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 deprecated; the permissible max is 0x%" UVXf ". This will be fatal in Perl 5.28";
42 #define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
45 =head1 Unicode Support
46 These are various utility functions for manipulating UTF8-encoded
47 strings. For the uninitiated, this is a method of representing arbitrary
48 Unicode characters as a variable number of bytes, in such a way that
49 characters in the ASCII range are unmodified, and a zero byte never appears
50 within non-zero characters.
56 Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
86 PL_curcop->cop_warnings = pWARN_ALL;
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
104 =for apidoc uvoffuni_to_utf8_flags
106 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
107 Instead, B<Almost all code should use L</uvchr_to_utf8> or
108 L</uvchr_to_utf8_flags>>.
110 This function is like them, but the input is a strict Unicode
111 (as opposed to native) code point. Only in very rare circumstances should code
112 not be using the native code point.
114 For details, see the description for L</uvchr_to_utf8_flags>.
119 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
121 if (flags & UNICODE_WARN_SURROGATE) { \
122 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
123 "UTF-16 surrogate U+%04" UVXf, uv); \
125 if (flags & UNICODE_DISALLOW_SURROGATE) { \
130 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
132 if (flags & UNICODE_WARN_NONCHAR) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
134 "Unicode non-character U+%04" UVXf " is not " \
135 "recommended for open interchange", uv); \
137 if (flags & UNICODE_DISALLOW_NONCHAR) { \
142 /* Use shorter names internally in this file */
143 #define SHIFT UTF_ACCUMULATION_SHIFT
145 #define MARK UTF_CONTINUATION_MARK
146 #define MASK UTF_CONTINUATION_MASK
149 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
151 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
153 if (OFFUNI_IS_INVARIANT(uv)) {
154 *d++ = LATIN1_TO_NATIVE(uv);
158 if (uv <= MAX_UTF8_TWO_BYTE) {
159 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
160 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
164 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
165 * below, the 16 is for start bytes E0-EF (which are all the possible ones
166 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
167 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
168 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
169 * 0x800-0xFFFF on ASCII */
170 if (uv < (16 * (1U << (2 * SHIFT)))) {
171 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
172 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
173 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
175 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
176 aren't tested here */
177 /* The most likely code points in this range are below the surrogates.
178 * Do an extra test to quickly exclude those. */
179 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
180 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
181 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
183 HANDLE_UNICODE_NONCHAR(uv, flags);
185 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
186 HANDLE_UNICODE_SURROGATE(uv, flags);
193 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
194 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
195 * happen starting with 4-byte characters on ASCII platforms. We unify the
196 * code for these with EBCDIC, even though some of them require 5-bytes on
197 * those, because khw believes the code saving is worth the very slight
198 * performance hit on these high EBCDIC code points. */
200 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
201 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
202 && ckWARN_d(WARN_DEPRECATED))
204 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
205 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
207 if ( (flags & UNICODE_WARN_SUPER)
208 || ( UNICODE_IS_ABOVE_31_BIT(uv)
209 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
211 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
213 /* Choose the more dire applicable warning */
214 (UNICODE_IS_ABOVE_31_BIT(uv))
215 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
216 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
219 if (flags & UNICODE_DISALLOW_SUPER
220 || ( UNICODE_IS_ABOVE_31_BIT(uv)
221 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
226 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
227 HANDLE_UNICODE_NONCHAR(uv, flags);
230 /* Test for and handle 4-byte result. In the test immediately below, the
231 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
232 * characters). The 3 is for 3 continuation bytes; these each contribute
233 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
234 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
235 * 0x1_0000-0x1F_FFFF on ASCII */
236 if (uv < (8 * (1U << (3 * SHIFT)))) {
237 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
238 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
239 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
240 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
242 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
243 characters. The end-plane non-characters for EBCDIC were
244 handled just above */
245 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
246 HANDLE_UNICODE_NONCHAR(uv, flags);
248 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
249 HANDLE_UNICODE_SURROGATE(uv, flags);
256 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
257 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
258 * format. The unrolled version above turns out to not save all that much
259 * time, and at these high code points (well above the legal Unicode range
260 * on ASCII platforms, and well above anything in common use in EBCDIC),
261 * khw believes that less code outweighs slight performance gains. */
264 STRLEN len = OFFUNISKIP(uv);
267 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
268 uv >>= UTF_ACCUMULATION_SHIFT;
270 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
276 =for apidoc uvchr_to_utf8
278 Adds the UTF-8 representation of the native code point C<uv> to the end
279 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
280 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
281 the byte after the end of the new character. In other words,
283 d = uvchr_to_utf8(d, uv);
285 is the recommended wide native character-aware way of saying
289 This function accepts any UV as input, but very high code points (above
290 C<IV_MAX> on the platform) will raise a deprecation warning. This is
291 typically 0x7FFF_FFFF in a 32-bit word.
293 It is possible to forbid or warn on non-Unicode code points, or those that may
294 be problematic by using L</uvchr_to_utf8_flags>.
299 /* This is also a macro */
300 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
303 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
305 return uvchr_to_utf8(d, uv);
309 =for apidoc uvchr_to_utf8_flags
311 Adds the UTF-8 representation of the native code point C<uv> to the end
312 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
313 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
314 the byte after the end of the new character. In other words,
316 d = uvchr_to_utf8_flags(d, uv, flags);
320 d = uvchr_to_utf8_flags(d, uv, 0);
322 This is the Unicode-aware way of saying
326 If C<flags> is 0, this function accepts any UV as input, but very high code
327 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
328 This is typically 0x7FFF_FFFF in a 32-bit word.
330 Specifying C<flags> can further restrict what is allowed and not warned on, as
333 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
334 the function will raise a warning, provided UTF8 warnings are enabled. If
335 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
336 NULL. If both flags are set, the function will both warn and return NULL.
338 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
339 affect how the function handles a Unicode non-character.
341 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
342 affect the handling of code points that are above the Unicode maximum of
343 0x10FFFF. Languages other than Perl may not be able to accept files that
346 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
347 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
348 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
349 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
350 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
351 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
352 above-Unicode and surrogate flags, but not the non-character ones, as
354 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
355 See L<perlunicode/Noncharacter code points>.
357 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
358 so using them is more problematic than other above-Unicode code points. Perl
359 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
360 likely that non-Perl languages will not be able to read files that contain
361 these that written by the perl interpreter; nor would Perl understand files
362 written by something that uses a different extension. For these reasons, there
363 is a separate set of flags that can warn and/or disallow these extremely high
364 code points, even if other above-Unicode ones are accepted. These are the
365 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
366 are entirely independent from the deprecation warning for code points above
367 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
368 code point that needs more than 31 bits to represent. When that happens,
369 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
370 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
371 above-Unicode code points, including these, as malformations; and
372 C<UNICODE_WARN_SUPER> warns on these.)
374 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
375 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
376 than on ASCII. Prior to that, code points 2**31 and higher were simply
377 unrepresentable, and a different, incompatible method was used to represent
378 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
379 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
380 platforms, warning and disallowing 2**31 and higher.
385 /* This is also a macro */
386 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
389 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
391 return uvchr_to_utf8_flags(d, uv, flags);
394 PERL_STATIC_INLINE bool
395 S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
397 /* Returns TRUE if the first code point represented by the Perl-extended-
398 * UTF-8-encoded string starting at 's', and looking no further than 'e -
399 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
401 * The function handles the case where the input bytes do not include all
402 * the ones necessary to represent a full character. That is, they may be
403 * the intial bytes of the representation of a code point, but possibly
404 * the final ones necessary for the complete representation may be beyond
407 * The function assumes that the sequence is well-formed UTF-8 as far as it
408 * goes, and is for a UTF-8 variant code point. If the sequence is
409 * incomplete, the function returns FALSE if there is any well-formed
410 * UTF-8 byte sequence that can complete it in such a way that a code point
411 * < 2**31 is produced; otherwise it returns TRUE.
413 * Getting this exactly right is slightly tricky, and has to be done in
414 * several places in this file, so is centralized here. It is based on the
417 * U+7FFFFFFF (2 ** 31 - 1)
418 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
419 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
420 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
421 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
422 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
423 * U+80000000 (2 ** 31):
424 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
425 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
426 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
427 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
428 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
429 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
434 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
435 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
436 const STRLEN prefix_len = sizeof(prefix) - 1;
437 const STRLEN len = e - s;
438 const STRLEN cmp_len = MIN(prefix_len, len - 1);
446 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
448 assert(! UTF8_IS_INVARIANT(*s));
452 /* Technically, a start byte of FE can be for a code point that fits into
453 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
459 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
460 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
461 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
463 if (*s != 0xFE || len == 1) {
467 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
469 return cBOOL(memGT(s + 1, prefix, cmp_len));
475 PERL_STATIC_INLINE bool
476 S_does_utf8_overflow(const U8 * const s, const U8 * e)
479 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
481 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
483 const STRLEN len = e - s;
487 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
488 * platform, that is if it represents a code point larger than the highest
489 * representable code point. (For ASCII platforms, we could use memcmp()
490 * because we don't have to convert each byte to I8, but it's very rare
491 * input indeed that would approach overflow, so the loop below will likely
492 * only get executed once.
494 * 'e' must not be beyond a full character. If it is less than a full
495 * character, the function returns FALSE if there is any input beyond 'e'
496 * that could result in a non-overflowing code point */
498 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
499 assert(s <= e && s + UTF8SKIP(s) >= e);
501 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
503 /* On 32 bit ASCII machines, many overlongs that start with FF don't
506 if (isFF_OVERLONG(s, len)) {
507 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
508 return memGE(s, max_32_bit_overlong,
509 MIN(len, sizeof(max_32_bit_overlong) - 1));
514 for (x = s; x < e; x++, y++) {
516 /* If this byte is larger than the corresponding highest UTF-8 byte, it
518 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
522 /* If not the same as this byte, it must be smaller, doesn't overflow */
523 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
528 /* Got to the end and all bytes are the same. If the input is a whole
529 * character, it doesn't overflow. And if it is a partial character,
530 * there's not enough information to tell, so assume doesn't overflow */
534 PERL_STATIC_INLINE bool
535 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
537 /* Overlongs can occur whenever the number of continuation bytes
538 * changes. That means whenever the number of leading 1 bits in a start
539 * byte increases from the next lower start byte. That happens for start
540 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
541 * illegal start bytes have already been excluded, so don't need to be
543 * ASCII platforms: C0, C1
544 * EBCDIC platforms C0, C1, C2, C3, C4, E0
546 * At least a second byte is required to determine if other sequences will
549 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
550 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
552 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
553 assert(len > 1 && UTF8_IS_START(*s));
555 /* Each platform has overlongs after the start bytes given above (expressed
556 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
557 * the logic is the same, except the E0 overlong has already been excluded
558 * on EBCDIC platforms. The values below were found by manually
559 * inspecting the UTF-8 patterns. See the tables in utf8.h and
563 # define F0_ABOVE_OVERLONG 0xB0
564 # define F8_ABOVE_OVERLONG 0xA8
565 # define FC_ABOVE_OVERLONG 0xA4
566 # define FE_ABOVE_OVERLONG 0xA2
567 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
571 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
575 # define F0_ABOVE_OVERLONG 0x90
576 # define F8_ABOVE_OVERLONG 0x88
577 # define FC_ABOVE_OVERLONG 0x84
578 # define FE_ABOVE_OVERLONG 0x82
579 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
583 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
584 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
585 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
586 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
591 /* Check for the FF overlong */
592 return isFF_OVERLONG(s, len);
595 PERL_STATIC_INLINE bool
596 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
598 PERL_ARGS_ASSERT_ISFF_OVERLONG;
600 /* Check for the FF overlong. This happens only if all these bytes match;
601 * what comes after them doesn't matter. See tables in utf8.h,
604 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
605 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
606 sizeof(FF_OVERLONG_PREFIX) - 1));
609 #undef F0_ABOVE_OVERLONG
610 #undef F8_ABOVE_OVERLONG
611 #undef FC_ABOVE_OVERLONG
612 #undef FE_ABOVE_OVERLONG
613 #undef FF_OVERLONG_PREFIX
616 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
621 /* A helper function that should not be called directly.
623 * This function returns non-zero if the string beginning at 's' and
624 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
625 * code point; otherwise it returns 0. The examination stops after the
626 * first code point in 's' is validated, not looking at the rest of the
627 * input. If 'e' is such that there are not enough bytes to represent a
628 * complete code point, this function will return non-zero anyway, if the
629 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
630 * excluded by 'flags'.
632 * A non-zero return gives the number of bytes required to represent the
633 * code point. Be aware that if the input is for a partial character, the
634 * return will be larger than 'e - s'.
636 * This function assumes that the code point represented is UTF-8 variant.
637 * The caller should have excluded this possibility before calling this
640 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
641 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
642 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
643 * disallowed by the flags. If the input is only for a partial character,
644 * the function will return non-zero if there is any sequence of
645 * well-formed UTF-8 that, when appended to the input sequence, could
646 * result in an allowed code point; otherwise it returns 0. Non characters
647 * cannot be determined based on partial character input. But many of the
648 * other excluded types can be determined with just the first one or two
653 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
655 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
656 |UTF8_DISALLOW_ABOVE_31_BIT)));
657 assert(! UTF8_IS_INVARIANT(*s));
659 /* A variant char must begin with a start byte */
660 if (UNLIKELY(! UTF8_IS_START(*s))) {
664 /* Examine a maximum of a single whole code point */
665 if (e - s > UTF8SKIP(s)) {
671 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
672 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
674 /* The code below is derived from this table. Keep in mind that legal
675 * continuation bytes range between \x80..\xBF for UTF-8, and
676 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
677 * Hence, we don't have to test the upper edge because if any of those
678 * are encountered, the sequence is malformed, and will fail elsewhere
680 * UTF-8 UTF-EBCDIC I8
681 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
682 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
683 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
687 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
688 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
689 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
691 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
693 && ((s1) & 0xFE ) == 0xB6)
695 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
696 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
697 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
700 if ( (flags & UTF8_DISALLOW_SUPER)
701 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
703 return 0; /* Above Unicode */
706 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
707 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
709 return 0; /* Above 31 bits */
713 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
715 if ( (flags & UTF8_DISALLOW_SUPER)
716 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
718 return 0; /* Above Unicode */
721 if ( (flags & UTF8_DISALLOW_SURROGATE)
722 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
724 return 0; /* Surrogate */
727 if ( (flags & UTF8_DISALLOW_NONCHAR)
728 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
730 return 0; /* Noncharacter code point */
735 /* Make sure that all that follows are continuation bytes */
736 for (x = s + 1; x < e; x++) {
737 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
742 /* Here is syntactically valid. Next, make sure this isn't the start of an
744 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
748 /* And finally, that the code point represented fits in a word on this
750 if (does_utf8_overflow(s, e)) {
758 Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
760 /* Returns a mortalized C string that is a displayable copy of the 'len'
761 * bytes starting at 's'. 'format' gives how to display each byte.
762 * Currently, there are only two formats, so it is currently a bool:
764 * 1 ab (that is a space between two hex digit bytes)
767 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
769 const U8 * const e = s + len;
773 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
775 Newx(output, output_len, char);
780 const unsigned high_nibble = (*s & 0xF0) >> 4;
781 const unsigned low_nibble = (*s & 0x0F);
791 if (high_nibble < 10) {
792 *d++ = high_nibble + '0';
795 *d++ = high_nibble - 10 + 'a';
798 if (low_nibble < 10) {
799 *d++ = low_nibble + '0';
802 *d++ = low_nibble - 10 + 'a';
810 PERL_STATIC_INLINE char *
811 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
813 /* How many bytes to print */
816 /* Which one is the non-continuation */
817 const STRLEN non_cont_byte_pos,
819 /* How many bytes should there be? */
820 const STRLEN expect_len)
822 /* Return the malformation warning text for an unexpected continuation
825 const char * const where = (non_cont_byte_pos == 1)
827 : Perl_form(aTHX_ "%d bytes",
828 (int) non_cont_byte_pos);
830 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
832 /* We don't need to pass this parameter, but since it has already been
833 * calculated, it's likely faster to pass it; verify under DEBUGGING */
834 assert(expect_len == UTF8SKIP(s));
836 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
837 " %s after start byte 0x%02x; need %d bytes, got %d)",
839 _byte_dump_string(s, print_len, 0),
840 *(s + non_cont_byte_pos),
844 (int) non_cont_byte_pos);
849 =for apidoc utf8n_to_uvchr
851 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
852 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
854 Bottom level UTF-8 decode routine.
855 Returns the native code point value of the first character in the string C<s>,
856 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
857 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
858 the length, in bytes, of that character.
860 The value of C<flags> determines the behavior when C<s> does not point to a
861 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
862 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
863 is the next possible position in C<s> that could begin a non-malformed
864 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
865 is raised. Some UTF-8 input sequences may contain multiple malformations.
866 This function tries to find every possible one in each call, so multiple
867 warnings can be raised for each sequence.
869 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
870 individual types of malformations, such as the sequence being overlong (that
871 is, when there is a shorter sequence that can express the same code point;
872 overlong sequences are expressly forbidden in the UTF-8 standard due to
873 potential security issues). Another malformation example is the first byte of
874 a character not being a legal first byte. See F<utf8.h> for the list of such
875 flags. Even if allowed, this function generally returns the Unicode
876 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
877 F<utf8.h> to override this behavior for the overlong malformations, but don't
878 do that except for very specialized purposes.
880 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
881 flags) malformation is found. If this flag is set, the routine assumes that
882 the caller will raise a warning, and this function will silently just set
883 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
885 Note that this API requires disambiguation between successful decoding a C<NUL>
886 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
887 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
888 be set to 1. To disambiguate, upon a zero return, see if the first byte of
889 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
890 error. Or you can use C<L</utf8n_to_uvchr_error>>.
892 Certain code points are considered problematic. These are Unicode surrogates,
893 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
894 By default these are considered regular code points, but certain situations
895 warrant special handling for them, which can be specified using the C<flags>
896 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
897 three classes are treated as malformations and handled as such. The flags
898 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
899 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
900 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
901 restricts the allowed inputs to the strict UTF-8 traditionally defined by
902 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
904 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
905 The difference between traditional strictness and C9 strictness is that the
906 latter does not forbid non-character code points. (They are still discouraged,
907 however.) For more discussion see L<perlunicode/Noncharacter code points>.
909 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
910 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
911 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
912 raised for their respective categories, but otherwise the code points are
913 considered valid (not malformations). To get a category to both be treated as
914 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
915 (But note that warnings are not raised if lexically disabled nor if
916 C<UTF8_CHECK_ONLY> is also specified.)
918 It is now deprecated to have very high code points (above C<IV_MAX> on the
919 platforms) and this function will raise a deprecation warning for these (unless
920 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
923 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
924 so using them is more problematic than other above-Unicode code points. Perl
925 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
926 likely that non-Perl languages will not be able to read files that contain
927 these; nor would Perl understand files
928 written by something that uses a different extension. For these reasons, there
929 is a separate set of flags that can warn and/or disallow these extremely high
930 code points, even if other above-Unicode ones are accepted. These are the
931 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
932 are entirely independent from the deprecation warning for code points above
933 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
934 code point that needs more than 31 bits to represent. When that happens,
935 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
936 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
937 above-Unicode code points, including these, as malformations; and
938 C<UTF8_WARN_SUPER> warns on these.)
940 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
941 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
942 than on ASCII. Prior to that, code points 2**31 and higher were simply
943 unrepresentable, and a different, incompatible method was used to represent
944 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
945 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
946 platforms, warning and disallowing 2**31 and higher.
948 All other code points corresponding to Unicode characters, including private
949 use and those yet to be assigned, are never considered malformed and never
954 Also implemented as a macro in utf8.h
958 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
963 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
965 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
970 =for apidoc utf8n_to_uvchr_error
972 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
973 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
975 This function is for code that needs to know what the precise malformation(s)
976 are when an error is found.
978 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
979 all the others, C<errors>. If this parameter is 0, this function behaves
980 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
981 to a C<U32> variable, which this function sets to indicate any errors found.
982 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
983 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
984 of these bits will be set if a malformation is found, even if the input
985 C<flags> parameter indicates that the given malformation is allowed; those
986 exceptions are noted:
990 =item C<UTF8_GOT_ABOVE_31_BIT>
992 The code point represented by the input UTF-8 sequence occupies more than 31
994 This bit is set only if the input C<flags> parameter contains either the
995 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
997 =item C<UTF8_GOT_CONTINUATION>
999 The input sequence was malformed in that the first byte was a a UTF-8
1002 =item C<UTF8_GOT_EMPTY>
1004 The input C<curlen> parameter was 0.
1006 =item C<UTF8_GOT_LONG>
1008 The input sequence was malformed in that there is some other sequence that
1009 evaluates to the same code point, but that sequence is shorter than this one.
1011 =item C<UTF8_GOT_NONCHAR>
1013 The code point represented by the input UTF-8 sequence is for a Unicode
1014 non-character code point.
1015 This bit is set only if the input C<flags> parameter contains either the
1016 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1018 =item C<UTF8_GOT_NON_CONTINUATION>
1020 The input sequence was malformed in that a non-continuation type byte was found
1021 in a position where only a continuation type one should be.
1023 =item C<UTF8_GOT_OVERFLOW>
1025 The input sequence was malformed in that it is for a code point that is not
1026 representable in the number of bits available in a UV on the current platform.
1028 =item C<UTF8_GOT_SHORT>
1030 The input sequence was malformed in that C<curlen> is smaller than required for
1031 a complete sequence. In other words, the input is for a partial character
1034 =item C<UTF8_GOT_SUPER>
1036 The input sequence was malformed in that it is for a non-Unicode code point;
1037 that is, one above the legal Unicode maximum.
1038 This bit is set only if the input C<flags> parameter contains either the
1039 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1041 =item C<UTF8_GOT_SURROGATE>
1043 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1045 This bit is set only if the input C<flags> parameter contains either the
1046 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1050 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1051 flag to suppress any warnings, and then examine the C<*errors> return.
1057 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1063 const U8 * const s0 = s;
1064 U8 * send = NULL; /* (initialized to silence compilers' wrong
1066 U32 possible_problems = 0; /* A bit is set here for each potential problem
1067 found as we go along */
1069 STRLEN expectlen = 0; /* How long should this sequence be?
1070 (initialized to silence compilers' wrong
1072 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1073 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1074 this gets set and discarded */
1076 /* The below are used only if there is both an overlong malformation and a
1077 * too short one. Otherwise the first two are set to 's0' and 'send', and
1078 * the third not used at all */
1079 U8 * adjusted_s0 = (U8 *) s0;
1080 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1082 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1083 routine; see [perl #130921] */
1084 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1086 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1092 errors = &discard_errors;
1095 /* The order of malformation tests here is important. We should consume as
1096 * few bytes as possible in order to not skip any valid character. This is
1097 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1098 * http://unicode.org/reports/tr36 for more discussion as to why. For
1099 * example, once we've done a UTF8SKIP, we can tell the expected number of
1100 * bytes, and could fail right off the bat if the input parameters indicate
1101 * that there are too few available. But it could be that just that first
1102 * byte is garbled, and the intended character occupies fewer bytes. If we
1103 * blindly assumed that the first byte is correct, and skipped based on
1104 * that number, we could skip over a valid input character. So instead, we
1105 * always examine the sequence byte-by-byte.
1107 * We also should not consume too few bytes, otherwise someone could inject
1108 * things. For example, an input could be deliberately designed to
1109 * overflow, and if this code bailed out immediately upon discovering that,
1110 * returning to the caller C<*retlen> pointing to the very next byte (one
1111 * which is actually part of of the overflowing sequence), that could look
1112 * legitimate to the caller, which could discard the initial partial
1113 * sequence and process the rest, inappropriately.
1115 * Some possible input sequences are malformed in more than one way. This
1116 * function goes to lengths to try to find all of them. This is necessary
1117 * for correctness, as the inputs may allow one malformation but not
1118 * another, and if we abandon searching for others after finding the
1119 * allowed one, we could allow in something that shouldn't have been.
1122 if (UNLIKELY(curlen == 0)) {
1123 possible_problems |= UTF8_GOT_EMPTY;
1125 uv = UNICODE_REPLACEMENT;
1126 goto ready_to_handle_errors;
1129 expectlen = UTF8SKIP(s);
1131 /* A well-formed UTF-8 character, as the vast majority of calls to this
1132 * function will be for, has this expected length. For efficiency, set
1133 * things up here to return it. It will be overriden only in those rare
1134 * cases where a malformation is found */
1136 *retlen = expectlen;
1139 /* An invariant is trivially well-formed */
1140 if (UTF8_IS_INVARIANT(uv)) {
1144 /* A continuation character can't start a valid sequence */
1145 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1146 possible_problems |= UTF8_GOT_CONTINUATION;
1148 uv = UNICODE_REPLACEMENT;
1149 goto ready_to_handle_errors;
1152 /* Here is not a continuation byte, nor an invariant. The only thing left
1153 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1154 * because it excludes start bytes like \xC0 that always lead to
1157 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1158 * that indicate the number of bytes in the character's whole UTF-8
1159 * sequence, leaving just the bits that are part of the value. */
1160 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1162 /* Setup the loop end point, making sure to not look past the end of the
1163 * input string, and flag it as too short if the size isn't big enough. */
1165 if (UNLIKELY(curlen < expectlen)) {
1166 possible_problems |= UTF8_GOT_SHORT;
1173 adjusted_send = send;
1175 /* Now, loop through the remaining bytes in the character's sequence,
1176 * accumulating each into the working value as we go. */
1177 for (s = s0 + 1; s < send; s++) {
1178 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1179 uv = UTF8_ACCUMULATE(uv, *s);
1183 /* Here, found a non-continuation before processing all expected bytes.
1184 * This byte indicates the beginning of a new character, so quit, even
1185 * if allowing this malformation. */
1186 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1188 } /* End of loop through the character's bytes */
1190 /* Save how many bytes were actually in the character */
1193 /* Note that there are two types of too-short malformation. One is when
1194 * there is actual wrong data before the normal termination of the
1195 * sequence. The other is that the sequence wasn't complete before the end
1196 * of the data we are allowed to look at, based on the input 'curlen'.
1197 * This means that we were passed data for a partial character, but it is
1198 * valid as far as we saw. The other is definitely invalid. This
1199 * distinction could be important to a caller, so the two types are kept
1202 * A convenience macro that matches either of the too-short conditions. */
1203 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1205 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1207 uv = UNICODE_REPLACEMENT;
1210 /* Check for overflow */
1211 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1212 possible_problems |= UTF8_GOT_OVERFLOW;
1213 uv = UNICODE_REPLACEMENT;
1216 /* Check for overlong. If no problems so far, 'uv' is the correct code
1217 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1218 * we must look at the UTF-8 byte sequence itself to see if it is for an
1220 if ( ( LIKELY(! possible_problems)
1221 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1222 || ( UNLIKELY( possible_problems)
1223 && ( UNLIKELY(! UTF8_IS_START(*s0))
1225 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1228 possible_problems |= UTF8_GOT_LONG;
1230 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1231 UV min_uv = uv_so_far;
1234 /* Here, the input is both overlong and is missing some trailing
1235 * bytes. There is no single code point it could be for, but there
1236 * may be enough information present to determine if what we have
1237 * so far is for an unallowed code point, such as for a surrogate.
1238 * The code below has the intelligence to determine this, but just
1239 * for non-overlong UTF-8 sequences. What we do here is calculate
1240 * the smallest code point the input could represent if there were
1241 * no too short malformation. Then we compute and save the UTF-8
1242 * for that, which is what the code below looks at instead of the
1243 * raw input. It turns out that the smallest such code point is
1245 for (i = curlen; i < expectlen; i++) {
1246 min_uv = UTF8_ACCUMULATE(min_uv,
1247 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1250 adjusted_s0 = temp_char_buf;
1251 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1255 /* Now check that the input isn't for a problematic code point not allowed
1256 * by the input parameters. */
1257 /* isn't problematic if < this */
1258 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1259 || ( UNLIKELY(possible_problems)
1261 /* if overflow, we know without looking further
1262 * precisely which of the problematic types it is,
1263 * and we deal with those in the overflow handling
1265 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1266 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1267 && ((flags & ( UTF8_DISALLOW_NONCHAR
1268 |UTF8_DISALLOW_SURROGATE
1269 |UTF8_DISALLOW_SUPER
1270 |UTF8_DISALLOW_ABOVE_31_BIT
1272 |UTF8_WARN_SURROGATE
1274 |UTF8_WARN_ABOVE_31_BIT))
1275 /* In case of a malformation, 'uv' is not valid, and has
1276 * been changed to something in the Unicode range.
1277 * Currently we don't output a deprecation message if there
1278 * is already a malformation, so we don't have to special
1279 * case the test immediately below */
1280 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1281 && ckWARN_d(WARN_DEPRECATED))))
1283 /* If there were no malformations, or the only malformation is an
1284 * overlong, 'uv' is valid */
1285 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1286 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1287 possible_problems |= UTF8_GOT_SURROGATE;
1289 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1290 possible_problems |= UTF8_GOT_SUPER;
1292 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1293 possible_problems |= UTF8_GOT_NONCHAR;
1296 else { /* Otherwise, need to look at the source UTF-8, possibly
1297 adjusted to be non-overlong */
1299 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1300 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1302 possible_problems |= UTF8_GOT_SUPER;
1304 else if (curlen > 1) {
1305 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1306 NATIVE_UTF8_TO_I8(*adjusted_s0),
1307 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1309 possible_problems |= UTF8_GOT_SUPER;
1311 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1312 NATIVE_UTF8_TO_I8(*adjusted_s0),
1313 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1315 possible_problems |= UTF8_GOT_SURROGATE;
1319 /* We need a complete well-formed UTF-8 character to discern
1320 * non-characters, so can't look for them here */
1324 ready_to_handle_errors:
1327 * curlen contains the number of bytes in the sequence that
1328 * this call should advance the input by.
