3 * This file contains definitions for use with the UTF-8 encoding. It
4 * actually also works with the variant UTF-8 encoding called UTF-EBCDIC, and
5 * hides almost all of the differences between these from the caller. In other
6 * words, someone should #include this file, and if the code is being compiled
7 * on an EBCDIC platform, things should mostly just work.
9 * Copyright (C) 2000, 2001, 2002, 2005, 2006, 2007, 2009,
10 * 2010, 2011 by Larry Wall and others
12 * You may distribute under the terms of either the GNU General Public
13 * License or the Artistic License, as specified in the README file.
17 #ifndef PERL_UTF8_H_ /* Guard against recursive inclusion */
18 #define PERL_UTF8_H_ 1
20 /* Use UTF-8 as the default script encoding?
21 * Turning this on will break scripts having non-UTF-8 binary
22 * data (such as Latin-1) in string literals. */
23 #ifdef USE_UTF8_SCRIPTS
24 # define USE_UTF8_IN_NAMES (!IN_BYTES)
26 # define USE_UTF8_IN_NAMES (PL_hints & HINT_UTF8)
29 #include "regcharclass.h"
30 #include "unicode_constants.h"
32 /* For to_utf8_fold_flags, q.v. */
33 #define FOLD_FLAGS_LOCALE 0x1
34 #define FOLD_FLAGS_FULL 0x2
35 #define FOLD_FLAGS_NOMIX_ASCII 0x4
37 /* For _core_swash_init(), internal core use only */
38 #define _CORE_SWASH_INIT_USER_DEFINED_PROPERTY 0x1
39 #define _CORE_SWASH_INIT_RETURN_IF_UNDEF 0x2
40 #define _CORE_SWASH_INIT_ACCEPT_INVLIST 0x4
43 =head1 Unicode Support
44 L<perlguts/Unicode Support> has an introduction to this API.
46 See also L</Character classification>,
47 and L</Character case changing>.
48 Various functions outside this section also work specially with Unicode.
49 Search for the string "utf8" in this document.
51 =for apidoc is_ascii_string
53 This is a misleadingly-named synonym for L</is_utf8_invariant_string>.
54 On ASCII-ish platforms, the name isn't misleading: the ASCII-range characters
55 are exactly the UTF-8 invariants. But EBCDIC machines have more invariants
56 than just the ASCII characters, so C<is_utf8_invariant_string> is preferred.
58 =for apidoc is_invariant_string
60 This is a somewhat misleadingly-named synonym for L</is_utf8_invariant_string>.
61 C<is_utf8_invariant_string> is preferred, as it indicates under what conditions
62 the string is invariant.
66 #define is_ascii_string(s, len) is_utf8_invariant_string(s, len)
67 #define is_invariant_string(s, len) is_utf8_invariant_string(s, len)
69 #define uvchr_to_utf8(a,b) uvchr_to_utf8_flags(a,b,0)
70 #define uvchr_to_utf8_flags(d,uv,flags) \
71 uvoffuni_to_utf8_flags(d,NATIVE_TO_UNI(uv),flags)
72 #define utf8_to_uvchr_buf(s, e, lenp) \
73 utf8n_to_uvchr(s, (U8*)(e) - (U8*)(s), lenp, \
74 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY)
75 #define utf8n_to_uvchr(s, len, lenp, flags) \
76 utf8n_to_uvchr_error(s, len, lenp, flags, 0)
78 #define to_uni_fold(c, p, lenp) _to_uni_fold_flags(c, p, lenp, FOLD_FLAGS_FULL)
80 #define to_utf8_fold(s, r, lenr) \
81 _to_utf8_fold_flags (s, NULL, r, lenr, FOLD_FLAGS_FULL, __FILE__, __LINE__)
82 #define to_utf8_lower(s, r, lenr) \
83 _to_utf8_lower_flags(s, NULL, r ,lenr, 0, __FILE__, __LINE__)
84 #define to_utf8_upper(s, r, lenr) \
85 _to_utf8_upper_flags(s, NULL, r, lenr, 0, __FILE__, __LINE__)
86 #define to_utf8_title(s, r, lenr) \
87 _to_utf8_title_flags(s, NULL, r, lenr ,0, __FILE__, __LINE__)
89 #define foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
90 foldEQ_utf8_flags(s1, pe1, l1, u1, s2, pe2, l2, u2, 0)
91 #define FOLDEQ_UTF8_NOMIX_ASCII (1 << 0)
92 #define FOLDEQ_LOCALE (1 << 1)
93 #define FOLDEQ_S1_ALREADY_FOLDED (1 << 2)
94 #define FOLDEQ_S2_ALREADY_FOLDED (1 << 3)
95 #define FOLDEQ_S1_FOLDS_SANE (1 << 4)
96 #define FOLDEQ_S2_FOLDS_SANE (1 << 5)
98 #define ibcmp_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
99 cBOOL(! foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2))
102 /* The equivalent of these macros but implementing UTF-EBCDIC
103 are in the following header file:
106 #include "utfebcdic.h"
111 /* How wide can a single UTF-8 encoded character become in bytes. */
112 /* NOTE: Strictly speaking Perl's UTF-8 should not be called UTF-8 since UTF-8
113 * is an encoding of Unicode, and Unicode's upper limit, 0x10FFFF, can be
114 * expressed with 4 bytes. However, Perl thinks of UTF-8 as a way to encode
115 * non-negative integers in a binary format, even those above Unicode */
116 #define UTF8_MAXBYTES 13
119 EXTCONST unsigned char PL_utf8skip[] = {
120 /* 0x00 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
121 /* 0x10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
122 /* 0x20 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
123 /* 0x30 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
124 /* 0x40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
125 /* 0x50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
126 /* 0x60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
127 /* 0x70 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
128 /* 0x80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
129 /* 0x90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
130 /* 0xA0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
131 /* 0xB0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
132 /* 0xC0 */ 2,2, /* overlong */
133 /* 0xC2 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0080 to U+03FF */
134 /* 0xD0 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0400 to U+07FF */
135 /* 0xE0 */ 3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, /* U+0800 to U+FFFF */
136 /* 0xF0 */ 4,4,4,4,4,4,4,4,5,5,5,5,6,6, /* above BMP to 2**31 - 1 */
137 /* Perl extended (never was official UTF-8). Up to 36 bit */
139 /* More extended, Up to 72 bits (64-bit + reserved) */
140 /* 0xFF */ UTF8_MAXBYTES
143 EXTCONST unsigned char PL_utf8skip[];
148 #if defined(_MSC_VER) && _MSC_VER < 1400
149 /* older MSVC versions have a smallish macro buffer */
150 #define PERL_SMALL_MACRO_BUFFER
153 /* Native character to/from iso-8859-1. Are the identity functions on ASCII
155 #ifdef PERL_SMALL_MACRO_BUFFER
156 #define NATIVE_TO_LATIN1(ch) ((U8)(ch))
157 #define LATIN1_TO_NATIVE(ch) ((U8)(ch))
159 #define NATIVE_TO_LATIN1(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
160 #define LATIN1_TO_NATIVE(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
163 /* I8 is an intermediate version of UTF-8 used only in UTF-EBCDIC. We thus
164 * consider it to be identical to UTF-8 on ASCII platforms. Strictly speaking
165 * UTF-8 and UTF-EBCDIC are two different things, but we often conflate them
166 * because they are 8-bit encodings that serve the same purpose in Perl, and
167 * rarely do we need to distinguish them. The term "NATIVE_UTF8" applies to
