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;
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, 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)) {
702 return 0; /* Above Unicode */
705 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
706 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
708 return 0; /* Above 31 bits */
712 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
714 if ( (flags & UTF8_DISALLOW_SUPER)
715 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
717 return 0; /* Above Unicode */
720 if ( (flags & UTF8_DISALLOW_SURROGATE)
721 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
723 return 0; /* Surrogate */
726 if ( (flags & UTF8_DISALLOW_NONCHAR)
727 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
729 return 0; /* Noncharacter code point */
734 /* Make sure that all that follows are continuation bytes */
735 for (x = s + 1; x < e; x++) {
736 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
741 /* Here is syntactically valid. Next, make sure this isn't the start of an
743 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
747 /* And finally, that the code point represented fits in a word on this
749 if (does_utf8_overflow(s, e)) {
757 S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
759 /* Returns a mortalized C string that is a displayable copy of the 'len'
760 * bytes starting at 's', each in a \xXY format. */
762 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
764 const U8 * const e = s + len;
768 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
770 Newx(output, output_len, char);
775 const unsigned high_nibble = (*s & 0xF0) >> 4;
776 const unsigned low_nibble = (*s & 0x0F);
781 if (high_nibble < 10) {
782 *d++ = high_nibble + '0';
785 *d++ = high_nibble - 10 + 'a';
788 if (low_nibble < 10) {
789 *d++ = low_nibble + '0';
792 *d++ = low_nibble - 10 + 'a';
800 PERL_STATIC_INLINE char *
801 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
803 /* How many bytes to print */
806 /* Which one is the non-continuation */
807 const STRLEN non_cont_byte_pos,
809 /* How many bytes should there be? */
810 const STRLEN expect_len)
812 /* Return the malformation warning text for an unexpected continuation
815 const char * const where = (non_cont_byte_pos == 1)
817 : Perl_form(aTHX_ "%d bytes",
818 (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 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
828 * length that is larger than is actually available in the buffer. If we
829 * print all the bytes based on that length, we will read past the buffer
830 * end. Often, the strings are NUL terminated, so to lower the chances of
831 * this happening, print the malformed bytes only up through any NUL. */
832 for (i = 1; i < print_len; i++) {
833 if (*(s + i) == '\0') {
834 print_len = i + 1; /* +1 gets the NUL printed */
839 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
840 " %s after start byte 0x%02x; need %d bytes, got %d)",
842 _byte_dump_string(s, print_len),
843 *(s + non_cont_byte_pos),
847 (int) non_cont_byte_pos);
852 =for apidoc utf8n_to_uvchr
854 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
855 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
857 Bottom level UTF-8 decode routine.
858 Returns the native code point value of the first character in the string C<s>,
859 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
860 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
861 the length, in bytes, of that character.
863 The value of C<flags> determines the behavior when C<s> does not point to a
864 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
865 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
866 is the next possible position in C<s> that could begin a non-malformed
867 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
868 is raised. Some UTF-8 input sequences may contain multiple malformations.
869 This function tries to find every possible one in each call, so multiple
870 warnings can be raised for each sequence.
872 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
873 individual types of malformations, such as the sequence being overlong (that
874 is, when there is a shorter sequence that can express the same code point;
875 overlong sequences are expressly forbidden in the UTF-8 standard due to
876 potential security issues). Another malformation example is the first byte of
877 a character not being a legal first byte. See F<utf8.h> for the list of such
878 flags. For allowed overlong sequences, the computed code point is returned;
879 for all other allowed malformations, the Unicode REPLACEMENT CHARACTER is
882 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
883 flags) malformation is found. If this flag is set, the routine assumes that
884 the caller will raise a warning, and this function will silently just set
885 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
887 Note that this API requires disambiguation between successful decoding a C<NUL>
888 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
889 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
890 be set to 1. To disambiguate, upon a zero return, see if the first byte of
891 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
892 error. Or you can use C<L</utf8n_to_uvchr_error>>.
894 Certain code points are considered problematic. These are Unicode surrogates,
895 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
896 By default these are considered regular code points, but certain situations
897 warrant special handling for them, which can be specified using the C<flags>
898 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
899 three classes are treated as malformations and handled as such. The flags
900 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
901 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
902 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
903 restricts the allowed inputs to the strict UTF-8 traditionally defined by
904 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
906 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
907 The difference between traditional strictness and C9 strictness is that the
908 latter does not forbid non-character code points. (They are still discouraged,
909 however.) For more discussion see L<perlunicode/Noncharacter code points>.
911 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
912 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
913 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
914 raised for their respective categories, but otherwise the code points are
915 considered valid (not malformations). To get a category to both be treated as
916 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
917 (But note that warnings are not raised if lexically disabled nor if
918 C<UTF8_CHECK_ONLY> is also specified.)
920 It is now deprecated to have very high code points (above C<IV_MAX> on the
921 platforms) and this function will raise a deprecation warning for these (unless
922 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
925 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
926 so using them is more problematic than other above-Unicode code points. Perl
927 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
928 likely that non-Perl languages will not be able to read files that contain
929 these; nor would Perl understand files
930 written by something that uses a different extension. For these reasons, there
931 is a separate set of flags that can warn and/or disallow these extremely high
932 code points, even if other above-Unicode ones are accepted. These are the
933 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
934 are entirely independent from the deprecation warning for code points above
935 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
936 code point that needs more than 31 bits to represent. When that happens,
937 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
938 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
939 above-Unicode code points, including these, as malformations; and
940 C<UTF8_WARN_SUPER> warns on these.)
942 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
943 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
944 than on ASCII. Prior to that, code points 2**31 and higher were simply
945 unrepresentable, and a different, incompatible method was used to represent
946 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
947 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
948 platforms, warning and disallowing 2**31 and higher.
950 All other code points corresponding to Unicode characters, including private
951 use and those yet to be assigned, are never considered malformed and never
956 Also implemented as a macro in utf8.h
960 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
965 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
967 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
972 =for apidoc utf8n_to_uvchr_error
974 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
975 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
977 This function is for code that needs to know what the precise malformation(s)
978 are when an error is found.
980 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
981 all the others, C<errors>. If this parameter is 0, this function behaves
982 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
983 to a C<U32> variable, which this function sets to indicate any errors found.
984 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
985 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
986 of these bits will be set if a malformation is found, even if the input
987 C<flags> parameter indicates that the given malformation is allowed; those
988 exceptions are noted:
992 =item C<UTF8_GOT_ABOVE_31_BIT>
994 The code point represented by the input UTF-8 sequence occupies more than 31
996 This bit is set only if the input C<flags> parameter contains either the
997 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
999 =item C<UTF8_GOT_CONTINUATION>
1001 The input sequence was malformed in that the first byte was a a UTF-8
1004 =item C<UTF8_GOT_EMPTY>
1006 The input C<curlen> parameter was 0.
1008 =item C<UTF8_GOT_LONG>
1010 The input sequence was malformed in that there is some other sequence that
1011 evaluates to the same code point, but that sequence is shorter than this one.
1013 =item C<UTF8_GOT_NONCHAR>
1015 The code point represented by the input UTF-8 sequence is for a Unicode
1016 non-character code point.
1017 This bit is set only if the input C<flags> parameter contains either the
1018 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1020 =item C<UTF8_GOT_NON_CONTINUATION>
1022 The input sequence was malformed in that a non-continuation type byte was found
1023 in a position where only a continuation type one should be.
1025 =item C<UTF8_GOT_OVERFLOW>
1027 The input sequence was malformed in that it is for a code point that is not
1028 representable in the number of bits available in a UV on the current platform.
1030 =item C<UTF8_GOT_SHORT>
1032 The input sequence was malformed in that C<curlen> is smaller than required for
1033 a complete sequence. In other words, the input is for a partial character
1036 =item C<UTF8_GOT_SUPER>
1038 The input sequence was malformed in that it is for a non-Unicode code point;
1039 that is, one above the legal Unicode maximum.
1040 This bit is set only if the input C<flags> parameter contains either the
1041 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1043 =item C<UTF8_GOT_SURROGATE>
1045 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1047 This bit is set only if the input C<flags> parameter contains either the
1048 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1052 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1053 flag to suppress any warnings, and then examine the C<*errors> return.
1059 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1065 const U8 * const s0 = s;
1066 U8 * send = NULL; /* (initialized to silence compilers' wrong
1068 U32 possible_problems = 0; /* A bit is set here for each potential problem
1069 found as we go along */
1071 STRLEN expectlen = 0; /* How long should this sequence be?
1072 (initialized to silence compilers' wrong
1074 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1075 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1076 this gets set and discarded */
1078 /* The below are used only if there is both an overlong malformation and a
1079 * too short one. Otherwise the first two are set to 's0' and 'send', and
1080 * the third not used at all */
1081 U8 * adjusted_s0 = (U8 *) s0;
1082 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1084 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1086 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1092 errors = &discard_errors;
1095 /* The order of malformation tests here is important. We should consume as
1096 * few bytes as possible in order to not skip any valid character. This is
1097 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1098 * http://unicode.org/reports/tr36 for more discussion as to why. For
1099 * example, once we've done a UTF8SKIP, we can tell the expected number of
1100 * bytes, and could fail right off the bat if the input parameters indicate
1101 * that there are too few available. But it could be that just that first
1102 * byte is garbled, and the intended character occupies fewer bytes. If we
1103 * blindly assumed that the first byte is correct, and skipped based on
1104 * that number, we could skip over a valid input character. So instead, we
1105 * always examine the sequence byte-by-byte.
1107 * We also should not consume too few bytes, otherwise someone could inject
1108 * things. For example, an input could be deliberately designed to
1109 * overflow, and if this code bailed out immediately upon discovering that,
1110 * returning to the caller C<*retlen> pointing to the very next byte (one
1111 * which is actually part of of the overflowing sequence), that could look
1112 * legitimate to the caller, which could discard the initial partial
1113 * sequence and process the rest, inappropriately.
1115 * Some possible input sequences are malformed in more than one way. This
1116 * function goes to lengths to try to find all of them. This is necessary
1117 * for correctness, as the inputs may allow one malformation but not
1118 * another, and if we abandon searching for others after finding the
1119 * allowed one, we could allow in something that shouldn't have been.
1122 if (UNLIKELY(curlen == 0)) {
1123 possible_problems |= UTF8_GOT_EMPTY;
1125 uv = UNICODE_REPLACEMENT;
1126 goto ready_to_handle_errors;
1129 expectlen = UTF8SKIP(s);
1131 /* A well-formed UTF-8 character, as the vast majority of calls to this
1132 * function will be for, has this expected length. For efficiency, set
1133 * things up here to return it. It will be overriden only in those rare
1134 * cases where a malformation is found */
1136 *retlen = expectlen;
1139 /* An invariant is trivially well-formed */
1140 if (UTF8_IS_INVARIANT(uv)) {
1144 /* A continuation character can't start a valid sequence */
1145 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1146 possible_problems |= UTF8_GOT_CONTINUATION;
1148 uv = UNICODE_REPLACEMENT;
1149 goto ready_to_handle_errors;
1152 /* Here is not a continuation byte, nor an invariant. The only thing left
1153 * is a start byte (possibly for an overlong) */
1155 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1156 * that indicate the number of bytes in the character's whole UTF-8
1157 * sequence, leaving just the bits that are part of the value. */
1158 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1160 /* Setup the loop end point, making sure to not look past the end of the
1161 * input string, and flag it as too short if the size isn't big enough. */
1163 if (UNLIKELY(curlen < expectlen)) {
1164 possible_problems |= UTF8_GOT_SHORT;
1171 adjusted_send = send;
1173 /* Now, loop through the remaining bytes in the character's sequence,
1174 * accumulating each into the working value as we go. */
1175 for (s = s0 + 1; s < send; s++) {
1176 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1177 uv = UTF8_ACCUMULATE(uv, *s);
1181 /* Here, found a non-continuation before processing all expected bytes.
1182 * This byte indicates the beginning of a new character, so quit, even
1183 * if allowing this malformation. */
1184 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1186 } /* End of loop through the character's bytes */
1188 /* Save how many bytes were actually in the character */
1191 /* A convenience macro that matches either of the too-short conditions. */
1192 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1194 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1196 uv = UNICODE_REPLACEMENT;
1199 /* Note that there are two types of too-short malformation. One is when
1200 * there is actual wrong data before the normal termination of the
1201 * sequence. The other is that the sequence wasn't complete before the end
1202 * of the data we are allowed to look at, based on the input 'curlen'.
1203 * This means that we were passed data for a partial character, but it is
1204 * valid as far as we saw. The other is definitely invalid. This
1205 * distinction could be important to a caller, so the two types are kept
1208 /* Check for overflow */
1209 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1210 possible_problems |= UTF8_GOT_OVERFLOW;
1211 uv = UNICODE_REPLACEMENT;
1214 /* Check for overlong. If no problems so far, 'uv' is the correct code
1215 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1216 * we must look at the UTF-8 byte sequence itself to see if it is for an
1218 if ( ( LIKELY(! possible_problems)
1219 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1220 || ( UNLIKELY( possible_problems)
1221 && ( UNLIKELY(! UTF8_IS_START(*s0))
1223 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1226 possible_problems |= UTF8_GOT_LONG;
1228 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1229 UV min_uv = uv_so_far;
1232 /* Here, the input is both overlong and is missing some trailing
1233 * bytes. There is no single code point it could be for, but there
1234 * may be enough information present to determine if what we have
1235 * so far is for an unallowed code point, such as for a surrogate.
1236 * The code below has the intelligence to determine this, but just
1237 * for non-overlong UTF-8 sequences. What we do here is calculate
1238 * the smallest code point the input could represent if there were
1239 * no too short malformation. Then we compute and save the UTF-8
1240 * for that, which is what the code below looks at instead of the
1241 * raw input. It turns out that the smallest such code point is
1243 for (i = curlen; i < expectlen; i++) {
1244 min_uv = UTF8_ACCUMULATE(min_uv,
1245 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1248 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1249 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1250 to free it ourselves if
1251 warnings are made fatal */
1252 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1256 /* Now check that the input isn't for a problematic code point not allowed
1257 * by the input parameters. */
1258 /* isn't problematic if < this */
1259 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1260 || ( UNLIKELY(possible_problems)
1261 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1262 && ((flags & ( UTF8_DISALLOW_NONCHAR
1263 |UTF8_DISALLOW_SURROGATE
1264 |UTF8_DISALLOW_SUPER
1265 |UTF8_DISALLOW_ABOVE_31_BIT
1267 |UTF8_WARN_SURROGATE
1269 |UTF8_WARN_ABOVE_31_BIT))
1270 /* In case of a malformation, 'uv' is not valid, and has
1271 * been changed to something in the Unicode range.
1272 * Currently we don't output a deprecation message if there
1273 * is already a malformation, so we don't have to special
1274 * case the test immediately below */
1275 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1276 && ckWARN_d(WARN_DEPRECATED))))
1278 /* If there were no malformations, or the only malformation is an
1279 * overlong, 'uv' is valid */
1280 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1281 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1282 possible_problems |= UTF8_GOT_SURROGATE;
1284 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1285 possible_problems |= UTF8_GOT_SUPER;
1287 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1288 possible_problems |= UTF8_GOT_NONCHAR;
1291 else { /* Otherwise, need to look at the source UTF-8, possibly
1292 adjusted to be non-overlong */
1294 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1295 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1297 possible_problems |= UTF8_GOT_SUPER;
1299 else if (curlen > 1) {
1300 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1301 NATIVE_UTF8_TO_I8(*adjusted_s0),
1302 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1304 possible_problems |= UTF8_GOT_SUPER;
1306 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1307 NATIVE_UTF8_TO_I8(*adjusted_s0),
1308 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1310 possible_problems |= UTF8_GOT_SURROGATE;
1314 /* We need a complete well-formed UTF-8 character to discern
1315 * non-characters, so can't look for them here */
1319 ready_to_handle_errors:
1322 * curlen contains the number of bytes in the sequence that
1323 * this call should advance the input by.
1324 * avail_len gives the available number of bytes passed in, but
1325 * only if this is less than the expected number of
1326 * bytes, based on the code point's start byte.
1327 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1328 * is set in it for each potential problem found.
1329 * uv contains the code point the input sequence
1330 * represents; or if there is a problem that prevents
1331 * a well-defined value from being computed, it is
1332 * some subsitute value, typically the REPLACEMENT
1334 * s0 points to the first byte of the character
1335 * send points to just after where that (potentially
1336 * partial) character ends
1337 * adjusted_s0 normally is the same as s0, but in case of an
1338 * overlong for which the UTF-8 matters below, it is
1339 * the first byte of the shortest form representation
1341 * adjusted_send normally is the same as 'send', but if adjusted_s0
1342 * is set to something other than s0, this points one
1346 if (UNLIKELY(possible_problems)) {
1347 bool disallowed = FALSE;
1348 const U32 orig_problems = possible_problems;
1350 while (possible_problems) { /* Handle each possible problem */
1352 char * message = NULL;
1354 /* Each 'if' clause handles one problem. They are ordered so that
1355 * the first ones' messages will be displayed before the later
1356 * ones; this is kinda in decreasing severity order */
1357 if (possible_problems & UTF8_GOT_OVERFLOW) {
1359 /* Overflow means also got a super and above 31 bits, but we
1360 * handle all three cases here */
1362 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1363 *errors |= UTF8_GOT_OVERFLOW;
1365 /* But the API says we flag all errors found */
1366 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1367 *errors |= UTF8_GOT_SUPER;
1369 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1370 *errors |= UTF8_GOT_ABOVE_31_BIT;
1375 /* The warnings code explicitly says it doesn't handle the case
1376 * of packWARN2 and two categories which have parent-child
1377 * relationship. Even if it works now to raise the warning if
1378 * either is enabled, it wouldn't necessarily do so in the
1379 * future. We output (only) the most dire warning*/
1380 if (! (flags & UTF8_CHECK_ONLY)) {
1381 if (ckWARN_d(WARN_UTF8)) {
1382 pack_warn = packWARN(WARN_UTF8);
1384 else if (ckWARN_d(WARN_NON_UNICODE)) {
1385 pack_warn = packWARN(WARN_NON_UNICODE);
1388 message = Perl_form(aTHX_ "%s: %s (overflows)",
1390 _byte_dump_string(s0, send - s0));
1394 else if (possible_problems & UTF8_GOT_EMPTY) {
1395 possible_problems &= ~UTF8_GOT_EMPTY;
1396 *errors |= UTF8_GOT_EMPTY;
1398 if (! (flags & UTF8_ALLOW_EMPTY)) {
1400 /* This so-called malformation is now treated as a bug in
1401 * the caller. If you have nothing to decode, skip calling
1406 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1407 pack_warn = packWARN(WARN_UTF8);
1408 message = Perl_form(aTHX_ "%s (empty string)",
1413 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1414 possible_problems &= ~UTF8_GOT_CONTINUATION;
1415 *errors |= UTF8_GOT_CONTINUATION;
1417 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1419 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1420 pack_warn = packWARN(WARN_UTF8);
1421 message = Perl_form(aTHX_
1422 "%s: %s (unexpected continuation byte 0x%02x,"
1423 " with no preceding start byte)",
1425 _byte_dump_string(s0, 1), *s0);
1429 else if (possible_problems & UTF8_GOT_SHORT) {
1430 possible_problems &= ~UTF8_GOT_SHORT;
1431 *errors |= UTF8_GOT_SHORT;
1433 if (! (flags & UTF8_ALLOW_SHORT)) {
1435 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1436 pack_warn = packWARN(WARN_UTF8);
1437 message = Perl_form(aTHX_
1438 "%s: %s (too short; %d byte%s available, need %d)",
1440 _byte_dump_string(s0, send - s0),
1442 avail_len == 1 ? "" : "s",
1448 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1449 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1450 *errors |= UTF8_GOT_NON_CONTINUATION;
1452 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1454 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1455 pack_warn = packWARN(WARN_UTF8);
1456 message = Perl_form(aTHX_ "%s",
1457 unexpected_non_continuation_text(s0,
1464 else if (possible_problems & UTF8_GOT_LONG) {
1465 possible_problems &= ~UTF8_GOT_LONG;
1466 *errors |= UTF8_GOT_LONG;
1468 if (! (flags & UTF8_ALLOW_LONG)) {
1471 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1472 pack_warn = packWARN(WARN_UTF8);
1474 /* These error types cause 'uv' to be something that
1475 * isn't what was intended, so can't use it in the
1476 * message. The other error types either can't
1477 * generate an overlong, or else the 'uv' is valid */
1479 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1481 message = Perl_form(aTHX_
1482 "%s: %s (any UTF-8 sequence that starts"
1483 " with \"%s\" is overlong which can and"
1484 " should be represented with a"
1485 " different, shorter sequence)",
1487 _byte_dump_string(s0, send - s0),
1488 _byte_dump_string(s0, curlen));
1491 U8 tmpbuf[UTF8_MAXBYTES+1];
1492 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1494 message = Perl_form(aTHX_
1495 "%s: %s (overlong; instead use %s to represent"
1498 _byte_dump_string(s0, send - s0),
1499 _byte_dump_string(tmpbuf, e - tmpbuf),
1500 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1501 small code points */
1507 else if (possible_problems & UTF8_GOT_SURROGATE) {
1508 possible_problems &= ~UTF8_GOT_SURROGATE;
1510 if (flags & UTF8_WARN_SURROGATE) {
1511 *errors |= UTF8_GOT_SURROGATE;
1513 if ( ! (flags & UTF8_CHECK_ONLY)
1514 && ckWARN_d(WARN_SURROGATE))
1516 pack_warn = packWARN(WARN_SURROGATE);
1518 /* These are the only errors that can occur with a
1519 * surrogate when the 'uv' isn't valid */
1520 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1521 message = Perl_form(aTHX_
1522 "UTF-16 surrogate (any UTF-8 sequence that"
1523 " starts with \"%s\" is for a surrogate)",
1524 _byte_dump_string(s0, curlen));
1527 message = Perl_form(aTHX_
1528 "UTF-16 surrogate U+%04" UVXf, uv);
1533 if (flags & UTF8_DISALLOW_SURROGATE) {
1535 *errors |= UTF8_GOT_SURROGATE;
1538 else if (possible_problems & UTF8_GOT_SUPER) {
1539 possible_problems &= ~UTF8_GOT_SUPER;
1541 if (flags & UTF8_WARN_SUPER) {
1542 *errors |= UTF8_GOT_SUPER;
1544 if ( ! (flags & UTF8_CHECK_ONLY)
1545 && ckWARN_d(WARN_NON_UNICODE))
1547 pack_warn = packWARN(WARN_NON_UNICODE);
1549 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1550 message = Perl_form(aTHX_
1551 "Any UTF-8 sequence that starts with"
1552 " \"%s\" is for a non-Unicode code point,"
1553 " may not be portable",
1554 _byte_dump_string(s0, curlen));
1557 message = Perl_form(aTHX_
1558 "Code point 0x%04" UVXf " is not"
1559 " Unicode, may not be portable",
1565 /* The maximum code point ever specified by a standard was
1566 * 2**31 - 1. Anything larger than that is a Perl extension
1567 * that very well may not be understood by other applications
1568 * (including earlier perl versions on EBCDIC platforms). We
1569 * test for these after the regular SUPER ones, and before
1570 * possibly bailing out, so that the slightly more dire warning
1571 * will override the regular one. */
1572 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1574 |UTF8_DISALLOW_ABOVE_31_BIT))
1575 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1576 && UNLIKELY(is_utf8_cp_above_31_bits(
1579 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1580 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1582 if ( ! (flags & UTF8_CHECK_ONLY)
1583 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1584 && ckWARN_d(WARN_UTF8))
1586 pack_warn = packWARN(WARN_UTF8);
1588 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1589 message = Perl_form(aTHX_
1590 "Any UTF-8 sequence that starts with"
1591 " \"%s\" is for a non-Unicode code"
1592 " point, and is not portable",
1593 _byte_dump_string(s0, curlen));
1596 message = Perl_form(aTHX_
1597 "Code point 0x%" UVXf " is not Unicode,"
1598 " and not portable",
1603 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1604 *errors |= UTF8_GOT_ABOVE_31_BIT;
1606 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1612 if (flags & UTF8_DISALLOW_SUPER) {
1613 *errors |= UTF8_GOT_SUPER;
1617 /* The deprecated warning overrides any non-deprecated one. If
1618 * there are other problems, a deprecation message is not
1619 * really helpful, so don't bother to raise it in that case.
1620 * This also keeps the code from having to handle the case
1621 * where 'uv' is not valid. */
1622 if ( ! (orig_problems
1623 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1624 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1625 && ckWARN_d(WARN_DEPRECATED))
1627 message = Perl_form(aTHX_ cp_above_legal_max,
1628 uv, MAX_NON_DEPRECATED_CP);
1629 pack_warn = packWARN(WARN_DEPRECATED);
1632 else if (possible_problems & UTF8_GOT_NONCHAR) {
1633 possible_problems &= ~UTF8_GOT_NONCHAR;
1635 if (flags & UTF8_WARN_NONCHAR) {
1636 *errors |= UTF8_GOT_NONCHAR;
1638 if ( ! (flags & UTF8_CHECK_ONLY)
1639 && ckWARN_d(WARN_NONCHAR))
1641 /* The code above should have guaranteed that we don't
1642 * get here with errors other than overlong */
1643 assert (! (orig_problems
1644 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1646 pack_warn = packWARN(WARN_NONCHAR);
1647 message = Perl_form(aTHX_ "Unicode non-character"
1648 " U+%04" UVXf " is not recommended"
1649 " for open interchange", uv);
1653 if (flags & UTF8_DISALLOW_NONCHAR) {
1655 *errors |= UTF8_GOT_NONCHAR;
1657 } /* End of looking through the possible flags */
1659 /* Display the message (if any) for the problem being handled in
1660 * this iteration of the loop */
1663 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1666 Perl_warner(aTHX_ pack_warn, "%s", message);
1668 } /* End of 'while (possible_problems) {' */
1670 /* Since there was a possible problem, the returned length may need to
1671 * be changed from the one stored at the beginning of this function.
1672 * Instead of trying to figure out if that's needed, just do it. */
1678 if (flags & UTF8_CHECK_ONLY && retlen) {
1679 *retlen = ((STRLEN) -1);
1685 return UNI_TO_NATIVE(uv);
1689 =for apidoc utf8_to_uvchr_buf
1691 Returns the native code point of the first character in the string C<s> which
1692 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1693 C<*retlen> will be set to the length, in bytes, of that character.
1695 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1696 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1697 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1698 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1699 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1700 the next possible position in C<s> that could begin a non-malformed character.
1701 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1704 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1705 unless those are turned off.
1709 Also implemented as a macro in utf8.h
1715 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1719 return utf8n_to_uvchr(s, send - s, retlen,
1720 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1723 /* This is marked as deprecated
1725 =for apidoc utf8_to_uvuni_buf
1727 Only in very rare circumstances should code need to be dealing in Unicode
1728 (as opposed to native) code points. In those few cases, use
1729 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1731 Returns the Unicode (not-native) code point of the first character in the
1733 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1734 C<retlen> will be set to the length, in bytes, of that character.
1736 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1737 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1738 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1739 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1740 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1741 next possible position in C<s> that could begin a non-malformed character.
1742 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1744 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1745 unless those are turned off.
1751 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1753 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1757 /* Call the low level routine, asking for checks */
1758 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1762 =for apidoc utf8_length
1764 Return the length of the UTF-8 char encoded string C<s> in characters.
1765 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1766 up past C<e>, croaks.
1772 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1776 PERL_ARGS_ASSERT_UTF8_LENGTH;
1778 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1779 * the bitops (especially ~) can create illegal UTF-8.
1780 * In other words: in Perl UTF-8 is not just for Unicode. */
1783 goto warn_and_return;
1793 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1794 "%s in %s", unees, OP_DESC(PL_op));
1796 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1803 =for apidoc bytes_cmp_utf8
1805 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1806 sequence of characters (stored as UTF-8)
1807 in C<u>, C<ulen>. Returns 0 if they are
1808 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1809 if the first string is greater than the second string.
1811 -1 or +1 is returned if the shorter string was identical to the start of the
1812 longer string. -2 or +2 is returned if
1813 there was a difference between characters
1820 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1822 const U8 *const bend = b + blen;
1823 const U8 *const uend = u + ulen;
1825 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1827 while (b < bend && u < uend) {
1829 if (!UTF8_IS_INVARIANT(c)) {
1830 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1833 if (UTF8_IS_CONTINUATION(c1)) {
1834 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1836 /* diag_listed_as: Malformed UTF-8 character%s */
1837 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1839 unexpected_non_continuation_text(u - 1, 2, 1, 2),
1840 PL_op ? " in " : "",
1841 PL_op ? OP_DESC(PL_op) : "");
1846 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1847 "%s in %s", unees, OP_DESC(PL_op));
1849 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1850 return -2; /* Really want to return undef :-) */
1857 return *b < c ? -2 : +2;
1862 if (b == bend && u == uend)
1865 return b < bend ? +1 : -1;
1869 =for apidoc utf8_to_bytes
1871 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1872 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1873 updates C<len> to contain the new length.
1874 Returns zero on failure, setting C<len> to -1.
1876 If you need a copy of the string, see L</bytes_from_utf8>.
1882 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1884 U8 * const save = s;
1885 U8 * const send = s + *len;
1888 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1889 PERL_UNUSED_CONTEXT;
1891 /* ensure valid UTF-8 and chars < 256 before updating string */
1893 if (! UTF8_IS_INVARIANT(*s)) {
1894 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1895 *len = ((STRLEN) -1);
1906 if (! UTF8_IS_INVARIANT(c)) {
1907 /* Then it is two-byte encoded */
1908 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1919 =for apidoc bytes_from_utf8
1921 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1922 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1923 the newly-created string, and updates C<len> to contain the new
1924 length. Returns the original string if no conversion occurs, C<len>
1925 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1926 0 if C<s> is converted or consisted entirely of characters that are invariant
1927 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1933 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1936 const U8 *start = s;
1940 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1941 PERL_UNUSED_CONTEXT;
1945 /* ensure valid UTF-8 and chars < 256 before converting string */
1946 for (send = s + *len; s < send;) {
1947 if (! UTF8_IS_INVARIANT(*s)) {
1948 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1959 Newx(d, (*len) - count + 1, U8);
1960 s = start; start = d;
1963 if (! UTF8_IS_INVARIANT(c)) {
1964 /* Then it is two-byte encoded */
1965 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1976 =for apidoc bytes_to_utf8
1978 Converts a string C<s> of length C<len> bytes from the native encoding into
1980 Returns a pointer to the newly-created string, and sets C<len> to
1981 reflect the new length in bytes.
1983 A C<NUL> character will be written after the end of the string.
1985 If you want to convert to UTF-8 from encodings other than
1986 the native (Latin1 or EBCDIC),
1987 see L</sv_recode_to_utf8>().
1992 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1993 likewise need duplication. */
1996 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1998 const U8 * const send = s + (*len);
2002 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2003 PERL_UNUSED_CONTEXT;
2005 Newx(d, (*len) * 2 + 1, U8);
2009 append_utf8_from_native_byte(*s, &d);
2018 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2020 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2021 * We optimize for native, for obvious reasons. */
2024 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2029 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2032 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
2037 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2039 if (OFFUNI_IS_INVARIANT(uv)) {
2040 *d++ = LATIN1_TO_NATIVE((U8) uv);
2043 if (uv <= MAX_UTF8_TWO_BYTE) {
2044 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2045 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2048 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2049 #define LAST_HIGH_SURROGATE 0xDBFF
2050 #define FIRST_LOW_SURROGATE 0xDC00
2051 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2053 /* This assumes that most uses will be in the first Unicode plane, not
2054 * needing surrogates */
2055 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2056 && uv <= UNICODE_SURROGATE_LAST))
2058 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2059 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2062 UV low = (p[0] << 8) + p[1];
2063 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2064 || UNLIKELY(low > LAST_LOW_SURROGATE))
2066 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2069 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2070 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2074 d = uvoffuni_to_utf8_flags(d, uv, 0);
2077 *d++ = (U8)(( uv >> 12) | 0xe0);
2078 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2079 *d++ = (U8)(( uv & 0x3f) | 0x80);
2083 *d++ = (U8)(( uv >> 18) | 0xf0);
2084 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2085 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2086 *d++ = (U8)(( uv & 0x3f) | 0x80);
2091 *newlen = d - dstart;
2095 /* Note: this one is slightly destructive of the source. */
2098 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2101 U8* const send = s + bytelen;
2103 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2106 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2110 const U8 tmp = s[0];
2115 return utf16_to_utf8(p, d, bytelen, newlen);
2119 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2121 U8 tmpbuf[UTF8_MAXBYTES+1];
2122 uvchr_to_utf8(tmpbuf, c);
2123 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2126 /* Internal function so we can deprecate the external one, and call
2127 this one from other deprecated functions in this file */
2130 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2132 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2136 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2140 Perl__is_uni_perl_idcont(pTHX_ UV c)
2142 U8 tmpbuf[UTF8_MAXBYTES+1];
2143 uvchr_to_utf8(tmpbuf, c);
2144 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2148 Perl__is_uni_perl_idstart(pTHX_ UV c)
2150 U8 tmpbuf[UTF8_MAXBYTES+1];
2151 uvchr_to_utf8(tmpbuf, c);
2152 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2156 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2158 /* We have the latin1-range values compiled into the core, so just use
2159 * those, converting the result to UTF-8. The only difference between upper
2160 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2161 * either "SS" or "Ss". Which one to use is passed into the routine in
2162 * 'S_or_s' to avoid a test */
2164 UV converted = toUPPER_LATIN1_MOD(c);
2166 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2168 assert(S_or_s == 'S' || S_or_s == 's');
2170 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2171 characters in this range */
2172 *p = (U8) converted;
2177 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2178 * which it maps to one of them, so as to only have to have one check for
2179 * it in the main case */
2180 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2182 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2183 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2186 converted = GREEK_CAPITAL_LETTER_MU;
2188 #if UNICODE_MAJOR_VERSION > 2 \
2189 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2190 && UNICODE_DOT_DOT_VERSION >= 8)
2191 case LATIN_SMALL_LETTER_SHARP_S:
2198 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2199 NOT_REACHED; /* NOTREACHED */
2203 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2204 *p = UTF8_TWO_BYTE_LO(converted);
2210 /* Call the function to convert a UTF-8 encoded character to the specified case.
2211 * Note that there may be more than one character in the result.
2212 * INP is a pointer to the first byte of the input character
2213 * OUTP will be set to the first byte of the string of changed characters. It
2214 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2215 * LENP will be set to the length in bytes of the string of changed characters
2217 * The functions return the ordinal of the first character in the string of OUTP */
2218 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2219 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2220 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2222 /* This additionally has the input parameter 'specials', which if non-zero will
2223 * cause this to use the specials hash for folding (meaning get full case
2224 * folding); otherwise, when zero, this implies a simple case fold */
2225 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2228 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2230 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2231 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2232 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2233 * the changed version may be longer than the original character.
2235 * The ordinal of the first character of the changed version is returned
2236 * (but note, as explained above, that there may be more.) */
2238 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2241 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2244 uvchr_to_utf8(p, c);
2245 return CALL_UPPER_CASE(c, p, p, lenp);
2249 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2251 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2254 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2257 uvchr_to_utf8(p, c);
2258 return CALL_TITLE_CASE(c, p, p, lenp);
2262 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2264 /* We have the latin1-range values compiled into the core, so just use
2265 * those, converting the result to UTF-8. Since the result is always just
2266 * one character, we allow <p> to be NULL */
2268 U8 converted = toLOWER_LATIN1(c);
2270 PERL_UNUSED_ARG(dummy);
2273 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2278 /* Result is known to always be < 256, so can use the EIGHT_BIT
2280 *p = UTF8_EIGHT_BIT_HI(converted);
2281 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2289 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2291 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2294 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2297 uvchr_to_utf8(p, c);
2298 return CALL_LOWER_CASE(c, p, p, lenp);
2302 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2304 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2305 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2306 * FOLD_FLAGS_FULL iff full folding is to be used;
2308 * Not to be used for locale folds
2313 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2314 PERL_UNUSED_CONTEXT;
2316 assert (! (flags & FOLD_FLAGS_LOCALE));
2318 if (UNLIKELY(c == MICRO_SIGN)) {
2319 converted = GREEK_SMALL_LETTER_MU;
2321 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2322 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2323 || UNICODE_DOT_DOT_VERSION > 0)
2324 else if ( (flags & FOLD_FLAGS_FULL)
2325 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2327 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2328 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2329 * under those circumstances. */
2330 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2331 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2332 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2334 return LATIN_SMALL_LETTER_LONG_S;
2344 else { /* In this range the fold of all other characters is their lower
2346 converted = toLOWER_LATIN1(c);
2349 if (UVCHR_IS_INVARIANT(converted)) {
2350 *p = (U8) converted;
2354 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2355 *p = UTF8_TWO_BYTE_LO(converted);
2363 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2366 /* Not currently externally documented, and subject to change
2367 * <flags> bits meanings:
2368 * FOLD_FLAGS_FULL iff full folding is to be used;
2369 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2370 * locale are to be used.
2371 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2374 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2376 if (flags & FOLD_FLAGS_LOCALE) {
2377 /* Treat a UTF-8 locale as not being in locale at all */
2378 if (IN_UTF8_CTYPE_LOCALE) {
2379 flags &= ~FOLD_FLAGS_LOCALE;
2382 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2383 goto needs_full_generality;
2388 return _to_fold_latin1((U8) c, p, lenp,
2389 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2392 /* Here, above 255. If no special needs, just use the macro */
2393 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2394 uvchr_to_utf8(p, c);
2395 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2397 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2398 the special flags. */
2399 U8 utf8_c[UTF8_MAXBYTES + 1];
2401 needs_full_generality:
2402 uvchr_to_utf8(utf8_c, c);
2403 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
2407 PERL_STATIC_INLINE bool
2408 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2409 const char *const swashname, SV* const invlist)
2411 /* returns a boolean giving whether or not the UTF8-encoded character that
2412 * starts at <p> is in the swash indicated by <swashname>. <swash>
2413 * contains a pointer to where the swash indicated by <swashname>
2414 * is to be stored; which this routine will do, so that future calls will
2415 * look at <*swash> and only generate a swash if it is not null. <invlist>
2416 * is NULL or an inversion list that defines the swash. If not null, it
2417 * saves time during initialization of the swash.
2419 * Note that it is assumed that the buffer length of <p> is enough to
2420 * contain all the bytes that comprise the character. Thus, <*p> should
2421 * have been checked before this call for mal-formedness enough to assure
2424 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2426 /* The API should have included a length for the UTF-8 character in <p>,
2427 * but it doesn't. We therefore assume that p has been validated at least
2428 * as far as there being enough bytes available in it to accommodate the
2429 * character without reading beyond the end, and pass that number on to the
2430 * validating routine */
2431 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2432 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2435 NOT_REACHED; /* NOTREACHED */
2439 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2440 *swash = _core_swash_init("utf8",
2442 /* Only use the name if there is no inversion
2443 * list; otherwise will go out to disk */
2444 (invlist) ? "" : swashname,
2446 &PL_sv_undef, 1, 0, invlist, &flags);
2449 return swash_fetch(*swash, p, TRUE) != 0;
2452 PERL_STATIC_INLINE bool
2453 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e, SV **swash,
2454 const char *const swashname, SV* const invlist)
2456 /* returns a boolean giving whether or not the UTF8-encoded character that
2457 * starts at <p>, and extending no further than <e - 1> is in the swash
2458 * indicated by <swashname>. <swash> contains a pointer to where the swash
2459 * indicated by <swashname> is to be stored; which this routine will do, so
2460 * that future calls will look at <*swash> and only generate a swash if it
2461 * is not null. <invlist> is NULL or an inversion list that defines the
2462 * swash. If not null, it saves time during initialization of the swash.
2465 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2467 if (! isUTF8_CHAR(p, e)) {
2468 _force_out_malformed_utf8_message(p, e, 0, 1);
2469 NOT_REACHED; /* NOTREACHED */
2473 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2474 *swash = _core_swash_init("utf8",
2476 /* Only use the name if there is no inversion
2477 * list; otherwise will go out to disk */
2478 (invlist) ? "" : swashname,
2480 &PL_sv_undef, 1, 0, invlist, &flags);
2483 return swash_fetch(*swash, p, TRUE) != 0;
2487 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2489 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2491 assert(classnum < _FIRST_NON_SWASH_CC);
2493 return is_utf8_common(p,
2494 &PL_utf8_swash_ptrs[classnum],
2495 swash_property_names[classnum],
2496 PL_XPosix_ptrs[classnum]);
2500 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2503 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2505 assert(classnum < _FIRST_NON_SWASH_CC);
2507 return is_utf8_common_with_len(p,
2509 &PL_utf8_swash_ptrs[classnum],
2510 swash_property_names[classnum],
2511 PL_XPosix_ptrs[classnum]);
2515 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
2519 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
2521 if (! PL_utf8_perl_idstart) {
2522 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2524 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
2528 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2532 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2534 if (! PL_utf8_perl_idstart) {
2535 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2537 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2538 "_Perl_IDStart", invlist);
2542 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2544 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2548 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2552 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2556 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2558 if (! PL_utf8_perl_idcont) {
2559 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2561 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
2565 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2569 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2571 if (! PL_utf8_perl_idcont) {
2572 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2574 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2575 "_Perl_IDCont", invlist);
2579 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2581 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2583 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2587 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2589 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2591 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2595 Perl__is_utf8_mark(pTHX_ const U8 *p)
2597 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2599 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2603 =for apidoc to_utf8_case
2605 Instead use the appropriate one of L</toUPPER_utf8>,
2610 C<p> contains the pointer to the UTF-8 string encoding
2611 the character that is being converted. This routine assumes that the character
2612 at C<p> is well-formed.
2614 C<ustrp> is a pointer to the character buffer to put the
2615 conversion result to. C<lenp> is a pointer to the length
2618 C<swashp> is a pointer to the swash to use.
2620 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2621 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
2622 but not always, a multicharacter mapping), is tried first.
2624 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2625 any special mappings; C<""> means to use the special mappings. Values other
2626 than these two are treated as the name of the hash containing the special
2627 mappings, like C<"utf8::ToSpecLower">.
2629 C<normal> is a string like C<"ToLower"> which means the swash
2632 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2633 unless those are turned off.
2638 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2639 SV **swashp, const char *normal, const char *special)
2641 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2643 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2646 /* change namve uv1 to 'from' */
2648 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2649 SV **swashp, const char *normal, const char *special)
2653 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2655 /* For code points that don't change case, we already know that the output
2656 * of this function is the unchanged input, so we can skip doing look-ups
2657 * for them. Unfortunately the case-changing code points are scattered
2658 * around. But there are some long consecutive ranges where there are no
2659 * case changing code points. By adding tests, we can eliminate the lookup
2660 * for all the ones in such ranges. This is currently done here only for
2661 * just a few cases where the scripts are in common use in modern commerce
2662 * (and scripts adjacent to those which can be included without additional
2665 if (uv1 >= 0x0590) {
2666 /* This keeps from needing further processing the code points most
2667 * likely to be used in the following non-cased scripts: Hebrew,
2668 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2669 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2670 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2675 /* The following largish code point ranges also don't have case
2676 * changes, but khw didn't think they warranted extra tests to speed
2677 * them up (which would slightly slow down everything else above them):
2678 * 1100..139F Hangul Jamo, Ethiopic
2679 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2680 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2681 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2682 * Combining Diacritical Marks Extended, Balinese,
2683 * Sundanese, Batak, Lepcha, Ol Chiki
2684 * 2000..206F General Punctuation
2687 if (uv1 >= 0x2D30) {
2689 /* This keeps the from needing further processing the code points
2690 * most likely to be used in the following non-cased major scripts:
2691 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2693 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2694 * event that Unicode eventually allocates the unused block as of
2695 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2696 * that the test suite will start having failures to alert you
2697 * should that happen) */
2702 if (uv1 >= 0xAC00) {
2703 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2704 if (ckWARN_d(WARN_SURROGATE)) {
2705 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2706 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2707 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2712 /* AC00..FAFF Catches Hangul syllables and private use, plus
2719 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2720 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2721 && ckWARN_d(WARN_DEPRECATED))
2723 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2724 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2726 if (ckWARN_d(WARN_NON_UNICODE)) {
2727 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2728 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2729 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2733 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2735 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2738 /* As of this writing, this means we avoid swash creation
2739 * for anything beyond low Plane 1 */
2746 /* Note that non-characters are perfectly legal, so no warning should
2747 * be given. There are so few of them, that it isn't worth the extra
2748 * tests to avoid swash creation */
2751 if (!*swashp) /* load on-demand */
2752 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2755 /* It might be "special" (sometimes, but not always,
2756 * a multicharacter mapping) */
2760 /* If passed in the specials name, use that; otherwise use any
2761 * given in the swash */
2762 if (*special != '\0') {
2763 hv = get_hv(special, 0);
2766 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2768 hv = MUTABLE_HV(SvRV(*svp));
2773 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2778 s = SvPV_const(*svp, len);
2781 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2783 Copy(s, ustrp, len, U8);
2788 if (!len && *swashp) {
2789 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2792 /* It was "normal" (a single character mapping). */
2793 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2801 return valid_utf8_to_uvchr(ustrp, 0);
2804 /* Here, there was no mapping defined, which means that the code point maps
2805 * to itself. Return the inputs */
2808 if (p != ustrp) { /* Don't copy onto itself */
2809 Copy(p, ustrp, len, U8);
2820 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2822 /* This is called when changing the case of a UTF-8-encoded character above
2823 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2824 * result contains a character that crosses the 255/256 boundary, disallow
2825 * the change, and return the original code point. See L<perlfunc/lc> for
2828 * p points to the original string whose case was changed; assumed
2829 * by this routine to be well-formed
2830 * result the code point of the first character in the changed-case string
2831 * ustrp points to the changed-case string (<result> represents its first char)
2832 * lenp points to the length of <ustrp> */
2834 UV original; /* To store the first code point of <p> */
2836 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2838 assert(UTF8_IS_ABOVE_LATIN1(*p));
2840 /* We know immediately if the first character in the string crosses the
2841 * boundary, so can skip */
2844 /* Look at every character in the result; if any cross the
2845 * boundary, the whole thing is disallowed */
2846 U8* s = ustrp + UTF8SKIP(ustrp);
2847 U8* e = ustrp + *lenp;
2849 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2855 /* Here, no characters crossed, result is ok as-is, but we warn. */
2856 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
2862 /* Failed, have to return the original */
2863 original = valid_utf8_to_uvchr(p, lenp);
2865 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2866 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2867 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2868 "resolved to \"\\x{%" UVXf "}\".",
2872 Copy(p, ustrp, *lenp, char);
2876 /* The process for changing the case is essentially the same for the four case
2877 * change types, except there are complications for folding. Otherwise the
2878 * difference is only which case to change to. To make sure that they all do
2879 * the same thing, the bodies of the functions are extracted out into the
2880 * following two macros. The functions are written with the same variable
2881 * names, and these are known and used inside these macros. It would be
2882 * better, of course, to have inline functions to do it, but since different
2883 * macros are called, depending on which case is being changed to, this is not
2884 * feasible in C (to khw's knowledge). Two macros are created so that the fold
2885 * function can start with the common start macro, then finish with its special
2886 * handling; while the other three cases can just use the common end macro.
2888 * The algorithm is to use the proper (passed in) macro or function to change
2889 * the case for code points that are below 256. The macro is used if using
2890 * locale rules for the case change; the function if not. If the code point is
2891 * above 255, it is computed from the input UTF-8, and another macro is called
2892 * to do the conversion. If necessary, the output is converted to UTF-8. If
2893 * using a locale, we have to check that the change did not cross the 255/256
2894 * boundary, see check_locale_boundary_crossing() for further details.
2896 * The macros are split with the correct case change for the below-256 case
2897 * stored into 'result', and in the middle of an else clause for the above-255
2898 * case. At that point in the 'else', 'result' is not the final result, but is
2899 * the input code point calculated from the UTF-8. The fold code needs to
2900 * realize all this and take it from there.
2902 * If you read the two macros as sequential, it's easier to understand what's
2904 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
2905 L1_func_extra_param) \
2906 if (flags & (locale_flags)) { \
2907 /* Treat a UTF-8 locale as not being in locale at all */ \
2908 if (IN_UTF8_CTYPE_LOCALE) { \
2909 flags &= ~(locale_flags); \
2912 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
2916 if (UTF8_IS_INVARIANT(*p)) { \
2917 if (flags & (locale_flags)) { \
2918 result = LC_L1_change_macro(*p); \
2921 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
2924 else if UTF8_IS_DOWNGRADEABLE_START(*p) { \
2925 if (flags & (locale_flags)) { \
2926 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
2930 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
2931 ustrp, lenp, L1_func_extra_param); \
2934 else { /* malformed UTF-8 */ \
2935 result = valid_utf8_to_uvchr(p, NULL); \
2937 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
2938 result = change_macro(result, p, ustrp, lenp); \
2940 if (flags & (locale_flags)) { \
2941 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
2946 /* Here, used locale rules. Convert back to UTF-8 */ \
2947 if (UTF8_IS_INVARIANT(result)) { \
2948 *ustrp = (U8) result; \
2952 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
2953 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
2960 =for apidoc to_utf8_upper
2962 Instead use L</toUPPER_utf8>.
2966 /* Not currently externally documented, and subject to change:
2967 * <flags> is set iff iff the rules from the current underlying locale are to
2971 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2975 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2977 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
2978 /* 2nd char of uc(U+DF) is 'S' */
2979 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
2980 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
2984 =for apidoc to_utf8_title
2986 Instead use L</toTITLE_utf8>.
2990 /* Not currently externally documented, and subject to change:
2991 * <flags> is set iff the rules from the current underlying locale are to be
2992 * used. Since titlecase is not defined in POSIX, for other than a
2993 * UTF-8 locale, uppercase is used instead for code points < 256.
2997 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
3001 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3003 /* 2nd char of ucfirst(U+DF) is 's' */
3004 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3005 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3009 =for apidoc to_utf8_lower
3011 Instead use L</toLOWER_utf8>.
3015 /* Not currently externally documented, and subject to change:
3016 * <flags> is set iff iff the rules from the current underlying locale are to
3021 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
3025 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3027 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3028 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3032 =for apidoc to_utf8_fold
3034 Instead use L</toFOLD_utf8>.
3038 /* Not currently externally documented, and subject to change,
3040 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3041 * locale are to be used.
3042 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3043 * otherwise simple folds
3044 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3049 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
3053 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3055 /* These are mutually exclusive */
3056 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3058 assert(p != ustrp); /* Otherwise overwrites */
3060 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3061 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3063 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3065 if (flags & FOLD_FLAGS_LOCALE) {
3067 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3068 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3070 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3071 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3073 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3075 /* Special case these two characters, as what normally gets
3076 * returned under locale doesn't work */
3077 if (UTF8SKIP(p) == cap_sharp_s_len
3078 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3080 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3081 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3082 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3083 "resolved to \"\\x{17F}\\x{17F}\".");
3088 if (UTF8SKIP(p) == long_s_t_len
3089 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3091 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3092 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3093 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3094 "resolved to \"\\x{FB06}\".");
3095 goto return_ligature_st;
3098 #if UNICODE_MAJOR_VERSION == 3 \
3099 && UNICODE_DOT_VERSION == 0 \
3100 && UNICODE_DOT_DOT_VERSION == 1
3101 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3103 /* And special case this on this Unicode version only, for the same
3104 * reaons the other two are special cased. They would cross the
3105 * 255/256 boundary which is forbidden under /l, and so the code
3106 * wouldn't catch that they are equivalent (which they are only in
3108 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3109 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3111 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3112 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3113 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3114 "resolved to \"\\x{0131}\".");
3115 goto return_dotless_i;
3119 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3121 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3125 /* This is called when changing the case of a UTF-8-encoded
3126 * character above the ASCII range, and the result should not
3127 * contain an ASCII character. */
3129 UV original; /* To store the first code point of <p> */
3131 /* Look at every character in the result; if any cross the
3132 * boundary, the whole thing is disallowed */
3134 U8* e = ustrp + *lenp;
3137 /* Crossed, have to return the original */
3138 original = valid_utf8_to_uvchr(p, lenp);
3140 /* But in these instances, there is an alternative we can
3141 * return that is valid */
3142 if (original == LATIN_SMALL_LETTER_SHARP_S
3143 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3144 || original == LATIN_CAPITAL_LETTER_SHARP_S
3149 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3150 goto return_ligature_st;
3152 #if UNICODE_MAJOR_VERSION == 3 \
3153 && UNICODE_DOT_VERSION == 0 \
3154 && UNICODE_DOT_DOT_VERSION == 1
3156 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3157 goto return_dotless_i;
3160 Copy(p, ustrp, *lenp, char);
3166 /* Here, no characters crossed, result is ok as-is */
3171 /* Here, used locale rules. Convert back to UTF-8 */
3172 if (UTF8_IS_INVARIANT(result)) {
3173 *ustrp = (U8) result;
3177 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3178 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3185 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3186 * folds to a string of two of these characters. By returning this
3187 * instead, then, e.g.,
3188 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3191 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3192 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3194 return LATIN_SMALL_LETTER_LONG_S;
3197 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3198 * have the other one fold to it */
3200 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3201 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3202 return LATIN_SMALL_LIGATURE_ST;
3204 #if UNICODE_MAJOR_VERSION == 3 \
3205 && UNICODE_DOT_VERSION == 0 \
3206 && UNICODE_DOT_DOT_VERSION == 1
3209 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3210 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3211 return LATIN_SMALL_LETTER_DOTLESS_I;
3218 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3219 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3220 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3224 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3226 PERL_ARGS_ASSERT_SWASH_INIT;
3228 /* Returns a copy of a swash initiated by the called function. This is the
3229 * public interface, and returning a copy prevents others from doing
3230 * mischief on the original */
3232 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3236 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3239 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3240 * use the following define */
3242 #define CORE_SWASH_INIT_RETURN(x) \
3243 PL_curpm= old_PL_curpm; \
3246 /* Initialize and return a swash, creating it if necessary. It does this
3247 * by calling utf8_heavy.pl in the general case. The returned value may be
3248 * the swash's inversion list instead if the input parameters allow it.
3249 * Which is returned should be immaterial to callers, as the only
3250 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3251 * and swash_to_invlist() handle both these transparently.
3253 * This interface should only be used by functions that won't destroy or
3254 * adversely change the swash, as doing so affects all other uses of the
3255 * swash in the program; the general public should use 'Perl_swash_init'
3258 * pkg is the name of the package that <name> should be in.
3259 * name is the name of the swash to find. Typically it is a Unicode
3260 * property name, including user-defined ones
3261 * listsv is a string to initialize the swash with. It must be of the form
3262 * documented as the subroutine return value in
3263 * L<perlunicode/User-Defined Character Properties>
3264 * minbits is the number of bits required to represent each data element.
3265 * It is '1' for binary properties.
3266 * none I (khw) do not understand this one, but it is used only in tr///.
3267 * invlist is an inversion list to initialize the swash with (or NULL)
3268 * flags_p if non-NULL is the address of various input and output flag bits
3269 * to the routine, as follows: ('I' means is input to the routine;
3270 * 'O' means output from the routine. Only flags marked O are
3271 * meaningful on return.)
3272 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3273 * came from a user-defined property. (I O)
3274 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3275 * when the swash cannot be located, to simply return NULL. (I)
3276 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3277 * return of an inversion list instead of a swash hash if this routine
3278 * thinks that would result in faster execution of swash_fetch() later
3281 * Thus there are three possible inputs to find the swash: <name>,
3282 * <listsv>, and <invlist>. At least one must be specified. The result
3283 * will be the union of the specified ones, although <listsv>'s various
3284 * actions can intersect, etc. what <name> gives. To avoid going out to
3285 * disk at all, <invlist> should specify completely what the swash should
3286 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3288 * <invlist> is only valid for binary properties */
3290 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3292 SV* retval = &PL_sv_undef;
3293 HV* swash_hv = NULL;
3294 const int invlist_swash_boundary =
3295 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3296 ? 512 /* Based on some benchmarking, but not extensive, see commit
3298 : -1; /* Never return just an inversion list */
3300 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3301 assert(! invlist || minbits == 1);
3303 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3304 that triggered the swash init and the swash init perl logic itself.
3307 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3309 if (listsv != &PL_sv_undef || strNE(name, "")) {
3311 const size_t pkg_len = strlen(pkg);
3312 const size_t name_len = strlen(name);
3313 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3317 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3319 PUSHSTACKi(PERLSI_MAGIC);
3323 /* We might get here via a subroutine signature which uses a utf8
3324 * parameter name, at which point PL_subname will have been set
3325 * but not yet used. */
3326 save_item(PL_subname);
3327 if (PL_parser && PL_parser->error_count)
3328 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3329 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3330 if (!method) { /* demand load UTF-8 */
3332 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3333 GvSV(PL_errgv) = NULL;
3334 #ifndef NO_TAINT_SUPPORT
3335 /* It is assumed that callers of this routine are not passing in
3336 * any user derived data. */
3337 /* Need to do this after save_re_context() as it will set
3338 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3339 * in Perl_magic_get). Even line to create errsv_save can turn on
3341 SAVEBOOL(TAINT_get);
3344 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3347 /* Not ERRSV, as there is no need to vivify a scalar we are
3348 about to discard. */
3349 SV * const errsv = GvSV(PL_errgv);
3350 if (!SvTRUE(errsv)) {
3351 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3352 SvREFCNT_dec(errsv);
3360 mPUSHp(pkg, pkg_len);
3361 mPUSHp(name, name_len);
3366 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3367 GvSV(PL_errgv) = NULL;
3368 /* If we already have a pointer to the method, no need to use
3369 * call_method() to repeat the lookup. */
3371 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3372 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3374 retval = *PL_stack_sp--;
3375 SvREFCNT_inc(retval);
3378 /* Not ERRSV. See above. */
3379 SV * const errsv = GvSV(PL_errgv);
3380 if (!SvTRUE(errsv)) {
3381 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3382 SvREFCNT_dec(errsv);
3387 if (IN_PERL_COMPILETIME) {
3388 CopHINTS_set(PL_curcop, PL_hints);
3390 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3391 if (SvPOK(retval)) {
3393 /* If caller wants to handle missing properties, let them */
3394 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3395 CORE_SWASH_INIT_RETURN(NULL);
3398 "Can't find Unicode property definition \"%" SVf "\"",
3400 NOT_REACHED; /* NOTREACHED */
3403 } /* End of calling the module to find the swash */
3405 /* If this operation fetched a swash, and we will need it later, get it */
3406 if (retval != &PL_sv_undef
3407 && (minbits == 1 || (flags_p
3409 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3411 swash_hv = MUTABLE_HV(SvRV(retval));
3413 /* If we don't already know that there is a user-defined component to
3414 * this swash, and the user has indicated they wish to know if there is
3415 * one (by passing <flags_p>), find out */
3416 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3417 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3418 if (user_defined && SvUV(*user_defined)) {
3419 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3424 /* Make sure there is an inversion list for binary properties */
3426 SV** swash_invlistsvp = NULL;
3427 SV* swash_invlist = NULL;
3428 bool invlist_in_swash_is_valid = FALSE;
3429 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3430 an unclaimed reference count */
3432 /* If this operation fetched a swash, get its already existing
3433 * inversion list, or create one for it */
3436 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3437 if (swash_invlistsvp) {
3438 swash_invlist = *swash_invlistsvp;
3439 invlist_in_swash_is_valid = TRUE;
3442 swash_invlist = _swash_to_invlist(retval);
3443 swash_invlist_unclaimed = TRUE;
3447 /* If an inversion list was passed in, have to include it */
3450 /* Any fetched swash will by now have an inversion list in it;
3451 * otherwise <swash_invlist> will be NULL, indicating that we
3452 * didn't fetch a swash */
3453 if (swash_invlist) {
3455 /* Add the passed-in inversion list, which invalidates the one
3456 * already stored in the swash */
3457 invlist_in_swash_is_valid = FALSE;
3458 SvREADONLY_off(swash_invlist); /* Turned on again below */
3459 _invlist_union(invlist, swash_invlist, &swash_invlist);
3463 /* Here, there is no swash already. Set up a minimal one, if
3464 * we are going to return a swash */
3465 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3467 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3469 swash_invlist = invlist;
3473 /* Here, we have computed the union of all the passed-in data. It may
3474 * be that there was an inversion list in the swash which didn't get
3475 * touched; otherwise save the computed one */
3476 if (! invlist_in_swash_is_valid
3477 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3479 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3481 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3483 /* We just stole a reference count. */
3484 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3485 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3488 /* The result is immutable. Forbid attempts to change it. */
3489 SvREADONLY_on(swash_invlist);
3491 /* Use the inversion list stand-alone if small enough */
3492 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3493 SvREFCNT_dec(retval);
3494 if (!swash_invlist_unclaimed)
3495 SvREFCNT_inc_simple_void_NN(swash_invlist);
3496 retval = newRV_noinc(swash_invlist);
3500 CORE_SWASH_INIT_RETURN(retval);
3501 #undef CORE_SWASH_INIT_RETURN
3505 /* This API is wrong for special case conversions since we may need to
3506 * return several Unicode characters for a single Unicode character
3507 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3508 * the lower-level routine, and it is similarly broken for returning
3509 * multiple values. --jhi
3510 * For those, you should use S__to_utf8_case() instead */
3511 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3514 * Returns the value of property/mapping C<swash> for the first character
3515 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3516 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3517 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3519 * A "swash" is a hash which contains initially the keys/values set up by
3520 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3521 * property for all possible code points. Things are stored in a compact form
3522 * (see utf8_heavy.pl) so that calculation is required to find the actual
3523 * property value for a given code point. As code points are looked up, new
3524 * key/value pairs are added to the hash, so that the calculation doesn't have
3525 * to ever be re-done. Further, each calculation is done, not just for the
3526 * desired one, but for a whole block of code points adjacent to that one.
3527 * For binary properties on ASCII machines, the block is usually for 64 code
3528 * points, starting with a code point evenly divisible by 64. Thus if the
3529 * property value for code point 257 is requested, the code goes out and
3530 * calculates the property values for all 64 code points between 256 and 319,
3531 * and stores these as a single 64-bit long bit vector, called a "swatch",
3532 * under the key for code point 256. The key is the UTF-8 encoding for code
3533 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3534 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3535 * for code point 258 is then requested, this code realizes that it would be
3536 * stored under the key for 256, and would find that value and extract the
3537 * relevant bit, offset from 256.
3539 * Non-binary properties are stored in as many bits as necessary to represent
3540 * their values (32 currently, though the code is more general than that), not
3541 * as single bits, but the principle is the same: the value for each key is a
3542 * vector that encompasses the property values for all code points whose UTF-8
3543 * representations are represented by the key. That is, for all code points
3544 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3548 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3550 HV *const hv = MUTABLE_HV(SvRV(swash));
3555 const U8 *tmps = NULL;
3559 PERL_ARGS_ASSERT_SWASH_FETCH;
3561 /* If it really isn't a hash, it isn't really swash; must be an inversion
3563 if (SvTYPE(hv) != SVt_PVHV) {
3564 return _invlist_contains_cp((SV*)hv,
3566 ? valid_utf8_to_uvchr(ptr, NULL)
3570 /* We store the values in a "swatch" which is a vec() value in a swash
3571 * hash. Code points 0-255 are a single vec() stored with key length
3572 * (klen) 0. All other code points have a UTF-8 representation
3573 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3574 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3575 * length for them is the length of the encoded char - 1. ptr[klen] is the
3576 * final byte in the sequence representing the character */
3577 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3582 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3585 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3588 klen = UTF8SKIP(ptr) - 1;
3590 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3591 * the vec is the final byte in the sequence. (In EBCDIC this is
3592 * converted to I8 to get consecutive values.) To help you visualize
3594 * Straight 1047 After final byte
3595 * UTF-8 UTF-EBCDIC I8 transform
3596 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3597 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3599 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3600 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3602 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3603 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3605 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3606 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3608 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3609 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3611 * (There are no discontinuities in the elided (...) entries.)
3612 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3613 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3614 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3615 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3616 * index into the vec() swatch (after subtracting 0x80, which we
3617 * actually do with an '&').
3618 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3619 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3620 * dicontinuities which go away by transforming it into I8, and we
3621 * effectively subtract 0xA0 to get the index. */
3622 needents = (1 << UTF_ACCUMULATION_SHIFT);
3623 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3627 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3628 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3629 * it's nothing to sniff at.) Pity we usually come through at least
3630 * two function calls to get here...
3632 * NB: this code assumes that swatches are never modified, once generated!
3635 if (hv == PL_last_swash_hv &&
3636 klen == PL_last_swash_klen &&
3637 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3639 tmps = PL_last_swash_tmps;
3640 slen = PL_last_swash_slen;
3643 /* Try our second-level swatch cache, kept in a hash. */
3644 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3646 /* If not cached, generate it via swatch_get */
3647 if (!svp || !SvPOK(*svp)
3648 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3651 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3652 swatch = swatch_get(swash,
3653 code_point & ~((UV)needents - 1),
3656 else { /* For the first 256 code points, the swatch has a key of
3658 swatch = swatch_get(swash, 0, needents);
3661 if (IN_PERL_COMPILETIME)
3662 CopHINTS_set(PL_curcop, PL_hints);
3664 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3666 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3667 || (slen << 3) < needents)
3668 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3669 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3670 svp, tmps, (UV)slen, (UV)needents);
3673 PL_last_swash_hv = hv;
3674 assert(klen <= sizeof(PL_last_swash_key));
3675 PL_last_swash_klen = (U8)klen;
3676 /* FIXME change interpvar.h? */
3677 PL_last_swash_tmps = (U8 *) tmps;
3678 PL_last_swash_slen = slen;
3680 Copy(ptr, PL_last_swash_key, klen, U8);
3683 switch ((int)((slen << 3) / needents)) {
3685 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3687 return ((UV) tmps[off]);
3691 ((UV) tmps[off ] << 8) +
3692 ((UV) tmps[off + 1]);
3696 ((UV) tmps[off ] << 24) +
3697 ((UV) tmps[off + 1] << 16) +
3698 ((UV) tmps[off + 2] << 8) +
3699 ((UV) tmps[off + 3]);
3701 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3702 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
3703 NORETURN_FUNCTION_END;
3706 /* Read a single line of the main body of the swash input text. These are of
3709 * where each number is hex. The first two numbers form the minimum and
3710 * maximum of a range, and the third is the value associated with the range.
3711 * Not all swashes should have a third number
3713 * On input: l points to the beginning of the line to be examined; it points
3714 * to somewhere in the string of the whole input text, and is
3715 * terminated by a \n or the null string terminator.
3716 * lend points to the null terminator of that string
3717 * wants_value is non-zero if the swash expects a third number
3718 * typestr is the name of the swash's mapping, like 'ToLower'
3719 * On output: *min, *max, and *val are set to the values read from the line.
3720 * returns a pointer just beyond the line examined. If there was no
3721 * valid min number on the line, returns lend+1
3725 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3726 const bool wants_value, const U8* const typestr)
3728 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3729 STRLEN numlen; /* Length of the number */
3730 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3731 | PERL_SCAN_DISALLOW_PREFIX
3732 | PERL_SCAN_SILENT_NON_PORTABLE;
3734 /* nl points to the next \n in the scan */
3735 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3737 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3739 /* Get the first number on the line: the range minimum */
3741 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3742 *max = *min; /* So can never return without setting max */
3743 if (numlen) /* If found a hex number, position past it */
3745 else if (nl) { /* Else, go handle next line, if any */
3746 return nl + 1; /* 1 is length of "\n" */
3748 else { /* Else, no next line */
3749 return lend + 1; /* to LIST's end at which \n is not found */
3752 /* The max range value follows, separated by a BLANK */
3755 flags = PERL_SCAN_SILENT_ILLDIGIT
3756 | PERL_SCAN_DISALLOW_PREFIX
3757 | PERL_SCAN_SILENT_NON_PORTABLE;
3759 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3762 else /* If no value here, it is a single element range */
3765 /* Non-binary tables have a third entry: what the first element of the
3766 * range maps to. The map for those currently read here is in hex */
3770 flags = PERL_SCAN_SILENT_ILLDIGIT
3771 | PERL_SCAN_DISALLOW_PREFIX
3772 | PERL_SCAN_SILENT_NON_PORTABLE;
3774 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3783 /* diag_listed_as: To%s: illegal mapping '%s' */
3784 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3790 *val = 0; /* bits == 1, then any val should be ignored */
3792 else { /* Nothing following range min, should be single element with no
3797 /* diag_listed_as: To%s: illegal mapping '%s' */
3798 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3802 *val = 0; /* bits == 1, then val should be ignored */
3805 /* Position to next line if any, or EOF */
3815 * Returns a swatch (a bit vector string) for a code point sequence
3816 * that starts from the value C<start> and comprises the number C<span>.
3817 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3818 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3821 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3824 U8 *l, *lend, *x, *xend, *s, *send;
3825 STRLEN lcur, xcur, scur;
3826 HV *const hv = MUTABLE_HV(SvRV(swash));
3827 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3829 SV** listsvp = NULL; /* The string containing the main body of the table */
3830 SV** extssvp = NULL;
3831 SV** invert_it_svp = NULL;
3834 STRLEN octets; /* if bits == 1, then octets == 0 */
3836 UV end = start + span;
3838 if (invlistsvp == NULL) {
3839 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3840 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3841 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3842 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3843 listsvp = hv_fetchs(hv, "LIST", FALSE);
3844 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3846 bits = SvUV(*bitssvp);
3847 none = SvUV(*nonesvp);
3848 typestr = (U8*)SvPV_nolen(*typesvp);
3854 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3856 PERL_ARGS_ASSERT_SWATCH_GET;
3858 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3859 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
3863 /* If overflowed, use the max possible */
3869 /* create and initialize $swatch */
3870 scur = octets ? (span * octets) : (span + 7) / 8;
3871 swatch = newSV(scur);
3873 s = (U8*)SvPVX(swatch);
3874 if (octets && none) {
3875 const U8* const e = s + scur;
3878 *s++ = (U8)(none & 0xff);
3879 else if (bits == 16) {
3880 *s++ = (U8)((none >> 8) & 0xff);
3881 *s++ = (U8)( none & 0xff);
3883 else if (bits == 32) {
3884 *s++ = (U8)((none >> 24) & 0xff);
3885 *s++ = (U8)((none >> 16) & 0xff);
3886 *s++ = (U8)((none >> 8) & 0xff);
3887 *s++ = (U8)( none & 0xff);
3893 (void)memzero((U8*)s, scur + 1);
3895 SvCUR_set(swatch, scur);
3896 s = (U8*)SvPVX(swatch);
3898 if (invlistsvp) { /* If has an inversion list set up use that */
3899 _invlist_populate_swatch(*invlistsvp, start, end, s);
3903 /* read $swash->{LIST} */
3904 l = (U8*)SvPV(*listsvp, lcur);
3907 UV min, max, val, upper;
3908 l = swash_scan_list_line(l, lend, &min, &max, &val,
3909 cBOOL(octets), typestr);
3914 /* If looking for something beyond this range, go try the next one */
3918 /* <end> is generally 1 beyond where we want to set things, but at the
3919 * platform's infinity, where we can't go any higher, we want to
3920 * include the code point at <end> */
3923 : (max != UV_MAX || end != UV_MAX)
3930 if (!none || val < none) {
3935 for (key = min; key <= upper; key++) {
3937 /* offset must be non-negative (start <= min <= key < end) */
3938 offset = octets * (key - start);
3940 s[offset] = (U8)(val & 0xff);
3941 else if (bits == 16) {
3942 s[offset ] = (U8)((val >> 8) & 0xff);
3943 s[offset + 1] = (U8)( val & 0xff);
3945 else if (bits == 32) {
3946 s[offset ] = (U8)((val >> 24) & 0xff);
3947 s[offset + 1] = (U8)((val >> 16) & 0xff);
3948 s[offset + 2] = (U8)((val >> 8) & 0xff);
3949 s[offset + 3] = (U8)( val & 0xff);
3952 if (!none || val < none)
3956 else { /* bits == 1, then val should be ignored */
3961 for (key = min; key <= upper; key++) {
3962 const STRLEN offset = (STRLEN)(key - start);
3963 s[offset >> 3] |= 1 << (offset & 7);
3968 /* Invert if the data says it should be. Assumes that bits == 1 */
3969 if (invert_it_svp && SvUV(*invert_it_svp)) {
3971 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3972 * be 0, and their inversion should also be 0, as we don't succeed any
3973 * Unicode property matches for non-Unicode code points */
3974 if (start <= PERL_UNICODE_MAX) {
3976 /* The code below assumes that we never cross the
3977 * Unicode/above-Unicode boundary in a range, as otherwise we would
3978 * have to figure out where to stop flipping the bits. Since this
3979 * boundary is divisible by a large power of 2, and swatches comes
3980 * in small powers of 2, this should be a valid assumption */
3981 assert(start + span - 1 <= PERL_UNICODE_MAX);
3991 /* read $swash->{EXTRAS}
3992 * This code also copied to swash_to_invlist() below */
3993 x = (U8*)SvPV(*extssvp, xcur);
4001 SV **otherbitssvp, *other;
4005 const U8 opc = *x++;
4009 nl = (U8*)memchr(x, '\n', xend - x);
4011 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4013 x = nl + 1; /* 1 is length of "\n" */
4017 x = xend; /* to EXTRAS' end at which \n is not found */
4024 namelen = nl - namestr;
4028 namelen = xend - namestr;
4032 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4033 otherhv = MUTABLE_HV(SvRV(*othersvp));
4034 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4035 otherbits = (STRLEN)SvUV(*otherbitssvp);
4036 if (bits < otherbits)
4037 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4038 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4040 /* The "other" swatch must be destroyed after. */
4041 other = swatch_get(*othersvp, start, span);
4042 o = (U8*)SvPV(other, olen);
4045 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4047 s = (U8*)SvPV(swatch, slen);
4048 if (bits == 1 && otherbits == 1) {
4050 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4051 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4052 (UV)slen, (UV)olen);
4076 STRLEN otheroctets = otherbits >> 3;
4078 U8* const send = s + slen;
4083 if (otherbits == 1) {
4084 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4088 STRLEN vlen = otheroctets;
4096 if (opc == '+' && otherval)
4097 NOOP; /* replace with otherval */
4098 else if (opc == '!' && !otherval)
4100 else if (opc == '-' && otherval)
4102 else if (opc == '&' && !otherval)
4105 s += octets; /* no replacement */
4110 *s++ = (U8)( otherval & 0xff);
4111 else if (bits == 16) {
4112 *s++ = (U8)((otherval >> 8) & 0xff);
4113 *s++ = (U8)( otherval & 0xff);
4115 else if (bits == 32) {
4116 *s++ = (U8)((otherval >> 24) & 0xff);
4117 *s++ = (U8)((otherval >> 16) & 0xff);
4118 *s++ = (U8)((otherval >> 8) & 0xff);
4119 *s++ = (U8)( otherval & 0xff);
4123 sv_free(other); /* through with it! */
4129 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4132 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4133 * Can't be used on a property that is subject to user override, as it
4134 * relies on the value of SPECIALS in the swash which would be set by
4135 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4136 * for overridden properties
4138 * Returns a hash which is the inversion and closure of a swash mapping.
4139 * For example, consider the input lines:
4144 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4145 * 006C. The value for each key is an array. For 006C, the array would
4146 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4147 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4149 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4150 * keys are only code points that are folded-to, so it isn't a full closure.
4152 * Essentially, for any code point, it gives all the code points that map to
4153 * it, or the list of 'froms' for that point.
4155 * Currently it ignores any additions or deletions from other swashes,
4156 * looking at just the main body of the swash, and if there are SPECIALS
4157 * in the swash, at that hash
4159 * The specials hash can be extra code points, and most likely consists of
4160 * maps from single code points to multiple ones (each expressed as a string
4161 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4162 * However consider this possible input in the specials hash:
4163 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4164 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4166 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4167 * currently handle. But it also means that FB05 and FB06 are equivalent in
4168 * a 1-1 mapping which we should handle, and this relationship may not be in
4169 * the main table. Therefore this function examines all the multi-char
4170 * sequences and adds the 1-1 mappings that come out of that.
4172 * XXX This function was originally intended to be multipurpose, but its
4173 * only use is quite likely to remain for constructing the inversion of
4174 * the CaseFolding (//i) property. If it were more general purpose for
4175 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4176 * because certain folds are prohibited under /iaa and /il. As an example,
4177 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4178 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4179 * prohibited, so we would not figure out that they fold to each other.
4180 * Code could be written to automatically figure this out, similar to the
4181 * code that does this for multi-character folds, but this is the only case
4182 * where something like this is ever likely to happen, as all the single
4183 * char folds to the 0-255 range are now quite settled. Instead there is a
4184 * little special code that is compiled only for this Unicode version. This
4185 * is smaller and didn't require much coding time to do. But this makes
4186 * this routine strongly tied to being used just for CaseFolding. If ever
4187 * it should be generalized, this would have to be fixed */
4191 HV *const hv = MUTABLE_HV(SvRV(swash));
4193 /* The string containing the main body of the table. This will have its
4194 * assertion fail if the swash has been converted to its inversion list */
4195 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4197 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4198 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4199 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4200 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4201 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4202 const STRLEN bits = SvUV(*bitssvp);
4203 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4204 const UV none = SvUV(*nonesvp);
4205 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4209 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4211 /* Must have at least 8 bits to get the mappings */
4212 if (bits != 8 && bits != 16 && bits != 32) {
4213 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4217 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4218 mapping to more than one character */
4220 /* Construct an inverse mapping hash for the specials */
4221 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4222 HV * specials_inverse = newHV();
4223 char *char_from; /* the lhs of the map */
4224 I32 from_len; /* its byte length */
4225 char *char_to; /* the rhs of the map */
4226 I32 to_len; /* its byte length */
4227 SV *sv_to; /* and in a sv */
4228 AV* from_list; /* list of things that map to each 'to' */
4230 hv_iterinit(specials_hv);
4232 /* The keys are the characters (in UTF-8) that map to the corresponding
4233 * UTF-8 string value. Iterate through the list creating the inverse
4235 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4237 if (! SvPOK(sv_to)) {
4238 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4239 "unexpectedly is not a string, flags=%lu",
4240 (unsigned long)SvFLAGS(sv_to));
4242 /*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)));*/
4244 /* Each key in the inverse list is a mapped-to value, and the key's
4245 * hash value is a list of the strings (each in UTF-8) that map to
4246 * it. Those strings are all one character long */
4247 if ((listp = hv_fetch(specials_inverse,
4251 from_list = (AV*) *listp;
4253 else { /* No entry yet for it: create one */
4254 from_list = newAV();
4255 if (! hv_store(specials_inverse,
4258 (SV*) from_list, 0))
4260 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4264 /* Here have the list associated with this 'to' (perhaps newly
4265 * created and empty). Just add to it. Note that we ASSUME that
4266 * the input is guaranteed to not have duplications, so we don't
4267 * check for that. Duplications just slow down execution time. */
4268 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4271 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4272 * it looking for cases like the FB05/FB06 examples above. There would
4273 * be an entry in the hash like
4274 * 'st' => [ FB05, FB06 ]
4275 * In this example we will create two lists that get stored in the
4276 * returned hash, 'ret':
4277 * FB05 => [ FB05, FB06 ]
4278 * FB06 => [ FB05, FB06 ]
4280 * Note that there is nothing to do if the array only has one element.
4281 * (In the normal 1-1 case handled below, we don't have to worry about
4282 * two lists, as everything gets tied to the single list that is
4283 * generated for the single character 'to'. But here, we are omitting
4284 * that list, ('st' in the example), so must have multiple lists.) */
4285 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4286 &char_to, &to_len)))
4288 if (av_tindex_nomg(from_list) > 0) {
4291 /* We iterate over all combinations of i,j to place each code
4292 * point on each list */
4293 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
4295 AV* i_list = newAV();
4296 SV** entryp = av_fetch(from_list, i, FALSE);
4297 if (entryp == NULL) {
4298 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4300 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4301 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4303 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4304 (SV*) i_list, FALSE))
4306 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4309 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4310 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
4311 entryp = av_fetch(from_list, j, FALSE);
4312 if (entryp == NULL) {