1329 * avail_len gives the available number of bytes passed in, but
1330 * only if this is less than the expected number of
1331 * bytes, based on the code point's start byte.
1332 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1333 * is set in it for each potential problem found.
1334 * uv contains the code point the input sequence
1335 * represents; or if there is a problem that prevents
1336 * a well-defined value from being computed, it is
1337 * some subsitute value, typically the REPLACEMENT
1339 * s0 points to the first byte of the character
1340 * send points to just after where that (potentially
1341 * partial) character ends
1342 * adjusted_s0 normally is the same as s0, but in case of an
1343 * overlong for which the UTF-8 matters below, it is
1344 * the first byte of the shortest form representation
1346 * adjusted_send normally is the same as 'send', but if adjusted_s0
1347 * is set to something other than s0, this points one
1351 if (UNLIKELY(possible_problems)) {
1352 bool disallowed = FALSE;
1353 const U32 orig_problems = possible_problems;
1355 while (possible_problems) { /* Handle each possible problem */
1357 char * message = NULL;
1359 /* Each 'if' clause handles one problem. They are ordered so that
1360 * the first ones' messages will be displayed before the later
1361 * ones; this is kinda in decreasing severity order */
1362 if (possible_problems & UTF8_GOT_OVERFLOW) {
1364 /* Overflow means also got a super and above 31 bits, but we
1365 * handle all three cases here */
1367 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1368 *errors |= UTF8_GOT_OVERFLOW;
1370 /* But the API says we flag all errors found */
1371 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1372 *errors |= UTF8_GOT_SUPER;
1375 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1377 *errors |= UTF8_GOT_ABOVE_31_BIT;
1380 /* Disallow if any of the three categories say to */
1381 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1382 || (flags & ( UTF8_DISALLOW_SUPER
1383 |UTF8_DISALLOW_ABOVE_31_BIT)))
1389 /* Likewise, warn if any say to, plus if deprecation warnings
1390 * are on, because this code point is above IV_MAX */
1391 if ( ckWARN_d(WARN_DEPRECATED)
1392 || ! (flags & UTF8_ALLOW_OVERFLOW)
1393 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1396 /* The warnings code explicitly says it doesn't handle the
1397 * case of packWARN2 and two categories which have
1398 * parent-child relationship. Even if it works now to
1399 * raise the warning if either is enabled, it wouldn't
1400 * necessarily do so in the future. We output (only) the
1401 * most dire warning*/
1402 if (! (flags & UTF8_CHECK_ONLY)) {
1403 if (ckWARN_d(WARN_UTF8)) {
1404 pack_warn = packWARN(WARN_UTF8);
1406 else if (ckWARN_d(WARN_NON_UNICODE)) {
1407 pack_warn = packWARN(WARN_NON_UNICODE);
1410 message = Perl_form(aTHX_ "%s: %s (overflows)",
1412 _byte_dump_string(s0, send - s0, 0));
1417 else if (possible_problems & UTF8_GOT_EMPTY) {
1418 possible_problems &= ~UTF8_GOT_EMPTY;
1419 *errors |= UTF8_GOT_EMPTY;
1421 if (! (flags & UTF8_ALLOW_EMPTY)) {
1423 /* This so-called malformation is now treated as a bug in
1424 * the caller. If you have nothing to decode, skip calling
1429 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1430 pack_warn = packWARN(WARN_UTF8);
1431 message = Perl_form(aTHX_ "%s (empty string)",
1436 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1437 possible_problems &= ~UTF8_GOT_CONTINUATION;
1438 *errors |= UTF8_GOT_CONTINUATION;
1440 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1442 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1443 pack_warn = packWARN(WARN_UTF8);
1444 message = Perl_form(aTHX_
1445 "%s: %s (unexpected continuation byte 0x%02x,"
1446 " with no preceding start byte)",
1448 _byte_dump_string(s0, 1, 0), *s0);
1452 else if (possible_problems & UTF8_GOT_SHORT) {
1453 possible_problems &= ~UTF8_GOT_SHORT;
1454 *errors |= UTF8_GOT_SHORT;
1456 if (! (flags & UTF8_ALLOW_SHORT)) {
1458 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1459 pack_warn = packWARN(WARN_UTF8);
1460 message = Perl_form(aTHX_
1461 "%s: %s (too short; %d byte%s available, need %d)",
1463 _byte_dump_string(s0, send - s0, 0),
1465 avail_len == 1 ? "" : "s",
1471 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1472 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1473 *errors |= UTF8_GOT_NON_CONTINUATION;
1475 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1477 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1479 /* If we don't know for sure that the input length is
1480 * valid, avoid as much as possible reading past the
1481 * end of the buffer */
1482 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1485 pack_warn = packWARN(WARN_UTF8);
1486 message = Perl_form(aTHX_ "%s",
1487 unexpected_non_continuation_text(s0,
1494 else if (possible_problems & UTF8_GOT_LONG) {
1495 possible_problems &= ~UTF8_GOT_LONG;
1496 *errors |= UTF8_GOT_LONG;
1498 if (flags & UTF8_ALLOW_LONG) {
1500 /* We don't allow the actual overlong value, unless the
1501 * special extra bit is also set */
1502 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1503 & ~UTF8_ALLOW_LONG)))
1505 uv = UNICODE_REPLACEMENT;
1511 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1512 pack_warn = packWARN(WARN_UTF8);
1514 /* These error types cause 'uv' to be something that
1515 * isn't what was intended, so can't use it in the
1516 * message. The other error types either can't
1517 * generate an overlong, or else the 'uv' is valid */
1519 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1521 message = Perl_form(aTHX_
1522 "%s: %s (any UTF-8 sequence that starts"
1523 " with \"%s\" is overlong which can and"
1524 " should be represented with a"
1525 " different, shorter sequence)",
1527 _byte_dump_string(s0, send - s0, 0),
1528 _byte_dump_string(s0, curlen, 0));
1531 U8 tmpbuf[UTF8_MAXBYTES+1];
1532 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1534 message = Perl_form(aTHX_
1535 "%s: %s (overlong; instead use %s to represent"
1538 _byte_dump_string(s0, send - s0, 0),
1539 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1540 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1541 small code points */
1547 else if (possible_problems & UTF8_GOT_SURROGATE) {
1548 possible_problems &= ~UTF8_GOT_SURROGATE;
1550 if (flags & UTF8_WARN_SURROGATE) {
1551 *errors |= UTF8_GOT_SURROGATE;
1553 if ( ! (flags & UTF8_CHECK_ONLY)
1554 && ckWARN_d(WARN_SURROGATE))
1556 pack_warn = packWARN(WARN_SURROGATE);
1558 /* These are the only errors that can occur with a
1559 * surrogate when the 'uv' isn't valid */
1560 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1561 message = Perl_form(aTHX_
1562 "UTF-16 surrogate (any UTF-8 sequence that"
1563 " starts with \"%s\" is for a surrogate)",
1564 _byte_dump_string(s0, curlen, 0));
1567 message = Perl_form(aTHX_
1568 "UTF-16 surrogate U+%04" UVXf, uv);
1573 if (flags & UTF8_DISALLOW_SURROGATE) {
1575 *errors |= UTF8_GOT_SURROGATE;
1578 else if (possible_problems & UTF8_GOT_SUPER) {
1579 possible_problems &= ~UTF8_GOT_SUPER;
1581 if (flags & UTF8_WARN_SUPER) {
1582 *errors |= UTF8_GOT_SUPER;
1584 if ( ! (flags & UTF8_CHECK_ONLY)
1585 && ckWARN_d(WARN_NON_UNICODE))
1587 pack_warn = packWARN(WARN_NON_UNICODE);
1589 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1590 message = Perl_form(aTHX_
1591 "Any UTF-8 sequence that starts with"
1592 " \"%s\" is for a non-Unicode code point,"
1593 " may not be portable",
1594 _byte_dump_string(s0, curlen, 0));
1597 message = Perl_form(aTHX_
1598 "Code point 0x%04" UVXf " is not"
1599 " Unicode, may not be portable",
1605 /* The maximum code point ever specified by a standard was
1606 * 2**31 - 1. Anything larger than that is a Perl extension
1607 * that very well may not be understood by other applications
1608 * (including earlier perl versions on EBCDIC platforms). We
1609 * test for these after the regular SUPER ones, and before
1610 * possibly bailing out, so that the slightly more dire warning
1611 * will override the regular one. */
1612 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1614 |UTF8_DISALLOW_ABOVE_31_BIT))
1615 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1616 && UNLIKELY(is_utf8_cp_above_31_bits(
1619 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1620 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1622 if ( ! (flags & UTF8_CHECK_ONLY)
1623 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1624 && ckWARN_d(WARN_UTF8))
1626 pack_warn = packWARN(WARN_UTF8);
1628 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1629 message = Perl_form(aTHX_
1630 "Any UTF-8 sequence that starts with"
1631 " \"%s\" is for a non-Unicode code"
1632 " point, and is not portable",
1633 _byte_dump_string(s0, curlen, 0));
1636 message = Perl_form(aTHX_
1637 "Code point 0x%" UVXf " is not Unicode,"
1638 " and not portable",
1643 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1644 |UTF8_DISALLOW_ABOVE_31_BIT))
1646 *errors |= UTF8_GOT_ABOVE_31_BIT;
1648 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1654 if (flags & UTF8_DISALLOW_SUPER) {
1655 *errors |= UTF8_GOT_SUPER;
1659 /* The deprecated warning overrides any non-deprecated one. If
1660 * there are other problems, a deprecation message is not
1661 * really helpful, so don't bother to raise it in that case.
1662 * This also keeps the code from having to handle the case
1663 * where 'uv' is not valid. */
1664 if ( ! (orig_problems
1665 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1666 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1667 && ckWARN_d(WARN_DEPRECATED))
1669 message = Perl_form(aTHX_ cp_above_legal_max,
1670 uv, MAX_NON_DEPRECATED_CP);
1671 pack_warn = packWARN(WARN_DEPRECATED);
1674 else if (possible_problems & UTF8_GOT_NONCHAR) {
1675 possible_problems &= ~UTF8_GOT_NONCHAR;
1677 if (flags & UTF8_WARN_NONCHAR) {
1678 *errors |= UTF8_GOT_NONCHAR;
1680 if ( ! (flags & UTF8_CHECK_ONLY)
1681 && ckWARN_d(WARN_NONCHAR))
1683 /* The code above should have guaranteed that we don't
1684 * get here with errors other than overlong */
1685 assert (! (orig_problems
1686 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1688 pack_warn = packWARN(WARN_NONCHAR);
1689 message = Perl_form(aTHX_ "Unicode non-character"
1690 " U+%04" UVXf " is not recommended"
1691 " for open interchange", uv);
1695 if (flags & UTF8_DISALLOW_NONCHAR) {
1697 *errors |= UTF8_GOT_NONCHAR;
1699 } /* End of looking through the possible flags */
1701 /* Display the message (if any) for the problem being handled in
1702 * this iteration of the loop */
1705 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1708 Perl_warner(aTHX_ pack_warn, "%s", message);
1710 } /* End of 'while (possible_problems)' */
1712 /* Since there was a possible problem, the returned length may need to
1713 * be changed from the one stored at the beginning of this function.
1714 * Instead of trying to figure out if that's needed, just do it. */
1720 if (flags & UTF8_CHECK_ONLY && retlen) {
1721 *retlen = ((STRLEN) -1);
1727 return UNI_TO_NATIVE(uv);
1731 =for apidoc utf8_to_uvchr_buf
1733 Returns the native code point of the first character in the string C<s> which
1734 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1735 C<*retlen> will be set to the length, in bytes, of that character.
1737 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1738 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1739 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1740 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1741 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1742 the next possible position in C<s> that could begin a non-malformed character.
1743 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1746 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1747 unless those are turned off.
1751 Also implemented as a macro in utf8.h
1757 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1759 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1763 return utf8n_to_uvchr(s, send - s, retlen,
1764 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1767 /* This is marked as deprecated
1769 =for apidoc utf8_to_uvuni_buf
1771 Only in very rare circumstances should code need to be dealing in Unicode
1772 (as opposed to native) code points. In those few cases, use
1773 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1775 Returns the Unicode (not-native) code point of the first character in the
1777 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1778 C<retlen> will be set to the length, in bytes, of that character.
1780 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1781 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1782 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1783 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1784 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1785 next possible position in C<s> that could begin a non-malformed character.
1786 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1788 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1789 unless those are turned off.
1795 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1797 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1801 /* Call the low level routine, asking for checks */
1802 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1806 =for apidoc utf8_length
1808 Return the length of the UTF-8 char encoded string C<s> in characters.
1809 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1810 up past C<e>, croaks.
1816 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1820 PERL_ARGS_ASSERT_UTF8_LENGTH;
1822 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1823 * the bitops (especially ~) can create illegal UTF-8.
1824 * In other words: in Perl UTF-8 is not just for Unicode. */
1827 goto warn_and_return;
1837 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1838 "%s in %s", unees, OP_DESC(PL_op));
1840 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1847 =for apidoc bytes_cmp_utf8
1849 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1850 sequence of characters (stored as UTF-8)
1851 in C<u>, C<ulen>. Returns 0 if they are
1852 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1853 if the first string is greater than the second string.
1855 -1 or +1 is returned if the shorter string was identical to the start of the
1856 longer string. -2 or +2 is returned if
1857 there was a difference between characters
1864 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1866 const U8 *const bend = b + blen;
1867 const U8 *const uend = u + ulen;
1869 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1871 while (b < bend && u < uend) {
1873 if (!UTF8_IS_INVARIANT(c)) {
1874 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1877 if (UTF8_IS_CONTINUATION(c1)) {
1878 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1880 /* diag_listed_as: Malformed UTF-8 character%s */
1881 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1883 unexpected_non_continuation_text(u - 1, 2, 1, 2),
1884 PL_op ? " in " : "",
1885 PL_op ? OP_DESC(PL_op) : "");
1890 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1891 "%s in %s", unees, OP_DESC(PL_op));
1893 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1894 return -2; /* Really want to return undef :-) */
1901 return *b < c ? -2 : +2;
1906 if (b == bend && u == uend)
1909 return b < bend ? +1 : -1;
1913 =for apidoc utf8_to_bytes
1915 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1916 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1917 updates C<len> to contain the new length.
1918 Returns zero on failure, setting C<len> to -1.
1920 If you need a copy of the string, see L</bytes_from_utf8>.
1926 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1928 U8 * const save = s;
1929 U8 * const send = s + *len;
1932 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1933 PERL_UNUSED_CONTEXT;
1935 /* ensure valid UTF-8 and chars < 256 before updating string */
1937 if (! UTF8_IS_INVARIANT(*s)) {
1938 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1939 *len = ((STRLEN) -1);
1950 if (! UTF8_IS_INVARIANT(c)) {
1951 /* Then it is two-byte encoded */
1952 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1963 =for apidoc bytes_from_utf8
1965 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1966 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1967 the newly-created string, and updates C<len> to contain the new
1968 length. Returns the original string if no conversion occurs, C<len>
1969 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1970 0 if C<s> is converted or consisted entirely of characters that are invariant
1971 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1977 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1980 const U8 *start = s;
1984 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1985 PERL_UNUSED_CONTEXT;
1989 /* ensure valid UTF-8 and chars < 256 before converting string */
1990 for (send = s + *len; s < send;) {
1991 if (! UTF8_IS_INVARIANT(*s)) {
1992 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2003 Newx(d, (*len) - count + 1, U8);
2004 s = start; start = d;
2007 if (! UTF8_IS_INVARIANT(c)) {
2008 /* Then it is two-byte encoded */
2009 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2020 =for apidoc bytes_to_utf8
2022 Converts a string C<s> of length C<len> bytes from the native encoding into
2024 Returns a pointer to the newly-created string, and sets C<len> to
2025 reflect the new length in bytes.
2027 A C<NUL> character will be written after the end of the string.
2029 If you want to convert to UTF-8 from encodings other than
2030 the native (Latin1 or EBCDIC),
2031 see L</sv_recode_to_utf8>().
2036 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2037 likewise need duplication. */
2040 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
2042 const U8 * const send = s + (*len);
2046 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2047 PERL_UNUSED_CONTEXT;
2049 Newx(d, (*len) * 2 + 1, U8);
2053 append_utf8_from_native_byte(*s, &d);
2062 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2064 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2065 * We optimize for native, for obvious reasons. */
2068 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2073 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2076 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
2081 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2083 if (OFFUNI_IS_INVARIANT(uv)) {
2084 *d++ = LATIN1_TO_NATIVE((U8) uv);
2087 if (uv <= MAX_UTF8_TWO_BYTE) {
2088 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2089 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2092 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2093 #define LAST_HIGH_SURROGATE 0xDBFF
2094 #define FIRST_LOW_SURROGATE 0xDC00
2095 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2097 /* This assumes that most uses will be in the first Unicode plane, not
2098 * needing surrogates */
2099 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2100 && uv <= UNICODE_SURROGATE_LAST))
2102 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2103 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2106 UV low = (p[0] << 8) + p[1];
2107 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2108 || UNLIKELY(low > LAST_LOW_SURROGATE))
2110 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2113 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2114 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2118 d = uvoffuni_to_utf8_flags(d, uv, 0);
2121 *d++ = (U8)(( uv >> 12) | 0xe0);
2122 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2123 *d++ = (U8)(( uv & 0x3f) | 0x80);
2127 *d++ = (U8)(( uv >> 18) | 0xf0);
2128 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2129 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2130 *d++ = (U8)(( uv & 0x3f) | 0x80);
2135 *newlen = d - dstart;
2139 /* Note: this one is slightly destructive of the source. */
2142 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2145 U8* const send = s + bytelen;
2147 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2150 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2154 const U8 tmp = s[0];
2159 return utf16_to_utf8(p, d, bytelen, newlen);
2163 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2165 U8 tmpbuf[UTF8_MAXBYTES+1];
2166 uvchr_to_utf8(tmpbuf, c);
2167 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2170 /* Internal function so we can deprecate the external one, and call
2171 this one from other deprecated functions in this file */
2174 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2176 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2180 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2184 Perl__is_uni_perl_idcont(pTHX_ UV c)
2186 U8 tmpbuf[UTF8_MAXBYTES+1];
2187 uvchr_to_utf8(tmpbuf, c);
2188 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2192 Perl__is_uni_perl_idstart(pTHX_ UV c)
2194 U8 tmpbuf[UTF8_MAXBYTES+1];
2195 uvchr_to_utf8(tmpbuf, c);
2196 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2200 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2202 /* We have the latin1-range values compiled into the core, so just use
2203 * those, converting the result to UTF-8. The only difference between upper
2204 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2205 * either "SS" or "Ss". Which one to use is passed into the routine in
2206 * 'S_or_s' to avoid a test */
2208 UV converted = toUPPER_LATIN1_MOD(c);
2210 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2212 assert(S_or_s == 'S' || S_or_s == 's');
2214 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2215 characters in this range */
2216 *p = (U8) converted;
2221 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2222 * which it maps to one of them, so as to only have to have one check for
2223 * it in the main case */
2224 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2226 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2227 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2230 converted = GREEK_CAPITAL_LETTER_MU;
2232 #if UNICODE_MAJOR_VERSION > 2 \
2233 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2234 && UNICODE_DOT_DOT_VERSION >= 8)
2235 case LATIN_SMALL_LETTER_SHARP_S:
2242 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2243 NOT_REACHED; /* NOTREACHED */
2247 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2248 *p = UTF8_TWO_BYTE_LO(converted);
2254 /* Call the function to convert a UTF-8 encoded character to the specified case.
2255 * Note that there may be more than one character in the result.
2256 * INP is a pointer to the first byte of the input character
2257 * OUTP will be set to the first byte of the string of changed characters. It
2258 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2259 * LENP will be set to the length in bytes of the string of changed characters
2261 * The functions return the ordinal of the first character in the string of OUTP */
2262 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2263 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2264 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2266 /* This additionally has the input parameter 'specials', which if non-zero will
2267 * cause this to use the specials hash for folding (meaning get full case
2268 * folding); otherwise, when zero, this implies a simple case fold */
2269 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2272 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2274 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2275 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2276 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2277 * the changed version may be longer than the original character.
2279 * The ordinal of the first character of the changed version is returned
2280 * (but note, as explained above, that there may be more.) */
2282 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2285 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2288 uvchr_to_utf8(p, c);
2289 return CALL_UPPER_CASE(c, p, p, lenp);
2293 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2295 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2298 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2301 uvchr_to_utf8(p, c);
2302 return CALL_TITLE_CASE(c, p, p, lenp);
2306 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2308 /* We have the latin1-range values compiled into the core, so just use
2309 * those, converting the result to UTF-8. Since the result is always just
2310 * one character, we allow <p> to be NULL */
2312 U8 converted = toLOWER_LATIN1(c);
2314 PERL_UNUSED_ARG(dummy);
2317 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2322 /* Result is known to always be < 256, so can use the EIGHT_BIT
2324 *p = UTF8_EIGHT_BIT_HI(converted);
2325 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2333 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2335 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2338 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2341 uvchr_to_utf8(p, c);
2342 return CALL_LOWER_CASE(c, p, p, lenp);
2346 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2348 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2349 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2350 * FOLD_FLAGS_FULL iff full folding is to be used;
2352 * Not to be used for locale folds
2357 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2358 PERL_UNUSED_CONTEXT;
2360 assert (! (flags & FOLD_FLAGS_LOCALE));
2362 if (UNLIKELY(c == MICRO_SIGN)) {
2363 converted = GREEK_SMALL_LETTER_MU;
2365 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2366 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2367 || UNICODE_DOT_DOT_VERSION > 0)
2368 else if ( (flags & FOLD_FLAGS_FULL)
2369 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2371 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2372 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2373 * under those circumstances. */
2374 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2375 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2376 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2378 return LATIN_SMALL_LETTER_LONG_S;
2388 else { /* In this range the fold of all other characters is their lower
2390 converted = toLOWER_LATIN1(c);
2393 if (UVCHR_IS_INVARIANT(converted)) {
2394 *p = (U8) converted;
2398 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2399 *p = UTF8_TWO_BYTE_LO(converted);
2407 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2410 /* Not currently externally documented, and subject to change
2411 * <flags> bits meanings:
2412 * FOLD_FLAGS_FULL iff full folding is to be used;
2413 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2414 * locale are to be used.
2415 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2418 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2420 if (flags & FOLD_FLAGS_LOCALE) {
2421 /* Treat a UTF-8 locale as not being in locale at all */
2422 if (IN_UTF8_CTYPE_LOCALE) {
2423 flags &= ~FOLD_FLAGS_LOCALE;
2426 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2427 goto needs_full_generality;
2432 return _to_fold_latin1((U8) c, p, lenp,
2433 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2436 /* Here, above 255. If no special needs, just use the macro */
2437 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2438 uvchr_to_utf8(p, c);
2439 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2441 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2442 the special flags. */
2443 U8 utf8_c[UTF8_MAXBYTES + 1];
2445 needs_full_generality:
2446 uvchr_to_utf8(utf8_c, c);
2447 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
2451 PERL_STATIC_INLINE bool
2452 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2453 const char *const swashname, SV* const invlist)
2455 /* returns a boolean giving whether or not the UTF8-encoded character that
2456 * starts at <p> is in the swash indicated by <swashname>. <swash>
2457 * contains a pointer to where the swash indicated by <swashname>
2458 * is to be stored; which this routine will do, so that future calls will
2459 * look at <*swash> and only generate a swash if it is not null. <invlist>
2460 * is NULL or an inversion list that defines the swash. If not null, it
2461 * saves time during initialization of the swash.
2463 * Note that it is assumed that the buffer length of <p> is enough to
2464 * contain all the bytes that comprise the character. Thus, <*p> should
2465 * have been checked before this call for mal-formedness enough to assure
2468 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2470 /* The API should have included a length for the UTF-8 character in <p>,
2471 * but it doesn't. We therefore assume that p has been validated at least
2472 * as far as there being enough bytes available in it to accommodate the
2473 * character without reading beyond the end, and pass that number on to the
2474 * validating routine */
2475 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2476 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2477 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2479 NOT_REACHED; /* NOTREACHED */
2483 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2484 *swash = _core_swash_init("utf8",
2486 /* Only use the name if there is no inversion
2487 * list; otherwise will go out to disk */
2488 (invlist) ? "" : swashname,
2490 &PL_sv_undef, 1, 0, invlist, &flags);
2493 return swash_fetch(*swash, p, TRUE) != 0;
2496 PERL_STATIC_INLINE bool
2497 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2498 const char *const swashname, SV* const invlist)
2500 /* returns a boolean giving whether or not the UTF8-encoded character that
2501 * starts at <p>, and extending no further than <e - 1> is in the swash
2502 * indicated by <swashname>. <swash> contains a pointer to where the swash
2503 * indicated by <swashname> is to be stored; which this routine will do, so
2504 * that future calls will look at <*swash> and only generate a swash if it
2505 * is not null. <invlist> is NULL or an inversion list that defines the
2506 * swash. If not null, it saves time during initialization of the swash.
2509 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2511 if (! isUTF8_CHAR(p, e)) {
2512 _force_out_malformed_utf8_message(p, e, 0, 1);
2513 NOT_REACHED; /* NOTREACHED */
2517 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2518 *swash = _core_swash_init("utf8",
2520 /* Only use the name if there is no inversion
2521 * list; otherwise will go out to disk */
2522 (invlist) ? "" : swashname,
2524 &PL_sv_undef, 1, 0, invlist, &flags);
2527 return swash_fetch(*swash, p, TRUE) != 0;
2531 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2532 const char * const alternative,
2533 const bool use_locale,
2534 const char * const file,
2535 const unsigned line)
2539 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2541 if (ckWARN_d(WARN_DEPRECATED)) {
2543 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2544 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2545 if (! PL_seen_deprecated_macro) {
2546 PL_seen_deprecated_macro = newHV();
2548 if (! hv_store(PL_seen_deprecated_macro, key,
2549 strlen(key), &PL_sv_undef, 0))
2551 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2554 if (instr(file, "mathoms.c")) {
2555 Perl_warner(aTHX_ WARN_DEPRECATED,
2556 "In %s, line %d, starting in Perl v5.30, %s()"
2557 " will be removed. Avoid this message by"
2558 " converting to use %s().\n",
2559 file, line, name, alternative);
2562 Perl_warner(aTHX_ WARN_DEPRECATED,
2563 "In %s, line %d, starting in Perl v5.30, %s() will"
2564 " require an additional parameter. Avoid this"
2565 " message by converting to use %s().\n",
2566 file, line, name, alternative);
2573 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2575 const char * const name,
2576 const char * const alternative,
2577 const bool use_utf8,
2578 const bool use_locale,
2579 const char * const file,
2580 const unsigned line)
2582 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2584 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2586 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2596 case _CC_ALPHANUMERIC:
2600 return is_utf8_common(p,
2601 &PL_utf8_swash_ptrs[classnum],
2602 swash_property_names[classnum],
2603 PL_XPosix_ptrs[classnum]);
2606 return is_XPERLSPACE_high(p);
2608 return is_HORIZWS_high(p);
2610 return is_XDIGIT_high(p);
2616 return is_VERTWS_high(p);
2618 if (! PL_utf8_perl_idstart) {
2619 PL_utf8_perl_idstart
2620 = _new_invlist_C_array(_Perl_IDStart_invlist);
2622 return is_utf8_common(p, &PL_utf8_perl_idstart,
2623 "_Perl_IDStart", NULL);
2625 if (! PL_utf8_perl_idcont) {
2627 = _new_invlist_C_array(_Perl_IDCont_invlist);
2629 return is_utf8_common(p, &PL_utf8_perl_idcont,
2630 "_Perl_IDCont", NULL);
2634 /* idcont is the same as wordchar below 256 */
2635 if (classnum == _CC_IDCONT) {
2636 classnum = _CC_WORDCHAR;
2638 else if (classnum == _CC_IDFIRST) {
2642 classnum = _CC_ALPHA;
2646 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2647 return _generic_isCC(*p, classnum);
2650 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2653 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2654 return isFOO_lc(classnum, *p);
2657 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2660 NOT_REACHED; /* NOTREACHED */
2664 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2667 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2669 assert(classnum < _FIRST_NON_SWASH_CC);
2671 return is_utf8_common_with_len(p,
2673 &PL_utf8_swash_ptrs[classnum],
2674 swash_property_names[classnum],
2675 PL_XPosix_ptrs[classnum]);
2679 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2683 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2685 if (! PL_utf8_perl_idstart) {
2686 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2688 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2689 "_Perl_IDStart", invlist);
2693 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2695 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2699 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2703 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2707 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2709 if (! PL_utf8_perl_idcont) {
2710 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2712 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2713 "_Perl_IDCont", invlist);
2717 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2719 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2721 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2725 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2727 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2729 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2733 Perl__is_utf8_mark(pTHX_ const U8 *p)
2735 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2737 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2741 =for apidoc to_utf8_case
2743 Instead use the appropriate one of L</toUPPER_utf8_safe>,
2744 L</toTITLE_utf8_safe>,
2745 L</toLOWER_utf8_safe>,
2746 or L</toFOLD_utf8_safe>.
2748 This function will be removed in Perl v5.28.
2750 C<p> contains the pointer to the UTF-8 string encoding
2751 the character that is being converted. This routine assumes that the character
2752 at C<p> is well-formed.
2754 C<ustrp> is a pointer to the character buffer to put the
2755 conversion result to. C<lenp> is a pointer to the length
2758 C<swashp> is a pointer to the swash to use.
2760 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2761 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
2762 but not always, a multicharacter mapping), is tried first.
2764 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2765 any special mappings; C<""> means to use the special mappings. Values other
2766 than these two are treated as the name of the hash containing the special
2767 mappings, like C<"utf8::ToSpecLower">.
2769 C<normal> is a string like C<"ToLower"> which means the swash
2772 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2773 unless those are turned off.
2778 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2779 SV **swashp, const char *normal, const char *special)
2783 const U8 * e = p + UTF8SKIP(p);
2785 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2787 cp = utf8n_to_uvchr(p, e - p, &len_cp, UTF8_CHECK_ONLY);
2788 if (len_cp == (STRLEN) -1) {
2789 _force_out_malformed_utf8_message(p, e,
2790 _UTF8_NO_CONFIDENCE_IN_CURLEN, 1 /* Die */ );
2793 return _to_utf8_case(cp, p, ustrp, lenp, swashp, normal, special);
2796 /* change namve uv1 to 'from' */
2798 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2799 SV **swashp, const char *normal, const char *special)
2803 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2805 /* For code points that don't change case, we already know that the output
2806 * of this function is the unchanged input, so we can skip doing look-ups
2807 * for them. Unfortunately the case-changing code points are scattered
2808 * around. But there are some long consecutive ranges where there are no
2809 * case changing code points. By adding tests, we can eliminate the lookup
2810 * for all the ones in such ranges. This is currently done here only for
2811 * just a few cases where the scripts are in common use in modern commerce
2812 * (and scripts adjacent to those which can be included without additional
2815 if (uv1 >= 0x0590) {
2816 /* This keeps from needing further processing the code points most
2817 * likely to be used in the following non-cased scripts: Hebrew,
2818 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2819 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2820 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2825 /* The following largish code point ranges also don't have case
2826 * changes, but khw didn't think they warranted extra tests to speed
2827 * them up (which would slightly slow down everything else above them):
2828 * 1100..139F Hangul Jamo, Ethiopic
2829 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2830 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2831 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2832 * Combining Diacritical Marks Extended, Balinese,
2833 * Sundanese, Batak, Lepcha, Ol Chiki
2834 * 2000..206F General Punctuation
2837 if (uv1 >= 0x2D30) {
2839 /* This keeps the from needing further processing the code points
2840 * most likely to be used in the following non-cased major scripts:
2841 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2843 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2844 * event that Unicode eventually allocates the unused block as of
2845 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2846 * that the test suite will start having failures to alert you
2847 * should that happen) */
2852 if (uv1 >= 0xAC00) {
2853 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2854 if (ckWARN_d(WARN_SURROGATE)) {
2855 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2856 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2857 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2862 /* AC00..FAFF Catches Hangul syllables and private use, plus
2869 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2870 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2871 && ckWARN_d(WARN_DEPRECATED))
2873 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2874 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2876 if (ckWARN_d(WARN_NON_UNICODE)) {
2877 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2878 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2879 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2883 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2885 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2888 /* As of this writing, this means we avoid swash creation
2889 * for anything beyond low Plane 1 */
2896 /* Note that non-characters are perfectly legal, so no warning should
2897 * be given. There are so few of them, that it isn't worth the extra
2898 * tests to avoid swash creation */
2901 if (!*swashp) /* load on-demand */
2902 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2905 /* It might be "special" (sometimes, but not always,
2906 * a multicharacter mapping) */
2910 /* If passed in the specials name, use that; otherwise use any
2911 * given in the swash */
2912 if (*special != '\0') {
2913 hv = get_hv(special, 0);
2916 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2918 hv = MUTABLE_HV(SvRV(*svp));
2923 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2928 s = SvPV_const(*svp, len);
2931 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2933 Copy(s, ustrp, len, U8);
2938 if (!len && *swashp) {
2939 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2942 /* It was "normal" (a single character mapping). */
2943 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2951 return valid_utf8_to_uvchr(ustrp, 0);
2954 /* Here, there was no mapping defined, which means that the code point maps
2955 * to itself. Return the inputs */
2958 if (p != ustrp) { /* Don't copy onto itself */
2959 Copy(p, ustrp, len, U8);
2970 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2972 /* This is called when changing the case of a UTF-8-encoded character above
2973 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2974 * result contains a character that crosses the 255/256 boundary, disallow
2975 * the change, and return the original code point. See L<perlfunc/lc> for
2978 * p points to the original string whose case was changed; assumed
2979 * by this routine to be well-formed
2980 * result the code point of the first character in the changed-case string
2981 * ustrp points to the changed-case string (<result> represents its first char)
2982 * lenp points to the length of <ustrp> */
2984 UV original; /* To store the first code point of <p> */
2986 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2988 assert(UTF8_IS_ABOVE_LATIN1(*p));
2990 /* We know immediately if the first character in the string crosses the
2991 * boundary, so can skip */
2994 /* Look at every character in the result; if any cross the
2995 * boundary, the whole thing is disallowed */
2996 U8* s = ustrp + UTF8SKIP(ustrp);
2997 U8* e = ustrp + *lenp;
2999 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3005 /* Here, no characters crossed, result is ok as-is, but we warn. */
3006 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3012 /* Failed, have to return the original */
3013 original = valid_utf8_to_uvchr(p, lenp);
3015 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3016 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3017 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
3018 "resolved to \"\\x{%" UVXf "}\".",
3022 Copy(p, ustrp, *lenp, char);
3027 S_check_and_deprecate(pTHX_ const U8 *p,
3029 const unsigned int type, /* See below */
3030 const bool use_locale, /* Is this a 'LC_'
3032 const char * const file,
3033 const unsigned line)
3035 /* This is a temporary function to deprecate the unsafe calls to the case
3036 * changing macros and functions. It keeps all the special stuff in just
3039 * It updates *e with the pointer to the end of the input string. If using
3040 * the old-style macros, *e is NULL on input, and so this function assumes
3041 * the input string is long enough to hold the entire UTF-8 sequence, and
3042 * sets *e accordingly, but it then returns a flag to pass the
3043 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3044 * using the full length if possible.
3046 * It also does the assert that *e > p when *e is not NULL. This should be
3047 * migrated to the callers when this function gets deleted.
3049 * The 'type' parameter is used for the caller to specify which case
3050 * changing function this is called from: */
3052 # define DEPRECATE_TO_UPPER 0
3053 # define DEPRECATE_TO_TITLE 1
3054 # define DEPRECATE_TO_LOWER 2
3055 # define DEPRECATE_TO_FOLD 3
3057 U32 utf8n_flags = 0;
3059 const char * alternative;
3061 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3064 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3065 *e = p + UTF8SKIP(p);
3067 /* For mathoms.c calls, we use the function name we know is stored
3068 * there. It could be part of a larger path */
3069 if (type == DEPRECATE_TO_UPPER) {
3070 name = instr(file, "mathoms.c")
3073 alternative = "toUPPER_utf8_safe";
3075 else if (type == DEPRECATE_TO_TITLE) {
3076 name = instr(file, "mathoms.c")
3079 alternative = "toTITLE_utf8_safe";
3081 else if (type == DEPRECATE_TO_LOWER) {
3082 name = instr(file, "mathoms.c")
3085 alternative = "toLOWER_utf8_safe";
3087 else if (type == DEPRECATE_TO_FOLD) {
3088 name = instr(file, "mathoms.c")
3091 alternative = "toFOLD_utf8_safe";
3093 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3095 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3104 /* The process for changing the case is essentially the same for the four case
3105 * change types, except there are complications for folding. Otherwise the
3106 * difference is only which case to change to. To make sure that they all do
3107 * the same thing, the bodies of the functions are extracted out into the
3108 * following two macros. The functions are written with the same variable
3109 * names, and these are known and used inside these macros. It would be
3110 * better, of course, to have inline functions to do it, but since different
3111 * macros are called, depending on which case is being changed to, this is not
3112 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3113 * function can start with the common start macro, then finish with its special
3114 * handling; while the other three cases can just use the common end macro.
3116 * The algorithm is to use the proper (passed in) macro or function to change
3117 * the case for code points that are below 256. The macro is used if using
3118 * locale rules for the case change; the function if not. If the code point is
3119 * above 255, it is computed from the input UTF-8, and another macro is called
3120 * to do the conversion. If necessary, the output is converted to UTF-8. If
3121 * using a locale, we have to check that the change did not cross the 255/256
3122 * boundary, see check_locale_boundary_crossing() for further details.
3124 * The macros are split with the correct case change for the below-256 case
3125 * stored into 'result', and in the middle of an else clause for the above-255
3126 * case. At that point in the 'else', 'result' is not the final result, but is
3127 * the input code point calculated from the UTF-8. The fold code needs to
3128 * realize all this and take it from there.
3130 * If you read the two macros as sequential, it's easier to understand what's
3132 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3133 L1_func_extra_param) \
3135 if (flags & (locale_flags)) { \
3136 /* Treat a UTF-8 locale as not being in locale at all */ \
3137 if (IN_UTF8_CTYPE_LOCALE) { \
3138 flags &= ~(locale_flags); \
3141 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3145 if (UTF8_IS_INVARIANT(*p)) { \
3146 if (flags & (locale_flags)) { \
3147 result = LC_L1_change_macro(*p); \
3150 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3153 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3154 if (flags & (locale_flags)) { \
3155 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3159 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3160 ustrp, lenp, L1_func_extra_param); \
3163 else { /* malformed UTF-8 or ord above 255 */ \
3164 STRLEN len_result; \
3165 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3166 if (len_result == (STRLEN) -1) { \
3167 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3171 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3172 result = change_macro(result, p, ustrp, lenp); \
3174 if (flags & (locale_flags)) { \
3175 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3180 /* Here, used locale rules. Convert back to UTF-8 */ \
3181 if (UTF8_IS_INVARIANT(result)) { \
3182 *ustrp = (U8) result; \
3186 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3187 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3194 =for apidoc to_utf8_upper
3196 Instead use L</toUPPER_utf8_safe>.
3200 /* Not currently externally documented, and subject to change:
3201 * <flags> is set iff iff the rules from the current underlying locale are to
3205 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3210 const char * const file,
3214 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3215 cBOOL(flags), file, line);
3217 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3219 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3220 /* 2nd char of uc(U+DF) is 'S' */
3221 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3222 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3226 =for apidoc to_utf8_title
3228 Instead use L</toTITLE_utf8_safe>.
3232 /* Not currently externally documented, and subject to change:
3233 * <flags> is set iff the rules from the current underlying locale are to be
3234 * used. Since titlecase is not defined in POSIX, for other than a
3235 * UTF-8 locale, uppercase is used instead for code points < 256.
3239 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3244 const char * const file,
3248 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3249 cBOOL(flags), file, line);
3251 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3253 /* 2nd char of ucfirst(U+DF) is 's' */
3254 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3255 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3259 =for apidoc to_utf8_lower
3261 Instead use L</toLOWER_utf8_safe>.
3265 /* Not currently externally documented, and subject to change:
3266 * <flags> is set iff iff the rules from the current underlying locale are to
3271 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3276 const char * const file,
3280 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3281 cBOOL(flags), file, line);
3283 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3285 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3286 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3290 =for apidoc to_utf8_fold
3292 Instead use L</toFOLD_utf8_safe>.
3296 /* Not currently externally documented, and subject to change,
3298 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3299 * locale are to be used.
3300 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3301 * otherwise simple folds
3302 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3307 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3312 const char * const file,
3316 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3317 cBOOL(flags), file, line);
3319 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3321 /* These are mutually exclusive */
3322 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3324 assert(p != ustrp); /* Otherwise overwrites */
3326 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3327 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3329 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3331 if (flags & FOLD_FLAGS_LOCALE) {
3333 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3334 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3336 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3337 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3339 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3341 /* Special case these two characters, as what normally gets
3342 * returned under locale doesn't work */
3343 if (UTF8SKIP(p) == cap_sharp_s_len
3344 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3346 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3347 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3348 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3349 "resolved to \"\\x{17F}\\x{17F}\".");
3354 if (UTF8SKIP(p) == long_s_t_len
3355 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3357 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3358 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3359 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3360 "resolved to \"\\x{FB06}\".");
3361 goto return_ligature_st;
3364 #if UNICODE_MAJOR_VERSION == 3 \
3365 && UNICODE_DOT_VERSION == 0 \
3366 && UNICODE_DOT_DOT_VERSION == 1
3367 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3369 /* And special case this on this Unicode version only, for the same
3370 * reaons the other two are special cased. They would cross the
3371 * 255/256 boundary which is forbidden under /l, and so the code
3372 * wouldn't catch that they are equivalent (which they are only in
3374 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3375 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3377 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3378 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3379 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3380 "resolved to \"\\x{0131}\".");
3381 goto return_dotless_i;
3385 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3387 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3391 /* This is called when changing the case of a UTF-8-encoded
3392 * character above the ASCII range, and the result should not
3393 * contain an ASCII character. */
3395 UV original; /* To store the first code point of <p> */
3397 /* Look at every character in the result; if any cross the
3398 * boundary, the whole thing is disallowed */
3400 U8* e = ustrp + *lenp;
3403 /* Crossed, have to return the original */
3404 original = valid_utf8_to_uvchr(p, lenp);
3406 /* But in these instances, there is an alternative we can
3407 * return that is valid */
3408 if (original == LATIN_SMALL_LETTER_SHARP_S
3409 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3410 || original == LATIN_CAPITAL_LETTER_SHARP_S
3415 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3416 goto return_ligature_st;
3418 #if UNICODE_MAJOR_VERSION == 3 \
3419 && UNICODE_DOT_VERSION == 0 \
3420 && UNICODE_DOT_DOT_VERSION == 1
3422 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3423 goto return_dotless_i;
3426 Copy(p, ustrp, *lenp, char);
3432 /* Here, no characters crossed, result is ok as-is */
3437 /* Here, used locale rules. Convert back to UTF-8 */
3438 if (UTF8_IS_INVARIANT(result)) {
3439 *ustrp = (U8) result;
3443 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3444 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3451 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3452 * folds to a string of two of these characters. By returning this
3453 * instead, then, e.g.,
3454 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3457 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3458 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3460 return LATIN_SMALL_LETTER_LONG_S;
3463 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3464 * have the other one fold to it */
3466 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3467 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3468 return LATIN_SMALL_LIGATURE_ST;
3470 #if UNICODE_MAJOR_VERSION == 3 \
3471 && UNICODE_DOT_VERSION == 0 \
3472 && UNICODE_DOT_DOT_VERSION == 1
3475 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3476 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3477 return LATIN_SMALL_LETTER_DOTLESS_I;
3484 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3485 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3486 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3490 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3492 PERL_ARGS_ASSERT_SWASH_INIT;
3494 /* Returns a copy of a swash initiated by the called function. This is the
3495 * public interface, and returning a copy prevents others from doing
3496 * mischief on the original */
3498 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3502 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3505 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3506 * use the following define */
3508 #define CORE_SWASH_INIT_RETURN(x) \
3509 PL_curpm= old_PL_curpm; \
3512 /* Initialize and return a swash, creating it if necessary. It does this
3513 * by calling utf8_heavy.pl in the general case. The returned value may be
3514 * the swash's inversion list instead if the input parameters allow it.
3515 * Which is returned should be immaterial to callers, as the only
3516 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3517 * and swash_to_invlist() handle both these transparently.
3519 * This interface should only be used by functions that won't destroy or
3520 * adversely change the swash, as doing so affects all other uses of the
3521 * swash in the program; the general public should use 'Perl_swash_init'
3524 * pkg is the name of the package that <name> should be in.
3525 * name is the name of the swash to find. Typically it is a Unicode
3526 * property name, including user-defined ones
3527 * listsv is a string to initialize the swash with. It must be of the form
3528 * documented as the subroutine return value in
3529 * L<perlunicode/User-Defined Character Properties>
3530 * minbits is the number of bits required to represent each data element.
3531 * It is '1' for binary properties.
3532 * none I (khw) do not understand this one, but it is used only in tr///.
3533 * invlist is an inversion list to initialize the swash with (or NULL)
3534 * flags_p if non-NULL is the address of various input and output flag bits
3535 * to the routine, as follows: ('I' means is input to the routine;
3536 * 'O' means output from the routine. Only flags marked O are
3537 * meaningful on return.)
3538 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3539 * came from a user-defined property. (I O)
3540 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3541 * when the swash cannot be located, to simply return NULL. (I)
3542 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3543 * return of an inversion list instead of a swash hash if this routine
3544 * thinks that would result in faster execution of swash_fetch() later
3547 * Thus there are three possible inputs to find the swash: <name>,
3548 * <listsv>, and <invlist>. At least one must be specified. The result
3549 * will be the union of the specified ones, although <listsv>'s various
3550 * actions can intersect, etc. what <name> gives. To avoid going out to
3551 * disk at all, <invlist> should specify completely what the swash should
3552 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3554 * <invlist> is only valid for binary properties */
3556 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3558 SV* retval = &PL_sv_undef;
3559 HV* swash_hv = NULL;
3560 const int invlist_swash_boundary =
3561 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3562 ? 512 /* Based on some benchmarking, but not extensive, see commit
3564 : -1; /* Never return just an inversion list */
3566 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3567 assert(! invlist || minbits == 1);
3569 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3570 that triggered the swash init and the swash init perl logic itself.
3573 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3575 if (listsv != &PL_sv_undef || strNE(name, "")) {
3577 const size_t pkg_len = strlen(pkg);
3578 const size_t name_len = strlen(name);
3579 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3583 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3585 PUSHSTACKi(PERLSI_MAGIC);
3589 /* We might get here via a subroutine signature which uses a utf8
3590 * parameter name, at which point PL_subname will have been set
3591 * but not yet used. */
3592 save_item(PL_subname);
3593 if (PL_parser && PL_parser->error_count)
3594 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3595 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3596 if (!method) { /* demand load UTF-8 */
3598 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3599 GvSV(PL_errgv) = NULL;
3600 #ifndef NO_TAINT_SUPPORT
3601 /* It is assumed that callers of this routine are not passing in
3602 * any user derived data. */
3603 /* Need to do this after save_re_context() as it will set
3604 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3605 * in Perl_magic_get). Even line to create errsv_save can turn on
3607 SAVEBOOL(TAINT_get);
3610 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3613 /* Not ERRSV, as there is no need to vivify a scalar we are
3614 about to discard. */
3615 SV * const errsv = GvSV(PL_errgv);
3616 if (!SvTRUE(errsv)) {
3617 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3618 SvREFCNT_dec(errsv);
3626 mPUSHp(pkg, pkg_len);
3627 mPUSHp(name, name_len);
3632 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3633 GvSV(PL_errgv) = NULL;
3634 /* If we already have a pointer to the method, no need to use
3635 * call_method() to repeat the lookup. */
3637 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3638 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3640 retval = *PL_stack_sp--;
3641 SvREFCNT_inc(retval);
3644 /* Not ERRSV. See above. */
3645 SV * const errsv = GvSV(PL_errgv);
3646 if (!SvTRUE(errsv)) {
3647 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3648 SvREFCNT_dec(errsv);
3653 if (IN_PERL_COMPILETIME) {
3654 CopHINTS_set(PL_curcop, PL_hints);
3656 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3657 if (SvPOK(retval)) {
3659 /* If caller wants to handle missing properties, let them */
3660 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3661 CORE_SWASH_INIT_RETURN(NULL);
3664 "Can't find Unicode property definition \"%" SVf "\"",
3666 NOT_REACHED; /* NOTREACHED */
3669 } /* End of calling the module to find the swash */
3671 /* If this operation fetched a swash, and we will need it later, get it */
3672 if (retval != &PL_sv_undef
3673 && (minbits == 1 || (flags_p
3675 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3677 swash_hv = MUTABLE_HV(SvRV(retval));
3679 /* If we don't already know that there is a user-defined component to
3680 * this swash, and the user has indicated they wish to know if there is
3681 * one (by passing <flags_p>), find out */
3682 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3683 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3684 if (user_defined && SvUV(*user_defined)) {
3685 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3690 /* Make sure there is an inversion list for binary properties */
3692 SV** swash_invlistsvp = NULL;
3693 SV* swash_invlist = NULL;
3694 bool invlist_in_swash_is_valid = FALSE;
3695 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3696 an unclaimed reference count */
3698 /* If this operation fetched a swash, get its already existing
3699 * inversion list, or create one for it */
3702 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3703 if (swash_invlistsvp) {
3704 swash_invlist = *swash_invlistsvp;
3705 invlist_in_swash_is_valid = TRUE;
3708 swash_invlist = _swash_to_invlist(retval);
3709 swash_invlist_unclaimed = TRUE;
3713 /* If an inversion list was passed in, have to include it */
3716 /* Any fetched swash will by now have an inversion list in it;
3717 * otherwise <swash_invlist> will be NULL, indicating that we
3718 * didn't fetch a swash */
3719 if (swash_invlist) {
3721 /* Add the passed-in inversion list, which invalidates the one
3722 * already stored in the swash */
3723 invlist_in_swash_is_valid = FALSE;
3724 SvREADONLY_off(swash_invlist); /* Turned on again below */
3725 _invlist_union(invlist, swash_invlist, &swash_invlist);
3729 /* Here, there is no swash already. Set up a minimal one, if
3730 * we are going to return a swash */
3731 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3733 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3735 swash_invlist = invlist;
3739 /* Here, we have computed the union of all the passed-in data. It may
3740 * be that there was an inversion list in the swash which didn't get
3741 * touched; otherwise save the computed one */
3742 if (! invlist_in_swash_is_valid
3743 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3745 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3747 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3749 /* We just stole a reference count. */
3750 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3751 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3754 /* The result is immutable. Forbid attempts to change it. */
3755 SvREADONLY_on(swash_invlist);
3757 /* Use the inversion list stand-alone if small enough */
3758 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3759 SvREFCNT_dec(retval);
3760 if (!swash_invlist_unclaimed)
3761 SvREFCNT_inc_simple_void_NN(swash_invlist);
3762 retval = newRV_noinc(swash_invlist);
3766 CORE_SWASH_INIT_RETURN(retval);
3767 #undef CORE_SWASH_INIT_RETURN
3771 /* This API is wrong for special case conversions since we may need to
3772 * return several Unicode characters for a single Unicode character
3773 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3774 * the lower-level routine, and it is similarly broken for returning
3775 * multiple values. --jhi
3776 * For those, you should use S__to_utf8_case() instead */
3777 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3780 * Returns the value of property/mapping C<swash> for the first character
3781 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3782 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3783 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3785 * A "swash" is a hash which contains initially the keys/values set up by
3786 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3787 * property for all possible code points. Things are stored in a compact form
3788 * (see utf8_heavy.pl) so that calculation is required to find the actual
3789 * property value for a given code point. As code points are looked up, new
3790 * key/value pairs are added to the hash, so that the calculation doesn't have
3791 * to ever be re-done. Further, each calculation is done, not just for the
3792 * desired one, but for a whole block of code points adjacent to that one.
3793 * For binary properties on ASCII machines, the block is usually for 64 code
3794 * points, starting with a code point evenly divisible by 64. Thus if the
3795 * property value for code point 257 is requested, the code goes out and
3796 * calculates the property values for all 64 code points between 256 and 319,
3797 * and stores these as a single 64-bit long bit vector, called a "swatch",
3798 * under the key for code point 256. The key is the UTF-8 encoding for code
3799 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3800 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3801 * for code point 258 is then requested, this code realizes that it would be
3802 * stored under the key for 256, and would find that value and extract the
3803 * relevant bit, offset from 256.
3805 * Non-binary properties are stored in as many bits as necessary to represent
3806 * their values (32 currently, though the code is more general than that), not
3807 * as single bits, but the principle is the same: the value for each key is a
3808 * vector that encompasses the property values for all code points whose UTF-8
3809 * representations are represented by the key. That is, for all code points
3810 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3814 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3816 HV *const hv = MUTABLE_HV(SvRV(swash));
3821 const U8 *tmps = NULL;
3825 PERL_ARGS_ASSERT_SWASH_FETCH;
3827 /* If it really isn't a hash, it isn't really swash; must be an inversion
3829 if (SvTYPE(hv) != SVt_PVHV) {
3830 return _invlist_contains_cp((SV*)hv,
3832 ? valid_utf8_to_uvchr(ptr, NULL)
3836 /* We store the values in a "swatch" which is a vec() value in a swash
3837 * hash. Code points 0-255 are a single vec() stored with key length
3838 * (klen) 0. All other code points have a UTF-8 representation
3839 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3840 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3841 * length for them is the length of the encoded char - 1. ptr[klen] is the
3842 * final byte in the sequence representing the character */
3843 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3848 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3851 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3854 klen = UTF8SKIP(ptr) - 1;
3856 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3857 * the vec is the final byte in the sequence. (In EBCDIC this is
3858 * converted to I8 to get consecutive values.) To help you visualize
3860 * Straight 1047 After final byte
3861 * UTF-8 UTF-EBCDIC I8 transform
3862 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3863 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3865 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3866 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3868 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3869 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3871 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3872 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3874 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3875 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3877 * (There are no discontinuities in the elided (...) entries.)
3878 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3879 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3880 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3881 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3882 * index into the vec() swatch (after subtracting 0x80, which we
3883 * actually do with an '&').
3884 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3885 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3886 * dicontinuities which go away by transforming it into I8, and we
3887 * effectively subtract 0xA0 to get the index. */
3888 needents = (1 << UTF_ACCUMULATION_SHIFT);
3889 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3893 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3894 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3895 * it's nothing to sniff at.) Pity we usually come through at least
3896 * two function calls to get here...
3898 * NB: this code assumes that swatches are never modified, once generated!
3901 if (hv == PL_last_swash_hv &&
3902 klen == PL_last_swash_klen &&
3903 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3905 tmps = PL_last_swash_tmps;
3906 slen = PL_last_swash_slen;
3909 /* Try our second-level swatch cache, kept in a hash. */
3910 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3912 /* If not cached, generate it via swatch_get */
3913 if (!svp || !SvPOK(*svp)
3914 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3917 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3918 swatch = swatch_get(swash,
3919 code_point & ~((UV)needents - 1),
3922 else { /* For the first 256 code points, the swatch has a key of
3924 swatch = swatch_get(swash, 0, needents);
3927 if (IN_PERL_COMPILETIME)
3928 CopHINTS_set(PL_curcop, PL_hints);
3930 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3932 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3933 || (slen << 3) < needents)
3934 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3935 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3936 svp, tmps, (UV)slen, (UV)needents);
3939 PL_last_swash_hv = hv;
3940 assert(klen <= sizeof(PL_last_swash_key));
3941 PL_last_swash_klen = (U8)klen;
3942 /* FIXME change interpvar.h? */
3943 PL_last_swash_tmps = (U8 *) tmps;
3944 PL_last_swash_slen = slen;
3946 Copy(ptr, PL_last_swash_key, klen, U8);
3949 switch ((int)((slen << 3) / needents)) {
3951 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3953 return ((UV) tmps[off]);
3957 ((UV) tmps[off ] << 8) +
3958 ((UV) tmps[off + 1]);
3962 ((UV) tmps[off ] << 24) +
3963 ((UV) tmps[off + 1] << 16) +
3964 ((UV) tmps[off + 2] << 8) +
3965 ((UV) tmps[off + 3]);
3967 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3968 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
3969 NORETURN_FUNCTION_END;
3972 /* Read a single line of the main body of the swash input text. These are of
3975 * where each number is hex. The first two numbers form the minimum and
3976 * maximum of a range, and the third is the value associated with the range.
3977 * Not all swashes should have a third number
3979 * On input: l points to the beginning of the line to be examined; it points
3980 * to somewhere in the string of the whole input text, and is
3981 * terminated by a \n or the null string terminator.
3982 * lend points to the null terminator of that string
3983 * wants_value is non-zero if the swash expects a third number
3984 * typestr is the name of the swash's mapping, like 'ToLower'
3985 * On output: *min, *max, and *val are set to the values read from the line.
3986 * returns a pointer just beyond the line examined. If there was no
3987 * valid min number on the line, returns lend+1
3991 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3992 const bool wants_value, const U8* const typestr)
3994 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3995 STRLEN numlen; /* Length of the number */
3996 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3997 | PERL_SCAN_DISALLOW_PREFIX
3998 | PERL_SCAN_SILENT_NON_PORTABLE;
4000 /* nl points to the next \n in the scan */
4001 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4003 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4005 /* Get the first number on the line: the range minimum */
4007 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4008 *max = *min; /* So can never return without setting max */
4009 if (numlen) /* If found a hex number, position past it */
4011 else if (nl) { /* Else, go handle next line, if any */
4012 return nl + 1; /* 1 is length of "\n" */
4014 else { /* Else, no next line */
4015 return lend + 1; /* to LIST's end at which \n is not found */
4018 /* The max range value follows, separated by a BLANK */
4021 flags = PERL_SCAN_SILENT_ILLDIGIT
4022 | PERL_SCAN_DISALLOW_PREFIX
4023 | PERL_SCAN_SILENT_NON_PORTABLE;
4025 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4028 else /* If no value here, it is a single element range */
4031 /* Non-binary tables have a third entry: what the first element of the
4032 * range maps to. The map for those currently read here is in hex */
4036 flags = PERL_SCAN_SILENT_ILLDIGIT
4037 | PERL_SCAN_DISALLOW_PREFIX
4038 | PERL_SCAN_SILENT_NON_PORTABLE;
4040 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4049 /* diag_listed_as: To%s: illegal mapping '%s' */
4050 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4056 *val = 0; /* bits == 1, then any val should be ignored */
4058 else { /* Nothing following range min, should be single element with no
4063 /* diag_listed_as: To%s: illegal mapping '%s' */
4064 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4068 *val = 0; /* bits == 1, then val should be ignored */
4071 /* Position to next line if any, or EOF */
4081 * Returns a swatch (a bit vector string) for a code point sequence
4082 * that starts from the value C<start> and comprises the number C<span>.
4083 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4084 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4087 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4090 U8 *l, *lend, *x, *xend, *s, *send;
4091 STRLEN lcur, xcur, scur;
4092 HV *const hv = MUTABLE_HV(SvRV(swash));
4093 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4095 SV** listsvp = NULL; /* The string containing the main body of the table */
4096 SV** extssvp = NULL;
4097 SV** invert_it_svp = NULL;
4100 STRLEN octets; /* if bits == 1, then octets == 0 */
4102 UV end = start + span;
4104 if (invlistsvp == NULL) {
4105 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4106 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4107 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4108 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4109 listsvp = hv_fetchs(hv, "LIST", FALSE);
4110 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4112 bits = SvUV(*bitssvp);
4113 none = SvUV(*nonesvp);
4114 typestr = (U8*)SvPV_nolen(*typesvp);
4120 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4122 PERL_ARGS_ASSERT_SWATCH_GET;
4124 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4125 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4129 /* If overflowed, use the max possible */
4135 /* create and initialize $swatch */
4136 scur = octets ? (span * octets) : (span + 7) / 8;
4137 swatch = newSV(scur);
4139 s = (U8*)SvPVX(swatch);
4140 if (octets && none) {
4141 const U8* const e = s + scur;
4144 *s++ = (U8)(none & 0xff);
4145 else if (bits == 16) {
4146 *s++ = (U8)((none >> 8) & 0xff);
4147 *s++ = (U8)( none & 0xff);
4149 else if (bits == 32) {
4150 *s++ = (U8)((none >> 24) & 0xff);
4151 *s++ = (U8)((none >> 16) & 0xff);
4152 *s++ = (U8)((none >> 8) & 0xff);
4153 *s++ = (U8)( none & 0xff);
4159 (void)memzero((U8*)s, scur + 1);
4161 SvCUR_set(swatch, scur);
4162 s = (U8*)SvPVX(swatch);
4164 if (invlistsvp) { /* If has an inversion list set up use that */
4165 _invlist_populate_swatch(*invlistsvp, start, end, s);
4169 /* read $swash->{LIST} */
4170 l = (U8*)SvPV(*listsvp, lcur);
4173 UV min, max, val, upper;
4174 l = swash_scan_list_line(l, lend, &min, &max, &val,
4175 cBOOL(octets), typestr);
4180 /* If looking for something beyond this range, go try the next one */
4184 /* <end> is generally 1 beyond where we want to set things, but at the
4185 * platform's infinity, where we can't go any higher, we want to
4186 * include the code point at <end> */
4189 : (max != UV_MAX || end != UV_MAX)
4196 if (!none || val < none) {
4201 for (key = min; key <= upper; key++) {
4203 /* offset must be non-negative (start <= min <= key < end) */
4204 offset = octets * (key - start);
4206 s[offset] = (U8)(val & 0xff);
4207 else if (bits == 16) {
4208 s[offset ] = (U8)((val >> 8) & 0xff);
4209 s[offset + 1] = (U8)( val & 0xff);
4211 else if (bits == 32) {
4212 s[offset ] = (U8)((val >> 24) & 0xff);
4213 s[offset + 1] = (U8)((val >> 16) & 0xff);
4214 s[offset + 2] = (U8)((val >> 8) & 0xff);
4215 s[offset + 3] = (U8)( val & 0xff);
4218 if (!none || val < none)
4222 else { /* bits == 1, then val should be ignored */
4227 for (key = min; key <= upper; key++) {
4228 const STRLEN offset = (STRLEN)(key - start);
4229 s[offset >> 3] |= 1 << (offset & 7);
4234 /* Invert if the data says it should be. Assumes that bits == 1 */
4235 if (invert_it_svp && SvUV(*invert_it_svp)) {
4237 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4238 * be 0, and their inversion should also be 0, as we don't succeed any
4239 * Unicode property matches for non-Unicode code points */
4240 if (start <= PERL_UNICODE_MAX) {
4242 /* The code below assumes that we never cross the
4243 * Unicode/above-Unicode boundary in a range, as otherwise we would
4244 * have to figure out where to stop flipping the bits. Since this
4245 * boundary is divisible by a large power of 2, and swatches comes
4246 * in small powers of 2, this should be a valid assumption */
4247 assert(start + span - 1 <= PERL_UNICODE_MAX);
4257 /* read $swash->{EXTRAS}
4258 * This code also copied to swash_to_invlist() below */
4259 x = (U8*)SvPV(*extssvp, xcur);
4267 SV **otherbitssvp, *other;
4271 const U8 opc = *x++;
4275 nl = (U8*)memchr(x, '\n', xend - x);
4277 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4279 x = nl + 1; /* 1 is length of "\n" */
4283 x = xend; /* to EXTRAS' end at which \n is not found */
4290 namelen = nl - namestr;
4294 namelen = xend - namestr;
4298 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4299 otherhv = MUTABLE_HV(SvRV(*othersvp));
4300 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4301 otherbits = (STRLEN)SvUV(*otherbitssvp);
4302 if (bits < otherbits)
4303 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4304 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4306 /* The "other" swatch must be destroyed after. */
4307 other = swatch_get(*othersvp, start, span);
4308 o = (U8*)SvPV(other, olen);
4311 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4313 s = (U8*)SvPV(swatch, slen);
4314 if (bits == 1 && otherbits == 1) {
4316 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4317 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4318 (UV)slen, (UV)olen);
4342 STRLEN otheroctets = otherbits >> 3;
4344 U8* const send = s + slen;
4349 if (otherbits == 1) {
4350 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4354 STRLEN vlen = otheroctets;
4362 if (opc == '+' && otherval)
4363 NOOP; /* replace with otherval */
4364 else if (opc == '!' && !otherval)
4366 else if (opc == '-' && otherval)
4368 else if (opc == '&' && !otherval)
4371 s += octets; /* no replacement */
4376 *s++ = (U8)( otherval & 0xff);
4377 else if (bits == 16) {
4378 *s++ = (U8)((otherval >> 8) & 0xff);
4379 *s++ = (U8)( otherval & 0xff);
4381 else if (bits == 32) {
4382 *s++ = (U8)((otherval >> 24) & 0xff);
4383 *s++ = (U8)((otherval >> 16) & 0xff);
4384 *s++ = (U8)((otherval >> 8) & 0xff);
4385 *s++ = (U8)( otherval & 0xff);
4389 sv_free(other); /* through with it! */
4395 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4398 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4399 * Can't be used on a property that is subject to user override, as it
4400 * relies on the value of SPECIALS in the swash which would be set by
4401 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4402 * for overridden properties
4404 * Returns a hash which is the inversion and closure of a swash mapping.
4405 * For example, consider the input lines:
4410 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4411 * 006C. The value for each key is an array. For 006C, the array would
4412 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4413 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4415 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4416 * keys are only code points that are folded-to, so it isn't a full closure.
4418 * Essentially, for any code point, it gives all the code points that map to
4419 * it, or the list of 'froms' for that point.
4421 * Currently it ignores any additions or deletions from other swashes,
4422 * looking at just the main body of the swash, and if there are SPECIALS
4423 * in the swash, at that hash
4425 * The specials hash can be extra code points, and most likely consists of
4426 * maps from single code points to multiple ones (each expressed as a string
4427 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4428 * However consider this possible input in the specials hash:
4429 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4430 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4432 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4433 * currently handle. But it also means that FB05 and FB06 are equivalent in
4434 * a 1-1 mapping which we should handle, and this relationship may not be in
4435 * the main table. Therefore this function examines all the multi-char
4436 * sequences and adds the 1-1 mappings that come out of that.
4438 * XXX This function was originally intended to be multipurpose, but its
4439 * only use is quite likely to remain for constructing the inversion of
4440 * the CaseFolding (//i) property. If it were more general purpose for
4441 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4442 * because certain folds are prohibited under /iaa and /il. As an example,
4443 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4444 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4445 * prohibited, so we would not figure out that they fold to each other.
4446 * Code could be written to automatically figure this out, similar to the
4447 * code that does this for multi-character folds, but this is the only case
4448 * where something like this is ever likely to happen, as all the single
4449 * char folds to the 0-255 range are now quite settled. Instead there is a
4450 * little special code that is compiled only for this Unicode version. This
4451 * is smaller and didn't require much coding time to do. But this makes
4452 * this routine strongly tied to being used just for CaseFolding. If ever
4453 * it should be generalized, this would have to be fixed */
4457 HV *const hv = MUTABLE_HV(SvRV(swash));
4459 /* The string containing the main body of the table. This will have its
4460 * assertion fail if the swash has been converted to its inversion list */
4461 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4463 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4464 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4465 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4466 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4467 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4468 const STRLEN bits = SvUV(*bitssvp);
4469 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4470 const UV none = SvUV(*nonesvp);
4471 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4475 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4477 /* Must have at least 8 bits to get the mappings */
4478 if (bits != 8 && bits != 16 && bits != 32) {
4479 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4483 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4484 mapping to more than one character */
4486 /* Construct an inverse mapping hash for the specials */
4487 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4488 HV * specials_inverse = newHV();
4489 char *char_from; /* the lhs of the map */
4490 I32 from_len; /* its byte length */
4491 char *char_to; /* the rhs of the map */
4492 I32 to_len; /* its byte length */
4493 SV *sv_to; /* and in a sv */
4494 AV* from_list; /* list of things that map to each 'to' */
4496 hv_iterinit(specials_hv);
4498 /* The keys are the characters (in UTF-8) that map to the corresponding
4499 * UTF-8 string value. Iterate through the list creating the inverse
4501 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4503 if (! SvPOK(sv_to)) {
4504 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4505 "unexpectedly is not a string, flags=%lu",
4506 (unsigned long)SvFLAGS(sv_to));
4508 /*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)));*/
4510 /* Each key in the inverse list is a mapped-to value, and the key's
4511 * hash value is a list of the strings (each in UTF-8) that map to
4512 * it. Those strings are all one character long */
4513 if ((listp = hv_fetch(specials_inverse,
4517 from_list = (AV*) *listp;
4519 else { /* No entry yet for it: create one */
4520 from_list = newAV();
4521 if (! hv_store(specials_inverse,
4524 (SV*) from_list, 0))
4526 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4530 /* Here have the list associated with this 'to' (perhaps newly
4531 * created and empty). Just add to it. Note that we ASSUME that
4532 * the input is guaranteed to not have duplications, so we don't
4533 * check for that. Duplications just slow down execution time. */
4534 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4537 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4538 * it looking for cases like the FB05/FB06 examples above. There would
4539 * be an entry in the hash like
4540 * 'st' => [ FB05, FB06 ]
4541 * In this example we will create two lists that get stored in the
4542 * returned hash, 'ret':
4543 * FB05 => [ FB05, FB06 ]
4544 * FB06 => [ FB05, FB06 ]
4546 * Note that there is nothing to do if the array only has one element.
4547 * (In the normal 1-1 case handled below, we don't have to worry about
4548 * two lists, as everything gets tied to the single list that is
4549 * generated for the single character 'to'. But here, we are omitting
4550 * that list, ('st' in the example), so must have multiple lists.) */
4551 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4552 &char_to, &to_len)))
4554 if (av_tindex_skip_len_mg(from_list) > 0) {
4557 /* We iterate over all combinations of i,j to place each code
4558 * point on each list */
4559 for (i = 0; i <= av_tindex_skip_len_mg(from_list); i++) {
4561 AV* i_list = newAV();
4562 SV** entryp = av_fetch(from_list, i, FALSE);
4563 if (entryp == NULL) {
4564 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4566 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4567 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4569 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4570 (SV*) i_list, FALSE))
4572 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4575 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4576 for (j = 0; j <= av_tindex_skip_len_mg(from_list); j++) {
4577 entryp = av_fetch(from_list, j, FALSE);
4578 if (entryp == NULL) {
4579 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4582 /* When i==j this adds itself to the list */
4583 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4584 (U8*) SvPVX(*entryp),
4585 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4587 /*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));*/
4592 SvREFCNT_dec(specials_inverse); /* done with it */
4593 } /* End of specials */
4595 /* read $swash->{LIST} */
4597 #if UNICODE_MAJOR_VERSION == 3 \
4598 && UNICODE_DOT_VERSION == 0 \
4599 && UNICODE_DOT_DOT_VERSION == 1
4601 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4602 * rule so that things work under /iaa and /il */
4604 SV * mod_listsv = sv_mortalcopy(*listsvp);
4605 sv_catpv(mod_listsv, "130\t130\t131\n");
4606 l = (U8*)SvPV(mod_listsv, lcur);
4610 l = (U8*)SvPV(*listsvp, lcur);
4616 /* Go through each input line */
4620 l = swash_scan_list_line(l, lend, &min, &max, &val,
4621 cBOOL(octets), typestr);
4626 /* Each element in the range is to be inverted */
4627 for (inverse = min; inverse <= max; inverse++) {
4631 bool found_key = FALSE;
4632 bool found_inverse = FALSE;
4634 /* The key is the inverse mapping */
4635 char key[UTF8_MAXBYTES+1];
4636 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4637 STRLEN key_len = key_end - key;
4639 /* Get the list for the map */
4640 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4641 list = (AV*) *listp;
4643 else { /* No entry yet for it: create one */
4645 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4646 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4650 /* Look through list to see if this inverse mapping already is
4651 * listed, or if there is a mapping to itself already */
4652 for (i = 0; i <= av_tindex_skip_len_mg(list); i++) {
4653 SV** entryp = av_fetch(list, i, FALSE);
4656 if (entryp == NULL) {
4657 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4661 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4665 if (uv == inverse) {
4666 found_inverse = TRUE;
4669 /* No need to continue searching if found everything we are
4671 if (found_key && found_inverse) {
4676 /* Make sure there is a mapping to itself on the list */
4678 av_push(list, newSVuv(val));
4679 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4683 /* Simply add the value to the list */
4684 if (! found_inverse) {
4685 av_push(list, newSVuv(inverse));
4686 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4689 /* swatch_get() increments the value of val for each element in the
4690 * range. That makes more compact tables possible. You can
4691 * express the capitalization, for example, of all consecutive
4692 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4693 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4694 * and it's not documented; it appears to be used only in
4695 * implementing tr//; I copied the semantics from swatch_get(), just
4697 if (!none || val < none) {
4707 Perl__swash_to_invlist(pTHX_ SV* const swash)
4710 /* Subject to change or removal. For use only in one place in regcomp.c.
4711 * Ownership is given to one reference count in the returned SV* */
4716 HV *const hv = MUTABLE_HV(SvRV(swash));
4717 UV elements = 0; /* Number of elements in the inversion list */
4727 STRLEN octets; /* if bits == 1, then octets == 0 */
4733 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4735 /* If not a hash, it must be the swash's inversion list instead */
4736 if (SvTYPE(hv) != SVt_PVHV) {
4737 return SvREFCNT_inc_simple_NN((SV*) hv);
4740 /* The string containing the main body of the table */
4741 listsvp = hv_fetchs(hv, "LIST", FALSE);
4742 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4743 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4744 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4745 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4747 typestr = (U8*)SvPV_nolen(*typesvp);
4748 bits = SvUV(*bitssvp);
4749 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4751 /* read $swash->{LIST} */
4752 if (SvPOK(*listsvp)) {
4753 l = (U8*)SvPV(*listsvp, lcur);
4756 /* LIST legitimately doesn't contain a string during compilation phases
4757 * of Perl itself, before the Unicode tables are generated. In this
4758 * case, just fake things up by creating an empty list */
4765 if (*l == 'V') { /* Inversion list format */
4766 const char *after_atou = (char *) lend;
4768 UV* other_elements_ptr;
4770 /* The first number is a count of the rest */
4772 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4773 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
4775 if (elements == 0) {
4776 invlist = _new_invlist(0);
4779 l = (U8 *) after_atou;
4781 /* Get the 0th element, which is needed to setup the inversion list */
4782 while (isSPACE(*l)) l++;
4783 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4784 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
4786 l = (U8 *) after_atou;
4787 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
4790 /* Then just populate the rest of the input */
4791 while (elements-- > 0) {
4793 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more elements than available", elements);
4795 while (isSPACE(*l)) l++;
4796 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
4797 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
4799 l = (U8 *) after_atou;
4805 /* Scan the input to count the number of lines to preallocate array
4806 * size based on worst possible case, which is each line in the input
4807 * creates 2 elements in the inversion list: 1) the beginning of a
4808 * range in the list; 2) the beginning of a range not in the list. */
4809 while ((loc = (strchr(loc, '\n'))) != NULL) {
4814 /* If the ending is somehow corrupt and isn't a new line, add another
4815 * element for the final range that isn't in the inversion list */
4816 if (! (*lend == '\n'
4817 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4822 invlist = _new_invlist(elements);
4824 /* Now go through the input again, adding each range to the list */
4827 UV val; /* Not used by this function */
4829 l = swash_scan_list_line(l, lend, &start, &end, &val,
4830 cBOOL(octets), typestr);
4836 invlist = _add_range_to_invlist(invlist, start, end);
4840 /* Invert if the data says it should be */
4841 if (invert_it_svp && SvUV(*invert_it_svp)) {
4842 _invlist_invert(invlist);
4845 /* This code is copied from swatch_get()
4846 * read $swash->{EXTRAS} */
4847 x = (U8*)SvPV(*extssvp, xcur);
4855 SV **otherbitssvp, *other;
4858 const U8 opc = *x++;
4862 nl = (U8*)memchr(x, '\n', xend - x);
4864 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4866 x = nl + 1; /* 1 is length of "\n" */
4870 x = xend; /* to EXTRAS' end at which \n is not found */
4877 namelen = nl - namestr;
4881 namelen = xend - namestr;
4885 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4886 otherhv = MUTABLE_HV(SvRV(*othersvp));
4887 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4888 otherbits = (STRLEN)SvUV(*otherbitssvp);
4890 if (bits != otherbits || bits != 1) {
4891 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4892 "properties, bits=%" UVuf ", otherbits=%" UVuf,
4893 (UV)bits, (UV)otherbits);
4896 /* The "other" swatch must be destroyed after. */
4897 other = _swash_to_invlist((SV *)*othersvp);
4899 /* End of code copied from swatch_get() */
4902 _invlist_union(invlist, other, &invlist);
4905 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4908 _invlist_subtract(invlist, other, &invlist);
4911 _invlist_intersection(invlist, other, &invlist);
4916 sv_free(other); /* through with it! */
4919 SvREADONLY_on(invlist);
4924 Perl__get_swash_invlist(pTHX_ SV* const swash)
4928 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4930 if (! SvROK(swash)) {
4934 /* If it really isn't a hash, it isn't really swash; must be an inversion
4936 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4940 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4949 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4951 /* May change: warns if surrogates, non-character code points, or
4952 * non-Unicode code points are in s which has length len bytes. Returns
4953 * TRUE if none found; FALSE otherwise. The only other validity check is
4954 * to make sure that this won't exceed the string's length.
4956 * Code points above the platform's C<IV_MAX> will raise a deprecation
4957 * warning, unless those are turned off. */
4959 const U8* const e = s + len;
4962 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4965 if (UTF8SKIP(s) > len) {
4966 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4967 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4970 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4971 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4972 if ( ckWARN_d(WARN_NON_UNICODE)
4973 || ( ckWARN_d(WARN_DEPRECATED)
4975 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
4976 #else /* Below is 64-bit words */
4977 /* 2**63 and up meet these conditions provided we have
4981 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
4984 /* s[1] being above 0x80 overflows */
4989 /* A side effect of this function will be to warn */
4990 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4994 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4995 if (ckWARN_d(WARN_SURROGATE)) {
4996 /* This has a different warning than the one the called
4997 * function would output, so can't just call it, unlike we
4998 * do for the non-chars and above-unicodes */
4999 UV uv = utf8_to_uvchr_buf(s, e, NULL);
5000 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
5001 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", uv);
5005 else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
5006 /* A side effect of this function will be to warn */
5007 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5018 =for apidoc pv_uni_display
5020 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5021 length C<len>, the displayable version being at most C<pvlim> bytes long
5022 (if longer, the rest is truncated and C<"..."> will be appended).
5024 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5025 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5026 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5027 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5028 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5029 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5031 The pointer to the PV of the C<dsv> is returned.
5033 See also L</sv_uni_display>.
5037 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
5042 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5046 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5048 /* This serves double duty as a flag and a character to print after
5049 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5053 if (pvlim && SvCUR(dsv) >= pvlim) {
5057 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5059 const unsigned char c = (unsigned char)u & 0xFF;
5060 if (flags & UNI_DISPLAY_BACKSLASH) {
5077 const char string = ok;
5078 sv_catpvs(dsv, "\\");
5079 sv_catpvn(dsv, &string, 1);
5082 /* isPRINT() is the locale-blind version. */
5083 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5084 const char string = c;
5085 sv_catpvn(dsv, &string, 1);
5090 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5093 sv_catpvs(dsv, "...");
5099 =for apidoc sv_uni_display
5101 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5102 the displayable version being at most C<pvlim> bytes long
5103 (if longer, the rest is truncated and "..." will be appended).
5105 The C<flags> argument is as in L</pv_uni_display>().
5107 The pointer to the PV of the C<dsv> is returned.
5112 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5114 const char * const ptr =
5115 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5117 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5119 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5120 SvCUR(ssv), pvlim, flags);
5124 =for apidoc foldEQ_utf8
5126 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
5127 of which may be in UTF-8) are the same case-insensitively; false otherwise.
5128 How far into the strings to compare is determined by other input parameters.
5130 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5131 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
5132 with respect to C<s2>.
5134 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
5135 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
5136 scan will not be considered to be a match unless the goal is reached, and
5137 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
5140 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
5141 considered an end pointer to the position 1 byte past the maximum point
5142 in C<s1> beyond which scanning will not continue under any circumstances.
5143 (This routine assumes that UTF-8 encoded input strings are not malformed;
5144 malformed input can cause it to read past C<pe1>).
5145 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
5146 is less than C<s1>+C<l1>, the match will never be successful because it can
5148 get as far as its goal (and in fact is asserted against). Correspondingly for
5149 C<pe2> with respect to C<s2>.
5151 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5152 C<l2> must be non-zero), and if both do, both have to be
5153 reached for a successful match. Also, if the fold of a character is multiple
5154 characters, all of them must be matched (see tr21 reference below for
5157 Upon a successful match, if C<pe1> is non-C<NULL>,
5158 it will be set to point to the beginning of the I<next> character of C<s1>
5159 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5161 For case-insensitiveness, the "casefolding" of Unicode is used
5162 instead of upper/lowercasing both the characters, see
5163 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5167 /* A flags parameter has been added which may change, and hence isn't
5168 * externally documented. Currently it is:
5169 * 0 for as-documented above
5170 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5171 ASCII one, to not match
5172 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5173 * locale are to be used.
5174 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5175 * routine. This allows that step to be skipped.
5176 * Currently, this requires s1 to be encoded as UTF-8
5177 * (u1 must be true), which is asserted for.
5178 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5179 * cross certain boundaries. Hence, the caller should
5180 * let this function do the folding instead of
5181 * pre-folding. This code contains an assertion to
5182 * that effect. However, if the caller knows what
5183 * it's doing, it can pass this flag to indicate that,
5184 * and the assertion is skipped.
5185 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5186 * FOLDEQ_S2_FOLDS_SANE
5189 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)
5191 const U8 *p1 = (const U8*)s1; /* Point to current char */
5192 const U8 *p2 = (const U8*)s2;
5193 const U8 *g1 = NULL; /* goal for s1 */
5194 const U8 *g2 = NULL;
5195 const U8 *e1 = NULL; /* Don't scan s1 past this */
5196 U8 *f1 = NULL; /* Point to current folded */
5197 const U8 *e2 = NULL;
5199 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5200 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5201 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5202 U8 flags_for_folder = FOLD_FLAGS_FULL;
5204 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5206 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5207 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5208 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5209 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5210 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5211 /* The algorithm is to trial the folds without regard to the flags on
5212 * the first line of the above assert(), and then see if the result
5213 * violates them. This means that the inputs can't be pre-folded to a
5214 * violating result, hence the assert. This could be changed, with the
5215 * addition of extra tests here for the already-folded case, which would
5216 * slow it down. That cost is more than any possible gain for when these
5217 * flags are specified, as the flags indicate /il or /iaa matching which
5218 * is less common than /iu, and I (khw) also believe that real-world /il
5219 * and /iaa matches are most likely to involve code points 0-255, and this
5220 * function only under rare conditions gets called for 0-255. */
5222 if (flags & FOLDEQ_LOCALE) {
5223 if (IN_UTF8_CTYPE_LOCALE) {
5224 flags &= ~FOLDEQ_LOCALE;
5227 flags_for_folder |= FOLD_FLAGS_LOCALE;
5236 g1 = (const U8*)s1 + l1;
5244 g2 = (const U8*)s2 + l2;
5247 /* Must have at least one goal */
5252 /* Will never match if goal is out-of-bounds */
5253 assert(! e1 || e1 >= g1);
5255 /* Here, there isn't an end pointer, or it is beyond the goal. We
5256 * only go as far as the goal */
5260 assert(e1); /* Must have an end for looking at s1 */
5263 /* Same for goal for s2 */
5265 assert(! e2 || e2 >= g2);
5272 /* If both operands are already folded, we could just do a memEQ on the
5273 * whole strings at once, but it would be better if the caller realized
5274 * this and didn't even call us */
5276 /* Look through both strings, a character at a time */
5277 while (p1 < e1 && p2 < e2) {
5279 /* If at the beginning of a new character in s1, get its fold to use
5280 * and the length of the fold. */
5282 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5288 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5290 /* We have to forbid mixing ASCII with non-ASCII if the
5291 * flags so indicate. And, we can short circuit having to
5292 * call the general functions for this common ASCII case,
5293 * all of whose non-locale folds are also ASCII, and hence
5294 * UTF-8 invariants, so the UTF8ness of the strings is not
5296 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5300 *foldbuf1 = toFOLD(*p1);
5303 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5305 else { /* Not UTF-8, get UTF-8 fold */
5306 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5312 if (n2 == 0) { /* Same for s2 */
5313 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5319 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5320 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5324 *foldbuf2 = toFOLD(*p2);
5327 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5330 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5336 /* Here f1 and f2 point to the beginning of the strings to compare.
5337 * These strings are the folds of the next character from each input
5338 * string, stored in UTF-8. */
5340 /* While there is more to look for in both folds, see if they
5341 * continue to match */
5343 U8 fold_length = UTF8SKIP(f1);
5344 if (fold_length != UTF8SKIP(f2)
5345 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5346 function call for single
5348 || memNE((char*)f1, (char*)f2, fold_length))
5350 return 0; /* mismatch */
5353 /* Here, they matched, advance past them */
5360 /* When reach the end of any fold, advance the input past it */
5362 p1 += u1 ? UTF8SKIP(p1) : 1;
5365 p2 += u2 ? UTF8SKIP(p2) : 1;
5367 } /* End of loop through both strings */
5369 /* A match is defined by each scan that specified an explicit length
5370 * reaching its final goal, and the other not having matched a partial
5371 * character (which can happen when the fold of a character is more than one
5373 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5377 /* Successful match. Set output pointers */
5387 /* XXX The next two functions should likely be moved to mathoms.c once all
5388 * occurrences of them are removed from the core; some cpan-upstream modules
5392 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5394 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5396 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5400 =for apidoc utf8n_to_uvuni
5402 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5404 This function was useful for code that wanted to handle both EBCDIC and
5405 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5406 distinctions between the platforms have mostly been made invisible to most
5407 code, so this function is quite unlikely to be what you want. If you do need
5408 this precise functionality, use instead
5409 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5410 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5416 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5418 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5420 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5424 =for apidoc uvuni_to_utf8_flags
5426 Instead you almost certainly want to use L</uvchr_to_utf8> or
5427 L</uvchr_to_utf8_flags>.
5429 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5430 which itself, while not deprecated, should be used only in isolated
5431 circumstances. These functions were useful for code that wanted to handle
5432 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5433 v5.20, the distinctions between the platforms have mostly been made invisible
5434 to most code, so this function is quite unlikely to be what you want.
5440 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5442 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5444 return uvoffuni_to_utf8_flags(d, uv, flags);
5448 * ex: set ts=8 sts=4 sw=4 et: