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 not allowed; the"
41 " permissible max is 0x%" UVXf;
43 #define MAX_EXTERNALLY_LEGAL_CP ((UV) (IV_MAX))
46 =head1 Unicode Support
47 These are various utility functions for manipulating UTF8-encoded
48 strings. For the uninitiated, this is a method of representing arbitrary
49 Unicode characters as a variable number of bytes, in such a way that
50 characters in the ASCII range are unmodified, and a zero byte never appears
51 within non-zero characters.
57 Perl__force_out_malformed_utf8_message(pTHX_
58 const U8 *const p, /* First byte in UTF-8 sequence */
59 const U8 * const e, /* Final byte in sequence (may include
61 const U32 flags, /* Flags to pass to utf8n_to_uvchr(),
62 usually 0, or some DISALLOW flags */
63 const bool die_here) /* If TRUE, this function does not return */
65 /* This core-only function is to be called when a malformed UTF-8 character
66 * is found, in order to output the detailed information about the
67 * malformation before dieing. The reason it exists is for the occasions
68 * when such a malformation is fatal, but warnings might be turned off, so
69 * that normally they would not be actually output. This ensures that they
70 * do get output. Because a sequence may be malformed in more than one
71 * way, multiple messages may be generated, so we can't make them fatal, as
72 * that would cause the first one to die.
74 * Instead we pretend -W was passed to perl, then die afterwards. The
75 * flexibility is here to return to the caller so they can finish up and
79 PERL_ARGS_ASSERT__FORCE_OUT_MALFORMED_UTF8_MESSAGE;
85 PL_dowarn = G_WARN_ALL_ON|G_WARN_ON;
87 PL_curcop->cop_warnings = pWARN_ALL;
90 (void) utf8n_to_uvchr_error(p, e - p, NULL, flags & ~UTF8_CHECK_ONLY, &errors);
95 Perl_croak(aTHX_ "panic: _force_out_malformed_utf8_message should"
96 " be called only when there are errors found");
100 Perl_croak(aTHX_ "Malformed UTF-8 character (fatal)");
105 =for apidoc uvoffuni_to_utf8_flags
107 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
108 Instead, B<Almost all code should use L</uvchr_to_utf8> or
109 L</uvchr_to_utf8_flags>>.
111 This function is like them, but the input is a strict Unicode
112 (as opposed to native) code point. Only in very rare circumstances should code
113 not be using the native code point.
115 For details, see the description for L</uvchr_to_utf8_flags>.
120 /* All these formats take a single UV code point argument */
121 const char surrogate_cp_format[] = "UTF-16 surrogate U+%04" UVXf;
122 const char nonchar_cp_format[] = "Unicode non-character U+%04" UVXf
123 " is not recommended for open interchange";
124 const char super_cp_format[] = "Code point 0x%" UVXf " is not Unicode,"
125 " may not be portable";
126 const char perl_extended_cp_format[] = "Code point 0x%" UVXf " is not" \
127 " Unicode, requires a Perl extension," \
128 " and so is not portable";
130 #define HANDLE_UNICODE_SURROGATE(uv, flags) \
132 if (flags & UNICODE_WARN_SURROGATE) { \
133 Perl_ck_warner_d(aTHX_ packWARN(WARN_SURROGATE), \
134 surrogate_cp_format, uv); \
136 if (flags & UNICODE_DISALLOW_SURROGATE) { \
141 #define HANDLE_UNICODE_NONCHAR(uv, flags) \
143 if (flags & UNICODE_WARN_NONCHAR) { \
144 Perl_ck_warner_d(aTHX_ packWARN(WARN_NONCHAR), \
145 nonchar_cp_format, uv); \
147 if (flags & UNICODE_DISALLOW_NONCHAR) { \
152 /* Use shorter names internally in this file */
153 #define SHIFT UTF_ACCUMULATION_SHIFT
155 #define MARK UTF_CONTINUATION_MARK
156 #define MASK UTF_CONTINUATION_MASK
159 Perl_uvoffuni_to_utf8_flags(pTHX_ U8 *d, UV uv, const UV flags)
161 PERL_ARGS_ASSERT_UVOFFUNI_TO_UTF8_FLAGS;
163 if (OFFUNI_IS_INVARIANT(uv)) {
164 *d++ = LATIN1_TO_NATIVE(uv);
168 if (uv <= MAX_UTF8_TWO_BYTE) {
169 *d++ = I8_TO_NATIVE_UTF8(( uv >> SHIFT) | UTF_START_MARK(2));
170 *d++ = I8_TO_NATIVE_UTF8(( uv & MASK) | MARK);
174 /* Not 2-byte; test for and handle 3-byte result. In the test immediately
175 * below, the 16 is for start bytes E0-EF (which are all the possible ones
176 * for 3 byte characters). The 2 is for 2 continuation bytes; these each
177 * contribute SHIFT bits. This yields 0x4000 on EBCDIC platforms, 0x1_0000
178 * on ASCII; so 3 bytes covers the range 0x400-0x3FFF on EBCDIC;
179 * 0x800-0xFFFF on ASCII */
180 if (uv < (16 * (1U << (2 * SHIFT)))) {
181 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((3 - 1) * SHIFT)) | UTF_START_MARK(3));
182 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
183 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
185 #ifndef EBCDIC /* These problematic code points are 4 bytes on EBCDIC, so
186 aren't tested here */
187 /* The most likely code points in this range are below the surrogates.
188 * Do an extra test to quickly exclude those. */
189 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST)) {
190 if (UNLIKELY( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)
191 || UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
193 HANDLE_UNICODE_NONCHAR(uv, flags);
195 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
196 HANDLE_UNICODE_SURROGATE(uv, flags);
203 /* Not 3-byte; that means the code point is at least 0x1_0000 on ASCII
204 * platforms, and 0x4000 on EBCDIC. There are problematic cases that can
205 * happen starting with 4-byte characters on ASCII platforms. We unify the
206 * code for these with EBCDIC, even though some of them require 5-bytes on
207 * those, because khw believes the code saving is worth the very slight
208 * performance hit on these high EBCDIC code points. */
210 if (UNLIKELY(UNICODE_IS_SUPER(uv))) {
211 if (UNLIKELY(uv > MAX_EXTERNALLY_LEGAL_CP)) {
212 Perl_croak(aTHX_ cp_above_legal_max, uv, MAX_EXTERNALLY_LEGAL_CP);
214 if ( (flags & UNICODE_WARN_SUPER)
215 || ( (flags & UNICODE_WARN_PERL_EXTENDED)
216 && UNICODE_IS_PERL_EXTENDED(uv)))
218 Perl_ck_warner_d(aTHX_ packWARN(WARN_NON_UNICODE),
220 /* Choose the more dire applicable warning */
221 (UNICODE_IS_PERL_EXTENDED(uv))
222 ? perl_extended_cp_format
226 if ( (flags & UNICODE_DISALLOW_SUPER)
227 || ( (flags & UNICODE_DISALLOW_PERL_EXTENDED)
228 && UNICODE_IS_PERL_EXTENDED(uv)))
233 else if (UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv))) {
234 HANDLE_UNICODE_NONCHAR(uv, flags);
237 /* Test for and handle 4-byte result. In the test immediately below, the
238 * 8 is for start bytes F0-F7 (which are all the possible ones for 4 byte
239 * characters). The 3 is for 3 continuation bytes; these each contribute
240 * SHIFT bits. This yields 0x4_0000 on EBCDIC platforms, 0x20_0000 on
241 * ASCII, so 4 bytes covers the range 0x4000-0x3_FFFF on EBCDIC;
242 * 0x1_0000-0x1F_FFFF on ASCII */
243 if (uv < (8 * (1U << (3 * SHIFT)))) {
244 *d++ = I8_TO_NATIVE_UTF8(( uv >> ((4 - 1) * SHIFT)) | UTF_START_MARK(4));
245 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((3 - 1) * SHIFT)) & MASK) | MARK);
246 *d++ = I8_TO_NATIVE_UTF8(((uv >> ((2 - 1) * SHIFT)) & MASK) | MARK);
247 *d++ = I8_TO_NATIVE_UTF8(( uv /* (1 - 1) */ & MASK) | MARK);
249 #ifdef EBCDIC /* These were handled on ASCII platforms in the code for 3-byte
250 characters. The end-plane non-characters for EBCDIC were
251 handled just above */
252 if (UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv))) {
253 HANDLE_UNICODE_NONCHAR(uv, flags);
255 else if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
256 HANDLE_UNICODE_SURROGATE(uv, flags);
263 /* Not 4-byte; that means the code point is at least 0x20_0000 on ASCII
264 * platforms, and 0x4000 on EBCDIC. At this point we switch to a loop
265 * format. The unrolled version above turns out to not save all that much
266 * time, and at these high code points (well above the legal Unicode range
267 * on ASCII platforms, and well above anything in common use in EBCDIC),
268 * khw believes that less code outweighs slight performance gains. */
271 STRLEN len = OFFUNISKIP(uv);
274 *p-- = I8_TO_NATIVE_UTF8((uv & UTF_CONTINUATION_MASK) | UTF_CONTINUATION_MARK);
275 uv >>= UTF_ACCUMULATION_SHIFT;
277 *p = I8_TO_NATIVE_UTF8((uv & UTF_START_MASK(len)) | UTF_START_MARK(len));
283 =for apidoc uvchr_to_utf8
285 Adds the UTF-8 representation of the native code point C<uv> to the end
286 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
287 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
288 the byte after the end of the new character. In other words,
290 d = uvchr_to_utf8(d, uv);
292 is the recommended wide native character-aware way of saying
296 This function accepts any code point from 0..C<IV_MAX> as input.
297 C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
299 It is possible to forbid or warn on non-Unicode code points, or those that may
300 be problematic by using L</uvchr_to_utf8_flags>.
305 /* This is also a macro */
306 PERL_CALLCONV U8* Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv);
309 Perl_uvchr_to_utf8(pTHX_ U8 *d, UV uv)
311 return uvchr_to_utf8(d, uv);
315 =for apidoc uvchr_to_utf8_flags
317 Adds the UTF-8 representation of the native code point C<uv> to the end
318 of the string C<d>; C<d> should have at least C<UVCHR_SKIP(uv)+1> (up to
319 C<UTF8_MAXBYTES+1>) free bytes available. The return value is the pointer to
320 the byte after the end of the new character. In other words,
322 d = uvchr_to_utf8_flags(d, uv, flags);
326 d = uvchr_to_utf8_flags(d, uv, 0);
328 This is the Unicode-aware way of saying
332 If C<flags> is 0, this function accepts any code point from 0..C<IV_MAX> as
333 input. C<IV_MAX> is typically 0x7FFF_FFFF in a 32-bit word.
335 Specifying C<flags> can further restrict what is allowed and not warned on, as
338 If C<uv> is a Unicode surrogate code point and C<UNICODE_WARN_SURROGATE> is set,
339 the function will raise a warning, provided UTF8 warnings are enabled. If
340 instead C<UNICODE_DISALLOW_SURROGATE> is set, the function will fail and return
341 NULL. If both flags are set, the function will both warn and return NULL.
343 Similarly, the C<UNICODE_WARN_NONCHAR> and C<UNICODE_DISALLOW_NONCHAR> flags
344 affect how the function handles a Unicode non-character.
346 And likewise, the C<UNICODE_WARN_SUPER> and C<UNICODE_DISALLOW_SUPER> flags
347 affect the handling of code points that are above the Unicode maximum of
348 0x10FFFF. Languages other than Perl may not be able to accept files that
351 The flag C<UNICODE_WARN_ILLEGAL_INTERCHANGE> selects all three of
352 the above WARN flags; and C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> selects all
353 three DISALLOW flags. C<UNICODE_DISALLOW_ILLEGAL_INTERCHANGE> restricts the
354 allowed inputs to the strict UTF-8 traditionally defined by Unicode.
355 Similarly, C<UNICODE_WARN_ILLEGAL_C9_INTERCHANGE> and
356 C<UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE> are shortcuts to select the
357 above-Unicode and surrogate flags, but not the non-character ones, as
359 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
360 See L<perlunicode/Noncharacter code points>.
362 Extremely high code points were never specified in any standard, and require an
363 extension to UTF-8 to express, which Perl does. It is likely that programs
364 written in something other than Perl would not be able to read files that
365 contain these; nor would Perl understand files written by something that uses a
366 different extension. For these reasons, there is a separate set of flags that
367 can warn and/or disallow these extremely high code points, even if other
368 above-Unicode ones are accepted. They are the C<UNICODE_WARN_PERL_EXTENDED>
369 and C<UNICODE_DISALLOW_PERL_EXTENDED> flags. For more information see
370 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UNICODE_DISALLOW_SUPER> will
371 treat all above-Unicode code points, including these, as malformations. (Note
372 that the Unicode standard considers anything above 0x10FFFF to be illegal, but
373 there are standards predating it that allow up to 0x7FFF_FFFF (2**31 -1))
375 A somewhat misleadingly named synonym for C<UNICODE_WARN_PERL_EXTENDED> is
376 retained for backward compatibility: C<UNICODE_WARN_ABOVE_31_BIT>. Similarly,
377 C<UNICODE_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
378 C<UNICODE_DISALLOW_PERL_EXTENDED>. The names are misleading because these
379 flags can apply to code points that actually do fit in 31 bits. This happens
380 on EBCDIC platforms, and sometimes when the L<overlong
381 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
382 describe the situation in all cases.
387 /* This is also a macro */
388 PERL_CALLCONV U8* Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags);
391 Perl_uvchr_to_utf8_flags(pTHX_ U8 *d, UV uv, UV flags)
393 return uvchr_to_utf8_flags(d, uv, flags);
399 S_is_utf8_cp_above_31_bits(const U8 * const s,
401 const bool consider_overlongs)
403 /* Returns TRUE if the first code point represented by the Perl-extended-
404 * UTF-8-encoded string starting at 's', and looking no further than 'e -
405 * 1' doesn't fit into 31 bytes. That is, that if it is >= 2**31.
407 * The function handles the case where the input bytes do not include all
408 * the ones necessary to represent a full character. That is, they may be
409 * the intial bytes of the representation of a code point, but possibly
410 * the final ones necessary for the complete representation may be beyond
413 * The function also can handle the case where the input is an overlong
414 * sequence. If 'consider_overlongs' is 0, the function assumes the
415 * input is not overlong, without checking, and will return based on that
416 * assumption. If this parameter is 1, the function will go to the trouble
417 * of figuring out if it actually evaluates to above or below 31 bits.
419 * The sequence is otherwise assumed to be well-formed, without checking.
422 const STRLEN len = e - s;
425 PERL_ARGS_ASSERT_IS_UTF8_CP_ABOVE_31_BITS;
427 assert(! UTF8_IS_INVARIANT(*s) && e > s);
431 PERL_UNUSED_ARG(consider_overlongs);
433 /* On the EBCDIC code pages we handle, only the native start byte 0xFE can
434 * mean a 32-bit or larger code point (0xFF is an invariant). 0xFE can
435 * also be the start byte for a 31-bit code point; we need at least 2
436 * bytes, and maybe up through 8 bytes, to determine that. (It can also be
437 * the start byte for an overlong sequence, but for 30-bit or smaller code
438 * points, so we don't have to worry about overlongs on EBCDIC.) */
449 /* On ASCII, FE and FF are the only start bytes that can evaluate to
450 * needing more than 31 bits. */
451 if (LIKELY(*s < 0xFE)) {
455 /* What we have left are FE and FF. Both of these require more than 31
456 * bits unless they are for overlongs. */
457 if (! consider_overlongs) {
461 /* Here, we have FE or FF. If the input isn't overlong, it evaluates to
462 * above 31 bits. But we need more than one byte to discern this, so if
463 * passed just the start byte, it could be an overlong evaluating to
469 /* Having excluded len==1, and knowing that FE and FF are both valid start
470 * bytes, we can call the function below to see if the sequence is
471 * overlong. (We don't need the full generality of the called function,
472 * but for these huge code points, speed shouldn't be a consideration, and
473 * the compiler does have enough information, since it's static to this
474 * file, to optimize to just the needed parts.) */
475 is_overlong = is_utf8_overlong_given_start_byte_ok(s, len);
477 /* If it isn't overlong, more than 31 bits are required. */
478 if (is_overlong == 0) {
482 /* If it is indeterminate if it is overlong, return that */
483 if (is_overlong < 0) {
487 /* Here is overlong. Such a sequence starting with FE is below 31 bits, as
488 * the max it can be is 2**31 - 1 */
495 /* Here, ASCII and EBCDIC rejoin:
496 * On ASCII: We have an overlong sequence starting with FF
497 * On EBCDIC: We have a sequence starting with FE. */
499 { /* For C89, use a block so the declaration can be close to its use */
503 /* U+7FFFFFFF (2 ** 31 - 1)
504 * [0] [1] [2] [3] [4] [5] [6] [7] [8] [9] 10 11 12 13
505 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x42\x73\x73\x73\x73\x73\x73
506 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x42\x72\x72\x72\x72\x72\x72
507 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x42\x75\x75\x75\x75\x75\x75
508 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA1\xBF\xBF\xBF\xBF\xBF\xBF
509 * U+80000000 (2 ** 31):
510 * IBM-1047: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
511 * IBM-037: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
512 * POSIX-BC: \xFE\x41\x41\x41\x41\x41\x41\x43\x41\x41\x41\x41\x41\x41
513 * I8: \xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA2\xA0\xA0\xA0\xA0\xA0\xA0
515 * and since we know that *s = \xfe, any continuation sequcence
516 * following it that is gt the below is above 31 bits
517 [0] [1] [2] [3] [4] [5] [6] */
518 const U8 conts_for_highest_30_bit[] = "\x41\x41\x41\x41\x41\x41\x42";
522 /* FF overlong for U+7FFFFFFF (2 ** 31 - 1)
523 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x81\xBF\xBF\xBF\xBF\xBF
524 * FF overlong for U+80000000 (2 ** 31):
525 * ASCII: \xFF\x80\x80\x80\x80\x80\x80\x82\x80\x80\x80\x80\x80
526 * and since we know that *s = \xff, any continuation sequcence
527 * following it that is gt the below is above 30 bits
528 [0] [1] [2] [3] [4] [5] [6] */
529 const U8 conts_for_highest_30_bit[] = "\x80\x80\x80\x80\x80\x80\x81";
533 const STRLEN conts_len = sizeof(conts_for_highest_30_bit) - 1;
534 const STRLEN cmp_len = MIN(conts_len, len - 1);
536 /* Now compare the continuation bytes in s with the ones we have
537 * compiled in that are for the largest 30 bit code point. If we have
538 * enough bytes available to determine the answer, or the bytes we do
539 * have differ from them, we can compare the two to get a definitive
540 * answer (Note that in UTF-EBCDIC, the two lowest possible
541 * continuation bytes are \x41 and \x42.) */
542 if (cmp_len >= conts_len || memNE(s + 1,
543 conts_for_highest_30_bit,
546 return cBOOL(memGT(s + 1, conts_for_highest_30_bit, cmp_len));
549 /* Here, all the bytes we have are the same as the highest 30-bit code
550 * point, but we are missing so many bytes that we can't make the
558 PERL_STATIC_INLINE int
559 S_is_utf8_overlong_given_start_byte_ok(const U8 * const s, const STRLEN len)
561 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
562 * 's' + 'len' - 1 is an overlong. It returns 1 if it is an overlong; 0 if
563 * it isn't, and -1 if there isn't enough information to tell. This last
564 * return value can happen if the sequence is incomplete, missing some
565 * trailing bytes that would form a complete character. If there are
566 * enough bytes to make a definitive decision, this function does so.
567 * Usually 2 bytes sufficient.
569 * Overlongs can occur whenever the number of continuation bytes changes.
570 * That means whenever the number of leading 1 bits in a start byte
571 * increases from the next lower start byte. That happens for start bytes
572 * C0, E0, F0, F8, FC, FE, and FF. On modern perls, the following illegal
573 * start bytes have already been excluded, so don't need to be tested here;
574 * ASCII platforms: C0, C1
575 * EBCDIC platforms C0, C1, C2, C3, C4, E0
578 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
579 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
581 PERL_ARGS_ASSERT_IS_UTF8_OVERLONG_GIVEN_START_BYTE_OK;
582 assert(len > 1 && UTF8_IS_START(*s));
584 /* Each platform has overlongs after the start bytes given above (expressed
585 * in I8 for EBCDIC). What constitutes an overlong varies by platform, but
586 * the logic is the same, except the E0 overlong has already been excluded
587 * on EBCDIC platforms. The values below were found by manually
588 * inspecting the UTF-8 patterns. See the tables in utf8.h and
592 # define F0_ABOVE_OVERLONG 0xB0
593 # define F8_ABOVE_OVERLONG 0xA8
594 # define FC_ABOVE_OVERLONG 0xA4
595 # define FE_ABOVE_OVERLONG 0xA2
596 # define FF_OVERLONG_PREFIX "\xfe\x41\x41\x41\x41\x41\x41\x41"
600 if (s0 == 0xE0 && UNLIKELY(s1 < 0xA0)) {
604 # define F0_ABOVE_OVERLONG 0x90
605 # define F8_ABOVE_OVERLONG 0x88
606 # define FC_ABOVE_OVERLONG 0x84
607 # define FE_ABOVE_OVERLONG 0x82
608 # define FF_OVERLONG_PREFIX "\xff\x80\x80\x80\x80\x80\x80"
612 if ( (s0 == 0xF0 && UNLIKELY(s1 < F0_ABOVE_OVERLONG))
613 || (s0 == 0xF8 && UNLIKELY(s1 < F8_ABOVE_OVERLONG))
614 || (s0 == 0xFC && UNLIKELY(s1 < FC_ABOVE_OVERLONG))
615 || (s0 == 0xFE && UNLIKELY(s1 < FE_ABOVE_OVERLONG)))
620 /* Check for the FF overlong */
621 return isFF_OVERLONG(s, len);
624 PERL_STATIC_INLINE int
625 S_isFF_OVERLONG(const U8 * const s, const STRLEN len)
627 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
628 * 'e' - 1 is an overlong beginning with \xFF. It returns 1 if it is; 0 if
629 * it isn't, and -1 if there isn't enough information to tell. This last
630 * return value can happen if the sequence is incomplete, missing some
631 * trailing bytes that would form a complete character. If there are
632 * enough bytes to make a definitive decision, this function does so. */
634 PERL_ARGS_ASSERT_ISFF_OVERLONG;
636 /* To be an FF overlong, all the available bytes must match */
637 if (LIKELY(memNE(s, FF_OVERLONG_PREFIX,
638 MIN(len, sizeof(FF_OVERLONG_PREFIX) - 1))))
643 /* To be an FF overlong sequence, all the bytes in FF_OVERLONG_PREFIX must
644 * be there; what comes after them doesn't matter. See tables in utf8.h,
646 if (len >= sizeof(FF_OVERLONG_PREFIX) - 1) {
650 /* The missing bytes could cause the result to go one way or the other, so
651 * the result is indeterminate */
655 #if defined(UV_IS_QUAD) /* These assume IV_MAX is 2**63-1 */
656 # ifdef EBCDIC /* Actually is I8 */
657 # define HIGHEST_REPRESENTABLE_UTF8 \
658 "\xFF\xA7\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
660 # define HIGHEST_REPRESENTABLE_UTF8 \
661 "\xFF\x80\x87\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
665 PERL_STATIC_INLINE int
666 S_does_utf8_overflow(const U8 * const s,
668 const bool consider_overlongs)
670 /* Returns an int indicating whether or not the UTF-8 sequence from 's' to
671 * 'e' - 1 would overflow an IV on this platform; that is if it represents
672 * a code point larger than the highest representable code point. It
673 * returns 1 if it does overflow; 0 if it doesn't, and -1 if there isn't
674 * enough information to tell. This last return value can happen if the
675 * sequence is incomplete, missing some trailing bytes that would form a
676 * complete character. If there are enough bytes to make a definitive
677 * decision, this function does so.
679 * If 'consider_overlongs' is TRUE, the function checks for the possibility
680 * that the sequence is an overlong that doesn't overflow. Otherwise, it
681 * assumes the sequence is not an overlong. This can give different
682 * results only on ASCII 32-bit platforms.
684 * (For ASCII platforms, we could use memcmp() because we don't have to
685 * convert each byte to I8, but it's very rare input indeed that would
686 * approach overflow, so the loop below will likely only get executed once.)
688 * 'e' - 1 must not be beyond a full character. */
691 PERL_ARGS_ASSERT_DOES_UTF8_OVERFLOW;
692 assert(s <= e && s + UTF8SKIP(s) >= e);
694 #if ! defined(UV_IS_QUAD)
696 return is_utf8_cp_above_31_bits(s, e, consider_overlongs);
700 PERL_UNUSED_ARG(consider_overlongs);
703 const STRLEN len = e - s;
705 const U8 * y = (const U8 *) HIGHEST_REPRESENTABLE_UTF8;
707 for (x = s; x < e; x++, y++) {
709 if (UNLIKELY(NATIVE_UTF8_TO_I8(*x) == *y)) {
713 /* If this byte is larger than the corresponding highest UTF-8
714 * byte, the sequence overflow; otherwise the byte is less than,
715 * and so the sequence doesn't overflow */
716 return NATIVE_UTF8_TO_I8(*x) > *y;
720 /* Got to the end and all bytes are the same. If the input is a whole
721 * character, it doesn't overflow. And if it is a partial character,
722 * there's not enough information to tell */
723 if (len < sizeof(HIGHEST_REPRESENTABLE_UTF8) - 1) {
736 /* This is the portions of the above function that deal with UV_MAX instead of
737 * IV_MAX. They are left here in case we want to combine them so that internal
738 * uses can have larger code points. The only logic difference is that the
739 * 32-bit EBCDIC platform is treate like the 64-bit, and the 32-bit ASCII has
743 /* Anything larger than this will overflow the word if it were converted into a UV */
744 #if defined(UV_IS_QUAD)
745 # ifdef EBCDIC /* Actually is I8 */
746 # define HIGHEST_REPRESENTABLE_UTF8 \
747 "\xFF\xAF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
749 # define HIGHEST_REPRESENTABLE_UTF8 \
750 "\xFF\x80\x8F\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF\xBF"
754 # define HIGHEST_REPRESENTABLE_UTF8 \
755 "\xFF\xA0\xA0\xA0\xA0\xA0\xA0\xA3\xBF\xBF\xBF\xBF\xBF\xBF"
757 # define HIGHEST_REPRESENTABLE_UTF8 "\xFE\x83\xBF\xBF\xBF\xBF\xBF"
761 #if ! defined(UV_IS_QUAD) && ! defined(EBCDIC)
763 /* On 32 bit ASCII machines, many overlongs that start with FF don't
765 if (consider_overlongs && isFF_OVERLONG(s, len) > 0) {
767 /* To be such an overlong, the first bytes of 's' must match
768 * FF_OVERLONG_PREFIX, which is "\xff\x80\x80\x80\x80\x80\x80". If we
769 * don't have any additional bytes available, the sequence, when
770 * completed might or might not fit in 32 bits. But if we have that
771 * next byte, we can tell for sure. If it is <= 0x83, then it does
773 if (len <= sizeof(FF_OVERLONG_PREFIX) - 1) {
777 return s[sizeof(FF_OVERLONG_PREFIX) - 1] > 0x83;
780 /* Starting with the #else, the rest of the function is identical except
781 * 1. we need to move the 'len' declaration to be global to the function
782 * 2. the endif move to just after the UNUSED_ARG.
783 * An empty endif is given just below to satisfy the preprocessor
789 #undef F0_ABOVE_OVERLONG
790 #undef F8_ABOVE_OVERLONG
791 #undef FC_ABOVE_OVERLONG
792 #undef FE_ABOVE_OVERLONG
793 #undef FF_OVERLONG_PREFIX
796 Perl__is_utf8_char_helper(const U8 * const s, const U8 * e, const U32 flags)
801 /* A helper function that should not be called directly.
803 * This function returns non-zero if the string beginning at 's' and
804 * looking no further than 'e - 1' is well-formed Perl-extended-UTF-8 for a
805 * code point; otherwise it returns 0. The examination stops after the
806 * first code point in 's' is validated, not looking at the rest of the
807 * input. If 'e' is such that there are not enough bytes to represent a
808 * complete code point, this function will return non-zero anyway, if the
809 * bytes it does have are well-formed UTF-8 as far as they go, and aren't
810 * excluded by 'flags'.
812 * A non-zero return gives the number of bytes required to represent the
813 * code point. Be aware that if the input is for a partial character, the
814 * return will be larger than 'e - s'.
816 * This function assumes that the code point represented is UTF-8 variant.
817 * The caller should have excluded the possibility of it being invariant
818 * before calling this function.
820 * 'flags' can be 0, or any combination of the UTF8_DISALLOW_foo flags
821 * accepted by L</utf8n_to_uvchr>. If non-zero, this function will return
822 * 0 if the code point represented is well-formed Perl-extended-UTF-8, but
823 * disallowed by the flags. If the input is only for a partial character,
824 * the function will return non-zero if there is any sequence of
825 * well-formed UTF-8 that, when appended to the input sequence, could
826 * result in an allowed code point; otherwise it returns 0. Non characters
827 * cannot be determined based on partial character input. But many of the
828 * other excluded types can be determined with just the first one or two
833 PERL_ARGS_ASSERT__IS_UTF8_CHAR_HELPER;
835 assert(0 == (flags & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE
836 |UTF8_DISALLOW_PERL_EXTENDED)));
837 assert(! UTF8_IS_INVARIANT(*s));
839 /* A variant char must begin with a start byte */
840 if (UNLIKELY(! UTF8_IS_START(*s))) {
844 /* Examine a maximum of a single whole code point */
845 if (e - s > UTF8SKIP(s)) {
851 if (flags && isUTF8_POSSIBLY_PROBLEMATIC(*s)) {
852 const U8 s0 = NATIVE_UTF8_TO_I8(s[0]);
854 /* Here, we are disallowing some set of largish code points, and the
855 * first byte indicates the sequence is for a code point that could be
856 * in the excluded set. We generally don't have to look beyond this or
857 * the second byte to see if the sequence is actually for one of the
858 * excluded classes. The code below is derived from this table:
860 * UTF-8 UTF-EBCDIC I8
861 * U+D800: \xED\xA0\x80 \xF1\xB6\xA0\xA0 First surrogate
862 * U+DFFF: \xED\xBF\xBF \xF1\xB7\xBF\xBF Final surrogate
863 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 First above Unicode
865 * Keep in mind that legal continuation bytes range between \x80..\xBF
866 * for UTF-8, and \xA0..\xBF for I8. Anything above those aren't
867 * continuation bytes. Hence, we don't have to test the upper edge
868 * because if any of those is encountered, the sequence is malformed,
869 * and would fail elsewhere in this function.
871 * The code here likewise assumes that there aren't other
872 * malformations; again the function should fail elsewhere because of
873 * these. For example, an overlong beginning with FC doesn't actually
874 * have to be a super; it could actually represent a small code point,
875 * even U+0000. But, since overlongs (and other malformations) are
876 * illegal, the function should return FALSE in either case.
879 #ifdef EBCDIC /* On EBCDIC, these are actually I8 bytes */
880 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xFA
881 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF9 && (s1) >= 0xA2)
883 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xF1 \
885 && ((s1) & 0xFE ) == 0xB6)
886 # define isUTF8_PERL_EXTENDED(s) (*s == I8_TO_NATIVE_UTF8(0xFF))
888 # define FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER 0xF5
889 # define IS_UTF8_2_BYTE_SUPER(s0, s1) ((s0) == 0xF4 && (s1) >= 0x90)
890 # define IS_UTF8_2_BYTE_SURROGATE(s0, s1) ((s0) == 0xED && (s1) >= 0xA0)
891 # define isUTF8_PERL_EXTENDED(s) (*s >= 0xFE)
894 if ( (flags & UTF8_DISALLOW_SUPER)
895 && UNLIKELY(s0 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
897 return 0; /* Above Unicode */
900 if ( (flags & UTF8_DISALLOW_PERL_EXTENDED)
901 && UNLIKELY(isUTF8_PERL_EXTENDED(s)))
907 const U8 s1 = NATIVE_UTF8_TO_I8(s[1]);
909 if ( (flags & UTF8_DISALLOW_SUPER)
910 && UNLIKELY(IS_UTF8_2_BYTE_SUPER(s0, s1)))
912 return 0; /* Above Unicode */
915 if ( (flags & UTF8_DISALLOW_SURROGATE)
916 && UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(s0, s1)))
918 return 0; /* Surrogate */
921 if ( (flags & UTF8_DISALLOW_NONCHAR)
922 && UNLIKELY(UTF8_IS_NONCHAR(s, e)))
924 return 0; /* Noncharacter code point */
929 /* Make sure that all that follows are continuation bytes */
930 for (x = s + 1; x < e; x++) {
931 if (UNLIKELY(! UTF8_IS_CONTINUATION(*x))) {
936 /* Here is syntactically valid. Next, make sure this isn't the start of an
938 if (len > 1 && is_utf8_overlong_given_start_byte_ok(s, len) > 0) {
942 /* And finally, that the code point represented fits in a word on this
944 if (0 < does_utf8_overflow(s, e,
945 0 /* Don't consider overlongs */
955 Perl__byte_dump_string(pTHX_ const U8 * const start, const STRLEN len, const bool format)
957 /* Returns a mortalized C string that is a displayable copy of the 'len'
958 * bytes starting at 'start'. 'format' gives how to display each byte.
959 * Currently, there are only two formats, so it is currently a bool:
961 * 1 ab (that is a space between two hex digit bytes)
964 const STRLEN output_len = 4 * len + 1; /* 4 bytes per each input, plus a
966 const U8 * s = start;
967 const U8 * const e = start + len;
971 PERL_ARGS_ASSERT__BYTE_DUMP_STRING;
973 Newx(output, output_len, char);
977 for (s = start; s < e; s++) {
978 const unsigned high_nibble = (*s & 0xF0) >> 4;
979 const unsigned low_nibble = (*s & 0x0F);
991 if (high_nibble < 10) {
992 *d++ = high_nibble + '0';
995 *d++ = high_nibble - 10 + 'a';
998 if (low_nibble < 10) {
999 *d++ = low_nibble + '0';
1002 *d++ = low_nibble - 10 + 'a';
1010 PERL_STATIC_INLINE char *
1011 S_unexpected_non_continuation_text(pTHX_ const U8 * const s,
1013 /* How many bytes to print */
1016 /* Which one is the non-continuation */
1017 const STRLEN non_cont_byte_pos,
1019 /* How many bytes should there be? */
1020 const STRLEN expect_len)
1022 /* Return the malformation warning text for an unexpected continuation
1025 const char * const where = (non_cont_byte_pos == 1)
1027 : Perl_form(aTHX_ "%d bytes",
1028 (int) non_cont_byte_pos);
1030 PERL_ARGS_ASSERT_UNEXPECTED_NON_CONTINUATION_TEXT;
1032 /* We don't need to pass this parameter, but since it has already been
1033 * calculated, it's likely faster to pass it; verify under DEBUGGING */
1034 assert(expect_len == UTF8SKIP(s));
1036 return Perl_form(aTHX_ "%s: %s (unexpected non-continuation byte 0x%02x,"
1037 " %s after start byte 0x%02x; need %d bytes, got %d)",
1039 _byte_dump_string(s, print_len, 0),
1040 *(s + non_cont_byte_pos),
1044 (int) non_cont_byte_pos);
1049 =for apidoc utf8n_to_uvchr
1051 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1052 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1054 Bottom level UTF-8 decode routine.
1055 Returns the native code point value of the first character in the string C<s>,
1056 which is assumed to be in UTF-8 (or UTF-EBCDIC) encoding, and no longer than
1057 C<curlen> bytes; C<*retlen> (if C<retlen> isn't NULL) will be set to
1058 the length, in bytes, of that character.
1060 The value of C<flags> determines the behavior when C<s> does not point to a
1061 well-formed UTF-8 character. If C<flags> is 0, encountering a malformation
1062 causes zero to be returned and C<*retlen> is set so that (S<C<s> + C<*retlen>>)
1063 is the next possible position in C<s> that could begin a non-malformed
1064 character. Also, if UTF-8 warnings haven't been lexically disabled, a warning
1065 is raised. Some UTF-8 input sequences may contain multiple malformations.
1066 This function tries to find every possible one in each call, so multiple
1067 warnings can be raised for the same sequence.
1069 Various ALLOW flags can be set in C<flags> to allow (and not warn on)
1070 individual types of malformations, such as the sequence being overlong (that
1071 is, when there is a shorter sequence that can express the same code point;
1072 overlong sequences are expressly forbidden in the UTF-8 standard due to
1073 potential security issues). Another malformation example is the first byte of
1074 a character not being a legal first byte. See F<utf8.h> for the list of such
1075 flags. Even if allowed, this function generally returns the Unicode
1076 REPLACEMENT CHARACTER when it encounters a malformation. There are flags in
1077 F<utf8.h> to override this behavior for the overlong malformations, but don't
1078 do that except for very specialized purposes.
1080 The C<UTF8_CHECK_ONLY> flag overrides the behavior when a non-allowed (by other
1081 flags) malformation is found. If this flag is set, the routine assumes that
1082 the caller will raise a warning, and this function will silently just set
1083 C<retlen> to C<-1> (cast to C<STRLEN>) and return zero.
1085 Note that this API requires disambiguation between successful decoding a C<NUL>
1086 character, and an error return (unless the C<UTF8_CHECK_ONLY> flag is set), as
1087 in both cases, 0 is returned, and, depending on the malformation, C<retlen> may
1088 be set to 1. To disambiguate, upon a zero return, see if the first byte of
1089 C<s> is 0 as well. If so, the input was a C<NUL>; if not, the input had an
1090 error. Or you can use C<L</utf8n_to_uvchr_error>>.
1092 Certain code points are considered problematic. These are Unicode surrogates,
1093 Unicode non-characters, and code points above the Unicode maximum of 0x10FFFF.
1094 By default these are considered regular code points, but certain situations
1095 warrant special handling for them, which can be specified using the C<flags>
1096 parameter. If C<flags> contains C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, all
1097 three classes are treated as malformations and handled as such. The flags
1098 C<UTF8_DISALLOW_SURROGATE>, C<UTF8_DISALLOW_NONCHAR>, and
1099 C<UTF8_DISALLOW_SUPER> (meaning above the legal Unicode maximum) can be set to
1100 disallow these categories individually. C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>
1101 restricts the allowed inputs to the strict UTF-8 traditionally defined by
1102 Unicode. Use C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE> to use the strictness
1104 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1105 The difference between traditional strictness and C9 strictness is that the
1106 latter does not forbid non-character code points. (They are still discouraged,
1107 however.) For more discussion see L<perlunicode/Noncharacter code points>.
1109 The flags C<UTF8_WARN_ILLEGAL_INTERCHANGE>,
1110 C<UTF8_WARN_ILLEGAL_C9_INTERCHANGE>, C<UTF8_WARN_SURROGATE>,
1111 C<UTF8_WARN_NONCHAR>, and C<UTF8_WARN_SUPER> will cause warning messages to be
1112 raised for their respective categories, but otherwise the code points are
1113 considered valid (not malformations). To get a category to both be treated as
1114 a malformation and raise a warning, specify both the WARN and DISALLOW flags.
1115 (But note that warnings are not raised if lexically disabled nor if
1116 C<UTF8_CHECK_ONLY> is also specified.)
1118 Extremely high code points were never specified in any standard, and require an
1119 extension to UTF-8 to express, which Perl does. It is likely that programs
1120 written in something other than Perl would not be able to read files that
1121 contain these; nor would Perl understand files written by something that uses a
1122 different extension. For these reasons, there is a separate set of flags that
1123 can warn and/or disallow these extremely high code points, even if other
1124 above-Unicode ones are accepted. They are the C<UTF8_WARN_PERL_EXTENDED> and
1125 C<UTF8_DISALLOW_PERL_EXTENDED> flags. For more information see
1126 L</C<UTF8_GOT_PERL_EXTENDED>>. Of course C<UTF8_DISALLOW_SUPER> will treat all
1127 above-Unicode code points, including these, as malformations.
1128 (Note that the Unicode standard considers anything above 0x10FFFF to be
1129 illegal, but there are standards predating it that allow up to 0x7FFF_FFFF
1132 A somewhat misleadingly named synonym for C<UTF8_WARN_PERL_EXTENDED> is
1133 retained for backward compatibility: C<UTF8_WARN_ABOVE_31_BIT>. Similarly,
1134 C<UTF8_DISALLOW_ABOVE_31_BIT> is usable instead of the more accurately named
1135 C<UTF8_DISALLOW_PERL_EXTENDED>. The names are misleading because these flags
1136 can apply to code points that actually do fit in 31 bits. This happens on
1137 EBCDIC platforms, and sometimes when the L<overlong
1138 malformation|/C<UTF8_GOT_LONG>> is also present. The new names accurately
1139 describe the situation in all cases.
1142 All other code points corresponding to Unicode characters, including private
1143 use and those yet to be assigned, are never considered malformed and never
1148 Also implemented as a macro in utf8.h
1152 Perl_utf8n_to_uvchr(pTHX_ const U8 *s,
1157 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR;
1159 return utf8n_to_uvchr_error(s, curlen, retlen, flags, NULL);
1164 =for apidoc utf8n_to_uvchr_error
1166 THIS FUNCTION SHOULD BE USED IN ONLY VERY SPECIALIZED CIRCUMSTANCES.
1167 Most code should use L</utf8_to_uvchr_buf>() rather than call this directly.
1169 This function is for code that needs to know what the precise malformation(s)
1170 are when an error is found.
1172 It is like C<L</utf8n_to_uvchr>> but it takes an extra parameter placed after
1173 all the others, C<errors>. If this parameter is 0, this function behaves
1174 identically to C<L</utf8n_to_uvchr>>. Otherwise, C<errors> should be a pointer
1175 to a C<U32> variable, which this function sets to indicate any errors found.
1176 Upon return, if C<*errors> is 0, there were no errors found. Otherwise,
1177 C<*errors> is the bit-wise C<OR> of the bits described in the list below. Some
1178 of these bits will be set if a malformation is found, even if the input
1179 C<flags> parameter indicates that the given malformation is allowed; those
1180 exceptions are noted:
1184 =item C<UTF8_GOT_PERL_EXTENDED>
1186 The input sequence is not standard UTF-8, but a Perl extension. This bit is
1187 set only if the input C<flags> parameter contains either the
1188 C<UTF8_DISALLOW_PERL_EXTENDED> or the C<UTF8_WARN_PERL_EXTENDED> flags.
1190 Code points above 0x7FFF_FFFF (2**31 - 1) were never specified in any standard,
1191 and so some extension must be used to express them. Perl uses a natural
1192 extension to UTF-8 to represent the ones up to 2**36-1, and invented a further
1193 extension to represent even higher ones, so that any code point that fits in a
1194 64-bit word can be represented. Text using these extensions is not likely to
1195 be portable to non-Perl code. We lump both of these extensions together and
1196 refer to them as Perl extended UTF-8. There exist other extensions that people
1197 have invented, incompatible with Perl's.
1199 On EBCDIC platforms starting in Perl v5.24, the Perl extension for representing
1200 extremely high code points kicks in at 0x3FFF_FFFF (2**30 -1), which is lower
1201 than on ASCII. Prior to that, code points 2**31 and higher were simply
1202 unrepresentable, and a different, incompatible method was used to represent
1203 code points between 2**30 and 2**31 - 1.
1205 On both platforms, ASCII and EBCDIC, C<UTF8_GOT_PERL_EXTENDED> is set if
1206 Perl extended UTF-8 is used.
1208 In earlier Perls, this bit was named C<UTF8_GOT_ABOVE_31_BIT>, which you still
1209 may use for backward compatibility. That name is misleading, as this flag may
1210 be set when the code point actually does fit in 31 bits. This happens on
1211 EBCDIC platforms, and sometimes when the L<overlong
1212 malformation|/C<UTF8_GOT_LONG>> is also present. The new name accurately
1213 describes the situation in all cases.
1215 =item C<UTF8_GOT_CONTINUATION>
1217 The input sequence was malformed in that the first byte was a a UTF-8
1220 =item C<UTF8_GOT_EMPTY>
1222 The input C<curlen> parameter was 0.
1224 =item C<UTF8_GOT_LONG>
1226 The input sequence was malformed in that there is some other sequence that
1227 evaluates to the same code point, but that sequence is shorter than this one.
1229 Until Unicode 3.1, it was legal for programs to accept this malformation, but
1230 it was discovered that this created security issues.
1232 =item C<UTF8_GOT_NONCHAR>
1234 The code point represented by the input UTF-8 sequence is for a Unicode
1235 non-character code point.
1236 This bit is set only if the input C<flags> parameter contains either the
1237 C<UTF8_DISALLOW_NONCHAR> or the C<UTF8_WARN_NONCHAR> flags.
1239 =item C<UTF8_GOT_NON_CONTINUATION>
1241 The input sequence was malformed in that a non-continuation type byte was found
1242 in a position where only a continuation type one should be.
1244 =item C<UTF8_GOT_OVERFLOW>
1246 The input sequence was malformed in that it is for a code point that is not
1247 representable in the number of bits available in an IV on the current platform.
1249 =item C<UTF8_GOT_SHORT>
1251 The input sequence was malformed in that C<curlen> is smaller than required for
1252 a complete sequence. In other words, the input is for a partial character
1255 =item C<UTF8_GOT_SUPER>
1257 The input sequence was malformed in that it is for a non-Unicode code point;
1258 that is, one above the legal Unicode maximum.
1259 This bit is set only if the input C<flags> parameter contains either the
1260 C<UTF8_DISALLOW_SUPER> or the C<UTF8_WARN_SUPER> flags.
1262 =item C<UTF8_GOT_SURROGATE>
1264 The input sequence was malformed in that it is for a -Unicode UTF-16 surrogate
1266 This bit is set only if the input C<flags> parameter contains either the
1267 C<UTF8_DISALLOW_SURROGATE> or the C<UTF8_WARN_SURROGATE> flags.
1271 To do your own error handling, call this function with the C<UTF8_CHECK_ONLY>
1272 flag to suppress any warnings, and then examine the C<*errors> return.
1278 Perl_utf8n_to_uvchr_error(pTHX_ const U8 *s,
1284 const U8 * const s0 = s;
1285 U8 * send = NULL; /* (initialized to silence compilers' wrong
1287 U32 possible_problems = 0; /* A bit is set here for each potential problem
1288 found as we go along */
1290 STRLEN expectlen = 0; /* How long should this sequence be?
1291 (initialized to silence compilers' wrong
1293 STRLEN avail_len = 0; /* When input is too short, gives what that is */
1294 U32 discard_errors = 0; /* Used to save branches when 'errors' is NULL;
1295 this gets set and discarded */
1297 /* The below are used only if there is both an overlong malformation and a
1298 * too short one. Otherwise the first two are set to 's0' and 'send', and
1299 * the third not used at all */
1300 U8 * adjusted_s0 = (U8 *) s0;
1301 U8 temp_char_buf[UTF8_MAXBYTES + 1]; /* Used to avoid a Newx in this
1302 routine; see [perl #130921] */
1303 UV uv_so_far = 0; /* (Initialized to silence compilers' wrong warning) */
1305 PERL_ARGS_ASSERT_UTF8N_TO_UVCHR_ERROR;
1311 errors = &discard_errors;
1314 /* The order of malformation tests here is important. We should consume as
1315 * few bytes as possible in order to not skip any valid character. This is
1316 * required by the Unicode Standard (section 3.9 of Unicode 6.0); see also
1317 * http://unicode.org/reports/tr36 for more discussion as to why. For
1318 * example, once we've done a UTF8SKIP, we can tell the expected number of
1319 * bytes, and could fail right off the bat if the input parameters indicate
1320 * that there are too few available. But it could be that just that first
1321 * byte is garbled, and the intended character occupies fewer bytes. If we
1322 * blindly assumed that the first byte is correct, and skipped based on
1323 * that number, we could skip over a valid input character. So instead, we
1324 * always examine the sequence byte-by-byte.
1326 * We also should not consume too few bytes, otherwise someone could inject
1327 * things. For example, an input could be deliberately designed to
1328 * overflow, and if this code bailed out immediately upon discovering that,
1329 * returning to the caller C<*retlen> pointing to the very next byte (one
1330 * which is actually part of of the overflowing sequence), that could look
1331 * legitimate to the caller, which could discard the initial partial
1332 * sequence and process the rest, inappropriately.
1334 * Some possible input sequences are malformed in more than one way. This
1335 * function goes to lengths to try to find all of them. This is necessary
1336 * for correctness, as the inputs may allow one malformation but not
1337 * another, and if we abandon searching for others after finding the
1338 * allowed one, we could allow in something that shouldn't have been.
1341 if (UNLIKELY(curlen == 0)) {
1342 possible_problems |= UTF8_GOT_EMPTY;
1344 uv = UNICODE_REPLACEMENT;
1345 goto ready_to_handle_errors;
1348 expectlen = UTF8SKIP(s);
1350 /* A well-formed UTF-8 character, as the vast majority of calls to this
1351 * function will be for, has this expected length. For efficiency, set
1352 * things up here to return it. It will be overriden only in those rare
1353 * cases where a malformation is found */
1355 *retlen = expectlen;
1358 /* An invariant is trivially well-formed */
1359 if (UTF8_IS_INVARIANT(uv)) {
1363 /* A continuation character can't start a valid sequence */
1364 if (UNLIKELY(UTF8_IS_CONTINUATION(uv))) {
1365 possible_problems |= UTF8_GOT_CONTINUATION;
1367 uv = UNICODE_REPLACEMENT;
1368 goto ready_to_handle_errors;
1371 /* Here is not a continuation byte, nor an invariant. The only thing left
1372 * is a start byte (possibly for an overlong). (We can't use UTF8_IS_START
1373 * because it excludes start bytes like \xC0 that always lead to
1376 /* Convert to I8 on EBCDIC (no-op on ASCII), then remove the leading bits
1377 * that indicate the number of bytes in the character's whole UTF-8
1378 * sequence, leaving just the bits that are part of the value. */
1379 uv = NATIVE_UTF8_TO_I8(uv) & UTF_START_MASK(expectlen);
1381 /* Setup the loop end point, making sure to not look past the end of the
1382 * input string, and flag it as too short if the size isn't big enough. */
1384 if (UNLIKELY(curlen < expectlen)) {
1385 possible_problems |= UTF8_GOT_SHORT;
1393 /* Now, loop through the remaining bytes in the character's sequence,
1394 * accumulating each into the working value as we go. */
1395 for (s = s0 + 1; s < send; s++) {
1396 if (LIKELY(UTF8_IS_CONTINUATION(*s))) {
1397 uv = UTF8_ACCUMULATE(uv, *s);
1401 /* Here, found a non-continuation before processing all expected bytes.
1402 * This byte indicates the beginning of a new character, so quit, even
1403 * if allowing this malformation. */
1404 possible_problems |= UTF8_GOT_NON_CONTINUATION;
1406 } /* End of loop through the character's bytes */
1408 /* Save how many bytes were actually in the character */
1411 /* Note that there are two types of too-short malformation. One is when
1412 * there is actual wrong data before the normal termination of the
1413 * sequence. The other is that the sequence wasn't complete before the end
1414 * of the data we are allowed to look at, based on the input 'curlen'.
1415 * This means that we were passed data for a partial character, but it is
1416 * valid as far as we saw. The other is definitely invalid. This
1417 * distinction could be important to a caller, so the two types are kept
1420 * A convenience macro that matches either of the too-short conditions. */
1421 # define UTF8_GOT_TOO_SHORT (UTF8_GOT_SHORT|UTF8_GOT_NON_CONTINUATION)
1423 if (UNLIKELY(possible_problems & UTF8_GOT_TOO_SHORT)) {
1425 uv = UNICODE_REPLACEMENT;
1428 /* Check for overflow. The algorithm requires us to not look past the end
1429 * of the current character, even if partial, so the upper limit is 's' */
1430 if (UNLIKELY(0 < does_utf8_overflow(s0, s,
1431 1 /* Do consider overlongs */
1434 possible_problems |= UTF8_GOT_OVERFLOW;
1435 uv = UNICODE_REPLACEMENT;
1438 /* Check for overlong. If no problems so far, 'uv' is the correct code
1439 * point value. Simply see if it is expressible in fewer bytes. Otherwise
1440 * we must look at the UTF-8 byte sequence itself to see if it is for an
1442 if ( ( LIKELY(! possible_problems)
1443 && UNLIKELY(expectlen > (STRLEN) OFFUNISKIP(uv)))
1444 || ( UNLIKELY(possible_problems)
1445 && ( UNLIKELY(! UTF8_IS_START(*s0))
1447 && UNLIKELY(0 < is_utf8_overlong_given_start_byte_ok(s0,
1450 possible_problems |= UTF8_GOT_LONG;
1452 if ( UNLIKELY( possible_problems & UTF8_GOT_TOO_SHORT)
1454 /* The calculation in the 'true' branch of this 'if'
1455 * below won't work if overflows, and isn't needed
1456 * anyway. Further below we handle all overflow
1458 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW)))
1460 UV min_uv = uv_so_far;
1463 /* Here, the input is both overlong and is missing some trailing
1464 * bytes. There is no single code point it could be for, but there
1465 * may be enough information present to determine if what we have
1466 * so far is for an unallowed code point, such as for a surrogate.
1467 * The code further below has the intelligence to determine this,
1468 * but just for non-overlong UTF-8 sequences. What we do here is
1469 * calculate the smallest code point the input could represent if
1470 * there were no too short malformation. Then we compute and save
1471 * the UTF-8 for that, which is what the code below looks at
1472 * instead of the raw input. It turns out that the smallest such
1473 * code point is all we need. */
1474 for (i = curlen; i < expectlen; i++) {
1475 min_uv = UTF8_ACCUMULATE(min_uv,
1476 I8_TO_NATIVE_UTF8(UTF_CONTINUATION_MARK));
1479 adjusted_s0 = temp_char_buf;
1480 (void) uvoffuni_to_utf8_flags(adjusted_s0, min_uv, 0);
1484 /* Here, we have found all the possible problems, except for when the input
1485 * is for a problematic code point not allowed by the input parameters. */
1487 /* uv is valid for overlongs */
1488 if ( ( ( LIKELY(! (possible_problems & ~UTF8_GOT_LONG))
1490 /* isn't problematic if < this */
1491 && uv >= UNICODE_SURROGATE_FIRST)
1492 || ( UNLIKELY(possible_problems)
1494 /* if overflow, we know without looking further
1495 * precisely which of the problematic types it is,
1496 * and we deal with those in the overflow handling
1498 && LIKELY(! (possible_problems & UTF8_GOT_OVERFLOW))
1499 && ( isUTF8_POSSIBLY_PROBLEMATIC(*adjusted_s0)
1500 || UNLIKELY(isUTF8_PERL_EXTENDED(s0)))))
1501 && ((flags & ( UTF8_DISALLOW_NONCHAR
1502 |UTF8_DISALLOW_SURROGATE
1503 |UTF8_DISALLOW_SUPER
1504 |UTF8_DISALLOW_PERL_EXTENDED
1506 |UTF8_WARN_SURROGATE
1508 |UTF8_WARN_PERL_EXTENDED))))
1510 /* If there were no malformations, or the only malformation is an
1511 * overlong, 'uv' is valid */
1512 if (LIKELY(! (possible_problems & ~UTF8_GOT_LONG))) {
1513 if (UNLIKELY(UNICODE_IS_SURROGATE(uv))) {
1514 possible_problems |= UTF8_GOT_SURROGATE;
1516 else if (UNLIKELY(uv > PERL_UNICODE_MAX)) {
1517 possible_problems |= UTF8_GOT_SUPER;
1519 else if (UNLIKELY(UNICODE_IS_NONCHAR(uv))) {
1520 possible_problems |= UTF8_GOT_NONCHAR;
1523 else { /* Otherwise, need to look at the source UTF-8, possibly
1524 adjusted to be non-overlong */
1526 if (UNLIKELY(NATIVE_UTF8_TO_I8(*adjusted_s0)
1527 >= FIRST_START_BYTE_THAT_IS_DEFINITELY_SUPER))
1529 possible_problems |= UTF8_GOT_SUPER;
1531 else if (curlen > 1) {
1532 if (UNLIKELY(IS_UTF8_2_BYTE_SUPER(
1533 NATIVE_UTF8_TO_I8(*adjusted_s0),
1534 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1536 possible_problems |= UTF8_GOT_SUPER;
1538 else if (UNLIKELY(IS_UTF8_2_BYTE_SURROGATE(
1539 NATIVE_UTF8_TO_I8(*adjusted_s0),
1540 NATIVE_UTF8_TO_I8(*(adjusted_s0 + 1)))))
1542 possible_problems |= UTF8_GOT_SURROGATE;
1546 /* We need a complete well-formed UTF-8 character to discern
1547 * non-characters, so can't look for them here */
1551 ready_to_handle_errors:
1554 * curlen contains the number of bytes in the sequence that
1555 * this call should advance the input by.
1556 * avail_len gives the available number of bytes passed in, but
1557 * only if this is less than the expected number of
1558 * bytes, based on the code point's start byte.
1559 * possible_problems' is 0 if there weren't any problems; otherwise a bit
1560 * is set in it for each potential problem found.
1561 * uv contains the code point the input sequence
1562 * represents; or if there is a problem that prevents
1563 * a well-defined value from being computed, it is
1564 * some subsitute value, typically the REPLACEMENT
1566 * s0 points to the first byte of the character
1567 * s points to just after were we left off processing
1569 * send points to just after where that character should
1570 * end, based on how many bytes the start byte tells
1571 * us should be in it, but no further than s0 +
1575 if (UNLIKELY(possible_problems)) {
1576 bool disallowed = FALSE;
1577 const U32 orig_problems = possible_problems;
1579 while (possible_problems) { /* Handle each possible problem */
1581 char * message = NULL;
1583 /* Each 'if' clause handles one problem. They are ordered so that
1584 * the first ones' messages will be displayed before the later
1585 * ones; this is kinda in decreasing severity order. But the
1586 * overlong must come last, as it changes 'uv' looked at by the
1588 if (possible_problems & UTF8_GOT_OVERFLOW) {
1590 /* Overflow means also got a super and are using Perl's
1591 * extended UTF-8, but we handle all three cases here */
1593 &= ~(UTF8_GOT_OVERFLOW|UTF8_GOT_SUPER|UTF8_GOT_PERL_EXTENDED);
1594 *errors |= UTF8_GOT_OVERFLOW;
1596 /* But the API says we flag all errors found */
1597 if (flags & (UTF8_WARN_SUPER|UTF8_DISALLOW_SUPER)) {
1598 *errors |= UTF8_GOT_SUPER;
1601 & (UTF8_WARN_PERL_EXTENDED|UTF8_DISALLOW_PERL_EXTENDED))
1603 *errors |= UTF8_GOT_PERL_EXTENDED;
1606 /* Disallow if any of the three categories say to */
1607 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1608 || (flags & ( UTF8_DISALLOW_SUPER
1609 |UTF8_DISALLOW_PERL_EXTENDED)))
1614 /* Likewise, warn if any say to */
1615 if ( ! (flags & UTF8_ALLOW_OVERFLOW)
1616 || (flags & (UTF8_WARN_SUPER|UTF8_WARN_PERL_EXTENDED)))
1619 /* The warnings code explicitly says it doesn't handle the
1620 * case of packWARN2 and two categories which have
1621 * parent-child relationship. Even if it works now to
1622 * raise the warning if either is enabled, it wouldn't
1623 * necessarily do so in the future. We output (only) the
1624 * most dire warning */
1625 if (! (flags & UTF8_CHECK_ONLY)) {
1626 if (ckWARN_d(WARN_UTF8)) {
1627 pack_warn = packWARN(WARN_UTF8);
1629 else if (ckWARN_d(WARN_NON_UNICODE)) {
1630 pack_warn = packWARN(WARN_NON_UNICODE);
1633 message = Perl_form(aTHX_ "%s: %s (overflows)",
1635 _byte_dump_string(s0, curlen, 0));
1640 else if (possible_problems & UTF8_GOT_EMPTY) {
1641 possible_problems &= ~UTF8_GOT_EMPTY;
1642 *errors |= UTF8_GOT_EMPTY;
1644 if (! (flags & UTF8_ALLOW_EMPTY)) {
1646 /* This so-called malformation is now treated as a bug in
1647 * the caller. If you have nothing to decode, skip calling
1652 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1653 pack_warn = packWARN(WARN_UTF8);
1654 message = Perl_form(aTHX_ "%s (empty string)",
1659 else if (possible_problems & UTF8_GOT_CONTINUATION) {
1660 possible_problems &= ~UTF8_GOT_CONTINUATION;
1661 *errors |= UTF8_GOT_CONTINUATION;
1663 if (! (flags & UTF8_ALLOW_CONTINUATION)) {
1665 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1666 pack_warn = packWARN(WARN_UTF8);
1667 message = Perl_form(aTHX_
1668 "%s: %s (unexpected continuation byte 0x%02x,"
1669 " with no preceding start byte)",
1671 _byte_dump_string(s0, 1, 0), *s0);
1675 else if (possible_problems & UTF8_GOT_SHORT) {
1676 possible_problems &= ~UTF8_GOT_SHORT;
1677 *errors |= UTF8_GOT_SHORT;
1679 if (! (flags & UTF8_ALLOW_SHORT)) {
1681 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1682 pack_warn = packWARN(WARN_UTF8);
1683 message = Perl_form(aTHX_
1684 "%s: %s (too short; %d byte%s available, need %d)",
1686 _byte_dump_string(s0, send - s0, 0),
1688 avail_len == 1 ? "" : "s",
1694 else if (possible_problems & UTF8_GOT_NON_CONTINUATION) {
1695 possible_problems &= ~UTF8_GOT_NON_CONTINUATION;
1696 *errors |= UTF8_GOT_NON_CONTINUATION;
1698 if (! (flags & UTF8_ALLOW_NON_CONTINUATION)) {
1700 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1702 /* If we don't know for sure that the input length is
1703 * valid, avoid as much as possible reading past the
1704 * end of the buffer */
1705 int printlen = (flags & _UTF8_NO_CONFIDENCE_IN_CURLEN)
1708 pack_warn = packWARN(WARN_UTF8);
1709 message = Perl_form(aTHX_ "%s",
1710 unexpected_non_continuation_text(s0,
1717 else if (possible_problems & UTF8_GOT_SURROGATE) {
1718 possible_problems &= ~UTF8_GOT_SURROGATE;
1720 if (flags & UTF8_WARN_SURROGATE) {
1721 *errors |= UTF8_GOT_SURROGATE;
1723 if ( ! (flags & UTF8_CHECK_ONLY)
1724 && ckWARN_d(WARN_SURROGATE))
1726 pack_warn = packWARN(WARN_SURROGATE);
1728 /* These are the only errors that can occur with a
1729 * surrogate when the 'uv' isn't valid */
1730 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1731 message = Perl_form(aTHX_
1732 "UTF-16 surrogate (any UTF-8 sequence that"
1733 " starts with \"%s\" is for a surrogate)",
1734 _byte_dump_string(s0, curlen, 0));
1737 message = Perl_form(aTHX_ surrogate_cp_format, uv);
1742 if (flags & UTF8_DISALLOW_SURROGATE) {
1744 *errors |= UTF8_GOT_SURROGATE;
1747 else if (possible_problems & UTF8_GOT_SUPER) {
1748 possible_problems &= ~UTF8_GOT_SUPER;
1750 if (flags & UTF8_WARN_SUPER) {
1751 *errors |= UTF8_GOT_SUPER;
1753 if ( ! (flags & UTF8_CHECK_ONLY)
1754 && ckWARN_d(WARN_NON_UNICODE))
1756 pack_warn = packWARN(WARN_NON_UNICODE);
1758 if (orig_problems & UTF8_GOT_TOO_SHORT) {
1759 message = Perl_form(aTHX_
1760 "Any UTF-8 sequence that starts with"
1761 " \"%s\" is for a non-Unicode code point,"
1762 " may not be portable",
1763 _byte_dump_string(s0, curlen, 0));
1766 message = Perl_form(aTHX_ super_cp_format, uv);
1771 /* Test for Perl's extended UTF-8 after the regular SUPER ones,
1772 * and before possibly bailing out, so that the more dire
1773 * warning will override the regular one. */
1774 if (UNLIKELY(isUTF8_PERL_EXTENDED(s0))) {
1775 if ( ! (flags & UTF8_CHECK_ONLY)
1776 && (flags & (UTF8_WARN_PERL_EXTENDED|UTF8_WARN_SUPER))
1777 && ckWARN_d(WARN_NON_UNICODE))
1779 pack_warn = packWARN(WARN_NON_UNICODE);
1781 /* If it is an overlong that evaluates to a code point
1782 * that doesn't have to use the Perl extended UTF-8, it
1783 * still used it, and so we output a message that
1784 * doesn't refer to the code point. The same is true
1785 * if there was a SHORT malformation where the code
1786 * point is not valid. In that case, 'uv' will have
1787 * been set to the REPLACEMENT CHAR, and the message
1788 * below without the code point in it will be selected
1790 if (UNICODE_IS_PERL_EXTENDED(uv)) {
1791 message = Perl_form(aTHX_
1792 perl_extended_cp_format, uv);
1795 message = Perl_form(aTHX_
1796 "Any UTF-8 sequence that starts with"
1797 " \"%s\" is a Perl extension, and"
1798 " so is not portable",
1799 _byte_dump_string(s0, curlen, 0));
1803 if (flags & ( UTF8_WARN_PERL_EXTENDED
1804 |UTF8_DISALLOW_PERL_EXTENDED))
1806 *errors |= UTF8_GOT_PERL_EXTENDED;
1808 if (flags & UTF8_DISALLOW_PERL_EXTENDED) {
1814 if (flags & UTF8_DISALLOW_SUPER) {
1815 *errors |= UTF8_GOT_SUPER;
1819 else if (possible_problems & UTF8_GOT_NONCHAR) {
1820 possible_problems &= ~UTF8_GOT_NONCHAR;
1822 if (flags & UTF8_WARN_NONCHAR) {
1823 *errors |= UTF8_GOT_NONCHAR;
1825 if ( ! (flags & UTF8_CHECK_ONLY)
1826 && ckWARN_d(WARN_NONCHAR))
1828 /* The code above should have guaranteed that we don't
1829 * get here with errors other than overlong */
1830 assert (! (orig_problems
1831 & ~(UTF8_GOT_LONG|UTF8_GOT_NONCHAR)));
1833 pack_warn = packWARN(WARN_NONCHAR);
1834 message = Perl_form(aTHX_ nonchar_cp_format, uv);
1838 if (flags & UTF8_DISALLOW_NONCHAR) {
1840 *errors |= UTF8_GOT_NONCHAR;
1843 else if (possible_problems & UTF8_GOT_LONG) {
1844 possible_problems &= ~UTF8_GOT_LONG;
1845 *errors |= UTF8_GOT_LONG;
1847 if (flags & UTF8_ALLOW_LONG) {
1849 /* We don't allow the actual overlong value, unless the
1850 * special extra bit is also set */
1851 if (! (flags & ( UTF8_ALLOW_LONG_AND_ITS_VALUE
1852 & ~UTF8_ALLOW_LONG)))
1854 uv = UNICODE_REPLACEMENT;
1860 if (ckWARN_d(WARN_UTF8) && ! (flags & UTF8_CHECK_ONLY)) {
1861 pack_warn = packWARN(WARN_UTF8);
1863 /* These error types cause 'uv' to be something that
1864 * isn't what was intended, so can't use it in the
1865 * message. The other error types either can't
1866 * generate an overlong, or else the 'uv' is valid */
1868 (UTF8_GOT_TOO_SHORT|UTF8_GOT_OVERFLOW))
1870 message = Perl_form(aTHX_
1871 "%s: %s (any UTF-8 sequence that starts"
1872 " with \"%s\" is overlong which can and"
1873 " should be represented with a"
1874 " different, shorter sequence)",
1876 _byte_dump_string(s0, send - s0, 0),
1877 _byte_dump_string(s0, curlen, 0));
1880 U8 tmpbuf[UTF8_MAXBYTES+1];
1881 const U8 * const e = uvoffuni_to_utf8_flags(tmpbuf,
1883 /* Don't use U+ for non-Unicode code points, which
1884 * includes those in the Latin1 range */
1885 const char * preface = ( uv > PERL_UNICODE_MAX
1892 message = Perl_form(aTHX_
1893 "%s: %s (overlong; instead use %s to represent"
1896 _byte_dump_string(s0, send - s0, 0),
1897 _byte_dump_string(tmpbuf, e - tmpbuf, 0),
1899 ((uv < 256) ? 2 : 4), /* Field width of 2 for
1900 small code points */
1905 } /* End of looking through the possible flags */
1907 /* Display the message (if any) for the problem being handled in
1908 * this iteration of the loop */
1911 Perl_warner(aTHX_ pack_warn, "%s in %s", message,
1914 Perl_warner(aTHX_ pack_warn, "%s", message);
1916 } /* End of 'while (possible_problems)' */
1918 /* Since there was a possible problem, the returned length may need to
1919 * be changed from the one stored at the beginning of this function.
1920 * Instead of trying to figure out if that's needed, just do it. */
1926 if (flags & UTF8_CHECK_ONLY && retlen) {
1927 *retlen = ((STRLEN) -1);
1933 return UNI_TO_NATIVE(uv);
1937 =for apidoc utf8_to_uvchr_buf
1939 Returns the native code point of the first character in the string C<s> which
1940 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1941 C<*retlen> will be set to the length, in bytes, of that character.
1943 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1944 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1945 C<NULL>) to -1. If those warnings are off, the computed value, if well-defined
1946 (or the Unicode REPLACEMENT CHARACTER if not), is silently returned, and
1947 C<*retlen> is set (if C<retlen> isn't C<NULL>) so that (S<C<s> + C<*retlen>>) is
1948 the next possible position in C<s> that could begin a non-malformed character.
1949 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is
1954 Also implemented as a macro in utf8.h
1960 Perl_utf8_to_uvchr_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1962 PERL_ARGS_ASSERT_UTF8_TO_UVCHR_BUF;
1966 return utf8n_to_uvchr(s, send - s, retlen,
1967 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY);
1970 /* This is marked as deprecated
1972 =for apidoc utf8_to_uvuni_buf
1974 Only in very rare circumstances should code need to be dealing in Unicode
1975 (as opposed to native) code points. In those few cases, use
1976 C<L<NATIVE_TO_UNI(utf8_to_uvchr_buf(...))|/utf8_to_uvchr_buf>> instead.
1978 Returns the Unicode (not-native) code point of the first character in the
1980 is assumed to be in UTF-8 encoding; C<send> points to 1 beyond the end of C<s>.
1981 C<retlen> will be set to the length, in bytes, of that character.
1983 If C<s> does not point to a well-formed UTF-8 character and UTF8 warnings are
1984 enabled, zero is returned and C<*retlen> is set (if C<retlen> isn't
1985 NULL) to -1. If those warnings are off, the computed value if well-defined (or
1986 the Unicode REPLACEMENT CHARACTER, if not) is silently returned, and C<*retlen>
1987 is set (if C<retlen> isn't NULL) so that (S<C<s> + C<*retlen>>) is the
1988 next possible position in C<s> that could begin a non-malformed character.
1989 See L</utf8n_to_uvchr> for details on when the REPLACEMENT CHARACTER is returned.
1995 Perl_utf8_to_uvuni_buf(pTHX_ const U8 *s, const U8 *send, STRLEN *retlen)
1997 PERL_ARGS_ASSERT_UTF8_TO_UVUNI_BUF;
2001 return NATIVE_TO_UNI(utf8_to_uvchr_buf(s, send, retlen));
2005 =for apidoc utf8_length
2007 Return the length of the UTF-8 char encoded string C<s> in characters.
2008 Stops at C<e> (i.e. the C<*e> byte does not form part of the character).
2009 If C<e E<lt> s> or if the scan would end up past C<e>, it croaks.
2015 Perl_utf8_length(pTHX_ const U8 *s, const U8 *e)
2019 PERL_ARGS_ASSERT_UTF8_LENGTH;
2021 /* Note: cannot use UTF8_IS_...() too eagerly here since e.g.
2022 * the bitops (especially ~) can create illegal UTF-8.
2023 * In other words: in Perl UTF-8 is not just for Unicode. */
2026 goto warn_and_return;
2036 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2037 "%s in %s", unees, OP_DESC(PL_op));
2039 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2046 =for apidoc bytes_cmp_utf8
2048 Compares the sequence of characters (stored as octets) in C<b>, C<blen> with the
2049 sequence of characters (stored as UTF-8)
2050 in C<u>, C<ulen>. Returns 0 if they are
2051 equal, -1 or -2 if the first string is less than the second string, +1 or +2
2052 if the first string is greater than the second string.
2054 -1 or +1 is returned if the shorter string was identical to the start of the
2055 longer string. -2 or +2 is returned if
2056 there was a difference between characters
2063 Perl_bytes_cmp_utf8(pTHX_ const U8 *b, STRLEN blen, const U8 *u, STRLEN ulen)
2065 const U8 *const bend = b + blen;
2066 const U8 *const uend = u + ulen;
2068 PERL_ARGS_ASSERT_BYTES_CMP_UTF8;
2070 while (b < bend && u < uend) {
2072 if (!UTF8_IS_INVARIANT(c)) {
2073 if (UTF8_IS_DOWNGRADEABLE_START(c)) {
2076 if (UTF8_IS_CONTINUATION(c1)) {
2077 c = EIGHT_BIT_UTF8_TO_NATIVE(c, c1);
2079 /* diag_listed_as: Malformed UTF-8 character%s */
2080 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2082 unexpected_non_continuation_text(u - 2, 2, 1, 2),
2083 PL_op ? " in " : "",
2084 PL_op ? OP_DESC(PL_op) : "");
2089 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8),
2090 "%s in %s", unees, OP_DESC(PL_op));
2092 Perl_ck_warner_d(aTHX_ packWARN(WARN_UTF8), "%s", unees);
2093 return -2; /* Really want to return undef :-) */
2100 return *b < c ? -2 : +2;
2105 if (b == bend && u == uend)
2108 return b < bend ? +1 : -1;
2112 =for apidoc utf8_to_bytes
2114 Converts a string C<"s"> of length C<*lenp> from UTF-8 into native byte encoding.
2115 Unlike L</bytes_to_utf8>, this over-writes the original string, and
2116 updates C<*lenp> to contain the new length.
2117 Returns zero on failure (leaving C<"s"> unchanged) setting C<*lenp> to -1.
2119 Upon successful return, the number of variants in the string can be computed by
2120 having saved the value of C<*lenp> before the call, and subtracting the
2121 after-call value of C<*lenp> from it.
2123 If you need a copy of the string, see L</bytes_from_utf8>.
2129 Perl_utf8_to_bytes(pTHX_ U8 *s, STRLEN *lenp)
2133 PERL_ARGS_ASSERT_UTF8_TO_BYTES;
2134 PERL_UNUSED_CONTEXT;
2136 /* This is a no-op if no variants at all in the input */
2137 if (is_utf8_invariant_string_loc(s, *lenp, (const U8 **) &first_variant)) {
2142 U8 * const save = s;
2143 U8 * const send = s + *lenp;
2146 /* Nothing before the first variant needs to be changed, so start the real
2150 if (! UTF8_IS_INVARIANT(*s)) {
2151 if (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, send)) {
2152 *lenp = ((STRLEN) -1);
2160 /* Is downgradable, so do it */
2161 d = s = first_variant;
2164 if (! UVCHR_IS_INVARIANT(c)) {
2165 /* Then it is two-byte encoded */
2166 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2179 =for apidoc bytes_from_utf8
2181 Converts a potentially UTF-8 encoded string C<s> of length C<*lenp> into native
2182 byte encoding. On input, the boolean C<*is_utf8p> gives whether or not C<s> is
2183 actually encoded in UTF-8.
2185 Unlike L</utf8_to_bytes> but like L</bytes_to_utf8>, this is non-destructive of
2188 Do nothing if C<*is_utf8p> is 0, or if there are code points in the string
2189 not expressible in native byte encoding. In these cases, C<*is_utf8p> and
2190 C<*lenp> are unchanged, and the return value is the original C<s>.
2192 Otherwise, C<*is_utf8p> is set to 0, and the return value is a pointer to a
2193 newly created string containing a downgraded copy of C<s>, and whose length is
2194 returned in C<*lenp>, updated. The new string is C<NUL>-terminated.
2196 Upon successful return, the number of variants in the string can be computed by
2197 having saved the value of C<*lenp> before the call, and subtracting the
2198 after-call value of C<*lenp> from it.
2202 There is a macro that avoids this function call, but this is retained for
2203 anyone who calls it with the Perl_ prefix */
2206 Perl_bytes_from_utf8(pTHX_ const U8 *s, STRLEN *lenp, bool *is_utf8p)
2208 PERL_ARGS_ASSERT_BYTES_FROM_UTF8;
2209 PERL_UNUSED_CONTEXT;
2211 return bytes_from_utf8_loc(s, lenp, is_utf8p, NULL);
2215 No = here because currently externally undocumented
2216 for apidoc bytes_from_utf8_loc
2218 Like C<L</bytes_from_utf8>()>, but takes an extra parameter, a pointer to where
2219 to store the location of the first character in C<"s"> that cannot be
2220 converted to non-UTF8.
2222 If that parameter is C<NULL>, this function behaves identically to
2225 Otherwise if C<*is_utf8p> is 0 on input, the function behaves identically to
2226 C<bytes_from_utf8>, except it also sets C<*first_non_downgradable> to C<NULL>.
2228 Otherwise, the function returns a newly created C<NUL>-terminated string
2229 containing the non-UTF8 equivalent of the convertible first portion of
2230 C<"s">. C<*lenp> is set to its length, not including the terminating C<NUL>.
2231 If the entire input string was converted, C<*is_utf8p> is set to a FALSE value,
2232 and C<*first_non_downgradable> is set to C<NULL>.
2234 Otherwise, C<*first_non_downgradable> set to point to the first byte of the
2235 first character in the original string that wasn't converted. C<*is_utf8p> is
2236 unchanged. Note that the new string may have length 0.
2238 Another way to look at it is, if C<*first_non_downgradable> is non-C<NULL> and
2239 C<*is_utf8p> is TRUE, this function starts at the beginning of C<"s"> and
2240 converts as many characters in it as possible stopping at the first one it
2241 finds that can't be converted to non-UTF-8. C<*first_non_downgradable> is
2242 set to point to that. The function returns the portion that could be converted
2243 in a newly created C<NUL>-terminated string, and C<*lenp> is set to its length,
2244 not including the terminating C<NUL>. If the very first character in the
2245 original could not be converted, C<*lenp> will be 0, and the new string will
2246 contain just a single C<NUL>. If the entire input string was converted,
2247 C<*is_utf8p> is set to FALSE and C<*first_non_downgradable> is set to C<NULL>.
2249 Upon successful return, the number of variants in the converted portion of the
2250 string can be computed by having saved the value of C<*lenp> before the call,
2251 and subtracting the after-call value of C<*lenp> from it.
2259 Perl_bytes_from_utf8_loc(const U8 *s, STRLEN *lenp, bool *is_utf8p, const U8** first_unconverted)
2262 const U8 *original = s;
2263 U8 *converted_start;
2264 const U8 *send = s + *lenp;
2266 PERL_ARGS_ASSERT_BYTES_FROM_UTF8_LOC;
2269 if (first_unconverted) {
2270 *first_unconverted = NULL;
2273 return (U8 *) original;
2276 Newx(d, (*lenp) + 1, U8);
2278 converted_start = d;
2281 if (! UTF8_IS_INVARIANT(c)) {
2283 /* Then it is multi-byte encoded. If the code point is above 0xFF,
2284 * have to stop now */
2285 if (UNLIKELY (! UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s - 1, send))) {
2286 if (first_unconverted) {
2287 *first_unconverted = s - 1;
2288 goto finish_and_return;
2291 Safefree(converted_start);
2292 return (U8 *) original;
2296 c = EIGHT_BIT_UTF8_TO_NATIVE(c, *s);
2302 /* Here, converted the whole of the input */
2304 if (first_unconverted) {
2305 *first_unconverted = NULL;
2310 *lenp = d - converted_start;
2312 /* Trim unused space */
2313 Renew(converted_start, *lenp + 1, U8);
2315 return converted_start;
2319 =for apidoc bytes_to_utf8
2321 Converts a string C<s> of length C<*lenp> bytes from the native encoding into
2323 Returns a pointer to the newly-created string, and sets C<*lenp> to
2324 reflect the new length in bytes.
2326 Upon successful return, the number of variants in the string can be computed by
2327 having saved the value of C<*lenp> before the call, and subtracting it from the
2328 after-call value of C<*lenp>.
2330 A C<NUL> character will be written after the end of the string.
2332 If you want to convert to UTF-8 from encodings other than
2333 the native (Latin1 or EBCDIC),
2334 see L</sv_recode_to_utf8>().
2340 Perl_bytes_to_utf8(pTHX_ const U8 *s, STRLEN *lenp)
2342 const U8 * const send = s + (*lenp);
2346 PERL_ARGS_ASSERT_BYTES_TO_UTF8;
2347 PERL_UNUSED_CONTEXT;
2349 Newx(d, (*lenp) * 2 + 1, U8);
2353 append_utf8_from_native_byte(*s, &d);
2360 /* Trim unused space */
2361 Renew(dst, *lenp + 1, U8);
2367 * Convert native (big-endian) UTF-16 to UTF-8. For reversed (little-endian),
2368 * use utf16_to_utf8_reversed().
2370 * UTF-16 requires 2 bytes for every code point below 0x10000; otherwise 4 bytes.
2371 * UTF-8 requires 1-3 bytes for every code point below 0x1000; otherwise 4 bytes.
2372 * UTF-EBCDIC requires 1-4 bytes for every code point below 0x1000; otherwise 4-5 bytes.
2374 * These functions don't check for overflow. The worst case is every code
2375 * point in the input is 2 bytes, and requires 4 bytes on output. (If the code
2376 * is never going to run in EBCDIC, it is 2 bytes requiring 3 on output.) Therefore the
2377 * destination must be pre-extended to 2 times the source length.
2379 * Do not use in-place. We optimize for native, for obvious reasons. */
2382 Perl_utf16_to_utf8(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2387 PERL_ARGS_ASSERT_UTF16_TO_UTF8;
2390 Perl_croak(aTHX_ "panic: utf16_to_utf8: odd bytelen %" UVuf,
2396 UV uv = (p[0] << 8) + p[1]; /* UTF-16BE */
2398 if (OFFUNI_IS_INVARIANT(uv)) {
2399 *d++ = LATIN1_TO_NATIVE((U8) uv);
2402 if (uv <= MAX_UTF8_TWO_BYTE) {
2403 *d++ = UTF8_TWO_BYTE_HI(UNI_TO_NATIVE(uv));
2404 *d++ = UTF8_TWO_BYTE_LO(UNI_TO_NATIVE(uv));
2408 #define FIRST_HIGH_SURROGATE UNICODE_SURROGATE_FIRST
2409 #define LAST_HIGH_SURROGATE 0xDBFF
2410 #define FIRST_LOW_SURROGATE 0xDC00
2411 #define LAST_LOW_SURROGATE UNICODE_SURROGATE_LAST
2412 #define FIRST_IN_PLANE1 0x10000
2414 /* This assumes that most uses will be in the first Unicode plane, not
2415 * needing surrogates */
2416 if (UNLIKELY(uv >= UNICODE_SURROGATE_FIRST
2417 && uv <= UNICODE_SURROGATE_LAST))
2419 if (UNLIKELY(p >= pend) || UNLIKELY(uv > LAST_HIGH_SURROGATE)) {
2420 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2423 UV low = (p[0] << 8) + p[1];
2424 if ( UNLIKELY(low < FIRST_LOW_SURROGATE)
2425 || UNLIKELY(low > LAST_LOW_SURROGATE))
2427 Perl_croak(aTHX_ "Malformed UTF-16 surrogate");
2430 uv = ((uv - FIRST_HIGH_SURROGATE) << 10)
2431 + (low - FIRST_LOW_SURROGATE) + FIRST_IN_PLANE1;
2435 d = uvoffuni_to_utf8_flags(d, uv, 0);
2437 if (uv < FIRST_IN_PLANE1) {
2438 *d++ = (U8)(( uv >> 12) | 0xe0);
2439 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2440 *d++ = (U8)(( uv & 0x3f) | 0x80);
2444 *d++ = (U8)(( uv >> 18) | 0xf0);
2445 *d++ = (U8)(((uv >> 12) & 0x3f) | 0x80);
2446 *d++ = (U8)(((uv >> 6) & 0x3f) | 0x80);
2447 *d++ = (U8)(( uv & 0x3f) | 0x80);
2452 *newlen = d - dstart;
2456 /* Note: this one is slightly destructive of the source. */
2459 Perl_utf16_to_utf8_reversed(pTHX_ U8* p, U8* d, I32 bytelen, I32 *newlen)
2462 U8* const send = s + bytelen;
2464 PERL_ARGS_ASSERT_UTF16_TO_UTF8_REVERSED;
2467 Perl_croak(aTHX_ "panic: utf16_to_utf8_reversed: odd bytelen %" UVuf,
2471 const U8 tmp = s[0];
2476 return utf16_to_utf8(p, d, bytelen, newlen);
2480 Perl__is_uni_FOO(pTHX_ const U8 classnum, const UV c)
2482 U8 tmpbuf[UTF8_MAXBYTES+1];
2483 uvchr_to_utf8(tmpbuf, c);
2484 return _is_utf8_FOO_with_len(classnum, tmpbuf, tmpbuf + sizeof(tmpbuf));
2487 /* Internal function so we can deprecate the external one, and call
2488 this one from other deprecated functions in this file */
2491 Perl__is_utf8_idstart(pTHX_ const U8 *p)
2493 PERL_ARGS_ASSERT__IS_UTF8_IDSTART;
2497 return is_utf8_common(p, &PL_utf8_idstart, "IdStart", NULL);
2501 Perl__is_uni_perl_idcont(pTHX_ UV c)
2503 U8 tmpbuf[UTF8_MAXBYTES+1];
2504 uvchr_to_utf8(tmpbuf, c);
2505 return _is_utf8_perl_idcont_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2509 Perl__is_uni_perl_idstart(pTHX_ UV c)
2511 U8 tmpbuf[UTF8_MAXBYTES+1];
2512 uvchr_to_utf8(tmpbuf, c);
2513 return _is_utf8_perl_idstart_with_len(tmpbuf, tmpbuf + sizeof(tmpbuf));
2517 Perl__to_upper_title_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2520 /* We have the latin1-range values compiled into the core, so just use
2521 * those, converting the result to UTF-8. The only difference between upper
2522 * and title case in this range is that LATIN_SMALL_LETTER_SHARP_S is
2523 * either "SS" or "Ss". Which one to use is passed into the routine in
2524 * 'S_or_s' to avoid a test */
2526 UV converted = toUPPER_LATIN1_MOD(c);
2528 PERL_ARGS_ASSERT__TO_UPPER_TITLE_LATIN1;
2530 assert(S_or_s == 'S' || S_or_s == 's');
2532 if (UVCHR_IS_INVARIANT(converted)) { /* No difference between the two for
2533 characters in this range */
2534 *p = (U8) converted;
2539 /* toUPPER_LATIN1_MOD gives the correct results except for three outliers,
2540 * which it maps to one of them, so as to only have to have one check for
2541 * it in the main case */
2542 if (UNLIKELY(converted == LATIN_SMALL_LETTER_Y_WITH_DIAERESIS)) {
2544 case LATIN_SMALL_LETTER_Y_WITH_DIAERESIS:
2545 converted = LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS;
2548 converted = GREEK_CAPITAL_LETTER_MU;
2550 #if UNICODE_MAJOR_VERSION > 2 \
2551 || (UNICODE_MAJOR_VERSION == 2 && UNICODE_DOT_VERSION >= 1 \
2552 && UNICODE_DOT_DOT_VERSION >= 8)
2553 case LATIN_SMALL_LETTER_SHARP_S:
2560 Perl_croak(aTHX_ "panic: to_upper_title_latin1 did not expect"
2561 " '%c' to map to '%c'",
2562 c, LATIN_SMALL_LETTER_Y_WITH_DIAERESIS);
2563 NOT_REACHED; /* NOTREACHED */
2567 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2568 *p = UTF8_TWO_BYTE_LO(converted);
2574 /* Call the function to convert a UTF-8 encoded character to the specified case.
2575 * Note that there may be more than one character in the result.
2576 * INP is a pointer to the first byte of the input character
2577 * OUTP will be set to the first byte of the string of changed characters. It
2578 * needs to have space for UTF8_MAXBYTES_CASE+1 bytes
2579 * LENP will be set to the length in bytes of the string of changed characters
2581 * The functions return the ordinal of the first character in the string of
2583 #define CALL_UPPER_CASE(uv, s, d, lenp) \
2584 _to_utf8_case(uv, s, d, lenp, &PL_utf8_toupper, "ToUc", "")
2585 #define CALL_TITLE_CASE(uv, s, d, lenp) \
2586 _to_utf8_case(uv, s, d, lenp, &PL_utf8_totitle, "ToTc", "")
2587 #define CALL_LOWER_CASE(uv, s, d, lenp) \
2588 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tolower, "ToLc", "")
2590 /* This additionally has the input parameter 'specials', which if non-zero will
2591 * cause this to use the specials hash for folding (meaning get full case
2592 * folding); otherwise, when zero, this implies a simple case fold */
2593 #define CALL_FOLD_CASE(uv, s, d, lenp, specials) \
2594 _to_utf8_case(uv, s, d, lenp, &PL_utf8_tofold, "ToCf", (specials) ? "" : NULL)
2597 Perl_to_uni_upper(pTHX_ UV c, U8* p, STRLEN *lenp)
2599 /* Convert the Unicode character whose ordinal is <c> to its uppercase
2600 * version and store that in UTF-8 in <p> and its length in bytes in <lenp>.
2601 * Note that the <p> needs to be at least UTF8_MAXBYTES_CASE+1 bytes since
2602 * the changed version may be longer than the original character.
2604 * The ordinal of the first character of the changed version is returned
2605 * (but note, as explained above, that there may be more.) */
2607 PERL_ARGS_ASSERT_TO_UNI_UPPER;
2610 return _to_upper_title_latin1((U8) c, p, lenp, 'S');
2613 uvchr_to_utf8(p, c);
2614 return CALL_UPPER_CASE(c, p, p, lenp);
2618 Perl_to_uni_title(pTHX_ UV c, U8* p, STRLEN *lenp)
2620 PERL_ARGS_ASSERT_TO_UNI_TITLE;
2623 return _to_upper_title_latin1((U8) c, p, lenp, 's');
2626 uvchr_to_utf8(p, c);
2627 return CALL_TITLE_CASE(c, p, p, lenp);
2631 S_to_lower_latin1(const U8 c, U8* p, STRLEN *lenp, const char dummy)
2633 /* We have the latin1-range values compiled into the core, so just use
2634 * those, converting the result to UTF-8. Since the result is always just
2635 * one character, we allow <p> to be NULL */
2637 U8 converted = toLOWER_LATIN1(c);
2639 PERL_UNUSED_ARG(dummy);
2642 if (NATIVE_BYTE_IS_INVARIANT(converted)) {
2647 /* Result is known to always be < 256, so can use the EIGHT_BIT
2649 *p = UTF8_EIGHT_BIT_HI(converted);
2650 *(p+1) = UTF8_EIGHT_BIT_LO(converted);
2658 Perl_to_uni_lower(pTHX_ UV c, U8* p, STRLEN *lenp)
2660 PERL_ARGS_ASSERT_TO_UNI_LOWER;
2663 return to_lower_latin1((U8) c, p, lenp, 0 /* 0 is a dummy arg */ );
2666 uvchr_to_utf8(p, c);
2667 return CALL_LOWER_CASE(c, p, p, lenp);
2671 Perl__to_fold_latin1(pTHX_ const U8 c, U8* p, STRLEN *lenp,
2672 const unsigned int flags)
2674 /* Corresponds to to_lower_latin1(); <flags> bits meanings:
2675 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2676 * FOLD_FLAGS_FULL iff full folding is to be used;
2678 * Not to be used for locale folds
2683 PERL_ARGS_ASSERT__TO_FOLD_LATIN1;
2684 PERL_UNUSED_CONTEXT;
2686 assert (! (flags & FOLD_FLAGS_LOCALE));
2688 if (UNLIKELY(c == MICRO_SIGN)) {
2689 converted = GREEK_SMALL_LETTER_MU;
2691 #if UNICODE_MAJOR_VERSION > 3 /* no multifolds in early Unicode */ \
2692 || (UNICODE_MAJOR_VERSION == 3 && ( UNICODE_DOT_VERSION > 0) \
2693 || UNICODE_DOT_DOT_VERSION > 0)
2694 else if ( (flags & FOLD_FLAGS_FULL)
2695 && UNLIKELY(c == LATIN_SMALL_LETTER_SHARP_S))
2697 /* If can't cross 127/128 boundary, can't return "ss"; instead return
2698 * two U+017F characters, as fc("\df") should eq fc("\x{17f}\x{17f}")
2699 * under those circumstances. */
2700 if (flags & FOLD_FLAGS_NOMIX_ASCII) {
2701 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
2702 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
2704 return LATIN_SMALL_LETTER_LONG_S;
2714 else { /* In this range the fold of all other characters is their lower
2716 converted = toLOWER_LATIN1(c);
2719 if (UVCHR_IS_INVARIANT(converted)) {
2720 *p = (U8) converted;
2724 *(p)++ = UTF8_TWO_BYTE_HI(converted);
2725 *p = UTF8_TWO_BYTE_LO(converted);
2733 Perl__to_uni_fold_flags(pTHX_ UV c, U8* p, STRLEN *lenp, U8 flags)
2736 /* Not currently externally documented, and subject to change
2737 * <flags> bits meanings:
2738 * FOLD_FLAGS_FULL iff full folding is to be used;
2739 * FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
2740 * locale are to be used.
2741 * FOLD_FLAGS_NOMIX_ASCII iff non-ASCII to ASCII folds are prohibited
2744 PERL_ARGS_ASSERT__TO_UNI_FOLD_FLAGS;
2746 if (flags & FOLD_FLAGS_LOCALE) {
2747 /* Treat a UTF-8 locale as not being in locale at all */
2748 if (IN_UTF8_CTYPE_LOCALE) {
2749 flags &= ~FOLD_FLAGS_LOCALE;
2752 _CHECK_AND_WARN_PROBLEMATIC_LOCALE;
2753 goto needs_full_generality;
2758 return _to_fold_latin1((U8) c, p, lenp,
2759 flags & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII));
2762 /* Here, above 255. If no special needs, just use the macro */
2763 if ( ! (flags & (FOLD_FLAGS_LOCALE|FOLD_FLAGS_NOMIX_ASCII))) {
2764 uvchr_to_utf8(p, c);
2765 return CALL_FOLD_CASE(c, p, p, lenp, flags & FOLD_FLAGS_FULL);
2767 else { /* Otherwise, _toFOLD_utf8_flags has the intelligence to deal with
2768 the special flags. */
2769 U8 utf8_c[UTF8_MAXBYTES + 1];
2771 needs_full_generality:
2772 uvchr_to_utf8(utf8_c, c);
2773 return _toFOLD_utf8_flags(utf8_c, utf8_c + sizeof(utf8_c),
2778 PERL_STATIC_INLINE bool
2779 S_is_utf8_common(pTHX_ const U8 *const p, SV **swash,
2780 const char *const swashname, SV* const invlist)
2782 /* returns a boolean giving whether or not the UTF8-encoded character that
2783 * starts at <p> is in the swash indicated by <swashname>. <swash>
2784 * contains a pointer to where the swash indicated by <swashname>
2785 * is to be stored; which this routine will do, so that future calls will
2786 * look at <*swash> and only generate a swash if it is not null. <invlist>
2787 * is NULL or an inversion list that defines the swash. If not null, it
2788 * saves time during initialization of the swash.
2790 * Note that it is assumed that the buffer length of <p> is enough to
2791 * contain all the bytes that comprise the character. Thus, <*p> should
2792 * have been checked before this call for mal-formedness enough to assure
2795 PERL_ARGS_ASSERT_IS_UTF8_COMMON;
2797 /* The API should have included a length for the UTF-8 character in <p>,
2798 * but it doesn't. We therefore assume that p has been validated at least
2799 * as far as there being enough bytes available in it to accommodate the
2800 * character without reading beyond the end, and pass that number on to the
2801 * validating routine */
2802 if (! isUTF8_CHAR(p, p + UTF8SKIP(p))) {
2803 _force_out_malformed_utf8_message(p, p + UTF8SKIP(p),
2804 _UTF8_NO_CONFIDENCE_IN_CURLEN,
2806 NOT_REACHED; /* NOTREACHED */
2810 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2811 *swash = _core_swash_init("utf8",
2813 /* Only use the name if there is no inversion
2814 * list; otherwise will go out to disk */
2815 (invlist) ? "" : swashname,
2817 &PL_sv_undef, 1, 0, invlist, &flags);
2820 return swash_fetch(*swash, p, TRUE) != 0;
2823 PERL_STATIC_INLINE bool
2824 S_is_utf8_common_with_len(pTHX_ const U8 *const p, const U8 * const e,
2825 SV **swash, const char *const swashname,
2828 /* returns a boolean giving whether or not the UTF8-encoded character that
2829 * starts at <p>, and extending no further than <e - 1> is in the swash
2830 * indicated by <swashname>. <swash> contains a pointer to where the swash
2831 * indicated by <swashname> is to be stored; which this routine will do, so
2832 * that future calls will look at <*swash> and only generate a swash if it
2833 * is not null. <invlist> is NULL or an inversion list that defines the
2834 * swash. If not null, it saves time during initialization of the swash.
2837 PERL_ARGS_ASSERT_IS_UTF8_COMMON_WITH_LEN;
2839 if (! isUTF8_CHAR(p, e)) {
2840 _force_out_malformed_utf8_message(p, e, 0, 1);
2841 NOT_REACHED; /* NOTREACHED */
2845 U8 flags = _CORE_SWASH_INIT_ACCEPT_INVLIST;
2846 *swash = _core_swash_init("utf8",
2848 /* Only use the name if there is no inversion
2849 * list; otherwise will go out to disk */
2850 (invlist) ? "" : swashname,
2852 &PL_sv_undef, 1, 0, invlist, &flags);
2855 return swash_fetch(*swash, p, TRUE) != 0;
2859 S_warn_on_first_deprecated_use(pTHX_ const char * const name,
2860 const char * const alternative,
2861 const bool use_locale,
2862 const char * const file,
2863 const unsigned line)
2867 PERL_ARGS_ASSERT_WARN_ON_FIRST_DEPRECATED_USE;
2869 if (ckWARN_d(WARN_DEPRECATED)) {
2871 key = Perl_form(aTHX_ "%s;%d;%s;%d", name, use_locale, file, line);
2872 if (! hv_fetch(PL_seen_deprecated_macro, key, strlen(key), 0)) {
2873 if (! PL_seen_deprecated_macro) {
2874 PL_seen_deprecated_macro = newHV();
2876 if (! hv_store(PL_seen_deprecated_macro, key,
2877 strlen(key), &PL_sv_undef, 0))
2879 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
2882 if (instr(file, "mathoms.c")) {
2883 Perl_warner(aTHX_ WARN_DEPRECATED,
2884 "In %s, line %d, starting in Perl v5.30, %s()"
2885 " will be removed. Avoid this message by"
2886 " converting to use %s().\n",
2887 file, line, name, alternative);
2890 Perl_warner(aTHX_ WARN_DEPRECATED,
2891 "In %s, line %d, starting in Perl v5.30, %s() will"
2892 " require an additional parameter. Avoid this"
2893 " message by converting to use %s().\n",
2894 file, line, name, alternative);
2901 Perl__is_utf8_FOO(pTHX_ U8 classnum,
2903 const char * const name,
2904 const char * const alternative,
2905 const bool use_utf8,
2906 const bool use_locale,
2907 const char * const file,
2908 const unsigned line)
2910 PERL_ARGS_ASSERT__IS_UTF8_FOO;
2912 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
2914 if (use_utf8 && UTF8_IS_ABOVE_LATIN1(*p)) {
2924 case _CC_ALPHANUMERIC:
2928 return is_utf8_common(p,
2929 &PL_utf8_swash_ptrs[classnum],
2930 swash_property_names[classnum],
2931 PL_XPosix_ptrs[classnum]);
2934 return is_XPERLSPACE_high(p);
2936 return is_HORIZWS_high(p);
2938 return is_XDIGIT_high(p);
2944 return is_VERTWS_high(p);
2946 if (! PL_utf8_perl_idstart) {
2947 PL_utf8_perl_idstart
2948 = _new_invlist_C_array(_Perl_IDStart_invlist);
2950 return is_utf8_common(p, &PL_utf8_perl_idstart,
2951 "_Perl_IDStart", NULL);
2953 if (! PL_utf8_perl_idcont) {
2955 = _new_invlist_C_array(_Perl_IDCont_invlist);
2957 return is_utf8_common(p, &PL_utf8_perl_idcont,
2958 "_Perl_IDCont", NULL);
2962 /* idcont is the same as wordchar below 256 */
2963 if (classnum == _CC_IDCONT) {
2964 classnum = _CC_WORDCHAR;
2966 else if (classnum == _CC_IDFIRST) {
2970 classnum = _CC_ALPHA;
2974 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2975 return _generic_isCC(*p, classnum);
2978 return _generic_isCC(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )), classnum);
2981 if (! use_utf8 || UTF8_IS_INVARIANT(*p)) {
2982 return isFOO_lc(classnum, *p);
2985 return isFOO_lc(classnum, EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p + 1 )));
2988 NOT_REACHED; /* NOTREACHED */
2992 Perl__is_utf8_FOO_with_len(pTHX_ const U8 classnum, const U8 *p,
2995 PERL_ARGS_ASSERT__IS_UTF8_FOO_WITH_LEN;
2997 assert(classnum < _FIRST_NON_SWASH_CC);
2999 return is_utf8_common_with_len(p,
3001 &PL_utf8_swash_ptrs[classnum],
3002 swash_property_names[classnum],
3003 PL_XPosix_ptrs[classnum]);
3007 Perl__is_utf8_perl_idstart_with_len(pTHX_ const U8 *p, const U8 * const e)
3011 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDSTART_WITH_LEN;
3013 if (! PL_utf8_perl_idstart) {
3014 invlist = _new_invlist_C_array(_Perl_IDStart_invlist);
3016 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idstart,
3017 "_Perl_IDStart", invlist);
3021 Perl__is_utf8_xidstart(pTHX_ const U8 *p)
3023 PERL_ARGS_ASSERT__IS_UTF8_XIDSTART;
3027 return is_utf8_common(p, &PL_utf8_xidstart, "XIdStart", NULL);
3031 Perl__is_utf8_perl_idcont_with_len(pTHX_ const U8 *p, const U8 * const e)
3035 PERL_ARGS_ASSERT__IS_UTF8_PERL_IDCONT_WITH_LEN;
3037 if (! PL_utf8_perl_idcont) {
3038 invlist = _new_invlist_C_array(_Perl_IDCont_invlist);
3040 return is_utf8_common_with_len(p, e, &PL_utf8_perl_idcont,
3041 "_Perl_IDCont", invlist);
3045 Perl__is_utf8_idcont(pTHX_ const U8 *p)
3047 PERL_ARGS_ASSERT__IS_UTF8_IDCONT;
3049 return is_utf8_common(p, &PL_utf8_idcont, "IdContinue", NULL);
3053 Perl__is_utf8_xidcont(pTHX_ const U8 *p)
3055 PERL_ARGS_ASSERT__IS_UTF8_XIDCONT;
3057 return is_utf8_common(p, &PL_utf8_idcont, "XIdContinue", NULL);
3061 Perl__is_utf8_mark(pTHX_ const U8 *p)
3063 PERL_ARGS_ASSERT__IS_UTF8_MARK;
3065 return is_utf8_common(p, &PL_utf8_mark, "IsM", NULL);
3068 /* change namve uv1 to 'from' */
3070 S__to_utf8_case(pTHX_ const UV uv1, const U8 *p, U8* ustrp, STRLEN *lenp,
3071 SV **swashp, const char *normal, const char *special)
3075 PERL_ARGS_ASSERT__TO_UTF8_CASE;
3077 /* For code points that don't change case, we already know that the output
3078 * of this function is the unchanged input, so we can skip doing look-ups
3079 * for them. Unfortunately the case-changing code points are scattered
3080 * around. But there are some long consecutive ranges where there are no
3081 * case changing code points. By adding tests, we can eliminate the lookup
3082 * for all the ones in such ranges. This is currently done here only for
3083 * just a few cases where the scripts are in common use in modern commerce
3084 * (and scripts adjacent to those which can be included without additional
3087 if (uv1 >= 0x0590) {
3088 /* This keeps from needing further processing the code points most
3089 * likely to be used in the following non-cased scripts: Hebrew,
3090 * Arabic, Syriac, Thaana, NKo, Samaritan, Mandaic, Devanagari,
3091 * Bengali, Gurmukhi, Gujarati, Oriya, Tamil, Telugu, Kannada,
3092 * Malayalam, Sinhala, Thai, Lao, Tibetan, Myanmar */
3097 /* The following largish code point ranges also don't have case
3098 * changes, but khw didn't think they warranted extra tests to speed
3099 * them up (which would slightly slow down everything else above them):
3100 * 1100..139F Hangul Jamo, Ethiopic
3101 * 1400..1CFF Unified Canadian Aboriginal Syllabics, Ogham, Runic,
3102 * Tagalog, Hanunoo, Buhid, Tagbanwa, Khmer, Mongolian,
3103 * Limbu, Tai Le, New Tai Lue, Buginese, Tai Tham,
3104 * Combining Diacritical Marks Extended, Balinese,
3105 * Sundanese, Batak, Lepcha, Ol Chiki
3106 * 2000..206F General Punctuation
3109 if (uv1 >= 0x2D30) {
3111 /* This keeps the from needing further processing the code points
3112 * most likely to be used in the following non-cased major scripts:
3113 * CJK, Katakana, Hiragana, plus some less-likely scripts.
3115 * (0x2D30 above might have to be changed to 2F00 in the unlikely
3116 * event that Unicode eventually allocates the unused block as of
3117 * v8.0 2FE0..2FEF to code points that are cased. khw has verified
3118 * that the test suite will start having failures to alert you
3119 * should that happen) */
3124 if (uv1 >= 0xAC00) {
3125 if (UNLIKELY(UNICODE_IS_SURROGATE(uv1))) {
3126 if (ckWARN_d(WARN_SURROGATE)) {
3127 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3128 Perl_warner(aTHX_ packWARN(WARN_SURROGATE),
3129 "Operation \"%s\" returns its argument for"
3130 " UTF-16 surrogate U+%04" UVXf, desc, uv1);
3135 /* AC00..FAFF Catches Hangul syllables and private use, plus
3142 if (UNLIKELY(UNICODE_IS_SUPER(uv1))) {
3143 if (UNLIKELY(uv1 > MAX_EXTERNALLY_LEGAL_CP)) {
3144 Perl_croak(aTHX_ cp_above_legal_max, uv1,
3145 MAX_EXTERNALLY_LEGAL_CP);
3147 if (ckWARN_d(WARN_NON_UNICODE)) {
3148 const char* desc = (PL_op) ? OP_DESC(PL_op) : normal;
3149 Perl_warner(aTHX_ packWARN(WARN_NON_UNICODE),
3150 "Operation \"%s\" returns its argument for"
3151 " non-Unicode code point 0x%04" UVXf, desc, uv1);
3155 #ifdef HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C
3157 > HIGHEST_CASE_CHANGING_CP_FOR_USE_ONLY_BY_UTF8_DOT_C))
3160 /* As of Unicode 10.0, this means we avoid swash creation
3161 * for anything beyond high Plane 1 (below emojis) */
3168 /* Note that non-characters are perfectly legal, so no warning should
3169 * be given. There are so few of them, that it isn't worth the extra
3170 * tests to avoid swash creation */
3173 if (!*swashp) /* load on-demand */
3174 *swashp = _core_swash_init("utf8", normal, &PL_sv_undef,
3178 /* It might be "special" (sometimes, but not always,
3179 * a multicharacter mapping) */
3183 /* If passed in the specials name, use that; otherwise use any
3184 * given in the swash */
3185 if (*special != '\0') {
3186 hv = get_hv(special, 0);
3189 svp = hv_fetchs(MUTABLE_HV(SvRV(*swashp)), "SPECIALS", 0);
3191 hv = MUTABLE_HV(SvRV(*svp));
3196 && (svp = hv_fetch(hv, (const char*)p, UVCHR_SKIP(uv1), FALSE))
3201 s = SvPV_const(*svp, len);
3204 len = uvchr_to_utf8(ustrp, *(U8*)s) - ustrp;
3206 Copy(s, ustrp, len, U8);
3211 if (!len && *swashp) {
3212 const UV uv2 = swash_fetch(*swashp, p, TRUE /* => is UTF-8 */);
3215 /* It was "normal" (a single character mapping). */
3216 len = uvchr_to_utf8(ustrp, uv2) - ustrp;
3224 return valid_utf8_to_uvchr(ustrp, 0);
3227 /* Here, there was no mapping defined, which means that the code point maps
3228 * to itself. Return the inputs */
3231 if (p != ustrp) { /* Don't copy onto itself */
3232 Copy(p, ustrp, len, U8);
3243 S_check_locale_boundary_crossing(pTHX_ const U8* const p, const UV result,
3244 U8* const ustrp, STRLEN *lenp)
3246 /* This is called when changing the case of a UTF-8-encoded character above
3247 * the Latin1 range, and the operation is in a non-UTF-8 locale. If the
3248 * result contains a character that crosses the 255/256 boundary, disallow
3249 * the change, and return the original code point. See L<perlfunc/lc> for
3252 * p points to the original string whose case was changed; assumed
3253 * by this routine to be well-formed
3254 * result the code point of the first character in the changed-case string
3255 * ustrp points to the changed-case string (<result> represents its
3257 * lenp points to the length of <ustrp> */
3259 UV original; /* To store the first code point of <p> */
3261 PERL_ARGS_ASSERT_CHECK_LOCALE_BOUNDARY_CROSSING;
3263 assert(UTF8_IS_ABOVE_LATIN1(*p));
3265 /* We know immediately if the first character in the string crosses the
3266 * boundary, so can skip */
3269 /* Look at every character in the result; if any cross the
3270 * boundary, the whole thing is disallowed */
3271 U8* s = ustrp + UTF8SKIP(ustrp);
3272 U8* e = ustrp + *lenp;
3274 if (! UTF8_IS_ABOVE_LATIN1(*s)) {
3280 /* Here, no characters crossed, result is ok as-is, but we warn. */
3281 _CHECK_AND_OUTPUT_WIDE_LOCALE_UTF8_MSG(p, p + UTF8SKIP(p));
3287 /* Failed, have to return the original */
3288 original = valid_utf8_to_uvchr(p, lenp);
3290 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3291 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3292 "Can't do %s(\"\\x{%" UVXf "}\") on non-UTF-8"
3293 " locale; resolved to \"\\x{%" UVXf "}\".",
3297 Copy(p, ustrp, *lenp, char);
3302 S_check_and_deprecate(pTHX_ const U8 *p,
3304 const unsigned int type, /* See below */
3305 const bool use_locale, /* Is this a 'LC_'
3307 const char * const file,
3308 const unsigned line)
3310 /* This is a temporary function to deprecate the unsafe calls to the case
3311 * changing macros and functions. It keeps all the special stuff in just
3314 * It updates *e with the pointer to the end of the input string. If using
3315 * the old-style macros, *e is NULL on input, and so this function assumes
3316 * the input string is long enough to hold the entire UTF-8 sequence, and
3317 * sets *e accordingly, but it then returns a flag to pass the
3318 * utf8n_to_uvchr(), to tell it that this size is a guess, and to avoid
3319 * using the full length if possible.
3321 * It also does the assert that *e > p when *e is not NULL. This should be
3322 * migrated to the callers when this function gets deleted.
3324 * The 'type' parameter is used for the caller to specify which case
3325 * changing function this is called from: */
3327 # define DEPRECATE_TO_UPPER 0
3328 # define DEPRECATE_TO_TITLE 1
3329 # define DEPRECATE_TO_LOWER 2
3330 # define DEPRECATE_TO_FOLD 3
3332 U32 utf8n_flags = 0;
3334 const char * alternative;
3336 PERL_ARGS_ASSERT_CHECK_AND_DEPRECATE;
3339 utf8n_flags = _UTF8_NO_CONFIDENCE_IN_CURLEN;
3340 *e = p + UTF8SKIP(p);
3342 /* For mathoms.c calls, we use the function name we know is stored
3343 * there. It could be part of a larger path */
3344 if (type == DEPRECATE_TO_UPPER) {
3345 name = instr(file, "mathoms.c")
3348 alternative = "toUPPER_utf8_safe";
3350 else if (type == DEPRECATE_TO_TITLE) {
3351 name = instr(file, "mathoms.c")
3354 alternative = "toTITLE_utf8_safe";
3356 else if (type == DEPRECATE_TO_LOWER) {
3357 name = instr(file, "mathoms.c")
3360 alternative = "toLOWER_utf8_safe";
3362 else if (type == DEPRECATE_TO_FOLD) {
3363 name = instr(file, "mathoms.c")
3366 alternative = "toFOLD_utf8_safe";
3368 else Perl_croak(aTHX_ "panic: Unexpected case change type");
3370 warn_on_first_deprecated_use(name, alternative, use_locale, file, line);
3379 /* The process for changing the case is essentially the same for the four case
3380 * change types, except there are complications for folding. Otherwise the
3381 * difference is only which case to change to. To make sure that they all do
3382 * the same thing, the bodies of the functions are extracted out into the
3383 * following two macros. The functions are written with the same variable
3384 * names, and these are known and used inside these macros. It would be
3385 * better, of course, to have inline functions to do it, but since different
3386 * macros are called, depending on which case is being changed to, this is not
3387 * feasible in C (to khw's knowledge). Two macros are created so that the fold
3388 * function can start with the common start macro, then finish with its special
3389 * handling; while the other three cases can just use the common end macro.
3391 * The algorithm is to use the proper (passed in) macro or function to change
3392 * the case for code points that are below 256. The macro is used if using
3393 * locale rules for the case change; the function if not. If the code point is
3394 * above 255, it is computed from the input UTF-8, and another macro is called
3395 * to do the conversion. If necessary, the output is converted to UTF-8. If
3396 * using a locale, we have to check that the change did not cross the 255/256
3397 * boundary, see check_locale_boundary_crossing() for further details.
3399 * The macros are split with the correct case change for the below-256 case
3400 * stored into 'result', and in the middle of an else clause for the above-255
3401 * case. At that point in the 'else', 'result' is not the final result, but is
3402 * the input code point calculated from the UTF-8. The fold code needs to
3403 * realize all this and take it from there.
3405 * If you read the two macros as sequential, it's easier to understand what's
3407 #define CASE_CHANGE_BODY_START(locale_flags, LC_L1_change_macro, L1_func, \
3408 L1_func_extra_param) \
3410 if (flags & (locale_flags)) { \
3411 /* Treat a UTF-8 locale as not being in locale at all */ \
3412 if (IN_UTF8_CTYPE_LOCALE) { \
3413 flags &= ~(locale_flags); \
3416 _CHECK_AND_WARN_PROBLEMATIC_LOCALE; \
3420 if (UTF8_IS_INVARIANT(*p)) { \
3421 if (flags & (locale_flags)) { \
3422 result = LC_L1_change_macro(*p); \
3425 return L1_func(*p, ustrp, lenp, L1_func_extra_param); \
3428 else if UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(p, e) { \
3429 if (flags & (locale_flags)) { \
3430 result = LC_L1_change_macro(EIGHT_BIT_UTF8_TO_NATIVE(*p, \
3434 return L1_func(EIGHT_BIT_UTF8_TO_NATIVE(*p, *(p+1)), \
3435 ustrp, lenp, L1_func_extra_param); \
3438 else { /* malformed UTF-8 or ord above 255 */ \
3439 STRLEN len_result; \
3440 result = utf8n_to_uvchr(p, e - p, &len_result, UTF8_CHECK_ONLY); \
3441 if (len_result == (STRLEN) -1) { \
3442 _force_out_malformed_utf8_message(p, e, utf8n_flags, \
3446 #define CASE_CHANGE_BODY_END(locale_flags, change_macro) \
3447 result = change_macro(result, p, ustrp, lenp); \
3449 if (flags & (locale_flags)) { \
3450 result = check_locale_boundary_crossing(p, result, ustrp, lenp); \
3455 /* Here, used locale rules. Convert back to UTF-8 */ \
3456 if (UTF8_IS_INVARIANT(result)) { \
3457 *ustrp = (U8) result; \
3461 *ustrp = UTF8_EIGHT_BIT_HI((U8) result); \
3462 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result); \
3469 =for apidoc to_utf8_upper
3471 Instead use L</toUPPER_utf8_safe>.
3475 /* Not currently externally documented, and subject to change:
3476 * <flags> is set iff iff the rules from the current underlying locale are to
3480 Perl__to_utf8_upper_flags(pTHX_ const U8 *p,
3485 const char * const file,
3489 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_UPPER,
3490 cBOOL(flags), file, line);
3492 PERL_ARGS_ASSERT__TO_UTF8_UPPER_FLAGS;
3494 /* ~0 makes anything non-zero in 'flags' mean we are using locale rules */
3495 /* 2nd char of uc(U+DF) is 'S' */
3496 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 'S');
3497 CASE_CHANGE_BODY_END (~0, CALL_UPPER_CASE);
3501 =for apidoc to_utf8_title
3503 Instead use L</toTITLE_utf8_safe>.
3507 /* Not currently externally documented, and subject to change:
3508 * <flags> is set iff the rules from the current underlying locale are to be
3509 * used. Since titlecase is not defined in POSIX, for other than a
3510 * UTF-8 locale, uppercase is used instead for code points < 256.
3514 Perl__to_utf8_title_flags(pTHX_ const U8 *p,
3519 const char * const file,
3523 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_TITLE,
3524 cBOOL(flags), file, line);
3526 PERL_ARGS_ASSERT__TO_UTF8_TITLE_FLAGS;
3528 /* 2nd char of ucfirst(U+DF) is 's' */
3529 CASE_CHANGE_BODY_START(~0, toUPPER_LC, _to_upper_title_latin1, 's');
3530 CASE_CHANGE_BODY_END (~0, CALL_TITLE_CASE);
3534 =for apidoc to_utf8_lower
3536 Instead use L</toLOWER_utf8_safe>.
3540 /* Not currently externally documented, and subject to change:
3541 * <flags> is set iff iff the rules from the current underlying locale are to
3546 Perl__to_utf8_lower_flags(pTHX_ const U8 *p,
3551 const char * const file,
3555 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_LOWER,
3556 cBOOL(flags), file, line);
3558 PERL_ARGS_ASSERT__TO_UTF8_LOWER_FLAGS;
3560 CASE_CHANGE_BODY_START(~0, toLOWER_LC, to_lower_latin1, 0 /* 0 is dummy */)
3561 CASE_CHANGE_BODY_END (~0, CALL_LOWER_CASE)
3565 =for apidoc to_utf8_fold
3567 Instead use L</toFOLD_utf8_safe>.
3571 /* Not currently externally documented, and subject to change,
3573 * bit FOLD_FLAGS_LOCALE is set iff the rules from the current underlying
3574 * locale are to be used.
3575 * bit FOLD_FLAGS_FULL is set iff full case folds are to be used;
3576 * otherwise simple folds
3577 * bit FOLD_FLAGS_NOMIX_ASCII is set iff folds of non-ASCII to ASCII are
3582 Perl__to_utf8_fold_flags(pTHX_ const U8 *p,
3587 const char * const file,
3591 const U32 utf8n_flags = check_and_deprecate(p, &e, DEPRECATE_TO_FOLD,
3592 cBOOL(flags), file, line);
3594 PERL_ARGS_ASSERT__TO_UTF8_FOLD_FLAGS;
3596 /* These are mutually exclusive */
3597 assert (! ((flags & FOLD_FLAGS_LOCALE) && (flags & FOLD_FLAGS_NOMIX_ASCII)));
3599 assert(p != ustrp); /* Otherwise overwrites */
3601 CASE_CHANGE_BODY_START(FOLD_FLAGS_LOCALE, toFOLD_LC, _to_fold_latin1,
3602 ((flags) & (FOLD_FLAGS_FULL | FOLD_FLAGS_NOMIX_ASCII)));
3604 result = CALL_FOLD_CASE(result, p, ustrp, lenp, flags & FOLD_FLAGS_FULL);
3606 if (flags & FOLD_FLAGS_LOCALE) {
3608 # define LONG_S_T LATIN_SMALL_LIGATURE_LONG_S_T_UTF8
3609 # ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3610 # define CAP_SHARP_S LATIN_CAPITAL_LETTER_SHARP_S_UTF8
3612 /* Special case these two characters, as what normally gets
3613 * returned under locale doesn't work */
3614 if (memEQs((char *) p, UTF8SKIP(p), CAP_SHARP_S))
3616 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3617 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3618 "Can't do fc(\"\\x{1E9E}\") on non-UTF-8 locale; "
3619 "resolved to \"\\x{17F}\\x{17F}\".");
3624 if (memEQs((char *) p, UTF8SKIP(p), LONG_S_T))
3626 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3627 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3628 "Can't do fc(\"\\x{FB05}\") on non-UTF-8 locale; "
3629 "resolved to \"\\x{FB06}\".");
3630 goto return_ligature_st;
3633 #if UNICODE_MAJOR_VERSION == 3 \
3634 && UNICODE_DOT_VERSION == 0 \
3635 && UNICODE_DOT_DOT_VERSION == 1
3636 # define DOTTED_I LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE_UTF8
3638 /* And special case this on this Unicode version only, for the same
3639 * reaons the other two are special cased. They would cross the
3640 * 255/256 boundary which is forbidden under /l, and so the code
3641 * wouldn't catch that they are equivalent (which they are only in
3643 else if (memEQs((char *) p, UTF8SKIP(p), DOTTED_I)) {
3644 /* diag_listed_as: Can't do %s("%s") on non-UTF-8 locale; resolved to "%s". */
3645 Perl_ck_warner(aTHX_ packWARN(WARN_LOCALE),
3646 "Can't do fc(\"\\x{0130}\") on non-UTF-8 locale; "
3647 "resolved to \"\\x{0131}\".");
3648 goto return_dotless_i;
3652 return check_locale_boundary_crossing(p, result, ustrp, lenp);
3654 else if (! (flags & FOLD_FLAGS_NOMIX_ASCII)) {
3658 /* This is called when changing the case of a UTF-8-encoded
3659 * character above the ASCII range, and the result should not
3660 * contain an ASCII character. */
3662 UV original; /* To store the first code point of <p> */
3664 /* Look at every character in the result; if any cross the
3665 * boundary, the whole thing is disallowed */
3667 U8* e = ustrp + *lenp;
3670 /* Crossed, have to return the original */
3671 original = valid_utf8_to_uvchr(p, lenp);
3673 /* But in these instances, there is an alternative we can
3674 * return that is valid */
3675 if (original == LATIN_SMALL_LETTER_SHARP_S
3676 #ifdef LATIN_CAPITAL_LETTER_SHARP_S /* not defined in early Unicode releases */
3677 || original == LATIN_CAPITAL_LETTER_SHARP_S
3682 else if (original == LATIN_SMALL_LIGATURE_LONG_S_T) {
3683 goto return_ligature_st;
3685 #if UNICODE_MAJOR_VERSION == 3 \
3686 && UNICODE_DOT_VERSION == 0 \
3687 && UNICODE_DOT_DOT_VERSION == 1
3689 else if (original == LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE) {
3690 goto return_dotless_i;
3693 Copy(p, ustrp, *lenp, char);
3699 /* Here, no characters crossed, result is ok as-is */
3704 /* Here, used locale rules. Convert back to UTF-8 */
3705 if (UTF8_IS_INVARIANT(result)) {
3706 *ustrp = (U8) result;
3710 *ustrp = UTF8_EIGHT_BIT_HI((U8) result);
3711 *(ustrp + 1) = UTF8_EIGHT_BIT_LO((U8) result);
3718 /* Certain folds to 'ss' are prohibited by the options, but they do allow
3719 * folds to a string of two of these characters. By returning this
3720 * instead, then, e.g.,
3721 * fc("\x{1E9E}") eq fc("\x{17F}\x{17F}")
3724 *lenp = 2 * sizeof(LATIN_SMALL_LETTER_LONG_S_UTF8) - 2;
3725 Copy(LATIN_SMALL_LETTER_LONG_S_UTF8 LATIN_SMALL_LETTER_LONG_S_UTF8,
3727 return LATIN_SMALL_LETTER_LONG_S;
3730 /* Two folds to 'st' are prohibited by the options; instead we pick one and
3731 * have the other one fold to it */
3733 *lenp = sizeof(LATIN_SMALL_LIGATURE_ST_UTF8) - 1;
3734 Copy(LATIN_SMALL_LIGATURE_ST_UTF8, ustrp, *lenp, U8);
3735 return LATIN_SMALL_LIGATURE_ST;
3737 #if UNICODE_MAJOR_VERSION == 3 \
3738 && UNICODE_DOT_VERSION == 0 \
3739 && UNICODE_DOT_DOT_VERSION == 1
3742 *lenp = sizeof(LATIN_SMALL_LETTER_DOTLESS_I_UTF8) - 1;
3743 Copy(LATIN_SMALL_LETTER_DOTLESS_I_UTF8, ustrp, *lenp, U8);
3744 return LATIN_SMALL_LETTER_DOTLESS_I;
3751 * Returns a "swash" which is a hash described in utf8.c:Perl_swash_fetch().
3752 * C<pkg> is a pointer to a package name for SWASHNEW, should be "utf8".
3753 * For other parameters, see utf8::SWASHNEW in lib/utf8_heavy.pl.
3757 Perl_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3758 I32 minbits, I32 none)
3760 PERL_ARGS_ASSERT_SWASH_INIT;
3762 /* Returns a copy of a swash initiated by the called function. This is the
3763 * public interface, and returning a copy prevents others from doing
3764 * mischief on the original */
3766 return newSVsv(_core_swash_init(pkg, name, listsv, minbits, none,
3771 Perl__core_swash_init(pTHX_ const char* pkg, const char* name, SV *listsv,
3772 I32 minbits, I32 none, SV* invlist,
3776 /*NOTE NOTE NOTE - If you want to use "return" in this routine you MUST
3777 * use the following define */
3779 #define CORE_SWASH_INIT_RETURN(x) \
3780 PL_curpm= old_PL_curpm; \
3783 /* Initialize and return a swash, creating it if necessary. It does this
3784 * by calling utf8_heavy.pl in the general case. The returned value may be
3785 * the swash's inversion list instead if the input parameters allow it.
3786 * Which is returned should be immaterial to callers, as the only
3787 * operations permitted on a swash, swash_fetch(), _get_swash_invlist(),
3788 * and swash_to_invlist() handle both these transparently.
3790 * This interface should only be used by functions that won't destroy or
3791 * adversely change the swash, as doing so affects all other uses of the
3792 * swash in the program; the general public should use 'Perl_swash_init'
3795 * pkg is the name of the package that <name> should be in.
3796 * name is the name of the swash to find. Typically it is a Unicode
3797 * property name, including user-defined ones
3798 * listsv is a string to initialize the swash with. It must be of the form
3799 * documented as the subroutine return value in
3800 * L<perlunicode/User-Defined Character Properties>
3801 * minbits is the number of bits required to represent each data element.
3802 * It is '1' for binary properties.
3803 * none I (khw) do not understand this one, but it is used only in tr///.
3804 * invlist is an inversion list to initialize the swash with (or NULL)
3805 * flags_p if non-NULL is the address of various input and output flag bits
3806 * to the routine, as follows: ('I' means is input to the routine;
3807 * 'O' means output from the routine. Only flags marked O are
3808 * meaningful on return.)
3809 * _CORE_SWASH_INIT_USER_DEFINED_PROPERTY indicates if the swash
3810 * came from a user-defined property. (I O)
3811 * _CORE_SWASH_INIT_RETURN_IF_UNDEF indicates that instead of croaking
3812 * when the swash cannot be located, to simply return NULL. (I)
3813 * _CORE_SWASH_INIT_ACCEPT_INVLIST indicates that the caller will accept a
3814 * return of an inversion list instead of a swash hash if this routine
3815 * thinks that would result in faster execution of swash_fetch() later
3818 * Thus there are three possible inputs to find the swash: <name>,
3819 * <listsv>, and <invlist>. At least one must be specified. The result
3820 * will be the union of the specified ones, although <listsv>'s various
3821 * actions can intersect, etc. what <name> gives. To avoid going out to
3822 * disk at all, <invlist> should specify completely what the swash should
3823 * have, and <listsv> should be &PL_sv_undef and <name> should be "".
3825 * <invlist> is only valid for binary properties */
3827 PMOP *old_PL_curpm= PL_curpm; /* save away the old PL_curpm */
3829 SV* retval = &PL_sv_undef;
3830 HV* swash_hv = NULL;
3831 const int invlist_swash_boundary =
3832 (flags_p && *flags_p & _CORE_SWASH_INIT_ACCEPT_INVLIST)
3833 ? 512 /* Based on some benchmarking, but not extensive, see commit
3835 : -1; /* Never return just an inversion list */
3837 assert(listsv != &PL_sv_undef || strNE(name, "") || invlist);
3838 assert(! invlist || minbits == 1);
3840 PL_curpm= NULL; /* reset PL_curpm so that we dont get confused between the
3841 regex that triggered the swash init and the swash init
3842 perl logic itself. See perl #122747 */
3844 /* If data was passed in to go out to utf8_heavy to find the swash of, do
3846 if (listsv != &PL_sv_undef || strNE(name, "")) {
3848 const size_t pkg_len = strlen(pkg);
3849 const size_t name_len = strlen(name);
3850 HV * const stash = gv_stashpvn(pkg, pkg_len, 0);
3854 PERL_ARGS_ASSERT__CORE_SWASH_INIT;
3856 PUSHSTACKi(PERLSI_MAGIC);
3860 /* We might get here via a subroutine signature which uses a utf8
3861 * parameter name, at which point PL_subname will have been set
3862 * but not yet used. */
3863 save_item(PL_subname);
3864 if (PL_parser && PL_parser->error_count)
3865 SAVEI8(PL_parser->error_count), PL_parser->error_count = 0;
3866 method = gv_fetchmeth(stash, "SWASHNEW", 8, -1);
3867 if (!method) { /* demand load UTF-8 */
3869 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3870 GvSV(PL_errgv) = NULL;
3871 #ifndef NO_TAINT_SUPPORT
3872 /* It is assumed that callers of this routine are not passing in
3873 * any user derived data. */
3874 /* Need to do this after save_re_context() as it will set
3875 * PL_tainted to 1 while saving $1 etc (see the code after getrx:
3876 * in Perl_magic_get). Even line to create errsv_save can turn on
3878 SAVEBOOL(TAINT_get);
3881 Perl_load_module(aTHX_ PERL_LOADMOD_NOIMPORT, newSVpvn(pkg,pkg_len),
3884 /* Not ERRSV, as there is no need to vivify a scalar we are
3885 about to discard. */
3886 SV * const errsv = GvSV(PL_errgv);
3887 if (!SvTRUE(errsv)) {
3888 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3889 SvREFCNT_dec(errsv);
3897 mPUSHp(pkg, pkg_len);
3898 mPUSHp(name, name_len);
3903 if ((errsv_save = GvSV(PL_errgv))) SAVEFREESV(errsv_save);
3904 GvSV(PL_errgv) = NULL;
3905 /* If we already have a pointer to the method, no need to use
3906 * call_method() to repeat the lookup. */
3908 ? call_sv(MUTABLE_SV(method), G_SCALAR)
3909 : call_sv(newSVpvs_flags("SWASHNEW", SVs_TEMP), G_SCALAR | G_METHOD))
3911 retval = *PL_stack_sp--;
3912 SvREFCNT_inc(retval);
3915 /* Not ERRSV. See above. */
3916 SV * const errsv = GvSV(PL_errgv);
3917 if (!SvTRUE(errsv)) {
3918 GvSV(PL_errgv) = SvREFCNT_inc_simple(errsv_save);
3919 SvREFCNT_dec(errsv);
3924 if (IN_PERL_COMPILETIME) {
3925 CopHINTS_set(PL_curcop, PL_hints);
3927 if (!SvROK(retval) || SvTYPE(SvRV(retval)) != SVt_PVHV) {
3928 if (SvPOK(retval)) {
3930 /* If caller wants to handle missing properties, let them */
3931 if (flags_p && *flags_p & _CORE_SWASH_INIT_RETURN_IF_UNDEF) {
3932 CORE_SWASH_INIT_RETURN(NULL);
3935 "Can't find Unicode property definition \"%" SVf "\"",
3937 NOT_REACHED; /* NOTREACHED */
3940 } /* End of calling the module to find the swash */
3942 /* If this operation fetched a swash, and we will need it later, get it */
3943 if (retval != &PL_sv_undef
3944 && (minbits == 1 || (flags_p
3946 & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY))))
3948 swash_hv = MUTABLE_HV(SvRV(retval));
3950 /* If we don't already know that there is a user-defined component to
3951 * this swash, and the user has indicated they wish to know if there is
3952 * one (by passing <flags_p>), find out */
3953 if (flags_p && ! (*flags_p & _CORE_SWASH_INIT_USER_DEFINED_PROPERTY)) {
3954 SV** user_defined = hv_fetchs(swash_hv, "USER_DEFINED", FALSE);
3955 if (user_defined && SvUV(*user_defined)) {
3956 *flags_p |= _CORE_SWASH_INIT_USER_DEFINED_PROPERTY;
3961 /* Make sure there is an inversion list for binary properties */
3963 SV** swash_invlistsvp = NULL;
3964 SV* swash_invlist = NULL;
3965 bool invlist_in_swash_is_valid = FALSE;
3966 bool swash_invlist_unclaimed = FALSE; /* whether swash_invlist has
3967 an unclaimed reference count */
3969 /* If this operation fetched a swash, get its already existing
3970 * inversion list, or create one for it */
3973 swash_invlistsvp = hv_fetchs(swash_hv, "V", FALSE);
3974 if (swash_invlistsvp) {
3975 swash_invlist = *swash_invlistsvp;
3976 invlist_in_swash_is_valid = TRUE;
3979 swash_invlist = _swash_to_invlist(retval);
3980 swash_invlist_unclaimed = TRUE;
3984 /* If an inversion list was passed in, have to include it */
3987 /* Any fetched swash will by now have an inversion list in it;
3988 * otherwise <swash_invlist> will be NULL, indicating that we
3989 * didn't fetch a swash */
3990 if (swash_invlist) {
3992 /* Add the passed-in inversion list, which invalidates the one
3993 * already stored in the swash */
3994 invlist_in_swash_is_valid = FALSE;
3995 SvREADONLY_off(swash_invlist); /* Turned on again below */
3996 _invlist_union(invlist, swash_invlist, &swash_invlist);
4000 /* Here, there is no swash already. Set up a minimal one, if
4001 * we are going to return a swash */
4002 if ((int) _invlist_len(invlist) > invlist_swash_boundary) {
4004 retval = newRV_noinc(MUTABLE_SV(swash_hv));
4006 swash_invlist = invlist;
4010 /* Here, we have computed the union of all the passed-in data. It may
4011 * be that there was an inversion list in the swash which didn't get
4012 * touched; otherwise save the computed one */
4013 if (! invlist_in_swash_is_valid
4014 && (int) _invlist_len(swash_invlist) > invlist_swash_boundary)
4016 if (! hv_stores(MUTABLE_HV(SvRV(retval)), "V", swash_invlist))
4018 Perl_croak(aTHX_ "panic: hv_store() unexpectedly failed");
4020 /* We just stole a reference count. */
4021 if (swash_invlist_unclaimed) swash_invlist_unclaimed = FALSE;
4022 else SvREFCNT_inc_simple_void_NN(swash_invlist);
4025 /* The result is immutable. Forbid attempts to change it. */
4026 SvREADONLY_on(swash_invlist);
4028 /* Use the inversion list stand-alone if small enough */
4029 if ((int) _invlist_len(swash_invlist) <= invlist_swash_boundary) {
4030 SvREFCNT_dec(retval);
4031 if (!swash_invlist_unclaimed)
4032 SvREFCNT_inc_simple_void_NN(swash_invlist);
4033 retval = newRV_noinc(swash_invlist);
4037 CORE_SWASH_INIT_RETURN(retval);
4038 #undef CORE_SWASH_INIT_RETURN
4042 /* This API is wrong for special case conversions since we may need to
4043 * return several Unicode characters for a single Unicode character
4044 * (see lib/unicore/SpecCase.txt) The SWASHGET in lib/utf8_heavy.pl is
4045 * the lower-level routine, and it is similarly broken for returning
4046 * multiple values. --jhi
4047 * For those, you should use S__to_utf8_case() instead */
4048 /* Now SWASHGET is recasted into S_swatch_get in this file. */
4051 * Returns the value of property/mapping C<swash> for the first character
4052 * of the string C<ptr>. If C<do_utf8> is true, the string C<ptr> is
4053 * assumed to be in well-formed UTF-8. If C<do_utf8> is false, the string C<ptr>
4054 * is assumed to be in native 8-bit encoding. Caches the swatch in C<swash>.
4056 * A "swash" is a hash which contains initially the keys/values set up by
4057 * SWASHNEW. The purpose is to be able to completely represent a Unicode
4058 * property for all possible code points. Things are stored in a compact form
4059 * (see utf8_heavy.pl) so that calculation is required to find the actual
4060 * property value for a given code point. As code points are looked up, new
4061 * key/value pairs are added to the hash, so that the calculation doesn't have
4062 * to ever be re-done. Further, each calculation is done, not just for the
4063 * desired one, but for a whole block of code points adjacent to that one.
4064 * For binary properties on ASCII machines, the block is usually for 64 code
4065 * points, starting with a code point evenly divisible by 64. Thus if the
4066 * property value for code point 257 is requested, the code goes out and
4067 * calculates the property values for all 64 code points between 256 and 319,
4068 * and stores these as a single 64-bit long bit vector, called a "swatch",
4069 * under the key for code point 256. The key is the UTF-8 encoding for code
4070 * point 256, minus the final byte. Thus, if the length of the UTF-8 encoding
4071 * for a code point is 13 bytes, the key will be 12 bytes long. If the value
4072 * for code point 258 is then requested, this code realizes that it would be
4073 * stored under the key for 256, and would find that value and extract the
4074 * relevant bit, offset from 256.
4076 * Non-binary properties are stored in as many bits as necessary to represent
4077 * their values (32 currently, though the code is more general than that), not
4078 * as single bits, but the principle is the same: the value for each key is a
4079 * vector that encompasses the property values for all code points whose UTF-8
4080 * representations are represented by the key. That is, for all code points
4081 * whose UTF-8 representations are length N bytes, and the key is the first N-1
4085 Perl_swash_fetch(pTHX_ SV *swash, const U8 *ptr, bool do_utf8)
4087 HV *const hv = MUTABLE_HV(SvRV(swash));
4092 const U8 *tmps = NULL;
4096 PERL_ARGS_ASSERT_SWASH_FETCH;
4098 /* If it really isn't a hash, it isn't really swash; must be an inversion
4100 if (SvTYPE(hv) != SVt_PVHV) {
4101 return _invlist_contains_cp((SV*)hv,
4103 ? valid_utf8_to_uvchr(ptr, NULL)
4107 /* We store the values in a "swatch" which is a vec() value in a swash
4108 * hash. Code points 0-255 are a single vec() stored with key length
4109 * (klen) 0. All other code points have a UTF-8 representation
4110 * 0xAA..0xYY,0xZZ. A vec() is constructed containing all of them which
4111 * share 0xAA..0xYY, which is the key in the hash to that vec. So the key
4112 * length for them is the length of the encoded char - 1. ptr[klen] is the
4113 * final byte in the sequence representing the character */
4114 if (!do_utf8 || UTF8_IS_INVARIANT(c)) {
4119 else if (UTF8_IS_DOWNGRADEABLE_START(c)) {
4122 off = EIGHT_BIT_UTF8_TO_NATIVE(c, *(ptr + 1));
4125 klen = UTF8SKIP(ptr) - 1;
4127 /* Each vec() stores 2**UTF_ACCUMULATION_SHIFT values. The offset into
4128 * the vec is the final byte in the sequence. (In EBCDIC this is
4129 * converted to I8 to get consecutive values.) To help you visualize
4131 * Straight 1047 After final byte
4132 * UTF-8 UTF-EBCDIC I8 transform
4133 * U+0400: \xD0\x80 \xB8\x41\x41 \xB8\x41\xA0
4134 * U+0401: \xD0\x81 \xB8\x41\x42 \xB8\x41\xA1
4136 * U+0409: \xD0\x89 \xB8\x41\x4A \xB8\x41\xA9
4137 * U+040A: \xD0\x8A \xB8\x41\x51 \xB8\x41\xAA
4139 * U+0412: \xD0\x92 \xB8\x41\x59 \xB8\x41\xB2
4140 * U+0413: \xD0\x93 \xB8\x41\x62 \xB8\x41\xB3
4142 * U+041B: \xD0\x9B \xB8\x41\x6A \xB8\x41\xBB
4143 * U+041C: \xD0\x9C \xB8\x41\x70 \xB8\x41\xBC
4145 * U+041F: \xD0\x9F \xB8\x41\x73 \xB8\x41\xBF
4146 * U+0420: \xD0\xA0 \xB8\x42\x41 \xB8\x42\x41
4148 * (There are no discontinuities in the elided (...) entries.)
4149 * The UTF-8 key for these 33 code points is '\xD0' (which also is the
4150 * key for the next 31, up through U+043F, whose UTF-8 final byte is
4151 * \xBF). Thus in UTF-8, each key is for a vec() for 64 code points.
4152 * The final UTF-8 byte, which ranges between \x80 and \xBF, is an
4153 * index into the vec() swatch (after subtracting 0x80, which we
4154 * actually do with an '&').
4155 * In UTF-EBCDIC, each key is for a 32 code point vec(). The first 32
4156 * code points above have key '\xB8\x41'. The final UTF-EBCDIC byte has
4157 * dicontinuities which go away by transforming it into I8, and we
4158 * effectively subtract 0xA0 to get the index. */
4159 needents = (1 << UTF_ACCUMULATION_SHIFT);
4160 off = NATIVE_UTF8_TO_I8(ptr[klen]) & UTF_CONTINUATION_MASK;
4164 * This single-entry cache saves about 1/3 of the UTF-8 overhead in test
4165 * suite. (That is, only 7-8% overall over just a hash cache. Still,
4166 * it's nothing to sniff at.) Pity we usually come through at least
4167 * two function calls to get here...
4169 * NB: this code assumes that swatches are never modified, once generated!
4172 if (hv == PL_last_swash_hv &&
4173 klen == PL_last_swash_klen &&
4174 (!klen || memEQ((char *)ptr, (char *)PL_last_swash_key, klen)) )
4176 tmps = PL_last_swash_tmps;
4177 slen = PL_last_swash_slen;
4180 /* Try our second-level swatch cache, kept in a hash. */
4181 SV** svp = hv_fetch(hv, (const char*)ptr, klen, FALSE);
4183 /* If not cached, generate it via swatch_get */
4184 if (!svp || !SvPOK(*svp)
4185 || !(tmps = (const U8*)SvPV_const(*svp, slen)))
4188 const UV code_point = valid_utf8_to_uvchr(ptr, NULL);
4189 swatch = swatch_get(swash,
4190 code_point & ~((UV)needents - 1),
4193 else { /* For the first 256 code points, the swatch has a key of
4195 swatch = swatch_get(swash, 0, needents);
4198 if (IN_PERL_COMPILETIME)
4199 CopHINTS_set(PL_curcop, PL_hints);
4201 svp = hv_store(hv, (const char *)ptr, klen, swatch, 0);
4203 if (!svp || !(tmps = (U8*)SvPV(*svp, slen))
4204 || (slen << 3) < needents)
4205 Perl_croak(aTHX_ "panic: swash_fetch got improper swatch, "
4206 "svp=%p, tmps=%p, slen=%" UVuf ", needents=%" UVuf,
4207 svp, tmps, (UV)slen, (UV)needents);
4210 PL_last_swash_hv = hv;
4211 assert(klen <= sizeof(PL_last_swash_key));
4212 PL_last_swash_klen = (U8)klen;
4213 /* FIXME change interpvar.h? */
4214 PL_last_swash_tmps = (U8 *) tmps;
4215 PL_last_swash_slen = slen;
4217 Copy(ptr, PL_last_swash_key, klen, U8);
4220 switch ((int)((slen << 3) / needents)) {
4222 return ((UV) tmps[off >> 3] & (1 << (off & 7))) != 0;
4224 return ((UV) tmps[off]);
4228 ((UV) tmps[off ] << 8) +
4229 ((UV) tmps[off + 1]);
4233 ((UV) tmps[off ] << 24) +
4234 ((UV) tmps[off + 1] << 16) +
4235 ((UV) tmps[off + 2] << 8) +
4236 ((UV) tmps[off + 3]);
4238 Perl_croak(aTHX_ "panic: swash_fetch got swatch of unexpected bit width, "
4239 "slen=%" UVuf ", needents=%" UVuf, (UV)slen, (UV)needents);
4240 NORETURN_FUNCTION_END;
4243 /* Read a single line of the main body of the swash input text. These are of
4246 * where each number is hex. The first two numbers form the minimum and
4247 * maximum of a range, and the third is the value associated with the range.
4248 * Not all swashes should have a third number
4250 * On input: l points to the beginning of the line to be examined; it points
4251 * to somewhere in the string of the whole input text, and is
4252 * terminated by a \n or the null string terminator.
4253 * lend points to the null terminator of that string
4254 * wants_value is non-zero if the swash expects a third number
4255 * typestr is the name of the swash's mapping, like 'ToLower'
4256 * On output: *min, *max, and *val are set to the values read from the line.
4257 * returns a pointer just beyond the line examined. If there was no
4258 * valid min number on the line, returns lend+1
4262 S_swash_scan_list_line(pTHX_ U8* l, U8* const lend, UV* min, UV* max, UV* val,
4263 const bool wants_value, const U8* const typestr)
4265 const int typeto = typestr[0] == 'T' && typestr[1] == 'o';
4266 STRLEN numlen; /* Length of the number */
4267 I32 flags = PERL_SCAN_SILENT_ILLDIGIT
4268 | PERL_SCAN_DISALLOW_PREFIX
4269 | PERL_SCAN_SILENT_NON_PORTABLE;
4271 /* nl points to the next \n in the scan */
4272 U8* const nl = (U8*)memchr(l, '\n', lend - l);
4274 PERL_ARGS_ASSERT_SWASH_SCAN_LIST_LINE;
4276 /* Get the first number on the line: the range minimum */
4278 *min = grok_hex((char *)l, &numlen, &flags, NULL);
4279 *max = *min; /* So can never return without setting max */
4280 if (numlen) /* If found a hex number, position past it */
4282 else if (nl) { /* Else, go handle next line, if any */
4283 return nl + 1; /* 1 is length of "\n" */
4285 else { /* Else, no next line */
4286 return lend + 1; /* to LIST's end at which \n is not found */
4289 /* The max range value follows, separated by a BLANK */
4292 flags = PERL_SCAN_SILENT_ILLDIGIT