168 * whichever one is applicable on the current platform */
169 #ifdef PERL_SMALL_MACRO_BUFFER
170 #define NATIVE_UTF8_TO_I8(ch) ((U8) (ch))
171 #define I8_TO_NATIVE_UTF8(ch) ((U8) (ch))
173 #define NATIVE_UTF8_TO_I8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
174 #define I8_TO_NATIVE_UTF8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
177 /* Transforms in wide UV chars */
178 #define UNI_TO_NATIVE(ch) ((UV) (ch))
179 #define NATIVE_TO_UNI(ch) ((UV) (ch))
183 The following table is from Unicode 3.2, plus the Perl extensions for above
186 Code Points 1st Byte 2nd Byte 3rd 4th 5th 6th 7th 8th-13th
188 U+0000..U+007F 00..7F
189 U+0080..U+07FF * C2..DF 80..BF
190 U+0800..U+0FFF E0 * A0..BF 80..BF
191 U+1000..U+CFFF E1..EC 80..BF 80..BF
192 U+D000..U+D7FF ED 80..9F 80..BF
193 U+D800..U+DFFF ED A0..BF 80..BF (surrogates)
194 U+E000..U+FFFF EE..EF 80..BF 80..BF
195 U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF
196 U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF
197 U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
198 Below are above-Unicode code points
199 U+110000..U+13FFFF F4 90..BF 80..BF 80..BF
200 U+110000..U+1FFFFF F5..F7 80..BF 80..BF 80..BF
201 U+200000..U+FFFFFF F8 * 88..BF 80..BF 80..BF 80..BF
202 U+1000000..U+3FFFFFF F9..FB 80..BF 80..BF 80..BF 80..BF
203 U+4000000..U+3FFFFFFF FC * 84..BF 80..BF 80..BF 80..BF 80..BF
204 U+40000000..U+7FFFFFFF FD 80..BF 80..BF 80..BF 80..BF 80..BF
205 U+80000000..U+FFFFFFFFF FE * 82..BF 80..BF 80..BF 80..BF 80..BF 80..BF
206 U+1000000000.. FF 80..BF 80..BF 80..BF 80..BF 80..BF * 81..BF 80..BF
208 Note the gaps before several of the byte entries above marked by '*'. These are
209 caused by legal UTF-8 avoiding non-shortest encodings: it is technically
210 possible to UTF-8-encode a single code point in different ways, but that is
211 explicitly forbidden, and the shortest possible encoding should always be used
212 (and that is what Perl does). The non-shortest ones are called 'overlongs'.
217 Another way to look at it, as bits:
219 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
222 0000 0bbb bbaa aaaa 110b bbbb 10aa aaaa
223 cccc bbbb bbaa aaaa 1110 cccc 10bb bbbb 10aa aaaa
224 00 000d ddcc cccc bbbb bbaa aaaa 1111 0ddd 10cc cccc 10bb bbbb 10aa aaaa
226 As you can see, the continuation bytes all begin with C<10>, and the
227 leading bits of the start byte tell how many bytes there are in the
230 Perl's extended UTF-8 means we can have start bytes up through FF, though any
231 beginning with FF yields a code point that is too large for 32-bit ASCII
232 platforms. FF signals to use 13 bytes for the encoded character. This breaks
233 the paradigm that the number of leading bits gives how many total bytes there
234 are in the character.
238 /* Is the representation of the Unicode code point 'cp' the same regardless of
239 * being encoded in UTF-8 or not? */
240 #define OFFUNI_IS_INVARIANT(cp) isASCII(cp)
243 =for apidoc Am|bool|UVCHR_IS_INVARIANT|UV cp
245 Evaluates to 1 if the representation of code point C<cp> is the same whether or
246 not it is encoded in UTF-8; otherwise evaluates to 0. UTF-8 invariant
247 characters can be copied as-is when converting to/from UTF-8, saving time.
248 C<cp> is Unicode if above 255; otherwise is platform-native.
253 #define UVCHR_IS_INVARIANT(cp) OFFUNI_IS_INVARIANT(cp)
255 /* This defines the bits that are to be in the continuation bytes of a multi-byte
256 * UTF-8 encoded character that mark it is a continuation byte. */
257 #define UTF_CONTINUATION_MARK 0x80
259 /* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence
260 * in UTF-8? This is the inverse of UTF8_IS_INVARIANT. The |0 makes sure this
261 * isn't mistakenly called with a ptr argument */
262 #define UTF8_IS_CONTINUED(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
263 ((U8)((c) | 0)) & UTF_CONTINUATION_MARK)
265 /* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence?
266 * This doesn't catch invariants (they are single-byte). It also excludes the
267 * illegal overlong sequences that begin with C0 and C1. The |0 makes sure
268 * this isn't mistakenly called with a ptr argument */
269 #define UTF8_IS_START(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
270 ((U8)((c) | 0)) >= 0xc2)
272 /* For use in UTF8_IS_CONTINUATION() below */
273 #define UTF_IS_CONTINUATION_MASK 0xC0
275 /* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the
276 * first byte thereof? The |0 makes sure this isn't mistakenly called with a
278 #define UTF8_IS_CONTINUATION(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
279 (((U8)((c) | 0)) & UTF_IS_CONTINUATION_MASK) == UTF_CONTINUATION_MARK)
281 /* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use
282 * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to
283 * be well-formed. Masking with 0xfe allows the low bit to be 0 or 1; thus
284 * this matches 0xc[23]. The |0 makes sure this isn't mistakenly called with a
286 #define UTF8_IS_DOWNGRADEABLE_START(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
287 (((U8)((c) | 0)) & 0xfe) == 0xc2)
289 /* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that
290 * represent a code point > 255? The |0 makes sure this isn't mistakenly
291 * called with a ptr argument */
292 #define UTF8_IS_ABOVE_LATIN1(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
293 ((U8)((c) | 0)) >= 0xc4)
295 /* This is the number of low-order bits a continuation byte in a UTF-8 encoded
296 * sequence contributes to the specification of the code point. In the bit
297 * maps above, you see that the first 2 bits are a constant '10', leaving 6 of
298 * real information */
299 #define UTF_ACCUMULATION_SHIFT 6
301 /* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL()
303 #define QUESTION_MARK_CTRL DEL_NATIVE
305 /* Surrogates, non-character code points and above-Unicode code points are
306 * problematic in some contexts. This allows code that needs to check for
307 * those to to quickly exclude the vast majority of code points it will
309 #define isUTF8_POSSIBLY_PROBLEMATIC(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
312 /* A helper macro for isUTF8_CHAR, so use that one instead of this. This was
313 * generated by regen/regcharclass.pl, and then moved here. Then it was
314 * hand-edited to add some LIKELY() calls, presuming that malformations are
315 * unlikely. The lines that generated it were then commented out. This was
316 * done because it takes on the order of 10 minutes to generate, and is never
317 * going to change, unless the generated code is improved, and figuring out
318 * the LIKELYs there would be hard.
320 UTF8_CHAR: Matches legal UTF-8 variant code points up through 0x1FFFFFF
324 /*** GENERATED CODE ***/
325 #define is_UTF8_CHAR_utf8_no_length_checks(s) \
326 ( ( 0xC2 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xDF ) ? \
327 ( LIKELY( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) ? 2 : 0 ) \
328 : ( 0xE0 == ((const U8*)s)[0] ) ? \
329 ( LIKELY( ( ( ((const U8*)s)[1] & 0xE0 ) == 0xA0 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
330 : ( 0xE1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xEF ) ? \
331 ( LIKELY( ( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
332 : ( 0xF0 == ((const U8*)s)[0] ) ? \
333 ( LIKELY( ( ( 0x90 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xBF ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
334 : ( ( ( ( 0xF1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xF7 ) && LIKELY( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) ) && LIKELY( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) && LIKELY( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )
336 /* The above macro handles UTF-8 that has this start byte as the maximum */
337 #define _IS_UTF8_CHAR_HIGHEST_START_BYTE 0xF7
339 /* A helper macro for isSTRICT_UTF8_CHAR, so use that one instead of this.
340 * Like is_UTF8_CHAR_utf8_no_length_checks(), this was moved here and LIKELYs
343 STRICT_UTF8_CHAR: Matches legal Unicode UTF-8 variant code points, no
344 surrrogates nor non-character code points
346 /*** GENERATED CODE ***/
347 #define is_STRICT_UTF8_CHAR_utf8_no_length_checks(s) \
348 ( ( 0xC2 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xDF ) ? \
349 ( LIKELY( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) ? 2 : 0 ) \
350 : ( 0xE0 == ((const U8*)s)[0] ) ? \
351 ( LIKELY( ( ( ((const U8*)s)[1] & 0xE0 ) == 0xA0 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
352 : ( ( 0xE1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xEC ) || 0xEE == ((const U8*)s)[0] ) ?\
353 ( ( ( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
354 : ( 0xED == ((const U8*)s)[0] ) ? \
355 ( LIKELY( ( ( ((const U8*)s)[1] & 0xE0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
356 : ( 0xEF == ((const U8*)s)[0] ) ? \
357 ( ( ( 0x80 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xB6 ) || ( 0xB8 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xBE ) ) ?\
358 ( LIKELY( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ? 3 : 0 ) \
359 : ( 0xB7 == ((const U8*)s)[1] ) ? \
360 ( LIKELY( ( ((const U8*)s)[2] & 0xF0 ) == 0x80 || ( ((const U8*)s)[2] & 0xF0 ) == 0xB0 ) ? 3 : 0 )\
361 : ( ( 0xBF == ((const U8*)s)[1] ) && ( 0x80 <= ((const U8*)s)[2] && ((const U8*)s)[2] <= 0xBD ) ) ? 3 : 0 )\
362 : ( 0xF0 == ((const U8*)s)[0] ) ? \
363 ( ( ( 0x90 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0x9E ) || ( 0xA0 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xAE ) || ( 0xB0 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xBE ) ) ?\
364 ( LIKELY( ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
365 : ( ((const U8*)s)[1] == 0x9F || ( ( ((const U8*)s)[1] & 0xEF ) == 0xAF ) ) ? \
366 ( ( 0x80 <= ((const U8*)s)[2] && ((const U8*)s)[2] <= 0xBE ) ? \
367 ( LIKELY( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ? 4 : 0 ) \
368 : LIKELY( ( 0xBF == ((const U8*)s)[2] ) && ( 0x80 <= ((const U8*)s)[3] && ((const U8*)s)[3] <= 0xBD ) ) ? 4 : 0 )\
370 : ( 0xF1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xF3 ) ? \
371 ( ( ( ( ((const U8*)s)[1] & 0xC8 ) == 0x80 ) || ( ( ((const U8*)s)[1] & 0xCC ) == 0x88 ) || ( ( ((const U8*)s)[1] & 0xCE ) == 0x8C ) || ( ( ((const U8*)s)[1] & 0xCF ) == 0x8E ) ) ?\
372 ( LIKELY( ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
373 : ( ( ((const U8*)s)[1] & 0xCF ) == 0x8F ) ? \
374 ( ( 0x80 <= ((const U8*)s)[2] && ((const U8*)s)[2] <= 0xBE ) ? \
375 ( LIKELY( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ? 4 : 0 ) \
376 : LIKELY( ( 0xBF == ((const U8*)s)[2] ) && ( 0x80 <= ((const U8*)s)[3] && ((const U8*)s)[3] <= 0xBD ) ) ? 4 : 0 )\
378 : ( 0xF4 == ((const U8*)s)[0] ) ? \
379 ( ( 0x80 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0x8E ) ? \
380 ( LIKELY( ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
381 : ( 0x8F == ((const U8*)s)[1] ) ? \
382 ( ( 0x80 <= ((const U8*)s)[2] && ((const U8*)s)[2] <= 0xBE ) ? \
383 ( LIKELY( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ? 4 : 0 ) \
384 : LIKELY( ( 0xBF == ((const U8*)s)[2] ) && ( 0x80 <= ((const U8*)s)[3] && ((const U8*)s)[3] <= 0xBD ) ) ? 4 : 0 )\
389 C9_STRICT_UTF8_CHAR: Matches legal Unicode UTF-8 variant code
390 points, no surrogates
394 /*** GENERATED CODE ***/
395 #define is_C9_STRICT_UTF8_CHAR_utf8_no_length_checks(s) \
396 ( ( 0xC2 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xDF ) ? \
397 ( LIKELY( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) ? 2 : 0 ) \
398 : ( 0xE0 == ((const U8*)s)[0] ) ? \
399 ( LIKELY( ( ( ((const U8*)s)[1] & 0xE0 ) == 0xA0 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
400 : ( ( 0xE1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xEC ) || ( ((const U8*)s)[0] & 0xFE ) == 0xEE ) ?\
401 ( LIKELY( ( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
402 : ( 0xED == ((const U8*)s)[0] ) ? \
403 ( LIKELY( ( ( ((const U8*)s)[1] & 0xE0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
404 : ( 0xF0 == ((const U8*)s)[0] ) ? \
405 ( LIKELY( ( ( 0x90 <= ((const U8*)s)[1] && ((const U8*)s)[1] <= 0xBF ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
406 : ( 0xF1 <= ((const U8*)s)[0] && ((const U8*)s)[0] <= 0xF3 ) ? \
407 ( LIKELY( ( ( ( ((const U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
408 : LIKELY( ( ( ( 0xF4 == ((const U8*)s)[0] ) && ( ( ((const U8*)s)[1] & 0xF0 ) == 0x80 ) ) && ( ( ((const U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((const U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )
410 #define UNICODE_IS_PERL_EXTENDED(uv) UNLIKELY((UV) (uv) > 0x7FFFFFFF)
412 #endif /* EBCDIC vs ASCII */
414 /* 2**UTF_ACCUMULATION_SHIFT - 1 */
415 #define UTF_CONTINUATION_MASK ((U8) ((1U << UTF_ACCUMULATION_SHIFT) - 1))
417 /* Internal macro to be used only in this file to aid in constructing other
418 * publicly accessible macros.
419 * The number of bytes required to express this uv in UTF-8, for just those
420 * uv's requiring 2 through 6 bytes, as these are common to all platforms and
421 * word sizes. The number of bytes needed is given by the number of leading 1
422 * bits in the start byte. There are 32 start bytes that have 2 initial 1 bits
423 * (C0-DF); there are 16 that have 3 initial 1 bits (E0-EF); 8 that have 4
424 * initial 1 bits (F0-F8); 4 that have 5 initial 1 bits (F9-FB), and 2 that
425 * have 6 initial 1 bits (FC-FD). The largest number a string of n bytes can
426 * represent is (the number of possible start bytes for 'n')
427 * * (the number of possiblities for each start byte
428 * The latter in turn is
429 * 2 ** ( (how many continuation bytes there are)
430 * * (the number of bits of information each
431 * continuation byte holds))
433 * If we were on a platform where we could use a fast find first set bit
434 * instruction (or count leading zeros instruction) this could be replaced by
435 * using that to find the log2 of the uv, and divide that by the number of bits
436 * of information in each continuation byte, adjusting for large cases and how
437 * much information is in a start byte for that length */
438 #define __COMMON_UNI_SKIP(uv) \
439 (UV) (uv) < (32 * (1U << ( UTF_ACCUMULATION_SHIFT))) ? 2 : \
440 (UV) (uv) < (16 * (1U << (2 * UTF_ACCUMULATION_SHIFT))) ? 3 : \
441 (UV) (uv) < ( 8 * (1U << (3 * UTF_ACCUMULATION_SHIFT))) ? 4 : \
442 (UV) (uv) < ( 4 * (1U << (4 * UTF_ACCUMULATION_SHIFT))) ? 5 : \
443 (UV) (uv) < ( 2 * (1U << (5 * UTF_ACCUMULATION_SHIFT))) ? 6 :
445 /* Internal macro to be used only in this file.
446 * This adds to __COMMON_UNI_SKIP the details at this platform's upper range.
447 * For any-sized EBCDIC platforms, or 64-bit ASCII ones, we need one more test
448 * to see if just 7 bytes is needed, or if the maximum is needed. For 32-bit
449 * ASCII platforms, everything is representable by 7 bytes */
450 #if defined(UV_IS_QUAD) || defined(EBCDIC)
451 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) \
452 (UV) (uv) < ((UV) 1U << (6 * UTF_ACCUMULATION_SHIFT)) ? 7 : UTF8_MAXBYTES)
454 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) 7)
457 /* The next two macros use the base macro defined above, and add in the tests
458 * at the low-end of the range, for just 1 byte, yielding complete macros,
459 * publicly accessible. */
461 /* Input is a true Unicode (not-native) code point */
462 #define OFFUNISKIP(uv) (OFFUNI_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
466 =for apidoc Am|STRLEN|UVCHR_SKIP|UV cp
467 returns the number of bytes required to represent the code point C<cp> when
468 encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than
469 255; a Unicode code point otherwise.
473 #define UVCHR_SKIP(uv) ( UVCHR_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
475 /* The largest code point representable by two UTF-8 bytes on this platform.
476 * As explained in the comments for __COMMON_UNI_SKIP, 32 start bytes with
477 * UTF_ACCUMULATION_SHIFT bits of information each */
478 #define MAX_UTF8_TWO_BYTE (32 * (1U << UTF_ACCUMULATION_SHIFT) - 1)
480 /* The largest code point representable by two UTF-8 bytes on any platform that
481 * Perl runs on. This value is constrained by EBCDIC which has 5 bits per
482 * continuation byte */
483 #define MAX_PORTABLE_UTF8_TWO_BYTE (32 * (1U << 5) - 1)
485 /* The maximum number of UTF-8 bytes a single Unicode character can
486 * uppercase/lowercase/fold into. Unicode guarantees that the maximum
487 * expansion is UTF8_MAX_FOLD_CHAR_EXPAND characters, but any above-Unicode
488 * code point will fold to itself, so we only have to look at the expansion of
489 * the maximum Unicode code point. But this number may be less than the space
490 * occupied by a very large code point under Perl's extended UTF-8. We have to
491 * make it large enough to fit any single character. (It turns out that ASCII
492 * and EBCDIC differ in which is larger) */
493 #define UTF8_MAXBYTES_CASE \
494 (UTF8_MAXBYTES >= (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)) \
496 : (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)))
498 /* Rest of these are attributes of Unicode and perl's internals rather than the
499 * encoding, or happen to be the same in both ASCII and EBCDIC (at least at
500 * this level; the macros that some of these call may have different
501 * definitions in the two encodings */
503 /* In domain restricted to ASCII, these may make more sense to the reader than
504 * the ones with Latin1 in the name */
505 #define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch)
506 #define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch)
508 /* More or less misleadingly-named defines, retained for back compat */
509 #define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch)
510 #define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch)
511 #define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
512 #define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
513 #define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch)
515 /* This defines the 1-bits that are to be in the first byte of a multi-byte
516 * UTF-8 encoded character that mark it as a start byte and give the number of
517 * bytes that comprise the character. 'len' is the number of bytes in the
518 * multi-byte sequence. */
519 #define UTF_START_MARK(len) (((len) > 7) ? 0xFF : (0xFF & (0xFE << (7-(len)))))
521 /* Masks out the initial one bits in a start byte, leaving the real data ones.
522 * Doesn't work on an invariant byte. 'len' is the number of bytes in the
523 * multi-byte sequence that comprises the character. */
524 #define UTF_START_MASK(len) (((len) >= 7) ? 0x00 : (0x1F >> ((len)-2)))
526 /* Adds a UTF8 continuation byte 'new' of information to a running total code
527 * point 'old' of all the continuation bytes so far. This is designed to be
528 * used in a loop to convert from UTF-8 to the code point represented. Note
529 * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is
530 * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC)
531 * code point in process of being generated */
532 #define UTF8_ACCUMULATE(old, new) (__ASSERT_(FITS_IN_8_BITS(new)) \
533 ((old) << UTF_ACCUMULATION_SHIFT) \
534 | ((NATIVE_UTF8_TO_I8((U8)new)) \
535 & UTF_CONTINUATION_MASK))
537 /* This works in the face of malformed UTF-8. */
538 #define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) \
539 ( UTF8_IS_DOWNGRADEABLE_START(*(s)) \
540 && ( (e) - (s) > 1) \
541 && UTF8_IS_CONTINUATION(*((s)+1)))
543 /* Number of bytes a code point occupies in UTF-8. */
544 #define NATIVE_SKIP(uv) UVCHR_SKIP(uv)
546 /* Most code which says UNISKIP is really thinking in terms of native code
547 * points (0-255) plus all those beyond. This is an imprecise term, but having
548 * it means existing code continues to work. For precision, use UVCHR_SKIP,
549 * NATIVE_SKIP, or OFFUNISKIP */
550 #define UNISKIP(uv) UVCHR_SKIP(uv)
552 /* Longer, but more accurate name */
553 #define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c)
555 /* Convert a UTF-8 variant Latin1 character to a native code point value.
556 * Needs just one iteration of accumulate. Should be used only if it is known
557 * that the code point is < 256, and is not UTF-8 invariant. Use the slower
558 * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point
559 * representable by two bytes (which turns out to be up through
560 * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are:
561 * HI: a downgradable start byte;
564 #define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \
565 ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \
566 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
567 LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \
568 NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO))))
570 /* Convert a two (not one) byte utf8 character to a native code point value.
571 * Needs just one iteration of accumulate. Should not be used unless it is
572 * known that the two bytes are legal: 1) two-byte start, and 2) continuation.
573 * Note that the result can be larger than 255 if the input character is not
575 #define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \
576 (__ASSERT_(FITS_IN_8_BITS(HI)) \
577 __ASSERT_(FITS_IN_8_BITS(LO)) \
578 __ASSERT_(PL_utf8skip[HI] == 2) \
579 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
580 UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \
583 /* Should never be used, and be deprecated */
584 #define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO))
588 =for apidoc Am|STRLEN|UTF8SKIP|char* s
589 returns the number of bytes in the UTF-8 encoded character whose first (perhaps
590 only) byte is pointed to by C<s>.
594 #define UTF8SKIP(s) PL_utf8skip[*(const U8*)(s)]
595 #define UTF8_SKIP(s) UTF8SKIP(s)
597 /* Most code that says 'UNI_' really means the native value for code points up
599 #define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp)
602 =for apidoc Am|bool|UTF8_IS_INVARIANT|char c
604 Evaluates to 1 if the byte C<c> represents the same character when encoded in
605 UTF-8 as when not; otherwise evaluates to 0. UTF-8 invariant characters can be
606 copied as-is when converting to/from UTF-8, saving time.
608 In spite of the name, this macro gives the correct result if the input string
609 from which C<c> comes is not encoded in UTF-8.
611 See C<L</UVCHR_IS_INVARIANT>> for checking if a UV is invariant.
615 The reason it works on both UTF-8 encoded strings and non-UTF-8 encoded, is
616 that it returns TRUE in each for the exact same set of bit patterns. It is
617 valid on a subset of what UVCHR_IS_INVARIANT is valid on, so can just use that;
618 and the compiler should optimize out anything extraneous given the
619 implementation of the latter. The |0 makes sure this isn't mistakenly called
622 #define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT((c) | 0)
624 /* Like the above, but its name implies a non-UTF8 input, which as the comments
625 * above show, doesn't matter as to its implementation */
626 #define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
628 /* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic
629 * bytes from an ordinal that is known to fit into exactly two (not one) bytes;
630 * it must be less than 0x3FF to work across both encodings. */
632 /* These two are helper macros for the other three sets, and should not be used
633 * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1
634 * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works
635 * for any code point */
636 #define __BASE_TWO_BYTE_HI(c, translate_function) \
637 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
638 I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \
639 | UTF_START_MARK(2)))
640 #define __BASE_TWO_BYTE_LO(c, translate_function) \
641 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
642 I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \
643 | UTF_CONTINUATION_MARK))
645 /* The next two macros should not be used. They were designed to be usable as
646 * the case label of a switch statement, but this doesn't work for EBCDIC. Use
647 * regen/unicode_constants.pl instead */
648 #define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)
649 #define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)
651 /* The next two macros are used when the source should be a single byte
652 * character; checked for under DEBUGGING */
653 #define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
654 ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1)))
655 #define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
656 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1)))
658 /* These final two macros in the series are used when the source can be any
659 * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient
660 * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~'
661 * operator instead of "<=" to avoid getting compiler warnings.
662 * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few
663 * places, so the ~ works */
664 #define UTF8_TWO_BYTE_HI(c) \
665 (__ASSERT_((sizeof(c) == 1) \
666 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
667 (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)))
668 #define UTF8_TWO_BYTE_LO(c) \
669 (__ASSERT_((sizeof(c) == 1) \
670 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
671 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)))
673 /* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII
674 * as it is only in overlongs. */
675 #define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1)
678 * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem
679 * from an earlier version of these macros in which they didn't call the
680 * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the
681 * beginning of a utf8 character. Now that foo_utf8() determines that itself,
682 * no need to do it again here
684 #define isIDFIRST_lazy_if(p,UTF) \
685 _is_utf8_FOO(_CC_IDFIRST, (const U8 *) p, "isIDFIRST_lazy_if", \
686 "isIDFIRST_lazy_if_safe", \
687 cBOOL(UTF && ! IN_BYTES), 0, __FILE__,__LINE__)
689 #define isIDFIRST_lazy_if_safe(p, e, UTF) \
690 ((IN_BYTES || !UTF) \
692 : isIDFIRST_utf8_safe(p, e))
694 #define isWORDCHAR_lazy_if(p,UTF) \
695 _is_utf8_FOO(_CC_IDFIRST, (const U8 *) p, "isWORDCHAR_lazy_if", \
696 "isWORDCHAR_lazy_if_safe", \
697 cBOOL(UTF && ! IN_BYTES), 0, __FILE__,__LINE__)
699 #define isWORDCHAR_lazy_if_safe(p, e, UTF) \
700 ((IN_BYTES || !UTF) \
702 : isWORDCHAR_utf8_safe((U8 *) p, (U8 *) e))
704 #define isALNUM_lazy_if(p,UTF) \
705 _is_utf8_FOO(_CC_IDFIRST, (const U8 *) p, "isALNUM_lazy_if", \
706 "isWORDCHAR_lazy_if_safe", \
707 cBOOL(UTF && ! IN_BYTES), 0, __FILE__,__LINE__)
709 #define UTF8_MAXLEN UTF8_MAXBYTES
711 /* A Unicode character can fold to up to 3 characters */
712 #define UTF8_MAX_FOLD_CHAR_EXPAND 3
714 #define IN_BYTES UNLIKELY(CopHINTS_get(PL_curcop) & HINT_BYTES)
718 =for apidoc Am|bool|DO_UTF8|SV* sv
719 Returns a bool giving whether or not the PV in C<sv> is to be treated as being
722 You should use this I<after> a call to C<SvPV()> or one of its variants, in
723 case any call to string overloading updates the internal UTF-8 encoding flag.
727 #define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES)
729 /* Should all strings be treated as Unicode, and not just UTF-8 encoded ones?
730 * Is so within 'feature unicode_strings' or 'locale :not_characters', and not
731 * within 'use bytes'. UTF-8 locales are not tested for here, but perhaps
733 #define IN_UNI_8_BIT \
734 (( ( (CopHINTS_get(PL_curcop) & HINT_UNI_8_BIT)) \
735 || ( CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \
736 /* -1 below is for :not_characters */ \
737 && _is_in_locale_category(FALSE, -1))) \
741 #define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */
742 #define UTF8_GOT_EMPTY UTF8_ALLOW_EMPTY
744 /* Allow first byte to be a continuation byte */
745 #define UTF8_ALLOW_CONTINUATION 0x0002
746 #define UTF8_GOT_CONTINUATION UTF8_ALLOW_CONTINUATION
748 /* Unexpected continuation byte */
749 #define UTF8_ALLOW_NON_CONTINUATION 0x0004
750 #define UTF8_GOT_NON_CONTINUATION UTF8_ALLOW_NON_CONTINUATION
752 /* expecting more bytes than were available in the string */
753 #define UTF8_ALLOW_SHORT 0x0008
754 #define UTF8_GOT_SHORT UTF8_ALLOW_SHORT
756 /* Overlong sequence; i.e., the code point can be specified in fewer bytes.
757 * First one will convert the overlong to the REPLACEMENT CHARACTER; second
758 * will return what the overlong evaluates to */
759 #define UTF8_ALLOW_LONG 0x0010
760 #define UTF8_ALLOW_LONG_AND_ITS_VALUE (UTF8_ALLOW_LONG|0x0020)
761 #define UTF8_GOT_LONG UTF8_ALLOW_LONG
763 #define UTF8_ALLOW_OVERFLOW 0x0080
764 #define UTF8_GOT_OVERFLOW UTF8_ALLOW_OVERFLOW
766 #define UTF8_DISALLOW_SURROGATE 0x0100 /* Unicode surrogates */
767 #define UTF8_GOT_SURROGATE UTF8_DISALLOW_SURROGATE
768 #define UTF8_WARN_SURROGATE 0x0200
770 /* Unicode non-character code points */
771 #define UTF8_DISALLOW_NONCHAR 0x0400
772 #define UTF8_GOT_NONCHAR UTF8_DISALLOW_NONCHAR
773 #define UTF8_WARN_NONCHAR 0x0800
775 /* Super-set of Unicode: code points above the legal max */
776 #define UTF8_DISALLOW_SUPER 0x1000
777 #define UTF8_GOT_SUPER UTF8_DISALLOW_SUPER
778 #define UTF8_WARN_SUPER 0x2000
780 /* The original UTF-8 standard did not define UTF-8 with start bytes of 0xFE or
781 * 0xFF, though UTF-EBCDIC did. This allowed both versions to represent code
782 * points up to 2 ** 31 - 1. Perl extends UTF-8 so that 0xFE and 0xFF are
783 * usable on ASCII platforms, and 0xFF means something different than
784 * UTF-EBCDIC defines. These changes allow code points of 64 bits (actually
785 * somewhat more) to be represented on both platforms. But these are Perl
786 * extensions, and not likely to be interchangeable with other languages. Note
787 * that on ASCII platforms, FE overflows a signed 32-bit word, and FF an
789 #define UTF8_DISALLOW_PERL_EXTENDED 0x4000
790 #define UTF8_GOT_PERL_EXTENDED UTF8_DISALLOW_PERL_EXTENDED
791 #define UTF8_WARN_PERL_EXTENDED 0x8000
793 /* For back compat, these old names are misleading for overlongs and
795 #define UTF8_DISALLOW_ABOVE_31_BIT UTF8_DISALLOW_PERL_EXTENDED
796 #define UTF8_GOT_ABOVE_31_BIT UTF8_GOT_PERL_EXTENDED
797 #define UTF8_WARN_ABOVE_31_BIT UTF8_WARN_PERL_EXTENDED
798 #define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_PERL_EXTENDED
799 #define UTF8_WARN_FE_FF UTF8_WARN_PERL_EXTENDED
801 #define UTF8_CHECK_ONLY 0x10000
802 #define _UTF8_NO_CONFIDENCE_IN_CURLEN 0x20000 /* Internal core use only */
804 /* For backwards source compatibility. They do nothing, as the default now
805 * includes what they used to mean. The first one's meaning was to allow the
806 * just the single non-character 0xFFFF */
807 #define UTF8_ALLOW_FFFF 0
808 #define UTF8_ALLOW_SURROGATE 0
810 /* C9 refers to Unicode Corrigendum #9: allows but discourages non-chars */
811 #define UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE \
812 (UTF8_DISALLOW_SUPER|UTF8_DISALLOW_SURROGATE)
813 #define UTF8_WARN_ILLEGAL_C9_INTERCHANGE (UTF8_WARN_SUPER|UTF8_WARN_SURROGATE)
815 #define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \
816 (UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE|UTF8_DISALLOW_NONCHAR)
817 #define UTF8_WARN_ILLEGAL_INTERCHANGE \
818 (UTF8_WARN_ILLEGAL_C9_INTERCHANGE|UTF8_WARN_NONCHAR)
820 /* This is typically used for code that processes UTF-8 input and doesn't want
821 * to have to deal with any malformations that might be present. All such will
822 * be safely replaced by the REPLACEMENT CHARACTER, unless other flags
823 * overriding this are also present. */
824 #define UTF8_ALLOW_ANY ( UTF8_ALLOW_CONTINUATION \
825 |UTF8_ALLOW_NON_CONTINUATION \
828 |UTF8_ALLOW_OVERFLOW)
830 /* Accept any Perl-extended UTF-8 that evaluates to any UV on the platform, but
831 * not any malformed. This is the default. (Note that UVs above IV_MAX are
833 #define UTF8_ALLOW_ANYUV 0
834 #define UTF8_ALLOW_DEFAULT UTF8_ALLOW_ANYUV
837 =for apidoc Am|bool|UTF8_IS_SURROGATE|const U8 *s|const U8 *e
839 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
840 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
841 of the Unicode surrogate code points; otherwise it evaluates to 0. If
842 non-zero, the value gives how many bytes starting at C<s> comprise the code
843 point's representation.
847 #define UTF8_IS_SURROGATE(s, e) is_SURROGATE_utf8_safe(s, e)
850 #define UTF8_IS_REPLACEMENT(s, send) is_REPLACEMENT_utf8_safe(s,send)
853 =for apidoc Am|bool|UTF8_IS_SUPER|const U8 *s|const U8 *e
855 Recall that Perl recognizes an extension to UTF-8 that can encode code
856 points larger than the ones defined by Unicode, which are 0..0x10FFFF.
858 This macro evaluates to non-zero if the first few bytes of the string starting
859 at C<s> and looking no further than S<C<e - 1>> are from this UTF-8 extension;
860 otherwise it evaluates to 0. If non-zero, the value gives how many bytes
861 starting at C<s> comprise the code point's representation.
863 0 is returned if the bytes are not well-formed extended UTF-8, or if they
864 represent a code point that cannot fit in a UV on the current platform. Hence
865 this macro can give different results when run on a 64-bit word machine than on
866 one with a 32-bit word size.
868 Note that it is deprecated to have code points that are larger than what can
869 fit in an IV on the current machine.
874 * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode
875 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0
876 * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1
879 # define UTF8_IS_SUPER(s, e) \
880 (( LIKELY((e) > (s) + 4) \
881 && NATIVE_UTF8_TO_I8(*(s)) >= 0xF9 \
882 && ( NATIVE_UTF8_TO_I8(*(s)) > 0xF9 \
883 || (NATIVE_UTF8_TO_I8(*((s) + 1)) >= 0xA2)) \
884 && LIKELY((s) + UTF8SKIP(s) <= (e))) \
885 ? _is_utf8_char_helper(s, s + UTF8SKIP(s), 0) : 0)
887 # define UTF8_IS_SUPER(s, e) \
888 (( LIKELY((e) > (s) + 3) \
889 && (*(U8*) (s)) >= 0xF4 \
890 && ((*(U8*) (s)) > 0xF4 || (*((U8*) (s) + 1) >= 0x90))\
891 && LIKELY((s) + UTF8SKIP(s) <= (e))) \
892 ? _is_utf8_char_helper(s, s + UTF8SKIP(s), 0) : 0)
895 /* These are now machine generated, and the 'given' clause is no longer
897 #define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \
898 cBOOL(is_NONCHAR_utf8_safe(s,e))
901 =for apidoc Am|bool|UTF8_IS_NONCHAR|const U8 *s|const U8 *e
903 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
904 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
905 of the Unicode non-character code points; otherwise it evaluates to 0. If
906 non-zero, the value gives how many bytes starting at C<s> comprise the code
907 point's representation.
911 #define UTF8_IS_NONCHAR(s, e) \
912 UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e)
914 #define UNICODE_SURROGATE_FIRST 0xD800
915 #define UNICODE_SURROGATE_LAST 0xDFFF
916 #define UNICODE_REPLACEMENT 0xFFFD
917 #define UNICODE_BYTE_ORDER_MARK 0xFEFF
919 /* Though our UTF-8 encoding can go beyond this,
920 * let's be conservative and do as Unicode says. */
921 #define PERL_UNICODE_MAX 0x10FFFF
923 #define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */
924 #define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */
925 #define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */
926 #define UNICODE_WARN_PERL_EXTENDED 0x0008 /* Above 0x7FFF_FFFF */
927 #define UNICODE_WARN_ABOVE_31_BIT UNICODE_WARN_PERL_EXTENDED
928 #define UNICODE_DISALLOW_SURROGATE 0x0010
929 #define UNICODE_DISALLOW_NONCHAR 0x0020
930 #define UNICODE_DISALLOW_SUPER 0x0040
931 #define UNICODE_DISALLOW_PERL_EXTENDED 0x0080
932 #define UNICODE_DISALLOW_ABOVE_31_BIT UNICODE_DISALLOW_PERL_EXTENDED
933 #define UNICODE_WARN_ILLEGAL_C9_INTERCHANGE \
934 (UNICODE_WARN_SURROGATE|UNICODE_WARN_SUPER)
935 #define UNICODE_WARN_ILLEGAL_INTERCHANGE \
936 (UNICODE_WARN_ILLEGAL_C9_INTERCHANGE|UNICODE_WARN_NONCHAR)
937 #define UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE \
938 (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_SUPER)
939 #define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \
940 (UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE|UNICODE_DISALLOW_NONCHAR)
942 /* For backward source compatibility, as are now the default */
943 #define UNICODE_ALLOW_SURROGATE 0
944 #define UNICODE_ALLOW_SUPER 0
945 #define UNICODE_ALLOW_ANY 0
947 /* This matches the 2048 code points between UNICODE_SURROGATE_FIRST (0xD800) and
948 * UNICODE_SURROGATE_LAST (0xDFFF) */
949 #define UNICODE_IS_SURROGATE(uv) (((UV) (uv) & (~0xFFFF | 0xF800)) \
952 #define UNICODE_IS_REPLACEMENT(uv) ((UV) (uv) == UNICODE_REPLACEMENT)
953 #define UNICODE_IS_BYTE_ORDER_MARK(uv) ((UV) (uv) == UNICODE_BYTE_ORDER_MARK)
955 /* Is 'uv' one of the 32 contiguous-range noncharacters? */
956 #define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) ((UV) (uv) >= 0xFDD0 \
957 && (UV) (uv) <= 0xFDEF)
959 /* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE,
960 * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above
961 * the Unicode legal max */
962 #define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \
963 (((UV) (uv) & 0xFFFE) == 0xFFFE)
965 #define UNICODE_IS_NONCHAR(uv) \
966 ( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \
967 || ( LIKELY( ! UNICODE_IS_SUPER(uv)) \
968 && UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
970 #define UNICODE_IS_SUPER(uv) ((UV) (uv) > PERL_UNICODE_MAX)
972 #define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE
973 #define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \
974 LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE
975 #define MICRO_SIGN MICRO_SIGN_NATIVE
976 #define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \
977 LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE
978 #define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \
979 LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE
980 #define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3
981 #define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2
982 #define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3
983 #define GREEK_SMALL_LETTER_MU 0x03BC
984 #define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case
986 #define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */
987 #ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
988 # define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E
990 #define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130
991 #define LATIN_SMALL_LETTER_DOTLESS_I 0x131
992 #define LATIN_SMALL_LETTER_LONG_S 0x017F
993 #define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05
994 #define LATIN_SMALL_LIGATURE_ST 0xFB06
995 #define KELVIN_SIGN 0x212A
996 #define ANGSTROM_SIGN 0x212B
998 #define UNI_DISPLAY_ISPRINT 0x0001
999 #define UNI_DISPLAY_BACKSLASH 0x0002
1000 #define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
1001 #define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
1003 #define ANYOF_FOLD_SHARP_S(node, input, end) \
1004 (ANYOF_BITMAP_TEST(node, LATIN_SMALL_LETTER_SHARP_S) && \
1005 (ANYOF_NONBITMAP(node)) && \
1006 (ANYOF_FLAGS(node) & ANYOF_LOC_NONBITMAP_FOLD) && \
1007 ((end) > (input) + 1) && \
1008 isALPHA_FOLD_EQ((input)[0], 's'))
1010 #define SHARP_S_SKIP 2
1014 =for apidoc Am|STRLEN|isUTF8_CHAR|const U8 *s|const U8 *e
1016 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1017 looking no further than S<C<e - 1>> are well-formed UTF-8, as extended by Perl,
1018 that represents some code point; otherwise it evaluates to 0. If non-zero, the
1019 value gives how many bytes starting at C<s> comprise the code point's
1020 representation. Any bytes remaining before C<e>, but beyond the ones needed to
1021 form the first code point in C<s>, are not examined.
1023 The code point can be any that will fit in a UV on this machine, using Perl's
1024 extension to official UTF-8 to represent those higher than the Unicode maximum
1025 of 0x10FFFF. That means that this macro is used to efficiently decide if the
1026 next few bytes in C<s> is legal UTF-8 for a single character.
1028 Use C<L</isSTRICT_UTF8_CHAR>> to restrict the acceptable code points to those
1029 defined by Unicode to be fully interchangeable across applications;
1030 C<L</isC9_STRICT_UTF8_CHAR>> to use the L<Unicode Corrigendum
1031 #9|http://www.unicode.org/versions/corrigendum9.html> definition of allowable
1032 code points; and C<L</isUTF8_CHAR_flags>> for a more customized definition.
1034 Use C<L</is_utf8_string>>, C<L</is_utf8_string_loc>>, and
1035 C<L</is_utf8_string_loclen>> to check entire strings.
1037 Note that it is deprecated to use code points higher than what will fit in an
1038 IV. This macro does not raise any warnings for such code points, treating them
1041 Note also that a UTF-8 INVARIANT character (i.e. ASCII on non-EBCDIC machines)
1042 is a valid UTF-8 character.
1047 #define isUTF8_CHAR(s, e) \
1048 (UNLIKELY((e) <= (s)) \
1050 : (UTF8_IS_INVARIANT(*s)) \
1052 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
1054 : LIKELY(NATIVE_UTF8_TO_I8(*s) <= _IS_UTF8_CHAR_HIGHEST_START_BYTE) \
1055 ? is_UTF8_CHAR_utf8_no_length_checks(s) \
1056 : _is_utf8_char_helper(s, e, 0))
1058 #define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end)
1059 #define bytes_from_utf8(s, lenp, is_utf8p) \
1060 bytes_from_utf8_loc(s, lenp, is_utf8p, 0)
1064 =for apidoc Am|STRLEN|isSTRICT_UTF8_CHAR|const U8 *s|const U8 *e
1066 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1067 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents some
1068 Unicode code point completely acceptable for open interchange between all
1069 applications; otherwise it evaluates to 0. If non-zero, the value gives how
1070 many bytes starting at C<s> comprise the code point's representation. Any
1071 bytes remaining before C<e>, but beyond the ones needed to form the first code
1072 point in C<s>, are not examined.
1074 The largest acceptable code point is the Unicode maximum 0x10FFFF, and must not
1075 be a surrogate nor a non-character code point. Thus this excludes any code
1076 point from Perl's extended UTF-8.
1078 This is used to efficiently decide if the next few bytes in C<s> is
1079 legal Unicode-acceptable UTF-8 for a single character.
1081 Use C<L</isC9_STRICT_UTF8_CHAR>> to use the L<Unicode Corrigendum
1082 #9|http://www.unicode.org/versions/corrigendum9.html> definition of allowable
1083 code points; C<L</isUTF8_CHAR>> to check for Perl's extended UTF-8;
1084 and C<L</isUTF8_CHAR_flags>> for a more customized definition.
1086 Use C<L</is_strict_utf8_string>>, C<L</is_strict_utf8_string_loc>>, and
1087 C<L</is_strict_utf8_string_loclen>> to check entire strings.
1092 #define isSTRICT_UTF8_CHAR(s, e) \
1093 (UNLIKELY((e) <= (s)) \
1095 : (UTF8_IS_INVARIANT(*s)) \
1097 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
1099 : is_STRICT_UTF8_CHAR_utf8_no_length_checks(s))
1103 =for apidoc Am|STRLEN|isC9_STRICT_UTF8_CHAR|const U8 *s|const U8 *e
1105 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1106 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents some
1107 Unicode non-surrogate code point; otherwise it evaluates to 0. If non-zero,
1108 the value gives how many bytes starting at C<s> comprise the code point's
1109 representation. Any bytes remaining before C<e>, but beyond the ones needed to
1110 form the first code point in C<s>, are not examined.
1112 The largest acceptable code point is the Unicode maximum 0x10FFFF. This
1113 differs from C<L</isSTRICT_UTF8_CHAR>> only in that it accepts non-character
1114 code points. This corresponds to
1115 L<Unicode Corrigendum #9|http://www.unicode.org/versions/corrigendum9.html>.
1116 which said that non-character code points are merely discouraged rather than
1117 completely forbidden in open interchange. See
1118 L<perlunicode/Noncharacter code points>.
1120 Use C<L</isUTF8_CHAR>> to check for Perl's extended UTF-8; and
1121 C<L</isUTF8_CHAR_flags>> for a more customized definition.
1123 Use C<L</is_c9strict_utf8_string>>, C<L</is_c9strict_utf8_string_loc>>, and
1124 C<L</is_c9strict_utf8_string_loclen>> to check entire strings.
1129 #define isC9_STRICT_UTF8_CHAR(s, e) \
1130 (UNLIKELY((e) <= (s)) \
1132 : (UTF8_IS_INVARIANT(*s)) \
1134 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
1136 : is_C9_STRICT_UTF8_CHAR_utf8_no_length_checks(s))
1140 =for apidoc Am|STRLEN|isUTF8_CHAR_flags|const U8 *s|const U8 *e| const U32 flags
1142 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1143 looking no further than S<C<e - 1>> are well-formed UTF-8, as extended by Perl,
1144 that represents some code point, subject to the restrictions given by C<flags>;
1145 otherwise it evaluates to 0. If non-zero, the value gives how many bytes
1146 starting at C<s> comprise the code point's representation. Any bytes remaining
1147 before C<e>, but beyond the ones needed to form the first code point in C<s>,
1150 If C<flags> is 0, this gives the same results as C<L</isUTF8_CHAR>>;
1151 if C<flags> is C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, this gives the same results
1152 as C<L</isSTRICT_UTF8_CHAR>>;
1153 and if C<flags> is C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE>, this gives
1154 the same results as C<L</isC9_STRICT_UTF8_CHAR>>.
1155 Otherwise C<flags> may be any combination of the C<UTF8_DISALLOW_I<foo>> flags
1156 understood by C<L</utf8n_to_uvchr>>, with the same meanings.
1158 The three alternative macros are for the most commonly needed validations; they
1159 are likely to run somewhat faster than this more general one, as they can be
1160 inlined into your code.
1162 Use L</is_utf8_string_flags>, L</is_utf8_string_loc_flags>, and
1163 L</is_utf8_string_loclen_flags> to check entire strings.
1168 #define isUTF8_CHAR_flags(s, e, flags) \
1169 (UNLIKELY((e) <= (s)) \
1171 : (UTF8_IS_INVARIANT(*s)) \
1173 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
1175 : _is_utf8_char_helper(s, e, flags))
1177 /* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is
1178 * retained solely for backwards compatibility */
1179 #define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n)
1181 #endif /* PERL_UTF8_H_ */
1184 * ex: set ts=8 sts=4 sw=4 et: