This is a live mirror of the Perl 5 development currently hosted at https://github.com/perl/perl5
Import perl5321delta.pod
[perl5.git] / utf8.h
CommitLineData
a0ed51b3
LW
1/* utf8.h
2 *
f3cb6f94
KW
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.
8 *
2eee27d7
SS
9 * Copyright (C) 2000, 2001, 2002, 2005, 2006, 2007, 2009,
10 * 2010, 2011 by Larry Wall and others
a0ed51b3
LW
11 *
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.
14 *
15 */
16
6a5bc5ac
KW
17#ifndef PERL_UTF8_H_ /* Guard against recursive inclusion */
18#define PERL_UTF8_H_ 1
57f0e7e2 19
39e02b42 20/* Use UTF-8 as the default script encoding?
1e54db1a 21 * Turning this on will break scripts having non-UTF-8 binary
39e02b42
JH
22 * data (such as Latin-1) in string literals. */
23#ifdef USE_UTF8_SCRIPTS
24# define USE_UTF8_IN_NAMES (!IN_BYTES)
25#else
26# define USE_UTF8_IN_NAMES (PL_hints & HINT_UTF8)
27#endif
28
3cd96634
KW
29#include "regcharclass.h"
30#include "unicode_constants.h"
31
051a06d4 32/* For to_utf8_fold_flags, q.v. */
e4f4ef45
KW
33#define FOLD_FLAGS_LOCALE 0x1
34#define FOLD_FLAGS_FULL 0x2
35#define FOLD_FLAGS_NOMIX_ASCII 0x4
051a06d4 36
7bbfa158 37/*
7bbfa158
KW
38=for apidoc is_ascii_string
39
8871a094 40This is a misleadingly-named synonym for L</is_utf8_invariant_string>.
7bbfa158
KW
41On ASCII-ish platforms, the name isn't misleading: the ASCII-range characters
42are exactly the UTF-8 invariants. But EBCDIC machines have more invariants
8871a094
KW
43than just the ASCII characters, so C<is_utf8_invariant_string> is preferred.
44
45=for apidoc is_invariant_string
46
47This is a somewhat misleadingly-named synonym for L</is_utf8_invariant_string>.
48C<is_utf8_invariant_string> is preferred, as it indicates under what conditions
49the string is invariant.
7bbfa158
KW
50
51=cut
52*/
8871a094
KW
53#define is_ascii_string(s, len) is_utf8_invariant_string(s, len)
54#define is_invariant_string(s, len) is_utf8_invariant_string(s, len)
7bbfa158 55
33f38593
KW
56#define uvoffuni_to_utf8_flags(d,uv,flags) \
57 uvoffuni_to_utf8_flags_msgs(d, uv, flags, 0)
de69f3af
KW
58#define uvchr_to_utf8(a,b) uvchr_to_utf8_flags(a,b,0)
59#define uvchr_to_utf8_flags(d,uv,flags) \
33f38593
KW
60 uvchr_to_utf8_flags_msgs(d,uv,flags, 0)
61#define uvchr_to_utf8_flags_msgs(d,uv,flags,msgs) \
62 uvoffuni_to_utf8_flags_msgs(d,NATIVE_TO_UNI(uv),flags, msgs)
de69f3af 63#define utf8_to_uvchr_buf(s, e, lenp) \
9a9a6c98 64 utf8_to_uvchr_buf_helper((const U8 *) (s), (const U8 *) e, lenp)
f9380377
KW
65#define utf8n_to_uvchr(s, len, lenp, flags) \
66 utf8n_to_uvchr_error(s, len, lenp, flags, 0)
37657a5b
KW
67#define utf8n_to_uvchr_error(s, len, lenp, flags, errors) \
68 utf8n_to_uvchr_msgs(s, len, lenp, flags, errors, 0)
de69f3af 69
a0270393 70#define to_uni_fold(c, p, lenp) _to_uni_fold_flags(c, p, lenp, FOLD_FLAGS_FULL)
a239b1e2 71
eda9cac1 72#define foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
1604cfb0 73 foldEQ_utf8_flags(s1, pe1, l1, u1, s2, pe2, l2, u2, 0)
baa60164 74#define FOLDEQ_UTF8_NOMIX_ASCII (1 << 0)
cea315b6 75#define FOLDEQ_LOCALE (1 << 1)
18f762c3
KW
76#define FOLDEQ_S1_ALREADY_FOLDED (1 << 2)
77#define FOLDEQ_S2_ALREADY_FOLDED (1 << 3)
d635b710
KW
78#define FOLDEQ_S1_FOLDS_SANE (1 << 4)
79#define FOLDEQ_S2_FOLDS_SANE (1 << 5)
a33c29bc 80
1d72bdf6
NIS
81#ifdef EBCDIC
82/* The equivalent of these macros but implementing UTF-EBCDIC
83 are in the following header file:
84 */
85
86#include "utfebcdic.h"
fd7cb289 87
d06134e5 88#else /* ! EBCDIC */
73c4f7a1
GS
89START_EXTERN_C
90
f4d83e55
KW
91/*
92
93=for apidoc AmnU|STRLEN|UTF8_MAXBYTES
94
95The maximum width of a single UTF-8 encoded character, in bytes.
96
97NOTE: Strictly speaking Perl's UTF-8 should not be called UTF-8 since UTF-8
98is an encoding of Unicode, and Unicode's upper limit, 0x10FFFF, can be
99expressed with 4 bytes. However, Perl thinks of UTF-8 as a way to encode
100non-negative integers in a binary format, even those above Unicode.
101
102=cut
103 */
111e8ed9
KW
104#define UTF8_MAXBYTES 13
105
a0ed51b3 106#ifdef DOINIT
6f06b55f 107EXTCONST unsigned char PL_utf8skip[] = {
b2635aa8
KW
108/* 0x00 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
109/* 0x10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
110/* 0x20 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
111/* 0x30 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
112/* 0x40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
113/* 0x50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
114/* 0x60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
115/* 0x70 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
116/* 0x80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
117/* 0x90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
118/* 0xA0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
119/* 0xB0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
120/* 0xC0 */ 2,2, /* overlong */
1ff3baa2 121/* 0xC2 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0080 to U+03FF */
b2635aa8
KW
122/* 0xD0 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0400 to U+07FF */
123/* 0xE0 */ 3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, /* U+0800 to U+FFFF */
124/* 0xF0 */ 4,4,4,4,4,4,4,4,5,5,5,5,6,6, /* above BMP to 2**31 - 1 */
6937f885
KW
125 /* Perl extended (never was official UTF-8). Up to 36 bit */
126/* 0xFE */ 7,
127 /* More extended, Up to 72 bits (64-bit + reserved) */
111e8ed9 128/* 0xFF */ UTF8_MAXBYTES
a0ed51b3
LW
129};
130#else
6f06b55f 131EXTCONST unsigned char PL_utf8skip[];
a0ed51b3
LW
132#endif
133
73c4f7a1 134END_EXTERN_C
7e2040f0 135
ef2a4c8f
KW
136/*
137
138=for apidoc Am|U8|NATIVE_TO_LATIN1|U8 ch
139
140Returns the Latin-1 (including ASCII and control characters) equivalent of the
141input native code point given by C<ch>. Thus, C<NATIVE_TO_LATIN1(193)> on
142EBCDIC platforms returns 65. These each represent the character C<"A"> on
143their respective platforms. On ASCII platforms no conversion is needed, so
144this macro expands to just its input, adding no time nor space requirements to
145the implementation.
146
147For conversion of code points potentially larger than will fit in a character,
148use L</NATIVE_TO_UNI>.
149
150=for apidoc Am|U8|LATIN1_TO_NATIVE|U8 ch
151
152Returns the native equivalent of the input Latin-1 code point (including ASCII
153and control characters) given by C<ch>. Thus, C<LATIN1_TO_NATIVE(66)> on
154EBCDIC platforms returns 194. These each represent the character C<"B"> on
155their respective platforms. On ASCII platforms no conversion is needed, so
156this macro expands to just its input, adding no time nor space requirements to
157the implementation.
158
159For conversion of code points potentially larger than will fit in a character,
160use L</UNI_TO_NATIVE>.
161
162=for apidoc Am|UV|NATIVE_TO_UNI|UV ch
163
164Returns the Unicode equivalent of the input native code point given by C<ch>.
165Thus, C<NATIVE_TO_UNI(195)> on EBCDIC platforms returns 67. These each
166represent the character C<"C"> on their respective platforms. On ASCII
167platforms no conversion is needed, so this macro expands to just its input,
168adding no time nor space requirements to the implementation.
169
170=for apidoc Am|UV|UNI_TO_NATIVE|UV ch
171
172Returns the native equivalent of the input Unicode code point given by C<ch>.
173Thus, C<UNI_TO_NATIVE(68)> on EBCDIC platforms returns 196. These each
174represent the character C<"D"> on their respective platforms. On ASCII
175platforms no conversion is needed, so this macro expands to just its input,
176adding no time nor space requirements to the implementation.
177
178=cut
179*/
180
e7ae132e
KW
181#define NATIVE_TO_LATIN1(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) ((ch) | 0)))
182#define LATIN1_TO_NATIVE(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) ((ch) | 0)))
59a449d5
KW
183
184/* I8 is an intermediate version of UTF-8 used only in UTF-EBCDIC. We thus
185 * consider it to be identical to UTF-8 on ASCII platforms. Strictly speaking
186 * UTF-8 and UTF-EBCDIC are two different things, but we often conflate them
187 * because they are 8-bit encodings that serve the same purpose in Perl, and
188 * rarely do we need to distinguish them. The term "NATIVE_UTF8" applies to
189 * whichever one is applicable on the current platform */
cb15eeb2
KW
190#define NATIVE_UTF8_TO_I8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) ((ch) | 0)))
191#define I8_TO_NATIVE_UTF8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) ((ch) | 0)))
59a449d5 192
cb15eeb2
KW
193#define UNI_TO_NATIVE(ch) ((UV) ((ch) | 0))
194#define NATIVE_TO_UNI(ch) ((UV) ((ch) | 0))
d7578b48 195
877d9f0d 196/*
9041c2e3 197
a14e0a36
KW
198 The following table is from Unicode 3.2, plus the Perl extensions for above
199 U+10FFFF
877d9f0d 200
a14e0a36 201 Code Points 1st Byte 2nd Byte 3rd 4th 5th 6th 7th 8th-13th
877d9f0d 202
375122d7 203 U+0000..U+007F 00..7F
e1b711da 204 U+0080..U+07FF * C2..DF 80..BF
a14e0a36
KW
205 U+0800..U+0FFF E0 * A0..BF 80..BF
206 U+1000..U+CFFF E1..EC 80..BF 80..BF
207 U+D000..U+D7FF ED 80..9F 80..BF
208 U+D800..U+DFFF ED A0..BF 80..BF (surrogates)
209 U+E000..U+FFFF EE..EF 80..BF 80..BF
210 U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF
211 U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF
212 U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
213 Below are above-Unicode code points
214 U+110000..U+13FFFF F4 90..BF 80..BF 80..BF
215 U+110000..U+1FFFFF F5..F7 80..BF 80..BF 80..BF
216 U+200000..U+FFFFFF F8 * 88..BF 80..BF 80..BF 80..BF
217U+1000000..U+3FFFFFF F9..FB 80..BF 80..BF 80..BF 80..BF
218U+4000000..U+3FFFFFFF FC * 84..BF 80..BF 80..BF 80..BF 80..BF
219U+40000000..U+7FFFFFFF FD 80..BF 80..BF 80..BF 80..BF 80..BF
220U+80000000..U+FFFFFFFFF FE * 82..BF 80..BF 80..BF 80..BF 80..BF 80..BF
221U+1000000000.. FF 80..BF 80..BF 80..BF 80..BF 80..BF * 81..BF 80..BF
877d9f0d 222
e1b711da 223Note the gaps before several of the byte entries above marked by '*'. These are
37e2e78e
KW
224caused by legal UTF-8 avoiding non-shortest encodings: it is technically
225possible to UTF-8-encode a single code point in different ways, but that is
226explicitly forbidden, and the shortest possible encoding should always be used
15824458 227(and that is what Perl does). The non-shortest ones are called 'overlongs'.
8c007b5a 228
877d9f0d
JH
229 */
230
8c007b5a
JH
231/*
232 Another way to look at it, as bits:
233
b2635aa8 234 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
8c007b5a 235
b2635aa8
KW
236 0aaa aaaa 0aaa aaaa
237 0000 0bbb bbaa aaaa 110b bbbb 10aa aaaa
238 cccc bbbb bbaa aaaa 1110 cccc 10bb bbbb 10aa aaaa
239 00 000d ddcc cccc bbbb bbaa aaaa 1111 0ddd 10cc cccc 10bb bbbb 10aa aaaa
8c007b5a
JH
240
241As you can see, the continuation bytes all begin with C<10>, and the
e1b711da 242leading bits of the start byte tell how many bytes there are in the
8c007b5a
JH
243encoded character.
244
df863e43
KW
245Perl's extended UTF-8 means we can have start bytes up through FF, though any
246beginning with FF yields a code point that is too large for 32-bit ASCII
247platforms. FF signals to use 13 bytes for the encoded character. This breaks
248the paradigm that the number of leading bits gives how many total bytes there
ab2e28c2 249are in the character. */
38953e5a 250
15824458
KW
251/* This is the number of low-order bits a continuation byte in a UTF-8 encoded
252 * sequence contributes to the specification of the code point. In the bit
253 * maps above, you see that the first 2 bits are a constant '10', leaving 6 of
254 * real information */
1d72bdf6 255#define UTF_ACCUMULATION_SHIFT 6
b2635aa8 256
fed423a5
KW
257/* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL()
258 * for more */
259#define QUESTION_MARK_CTRL DEL_NATIVE
260
261/* Surrogates, non-character code points and above-Unicode code points are
262 * problematic in some contexts. This allows code that needs to check for
a3815e44 263 * those to quickly exclude the vast majority of code points it will
fed423a5 264 * encounter */
a6951642
KW
265#define isUTF8_POSSIBLY_PROBLEMATIC(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
266 (U8) c >= 0xED)
fed423a5 267
57ff5f59
KW
268#define UNICODE_IS_PERL_EXTENDED(uv) UNLIKELY((UV) (uv) > 0x7FFFFFFF)
269
fed423a5
KW
270#endif /* EBCDIC vs ASCII */
271
d7bcd45a
KW
272/* 2**UTF_ACCUMULATION_SHIFT - 1. This masks out all but the bits that carry
273 * real information in a continuation byte. This turns out to be 0x3F in
274 * UTF-8, 0x1F in UTF-EBCDIC. */
d223e1ea 275#define UTF_CONTINUATION_MASK ((U8) (nBIT_MASK(UTF_ACCUMULATION_SHIFT)))
fed423a5 276
9f3cfb7a
KW
277/* For use in UTF8_IS_CONTINUATION(). This turns out to be 0xC0 in UTF-8,
278 * E0 in UTF-EBCDIC */
279#define UTF_IS_CONTINUATION_MASK ((U8) (0xFF << UTF_ACCUMULATION_SHIFT))
280
38f458ff
KW
281/* This defines the bits that are to be in the continuation bytes of a
282 * multi-byte UTF-8 encoded character that mark it is a continuation byte.
283 * This turns out to be 0x80 in UTF-8, 0xA0 in UTF-EBCDIC. (khw doesn't know
284 * the underlying reason that B0 works here) */
285#define UTF_CONTINUATION_MARK (UTF_IS_CONTINUATION_MASK & 0xB0)
286
f4225fa0
KW
287/* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the
288 * first byte thereof? */
289#define UTF8_IS_CONTINUATION(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
290 (((NATIVE_UTF8_TO_I8(c) & UTF_IS_CONTINUATION_MASK) \
291 == UTF_CONTINUATION_MARK)))
292
2dc97505
KW
293/* Is the representation of the Unicode code point 'cp' the same regardless of
294 * being encoded in UTF-8 or not? This is a fundamental property of
295 * UTF-8,EBCDIC */
296#define OFFUNI_IS_INVARIANT(c) (((WIDEST_UTYPE)(c)) < UTF_CONTINUATION_MARK)
297
ab2e28c2
KW
298/*
299=for apidoc Am|bool|UVCHR_IS_INVARIANT|UV cp
300
301Evaluates to 1 if the representation of code point C<cp> is the same whether or
302not it is encoded in UTF-8; otherwise evaluates to 0. UTF-8 invariant
303characters can be copied as-is when converting to/from UTF-8, saving time.
304C<cp> is Unicode if above 255; otherwise is platform-native.
305
306=cut
307 */
308#define UVCHR_IS_INVARIANT(cp) (OFFUNI_IS_INVARIANT(NATIVE_TO_UNI(cp)))
309
7028aeba
KW
310/* Internal macro to be used only in this file to aid in constructing other
311 * publicly accessible macros.
312 * The number of bytes required to express this uv in UTF-8, for just those
313 * uv's requiring 2 through 6 bytes, as these are common to all platforms and
314 * word sizes. The number of bytes needed is given by the number of leading 1
315 * bits in the start byte. There are 32 start bytes that have 2 initial 1 bits
316 * (C0-DF); there are 16 that have 3 initial 1 bits (E0-EF); 8 that have 4
317 * initial 1 bits (F0-F8); 4 that have 5 initial 1 bits (F9-FB), and 2 that
318 * have 6 initial 1 bits (FC-FD). The largest number a string of n bytes can
319 * represent is (the number of possible start bytes for 'n')
320 * * (the number of possiblities for each start byte
321 * The latter in turn is
322 * 2 ** ( (how many continuation bytes there are)
323 * * (the number of bits of information each
324 * continuation byte holds))
325 *
326 * If we were on a platform where we could use a fast find first set bit
327 * instruction (or count leading zeros instruction) this could be replaced by
328 * using that to find the log2 of the uv, and divide that by the number of bits
329 * of information in each continuation byte, adjusting for large cases and how
330 * much information is in a start byte for that length */
72164d3a 331#define __COMMON_UNI_SKIP(uv) \
7028aeba
KW
332 (UV) (uv) < (32 * (1U << ( UTF_ACCUMULATION_SHIFT))) ? 2 : \
333 (UV) (uv) < (16 * (1U << (2 * UTF_ACCUMULATION_SHIFT))) ? 3 : \
334 (UV) (uv) < ( 8 * (1U << (3 * UTF_ACCUMULATION_SHIFT))) ? 4 : \
335 (UV) (uv) < ( 4 * (1U << (4 * UTF_ACCUMULATION_SHIFT))) ? 5 : \
336 (UV) (uv) < ( 2 * (1U << (5 * UTF_ACCUMULATION_SHIFT))) ? 6 :
72164d3a
KW
337
338/* Internal macro to be used only in this file.
339 * This adds to __COMMON_UNI_SKIP the details at this platform's upper range.
fed423a5 340 * For any-sized EBCDIC platforms, or 64-bit ASCII ones, we need one more test
72164d3a
KW
341 * to see if just 7 bytes is needed, or if the maximum is needed. For 32-bit
342 * ASCII platforms, everything is representable by 7 bytes */
fed423a5 343#if defined(UV_IS_QUAD) || defined(EBCDIC)
72164d3a 344# define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) \
8974941d
KW
345 LIKELY((UV) (uv) < ((UV) 1U << (6 * UTF_ACCUMULATION_SHIFT))) \
346 ? 7 \
347 : UTF8_MAXBYTES)
1d68d6cd 348#else
72164d3a 349# define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) 7)
1d68d6cd
SC
350#endif
351
fed423a5
KW
352/* The next two macros use the base macro defined above, and add in the tests
353 * at the low-end of the range, for just 1 byte, yielding complete macros,
354 * publicly accessible. */
355
356/* Input is a true Unicode (not-native) code point */
357#define OFFUNISKIP(uv) (OFFUNI_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
2084b489 358
5352a763
KW
359/*
360
361=for apidoc Am|STRLEN|UVCHR_SKIP|UV cp
362returns the number of bytes required to represent the code point C<cp> when
363encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than
364255; a Unicode code point otherwise.
365
366=cut
367 */
fdb6583d 368#define UVCHR_SKIP(uv) ( UVCHR_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
5352a763 369
4bab39bc
KW
370#define UTF_MIN_START_BYTE \
371 ((UTF_CONTINUATION_MARK >> UTF_ACCUMULATION_SHIFT) | UTF_START_MARK(2))
372
373/* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence?
59645eb1 374 * This excludes invariants (they are single-byte). It also excludes the
4bab39bc 375 * illegal overlong sequences that begin with C0 and C1 on ASCII platforms, and
59645eb1
KW
376 * C0-C4 I8 start bytes on EBCDIC ones. On EBCDIC E0 can't start a
377 * non-overlong sequence, so we define a base macro and for those platforms,
378 * extend it to also exclude E0 */
379#define UTF8_IS_START_base(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
4bab39bc 380 (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_START_BYTE))
59645eb1
KW
381#ifdef EBCDIC
382# define UTF8_IS_START(c) \
383 (UTF8_IS_START_base(c) && (c) != I8_TO_NATIVE_UTF8(0xE0))
384#else
385# define UTF8_IS_START(c) UTF8_IS_START_base(c)
386#endif
4bab39bc 387
1df63428
KW
388#define UTF_MIN_ABOVE_LATIN1_BYTE \
389 ((0x100 >> UTF_ACCUMULATION_SHIFT) | UTF_START_MARK(2))
390
391/* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that
392 * represent a code point > 255? */
393#define UTF8_IS_ABOVE_LATIN1(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
394 (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_ABOVE_LATIN1_BYTE))
395
7c88d61e
KW
396/* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use
397 * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to
398 * be well-formed. */
399#define UTF8_IS_DOWNGRADEABLE_START(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
400 inRANGE(NATIVE_UTF8_TO_I8(c), \
401 UTF_MIN_START_BYTE, UTF_MIN_ABOVE_LATIN1_BYTE - 1))
402
b651802e
KW
403/* The largest code point representable by two UTF-8 bytes on this platform.
404 * As explained in the comments for __COMMON_UNI_SKIP, 32 start bytes with
fed423a5 405 * UTF_ACCUMULATION_SHIFT bits of information each */
aa206fb7
KW
406#define MAX_UTF8_TWO_BYTE (32 * (1U << UTF_ACCUMULATION_SHIFT) - 1)
407
b651802e
KW
408/* The largest code point representable by two UTF-8 bytes on any platform that
409 * Perl runs on. This value is constrained by EBCDIC which has 5 bits per
410 * continuation byte */
924b0bfd 411#define MAX_PORTABLE_UTF8_TWO_BYTE (32 * nBIT_UMAX(5))
aa206fb7 412
f2c50040
KW
413/*
414
415=for apidoc AmnU|STRLEN|UTF8_MAXBYTES_CASE
416
417The maximum number of UTF-8 bytes a single Unicode character can
418uppercase/lowercase/titlecase/fold into.
419
420=cut
421
422 * Unicode guarantees that the maximum expansion is UTF8_MAX_FOLD_CHAR_EXPAND
423 * characters, but any above-Unicode code point will fold to itself, so we only
424 * have to look at the expansion of the maximum Unicode code point. But this
425 * number may be less than the space occupied by a very large code point under
426 * Perl's extended UTF-8. We have to make it large enough to fit any single
427 * character. (It turns out that ASCII and EBCDIC differ in which is larger)
428 *
429=cut
430*/
c03c0950 431#define UTF8_MAXBYTES_CASE \
ae9a9edb 432 MAX(UTF8_MAXBYTES, UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF))
c03c0950 433
d06134e5
KW
434/* Rest of these are attributes of Unicode and perl's internals rather than the
435 * encoding, or happen to be the same in both ASCII and EBCDIC (at least at
436 * this level; the macros that some of these call may have different
437 * definitions in the two encodings */
438
59a449d5
KW
439/* In domain restricted to ASCII, these may make more sense to the reader than
440 * the ones with Latin1 in the name */
441#define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch)
442#define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch)
443
444/* More or less misleadingly-named defines, retained for back compat */
445#define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch)
446#define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch)
447#define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
448#define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
449#define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch)
d06134e5 450
c0236afe 451/* This defines the 1-bits that are to be in the first byte of a multi-byte
97d0ceda
KW
452 * UTF-8 encoded character that mark it as a start byte and give the number of
453 * bytes that comprise the character. 'len' is the number of bytes in the
454 * multi-byte sequence. */
8974941d
KW
455#define UTF_START_MARK(len) (UNLIKELY((len) > 7) \
456 ? 0xFF \
457 : ((U8) (0xFE << (7-(len)))))
c0236afe
KW
458
459/* Masks out the initial one bits in a start byte, leaving the real data ones.
460 * Doesn't work on an invariant byte. 'len' is the number of bytes in the
461 * multi-byte sequence that comprises the character. */
8974941d 462#define UTF_START_MASK(len) (UNLIKELY((len) >= 7) ? 0x00 : (0x1F >> ((len)-2)))
c0236afe 463
537124e4
KW
464/* Adds a UTF8 continuation byte 'new' of information to a running total code
465 * point 'old' of all the continuation bytes so far. This is designed to be
155d2738
KW
466 * used in a loop to convert from UTF-8 to the code point represented. Note
467 * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is
468 * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC)
469 * code point in process of being generated */
a6951642
KW
470#define UTF8_ACCUMULATE(old, new) (__ASSERT_(FITS_IN_8_BITS(new)) \
471 ((old) << UTF_ACCUMULATION_SHIFT) \
009097b1 472 | ((NATIVE_UTF8_TO_I8(new)) \
155d2738 473 & UTF_CONTINUATION_MASK))
d06134e5 474
4ab10950 475/* This works in the face of malformed UTF-8. */
4e1ed312
KW
476#define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) \
477 ( UTF8_IS_DOWNGRADEABLE_START(*(s)) \
478 && ( (e) - (s) > 1) \
479 && UTF8_IS_CONTINUATION(*((s)+1)))
4ab10950 480
5aaebcb3 481/* Number of bytes a code point occupies in UTF-8. */
5352a763 482#define NATIVE_SKIP(uv) UVCHR_SKIP(uv)
bd18bd40 483
5aaebcb3
KW
484/* Most code which says UNISKIP is really thinking in terms of native code
485 * points (0-255) plus all those beyond. This is an imprecise term, but having
2accb712 486 * it means existing code continues to work. For precision, use UVCHR_SKIP,
5352a763
KW
487 * NATIVE_SKIP, or OFFUNISKIP */
488#define UNISKIP(uv) UVCHR_SKIP(uv)
5aaebcb3 489
3c0792e4
KW
490/* Longer, but more accurate name */
491#define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c)
492
a62b247b
KW
493/* Convert a UTF-8 variant Latin1 character to a native code point value.
494 * Needs just one iteration of accumulate. Should be used only if it is known
495 * that the code point is < 256, and is not UTF-8 invariant. Use the slower
496 * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point
497 * representable by two bytes (which turns out to be up through
498 * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are:
499 * HI: a downgradable start byte;
500 * LO: continuation.
501 * */
502#define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \
503 ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \
504 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
505 LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \
506 NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO))))
507
94bb8c36 508/* Convert a two (not one) byte utf8 character to a native code point value.
2950f2a7
KW
509 * Needs just one iteration of accumulate. Should not be used unless it is
510 * known that the two bytes are legal: 1) two-byte start, and 2) continuation.
511 * Note that the result can be larger than 255 if the input character is not
512 * downgradable */
94bb8c36 513#define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \
a6951642
KW
514 (__ASSERT_(FITS_IN_8_BITS(HI)) \
515 __ASSERT_(FITS_IN_8_BITS(LO)) \
516 __ASSERT_(PL_utf8skip[HI] == 2) \
517 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
94bb8c36 518 UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \
635e76f5 519 (LO))))
94bb8c36
KW
520
521/* Should never be used, and be deprecated */
522#define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO))
2950f2a7 523
bd18bd40
KW
524/*
525
526=for apidoc Am|STRLEN|UTF8SKIP|char* s
ee0ff0f5
KW
527returns the number of bytes a non-malformed UTF-8 encoded character whose first
528(perhaps only) byte is pointed to by C<s>.
529
530If there is a possibility of malformed input, use instead:
531
532=over
533
eb992c6f 534=item C<L</UTF8_SAFE_SKIP>> if you know the maximum ending pointer in the
ee0ff0f5
KW
535buffer pointed to by C<s>; or
536
eb992c6f 537=item C<L</UTF8_CHK_SKIP>> if you don't know it.
ee0ff0f5
KW
538
539=back
540
541It is better to restructure your code so the end pointer is passed down so that
542you know what it actually is at the point of this call, but if that isn't
eb992c6f 543possible, C<L</UTF8_CHK_SKIP>> can minimize the chance of accessing beyond the end
ee0ff0f5 544of the input buffer.
bd18bd40
KW
545
546=cut
547 */
2a70536e 548#define UTF8SKIP(s) PL_utf8skip[*(const U8*)(s)]
a281f16c
KW
549
550/*
551=for apidoc Am|STRLEN|UTF8_SKIP|char* s
eb992c6f 552This is a synonym for C<L</UTF8SKIP>>
a281f16c
KW
553
554=cut
555*/
556
2a70536e 557#define UTF8_SKIP(s) UTF8SKIP(s)
d06134e5 558
85fcc8f2 559/*
ee0ff0f5
KW
560=for apidoc Am|STRLEN|UTF8_CHK_SKIP|char* s
561
eb992c6f
KW
562This is a safer version of C<L</UTF8SKIP>>, but still not as safe as
563C<L</UTF8_SAFE_SKIP>>. This version doesn't blindly assume that the input
ee0ff0f5
KW
564string pointed to by C<s> is well-formed, but verifies that there isn't a NUL
565terminating character before the expected end of the next character in C<s>.
566The length C<UTF8_CHK_SKIP> returns stops just before any such NUL.
567
568Perl tends to add NULs, as an insurance policy, after the end of strings in
569SV's, so it is likely that using this macro will prevent inadvertent reading
570beyond the end of the input buffer, even if it is malformed UTF-8.
571
572This macro is intended to be used by XS modules where the inputs could be
573malformed, and it isn't feasible to restructure to use the safer
eb992c6f 574C<L</UTF8_SAFE_SKIP>>, for example when interfacing with a C library.
ee0ff0f5
KW
575
576=cut
577*/
578
579#define UTF8_CHK_SKIP(s) \
8974941d 580 (UNLIKELY(s[0] == '\0') ? 1 : MIN(UTF8SKIP(s), \
f87d8789 581 my_strnlen((char *) (s), UTF8SKIP(s))))
ee0ff0f5 582/*
85fcc8f2
KW
583
584=for apidoc Am|STRLEN|UTF8_SAFE_SKIP|char* s|char* e
45671da2
KW
585returns 0 if S<C<s E<gt>= e>>; otherwise returns the number of bytes in the
586UTF-8 encoded character whose first byte is pointed to by C<s>. But it never
587returns beyond C<e>. On DEBUGGING builds, it asserts that S<C<s E<lt>= e>>.
85fcc8f2
KW
588
589=cut
590 */
45671da2 591#define UTF8_SAFE_SKIP(s, e) (__ASSERT_((e) >= (s)) \
8974941d 592 UNLIKELY(((e) - (s)) <= 0) \
45671da2
KW
593 ? 0 \
594 : MIN(((e) - (s)), UTF8_SKIP(s)))
85fcc8f2 595
2d1545e5
KW
596/* Most code that says 'UNI_' really means the native value for code points up
597 * through 255 */
598#define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp)
599
c2b32798
KW
600/*
601=for apidoc Am|bool|UTF8_IS_INVARIANT|char c
602
603Evaluates to 1 if the byte C<c> represents the same character when encoded in
604UTF-8 as when not; otherwise evaluates to 0. UTF-8 invariant characters can be
605copied as-is when converting to/from UTF-8, saving time.
606
607In spite of the name, this macro gives the correct result if the input string
608from which C<c> comes is not encoded in UTF-8.
609
610See C<L</UVCHR_IS_INVARIANT>> for checking if a UV is invariant.
611
612=cut
613
614The reason it works on both UTF-8 encoded strings and non-UTF-8 encoded, is
615that it returns TRUE in each for the exact same set of bit patterns. It is
616valid on a subset of what UVCHR_IS_INVARIANT is valid on, so can just use that;
617and the compiler should optimize out anything extraneous given the
618implementation of the latter. The |0 makes sure this isn't mistakenly called
619with a ptr argument.
620*/
5c06326b 621#define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT((c) | 0)
5fc230f1
KW
622
623/* Like the above, but its name implies a non-UTF8 input, which as the comments
624 * above show, doesn't matter as to its implementation */
38953e5a 625#define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
d06134e5 626
2c03e801
KW
627/* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence
628 * in UTF-8? This is the inverse of UTF8_IS_INVARIANT. */
629#define UTF8_IS_CONTINUED(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
630 (! UTF8_IS_INVARIANT(c)))
631
48ccf5e1
KW
632/* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic
633 * bytes from an ordinal that is known to fit into exactly two (not one) bytes;
634 * it must be less than 0x3FF to work across both encodings. */
635
636/* These two are helper macros for the other three sets, and should not be used
637 * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1
1ff3baa2
KW
638 * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works
639 * for any code point */
48ccf5e1 640#define __BASE_TWO_BYTE_HI(c, translate_function) \
2863dafa 641 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
48ccf5e1 642 I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \
2863dafa 643 | UTF_START_MARK(2)))
48ccf5e1 644#define __BASE_TWO_BYTE_LO(c, translate_function) \
2863dafa 645 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
48ccf5e1 646 I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \
2863dafa 647 | UTF_CONTINUATION_MARK))
48ccf5e1 648
48ccf5e1
KW
649/* The next two macros should not be used. They were designed to be usable as
650 * the case label of a switch statement, but this doesn't work for EBCDIC. Use
9d0d3a03 651 * regen/unicode_constants.pl instead */
48ccf5e1
KW
652#define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)
653#define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)
654
655/* The next two macros are used when the source should be a single byte
656 * character; checked for under DEBUGGING */
657#define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
4c8cd605 658 ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1)))
48ccf5e1 659#define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
4c8cd605 660 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1)))
48ccf5e1
KW
661
662/* These final two macros in the series are used when the source can be any
663 * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient
664 * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~'
665 * operator instead of "<=" to avoid getting compiler warnings.
d52b8576 666 * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few
48ccf5e1
KW
667 * places, so the ~ works */
668#define UTF8_TWO_BYTE_HI(c) \
669 (__ASSERT_((sizeof(c) == 1) \
d52b8576 670 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
4c8cd605 671 (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)))
48ccf5e1
KW
672#define UTF8_TWO_BYTE_LO(c) \
673 (__ASSERT_((sizeof(c) == 1) \
d52b8576 674 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
4c8cd605 675 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)))
d06134e5 676
e7214ce8
KW
677/* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII
678 * as it is only in overlongs. */
679#define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1)
680
7e2040f0 681/*
e3036cf4 682 * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem
20df05f4
KW
683 * from an earlier version of these macros in which they didn't call the
684 * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the
685 * beginning of a utf8 character. Now that foo_utf8() determines that itself,
686 * no need to do it again here
7e2040f0 687 */
da8c1a98
KW
688#define isIDFIRST_lazy_if_safe(p, e, UTF) \
689 ((IN_BYTES || !UTF) \
690 ? isIDFIRST(*(p)) \
691 : isIDFIRST_utf8_safe(p, e))
da8c1a98
KW
692#define isWORDCHAR_lazy_if_safe(p, e, UTF) \
693 ((IN_BYTES || !UTF) \
694 ? isWORDCHAR(*(p)) \
695 : isWORDCHAR_utf8_safe((U8 *) p, (U8 *) e))
4c1d9526 696#define isALNUM_lazy_if_safe(p, e, UTF) isWORDCHAR_lazy_if_safe(p, e, UTF)
da8c1a98 697
89ebb4a3
JH
698#define UTF8_MAXLEN UTF8_MAXBYTES
699
8cb75cc8
KW
700/* A Unicode character can fold to up to 3 characters */
701#define UTF8_MAX_FOLD_CHAR_EXPAND 3
702
d3481830 703#define IN_BYTES UNLIKELY(CopHINTS_get(PL_curcop) & HINT_BYTES)
bd18bd40
KW
704
705/*
706
707=for apidoc Am|bool|DO_UTF8|SV* sv
708Returns a bool giving whether or not the PV in C<sv> is to be treated as being
709encoded in UTF-8.
710
711You should use this I<after> a call to C<SvPV()> or one of its variants, in
712case any call to string overloading updates the internal UTF-8 encoding flag.
713
714=cut
715*/
0064a8a9 716#define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES)
1ff3baa2
KW
717
718/* Should all strings be treated as Unicode, and not just UTF-8 encoded ones?
719 * Is so within 'feature unicode_strings' or 'locale :not_characters', and not
720 * within 'use bytes'. UTF-8 locales are not tested for here, but perhaps
721 * could be */
70844984 722#define IN_UNI_8_BIT \
1604cfb0 723 (( ( (CopHINTS_get(PL_curcop) & HINT_UNI_8_BIT)) \
70844984
KW
724 || ( CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \
725 /* -1 below is for :not_characters */ \
726 && _is_in_locale_category(FALSE, -1))) \
727 && (! IN_BYTES))
b36bf33f 728
1d72bdf6 729
c76687c5 730#define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */
2b5e7bc2 731#define UTF8_GOT_EMPTY UTF8_ALLOW_EMPTY
c76687c5
KW
732
733/* Allow first byte to be a continuation byte */
1d72bdf6 734#define UTF8_ALLOW_CONTINUATION 0x0002
2b5e7bc2 735#define UTF8_GOT_CONTINUATION UTF8_ALLOW_CONTINUATION
c76687c5 736
cd01d3b1 737/* Unexpected non-continuation byte */
1d72bdf6 738#define UTF8_ALLOW_NON_CONTINUATION 0x0004
2b5e7bc2 739#define UTF8_GOT_NON_CONTINUATION UTF8_ALLOW_NON_CONTINUATION
949cf498
KW
740
741/* expecting more bytes than were available in the string */
742#define UTF8_ALLOW_SHORT 0x0008
2b5e7bc2 743#define UTF8_GOT_SHORT UTF8_ALLOW_SHORT
949cf498 744
94953955
KW
745/* Overlong sequence; i.e., the code point can be specified in fewer bytes.
746 * First one will convert the overlong to the REPLACEMENT CHARACTER; second
747 * will return what the overlong evaluates to */
949cf498 748#define UTF8_ALLOW_LONG 0x0010
94953955 749#define UTF8_ALLOW_LONG_AND_ITS_VALUE (UTF8_ALLOW_LONG|0x0020)
2b5e7bc2
KW
750#define UTF8_GOT_LONG UTF8_ALLOW_LONG
751
d60baaa7
KW
752#define UTF8_ALLOW_OVERFLOW 0x0080
753#define UTF8_GOT_OVERFLOW UTF8_ALLOW_OVERFLOW
949cf498 754
f180b292 755#define UTF8_DISALLOW_SURROGATE 0x0100 /* Unicode surrogates */
2b5e7bc2 756#define UTF8_GOT_SURROGATE UTF8_DISALLOW_SURROGATE
f180b292 757#define UTF8_WARN_SURROGATE 0x0200
949cf498 758
c4e96019
KW
759/* Unicode non-character code points */
760#define UTF8_DISALLOW_NONCHAR 0x0400
2b5e7bc2 761#define UTF8_GOT_NONCHAR UTF8_DISALLOW_NONCHAR
c4e96019 762#define UTF8_WARN_NONCHAR 0x0800
949cf498 763
c4e96019
KW
764/* Super-set of Unicode: code points above the legal max */
765#define UTF8_DISALLOW_SUPER 0x1000
2b5e7bc2 766#define UTF8_GOT_SUPER UTF8_DISALLOW_SUPER
c4e96019
KW
767#define UTF8_WARN_SUPER 0x2000
768
769/* The original UTF-8 standard did not define UTF-8 with start bytes of 0xFE or
770 * 0xFF, though UTF-EBCDIC did. This allowed both versions to represent code
771 * points up to 2 ** 31 - 1. Perl extends UTF-8 so that 0xFE and 0xFF are
772 * usable on ASCII platforms, and 0xFF means something different than
773 * UTF-EBCDIC defines. These changes allow code points of 64 bits (actually
774 * somewhat more) to be represented on both platforms. But these are Perl
775 * extensions, and not likely to be interchangeable with other languages. Note
776 * that on ASCII platforms, FE overflows a signed 32-bit word, and FF an
777 * unsigned one. */
d044b7a7
KW
778#define UTF8_DISALLOW_PERL_EXTENDED 0x4000
779#define UTF8_GOT_PERL_EXTENDED UTF8_DISALLOW_PERL_EXTENDED
780#define UTF8_WARN_PERL_EXTENDED 0x8000
d35f2ca5 781
57ff5f59
KW
782/* For back compat, these old names are misleading for overlongs and
783 * UTF_EBCDIC. */
d044b7a7
KW
784#define UTF8_DISALLOW_ABOVE_31_BIT UTF8_DISALLOW_PERL_EXTENDED
785#define UTF8_GOT_ABOVE_31_BIT UTF8_GOT_PERL_EXTENDED
786#define UTF8_WARN_ABOVE_31_BIT UTF8_WARN_PERL_EXTENDED
787#define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_PERL_EXTENDED
788#define UTF8_WARN_FE_FF UTF8_WARN_PERL_EXTENDED
949cf498 789
f180b292 790#define UTF8_CHECK_ONLY 0x10000
99a765e9 791#define _UTF8_NO_CONFIDENCE_IN_CURLEN 0x20000 /* Internal core use only */
949cf498
KW
792
793/* For backwards source compatibility. They do nothing, as the default now
794 * includes what they used to mean. The first one's meaning was to allow the
795 * just the single non-character 0xFFFF */
796#define UTF8_ALLOW_FFFF 0
c825ef8c 797#define UTF8_ALLOW_FE_FF 0
949cf498
KW
798#define UTF8_ALLOW_SURROGATE 0
799
ecc1615f
KW
800/* C9 refers to Unicode Corrigendum #9: allows but discourages non-chars */
801#define UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE \
802 (UTF8_DISALLOW_SUPER|UTF8_DISALLOW_SURROGATE)
803#define UTF8_WARN_ILLEGAL_C9_INTERCHANGE (UTF8_WARN_SUPER|UTF8_WARN_SURROGATE)
804
d35f2ca5 805#define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \
ecc1615f 806 (UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE|UTF8_DISALLOW_NONCHAR)
949cf498 807#define UTF8_WARN_ILLEGAL_INTERCHANGE \
ecc1615f
KW
808 (UTF8_WARN_ILLEGAL_C9_INTERCHANGE|UTF8_WARN_NONCHAR)
809
0eb3d6a0
KW
810/* This is typically used for code that processes UTF-8 input and doesn't want
811 * to have to deal with any malformations that might be present. All such will
812 * be safely replaced by the REPLACEMENT CHARACTER, unless other flags
813 * overriding this are also present. */
2d532c27
KW
814#define UTF8_ALLOW_ANY ( UTF8_ALLOW_CONTINUATION \
815 |UTF8_ALLOW_NON_CONTINUATION \
816 |UTF8_ALLOW_SHORT \
d60baaa7
KW
817 |UTF8_ALLOW_LONG \
818 |UTF8_ALLOW_OVERFLOW)
2d532c27
KW
819
820/* Accept any Perl-extended UTF-8 that evaluates to any UV on the platform, but
cd01d3b1 821 * not any malformed. This is the default. */
2d532c27
KW
822#define UTF8_ALLOW_ANYUV 0
823#define UTF8_ALLOW_DEFAULT UTF8_ALLOW_ANYUV
1d72bdf6 824
89d986df
KW
825/*
826=for apidoc Am|bool|UTF8_IS_SURROGATE|const U8 *s|const U8 *e
827
828Evaluates to non-zero if the first few bytes of the string starting at C<s> and
829looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
830of the Unicode surrogate code points; otherwise it evaluates to 0. If
831non-zero, the value gives how many bytes starting at C<s> comprise the code
832point's representation.
833
834=cut
835 */
836#define UTF8_IS_SURROGATE(s, e) is_SURROGATE_utf8_safe(s, e)
837
838
839#define UTF8_IS_REPLACEMENT(s, send) is_REPLACEMENT_utf8_safe(s,send)
840
285aa1f0 841#define MAX_LEGAL_CP ((UV)IV_MAX)
40606899 842
89d986df
KW
843/*
844=for apidoc Am|bool|UTF8_IS_SUPER|const U8 *s|const U8 *e
845
846Recall that Perl recognizes an extension to UTF-8 that can encode code
847points larger than the ones defined by Unicode, which are 0..0x10FFFF.
848
849This macro evaluates to non-zero if the first few bytes of the string starting
850at C<s> and looking no further than S<C<e - 1>> are from this UTF-8 extension;
851otherwise it evaluates to 0. If non-zero, the value gives how many bytes
852starting at C<s> comprise the code point's representation.
853
8540 is returned if the bytes are not well-formed extended UTF-8, or if they
855represent a code point that cannot fit in a UV on the current platform. Hence
856this macro can give different results when run on a 64-bit word machine than on
857one with a 32-bit word size.
0c58a72b 858
891fd405 859Note that it is illegal to have code points that are larger than what can
89d986df 860fit in an IV on the current machine.
7131f24d 861
89d986df
KW
862=cut
863
864 * ASCII EBCDIC I8
7131f24d
KW
865 * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode
866 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0
867 * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1
89d986df
KW
868 */
869#ifdef EBCDIC
a14e0a36 870# define UTF8_IS_SUPER(s, e) \
8974941d
KW
871 (( ((e) > (s) + 4) \
872 && (NATIVE_UTF8_TO_I8(*(s)) >= 0xF9) \
873 && UNLIKELY( NATIVE_UTF8_TO_I8(*(s)) > 0xF9 \
874 || (NATIVE_UTF8_TO_I8(*((s) + 1)) >= 0xA2)) \
875 && LIKELY((s) + UTF8SKIP(s) <= (e))) \
0cf25474 876 ? is_utf8_char_helper(s, s + UTF8SKIP(s), 0) : 0)
7131f24d 877#else
a14e0a36 878# define UTF8_IS_SUPER(s, e) \
8974941d 879 (( ((e) > (s) + 3) \
89d986df 880 && (*(U8*) (s)) >= 0xF4 \
8974941d
KW
881 && (UNLIKELY( ((*(U8*) (s)) > 0xF4) \
882 || (*((U8*) (s) + 1) >= 0x90))) \
89d986df 883 && LIKELY((s) + UTF8SKIP(s) <= (e))) \
1376b35c 884 ? is_utf8_char_helper(s, s + UTF8SKIP(s), 0) : 0)
7131f24d
KW
885#endif
886
b96a92fb
KW
887/* These are now machine generated, and the 'given' clause is no longer
888 * applicable */
0c58a72b 889#define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \
89d986df
KW
890 cBOOL(is_NONCHAR_utf8_safe(s,e))
891
892/*
893=for apidoc Am|bool|UTF8_IS_NONCHAR|const U8 *s|const U8 *e
894
895Evaluates to non-zero if the first few bytes of the string starting at C<s> and
896looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
897of the Unicode non-character code points; otherwise it evaluates to 0. If
898non-zero, the value gives how many bytes starting at C<s> comprise the code
899point's representation.
900
d296fe14
KW
901=for apidoc AmnU|UV|UNICODE_REPLACEMENT
902
903Evaluates to 0xFFFD, the code point of the Unicode REPLACEMENT CHARACTER
904
89d986df
KW
905=cut
906 */
0c58a72b
KW
907#define UTF8_IS_NONCHAR(s, e) \
908 UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e)
7131f24d 909
c867b360
JH
910#define UNICODE_SURROGATE_FIRST 0xD800
911#define UNICODE_SURROGATE_LAST 0xDFFF
912#define UNICODE_REPLACEMENT 0xFFFD
913#define UNICODE_BYTE_ORDER_MARK 0xFEFF
1d72bdf6 914
b851fbc1 915/* Though our UTF-8 encoding can go beyond this,
c76687c5 916 * let's be conservative and do as Unicode says. */
b851fbc1
JH
917#define PERL_UNICODE_MAX 0x10FFFF
918
d044b7a7
KW
919#define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */
920#define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */
921#define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */
922#define UNICODE_WARN_PERL_EXTENDED 0x0008 /* Above 0x7FFF_FFFF */
923#define UNICODE_WARN_ABOVE_31_BIT UNICODE_WARN_PERL_EXTENDED
924#define UNICODE_DISALLOW_SURROGATE 0x0010
925#define UNICODE_DISALLOW_NONCHAR 0x0020
926#define UNICODE_DISALLOW_SUPER 0x0040
927#define UNICODE_DISALLOW_PERL_EXTENDED 0x0080
24b4c303
KW
928
929#ifdef PERL_CORE
930# define UNICODE_ALLOW_ABOVE_IV_MAX 0x0100
931#endif
d044b7a7 932#define UNICODE_DISALLOW_ABOVE_31_BIT UNICODE_DISALLOW_PERL_EXTENDED
33f38593
KW
933
934#define UNICODE_GOT_SURROGATE UNICODE_DISALLOW_SURROGATE
935#define UNICODE_GOT_NONCHAR UNICODE_DISALLOW_NONCHAR
936#define UNICODE_GOT_SUPER UNICODE_DISALLOW_SUPER
937#define UNICODE_GOT_PERL_EXTENDED UNICODE_DISALLOW_PERL_EXTENDED
938
ecc1615f
KW
939#define UNICODE_WARN_ILLEGAL_C9_INTERCHANGE \
940 (UNICODE_WARN_SURROGATE|UNICODE_WARN_SUPER)
bb88be5f 941#define UNICODE_WARN_ILLEGAL_INTERCHANGE \
ecc1615f
KW
942 (UNICODE_WARN_ILLEGAL_C9_INTERCHANGE|UNICODE_WARN_NONCHAR)
943#define UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE \
944 (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_SUPER)
bb88be5f 945#define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \
ecc1615f 946 (UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE|UNICODE_DISALLOW_NONCHAR)
949cf498
KW
947
948/* For backward source compatibility, as are now the default */
949#define UNICODE_ALLOW_SURROGATE 0
950#define UNICODE_ALLOW_SUPER 0
951#define UNICODE_ALLOW_ANY 0
b851fbc1 952
2d6b3d38
KW
953/* This matches the 2048 code points between UNICODE_SURROGATE_FIRST (0xD800) and
954 * UNICODE_SURROGATE_LAST (0xDFFF) */
8974941d 955#define UNICODE_IS_SURROGATE(uv) UNLIKELY(((UV) (uv) & (~0xFFFF | 0xF800)) \
2d6b3d38
KW
956 == 0xD800)
957
8974941d
KW
958#define UNICODE_IS_REPLACEMENT(uv) UNLIKELY((UV) (uv) == UNICODE_REPLACEMENT)
959#define UNICODE_IS_BYTE_ORDER_MARK(uv) UNLIKELY((UV) (uv) \
960 == UNICODE_BYTE_ORDER_MARK)
c149ab20
KW
961
962/* Is 'uv' one of the 32 contiguous-range noncharacters? */
8974941d
KW
963#define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) UNLIKELY((UV) (uv) >= 0xFDD0 \
964 && (UV) (uv) <= 0xFDEF)
c149ab20
KW
965
966/* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE,
967 * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above
968 * the Unicode legal max */
969#define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \
8974941d 970 UNLIKELY(((UV) (uv) & 0xFFFE) == 0xFFFE)
c149ab20
KW
971
972#define UNICODE_IS_NONCHAR(uv) \
973 ( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \
974 || ( LIKELY( ! UNICODE_IS_SUPER(uv)) \
975 && UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
976
8974941d 977#define UNICODE_IS_SUPER(uv) UNLIKELY((UV) (uv) > PERL_UNICODE_MAX)
1d72bdf6 978
ec34087a
KW
979#define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE
980#define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \
981 LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE
982#define MICRO_SIGN MICRO_SIGN_NATIVE
983#define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \
984 LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE
985#define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \
986 LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE
09091399
JH
987#define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3
988#define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2
989#define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3
9dcbe121 990#define GREEK_SMALL_LETTER_MU 0x03BC
9e682c18
KW
991#define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case
992 of MICRON */
993#define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */
0766489e
KW
994#ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
995# define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E
996#endif
74894415
KW
997#define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130
998#define LATIN_SMALL_LETTER_DOTLESS_I 0x131
9e682c18 999#define LATIN_SMALL_LETTER_LONG_S 0x017F
a9f50d33
KW
1000#define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05
1001#define LATIN_SMALL_LIGATURE_ST 0xFB06
9e682c18
KW
1002#define KELVIN_SIGN 0x212A
1003#define ANGSTROM_SIGN 0x212B
09091399 1004
9e55ce06 1005#define UNI_DISPLAY_ISPRINT 0x0001
c728cb41 1006#define UNI_DISPLAY_BACKSLASH 0x0002
daf6caf1
KW
1007#define UNI_DISPLAY_BACKSPACE 0x0004 /* Allow \b when also
1008 UNI_DISPLAY_BACKSLASH */
1009#define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT \
1010 |UNI_DISPLAY_BACKSLASH \
1011 |UNI_DISPLAY_BACKSPACE)
1012
1013/* Character classes could also allow \b, but not patterns in general */
c728cb41 1014#define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
9e55ce06 1015
e0ffa6d6 1016/* Should be removed; maybe deprecated, but not used in CPAN */
ebc501f0 1017#define SHARP_S_SKIP 2
3b0fc154 1018
3cedd9d9 1019#define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end)
976c1b08
KW
1020#define bytes_from_utf8(s, lenp, is_utf8p) \
1021 bytes_from_utf8_loc(s, lenp, is_utf8p, 0)
3cedd9d9 1022
e23e8bc1
KW
1023/*
1024
25e3a4e0
KW
1025=for apidoc Am|STRLEN|isUTF8_CHAR_flags|const U8 *s|const U8 *e| const U32 flags
1026
1027Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1028looking no further than S<C<e - 1>> are well-formed UTF-8, as extended by Perl,
1029that represents some code point, subject to the restrictions given by C<flags>;
1030otherwise it evaluates to 0. If non-zero, the value gives how many bytes
2717076a
KW
1031starting at C<s> comprise the code point's representation. Any bytes remaining
1032before C<e>, but beyond the ones needed to form the first code point in C<s>,
1033are not examined.
25e3a4e0
KW
1034
1035If C<flags> is 0, this gives the same results as C<L</isUTF8_CHAR>>;
1036if C<flags> is C<UTF8_DISALLOW_ILLEGAL_INTERCHANGE>, this gives the same results
1037as C<L</isSTRICT_UTF8_CHAR>>;
1038and if C<flags> is C<UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE>, this gives
1039the same results as C<L</isC9_STRICT_UTF8_CHAR>>.
1040Otherwise C<flags> may be any combination of the C<UTF8_DISALLOW_I<foo>> flags
1041understood by C<L</utf8n_to_uvchr>>, with the same meanings.
1042
1043The three alternative macros are for the most commonly needed validations; they
1044are likely to run somewhat faster than this more general one, as they can be
1045inlined into your code.
1046
9f2abfde
KW
1047Use L</is_utf8_string_flags>, L</is_utf8_string_loc_flags>, and
1048L</is_utf8_string_loclen_flags> to check entire strings.
1049
25e3a4e0
KW
1050=cut
1051*/
1052
1053#define isUTF8_CHAR_flags(s, e, flags) \
1054 (UNLIKELY((e) <= (s)) \
1055 ? 0 \
1056 : (UTF8_IS_INVARIANT(*s)) \
1057 ? 1 \
1058 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
1059 ? 0 \
1376b35c 1060 : is_utf8_char_helper(s, e, flags))
25e3a4e0 1061
6302f837
KW
1062/* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is
1063 * retained solely for backwards compatibility */
1064#define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n)
e9a8c099 1065
6a5bc5ac 1066#endif /* PERL_UTF8_H_ */
57f0e7e2 1067
e9a8c099 1068/*
14d04a33 1069 * ex: set ts=8 sts=4 sw=4 et:
e9a8c099 1070 */