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 S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
760 /* Returns a mortalized C string that is a displayable copy of the 'len'
761 * bytes starting at 's', each in a \xXY format. */
763 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
765 const U8 * const e = s + len;
769 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
771 Newx(output, output_len, char);
776 const unsigned high_nibble = (*s & 0xF0) >> 4;
777 const unsigned low_nibble = (*s & 0x0F);
782 if (high_nibble < 10) {
783 *d++ = high_nibble + '0';
786 *d++ = high_nibble - 10 + 'a';
789 if (low_nibble < 10) {
790 *d++ = low_nibble + '0';
793 *d++ = low_nibble - 10 + 'a';
801 PERL_STATIC_INLINE char *
802 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
804 /* How many bytes to print */
807 /* Which one is the non-continuation */
808 const STRLEN non_cont_byte_pos,
810 /* How many bytes should there be? */
811 const STRLEN expect_len)
813 /* Return the malformation warning text for an unexpected continuation
816 const char * const where = (non_cont_byte_pos == 1)
818 : Perl_form(aTHX_ "%d bytes",
819 (int) non_cont_byte_pos);
821 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
823 /* We don't need to pass this parameter, but since it has already been
824 * calculated, it's likely faster to pass it; verify under DEBUGGING */
825 assert(expect_len == UTF8SKIP(s));
827 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
828 " %s after start byte 0x%02x; need %d bytes, got %d)",
830 _byte_dump_string(s, print_len),
831 *(s + non_cont_byte_pos),
835 (int) non_cont_byte_pos);
840 =for apidoc utf8n_to_uvchr
842 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
843 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
845 Bottom level UTF-8 decode routine.
846 Returns the native code point value of the first character in the string C<s>,
847 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
848 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
849 the length, in bytes, of that character.
851 The value of C<flags> determines the behavior when C<s> does not point to a
852 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
853 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
854 is the next possible position in C<s> that could begin a non-malformed
855 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
856 is raised. Some UTF-8 input sequences may contain multiple malformations.
857 This function tries to find every possible one in each call, so multiple
858 warnings can be raised for each sequence.
860 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
861 individual types of malformations, such as the sequence being overlong (that
862 is, when there is a shorter sequence that can express the same code point;
863 overlong sequences are expressly forbidden in the UTF-8 standard due to
864 potential security issues). Another malformation example is the first byte of
865 a character not being a legal first byte. See F<utf8.h> for the list of such
866 flags. Even if allowed, this function generally returns the Unicode
867 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
868 F<utf8.h> to override this behavior for the overlong malformations, but don't
869 do that except for very specialized purposes.
871 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
872 flags) malformation is found. If this flag is set, the routine assumes that
873 the caller will raise a warning, and this function will silently just set
874 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
876 Note that this API requires disambiguation between successful decoding a C<NUL>
877 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
878 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
879 be set to 1. To disambiguate, upon a zero return, see if the first byte of
880 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
881 error. Or you can use C<L</utf8n_to_uvchr_error>>.
883 Certain code points are considered problematic. These are Unicode surrogates,
884 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
885 By default these are considered regular code points, but certain situations
886 warrant special handling for them, which can be specified using the C<flags>
887 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
888 three classes are treated as malformations and handled as such. The flags
889 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
890 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
891 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
892 restricts the allowed inputs to the strict UTF-8 traditionally defined by
893 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
895 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
896 The difference between traditional strictness and C9 strictness is that the
897 latter does not forbid non-character code points. (They are still discouraged,
898 however.) For more discussion see L<perlunicode/Noncharacter code points>.
900 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
901 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
902 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
903 raised for their respective categories, but otherwise the code points are
904 considered valid (not malformations). To get a category to both be treated as
905 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
906 (But note that warnings are not raised if lexically disabled nor if
907 C<UTF8_CHECK_ONLY> is also specified.)
909 It is now deprecated to have very high code points (above C<IV_MAX> on the
910 platforms) and this function will raise a deprecation warning for these (unless
911 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
914 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
915 so using them is more problematic than other above-Unicode code points. Perl
916 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
917 likely that non-Perl languages will not be able to read files that contain
918 these; nor would Perl understand files
919 written by something that uses a different extension. For these reasons, there
920 is a separate set of flags that can warn and/or disallow these extremely high
921 code points, even if other above-Unicode ones are accepted. These are the
922 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
923 are entirely independent from the deprecation warning for code points above
924 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
925 code point that needs more than 31 bits to represent. When that happens,
926 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
927 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
928 above-Unicode code points, including these, as malformations; and
929 C<UTF8_WARN_SUPER> warns on these.)
931 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
932 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
933 than on ASCII. Prior to that, code points 2**31 and higher were simply
934 unrepresentable, and a different, incompatible method was used to represent
935 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
936 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
937 platforms, warning and disallowing 2**31 and higher.
939 All other code points corresponding to Unicode characters, including private
940 use and those yet to be assigned, are never considered malformed and never
945 Also implemented as a macro in utf8.h
949 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
954 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
956 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
961 =for apidoc utf8n_to_uvchr_error
963 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
964 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
966 This function is for code that needs to know what the precise malformation(s)
967 are when an error is found.
969 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
970 all the others, C<errors>. If this parameter is 0, this function behaves
971 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
972 to a C<U32> variable, which this function sets to indicate any errors found.
973 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
974 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
975 of these bits will be set if a malformation is found, even if the input
976 C<flags> parameter indicates that the given malformation is allowed; those
977 exceptions are noted:
981 =item C<UTF8_GOT_ABOVE_31_BIT>
983 The code point represented by the input UTF-8 sequence occupies more than 31
985 This bit is set only if the input C<flags> parameter contains either the
986 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
988 =item C<UTF8_GOT_CONTINUATION>
990 The input sequence was malformed in that the first byte was a a UTF-8
993 =item C<UTF8_GOT_EMPTY>
995 The input C<curlen> parameter was 0.
997 =item C<UTF8_GOT_LONG>
999 The input sequence was malformed in that there is some other sequence that
1000 evaluates to the same code point, but that sequence is shorter than this one.
1002 =item C<UTF8_GOT_NONCHAR>
1004 The code point represented by the input UTF-8 sequence is for a Unicode
1005 non-character code point.
1006 This bit is set only if the input C<flags> parameter contains either the
1007 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1009 =item C<UTF8_GOT_NON_CONTINUATION>
1011 The input sequence was malformed in that a non-continuation type byte was found
1012 in a position where only a continuation type one should be.
1014 =item C<UTF8_GOT_OVERFLOW>
1016 The input sequence was malformed in that it is for a code point that is not
1017 representable in the number of bits available in a UV on the current platform.
1019 =item C<UTF8_GOT_SHORT>
1021 The input sequence was malformed in that C<curlen> is smaller than required for
1022 a complete sequence. In other words, the input is for a partial character
1025 =item C<UTF8_GOT_SUPER>
1027 The input sequence was malformed in that it is for a non-Unicode code point;
1028 that is, one above the legal Unicode maximum.
1029 This bit is set only if the input C<flags> parameter contains either the
1030 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1032 =item C<UTF8_GOT_SURROGATE>
1034 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1036 This bit is set only if the input C<flags> parameter contains either the
1037 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1041 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1042 flag to suppress any warnings, and then examine the C<*errors> return.
1048 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1054 const U8 * const s0 = s;
1055 U8 * send = NULL; /* (initialized to silence compilers' wrong
1057 U32 possible_problems = 0; /* A bit is set here for each potential problem
1058 found as we go along */
1060 STRLEN expectlen = 0; /* How long should this sequence be?
1061 (initialized to silence compilers' wrong
1063 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1064 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1065 this gets set and discarded */
1067 /* The below are used only if there is both an overlong malformation and a
1068 * too short one. Otherwise the first two are set to 's0' and 'send', and
1069 * the third not used at all */
1070 U8 * adjusted_s0 = (U8 *) s0;
1071 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1073 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1075 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1081 errors = &discard_errors;
1084 /* The order of malformation tests here is important. We should consume as
1085 * few bytes as possible in order to not skip any valid character. This is
1086 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1087 * http://unicode.org/reports/tr36 for more discussion as to why. For
1088 * example, once we've done a UTF8SKIP, we can tell the expected number of
1089 * bytes, and could fail right off the bat if the input parameters indicate
1090 * that there are too few available. But it could be that just that first
1091 * byte is garbled, and the intended character occupies fewer bytes. If we
1092 * blindly assumed that the first byte is correct, and skipped based on
1093 * that number, we could skip over a valid input character. So instead, we
1094 * always examine the sequence byte-by-byte.
1096 * We also should not consume too few bytes, otherwise someone could inject
1097 * things. For example, an input could be deliberately designed to
1098 * overflow, and if this code bailed out immediately upon discovering that,
1099 * returning to the caller C<*retlen> pointing to the very next byte (one
1100 * which is actually part of of the overflowing sequence), that could look
1101 * legitimate to the caller, which could discard the initial partial
1102 * sequence and process the rest, inappropriately.
1104 * Some possible input sequences are malformed in more than one way. This
1105 * function goes to lengths to try to find all of them. This is necessary
1106 * for correctness, as the inputs may allow one malformation but not
1107 * another, and if we abandon searching for others after finding the
1108 * allowed one, we could allow in something that shouldn't have been.
1111 if (UNLIKELY(curlen == 0)) {
1112 possible_problems |= UTF8_GOT_EMPTY;
1114 uv = UNICODE_REPLACEMENT;
1115 goto ready_to_handle_errors;
1118 expectlen = UTF8SKIP(s);
1120 /* A well-formed UTF-8 character, as the vast majority of calls to this
1121 * function will be for, has this expected length. For efficiency, set
1122 * things up here to return it. It will be overriden only in those rare
1123 * cases where a malformation is found */
1125 *retlen = expectlen;
1128 /* An invariant is trivially well-formed */
1129 if (UTF8_IS_INVARIANT(uv)) {
1133 /* A continuation character can't start a valid sequence */
1134 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1135 possible_problems |= UTF8_GOT_CONTINUATION;
1137 uv = UNICODE_REPLACEMENT;
1138 goto ready_to_handle_errors;
1141 /* Here is not a continuation byte, nor an invariant. The only thing left
1142 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1143 * because it excludes start bytes like \xC0 that always lead to
1146 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1147 * that indicate the number of bytes in the character's whole UTF-8
1148 * sequence, leaving just the bits that are part of the value. */
1149 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1151 /* Setup the loop end point, making sure to not look past the end of the
1152 * input string, and flag it as too short if the size isn't big enough. */
1154 if (UNLIKELY(curlen < expectlen)) {
1155 possible_problems |= UTF8_GOT_SHORT;
1162 adjusted_send = send;
1164 /* Now, loop through the remaining bytes in the character's sequence,
1165 * accumulating each into the working value as we go. */
1166 for (s = s0 + 1; s < send; s++) {
1167 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1168 uv = UTF8_ACCUMULATE(uv, *s);
1172 /* Here, found a non-continuation before processing all expected bytes.
1173 * This byte indicates the beginning of a new character, so quit, even
1174 * if allowing this malformation. */
1175 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1177 } /* End of loop through the character's bytes */
1179 /* Save how many bytes were actually in the character */
1182 /* A convenience macro that matches either of the too-short conditions. */
1183 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1185 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1187 uv = UNICODE_REPLACEMENT;
1190 /* Note that there are two types of too-short malformation. One is when
1191 * there is actual wrong data before the normal termination of the
1192 * sequence. The other is that the sequence wasn't complete before the end
1193 * of the data we are allowed to look at, based on the input 'curlen'.
1194 * This means that we were passed data for a partial character, but it is
1195 * valid as far as we saw. The other is definitely invalid. This
1196 * distinction could be important to a caller, so the two types are kept
1199 /* Check for overflow */
1200 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1201 possible_problems |= UTF8_GOT_OVERFLOW;
1202 uv = UNICODE_REPLACEMENT;
1205 /* Check for overlong. If no problems so far, 'uv' is the correct code
1206 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1207 * we must look at the UTF-8 byte sequence itself to see if it is for an
1209 if ( ( LIKELY(! possible_problems)
1210 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1211 || ( UNLIKELY( possible_problems)
1212 && ( UNLIKELY(! UTF8_IS_START(*s0))
1214 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1217 possible_problems |= UTF8_GOT_LONG;
1219 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1220 UV min_uv = uv_so_far;
1223 /* Here, the input is both overlong and is missing some trailing
1224 * bytes. There is no single code point it could be for, but there
1225 * may be enough information present to determine if what we have
1226 * so far is for an unallowed code point, such as for a surrogate.
1227 * The code below has the intelligence to determine this, but just
1228 * for non-overlong UTF-8 sequences. What we do here is calculate
1229 * the smallest code point the input could represent if there were
1230 * no too short malformation. Then we compute and save the UTF-8
1231 * for that, which is what the code below looks at instead of the
1232 * raw input. It turns out that the smallest such code point is
1234 for (i = curlen; i < expectlen; i++) {
1235 min_uv = UTF8_ACCUMULATE(min_uv,
1236 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1239 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1240 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1241 to free it ourselves if
1242 warnings are made fatal */
1243 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1247 /* Now check that the input isn't for a problematic code point not allowed
1248 * by the input parameters. */
1249 /* isn't problematic if < this */
1250 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1251 || ( UNLIKELY(possible_problems)
1253 /* if overflow, we know without looking further
1254 * precisely which of the problematic types it is,
1255 * and we deal with those in the overflow handling
1257 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1258 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1259 && ((flags & ( UTF8_DISALLOW_NONCHAR
1260 |UTF8_DISALLOW_SURROGATE
1261 |UTF8_DISALLOW_SUPER
1262 |UTF8_DISALLOW_ABOVE_31_BIT
1264 |UTF8_WARN_SURROGATE
1266 |UTF8_WARN_ABOVE_31_BIT))
1267 /* In case of a malformation, 'uv' is not valid, and has
1268 * been changed to something in the Unicode range.
1269 * Currently we don't output a deprecation message if there
1270 * is already a malformation, so we don't have to special
1271 * case the test immediately below */
1272 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1273 && ckWARN_d(WARN_DEPRECATED))))
1275 /* If there were no malformations, or the only malformation is an
1276 * overlong, 'uv' is valid */
1277 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1278 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1279 possible_problems |= UTF8_GOT_SURROGATE;
1281 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1282 possible_problems |= UTF8_GOT_SUPER;
1284 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1285 possible_problems |= UTF8_GOT_NONCHAR;
1288 else { /* Otherwise, need to look at the source UTF-8, possibly
1289 adjusted to be non-overlong */
1291 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1292 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1294 possible_problems |= UTF8_GOT_SUPER;
1296 else if (curlen > 1) {
1297 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1298 NATIVE_UTF8_TO_I8(*adjusted_s0),
1299 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1301 possible_problems |= UTF8_GOT_SUPER;
1303 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1304 NATIVE_UTF8_TO_I8(*adjusted_s0),
1305 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1307 possible_problems |= UTF8_GOT_SURROGATE;
1311 /* We need a complete well-formed UTF-8 character to discern
1312 * non-characters, so can't look for them here */
1316 ready_to_handle_errors:
1319 * curlen contains the number of bytes in the sequence that
1320 * this call should advance the input by.
1321 * avail_len gives the available number of bytes passed in, but
1322 * only if this is less than the expected number of
1323 * bytes, based on the code point's start byte.
1324 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1325 * is set in it for each potential problem found.
1326 * uv contains the code point the input sequence
1327 * represents; or if there is a problem that prevents
1328 * a well-defined value from being computed, it is
1329 * some subsitute value, typically the REPLACEMENT
1331 * s0 points to the first byte of the character
1332 * send points to just after where that (potentially
1333 * partial) character ends
1334 * adjusted_s0 normally is the same as s0, but in case of an
1335 * overlong for which the UTF-8 matters below, it is
1336 * the first byte of the shortest form representation
1338 * adjusted_send normally is the same as 'send', but if adjusted_s0
1339 * is set to something other than s0, this points one
1343 if (UNLIKELY(possible_problems)) {
1344 bool disallowed = FALSE;
1345 const U32 orig_problems = possible_problems;
1347 while (possible_problems) { /* Handle each possible problem */
1349 char * message = NULL;
1351 /* Each 'if' clause handles one problem. They are ordered so that
1352 * the first ones' messages will be displayed before the later
1353 * ones; this is kinda in decreasing severity order */
1354 if (possible_problems & UTF8_GOT_OVERFLOW) {
1356 /* Overflow means also got a super and above 31 bits, but we
1357 * handle all three cases here */
1359 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1360 *errors |= UTF8_GOT_OVERFLOW;
1362 /* But the API says we flag all errors found */
1363 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1364 *errors |= UTF8_GOT_SUPER;
1367 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1369 *errors |= UTF8_GOT_ABOVE_31_BIT;
1372 /* Disallow if any of the three categories say to */
1373 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1374 || (flags & ( UTF8_DISALLOW_SUPER
1375 |UTF8_DISALLOW_ABOVE_31_BIT)))
1381 /* Likewise, warn if any say to, plus if deprecation warnings
1382 * are on, because this code point is above IV_MAX */
1383 if ( ckWARN_d(WARN_DEPRECATED)
1384 || ! (flags & UTF8_ALLOW_OVERFLOW)
1385 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1388 /* The warnings code explicitly says it doesn't handle the
1389 * case of packWARN2 and two categories which have
1390 * parent-child relationship. Even if it works now to
1391 * raise the warning if either is enabled, it wouldn't
1392 * necessarily do so in the future. We output (only) the
1393 * most dire warning*/
1394 if (! (flags & UTF8_CHECK_ONLY)) {
1395 if (ckWARN_d(WARN_UTF8)) {
1396 pack_warn = packWARN(WARN_UTF8);
1398 else if (ckWARN_d(WARN_NON_UNICODE)) {
1399 pack_warn = packWARN(WARN_NON_UNICODE);
1402 message = Perl_form(aTHX_ "%s: %s (overflows)",
1404 _byte_dump_string(s0, send - s0));
1409 else if (possible_problems & UTF8_GOT_EMPTY) {
1410 possible_problems &= ~UTF8_GOT_EMPTY;
1411 *errors |= UTF8_GOT_EMPTY;
1413 if (! (flags & UTF8_ALLOW_EMPTY)) {
1415 /* This so-called malformation is now treated as a bug in
1416 * the caller. If you have nothing to decode, skip calling
1421 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1422 pack_warn = packWARN(WARN_UTF8);
1423 message = Perl_form(aTHX_ "%s (empty string)",
1428 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1429 possible_problems &= ~UTF8_GOT_CONTINUATION;
1430 *errors |= UTF8_GOT_CONTINUATION;
1432 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1434 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1435 pack_warn = packWARN(WARN_UTF8);
1436 message = Perl_form(aTHX_
1437 "%s: %s (unexpected continuation byte 0x%02x,"
1438 " with no preceding start byte)",
1440 _byte_dump_string(s0, 1), *s0);
1444 else if (possible_problems & UTF8_GOT_SHORT) {
1445 possible_problems &= ~UTF8_GOT_SHORT;
1446 *errors |= UTF8_GOT_SHORT;
1448 if (! (flags & UTF8_ALLOW_SHORT)) {
1450 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1451 pack_warn = packWARN(WARN_UTF8);
1452 message = Perl_form(aTHX_
1453 "%s: %s (too short; %d byte%s available, need %d)",
1455 _byte_dump_string(s0, send - s0),
1457 avail_len == 1 ? "" : "s",
1463 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1464 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1465 *errors |= UTF8_GOT_NON_CONTINUATION;
1467 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1469 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1471 /* If we don't know for sure that the input length is
1472 * valid, avoid as much as possible reading past the
1473 * end of the buffer */
1474 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1477 pack_warn = packWARN(WARN_UTF8);
1478 message = Perl_form(aTHX_ "%s",
1479 unexpected_non_continuation_text(s0,
1486 else if (possible_problems & UTF8_GOT_LONG) {
1487 possible_problems &= ~UTF8_GOT_LONG;
1488 *errors |= UTF8_GOT_LONG;
1490 if (flags & UTF8_ALLOW_LONG) {
1492 /* We don't allow the actual overlong value, unless the
1493 * special extra bit is also set */
1494 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1495 & ~UTF8_ALLOW_LONG)))
1497 uv = UNICODE_REPLACEMENT;
1503 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1504 pack_warn = packWARN(WARN_UTF8);
1506 /* These error types cause 'uv' to be something that
1507 * isn't what was intended, so can't use it in the
1508 * message. The other error types either can't
1509 * generate an overlong, or else the 'uv' is valid */
1511 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1513 message = Perl_form(aTHX_
1514 "%s: %s (any UTF-8 sequence that starts"
1515 " with \"%s\" is overlong which can and"
1516 " should be represented with a"
1517 " different, shorter sequence)",
1519 _byte_dump_string(s0, send - s0),
1520 _byte_dump_string(s0, curlen));
1523 U8 tmpbuf[UTF8_MAXBYTES+1];
1524 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1526 message = Perl_form(aTHX_
1527 "%s: %s (overlong; instead use %s to represent"
1530 _byte_dump_string(s0, send - s0),
1531 _byte_dump_string(tmpbuf, e - tmpbuf),
1532 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1533 small code points */
1539 else if (possible_problems & UTF8_GOT_SURROGATE) {
1540 possible_problems &= ~UTF8_GOT_SURROGATE;
1542 if (flags & UTF8_WARN_SURROGATE) {
1543 *errors |= UTF8_GOT_SURROGATE;
1545 if ( ! (flags & UTF8_CHECK_ONLY)
1546 && ckWARN_d(WARN_SURROGATE))
1548 pack_warn = packWARN(WARN_SURROGATE);
1550 /* These are the only errors that can occur with a
1551 * surrogate when the 'uv' isn't valid */
1552 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1553 message = Perl_form(aTHX_
1554 "UTF-16 surrogate (any UTF-8 sequence that"
1555 " starts with \"%s\" is for a surrogate)",
1556 _byte_dump_string(s0, curlen));
1559 message = Perl_form(aTHX_
1560 "UTF-16 surrogate U+%04" UVXf, uv);
1565 if (flags & UTF8_DISALLOW_SURROGATE) {
1567 *errors |= UTF8_GOT_SURROGATE;
1570 else if (possible_problems & UTF8_GOT_SUPER) {
1571 possible_problems &= ~UTF8_GOT_SUPER;
1573 if (flags & UTF8_WARN_SUPER) {
1574 *errors |= UTF8_GOT_SUPER;
1576 if ( ! (flags & UTF8_CHECK_ONLY)
1577 && ckWARN_d(WARN_NON_UNICODE))
1579 pack_warn = packWARN(WARN_NON_UNICODE);
1581 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1582 message = Perl_form(aTHX_
1583 "Any UTF-8 sequence that starts with"
1584 " \"%s\" is for a non-Unicode code point,"
1585 " may not be portable",
1586 _byte_dump_string(s0, curlen));
1589 message = Perl_form(aTHX_
1590 "Code point 0x%04" UVXf " is not"
1591 " Unicode, may not be portable",
1597 /* The maximum code point ever specified by a standard was
1598 * 2**31 - 1. Anything larger than that is a Perl extension
1599 * that very well may not be understood by other applications
1600 * (including earlier perl versions on EBCDIC platforms). We
1601 * test for these after the regular SUPER ones, and before
1602 * possibly bailing out, so that the slightly more dire warning
1603 * will override the regular one. */
1604 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1606 |UTF8_DISALLOW_ABOVE_31_BIT))
1607 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1608 && UNLIKELY(is_utf8_cp_above_31_bits(
1611 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1612 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1614 if ( ! (flags & UTF8_CHECK_ONLY)
1615 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1616 && ckWARN_d(WARN_UTF8))
1618 pack_warn = packWARN(WARN_UTF8);
1620 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1621 message = Perl_form(aTHX_
1622 "Any UTF-8 sequence that starts with"
1623 " \"%s\" is for a non-Unicode code"
1624 " point, and is not portable",
1625 _byte_dump_string(s0, curlen));
1628 message = Perl_form(aTHX_
1629 "Code point 0x%" UVXf " is not Unicode,"
1630 " and not portable",
1635 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1636 |UTF8_DISALLOW_ABOVE_31_BIT))
1638 *errors |= UTF8_GOT_ABOVE_31_BIT;
1640 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1646 if (flags & UTF8_DISALLOW_SUPER) {
1647 *errors |= UTF8_GOT_SUPER;
1651 /* The deprecated warning overrides any non-deprecated one. If
1652 * there are other problems, a deprecation message is not
1653 * really helpful, so don't bother to raise it in that case.
1654 * This also keeps the code from having to handle the case
1655 * where 'uv' is not valid. */
1656 if ( ! (orig_problems
1657 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1658 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1659 && ckWARN_d(WARN_DEPRECATED))
1661 message = Perl_form(aTHX_ cp_above_legal_max,
1662 uv, MAX_NON_DEPRECATED_CP);
1663 pack_warn = packWARN(WARN_DEPRECATED);
1666 else if (possible_problems & UTF8_GOT_NONCHAR) {
1667 possible_problems &= ~UTF8_GOT_NONCHAR;
1669 if (flags & UTF8_WARN_NONCHAR) {
1670 *errors |= UTF8_GOT_NONCHAR;
1672 if ( ! (flags & UTF8_CHECK_ONLY)
1673 && ckWARN_d(WARN_NONCHAR))
1675 /* The code above should have guaranteed that we don't
1676 * get here with errors other than overlong */
1677 assert (! (orig_problems
1678 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1680 pack_warn = packWARN(WARN_NONCHAR);
1681 message = Perl_form(aTHX_ "Unicode non-character"
1682 " U+%04" UVXf " is not recommended"
1683 " for open interchange", uv);
1687 if (flags & UTF8_DISALLOW_NONCHAR) {
1689 *errors |= UTF8_GOT_NONCHAR;
1691 } /* End of looking through the possible flags */
1693 /* Display the message (if any) for the problem being handled in
1694 * this iteration of the loop */
1697 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1700 Perl_warner(aTHX_ pack_warn, "%s", message);
1702 } /* End of 'while (possible_problems)' */
1704 /* Since there was a possible problem, the returned length may need to
1705 * be changed from the one stored at the beginning of this function.
1706 * Instead of trying to figure out if that's needed, just do it. */
1712 if (flags & UTF8_CHECK_ONLY && retlen) {
1713 *retlen = ((STRLEN) -1);
1719 return UNI_TO_NATIVE(uv);
1723 =for apidoc utf8_to_uvchr_buf
1725 Returns the native code point of the first character in the string C<s> which
1726 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1727 C<*retlen> will be set to the length, in bytes, of that character.
1729 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1730 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1731 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1732 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1733 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1734 the next possible position in C<s> that could begin a non-malformed character.
1735 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1738 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1739 unless those are turned off.
1743 Also implemented as a macro in utf8.h
1749 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1751 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1755 return utf8n_to_uvchr(s, send - s, retlen,
1756 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1759 /* This is marked as deprecated
1761 =for apidoc utf8_to_uvuni_buf
1763 Only in very rare circumstances should code need to be dealing in Unicode
1764 (as opposed to native) code points. In those few cases, use
1765 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1767 Returns the Unicode (not-native) code point of the first character in the
1769 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1770 C<retlen> will be set to the length, in bytes, of that character.
1772 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1773 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1774 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1775 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1776 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1777 next possible position in C<s> that could begin a non-malformed character.
1778 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1780 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1781 unless those are turned off.
1787 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1789 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1793 /* Call the low level routine, asking for checks */
1794 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1798 =for apidoc utf8_length
1800 Return the length of the UTF-8 char encoded string C<s> in characters.
1801 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1802 up past C<e>, croaks.
1808 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1812 PERL_ARGS_ASSERT_UTF8_LENGTH;
1814 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1815 * the bitops (especially ~) can create illegal UTF-8.
1816 * In other words: in Perl UTF-8 is not just for Unicode. */
1819 goto warn_and_return;
1829 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1830 "%s in %s", unees, OP_DESC(PL_op));
1832 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1839 =for apidoc bytes_cmp_utf8
1841 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1842 sequence of characters (stored as UTF-8)
1843 in C<u>, C<ulen>. Returns 0 if they are
1844 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1845 if the first string is greater than the second string.
1847 -1 or +1 is returned if the shorter string was identical to the start of the
1848 longer string. -2 or +2 is returned if
1849 there was a difference between characters
1856 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1858 const U8 *const bend = b + blen;
1859 const U8 *const uend = u + ulen;
1861 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1863 while (b < bend && u < uend) {
1865 if (!UTF8_IS_INVARIANT(c)) {
1866 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1869 if (UTF8_IS_CONTINUATION(c1)) {
1870 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1872 /* diag_listed_as: Malformed UTF-8 character%s */
1873 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1875 unexpected_non_continuation_text(u - 1, 2, 1, 2),
1876 PL_op ? " in " : "",
1877 PL_op ? OP_DESC(PL_op) : "");
1882 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1883 "%s in %s", unees, OP_DESC(PL_op));
1885 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1886 return -2; /* Really want to return undef :-) */
1893 return *b < c ? -2 : +2;
1898 if (b == bend && u == uend)
1901 return b < bend ? +1 : -1;
1905 =for apidoc utf8_to_bytes
1907 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1908 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1909 updates C<len> to contain the new length.
1910 Returns zero on failure, setting C<len> to -1.
1912 If you need a copy of the string, see L</bytes_from_utf8>.
1918 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1920 U8 * const save = s;
1921 U8 * const send = s + *len;
1924 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1925 PERL_UNUSED_CONTEXT;
1927 /* ensure valid UTF-8 and chars < 256 before updating string */
1929 if (! UTF8_IS_INVARIANT(*s)) {
1930 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1931 *len = ((STRLEN) -1);
1942 if (! UTF8_IS_INVARIANT(c)) {
1943 /* Then it is two-byte encoded */
1944 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1955 =for apidoc bytes_from_utf8
1957 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1958 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1959 the newly-created string, and updates C<len> to contain the new
1960 length. Returns the original string if no conversion occurs, C<len>
1961 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1962 0 if C<s> is converted or consisted entirely of characters that are invariant
1963 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1969 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1972 const U8 *start = s;
1976 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1977 PERL_UNUSED_CONTEXT;
1981 /* ensure valid UTF-8 and chars < 256 before converting string */
1982 for (send = s + *len; s < send;) {
1983 if (! UTF8_IS_INVARIANT(*s)) {
1984 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1995 Newx(d, (*len) - count + 1, U8);
1996 s = start; start = d;
1999 if (! UTF8_IS_INVARIANT(c)) {
2000 /* Then it is two-byte encoded */
2001 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2012 =for apidoc bytes_to_utf8
2014 Converts a string C<s> of length C<len> bytes from the native encoding into
2016 Returns a pointer to the newly-created string, and sets C<len> to
2017 reflect the new length in bytes.
2019 A C<NUL> character will be written after the end of the string.
2021 If you want to convert to UTF-8 from encodings other than
2022 the native (Latin1 or EBCDIC),
2023 see L</sv_recode_to_utf8>().
2028 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
2029 likewise need duplication. */
2032 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
2034 const U8 * const send = s + (*len);
2038 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2039 PERL_UNUSED_CONTEXT;
2041 Newx(d, (*len) * 2 + 1, U8);
2045 append_utf8_from_native_byte(*s, &d);
2054 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2056 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2057 * We optimize for native, for obvious reasons. */
2060 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2065 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2068 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
2073 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2075 if (OFFUNI_IS_INVARIANT(uv)) {
2076 *d++ = LATIN1_TO_NATIVE((U8) uv);
2079 if (uv <= MAX_UTF8_TWO_BYTE) {
2080 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2081 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2084 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2085 #define LAST_HIGH_SURROGATE 0xDBFF
2086 #define FIRST_LOW_SURROGATE 0xDC00
2087 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2089 /* This assumes that most uses will be in the first Unicode plane, not
2090 * needing surrogates */
2091 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2092 && uv <= UNICODE_SURROGATE_LAST))
2094 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2095 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2098 UV low = (p[0] << 8) + p[1];
2099 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2100 || UNLIKELY(low > LAST_LOW_SURROGATE))
2102 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2105 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2106 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2110 d = uvoffuni_to_utf8_flags(d, uv, 0);
2113 *d++ = (U8)(( uv >> 12) | 0xe0);
2114 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2115 *d++ = (U8)(( uv & 0x3f) | 0x80);
2119 *d++ = (U8)(( uv >> 18) | 0xf0);
2120 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2121 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2122 *d++ = (U8)(( uv & 0x3f) | 0x80);
2127 *newlen = d - dstart;
2131 /* Note: this one is slightly destructive of the source. */
2134 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2137 U8* const send = s + bytelen;
2139 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2142 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2146 const U8 tmp = s[0];
2151 return utf16_to_utf8(p, d, bytelen, newlen);
2155 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2157 U8 tmpbuf[UTF8_MAXBYTES+1];
2158 uvchr_to_utf8(tmpbuf, c);
2159 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2162 /* Internal function so we can deprecate the external one, and call
2163 this one from other deprecated functions in this file */
2166 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2168 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2172 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2176 Perl__is_uni_perl_idcont(pTHX_ UV c)
2178 U8 tmpbuf[UTF8_MAXBYTES+1];
2179 uvchr_to_utf8(tmpbuf, c);
2180 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2184 Perl__is_uni_perl_idstart(pTHX_ UV c)
2186 U8 tmpbuf[UTF8_MAXBYTES+1];
2187 uvchr_to_utf8(tmpbuf, c);
2188 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2192 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2194 /* We have the latin1-range values compiled into the core, so just use
2195 * those, converting the result to UTF-8. The only difference between upper
2196 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2197 * either "SS" or "Ss". Which one to use is passed into the routine in
2198 * 'S_or_s' to avoid a test */
2200 UV converted = toUPPER_LATIN1_MOD(c);
2202 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2204 assert(S_or_s == 'S' || S_or_s == 's');
2206 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2207 characters in this range */
2208 *p = (U8) converted;
2213 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2214 * which it maps to one of them, so as to only have to have one check for
2215 * it in the main case */
2216 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2218 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2219 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2222 converted = GREEK_CAPITAL_LETTER_MU;
2224 #if UNICODE_MAJOR_VERSION > 2 \
2225 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2226 && UNICODE_DOT_DOT_VERSION >= 8)
2227 case LATIN_SMALL_LETTER_SHARP_S:
2234 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2235 NOT_REACHED; /* NOTREACHED */
2239 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2240 *p = UTF8_TWO_BYTE_LO(converted);
2246 /* Call the function to convert a UTF-8 encoded character to the specified case.
2247 * Note that there may be more than one character in the result.
2248 * INP is a pointer to the first byte of the input character
2249 * OUTP will be set to the first byte of the string of changed characters. It
2250 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2251 * LENP will be set to the length in bytes of the string of changed characters
2253 * The functions return the ordinal of the first character in the string of OUTP */
2254 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2255 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2256 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2258 /* This additionally has the input parameter 'specials', which if non-zero will
2259 * cause this to use the specials hash for folding (meaning get full case
2260 * folding); otherwise, when zero, this implies a simple case fold */
2261 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2264 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2266 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2267 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2268 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2269 * the changed version may be longer than the original character.
2271 * The ordinal of the first character of the changed version is returned
2272 * (but note, as explained above, that there may be more.) */
2274 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2277 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2280 uvchr_to_utf8(p, c);
2281 return CALL_UPPER_CASE(c, p, p, lenp);
2285 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2287 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2290 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2293 uvchr_to_utf8(p, c);
2294 return CALL_TITLE_CASE(c, p, p, lenp);
2298 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2300 /* We have the latin1-range values compiled into the core, so just use
2301 * those, converting the result to UTF-8. Since the result is always just
2302 * one character, we allow <p> to be NULL */
2304 U8 converted = toLOWER_LATIN1(c);
2306 PERL_UNUSED_ARG(dummy);
2309 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2314 /* Result is known to always be < 256, so can use the EIGHT_BIT
2316 *p = UTF8_EIGHT_BIT_HI(converted);
2317 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2325 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2327 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2330 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2333 uvchr_to_utf8(p, c);
2334 return CALL_LOWER_CASE(c, p, p, lenp);
2338 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2340 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2341 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2342 * FOLD_FLAGS_FULL iff full folding is to be used;
2344 * Not to be used for locale folds
2349 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2350 PERL_UNUSED_CONTEXT;
2352 assert (! (flags & FOLD_FLAGS_LOCALE));
2354 if (UNLIKELY(c == MICRO_SIGN)) {
2355 converted = GREEK_SMALL_LETTER_MU;
2357 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2358 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2359 || UNICODE_DOT_DOT_VERSION > 0)
2360 else if ( (flags & FOLD_FLAGS_FULL)
2361 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2363 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2364 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2365 * under those circumstances. */
2366 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2367 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2368 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2370 return LATIN_SMALL_LETTER_LONG_S;
2380 else { /* In this range the fold of all other characters is their lower
2382 converted = toLOWER_LATIN1(c);
2385 if (UVCHR_IS_INVARIANT(converted)) {
2386 *p = (U8) converted;
2390 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2391 *p = UTF8_TWO_BYTE_LO(converted);
2399 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2402 /* Not currently externally documented, and subject to change
2403 * <flags> bits meanings:
2404 * FOLD_FLAGS_FULL iff full folding is to be used;
2405 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2406 * locale are to be used.
2407 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2410 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2412 if (flags & FOLD_FLAGS_LOCALE) {
2413 /* Treat a UTF-8 locale as not being in locale at all */
2414 if (IN_UTF8_CTYPE_LOCALE) {
2415 flags &= ~FOLD_FLAGS_LOCALE;
2418 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2419 goto needs_full_generality;
2424 return _to_fold_latin1((U8) c, p, lenp,
2425 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2428 /* Here, above 255. If no special needs, just use the macro */
2429 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2430 uvchr_to_utf8(p, c);
2431 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2433 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2434 the special flags. */
2435 U8 utf8_c[UTF8_MAXBYTES + 1];
2437 needs_full_generality:
2438 uvchr_to_utf8(utf8_c, c);
2439 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c), p, lenp, flags);
2443 PERL_STATIC_INLINE bool
2444 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2445 const char *const swashname, SV* const invlist)
2447 /* returns a boolean giving whether or not the UTF8-encoded character that
2448 * starts at <p> is in the swash indicated by <swashname>. <swash>
2449 * contains a pointer to where the swash indicated by <swashname>
2450 * is to be stored; which this routine will do, so that future calls will
2451 * look at <*swash> and only generate a swash if it is not null. <invlist>
2452 * is NULL or an inversion list that defines the swash. If not null, it
2453 * saves time during initialization of the swash.
2455 * Note that it is assumed that the buffer length of <p> is enough to
2456 * contain all the bytes that comprise the character. Thus, <*p> should
2457 * have been checked before this call for mal-formedness enough to assure
2460 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2462 /* The API should have included a length for the UTF-8 character in <p>,
2463 * but it doesn't. We therefore assume that p has been validated at least
2464 * as far as there being enough bytes available in it to accommodate the
2465 * character without reading beyond the end, and pass that number on to the
2466 * validating routine */
2467 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2468 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2469 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2471 NOT_REACHED; /* NOTREACHED */
2475 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2476 *swash = _core_swash_init("utf8",
2478 /* Only use the name if there is no inversion
2479 * list; otherwise will go out to disk */
2480 (invlist) ? "" : swashname,
2482 &PL_sv_undef, 1, 0, invlist, &flags);
2485 return swash_fetch(*swash, p, TRUE) != 0;
2488 PERL_STATIC_INLINE bool
2489 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2490 const char *const swashname, SV* const invlist)
2492 /* returns a boolean giving whether or not the UTF8-encoded character that
2493 * starts at <p>, and extending no further than <e - 1> is in the swash
2494 * indicated by <swashname>. <swash> contains a pointer to where the swash
2495 * indicated by <swashname> is to be stored; which this routine will do, so
2496 * that future calls will look at <*swash> and only generate a swash if it
2497 * is not null. <invlist> is NULL or an inversion list that defines the
2498 * swash. If not null, it saves time during initialization of the swash.
2501 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2503 if (! isUTF8_CHAR(p, e)) {
2504 _force_out_malformed_utf8_message(p, e, 0, 1);
2505 NOT_REACHED; /* NOTREACHED */
2509 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2510 *swash = _core_swash_init("utf8",
2512 /* Only use the name if there is no inversion
2513 * list; otherwise will go out to disk */
2514 (invlist) ? "" : swashname,
2516 &PL_sv_undef, 1, 0, invlist, &flags);
2519 return swash_fetch(*swash, p, TRUE) != 0;
2523 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2524 const char * const alternative,
2525 const bool use_locale,
2526 const char * const file,
2527 const unsigned line)
2531 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2533 if (ckWARN_d(WARN_DEPRECATED)) {
2535 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2536 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2537 if (! PL_seen_deprecated_macro) {
2538 PL_seen_deprecated_macro = newHV();
2540 if (! hv_store(PL_seen_deprecated_macro, key,
2541 strlen(key), &PL_sv_undef, 0))
2543 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2546 if (instr(file, "mathoms.c")) {
2547 Perl_warner(aTHX_ WARN_DEPRECATED,
2548 "In %s, line %d, starting in Perl v5.30, %s()"
2549 " will be removed. Avoid this message by"
2550 " converting to use %s().\n",
2551 file, line, name, alternative);
2554 Perl_warner(aTHX_ WARN_DEPRECATED,
2555 "In %s, line %d, starting in Perl v5.30, %s() will"
2556 " require an additional parameter. Avoid this"
2557 " message by converting to use %s().\n",
2558 file, line, name, alternative);
2565 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2567 const char * const name,
2568 const char * const alternative,
2569 const bool use_utf8,
2570 const bool use_locale,
2571 const char * const file,
2572 const unsigned line)
2574 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2576 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2578 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2588 case _CC_ALPHANUMERIC:
2592 return is_utf8_common(p,
2593 &PL_utf8_swash_ptrs[classnum],
2594 swash_property_names[classnum],
2595 PL_XPosix_ptrs[classnum]);
2598 return is_XPERLSPACE_high(p);
2600 return is_HORIZWS_high(p);
2602 return is_XDIGIT_high(p);
2608 return is_VERTWS_high(p);
2610 if (! PL_utf8_perl_idstart) {
2611 PL_utf8_perl_idstart
2612 = _new_invlist_C_array(_Perl_IDStart_invlist);
2614 return is_utf8_common(p, &PL_utf8_perl_idstart,
2615 "_Perl_IDStart", NULL);
2617 if (! PL_utf8_perl_idcont) {
2619 = _new_invlist_C_array(_Perl_IDCont_invlist);
2621 return is_utf8_common(p, &PL_utf8_perl_idcont,
2622 "_Perl_IDCont", NULL);
2626 /* idcont is the same as wordchar below 256 */
2627 if (classnum == _CC_IDCONT) {
2628 classnum = _CC_WORDCHAR;
2630 else if (classnum == _CC_IDFIRST) {
2634 classnum = _CC_ALPHA;
2638 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2639 return _generic_isCC(*p, classnum);
2642 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2645 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2646 return isFOO_lc(classnum, *p);
2649 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2652 NOT_REACHED; /* NOTREACHED */
2656 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2659 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2661 assert(classnum < _FIRST_NON_SWASH_CC);
2663 return is_utf8_common_with_len(p,
2665 &PL_utf8_swash_ptrs[classnum],
2666 swash_property_names[classnum],
2667 PL_XPosix_ptrs[classnum]);
2671 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2675 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2677 if (! PL_utf8_perl_idstart) {
2678 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2680 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2681 "_Perl_IDStart", invlist);
2685 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2687 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2691 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2695 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2699 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2701 if (! PL_utf8_perl_idcont) {
2702 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2704 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2705 "_Perl_IDCont", invlist);
2709 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2711 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2713 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2717 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2719 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2721 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2725 Perl__is_utf8_mark(pTHX_ const U8 *p)
2727 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2729 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2733 =for apidoc to_utf8_case
2735 Instead use the appropriate one of L</toUPPER_utf8_safe>,
2736 L</toTITLE_utf8_safe>,
2737 L</toLOWER_utf8_safe>,
2738 or L</toFOLD_utf8_safe>.
2740 This function will be removed in Perl v5.28.
2742 C<p> contains the pointer to the UTF-8 string encoding
2743 the character that is being converted. This routine assumes that the character
2744 at C<p> is well-formed.
2746 C<ustrp> is a pointer to the character buffer to put the
2747 conversion result to. C<lenp> is a pointer to the length
2750 C<swashp> is a pointer to the swash to use.
2752 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2753 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
2754 but not always, a multicharacter mapping), is tried first.
2756 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2757 any special mappings; C<""> means to use the special mappings. Values other
2758 than these two are treated as the name of the hash containing the special
2759 mappings, like C<"utf8::ToSpecLower">.
2761 C<normal> is a string like C<"ToLower"> which means the swash
2764 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2765 unless those are turned off.
2770 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2771 SV **swashp, const char *normal, const char *special)
2775 const U8 * e = p + UTF8SKIP(p);
2777 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2779 cp = utf8n_to_uvchr(p, e - p, &len_cp, UTF8_CHECK_ONLY);
2780 if (len_cp == (STRLEN) -1) {
2781 _force_out_malformed_utf8_message(p, e,
2782 _UTF8_NO_CONFIDENCE_IN_CURLEN, 1 /* Die */ );
2785 return _to_utf8_case(cp, p, ustrp, lenp, swashp, normal, special);
2788 /* change namve uv1 to 'from' */
2790 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2791 SV **swashp, const char *normal, const char *special)
2795 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2797 /* For code points that don't change case, we already know that the output
2798 * of this function is the unchanged input, so we can skip doing look-ups
2799 * for them. Unfortunately the case-changing code points are scattered
2800 * around. But there are some long consecutive ranges where there are no
2801 * case changing code points. By adding tests, we can eliminate the lookup
2802 * for all the ones in such ranges. This is currently done here only for
2803 * just a few cases where the scripts are in common use in modern commerce
2804 * (and scripts adjacent to those which can be included without additional
2807 if (uv1 >= 0x0590) {
2808 /* This keeps from needing further processing the code points most
2809 * likely to be used in the following non-cased scripts: Hebrew,
2810 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2811 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2812 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2817 /* The following largish code point ranges also don't have case
2818 * changes, but khw didn't think they warranted extra tests to speed
2819 * them up (which would slightly slow down everything else above them):
2820 * 1100..139F Hangul Jamo, Ethiopic
2821 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2822 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2823 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2824 * Combining Diacritical Marks Extended, Balinese,
2825 * Sundanese, Batak, Lepcha, Ol Chiki
2826 * 2000..206F General Punctuation
2829 if (uv1 >= 0x2D30) {
2831 /* This keeps the from needing further processing the code points
2832 * most likely to be used in the following non-cased major scripts:
2833 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2835 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2836 * event that Unicode eventually allocates the unused block as of
2837 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2838 * that the test suite will start having failures to alert you
2839 * should that happen) */
2844 if (uv1 >= 0xAC00) {
2845 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2846 if (ckWARN_d(WARN_SURROGATE)) {
2847 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2848 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2849 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2854 /* AC00..FAFF Catches Hangul syllables and private use, plus
2861 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2862 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2863 && ckWARN_d(WARN_DEPRECATED))
2865 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2866 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2868 if (ckWARN_d(WARN_NON_UNICODE)) {
2869 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2870 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2871 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2875 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2877 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2880 /* As of this writing, this means we avoid swash creation
2881 * for anything beyond low Plane 1 */
2888 /* Note that non-characters are perfectly legal, so no warning should
2889 * be given. There are so few of them, that it isn't worth the extra
2890 * tests to avoid swash creation */
2893 if (!*swashp) /* load on-demand */
2894 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2897 /* It might be "special" (sometimes, but not always,
2898 * a multicharacter mapping) */
2902 /* If passed in the specials name, use that; otherwise use any
2903 * given in the swash */
2904 if (*special != '\0') {
2905 hv = get_hv(special, 0);
2908 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2910 hv = MUTABLE_HV(SvRV(*svp));
2915 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2920 s = SvPV_const(*svp, len);
2923 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2925 Copy(s, ustrp, len, U8);
2930 if (!len && *swashp) {
2931 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2934 /* It was "normal" (a single character mapping). */
2935 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2943 return valid_utf8_to_uvchr(ustrp, 0);
2946 /* Here, there was no mapping defined, which means that the code point maps
2947 * to itself. Return the inputs */
2950 if (p != ustrp) { /* Don't copy onto itself */
2951 Copy(p, ustrp, len, U8);
2962 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2964 /* This is called when changing the case of a UTF-8-encoded character above
2965 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2966 * result contains a character that crosses the 255/256 boundary, disallow
2967 * the change, and return the original code point. See L<perlfunc/lc> for
2970 * p points to the original string whose case was changed; assumed
2971 * by this routine to be well-formed
2972 * result the code point of the first character in the changed-case string
2973 * ustrp points to the changed-case string (<result> represents its first char)
2974 * lenp points to the length of <ustrp> */
2976 UV original; /* To store the first code point of <p> */
2978 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2980 assert(UTF8_IS_ABOVE_LATIN1(*p));
2982 /* We know immediately if the first character in the string crosses the
2983 * boundary, so can skip */
2986 /* Look at every character in the result; if any cross the
2987 * boundary, the whole thing is disallowed */
2988 U8* s = ustrp + UTF8SKIP(ustrp);
2989 U8* e = ustrp + *lenp;
2991 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2997 /* Here, no characters crossed, result is ok as-is, but we warn. */
2998 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3004 /* Failed, have to return the original */
3005 original = valid_utf8_to_uvchr(p, lenp);
3007 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3008 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3009 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
3010 "resolved to \"\\x{%" UVXf "}\".",
3014 Copy(p, ustrp, *lenp, char);
3019 S_check_and_deprecate(pTHX_ const U8 *p,
3021 const unsigned int type, /* See below */
3022 const bool use_locale, /* Is this a 'LC_'
3024 const char * const file,
3025 const unsigned line)
3027 /* This is a temporary function to deprecate the unsafe calls to the case
3028 * changing macros and functions. It keeps all the special stuff in just
3031 * It updates *e with the pointer to the end of the input string. If using
3032 * the old-style macros, *e is NULL on input, and so this function assumes
3033 * the input string is long enough to hold the entire UTF-8 sequence, and
3034 * sets *e accordingly, but it then returns a flag to pass the
3035 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3036 * using the full length if possible.
3038 * It also does the assert that *e > p when *e is not NULL. This should be
3039 * migrated to the callers when this function gets deleted.
3041 * The 'type' parameter is used for the caller to specify which case
3042 * changing function this is called from: */
3044 # define DEPRECATE_TO_UPPER 0
3045 # define DEPRECATE_TO_TITLE 1
3046 # define DEPRECATE_TO_LOWER 2
3047 # define DEPRECATE_TO_FOLD 3
3049 U32 utf8n_flags = 0;
3051 const char * alternative;
3053 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3056 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3057 *e = p + UTF8SKIP(p);
3059 /* For mathoms.c calls, we use the function name we know is stored
3060 * there. It could be part of a larger path */
3061 if (type == DEPRECATE_TO_UPPER) {
3062 name = instr(file, "mathoms.c")
3065 alternative = "toUPPER_utf8_safe";
3067 else if (type == DEPRECATE_TO_TITLE) {
3068 name = instr(file, "mathoms.c")
3071 alternative = "toTITLE_utf8_safe";
3073 else if (type == DEPRECATE_TO_LOWER) {
3074 name = instr(file, "mathoms.c")
3077 alternative = "toLOWER_utf8_safe";
3079 else if (type == DEPRECATE_TO_FOLD) {
3080 name = instr(file, "mathoms.c")
3083 alternative = "toFOLD_utf8_safe";
3085 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3087 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3096 /* The process for changing the case is essentially the same for the four case
3097 * change types, except there are complications for folding. Otherwise the
3098 * difference is only which case to change to. To make sure that they all do
3099 * the same thing, the bodies of the functions are extracted out into the
3100 * following two macros. The functions are written with the same variable
3101 * names, and these are known and used inside these macros. It would be
3102 * better, of course, to have inline functions to do it, but since different
3103 * macros are called, depending on which case is being changed to, this is not
3104 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3105 * function can start with the common start macro, then finish with its special
3106 * handling; while the other three cases can just use the common end macro.
3108 * The algorithm is to use the proper (passed in) macro or function to change
3109 * the case for code points that are below 256. The macro is used if using
3110 * locale rules for the case change; the function if not. If the code point is
3111 * above 255, it is computed from the input UTF-8, and another macro is called
3112 * to do the conversion. If necessary, the output is converted to UTF-8. If
3113 * using a locale, we have to check that the change did not cross the 255/256
3114 * boundary, see check_locale_boundary_crossing() for further details.
3116 * The macros are split with the correct case change for the below-256 case
3117 * stored into 'result', and in the middle of an else clause for the above-255
3118 * case. At that point in the 'else', 'result' is not the final result, but is
3119 * the input code point calculated from the UTF-8. The fold code needs to
3120 * realize all this and take it from there.
3122 * If you read the two macros as sequential, it's easier to understand what's
3124 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3125 L1_func_extra_param) \
3127 if (flags & (locale_flags)) { \
3128 /* Treat a UTF-8 locale as not being in locale at all */ \
3129 if (IN_UTF8_CTYPE_LOCALE) { \
3130 flags &= ~(locale_flags); \
3133 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3137 if (UTF8_IS_INVARIANT(*p)) { \
3138 if (flags & (locale_flags)) { \
3139 result = LC_L1_change_macro(*p); \
3142 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3145 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3146 if (flags & (locale_flags)) { \
3147 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3151 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3152 ustrp, lenp, L1_func_extra_param); \
3155 else { /* malformed UTF-8 or ord above 255 */ \
3156 STRLEN len_result; \
3157 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3158 if (len_result == (STRLEN) -1) { \
3159 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3163 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3164 result = change_macro(result, p, ustrp, lenp); \
3166 if (flags & (locale_flags)) { \
3167 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3172 /* Here, used locale rules. Convert back to UTF-8 */ \
3173 if (UTF8_IS_INVARIANT(result)) { \
3174 *ustrp = (U8) result; \
3178 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3179 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3186 =for apidoc to_utf8_upper
3188 Instead use L</toUPPER_utf8_safe>.
3192 /* Not currently externally documented, and subject to change:
3193 * <flags> is set iff iff the rules from the current underlying locale are to
3197 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3202 const char * const file,
3206 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3207 cBOOL(flags), file, line);
3209 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3211 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3212 /* 2nd char of uc(U+DF) is 'S' */
3213 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3214 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3218 =for apidoc to_utf8_title
3220 Instead use L</toTITLE_utf8_safe>.
3224 /* Not currently externally documented, and subject to change:
3225 * <flags> is set iff the rules from the current underlying locale are to be
3226 * used. Since titlecase is not defined in POSIX, for other than a
3227 * UTF-8 locale, uppercase is used instead for code points < 256.
3231 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3236 const char * const file,
3240 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3241 cBOOL(flags), file, line);
3243 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3245 /* 2nd char of ucfirst(U+DF) is 's' */
3246 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3247 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3251 =for apidoc to_utf8_lower
3253 Instead use L</toLOWER_utf8_safe>.
3257 /* Not currently externally documented, and subject to change:
3258 * <flags> is set iff iff the rules from the current underlying locale are to
3263 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3268 const char * const file,
3272 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3273 cBOOL(flags), file, line);
3275 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3277 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3278 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3282 =for apidoc to_utf8_fold
3284 Instead use L</toFOLD_utf8_safe>.
3288 /* Not currently externally documented, and subject to change,
3290 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3291 * locale are to be used.
3292 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3293 * otherwise simple folds
3294 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3299 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3304 const char * const file,
3308 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3309 cBOOL(flags), file, line);
3311 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3313 /* These are mutually exclusive */
3314 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3316 assert(p != ustrp); /* Otherwise overwrites */
3318 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3319 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3321 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3323 if (flags & FOLD_FLAGS_LOCALE) {
3325 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3326 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3328 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3329 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3331 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3333 /* Special case these two characters, as what normally gets
3334 * returned under locale doesn't work */
3335 if (UTF8SKIP(p) == cap_sharp_s_len
3336 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3338 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3339 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3340 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3341 "resolved to \"\\x{17F}\\x{17F}\".");
3346 if (UTF8SKIP(p) == long_s_t_len
3347 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3349 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3350 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3351 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3352 "resolved to \"\\x{FB06}\".");
3353 goto return_ligature_st;
3356 #if UNICODE_MAJOR_VERSION == 3 \
3357 && UNICODE_DOT_VERSION == 0 \
3358 && UNICODE_DOT_DOT_VERSION == 1
3359 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3361 /* And special case this on this Unicode version only, for the same
3362 * reaons the other two are special cased. They would cross the
3363 * 255/256 boundary which is forbidden under /l, and so the code
3364 * wouldn't catch that they are equivalent (which they are only in
3366 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3367 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3369 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3370 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3371 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3372 "resolved to \"\\x{0131}\".");
3373 goto return_dotless_i;
3377 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3379 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3383 /* This is called when changing the case of a UTF-8-encoded
3384 * character above the ASCII range, and the result should not
3385 * contain an ASCII character. */
3387 UV original; /* To store the first code point of <p> */
3389 /* Look at every character in the result; if any cross the
3390 * boundary, the whole thing is disallowed */
3392 U8* e = ustrp + *lenp;
3395 /* Crossed, have to return the original */
3396 original = valid_utf8_to_uvchr(p, lenp);
3398 /* But in these instances, there is an alternative we can
3399 * return that is valid */
3400 if (original == LATIN_SMALL_LETTER_SHARP_S
3401 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3402 || original == LATIN_CAPITAL_LETTER_SHARP_S
3407 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3408 goto return_ligature_st;
3410 #if UNICODE_MAJOR_VERSION == 3 \
3411 && UNICODE_DOT_VERSION == 0 \
3412 && UNICODE_DOT_DOT_VERSION == 1
3414 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3415 goto return_dotless_i;
3418 Copy(p, ustrp, *lenp, char);
3424 /* Here, no characters crossed, result is ok as-is */
3429 /* Here, used locale rules. Convert back to UTF-8 */
3430 if (UTF8_IS_INVARIANT(result)) {
3431 *ustrp = (U8) result;
3435 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3436 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3443 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3444 * folds to a string of two of these characters. By returning this
3445 * instead, then, e.g.,
3446 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3449 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3450 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3452 return LATIN_SMALL_LETTER_LONG_S;
3455 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3456 * have the other one fold to it */
3458 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3459 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3460 return LATIN_SMALL_LIGATURE_ST;
3462 #if UNICODE_MAJOR_VERSION == 3 \
3463 && UNICODE_DOT_VERSION == 0 \
3464 && UNICODE_DOT_DOT_VERSION == 1
3467 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3468 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3469 return LATIN_SMALL_LETTER_DOTLESS_I;
3476 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3477 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3478 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3482 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3484 PERL_ARGS_ASSERT_SWASH_INIT;
3486 /* Returns a copy of a swash initiated by the called function. This is the
3487 * public interface, and returning a copy prevents others from doing
3488 * mischief on the original */
3490 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3494 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3497 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3498 * use the following define */
3500 #define CORE_SWASH_INIT_RETURN(x) \
3501 PL_curpm= old_PL_curpm; \
3504 /* Initialize and return a swash, creating it if necessary. It does this
3505 * by calling utf8_heavy.pl in the general case. The returned value may be
3506 * the swash's inversion list instead if the input parameters allow it.
3507 * Which is returned should be immaterial to callers, as the only
3508 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3509 * and swash_to_invlist() handle both these transparently.
3511 * This interface should only be used by functions that won't destroy or
3512 * adversely change the swash, as doing so affects all other uses of the
3513 * swash in the program; the general public should use 'Perl_swash_init'
3516 * pkg is the name of the package that <name> should be in.
3517 * name is the name of the swash to find. Typically it is a Unicode
3518 * property name, including user-defined ones
3519 * listsv is a string to initialize the swash with. It must be of the form
3520 * documented as the subroutine return value in
3521 * L<perlunicode/User-Defined Character Properties>
3522 * minbits is the number of bits required to represent each data element.
3523 * It is '1' for binary properties.
3524 * none I (khw) do not understand this one, but it is used only in tr///.
3525 * invlist is an inversion list to initialize the swash with (or NULL)
3526 * flags_p if non-NULL is the address of various input and output flag bits
3527 * to the routine, as follows: ('I' means is input to the routine;
3528 * 'O' means output from the routine. Only flags marked O are
3529 * meaningful on return.)
3530 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3531 * came from a user-defined property. (I O)
3532 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3533 * when the swash cannot be located, to simply return NULL. (I)
3534 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3535 * return of an inversion list instead of a swash hash if this routine
3536 * thinks that would result in faster execution of swash_fetch() later
3539 * Thus there are three possible inputs to find the swash: <name>,
3540 * <listsv>, and <invlist>. At least one must be specified. The result
3541 * will be the union of the specified ones, although <listsv>'s various
3542 * actions can intersect, etc. what <name> gives. To avoid going out to
3543 * disk at all, <invlist> should specify completely what the swash should
3544 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3546 * <invlist> is only valid for binary properties */
3548 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3550 SV* retval = &PL_sv_undef;
3551 HV* swash_hv = NULL;
3552 const int invlist_swash_boundary =
3553 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3554 ? 512 /* Based on some benchmarking, but not extensive, see commit
3556 : -1; /* Never return just an inversion list */
3558 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3559 assert(! invlist || minbits == 1);
3561 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3562 that triggered the swash init and the swash init perl logic itself.
3565 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3567 if (listsv != &PL_sv_undef || strNE(name, "")) {
3569 const size_t pkg_len = strlen(pkg);
3570 const size_t name_len = strlen(name);
3571 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3575 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3577 PUSHSTACKi(PERLSI_MAGIC);
3581 /* We might get here via a subroutine signature which uses a utf8
3582 * parameter name, at which point PL_subname will have been set
3583 * but not yet used. */
3584 save_item(PL_subname);
3585 if (PL_parser && PL_parser->error_count)
3586 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3587 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3588 if (!method) { /* demand load UTF-8 */
3590 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3591 GvSV(PL_errgv) = NULL;
3592 #ifndef NO_TAINT_SUPPORT
3593 /* It is assumed that callers of this routine are not passing in
3594 * any user derived data. */
3595 /* Need to do this after save_re_context() as it will set
3596 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3597 * in Perl_magic_get). Even line to create errsv_save can turn on
3599 SAVEBOOL(TAINT_get);
3602 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3605 /* Not ERRSV, as there is no need to vivify a scalar we are
3606 about to discard. */
3607 SV * const errsv = GvSV(PL_errgv);
3608 if (!SvTRUE(errsv)) {
3609 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3610 SvREFCNT_dec(errsv);
3618 mPUSHp(pkg, pkg_len);
3619 mPUSHp(name, name_len);
3624 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3625 GvSV(PL_errgv) = NULL;
3626 /* If we already have a pointer to the method, no need to use
3627 * call_method() to repeat the lookup. */
3629 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3630 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3632 retval = *PL_stack_sp--;
3633 SvREFCNT_inc(retval);
3636 /* Not ERRSV. See above. */
3637 SV * const errsv = GvSV(PL_errgv);
3638 if (!SvTRUE(errsv)) {
3639 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3640 SvREFCNT_dec(errsv);
3645 if (IN_PERL_COMPILETIME) {
3646 CopHINTS_set(PL_curcop, PL_hints);
3648 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3649 if (SvPOK(retval)) {
3651 /* If caller wants to handle missing properties, let them */
3652 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3653 CORE_SWASH_INIT_RETURN(NULL);
3656 "Can't find Unicode property definition \"%" SVf "\"",
3658 NOT_REACHED; /* NOTREACHED */
3661 } /* End of calling the module to find the swash */
3663 /* If this operation fetched a swash, and we will need it later, get it */
3664 if (retval != &PL_sv_undef
3665 && (minbits == 1 || (flags_p
3667 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3669 swash_hv = MUTABLE_HV(SvRV(retval));
3671 /* If we don't already know that there is a user-defined component to
3672 * this swash, and the user has indicated they wish to know if there is
3673 * one (by passing <flags_p>), find out */
3674 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3675 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3676 if (user_defined && SvUV(*user_defined)) {
3677 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3682 /* Make sure there is an inversion list for binary properties */
3684 SV** swash_invlistsvp = NULL;
3685 SV* swash_invlist = NULL;
3686 bool invlist_in_swash_is_valid = FALSE;
3687 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3688 an unclaimed reference count */
3690 /* If this operation fetched a swash, get its already existing
3691 * inversion list, or create one for it */
3694 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3695 if (swash_invlistsvp) {
3696 swash_invlist = *swash_invlistsvp;
3697 invlist_in_swash_is_valid = TRUE;
3700 swash_invlist = _swash_to_invlist(retval);
3701 swash_invlist_unclaimed = TRUE;
3705 /* If an inversion list was passed in, have to include it */
3708 /* Any fetched swash will by now have an inversion list in it;
3709 * otherwise <swash_invlist> will be NULL, indicating that we
3710 * didn't fetch a swash */
3711 if (swash_invlist) {
3713 /* Add the passed-in inversion list, which invalidates the one
3714 * already stored in the swash */
3715 invlist_in_swash_is_valid = FALSE;
3716 SvREADONLY_off(swash_invlist); /* Turned on again below */
3717 _invlist_union(invlist, swash_invlist, &swash_invlist);
3721 /* Here, there is no swash already. Set up a minimal one, if
3722 * we are going to return a swash */
3723 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3725 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3727 swash_invlist = invlist;
3731 /* Here, we have computed the union of all the passed-in data. It may
3732 * be that there was an inversion list in the swash which didn't get
3733 * touched; otherwise save the computed one */
3734 if (! invlist_in_swash_is_valid
3735 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3737 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3739 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3741 /* We just stole a reference count. */
3742 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3743 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3746 /* The result is immutable. Forbid attempts to change it. */
3747 SvREADONLY_on(swash_invlist);
3749 /* Use the inversion list stand-alone if small enough */
3750 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3751 SvREFCNT_dec(retval);
3752 if (!swash_invlist_unclaimed)
3753 SvREFCNT_inc_simple_void_NN(swash_invlist);
3754 retval = newRV_noinc(swash_invlist);
3758 CORE_SWASH_INIT_RETURN(retval);
3759 #undef CORE_SWASH_INIT_RETURN
3763 /* This API is wrong for special case conversions since we may need to
3764 * return several Unicode characters for a single Unicode character
3765 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3766 * the lower-level routine, and it is similarly broken for returning
3767 * multiple values. --jhi
3768 * For those, you should use S__to_utf8_case() instead */
3769 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3772 * Returns the value of property/mapping C<swash> for the first character
3773 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3774 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3775 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3777 * A "swash" is a hash which contains initially the keys/values set up by
3778 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3779 * property for all possible code points. Things are stored in a compact form
3780 * (see utf8_heavy.pl) so that calculation is required to find the actual
3781 * property value for a given code point. As code points are looked up, new
3782 * key/value pairs are added to the hash, so that the calculation doesn't have
3783 * to ever be re-done. Further, each calculation is done, not just for the
3784 * desired one, but for a whole block of code points adjacent to that one.
3785 * For binary properties on ASCII machines, the block is usually for 64 code
3786 * points, starting with a code point evenly divisible by 64. Thus if the
3787 * property value for code point 257 is requested, the code goes out and
3788 * calculates the property values for all 64 code points between 256 and 319,
3789 * and stores these as a single 64-bit long bit vector, called a "swatch",
3790 * under the key for code point 256. The key is the UTF-8 encoding for code
3791 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3792 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3793 * for code point 258 is then requested, this code realizes that it would be
3794 * stored under the key for 256, and would find that value and extract the
3795 * relevant bit, offset from 256.
3797 * Non-binary properties are stored in as many bits as necessary to represent
3798 * their values (32 currently, though the code is more general than that), not
3799 * as single bits, but the principle is the same: the value for each key is a
3800 * vector that encompasses the property values for all code points whose UTF-8
3801 * representations are represented by the key. That is, for all code points
3802 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3806 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3808 HV *const hv = MUTABLE_HV(SvRV(swash));
3813 const U8 *tmps = NULL;
3817 PERL_ARGS_ASSERT_SWASH_FETCH;
3819 /* If it really isn't a hash, it isn't really swash; must be an inversion
3821 if (SvTYPE(hv) != SVt_PVHV) {
3822 return _invlist_contains_cp((SV*)hv,
3824 ? valid_utf8_to_uvchr(ptr, NULL)
3828 /* We store the values in a "swatch" which is a vec() value in a swash
3829 * hash. Code points 0-255 are a single vec() stored with key length
3830 * (klen) 0. All other code points have a UTF-8 representation
3831 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3832 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3833 * length for them is the length of the encoded char - 1. ptr[klen] is the
3834 * final byte in the sequence representing the character */
3835 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3840 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3843 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3846 klen = UTF8SKIP(ptr) - 1;
3848 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3849 * the vec is the final byte in the sequence. (In EBCDIC this is
3850 * converted to I8 to get consecutive values.) To help you visualize
3852 * Straight 1047 After final byte
3853 * UTF-8 UTF-EBCDIC I8 transform
3854 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3855 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3857 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3858 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3860 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3861 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3863 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3864 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3866 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3867 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3869 * (There are no discontinuities in the elided (...) entries.)
3870 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3871 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3872 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3873 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3874 * index into the vec() swatch (after subtracting 0x80, which we
3875 * actually do with an '&').
3876 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3877 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3878 * dicontinuities which go away by transforming it into I8, and we
3879 * effectively subtract 0xA0 to get the index. */
3880 needents = (1 << UTF_ACCUMULATION_SHIFT);
3881 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3885 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3886 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3887 * it's nothing to sniff at.) Pity we usually come through at least
3888 * two function calls to get here...
3890 * NB: this code assumes that swatches are never modified, once generated!
3893 if (hv == PL_last_swash_hv &&
3894 klen == PL_last_swash_klen &&
3895 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3897 tmps = PL_last_swash_tmps;
3898 slen = PL_last_swash_slen;
3901 /* Try our second-level swatch cache, kept in a hash. */
3902 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3904 /* If not cached, generate it via swatch_get */
3905 if (!svp || !SvPOK(*svp)
3906 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3909 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3910 swatch = swatch_get(swash,
3911 code_point & ~((UV)needents - 1),
3914 else { /* For the first 256 code points, the swatch has a key of
3916 swatch = swatch_get(swash, 0, needents);
3919 if (IN_PERL_COMPILETIME)
3920 CopHINTS_set(PL_curcop, PL_hints);
3922 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3924 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3925 || (slen << 3) < needents)
3926 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3927 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3928 svp, tmps, (UV)slen, (UV)needents);
3931 PL_last_swash_hv = hv;
3932 assert(klen <= sizeof(PL_last_swash_key));
3933 PL_last_swash_klen = (U8)klen;
3934 /* FIXME change interpvar.h? */
3935 PL_last_swash_tmps = (U8 *) tmps;
3936 PL_last_swash_slen = slen;
3938 Copy(ptr, PL_last_swash_key, klen, U8);
3941 switch ((int)((slen << 3) / needents)) {
3943 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3945 return ((UV) tmps[off]);
3949 ((UV) tmps[off ] << 8) +
3950 ((UV) tmps[off + 1]);
3954 ((UV) tmps[off ] << 24) +
3955 ((UV) tmps[off + 1] << 16) +
3956 ((UV) tmps[off + 2] << 8) +
3957 ((UV) tmps[off + 3]);
3959 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3960 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
3961 NORETURN_FUNCTION_END;
3964 /* Read a single line of the main body of the swash input text. These are of
3967 * where each number is hex. The first two numbers form the minimum and
3968 * maximum of a range, and the third is the value associated with the range.
3969 * Not all swashes should have a third number
3971 * On input: l points to the beginning of the line to be examined; it points
3972 * to somewhere in the string of the whole input text, and is
3973 * terminated by a \n or the null string terminator.
3974 * lend points to the null terminator of that string
3975 * wants_value is non-zero if the swash expects a third number
3976 * typestr is the name of the swash's mapping, like 'ToLower'
3977 * On output: *min, *max, and *val are set to the values read from the line.
3978 * returns a pointer just beyond the line examined. If there was no
3979 * valid min number on the line, returns lend+1
3983 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3984 const bool wants_value, const U8* const typestr)
3986 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3987 STRLEN numlen; /* Length of the number */
3988 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3989 | PERL_SCAN_DISALLOW_PREFIX
3990 | PERL_SCAN_SILENT_NON_PORTABLE;
3992 /* nl points to the next \n in the scan */
3993 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3995 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3997 /* Get the first number on the line: the range minimum */
3999 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4000 *max = *min; /* So can never return without setting max */
4001 if (numlen) /* If found a hex number, position past it */
4003 else if (nl) { /* Else, go handle next line, if any */
4004 return nl + 1; /* 1 is length of "\n" */
4006 else { /* Else, no next line */
4007 return lend + 1; /* to LIST's end at which \n is not found */
4010 /* The max range value follows, separated by a BLANK */
4013 flags = PERL_SCAN_SILENT_ILLDIGIT
4014 | PERL_SCAN_DISALLOW_PREFIX
4015 | PERL_SCAN_SILENT_NON_PORTABLE;
4017 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4020 else /* If no value here, it is a single element range */
4023 /* Non-binary tables have a third entry: what the first element of the
4024 * range maps to. The map for those currently read here is in hex */
4028 flags = PERL_SCAN_SILENT_ILLDIGIT
4029 | PERL_SCAN_DISALLOW_PREFIX
4030 | PERL_SCAN_SILENT_NON_PORTABLE;
4032 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4041 /* diag_listed_as: To%s: illegal mapping '%s' */
4042 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4048 *val = 0; /* bits == 1, then any val should be ignored */
4050 else { /* Nothing following range min, should be single element with no
4055 /* diag_listed_as: To%s: illegal mapping '%s' */
4056 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4060 *val = 0; /* bits == 1, then val should be ignored */
4063 /* Position to next line if any, or EOF */
4073 * Returns a swatch (a bit vector string) for a code point sequence
4074 * that starts from the value C<start> and comprises the number C<span>.
4075 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4076 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4079 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4082 U8 *l, *lend, *x, *xend, *s, *send;
4083 STRLEN lcur, xcur, scur;
4084 HV *const hv = MUTABLE_HV(SvRV(swash));
4085 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4087 SV** listsvp = NULL; /* The string containing the main body of the table */
4088 SV** extssvp = NULL;
4089 SV** invert_it_svp = NULL;
4092 STRLEN octets; /* if bits == 1, then octets == 0 */
4094 UV end = start + span;
4096 if (invlistsvp == NULL) {
4097 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4098 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4099 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4100 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4101 listsvp = hv_fetchs(hv, "LIST", FALSE);
4102 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4104 bits = SvUV(*bitssvp);
4105 none = SvUV(*nonesvp);
4106 typestr = (U8*)SvPV_nolen(*typesvp);
4112 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4114 PERL_ARGS_ASSERT_SWATCH_GET;
4116 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4117 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4121 /* If overflowed, use the max possible */
4127 /* create and initialize $swatch */
4128 scur = octets ? (span * octets) : (span + 7) / 8;
4129 swatch = newSV(scur);
4131 s = (U8*)SvPVX(swatch);
4132 if (octets && none) {
4133 const U8* const e = s + scur;
4136 *s++ = (U8)(none & 0xff);
4137 else if (bits == 16) {
4138 *s++ = (U8)((none >> 8) & 0xff);
4139 *s++ = (U8)( none & 0xff);
4141 else if (bits == 32) {
4142 *s++ = (U8)((none >> 24) & 0xff);
4143 *s++ = (U8)((none >> 16) & 0xff);
4144 *s++ = (U8)((none >> 8) & 0xff);
4145 *s++ = (U8)( none & 0xff);
4151 (void)memzero((U8*)s, scur + 1);
4153 SvCUR_set(swatch, scur);
4154 s = (U8*)SvPVX(swatch);
4156 if (invlistsvp) { /* If has an inversion list set up use that */
4157 _invlist_populate_swatch(*invlistsvp, start, end, s);
4161 /* read $swash->{LIST} */
4162 l = (U8*)SvPV(*listsvp, lcur);
4165 UV min, max, val, upper;
4166 l = swash_scan_list_line(l, lend, &min, &max, &val,
4167 cBOOL(octets), typestr);
4172 /* If looking for something beyond this range, go try the next one */
4176 /* <end> is generally 1 beyond where we want to set things, but at the
4177 * platform's infinity, where we can't go any higher, we want to
4178 * include the code point at <end> */
4181 : (max != UV_MAX || end != UV_MAX)
4188 if (!none || val < none) {
4193 for (key = min; key <= upper; key++) {
4195 /* offset must be non-negative (start <= min <= key < end) */
4196 offset = octets * (key - start);
4198 s[offset] = (U8)(val & 0xff);
4199 else if (bits == 16) {
4200 s[offset ] = (U8)((val >> 8) & 0xff);
4201 s[offset + 1] = (U8)( val & 0xff);
4203 else if (bits == 32) {
4204 s[offset ] = (U8)((val >> 24) & 0xff);
4205 s[offset + 1] = (U8)((val >> 16) & 0xff);
4206 s[offset + 2] = (U8)((val >> 8) & 0xff);
4207 s[offset + 3] = (U8)( val & 0xff);
4210 if (!none || val < none)
4214 else { /* bits == 1, then val should be ignored */
4219 for (key = min; key <= upper; key++) {
4220 const STRLEN offset = (STRLEN)(key - start);
4221 s[offset >> 3] |= 1 << (offset & 7);
4226 /* Invert if the data says it should be. Assumes that bits == 1 */
4227 if (invert_it_svp && SvUV(*invert_it_svp)) {
4229 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4230 * be 0, and their inversion should also be 0, as we don't succeed any
4231 * Unicode property matches for non-Unicode code points */
4232 if (start <= PERL_UNICODE_MAX) {
4234 /* The code below assumes that we never cross the
4235 * Unicode/above-Unicode boundary in a range, as otherwise we would
4236 * have to figure out where to stop flipping the bits. Since this
4237 * boundary is divisible by a large power of 2, and swatches comes
4238 * in small powers of 2, this should be a valid assumption */
4239 assert(start + span - 1 <= PERL_UNICODE_MAX);
4249 /* read $swash->{EXTRAS}
4250 * This code also copied to swash_to_invlist() below */
4251 x = (U8*)SvPV(*extssvp, xcur);
4259 SV **otherbitssvp, *other;
4263 const U8 opc = *x++;
4267 nl = (U8*)memchr(x, '\n', xend - x);
4269 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4271 x = nl + 1; /* 1 is length of "\n" */
4275 x = xend; /* to EXTRAS' end at which \n is not found */
4282 namelen = nl - namestr;
4286 namelen = xend - namestr;
4290 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4291 otherhv = MUTABLE_HV(SvRV(*othersvp));
4292 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4293 otherbits = (STRLEN)SvUV(*otherbitssvp);
4294 if (bits < otherbits)
4295 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4296 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4298 /* The "other" swatch must be destroyed after. */
4299 other = swatch_get(*othersvp, start, span);
4300 o = (U8*)SvPV(other, olen);
4303 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4305 s = (U8*)SvPV(swatch, slen);
4306 if (bits == 1 && otherbits == 1) {
4308 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4309 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4310 (UV)slen, (UV)olen);
4334 STRLEN otheroctets = otherbits >> 3;
4336 U8* const send = s + slen;
4341 if (otherbits == 1) {
4342 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4346 STRLEN vlen = otheroctets;
4354 if (opc == '+' && otherval)
4355 NOOP; /* replace with otherval */
4356 else if (opc == '!' && !otherval)
4358 else if (opc == '-' && otherval)
4360 else if (opc == '&' && !otherval)
4363 s += octets; /* no replacement */
4368 *s++ = (U8)( otherval & 0xff);
4369 else if (bits == 16) {
4370 *s++ = (U8)((otherval >> 8) & 0xff);
4371 *s++ = (U8)( otherval & 0xff);
4373 else if (bits == 32) {
4374 *s++ = (U8)((otherval >> 24) & 0xff);
4375 *s++ = (U8)((otherval >> 16) & 0xff);
4376 *s++ = (U8)((otherval >> 8) & 0xff);
4377 *s++ = (U8)( otherval & 0xff);
4381 sv_free(other); /* through with it! */
4387 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4390 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4391 * Can't be used on a property that is subject to user override, as it
4392 * relies on the value of SPECIALS in the swash which would be set by
4393 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4394 * for overridden properties
4396 * Returns a hash which is the inversion and closure of a swash mapping.
4397 * For example, consider the input lines:
4402 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4403 * 006C. The value for each key is an array. For 006C, the array would
4404 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4405 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4407 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4408 * keys are only code points that are folded-to, so it isn't a full closure.
4410 * Essentially, for any code point, it gives all the code points that map to
4411 * it, or the list of 'froms' for that point.
4413 * Currently it ignores any additions or deletions from other swashes,
4414 * looking at just the main body of the swash, and if there are SPECIALS
4415 * in the swash, at that hash
4417 * The specials hash can be extra code points, and most likely consists of
4418 * maps from single code points to multiple ones (each expressed as a string
4419 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4420 * However consider this possible input in the specials hash:
4421 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4422 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4424 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4425 * currently handle. But it also means that FB05 and FB06 are equivalent in
4426 * a 1-1 mapping which we should handle, and this relationship may not be in
4427 * the main table. Therefore this function examines all the multi-char
4428 * sequences and adds the 1-1 mappings that come out of that.
4430 * XXX This function was originally intended to be multipurpose, but its
4431 * only use is quite likely to remain for constructing the inversion of
4432 * the CaseFolding (//i) property. If it were more general purpose for
4433 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4434 * because certain folds are prohibited under /iaa and /il. As an example,
4435 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4436 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4437 * prohibited, so we would not figure out that they fold to each other.
4438 * Code could be written to automatically figure this out, similar to the
4439 * code that does this for multi-character folds, but this is the only case
4440 * where something like this is ever likely to happen, as all the single
4441 * char folds to the 0-255 range are now quite settled. Instead there is a
4442 * little special code that is compiled only for this Unicode version. This
4443 * is smaller and didn't require much coding time to do. But this makes
4444 * this routine strongly tied to being used just for CaseFolding. If ever
4445 * it should be generalized, this would have to be fixed */
4449 HV *const hv = MUTABLE_HV(SvRV(swash));
4451 /* The string containing the main body of the table. This will have its
4452 * assertion fail if the swash has been converted to its inversion list */
4453 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4455 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4456 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4457 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4458 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4459 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4460 const STRLEN bits = SvUV(*bitssvp);
4461 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4462 const UV none = SvUV(*nonesvp);
4463 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4467 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4469 /* Must have at least 8 bits to get the mappings */
4470 if (bits != 8 && bits != 16 && bits != 32) {
4471 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4475 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4476 mapping to more than one character */
4478 /* Construct an inverse mapping hash for the specials */
4479 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4480 HV * specials_inverse = newHV();
4481 char *char_from; /* the lhs of the map */
4482 I32 from_len; /* its byte length */
4483 char *char_to; /* the rhs of the map */
4484 I32 to_len; /* its byte length */
4485 SV *sv_to; /* and in a sv */
4486 AV* from_list; /* list of things that map to each 'to' */
4488 hv_iterinit(specials_hv);
4490 /* The keys are the characters (in UTF-8) that map to the corresponding
4491 * UTF-8 string value. Iterate through the list creating the inverse
4493 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4495 if (! SvPOK(sv_to)) {
4496 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4497 "unexpectedly is not a string, flags=%lu",
4498 (unsigned long)SvFLAGS(sv_to));
4500 /*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)));*/
4502 /* Each key in the inverse list is a mapped-to value, and the key's
4503 * hash value is a list of the strings (each in UTF-8) that map to
4504 * it. Those strings are all one character long */
4505 if ((listp = hv_fetch(specials_inverse,
4509 from_list = (AV*) *listp;
4511 else { /* No entry yet for it: create one */
4512 from_list = newAV();
4513 if (! hv_store(specials_inverse,
4516 (SV*) from_list, 0))
4518 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4522 /* Here have the list associated with this 'to' (perhaps newly
4523 * created and empty). Just add to it. Note that we ASSUME that
4524 * the input is guaranteed to not have duplications, so we don't
4525 * check for that. Duplications just slow down execution time. */
4526 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4529 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4530 * it looking for cases like the FB05/FB06 examples above. There would
4531 * be an entry in the hash like
4532 * 'st' => [ FB05, FB06 ]
4533 * In this example we will create two lists that get stored in the
4534 * returned hash, 'ret':
4535 * FB05 => [ FB05, FB06 ]
4536 * FB06 => [ FB05, FB06 ]
4538 * Note that there is nothing to do if the array only has one element.
4539 * (In the normal 1-1 case handled below, we don't have to worry about
4540 * two lists, as everything gets tied to the single list that is
4541 * generated for the single character 'to'. But here, we are omitting
4542 * that list, ('st' in the example), so must have multiple lists.) */
4543 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4544 &char_to, &to_len)))
4546 if (av_tindex_nomg(from_list) > 0) {
4549 /* We iterate over all combinations of i,j to place each code
4550 * point on each list */
4551 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
4553 AV* i_list = newAV();
4554 SV** entryp = av_fetch(from_list, i, FALSE);
4555 if (entryp == NULL) {
4556 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4558 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4559 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4561 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4562 (SV*) i_list, FALSE))
4564 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4567 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4568 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
4569 entryp = av_fetch(from_list, j, FALSE);
4570 if (entryp == NULL) {
4571 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4574 /* When i==j this adds itself to the list */
4575 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4576 (U8*) SvPVX(*entryp),
4577 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4579 /*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));*/
4584 SvREFCNT_dec(specials_inverse); /* done with it */
4585 } /* End of specials */
4587 /* read $swash->{LIST} */
4589 #if UNICODE_MAJOR_VERSION == 3 \
4590 && UNICODE_DOT_VERSION == 0 \
4591 && UNICODE_DOT_DOT_VERSION == 1
4593 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4594 * rule so that things work under /iaa and /il */
4596 SV * mod_listsv = sv_mortalcopy(*listsvp);
4597 sv_catpv(mod_listsv, "130\t130\t131\n");
4598 l = (U8*)SvPV(mod_listsv, lcur);
4602 l = (U8*)SvPV(*listsvp, lcur);
4608 /* Go through each input line */
4612 l = swash_scan_list_line(l, lend, &min, &max, &val,
4613 cBOOL(octets), typestr);
4618 /* Each element in the range is to be inverted */
4619 for (inverse = min; inverse <= max; inverse++) {
4623 bool found_key = FALSE;
4624 bool found_inverse = FALSE;
4626 /* The key is the inverse mapping */
4627 char key[UTF8_MAXBYTES+1];
4628 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4629 STRLEN key_len = key_end - key;
4631 /* Get the list for the map */
4632 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4633 list = (AV*) *listp;
4635 else { /* No entry yet for it: create one */
4637 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4638 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4642 /* Look through list to see if this inverse mapping already is
4643 * listed, or if there is a mapping to itself already */
4644 for (i = 0; i <= av_tindex_nomg(list); i++) {
4645 SV** entryp = av_fetch(list, i, FALSE);
4648 if (entryp == NULL) {
4649 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4653 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4657 if (uv == inverse) {
4658 found_inverse = TRUE;
4661 /* No need to continue searching if found everything we are
4663 if (found_key && found_inverse) {
4668 /* Make sure there is a mapping to itself on the list */
4670 av_push(list, newSVuv(val));
4671 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4675 /* Simply add the value to the list */
4676 if (! found_inverse) {
4677 av_push(list, newSVuv(inverse));
4678 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4681 /* swatch_get() increments the value of val for each element in the
4682 * range. That makes more compact tables possible. You can
4683 * express the capitalization, for example, of all consecutive
4684 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4685 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4686 * and it's not documented; it appears to be used only in
4687 * implementing tr//; I copied the semantics from swatch_get(), just
4689 if (!none || val < none) {
4699 Perl__swash_to_invlist(pTHX_ SV* const swash)
4702 /* Subject to change or removal. For use only in one place in regcomp.c.
4703 * Ownership is given to one reference count in the returned SV* */
4708 HV *const hv = MUTABLE_HV(SvRV(swash));
4709 UV elements = 0; /* Number of elements in the inversion list */
4719 STRLEN octets; /* if bits == 1, then octets == 0 */
4725 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4727 /* If not a hash, it must be the swash's inversion list instead */
4728 if (SvTYPE(hv) != SVt_PVHV) {
4729 return SvREFCNT_inc_simple_NN((SV*) hv);
4732 /* The string containing the main body of the table */
4733 listsvp = hv_fetchs(hv, "LIST", FALSE);
4734 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4735 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4736 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4737 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4739 typestr = (U8*)SvPV_nolen(*typesvp);
4740 bits = SvUV(*bitssvp);
4741 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4743 /* read $swash->{LIST} */
4744 if (SvPOK(*listsvp)) {
4745 l = (U8*)SvPV(*listsvp, lcur);
4748 /* LIST legitimately doesn't contain a string during compilation phases
4749 * of Perl itself, before the Unicode tables are generated. In this
4750 * case, just fake things up by creating an empty list */
4757 if (*l == 'V') { /* Inversion list format */
4758 const char *after_atou = (char *) lend;
4760 UV* other_elements_ptr;
4762 /* The first number is a count of the rest */
4764 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4765 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
4767 if (elements == 0) {
4768 invlist = _new_invlist(0);
4771 l = (U8 *) after_atou;
4773 /* Get the 0th element, which is needed to setup the inversion list */
4774 while (isSPACE(*l)) l++;
4775 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4776 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
4778 l = (U8 *) after_atou;
4779 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
4782 /* Then just populate the rest of the input */
4783 while (elements-- > 0) {
4785 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more elements than available", elements);
4787 while (isSPACE(*l)) l++;
4788 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
4789 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
4791 l = (U8 *) after_atou;
4797 /* Scan the input to count the number of lines to preallocate array
4798 * size based on worst possible case, which is each line in the input
4799 * creates 2 elements in the inversion list: 1) the beginning of a
4800 * range in the list; 2) the beginning of a range not in the list. */
4801 while ((loc = (strchr(loc, '\n'))) != NULL) {
4806 /* If the ending is somehow corrupt and isn't a new line, add another
4807 * element for the final range that isn't in the inversion list */
4808 if (! (*lend == '\n'
4809 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4814 invlist = _new_invlist(elements);
4816 /* Now go through the input again, adding each range to the list */
4819 UV val; /* Not used by this function */
4821 l = swash_scan_list_line(l, lend, &start, &end, &val,
4822 cBOOL(octets), typestr);
4828 invlist = _add_range_to_invlist(invlist, start, end);
4832 /* Invert if the data says it should be */
4833 if (invert_it_svp && SvUV(*invert_it_svp)) {
4834 _invlist_invert(invlist);
4837 /* This code is copied from swatch_get()
4838 * read $swash->{EXTRAS} */
4839 x = (U8*)SvPV(*extssvp, xcur);
4847 SV **otherbitssvp, *other;
4850 const U8 opc = *x++;
4854 nl = (U8*)memchr(x, '\n', xend - x);
4856 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4858 x = nl + 1; /* 1 is length of "\n" */
4862 x = xend; /* to EXTRAS' end at which \n is not found */
4869 namelen = nl - namestr;
4873 namelen = xend - namestr;
4877 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4878 otherhv = MUTABLE_HV(SvRV(*othersvp));
4879 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4880 otherbits = (STRLEN)SvUV(*otherbitssvp);
4882 if (bits != otherbits || bits != 1) {
4883 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4884 "properties, bits=%" UVuf ", otherbits=%" UVuf,
4885 (UV)bits, (UV)otherbits);
4888 /* The "other" swatch must be destroyed after. */
4889 other = _swash_to_invlist((SV *)*othersvp);
4891 /* End of code copied from swatch_get() */
4894 _invlist_union(invlist, other, &invlist);
4897 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4900 _invlist_subtract(invlist, other, &invlist);
4903 _invlist_intersection(invlist, other, &invlist);
4908 sv_free(other); /* through with it! */
4911 SvREADONLY_on(invlist);
4916 Perl__get_swash_invlist(pTHX_ SV* const swash)
4920 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4922 if (! SvROK(swash)) {
4926 /* If it really isn't a hash, it isn't really swash; must be an inversion
4928 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4932 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4941 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4943 /* May change: warns if surrogates, non-character code points, or
4944 * non-Unicode code points are in s which has length len bytes. Returns
4945 * TRUE if none found; FALSE otherwise. The only other validity check is
4946 * to make sure that this won't exceed the string's length.
4948 * Code points above the platform's C<IV_MAX> will raise a deprecation
4949 * warning, unless those are turned off. */
4951 const U8* const e = s + len;
4954 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4957 if (UTF8SKIP(s) > len) {
4958 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4959 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4962 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4963 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4964 if ( ckWARN_d(WARN_NON_UNICODE)
4965 || ( ckWARN_d(WARN_DEPRECATED)
4967 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
4968 #else /* Below is 64-bit words */
4969 /* 2**63 and up meet these conditions provided we have
4973 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
4976 /* s[1] being above 0x80 overflows */
4981 /* A side effect of this function will be to warn */
4982 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4986 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4987 if (ckWARN_d(WARN_SURROGATE)) {
4988 /* This has a different warning than the one the called
4989 * function would output, so can't just call it, unlike we
4990 * do for the non-chars and above-unicodes */
4991 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4992 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4993 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", uv);
4997 else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
4998 /* A side effect of this function will be to warn */
4999 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5010 =for apidoc pv_uni_display
5012 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5013 length C<len>, the displayable version being at most C<pvlim> bytes long
5014 (if longer, the rest is truncated and C<"..."> will be appended).
5016 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5017 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5018 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5019 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5020 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5021 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5023 The pointer to the PV of the C<dsv> is returned.
5025 See also L</sv_uni_display>.
5029 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
5034 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5038 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5040 /* This serves double duty as a flag and a character to print after
5041 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5045 if (pvlim && SvCUR(dsv) >= pvlim) {
5049 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5051 const unsigned char c = (unsigned char)u & 0xFF;
5052 if (flags & UNI_DISPLAY_BACKSLASH) {
5069 const char string = ok;
5070 sv_catpvs(dsv, "\\");
5071 sv_catpvn(dsv, &string, 1);
5074 /* isPRINT() is the locale-blind version. */
5075 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5076 const char string = c;
5077 sv_catpvn(dsv, &string, 1);
5082 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5085 sv_catpvs(dsv, "...");
5091 =for apidoc sv_uni_display
5093 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5094 the displayable version being at most C<pvlim> bytes long
5095 (if longer, the rest is truncated and "..." will be appended).
5097 The C<flags> argument is as in L</pv_uni_display>().
5099 The pointer to the PV of the C<dsv> is returned.
5104 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5106 const char * const ptr =
5107 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5109 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5111 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5112 SvCUR(ssv), pvlim, flags);
5116 =for apidoc foldEQ_utf8
5118 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
5119 of which may be in UTF-8) are the same case-insensitively; false otherwise.
5120 How far into the strings to compare is determined by other input parameters.
5122 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5123 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
5124 with respect to C<s2>.
5126 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
5127 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
5128 scan will not be considered to be a match unless the goal is reached, and
5129 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
5132 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
5133 considered an end pointer to the position 1 byte past the maximum point
5134 in C<s1> beyond which scanning will not continue under any circumstances.
5135 (This routine assumes that UTF-8 encoded input strings are not malformed;
5136 malformed input can cause it to read past C<pe1>).
5137 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
5138 is less than C<s1>+C<l1>, the match will never be successful because it can
5140 get as far as its goal (and in fact is asserted against). Correspondingly for
5141 C<pe2> with respect to C<s2>.
5143 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5144 C<l2> must be non-zero), and if both do, both have to be
5145 reached for a successful match. Also, if the fold of a character is multiple
5146 characters, all of them must be matched (see tr21 reference below for
5149 Upon a successful match, if C<pe1> is non-C<NULL>,
5150 it will be set to point to the beginning of the I<next> character of C<s1>
5151 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5153 For case-insensitiveness, the "casefolding" of Unicode is used
5154 instead of upper/lowercasing both the characters, see
5155 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5159 /* A flags parameter has been added which may change, and hence isn't
5160 * externally documented. Currently it is:
5161 * 0 for as-documented above
5162 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5163 ASCII one, to not match
5164 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5165 * locale are to be used.
5166 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5167 * routine. This allows that step to be skipped.
5168 * Currently, this requires s1 to be encoded as UTF-8
5169 * (u1 must be true), which is asserted for.
5170 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5171 * cross certain boundaries. Hence, the caller should
5172 * let this function do the folding instead of
5173 * pre-folding. This code contains an assertion to
5174 * that effect. However, if the caller knows what
5175 * it's doing, it can pass this flag to indicate that,
5176 * and the assertion is skipped.
5177 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5178 * FOLDEQ_S2_FOLDS_SANE
5181 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)
5183 const U8 *p1 = (const U8*)s1; /* Point to current char */
5184 const U8 *p2 = (const U8*)s2;
5185 const U8 *g1 = NULL; /* goal for s1 */
5186 const U8 *g2 = NULL;
5187 const U8 *e1 = NULL; /* Don't scan s1 past this */
5188 U8 *f1 = NULL; /* Point to current folded */
5189 const U8 *e2 = NULL;
5191 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5192 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5193 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5194 U8 flags_for_folder = FOLD_FLAGS_FULL;
5196 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5198 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5199 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5200 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5201 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5202 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5203 /* The algorithm is to trial the folds without regard to the flags on
5204 * the first line of the above assert(), and then see if the result
5205 * violates them. This means that the inputs can't be pre-folded to a
5206 * violating result, hence the assert. This could be changed, with the
5207 * addition of extra tests here for the already-folded case, which would
5208 * slow it down. That cost is more than any possible gain for when these
5209 * flags are specified, as the flags indicate /il or /iaa matching which
5210 * is less common than /iu, and I (khw) also believe that real-world /il
5211 * and /iaa matches are most likely to involve code points 0-255, and this
5212 * function only under rare conditions gets called for 0-255. */
5214 if (flags & FOLDEQ_LOCALE) {
5215 if (IN_UTF8_CTYPE_LOCALE) {
5216 flags &= ~FOLDEQ_LOCALE;
5219 flags_for_folder |= FOLD_FLAGS_LOCALE;
5228 g1 = (const U8*)s1 + l1;
5236 g2 = (const U8*)s2 + l2;
5239 /* Must have at least one goal */
5244 /* Will never match if goal is out-of-bounds */
5245 assert(! e1 || e1 >= g1);
5247 /* Here, there isn't an end pointer, or it is beyond the goal. We
5248 * only go as far as the goal */
5252 assert(e1); /* Must have an end for looking at s1 */
5255 /* Same for goal for s2 */
5257 assert(! e2 || e2 >= g2);
5264 /* If both operands are already folded, we could just do a memEQ on the
5265 * whole strings at once, but it would be better if the caller realized
5266 * this and didn't even call us */
5268 /* Look through both strings, a character at a time */
5269 while (p1 < e1 && p2 < e2) {
5271 /* If at the beginning of a new character in s1, get its fold to use
5272 * and the length of the fold. */
5274 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5280 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5282 /* We have to forbid mixing ASCII with non-ASCII if the
5283 * flags so indicate. And, we can short circuit having to
5284 * call the general functions for this common ASCII case,
5285 * all of whose non-locale folds are also ASCII, and hence
5286 * UTF-8 invariants, so the UTF8ness of the strings is not
5288 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5292 *foldbuf1 = toFOLD(*p1);
5295 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5297 else { /* Not UTF-8, get UTF-8 fold */
5298 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5304 if (n2 == 0) { /* Same for s2 */
5305 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5311 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5312 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5316 *foldbuf2 = toFOLD(*p2);
5319 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5322 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5328 /* Here f1 and f2 point to the beginning of the strings to compare.
5329 * These strings are the folds of the next character from each input
5330 * string, stored in UTF-8. */
5332 /* While there is more to look for in both folds, see if they
5333 * continue to match */
5335 U8 fold_length = UTF8SKIP(f1);
5336 if (fold_length != UTF8SKIP(f2)
5337 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5338 function call for single
5340 || memNE((char*)f1, (char*)f2, fold_length))
5342 return 0; /* mismatch */
5345 /* Here, they matched, advance past them */
5352 /* When reach the end of any fold, advance the input past it */
5354 p1 += u1 ? UTF8SKIP(p1) : 1;
5357 p2 += u2 ? UTF8SKIP(p2) : 1;
5359 } /* End of loop through both strings */
5361 /* A match is defined by each scan that specified an explicit length
5362 * reaching its final goal, and the other not having matched a partial
5363 * character (which can happen when the fold of a character is more than one
5365 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5369 /* Successful match. Set output pointers */
5379 /* XXX The next two functions should likely be moved to mathoms.c once all
5380 * occurrences of them are removed from the core; some cpan-upstream modules
5384 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5386 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5388 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5392 =for apidoc utf8n_to_uvuni
5394 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5396 This function was useful for code that wanted to handle both EBCDIC and
5397 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5398 distinctions between the platforms have mostly been made invisible to most
5399 code, so this function is quite unlikely to be what you want. If you do need
5400 this precise functionality, use instead
5401 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5402 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5408 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5410 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5412 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5416 =for apidoc uvuni_to_utf8_flags
5418 Instead you almost certainly want to use L</uvchr_to_utf8> or
5419 L</uvchr_to_utf8_flags>.
5421 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5422 which itself, while not deprecated, should be used only in isolated
5423 circumstances. These functions were useful for code that wanted to handle
5424 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5425 v5.20, the distinctions between the platforms have mostly been made invisible
5426 to most code, so this function is quite unlikely to be what you want.
5432 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5434 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5436 return uvoffuni_to_utf8_flags(d, uv, flags);
5440 * ex: set ts=8 sts=4 sw=4 et: