3 * Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006, 2007, 2008
4 * by Larry Wall and others
6 * You may distribute under the terms of either the GNU General Public
7 * License or the Artistic License, as specified in the README file.
12 * 'What a fix!' said Sam. 'That's the one place in all the lands we've ever
13 * heard of that we don't want to see any closer; and that's the one place
14 * we're trying to get to! And that's just where we can't get, nohow.'
16 * [p.603 of _The Lord of the Rings_, IV/I: "The Taming of Sméagol"]
18 * 'Well do I understand your speech,' he answered in the same language;
19 * 'yet few strangers do so. Why then do you not speak in the Common Tongue,
20 * as is the custom in the West, if you wish to be answered?'
21 * --Gandalf, addressing Théoden's door wardens
23 * [p.508 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
25 * ...the travellers perceived that the floor was paved with stones of many
26 * hues; branching runes and strange devices intertwined beneath their feet.
28 * [p.512 of _The Lord of the Rings_, III/vi: "The King of the Golden Hall"]
32 #define PERL_IN_UTF8_C
34 #include "invlist_inline.h"
36 static const char malformed_text[] = "Malformed UTF-8 character";
37 static const char unees[] =
38 "Malformed UTF-8 character (unexpected end of string)";
39 static const char cp_above_legal_max[] =
40 "Use of code point 0x%" UVXf " is deprecated; the permissible max is 0x%" UVXf ". This will be fatal in Perl 5.28";
42 #define MAX_NON_DEPRECATED_CP ((UV) (IV_MAX))
45 =head1 Unicode Support
46 These are various utility functions for manipulating UTF8-encoded
47 strings. For the uninitiated, this is a method of representing arbitrary
48 Unicode characters as a variable number of bytes, in such a way that
49 characters in the ASCII range are unmodified, and a zero byte never appears
50 within non-zero characters.
56 Perl__force_out_malformed_utf8_message(pTHX_
57 const U8 *const p, /* First byte in UTF-8 sequence */
58 const U8 * const e, /* Final byte in sequence (may include
60 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
61 usually 0, or some DISALLOW flags */
62 const bool die_here) /* If TRUE, this function does not return */
64 /* This core-only function is to be called when a malformed UTF-8 character
65 * is found, in order to output the detailed information about the
66 * malformation before dieing. The reason it exists is for the occasions
67 * when such a malformation is fatal, but warnings might be turned off, so
68 * that normally they would not be actually output. This ensures that they
69 * do get output. Because a sequence may be malformed in more than one
70 * way, multiple messages may be generated, so we can't make them fatal, as
71 * that would cause the first one to die.
73 * Instead we pretend -W was passed to perl, then die afterwards. The
74 * flexibility is here to return to the caller so they can finish up and
78 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
84 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
86 PL_curcop->cop_warnings = pWARN_ALL;
89 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
94 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
95 " be called only when there are errors found");
99 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
104 =for apidoc uvoffuni_to_utf8_flags
106 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
107 Instead, B<Almost all code should use L</uvchr_to_utf8> or
108 L</uvchr_to_utf8_flags>>.
110 This function is like them, but the input is a strict Unicode
111 (as opposed to native) code point. Only in very rare circumstances should code
112 not be using the native code point.
114 For details, see the description for L</uvchr_to_utf8_flags>.
119 /* All these formats take a single UV code point argument */
120 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
121 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
122 " is not recommended for open interchange";
123 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
124 " may not be portable";
125 const char above_31_bit_cp_format[] = "Code point 0x%" UVXf " is not"
126 " Unicode, and not portable";
128 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
130 if (flags & UNICODE_WARN_SURROGATE) { \
131 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
132 surrogate_cp_format, uv); \
134 if (flags & UNICODE_DISALLOW_SURROGATE) { \
139 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
141 if (flags & UNICODE_WARN_NONCHAR) { \
142 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
143 nonchar_cp_format, uv); \
145 if (flags & UNICODE_DISALLOW_NONCHAR) { \
150 /* Use shorter names internally in this file */
151 #define SHIFT UTF_ACCUMULATION_SHIFT
153 #define MARK UTF_CONTINUATION_MARK
154 #define MASK UTF_CONTINUATION_MASK
157 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
159 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
161 if (OFFUNI_IS_INVARIANT(uv)) {
162 *d++ = LATIN1_TO_NATIVE(uv);
166 if (uv <= MAX_UTF8_TWO_BYTE) {
167 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
168 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
172 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
173 * below, the 16 is for start bytes E0-EF (which are all the possible ones
174 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
175 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
176 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
177 * 0x800-0xFFFF on ASCII */
178 if (uv < (16 * (1U << (2 * SHIFT)))) {
179 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
180 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
181 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
183 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
184 aren't tested here */
185 /* The most likely code points in this range are below the surrogates.
186 * Do an extra test to quickly exclude those. */
187 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
188 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
189 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
191 HANDLE_UNICODE_NONCHAR(uv, flags);
193 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
194 HANDLE_UNICODE_SURROGATE(uv, flags);
201 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
202 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
203 * happen starting with 4-byte characters on ASCII platforms. We unify the
204 * code for these with EBCDIC, even though some of them require 5-bytes on
205 * those, because khw believes the code saving is worth the very slight
206 * performance hit on these high EBCDIC code points. */
208 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
209 if ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
210 && ckWARN_d(WARN_DEPRECATED))
212 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
213 cp_above_legal_max, uv, MAX_NON_DEPRECATED_CP);
215 if ( (flags & UNICODE_WARN_SUPER)
216 || ( UNICODE_IS_ABOVE_31_BIT(uv)
217 && (flags & UNICODE_WARN_ABOVE_31_BIT)))
219 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
221 /* Choose the more dire applicable warning */
222 (UNICODE_IS_ABOVE_31_BIT(uv))
223 ? above_31_bit_cp_format
227 if ( (flags & UNICODE_DISALLOW_SUPER)
228 || ( UNICODE_IS_ABOVE_31_BIT(uv)
229 && (flags & UNICODE_DISALLOW_ABOVE_31_BIT)))
234 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
235 HANDLE_UNICODE_NONCHAR(uv, flags);
238 /* Test for and handle 4-byte result. In the test immediately below, the
239 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
240 * characters). The 3 is for 3 continuation bytes; these each contribute
241 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
242 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
243 * 0x1_0000-0x1F_FFFF on ASCII */
244 if (uv < (8 * (1U << (3 * SHIFT)))) {
245 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
246 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
247 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
248 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
250 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
251 characters. The end-plane non-characters for EBCDIC were
252 handled just above */
253 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
254 HANDLE_UNICODE_NONCHAR(uv, flags);
256 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
257 HANDLE_UNICODE_SURROGATE(uv, flags);
264 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
265 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
266 * format. The unrolled version above turns out to not save all that much
267 * time, and at these high code points (well above the legal Unicode range
268 * on ASCII platforms, and well above anything in common use in EBCDIC),
269 * khw believes that less code outweighs slight performance gains. */
272 STRLEN len = OFFUNISKIP(uv);
275 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
276 uv >>= UTF_ACCUMULATION_SHIFT;
278 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
284 =for apidoc uvchr_to_utf8
286 Adds the UTF-8 representation of the native code point C<uv> to the end
287 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
288 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
289 the byte after the end of the new character. In other words,
291 d = uvchr_to_utf8(d, uv);
293 is the recommended wide native character-aware way of saying
297 This function accepts any UV as input, but very high code points (above
298 C<IV_MAX> on the platform) will raise a deprecation warning. This is
299 typically 0x7FFF_FFFF in a 32-bit word.
301 It is possible to forbid or warn on non-Unicode code points, or those that may
302 be problematic by using L</uvchr_to_utf8_flags>.
307 /* This is also a macro */
308 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
311 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
313 return uvchr_to_utf8(d, uv);
317 =for apidoc uvchr_to_utf8_flags
319 Adds the UTF-8 representation of the native code point C<uv> to the end
320 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
321 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
322 the byte after the end of the new character. In other words,
324 d = uvchr_to_utf8_flags(d, uv, flags);
328 d = uvchr_to_utf8_flags(d, uv, 0);
330 This is the Unicode-aware way of saying
334 If C<flags> is 0, this function accepts any UV as input, but very high code
335 points (above C<IV_MAX> for the platform) will raise a deprecation warning.
336 This is typically 0x7FFF_FFFF in a 32-bit word.
338 Specifying C<flags> can further restrict what is allowed and not warned on, as
341 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
342 the function will raise a warning, provided UTF8 warnings are enabled. If
343 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
344 NULL. If both flags are set, the function will both warn and return NULL.
346 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
347 affect how the function handles a Unicode non-character.
349 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
350 affect the handling of code points that are above the Unicode maximum of
351 0x10FFFF. Languages other than Perl may not be able to accept files that
354 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
355 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
356 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
357 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
358 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
359 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
360 above-Unicode and surrogate flags, but not the non-character ones, as
362 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
363 See L<perlunicode/Noncharacter code points>.
365 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
366 so using them is more problematic than other above-Unicode code points. Perl
367 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
368 likely that non-Perl languages will not be able to read files that contain
369 these that written by the perl interpreter; nor would Perl understand files
370 written by something that uses a different extension. For these reasons, there
371 is a separate set of flags that can warn and/or disallow these extremely high
372 code points, even if other above-Unicode ones are accepted. These are the
373 C<UNICODE_WARN_ABOVE_31_BIT> and C<UNICODE_DISALLOW_ABOVE_31_BIT> flags. These
374 are entirely independent from the deprecation warning for code points above
375 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
376 code point that needs more than 31 bits to represent. When that happens,
377 effectively the C<UNICODE_DISALLOW_ABOVE_31_BIT> flag will always be set on
378 32-bit machines. (Of course C<UNICODE_DISALLOW_SUPER> will treat all
379 above-Unicode code points, including these, as malformations; and
380 C<UNICODE_WARN_SUPER> warns on these.)
382 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
383 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
384 than on ASCII. Prior to that, code points 2**31 and higher were simply
385 unrepresentable, and a different, incompatible method was used to represent
386 code points between 2**30 and 2**31 - 1. The flags C<UNICODE_WARN_ABOVE_31_BIT>
387 and C<UNICODE_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
388 platforms, warning and disallowing 2**31 and higher.
393 /* This is also a macro */
394 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
397 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
399 return uvchr_to_utf8_flags(d, uv, flags);
402 PERL_STATIC_INLINE bool
403 S_is_utf8_cp_above_31_bits(const U8 * const s, const U8 * const e)
405 /* Returns TRUE if the first code point represented by the Perl-extended-
406 * UTF-8-encoded string starting at 's', and looking no further than 'e -
407 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
409 * The function handles the case where the input bytes do not include all
410 * the ones necessary to represent a full character. That is, they may be
411 * the intial bytes of the representation of a code point, but possibly
412 * the final ones necessary for the complete representation may be beyond
415 * The function assumes that the sequence is well-formed UTF-8 as far as it
416 * goes, and is for a UTF-8 variant code point. If the sequence is
417 * incomplete, the function returns FALSE if there is any well-formed
418 * UTF-8 byte sequence that can complete it in such a way that a code point
419 * < 2**31 is produced; otherwise it returns TRUE.
421 * Getting this exactly right is slightly tricky, and has to be done in
422 * several places in this file, so is centralized here. It is based on the
425 * U+7FFFFFFF (2 ** 31 - 1)
426 * ASCII: \xFD\xBF\xBF\xBF\xBF\xBF
427 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
428 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
429 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
430 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
431 * U+80000000 (2 ** 31):
432 * ASCII: \xFE\x82\x80\x80\x80\x80\x80
433 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
434 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
435 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
436 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
437 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
442 /* [0] is start byte [1] [2] [3] [4] [5] [6] [7] */
443 const U8 prefix[] = "\x41\x41\x41\x41\x41\x41\x42";
444 const STRLEN prefix_len = sizeof(prefix) - 1;
445 const STRLEN len = e - s;
446 const STRLEN cmp_len = MIN(prefix_len, len - 1);
454 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
456 assert(! UTF8_IS_INVARIANT(*s));
460 /* Technically, a start byte of FE can be for a code point that fits into
461 * 31 bytes, but not for well-formed UTF-8: doing that requires an overlong
467 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
468 * mean a 32-bit or larger code point (0xFF is an invariant). For 0xFE, we
469 * need at least 2 bytes, and maybe up through 8 bytes, to be sure that the
470 * value is above 31 bits. */
471 if (*s != 0xFE || len == 1) {
475 /* Note that in UTF-EBCDIC, the two lowest possible continuation bytes are
477 return cBOOL(memGT(s + 1, prefix, cmp_len));
483 /* Anything larger than this will overflow the word if it were converted into a UV */
484 #if defined(UV_IS_QUAD)
485 # ifdef EBCDIC /* Actually is I8 */
486 # define HIGHEST_REPRESENTABLE_UTF8 \
487 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
489 # define HIGHEST_REPRESENTABLE_UTF8 \
490 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
494 # define HIGHEST_REPRESENTABLE_UTF8 \
495 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
497 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
501 PERL_STATIC_INLINE bool
502 S_does_utf8_overflow(const U8 * const s, const U8 * e)
505 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
507 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
509 const STRLEN len = e - s;
513 /* Returns a boolean as to if this UTF-8 string would overflow a UV on this
514 * platform, that is if it represents a code point larger than the highest
515 * representable code point. (For ASCII platforms, we could use memcmp()
516 * because we don't have to convert each byte to I8, but it's very rare
517 * input indeed that would approach overflow, so the loop below will likely
518 * only get executed once.
520 * 'e' must not be beyond a full character. If it is less than a full
521 * character, the function returns FALSE if there is any input beyond 'e'
522 * that could result in a non-overflowing code point */
524 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
525 assert(s <= e && s + UTF8SKIP(s) >= e);
527 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
529 /* On 32 bit ASCII machines, many overlongs that start with FF don't
532 if (isFF_OVERLONG(s, len)) {
533 const U8 max_32_bit_overlong[] = "\xFF\x80\x80\x80\x80\x80\x80\x84";
534 return memGE(s, max_32_bit_overlong,
535 MIN(len, sizeof(max_32_bit_overlong) - 1));
540 for (x = s; x < e; x++, y++) {
542 /* If this byte is larger than the corresponding highest UTF-8
543 * byte, it overflows */
544 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) > *y)) {
548 /* If not the same as this byte, it must be smaller, doesn't
550 if (LIKELY(NATIVE_UTF8_TO_I8(*x) != *y)) {
555 /* Got to the end and all bytes are the same. If the input is a whole
556 * character, it doesn't overflow. And if it is a partial character,
557 * there's not enough information to tell, so assume doesn't overflow */
561 PERL_STATIC_INLINE bool
562 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
564 /* Overlongs can occur whenever the number of continuation bytes
565 * changes. That means whenever the number of leading 1 bits in a start
566 * byte increases from the next lower start byte. That happens for start
567 * bytes C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following
568 * illegal start bytes have already been excluded, so don't need to be
570 * ASCII platforms: C0, C1
571 * EBCDIC platforms C0, C1, C2, C3, C4, E0
573 * At least a second byte is required to determine if other sequences will
576 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
577 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
579 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
580 assert(len > 1 && UTF8_IS_START(*s));
582 /* Each platform has overlongs after the start bytes given above (expressed
583 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
584 * the logic is the same, except the E0 overlong has already been excluded
585 * on EBCDIC platforms. The values below were found by manually
586 * inspecting the UTF-8 patterns. See the tables in utf8.h and
590 # define F0_ABOVE_OVERLONG 0xB0
591 # define F8_ABOVE_OVERLONG 0xA8
592 # define FC_ABOVE_OVERLONG 0xA4
593 # define FE_ABOVE_OVERLONG 0xA2
594 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
598 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
602 # define F0_ABOVE_OVERLONG 0x90
603 # define F8_ABOVE_OVERLONG 0x88
604 # define FC_ABOVE_OVERLONG 0x84
605 # define FE_ABOVE_OVERLONG 0x82
606 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
610 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
611 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
612 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
613 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
618 /* Check for the FF overlong */
619 return isFF_OVERLONG(s, len);
622 PERL_STATIC_INLINE bool
623 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
625 PERL_ARGS_ASSERT_ISFF_OVERLONG;
627 /* Check for the FF overlong. This happens only if all these bytes match;
628 * what comes after them doesn't matter. See tables in utf8.h,
631 return len >= sizeof(FF_OVERLONG_PREFIX) - 1
632 && UNLIKELY(memEQ(s, FF_OVERLONG_PREFIX,
633 sizeof(FF_OVERLONG_PREFIX) - 1));
636 #undef F0_ABOVE_OVERLONG
637 #undef F8_ABOVE_OVERLONG
638 #undef FC_ABOVE_OVERLONG
639 #undef FE_ABOVE_OVERLONG
640 #undef FF_OVERLONG_PREFIX
643 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
648 /* A helper function that should not be called directly.
650 * This function returns non-zero if the string beginning at 's' and
651 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
652 * code point; otherwise it returns 0. The examination stops after the
653 * first code point in 's' is validated, not looking at the rest of the
654 * input. If 'e' is such that there are not enough bytes to represent a
655 * complete code point, this function will return non-zero anyway, if the
656 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
657 * excluded by 'flags'.
659 * A non-zero return gives the number of bytes required to represent the
660 * code point. Be aware that if the input is for a partial character, the
661 * return will be larger than 'e - s'.
663 * This function assumes that the code point represented is UTF-8 variant.
664 * The caller should have excluded the possibility of it being invariant
665 * before calling this function.
667 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
668 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
669 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
670 * disallowed by the flags. If the input is only for a partial character,
671 * the function will return non-zero if there is any sequence of
672 * well-formed UTF-8 that, when appended to the input sequence, could
673 * result in an allowed code point; otherwise it returns 0. Non characters
674 * cannot be determined based on partial character input. But many of the
675 * other excluded types can be determined with just the first one or two
680 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
682 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
683 |UTF8_DISALLOW_ABOVE_31_BIT)));
684 assert(! UTF8_IS_INVARIANT(*s));
686 /* A variant char must begin with a start byte */
687 if (UNLIKELY(! UTF8_IS_START(*s))) {
691 /* Examine a maximum of a single whole code point */
692 if (e - s > UTF8SKIP(s)) {
698 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
699 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
701 /* Here, we are disallowing some set of largish code points, and the
702 * first byte indicates the sequence is for a code point that could be
703 * in the excluded set. We generally don't have to look beyond this or
704 * the second byte to see if the sequence is actually for one of the
705 * excluded classes. The code below is derived from this table:
707 * UTF-8 UTF-EBCDIC I8
708 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
709 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
710 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
712 * Keep in mind that legal continuation bytes range between \x80..\xBF
713 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
714 * continuation bytes. Hence, we don't have to test the upper edge
715 * because if any of those is encountered, the sequence is malformed,
716 * and would fail elsewhere in this function.
718 * The code here likewise assumes that there aren't other
719 * malformations; again the function should fail elsewhere because of
720 * these. For example, an overlong beginning with FC doesn't actually
721 * have to be a super; it could actually represent a small code point,
722 * even U+0000. But, since overlongs (and other malformations) are
723 * illegal, the function should return FALSE in either case.
726 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
727 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
728 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
730 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
732 && ((s1) & 0xFE ) == 0xB6)
734 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
735 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
736 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
739 if ( (flags & UTF8_DISALLOW_SUPER)
740 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
742 return 0; /* Above Unicode */
745 if ( (flags & UTF8_DISALLOW_ABOVE_31_BIT)
746 && UNLIKELY(is_utf8_cp_above_31_bits(s, e)))
748 return 0; /* Above 31 bits */
752 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
754 if ( (flags & UTF8_DISALLOW_SUPER)
755 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
757 return 0; /* Above Unicode */
760 if ( (flags & UTF8_DISALLOW_SURROGATE)
761 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
763 return 0; /* Surrogate */
766 if ( (flags & UTF8_DISALLOW_NONCHAR)
767 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
769 return 0; /* Noncharacter code point */
774 /* Make sure that all that follows are continuation bytes */
775 for (x = s + 1; x < e; x++) {
776 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
781 /* Here is syntactically valid. Next, make sure this isn't the start of an
783 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len)) {
787 /* And finally, that the code point represented fits in a word on this
789 if (does_utf8_overflow(s, e)) {
797 Perl__byte_dump_string(pTHX_ const U8 * s, const STRLEN len, const bool format)
799 /* Returns a mortalized C string that is a displayable copy of the 'len'
800 * bytes starting at 's'. 'format' gives how to display each byte.
801 * Currently, there are only two formats, so it is currently a bool:
803 * 1 ab (that is a space between two hex digit bytes)
806 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
808 const U8 * const e = s + len;
812 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
814 Newx(output, output_len, char);
819 const unsigned high_nibble = (*s & 0xF0) >> 4;
820 const unsigned low_nibble = (*s & 0x0F);
830 if (high_nibble < 10) {
831 *d++ = high_nibble + '0';
834 *d++ = high_nibble - 10 + 'a';
837 if (low_nibble < 10) {
838 *d++ = low_nibble + '0';
841 *d++ = low_nibble - 10 + 'a';
849 PERL_STATIC_INLINE char *
850 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
852 /* How many bytes to print */
855 /* Which one is the non-continuation */
856 const STRLEN non_cont_byte_pos,
858 /* How many bytes should there be? */
859 const STRLEN expect_len)
861 /* Return the malformation warning text for an unexpected continuation
864 const char * const where = (non_cont_byte_pos == 1)
866 : Perl_form(aTHX_ "%d bytes",
867 (int) non_cont_byte_pos);
869 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
871 /* We don't need to pass this parameter, but since it has already been
872 * calculated, it's likely faster to pass it; verify under DEBUGGING */
873 assert(expect_len == UTF8SKIP(s));
875 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
876 " %s after start byte 0x%02x; need %d bytes, got %d)",
878 _byte_dump_string(s, print_len, 0),
879 *(s + non_cont_byte_pos),
883 (int) non_cont_byte_pos);
888 =for apidoc utf8n_to_uvchr
890 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
891 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
893 Bottom level UTF-8 decode routine.
894 Returns the native code point value of the first character in the string C<s>,
895 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
896 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
897 the length, in bytes, of that character.
899 The value of C<flags> determines the behavior when C<s> does not point to a
900 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
901 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
902 is the next possible position in C<s> that could begin a non-malformed
903 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
904 is raised. Some UTF-8 input sequences may contain multiple malformations.
905 This function tries to find every possible one in each call, so multiple
906 warnings can be raised for the same sequence.
908 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
909 individual types of malformations, such as the sequence being overlong (that
910 is, when there is a shorter sequence that can express the same code point;
911 overlong sequences are expressly forbidden in the UTF-8 standard due to
912 potential security issues). Another malformation example is the first byte of
913 a character not being a legal first byte. See F<utf8.h> for the list of such
914 flags. Even if allowed, this function generally returns the Unicode
915 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
916 F<utf8.h> to override this behavior for the overlong malformations, but don't
917 do that except for very specialized purposes.
919 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
920 flags) malformation is found. If this flag is set, the routine assumes that
921 the caller will raise a warning, and this function will silently just set
922 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
924 Note that this API requires disambiguation between successful decoding a C<NUL>
925 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
926 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
927 be set to 1. To disambiguate, upon a zero return, see if the first byte of
928 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
929 error. Or you can use C<L</utf8n_to_uvchr_error>>.
931 Certain code points are considered problematic. These are Unicode surrogates,
932 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
933 By default these are considered regular code points, but certain situations
934 warrant special handling for them, which can be specified using the C<flags>
935 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
936 three classes are treated as malformations and handled as such. The flags
937 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
938 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
939 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
940 restricts the allowed inputs to the strict UTF-8 traditionally defined by
941 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
943 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
944 The difference between traditional strictness and C9 strictness is that the
945 latter does not forbid non-character code points. (They are still discouraged,
946 however.) For more discussion see L<perlunicode/Noncharacter code points>.
948 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
949 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
950 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
951 raised for their respective categories, but otherwise the code points are
952 considered valid (not malformations). To get a category to both be treated as
953 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
954 (But note that warnings are not raised if lexically disabled nor if
955 C<UTF8_CHECK_ONLY> is also specified.)
957 It is now deprecated to have very high code points (above C<IV_MAX> on the
958 platforms) and this function will raise a deprecation warning for these (unless
959 such warnings are turned off). This value is typically 0x7FFF_FFFF (2**31 -1)
962 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
963 so using them is more problematic than other above-Unicode code points. Perl
964 invented an extension to UTF-8 to represent the ones above 2**36-1, so it is
965 likely that non-Perl languages will not be able to read files that contain
966 these; nor would Perl understand files
967 written by something that uses a different extension. For these reasons, there
968 is a separate set of flags that can warn and/or disallow these extremely high
969 code points, even if other above-Unicode ones are accepted. These are the
970 C<UTF8_WARN_ABOVE_31_BIT> and C<UTF8_DISALLOW_ABOVE_31_BIT> flags. These
971 are entirely independent from the deprecation warning for code points above
972 C<IV_MAX>. On 32-bit machines, it will eventually be forbidden to have any
973 code point that needs more than 31 bits to represent. When that happens,
974 effectively the C<UTF8_DISALLOW_ABOVE_31_BIT> flag will always be set on
975 32-bit machines. (Of course C<UTF8_DISALLOW_SUPER> will treat all
976 above-Unicode code points, including these, as malformations; and
977 C<UTF8_WARN_SUPER> warns on these.)
979 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
980 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
981 than on ASCII. Prior to that, code points 2**31 and higher were simply
982 unrepresentable, and a different, incompatible method was used to represent
983 code points between 2**30 and 2**31 - 1. The flags C<UTF8_WARN_ABOVE_31_BIT>
984 and C<UTF8_DISALLOW_ABOVE_31_BIT> have the same function as on ASCII
985 platforms, warning and disallowing 2**31 and higher.
987 All other code points corresponding to Unicode characters, including private
988 use and those yet to be assigned, are never considered malformed and never
993 Also implemented as a macro in utf8.h
997 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1002 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1004 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1009 =for apidoc utf8n_to_uvchr_error
1011 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1012 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1014 This function is for code that needs to know what the precise malformation(s)
1015 are when an error is found.
1017 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1018 all the others, C<errors>. If this parameter is 0, this function behaves
1019 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1020 to a C<U32> variable, which this function sets to indicate any errors found.
1021 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1022 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1023 of these bits will be set if a malformation is found, even if the input
1024 C<flags> parameter indicates that the given malformation is allowed; those
1025 exceptions are noted:
1029 =item C<UTF8_GOT_ABOVE_31_BIT>
1031 The code point represented by the input UTF-8 sequence occupies more than 31
1033 This bit is set only if the input C<flags> parameter contains either the
1034 C<UTF8_DISALLOW_ABOVE_31_BIT> or the C<UTF8_WARN_ABOVE_31_BIT> flags.
1036 =item C<UTF8_GOT_CONTINUATION>
1038 The input sequence was malformed in that the first byte was a a UTF-8
1041 =item C<UTF8_GOT_EMPTY>
1043 The input C<curlen> parameter was 0.
1045 =item C<UTF8_GOT_LONG>
1047 The input sequence was malformed in that there is some other sequence that
1048 evaluates to the same code point, but that sequence is shorter than this one.
1050 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1051 it was discovered that this created security issues.
1053 =item C<UTF8_GOT_NONCHAR>
1055 The code point represented by the input UTF-8 sequence is for a Unicode
1056 non-character code point.
1057 This bit is set only if the input C<flags> parameter contains either the
1058 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1060 =item C<UTF8_GOT_NON_CONTINUATION>
1062 The input sequence was malformed in that a non-continuation type byte was found
1063 in a position where only a continuation type one should be.
1065 =item C<UTF8_GOT_OVERFLOW>
1067 The input sequence was malformed in that it is for a code point that is not
1068 representable in the number of bits available in a UV on the current platform.
1070 =item C<UTF8_GOT_SHORT>
1072 The input sequence was malformed in that C<curlen> is smaller than required for
1073 a complete sequence. In other words, the input is for a partial character
1076 =item C<UTF8_GOT_SUPER>
1078 The input sequence was malformed in that it is for a non-Unicode code point;
1079 that is, one above the legal Unicode maximum.
1080 This bit is set only if the input C<flags> parameter contains either the
1081 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1083 =item C<UTF8_GOT_SURROGATE>
1085 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1087 This bit is set only if the input C<flags> parameter contains either the
1088 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1092 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1093 flag to suppress any warnings, and then examine the C<*errors> return.
1099 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1105 const U8 * const s0 = s;
1106 U8 * send = NULL; /* (initialized to silence compilers' wrong
1108 U32 possible_problems = 0; /* A bit is set here for each potential problem
1109 found as we go along */
1111 STRLEN expectlen = 0; /* How long should this sequence be?
1112 (initialized to silence compilers' wrong
1114 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1115 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1116 this gets set and discarded */
1118 /* The below are used only if there is both an overlong malformation and a
1119 * too short one. Otherwise the first two are set to 's0' and 'send', and
1120 * the third not used at all */
1121 U8 * adjusted_s0 = (U8 *) s0;
1122 U8 * adjusted_send = NULL; /* (Initialized to silence compilers' wrong
1124 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1125 routine; see [perl #130921] */
1126 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1128 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1134 errors = &discard_errors;
1137 /* The order of malformation tests here is important. We should consume as
1138 * few bytes as possible in order to not skip any valid character. This is
1139 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1140 * http://unicode.org/reports/tr36 for more discussion as to why. For
1141 * example, once we've done a UTF8SKIP, we can tell the expected number of
1142 * bytes, and could fail right off the bat if the input parameters indicate
1143 * that there are too few available. But it could be that just that first
1144 * byte is garbled, and the intended character occupies fewer bytes. If we
1145 * blindly assumed that the first byte is correct, and skipped based on
1146 * that number, we could skip over a valid input character. So instead, we
1147 * always examine the sequence byte-by-byte.
1149 * We also should not consume too few bytes, otherwise someone could inject
1150 * things. For example, an input could be deliberately designed to
1151 * overflow, and if this code bailed out immediately upon discovering that,
1152 * returning to the caller C<*retlen> pointing to the very next byte (one
1153 * which is actually part of of the overflowing sequence), that could look
1154 * legitimate to the caller, which could discard the initial partial
1155 * sequence and process the rest, inappropriately.
1157 * Some possible input sequences are malformed in more than one way. This
1158 * function goes to lengths to try to find all of them. This is necessary
1159 * for correctness, as the inputs may allow one malformation but not
1160 * another, and if we abandon searching for others after finding the
1161 * allowed one, we could allow in something that shouldn't have been.
1164 if (UNLIKELY(curlen == 0)) {
1165 possible_problems |= UTF8_GOT_EMPTY;
1167 uv = UNICODE_REPLACEMENT;
1168 goto ready_to_handle_errors;
1171 expectlen = UTF8SKIP(s);
1173 /* A well-formed UTF-8 character, as the vast majority of calls to this
1174 * function will be for, has this expected length. For efficiency, set
1175 * things up here to return it. It will be overriden only in those rare
1176 * cases where a malformation is found */
1178 *retlen = expectlen;
1181 /* An invariant is trivially well-formed */
1182 if (UTF8_IS_INVARIANT(uv)) {
1186 /* A continuation character can't start a valid sequence */
1187 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1188 possible_problems |= UTF8_GOT_CONTINUATION;
1190 uv = UNICODE_REPLACEMENT;
1191 goto ready_to_handle_errors;
1194 /* Here is not a continuation byte, nor an invariant. The only thing left
1195 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1196 * because it excludes start bytes like \xC0 that always lead to
1199 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1200 * that indicate the number of bytes in the character's whole UTF-8
1201 * sequence, leaving just the bits that are part of the value. */
1202 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1204 /* Setup the loop end point, making sure to not look past the end of the
1205 * input string, and flag it as too short if the size isn't big enough. */
1207 if (UNLIKELY(curlen < expectlen)) {
1208 possible_problems |= UTF8_GOT_SHORT;
1215 adjusted_send = send;
1217 /* Now, loop through the remaining bytes in the character's sequence,
1218 * accumulating each into the working value as we go. */
1219 for (s = s0 + 1; s < send; s++) {
1220 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1221 uv = UTF8_ACCUMULATE(uv, *s);
1225 /* Here, found a non-continuation before processing all expected bytes.
1226 * This byte indicates the beginning of a new character, so quit, even
1227 * if allowing this malformation. */
1228 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1230 } /* End of loop through the character's bytes */
1232 /* Save how many bytes were actually in the character */
1235 /* Note that there are two types of too-short malformation. One is when
1236 * there is actual wrong data before the normal termination of the
1237 * sequence. The other is that the sequence wasn't complete before the end
1238 * of the data we are allowed to look at, based on the input 'curlen'.
1239 * This means that we were passed data for a partial character, but it is
1240 * valid as far as we saw. The other is definitely invalid. This
1241 * distinction could be important to a caller, so the two types are kept
1244 * A convenience macro that matches either of the too-short conditions. */
1245 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1247 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1249 uv = UNICODE_REPLACEMENT;
1252 /* Check for overflow. */
1253 if (UNLIKELY(does_utf8_overflow(s0, send))) {
1254 possible_problems |= UTF8_GOT_OVERFLOW;
1255 uv = UNICODE_REPLACEMENT;
1258 /* Check for overlong. If no problems so far, 'uv' is the correct code
1259 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1260 * we must look at the UTF-8 byte sequence itself to see if it is for an
1262 if ( ( LIKELY(! possible_problems)
1263 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1264 || ( UNLIKELY(possible_problems)
1265 && ( UNLIKELY(! UTF8_IS_START(*s0))
1267 && UNLIKELY(is_utf8_overlong_given_start_byte_ok(s0,
1270 possible_problems |= UTF8_GOT_LONG;
1272 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1274 /* The calculation in the 'true' branch of this 'if'
1275 * below won't work if overflows, and isn't needed
1276 * anyway. Further below we handle all overflow
1278 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1280 UV min_uv = uv_so_far;
1283 /* Here, the input is both overlong and is missing some trailing
1284 * bytes. There is no single code point it could be for, but there
1285 * may be enough information present to determine if what we have
1286 * so far is for an unallowed code point, such as for a surrogate.
1287 * The code further below has the intelligence to determine this,
1288 * but just for non-overlong UTF-8 sequences. What we do here is
1289 * calculate the smallest code point the input could represent if
1290 * there were no too short malformation. Then we compute and save
1291 * the UTF-8 for that, which is what the code below looks at
1292 * instead of the raw input. It turns out that the smallest such
1293 * code point is all we need. */
1294 for (i = curlen; i < expectlen; i++) {
1295 min_uv = UTF8_ACCUMULATE(min_uv,
1296 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1299 adjusted_s0 = temp_char_buf;
1300 adjusted_send = uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1304 /* Here, we have found all the possible problems, except for when the input
1305 * is for a problematic code point not allowed by the input parameters. */
1307 /* uv is valid for overlongs */
1308 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1310 /* isn't problematic if < this */
1311 && uv >= UNICODE_SURROGATE_FIRST)
1312 || ( UNLIKELY(possible_problems)
1314 /* if overflow, we know without looking further
1315 * precisely which of the problematic types it is,
1316 * and we deal with those in the overflow handling
1318 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1319 && isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)))
1320 && ((flags & ( UTF8_DISALLOW_NONCHAR
1321 |UTF8_DISALLOW_SURROGATE
1322 |UTF8_DISALLOW_SUPER
1323 |UTF8_DISALLOW_ABOVE_31_BIT
1325 |UTF8_WARN_SURROGATE
1327 |UTF8_WARN_ABOVE_31_BIT))
1328 /* In case of a malformation, 'uv' is not valid, and has
1329 * been changed to something in the Unicode range.
1330 * Currently we don't output a deprecation message if there
1331 * is already a malformation, so we don't have to special
1332 * case the test immediately below */
1333 || ( UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1334 && ckWARN_d(WARN_DEPRECATED))))
1336 /* If there were no malformations, or the only malformation is an
1337 * overlong, 'uv' is valid */
1338 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1339 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1340 possible_problems |= UTF8_GOT_SURROGATE;
1342 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1343 possible_problems |= UTF8_GOT_SUPER;
1345 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1346 possible_problems |= UTF8_GOT_NONCHAR;
1349 else { /* Otherwise, need to look at the source UTF-8, possibly
1350 adjusted to be non-overlong */
1352 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1353 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1355 possible_problems |= UTF8_GOT_SUPER;
1357 else if (curlen > 1) {
1358 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1359 NATIVE_UTF8_TO_I8(*adjusted_s0),
1360 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1362 possible_problems |= UTF8_GOT_SUPER;
1364 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1365 NATIVE_UTF8_TO_I8(*adjusted_s0),
1366 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1368 possible_problems |= UTF8_GOT_SURROGATE;
1372 /* We need a complete well-formed UTF-8 character to discern
1373 * non-characters, so can't look for them here */
1377 ready_to_handle_errors:
1380 * curlen contains the number of bytes in the sequence that
1381 * this call should advance the input by.
1382 * avail_len gives the available number of bytes passed in, but
1383 * only if this is less than the expected number of
1384 * bytes, based on the code point's start byte.
1385 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1386 * is set in it for each potential problem found.
1387 * uv contains the code point the input sequence
1388 * represents; or if there is a problem that prevents
1389 * a well-defined value from being computed, it is
1390 * some subsitute value, typically the REPLACEMENT
1392 * s0 points to the first byte of the character
1393 * s points to just after were we left off processing
1395 * send points to just after where that character should
1396 * end, based on how many bytes the start byte tells
1397 * us should be in it, but no further than s0 +
1399 * adjusted_s0 normally is the same as s0, but in case of an
1400 * overlong for which the UTF-8 matters below, it is
1401 * the first byte of the shortest form representation
1403 * adjusted_send normally is the same as 'send', but if adjusted_s0
1404 * is set to something other than s0, this points one
1408 if (UNLIKELY(possible_problems)) {
1409 bool disallowed = FALSE;
1410 const U32 orig_problems = possible_problems;
1412 while (possible_problems) { /* Handle each possible problem */
1414 char * message = NULL;
1416 /* Each 'if' clause handles one problem. They are ordered so that
1417 * the first ones' messages will be displayed before the later
1418 * ones; this is kinda in decreasing severity order */
1419 if (possible_problems & UTF8_GOT_OVERFLOW) {
1421 /* Overflow means also got a super and are using Perl's
1422 * extended UTF-8, but we handle all three cases here */
1424 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_ABOVE_31_BIT);
1425 *errors |= UTF8_GOT_OVERFLOW;
1427 /* But the API says we flag all errors found */
1428 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1429 *errors |= UTF8_GOT_SUPER;
1432 & (UTF8_WARN_ABOVE_31_BIT|UTF8_DISALLOW_ABOVE_31_BIT))
1434 *errors |= UTF8_GOT_ABOVE_31_BIT;
1437 /* Disallow if any of the three categories say to */
1438 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1439 || (flags & ( UTF8_DISALLOW_SUPER
1440 |UTF8_DISALLOW_ABOVE_31_BIT)))
1445 /* Likewise, warn if any say to, plus if deprecation warnings
1446 * are on, because this code point is above IV_MAX */
1447 if ( ckWARN_d(WARN_DEPRECATED)
1448 || ! (flags & UTF8_ALLOW_OVERFLOW)
1449 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_ABOVE_31_BIT)))
1452 /* The warnings code explicitly says it doesn't handle the
1453 * case of packWARN2 and two categories which have
1454 * parent-child relationship. Even if it works now to
1455 * raise the warning if either is enabled, it wouldn't
1456 * necessarily do so in the future. We output (only) the
1457 * most dire warning */
1458 if (! (flags & UTF8_CHECK_ONLY)) {
1459 if (ckWARN_d(WARN_UTF8)) {
1460 pack_warn = packWARN(WARN_UTF8);
1462 else if (ckWARN_d(WARN_NON_UNICODE)) {
1463 pack_warn = packWARN(WARN_NON_UNICODE);
1466 message = Perl_form(aTHX_ "%s: %s (overflows)",
1468 _byte_dump_string(s0, curlen, 0));
1473 else if (possible_problems & UTF8_GOT_EMPTY) {
1474 possible_problems &= ~UTF8_GOT_EMPTY;
1475 *errors |= UTF8_GOT_EMPTY;
1477 if (! (flags & UTF8_ALLOW_EMPTY)) {
1479 /* This so-called malformation is now treated as a bug in
1480 * the caller. If you have nothing to decode, skip calling
1485 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1486 pack_warn = packWARN(WARN_UTF8);
1487 message = Perl_form(aTHX_ "%s (empty string)",
1492 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1493 possible_problems &= ~UTF8_GOT_CONTINUATION;
1494 *errors |= UTF8_GOT_CONTINUATION;
1496 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1498 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1499 pack_warn = packWARN(WARN_UTF8);
1500 message = Perl_form(aTHX_
1501 "%s: %s (unexpected continuation byte 0x%02x,"
1502 " with no preceding start byte)",
1504 _byte_dump_string(s0, 1, 0), *s0);
1508 else if (possible_problems & UTF8_GOT_SHORT) {
1509 possible_problems &= ~UTF8_GOT_SHORT;
1510 *errors |= UTF8_GOT_SHORT;
1512 if (! (flags & UTF8_ALLOW_SHORT)) {
1514 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1515 pack_warn = packWARN(WARN_UTF8);
1516 message = Perl_form(aTHX_
1517 "%s: %s (too short; %d byte%s available, need %d)",
1519 _byte_dump_string(s0, send - s0, 0),
1521 avail_len == 1 ? "" : "s",
1527 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1528 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1529 *errors |= UTF8_GOT_NON_CONTINUATION;
1531 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1533 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1535 /* If we don't know for sure that the input length is
1536 * valid, avoid as much as possible reading past the
1537 * end of the buffer */
1538 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1541 pack_warn = packWARN(WARN_UTF8);
1542 message = Perl_form(aTHX_ "%s",
1543 unexpected_non_continuation_text(s0,
1550 else if (possible_problems & UTF8_GOT_LONG) {
1551 possible_problems &= ~UTF8_GOT_LONG;
1552 *errors |= UTF8_GOT_LONG;
1554 if (flags & UTF8_ALLOW_LONG) {
1556 /* We don't allow the actual overlong value, unless the
1557 * special extra bit is also set */
1558 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1559 & ~UTF8_ALLOW_LONG)))
1561 uv = UNICODE_REPLACEMENT;
1567 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1568 pack_warn = packWARN(WARN_UTF8);
1570 /* These error types cause 'uv' to be something that
1571 * isn't what was intended, so can't use it in the
1572 * message. The other error types either can't
1573 * generate an overlong, or else the 'uv' is valid */
1575 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1577 message = Perl_form(aTHX_
1578 "%s: %s (any UTF-8 sequence that starts"
1579 " with \"%s\" is overlong which can and"
1580 " should be represented with a"
1581 " different, shorter sequence)",
1583 _byte_dump_string(s0, send - s0, 0),
1584 _byte_dump_string(s0, curlen, 0));
1587 U8 tmpbuf[UTF8_MAXBYTES+1];
1588 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1590 message = Perl_form(aTHX_
1591 "%s: %s (overlong; instead use %s to represent"
1594 _byte_dump_string(s0, curlen, 0),
1595 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1596 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1597 small code points */
1603 else if (possible_problems & UTF8_GOT_SURROGATE) {
1604 possible_problems &= ~UTF8_GOT_SURROGATE;
1606 if (flags & UTF8_WARN_SURROGATE) {
1607 *errors |= UTF8_GOT_SURROGATE;
1609 if ( ! (flags & UTF8_CHECK_ONLY)
1610 && ckWARN_d(WARN_SURROGATE))
1612 pack_warn = packWARN(WARN_SURROGATE);
1614 /* These are the only errors that can occur with a
1615 * surrogate when the 'uv' isn't valid */
1616 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1617 message = Perl_form(aTHX_
1618 "UTF-16 surrogate (any UTF-8 sequence that"
1619 " starts with \"%s\" is for a surrogate)",
1620 _byte_dump_string(s0, curlen, 0));
1623 message = Perl_form(aTHX_ surrogate_cp_format, uv);
1628 if (flags & UTF8_DISALLOW_SURROGATE) {
1630 *errors |= UTF8_GOT_SURROGATE;
1633 else if (possible_problems & UTF8_GOT_SUPER) {
1634 possible_problems &= ~UTF8_GOT_SUPER;
1636 if (flags & UTF8_WARN_SUPER) {
1637 *errors |= UTF8_GOT_SUPER;
1639 if ( ! (flags & UTF8_CHECK_ONLY)
1640 && ckWARN_d(WARN_NON_UNICODE))
1642 pack_warn = packWARN(WARN_NON_UNICODE);
1644 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1645 message = Perl_form(aTHX_
1646 "Any UTF-8 sequence that starts with"
1647 " \"%s\" is for a non-Unicode code point,"
1648 " may not be portable",
1649 _byte_dump_string(s0, curlen, 0));
1652 message = Perl_form(aTHX_ super_cp_format, uv);
1657 /* The maximum code point ever specified by a standard was
1658 * 2**31 - 1. Anything larger than that is a Perl extension
1659 * that very well may not be understood by other applications
1660 * (including earlier perl versions on EBCDIC platforms). We
1661 * test for these after the regular SUPER ones, and before
1662 * possibly bailing out, so that the slightly more dire warning
1663 * will override the regular one. */
1664 if ( (flags & (UTF8_WARN_ABOVE_31_BIT
1666 |UTF8_DISALLOW_ABOVE_31_BIT))
1667 && ( ( UNLIKELY(orig_problems & UTF8_GOT_TOO_SHORT)
1668 && UNLIKELY(is_utf8_cp_above_31_bits(
1671 || ( LIKELY(! (orig_problems & UTF8_GOT_TOO_SHORT))
1672 && UNLIKELY(UNICODE_IS_ABOVE_31_BIT(uv)))))
1674 if ( ! (flags & UTF8_CHECK_ONLY)
1675 && (flags & (UTF8_WARN_ABOVE_31_BIT|UTF8_WARN_SUPER))
1676 && ckWARN_d(WARN_UTF8))
1678 pack_warn = packWARN(WARN_UTF8);
1680 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1681 message = Perl_form(aTHX_
1682 "Any UTF-8 sequence that starts with"
1683 " \"%s\" is for a non-Unicode code"
1684 " point, and is not portable",
1685 _byte_dump_string(s0, curlen, 0));
1688 message = Perl_form(aTHX_
1689 above_31_bit_cp_format, uv);
1693 if (flags & ( UTF8_WARN_ABOVE_31_BIT
1694 |UTF8_DISALLOW_ABOVE_31_BIT))
1696 *errors |= UTF8_GOT_ABOVE_31_BIT;
1698 if (flags & UTF8_DISALLOW_ABOVE_31_BIT) {
1704 if (flags & UTF8_DISALLOW_SUPER) {
1705 *errors |= UTF8_GOT_SUPER;
1709 /* The deprecated warning overrides any non-deprecated one. If
1710 * there are other problems, a deprecation message is not
1711 * really helpful, so don't bother to raise it in that case.
1712 * This also keeps the code from having to handle the case
1713 * where 'uv' is not valid. */
1714 if ( ! (orig_problems
1715 & (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1716 && UNLIKELY(uv > MAX_NON_DEPRECATED_CP)
1717 && ckWARN_d(WARN_DEPRECATED))
1719 message = Perl_form(aTHX_ cp_above_legal_max,
1720 uv, MAX_NON_DEPRECATED_CP);
1721 pack_warn = packWARN(WARN_DEPRECATED);
1724 else if (possible_problems & UTF8_GOT_NONCHAR) {
1725 possible_problems &= ~UTF8_GOT_NONCHAR;
1727 if (flags & UTF8_WARN_NONCHAR) {
1728 *errors |= UTF8_GOT_NONCHAR;
1730 if ( ! (flags & UTF8_CHECK_ONLY)
1731 && ckWARN_d(WARN_NONCHAR))
1733 /* The code above should have guaranteed that we don't
1734 * get here with errors other than overlong */
1735 assert (! (orig_problems
1736 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1738 pack_warn = packWARN(WARN_NONCHAR);
1739 message = Perl_form(aTHX_ nonchar_cp_format, uv);
1743 if (flags & UTF8_DISALLOW_NONCHAR) {
1745 *errors |= UTF8_GOT_NONCHAR;
1747 } /* End of looking through the possible flags */
1749 /* Display the message (if any) for the problem being handled in
1750 * this iteration of the loop */
1753 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1756 Perl_warner(aTHX_ pack_warn, "%s", message);
1758 } /* End of 'while (possible_problems)' */
1760 /* Since there was a possible problem, the returned length may need to
1761 * be changed from the one stored at the beginning of this function.
1762 * Instead of trying to figure out if that's needed, just do it. */
1768 if (flags & UTF8_CHECK_ONLY && retlen) {
1769 *retlen = ((STRLEN) -1);
1775 return UNI_TO_NATIVE(uv);
1779 =for apidoc utf8_to_uvchr_buf
1781 Returns the native code point of the first character in the string C<s> which
1782 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1783 C<*retlen> will be set to the length, in bytes, of that character.
1785 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1786 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1787 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1788 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1789 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1790 the next possible position in C<s> that could begin a non-malformed character.
1791 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1794 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1795 unless those are turned off.
1799 Also implemented as a macro in utf8.h
1805 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1807 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1811 return utf8n_to_uvchr(s, send - s, retlen,
1812 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1815 /* This is marked as deprecated
1817 =for apidoc utf8_to_uvuni_buf
1819 Only in very rare circumstances should code need to be dealing in Unicode
1820 (as opposed to native) code points. In those few cases, use
1821 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1823 Returns the Unicode (not-native) code point of the first character in the
1825 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1826 C<retlen> will be set to the length, in bytes, of that character.
1828 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1829 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1830 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1831 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1832 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1833 next possible position in C<s> that could begin a non-malformed character.
1834 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1836 Code points above the platform's C<IV_MAX> will raise a deprecation warning,
1837 unless those are turned off.
1843 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1845 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
1849 /* Call the low level routine, asking for checks */
1850 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
1854 =for apidoc utf8_length
1856 Return the length of the UTF-8 char encoded string C<s> in characters.
1857 Stops at C<e> (inclusive). If C<e E<lt> s> or if the scan would end
1858 up past C<e>, croaks.
1864 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
1868 PERL_ARGS_ASSERT_UTF8_LENGTH;
1870 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
1871 * the bitops (especially ~) can create illegal UTF-8.
1872 * In other words: in Perl UTF-8 is not just for Unicode. */
1875 goto warn_and_return;
1885 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1886 "%s in %s", unees, OP_DESC(PL_op));
1888 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1895 =for apidoc bytes_cmp_utf8
1897 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
1898 sequence of characters (stored as UTF-8)
1899 in C<u>, C<ulen>. Returns 0 if they are
1900 equal, -1 or -2 if the first string is less than the second string, +1 or +2
1901 if the first string is greater than the second string.
1903 -1 or +1 is returned if the shorter string was identical to the start of the
1904 longer string. -2 or +2 is returned if
1905 there was a difference between characters
1912 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
1914 const U8 *const bend = b + blen;
1915 const U8 *const uend = u + ulen;
1917 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
1919 while (b < bend && u < uend) {
1921 if (!UTF8_IS_INVARIANT(c)) {
1922 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
1925 if (UTF8_IS_CONTINUATION(c1)) {
1926 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
1928 /* diag_listed_as: Malformed UTF-8 character%s */
1929 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1931 unexpected_non_continuation_text(u - 2, 2, 1, 2),
1932 PL_op ? " in " : "",
1933 PL_op ? OP_DESC(PL_op) : "");
1938 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
1939 "%s in %s", unees, OP_DESC(PL_op));
1941 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
1942 return -2; /* Really want to return undef :-) */
1949 return *b < c ? -2 : +2;
1954 if (b == bend && u == uend)
1957 return b < bend ? +1 : -1;
1961 =for apidoc utf8_to_bytes
1963 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
1964 Unlike L</bytes_to_utf8>, this over-writes the original string, and
1965 updates C<*lenp> to contain the new length.
1966 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
1968 Upon successful return, the number of variants in the string can be computed by
1969 having saved the value of C<*lenp> before the call, and subtracting the
1970 after-call value of C<*lenp> from it.
1972 If you need a copy of the string, see L</bytes_from_utf8>.
1978 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
1982 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
1983 PERL_UNUSED_CONTEXT;
1985 /* This is a no-op if no variants at all in the input */
1986 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
1991 U8 * const save = s;
1992 U8 * const send = s + *lenp;
1995 /* Nothing before the first variant needs to be changed, so start the real
1999 if (! UTF8_IS_INVARIANT(*s)) {
2000 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2001 *lenp = ((STRLEN) -1);
2009 /* Is downgradable, so do it */
2010 d = s = first_variant;
2013 if (! UVCHR_IS_INVARIANT(c)) {
2014 /* Then it is two-byte encoded */
2015 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2028 =for apidoc bytes_from_utf8
2030 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2031 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2032 actually encoded in UTF-8.
2034 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2037 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2038 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2039 C<*lenp> are unchanged, and the return value is the original C<s>.
2041 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2042 newly created string containing a downgraded copy of C<s>, and whose length is
2043 returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
2045 Upon successful return, the number of variants in the string can be computed by
2046 having saved the value of C<*lenp> before the call, and subtracting the
2047 after-call value of C<*lenp> from it.
2051 There is a macro that avoids this function call, but this is retained for
2052 anyone who calls it with the Perl_ prefix */
2055 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2057 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2058 PERL_UNUSED_CONTEXT;
2060 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2064 No = here because currently externally undocumented
2065 for apidoc bytes_from_utf8_loc
2067 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2068 to store the location of the first character in C<"s"> that cannot be
2069 converted to non-UTF8.
2071 If that parameter is C<NULL>, this function behaves identically to
2074 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2075 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2077 Otherwise, the function returns a newly created C<NUL>-terminated string
2078 containing the non-UTF8 equivalent of the convertible first portion of
2079 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2080 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2081 and C<*first_non_downgradable> is set to C<NULL>.
2083 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2084 first character in the original string that wasn't converted. C<*is_utf8p> is
2085 unchanged. Note that the new string may have length 0.
2087 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2088 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2089 converts as many characters in it as possible stopping at the first one it
2090 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2091 set to point to that. The function returns the portion that could be converted
2092 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2093 not including the terminating C<NUL>. If the very first character in the
2094 original could not be converted, C<*lenp> will be 0, and the new string will
2095 contain just a single C<NUL>. If the entire input string was converted,
2096 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2098 Upon successful return, the number of variants in the converted portion of the
2099 string can be computed by having saved the value of C<*lenp> before the call,
2100 and subtracting the after-call value of C<*lenp> from it.
2108 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2111 const U8 *original = s;
2112 U8 *converted_start;
2113 const U8 *send = s + *lenp;
2115 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2118 if (first_unconverted) {
2119 *first_unconverted = NULL;
2122 return (U8 *) original;
2125 Newx(d, (*lenp) + 1, U8);
2127 converted_start = d;
2130 if (! UTF8_IS_INVARIANT(c)) {
2132 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2133 * have to stop now */
2134 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2135 if (first_unconverted) {
2136 *first_unconverted = s - 1;
2137 goto finish_and_return;
2140 Safefree(converted_start);
2141 return (U8 *) original;
2145 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2151 /* Here, converted the whole of the input */
2153 if (first_unconverted) {
2154 *first_unconverted = NULL;
2159 *lenp = d - converted_start;
2161 /* Trim unused space */
2162 Renew(converted_start, *lenp + 1, U8);
2164 return converted_start;
2168 =for apidoc bytes_to_utf8
2170 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2172 Returns a pointer to the newly-created string, and sets C<*lenp> to
2173 reflect the new length in bytes.
2175 Upon successful return, the number of variants in the string can be computed by
2176 having saved the value of C<*lenp> before the call, and subtracting it from the
2177 after-call value of C<*lenp>.
2179 A C<NUL> character will be written after the end of the string.
2181 If you want to convert to UTF-8 from encodings other than
2182 the native (Latin1 or EBCDIC),
2183 see L</sv_recode_to_utf8>().
2189 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2191 const U8 * const send = s + (*lenp);
2195 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2196 PERL_UNUSED_CONTEXT;
2198 Newx(d, (*lenp) * 2 + 1, U8);
2202 append_utf8_from_native_byte(*s, &d);
2211 * Convert native (big-endian) or reversed (little-endian) UTF-16 to UTF-8.
2213 * Destination must be pre-extended to 3/2 source. Do not use in-place.
2214 * We optimize for native, for obvious reasons. */
2217 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2222 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2225 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2231 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2233 if (OFFUNI_IS_INVARIANT(uv)) {
2234 *d++ = LATIN1_TO_NATIVE((U8) uv);
2237 if (uv <= MAX_UTF8_TWO_BYTE) {
2238 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2239 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2242 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2243 #define LAST_HIGH_SURROGATE 0xDBFF
2244 #define FIRST_LOW_SURROGATE 0xDC00
2245 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2247 /* This assumes that most uses will be in the first Unicode plane, not
2248 * needing surrogates */
2249 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2250 && uv <= UNICODE_SURROGATE_LAST))
2252 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2253 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2256 UV low = (p[0] << 8) + p[1];
2257 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2258 || UNLIKELY(low > LAST_LOW_SURROGATE))
2260 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2263 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2264 + (low - FIRST_LOW_SURROGATE) + 0x10000;
2268 d = uvoffuni_to_utf8_flags(d, uv, 0);
2271 *d++ = (U8)(( uv >> 12) | 0xe0);
2272 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2273 *d++ = (U8)(( uv & 0x3f) | 0x80);
2277 *d++ = (U8)(( uv >> 18) | 0xf0);
2278 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2279 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2280 *d++ = (U8)(( uv & 0x3f) | 0x80);
2285 *newlen = d - dstart;
2289 /* Note: this one is slightly destructive of the source. */
2292 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2295 U8* const send = s + bytelen;
2297 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2300 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2304 const U8 tmp = s[0];
2309 return utf16_to_utf8(p, d, bytelen, newlen);
2313 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2315 U8 tmpbuf[UTF8_MAXBYTES+1];
2316 uvchr_to_utf8(tmpbuf, c);
2317 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2320 /* Internal function so we can deprecate the external one, and call
2321 this one from other deprecated functions in this file */
2324 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2326 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2330 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2334 Perl__is_uni_perl_idcont(pTHX_ UV c)
2336 U8 tmpbuf[UTF8_MAXBYTES+1];
2337 uvchr_to_utf8(tmpbuf, c);
2338 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2342 Perl__is_uni_perl_idstart(pTHX_ UV c)
2344 U8 tmpbuf[UTF8_MAXBYTES+1];
2345 uvchr_to_utf8(tmpbuf, c);
2346 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2350 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2353 /* We have the latin1-range values compiled into the core, so just use
2354 * those, converting the result to UTF-8. The only difference between upper
2355 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2356 * either "SS" or "Ss". Which one to use is passed into the routine in
2357 * 'S_or_s' to avoid a test */
2359 UV converted = toUPPER_LATIN1_MOD(c);
2361 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2363 assert(S_or_s == 'S' || S_or_s == 's');
2365 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2366 characters in this range */
2367 *p = (U8) converted;
2372 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2373 * which it maps to one of them, so as to only have to have one check for
2374 * it in the main case */
2375 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2377 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2378 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2381 converted = GREEK_CAPITAL_LETTER_MU;
2383 #if UNICODE_MAJOR_VERSION > 2 \
2384 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2385 && UNICODE_DOT_DOT_VERSION >= 8)
2386 case LATIN_SMALL_LETTER_SHARP_S:
2393 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2394 " '%c' to map to '%c'",
2395 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2396 NOT_REACHED; /* NOTREACHED */
2400 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2401 *p = UTF8_TWO_BYTE_LO(converted);
2407 /* Call the function to convert a UTF-8 encoded character to the specified case.
2408 * Note that there may be more than one character in the result.
2409 * INP is a pointer to the first byte of the input character
2410 * OUTP will be set to the first byte of the string of changed characters. It
2411 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2412 * LENP will be set to the length in bytes of the string of changed characters
2414 * The functions return the ordinal of the first character in the string of
2416 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2417 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2418 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2419 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2420 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2421 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2423 /* This additionally has the input parameter 'specials', which if non-zero will
2424 * cause this to use the specials hash for folding (meaning get full case
2425 * folding); otherwise, when zero, this implies a simple case fold */
2426 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2427 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2430 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2432 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2433 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2434 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2435 * the changed version may be longer than the original character.
2437 * The ordinal of the first character of the changed version is returned
2438 * (but note, as explained above, that there may be more.) */
2440 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2443 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2446 uvchr_to_utf8(p, c);
2447 return CALL_UPPER_CASE(c, p, p, lenp);
2451 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2453 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2456 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2459 uvchr_to_utf8(p, c);
2460 return CALL_TITLE_CASE(c, p, p, lenp);
2464 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2466 /* We have the latin1-range values compiled into the core, so just use
2467 * those, converting the result to UTF-8. Since the result is always just
2468 * one character, we allow <p> to be NULL */
2470 U8 converted = toLOWER_LATIN1(c);
2472 PERL_UNUSED_ARG(dummy);
2475 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2480 /* Result is known to always be < 256, so can use the EIGHT_BIT
2482 *p = UTF8_EIGHT_BIT_HI(converted);
2483 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2491 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2493 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2496 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2499 uvchr_to_utf8(p, c);
2500 return CALL_LOWER_CASE(c, p, p, lenp);
2504 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2505 const unsigned int flags)
2507 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2508 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2509 * FOLD_FLAGS_FULL iff full folding is to be used;
2511 * Not to be used for locale folds
2516 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2517 PERL_UNUSED_CONTEXT;
2519 assert (! (flags & FOLD_FLAGS_LOCALE));
2521 if (UNLIKELY(c == MICRO_SIGN)) {
2522 converted = GREEK_SMALL_LETTER_MU;
2524 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2525 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2526 || UNICODE_DOT_DOT_VERSION > 0)
2527 else if ( (flags & FOLD_FLAGS_FULL)
2528 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2530 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2531 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2532 * under those circumstances. */
2533 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2534 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2535 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2537 return LATIN_SMALL_LETTER_LONG_S;
2547 else { /* In this range the fold of all other characters is their lower
2549 converted = toLOWER_LATIN1(c);
2552 if (UVCHR_IS_INVARIANT(converted)) {
2553 *p = (U8) converted;
2557 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2558 *p = UTF8_TWO_BYTE_LO(converted);
2566 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2569 /* Not currently externally documented, and subject to change
2570 * <flags> bits meanings:
2571 * FOLD_FLAGS_FULL iff full folding is to be used;
2572 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2573 * locale are to be used.
2574 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2577 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2579 if (flags & FOLD_FLAGS_LOCALE) {
2580 /* Treat a UTF-8 locale as not being in locale at all */
2581 if (IN_UTF8_CTYPE_LOCALE) {
2582 flags &= ~FOLD_FLAGS_LOCALE;
2585 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2586 goto needs_full_generality;
2591 return _to_fold_latin1((U8) c, p, lenp,
2592 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2595 /* Here, above 255. If no special needs, just use the macro */
2596 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2597 uvchr_to_utf8(p, c);
2598 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2600 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2601 the special flags. */
2602 U8 utf8_c[UTF8_MAXBYTES + 1];
2604 needs_full_generality:
2605 uvchr_to_utf8(utf8_c, c);
2606 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2611 PERL_STATIC_INLINE bool
2612 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2613 const char *const swashname, SV* const invlist)
2615 /* returns a boolean giving whether or not the UTF8-encoded character that
2616 * starts at <p> is in the swash indicated by <swashname>. <swash>
2617 * contains a pointer to where the swash indicated by <swashname>
2618 * is to be stored; which this routine will do, so that future calls will
2619 * look at <*swash> and only generate a swash if it is not null. <invlist>
2620 * is NULL or an inversion list that defines the swash. If not null, it
2621 * saves time during initialization of the swash.
2623 * Note that it is assumed that the buffer length of <p> is enough to
2624 * contain all the bytes that comprise the character. Thus, <*p> should
2625 * have been checked before this call for mal-formedness enough to assure
2628 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2630 /* The API should have included a length for the UTF-8 character in <p>,
2631 * but it doesn't. We therefore assume that p has been validated at least
2632 * as far as there being enough bytes available in it to accommodate the
2633 * character without reading beyond the end, and pass that number on to the
2634 * validating routine */
2635 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2636 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2637 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2639 NOT_REACHED; /* NOTREACHED */
2643 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2644 *swash = _core_swash_init("utf8",
2646 /* Only use the name if there is no inversion
2647 * list; otherwise will go out to disk */
2648 (invlist) ? "" : swashname,
2650 &PL_sv_undef, 1, 0, invlist, &flags);
2653 return swash_fetch(*swash, p, TRUE) != 0;
2656 PERL_STATIC_INLINE bool
2657 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2658 SV **swash, const char *const swashname,
2661 /* returns a boolean giving whether or not the UTF8-encoded character that
2662 * starts at <p>, and extending no further than <e - 1> is in the swash
2663 * indicated by <swashname>. <swash> contains a pointer to where the swash
2664 * indicated by <swashname> is to be stored; which this routine will do, so
2665 * that future calls will look at <*swash> and only generate a swash if it
2666 * is not null. <invlist> is NULL or an inversion list that defines the
2667 * swash. If not null, it saves time during initialization of the swash.
2670 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2672 if (! isUTF8_CHAR(p, e)) {
2673 _force_out_malformed_utf8_message(p, e, 0, 1);
2674 NOT_REACHED; /* NOTREACHED */
2678 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2679 *swash = _core_swash_init("utf8",
2681 /* Only use the name if there is no inversion
2682 * list; otherwise will go out to disk */
2683 (invlist) ? "" : swashname,
2685 &PL_sv_undef, 1, 0, invlist, &flags);
2688 return swash_fetch(*swash, p, TRUE) != 0;
2692 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2693 const char * const alternative,
2694 const bool use_locale,
2695 const char * const file,
2696 const unsigned line)
2700 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2702 if (ckWARN_d(WARN_DEPRECATED)) {
2704 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2705 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2706 if (! PL_seen_deprecated_macro) {
2707 PL_seen_deprecated_macro = newHV();
2709 if (! hv_store(PL_seen_deprecated_macro, key,
2710 strlen(key), &PL_sv_undef, 0))
2712 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2715 if (instr(file, "mathoms.c")) {
2716 Perl_warner(aTHX_ WARN_DEPRECATED,
2717 "In %s, line %d, starting in Perl v5.30, %s()"
2718 " will be removed. Avoid this message by"
2719 " converting to use %s().\n",
2720 file, line, name, alternative);
2723 Perl_warner(aTHX_ WARN_DEPRECATED,
2724 "In %s, line %d, starting in Perl v5.30, %s() will"
2725 " require an additional parameter. Avoid this"
2726 " message by converting to use %s().\n",
2727 file, line, name, alternative);
2734 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2736 const char * const name,
2737 const char * const alternative,
2738 const bool use_utf8,
2739 const bool use_locale,
2740 const char * const file,
2741 const unsigned line)
2743 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2745 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2747 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2757 case _CC_ALPHANUMERIC:
2761 return is_utf8_common(p,
2762 &PL_utf8_swash_ptrs[classnum],
2763 swash_property_names[classnum],
2764 PL_XPosix_ptrs[classnum]);
2767 return is_XPERLSPACE_high(p);
2769 return is_HORIZWS_high(p);
2771 return is_XDIGIT_high(p);
2777 return is_VERTWS_high(p);
2779 if (! PL_utf8_perl_idstart) {
2780 PL_utf8_perl_idstart
2781 = _new_invlist_C_array(_Perl_IDStart_invlist);
2783 return is_utf8_common(p, &PL_utf8_perl_idstart,
2784 "_Perl_IDStart", NULL);
2786 if (! PL_utf8_perl_idcont) {
2788 = _new_invlist_C_array(_Perl_IDCont_invlist);
2790 return is_utf8_common(p, &PL_utf8_perl_idcont,
2791 "_Perl_IDCont", NULL);
2795 /* idcont is the same as wordchar below 256 */
2796 if (classnum == _CC_IDCONT) {
2797 classnum = _CC_WORDCHAR;
2799 else if (classnum == _CC_IDFIRST) {
2803 classnum = _CC_ALPHA;
2807 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2808 return _generic_isCC(*p, classnum);
2811 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2814 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2815 return isFOO_lc(classnum, *p);
2818 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2821 NOT_REACHED; /* NOTREACHED */
2825 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2828 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2830 assert(classnum < _FIRST_NON_SWASH_CC);
2832 return is_utf8_common_with_len(p,
2834 &PL_utf8_swash_ptrs[classnum],
2835 swash_property_names[classnum],
2836 PL_XPosix_ptrs[classnum]);
2840 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
2844 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
2846 if (! PL_utf8_perl_idstart) {
2847 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
2849 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
2850 "_Perl_IDStart", invlist);
2854 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
2856 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
2860 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
2864 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
2868 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
2870 if (! PL_utf8_perl_idcont) {
2871 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
2873 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
2874 "_Perl_IDCont", invlist);
2878 Perl__is_utf8_idcont(pTHX_ const U8 *p)
2880 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
2882 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
2886 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
2888 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
2890 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
2894 Perl__is_utf8_mark(pTHX_ const U8 *p)
2896 PERL_ARGS_ASSERT__IS_UTF8_MARK;
2898 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
2901 /* change namve uv1 to 'from' */
2903 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
2904 SV **swashp, const char *normal, const char *special)
2908 PERL_ARGS_ASSERT__TO_UTF8_CASE;
2910 /* For code points that don't change case, we already know that the output
2911 * of this function is the unchanged input, so we can skip doing look-ups
2912 * for them. Unfortunately the case-changing code points are scattered
2913 * around. But there are some long consecutive ranges where there are no
2914 * case changing code points. By adding tests, we can eliminate the lookup
2915 * for all the ones in such ranges. This is currently done here only for
2916 * just a few cases where the scripts are in common use in modern commerce
2917 * (and scripts adjacent to those which can be included without additional
2920 if (uv1 >= 0x0590) {
2921 /* This keeps from needing further processing the code points most
2922 * likely to be used in the following non-cased scripts: Hebrew,
2923 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
2924 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
2925 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
2930 /* The following largish code point ranges also don't have case
2931 * changes, but khw didn't think they warranted extra tests to speed
2932 * them up (which would slightly slow down everything else above them):
2933 * 1100..139F Hangul Jamo, Ethiopic
2934 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
2935 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
2936 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
2937 * Combining Diacritical Marks Extended, Balinese,
2938 * Sundanese, Batak, Lepcha, Ol Chiki
2939 * 2000..206F General Punctuation
2942 if (uv1 >= 0x2D30) {
2944 /* This keeps the from needing further processing the code points
2945 * most likely to be used in the following non-cased major scripts:
2946 * CJK, Katakana, Hiragana, plus some less-likely scripts.
2948 * (0x2D30 above might have to be changed to 2F00 in the unlikely
2949 * event that Unicode eventually allocates the unused block as of
2950 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
2951 * that the test suite will start having failures to alert you
2952 * should that happen) */
2957 if (uv1 >= 0xAC00) {
2958 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
2959 if (ckWARN_d(WARN_SURROGATE)) {
2960 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2961 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
2962 "Operation \"%s\" returns its argument for"
2963 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
2968 /* AC00..FAFF Catches Hangul syllables and private use, plus
2975 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
2976 if ( UNLIKELY(uv1 > MAX_NON_DEPRECATED_CP)
2977 && ckWARN_d(WARN_DEPRECATED))
2979 Perl_warner(aTHX_ packWARN(WARN_DEPRECATED),
2980 cp_above_legal_max, uv1, MAX_NON_DEPRECATED_CP);
2982 if (ckWARN_d(WARN_NON_UNICODE)) {
2983 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
2984 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
2985 "Operation \"%s\" returns its argument for"
2986 " non-Unicode code point 0x%04" UVXf, desc, uv1);
2990 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
2992 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
2995 /* As of Unicode 10.0, this means we avoid swash creation
2996 * for anything beyond high Plane 1 (below emojis) */
3003 /* Note that non-characters are perfectly legal, so no warning should
3004 * be given. There are so few of them, that it isn't worth the extra
3005 * tests to avoid swash creation */
3008 if (!*swashp) /* load on-demand */
3009 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3013 /* It might be "special" (sometimes, but not always,
3014 * a multicharacter mapping) */
3018 /* If passed in the specials name, use that; otherwise use any
3019 * given in the swash */
3020 if (*special != '\0') {
3021 hv = get_hv(special, 0);
3024 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3026 hv = MUTABLE_HV(SvRV(*svp));
3031 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
3036 s = SvPV_const(*svp, len);
3039 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
3041 Copy(s, ustrp, len, U8);
3046 if (!len && *swashp) {
3047 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
3050 /* It was "normal" (a single character mapping). */
3051 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
3059 return valid_utf8_to_uvchr(ustrp, 0);
3062 /* Here, there was no mapping defined, which means that the code point maps
3063 * to itself. Return the inputs */
3066 if (p != ustrp) { /* Don't copy onto itself */
3067 Copy(p, ustrp, len, U8);
3078 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3079 U8* const ustrp, STRLEN *lenp)
3081 /* This is called when changing the case of a UTF-8-encoded character above
3082 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3083 * result contains a character that crosses the 255/256 boundary, disallow
3084 * the change, and return the original code point. See L<perlfunc/lc> for
3087 * p points to the original string whose case was changed; assumed
3088 * by this routine to be well-formed
3089 * result the code point of the first character in the changed-case string
3090 * ustrp points to the changed-case string (<result> represents its
3092 * lenp points to the length of <ustrp> */
3094 UV original; /* To store the first code point of <p> */
3096 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3098 assert(UTF8_IS_ABOVE_LATIN1(*p));
3100 /* We know immediately if the first character in the string crosses the
3101 * boundary, so can skip */
3104 /* Look at every character in the result; if any cross the
3105 * boundary, the whole thing is disallowed */
3106 U8* s = ustrp + UTF8SKIP(ustrp);
3107 U8* e = ustrp + *lenp;
3109 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3115 /* Here, no characters crossed, result is ok as-is, but we warn. */
3116 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3122 /* Failed, have to return the original */
3123 original = valid_utf8_to_uvchr(p, lenp);
3125 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3126 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3127 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3128 " locale; resolved to \"\\x{%" UVXf "}\".",
3132 Copy(p, ustrp, *lenp, char);
3137 S_check_and_deprecate(pTHX_ const U8 *p,
3139 const unsigned int type, /* See below */
3140 const bool use_locale, /* Is this a 'LC_'
3142 const char * const file,
3143 const unsigned line)
3145 /* This is a temporary function to deprecate the unsafe calls to the case
3146 * changing macros and functions. It keeps all the special stuff in just
3149 * It updates *e with the pointer to the end of the input string. If using
3150 * the old-style macros, *e is NULL on input, and so this function assumes
3151 * the input string is long enough to hold the entire UTF-8 sequence, and
3152 * sets *e accordingly, but it then returns a flag to pass the
3153 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3154 * using the full length if possible.
3156 * It also does the assert that *e > p when *e is not NULL. This should be
3157 * migrated to the callers when this function gets deleted.
3159 * The 'type' parameter is used for the caller to specify which case
3160 * changing function this is called from: */
3162 # define DEPRECATE_TO_UPPER 0
3163 # define DEPRECATE_TO_TITLE 1
3164 # define DEPRECATE_TO_LOWER 2
3165 # define DEPRECATE_TO_FOLD 3
3167 U32 utf8n_flags = 0;
3169 const char * alternative;
3171 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3174 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3175 *e = p + UTF8SKIP(p);
3177 /* For mathoms.c calls, we use the function name we know is stored
3178 * there. It could be part of a larger path */
3179 if (type == DEPRECATE_TO_UPPER) {
3180 name = instr(file, "mathoms.c")
3183 alternative = "toUPPER_utf8_safe";
3185 else if (type == DEPRECATE_TO_TITLE) {
3186 name = instr(file, "mathoms.c")
3189 alternative = "toTITLE_utf8_safe";
3191 else if (type == DEPRECATE_TO_LOWER) {
3192 name = instr(file, "mathoms.c")
3195 alternative = "toLOWER_utf8_safe";
3197 else if (type == DEPRECATE_TO_FOLD) {
3198 name = instr(file, "mathoms.c")
3201 alternative = "toFOLD_utf8_safe";
3203 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3205 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3214 /* The process for changing the case is essentially the same for the four case
3215 * change types, except there are complications for folding. Otherwise the
3216 * difference is only which case to change to. To make sure that they all do
3217 * the same thing, the bodies of the functions are extracted out into the
3218 * following two macros. The functions are written with the same variable
3219 * names, and these are known and used inside these macros. It would be
3220 * better, of course, to have inline functions to do it, but since different
3221 * macros are called, depending on which case is being changed to, this is not
3222 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3223 * function can start with the common start macro, then finish with its special
3224 * handling; while the other three cases can just use the common end macro.
3226 * The algorithm is to use the proper (passed in) macro or function to change
3227 * the case for code points that are below 256. The macro is used if using
3228 * locale rules for the case change; the function if not. If the code point is
3229 * above 255, it is computed from the input UTF-8, and another macro is called
3230 * to do the conversion. If necessary, the output is converted to UTF-8. If
3231 * using a locale, we have to check that the change did not cross the 255/256
3232 * boundary, see check_locale_boundary_crossing() for further details.
3234 * The macros are split with the correct case change for the below-256 case
3235 * stored into 'result', and in the middle of an else clause for the above-255
3236 * case. At that point in the 'else', 'result' is not the final result, but is
3237 * the input code point calculated from the UTF-8. The fold code needs to
3238 * realize all this and take it from there.
3240 * If you read the two macros as sequential, it's easier to understand what's
3242 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3243 L1_func_extra_param) \
3245 if (flags & (locale_flags)) { \
3246 /* Treat a UTF-8 locale as not being in locale at all */ \
3247 if (IN_UTF8_CTYPE_LOCALE) { \
3248 flags &= ~(locale_flags); \
3251 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3255 if (UTF8_IS_INVARIANT(*p)) { \
3256 if (flags & (locale_flags)) { \
3257 result = LC_L1_change_macro(*p); \
3260 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3263 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3264 if (flags & (locale_flags)) { \
3265 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3269 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3270 ustrp, lenp, L1_func_extra_param); \
3273 else { /* malformed UTF-8 or ord above 255 */ \
3274 STRLEN len_result; \
3275 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3276 if (len_result == (STRLEN) -1) { \
3277 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3281 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3282 result = change_macro(result, p, ustrp, lenp); \
3284 if (flags & (locale_flags)) { \
3285 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3290 /* Here, used locale rules. Convert back to UTF-8 */ \
3291 if (UTF8_IS_INVARIANT(result)) { \
3292 *ustrp = (U8) result; \
3296 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3297 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3304 =for apidoc to_utf8_upper
3306 Instead use L</toUPPER_utf8_safe>.
3310 /* Not currently externally documented, and subject to change:
3311 * <flags> is set iff iff the rules from the current underlying locale are to
3315 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3320 const char * const file,
3324 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3325 cBOOL(flags), file, line);
3327 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3329 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3330 /* 2nd char of uc(U+DF) is 'S' */
3331 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3332 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3336 =for apidoc to_utf8_title
3338 Instead use L</toTITLE_utf8_safe>.
3342 /* Not currently externally documented, and subject to change:
3343 * <flags> is set iff the rules from the current underlying locale are to be
3344 * used. Since titlecase is not defined in POSIX, for other than a
3345 * UTF-8 locale, uppercase is used instead for code points < 256.
3349 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3354 const char * const file,
3358 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3359 cBOOL(flags), file, line);
3361 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3363 /* 2nd char of ucfirst(U+DF) is 's' */
3364 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3365 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3369 =for apidoc to_utf8_lower
3371 Instead use L</toLOWER_utf8_safe>.
3375 /* Not currently externally documented, and subject to change:
3376 * <flags> is set iff iff the rules from the current underlying locale are to
3381 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3386 const char * const file,
3390 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3391 cBOOL(flags), file, line);
3393 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3395 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3396 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3400 =for apidoc to_utf8_fold
3402 Instead use L</toFOLD_utf8_safe>.
3406 /* Not currently externally documented, and subject to change,
3408 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3409 * locale are to be used.
3410 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3411 * otherwise simple folds
3412 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3417 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3422 const char * const file,
3426 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3427 cBOOL(flags), file, line);
3429 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3431 /* These are mutually exclusive */
3432 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3434 assert(p != ustrp); /* Otherwise overwrites */
3436 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3437 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3439 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3441 if (flags & FOLD_FLAGS_LOCALE) {
3443 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3444 const unsigned int long_s_t_len = sizeof(LONG_S_T) - 1;
3446 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3447 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3449 const unsigned int cap_sharp_s_len = sizeof(CAP_SHARP_S) - 1;
3451 /* Special case these two characters, as what normally gets
3452 * returned under locale doesn't work */
3453 if (UTF8SKIP(p) == cap_sharp_s_len
3454 && memEQ((char *) p, CAP_SHARP_S, cap_sharp_s_len))
3456 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3457 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3458 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3459 "resolved to \"\\x{17F}\\x{17F}\".");
3464 if (UTF8SKIP(p) == long_s_t_len
3465 && memEQ((char *) p, LONG_S_T, long_s_t_len))
3467 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3468 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3469 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3470 "resolved to \"\\x{FB06}\".");
3471 goto return_ligature_st;
3474 #if UNICODE_MAJOR_VERSION == 3 \
3475 && UNICODE_DOT_VERSION == 0 \
3476 && UNICODE_DOT_DOT_VERSION == 1
3477 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3479 /* And special case this on this Unicode version only, for the same
3480 * reaons the other two are special cased. They would cross the
3481 * 255/256 boundary which is forbidden under /l, and so the code
3482 * wouldn't catch that they are equivalent (which they are only in
3484 else if (UTF8SKIP(p) == sizeof(DOTTED_I) - 1
3485 && memEQ((char *) p, DOTTED_I, sizeof(DOTTED_I) - 1))
3487 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3488 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3489 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3490 "resolved to \"\\x{0131}\".");
3491 goto return_dotless_i;
3495 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3497 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3501 /* This is called when changing the case of a UTF-8-encoded
3502 * character above the ASCII range, and the result should not
3503 * contain an ASCII character. */
3505 UV original; /* To store the first code point of <p> */
3507 /* Look at every character in the result; if any cross the
3508 * boundary, the whole thing is disallowed */
3510 U8* e = ustrp + *lenp;
3513 /* Crossed, have to return the original */
3514 original = valid_utf8_to_uvchr(p, lenp);
3516 /* But in these instances, there is an alternative we can
3517 * return that is valid */
3518 if (original == LATIN_SMALL_LETTER_SHARP_S
3519 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3520 || original == LATIN_CAPITAL_LETTER_SHARP_S
3525 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3526 goto return_ligature_st;
3528 #if UNICODE_MAJOR_VERSION == 3 \
3529 && UNICODE_DOT_VERSION == 0 \
3530 && UNICODE_DOT_DOT_VERSION == 1
3532 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3533 goto return_dotless_i;
3536 Copy(p, ustrp, *lenp, char);
3542 /* Here, no characters crossed, result is ok as-is */
3547 /* Here, used locale rules. Convert back to UTF-8 */
3548 if (UTF8_IS_INVARIANT(result)) {
3549 *ustrp = (U8) result;
3553 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3554 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3561 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3562 * folds to a string of two of these characters. By returning this
3563 * instead, then, e.g.,
3564 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3567 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3568 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3570 return LATIN_SMALL_LETTER_LONG_S;
3573 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3574 * have the other one fold to it */
3576 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3577 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3578 return LATIN_SMALL_LIGATURE_ST;
3580 #if UNICODE_MAJOR_VERSION == 3 \
3581 && UNICODE_DOT_VERSION == 0 \
3582 && UNICODE_DOT_DOT_VERSION == 1
3585 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3586 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3587 return LATIN_SMALL_LETTER_DOTLESS_I;
3594 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3595 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3596 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3600 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3601 I32 minbits, I32 none)
3603 PERL_ARGS_ASSERT_SWASH_INIT;
3605 /* Returns a copy of a swash initiated by the called function. This is the
3606 * public interface, and returning a copy prevents others from doing
3607 * mischief on the original */
3609 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3614 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3615 I32 minbits, I32 none, SV* invlist,
3619 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3620 * use the following define */
3622 #define CORE_SWASH_INIT_RETURN(x) \
3623 PL_curpm= old_PL_curpm; \
3626 /* Initialize and return a swash, creating it if necessary. It does this
3627 * by calling utf8_heavy.pl in the general case. The returned value may be
3628 * the swash's inversion list instead if the input parameters allow it.
3629 * Which is returned should be immaterial to callers, as the only
3630 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3631 * and swash_to_invlist() handle both these transparently.
3633 * This interface should only be used by functions that won't destroy or
3634 * adversely change the swash, as doing so affects all other uses of the
3635 * swash in the program; the general public should use 'Perl_swash_init'
3638 * pkg is the name of the package that <name> should be in.
3639 * name is the name of the swash to find. Typically it is a Unicode
3640 * property name, including user-defined ones
3641 * listsv is a string to initialize the swash with. It must be of the form
3642 * documented as the subroutine return value in
3643 * L<perlunicode/User-Defined Character Properties>
3644 * minbits is the number of bits required to represent each data element.
3645 * It is '1' for binary properties.
3646 * none I (khw) do not understand this one, but it is used only in tr///.
3647 * invlist is an inversion list to initialize the swash with (or NULL)
3648 * flags_p if non-NULL is the address of various input and output flag bits
3649 * to the routine, as follows: ('I' means is input to the routine;
3650 * 'O' means output from the routine. Only flags marked O are
3651 * meaningful on return.)
3652 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3653 * came from a user-defined property. (I O)
3654 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3655 * when the swash cannot be located, to simply return NULL. (I)
3656 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3657 * return of an inversion list instead of a swash hash if this routine
3658 * thinks that would result in faster execution of swash_fetch() later
3661 * Thus there are three possible inputs to find the swash: <name>,
3662 * <listsv>, and <invlist>. At least one must be specified. The result
3663 * will be the union of the specified ones, although <listsv>'s various
3664 * actions can intersect, etc. what <name> gives. To avoid going out to
3665 * disk at all, <invlist> should specify completely what the swash should
3666 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3668 * <invlist> is only valid for binary properties */
3670 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3672 SV* retval = &PL_sv_undef;
3673 HV* swash_hv = NULL;
3674 const int invlist_swash_boundary =
3675 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3676 ? 512 /* Based on some benchmarking, but not extensive, see commit
3678 : -1; /* Never return just an inversion list */
3680 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3681 assert(! invlist || minbits == 1);
3683 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3684 regex that triggered the swash init and the swash init
3685 perl logic itself. See perl #122747 */
3687 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3689 if (listsv != &PL_sv_undef || strNE(name, "")) {
3691 const size_t pkg_len = strlen(pkg);
3692 const size_t name_len = strlen(name);
3693 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3697 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3699 PUSHSTACKi(PERLSI_MAGIC);
3703 /* We might get here via a subroutine signature which uses a utf8
3704 * parameter name, at which point PL_subname will have been set
3705 * but not yet used. */
3706 save_item(PL_subname);
3707 if (PL_parser && PL_parser->error_count)
3708 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3709 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3710 if (!method) { /* demand load UTF-8 */
3712 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3713 GvSV(PL_errgv) = NULL;
3714 #ifndef NO_TAINT_SUPPORT
3715 /* It is assumed that callers of this routine are not passing in
3716 * any user derived data. */
3717 /* Need to do this after save_re_context() as it will set
3718 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3719 * in Perl_magic_get). Even line to create errsv_save can turn on
3721 SAVEBOOL(TAINT_get);
3724 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3727 /* Not ERRSV, as there is no need to vivify a scalar we are
3728 about to discard. */
3729 SV * const errsv = GvSV(PL_errgv);
3730 if (!SvTRUE(errsv)) {
3731 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3732 SvREFCNT_dec(errsv);
3740 mPUSHp(pkg, pkg_len);
3741 mPUSHp(name, name_len);
3746 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3747 GvSV(PL_errgv) = NULL;
3748 /* If we already have a pointer to the method, no need to use
3749 * call_method() to repeat the lookup. */
3751 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3752 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3754 retval = *PL_stack_sp--;
3755 SvREFCNT_inc(retval);
3758 /* Not ERRSV. See above. */
3759 SV * const errsv = GvSV(PL_errgv);
3760 if (!SvTRUE(errsv)) {
3761 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3762 SvREFCNT_dec(errsv);
3767 if (IN_PERL_COMPILETIME) {
3768 CopHINTS_set(PL_curcop, PL_hints);
3770 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3771 if (SvPOK(retval)) {
3773 /* If caller wants to handle missing properties, let them */
3774 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3775 CORE_SWASH_INIT_RETURN(NULL);
3778 "Can't find Unicode property definition \"%" SVf "\"",
3780 NOT_REACHED; /* NOTREACHED */
3783 } /* End of calling the module to find the swash */
3785 /* If this operation fetched a swash, and we will need it later, get it */
3786 if (retval != &PL_sv_undef
3787 && (minbits == 1 || (flags_p
3789 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3791 swash_hv = MUTABLE_HV(SvRV(retval));
3793 /* If we don't already know that there is a user-defined component to
3794 * this swash, and the user has indicated they wish to know if there is
3795 * one (by passing <flags_p>), find out */
3796 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3797 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3798 if (user_defined && SvUV(*user_defined)) {
3799 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3804 /* Make sure there is an inversion list for binary properties */
3806 SV** swash_invlistsvp = NULL;
3807 SV* swash_invlist = NULL;
3808 bool invlist_in_swash_is_valid = FALSE;
3809 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3810 an unclaimed reference count */
3812 /* If this operation fetched a swash, get its already existing
3813 * inversion list, or create one for it */
3816 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3817 if (swash_invlistsvp) {
3818 swash_invlist = *swash_invlistsvp;
3819 invlist_in_swash_is_valid = TRUE;
3822 swash_invlist = _swash_to_invlist(retval);
3823 swash_invlist_unclaimed = TRUE;
3827 /* If an inversion list was passed in, have to include it */
3830 /* Any fetched swash will by now have an inversion list in it;
3831 * otherwise <swash_invlist> will be NULL, indicating that we
3832 * didn't fetch a swash */
3833 if (swash_invlist) {
3835 /* Add the passed-in inversion list, which invalidates the one
3836 * already stored in the swash */
3837 invlist_in_swash_is_valid = FALSE;
3838 SvREADONLY_off(swash_invlist); /* Turned on again below */
3839 _invlist_union(invlist, swash_invlist, &swash_invlist);
3843 /* Here, there is no swash already. Set up a minimal one, if
3844 * we are going to return a swash */
3845 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
3847 retval = newRV_noinc(MUTABLE_SV(swash_hv));
3849 swash_invlist = invlist;
3853 /* Here, we have computed the union of all the passed-in data. It may
3854 * be that there was an inversion list in the swash which didn't get
3855 * touched; otherwise save the computed one */
3856 if (! invlist_in_swash_is_valid
3857 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
3859 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
3861 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
3863 /* We just stole a reference count. */
3864 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
3865 else SvREFCNT_inc_simple_void_NN(swash_invlist);
3868 /* The result is immutable. Forbid attempts to change it. */
3869 SvREADONLY_on(swash_invlist);
3871 /* Use the inversion list stand-alone if small enough */
3872 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
3873 SvREFCNT_dec(retval);
3874 if (!swash_invlist_unclaimed)
3875 SvREFCNT_inc_simple_void_NN(swash_invlist);
3876 retval = newRV_noinc(swash_invlist);
3880 CORE_SWASH_INIT_RETURN(retval);
3881 #undef CORE_SWASH_INIT_RETURN
3885 /* This API is wrong for special case conversions since we may need to
3886 * return several Unicode characters for a single Unicode character
3887 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
3888 * the lower-level routine, and it is similarly broken for returning
3889 * multiple values. --jhi
3890 * For those, you should use S__to_utf8_case() instead */
3891 /* Now SWASHGET is recasted into S_swatch_get in this file. */
3894 * Returns the value of property/mapping C<swash> for the first character
3895 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
3896 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
3897 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
3899 * A "swash" is a hash which contains initially the keys/values set up by
3900 * SWASHNEW. The purpose is to be able to completely represent a Unicode
3901 * property for all possible code points. Things are stored in a compact form
3902 * (see utf8_heavy.pl) so that calculation is required to find the actual
3903 * property value for a given code point. As code points are looked up, new
3904 * key/value pairs are added to the hash, so that the calculation doesn't have
3905 * to ever be re-done. Further, each calculation is done, not just for the
3906 * desired one, but for a whole block of code points adjacent to that one.
3907 * For binary properties on ASCII machines, the block is usually for 64 code
3908 * points, starting with a code point evenly divisible by 64. Thus if the
3909 * property value for code point 257 is requested, the code goes out and
3910 * calculates the property values for all 64 code points between 256 and 319,
3911 * and stores these as a single 64-bit long bit vector, called a "swatch",
3912 * under the key for code point 256. The key is the UTF-8 encoding for code
3913 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
3914 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
3915 * for code point 258 is then requested, this code realizes that it would be
3916 * stored under the key for 256, and would find that value and extract the
3917 * relevant bit, offset from 256.
3919 * Non-binary properties are stored in as many bits as necessary to represent
3920 * their values (32 currently, though the code is more general than that), not
3921 * as single bits, but the principle is the same: the value for each key is a
3922 * vector that encompasses the property values for all code points whose UTF-8
3923 * representations are represented by the key. That is, for all code points
3924 * whose UTF-8 representations are length N bytes, and the key is the first N-1
3928 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
3930 HV *const hv = MUTABLE_HV(SvRV(swash));
3935 const U8 *tmps = NULL;
3939 PERL_ARGS_ASSERT_SWASH_FETCH;
3941 /* If it really isn't a hash, it isn't really swash; must be an inversion
3943 if (SvTYPE(hv) != SVt_PVHV) {
3944 return _invlist_contains_cp((SV*)hv,
3946 ? valid_utf8_to_uvchr(ptr, NULL)
3950 /* We store the values in a "swatch" which is a vec() value in a swash
3951 * hash. Code points 0-255 are a single vec() stored with key length
3952 * (klen) 0. All other code points have a UTF-8 representation
3953 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
3954 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
3955 * length for them is the length of the encoded char - 1. ptr[klen] is the
3956 * final byte in the sequence representing the character */
3957 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
3962 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
3965 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
3968 klen = UTF8SKIP(ptr) - 1;
3970 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
3971 * the vec is the final byte in the sequence. (In EBCDIC this is
3972 * converted to I8 to get consecutive values.) To help you visualize
3974 * Straight 1047 After final byte
3975 * UTF-8 UTF-EBCDIC I8 transform
3976 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
3977 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
3979 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
3980 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
3982 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
3983 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
3985 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
3986 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
3988 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
3989 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
3991 * (There are no discontinuities in the elided (...) entries.)
3992 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
3993 * key for the next 31, up through U+043F, whose UTF-8 final byte is
3994 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
3995 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
3996 * index into the vec() swatch (after subtracting 0x80, which we
3997 * actually do with an '&').
3998 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
3999 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4000 * dicontinuities which go away by transforming it into I8, and we
4001 * effectively subtract 0xA0 to get the index. */
4002 needents = (1 << UTF_ACCUMULATION_SHIFT);
4003 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4007 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4008 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4009 * it's nothing to sniff at.) Pity we usually come through at least
4010 * two function calls to get here...
4012 * NB: this code assumes that swatches are never modified, once generated!
4015 if (hv == PL_last_swash_hv &&
4016 klen == PL_last_swash_klen &&
4017 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4019 tmps = PL_last_swash_tmps;
4020 slen = PL_last_swash_slen;
4023 /* Try our second-level swatch cache, kept in a hash. */
4024 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4026 /* If not cached, generate it via swatch_get */
4027 if (!svp || !SvPOK(*svp)
4028 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4031 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4032 swatch = swatch_get(swash,
4033 code_point & ~((UV)needents - 1),
4036 else { /* For the first 256 code points, the swatch has a key of
4038 swatch = swatch_get(swash, 0, needents);
4041 if (IN_PERL_COMPILETIME)
4042 CopHINTS_set(PL_curcop, PL_hints);
4044 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4046 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4047 || (slen << 3) < needents)
4048 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4049 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4050 svp, tmps, (UV)slen, (UV)needents);
4053 PL_last_swash_hv = hv;
4054 assert(klen <= sizeof(PL_last_swash_key));
4055 PL_last_swash_klen = (U8)klen;
4056 /* FIXME change interpvar.h? */
4057 PL_last_swash_tmps = (U8 *) tmps;
4058 PL_last_swash_slen = slen;
4060 Copy(ptr, PL_last_swash_key, klen, U8);
4063 switch ((int)((slen << 3) / needents)) {
4065 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4067 return ((UV) tmps[off]);
4071 ((UV) tmps[off ] << 8) +
4072 ((UV) tmps[off + 1]);
4076 ((UV) tmps[off ] << 24) +
4077 ((UV) tmps[off + 1] << 16) +
4078 ((UV) tmps[off + 2] << 8) +
4079 ((UV) tmps[off + 3]);
4081 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4082 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4083 NORETURN_FUNCTION_END;
4086 /* Read a single line of the main body of the swash input text. These are of
4089 * where each number is hex. The first two numbers form the minimum and
4090 * maximum of a range, and the third is the value associated with the range.
4091 * Not all swashes should have a third number
4093 * On input: l points to the beginning of the line to be examined; it points
4094 * to somewhere in the string of the whole input text, and is
4095 * terminated by a \n or the null string terminator.
4096 * lend points to the null terminator of that string
4097 * wants_value is non-zero if the swash expects a third number
4098 * typestr is the name of the swash's mapping, like 'ToLower'
4099 * On output: *min, *max, and *val are set to the values read from the line.
4100 * returns a pointer just beyond the line examined. If there was no
4101 * valid min number on the line, returns lend+1
4105 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4106 const bool wants_value, const U8* const typestr)
4108 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4109 STRLEN numlen; /* Length of the number */
4110 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4111 | PERL_SCAN_DISALLOW_PREFIX
4112 | PERL_SCAN_SILENT_NON_PORTABLE;
4114 /* nl points to the next \n in the scan */
4115 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4117 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4119 /* Get the first number on the line: the range minimum */
4121 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4122 *max = *min; /* So can never return without setting max */
4123 if (numlen) /* If found a hex number, position past it */
4125 else if (nl) { /* Else, go handle next line, if any */
4126 return nl + 1; /* 1 is length of "\n" */
4128 else { /* Else, no next line */
4129 return lend + 1; /* to LIST's end at which \n is not found */
4132 /* The max range value follows, separated by a BLANK */
4135 flags = PERL_SCAN_SILENT_ILLDIGIT
4136 | PERL_SCAN_DISALLOW_PREFIX
4137 | PERL_SCAN_SILENT_NON_PORTABLE;
4139 *max = grok_hex((char *)l, &numlen, &flags, NULL);
4142 else /* If no value here, it is a single element range */
4145 /* Non-binary tables have a third entry: what the first element of the
4146 * range maps to. The map for those currently read here is in hex */
4150 flags = PERL_SCAN_SILENT_ILLDIGIT
4151 | PERL_SCAN_DISALLOW_PREFIX
4152 | PERL_SCAN_SILENT_NON_PORTABLE;
4154 *val = grok_hex((char *)l, &numlen, &flags, NULL);
4163 /* diag_listed_as: To%s: illegal mapping '%s' */
4164 Perl_croak(aTHX_ "%s: illegal mapping '%s'",
4170 *val = 0; /* bits == 1, then any val should be ignored */
4172 else { /* Nothing following range min, should be single element with no
4177 /* diag_listed_as: To%s: illegal mapping '%s' */
4178 Perl_croak(aTHX_ "%s: illegal mapping '%s'", typestr, l);
4182 *val = 0; /* bits == 1, then val should be ignored */
4185 /* Position to next line if any, or EOF */
4195 * Returns a swatch (a bit vector string) for a code point sequence
4196 * that starts from the value C<start> and comprises the number C<span>.
4197 * A C<swash> must be an object created by SWASHNEW (see lib/utf8_heavy.pl).
4198 * Should be used via swash_fetch, which will cache the swatch in C<swash>.
4201 S_swatch_get(pTHX_ SV* swash, UV start, UV span)
4204 U8 *l, *lend, *x, *xend, *s, *send;
4205 STRLEN lcur, xcur, scur;
4206 HV *const hv = MUTABLE_HV(SvRV(swash));
4207 SV** const invlistsvp = hv_fetchs(hv, "V", FALSE);
4209 SV** listsvp = NULL; /* The string containing the main body of the table */
4210 SV** extssvp = NULL;
4211 SV** invert_it_svp = NULL;
4214 STRLEN octets; /* if bits == 1, then octets == 0 */
4216 UV end = start + span;
4218 if (invlistsvp == NULL) {
4219 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4220 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4221 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4222 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4223 listsvp = hv_fetchs(hv, "LIST", FALSE);
4224 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4226 bits = SvUV(*bitssvp);
4227 none = SvUV(*nonesvp);
4228 typestr = (U8*)SvPV_nolen(*typesvp);
4234 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4236 PERL_ARGS_ASSERT_SWATCH_GET;
4238 if (bits != 1 && bits != 8 && bits != 16 && bits != 32) {
4239 Perl_croak(aTHX_ "panic: swatch_get doesn't expect bits %" UVuf,
4243 /* If overflowed, use the max possible */
4249 /* create and initialize $swatch */
4250 scur = octets ? (span * octets) : (span + 7) / 8;
4251 swatch = newSV(scur);
4253 s = (U8*)SvPVX(swatch);
4254 if (octets && none) {
4255 const U8* const e = s + scur;
4258 *s++ = (U8)(none & 0xff);
4259 else if (bits == 16) {
4260 *s++ = (U8)((none >> 8) & 0xff);
4261 *s++ = (U8)( none & 0xff);
4263 else if (bits == 32) {
4264 *s++ = (U8)((none >> 24) & 0xff);
4265 *s++ = (U8)((none >> 16) & 0xff);
4266 *s++ = (U8)((none >> 8) & 0xff);
4267 *s++ = (U8)( none & 0xff);
4273 (void)memzero((U8*)s, scur + 1);
4275 SvCUR_set(swatch, scur);
4276 s = (U8*)SvPVX(swatch);
4278 if (invlistsvp) { /* If has an inversion list set up use that */
4279 _invlist_populate_swatch(*invlistsvp, start, end, s);
4283 /* read $swash->{LIST} */
4284 l = (U8*)SvPV(*listsvp, lcur);
4287 UV min, max, val, upper;
4288 l = swash_scan_list_line(l, lend, &min, &max, &val,
4289 cBOOL(octets), typestr);
4294 /* If looking for something beyond this range, go try the next one */
4298 /* <end> is generally 1 beyond where we want to set things, but at the
4299 * platform's infinity, where we can't go any higher, we want to
4300 * include the code point at <end> */
4303 : (max != UV_MAX || end != UV_MAX)
4310 if (!none || val < none) {
4315 for (key = min; key <= upper; key++) {
4317 /* offset must be non-negative (start <= min <= key < end) */
4318 offset = octets * (key - start);
4320 s[offset] = (U8)(val & 0xff);
4321 else if (bits == 16) {
4322 s[offset ] = (U8)((val >> 8) & 0xff);
4323 s[offset + 1] = (U8)( val & 0xff);
4325 else if (bits == 32) {
4326 s[offset ] = (U8)((val >> 24) & 0xff);
4327 s[offset + 1] = (U8)((val >> 16) & 0xff);
4328 s[offset + 2] = (U8)((val >> 8) & 0xff);
4329 s[offset + 3] = (U8)( val & 0xff);
4332 if (!none || val < none)
4336 else { /* bits == 1, then val should be ignored */
4341 for (key = min; key <= upper; key++) {
4342 const STRLEN offset = (STRLEN)(key - start);
4343 s[offset >> 3] |= 1 << (offset & 7);
4348 /* Invert if the data says it should be. Assumes that bits == 1 */
4349 if (invert_it_svp && SvUV(*invert_it_svp)) {
4351 /* Unicode properties should come with all bits above PERL_UNICODE_MAX
4352 * be 0, and their inversion should also be 0, as we don't succeed any
4353 * Unicode property matches for non-Unicode code points */
4354 if (start <= PERL_UNICODE_MAX) {
4356 /* The code below assumes that we never cross the
4357 * Unicode/above-Unicode boundary in a range, as otherwise we would
4358 * have to figure out where to stop flipping the bits. Since this
4359 * boundary is divisible by a large power of 2, and swatches comes
4360 * in small powers of 2, this should be a valid assumption */
4361 assert(start + span - 1 <= PERL_UNICODE_MAX);
4371 /* read $swash->{EXTRAS}
4372 * This code also copied to swash_to_invlist() below */
4373 x = (U8*)SvPV(*extssvp, xcur);
4381 SV **otherbitssvp, *other;
4385 const U8 opc = *x++;
4389 nl = (U8*)memchr(x, '\n', xend - x);
4391 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4393 x = nl + 1; /* 1 is length of "\n" */
4397 x = xend; /* to EXTRAS' end at which \n is not found */
4404 namelen = nl - namestr;
4408 namelen = xend - namestr;
4412 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
4413 otherhv = MUTABLE_HV(SvRV(*othersvp));
4414 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
4415 otherbits = (STRLEN)SvUV(*otherbitssvp);
4416 if (bits < otherbits)
4417 Perl_croak(aTHX_ "panic: swatch_get found swatch size mismatch, "
4418 "bits=%" UVuf ", otherbits=%" UVuf, (UV)bits, (UV)otherbits);
4420 /* The "other" swatch must be destroyed after. */
4421 other = swatch_get(*othersvp, start, span);
4422 o = (U8*)SvPV(other, olen);
4425 Perl_croak(aTHX_ "panic: swatch_get got improper swatch");
4427 s = (U8*)SvPV(swatch, slen);
4428 if (bits == 1 && otherbits == 1) {
4430 Perl_croak(aTHX_ "panic: swatch_get found swatch length "
4431 "mismatch, slen=%" UVuf ", olen=%" UVuf,
4432 (UV)slen, (UV)olen);
4456 STRLEN otheroctets = otherbits >> 3;
4458 U8* const send = s + slen;
4463 if (otherbits == 1) {
4464 otherval = (o[offset >> 3] >> (offset & 7)) & 1;
4468 STRLEN vlen = otheroctets;
4476 if (opc == '+' && otherval)
4477 NOOP; /* replace with otherval */
4478 else if (opc == '!' && !otherval)
4480 else if (opc == '-' && otherval)
4482 else if (opc == '&' && !otherval)
4485 s += octets; /* no replacement */
4490 *s++ = (U8)( otherval & 0xff);
4491 else if (bits == 16) {
4492 *s++ = (U8)((otherval >> 8) & 0xff);
4493 *s++ = (U8)( otherval & 0xff);
4495 else if (bits == 32) {
4496 *s++ = (U8)((otherval >> 24) & 0xff);
4497 *s++ = (U8)((otherval >> 16) & 0xff);
4498 *s++ = (U8)((otherval >> 8) & 0xff);
4499 *s++ = (U8)( otherval & 0xff);
4503 sv_free(other); /* through with it! */
4509 Perl__swash_inversion_hash(pTHX_ SV* const swash)
4512 /* Subject to change or removal. For use only in regcomp.c and regexec.c
4513 * Can't be used on a property that is subject to user override, as it
4514 * relies on the value of SPECIALS in the swash which would be set by
4515 * utf8_heavy.pl to the hash in the non-overriden file, and hence is not set
4516 * for overridden properties
4518 * Returns a hash which is the inversion and closure of a swash mapping.
4519 * For example, consider the input lines:
4524 * The returned hash would have two keys, the UTF-8 for 006B and the UTF-8 for
4525 * 006C. The value for each key is an array. For 006C, the array would
4526 * have two elements, the UTF-8 for itself, and for 004C. For 006B, there
4527 * would be three elements in its array, the UTF-8 for 006B, 004B and 212A.
4529 * Note that there are no elements in the hash for 004B, 004C, 212A. The
4530 * keys are only code points that are folded-to, so it isn't a full closure.
4532 * Essentially, for any code point, it gives all the code points that map to
4533 * it, or the list of 'froms' for that point.
4535 * Currently it ignores any additions or deletions from other swashes,
4536 * looking at just the main body of the swash, and if there are SPECIALS
4537 * in the swash, at that hash
4539 * The specials hash can be extra code points, and most likely consists of
4540 * maps from single code points to multiple ones (each expressed as a string
4541 * of UTF-8 characters). This function currently returns only 1-1 mappings.
4542 * However consider this possible input in the specials hash:
4543 * "\xEF\xAC\x85" => "\x{0073}\x{0074}", # U+FB05 => 0073 0074
4544 * "\xEF\xAC\x86" => "\x{0073}\x{0074}", # U+FB06 => 0073 0074
4546 * Both FB05 and FB06 map to the same multi-char sequence, which we don't
4547 * currently handle. But it also means that FB05 and FB06 are equivalent in
4548 * a 1-1 mapping which we should handle, and this relationship may not be in
4549 * the main table. Therefore this function examines all the multi-char
4550 * sequences and adds the 1-1 mappings that come out of that.
4552 * XXX This function was originally intended to be multipurpose, but its
4553 * only use is quite likely to remain for constructing the inversion of
4554 * the CaseFolding (//i) property. If it were more general purpose for
4555 * regex patterns, it would have to do the FB05/FB06 game for simple folds,
4556 * because certain folds are prohibited under /iaa and /il. As an example,
4557 * in Unicode 3.0.1 both U+0130 and U+0131 fold to 'i', and hence are both
4558 * equivalent under /i. But under /iaa and /il, the folds to 'i' are
4559 * prohibited, so we would not figure out that they fold to each other.
4560 * Code could be written to automatically figure this out, similar to the
4561 * code that does this for multi-character folds, but this is the only case
4562 * where something like this is ever likely to happen, as all the single
4563 * char folds to the 0-255 range are now quite settled. Instead there is a
4564 * little special code that is compiled only for this Unicode version. This
4565 * is smaller and didn't require much coding time to do. But this makes
4566 * this routine strongly tied to being used just for CaseFolding. If ever
4567 * it should be generalized, this would have to be fixed */
4571 HV *const hv = MUTABLE_HV(SvRV(swash));
4573 /* The string containing the main body of the table. This will have its
4574 * assertion fail if the swash has been converted to its inversion list */
4575 SV** const listsvp = hv_fetchs(hv, "LIST", FALSE);
4577 SV** const typesvp = hv_fetchs(hv, "TYPE", FALSE);
4578 SV** const bitssvp = hv_fetchs(hv, "BITS", FALSE);
4579 SV** const nonesvp = hv_fetchs(hv, "NONE", FALSE);
4580 /*SV** const extssvp = hv_fetchs(hv, "EXTRAS", FALSE);*/
4581 const U8* const typestr = (U8*)SvPV_nolen(*typesvp);
4582 const STRLEN bits = SvUV(*bitssvp);
4583 const STRLEN octets = bits >> 3; /* if bits == 1, then octets == 0 */
4584 const UV none = SvUV(*nonesvp);
4585 SV **specials_p = hv_fetchs(hv, "SPECIALS", 0);
4589 PERL_ARGS_ASSERT__SWASH_INVERSION_HASH;
4591 /* Must have at least 8 bits to get the mappings */
4592 if (bits != 8 && bits != 16 && bits != 32) {
4593 Perl_croak(aTHX_ "panic: swash_inversion_hash doesn't expect bits %"
4597 if (specials_p) { /* It might be "special" (sometimes, but not always, a
4598 mapping to more than one character */
4600 /* Construct an inverse mapping hash for the specials */
4601 HV * const specials_hv = MUTABLE_HV(SvRV(*specials_p));
4602 HV * specials_inverse = newHV();
4603 char *char_from; /* the lhs of the map */
4604 I32 from_len; /* its byte length */
4605 char *char_to; /* the rhs of the map */
4606 I32 to_len; /* its byte length */
4607 SV *sv_to; /* and in a sv */
4608 AV* from_list; /* list of things that map to each 'to' */
4610 hv_iterinit(specials_hv);
4612 /* The keys are the characters (in UTF-8) that map to the corresponding
4613 * UTF-8 string value. Iterate through the list creating the inverse
4615 while ((sv_to = hv_iternextsv(specials_hv, &char_from, &from_len))) {
4617 if (! SvPOK(sv_to)) {
4618 Perl_croak(aTHX_ "panic: value returned from hv_iternextsv() "
4619 "unexpectedly is not a string, flags=%lu",
4620 (unsigned long)SvFLAGS(sv_to));
4622 /*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)));*/
4624 /* Each key in the inverse list is a mapped-to value, and the key's
4625 * hash value is a list of the strings (each in UTF-8) that map to
4626 * it. Those strings are all one character long */
4627 if ((listp = hv_fetch(specials_inverse,
4631 from_list = (AV*) *listp;
4633 else { /* No entry yet for it: create one */
4634 from_list = newAV();
4635 if (! hv_store(specials_inverse,
4638 (SV*) from_list, 0))
4640 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4644 /* Here have the list associated with this 'to' (perhaps newly
4645 * created and empty). Just add to it. Note that we ASSUME that
4646 * the input is guaranteed to not have duplications, so we don't
4647 * check for that. Duplications just slow down execution time. */
4648 av_push(from_list, newSVpvn_utf8(char_from, from_len, TRUE));
4651 /* Here, 'specials_inverse' contains the inverse mapping. Go through
4652 * it looking for cases like the FB05/FB06 examples above. There would
4653 * be an entry in the hash like
4654 * 'st' => [ FB05, FB06 ]
4655 * In this example we will create two lists that get stored in the
4656 * returned hash, 'ret':
4657 * FB05 => [ FB05, FB06 ]
4658 * FB06 => [ FB05, FB06 ]
4660 * Note that there is nothing to do if the array only has one element.
4661 * (In the normal 1-1 case handled below, we don't have to worry about
4662 * two lists, as everything gets tied to the single list that is
4663 * generated for the single character 'to'. But here, we are omitting
4664 * that list, ('st' in the example), so must have multiple lists.) */
4665 while ((from_list = (AV *) hv_iternextsv(specials_inverse,
4666 &char_to, &to_len)))
4668 if (av_tindex_skip_len_mg(from_list) > 0) {
4671 /* We iterate over all combinations of i,j to place each code
4672 * point on each list */
4673 for (i = 0; i <= av_tindex_skip_len_mg(from_list); i++) {
4675 AV* i_list = newAV();
4676 SV** entryp = av_fetch(from_list, i, FALSE);
4677 if (entryp == NULL) {
4678 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly"
4681 if (hv_fetch(ret, SvPVX(*entryp), SvCUR(*entryp), FALSE)) {
4682 Perl_croak(aTHX_ "panic: unexpected entry for %s",
4685 if (! hv_store(ret, SvPVX(*entryp), SvCUR(*entryp),
4686 (SV*) i_list, FALSE))
4688 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4691 /* For DEBUG_U: UV u = valid_utf8_to_uvchr((U8*) SvPVX(*entryp), 0);*/
4692 for (j = 0; j <= av_tindex_skip_len_mg(from_list); j++) {
4693 entryp = av_fetch(from_list, j, FALSE);
4694 if (entryp == NULL) {
4695 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4698 /* When i==j this adds itself to the list */
4699 av_push(i_list, newSVuv(utf8_to_uvchr_buf(
4700 (U8*) SvPVX(*entryp),
4701 (U8*) SvPVX(*entryp) + SvCUR(*entryp),
4703 /*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));*/
4708 SvREFCNT_dec(specials_inverse); /* done with it */
4709 } /* End of specials */
4711 /* read $swash->{LIST} */
4713 #if UNICODE_MAJOR_VERSION == 3 \
4714 && UNICODE_DOT_VERSION == 0 \
4715 && UNICODE_DOT_DOT_VERSION == 1
4717 /* For this version only U+130 and U+131 are equivalent under qr//i. Add a
4718 * rule so that things work under /iaa and /il */
4720 SV * mod_listsv = sv_mortalcopy(*listsvp);
4721 sv_catpv(mod_listsv, "130\t130\t131\n");
4722 l = (U8*)SvPV(mod_listsv, lcur);
4726 l = (U8*)SvPV(*listsvp, lcur);
4732 /* Go through each input line */
4736 l = swash_scan_list_line(l, lend, &min, &max, &val,
4737 cBOOL(octets), typestr);
4742 /* Each element in the range is to be inverted */
4743 for (inverse = min; inverse <= max; inverse++) {
4747 bool found_key = FALSE;
4748 bool found_inverse = FALSE;
4750 /* The key is the inverse mapping */
4751 char key[UTF8_MAXBYTES+1];
4752 char* key_end = (char *) uvchr_to_utf8((U8*) key, val);
4753 STRLEN key_len = key_end - key;
4755 /* Get the list for the map */
4756 if ((listp = hv_fetch(ret, key, key_len, FALSE))) {
4757 list = (AV*) *listp;
4759 else { /* No entry yet for it: create one */
4761 if (! hv_store(ret, key, key_len, (SV*) list, FALSE)) {
4762 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4766 /* Look through list to see if this inverse mapping already is
4767 * listed, or if there is a mapping to itself already */
4768 for (i = 0; i <= av_tindex_skip_len_mg(list); i++) {
4769 SV** entryp = av_fetch(list, i, FALSE);
4772 if (entryp == NULL) {
4773 Perl_croak(aTHX_ "panic: av_fetch() unexpectedly failed");
4777 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "list for %" UVXf " contains %" UVXf "\n", val, uv));*/
4781 if (uv == inverse) {
4782 found_inverse = TRUE;
4785 /* No need to continue searching if found everything we are
4787 if (found_key && found_inverse) {
4792 /* Make sure there is a mapping to itself on the list */
4794 av_push(list, newSVuv(val));
4795 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, val, val));*/
4799 /* Simply add the value to the list */
4800 if (! found_inverse) {
4801 av_push(list, newSVuv(inverse));
4802 /*DEBUG_U(PerlIO_printf(Perl_debug_log, "%s: %d: Adding %" UVXf " to list for %" UVXf "\n", __FILE__, __LINE__, inverse, val));*/
4805 /* swatch_get() increments the value of val for each element in the
4806 * range. That makes more compact tables possible. You can
4807 * express the capitalization, for example, of all consecutive
4808 * letters with a single line: 0061\t007A\t0041 This maps 0061 to
4809 * 0041, 0062 to 0042, etc. I (khw) have never understood 'none',
4810 * and it's not documented; it appears to be used only in
4811 * implementing tr//; I copied the semantics from swatch_get(), just
4813 if (!none || val < none) {
4823 Perl__swash_to_invlist(pTHX_ SV* const swash)
4826 /* Subject to change or removal. For use only in one place in regcomp.c.
4827 * Ownership is given to one reference count in the returned SV* */
4832 HV *const hv = MUTABLE_HV(SvRV(swash));
4833 UV elements = 0; /* Number of elements in the inversion list */
4843 STRLEN octets; /* if bits == 1, then octets == 0 */
4849 PERL_ARGS_ASSERT__SWASH_TO_INVLIST;
4851 /* If not a hash, it must be the swash's inversion list instead */
4852 if (SvTYPE(hv) != SVt_PVHV) {
4853 return SvREFCNT_inc_simple_NN((SV*) hv);
4856 /* The string containing the main body of the table */
4857 listsvp = hv_fetchs(hv, "LIST", FALSE);
4858 typesvp = hv_fetchs(hv, "TYPE", FALSE);
4859 bitssvp = hv_fetchs(hv, "BITS", FALSE);
4860 extssvp = hv_fetchs(hv, "EXTRAS", FALSE);
4861 invert_it_svp = hv_fetchs(hv, "INVERT_IT", FALSE);
4863 typestr = (U8*)SvPV_nolen(*typesvp);
4864 bits = SvUV(*bitssvp);
4865 octets = bits >> 3; /* if bits == 1, then octets == 0 */
4867 /* read $swash->{LIST} */
4868 if (SvPOK(*listsvp)) {
4869 l = (U8*)SvPV(*listsvp, lcur);
4872 /* LIST legitimately doesn't contain a string during compilation phases
4873 * of Perl itself, before the Unicode tables are generated. In this
4874 * case, just fake things up by creating an empty list */
4881 if (*l == 'V') { /* Inversion list format */
4882 const char *after_atou = (char *) lend;
4884 UV* other_elements_ptr;
4886 /* The first number is a count of the rest */
4888 if (!grok_atoUV((const char *)l, &elements, &after_atou)) {
4889 Perl_croak(aTHX_ "panic: Expecting a valid count of elements"
4890 " at start of inversion list");
4892 if (elements == 0) {
4893 invlist = _new_invlist(0);
4896 l = (U8 *) after_atou;
4898 /* Get the 0th element, which is needed to setup the inversion list
4900 while (isSPACE(*l)) l++;
4901 if (!grok_atoUV((const char *)l, &element0, &after_atou)) {
4902 Perl_croak(aTHX_ "panic: Expecting a valid 0th element for"
4905 l = (U8 *) after_atou;
4906 invlist = _setup_canned_invlist(elements, element0,
4907 &other_elements_ptr);
4910 /* Then just populate the rest of the input */
4911 while (elements-- > 0) {
4913 Perl_croak(aTHX_ "panic: Expecting %" UVuf " more"
4914 " elements than available", elements);
4916 while (isSPACE(*l)) l++;
4917 if (!grok_atoUV((const char *)l, other_elements_ptr++,
4920 Perl_croak(aTHX_ "panic: Expecting a valid element"
4921 " in inversion list");
4923 l = (U8 *) after_atou;
4929 /* Scan the input to count the number of lines to preallocate array
4930 * size based on worst possible case, which is each line in the input
4931 * creates 2 elements in the inversion list: 1) the beginning of a
4932 * range in the list; 2) the beginning of a range not in the list. */
4933 while ((loc = (strchr(loc, '\n'))) != NULL) {
4938 /* If the ending is somehow corrupt and isn't a new line, add another
4939 * element for the final range that isn't in the inversion list */
4940 if (! (*lend == '\n'
4941 || (*lend == '\0' && (lcur == 0 || *(lend - 1) == '\n'))))
4946 invlist = _new_invlist(elements);
4948 /* Now go through the input again, adding each range to the list */
4951 UV val; /* Not used by this function */
4953 l = swash_scan_list_line(l, lend, &start, &end, &val,
4954 cBOOL(octets), typestr);
4960 invlist = _add_range_to_invlist(invlist, start, end);
4964 /* Invert if the data says it should be */
4965 if (invert_it_svp && SvUV(*invert_it_svp)) {
4966 _invlist_invert(invlist);
4969 /* This code is copied from swatch_get()
4970 * read $swash->{EXTRAS} */
4971 x = (U8*)SvPV(*extssvp, xcur);
4979 SV **otherbitssvp, *other;
4982 const U8 opc = *x++;
4986 nl = (U8*)memchr(x, '\n', xend - x);
4988 if (opc != '-' && opc != '+' && opc != '!' && opc != '&') {
4990 x = nl + 1; /* 1 is length of "\n" */
4994 x = xend; /* to EXTRAS' end at which \n is not found */
5001 namelen = nl - namestr;
5005 namelen = xend - namestr;
5009 othersvp = hv_fetch(hv, (char *)namestr, namelen, FALSE);
5010 otherhv = MUTABLE_HV(SvRV(*othersvp));
5011 otherbitssvp = hv_fetchs(otherhv, "BITS", FALSE);
5012 otherbits = (STRLEN)SvUV(*otherbitssvp);
5014 if (bits != otherbits || bits != 1) {
5015 Perl_croak(aTHX_ "panic: _swash_to_invlist only operates on boolean "
5016 "properties, bits=%" UVuf ", otherbits=%" UVuf,
5017 (UV)bits, (UV)otherbits);
5020 /* The "other" swatch must be destroyed after. */
5021 other = _swash_to_invlist((SV *)*othersvp);
5023 /* End of code copied from swatch_get() */
5026 _invlist_union(invlist, other, &invlist);
5029 _invlist_union_maybe_complement_2nd(invlist, other, TRUE, &invlist);
5032 _invlist_subtract(invlist, other, &invlist);
5035 _invlist_intersection(invlist, other, &invlist);
5040 sv_free(other); /* through with it! */
5043 SvREADONLY_on(invlist);
5048 Perl__get_swash_invlist(pTHX_ SV* const swash)
5052 PERL_ARGS_ASSERT__GET_SWASH_INVLIST;
5054 if (! SvROK(swash)) {
5058 /* If it really isn't a hash, it isn't really swash; must be an inversion
5060 if (SvTYPE(SvRV(swash)) != SVt_PVHV) {
5064 ptr = hv_fetchs(MUTABLE_HV(SvRV(swash)), "V", FALSE);
5073 Perl_check_utf8_print(pTHX_ const U8* s, const STRLEN len)
5075 /* May change: warns if surrogates, non-character code points, or
5076 * non-Unicode code points are in 's' which has length 'len' bytes.
5077 * Returns TRUE if none found; FALSE otherwise. The only other validity
5078 * check is to make sure that this won't exceed the string's length.
5080 * Code points above the platform's C<IV_MAX> will raise a deprecation
5081 * warning, unless those are turned off. */
5083 const U8* const e = s + len;
5086 PERL_ARGS_ASSERT_CHECK_UTF8_PRINT;
5089 if (UTF8SKIP(s) > len) {
5090 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
5091 "%s in %s", unees, PL_op ? OP_DESC(PL_op) : "print");
5094 if (UNLIKELY(isUTF8_POSSIBLY_PROBLEMATIC(*s))) {
5095 if (UNLIKELY(UTF8_IS_SUPER(s, e))) {
5096 if ( ckWARN_d(WARN_NON_UNICODE)
5097 || ( ckWARN_d(WARN_DEPRECATED)
5099 && UNLIKELY(is_utf8_cp_above_31_bits(s, e))
5100 #else /* Below is 64-bit words */
5101 /* 2**63 and up meet these conditions provided we have
5105 && NATIVE_UTF8_TO_I8(s[1]) >= 0xA8
5108 /* s[1] being above 0x80 overflows */
5113 /* A side effect of this function will be to warn */
5114 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_SUPER);
5118 else if (UNLIKELY(UTF8_IS_SURROGATE(s, e))) {
5119 if (ckWARN_d(WARN_SURROGATE)) {
5120 /* This has a different warning than the one the called
5121 * function would output, so can't just call it, unlike we
5122 * do for the non-chars and above-unicodes */
5123 UV uv = utf8_to_uvchr_buf(s, e, NULL);
5124 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
5125 "Unicode surrogate U+%04" UVXf " is illegal in UTF-8",
5130 else if ( UNLIKELY(UTF8_IS_NONCHAR(s, e))
5131 && (ckWARN_d(WARN_NONCHAR)))
5133 /* A side effect of this function will be to warn */
5134 (void) utf8n_to_uvchr(s, e - s, NULL, UTF8_WARN_NONCHAR);
5145 =for apidoc pv_uni_display
5147 Build to the scalar C<dsv> a displayable version of the string C<spv>,
5148 length C<len>, the displayable version being at most C<pvlim> bytes long
5149 (if longer, the rest is truncated and C<"..."> will be appended).
5151 The C<flags> argument can have C<UNI_DISPLAY_ISPRINT> set to display
5152 C<isPRINT()>able characters as themselves, C<UNI_DISPLAY_BACKSLASH>
5153 to display the C<\\[nrfta\\]> as the backslashed versions (like C<"\n">)
5154 (C<UNI_DISPLAY_BACKSLASH> is preferred over C<UNI_DISPLAY_ISPRINT> for C<"\\">).
5155 C<UNI_DISPLAY_QQ> (and its alias C<UNI_DISPLAY_REGEX>) have both
5156 C<UNI_DISPLAY_BACKSLASH> and C<UNI_DISPLAY_ISPRINT> turned on.
5158 The pointer to the PV of the C<dsv> is returned.
5160 See also L</sv_uni_display>.
5164 Perl_pv_uni_display(pTHX_ SV *dsv, const U8 *spv, STRLEN len, STRLEN pvlim,
5170 PERL_ARGS_ASSERT_PV_UNI_DISPLAY;
5174 for (s = (const char *)spv, e = s + len; s < e; s += UTF8SKIP(s)) {
5176 /* This serves double duty as a flag and a character to print after
5177 a \ when flags & UNI_DISPLAY_BACKSLASH is true.
5181 if (pvlim && SvCUR(dsv) >= pvlim) {
5185 u = utf8_to_uvchr_buf((U8*)s, (U8*)e, 0);
5187 const unsigned char c = (unsigned char)u & 0xFF;
5188 if (flags & UNI_DISPLAY_BACKSLASH) {
5205 const char string = ok;
5206 sv_catpvs(dsv, "\\");
5207 sv_catpvn(dsv, &string, 1);
5210 /* isPRINT() is the locale-blind version. */
5211 if (!ok && (flags & UNI_DISPLAY_ISPRINT) && isPRINT(c)) {
5212 const char string = c;
5213 sv_catpvn(dsv, &string, 1);
5218 Perl_sv_catpvf(aTHX_ dsv, "\\x{%" UVxf "}", u);
5221 sv_catpvs(dsv, "...");
5227 =for apidoc sv_uni_display
5229 Build to the scalar C<dsv> a displayable version of the scalar C<sv>,
5230 the displayable version being at most C<pvlim> bytes long
5231 (if longer, the rest is truncated and "..." will be appended).
5233 The C<flags> argument is as in L</pv_uni_display>().
5235 The pointer to the PV of the C<dsv> is returned.
5240 Perl_sv_uni_display(pTHX_ SV *dsv, SV *ssv, STRLEN pvlim, UV flags)
5242 const char * const ptr =
5243 isREGEXP(ssv) ? RX_WRAPPED((REGEXP*)ssv) : SvPVX_const(ssv);
5245 PERL_ARGS_ASSERT_SV_UNI_DISPLAY;
5247 return Perl_pv_uni_display(aTHX_ dsv, (const U8*)ptr,
5248 SvCUR(ssv), pvlim, flags);
5252 =for apidoc foldEQ_utf8
5254 Returns true if the leading portions of the strings C<s1> and C<s2> (either or
5255 both of which may be in UTF-8) are the same case-insensitively; false
5256 otherwise. How far into the strings to compare is determined by other input
5259 If C<u1> is true, the string C<s1> is assumed to be in UTF-8-encoded Unicode;
5260 otherwise it is assumed to be in native 8-bit encoding. Correspondingly for
5261 C<u2> with respect to C<s2>.
5263 If the byte length C<l1> is non-zero, it says how far into C<s1> to check for
5264 fold equality. In other words, C<s1>+C<l1> will be used as a goal to reach.
5265 The scan will not be considered to be a match unless the goal is reached, and
5266 scanning won't continue past that goal. Correspondingly for C<l2> with respect
5269 If C<pe1> is non-C<NULL> and the pointer it points to is not C<NULL>, that
5270 pointer is considered an end pointer to the position 1 byte past the maximum
5271 point in C<s1> beyond which scanning will not continue under any circumstances.
5272 (This routine assumes that UTF-8 encoded input strings are not malformed;
5273 malformed input can cause it to read past C<pe1>). This means that if both
5274 C<l1> and C<pe1> are specified, and C<pe1> is less than C<s1>+C<l1>, the match
5275 will never be successful because it can never
5276 get as far as its goal (and in fact is asserted against). Correspondingly for
5277 C<pe2> with respect to C<s2>.
5279 At least one of C<s1> and C<s2> must have a goal (at least one of C<l1> and
5280 C<l2> must be non-zero), and if both do, both have to be
5281 reached for a successful match. Also, if the fold of a character is multiple
5282 characters, all of them must be matched (see tr21 reference below for
5285 Upon a successful match, if C<pe1> is non-C<NULL>,
5286 it will be set to point to the beginning of the I<next> character of C<s1>
5287 beyond what was matched. Correspondingly for C<pe2> and C<s2>.
5289 For case-insensitiveness, the "casefolding" of Unicode is used
5290 instead of upper/lowercasing both the characters, see
5291 L<http://www.unicode.org/unicode/reports/tr21/> (Case Mappings).
5295 /* A flags parameter has been added which may change, and hence isn't
5296 * externally documented. Currently it is:
5297 * 0 for as-documented above
5298 * FOLDEQ_UTF8_NOMIX_ASCII meaning that if a non-ASCII character folds to an
5299 ASCII one, to not match
5300 * FOLDEQ_LOCALE is set iff the rules from the current underlying
5301 * locale are to be used.
5302 * FOLDEQ_S1_ALREADY_FOLDED s1 has already been folded before calling this
5303 * routine. This allows that step to be skipped.
5304 * Currently, this requires s1 to be encoded as UTF-8
5305 * (u1 must be true), which is asserted for.
5306 * FOLDEQ_S1_FOLDS_SANE With either NOMIX_ASCII or LOCALE, no folds may
5307 * cross certain boundaries. Hence, the caller should
5308 * let this function do the folding instead of
5309 * pre-folding. This code contains an assertion to
5310 * that effect. However, if the caller knows what
5311 * it's doing, it can pass this flag to indicate that,
5312 * and the assertion is skipped.
5313 * FOLDEQ_S2_ALREADY_FOLDED Similarly.
5314 * FOLDEQ_S2_FOLDS_SANE
5317 Perl_foldEQ_utf8_flags(pTHX_ const char *s1, char **pe1, UV l1, bool u1,
5318 const char *s2, char **pe2, UV l2, bool u2,
5321 const U8 *p1 = (const U8*)s1; /* Point to current char */
5322 const U8 *p2 = (const U8*)s2;
5323 const U8 *g1 = NULL; /* goal for s1 */
5324 const U8 *g2 = NULL;
5325 const U8 *e1 = NULL; /* Don't scan s1 past this */
5326 U8 *f1 = NULL; /* Point to current folded */
5327 const U8 *e2 = NULL;
5329 STRLEN n1 = 0, n2 = 0; /* Number of bytes in current char */
5330 U8 foldbuf1[UTF8_MAXBYTES_CASE+1];
5331 U8 foldbuf2[UTF8_MAXBYTES_CASE+1];
5332 U8 flags_for_folder = FOLD_FLAGS_FULL;
5334 PERL_ARGS_ASSERT_FOLDEQ_UTF8_FLAGS;
5336 assert( ! ((flags & (FOLDEQ_UTF8_NOMIX_ASCII | FOLDEQ_LOCALE))
5337 && (((flags & FOLDEQ_S1_ALREADY_FOLDED)
5338 && !(flags & FOLDEQ_S1_FOLDS_SANE))
5339 || ((flags & FOLDEQ_S2_ALREADY_FOLDED)
5340 && !(flags & FOLDEQ_S2_FOLDS_SANE)))));
5341 /* The algorithm is to trial the folds without regard to the flags on
5342 * the first line of the above assert(), and then see if the result
5343 * violates them. This means that the inputs can't be pre-folded to a
5344 * violating result, hence the assert. This could be changed, with the
5345 * addition of extra tests here for the already-folded case, which would
5346 * slow it down. That cost is more than any possible gain for when these
5347 * flags are specified, as the flags indicate /il or /iaa matching which
5348 * is less common than /iu, and I (khw) also believe that real-world /il
5349 * and /iaa matches are most likely to involve code points 0-255, and this
5350 * function only under rare conditions gets called for 0-255. */
5352 if (flags & FOLDEQ_LOCALE) {
5353 if (IN_UTF8_CTYPE_LOCALE) {
5354 flags &= ~FOLDEQ_LOCALE;
5357 flags_for_folder |= FOLD_FLAGS_LOCALE;
5366 g1 = (const U8*)s1 + l1;
5374 g2 = (const U8*)s2 + l2;
5377 /* Must have at least one goal */
5382 /* Will never match if goal is out-of-bounds */
5383 assert(! e1 || e1 >= g1);
5385 /* Here, there isn't an end pointer, or it is beyond the goal. We
5386 * only go as far as the goal */
5390 assert(e1); /* Must have an end for looking at s1 */
5393 /* Same for goal for s2 */
5395 assert(! e2 || e2 >= g2);
5402 /* If both operands are already folded, we could just do a memEQ on the
5403 * whole strings at once, but it would be better if the caller realized
5404 * this and didn't even call us */
5406 /* Look through both strings, a character at a time */
5407 while (p1 < e1 && p2 < e2) {
5409 /* If at the beginning of a new character in s1, get its fold to use
5410 * and the length of the fold. */
5412 if (flags & FOLDEQ_S1_ALREADY_FOLDED) {
5418 if (isASCII(*p1) && ! (flags & FOLDEQ_LOCALE)) {
5420 /* We have to forbid mixing ASCII with non-ASCII if the
5421 * flags so indicate. And, we can short circuit having to
5422 * call the general functions for this common ASCII case,
5423 * all of whose non-locale folds are also ASCII, and hence
5424 * UTF-8 invariants, so the UTF8ness of the strings is not
5426 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p2)) {
5430 *foldbuf1 = toFOLD(*p1);
5433 _toFOLD_utf8_flags(p1, e1, foldbuf1, &n1, flags_for_folder);
5435 else { /* Not UTF-8, get UTF-8 fold */
5436 _to_uni_fold_flags(*p1, foldbuf1, &n1, flags_for_folder);
5442 if (n2 == 0) { /* Same for s2 */
5443 if (flags & FOLDEQ_S2_ALREADY_FOLDED) {
5449 if (isASCII(*p2) && ! (flags & FOLDEQ_LOCALE)) {
5450 if ((flags & FOLDEQ_UTF8_NOMIX_ASCII) && ! isASCII(*p1)) {
5454 *foldbuf2 = toFOLD(*p2);
5457 _toFOLD_utf8_flags(p2, e2, foldbuf2, &n2, flags_for_folder);
5460 _to_uni_fold_flags(*p2, foldbuf2, &n2, flags_for_folder);
5466 /* Here f1 and f2 point to the beginning of the strings to compare.
5467 * These strings are the folds of the next character from each input
5468 * string, stored in UTF-8. */
5470 /* While there is more to look for in both folds, see if they
5471 * continue to match */
5473 U8 fold_length = UTF8SKIP(f1);
5474 if (fold_length != UTF8SKIP(f2)
5475 || (fold_length == 1 && *f1 != *f2) /* Short circuit memNE
5476 function call for single
5478 || memNE((char*)f1, (char*)f2, fold_length))
5480 return 0; /* mismatch */
5483 /* Here, they matched, advance past them */
5490 /* When reach the end of any fold, advance the input past it */
5492 p1 += u1 ? UTF8SKIP(p1) : 1;
5495 p2 += u2 ? UTF8SKIP(p2) : 1;
5497 } /* End of loop through both strings */
5499 /* A match is defined by each scan that specified an explicit length
5500 * reaching its final goal, and the other not having matched a partial
5501 * character (which can happen when the fold of a character is more than one
5503 if (! ((g1 == 0 || p1 == g1) && (g2 == 0 || p2 == g2)) || n1 || n2) {
5507 /* Successful match. Set output pointers */
5517 /* XXX The next two functions should likely be moved to mathoms.c once all
5518 * occurrences of them are removed from the core; some cpan-upstream modules
5522 Perl_uvuni_to_utf8(pTHX_ U8 *d, UV uv)
5524 PERL_ARGS_ASSERT_UVUNI_TO_UTF8;
5526 return Perl_uvoffuni_to_utf8_flags(aTHX_ d, uv, 0);
5530 =for apidoc utf8n_to_uvuni
5532 Instead use L</utf8_to_uvchr_buf>, or rarely, L</utf8n_to_uvchr>.
5534 This function was useful for code that wanted to handle both EBCDIC and
5535 ASCII platforms with Unicode properties, but starting in Perl v5.20, the
5536 distinctions between the platforms have mostly been made invisible to most
5537 code, so this function is quite unlikely to be what you want. If you do need
5538 this precise functionality, use instead
5539 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>>
5540 or C<L<NATIVE_TO_UNI(utf8n_to_uvchr(...))|/utf8n_to_uvchr>>.
5546 Perl_utf8n_to_uvuni(pTHX_ const U8 *s, STRLEN curlen, STRLEN *retlen, U32 flags)
5548 PERL_ARGS_ASSERT_UTF8N_TO_UVUNI;
5550 return NATIVE_TO_UNI(utf8n_to_uvchr(s, curlen, retlen, flags));
5554 =for apidoc uvuni_to_utf8_flags
5556 Instead you almost certainly want to use L</uvchr_to_utf8> or
5557 L</uvchr_to_utf8_flags>.
5559 This function is a deprecated synonym for L</uvoffuni_to_utf8_flags>,
5560 which itself, while not deprecated, should be used only in isolated
5561 circumstances. These functions were useful for code that wanted to handle
5562 both EBCDIC and ASCII platforms with Unicode properties, but starting in Perl
5563 v5.20, the distinctions between the platforms have mostly been made invisible
5564 to most code, so this function is quite unlikely to be what you want.
5570 Perl_uvuni_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
5572 PERL_ARGS_ASSERT_UVUNI_TO_UTF8_FLAGS;
5574 return uvoffuni_to_utf8_flags(d, uv, flags);
5578 * ex: set ts=8 sts=4 sw=4 et: