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 =for apidoc uvoffuni_to_utf8_flags
58 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
59 Instead, B<Almost all code should use L</uvchr_to_utf8> or
60 L</uvchr_to_utf8_flags>>.
62 This function is like them, but the input is a strict Unicode
63 (as opposed to native) code point. Only in very rare circumstances should code
64 not be using the native code point.
66 For details, see the description for L</uvchr_to_utf8_flags>.
71 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
73 if (flags & UNICODE_WARN_SURROGATE) { \
74 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
75 "UTF-16 surrogate U+%04" UVXf, uv); \
77 if (flags & UNICODE_DISALLOW_SURROGATE) { \
82 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
84 if (flags & UNICODE_WARN_NONCHAR) { \
85 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
86 "Unicode non-character U+%04" UVXf " is not " \
87 "recommended for open interchange", uv); \
89 if (flags & UNICODE_DISALLOW_NONCHAR) { \
94 /* Use shorter names internally in this file */
95 #define SHIFT UTF_ACCUMULATION_SHIFT
97 #define MARK UTF_CONTINUATION_MARK
98 #define MASK UTF_CONTINUATION_MASK
101 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
103 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
105 if (OFFUNI_IS_INVARIANT(uv)) {
106 *d++ = LATIN1_TO_NATIVE(uv);
110 if (uv <= MAX_UTF8_TWO_BYTE) {
111 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
112 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
116 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
117 * below, the 16 is for start bytes E0-EF (which are all the possible ones
118 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
119 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
120 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
121 * 0x800-0xFFFF on ASCII */
122 if (uv < (16 * (1U << (2 * SHIFT)))) {
123 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
124 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
125 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
127 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
128 aren't tested here */
129 /* The most likely code points in this range are below the surrogates.
130 * Do an extra test to quickly exclude those. */
131 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
132 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
133 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
135 HANDLE_UNICODE_NONCHAR(uv, flags);
137 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
138 HANDLE_UNICODE_SURROGATE(uv, flags);
145 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
146 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
147 * happen starting with 4-byte characters on ASCII platforms. We unify the
148 * code for these with EBCDIC, even though some of them require 5-bytes on
149 * those, because khw believes the code saving is worth the very slight
150 * performance hit on these high EBCDIC code points. */
152 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
153 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
154 && ckWARN_d(WARN_DEPRECATED))
156 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
157 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
159 if ( (flags & UNICODE_WARN_SUPER)
160 || ( UNICODE_IS_ABOVE_31_BIT(uv)
161 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
163 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
165 /* Choose the more dire applicable warning */
166 (UNICODE_IS_ABOVE_31_BIT(uv))
167 ? "Code point 0x%" UVXf " is not Unicode, and not portable"
168 : "Code point 0x%" UVXf " is not Unicode, may not be portable",
171 if (flags & UNICODE_DISALLOW_SUPER
172 || ( UNICODE_IS_ABOVE_31_BIT(uv)
173 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
178 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
179 HANDLE_UNICODE_NONCHAR(uv, flags);
182 /* Test for and handle 4-byte result. In the test immediately below, the
183 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
184 * characters). The 3 is for 3 continuation bytes; these each contribute
185 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
186 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
187 * 0x1_0000-0x1F_FFFF on ASCII */
188 if (uv < (8 * (1U << (3 * SHIFT)))) {
189 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
190 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
191 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
192 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
194 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
195 characters. The end-plane non-characters for EBCDIC were
196 handled just above */
197 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
198 HANDLE_UNICODE_NONCHAR(uv, flags);
200 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
201 HANDLE_UNICODE_SURROGATE(uv, flags);
208 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
209 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
210 * format. The unrolled version above turns out to not save all that much
211 * time, and at these high code points (well above the legal Unicode range
212 * on ASCII platforms, and well above anything in common use in EBCDIC),
213 * khw believes that less code outweighs slight performance gains. */
216 STRLEN len = OFFUNISKIP(uv);
219 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
220 uv >>= UTF_ACCUMULATION_SHIFT;
222 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
228 =for apidoc uvchr_to_utf8
230 Adds the UTF-8 representation of the native code point C<uv> to the end
231 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
232 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
233 the byte after the end of the new character. In other words,
235 d = uvchr_to_utf8(d, uv);
237 is the recommended wide native character-aware way of saying
241 This function accepts any UV as input, but very high code points (above
242 C<IV_MAX> on the platform) will raise a deprecation warning. This is
243 typically 0x7FFF_FFFF in a 32-bit word.
245 It is possible to forbid or warn on non-Unicode code points, or those that may
246 be problematic by using L</uvchr_to_utf8_flags>.
251 /* This is also a macro */
252 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
255 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
257 return uvchr_to_utf8(d, uv);
261 =for apidoc uvchr_to_utf8_flags
263 Adds the UTF-8 representation of the native code point C<uv> to the end
264 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
265 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
266 the byte after the end of the new character. In other words,
268 d = uvchr_to_utf8_flags(d, uv, flags);
272 d = uvchr_to_utf8_flags(d, uv, 0);
274 This is the Unicode-aware way of saying
278 If C<flags> is 0, this function accepts any UV as input, but very high code
279 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
280 This is typically 0x7FFF_FFFF in a 32-bit word.
282 Specifying C<flags> can further restrict what is allowed and not warned on, as
285 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
286 the function will raise a warning, provided UTF8 warnings are enabled. If
287 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
288 NULL. If both flags are set, the function will both warn and return NULL.
290 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
291 affect how the function handles a Unicode non-character.
293 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
294 affect the handling of code points that are above the Unicode maximum of
295 0x10FFFF. Languages other than Perl may not be able to accept files that
298 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
299 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
300 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
301 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
302 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
303 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
304 above-Unicode and surrogate flags, but not the non-character ones, as
306 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
307 See L<perlunicode/Noncharacter code points>.
309 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
310 so using them is more problematic than other above-Unicode code points. Perl
311 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
312 likely that non-Perl languages will not be able to read files that contain
313 these that written by the perl interpreter; nor would Perl understand files
314 written by something that uses a different extension. For these reasons, there
315 is a separate set of flags that can warn and/or disallow these extremely high
316 code points, even if other above-Unicode ones are accepted. These are the
317 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
318 are entirely independent from the deprecation warning for code points above
319 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
320 code point that needs more than 31 bits to represent. When that happens,
321 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
322 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
323 above-Unicode code points, including these, as malformations; and
324 C<UNICODE_WARN_SUPER> warns on these.)
326 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
327 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
328 than on ASCII. Prior to that, code points 2**31 and higher were simply
329 unrepresentable, and a different, incompatible method was used to represent
330 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
331 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
332 platforms, warning and disallowing 2**31 and higher.
337 /* This is also a macro */
338 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
341 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
343 return uvchr_to_utf8_flags(d, uv, flags);
346 PERL_STATIC_INLINE bool
347 S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
349 /* Returns TRUE if the first code point represented by the Perl-extended-
350 * UTF-8-encoded string starting at 's', and looking no further than 'e -
351 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
353 * The function handles the case where the input bytes do not include all
354 * the ones necessary to represent a full character. That is, they may be
355 * the intial bytes of the representation of a code point, but possibly
356 * the final ones necessary for the complete representation may be beyond
359 * The function assumes that the sequence is well-formed UTF-8 as far as it
360 * goes, and is for a UTF-8 variant code point. If the sequence is
361 * incomplete, the function returns FALSE if there is any well-formed
362 * UTF-8 byte sequence that can complete it in such a way that a code point
363 * < 2**31 is produced; otherwise it returns TRUE.
365 * Getting this exactly right is slightly tricky, and has to be done in
366 * several places in this file, so is centralized here. It is based on the
369 * U+7FFFFFFF (2 ** 31 - 1)
370 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
371 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
372 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
373 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
374 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
375 * U+80000000 (2 ** 31):
376 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
377 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
378 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
379 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
380 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
381 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
386 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
387 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
388 const STRLEN prefix_len = sizeof(prefix) - 1;
389 const STRLEN len = e - s;
390 const STRLEN cmp_len = MIN(prefix_len, len - 1);
398 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
400 assert(! UTF8_IS_INVARIANT(*s));
404 /* Technically, a start byte of FE can be for a code point that fits into
405 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
411 /* On the EBCDIC code pages we handle, only 0xFE can mean a 32-bit or
412 * larger code point (0xFF is an invariant). For 0xFE, we need at least 2
413 * bytes, and maybe up through 8 bytes, to be sure if the value is above 31
415 if (*s != 0xFE || len == 1) {
419 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
421 return cBOOL(memGT(s + 1, prefix, cmp_len));
427 PERL_STATIC_INLINE bool
428 S_does_utf8_overflow(const U8 * const s, const U8 * e)
431 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
433 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
435 const STRLEN len = e - s;
439 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
440 * platform, that is if it represents a code point larger than the highest
441 * representable code point. (For ASCII platforms, we could use memcmp()
442 * because we don't have to convert each byte to I8, but it's very rare
443 * input indeed that would approach overflow, so the loop below will likely
444 * only get executed once.
446 * 'e' must not be beyond a full character. If it is less than a full
447 * character, the function returns FALSE if there is any input beyond 'e'
448 * that could result in a non-overflowing code point */
450 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
451 assert(s <= e && s + UTF8SKIP(s) >= e);
453 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
455 /* On 32 bit ASCII machines, many overlongs that start with FF don't
458 if (isFF_OVERLONG(s, len)) {
459 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
460 return memGE(s, max_32_bit_overlong,
461 MIN(len, sizeof(max_32_bit_overlong) - 1));
466 for (x = s; x < e; x++, y++) {
468 /* If this byte is larger than the corresponding highest UTF-8 byte, it
470 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
474 /* If not the same as this byte, it must be smaller, doesn't overflow */
475 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
480 /* Got to the end and all bytes are the same. If the input is a whole
481 * character, it doesn't overflow. And if it is a partial character,
482 * there's not enough information to tell, so assume doesn't overflow */
486 PERL_STATIC_INLINE bool
487 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
489 /* Overlongs can occur whenever the number of continuation bytes
490 * changes. That means whenever the number of leading 1 bits in a start
491 * byte increases from the next lower start byte. That happens for start
492 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
493 * illegal start bytes have already been excluded, so don't need to be
495 * ASCII platforms: C0, C1
496 * EBCDIC platforms C0, C1, C2, C3, C4, E0
498 * At least a second byte is required to determine if other sequences will
501 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
502 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
504 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
505 assert(len > 1 && UTF8_IS_START(*s));
507 /* Each platform has overlongs after the start bytes given above (expressed
508 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
509 * the logic is the same, except the E0 overlong has already been excluded
510 * on EBCDIC platforms. The values below were found by manually
511 * inspecting the UTF-8 patterns. See the tables in utf8.h and
515 # define F0_ABOVE_OVERLONG 0xB0
516 # define F8_ABOVE_OVERLONG 0xA8
517 # define FC_ABOVE_OVERLONG 0xA4
518 # define FE_ABOVE_OVERLONG 0xA2
519 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
523 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
527 # define F0_ABOVE_OVERLONG 0x90
528 # define F8_ABOVE_OVERLONG 0x88
529 # define FC_ABOVE_OVERLONG 0x84
530 # define FE_ABOVE_OVERLONG 0x82
531 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
535 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
536 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
537 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
538 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
543 /* Check for the FF overlong */
544 return isFF_OVERLONG(s, len);
547 PERL_STATIC_INLINE bool
548 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
550 PERL_ARGS_ASSERT_ISFF_OVERLONG;
552 /* Check for the FF overlong. This happens only if all these bytes match;
553 * what comes after them doesn't matter. See tables in utf8.h,
556 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
557 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
558 sizeof(FF_OVERLONG_PREFIX) - 1));
561 #undef F0_ABOVE_OVERLONG
562 #undef F8_ABOVE_OVERLONG
563 #undef FC_ABOVE_OVERLONG
564 #undef FE_ABOVE_OVERLONG
565 #undef FF_OVERLONG_PREFIX
568 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
573 /* A helper function that should not be called directly.
575 * This function returns non-zero if the string beginning at 's' and
576 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
577 * code point; otherwise it returns 0. The examination stops after the
578 * first code point in 's' is validated, not looking at the rest of the
579 * input. If 'e' is such that there are not enough bytes to represent a
580 * complete code point, this function will return non-zero anyway, if the
581 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
582 * excluded by 'flags'.
584 * A non-zero return gives the number of bytes required to represent the
585 * code point. Be aware that if the input is for a partial character, the
586 * return will be larger than 'e - s'.
588 * This function assumes that the code point represented is UTF-8 variant.
589 * The caller should have excluded this possibility before calling this
592 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
593 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
594 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
595 * disallowed by the flags. If the input is only for a partial character,
596 * the function will return non-zero if there is any sequence of
597 * well-formed UTF-8 that, when appended to the input sequence, could
598 * result in an allowed code point; otherwise it returns 0. Non characters
599 * cannot be determined based on partial character input. But many of the
600 * other excluded types can be determined with just the first one or two
605 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
607 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
608 |UTF8_DISALLOW_ABOVE_31_BIT)));
609 assert(! UTF8_IS_INVARIANT(*s));
611 /* A variant char must begin with a start byte */
612 if (UNLIKELY(! UTF8_IS_START(*s))) {
616 /* Examine a maximum of a single whole code point */
617 if (e - s > UTF8SKIP(s)) {
623 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
624 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
626 /* The code below is derived from this table. Keep in mind that legal
627 * continuation bytes range between \x80..\xBF for UTF-8, and
628 * \xA0..\xBF for I8. Anything above those aren't continuation bytes.
629 * Hence, we don't have to test the upper edge because if any of those
630 * are encountered, the sequence is malformed, and will fail elsewhere
632 * UTF-8 UTF-EBCDIC I8
633 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
634 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
635 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
639 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
640 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
641 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
643 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
645 && ((s1) & 0xFE ) == 0xB6)
647 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
648 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
649 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
652 if ( (flags & UTF8_DISALLOW_SUPER)
653 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER)) {
654 return 0; /* Above Unicode */
657 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
658 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
660 return 0; /* Above 31 bits */
664 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
666 if ( (flags & UTF8_DISALLOW_SUPER)
667 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
669 return 0; /* Above Unicode */
672 if ( (flags & UTF8_DISALLOW_SURROGATE)
673 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
675 return 0; /* Surrogate */
678 if ( (flags & UTF8_DISALLOW_NONCHAR)
679 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
681 return 0; /* Noncharacter code point */
686 /* Make sure that all that follows are continuation bytes */
687 for (x = s + 1; x < e; x++) {
688 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
693 /* Here is syntactically valid. Next, make sure this isn't the start of an
695 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
699 /* And finally, that the code point represented fits in a word on this
701 if (does_utf8_overflow(s, e)) {
709 S__byte_dump_string(pTHX_ const U8 * s, const STRLEN len)
711 /* Returns a mortalized C string that is a displayable copy of the 'len'
712 * bytes starting at 's', each in a \xXY format. */
714 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
716 const U8 * const e = s + len;
720 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
722 Newx(output, output_len, char);
727 const unsigned high_nibble = (*s & 0xF0) >> 4;
728 const unsigned low_nibble = (*s & 0x0F);
733 if (high_nibble < 10) {
734 *d++ = high_nibble + '0';
737 *d++ = high_nibble - 10 + 'a';
740 if (low_nibble < 10) {
741 *d++ = low_nibble + '0';
744 *d++ = low_nibble - 10 + 'a';
752 PERL_STATIC_INLINE char *
753 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
755 /* How many bytes to print */
758 /* Which one is the non-continuation */
759 const STRLEN non_cont_byte_pos,
761 /* How many bytes should there be? */
762 const STRLEN expect_len)
764 /* Return the malformation warning text for an unexpected continuation
767 const char * const where = (non_cont_byte_pos == 1)
769 : Perl_form(aTHX_ "%d bytes",
770 (int) non_cont_byte_pos);
773 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
775 /* We don't need to pass this parameter, but since it has already been
776 * calculated, it's likely faster to pass it; verify under DEBUGGING */
777 assert(expect_len == UTF8SKIP(s));
779 /* It is possible that utf8n_to_uvchr() was called incorrectly, with a
780 * length that is larger than is actually available in the buffer. If we
781 * print all the bytes based on that length, we will read past the buffer
782 * end. Often, the strings are NUL terminated, so to lower the chances of
783 * this happening, print the malformed bytes only up through any NUL. */
784 for (i = 1; i < print_len; i++) {
785 if (*(s + i) == '\0') {
786 print_len = i + 1; /* +1 gets the NUL printed */
791 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
792 " %s after start byte 0x%02x; need %d bytes, got %d)",
794 _byte_dump_string(s, print_len),
795 *(s + non_cont_byte_pos),
799 (int) non_cont_byte_pos);
804 =for apidoc utf8n_to_uvchr
806 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
807 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
809 Bottom level UTF-8 decode routine.
810 Returns the native code point value of the first character in the string C<s>,
811 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
812 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
813 the length, in bytes, of that character.
815 The value of C<flags> determines the behavior when C<s> does not point to a
816 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
817 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
818 is the next possible position in C<s> that could begin a non-malformed
819 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
820 is raised. Some UTF-8 input sequences may contain multiple malformations.
821 This function tries to find every possible one in each call, so multiple
822 warnings can be raised for each sequence.
824 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
825 individual types of malformations, such as the sequence being overlong (that
826 is, when there is a shorter sequence that can express the same code point;
827 overlong sequences are expressly forbidden in the UTF-8 standard due to
828 potential security issues). Another malformation example is the first byte of
829 a character not being a legal first byte. See F<utf8.h> for the list of such
830 flags. For allowed 0 length strings, this function returns 0; for allowed
831 overlong sequences, the computed code point is returned; for all other allowed
832 malformations, the Unicode REPLACEMENT CHARACTER is returned, as these have no
833 determinable reasonable value.
835 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
836 flags) malformation is found. If this flag is set, the routine assumes that
837 the caller will raise a warning, and this function will silently just set
838 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
840 Note that this API requires disambiguation between successful decoding a C<NUL>
841 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
842 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
843 be set to 1. To disambiguate, upon a zero return, see if the first byte of
844 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
845 error. Or you can use C<L</utf8n_to_uvchr_error>>.
847 Certain code points are considered problematic. These are Unicode surrogates,
848 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
849 By default these are considered regular code points, but certain situations
850 warrant special handling for them, which can be specified using the C<flags>
851 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
852 three classes are treated as malformations and handled as such. The flags
853 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
854 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
855 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
856 restricts the allowed inputs to the strict UTF-8 traditionally defined by
857 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
859 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
860 The difference between traditional strictness and C9 strictness is that the
861 latter does not forbid non-character code points. (They are still discouraged,
862 however.) For more discussion see L<perlunicode/Noncharacter code points>.
864 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
865 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
866 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
867 raised for their respective categories, but otherwise the code points are
868 considered valid (not malformations). To get a category to both be treated as
869 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
870 (But note that warnings are not raised if lexically disabled nor if
871 C<UTF8_CHECK_ONLY> is also specified.)
873 It is now deprecated to have very high code points (above C<IV_MAX> on the
874 platforms) and this function will raise a deprecation warning for these (unless
875 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
878 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
879 so using them is more problematic than other above-Unicode code points. Perl
880 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
881 likely that non-Perl languages will not be able to read files that contain
882 these; nor would Perl understand files
883 written by something that uses a different extension. For these reasons, there
884 is a separate set of flags that can warn and/or disallow these extremely high
885 code points, even if other above-Unicode ones are accepted. These are the
886 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
887 are entirely independent from the deprecation warning for code points above
888 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
889 code point that needs more than 31 bits to represent. When that happens,
890 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
891 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
892 above-Unicode code points, including these, as malformations; and
893 C<UTF8_WARN_SUPER> warns on these.)
895 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
896 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
897 than on ASCII. Prior to that, code points 2**31 and higher were simply
898 unrepresentable, and a different, incompatible method was used to represent
899 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
900 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
901 platforms, warning and disallowing 2**31 and higher.
903 All other code points corresponding to Unicode characters, including private
904 use and those yet to be assigned, are never considered malformed and never
909 Also implemented as a macro in utf8.h
913 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
918 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
920 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
925 =for apidoc utf8n_to_uvchr_error
927 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
928 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
930 This function is for code that needs to know what the precise malformation(s)
931 are when an error is found.
933 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
934 all the others, C<errors>. If this parameter is 0, this function behaves
935 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
936 to a C<U32> variable, which this function sets to indicate any errors found.
937 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
938 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
939 of these bits will be set if a malformation is found, even if the input
940 C<flags> parameter indicates that the given malformation is allowed; those
941 exceptions are noted:
945 =item C<UTF8_GOT_ABOVE_31_BIT>
947 The code point represented by the input UTF-8 sequence occupies more than 31
949 This bit is set only if the input C<flags> parameter contains either the
950 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
952 =item C<UTF8_GOT_CONTINUATION>
954 The input sequence was malformed in that the first byte was a a UTF-8
957 =item C<UTF8_GOT_EMPTY>
959 The input C<curlen> parameter was 0.
961 =item C<UTF8_GOT_LONG>
963 The input sequence was malformed in that there is some other sequence that
964 evaluates to the same code point, but that sequence is shorter than this one.
966 =item C<UTF8_GOT_NONCHAR>
968 The code point represented by the input UTF-8 sequence is for a Unicode
969 non-character code point.
970 This bit is set only if the input C<flags> parameter contains either the
971 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
973 =item C<UTF8_GOT_NON_CONTINUATION>
975 The input sequence was malformed in that a non-continuation type byte was found
976 in a position where only a continuation type one should be.
978 =item C<UTF8_GOT_OVERFLOW>
980 The input sequence was malformed in that it is for a code point that is not
981 representable in the number of bits available in a UV on the current platform.
983 =item C<UTF8_GOT_SHORT>
985 The input sequence was malformed in that C<curlen> is smaller than required for
986 a complete sequence. In other words, the input is for a partial character
989 =item C<UTF8_GOT_SUPER>
991 The input sequence was malformed in that it is for a non-Unicode code point;
992 that is, one above the legal Unicode maximum.
993 This bit is set only if the input C<flags> parameter contains either the
994 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
996 =item C<UTF8_GOT_SURROGATE>
998 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1000 This bit is set only if the input C<flags> parameter contains either the
1001 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1005 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1006 flag to suppress any warnings, and then examine the C<*errors> return.
1012 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1018 const U8 * const s0 = s;
1019 U8 * send = NULL; /* (initialized to silence compilers' wrong
1021 U32 possible_problems = 0; /* A bit is set here for each potential problem
1022 found as we go along */
1024 STRLEN expectlen = 0; /* How long should this sequence be?
1025 (initialized to silence compilers' wrong
1027 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1028 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1029 this gets set and discarded */
1031 /* The below are used only if there is both an overlong malformation and a
1032 * too short one. Otherwise the first two are set to 's0' and 'send', and
1033 * the third not used at all */
1034 U8 * adjusted_s0 = (U8 *) s0;
1035 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1037 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1039 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1045 errors = &discard_errors;
1048 /* The order of malformation tests here is important. We should consume as
1049 * few bytes as possible in order to not skip any valid character. This is
1050 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1051 * http://unicode.org/reports/tr36 for more discussion as to why. For
1052 * example, once we've done a UTF8SKIP, we can tell the expected number of
1053 * bytes, and could fail right off the bat if the input parameters indicate
1054 * that there are too few available. But it could be that just that first
1055 * byte is garbled, and the intended character occupies fewer bytes. If we
1056 * blindly assumed that the first byte is correct, and skipped based on
1057 * that number, we could skip over a valid input character. So instead, we
1058 * always examine the sequence byte-by-byte.
1060 * We also should not consume too few bytes, otherwise someone could inject
1061 * things. For example, an input could be deliberately designed to
1062 * overflow, and if this code bailed out immediately upon discovering that,
1063 * returning to the caller C<*retlen> pointing to the very next byte (one
1064 * which is actually part of of the overflowing sequence), that could look
1065 * legitimate to the caller, which could discard the initial partial
1066 * sequence and process the rest, inappropriately.
1068 * Some possible input sequences are malformed in more than one way. This
1069 * function goes to lengths to try to find all of them. This is necessary
1070 * for correctness, as the inputs may allow one malformation but not
1071 * another, and if we abandon searching for others after finding the
1072 * allowed one, we could allow in something that shouldn't have been.
1075 if (UNLIKELY(curlen == 0)) {
1076 possible_problems |= UTF8_GOT_EMPTY;
1078 uv = 0; /* XXX It could be argued that this should be
1079 UNICODE_REPLACEMENT? */
1080 goto ready_to_handle_errors;
1083 expectlen = UTF8SKIP(s);
1085 /* A well-formed UTF-8 character, as the vast majority of calls to this
1086 * function will be for, has this expected length. For efficiency, set
1087 * things up here to return it. It will be overriden only in those rare
1088 * cases where a malformation is found */
1090 *retlen = expectlen;
1093 /* An invariant is trivially well-formed */
1094 if (UTF8_IS_INVARIANT(uv)) {
1098 /* A continuation character can't start a valid sequence */
1099 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1100 possible_problems |= UTF8_GOT_CONTINUATION;
1102 uv = UNICODE_REPLACEMENT;
1103 goto ready_to_handle_errors;
1106 /* Here is not a continuation byte, nor an invariant. The only thing left
1107 * is a start byte (possibly for an overlong) */
1109 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1110 * that indicate the number of bytes in the character's whole UTF-8
1111 * sequence, leaving just the bits that are part of the value. */
1112 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1114 /* Setup the loop end point, making sure to not look past the end of the
1115 * input string, and flag it as too short if the size isn't big enough. */
1117 if (UNLIKELY(curlen < expectlen)) {
1118 possible_problems |= UTF8_GOT_SHORT;
1125 adjusted_send = send;
1127 /* Now, loop through the remaining bytes in the character's sequence,
1128 * accumulating each into the working value as we go. */
1129 for (s = s0 + 1; s < send; s++) {
1130 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1131 uv = UTF8_ACCUMULATE(uv, *s);
1135 /* Here, found a non-continuation before processing all expected bytes.
1136 * This byte indicates the beginning of a new character, so quit, even
1137 * if allowing this malformation. */
1138 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1140 } /* End of loop through the character's bytes */
1142 /* Save how many bytes were actually in the character */
1145 /* A convenience macro that matches either of the too-short conditions. */
1146 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1148 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1150 uv = UNICODE_REPLACEMENT;
1153 /* Note that there are two types of too-short malformation. One is when
1154 * there is actual wrong data before the normal termination of the
1155 * sequence. The other is that the sequence wasn't complete before the end
1156 * of the data we are allowed to look at, based on the input 'curlen'.
1157 * This means that we were passed data for a partial character, but it is
1158 * valid as far as we saw. The other is definitely invalid. This
1159 * distinction could be important to a caller, so the two types are kept
1162 /* Check for overflow */
1163 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1164 possible_problems |= UTF8_GOT_OVERFLOW;
1165 uv = UNICODE_REPLACEMENT;
1168 /* Check for overlong. If no problems so far, 'uv' is the correct code
1169 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1170 * we must look at the UTF-8 byte sequence itself to see if it is for an
1172 if ( ( LIKELY(! possible_problems)
1173 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1174 || ( UNLIKELY( possible_problems)
1175 && ( UNLIKELY(! UTF8_IS_START(*s0))
1177 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1180 possible_problems |= UTF8_GOT_LONG;
1182 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1183 UV min_uv = uv_so_far;
1186 /* Here, the input is both overlong and is missing some trailing
1187 * bytes. There is no single code point it could be for, but there
1188 * may be enough information present to determine if what we have
1189 * so far is for an unallowed code point, such as for a surrogate.
1190 * The code below has the intelligence to determine this, but just
1191 * for non-overlong UTF-8 sequences. What we do here is calculate
1192 * the smallest code point the input could represent if there were
1193 * no too short malformation. Then we compute and save the UTF-8
1194 * for that, which is what the code below looks at instead of the
1195 * raw input. It turns out that the smallest such code point is
1197 for (i = curlen; i < expectlen; i++) {
1198 min_uv = UTF8_ACCUMULATE(min_uv,
1199 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1202 Newx(adjusted_s0, OFFUNISKIP(min_uv) + 1, U8);
1203 SAVEFREEPV((U8 *) adjusted_s0); /* Needed because we may not get
1204 to free it ourselves if
1205 warnings are made fatal */
1206 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1210 /* Now check that the input isn't for a problematic code point not allowed
1211 * by the input parameters. */
1212 /* isn't problematic if < this */
1213 if ( ( ( LIKELY(! possible_problems) && uv >= UNICODE_SURROGATE_FIRST)
1214 || ( UNLIKELY(possible_problems)
1215 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1216 && ((flags & ( UTF8_DISALLOW_NONCHAR
1217 |UTF8_DISALLOW_SURROGATE
1218 |UTF8_DISALLOW_SUPER
1219 |UTF8_DISALLOW_ABOVE_31_BIT
1221 |UTF8_WARN_SURROGATE
1223 |UTF8_WARN_ABOVE_31_BIT))
1224 /* In case of a malformation, 'uv' is not valid, and has
1225 * been changed to something in the Unicode range.
1226 * Currently we don't output a deprecation message if there
1227 * is already a malformation, so we don't have to special
1228 * case the test immediately below */
1229 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1230 && ckWARN_d(WARN_DEPRECATED))))
1232 /* If there were no malformations, or the only malformation is an
1233 * overlong, 'uv' is valid */
1234 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1235 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1236 possible_problems |= UTF8_GOT_SURROGATE;
1238 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1239 possible_problems |= UTF8_GOT_SUPER;
1241 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1242 possible_problems |= UTF8_GOT_NONCHAR;
1245 else { /* Otherwise, need to look at the source UTF-8, possibly
1246 adjusted to be non-overlong */
1248 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1249 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1251 possible_problems |= UTF8_GOT_SUPER;
1253 else if (curlen > 1) {
1254 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1255 NATIVE_UTF8_TO_I8(*adjusted_s0),
1256 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1258 possible_problems |= UTF8_GOT_SUPER;
1260 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1261 NATIVE_UTF8_TO_I8(*adjusted_s0),
1262 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1264 possible_problems |= UTF8_GOT_SURROGATE;
1268 /* We need a complete well-formed UTF-8 character to discern
1269 * non-characters, so can't look for them here */
1273 ready_to_handle_errors:
1276 * curlen contains the number of bytes in the sequence that
1277 * this call should advance the input by.
1278 * avail_len gives the available number of bytes passed in, but
1279 * only if this is less than the expected number of
1280 * bytes, based on the code point's start byte.
1281 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1282 * is set in it for each potential problem found.
1283 * uv contains the code point the input sequence
1284 * represents; or if there is a problem that prevents
1285 * a well-defined value from being computed, it is
1286 * some subsitute value, typically the REPLACEMENT
1288 * s0 points to the first byte of the character
1289 * send points to just after where that (potentially
1290 * partial) character ends
1291 * adjusted_s0 normally is the same as s0, but in case of an
1292 * overlong for which the UTF-8 matters below, it is
1293 * the first byte of the shortest form representation
1295 * adjusted_send normally is the same as 'send', but if adjusted_s0
1296 * is set to something other than s0, this points one
1300 if (UNLIKELY(possible_problems)) {
1301 bool disallowed = FALSE;
1302 const U32 orig_problems = possible_problems;
1304 while (possible_problems) { /* Handle each possible problem */
1306 char * message = NULL;
1308 /* Each 'if' clause handles one problem. They are ordered so that
1309 * the first ones' messages will be displayed before the later
1310 * ones; this is kinda in decreasing severity order */
1311 if (possible_problems & UTF8_GOT_OVERFLOW) {
1313 /* Overflow means also got a super and above 31 bits, but we
1314 * handle all three cases here */
1316 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1317 *errors |= UTF8_GOT_OVERFLOW;
1319 /* But the API says we flag all errors found */
1320 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1321 *errors |= UTF8_GOT_SUPER;
1323 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1324 *errors |= UTF8_GOT_ABOVE_31_BIT;
1329 /* The warnings code explicitly says it doesn't handle the case
1330 * of packWARN2 and two categories which have parent-child
1331 * relationship. Even if it works now to raise the warning if
1332 * either is enabled, it wouldn't necessarily do so in the
1333 * future. We output (only) the most dire warning*/
1334 if (! (flags & UTF8_CHECK_ONLY)) {
1335 if (ckWARN_d(WARN_UTF8)) {
1336 pack_warn = packWARN(WARN_UTF8);
1338 else if (ckWARN_d(WARN_NON_UNICODE)) {
1339 pack_warn = packWARN(WARN_NON_UNICODE);
1342 message = Perl_form(aTHX_ "%s: %s (overflows)",
1344 _byte_dump_string(s0, send - s0));
1348 else if (possible_problems & UTF8_GOT_EMPTY) {
1349 possible_problems &= ~UTF8_GOT_EMPTY;
1350 *errors |= UTF8_GOT_EMPTY;
1352 if (! (flags & UTF8_ALLOW_EMPTY)) {
1354 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1355 pack_warn = packWARN(WARN_UTF8);
1356 message = Perl_form(aTHX_ "%s (empty string)",
1361 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1362 possible_problems &= ~UTF8_GOT_CONTINUATION;
1363 *errors |= UTF8_GOT_CONTINUATION;
1365 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1367 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1368 pack_warn = packWARN(WARN_UTF8);
1369 message = Perl_form(aTHX_
1370 "%s: %s (unexpected continuation byte 0x%02x,"
1371 " with no preceding start byte)",
1373 _byte_dump_string(s0, 1), *s0);
1377 else if (possible_problems & UTF8_GOT_SHORT) {
1378 possible_problems &= ~UTF8_GOT_SHORT;
1379 *errors |= UTF8_GOT_SHORT;
1381 if (! (flags & UTF8_ALLOW_SHORT)) {
1383 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1384 pack_warn = packWARN(WARN_UTF8);
1385 message = Perl_form(aTHX_
1386 "%s: %s (too short; %d byte%s available, need %d)",
1388 _byte_dump_string(s0, send - s0),
1390 avail_len == 1 ? "" : "s",
1396 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1397 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1398 *errors |= UTF8_GOT_NON_CONTINUATION;
1400 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1402 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1403 pack_warn = packWARN(WARN_UTF8);
1404 message = Perl_form(aTHX_ "%s",
1405 unexpected_non_continuation_text(s0,
1412 else if (possible_problems & UTF8_GOT_LONG) {
1413 possible_problems &= ~UTF8_GOT_LONG;
1414 *errors |= UTF8_GOT_LONG;
1416 if (! (flags & UTF8_ALLOW_LONG)) {
1419 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1420 pack_warn = packWARN(WARN_UTF8);
1422 /* These error types cause 'uv' to be something that
1423 * isn't what was intended, so can't use it in the
1424 * message. The other error types either can't
1425 * generate an overlong, or else the 'uv' is valid */
1427 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1429 message = Perl_form(aTHX_
1430 "%s: %s (any UTF-8 sequence that starts"
1431 " with \"%s\" is overlong which can and"
1432 " should be represented with a"
1433 " different, shorter sequence)",
1435 _byte_dump_string(s0, send - s0),
1436 _byte_dump_string(s0, curlen));
1439 U8 tmpbuf[UTF8_MAXBYTES+1];
1440 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1442 message = Perl_form(aTHX_
1443 "%s: %s (overlong; instead use %s to represent"
1446 _byte_dump_string(s0, send - s0),
1447 _byte_dump_string(tmpbuf, e - tmpbuf),
1448 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1449 small code points */
1455 else if (possible_problems & UTF8_GOT_SURROGATE) {
1456 possible_problems &= ~UTF8_GOT_SURROGATE;
1458 if (flags & UTF8_WARN_SURROGATE) {
1459 *errors |= UTF8_GOT_SURROGATE;
1461 if ( ! (flags & UTF8_CHECK_ONLY)
1462 && ckWARN_d(WARN_SURROGATE))
1464 pack_warn = packWARN(WARN_SURROGATE);
1466 /* These are the only errors that can occur with a
1467 * surrogate when the 'uv' isn't valid */
1468 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1469 message = Perl_form(aTHX_
1470 "UTF-16 surrogate (any UTF-8 sequence that"
1471 " starts with \"%s\" is for a surrogate)",
1472 _byte_dump_string(s0, curlen));
1475 message = Perl_form(aTHX_
1476 "UTF-16 surrogate U+%04" UVXf, uv);
1481 if (flags & UTF8_DISALLOW_SURROGATE) {
1483 *errors |= UTF8_GOT_SURROGATE;
1486 else if (possible_problems & UTF8_GOT_SUPER) {
1487 possible_problems &= ~UTF8_GOT_SUPER;
1489 if (flags & UTF8_WARN_SUPER) {
1490 *errors |= UTF8_GOT_SUPER;
1492 if ( ! (flags & UTF8_CHECK_ONLY)
1493 && ckWARN_d(WARN_NON_UNICODE))
1495 pack_warn = packWARN(WARN_NON_UNICODE);
1497 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1498 message = Perl_form(aTHX_
1499 "Any UTF-8 sequence that starts with"
1500 " \"%s\" is for a non-Unicode code point,"
1501 " may not be portable",
1502 _byte_dump_string(s0, curlen));
1505 message = Perl_form(aTHX_
1506 "Code point 0x%04" UVXf " is not"
1507 " Unicode, may not be portable",
1513 /* The maximum code point ever specified by a standard was
1514 * 2**31 - 1. Anything larger than that is a Perl extension
1515 * that very well may not be understood by other applications
1516 * (including earlier perl versions on EBCDIC platforms). We
1517 * test for these after the regular SUPER ones, and before
1518 * possibly bailing out, so that the slightly more dire warning
1519 * will override the regular one. */
1520 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1522 |UTF8_DISALLOW_ABOVE_31_BIT))
1523 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1524 && UNLIKELY(is_utf8_cp_above_31_bits(
1527 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1528 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1530 if ( ! (flags & UTF8_CHECK_ONLY)
1531 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1532 && ckWARN_d(WARN_UTF8))
1534 pack_warn = packWARN(WARN_UTF8);
1536 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1537 message = Perl_form(aTHX_
1538 "Any UTF-8 sequence that starts with"
1539 " \"%s\" is for a non-Unicode code"
1540 " point, and is not portable",
1541 _byte_dump_string(s0, curlen));
1544 message = Perl_form(aTHX_
1545 "Code point 0x%" UVXf " is not Unicode,"
1546 " and not portable",
1551 if (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT)) {
1552 *errors |= UTF8_GOT_ABOVE_31_BIT;
1554 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1560 if (flags & UTF8_DISALLOW_SUPER) {
1561 *errors |= UTF8_GOT_SUPER;
1565 /* The deprecated warning overrides any non-deprecated one. If
1566 * there are other problems, a deprecation message is not
1567 * really helpful, so don't bother to raise it in that case.
1568 * This also keeps the code from having to handle the case
1569 * where 'uv' is not valid. */
1570 if ( ! (orig_problems
1571 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1572 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1573 && ckWARN_d(WARN_DEPRECATED))
1575 message = Perl_form(aTHX_ cp_above_legal_max,
1576 uv, MAX_NON_DEPRECATED_CP);
1577 pack_warn = packWARN(WARN_DEPRECATED);
1580 else if (possible_problems & UTF8_GOT_NONCHAR) {
1581 possible_problems &= ~UTF8_GOT_NONCHAR;
1583 if (flags & UTF8_WARN_NONCHAR) {
1584 *errors |= UTF8_GOT_NONCHAR;
1586 if ( ! (flags & UTF8_CHECK_ONLY)
1587 && ckWARN_d(WARN_NONCHAR))
1589 /* The code above should have guaranteed that we don't
1590 * get here with errors other than overlong */
1591 assert (! (orig_problems
1592 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1594 pack_warn = packWARN(WARN_NONCHAR);
1595 message = Perl_form(aTHX_ "Unicode non-character"
1596 " U+%04" UVXf " is not recommended"
1597 " for open interchange", uv);
1601 if (flags & UTF8_DISALLOW_NONCHAR) {
1603 *errors |= UTF8_GOT_NONCHAR;
1605 } /* End of looking through the possible flags */
1607 /* Display the message (if any) for the problem being handled in
1608 * this iteration of the loop */
1611 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1614 Perl_warner(aTHX_ pack_warn, "%s", message);
1616 } /* End of 'while (possible_problems) {' */
1618 /* Since there was a possible problem, the returned length may need to
1619 * be changed from the one stored at the beginning of this function.
1620 * Instead of trying to figure out if that's needed, just do it. */
1626 if (flags & UTF8_CHECK_ONLY && retlen) {
1627 *retlen = ((STRLEN) -1);
1633 return UNI_TO_NATIVE(uv);
1637 =for apidoc utf8_to_uvchr_buf
1639 Returns the native code point of the first character in the string C<s> which
1640 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1641 C<*retlen> will be set to the length, in bytes, of that character.
1643 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1644 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1645 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1646 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1647 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1648 the next possible position in C<s> that could begin a non-malformed character.
1649 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1652 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1653 unless those are turned off.
1657 Also implemented as a macro in utf8.h
1663 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1667 return utf8n_to_uvchr(s, send - s, retlen,
1668 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1671 /* This is marked as deprecated
1673 =for apidoc utf8_to_uvuni_buf
1675 Only in very rare circumstances should code need to be dealing in Unicode
1676 (as opposed to native) code points. In those few cases, use
1677 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1679 Returns the Unicode (not-native) code point of the first character in the
1681 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1682 C<retlen> will be set to the length, in bytes, of that character.
1684 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1685 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1686 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1687 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1688 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1689 next possible position in C<s> that could begin a non-malformed character.
1690 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1692 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1693 unless those are turned off.
1699 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1701 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1705 /* Call the low level routine, asking for checks */
1706 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1710 =for apidoc utf8_length
1712 Return the length of the UTF-8 char encoded string C<s> in characters.
1713 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1714 up past C<e>, croaks.
1720 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1724 PERL_ARGS_ASSERT_UTF8_LENGTH;
1726 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1727 * the bitops (especially ~) can create illegal UTF-8.
1728 * In other words: in Perl UTF-8 is not just for Unicode. */
1731 goto warn_and_return;
1741 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1742 "%s in %s", unees, OP_DESC(PL_op));
1744 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1751 =for apidoc bytes_cmp_utf8
1753 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1754 sequence of characters (stored as UTF-8)
1755 in C<u>, C<ulen>. Returns 0 if they are
1756 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1757 if the first string is greater than the second string.
1759 -1 or +1 is returned if the shorter string was identical to the start of the
1760 longer string. -2 or +2 is returned if
1761 there was a difference between characters
1768 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1770 const U8 *const bend = b + blen;
1771 const U8 *const uend = u + ulen;
1773 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1775 while (b < bend && u < uend) {
1777 if (!UTF8_IS_INVARIANT(c)) {
1778 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1781 if (UTF8_IS_CONTINUATION(c1)) {
1782 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1784 /* diag_listed_as: Malformed UTF-8 character%s */
1785 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1787 unexpected_non_continuation_text(u - 1, 2, 1, 2),
1788 PL_op ? " in " : "",
1789 PL_op ? OP_DESC(PL_op) : "");
1794 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1795 "%s in %s", unees, OP_DESC(PL_op));
1797 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1798 return -2; /* Really want to return undef :-) */
1805 return *b < c ? -2 : +2;
1810 if (b == bend && u == uend)
1813 return b < bend ? +1 : -1;
1817 =for apidoc utf8_to_bytes
1819 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1820 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1821 updates C<len> to contain the new length.
1822 Returns zero on failure, setting C<len> to -1.
1824 If you need a copy of the string, see L</bytes_from_utf8>.
1830 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *len)
1832 U8 * const save = s;
1833 U8 * const send = s + *len;
1836 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1837 PERL_UNUSED_CONTEXT;
1839 /* ensure valid UTF-8 and chars < 256 before updating string */
1841 if (! UTF8_IS_INVARIANT(*s)) {
1842 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1843 *len = ((STRLEN) -1);
1854 if (! UTF8_IS_INVARIANT(c)) {
1855 /* Then it is two-byte encoded */
1856 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1867 =for apidoc bytes_from_utf8
1869 Converts a string C<s> of length C<len> from UTF-8 into native byte encoding.
1870 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, returns a pointer to
1871 the newly-created string, and updates C<len> to contain the new
1872 length. Returns the original string if no conversion occurs, C<len>
1873 is unchanged. Do nothing if C<is_utf8> points to 0. Sets C<is_utf8> to
1874 0 if C<s> is converted or consisted entirely of characters that are invariant
1875 in UTF-8 (i.e., US-ASCII on non-EBCDIC machines).
1881 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *len, bool *is_utf8)
1884 const U8 *start = s;
1888 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
1889 PERL_UNUSED_CONTEXT;
1893 /* ensure valid UTF-8 and chars < 256 before converting string */
1894 for (send = s + *len; s < send;) {
1895 if (! UTF8_IS_INVARIANT(*s)) {
1896 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
1907 Newx(d, (*len) - count + 1, U8);
1908 s = start; start = d;
1911 if (! UTF8_IS_INVARIANT(c)) {
1912 /* Then it is two-byte encoded */
1913 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
1924 =for apidoc bytes_to_utf8
1926 Converts a string C<s> of length C<len> bytes from the native encoding into
1928 Returns a pointer to the newly-created string, and sets C<len> to
1929 reflect the new length in bytes.
1931 A C<NUL> character will be written after the end of the string.
1933 If you want to convert to UTF-8 from encodings other than
1934 the native (Latin1 or EBCDIC),
1935 see L</sv_recode_to_utf8>().
1940 /* This logic is duplicated in sv_catpvn_flags, so any bug fixes will
1941 likewise need duplication. */
1944 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *len)
1946 const U8 * const send = s + (*len);
1950 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
1951 PERL_UNUSED_CONTEXT;
1953 Newx(d, (*len) * 2 + 1, U8);
1957 append_utf8_from_native_byte(*s, &d);
1966 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
1968 * Destination must be pre-extended to 3/2 source. Do not use in-place.
1969 * We optimize for native, for obvious reasons. */
1972 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
1977 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
1980 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf, (UV)bytelen);
1985 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
1987 if (OFFUNI_IS_INVARIANT(uv)) {
1988 *d++ = LATIN1_TO_NATIVE((U8) uv);
1991 if (uv <= MAX_UTF8_TWO_BYTE) {
1992 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
1993 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
1996 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
1997 #define LAST_HIGH_SURROGATE 0xDBFF
1998 #define FIRST_LOW_SURROGATE 0xDC00
1999 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2001 /* This assumes that most uses will be in the first Unicode plane, not
2002 * needing surrogates */
2003 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2004 && uv <= UNICODE_SURROGATE_LAST))
2006 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2007 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2010 UV low = (p[0] << 8) + p[1];
2011 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2012 || UNLIKELY(low > LAST_LOW_SURROGATE))
2014 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2017 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2018 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2022 d = uvoffuni_to_utf8_flags(d, uv, 0);
2025 *d++ = (U8)(( uv >> 12) | 0xe0);
2026 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2027 *d++ = (U8)(( uv & 0x3f) | 0x80);
2031 *d++ = (U8)(( uv >> 18) | 0xf0);
2032 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2033 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2034 *d++ = (U8)(( uv & 0x3f) | 0x80);
2039 *newlen = d - dstart;
2043 /* Note: this one is slightly destructive of the source. */
2046 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2049 U8* const send = s + bytelen;
2051 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2054 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2058 const U8 tmp = s[0];
2063 return utf16_to_utf8(p, d, bytelen, newlen);
2067 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2069 U8 tmpbuf[UTF8_MAXBYTES+1];
2070 uvchr_to_utf8(tmpbuf, c);
2071 return _is_utf8_FOO(classnum, tmpbuf);
2074 /* Internal function so we can deprecate the external one, and call
2075 this one from other deprecated functions in this file */
2078 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2080 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2084 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2088 Perl__is_uni_perl_idcont(pTHX_ UV c)
2090 U8 tmpbuf[UTF8_MAXBYTES+1];
2091 uvchr_to_utf8(tmpbuf, c);
2092 return _is_utf8_perl_idcont(tmpbuf);
2096 Perl__is_uni_perl_idstart(pTHX_ UV c)
2098 U8 tmpbuf[UTF8_MAXBYTES+1];
2099 uvchr_to_utf8(tmpbuf, c);
2100 return _is_utf8_perl_idstart(tmpbuf);
2104 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const char S_or_s)
2106 /* We have the latin1-range values compiled into the core, so just use
2107 * those, converting the result to UTF-8. The only difference between upper
2108 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2109 * either "SS" or "Ss". Which one to use is passed into the routine in
2110 * 'S_or_s' to avoid a test */
2112 UV converted = toUPPER_LATIN1_MOD(c);
2114 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2116 assert(S_or_s == 'S' || S_or_s == 's');
2118 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2119 characters in this range */
2120 *p = (U8) converted;
2125 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2126 * which it maps to one of them, so as to only have to have one check for
2127 * it in the main case */
2128 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2130 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2131 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2134 converted = GREEK_CAPITAL_LETTER_MU;
2136 #if UNICODE_MAJOR_VERSION > 2 \
2137 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2138 && UNICODE_DOT_DOT_VERSION >= 8)
2139 case LATIN_SMALL_LETTER_SHARP_S:
2146 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect '%c' to map to '%c'", c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2147 NOT_REACHED; /* NOTREACHED */
2151 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2152 *p = UTF8_TWO_BYTE_LO(converted);
2158 /* Call the function to convert a UTF-8 encoded character to the specified case.
2159 * Note that there may be more than one character in the result.
2160 * INP is a pointer to the first byte of the input character
2161 * OUTP will be set to the first byte of the string of changed characters. It
2162 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2163 * LENP will be set to the length in bytes of the string of changed characters
2165 * The functions return the ordinal of the first character in the string of OUTP */
2166 #define CALL_UPPER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2167 #define CALL_TITLE_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2168 #define CALL_LOWER_CASE(uv, s, d, lenp) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2170 /* This additionally has the input parameter 'specials', which if non-zero will
2171 * cause this to use the specials hash for folding (meaning get full case
2172 * folding); otherwise, when zero, this implies a simple case fold */
2173 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2176 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2178 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2179 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2180 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2181 * the changed version may be longer than the original character.
2183 * The ordinal of the first character of the changed version is returned
2184 * (but note, as explained above, that there may be more.) */
2186 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2189 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2192 uvchr_to_utf8(p, c);
2193 return CALL_UPPER_CASE(c, p, p, lenp);
2197 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2199 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2202 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2205 uvchr_to_utf8(p, c);
2206 return CALL_TITLE_CASE(c, p, p, lenp);
2210 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp)
2212 /* We have the latin1-range values compiled into the core, so just use
2213 * those, converting the result to UTF-8. Since the result is always just
2214 * one character, we allow <p> to be NULL */
2216 U8 converted = toLOWER_LATIN1(c);
2219 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2224 /* Result is known to always be < 256, so can use the EIGHT_BIT
2226 *p = UTF8_EIGHT_BIT_HI(converted);
2227 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2235 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2237 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2240 return to_lower_latin1((U8) c, p, lenp);
2243 uvchr_to_utf8(p, c);
2244 return CALL_LOWER_CASE(c, p, p, lenp);
2248 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp, const unsigned int flags)
2250 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2251 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2252 * FOLD_FLAGS_FULL iff full folding is to be used;
2254 * Not to be used for locale folds
2259 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2260 PERL_UNUSED_CONTEXT;
2262 assert (! (flags & FOLD_FLAGS_LOCALE));
2264 if (UNLIKELY(c == MICRO_SIGN)) {
2265 converted = GREEK_SMALL_LETTER_MU;
2267 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2268 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2269 || UNICODE_DOT_DOT_VERSION > 0)
2270 else if ( (flags & FOLD_FLAGS_FULL)
2271 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2273 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2274 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2275 * under those circumstances. */
2276 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2277 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2278 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2280 return LATIN_SMALL_LETTER_LONG_S;
2290 else { /* In this range the fold of all other characters is their lower
2292 converted = toLOWER_LATIN1(c);
2295 if (UVCHR_IS_INVARIANT(converted)) {
2296 *p = (U8) converted;
2300 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2301 *p = UTF8_TWO_BYTE_LO(converted);
2309 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2312 /* Not currently externally documented, and subject to change
2313 * <flags> bits meanings:
2314 * FOLD_FLAGS_FULL iff full folding is to be used;
2315 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2316 * locale are to be used.
2317 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2320 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2322 if (flags & FOLD_FLAGS_LOCALE) {
2323 /* Treat a UTF-8 locale as not being in locale at all */
2324 if (IN_UTF8_CTYPE_LOCALE) {
2325 flags &= ~FOLD_FLAGS_LOCALE;
2328 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2329 goto needs_full_generality;
2334 return _to_fold_latin1((U8) c, p, lenp,
2335 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2338 /* Here, above 255. If no special needs, just use the macro */
2339 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2340 uvchr_to_utf8(p, c);
2341 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2343 else { /* Otherwise, _to_utf8_fold_flags has the intelligence to deal with
2344 the special flags. */
2345 U8 utf8_c[UTF8_MAXBYTES + 1];
2347 needs_full_generality:
2348 uvchr_to_utf8(utf8_c, c);
2349 return _to_utf8_fold_flags(utf8_c, p, lenp, flags);
2353 PERL_STATIC_INLINE bool
2354 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2355 const char *const swashname, SV* const invlist)
2357 /* returns a boolean giving whether or not the UTF8-encoded character that
2358 * starts at <p> is in the swash indicated by <swashname>. <swash>
2359 * contains a pointer to where the swash indicated by <swashname>
2360 * is to be stored; which this routine will do, so that future calls will
2361 * look at <*swash> and only generate a swash if it is not null. <invlist>
2362 * is NULL or an inversion list that defines the swash. If not null, it
2363 * saves time during initialization of the swash.
2365 * Note that it is assumed that the buffer length of <p> is enough to
2366 * contain all the bytes that comprise the character. Thus, <*p> should
2367 * have been checked before this call for mal-formedness enough to assure
2370 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2372 /* The API should have included a length for the UTF-8 character in <p>,
2373 * but it doesn't. We therefore assume that p has been validated at least
2374 * as far as there being enough bytes available in it to accommodate the
2375 * character without reading beyond the end, and pass that number on to the
2376 * validating routine */
2377 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2378 if (ckWARN_d(WARN_UTF8)) {
2379 Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED,WARN_UTF8),
2380 "Passing malformed UTF-8 to \"%s\" is deprecated", swashname);
2381 if (ckWARN(WARN_UTF8)) { /* This will output details as to the
2382 what the malformation is */
2383 utf8_to_uvchr_buf(p, p + UTF8SKIP(p), NULL);
2389 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2390 *swash = _core_swash_init("utf8",
2392 /* Only use the name if there is no inversion
2393 * list; otherwise will go out to disk */
2394 (invlist) ? "" : swashname,
2396 &PL_sv_undef, 1, 0, invlist, &flags);
2399 return swash_fetch(*swash, p, TRUE) != 0;
2403 Perl__is_utf8_FOO(pTHX_ const U8 classnum, const U8 *p)
2405 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2407 assert(classnum < _FIRST_NON_SWASH_CC);
2409 return is_utf8_common(p,
2410 &PL_utf8_swash_ptrs[classnum],
2411 swash_property_names[classnum],
2412 PL_XPosix_ptrs[classnum]);
2416 Perl__is_utf8_perl_idstart(pTHX_ const U8 *p)
2420 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART;
2422 if (! PL_utf8_perl_idstart) {
2423 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2425 return is_utf8_common(p, &PL_utf8_perl_idstart, "_Perl_IDStart", invlist);
2429 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2431 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2435 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2439 Perl__is_utf8_perl_idcont(pTHX_ const U8 *p)
2443 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT;
2445 if (! PL_utf8_perl_idcont) {
2446 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2448 return is_utf8_common(p, &PL_utf8_perl_idcont, "_Perl_IDCont", invlist);
2452 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2454 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2456 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2460 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2462 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2464 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2468 Perl__is_utf8_mark(pTHX_ const U8 *p)
2470 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2472 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2476 =for apidoc to_utf8_case
2478 Instead use the appropriate one of L</toUPPER_utf8>,
2483 C<p> contains the pointer to the UTF-8 string encoding
2484 the character that is being converted. This routine assumes that the character
2485 at C<p> is well-formed.
2487 C<ustrp> is a pointer to the character buffer to put the
2488 conversion result to. C<lenp> is a pointer to the length
2491 C<swashp> is a pointer to the swash to use.
2493 Both the special and normal mappings are stored in F<lib/unicore/To/Foo.pl>,
2494 and loaded by C<SWASHNEW>, using F<lib/utf8_heavy.pl>. C<special> (usually,
2495 but not always, a multicharacter mapping), is tried first.
2497 C<special> is a string, normally C<NULL> or C<"">. C<NULL> means to not use
2498 any special mappings; C<""> means to use the special mappings. Values other
2499 than these two are treated as the name of the hash containing the special
2500 mappings, like C<"utf8::ToSpecLower">.
2502 C<normal> is a string like C<"ToLower"> which means the swash
2505 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
2506 unless those are turned off.
2511 Perl_to_utf8_case(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp,
2512 SV **swashp, const char *normal, const char *special)
2514 PERL_ARGS_ASSERT_TO_UTF8_CASE;
2516 return _to_utf8_case(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, swashp, normal, special);
2519 /* change namve uv1 to 'from' */
2521 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2522 SV **swashp, const char *normal, const char *special)
2526 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2528 /* For code points that don't change case, we already know that the output
2529 * of this function is the unchanged input, so we can skip doing look-ups
2530 * for them. Unfortunately the case-changing code points are scattered
2531 * around. But there are some long consecutive ranges where there are no
2532 * case changing code points. By adding tests, we can eliminate the lookup
2533 * for all the ones in such ranges. This is currently done here only for
2534 * just a few cases where the scripts are in common use in modern commerce
2535 * (and scripts adjacent to those which can be included without additional
2538 if (uv1 >= 0x0590) {
2539 /* This keeps from needing further processing the code points most
2540 * likely to be used in the following non-cased scripts: Hebrew,
2541 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2542 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2543 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2548 /* The following largish code point ranges also don't have case
2549 * changes, but khw didn't think they warranted extra tests to speed
2550 * them up (which would slightly slow down everything else above them):
2551 * 1100..139F Hangul Jamo, Ethiopic
2552 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2553 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2554 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2555 * Combining Diacritical Marks Extended, Balinese,
2556 * Sundanese, Batak, Lepcha, Ol Chiki
2557 * 2000..206F General Punctuation
2560 if (uv1 >= 0x2D30) {
2562 /* This keeps the from needing further processing the code points
2563 * most likely to be used in the following non-cased major scripts:
2564 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2566 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2567 * event that Unicode eventually allocates the unused block as of
2568 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2569 * that the test suite will start having failures to alert you
2570 * should that happen) */
2575 if (uv1 >= 0xAC00) {
2576 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2577 if (ckWARN_d(WARN_SURROGATE)) {
2578 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2579 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2580 "Operation \"%s\" returns its argument for UTF-16 surrogate U+%04" UVXf, desc, uv1);
2585 /* AC00..FAFF Catches Hangul syllables and private use, plus
2592 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2593 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2594 && ckWARN_d(WARN_DEPRECATED))
2596 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2597 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2599 if (ckWARN_d(WARN_NON_UNICODE)) {
2600 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2601 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2602 "Operation \"%s\" returns its argument for non-Unicode code point 0x%04" UVXf, desc, uv1);
2606 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2608 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2611 /* As of this writing, this means we avoid swash creation
2612 * for anything beyond low Plane 1 */
2619 /* Note that non-characters are perfectly legal, so no warning should
2620 * be given. There are so few of them, that it isn't worth the extra
2621 * tests to avoid swash creation */
2624 if (!*swashp) /* load on-demand */
2625 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef, 4, 0, NULL, NULL);
2628 /* It might be "special" (sometimes, but not always,
2629 * a multicharacter mapping) */
2633 /* If passed in the specials name, use that; otherwise use any
2634 * given in the swash */
2635 if (*special != '\0') {
2636 hv = get_hv(special, 0);
2639 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
2641 hv = MUTABLE_HV(SvRV(*svp));
2646 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
2651 s = SvPV_const(*svp, len);
2654 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
2656 Copy(s, ustrp, len, U8);
2661 if (!len && *swashp) {
2662 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
2665 /* It was "normal" (a single character mapping). */
2666 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
2674 return valid_utf8_to_uvchr(ustrp, 0);
2677 /* Here, there was no mapping defined, which means that the code point maps
2678 * to itself. Return the inputs */
2681 if (p != ustrp) { /* Don't copy onto itself */
2682 Copy(p, ustrp, len, U8);
2693 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result, U8* const ustrp, STRLEN *lenp)
2695 /* This is called when changing the case of a UTF-8-encoded character above
2696 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
2697 * result contains a character that crosses the 255/256 boundary, disallow
2698 * the change, and return the original code point. See L<perlfunc/lc> for
2701 * p points to the original string whose case was changed; assumed
2702 * by this routine to be well-formed
2703 * result the code point of the first character in the changed-case string
2704 * ustrp points to the changed-case string (<result> represents its first char)
2705 * lenp points to the length of <ustrp> */
2707 UV original; /* To store the first code point of <p> */
2709 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
2711 assert(UTF8_IS_ABOVE_LATIN1(*p));
2713 /* We know immediately if the first character in the string crosses the
2714 * boundary, so can skip */
2717 /* Look at every character in the result; if any cross the
2718 * boundary, the whole thing is disallowed */
2719 U8* s = ustrp + UTF8SKIP(ustrp);
2720 U8* e = ustrp + *lenp;
2722 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
2728 /* Here, no characters crossed, result is ok as-is, but we warn. */
2729 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
2735 /* Failed, have to return the original */
2736 original = valid_utf8_to_uvchr(p, lenp);
2738 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
2739 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
2740 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8 locale; "
2741 "resolved to \"\\x{%" UVXf "}\".",
2745 Copy(p, ustrp, *lenp, char);
2750 =for apidoc to_utf8_upper
2752 Instead use L</toUPPER_utf8>.
2756 /* Not currently externally documented, and subject to change:
2757 * <flags> is set iff iff the rules from the current underlying locale are to
2761 Perl__to_utf8_upper_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2765 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
2768 /* Treat a UTF-8 locale as not being in locale at all */
2769 if (IN_UTF8_CTYPE_LOCALE) {
2773 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2777 if (UTF8_IS_INVARIANT(*p)) {
2779 result = toUPPER_LC(*p);
2782 return _to_upper_title_latin1(*p, ustrp, lenp, 'S');
2785 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2787 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2788 result = toUPPER_LC(c);
2791 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2795 else { /* UTF-8, ord above 255 */
2796 result = CALL_UPPER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
2799 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2804 /* Here, used locale rules. Convert back to UTF-8 */
2805 if (UTF8_IS_INVARIANT(result)) {
2806 *ustrp = (U8) result;
2810 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2811 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2819 =for apidoc to_utf8_title
2821 Instead use L</toTITLE_utf8>.
2825 /* Not currently externally documented, and subject to change:
2826 * <flags> is set iff the rules from the current underlying locale are to be
2827 * used. Since titlecase is not defined in POSIX, for other than a
2828 * UTF-8 locale, uppercase is used instead for code points < 256.
2832 Perl__to_utf8_title_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2836 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
2839 /* Treat a UTF-8 locale as not being in locale at all */
2840 if (IN_UTF8_CTYPE_LOCALE) {
2844 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2848 if (UTF8_IS_INVARIANT(*p)) {
2850 result = toUPPER_LC(*p);
2853 return _to_upper_title_latin1(*p, ustrp, lenp, 's');
2856 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2858 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2859 result = toUPPER_LC(c);
2862 return _to_upper_title_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2866 else { /* UTF-8, ord above 255 */
2867 result = CALL_TITLE_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
2870 result = check_locale_boundary_crossing(p, result, ustrp, lenp);
2875 /* Here, used locale rules. Convert back to UTF-8 */
2876 if (UTF8_IS_INVARIANT(result)) {
2877 *ustrp = (U8) result;
2881 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
2882 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
2890 =for apidoc to_utf8_lower
2892 Instead use L</toLOWER_utf8>.
2896 /* Not currently externally documented, and subject to change:
2897 * <flags> is set iff iff the rules from the current underlying locale are to
2902 Perl__to_utf8_lower_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, bool flags)
2906 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
2909 /* Treat a UTF-8 locale as not being in locale at all */
2910 if (IN_UTF8_CTYPE_LOCALE) {
2914 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2918 if (UTF8_IS_INVARIANT(*p)) {
2920 result = toLOWER_LC(*p);
2923 return to_lower_latin1(*p, ustrp, lenp);
2926 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
2928 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
2929 result = toLOWER_LC(c);
2932 return to_lower_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
2936 else { /* UTF-8, ord above 255 */
2937 result = CALL_LOWER_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp);
2940 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);
2961 =for apidoc to_utf8_fold
2963 Instead use L</toFOLD_utf8>.
2967 /* Not currently externally documented, and subject to change,
2969 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2970 * locale are to be used.
2971 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
2972 * otherwise simple folds
2973 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
2978 Perl__to_utf8_fold_flags(pTHX_ const U8 *p, U8* ustrp, STRLEN *lenp, U8 flags)
2982 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
2984 /* These are mutually exclusive */
2985 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
2987 assert(p != ustrp); /* Otherwise overwrites */
2989 if (flags & FOLD_FLAGS_LOCALE) {
2990 /* Treat a UTF-8 locale as not being in locale at all */
2991 if (IN_UTF8_CTYPE_LOCALE) {
2992 flags &= ~FOLD_FLAGS_LOCALE;
2995 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2999 if (UTF8_IS_INVARIANT(*p)) {
3000 if (flags & FOLD_FLAGS_LOCALE) {
3001 result = toFOLD_LC(*p);
3004 return _to_fold_latin1(*p, ustrp, lenp,
3005 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3008 else if UTF8_IS_DOWNGRADEABLE_START(*p) {
3009 if (flags & FOLD_FLAGS_LOCALE) {
3010 U8 c = EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1));
3011 result = toFOLD_LC(c);
3014 return _to_fold_latin1(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)),
3016 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
3019 else { /* UTF-8, ord above 255 */
3020 result = CALL_FOLD_CASE(valid_utf8_to_uvchr(p, NULL), p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3022 if (flags & FOLD_FLAGS_LOCALE) {
3024 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3025 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3027 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3028 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3030 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3032 /* Special case these two characters, as what normally gets
3033 * returned under locale doesn't work */
3034 if (UTF8SKIP(p) == cap_sharp_s_len
3035 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3037 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3038 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3039 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3040 "resolved to \"\\x{17F}\\x{17F}\".");
3045 if (UTF8SKIP(p) == long_s_t_len
3046 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3048 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3049 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3050 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3051 "resolved to \"\\x{FB06}\".");
3052 goto return_ligature_st;
3055 #if UNICODE_MAJOR_VERSION == 3 \
3056 && UNICODE_DOT_VERSION == 0 \
3057 && UNICODE_DOT_DOT_VERSION == 1
3058 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3060 /* And special case this on this Unicode version only, for the same
3061 * reaons the other two are special cased. They would cross the
3062 * 255/256 boundary which is forbidden under /l, and so the code
3063 * wouldn't catch that they are equivalent (which they are only in
3065 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3066 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3068 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3069 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3070 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3071 "resolved to \"\\x{0131}\".");
3072 goto return_dotless_i;
3076 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3078 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3082 /* This is called when changing the case of a UTF-8-encoded
3083 * character above the ASCII range, and the result should not
3084 * contain an ASCII character. */
3086 UV original; /* To store the first code point of <p> */
3088 /* Look at every character in the result; if any cross the
3089 * boundary, the whole thing is disallowed */
3091 U8* e = ustrp + *lenp;
3094 /* Crossed, have to return the original */
3095 original = valid_utf8_to_uvchr(p, lenp);
3097 /* But in these instances, there is an alternative we can
3098 * return that is valid */
3099 if (original == LATIN_SMALL_LETTER_SHARP_S
3100 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3101 || original == LATIN_CAPITAL_LETTER_SHARP_S
3106 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3107 goto return_ligature_st;
3109 #if UNICODE_MAJOR_VERSION == 3 \
3110 && UNICODE_DOT_VERSION == 0 \
3111 && UNICODE_DOT_DOT_VERSION == 1
3113 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3114 goto return_dotless_i;
3117 Copy(p, ustrp, *lenp, char);
3123 /* Here, no characters crossed, result is ok as-is */
3128 /* Here, used locale rules. Convert back to UTF-8 */
3129 if (UTF8_IS_INVARIANT(result)) {
3130 *ustrp = (U8) result;
3134 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3135 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3142 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3143 * folds to a string of two of these characters. By returning this
3144 * instead, then, e.g.,
3145 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3148 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3149 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3151 return LATIN_SMALL_LETTER_LONG_S;
3154 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3155 * have the other one fold to it */
3157 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3158 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3159 return LATIN_SMALL_LIGATURE_ST;
3161 #if UNICODE_MAJOR_VERSION == 3 \
3162 && UNICODE_DOT_VERSION == 0 \
3163 && UNICODE_DOT_DOT_VERSION == 1
3166 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3167 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3168 return LATIN_SMALL_LETTER_DOTLESS_I;
3175 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3176 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3177 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3181 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none)
3183 PERL_ARGS_ASSERT_SWASH_INIT;
3185 /* Returns a copy of a swash initiated by the called function. This is the
3186 * public interface, and returning a copy prevents others from doing
3187 * mischief on the original */
3189 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none, NULL, NULL));
3193 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv, I32 minbits, I32 none, SV* invlist, U8* const flags_p)
3196 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3197 * use the following define */
3199 #define CORE_SWASH_INIT_RETURN(x) \
3200 PL_curpm= old_PL_curpm; \
3203 /* Initialize and return a swash, creating it if necessary. It does this
3204 * by calling utf8_heavy.pl in the general case. The returned value may be
3205 * the swash's inversion list instead if the input parameters allow it.
3206 * Which is returned should be immaterial to callers, as the only
3207 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3208 * and swash_to_invlist() handle both these transparently.
3210 * This interface should only be used by functions that won't destroy or
3211 * adversely change the swash, as doing so affects all other uses of the
3212 * swash in the program; the general public should use 'Perl_swash_init'
3215 * pkg is the name of the package that <name> should be in.
3216 * name is the name of the swash to find. Typically it is a Unicode
3217 * property name, including user-defined ones
3218 * listsv is a string to initialize the swash with. It must be of the form
3219 * documented as the subroutine return value in
3220 * L<perlunicode/User-Defined Character Properties>
3221 * minbits is the number of bits required to represent each data element.
3222 * It is '1' for binary properties.
3223 * none I (khw) do not understand this one, but it is used only in tr///.
3224 * invlist is an inversion list to initialize the swash with (or NULL)
3225 * flags_p if non-NULL is the address of various input and output flag bits
3226 * to the routine, as follows: ('I' means is input to the routine;
3227 * 'O' means output from the routine. Only flags marked O are
3228 * meaningful on return.)
3229 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3230 * came from a user-defined property. (I O)
3231 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3232 * when the swash cannot be located, to simply return NULL. (I)
3233 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3234 * return of an inversion list instead of a swash hash if this routine
3235 * thinks that would result in faster execution of swash_fetch() later
3238 * Thus there are three possible inputs to find the swash: <name>,
3239 * <listsv>, and <invlist>. At least one must be specified. The result
3240 * will be the union of the specified ones, although <listsv>'s various
3241 * actions can intersect, etc. what <name> gives. To avoid going out to
3242 * disk at all, <invlist> should specify completely what the swash should
3243 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3245 * <invlist> is only valid for binary properties */
3247 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3249 SV* retval = &PL_sv_undef;
3250 HV* swash_hv = NULL;
3251 const int invlist_swash_boundary =
3252 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3253 ? 512 /* Based on some benchmarking, but not extensive, see commit
3255 : -1; /* Never return just an inversion list */
3257 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3258 assert(! invlist || minbits == 1);
3260 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the regex
3261 that triggered the swash init and the swash init perl logic itself.
3264 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3266 if (listsv != &PL_sv_undef || strNE(name, "")) {
3268 const size_t pkg_len = strlen(pkg);
3269 const size_t name_len = strlen(name);
3270 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3274 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3276 PUSHSTACKi(PERLSI_MAGIC);
3280 /* We might get here via a subroutine signature which uses a utf8
3281 * parameter name, at which point PL_subname will have been set
3282 * but not yet used. */
3283 save_item(PL_subname);
3284 if (PL_parser && PL_parser->error_count)
3285 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3286 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3287 if (!method) { /* demand load UTF-8 */
3289 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3290 GvSV(PL_errgv) = NULL;
3291 #ifndef NO_TAINT_SUPPORT
3292 /* It is assumed that callers of this routine are not passing in
3293 * any user derived data. */
3294 /* Need to do this after save_re_context() as it will set
3295 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3296 * in Perl_magic_get). Even line to create errsv_save can turn on
3298 SAVEBOOL(TAINT_get);
3301 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3304 /* Not ERRSV, as there is no need to vivify a scalar we are
3305 about to discard. */
3306 SV * const errsv = GvSV(PL_errgv);
3307 if (!SvTRUE(errsv)) {
3308 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3309 SvREFCNT_dec(errsv);
3317 mPUSHp(pkg, pkg_len);
3318 mPUSHp(name, name_len);
3323 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3324 GvSV(PL_errgv) = NULL;
3325 /* If we already have a pointer to the method, no need to use
3326 * call_method() to repeat the lookup. */
3328 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3329 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3331 retval = *PL_stack_sp--;
3332 SvREFCNT_inc(retval);
3335 /* Not ERRSV. See above. */
3336 SV * const errsv = GvSV(PL_errgv);
3337 if (!SvTRUE(errsv)) {
3338 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3339 SvREFCNT_dec(errsv);
3344 if (IN_PERL_COMPILETIME) {
3345 CopHINTS_set(PL_curcop, PL_hints);
3347 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3348 if (SvPOK(retval)) {
3350 /* If caller wants to handle missing properties, let them */
3351 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3352 CORE_SWASH_INIT_RETURN(NULL);
3355 "Can't find Unicode property definition \"%" SVf "\"",
3357 NOT_REACHED; /* NOTREACHED */
3360 } /* End of calling the module to find the swash */
3362 /* If this operation fetched a swash, and we will need it later, get it */
3363 if (retval != &PL_sv_undef
3364 && (minbits == 1 || (flags_p
3366 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3368 swash_hv = MUTABLE_HV(SvRV(retval));
3370 /* If we don't already know that there is a user-defined component to
3371 * this swash, and the user has indicated they wish to know if there is
3372 * one (by passing <flags_p>), find out */
3373 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3374 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3375 if (user_defined && SvUV(*user_defined)) {
3376 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3381 /* Make sure there is an inversion list for binary properties */
3383 SV** swash_invlistsvp = NULL;
3384 SV* swash_invlist = NULL;
3385 bool invlist_in_swash_is_valid = FALSE;
3386 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3387 an unclaimed reference count */
3389 /* If this operation fetched a swash, get its already existing
3390 * inversion list, or create one for it */
3393 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3394 if (swash_invlistsvp) {
3395 swash_invlist = *swash_invlistsvp;
3396 invlist_in_swash_is_valid = TRUE;
3399 swash_invlist = _swash_to_invlist(retval);
3400 swash_invlist_unclaimed = TRUE;
3404 /* If an inversion list was passed in, have to include it */
3407 /* Any fetched swash will by now have an inversion list in it;
3408 * otherwise <swash_invlist> will be NULL, indicating that we
3409 * didn't fetch a swash */
3410 if (swash_invlist) {
3412 /* Add the passed-in inversion list, which invalidates the one
3413 * already stored in the swash */
3414 invlist_in_swash_is_valid = FALSE;
3415 SvREADONLY_off(swash_invlist); /* Turned on again below */
3416 _invlist_union(invlist, swash_invlist, &swash_invlist);
3420 /* Here, there is no swash already. Set up a minimal one, if
3421 * we are going to return a swash */
3422 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3424 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3426 swash_invlist = invlist;
3430 /* Here, we have computed the union of all the passed-in data. It may
3431 * be that there was an inversion list in the swash which didn't get
3432 * touched; otherwise save the computed one */
3433 if (! invlist_in_swash_is_valid
3434 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3436 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3438 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3440 /* We just stole a reference count. */
3441 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3442 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3445 /* The result is immutable. Forbid attempts to change it. */
3446 SvREADONLY_on(swash_invlist);
3448 /* Use the inversion list stand-alone if small enough */
3449 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3450 SvREFCNT_dec(retval);
3451 if (!swash_invlist_unclaimed)
3452 SvREFCNT_inc_simple_void_NN(swash_invlist);
3453 retval = newRV_noinc(swash_invlist);
3457 CORE_SWASH_INIT_RETURN(retval);
3458 #undef CORE_SWASH_INIT_RETURN
3462 /* This API is wrong for special case conversions since we may need to
3463 * return several Unicode characters for a single Unicode character
3464 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3465 * the lower-level routine, and it is similarly broken for returning
3466 * multiple values. --jhi
3467 * For those, you should use S__to_utf8_case() instead */
3468 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3471 * Returns the value of property/mapping C<swash> for the first character
3472 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3473 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3474 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3476 * A "swash" is a hash which contains initially the keys/values set up by
3477 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3478 * property for all possible code points. Things are stored in a compact form
3479 * (see utf8_heavy.pl) so that calculation is required to find the actual
3480 * property value for a given code point. As code points are looked up, new
3481 * key/value pairs are added to the hash, so that the calculation doesn't have
3482 * to ever be re-done. Further, each calculation is done, not just for the
3483 * desired one, but for a whole block of code points adjacent to that one.
3484 * For binary properties on ASCII machines, the block is usually for 64 code
3485 * points, starting with a code point evenly divisible by 64. Thus if the
3486 * property value for code point 257 is requested, the code goes out and
3487 * calculates the property values for all 64 code points between 256 and 319,
3488 * and stores these as a single 64-bit long bit vector, called a "swatch",
3489 * under the key for code point 256. The key is the UTF-8 encoding for code
3490 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3491 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3492 * for code point 258 is then requested, this code realizes that it would be
3493 * stored under the key for 256, and would find that value and extract the
3494 * relevant bit, offset from 256.
3496 * Non-binary properties are stored in as many bits as necessary to represent
3497 * their values (32 currently, though the code is more general than that), not
3498 * as single bits, but the principle is the same: the value for each key is a
3499 * vector that encompasses the property values for all code points whose UTF-8
3500 * representations are represented by the key. That is, for all code points
3501 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3505 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3507 HV *const hv = MUTABLE_HV(SvRV(swash));
3512 const U8 *tmps = NULL;
3516 PERL_ARGS_ASSERT_SWASH_FETCH;
3518 /* If it really isn't a hash, it isn't really swash; must be an inversion
3520 if (SvTYPE(hv) != SVt_PVHV) {
3521 return _invlist_contains_cp((SV*)hv,
3523 ? valid_utf8_to_uvchr(ptr, NULL)
3527 /* We store the values in a "swatch" which is a vec() value in a swash
3528 * hash. Code points 0-255 are a single vec() stored with key length
3529 * (klen) 0. All other code points have a UTF-8 representation
3530 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3531 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3532 * length for them is the length of the encoded char - 1. ptr[klen] is the
3533 * final byte in the sequence representing the character */
3534 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3539 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3542 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3545 klen = UTF8SKIP(ptr) - 1;
3547 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3548 * the vec is the final byte in the sequence. (In EBCDIC this is
3549 * converted to I8 to get consecutive values.) To help you visualize
3551 * Straight 1047 After final byte
3552 * UTF-8 UTF-EBCDIC I8 transform
3553 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3554 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3556 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3557 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3559 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3560 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3562 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3563 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3565 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3566 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3568 * (There are no discontinuities in the elided (...) entries.)
3569 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3570 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3571 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3572 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3573 * index into the vec() swatch (after subtracting 0x80, which we
3574 * actually do with an '&').
3575 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3576 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
3577 * dicontinuities which go away by transforming it into I8, and we
3578 * effectively subtract 0xA0 to get the index. */
3579 needents = (1 << UTF_ACCUMULATION_SHIFT);
3580 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
3584 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
3585 * suite. (That is, only 7-8% overall over just a hash cache. Still,
3586 * it's nothing to sniff at.) Pity we usually come through at least
3587 * two function calls to get here...
3589 * NB: this code assumes that swatches are never modified, once generated!
3592 if (hv == PL_last_swash_hv &&
3593 klen == PL_last_swash_klen &&
3594 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
3596 tmps = PL_last_swash_tmps;
3597 slen = PL_last_swash_slen;
3600 /* Try our second-level swatch cache, kept in a hash. */
3601 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
3603 /* If not cached, generate it via swatch_get */
3604 if (!svp || !SvPOK(*svp)
3605 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
3608 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
3609 swatch = swatch_get(swash,
3610 code_point & ~((UV)needents - 1),
3613 else { /* For the first 256 code points, the swatch has a key of
3615 swatch = swatch_get(swash, 0, needents);
3618 if (IN_PERL_COMPILETIME)
3619 CopHINTS_set(PL_curcop, PL_hints);
3621 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
3623 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
3624 || (slen << 3) < needents)
3625 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
3626 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
3627 svp, tmps, (UV)slen, (UV)needents);
3630 PL_last_swash_hv = hv;
3631 assert(klen <= sizeof(PL_last_swash_key));
3632 PL_last_swash_klen = (U8)klen;
3633 /* FIXME change interpvar.h? */
3634 PL_last_swash_tmps = (U8 *) tmps;
3635 PL_last_swash_slen = slen;
3637 Copy(ptr, PL_last_swash_key, klen, U8);
3640 switch ((int)((slen << 3) / needents)) {
3642 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
3644 return ((UV) tmps[off]);
3648 ((UV) tmps[off ] << 8) +
3649 ((UV) tmps[off + 1]);
3653 ((UV) tmps[off ] << 24) +
3654 ((UV) tmps[off + 1] << 16) +
3655 ((UV) tmps[off + 2] << 8) +
3656 ((UV) tmps[off + 3]);
3658 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
3659 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
3660 NORETURN_FUNCTION_END;
3663 /* Read a single line of the main body of the swash input text. These are of
3666 * where each number is hex. The first two numbers form the minimum and
3667 * maximum of a range, and the third is the value associated with the range.
3668 * Not all swashes should have a third number
3670 * On input: l points to the beginning of the line to be examined; it points
3671 * to somewhere in the string of the whole input text, and is
3672 * terminated by a \n or the null string terminator.
3673 * lend points to the null terminator of that string
3674 * wants_value is non-zero if the swash expects a third number
3675 * typestr is the name of the swash's mapping, like 'ToLower'
3676 * On output: *min, *max, and *val are set to the values read from the line.
3677 * returns a pointer just beyond the line examined. If there was no
3678 * valid min number on the line, returns lend+1
3682 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
3683 const bool wants_value, const U8* const typestr)
3685 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
3686 STRLEN numlen; /* Length of the number */
3687 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
3688 | PERL_SCAN_DISALLOW_PREFIX
3689 | PERL_SCAN_SILENT_NON_PORTABLE;
3691 /* nl points to the next \n in the scan */
3692 U8* const nl = (U8*)memchr(l, '\n', lend - l);
3694 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
3696 /* Get the first number on the line: the range minimum */
3698 *min = grok_hex((char *)l, &numlen, &flags, NULL);
3699 *max = *min; /* So can never return without setting max */
3700 if (numlen) /* If found a hex number, position past it */
3702 else if (nl) { /* Else, go handle next line, if any */
3703 return nl + 1; /* 1 is length of "\n" */
3705 else { /* Else, no next line */
3706 return lend + 1; /* to LIST's end at which \n is not found */
3709 /* The max range value follows, separated by a BLANK */
3712 flags = PERL_SCAN_SILENT_ILLDIGIT
3713 | PERL_SCAN_DISALLOW_PREFIX
3714 | PERL_SCAN_SILENT_NON_PORTABLE;
3716 *max = grok_hex((char *)l, &numlen, &flags, NULL);
3719 else /* If no value here, it is a single element range */
3722 /* Non-binary tables have a third entry: what the first element of the
3723 * range maps to. The map for those currently read here is in hex */
3727 flags = PERL_SCAN_SILENT_ILLDIGIT
3728 | PERL_SCAN_DISALLOW_PREFIX
3729 | PERL_SCAN_SILENT_NON_PORTABLE;
3731 *val = grok_hex((char *)l, &numlen, &flags, NULL);
3740 /* diag_listed_as: To%s: illegal mapping '%s' */
3741 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
3747 *val = 0; /* bits == 1, then any val should be ignored */
3749 else { /* Nothing following range min, should be single element with no
3754 /* diag_listed_as: To%s: illegal mapping '%s' */
3755 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
3759 *val = 0; /* bits == 1, then val should be ignored */
3762 /* Position to next line if any, or EOF */
3772 * Returns a swatch (a bit vector string) for a code point sequence
3773 * that starts from the value C<start> and comprises the number C<span>.
3774 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
3775 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
3778 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
3781 U8 *l, *lend, *x, *xend, *s, *send;
3782 STRLEN lcur, xcur, scur;
3783 HV *const hv = MUTABLE_HV(SvRV(swash));
3784 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
3786 SV** listsvp = NULL; /* The string containing the main body of the table */
3787 SV** extssvp = NULL;
3788 SV** invert_it_svp = NULL;
3791 STRLEN octets; /* if bits == 1, then octets == 0 */
3793 UV end = start + span;
3795 if (invlistsvp == NULL) {
3796 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
3797 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
3798 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
3799 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
3800 listsvp = hv_fetchs(hv, "LIST", FALSE);
3801 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
3803 bits = SvUV(*bitssvp);
3804 none = SvUV(*nonesvp);
3805 typestr = (U8*)SvPV_nolen(*typesvp);
3811 octets = bits >> 3; /* if bits == 1, then octets == 0 */
3813 PERL_ARGS_ASSERT_SWATCH_GET;
3815 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
3816 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
3820 /* If overflowed, use the max possible */
3826 /* create and initialize $swatch */
3827 scur = octets ? (span * octets) : (span + 7) / 8;
3828 swatch = newSV(scur);
3830 s = (U8*)SvPVX(swatch);
3831 if (octets && none) {
3832 const U8* const e = s + scur;
3835 *s++ = (U8)(none & 0xff);
3836 else if (bits == 16) {
3837 *s++ = (U8)((none >> 8) & 0xff);
3838 *s++ = (U8)( none & 0xff);
3840 else if (bits == 32) {
3841 *s++ = (U8)((none >> 24) & 0xff);
3842 *s++ = (U8)((none >> 16) & 0xff);
3843 *s++ = (U8)((none >> 8) & 0xff);
3844 *s++ = (U8)( none & 0xff);
3850 (void)memzero((U8*)s, scur + 1);
3852 SvCUR_set(swatch, scur);
3853 s = (U8*)SvPVX(swatch);
3855 if (invlistsvp) { /* If has an inversion list set up use that */
3856 _invlist_populate_swatch(*invlistsvp, start, end, s);
3860 /* read $swash->{LIST} */
3861 l = (U8*)SvPV(*listsvp, lcur);
3864 UV min, max, val, upper;
3865 l = swash_scan_list_line(l, lend, &min, &max, &val,
3866 cBOOL(octets), typestr);
3871 /* If looking for something beyond this range, go try the next one */
3875 /* <end> is generally 1 beyond where we want to set things, but at the
3876 * platform's infinity, where we can't go any higher, we want to
3877 * include the code point at <end> */
3880 : (max != UV_MAX || end != UV_MAX)
3887 if (!none || val < none) {
3892 for (key = min; key <= upper; key++) {
3894 /* offset must be non-negative (start <= min <= key < end) */
3895 offset = octets * (key - start);
3897 s[offset] = (U8)(val & 0xff);
3898 else if (bits == 16) {
3899 s[offset ] = (U8)((val >> 8) & 0xff);
3900 s[offset + 1] = (U8)( val & 0xff);
3902 else if (bits == 32) {
3903 s[offset ] = (U8)((val >> 24) & 0xff);
3904 s[offset + 1] = (U8)((val >> 16) & 0xff);
3905 s[offset + 2] = (U8)((val >> 8) & 0xff);
3906 s[offset + 3] = (U8)( val & 0xff);
3909 if (!none || val < none)
3913 else { /* bits == 1, then val should be ignored */
3918 for (key = min; key <= upper; key++) {
3919 const STRLEN offset = (STRLEN)(key - start);
3920 s[offset >> 3] |= 1 << (offset & 7);
3925 /* Invert if the data says it should be. Assumes that bits == 1 */
3926 if (invert_it_svp && SvUV(*invert_it_svp)) {
3928 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
3929 * be 0, and their inversion should also be 0, as we don't succeed any
3930 * Unicode property matches for non-Unicode code points */
3931 if (start <= PERL_UNICODE_MAX) {
3933 /* The code below assumes that we never cross the
3934 * Unicode/above-Unicode boundary in a range, as otherwise we would
3935 * have to figure out where to stop flipping the bits. Since this
3936 * boundary is divisible by a large power of 2, and swatches comes
3937 * in small powers of 2, this should be a valid assumption */
3938 assert(start + span - 1 <= PERL_UNICODE_MAX);
3948 /* read $swash->{EXTRAS}
3949 * This code also copied to swash_to_invlist() below */
3950 x = (U8*)SvPV(*extssvp, xcur);
3958 SV **otherbitssvp, *other;
3962 const U8 opc = *x++;
3966 nl = (U8*)memchr(x, '\n', xend - x);
3968 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
3970 x = nl + 1; /* 1 is length of "\n" */
3974 x = xend; /* to EXTRAS' end at which \n is not found */
3981 namelen = nl - namestr;
3985 namelen = xend - namestr;
3989 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
3990 otherhv = MUTABLE_HV(SvRV(*othersvp));
3991 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
3992 otherbits = (STRLEN)SvUV(*otherbitssvp);
3993 if (bits < otherbits)
3994 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
3995 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
3997 /* The "other" swatch must be destroyed after. */
3998 other = swatch_get(*othersvp, start, span);
3999 o = (U8*)SvPV(other, olen);
4002 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4004 s = (U8*)SvPV(swatch, slen);
4005 if (bits == 1 && otherbits == 1) {
4007 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4008 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4009 (UV)slen, (UV)olen);
4033 STRLEN otheroctets = otherbits >> 3;
4035 U8* const send = s + slen;
4040 if (otherbits == 1) {
4041 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4045 STRLEN vlen = otheroctets;
4053 if (opc == '+' && otherval)
4054 NOOP; /* replace with otherval */
4055 else if (opc == '!' && !otherval)
4057 else if (opc == '-' && otherval)
4059 else if (opc == '&' && !otherval)
4062 s += octets; /* no replacement */
4067 *s++ = (U8)( otherval & 0xff);
4068 else if (bits == 16) {
4069 *s++ = (U8)((otherval >> 8) & 0xff);
4070 *s++ = (U8)( otherval & 0xff);
4072 else if (bits == 32) {
4073 *s++ = (U8)((otherval >> 24) & 0xff);
4074 *s++ = (U8)((otherval >> 16) & 0xff);
4075 *s++ = (U8)((otherval >> 8) & 0xff);
4076 *s++ = (U8)( otherval & 0xff);
4080 sv_free(other); /* through with it! */
4086 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4089 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4090 * Can't be used on a property that is subject to user override, as it
4091 * relies on the value of SPECIALS in the swash which would be set by
4092 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4093 * for overridden properties
4095 * Returns a hash which is the inversion and closure of a swash mapping.
4096 * For example, consider the input lines:
4101 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4102 * 006C. The value for each key is an array. For 006C, the array would
4103 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4104 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4106 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4107 * keys are only code points that are folded-to, so it isn't a full closure.
4109 * Essentially, for any code point, it gives all the code points that map to
4110 * it, or the list of 'froms' for that point.
4112 * Currently it ignores any additions or deletions from other swashes,
4113 * looking at just the main body of the swash, and if there are SPECIALS
4114 * in the swash, at that hash
4116 * The specials hash can be extra code points, and most likely consists of
4117 * maps from single code points to multiple ones (each expressed as a string
4118 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4119 * However consider this possible input in the specials hash:
4120 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4121 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4123 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4124 * currently handle. But it also means that FB05 and FB06 are equivalent in
4125 * a 1-1 mapping which we should handle, and this relationship may not be in
4126 * the main table. Therefore this function examines all the multi-char
4127 * sequences and adds the 1-1 mappings that come out of that.
4129 * XXX This function was originally intended to be multipurpose, but its
4130 * only use is quite likely to remain for constructing the inversion of
4131 * the CaseFolding (//i) property. If it were more general purpose for
4132 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4133 * because certain folds are prohibited under /iaa and /il. As an example,
4134 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4135 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4136 * prohibited, so we would not figure out that they fold to each other.
4137 * Code could be written to automatically figure this out, similar to the
4138 * code that does this for multi-character folds, but this is the only case
4139 * where something like this is ever likely to happen, as all the single
4140 * char folds to the 0-255 range are now quite settled. Instead there is a
4141 * little special code that is compiled only for this Unicode version. This
4142 * is smaller and didn't require much coding time to do. But this makes
4143 * this routine strongly tied to being used just for CaseFolding. If ever
4144 * it should be generalized, this would have to be fixed */
4148 HV *const hv = MUTABLE_HV(SvRV(swash));
4150 /* The string containing the main body of the table. This will have its
4151 * assertion fail if the swash has been converted to its inversion list */
4152 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4154 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4155 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4156 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4157 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4158 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4159 const STRLEN bits = SvUV(*bitssvp);
4160 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4161 const UV none = SvUV(*nonesvp);
4162 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4166 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4168 /* Must have at least 8 bits to get the mappings */
4169 if (bits != 8 && bits != 16 && bits != 32) {
4170 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %" UVuf,
4174 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4175 mapping to more than one character */
4177 /* Construct an inverse mapping hash for the specials */
4178 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4179 HV * specials_inverse = newHV();
4180 char *char_from; /* the lhs of the map */
4181 I32 from_len; /* its byte length */
4182 char *char_to; /* the rhs of the map */
4183 I32 to_len; /* its byte length */
4184 SV *sv_to; /* and in a sv */
4185 AV* from_list; /* list of things that map to each 'to' */
4187 hv_iterinit(specials_hv);
4189 /* The keys are the characters (in UTF-8) that map to the corresponding
4190 * UTF-8 string value. Iterate through the list creating the inverse
4192 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4194 if (! SvPOK(sv_to)) {
4195 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4196 "unexpectedly is not a string, flags=%lu",
4197 (unsigned long)SvFLAGS(sv_to));
4199 /*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)));*/
4201 /* Each key in the inverse list is a mapped-to value, and the key's
4202 * hash value is a list of the strings (each in UTF-8) that map to
4203 * it. Those strings are all one character long */
4204 if ((listp = hv_fetch(specials_inverse,
4208 from_list = (AV*) *listp;
4210 else { /* No entry yet for it: create one */
4211 from_list = newAV();
4212 if (! hv_store(specials_inverse,
4215 (SV*) from_list, 0))
4217 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4221 /* Here have the list associated with this 'to' (perhaps newly
4222 * created and empty). Just add to it. Note that we ASSUME that
4223 * the input is guaranteed to not have duplications, so we don't
4224 * check for that. Duplications just slow down execution time. */
4225 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4228 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4229 * it looking for cases like the FB05/FB06 examples above. There would
4230 * be an entry in the hash like
4231 * 'st' => [ FB05, FB06 ]
4232 * In this example we will create two lists that get stored in the
4233 * returned hash, 'ret':
4234 * FB05 => [ FB05, FB06 ]
4235 * FB06 => [ FB05, FB06 ]
4237 * Note that there is nothing to do if the array only has one element.
4238 * (In the normal 1-1 case handled below, we don't have to worry about
4239 * two lists, as everything gets tied to the single list that is
4240 * generated for the single character 'to'. But here, we are omitting
4241 * that list, ('st' in the example), so must have multiple lists.) */
4242 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4243 &char_to, &to_len)))
4245 if (av_tindex_nomg(from_list) > 0) {
4248 /* We iterate over all combinations of i,j to place each code
4249 * point on each list */
4250 for (i = 0; i <= av_tindex_nomg(from_list); i++) {
4252 AV* i_list = newAV();
4253 SV** entryp = av_fetch(from_list, i, FALSE);
4254 if (entryp == NULL) {
4255 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4257 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4258 Perl_croak(aTHX_ "panic: unexpected entry for %s", SvPVX(*entryp));
4260 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4261 (SV*) i_list, FALSE))
4263 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4266 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4267 for (j = 0; j <= av_tindex_nomg(from_list); j++) {
4268 entryp = av_fetch(from_list, j, FALSE);
4269 if (entryp == NULL) {
4270 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4273 /* When i==j this adds itself to the list */
4274 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4275 (U8*) SvPVX(*entryp),
4276 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4278 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0), u));*/
4283 SvREFCNT_dec(specials_inverse); /* done with it */
4284 } /* End of specials */
4286 /* read $swash->{LIST} */
4288 #if UNICODE_MAJOR_VERSION == 3 \
4289 && UNICODE_DOT_VERSION == 0 \
4290 && UNICODE_DOT_DOT_VERSION == 1
4292 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4293 * rule so that things work under /iaa and /il */
4295 SV * mod_listsv = sv_mortalcopy(*listsvp);
4296 sv_catpv(mod_listsv, "130\t130\t131\n");
4297 l = (U8*)SvPV(mod_listsv, lcur);
4301 l = (U8*)SvPV(*listsvp, lcur);
4307 /* Go through each input line */
4311 l = swash_scan_list_line(l, lend, &min, &max, &val,
4312 cBOOL(octets), typestr);
4317 /* Each element in the range is to be inverted */
4318 for (inverse = min; inverse <= max; inverse++) {
4322 bool found_key = FALSE;
4323 bool found_inverse = FALSE;
4325 /* The key is the inverse mapping */
4326 char key[UTF8_MAXBYTES+1];
4327 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4328 STRLEN key_len = key_end - key;
4330 /* Get the list for the map */
4331 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4332 list = (AV*) *listp;
4334 else { /* No entry yet for it: create one */
4336 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4337 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4341 /* Look through list to see if this inverse mapping already is
4342 * listed, or if there is a mapping to itself already */
4343 for (i = 0; i <= av_tindex_nomg(list); i++) {
4344 SV** entryp = av_fetch(list, i, FALSE);
4347 if (entryp == NULL) {
4348 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4352 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4356 if (uv == inverse) {
4357 found_inverse = TRUE;
4360 /* No need to continue searching if found everything we are
4362 if (found_key && found_inverse) {
4367 /* Make sure there is a mapping to itself on the list */
4369 av_push(list, newSVuv(val));
4370 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4374 /* Simply add the value to the list */
4375 if (! found_inverse) {
4376 av_push(list, newSVuv(inverse));
4377 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4380 /* swatch_get() increments the value of val for each element in the
4381 * range. That makes more compact tables possible. You can
4382 * express the capitalization, for example, of all consecutive
4383 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4384 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4385 * and it's not documented; it appears to be used only in
4386 * implementing tr//; I copied the semantics from swatch_get(), just
4388 if (!none || val < none) {
4398 Perl__swash_to_invlist(pTHX_ SV* const swash)
4401 /* Subject to change or removal. For use only in one place in regcomp.c.
4402 * Ownership is given to one reference count in the returned SV* */
4407 HV *const hv = MUTABLE_HV(SvRV(swash));
4408 UV elements = 0; /* Number of elements in the inversion list */
4418 STRLEN octets; /* if bits == 1, then octets == 0 */
4424 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4426 /* If not a hash, it must be the swash's inversion list instead */
4427 if (SvTYPE(hv) != SVt_PVHV) {
4428 return SvREFCNT_inc_simple_NN((SV*) hv);
4431 /* The string containing the main body of the table */
4432 listsvp = hv_fetchs(hv, "LIST", FALSE);
4433 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4434 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4435 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4436 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4438 typestr = (U8*)SvPV_nolen(*typesvp);
4439 bits = SvUV(*bitssvp);
4440 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4442 /* read $swash->{LIST} */
4443 if (SvPOK(*listsvp)) {
4444 l = (U8*)SvPV(*listsvp, lcur);
4447 /* LIST legitimately doesn't contain a string during compilation phases
4448 * of Perl itself, before the Unicode tables are generated. In this
4449 * case, just fake things up by creating an empty list */
4456 if (*l == 'V') { /* Inversion list format */
4457 const char *after_atou = (char *) lend;
4459 UV* other_elements_ptr;
4461 /* The first number is a count of the rest */
4463 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4464 Perl_croak(aTHX_ "panic: Expecting a valid count of elements at start of inversion list");
4466 if (elements == 0) {
4467 invlist = _new_invlist(0);
4470 while (isSPACE(*l)) l++;
4471 l = (U8 *) after_atou;
4473 /* Get the 0th element, which is needed to setup the inversion list */
4474 while (isSPACE(*l)) l++;
4475 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4476 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for inversion list");
4478 l = (U8 *) after_atou;
4479 invlist = _setup_canned_invlist(elements, element0, &other_elements_ptr);
4482 /* Then just populate the rest of the input */
4483 while (elements-- > 0) {
4485 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more elements than available", elements);
4487 while (isSPACE(*l)) l++;
4488 if (!grok_atoUV((const char *)l, other_elements_ptr++, &after_atou)) {
4489 Perl_croak(aTHX_ "panic: Expecting a valid element in inversion list");
4491 l = (U8 *) after_atou;
4497 /* Scan the input to count the number of lines to preallocate array
4498 * size based on worst possible case, which is each line in the input
4499 * creates 2 elements in the inversion list: 1) the beginning of a
4500 * range in the list; 2) the beginning of a range not in the list. */
4501 while ((loc = (strchr(loc, '\n'))) != NULL) {
4506 /* If the ending is somehow corrupt and isn't a new line, add another
4507 * element for the final range that isn't in the inversion list */
4508 if (! (*lend == '\n'
4509 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4514 invlist = _new_invlist(elements);
4516 /* Now go through the input again, adding each range to the list */
4519 UV val; /* Not used by this function */
4521 l = swash_scan_list_line(l, lend, &start, &end, &val,
4522 cBOOL(octets), typestr);
4528 invlist = _add_range_to_invlist(invlist, start, end);
4532 /* Invert if the data says it should be */
4533 if (invert_it_svp && SvUV(*invert_it_svp)) {
4534 _invlist_invert(invlist);
4537 /* This code is copied from swatch_get()
4538 * read $swash->{EXTRAS} */
4539 x = (U8*)SvPV(*extssvp, xcur);
4547 SV **otherbitssvp, *other;
4550 const U8 opc = *x++;
4554 nl = (U8*)memchr(x, '\n', xend - x);
4556 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4558 x = nl + 1; /* 1 is length of "\n" */
4562 x = xend; /* to EXTRAS' end at which \n is not found */
4569 namelen = nl - namestr;
4573 namelen = xend - namestr;
4577 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4578 otherhv = MUTABLE_HV(SvRV(*othersvp));
4579 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4580 otherbits = (STRLEN)SvUV(*otherbitssvp);
4582 if (bits != otherbits || bits != 1) {
4583 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
4584 "properties, bits=%" UVuf ", otherbits=%" UVuf,
4585 (UV)bits, (UV)otherbits);
4588 /* The "other" swatch must be destroyed after. */
4589 other = _swash_to_invlist((SV *)*othersvp);
4591 /* End of code copied from swatch_get() */
4594 _invlist_union(invlist, other, &invlist);
4597 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
4600 _invlist_subtract(invlist, other, &invlist);
4603 _invlist_intersection(invlist, other, &invlist);
4608 sv_free(other); /* through with it! */
4611 SvREADONLY_on(invlist);
4616 Perl__get_swash_invlist(pTHX_ SV* const swash)
4620 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
4622 if (! SvROK(swash)) {
4626 /* If it really isn't a hash, it isn't really swash; must be an inversion
4628 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
4632 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
4641 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
4643 /* May change: warns if surrogates, non-character code points, or
4644 * non-Unicode code points are in s which has length len bytes. Returns
4645 * TRUE if none found; FALSE otherwise. The only other validity check is
4646 * to make sure that this won't exceed the string's length.
4648 * Code points above the platform's C<IV_MAX> will raise a deprecation
4649 * warning, unless those are turned off. */
4651 const U8* const e = s + len;
4654 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
4657 if (UTF8SKIP(s) > len) {
4658 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
4659 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
4662 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
4663 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
4664 if ( ckWARN_d(WARN_NON_UNICODE)
4665 || ( ckWARN_d(WARN_DEPRECATED)
4667 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
4668 #else /* Below is 64-bit words */
4669 /* 2**63 and up meet these conditions provided we have
4673 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
4676 /* s[1] being above 0x80 overflows */
4681 /* A side effect of this function will be to warn */
4682 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
4686 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
4687 if (ckWARN_d(WARN_SURROGATE)) {
4688 /* This has a different warning than the one the called
4689 * function would output, so can't just call it, unlike we
4690 * do for the non-chars and above-unicodes */
4691 UV uv = utf8_to_uvchr_buf(s, e, NULL);
4692 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
4693 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8", uv);
4697 else if (UNLIKELY(UTF8_IS_NONCHAR(s, e)) && (ckWARN_d(WARN_NONCHAR))) {
4698 /* A side effect of this function will be to warn */
4699 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
4710 =for apidoc pv_uni_display
4712 Build to the scalar C<dsv> a displayable version of the string C<spv>,
4713 length C<len>, the displayable version being at most C<pvlim> bytes long
4714 (if longer, the rest is truncated and C<"..."> will be appended).
4716 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
4717 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
4718 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
4719 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
4720 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
4721 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
4723 The pointer to the PV of the C<dsv> is returned.
4725 See also L</sv_uni_display>.
4729 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim, UV flags)
4734 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
4738 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
4740 /* This serves double duty as a flag and a character to print after
4741 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
4745 if (pvlim && SvCUR(dsv) >= pvlim) {
4749 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
4751 const unsigned char c = (unsigned char)u & 0xFF;
4752 if (flags & UNI_DISPLAY_BACKSLASH) {
4769 const char string = ok;
4770 sv_catpvs(dsv, "\\");
4771 sv_catpvn(dsv, &string, 1);
4774 /* isPRINT() is the locale-blind version. */
4775 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
4776 const char string = c;
4777 sv_catpvn(dsv, &string, 1);
4782 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
4785 sv_catpvs(dsv, "...");
4791 =for apidoc sv_uni_display
4793 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
4794 the displayable version being at most C<pvlim> bytes long
4795 (if longer, the rest is truncated and "..." will be appended).
4797 The C<flags> argument is as in L</pv_uni_display>().
4799 The pointer to the PV of the C<dsv> is returned.
4804 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
4806 const char * const ptr =
4807 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
4809 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
4811 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
4812 SvCUR(ssv), pvlim, flags);
4816 =for apidoc foldEQ_utf8
4818 Returns true if the leading portions of the strings C<s1> and C<s2> (either or both
4819 of which may be in UTF-8) are the same case-insensitively; false otherwise.
4820 How far into the strings to compare is determined by other input parameters.
4822 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
4823 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for C<u2>
4824 with respect to C<s2>.
4826 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for fold
4827 equality. In other words, C<s1>+C<l1> will be used as a goal to reach. The
4828 scan will not be considered to be a match unless the goal is reached, and
4829 scanning won't continue past that goal. Correspondingly for C<l2> with respect to
4832 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that pointer is
4833 considered an end pointer to the position 1 byte past the maximum point
4834 in C<s1> beyond which scanning will not continue under any circumstances.
4835 (This routine assumes that UTF-8 encoded input strings are not malformed;
4836 malformed input can cause it to read past C<pe1>).
4837 This means that if both C<l1> and C<pe1> are specified, and C<pe1>
4838 is less than C<s1>+C<l1>, the match will never be successful because it can
4840 get as far as its goal (and in fact is asserted against). Correspondingly for
4841 C<pe2> with respect to C<s2>.
4843 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
4844 C<l2> must be non-zero), and if both do, both have to be
4845 reached for a successful match. Also, if the fold of a character is multiple
4846 characters, all of them must be matched (see tr21 reference below for
4849 Upon a successful match, if C<pe1> is non-C<NULL>,
4850 it will be set to point to the beginning of the I<next> character of C<s1>
4851 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
4853 For case-insensitiveness, the "casefolding" of Unicode is used
4854 instead of upper/lowercasing both the characters, see
4855 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
4859 /* A flags parameter has been added which may change, and hence isn't
4860 * externally documented. Currently it is:
4861 * 0 for as-documented above
4862 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
4863 ASCII one, to not match
4864 * FOLDEQ_LOCALE is set iff the rules from the current underlying
4865 * locale are to be used.
4866 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
4867 * routine. This allows that step to be skipped.
4868 * Currently, this requires s1 to be encoded as UTF-8
4869 * (u1 must be true), which is asserted for.
4870 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
4871 * cross certain boundaries. Hence, the caller should
4872 * let this function do the folding instead of
4873 * pre-folding. This code contains an assertion to
4874 * that effect. However, if the caller knows what
4875 * it's doing, it can pass this flag to indicate that,
4876 * and the assertion is skipped.
4877 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
4878 * FOLDEQ_S2_FOLDS_SANE
4881 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1, const char *s2, char **pe2, UV l2, bool u2, U32 flags)
4883 const U8 *p1 = (const U8*)s1; /* Point to current char */
4884 const U8 *p2 = (const U8*)s2;
4885 const U8 *g1 = NULL; /* goal for s1 */
4886 const U8 *g2 = NULL;
4887 const U8 *e1 = NULL; /* Don't scan s1 past this */
4888 U8 *f1 = NULL; /* Point to current folded */
4889 const U8 *e2 = NULL;
4891 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
4892 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
4893 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
4894 U8 flags_for_folder = FOLD_FLAGS_FULL;
4896 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
4898 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
4899 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
4900 && !(flags & FOLDEQ_S1_FOLDS_SANE))
4901 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
4902 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
4903 /* The algorithm is to trial the folds without regard to the flags on
4904 * the first line of the above assert(), and then see if the result
4905 * violates them. This means that the inputs can't be pre-folded to a
4906 * violating result, hence the assert. This could be changed, with the
4907 * addition of extra tests here for the already-folded case, which would
4908 * slow it down. That cost is more than any possible gain for when these
4909 * flags are specified, as the flags indicate /il or /iaa matching which
4910 * is less common than /iu, and I (khw) also believe that real-world /il
4911 * and /iaa matches are most likely to involve code points 0-255, and this
4912 * function only under rare conditions gets called for 0-255. */
4914 if (flags & FOLDEQ_LOCALE) {
4915 if (IN_UTF8_CTYPE_LOCALE) {
4916 flags &= ~FOLDEQ_LOCALE;
4919 flags_for_folder |= FOLD_FLAGS_LOCALE;
4928 g1 = (const U8*)s1 + l1;
4936 g2 = (const U8*)s2 + l2;
4939 /* Must have at least one goal */
4944 /* Will never match if goal is out-of-bounds */
4945 assert(! e1 || e1 >= g1);
4947 /* Here, there isn't an end pointer, or it is beyond the goal. We
4948 * only go as far as the goal */
4952 assert(e1); /* Must have an end for looking at s1 */
4955 /* Same for goal for s2 */
4957 assert(! e2 || e2 >= g2);
4964 /* If both operands are already folded, we could just do a memEQ on the
4965 * whole strings at once, but it would be better if the caller realized
4966 * this and didn't even call us */
4968 /* Look through both strings, a character at a time */
4969 while (p1 < e1 && p2 < e2) {
4971 /* If at the beginning of a new character in s1, get its fold to use
4972 * and the length of the fold. */
4974 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
4980 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
4982 /* We have to forbid mixing ASCII with non-ASCII if the
4983 * flags so indicate. And, we can short circuit having to
4984 * call the general functions for this common ASCII case,
4985 * all of whose non-locale folds are also ASCII, and hence
4986 * UTF-8 invariants, so the UTF8ness of the strings is not
4988 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
4992 *foldbuf1 = toFOLD(*p1);
4995 _to_utf8_fold_flags(p1, foldbuf1, &n1, flags_for_folder);
4997 else { /* Not UTF-8, get UTF-8 fold */
4998 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5004 if (n2 == 0) { /* Same for s2 */
5005 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5011 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5012 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5016 *foldbuf2 = toFOLD(*p2);
5019 _to_utf8_fold_flags(p2, foldbuf2, &n2, flags_for_folder);
5022 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5028 /* Here f1 and f2 point to the beginning of the strings to compare.
5029 * These strings are the folds of the next character from each input
5030 * string, stored in UTF-8. */
5032 /* While there is more to look for in both folds, see if they
5033 * continue to match */
5035 U8 fold_length = UTF8SKIP(f1);
5036 if (fold_length != UTF8SKIP(f2)
5037 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5038 function call for single
5040 || memNE((char*)f1, (char*)f2, fold_length))
5042 return 0; /* mismatch */
5045 /* Here, they matched, advance past them */
5052 /* When reach the end of any fold, advance the input past it */
5054 p1 += u1 ? UTF8SKIP(p1) : 1;
5057 p2 += u2 ? UTF8SKIP(p2) : 1;
5059 } /* End of loop through both strings */
5061 /* A match is defined by each scan that specified an explicit length
5062 * reaching its final goal, and the other not having matched a partial
5063 * character (which can happen when the fold of a character is more than one
5065 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5069 /* Successful match. Set output pointers */
5079 /* XXX The next two functions should likely be moved to mathoms.c once all
5080 * occurrences of them are removed from the core; some cpan-upstream modules
5084 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5086 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5088 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5092 =for apidoc utf8n_to_uvuni
5094 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5096 This function was useful for code that wanted to handle both EBCDIC and
5097 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5098 distinctions between the platforms have mostly been made invisible to most
5099 code, so this function is quite unlikely to be what you want. If you do need
5100 this precise functionality, use instead
5101 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5102 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5108 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5110 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5112 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5116 =for apidoc uvuni_to_utf8_flags
5118 Instead you almost certainly want to use L</uvchr_to_utf8> or
5119 L</uvchr_to_utf8_flags>.
5121 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5122 which itself, while not deprecated, should be used only in isolated
5123 circumstances. These functions were useful for code that wanted to handle
5124 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5125 v5.20, the distinctions between the platforms have mostly been made invisible
5126 to most code, so this function is quite unlikely to be what you want.
5132 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5134 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5136 return uvoffuni_to_utf8_flags(d, uv, flags);
5140 * ex: set ts=8 sts=4 sw=4 et: