization of the swash.
2559 * Note that it is assumed that the buffer length of <p> is enough to
2560 * contain all the bytes that comprise the character. Thus, <*p> should
2561 * have been checked before this call for mal-formedness enough to assure
2564 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2566 /* The API should have included a length for the UTF-8 character in <p>,
2567 * but it doesn't. We therefore assume that p has been validated at least
2568 * as far as there being enough bytes available in it to accommodate the
2569 * character without reading beyond the end, and pass that number on to the
2570 * validating routine */
2571 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2572 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2573 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2575 NOT_REACHED; /* NOTREACHED */
2579 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2580 *swash = _core_swash_init("utf8",
2582 /* Only use the name if there is no inversion
2583 * list; otherwise will go out to disk */
2584 (invlist) ? "" : swashname,
2586 &PL_sv_undef, 1, 0, invlist, &flags);
2589 return swash_fetch(*swash, p, TRUE) != 0;
2592 PERL_STATIC_INLINE bool
2593 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2594 const char *const swashname, SV* const invlist)
2596 /* returns a boolean giving whether or not the UTF8-encoded character that
2597 * starts at <p>, and extending no further than <e - 1> is in the swash
2598 * indicated by <swashname>. <swash> contains a pointer to where the swash
2599 * indicated by <swashname> is to be stored; which this routine will do, so
2600 * that future calls will look at <*swash> and only generate a swash if it
2601 * is not null. <invlist> is NULL or an inversion list that defines the
2602 * swash. If not null, it saves time during initialization of the swash.
2605 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2607 if (! isUTF8_CHAR(p, e)) {
2608 _force_out_malformed_utf8_message(p, e, 0, 1);
2609 NOT_REACHED; /* NOTREACHED */
2613 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2614 *swash = _core_swash_init("utf8",
2616 /* Only use the name if there is no inversion
2617 * list; otherwise will go out to disk */
2618 (invlist) ? "" : swashname,
2620 &PL_sv_undef, 1, 0, invlist, &flags);
2623 return swash_fetch(*swash, p, TRUE) != 0;
2627 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2628 const char * const alternative,
2629 const bool use_locale,
2630 const char * const file,
2631 const unsigned line)
2635 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2637 if (ckWARN_d(WARN_DEPRECATED)) {
2639 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2640 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2641 if (! PL_seen_deprecated_macro) {
2642 PL_seen_deprecated_macro = newHV();
2644 if (! hv_store(PL_seen_deprecated_macro, key,
2645 strlen(key), &PL_sv_undef, 0))
2647 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2650 if (instr(file, "mathoms.c")) {
2651 Perl_warner(aTHX_ WARN_DEPRECATED,
2652 "In %s, line %d, starting in Perl v5.30, %s()"
2653 " will be removed. Avoid this message by"
2654 " converting to use %s().\n",
2655 file, line, name, alternative);
2658 Perl_warner(aTHX_ WARN_DEPRECATED,
2659 "In %s, line %d, starting in Perl v5.30, %s() will"
2660 " require an additional parameter. Avoid this"
2661 " message by converting to use %s().\n",
2662 file, line, name, alternative);
2669 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2671 const char * const name,
2672 const char * const alternative,
2673 const bool use_utf8,
2674 const bool use_locale,
2675 const char * const file,
2676 const unsigned line)
2678 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2680 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2682 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2692 case _CC_ALPHANUMERIC:
2696 return is_utf8_common(p,
2697 &PL_utf8_swash_ptrs[classnum],
2698 swash_property_names[classnum],
2699 PL_XPosix_ptrs[classnum]);
2702 return is_XPERLSPACE_high(p);
2704 return is_HORIZWS_high(p);
2706 return is_XDIGIT_high(p);
2712 return is_VERTWS_high(p);
2714 if (! PL_utf8_perl_idstart) {
2715 PL_utf8_perl_idstart
2716 = _new_invlist_C_array(_Perl_IDStart_invlist);
2718 return is_utf8_common(p, &PL_utf8_perl_idstart,
2719 "_Perl_IDStart", NULL);
2721 if (! PL_utf8_perl_idcont) {
2723 = _new_invlist_C_array(_Perl_IDCont_invlist);
2725 return is_utf8_common(p, &PL_utf8_perl_idcont,
2726 "_Perl_IDCont", NULL);
2730 /* idcont is the same as wordchar below 256 */
2731 if (classnum == _CC_IDCONT) {
2732 classnum = _CC_WORDCHAR;
2734 else if (classnum == _CC_IDFIRST) {
2738 classnum = _CC_ALPHA;
2742 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2743 return _generic_isCC(*p, classnum);
2746 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2749 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2750 return isFOO_lc(classnum, *p);
2753 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2756 NOT_REACHED; /* NOTREACHED */
2760 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2763 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2765 assert(classnum < _FIRST_NON_SWASH_CC);
2767 return is_utf8_common_with_len(p,
2769 &PL_utf8_swash_ptrs[classnum],
2770 swash_property_names[classnum],
2771 PL_XPosix_ptrs[classnum]);
2775 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2779 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2781 if (! PL_utf8_perl_idstart) {
2782 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2784 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2785 "_Perl_IDStart", invlist);
2789 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2791 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2795 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2799 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2803 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2805 if (! PL_utf8_perl_idcont) {
2806 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2808 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2809 "_Perl_IDCont", invlist);
2813 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2815 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2817 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2821 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2823 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2825 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2829 Perl__is_utf8_mark(pTHX_ const U8 *p)
2831 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2833 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2836 /* change namve uv1 to 'from' */
2838 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2839 SV **swashp, const char *normal, const char *special)
2843 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2845 /* For code points that don't change case, we already know that the output
2846 * of this function is the unchanged input, so we can skip doing look-ups
2847 * for them. Unfortunately the case-changing code points are scattered
2848 * around. But there are some long consecutive ranges where there are no
2849 * case changing code points. By adding tests, we can eliminate the lookup
2850 * for all the ones in such ranges. This is currently done here only for
2851 * just a few cases where the scripts are in common use in modern commerce
2852 * (and scripts adjacent to those which can be included without additional
2855 if (uv1 >= 0x0590) {
2856 /* This keeps from needing further processing the code points most
2857 * likely to be used in the following non-cased scripts: Hebrew,
2858 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2859 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2860 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2865 /* The following largish code point ranges also don't have case
2866 * changes, but khw didn't think they warranted extra tests to speed
2867 * them up (which would slightly slow down everything else above them):
2868 * 1100..139F Hangul Jamo, Ethiopic
2869 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2870 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2871 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2872 * Combining Diacritical Marks Extended, Balinese,
2873 * Sundanese, Batak, Lepcha, Ol Chiki
2874 * 2000..206F General Punctuation
2877 if (uv1 >= 0x2D30) {
2879 /* This keeps the from needing further processing the code points
2880 * most likely to be used in the following non-cased major scripts:
2881 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2883 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2884 * event that Unicode eventually allocates the unused block as of
2885 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2886 * that the test suite will start having failures to alert you
2887 * should that happen) */
2892 if (uv1 >= 0xAC00) {
2893 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2894 if (ckWARN_d(WARN_SURROGATE)) {
2895 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2896 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2897 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2902 /* AC00..FAFF Catches Hangul syllables and private use, plus
2909 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2910 if (UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)) {
2911 Perl_croak(aTHX_ cp_above_legal_max, uv1,
2912 MAX_NON_DEPRECATED_CP);
2914 if (ckWARN_d(WARN_NON_UNICODE)) {
2915 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2916 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2917 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2921 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2923 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2926 /* As of this writing, this means we avoid swash creation
2927 * for anything beyond low Plane 1 */
2934 /* Note that non-characters are perfectly legal, so no warning should
2935 * be given. There are so few of them, that it isn't worth the extra
2936 * tests to avoid swash creation */
2939 if (!*swashp) /* load on-demand */
2940 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2943 /* It might be "special" (sometimes, but not always,
2944 * a multicharacter mapping) */
2948 /* If passed in the specials name, use that; otherwise use any
2949 * given in the swash */
2950 if (*special != '\0') {
2951 hv = get_hv(special, 0);
2954 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2956 hv = MUTABLE_HV(SvRV(*svp));
2961 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2966 s = SvPV_const(*svp, len);
2969 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2971 Copy(s, ustrp, len, U8);
2976 if (!len && *swashp) {
2977 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2980 /* It was "normal" (a single character mapping). */
2981 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2989 return valid_utf8_to_uvchr(ustrp, 0);
2992 /* Here, there was no mapping defined, which means that the code point maps
2993 * to itself. Return the inputs */
2996 if (p != ustrp) { /* Don't copy onto itself */
2997 Copy(p, ustrp, len, U8);
3008 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
3010 /* This is called when changing the case of a UTF-8-encoded character above
3011 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3012 * result contains a character that crosses the 255/256 boundary, disallow
3013 * the change, and return the original code point. See L<perlfunc/lc> for
3016 * p points to the original string whose case was changed; assumed
3017 * by this routine to be well-formed
3018 * result the code point of the first character in the changed-case string
3019 * ustrp points to the changed-case string (<result> represents its first char)
3020 * lenp points to the length of <ustrp> */
3022 UV original; /* To store the first code point of <p> */
3024 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3026 assert(UTF8_IS_ABOVE_LATIN1(*p));
3028 /* We know immediately if the first character in the string crosses the
3029 * boundary, so can skip */
3032 /* Look at every character in the result; if any cross the
3033 * boundary, the whole thing is disallowed */
3034 U8* s = ustrp + UTF8SKIP(ustrp);
3035 U8* e = ustrp + *lenp;
3037 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3043 /* Here, no characters crossed, result is ok as-is, but we warn. */
3044 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3050 /* Failed, have to return the original */
3051 original = valid_utf8_to_uvchr(p, lenp);
3053 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3054 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3055 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
3056 "resolved to \"\\x{%" UVXf "}\".",
3060 Copy(p, ustrp, *lenp, char);
3065 S_check_and_deprecate(pTHX_ const U8 *p,
3067 const unsigned int type, /* See below */
3068 const bool use_locale, /* Is this a 'LC_'
3070 const char * const file,
3071 const unsigned line)
3073 /* This is a temporary function to deprecate the unsafe calls to the case
3074 * changing macros and functions. It keeps all the special stuff in just
3077 * It updates *e with the pointer to the end of the input string. If using
3078 * the old-style macros, *e is NULL on input, and so this function assumes
3079 * the input string is long enough to hold the entire UTF-8 sequence, and
3080 * sets *e accordingly, but it then returns a flag to pass the
3081 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3082 * using the full length if possible.
3084 * It also does the assert that *e > p when *e is not NULL. This should be
3085 * migrated to the callers when this function gets deleted.
3087 * The 'type' parameter is used for the caller to specify which case
3088 * changing function this is called from: */
3090 # define DEPRECATE_TO_UPPER 0
3091 # define DEPRECATE_TO_TITLE 1
3092 # define DEPRECATE_TO_LOWER 2
3093 # define DEPRECATE_TO_FOLD 3
3095 U32 utf8n_flags = 0;
3097 const char * alternative;
3099 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3102 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3103 *e = p + UTF8SKIP(p);
3105 /* For mathoms.c calls, we use the function name we know is stored
3106 * there. It could be part of a larger path */
3107 if (type == DEPRECATE_TO_UPPER) {
3108 name = instr(file, "mathoms.c")
3111 alternative = "toUPPER_utf8_safe";
3113 else if (type == DEPRECATE_TO_TITLE) {
3114 name = instr(file, "mathoms.c")
3117 alternative = "toTITLE_utf8_safe";
3119 else if (type == DEPRECATE_TO_LOWER) {
3120 name = instr(file, "mathoms.c")
3123 alternative = "toLOWER_utf8_safe";
3125 else if (type == DEPRECATE_TO_FOLD) {
3126 name = instr(file, "mathoms.c")
3129 alternative = "toFOLD_utf8_safe";
3131 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3133 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3142 /* The process for changing the case is essentially the same for the four case
3143 * change types, except there are complications for folding. Otherwise the
3144 * difference is only which case to change to. To make sure that they all do
3145 * the same thing, the bodies of the functions are extracted out into the
3146 * following two macros. The functions are written with the same variable
3147 * names, and these are known and used inside these macros. It would be
3148 * better, of course, to have inline functions to do it, but since different
3149 * macros are called, depending on which case is being changed to, this is not
3150 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3151 * function can start with the common start macro, then finish with its special
3152 * handling; while the other three cases can just use the common end macro.
3154 * The algorithm is to use the proper (passed in) macro or function to change
3155 * the case for code points that are below 256. The macro is used if using
3156 * locale rules for the case change; the function if not. If the code point is
3157 * above 255, it is computed from the input UTF-8, and another macro is called
3158 * to do the conversion. If necessary, the output is converted to UTF-8. If
3159 * using a locale, we have to check that the change did not cross the 255/256
3160 * boundary, see check_locale_boundary_crossing() for further details.
3162 * The macros are split with the correct case change for the below-256 case
3163 * stored into 'result', and in the middle of an else clause for the above-255
3164 * case. At that point in the 'else', 'result' is not the final result, but is
3165 * the input code point calculated from the UTF-8. The fold code needs to
3166 * realize all this and take it from there.
3168 * If you read the two macros as sequential, it's easier to understand what's
3170 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3171 L1_func_extra_param) \
3173 if (flags & (locale_flags)) { \
3174 /* Treat a UTF-8 locale as not being in locale at all */ \
3175 if (IN_UTF8_CTYPE_LOCALE) { \
3176 flags &= ~(locale_flags); \
3179 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3183 if (UTF8_IS_INVARIANT(*p)) { \
3184 if (flags & (locale_flags)) { \
3185 result = LC_L1_change_macro(*p); \
3188 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3191 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3192 if (flags & (locale_flags)) { \
3193 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3197 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3198 ustrp, lenp, L1_func_extra_param); \
3201 else { /* malformed UTF-8 or ord above 255 */ \
3202 STRLEN len_result; \
3203 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3204 if (len_result == (STRLEN) -1) { \
3205 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3209 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3210 result = change_macro(result, p, ustrp, lenp); \
3212 if (flags & (locale_flags)) { \
3213 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3218 /* Here, used locale rules. Convert back to UTF-8 */ \
3219 if (UTF8_IS_INVARIANT(result)) { \
3220 *ustrp = (U8) result; \
3224 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3225 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3232 =for apidoc to_utf8_upper
3234 Instead use L</toUPPER_utf8_safe>.
3238 /* Not currently externally documented, and subject to change:
3239 * <flags> is set iff iff the rules from the current underlying locale are to
3243 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3248 const char * const file,
3252 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3253 cBOOL(flags), file, line);
3255 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3257 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3258 /* 2nd char of uc(U+DF) is 'S' */
3259 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3260 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3264 =for apidoc to_utf8_title
3266 Instead use L</toTITLE_utf8_safe>.
3270 /* Not currently externally documented, and subject to change:
3271 * <flags> is set iff the rules from the current underlying locale are to be
3272 * used. Since titlecase is not defined in POSIX, for other than a
3273 * UTF-8 locale, uppercase is used instead for code points < 256.
3277 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3282 const char * const file,
3286 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3287 cBOOL(flags), file, line);
3289 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3291 /* 2nd char of ucfirst(U+DF) is 's' */
3292 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3293 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3297 =for apidoc to_utf8_lower
3299 Instead use L</toLOWER_utf8_safe>.
3303 /* Not currently externally documented, and subject to change:
3304 * <flags> is set iff iff the rules from the current underlying locale are to
3309 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3314 const char * const file,
3318 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3319 cBOOL(flags), file, line);
3321 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3323 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3324 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3328 =for apidoc to_utf8_fold
3330 Instead use L</toFOLD_utf8_safe>.
3334 /* Not currently externally documented, and subject to change,
3336 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3337 * locale are to be used.
3338 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3339 * otherwise simple folds
3340 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3345 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3350 const char * const file,
3354 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3355 cBOOL(flags), file, line);
3357 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3359 /* These are mutually exclusive */
3360 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3362 assert(p != ustrp); /* Otherwise overwrites */
3364 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3365 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3367 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3369 if (flags & FOLD_FLAGS_LOCALE) {
3371 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3372 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3374 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3375 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3377 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3379 /* Special case these two characters, as what normally gets
3380 * returned under locale doesn't work */
3381 if (UTF8SKIP(p) == cap_sharp_s_len
3382 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3384 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3385 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3386 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3387 "resolved to \"\\x{17F}\\x{17F}\".");
3392 if (UTF8SKIP(p) == long_s_t_len
3393 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3395 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3396 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3397 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3398 "resolved to \"\\x{FB06}\".");
3399 goto return_ligature_st;
3402 #if UNICODE_MAJOR_VERSION == 3 \
3403 && UNICODE_DOT_VERSION == 0 \
3404 && UNICODE_DOT_DOT_VERSION == 1
3405 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3407 /* And special case this on this Unicode version only, for the same
3408 * reaons the other two are special cased. They would cross the
3409 * 255/256 boundary which is forbidden under /l, and so the code
3410 * wouldn't catch that they are equivalent (which they are only in
3412 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3413 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3415 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3416 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3417 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3418 "resolved to \"\\x{0131}\".");
3419 goto return_dotless_i;
3423 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3425 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3429 /* This is called when changing the case of a UTF-8-encoded
3430 * character above the ASCII range, and the result should not
3431 * contain an ASCII character. */
3433 UV original; /* To store the first code point of <p> */
3435 /* Look at every character in the result; if any cross the
3436 * boundary, the whole thing is disallowed */
3438 U8* e = ustrp + *lenp;
3441 /* Crossed, have to return the original */
3442 original = valid_utf8_to_uvchr(p, lenp);
3444 /* But in these instances, there is an alternative we can
3445 * return that is valid */
3446 if (original == LATIN_SMALL_LETTER_SHARP_S
3447 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3448 || original == LATIN_CAPITAL_LETTER_SHARP_S
3453 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3454 goto return_ligature_st;
3456 #if UNICODE_MAJOR_VERSION == 3 \
3457 && UNICODE_DOT_VERSION == 0 \
3458 && UNICODE_DOT_DOT_VERSION == 1
3460 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3461 goto return_dotless_i;
3464 Copy(p, ustrp, *lenp, char);
3470 /* Here, no characters crossed, result is ok as-is */
3475 /* Here, used locale rules. Convert back to UTF-8 */
3476 if (UTF8_IS_INVARIANT(result)) {
3477 *ustrp = (U8) result;
3481 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3482 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3489 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3490 * folds to a string of two of these characters. By returning this
3491 * instead, then, e.g.,
3492 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3495 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3496 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3498 return LATIN_SMALL_LETTER_LONG_S;
3501 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3502 * have the other one fold to it */
3504 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3505 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3506 return LATIN_SMALL_LIGATURE_ST;
3508 #if UNICODE_MAJOR_VERSION == 3 \
3509 && UNICODE_DOT_VERSION == 0 \
3510 && UNICODE_DOT_DOT_VERSION == 1
3513 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3514 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3515 return LATIN_SMALL_LETTER_DOTLESS_I;
3522 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3523 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3524 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3528 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3530 PERL_ARGS_ASSERT_SWASH_INIT;
3532 /* Returns a copy of a swash initiated by the called function. This is the
3533 * public interface, and returning a copy prevents others from doing
3534 * mischief on the original */
3536 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3540 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3543 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3544 * use the following define */
3546 #define CORE_SWASH_INIT_RETURN(x) \
3547 PL_curpm= old_PL_curpm; \
3550 /* Initialize and return a swash, creating it if necessary. It does this
3551 * by calling utf8_heavy.pl in the general case. The returned value may be
3552 * the swash's inversion list instead if the input parameters allow it.
3553 * Which is returned should be immaterial to callers, as the only
3554 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3555 * and swash_to_invlist() handle both these transparently.
3557 * This interface should only be used by functions that won't destroy or
3558 * adversely change the swash, as doing so affects all other uses of the
3559 * swash in the program; the general public should use 'Perl_swash_init'
3562 * pkg is the name of the package that <name> should be in.
3563 * name is the name of the swash to find. Typically it is a Unicode
3564 * property name, including user-defined ones
3565 * listsv is a string to initialize the swash with. It must be of the form
3566 * documented as the subroutine return value in
3567 * L<perlunicode/User-Defined Character Properties>
3568 * minbits is the number of bits required to represent each data element.
3569 * It is '1' for binary properties.
3570 * none I (khw) do not understand this one, but it is used only in tr///.
3571 * invlist is an inversion list to initialize the swash with (or NULL)
3572 * flags_p if non-NULL is the address of various input and output flag bits
3573 * to the routine, as follows: ('I' means is input to the routine;
3574 * 'O' means output from the routine. Only flags marked O are
3575 * meaningful on return.)
3576 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3577 * came from a user-defined property. (I O)
3578 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3579 * when the swash cannot be located, to simply return NULL. (I)
3580 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3581 * return of an inversion list instead of a swash hash if this routine
3582 * thinks that would result in faster execution of swash_fetch() later
3585 * Thus there are three possible inputs to find the swash: <name>,
3586 * <listsv>, and <invlist>. At least one must be specified. The result
3587 * will be the union of the specified ones, although <listsv>'s various
3588 * actions can intersect, etc. what <name> gives. To avoid going out to
3589 * disk at all, <invlist> should specify completely what the swash should
3590 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3592 * <invlist> is only valid for binary properties */
3594 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3596 SV* retval = &PL_sv_undef;
3597 HV* swash_hv = NULL;
3598 const int invlist_swash_boundary =
3599 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3600 ? 512 /* Based on some benchmarking, but not extensive, see commit
3602 : -1; /* Never return just an inversion list */
3604 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3605 assert(! invlist || minbits == 1);
3607 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3608 that triggered the swash init and the swash init perl logic itself.
3611 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3613 if (listsv != &PL_sv_undef || strNE(name, "")) {
3615 const size_t pkg_len = strlen(pkg);
3616 const size_t name_len = strlen(name);
3617 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3621 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3623 PUSHSTACKi(PERLSI_MAGIC);
3627 /* We might get here via a subroutine signature which uses a utf8
3628 * parameter name, at which point PL_subname will have been set
3629 * but not yet used. */
3630 save_item(PL_subname);
3631 if (PL_parser && PL_parser->error_count)
3632 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3633 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3634 if (!method) { /* demand load UTF-8 */
3636 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3637 GvSV(PL_errgv) = NULL;
3638 #ifndef NO_TAINT_SUPPORT
3639 /* It is assumed that callers of this routine are not passing in
3640 * any user derived data. */
3641 /* Need to do this after save_re_context() as it will set
3642 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3643 * in Perl_magic_get). Even line to create errsv_save can turn on
3645 SAVEBOOL(TAINT_get);
3648 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3651 /* Not ERRSV, as there is no need to vivify a scalar we are
3652 about to discard. */
3653 SV * const errsv = GvSV(PL_errgv);
3654 if (!SvTRUE(errsv)) {
3655 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3656 SvREFCNT_dec(errsv);
3664 mPUSHp(pkg, pkg_len);
3665 mPUSHp(name, name_len);
3670 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3671 GvSV(PL_errgv) = NULL;
3672 /* If we already have a pointer to the method, no need to use
3673 * call_method() to repeat the lookup. */
3675 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3676 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3678 retval = *PL_stack_sp--;
3679 SvREFCNT_inc(retval);
3682 /* Not ERRSV. See above. */
3683 SV * const errsv = GvSV(PL_errgv);
3684 if (!SvTRUE(errsv)) {
3685 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3686 SvREFCNT_dec(errsv);
3691 if (IN_PERL_COMPILETIME) {
3692 CopHINTS_set(PL_curcop, PL_hints);
3694 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3695 if (SvPOK(retval)) {
3697 /* If caller wants to handle missing properties, let them */
3698 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3699 CORE_SWASH_INIT_RETURN(NULL);
3702 "Can't find Unicode property definition \"%" SVf "\"",
3704 NOT_REACHED; /* NOTREACHED */
3707 } /* End of calling the module to find the swash */
3709 /* If this operation fetched a swash, and we will need it later, get it */
3710 if (retval != &PL_sv_undef
3711 && (minbits == 1 || (flags_p
3713 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3715 swash_hv = MUTABLE_HV(SvRV(retval));
3717 /* If we don't already know that there is a user-defined component to
3718 * this swash, and the user has indicated they wish to know if there is
3719 * one (by passing <flags_p>), find out */
3720 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3721 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3722 if (user_defined && SvUV(*user_defined)) {
3723 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3728 /* Make sure there is an inversion list for binary properties */
3730 SV** swash_invlistsvp = NULL;
3731 SV* swash_invlist = NULL;
3732 bool invlist_in_swash_is_valid = FALSE;
3733 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3734 an unclaimed reference count */
3736 /* If this operation fetched a swash, get its already existing
3737 * inversion list, or create one for it */
3740 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3741 if (swash_invlistsvp) {
3742 swash_invlist = *swash_invlistsvp;
3743 invlist_in_swash_is_valid = TRUE;
3746 swash_invlist = _swash_to_invlist(retval);
3747 swash_invlist_unclaimed = TRUE;
3751 /* If an inversion list was passed in, have to include it */
3754 /* Any fetched swash will by now have an inversion list in it;
3755 * otherwise <swash_invlist> will be NULL, indicating that we
3756 * didn't fetch a swash */
3757 if (swash_invlist) {
3759 /* Add the passed-in inversion list, which invalidates the one
3760 * already stored in the swash */
3761 invlist_in_swash_is_valid = FALSE;
3762 SvREADONLY_off(swash_invlist); /* Turned on again below */
3763 _invlist_union(invlist, swash_invlist, &swash_invlist);
3767 /* Here, there is no swash already. Set up a minimal one, if
3768 * we are going to return a swash */
3769 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3771 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3773 swash_invlist = invlist;
3777 /* Here, we have computed the union of all the passed-in data. It may
3778 * be that there was an inversion list in the swash which didn't get
3779 * touched; otherwise save the computed one */
3780 if (! invlist_in_swash_is_valid
3781 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3783 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3785 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3787 /* We just stole a reference count. */
3788 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3789 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3792 /* The result is immutable. Forbid attempts to change it. */
3793 SvREADONLY_on(swash_invlist);
3795 /* Use the inversion list stand-alone if small enough */
3796 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3797 SvREFCNT_dec(retval);
3798 if (!swash_invlist_unclaimed)
3799 SvREFCNT_inc_simple_void_NN(swash_invlist);
3800 retval = newRV_noinc(swash_invlist);
3804 CORE_SWASH_INIT_RETURN(retval);
3805 #undef CORE_SWASH_INIT_RETURN
3809 /* This API is wrong for special case conversions since we may need to
3810 * return several Unicode characters for a single Unicode character
3811 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3812 * the lower-level routine, and it is similarly broken for returning
3813 * multiple values. --jhi
3814 * For those, you should use S__to_utf8_case() instead */
3815 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3818 * Returns the value of property/mapping C<swash> for the first character
3819 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3820 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3821 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3823 * A "swash" is a hash which contains initially the keys/values set up by
3824 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3825 * property for all possible code points. Things are stored in a compact form
3826 * (see utf8_heavy.pl) so that calculation is required to find the actual
3827 * property value for a given code point. As code points are looked up, new
3828 * key/value pairs are added to the hash, so that the calculation doesn't have
3829 * to ever be re-done. Further, each calculation is done, not just for the
3830 * desired one, but for a whole block of code points adjacent to that one.
3831 * For binary properties on ASCII machines, the block is usually for 64 code
3832 * points, starting with a code point evenly divisible by 64. Thus if the
3833 * property value for code point 257 is requested, the code goes out and
3834 * calculates the property values for all 64 code points between 256 and 319,
3835 * and stores these as a single 64-bit long bit vector, called a "swatch",
3836 * under the key for code point 256. The key is the UTF-8 encoding for code
3837 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3838 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3839 * for code point 258 is then requested, this code realizes that it would be
3840 * stored under the key for 256, and would find that value and extract the
3841 * relevant bit, offset from 256.
3843 * Non-binary properties are stored in as many bits as necessary to represent
3844 * their values (32 currently, though the code is more general than that), not
3845 * as single bits, but the principle is the same: the value for each key is a
3846 * vector that encompasses the property values for all code points whose UTF-8
3847 * representations are represented by the key. That is, for all code points
3848 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3852 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3854 HV *const hv = MUTABLE_HV(SvRV(swash));
3859 const U8 *tmps = NULL;
3863 PERL_ARGS_ASSERT_SWASH_FETCH;
3865 /* If it really isn't a hash, it isn't really swash; must be an inversion
3867 if (SvTYPE(hv) != SVt_PVHV) {
3868 return _invlist_contains_cp((SV*)hv,
3870 ? valid_utf8_to_uvchr(ptr, NULL)
3874 /* We store the values in a "swatch" which is a vec() value in a swash
3875 * hash. Code points 0-255 are a single vec() stored with key length
3876 * (klen) 0. All other code points have a UTF-8 representation
3877 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3878 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3879 * length for them is the length of the encoded char - 1. ptr[klen] is the
3880 * final byte in the sequence representing the character */
3881 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3886 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3889 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3892 klen = UTF8SKIP(ptr) - 1;
3894 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3895 * the vec is the final byte in the sequence. (In EBCDIC this is
3896 * converted to I8 to get consecutive values.) To help you visualize
3898 * Straight 1047 After final byte
3899 * UTF-8 UTF-EBCDIC I8 transform
3900 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3901 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3903 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3904 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3906 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3907 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3909 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3910 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3912 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3913 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3915 * (There are no discontinuities in the elided (...) entries.)
3916 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3917 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3918 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3919 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3920 * index into the vec() swatch (after subtracting 0x80, which we
3921 * actually do with an '&').
3922 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3923 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3924 * dicontinuities which go away by transforming it into I8, and we
3925 * effectively subtract 0xA0 to get the index. */
3926 needents = (1 << UTF_ACCUMULATION_SHIFT);
3927 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3931 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3932 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3933 * it's nothing to sniff at.) Pity we usually come through at least
3934 * two function calls to get here...
3936 * NB: this code assumes that swatches are never modified, once generated!
3939 if (hv == PL_last_swash_hv &&
3940 klen == PL_last_swash_klen &&
3941 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3943 tmps = PL_last_swash_tmps;
3944 slen = PL_last_swash_slen;
3947 /* Try our second-level swatch cache, kept in a hash. */
3948 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3950 /* If not cached, generate it via swatch_get */
3951 if (!svp || !SvPOK(*svp)
3952 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3955 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3956 swatch = swatch_get(swash,
3957 code_point & ~((UV)needents - 1),
3960 else { /* For the first 256 code points, the swatch has a key of
3962 swatch = swatch_get(swash, 0, needents);
3965 if (IN_PERL_COMPILETIME)
3966 CopHINTS_set(PL_curcop, PL_hints);
3968 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3970 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3971 || (slen << 3) < needents)
3972 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3973 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3974 svp, tmps, (UV)slen, (UV)needents);
3977 PL_last_swash_hv = hv;
3978 assert(klen <= sizeof(PL_last_swash_key));
3979 PL_last_swash_klen = (U8)klen;
3980 /* FIXME change interpvar.h? */
3981 PL_last_swash_tmps = (U8 *) tmps;
3982 PL_last_swash_slen = slen;
3984 Copy(ptr, PL_last_swash_key, klen, U8);
3987 switch ((int)((slen << 3) / needents)) {
3989 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3991 return ((UV) tmps[off]);
3995 ((UV) tmps[off ] << 8) +
3996 ((UV) tmps[off + 1]);
4000 ((UV) tmps[off ] << 24) +
4001 ((UV) tmps[off + 1] << 16) +
4002 ((UV) tmps[off + 2] << 8) +
4003 ((UV) tmps[off + 3]);
4005 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4006 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4007 NORETURN_FUNCTION_END;
4010 /* Read a single line of the main body of the swash input text. These are of
4013 * where each number is hex. The first two numbers form the minimum and
4014 * maximum of a range, and the third is the value associated with the range.
4015 * Not all swashes should have a third number
4017 * On input: l points to the beginning of the line to be examined; it points
4018 * to somewhere in the string of the whole input text, and is
4019 * terminated by a \n or the null string terminator.
4020 * lend points to the null terminator of that string
4021 * wants_value is non-zero if the swash expects a third number
4022 * typestr is the name of the swash's mapping, like 'ToLower'
4023 * On output: *min, *max, and *val are set to the values read from the line.
4024 * returns a pointer just beyond the line examined. If there was no
4025 * valid min number on the line, returns lend+1
4029 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4030 const bool wants_value, const U8* const typestr)
4032 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4033 STRLEN numlen; /* Length of the number */
4034 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4035 | PERL_SCAN_DISALLOW_PREFIX
4036 | PERL_SCAN_SILENT_NON_PORTABLE;
4038 /* nl points to the next \n in the scan */
4039 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4041 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4043 /* Get the first number on the line: the range minimum */
4045 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4046 *max = *min; /* So can never return without setting max */
4047 if (numlen) /* If found a hex number, position past it */
4049 else if (nl) { /* Else, go handle next line, if any */
4050 return nl + 1; /* 1 is length of "\n" */
4052 else { /* Else, no next line */
4053 return lend + 1; /* to LIST's end at which \n is not found */
4056 /* The max range value follows, separated by a BLANK */
4059 flags = PERL_SCAN_SILENT_ILLDIGIT
4060 | PERL_SCAN_DISALLOW_PREFIX
4061 | PERL_SCAN_SILENT_NON_PORTABLE;
4063 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4066 else /* If no value here, it is a single element range */
4069 /* Non-binary tables have a third entry: what the first element of the
4070 * range maps to. The map for those currently read here is in hex */
4074 flags = PERL_SCAN_SILENT_ILLDIGIT
4075 | PERL_SCAN_DISALLOW_PREFIX
4076 | PERL_SCAN_SILENT_NON_PORTABLE;
4078 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4087 /* diag_listed_as: To%s: illegal mapping '%s' */
4088 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4094 *val = 0; /* bits == 1, then any val should be ignored */
4096 else { /* Nothing following range min, should be single element with no
4101 /* diag_listed_as: To%s: illegal mapping '%s' */
4102 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4106 *val = 0; /* bits == 1, then val should be ignored */
4109 /* Position to next line if any, or EOF */
4119 * Returns a swatch (a bit vector string) for a code point sequence
4120 * that starts from the value C<start> and comprises the number C<span>.
4121 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4122 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4125 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4128 U8 *l, *lend, *x, *xend, *s, *send;
4129 STRLEN lcur, xcur, scur;
4130 HV *const hv = MUTABLE_HV(SvRV(swash));
4131 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4133 SV** listsvp = NULL; /* The string containing the main body of the table */
4134 SV** extssvp = NULL;
4135 SV** invert_it_svp = NULL;
4138 STRLEN octets; /* if bits == 1, then octets == 0 */
4140 UV end = start + span;
4142 if (invlistsvp == NULL) {
4143 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4144 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4145 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4146 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4147 listsvp = hv_fetchs(hv, "LIST", FALSE);
4148 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4150 bits = SvUV(*bitssvp);
4151 none = SvUV(*nonesvp);
4152 typestr = (U8*)SvPV_nolen(*typesvp);
4158 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4160 PERL_ARGS_ASSERT_SWATCH_GET;
4162 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4163 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4167 /* If overflowed, use the max possible */
4173 /* create and initialize $swatch */
4174 scur = octets ? (span * octets) : (span + 7) / 8;
4175 swatch = newSV(scur);
4177 s = (U8*)SvPVX(swatch);
4178 if (octets && none) {
4179 const U8* const e = s + scur;
4182 *s++ = (U8)(none & 0xff);
4183 else if (bits == 16) {
4184 *s++ = (U8)((none >> 8) & 0xff);
4185 *s++ = (U8)( none & 0xff);
4187 else if (bits == 32) {
4188 *s++ = (U8)((none >> 24) & 0xff);
4189 *s++ = (U8)((none >> 16) & 0xff);
4190 *s++ = (U8)((none >> 8) & 0xff);
4191 *s++ = (U8)( none & 0xff);
4197 (void)memzero((U8*)s, scur + 1);
4199 SvCUR_set(swatch, scur);
4200 s = (U8*)SvPVX(swatch);
4202 if (invlistsvp) { /* If has an inversion list set up use that */
4203 _invlist_populate_swatch(*invlistsvp, start, end, s);
4207 /* read $swash->{LIST} */
4208 l = (U8*)SvPV(*listsvp, lcur);
4211 UV min, max, val, upper;
4212 l = swash_scan_list_line(l, lend, &min, &max, &val,
4213 cBOOL(octets), typestr);
4218 /* If looking for something beyond this range, go try the next one */
4222 /* <end> is generally 1 beyond where we want to set things, but at the
4223 * platform's infinity, where we can't go any higher, we want to
4224 * include the code point at <end> */
4227 : (max != UV_MAX || end != UV_MAX)
4234 if (!none || val < none) {
4239 for (key = min; key <= upper; key++) {
4241 /* offset must be non-negative (start <= min <= key < end) */
4242 offset = octets * (key - start);
4244 s[offset] = (U8)(val & 0xff);
4245 else if (bits == 16) {
4246 s[offset ] = (U8)((val >> 8) & 0xff);
4247 s[offset + 1] = (U8)( val & 0xff);
4249 else if (bits == 32) {
4250 s[offset ] = (U8)((val >> 24) & 0xff);
4251 s[offset + 1] = (U8)((val >> 16) & 0xff);
4252 s[offset + 2] = (U8)((val >> 8) & 0xff);
4253 s[offset + 3] = (U8)( val & 0xff);
4256 if (!none || val < none)
4260 else { /* bits == 1, then val should be ignored */
4265 for (key = min; key <= upper; key++) {
4266 const STRLEN offset = (STRLEN)(key - start);
4267 s[offset >> 3] |= 1 << (offset & 7);
4272 /* Invert if the data says it should be. Assumes that bits == 1 */
4273 if (invert_it_svp && SvUV(*invert_it_svp)) {
4275 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4276 * be 0, and their inversion should also be 0, as we don't succeed any
4277 * Unicode property matches for non-Unicode code points */
4278 if (start <= PERL_UNICODE_MAX) {
4280 /* The code below assumes that we never cross the
4281 * Unicode/above-Unicode boundary in a range, as otherwise we would
4282 * have to figure out where to stop flipping the bits. Since this
4283 * boundary is divisible by a large power of 2, and swatches comes
4284 * in small powers of 2, this should be a valid assumption */
4285 assert(start + span - 1 <= PERL_UNICODE_MAX);
4295 /* read $swash->{EXTRAS}
4296 * This code also copied to swash_to_invlist() below */
4297 x = (U8*)SvPV(*extssvp, xcur);
4305 SV **otherbitssvp, *other;
4309 const U8 opc = *x++;
4313 nl = (U8*)memchr(x, '\n', xend - x);
4315 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4317 x = nl + 1; /* 1 is length of "\n" */
4321 x = xend; /* to EXTRAS' end at which \n is not found */
4328 namelen = nl - namestr;
4332 namelen = xend - namestr;
4336 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4337 otherhv = MUTABLE_HV(SvRV(*othersvp));
4338 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4339 otherbits = (STRLEN)SvUV(*otherbitssvp);
4340 if (bits < otherbits)
4341 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4342 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4344 /* The "other" swatch must be destroyed after. */
4345 other = swatch_get(*othersvp, start, span);
4346 o = (U8*)SvPV(other, olen);
4349 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4351 s = (U8*)SvPV(swatch, slen);
4352 if (bits == 1 && otherbits == 1) {
4354 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4355 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4356 (UV)slen, (UV)olen);
4380 STRLEN otheroctets = otherbits >> 3;
4382 U8* const send = s + slen;
4387 if (otherbits == 1) {
4388 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4392 STRLEN vlen = otheroctets;
4400 if (opc == '+' && otherval)
4401 NOOP; /* replace with otherval */
4402 else if (opc == '!' && !otherval)
4404 else if (opc == '-' && otherval)
4406 else if (opc == '&' && !otherval)
4409 s += octets; /* no replacement */
4414 *s++ = (U8)( otherval & 0xff);
4415 else if (bits == 16) {
4416 *s++ = (U8)((otherval >> 8) & 0xff);
4417 *s++ = (U8)( otherval & 0xff);
4419 else if (bits == 32) {
4420 *s++ = (U8)((otherval >> 24) & 0xff);
4421 *s++ = (U8)((otherval >> 16) & 0xff);
4422 *s++ = (U8)((otherval >> 8) & 0xff);
4423 *s++ = (U8)( otherval & 0xff);
4427 sv_free(other); /* through with it! */
4433 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4436 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4437 * Can't be used on a property that is subject to user override, as it
4438 * relies on the value of SPECIALS in the swash which would be set by
4439 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4440 * for overridden properties
4442 * Returns a hash which is the inversion and closure of a swash mapping.
4443 * For example, consider the input lines:
4448 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4449 * 006C. The value for each key is an array. For 006C, the array would
4450 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4451 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4453 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4454 * keys are only code points that are folded-to, so it isn't a full closure.
4456 * Essentially, for any code point, it gives all the code points that map to
4457 * it, or the list of 'froms' for that point.
4459 * Currently it ignores any additions or deletions from other swashes,
4460 * looking at just the main body of the swash, and if there are SPECIALS
4461 * in the swash, at that hash
4463 * The specials hash can be extra code points, and most likely consists of
4464 * maps from single code points to multiple ones (each expressed as a string
4465 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4466 * However consider this possible input in the specials hash:
4467 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4468 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4470 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4471 * currently handle. But it also means that FB05 and FB06 are equivalent in
4472 * a 1-1 mapping which we should handle, and this relationship may not be in
4473 * the main table. Therefore this function examines all the multi-char
4474 * sequences and adds the 1-1 mappings that come out of that.
4476 * XXX This function was originally intended to be multipurpose, but its
4477 * only use is quite likely to remain for constructing the inversion of
4478 * the CaseFolding (//i) property. If it were more general purpose for
4479 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4480 * because certain folds are prohibited under /iaa and /il. As an example,
4481 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4482 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4483 * prohibited, so we would not figure out that they fold to each other.
4484 * Code could be written to automatically figure this out, similar to the
4485 * code that does this for multi-character folds, but this is the only case
4486 * where something like this is ever likely to happen, as all the single
4487 * char folds to the 0-255 range are now quite settled. Instead there is a
4488 * little special code that is compiled only for this Unicode version. This
4489 * is smaller and didn't require much coding time to do. But this makes
4490 * this routine strongly tied to being used just for CaseFolding. If ever
4491 * it should be generalized, this would have to be fixed */
4495 HV *const hv = MUTABLE_HV(SvRV(swash));
4497 /* The string containing the main body of the table. This will have its
4498 * assertion fail if the swash has been converted to its inversion list */
4499 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4501 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4502 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4503 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4504 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4505 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4506 const STRLEN bits = SvUV(*bitssvp);
4507 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4508 const UV none = SvUV(*nonesvp);
4509 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4513 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4515 /* Must have at least 8 bits to get the mappings */
4516 if (bits != 8 && bits != 16 && bits != 32) {
4517 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4521 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4522 mapping to more than one character */
4524 /* Construct an inverse mapping hash for the specials */
4525 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4526 HV * specials_inverse = newHV();
4527 char *char_from; /* the lhs of the map */
4528 I32 from_len; /* its byte length */
4529 char *char_to; /* the rhs of the map */
4530 I32 to_len; /* its byte length */
4531 SV *sv_to; /* and in a sv */
4532 AV* from_list; /* list of things that map to each 'to' */
4534 hv_iterinit(specials_hv);
4536 /* The keys are the characters (in UTF-8) that map to the corresponding
4537 * UTF-8 string value. Iterate through the list creating the inverse
4539 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4541 if (! SvPOK(sv_to)) {
4542 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4543 "unexpectedly is not a string, flags=%lu",
4544 (unsigned long)SvFLAGS(sv_to));
4546 /*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)));*/
4548 /* Each key in the inverse list is a mapped-to value, and the key's
4549 * hash value is a list of the strings (each in UTF-8) that map to
4550 * it. Those strings are all one character long */
4551 if ((listp = hv_fetch(specials_inverse,
4555 from_list = (AV*) *listp;
4557 else { /* No entry yet for it: create one */
4558 from_list = newAV();
4559 if (! hv_store(specials_inverse,
4562 (SV*) from_list, 0))
4564 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4568 /* Here have the list associated with this 'to' (perhaps newly
4569 * created and empty). Just add to it. Note that we ASSUME that
4570 * the input is guaranteed to not have duplications, so we don't
4571 * check for that. Duplications just slow down execution time. */
4572 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4575 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4576 * it looking for cases like the FB05/FB06 examples above. There would
4577 * be an entry in the hash like
4578 * 'st' => [ FB05, FB06 ]
4579 * In this example we will create two lists that get stored in the
4580 * returned hash, 'ret':
4581 * FB05 => [ FB05, FB06 ]
4582 * FB06 => [ FB05, FB06 ]
4584 * Note that there is nothing to do if the array only has one element.
4585 * (In the normal 1-1 case handled below, we don't have to worry about
4586 * two lists, as everything gets tied to the single list that is
4587 * generated for the single character 'to'. But here, we are omitting
4588 * that list, ('st' in the example), so must have multiple lists.) */
4589 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4590 &char_to, &to_len)))
4592 if (av_tindex_skip_len_mg(from_list) > 0) {
4595 /* We iterate over all combinations of i,j to place each code
4596 * point on each list */
4597 for (i = 0; i <= av_tindex_skip_len_mg(from_list); i++) {
4599 AV* i_list = newAV();
4600 SV** entryp = av_fetch(from_list, i, FALSE);
4601 if (entryp == NULL) {
4602 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4604 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4605 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4607 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4608 (SV*) i_list, FALSE))
4610 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4613 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4614 for (j = 0; j <= av_tindex_skip_len_mg(from_list); j++) {
4615 entryp = av_fetch(from_list, j, FALSE);
4616 if (entryp == NULL) {
4617 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4620 /* When i==j this adds itself to the list */
4621 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4622 (U8*) SvPVX(*entryp),
4623 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4625 /*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));*/
4630 SvREFCNT_dec(specials_inverse); /* done with it */
4631 } /* End of specials */
4633 /* read $swash->{LIST} */
4635 #if UNICODE_MAJOR_VERSION == 3 \
4636 && UNICODE_DOT_VERSION == 0 \
4637 && UNICODE_DOT_DOT_VERSION == 1
4639 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4640 * rule so that things work under /iaa and /il */
4642 SV * mod_listsv = sv_mortalcopy(*listsvp);
4643 sv_catpv(mod_listsv, "130\t130\t131\n");
4644 l = (U8*)SvPV(mod_listsv, lcur);
4648 l = (U8*)SvPV(*listsvp, lcur);
4654 /* Go through each input line */
4658 l = swash_scan_list_line(l, lend, &min, &max, &val,
4659 cBOOL(octets), typestr);
4664 /* Each element in the range is to be inverted */
4665 for (inverse = min; inverse <= max; inverse++) {
4669 bool found_key = FALSE;
4670 bool found_inverse = FALSE;
4672 /* The key is the inverse mapping */
4673 char key[UTF8_MAXBYTES+1];
4674 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4675 STRLEN key_len = key_end - key;
4677 /* Get the list for the map */
4678 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4679 list = (AV*) *listp;
4681 else { /* No entry yet for it: create one */
4683 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4684 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4688 /* Look through list to see if this inverse mapping already is
4689 * listed, or if there is a mapping to itself already */
4690 for (i = 0; i <= av_tindex_skip_len_mg(list); i++) {
4691 SV** entryp = av_fetch(list, i, FALSE);
4694 if (entryp == NULL) {
4695 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4699 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4703 if (uv == inverse) {
4704 found_inverse = TRUE;
4707 /* No need to continue searching if found everything we are
4709 if (found_key && found_inverse) {
4714 /* Make sure there is a mapping to itself on the list */
4716 av_push(list, newSVuv(val));
4717 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4721 /* Simply add the value to the list */
4722 if (! found_inverse) {
4723 av_push(list, newSVuv(inverse));
4724 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4727 /* swatch_get() increments the value of val for each element in the
4728 * range. That makes more compact tables possible. You can
4729 * express the capitalization, for example, of all consecutive
4730 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4731 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4732 * and it's not documented; it appears to be used only in
4733 * implementing tr//; I copied the semantics from swatch_get(), just
4735 if (!none || val < none) {
4745 Perl__swash_to_invlist(pTHX_ SV* const swash)
4748 /* Subject to change or removal. For use only in one place in regcomp.c.
4749 * Ownership is given to one reference count in the returned SV* */
4754 HV *const hv = MUTABLE_HV(SvRV(swash));
4755 UV elements = 0; /* Number of elements in the inversion list */
4765 STRLEN octets; /* if bits == 1, then octets == 0 */
4771 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4773 /* If not a hash, it must be the swash's inversion list instead */
4774 if (SvTYPE(hv) != SVt_PVHV) {
4775 return SvREFCNT_inc_simple_NN((SV*) hv);
4778 /* The string containing the main body of the table */
4779 listsvp = hv_fetchs(hv, "LIST", FALSE);
4780 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4781 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4782 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4783 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4785 typestr = (U8*)SvPV_nolen(*typesvp);
4786 bits = SvUV(*bitssvp);
4787 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4789 /* read $swash->{LIST} */
4790 if (SvPOK(*listsvp)) {
4791 l = (U8*)SvPV(*listsvp, lcur);
4794 /* LIST legitimately doesn't contain a string during compilation phases
4795 * of Perl itself, before the Unicode tables are generated. In this
4796 * case, just fake things up by creating an empty list */
4803 if (*l == 'V') { /* Inversion list format */
4804 const char *after_atou = (char *) lend;
4806 UV* other_elements_ptr;
4808 /* The first number is a count of the rest */
4810 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4811 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
4813 if (elements == 0) {
4814 invlist = _new_invlist(0);
4817 l = (U8 *) after_atou;
4819 /* Get the 0th element, which is needed to setup the inversion list */
4820 while (isSPACE(*l)) l++;
4821 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4822 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
4824 l = (U8 *) after_atou;
4825 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
4828 /* Then just populate the rest of the input */
4829 while (elements-- > 0) {
4831 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more elements than available", elements);
4833 while (isSPACE(*l)) l++;
4834 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
4835 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
4837 l = (U8 *) after_atou;
4843 /* Scan the input to count the number of lines to preallocate array
4844 * size based on worst possible case, which is each line in the input
4845 * creates 2 elements in the inversion list: 1) the beginning of a
4846 * range in the list; 2) the beginning of a range not in the list. */
4847 while ((loc = (strchr(loc, '\n'))) != NULL) {
4852 /* If the ending is somehow corrupt and isn't a new line, add another
4853 * element for the final range that isn't in the inversion list */
4854 if (! (*lend == '\n'
4855 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4860 invlist = _new_invlist(elements);
4862 /* Now go through the input again, adding each range to the list */
4865 UV val; /* Not used by this function */
4867 l = swash_scan_list_line(l, lend, &start, &end, &val,
4868 cBOOL(octets), typestr);
4874 invlist = _add_range_to_invlist(invlist, start, end);
4878 /* Invert if the data says it should be */
4879 if (invert_it_svp && SvUV(*invert_it_svp)) {
4880 _invlist_invert(invlist);
4883 /* This code is copied from swatch_get()
4884 * read $swash->{EXTRAS} */
4885 x = (U8*)SvPV(*extssvp, xcur);
4893 SV **otherbitssvp, *other;
4896 const U8 opc = *x++;
4900 nl = (U8*)memchr(x, '\n', xend - x);
4902 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4904 x = nl + 1; /* 1 is length of "\n" */
4908 x = xend; /* to EXTRAS' end at which \n is not found */
4915 namelen = nl - namestr;
4919 namelen = xend - namestr;
4923 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4924 otherhv = MUTABLE_HV(SvRV(*othersvp));
4925 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4926 otherbits = (STRLEN)SvUV(*otherbitssvp);
4928 if (bits != otherbits || bits != 1) {
4929 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4930 "properties, bits=%" UVuf ", otherbits=%" UVuf,
4931 (UV)bits, (UV)otherbits);
4934 /* The "other" swatch must be destroyed after. */
4935 other = _swash_to_invlist((SV *)*othersvp);
4937 /* End of code copied from swatch_get() */
4940 _invlist_union(invlist, other, &invlist);
4943 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4946 _invlist_subtract(invlist, other, &invlist);
4949 _invlist_intersection(invlist, other, &invlist);
4954 sv_free(other); /* through with it! */
4957 SvREADONLY_on(invlist);
4962 Perl__get_swash_invlist(pTHX_ SV* const swash)
4966 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4968 if (! SvROK(swash)) {
4972 /* If it really isn't a hash, it isn't really swash; must be an inversion
4974 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4978 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4987 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4989 /* May change: warns if surrogates, non-character code points, or
4990 * non-Unicode code points are in s which has length len bytes. Returns
4991 * TRUE if none found; FALSE otherwise. The only other validity check is
4992 * to make sure that this won't exceed the string's length.
4994 * Code points above the platform's C<IV_MAX> will raise a deprecation
4995 * warning, unless those are turned off. */
4997 const U8* const e = s + len;
5000 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
5003 if (UTF8SKIP(s) > len) {
5004 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
5005 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
5008 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
5009 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
5010 if ( ckWARN_d(WARN_NON_UNICODE)
5011 || ( ckWARN_d(WARN_DEPRECATED)
5013 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
5014 #else /* Below is 64-bit words */
5015 /* 2**63 and up meet these conditions provided we have
5019 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
5022 /* s[1] being above 0x80 overflows */
5027 /* A side effect of this function will be to warn */
5028 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
5032 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
5033 if (ckWARN_d(WARN_SURROGATE)) {
5034 /* This has a different warning than the one the called
5035 * function would output, so can't just call it, unlike we
5036 * do for the non-chars and above-unicodes */
5037 UV uv = utf8_to_uvchr_buf(s, e, NULL);
5038 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
5039 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", uv);
5043 else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
5044 /* A side effect of this function will be to warn */
5045 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5056 =for apidoc pv_uni_display
5058 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5059 length C<len>, the displayable version being at most C<pvlim> bytes long
5060 (if longer, the rest is truncated and C<"..."> will be appended).
5062 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5063 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5064 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5065 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5066 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5067 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5069 The pointer to the PV of the C<dsv> is returned.
5071 See also L</sv_uni_display>.
5075 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
5080 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5084 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5086 /* This serves double duty as a flag and a character to print after
5087 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5091 if (pvlim && SvCUR(dsv) >= pvlim) {
5095 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5097 const unsigned char c = (unsigned char)u & 0xFF;
5098 if (flags & UNI_DISPLAY_BACKSLASH) {
5115 const char string = ok;
5116 sv_catpvs(dsv, "\\");
5117 sv_catpvn(dsv, &string, 1);
5120 /* isPRINT() is the locale-blind version. */
5121 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5122 const char string = c;
5123 sv_catpvn(dsv, &string, 1);
5128 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5131 sv_catpvs(dsv, "...");
5137 =for apidoc sv_uni_display
5139 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5140 the displayable version being at most C<pvlim> bytes long
5141 (if longer, the rest is truncated and "..." will be appended).
5143 The C<flags> argument is as in L</pv_uni_display>().
5145 The pointer to the PV of the C<dsv> is returned.
5150 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5152 const char * const ptr =
5153 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5155 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5157 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5158 SvCUR(ssv), pvlim, flags);
5162 =for apidoc foldEQ_utf8
5164 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
5165 of which may be in UTF-8) are the same case-insensitively; false otherwise.
5166 How far into the strings to compare is determined by other input parameters.
5168 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5169 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
5170 with respect to C<s2>.
5172 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
5173 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
5174 scan will not be considered to be a match unless the goal is reached, and
5175 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
5178 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
5179 considered an end pointer to the position 1 byte past the maximum point
5180 in C<s1> beyond which scanning will not continue under any circumstances.
5181 (This routine assumes that UTF-8 encoded input strings are not malformed;
5182 malformed input can cause it to read past C<pe1>).
5183 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
5184 is less than C<s1>+C<l1>, the match will never be successful because it can
5186 get as far as its goal (and in fact is asserted against). Correspondingly for
5187 C<pe2> with respect to C<s2>.
5189 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5190 C<l2> must be non-zero), and if both do, both have to be
5191 reached for a successful match. Also, if the fold of a character is multiple
5192 characters, all of them must be matched (see tr21 reference below for
5195 Upon a successful match, if C<pe1> is non-C<NULL>,
5196 it will be set to point to the beginning of the I<next> character of C<s1>
5197 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5199 For case-insensitiveness, the "casefolding" of Unicode is used
5200 instead of upper/lowercasing both the characters, see
5201 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5205 /* A flags parameter has been added which may change, and hence isn't
5206 * externally documented. Currently it is:
5207 * 0 for as-documented above
5208 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5209 ASCII one, to not match
5210 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5211 * locale are to be used.
5212 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5213 * routine. This allows that step to be skipped.
5214 * Currently, this requires s1 to be encoded as UTF-8
5215 * (u1 must be true), which is asserted for.
5216 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5217 * cross certain boundaries. Hence, the caller should
5218 * let this function do the folding instead of
5219 * pre-folding. This code contains an assertion to
5220 * that effect. However, if the caller knows what
5221 * it's doing, it can pass this flag to indicate that,
5222 * and the assertion is skipped.
5223 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5224 * FOLDEQ_S2_FOLDS_SANE
5227 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)
5229 const U8 *p1 = (const U8*)s1; /* Point to current char */
5230 const U8 *p2 = (const U8*)s2;
5231 const U8 *g1 = NULL; /* goal for s1 */
5232 const U8 *g2 = NULL;
5233 const U8 *e1 = NULL; /* Don't scan s1 past this */
5234 U8 *f1 = NULL; /* Point to current folded */
5235 const U8 *e2 = NULL;
5237 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5238 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5239 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5240 U8 flags_for_folder = FOLD_FLAGS_FULL;
5242 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5244 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5245 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5246 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5247 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5248 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5249 /* The algorithm is to trial the folds without regard to the flags on
5250 * the first line of the above assert(), and then see if the result
5251 * violates them. This means that the inputs can't be pre-folded to a
5252 * violating result, hence the assert. This could be changed, with the
5253 * addition of extra tests here for the already-folded case, which would
5254 * slow it down. That cost is more than any possible gain for when these
5255 * flags are specified, as the flags indicate /il or /iaa matching which
5256 * is less common than /iu, and I (khw) also believe that real-world /il
5257 * and /iaa matches are most likely to involve code points 0-255, and this
5258 * function only under rare conditions gets called for 0-255. */
5260 if (flags & FOLDEQ_LOCALE) {
5261 if (IN_UTF8_CTYPE_LOCALE) {
5262 flags &= ~FOLDEQ_LOCALE;
5265 flags_for_folder |= FOLD_FLAGS_LOCALE;
5274 g1 = (const U8*)s1 + l1;
5282 g2 = (const U8*)s2 + l2;
5285 /* Must have at least one goal */
5290 /* Will never match if goal is out-of-bounds */
5291 assert(! e1 || e1 >= g1);
5293 /* Here, there isn't an end pointer, or it is beyond the goal. We
5294 * only go as far as the goal */
5298 assert(e1); /* Must have an end for looking at s1 */
5301 /* Same for goal for s2 */
5303 assert(! e2 || e2 >= g2);
5310 /* If both operands are already folded, we could just do a memEQ on the
5311 * whole strings at once, but it would be better if the caller realized
5312 * this and didn't even call us */
5314 /* Look through both strings, a character at a time */
5315 while (p1 < e1 && p2 < e2) {
5317 /* If at the beginning of a new character in s1, get its fold to use
5318 * and the length of the fold. */
5320 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5326 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5328 /* We have to forbid mixing ASCII with non-ASCII if the
5329 * flags so indicate. And, we can short circuit having to
5330 * call the general functions for this common ASCII case,
5331 * all of whose non-locale folds are also ASCII, and hence
5332 * UTF-8 invariants, so the UTF8ness of the strings is not
5334 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5338 *foldbuf1 = toFOLD(*p1);
5341 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5343 else { /* Not UTF-8, get UTF-8 fold */
5344 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5350 if (n2 == 0) { /* Same for s2 */
5351 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5357 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5358 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5362 *foldbuf2 = toFOLD(*p2);
5365 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5368 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5374 /* Here f1 and f2 point to the beginning of the strings to compare.
5375 * These strings are the folds of the next character from each input
5376 * string, stored in UTF-8. */
5378 /* While there is more to look for in both folds, see if they
5379 * continue to match */
5381 U8 fold_length = UTF8SKIP(f1);
5382 if (fold_length != UTF8SKIP(f2)
5383 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5384 function call for single
5386 || memNE((char*)f1, (char*)f2, fold_length))
5388 return 0; /* mismatch */
5391 /* Here, they matched, advance past them */
5398 /* When reach the end of any fold, advance the input past it */
5400 p1 += u1 ? UTF8SKIP(p1) : 1;
5403 p2 += u2 ? UTF8SKIP(p2) : 1;
5405 } /* End of loop through both strings */
5407 /* A match is defined by each scan that specified an explicit length
5408 * reaching its final goal, and the other not having matched a partial
5409 * character (which can happen when the fold of a character is more than one
5411 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5415 /* Successful match. Set output pointers */
5425 /* XXX The next two functions should likely be moved to mathoms.c once all
5426 * occurrences of them are removed from the core; some cpan-upstream modules
5430 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5432 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5434 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5438 =for apidoc utf8n_to_uvuni
5440 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5442 This function was useful for code that wanted to handle both EBCDIC and
5443 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5444 distinctions between the platforms have mostly been made invisible to most
5445 code, so this function is quite unlikely to be what you want. If you do need
5446 this precise functionality, use instead
5447 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5448 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5454 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5456 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5458 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5462 =for apidoc uvuni_to_utf8_flags
5464 Instead you almost certainly want to use L</uvchr_to_utf8> or
5465 L</uvchr_to_utf8_flags>.
5467 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5468 which itself, while not deprecated, should be used only in isolated
5469 circumstances. These functions were useful for code that wanted to handle
5470 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5471 v5.20, the distinctions between the platforms have mostly been made invisible
5472 to most code, so this function is quite unlikely to be what you want.
5478 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5480 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5482 return uvoffuni_to_utf8_flags(d, uv, flags);
5486 * ex: set ts=8 sts=4 sw=4 et: