Commit | Line | Data |
---|---|---|
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 | * | |
42b360b2 KW |
15 | * A note on nomenclature: The term UTF-8 is used loosely and inconsistently |
16 | * in Perl documentation. For one, perl uses an extension of UTF-8 to | |
17 | * represent code points that Unicode considers illegal. For another, ASCII | |
18 | * platform UTF-8 is usually conflated with EBCDIC platform UTF-EBCDIC, because | |
19 | * outside some of the macros in this this file, the differences are hopefully | |
20 | * invisible at the semantic level. | |
21 | * | |
22 | * UTF-EBCDIC has an isomorphic translation named I8 (for "Intermediate eight") | |
23 | * which differs from UTF-8 only in a few details. It is often useful to | |
24 | * translate UTF-EBCDIC into this form for processing. In general, macros and | |
25 | * functions that are expecting their inputs to be either in I8 or UTF-8 are | |
26 | * named UTF_foo (without an '8'), to indicate this. | |
27 | * | |
28 | * Unfortunately there are inconsistencies. | |
29 | * | |
a0ed51b3 LW |
30 | */ |
31 | ||
6a5bc5ac KW |
32 | #ifndef PERL_UTF8_H_ /* Guard against recursive inclusion */ |
33 | #define PERL_UTF8_H_ 1 | |
57f0e7e2 | 34 | |
1cdf84d2 KW |
35 | /* |
36 | =for apidoc Ay||utf8ness_t | |
37 | ||
38 | This typedef is used by several core functions that return PV strings, to | |
39 | indicate the UTF-8ness of those strings. | |
40 | ||
41 | (If you write a new function, you probably should instead return the PV in an | |
42 | SV with the UTF-8 flag of the SV properly set, rather than use this mechanism.) | |
43 | ||
44 | The possible values this can be are: | |
45 | ||
46 | =over | |
47 | ||
48 | =item C<UTF8NESS_YES> | |
49 | ||
50 | This means the string definitely should be treated as a sequence of | |
51 | UTF-8-encoded characters. | |
52 | ||
53 | Most code that needs to handle this typedef should be of the form: | |
54 | ||
55 | if (utf8ness_flag == UTF8NESS_YES) { | |
56 | treat as utf8; // like turning on an SV UTF-8 flag | |
57 | } | |
58 | ||
59 | =item C<UTF8NESS_NO> | |
60 | ||
61 | This means the string definitely should be treated as a sequence of bytes, not | |
62 | encoded as UTF-8. | |
63 | ||
64 | =item C<UTF8NESS_IMMATERIAL> | |
65 | ||
66 | This means it is equally valid to treat the string as bytes, or as UTF-8 | |
67 | characters; use whichever way you want. This happens when the string consists | |
68 | entirely of characters which have the same representation whether encoded in | |
69 | UTF-8 or not. | |
70 | ||
71 | =item C<UTF8NESS_UNKNOWN> | |
72 | ||
73 | This means it is unknown how the string should be treated. No core function | |
74 | will ever return this value to a non-core caller. Instead, it is used by the | |
75 | caller to initialize a variable to a non-legal value. A typical call will look like: | |
76 | ||
77 | utf8ness_t string_is_utf8 = UTF8NESS_UNKNOWN | |
78 | const char * string = foo(arg1, arg2, ..., &string_is_utf8); | |
79 | if (string_is_utf8 == UTF8NESS_YES) { | |
80 | do something for UTF-8; | |
81 | } | |
82 | ||
83 | =back | |
84 | ||
85 | The following relationships hold between the enum values: | |
86 | ||
87 | =over | |
88 | ||
89 | =item S<C<0 E<lt>= I<enum value> E<lt>= UTF8NESS_IMMATERIAL>> | |
90 | ||
91 | the string may be treated in code as non-UTF8 | |
92 | ||
93 | =item S<C<UTF8NESS_IMMATERIAL E<lt>= <I<enum value>>> | |
94 | ||
95 | the string may be treated in code as encoded in UTF-8 | |
96 | ||
97 | =back | |
98 | ||
99 | =cut | |
100 | */ | |
101 | ||
102 | typedef enum { | |
103 | UTF8NESS_NO = 0, /* Definitely not UTF-8 */ | |
104 | UTF8NESS_IMMATERIAL = 1, /* Representation is the same in UTF-8 as | |
105 | not, so the UTF8ness doesn't actually | |
106 | matter */ | |
107 | UTF8NESS_YES = 2, /* Defintely is UTF-8, wideness | |
108 | unspecified */ | |
109 | UTF8NESS_UNKNOWN = (STRLEN) -1, /* Undetermined so far */ | |
110 | } utf8ness_t; | |
111 | ||
39e02b42 | 112 | /* Use UTF-8 as the default script encoding? |
1e54db1a | 113 | * Turning this on will break scripts having non-UTF-8 binary |
39e02b42 JH |
114 | * data (such as Latin-1) in string literals. */ |
115 | #ifdef USE_UTF8_SCRIPTS | |
116 | # define USE_UTF8_IN_NAMES (!IN_BYTES) | |
117 | #else | |
118 | # define USE_UTF8_IN_NAMES (PL_hints & HINT_UTF8) | |
119 | #endif | |
120 | ||
3cd96634 KW |
121 | #include "regcharclass.h" |
122 | #include "unicode_constants.h" | |
123 | ||
051a06d4 | 124 | /* For to_utf8_fold_flags, q.v. */ |
e4f4ef45 KW |
125 | #define FOLD_FLAGS_LOCALE 0x1 |
126 | #define FOLD_FLAGS_FULL 0x2 | |
127 | #define FOLD_FLAGS_NOMIX_ASCII 0x4 | |
051a06d4 | 128 | |
7bbfa158 | 129 | /* |
7bbfa158 KW |
130 | =for apidoc is_ascii_string |
131 | ||
8871a094 | 132 | This is a misleadingly-named synonym for L</is_utf8_invariant_string>. |
7bbfa158 KW |
133 | On ASCII-ish platforms, the name isn't misleading: the ASCII-range characters |
134 | are exactly the UTF-8 invariants. But EBCDIC machines have more invariants | |
8871a094 KW |
135 | than just the ASCII characters, so C<is_utf8_invariant_string> is preferred. |
136 | ||
137 | =for apidoc is_invariant_string | |
138 | ||
139 | This is a somewhat misleadingly-named synonym for L</is_utf8_invariant_string>. | |
140 | C<is_utf8_invariant_string> is preferred, as it indicates under what conditions | |
141 | the string is invariant. | |
7bbfa158 KW |
142 | |
143 | =cut | |
144 | */ | |
8871a094 KW |
145 | #define is_ascii_string(s, len) is_utf8_invariant_string(s, len) |
146 | #define is_invariant_string(s, len) is_utf8_invariant_string(s, len) | |
7bbfa158 | 147 | |
33f38593 KW |
148 | #define uvoffuni_to_utf8_flags(d,uv,flags) \ |
149 | uvoffuni_to_utf8_flags_msgs(d, uv, flags, 0) | |
de69f3af KW |
150 | #define uvchr_to_utf8(a,b) uvchr_to_utf8_flags(a,b,0) |
151 | #define uvchr_to_utf8_flags(d,uv,flags) \ | |
33f38593 KW |
152 | uvchr_to_utf8_flags_msgs(d,uv,flags, 0) |
153 | #define uvchr_to_utf8_flags_msgs(d,uv,flags,msgs) \ | |
154 | uvoffuni_to_utf8_flags_msgs(d,NATIVE_TO_UNI(uv),flags, msgs) | |
de69f3af | 155 | #define utf8_to_uvchr_buf(s, e, lenp) \ |
9a9a6c98 | 156 | utf8_to_uvchr_buf_helper((const U8 *) (s), (const U8 *) e, lenp) |
f9380377 KW |
157 | #define utf8n_to_uvchr(s, len, lenp, flags) \ |
158 | utf8n_to_uvchr_error(s, len, lenp, flags, 0) | |
37657a5b KW |
159 | #define utf8n_to_uvchr_error(s, len, lenp, flags, errors) \ |
160 | utf8n_to_uvchr_msgs(s, len, lenp, flags, errors, 0) | |
de69f3af | 161 | |
5fd26678 KW |
162 | #define utf16_to_utf8(p, d, bytelen, newlen) \ |
163 | utf16_to_utf8_base(p, d, bytelen, newlen, 0, 1) | |
164 | #define utf16_to_utf8_reversed(p, d, bytelen, newlen) \ | |
165 | utf16_to_utf8_base(p, d, bytelen, newlen, 1, 0) | |
6af810c4 KW |
166 | #define utf8_to_utf16(p, d, bytelen, newlen) \ |
167 | utf8_to_utf16_base(p, d, bytelen, newlen, 0, 1) | |
168 | #define utf8_to_utf16_reversed(p, d, bytelen, newlen) \ | |
169 | utf8_to_utf16_base(p, d, bytelen, newlen, 1, 0) | |
5fd26678 | 170 | |
a0270393 | 171 | #define to_uni_fold(c, p, lenp) _to_uni_fold_flags(c, p, lenp, FOLD_FLAGS_FULL) |
a239b1e2 | 172 | |
eda9cac1 | 173 | #define foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \ |
1604cfb0 | 174 | foldEQ_utf8_flags(s1, pe1, l1, u1, s2, pe2, l2, u2, 0) |
baa60164 | 175 | #define FOLDEQ_UTF8_NOMIX_ASCII (1 << 0) |
cea315b6 | 176 | #define FOLDEQ_LOCALE (1 << 1) |
18f762c3 KW |
177 | #define FOLDEQ_S1_ALREADY_FOLDED (1 << 2) |
178 | #define FOLDEQ_S2_ALREADY_FOLDED (1 << 3) | |
d635b710 KW |
179 | #define FOLDEQ_S1_FOLDS_SANE (1 << 4) |
180 | #define FOLDEQ_S2_FOLDS_SANE (1 << 5) | |
a33c29bc | 181 | |
fcd03d92 KW |
182 | /* This will be described more fully below, but it turns out that the |
183 | * fundamental difference between UTF-8 and UTF-EBCDIC is that the former has | |
184 | * the upper 2 bits of a continuation byte be '10', and the latter has the | |
185 | * upper 3 bits be '101', leaving 6 and 5 significant bits respectively. | |
186 | * | |
187 | * It is helpful to know the EBCDIC value on ASCII platforms, mainly to avoid | |
188 | * some #ifdef's */ | |
189 | #define UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS 5 | |
190 | ||
e77f0df2 KW |
191 | /* See explanation below at 'UTF8_MAXBYTES' */ |
192 | #define ASCII_PLATFORM_UTF8_MAXBYTES 13 | |
f4d83e55 | 193 | |
6736ce80 | 194 | #ifdef EBCDIC |
f4d83e55 | 195 | |
6736ce80 KW |
196 | /* The equivalent of the next few macros but implementing UTF-EBCDIC are in the |
197 | * following header file: */ | |
198 | # include "utfebcdic.h" | |
f4d83e55 | 199 | |
030c8206 | 200 | # else /* ! EBCDIC */ |
f4d83e55 | 201 | |
6736ce80 | 202 | START_EXTERN_C |
111e8ed9 | 203 | |
6736ce80 KW |
204 | # ifndef DOINIT |
205 | EXTCONST unsigned char PL_utf8skip[]; | |
206 | # else | |
6f06b55f | 207 | EXTCONST unsigned char PL_utf8skip[] = { |
b2635aa8 KW |
208 | /* 0x00 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ |
209 | /* 0x10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
210 | /* 0x20 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
211 | /* 0x30 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
212 | /* 0x40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
213 | /* 0x50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
214 | /* 0x60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
215 | /* 0x70 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */ | |
216 | /* 0x80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */ | |
217 | /* 0x90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */ | |
218 | /* 0xA0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */ | |
219 | /* 0xB0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */ | |
220 | /* 0xC0 */ 2,2, /* overlong */ | |
1ff3baa2 | 221 | /* 0xC2 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0080 to U+03FF */ |
b2635aa8 KW |
222 | /* 0xD0 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0400 to U+07FF */ |
223 | /* 0xE0 */ 3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, /* U+0800 to U+FFFF */ | |
224 | /* 0xF0 */ 4,4,4,4,4,4,4,4,5,5,5,5,6,6, /* above BMP to 2**31 - 1 */ | |
6937f885 KW |
225 | /* Perl extended (never was official UTF-8). Up to 36 bit */ |
226 | /* 0xFE */ 7, | |
227 | /* More extended, Up to 72 bits (64-bit + reserved) */ | |
e77f0df2 | 228 | /* 0xFF */ ASCII_PLATFORM_UTF8_MAXBYTES |
a0ed51b3 | 229 | }; |
6736ce80 | 230 | # endif |
a0ed51b3 | 231 | |
73c4f7a1 | 232 | END_EXTERN_C |
7e2040f0 | 233 | |
ef2a4c8f KW |
234 | /* |
235 | ||
236 | =for apidoc Am|U8|NATIVE_TO_LATIN1|U8 ch | |
237 | ||
238 | Returns the Latin-1 (including ASCII and control characters) equivalent of the | |
239 | input native code point given by C<ch>. Thus, C<NATIVE_TO_LATIN1(193)> on | |
240 | EBCDIC platforms returns 65. These each represent the character C<"A"> on | |
241 | their respective platforms. On ASCII platforms no conversion is needed, so | |
242 | this macro expands to just its input, adding no time nor space requirements to | |
243 | the implementation. | |
244 | ||
245 | For conversion of code points potentially larger than will fit in a character, | |
246 | use L</NATIVE_TO_UNI>. | |
247 | ||
248 | =for apidoc Am|U8|LATIN1_TO_NATIVE|U8 ch | |
249 | ||
250 | Returns the native equivalent of the input Latin-1 code point (including ASCII | |
251 | and control characters) given by C<ch>. Thus, C<LATIN1_TO_NATIVE(66)> on | |
252 | EBCDIC platforms returns 194. These each represent the character C<"B"> on | |
253 | their respective platforms. On ASCII platforms no conversion is needed, so | |
254 | this macro expands to just its input, adding no time nor space requirements to | |
255 | the implementation. | |
256 | ||
257 | For conversion of code points potentially larger than will fit in a character, | |
258 | use L</UNI_TO_NATIVE>. | |
259 | ||
260 | =for apidoc Am|UV|NATIVE_TO_UNI|UV ch | |
261 | ||
262 | Returns the Unicode equivalent of the input native code point given by C<ch>. | |
263 | Thus, C<NATIVE_TO_UNI(195)> on EBCDIC platforms returns 67. These each | |
264 | represent the character C<"C"> on their respective platforms. On ASCII | |
265 | platforms no conversion is needed, so this macro expands to just its input, | |
266 | adding no time nor space requirements to the implementation. | |
267 | ||
268 | =for apidoc Am|UV|UNI_TO_NATIVE|UV ch | |
269 | ||
270 | Returns the native equivalent of the input Unicode code point given by C<ch>. | |
271 | Thus, C<UNI_TO_NATIVE(68)> on EBCDIC platforms returns 196. These each | |
272 | represent the character C<"D"> on their respective platforms. On ASCII | |
273 | platforms no conversion is needed, so this macro expands to just its input, | |
274 | adding no time nor space requirements to the implementation. | |
275 | ||
276 | =cut | |
277 | */ | |
278 | ||
e7285277 KW |
279 | #define NATIVE_TO_LATIN1(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch))) |
280 | #define LATIN1_TO_NATIVE(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch))) | |
59a449d5 KW |
281 | |
282 | /* I8 is an intermediate version of UTF-8 used only in UTF-EBCDIC. We thus | |
283 | * consider it to be identical to UTF-8 on ASCII platforms. Strictly speaking | |
284 | * UTF-8 and UTF-EBCDIC are two different things, but we often conflate them | |
285 | * because they are 8-bit encodings that serve the same purpose in Perl, and | |
286 | * rarely do we need to distinguish them. The term "NATIVE_UTF8" applies to | |
287 | * whichever one is applicable on the current platform */ | |
e7285277 KW |
288 | #define NATIVE_UTF8_TO_I8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch))) |
289 | #define I8_TO_NATIVE_UTF8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch))) | |
59a449d5 | 290 | |
296969d3 KW |
291 | #define UNI_TO_NATIVE(ch) ((UV) ASSERT_NOT_PTR(ch)) |
292 | #define NATIVE_TO_UNI(ch) ((UV) ASSERT_NOT_PTR(ch)) | |
d7578b48 | 293 | |
877d9f0d | 294 | /* |
9041c2e3 | 295 | |
a14e0a36 KW |
296 | The following table is from Unicode 3.2, plus the Perl extensions for above |
297 | U+10FFFF | |
877d9f0d | 298 | |
a14e0a36 | 299 | Code Points 1st Byte 2nd Byte 3rd 4th 5th 6th 7th 8th-13th |
877d9f0d | 300 | |
375122d7 | 301 | U+0000..U+007F 00..7F |
e1b711da | 302 | U+0080..U+07FF * C2..DF 80..BF |
a14e0a36 KW |
303 | U+0800..U+0FFF E0 * A0..BF 80..BF |
304 | U+1000..U+CFFF E1..EC 80..BF 80..BF | |
305 | U+D000..U+D7FF ED 80..9F 80..BF | |
306 | U+D800..U+DFFF ED A0..BF 80..BF (surrogates) | |
307 | U+E000..U+FFFF EE..EF 80..BF 80..BF | |
308 | U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF | |
309 | U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF | |
310 | U+100000..U+10FFFF F4 80..8F 80..BF 80..BF | |
311 | Below are above-Unicode code points | |
312 | U+110000..U+13FFFF F4 90..BF 80..BF 80..BF | |
313 | U+110000..U+1FFFFF F5..F7 80..BF 80..BF 80..BF | |
314 | U+200000..U+FFFFFF F8 * 88..BF 80..BF 80..BF 80..BF | |
315 | U+1000000..U+3FFFFFF F9..FB 80..BF 80..BF 80..BF 80..BF | |
316 | U+4000000..U+3FFFFFFF FC * 84..BF 80..BF 80..BF 80..BF 80..BF | |
317 | U+40000000..U+7FFFFFFF FD 80..BF 80..BF 80..BF 80..BF 80..BF | |
318 | U+80000000..U+FFFFFFFFF FE * 82..BF 80..BF 80..BF 80..BF 80..BF 80..BF | |
319 | U+1000000000.. FF 80..BF 80..BF 80..BF 80..BF 80..BF * 81..BF 80..BF | |
877d9f0d | 320 | |
e1b711da | 321 | Note the gaps before several of the byte entries above marked by '*'. These are |
37e2e78e KW |
322 | caused by legal UTF-8 avoiding non-shortest encodings: it is technically |
323 | possible to UTF-8-encode a single code point in different ways, but that is | |
324 | explicitly forbidden, and the shortest possible encoding should always be used | |
15824458 | 325 | (and that is what Perl does). The non-shortest ones are called 'overlongs'. |
8c007b5a | 326 | |
42b360b2 | 327 | Another way to look at it, as bits: |
8c007b5a | 328 | |
b2635aa8 | 329 | Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte |
8c007b5a | 330 | |
b2635aa8 KW |
331 | 0aaa aaaa 0aaa aaaa |
332 | 0000 0bbb bbaa aaaa 110b bbbb 10aa aaaa | |
333 | cccc bbbb bbaa aaaa 1110 cccc 10bb bbbb 10aa aaaa | |
334 | 00 000d ddcc cccc bbbb bbaa aaaa 1111 0ddd 10cc cccc 10bb bbbb 10aa aaaa | |
8c007b5a JH |
335 | |
336 | As you can see, the continuation bytes all begin with C<10>, and the | |
e1b711da | 337 | leading bits of the start byte tell how many bytes there are in the |
8c007b5a JH |
338 | encoded character. |
339 | ||
df863e43 KW |
340 | Perl's extended UTF-8 means we can have start bytes up through FF, though any |
341 | beginning with FF yields a code point that is too large for 32-bit ASCII | |
342 | platforms. FF signals to use 13 bytes for the encoded character. This breaks | |
343 | the paradigm that the number of leading bits gives how many total bytes there | |
ab2e28c2 | 344 | are in the character. */ |
38953e5a | 345 | |
15824458 KW |
346 | /* This is the number of low-order bits a continuation byte in a UTF-8 encoded |
347 | * sequence contributes to the specification of the code point. In the bit | |
348 | * maps above, you see that the first 2 bits are a constant '10', leaving 6 of | |
349 | * real information */ | |
36da1e17 | 350 | # define UTF_CONTINUATION_BYTE_INFO_BITS 6 |
b2635aa8 | 351 | |
fed423a5 KW |
352 | /* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL() |
353 | * for more */ | |
030c8206 | 354 | # define QUESTION_MARK_CTRL DEL_NATIVE |
fed423a5 | 355 | |
fed423a5 KW |
356 | #endif /* EBCDIC vs ASCII */ |
357 | ||
28ca3ab5 KW |
358 | /* It turns out that in a number of cases, that handling ASCII vs EBCDIC is a |
359 | * matter of being off-by-one. So this is a convenience macro, used to avoid | |
360 | * some #ifdefs. */ | |
361 | #define ONE_IF_EBCDIC_ZERO_IF_NOT \ | |
362 | (UTF_CONTINUATION_BYTE_INFO_BITS == UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS) | |
363 | ||
36da1e17 KW |
364 | /* Since the significant bits in a continuation byte are stored in the |
365 | * least-significant positions, we often find ourselves shifting by that | |
366 | * amount. This is a clearer name in such situations */ | |
367 | #define UTF_ACCUMULATION_SHIFT UTF_CONTINUATION_BYTE_INFO_BITS | |
368 | ||
42b360b2 KW |
369 | /* 2**info_bits - 1. This masks out all but the bits that carry real |
370 | * information in a continuation byte. This turns out to be 0x3F in UTF-8, | |
371 | * 0x1F in UTF-EBCDIC. */ | |
36da1e17 KW |
372 | #define UTF_CONTINUATION_MASK \ |
373 | ((U8) nBIT_MASK(UTF_CONTINUATION_BYTE_INFO_BITS)) | |
fed423a5 | 374 | |
9f3cfb7a KW |
375 | /* For use in UTF8_IS_CONTINUATION(). This turns out to be 0xC0 in UTF-8, |
376 | * E0 in UTF-EBCDIC */ | |
377 | #define UTF_IS_CONTINUATION_MASK ((U8) (0xFF << UTF_ACCUMULATION_SHIFT)) | |
378 | ||
38f458ff KW |
379 | /* This defines the bits that are to be in the continuation bytes of a |
380 | * multi-byte UTF-8 encoded character that mark it is a continuation byte. | |
381 | * This turns out to be 0x80 in UTF-8, 0xA0 in UTF-EBCDIC. (khw doesn't know | |
fcd03d92 KW |
382 | * the underlying reason that B0 works here, except it just happens to work. |
383 | * One could solve for two linear equations and come up with it.) */ | |
38f458ff KW |
384 | #define UTF_CONTINUATION_MARK (UTF_IS_CONTINUATION_MASK & 0xB0) |
385 | ||
4a2c769e KW |
386 | /* This value is clearer in some contexts */ |
387 | #define UTF_MIN_CONTINUATION_BYTE UTF_CONTINUATION_MARK | |
388 | ||
f4225fa0 KW |
389 | /* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the |
390 | * first byte thereof? */ | |
391 | #define UTF8_IS_CONTINUATION(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
392 | (((NATIVE_UTF8_TO_I8(c) & UTF_IS_CONTINUATION_MASK) \ | |
393 | == UTF_CONTINUATION_MARK))) | |
394 | ||
2dc97505 KW |
395 | /* Is the representation of the Unicode code point 'cp' the same regardless of |
396 | * being encoded in UTF-8 or not? This is a fundamental property of | |
397 | * UTF-8,EBCDIC */ | |
4a2c769e KW |
398 | #define OFFUNI_IS_INVARIANT(c) \ |
399 | (((WIDEST_UTYPE)(c)) < UTF_MIN_CONTINUATION_BYTE) | |
2dc97505 | 400 | |
ab2e28c2 KW |
401 | /* |
402 | =for apidoc Am|bool|UVCHR_IS_INVARIANT|UV cp | |
403 | ||
404 | Evaluates to 1 if the representation of code point C<cp> is the same whether or | |
405 | not it is encoded in UTF-8; otherwise evaluates to 0. UTF-8 invariant | |
406 | characters can be copied as-is when converting to/from UTF-8, saving time. | |
407 | C<cp> is Unicode if above 255; otherwise is platform-native. | |
408 | ||
409 | =cut | |
410 | */ | |
411 | #define UVCHR_IS_INVARIANT(cp) (OFFUNI_IS_INVARIANT(NATIVE_TO_UNI(cp))) | |
412 | ||
c26e6896 KW |
413 | /* This defines the 1-bits that are to be in the first byte of a multi-byte |
414 | * UTF-8 encoded character that mark it as a start byte and give the number of | |
415 | * bytes that comprise the character. 'len' is that number. | |
416 | * | |
417 | * To illustrate: len = 2 => ((U8) ~ 0b0011_1111) or 1100_0000 | |
418 | * 7 => ((U8) ~ 0b0000_0001) or 1111_1110 | |
419 | * > 7 => 0xFF | |
420 | * | |
421 | * This is not to be used on a single-byte character. As in many places in | |
422 | * perl, U8 must be 8 bits | |
423 | */ | |
424 | #define UTF_START_MARK(len) ((U8) ~(0xFF >> (len))) | |
425 | ||
0176a72c KW |
426 | /* Masks out the initial one bits in a start byte, leaving the following 0 bit |
427 | * and the real data bits. 'len' is the number of bytes in the multi-byte | |
428 | * sequence that comprises the character. | |
429 | * | |
430 | * To illustrate: len = 2 => 0b0011_1111 works on start byte 110xxxxx | |
431 | * 6 => 0b0000_0011 works on start byte 1111110x | |
432 | * >= 7 => There are no data bits in the start byte | |
433 | * Note that on ASCII platforms, this can be passed a len=1 byte; and all the | |
434 | * real data bits will be returned: | |
435 | len = 1 => 0b0111_1111 | |
436 | * This isn't true on EBCDIC platforms, where some len=1 bytes are of the form | |
437 | * 0b101x_xxxx, so this can't be used there on single-byte characters. */ | |
438 | #define UTF_START_MASK(len) (0xFF >> (len)) | |
439 | ||
e77f0df2 KW |
440 | /* |
441 | ||
442 | =for apidoc AmnU|STRLEN|UTF8_MAXBYTES | |
443 | ||
444 | The maximum width of a single UTF-8 encoded character, in bytes. | |
445 | ||
446 | NOTE: Strictly speaking Perl's UTF-8 should not be called UTF-8 since UTF-8 | |
447 | is an encoding of Unicode, and Unicode's upper limit, 0x10FFFF, can be | |
448 | expressed with 4 bytes. However, Perl thinks of UTF-8 as a way to encode | |
449 | non-negative integers in a binary format, even those above Unicode. | |
450 | ||
451 | =cut | |
452 | ||
453 | The start byte 0xFE, never used in any ASCII platform UTF-8 specification, has | |
454 | an obvious meaning, namely it has its upper 7 bits set, so it should start a | |
455 | sequence of 7 bytes. And in fact, this is exactly what standard UTF-EBCDIC | |
456 | does. | |
457 | ||
458 | The start byte FF, on the other hand could have several different plausible | |
459 | meanings: | |
460 | 1) The meaning in standard UTF-EBCDIC, namely as an FE start byte, with the | |
461 | bottom bit that should be a fixed '0' to form FE, instead acting as an | |
462 | info bit, 0 or 1. | |
463 | 2) That the sequence should have exactly 8 bytes. | |
464 | 3) That the next byte is to be treated as a sort of extended start byte, | |
465 | which in combination with this one gives the total length of the sequence. | |
466 | There are published UTF-8 extensions that do this, some string together | |
467 | multiple initial FF start bytes to achieve arbitrary precision. | |
468 | 4) That the sequence has exactly n bytes, where n is what the implementation | |
469 | chooses. | |
470 | ||
471 | Perl has chosen 4). | |
472 | The goal is to be able to represent 64-bit values in UTF-8 or UTF-EBCDIC. That | |
473 | rules out items 1) and 2). Item 3) has the deal-breaking disadvantage of | |
474 | requiring one to read more than one byte to determine the total length of the | |
475 | sequence. So in Perl, a start byte of FF indicates a UTF-8 string consisting | |
476 | of the start byte, plus enough continuation bytes to encode a 64 bit value. | |
477 | This turns out to be 13 total bytes in UTF-8 and 14 in UTF-EBCDIC. This is | |
478 | because we get zero info bits from the start byte, plus | |
479 | 12 * 6 bits of info per continuation byte (could encode 72-bit numbers) on | |
480 | UTF-8 (khw knows not why 11, which would encode 66 bits wasn't | |
481 | chosen instead); and | |
482 | 13 * 5 bits of info per byte (could encode 65-bit numbers) on UTF-EBCDIC | |
483 | ||
484 | The disadvantages of this method are: | |
485 | 1) There's potentially a lot of wasted bytes for all but the largest values. | |
486 | For example, something that could be represented by 7 continuation bytes, | |
487 | instead requires the full 12 or 13. | |
488 | 2) There would be problems should larger values, 128-bit say, ever need to be | |
489 | represented. | |
490 | ||
491 | WARNING: This number must be in sync with the value in | |
492 | regen/charset_translations.pl. */ | |
493 | #define UTF8_MAXBYTES \ | |
494 | (ASCII_PLATFORM_UTF8_MAXBYTES + ONE_IF_EBCDIC_ZERO_IF_NOT) | |
495 | ||
7bf011a1 KW |
496 | /* Calculate how many bytes are necessary to represent a value whose most |
497 | * significant 1 bit is in bit position 'pos' of the word. For 0x1, 'pos would | |
498 | * be 0; and for 0x400, 'pos' would be 10, and the result would be: | |
499 | * EBCDIC floor((-1 + (10 + 5 - 1 - 1)) / (5 - 1)) | |
500 | * = floor((-1 + (13)) / 4) | |
501 | * = floor(12 / 4) | |
502 | * = 3 | |
503 | * ASCII floor(( 0 + (10 + 6 - 1 - 1)) / (6 - 1)) | |
504 | * = floor(14 / 5) | |
505 | * = 2 | |
506 | * The reason this works is because the number of bits needed to represent a | |
507 | * value is proportional to (UTF_CONTINUATION_BYTE_INFO_BITS - 1). The -1 is | |
508 | * because each new continuation byte removes one bit of information from the | |
509 | * start byte. | |
510 | * | |
511 | * This is a step function (we need to allocate a full extra byte if we | |
512 | * overflow by just a single bit) | |
513 | * | |
514 | * The caller is responsible for making sure 'pos' is at least 8 (occupies 9 | |
515 | * bits), as it breaks down at the lower edge. At the high end, if it returns | |
516 | * 8 or more, Perl instead anomalously uses MAX_BYTES, so this would be wrong. | |
517 | * */ | |
518 | #define UNISKIP_BY_MSB_(pos) \ | |
519 | ( ( -ONE_IF_EBCDIC_ZERO_IF_NOT /* platform break pos's are off-by-one */ \ | |
520 | + (pos) + ((UTF_CONTINUATION_BYTE_INFO_BITS - 1) - 1)) /* Step fcn */ \ | |
521 | / (UTF_CONTINUATION_BYTE_INFO_BITS - 1)) /* take floor of */ | |
522 | ||
787e8384 KW |
523 | /* Compute the number of UTF-8 bytes required for representing the input uv, |
524 | * which must be a Unicode, not native value. | |
7028aeba | 525 | * |
787e8384 KW |
526 | * This uses msbit_pos() which doesn't work on NUL, and UNISKIP_BY_MSB_ breaks |
527 | * down for small code points. So first check if the input is invariant to get | |
528 | * around that, and use a helper for high code points to accommodate the fact | |
529 | * that above 7 btyes, the value is anomalous. The helper is empty on | |
530 | * platforms that don't go that high */ | |
531 | #define OFFUNISKIP(uv) \ | |
532 | ((OFFUNI_IS_INVARIANT(uv)) \ | |
533 | ? 1 \ | |
534 | : (OFFUNISKIP_helper_(uv) UNISKIP_BY_MSB_(msbit_pos(uv)))) | |
535 | ||
536 | /* We need to go to MAX_BYTES when we can't represent 'uv' by the number of | |
537 | * information bits in 6 continuation bytes (when we get to 6, the start byte | |
538 | * has no information bits to add to the total). But on 32-bit ASCII | |
539 | * platforms, that doesn't happen until 6*6 bits, so on those platforms, this | |
540 | * will always be false */ | |
541 | #if UVSIZE * CHARBITS > (6 * UTF_CONTINUATION_BYTE_INFO_BITS) | |
788cdc67 | 542 | # define HAS_EXTRA_LONG_UTF8 |
787e8384 KW |
543 | # define OFFUNISKIP_helper_(uv) \ |
544 | UNLIKELY(uv > nBIT_UMAX(6 * UTF_CONTINUATION_BYTE_INFO_BITS)) \ | |
545 | ? UTF8_MAXBYTES : | |
1d68d6cd | 546 | #else |
787e8384 | 547 | # define OFFUNISKIP_helper_(uv) |
1d68d6cd SC |
548 | #endif |
549 | ||
5352a763 KW |
550 | /* |
551 | ||
552 | =for apidoc Am|STRLEN|UVCHR_SKIP|UV cp | |
553 | returns the number of bytes required to represent the code point C<cp> when | |
554 | encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than | |
555 | 255; a Unicode code point otherwise. | |
556 | ||
557 | =cut | |
558 | */ | |
787e8384 | 559 | #define UVCHR_SKIP(uv) OFFUNISKIP(NATIVE_TO_UNI(uv)) |
5352a763 | 560 | |
222d7630 KW |
561 | #define NATIVE_SKIP(uv) UVCHR_SKIP(uv) /* Old terminology */ |
562 | ||
563 | /* Most code which says UNISKIP is really thinking in terms of native code | |
564 | * points (0-255) plus all those beyond. This is an imprecise term, but having | |
565 | * it means existing code continues to work. For precision, use UVCHR_SKIP, | |
566 | * NATIVE_SKIP, or OFFUNISKIP */ | |
567 | #define UNISKIP(uv) UVCHR_SKIP(uv) | |
568 | ||
bdcc1e93 KW |
569 | /* Compute the start byte for a given code point. This requires the log2 of |
570 | * the code point, which is hard to compute at compile time, which this macro | |
571 | * wants to be. (Perhaps deBruijn sequences could be used.) So a parameter | |
572 | * for the number of bits the value occupies is passed in, which the programmer | |
573 | * has had to figure out to get compile-time effect. And asserts are used to | |
574 | * make sure the value is correct. | |
575 | * | |
576 | * Since we are interested only in the start byte, we ignore the lower bits | |
577 | * accounted for by the continuation bytes. Each continuation byte eats up | |
578 | * UTF_CONTINUATION_BYTE_INFO_BITS bits, so the number of continuation bytes | |
579 | * needed is floor(bits / UTF_CONTINUATION_BYTE_INFO_BITS). That number is fed | |
580 | * to UTF_START_MARK() to get the upper part of the start byte. The left over | |
581 | * bits form the lower part which is OR'd with the mark | |
582 | * | |
583 | * Note that on EBCDIC platforms, this is actually the I8 */ | |
584 | #define UTF_START_BYTE(uv, bits) \ | |
585 | (__ASSERT_((uv) >> ((bits) - 1)) /* At least 'bits' */ \ | |
586 | __ASSERT_(((uv) & ~nBIT_MASK(bits)) == 0) /* No extra bits */ \ | |
587 | UTF_START_MARK(UNISKIP_BY_MSB_((bits) - 1)) \ | |
588 | | ((uv) >> (((bits) / UTF_CONTINUATION_BYTE_INFO_BITS) \ | |
589 | * UTF_CONTINUATION_BYTE_INFO_BITS))) | |
590 | ||
591 | /* Compute the first continuation byte for a given code point. This is mostly | |
592 | * for compile-time, so how many bits it occupies is also passed in). | |
593 | * | |
594 | * We are interested in the first continuation byte, so we ignore the lower | |
595 | * bits accounted for by the rest of the continuation bytes by right shifting | |
596 | * out their info bit, and mask out the higher bits that will go into the start | |
597 | * byte. | |
598 | * | |
599 | * Note that on EBCDIC platforms, this is actually the I8 */ | |
600 | #define UTF_FIRST_CONT_BYTE(uv, bits) \ | |
601 | (__ASSERT_((uv) >> ((bits) - 1)) /* At least 'bits' */ \ | |
602 | __ASSERT_(((uv) & ~nBIT_MASK(bits)) == 0) /* No extra bits */ \ | |
603 | UTF_CONTINUATION_MARK \ | |
604 | | ( UTF_CONTINUATION_MASK \ | |
605 | & ((uv) >> ((((bits) / UTF_CONTINUATION_BYTE_INFO_BITS) - 1) \ | |
606 | * UTF_CONTINUATION_BYTE_INFO_BITS)))) | |
607 | ||
608 | #define UTF_MIN_START_BYTE UTF_START_BYTE(UTF_MIN_CONTINUATION_BYTE, 8) | |
4bab39bc KW |
609 | |
610 | /* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence? | |
59645eb1 | 611 | * This excludes invariants (they are single-byte). It also excludes the |
4bab39bc | 612 | * illegal overlong sequences that begin with C0 and C1 on ASCII platforms, and |
59645eb1 KW |
613 | * C0-C4 I8 start bytes on EBCDIC ones. On EBCDIC E0 can't start a |
614 | * non-overlong sequence, so we define a base macro and for those platforms, | |
615 | * extend it to also exclude E0 */ | |
616 | #define UTF8_IS_START_base(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
4bab39bc | 617 | (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_START_BYTE)) |
59645eb1 KW |
618 | #ifdef EBCDIC |
619 | # define UTF8_IS_START(c) \ | |
620 | (UTF8_IS_START_base(c) && (c) != I8_TO_NATIVE_UTF8(0xE0)) | |
621 | #else | |
622 | # define UTF8_IS_START(c) UTF8_IS_START_base(c) | |
623 | #endif | |
4bab39bc | 624 | |
bdcc1e93 | 625 | #define UTF_MIN_ABOVE_LATIN1_BYTE UTF_START_BYTE(0x100, 9) |
1df63428 KW |
626 | |
627 | /* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that | |
628 | * represent a code point > 255? */ | |
629 | #define UTF8_IS_ABOVE_LATIN1(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
630 | (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_ABOVE_LATIN1_BYTE)) | |
631 | ||
7c88d61e KW |
632 | /* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use |
633 | * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to | |
634 | * be well-formed. */ | |
635 | #define UTF8_IS_DOWNGRADEABLE_START(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
03dc0b1b | 636 | inRANGE_helper_(U8, NATIVE_UTF8_TO_I8(c), \ |
7c88d61e KW |
637 | UTF_MIN_START_BYTE, UTF_MIN_ABOVE_LATIN1_BYTE - 1)) |
638 | ||
b651802e | 639 | /* The largest code point representable by two UTF-8 bytes on this platform. |
03a8ddc0 KW |
640 | * The binary for that code point is: |
641 | * 1101_1111 10xx_xxxx in UTF-8, and | |
642 | * 1101_1111 101y_yyyy in UTF-EBCDIC I8. | |
643 | * where both x and y are 1, and shown this way to indicate there is one more x | |
644 | * than there is y. The number of x and y bits are their platform's respective | |
645 | * UTF_CONTINUATION_BYTE_INFO_BITS. Squeezing out the bits that don't | |
646 | * contribute to the value, these evaluate to: | |
647 | * 1_1111 xx_xxxx in UTF-8, and | |
648 | * 1_1111 y_yyyy in UTF-EBCDIC I8. | |
649 | * or, the maximum value of an unsigned with (5 + info_bit_count) bits */ | |
650 | #define MAX_UTF8_TWO_BYTE nBIT_UMAX(5 + UTF_CONTINUATION_BYTE_INFO_BITS) | |
aa206fb7 | 651 | |
b651802e | 652 | /* The largest code point representable by two UTF-8 bytes on any platform that |
fcd03d92 KW |
653 | * Perl runs on. */ |
654 | #define MAX_PORTABLE_UTF8_TWO_BYTE \ | |
655 | nBIT_UMAX(5 + MIN( UTF_CONTINUATION_BYTE_INFO_BITS, \ | |
656 | UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS)) | |
aa206fb7 | 657 | |
f2c50040 KW |
658 | /* |
659 | ||
660 | =for apidoc AmnU|STRLEN|UTF8_MAXBYTES_CASE | |
661 | ||
662 | The maximum number of UTF-8 bytes a single Unicode character can | |
663 | uppercase/lowercase/titlecase/fold into. | |
664 | ||
665 | =cut | |
666 | ||
667 | * Unicode guarantees that the maximum expansion is UTF8_MAX_FOLD_CHAR_EXPAND | |
668 | * characters, but any above-Unicode code point will fold to itself, so we only | |
669 | * have to look at the expansion of the maximum Unicode code point. But this | |
670 | * number may be less than the space occupied by a very large code point under | |
671 | * Perl's extended UTF-8. We have to make it large enough to fit any single | |
672 | * character. (It turns out that ASCII and EBCDIC differ in which is larger) | |
673 | * | |
674 | =cut | |
675 | */ | |
7bf011a1 KW |
676 | #define UTF8_MAXBYTES_CASE \ |
677 | MAX(UTF8_MAXBYTES, UTF8_MAX_FOLD_CHAR_EXPAND * UNISKIP_BY_MSB_(20)) | |
c03c0950 | 678 | |
d06134e5 KW |
679 | /* Rest of these are attributes of Unicode and perl's internals rather than the |
680 | * encoding, or happen to be the same in both ASCII and EBCDIC (at least at | |
681 | * this level; the macros that some of these call may have different | |
682 | * definitions in the two encodings */ | |
683 | ||
59a449d5 KW |
684 | /* In domain restricted to ASCII, these may make more sense to the reader than |
685 | * the ones with Latin1 in the name */ | |
686 | #define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch) | |
687 | #define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch) | |
688 | ||
689 | /* More or less misleadingly-named defines, retained for back compat */ | |
690 | #define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch) | |
691 | #define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch) | |
692 | #define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch) | |
693 | #define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch) | |
694 | #define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch) | |
d06134e5 | 695 | |
537124e4 KW |
696 | /* Adds a UTF8 continuation byte 'new' of information to a running total code |
697 | * point 'old' of all the continuation bytes so far. This is designed to be | |
155d2738 KW |
698 | * used in a loop to convert from UTF-8 to the code point represented. Note |
699 | * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is | |
700 | * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC) | |
701 | * code point in process of being generated */ | |
a6951642 KW |
702 | #define UTF8_ACCUMULATE(old, new) (__ASSERT_(FITS_IN_8_BITS(new)) \ |
703 | ((old) << UTF_ACCUMULATION_SHIFT) \ | |
009097b1 | 704 | | ((NATIVE_UTF8_TO_I8(new)) \ |
155d2738 | 705 | & UTF_CONTINUATION_MASK)) |
d06134e5 | 706 | |
4ab10950 | 707 | /* This works in the face of malformed UTF-8. */ |
4e1ed312 | 708 | #define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) \ |
0b47b392 KW |
709 | ( ( (e) - (s) > 1) \ |
710 | && UTF8_IS_DOWNGRADEABLE_START(*(s)) \ | |
4e1ed312 | 711 | && UTF8_IS_CONTINUATION(*((s)+1))) |
4ab10950 | 712 | |
3c0792e4 KW |
713 | /* Longer, but more accurate name */ |
714 | #define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c) | |
715 | ||
a62b247b KW |
716 | /* Convert a UTF-8 variant Latin1 character to a native code point value. |
717 | * Needs just one iteration of accumulate. Should be used only if it is known | |
718 | * that the code point is < 256, and is not UTF-8 invariant. Use the slower | |
719 | * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point | |
720 | * representable by two bytes (which turns out to be up through | |
721 | * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are: | |
722 | * HI: a downgradable start byte; | |
723 | * LO: continuation. | |
724 | * */ | |
725 | #define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \ | |
726 | ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \ | |
727 | __ASSERT_(UTF8_IS_CONTINUATION(LO)) \ | |
728 | LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \ | |
729 | NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO)))) | |
730 | ||
94bb8c36 | 731 | /* Convert a two (not one) byte utf8 character to a native code point value. |
2950f2a7 KW |
732 | * Needs just one iteration of accumulate. Should not be used unless it is |
733 | * known that the two bytes are legal: 1) two-byte start, and 2) continuation. | |
734 | * Note that the result can be larger than 255 if the input character is not | |
735 | * downgradable */ | |
94bb8c36 | 736 | #define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \ |
a6951642 KW |
737 | (__ASSERT_(FITS_IN_8_BITS(HI)) \ |
738 | __ASSERT_(FITS_IN_8_BITS(LO)) \ | |
46b0a3a2 | 739 | __ASSERT_(PL_utf8skip[(U8) HI] == 2) \ |
a6951642 | 740 | __ASSERT_(UTF8_IS_CONTINUATION(LO)) \ |
94bb8c36 | 741 | UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \ |
635e76f5 | 742 | (LO)))) |
94bb8c36 KW |
743 | |
744 | /* Should never be used, and be deprecated */ | |
745 | #define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO)) | |
2950f2a7 | 746 | |
bd18bd40 KW |
747 | /* |
748 | ||
749 | =for apidoc Am|STRLEN|UTF8SKIP|char* s | |
ee0ff0f5 KW |
750 | returns the number of bytes a non-malformed UTF-8 encoded character whose first |
751 | (perhaps only) byte is pointed to by C<s>. | |
752 | ||
753 | If there is a possibility of malformed input, use instead: | |
754 | ||
755 | =over | |
756 | ||
eb992c6f | 757 | =item C<L</UTF8_SAFE_SKIP>> if you know the maximum ending pointer in the |
ee0ff0f5 KW |
758 | buffer pointed to by C<s>; or |
759 | ||
eb992c6f | 760 | =item C<L</UTF8_CHK_SKIP>> if you don't know it. |
ee0ff0f5 KW |
761 | |
762 | =back | |
763 | ||
764 | It is better to restructure your code so the end pointer is passed down so that | |
765 | you know what it actually is at the point of this call, but if that isn't | |
eb992c6f | 766 | possible, C<L</UTF8_CHK_SKIP>> can minimize the chance of accessing beyond the end |
ee0ff0f5 | 767 | of the input buffer. |
bd18bd40 KW |
768 | |
769 | =cut | |
770 | */ | |
08e4e7bf | 771 | #define UTF8SKIP(s) PL_utf8skip[*(const U8*)(ASSERT_IS_PTR(s))] |
a281f16c KW |
772 | |
773 | /* | |
774 | =for apidoc Am|STRLEN|UTF8_SKIP|char* s | |
eb992c6f | 775 | This is a synonym for C<L</UTF8SKIP>> |
a281f16c KW |
776 | |
777 | =cut | |
778 | */ | |
779 | ||
2a70536e | 780 | #define UTF8_SKIP(s) UTF8SKIP(s) |
d06134e5 | 781 | |
85fcc8f2 | 782 | /* |
ee0ff0f5 KW |
783 | =for apidoc Am|STRLEN|UTF8_CHK_SKIP|char* s |
784 | ||
eb992c6f KW |
785 | This is a safer version of C<L</UTF8SKIP>>, but still not as safe as |
786 | C<L</UTF8_SAFE_SKIP>>. This version doesn't blindly assume that the input | |
ee0ff0f5 KW |
787 | string pointed to by C<s> is well-formed, but verifies that there isn't a NUL |
788 | terminating character before the expected end of the next character in C<s>. | |
789 | The length C<UTF8_CHK_SKIP> returns stops just before any such NUL. | |
790 | ||
791 | Perl tends to add NULs, as an insurance policy, after the end of strings in | |
792 | SV's, so it is likely that using this macro will prevent inadvertent reading | |
793 | beyond the end of the input buffer, even if it is malformed UTF-8. | |
794 | ||
795 | This macro is intended to be used by XS modules where the inputs could be | |
796 | malformed, and it isn't feasible to restructure to use the safer | |
eb992c6f | 797 | C<L</UTF8_SAFE_SKIP>>, for example when interfacing with a C library. |
ee0ff0f5 KW |
798 | |
799 | =cut | |
800 | */ | |
801 | ||
802 | #define UTF8_CHK_SKIP(s) \ | |
8974941d | 803 | (UNLIKELY(s[0] == '\0') ? 1 : MIN(UTF8SKIP(s), \ |
f87d8789 | 804 | my_strnlen((char *) (s), UTF8SKIP(s)))) |
ee0ff0f5 | 805 | /* |
85fcc8f2 KW |
806 | |
807 | =for apidoc Am|STRLEN|UTF8_SAFE_SKIP|char* s|char* e | |
45671da2 KW |
808 | returns 0 if S<C<s E<gt>= e>>; otherwise returns the number of bytes in the |
809 | UTF-8 encoded character whose first byte is pointed to by C<s>. But it never | |
810 | returns beyond C<e>. On DEBUGGING builds, it asserts that S<C<s E<lt>= e>>. | |
85fcc8f2 KW |
811 | |
812 | =cut | |
813 | */ | |
45671da2 | 814 | #define UTF8_SAFE_SKIP(s, e) (__ASSERT_((e) >= (s)) \ |
8974941d | 815 | UNLIKELY(((e) - (s)) <= 0) \ |
45671da2 KW |
816 | ? 0 \ |
817 | : MIN(((e) - (s)), UTF8_SKIP(s))) | |
85fcc8f2 | 818 | |
2d1545e5 KW |
819 | /* Most code that says 'UNI_' really means the native value for code points up |
820 | * through 255 */ | |
821 | #define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp) | |
822 | ||
c2b32798 KW |
823 | /* |
824 | =for apidoc Am|bool|UTF8_IS_INVARIANT|char c | |
825 | ||
826 | Evaluates to 1 if the byte C<c> represents the same character when encoded in | |
827 | UTF-8 as when not; otherwise evaluates to 0. UTF-8 invariant characters can be | |
828 | copied as-is when converting to/from UTF-8, saving time. | |
829 | ||
830 | In spite of the name, this macro gives the correct result if the input string | |
831 | from which C<c> comes is not encoded in UTF-8. | |
832 | ||
833 | See C<L</UVCHR_IS_INVARIANT>> for checking if a UV is invariant. | |
834 | ||
835 | =cut | |
836 | ||
837 | The reason it works on both UTF-8 encoded strings and non-UTF-8 encoded, is | |
838 | that it returns TRUE in each for the exact same set of bit patterns. It is | |
839 | valid on a subset of what UVCHR_IS_INVARIANT is valid on, so can just use that; | |
840 | and the compiler should optimize out anything extraneous given the | |
296969d3 KW |
841 | implementation of the latter. */ |
842 | #define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT(ASSERT_NOT_PTR(c)) | |
5fc230f1 KW |
843 | |
844 | /* Like the above, but its name implies a non-UTF8 input, which as the comments | |
845 | * above show, doesn't matter as to its implementation */ | |
38953e5a | 846 | #define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c) |
d06134e5 | 847 | |
2c03e801 KW |
848 | /* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence |
849 | * in UTF-8? This is the inverse of UTF8_IS_INVARIANT. */ | |
850 | #define UTF8_IS_CONTINUED(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
851 | (! UTF8_IS_INVARIANT(c))) | |
852 | ||
48ccf5e1 KW |
853 | /* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic |
854 | * bytes from an ordinal that is known to fit into exactly two (not one) bytes; | |
855 | * it must be less than 0x3FF to work across both encodings. */ | |
856 | ||
857 | /* These two are helper macros for the other three sets, and should not be used | |
858 | * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1 | |
1ff3baa2 KW |
859 | * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works |
860 | * for any code point */ | |
48ccf5e1 | 861 | #define __BASE_TWO_BYTE_HI(c, translate_function) \ |
2863dafa | 862 | (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \ |
48ccf5e1 | 863 | I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \ |
2863dafa | 864 | | UTF_START_MARK(2))) |
48ccf5e1 | 865 | #define __BASE_TWO_BYTE_LO(c, translate_function) \ |
2863dafa | 866 | (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \ |
48ccf5e1 | 867 | I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \ |
2863dafa | 868 | | UTF_CONTINUATION_MARK)) |
48ccf5e1 | 869 | |
48ccf5e1 KW |
870 | /* The next two macros should not be used. They were designed to be usable as |
871 | * the case label of a switch statement, but this doesn't work for EBCDIC. Use | |
9d0d3a03 | 872 | * regen/unicode_constants.pl instead */ |
48ccf5e1 KW |
873 | #define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI) |
874 | #define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI) | |
875 | ||
876 | /* The next two macros are used when the source should be a single byte | |
877 | * character; checked for under DEBUGGING */ | |
878 | #define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ | |
4c8cd605 | 879 | ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1))) |
48ccf5e1 | 880 | #define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \ |
4c8cd605 | 881 | (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1))) |
48ccf5e1 KW |
882 | |
883 | /* These final two macros in the series are used when the source can be any | |
884 | * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient | |
885 | * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~' | |
886 | * operator instead of "<=" to avoid getting compiler warnings. | |
d52b8576 | 887 | * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few |
48ccf5e1 KW |
888 | * places, so the ~ works */ |
889 | #define UTF8_TWO_BYTE_HI(c) \ | |
890 | (__ASSERT_((sizeof(c) == 1) \ | |
d52b8576 | 891 | || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \ |
4c8cd605 | 892 | (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI))) |
48ccf5e1 KW |
893 | #define UTF8_TWO_BYTE_LO(c) \ |
894 | (__ASSERT_((sizeof(c) == 1) \ | |
d52b8576 | 895 | || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \ |
4c8cd605 | 896 | (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI))) |
d06134e5 | 897 | |
e7214ce8 KW |
898 | /* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII |
899 | * as it is only in overlongs. */ | |
900 | #define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1) | |
901 | ||
7e2040f0 | 902 | /* |
e3036cf4 | 903 | * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem |
20df05f4 KW |
904 | * from an earlier version of these macros in which they didn't call the |
905 | * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the | |
906 | * beginning of a utf8 character. Now that foo_utf8() determines that itself, | |
907 | * no need to do it again here | |
7e2040f0 | 908 | */ |
da8c1a98 KW |
909 | #define isIDFIRST_lazy_if_safe(p, e, UTF) \ |
910 | ((IN_BYTES || !UTF) \ | |
911 | ? isIDFIRST(*(p)) \ | |
912 | : isIDFIRST_utf8_safe(p, e)) | |
da8c1a98 KW |
913 | #define isWORDCHAR_lazy_if_safe(p, e, UTF) \ |
914 | ((IN_BYTES || !UTF) \ | |
915 | ? isWORDCHAR(*(p)) \ | |
916 | : isWORDCHAR_utf8_safe((U8 *) p, (U8 *) e)) | |
4c1d9526 | 917 | #define isALNUM_lazy_if_safe(p, e, UTF) isWORDCHAR_lazy_if_safe(p, e, UTF) |
da8c1a98 | 918 | |
030c8206 | 919 | #define UTF8_MAXLEN UTF8_MAXBYTES |
89ebb4a3 | 920 | |
8cb75cc8 | 921 | /* A Unicode character can fold to up to 3 characters */ |
030c8206 | 922 | #define UTF8_MAX_FOLD_CHAR_EXPAND 3 |
8cb75cc8 | 923 | |
030c8206 | 924 | #define IN_BYTES UNLIKELY(CopHINTS_get(PL_curcop) & HINT_BYTES) |
bd18bd40 KW |
925 | |
926 | /* | |
927 | ||
928 | =for apidoc Am|bool|DO_UTF8|SV* sv | |
929 | Returns a bool giving whether or not the PV in C<sv> is to be treated as being | |
930 | encoded in UTF-8. | |
931 | ||
932 | You should use this I<after> a call to C<SvPV()> or one of its variants, in | |
933 | case any call to string overloading updates the internal UTF-8 encoding flag. | |
934 | ||
935 | =cut | |
936 | */ | |
0064a8a9 | 937 | #define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES) |
1ff3baa2 KW |
938 | |
939 | /* Should all strings be treated as Unicode, and not just UTF-8 encoded ones? | |
940 | * Is so within 'feature unicode_strings' or 'locale :not_characters', and not | |
941 | * within 'use bytes'. UTF-8 locales are not tested for here, but perhaps | |
942 | * could be */ | |
70844984 | 943 | #define IN_UNI_8_BIT \ |
1604cfb0 | 944 | (( ( (CopHINTS_get(PL_curcop) & HINT_UNI_8_BIT)) \ |
70844984 KW |
945 | || ( CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \ |
946 | /* -1 below is for :not_characters */ \ | |
947 | && _is_in_locale_category(FALSE, -1))) \ | |
948 | && (! IN_BYTES)) | |
b36bf33f | 949 | |
6110285c KW |
950 | #define UNICODE_SURROGATE_FIRST 0xD800 |
951 | #define UNICODE_SURROGATE_LAST 0xDFFF | |
952 | ||
953 | /* | |
39fafb79 KW |
954 | =for apidoc Am|bool|UNICODE_IS_SURROGATE|const UV uv |
955 | ||
956 | Returns a boolean as to whether or not C<uv> is one of the Unicode surrogate | |
957 | code points | |
958 | ||
6110285c KW |
959 | =for apidoc Am|bool|UTF8_IS_SURROGATE|const U8 *s|const U8 *e |
960 | ||
961 | Evaluates to non-zero if the first few bytes of the string starting at C<s> and | |
962 | looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one | |
963 | of the Unicode surrogate code points; otherwise it evaluates to 0. If | |
964 | non-zero, the value gives how many bytes starting at C<s> comprise the code | |
965 | point's representation. | |
966 | ||
967 | =cut | |
968 | */ | |
969 | ||
6110285c KW |
970 | #define UNICODE_IS_SURROGATE(uv) UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST, \ |
971 | UNICODE_SURROGATE_LAST)) | |
972 | #define UTF8_IS_SURROGATE(s, e) is_SURROGATE_utf8_safe(s, e) | |
973 | ||
974 | /* | |
975 | ||
976 | =for apidoc AmnU|UV|UNICODE_REPLACEMENT | |
977 | ||
978 | Evaluates to 0xFFFD, the code point of the Unicode REPLACEMENT CHARACTER | |
979 | ||
39fafb79 KW |
980 | =for apidoc Am|bool|UNICODE_IS_REPLACEMENT|const UV uv |
981 | ||
982 | Returns a boolean as to whether or not C<uv> is the Unicode REPLACEMENT | |
983 | CHARACTER | |
984 | ||
985 | =for apidoc Am|bool|UTF8_IS_REPLACEMENT|const U8 *s|const U8 *e | |
986 | ||
987 | Evaluates to non-zero if the first few bytes of the string starting at C<s> and | |
988 | looking no further than S<C<e - 1>> are well-formed UTF-8 that represents the | |
989 | Unicode REPLACEMENT CHARACTER; otherwise it evaluates to 0. If non-zero, the | |
990 | value gives how many bytes starting at C<s> comprise the code point's | |
991 | representation. | |
992 | ||
6110285c KW |
993 | =cut |
994 | */ | |
995 | #define UNICODE_REPLACEMENT 0xFFFD | |
996 | #define UNICODE_IS_REPLACEMENT(uv) UNLIKELY((UV) (uv) == UNICODE_REPLACEMENT) | |
3c9bbd85 YO |
997 | #define UTF8_IS_REPLACEMENT(s, send) \ |
998 | UNLIKELY( \ | |
999 | ((send) - (s)) >= ((SSize_t)(sizeof(REPLACEMENT_CHARACTER_UTF8) - 1))\ | |
1000 | && memEQ((s), REPLACEMENT_CHARACTER_UTF8, \ | |
ae005663 | 1001 | sizeof(REPLACEMENT_CHARACTER_UTF8) - 1)) |
6110285c | 1002 | |
42b360b2 | 1003 | /* Max legal code point according to Unicode */ |
6110285c KW |
1004 | #define PERL_UNICODE_MAX 0x10FFFF |
1005 | ||
39fafb79 KW |
1006 | /* |
1007 | ||
1008 | =for apidoc Am|bool|UNICODE_IS_SUPER|const UV uv | |
1009 | ||
1010 | Returns a boolean as to whether or not C<uv> is above the maximum legal Unicode | |
1011 | code point of U+10FFFF. | |
1012 | ||
1013 | =cut | |
1014 | */ | |
1015 | ||
030c8206 | 1016 | #define UNICODE_IS_SUPER(uv) UNLIKELY((UV) (uv) > PERL_UNICODE_MAX) |
6110285c KW |
1017 | |
1018 | /* | |
1019 | =for apidoc Am|bool|UTF8_IS_SUPER|const U8 *s|const U8 *e | |
1020 | ||
1021 | Recall that Perl recognizes an extension to UTF-8 that can encode code | |
1022 | points larger than the ones defined by Unicode, which are 0..0x10FFFF. | |
1023 | ||
1024 | This macro evaluates to non-zero if the first few bytes of the string starting | |
1025 | at C<s> and looking no further than S<C<e - 1>> are from this UTF-8 extension; | |
7ce5b055 | 1026 | otherwise it evaluates to 0. If non-zero, the return is how many bytes |
6110285c KW |
1027 | starting at C<s> comprise the code point's representation. |
1028 | ||
1029 | 0 is returned if the bytes are not well-formed extended UTF-8, or if they | |
1030 | represent a code point that cannot fit in a UV on the current platform. Hence | |
1031 | this macro can give different results when run on a 64-bit word machine than on | |
1032 | one with a 32-bit word size. | |
1033 | ||
7ce5b055 KW |
1034 | Note that it is illegal in Perl to have code points that are larger than what can |
1035 | fit in an IV on the current machine; and illegal in Unicode to have any that | |
1036 | this macro matches | |
6110285c KW |
1037 | |
1038 | =cut | |
1039 | ||
1040 | * ASCII EBCDIC I8 | |
1041 | * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode | |
1042 | * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0 | |
1043 | * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1 | |
1044 | */ | |
7ce5b055 KW |
1045 | #define UTF_START_BYTE_110000_ UTF_START_BYTE(PERL_UNICODE_MAX + 1, 21) |
1046 | #define UTF_FIRST_CONT_BYTE_110000_ \ | |
1047 | UTF_FIRST_CONT_BYTE(PERL_UNICODE_MAX + 1, 21) | |
1048 | #define UTF8_IS_SUPER(s, e) \ | |
1049 | ( ((e) - (s)) >= UNISKIP_BY_MSB_(20) \ | |
1050 | && ( NATIVE_UTF8_TO_I8(s[0]) >= UTF_START_BYTE_110000_ \ | |
1051 | && ( NATIVE_UTF8_TO_I8(s[0]) > UTF_START_BYTE_110000_ \ | |
1052 | || NATIVE_UTF8_TO_I8(s[1]) >= UTF_FIRST_CONT_BYTE_110000_))) \ | |
1053 | ? isUTF8_CHAR(s, e) \ | |
1054 | : 0 | |
6110285c | 1055 | |
39fafb79 KW |
1056 | /* |
1057 | =for apidoc Am|bool|UNICODE_IS_NONCHAR|const UV uv | |
1058 | ||
1059 | Returns a boolean as to whether or not C<uv> is one of the Unicode | |
1060 | non-character code points | |
1061 | ||
1062 | =cut | |
1063 | */ | |
1064 | ||
6110285c KW |
1065 | /* Is 'uv' one of the 32 contiguous-range noncharacters? */ |
1066 | #define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \ | |
1067 | UNLIKELY(inRANGE(uv, 0xFDD0, 0xFDEF)) | |
1068 | ||
1069 | /* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE, | |
1070 | * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above | |
1071 | * the Unicode legal max */ | |
030c8206 KW |
1072 | #define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \ |
1073 | UNLIKELY(((UV) (uv) & 0xFFFE) == 0xFFFE) | |
6110285c | 1074 | |
030c8206 KW |
1075 | #define UNICODE_IS_NONCHAR(uv) \ |
1076 | ( UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)) \ | |
1077 | || ( UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)) \ | |
6110285c KW |
1078 | && LIKELY(! UNICODE_IS_SUPER(uv)))) |
1079 | ||
6110285c KW |
1080 | /* |
1081 | =for apidoc Am|bool|UTF8_IS_NONCHAR|const U8 *s|const U8 *e | |
1082 | ||
1083 | Evaluates to non-zero if the first few bytes of the string starting at C<s> and | |
1084 | looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one | |
1085 | of the Unicode non-character code points; otherwise it evaluates to 0. If | |
1086 | non-zero, the value gives how many bytes starting at C<s> comprise the code | |
1087 | point's representation. | |
1088 | ||
1089 | =cut | |
1090 | */ | |
65e4aa05 KW |
1091 | #define UTF8_IS_NONCHAR(s, e) is_NONCHAR_utf8_safe(s,e) |
1092 | ||
1093 | /* This is now machine generated, and the 'given' clause is no longer | |
1094 | * applicable */ | |
1095 | #define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \ | |
1096 | UTF8_IS_NONCHAR(s, e) | |
6110285c | 1097 | |
6f8b1f93 KW |
1098 | /* Surrogates, non-character code points and above-Unicode code points are |
1099 | * problematic in some contexts. These macros allow code that needs to check | |
1100 | * for those to quickly exclude the vast majority of code points it will | |
1101 | * encounter. | |
1102 | * | |
1103 | * The lowest such code point is the smallest surrogate, U+D800. We calculate | |
1104 | * the start byte of that. 0xD800 occupies 16 bits. */ | |
1105 | #define isUNICODE_POSSIBLY_PROBLEMATIC(uv) ((uv) >= UNICODE_SURROGATE_FIRST) | |
1106 | #define isUTF8_POSSIBLY_PROBLEMATIC(c) \ | |
1107 | (NATIVE_UTF8_TO_I8(c) >= UTF_START_BYTE(UNICODE_SURROGATE_FIRST, 16)) | |
1d72bdf6 | 1108 | |
99904f65 KW |
1109 | /* Perl extends Unicode so that it is possible to encode (as extended UTF-8 or |
1110 | * UTF-EBCDIC) any 64-bit value. No standard known to khw ever encoded higher | |
1111 | * than a 31 bit value. On ASCII platforms this just meant arbitrarily saying | |
1112 | * nothing could be higher than this. On these the start byte FD gets you to | |
1113 | * 31 bits, and FE and FF are forbidden as start bytes. On EBCDIC platforms, | |
1114 | * FD gets you only to 26 bits; adding FE to mean 7 total bytes gets you to 30 | |
1115 | * bits. To get to 31 bits, they treated an initial FF byte idiosyncratically. | |
1116 | * It was considered to be the start byte FE meaning it had 7 total bytes, and | |
1117 | * the final 1 was treated as an information bit, getting you to 31 bits. | |
1118 | * | |
1119 | * Perl used to accept this idiosyncratic interpretation of FF, but now rejects | |
1120 | * it in order to get to being able to encode 64 bits. The bottom line is that | |
1121 | * it is a Perl extension to use the start bytes FE and FF on ASCII platforms, | |
1122 | * and the start byte FF on EBCDIC ones. That translates into that it is a | |
1123 | * Perl extension to represent anything occupying more than 31 bits on ASCII | |
1124 | * platforms; 30 bits on EBCDIC. */ | |
1125 | #define UNICODE_IS_PERL_EXTENDED(uv) \ | |
1126 | UNLIKELY((UV) (uv) > nBIT_UMAX(31 - ONE_IF_EBCDIC_ZERO_IF_NOT)) | |
43732c4f KW |
1127 | #define UTF8_IS_PERL_EXTENDED(s) \ |
1128 | (UTF8SKIP(s) > 6 + ONE_IF_EBCDIC_ZERO_IF_NOT) | |
99904f65 | 1129 | |
42b360b2 | 1130 | /* Largest code point we accept from external sources */ |
6110285c KW |
1131 | #define MAX_LEGAL_CP ((UV)IV_MAX) |
1132 | ||
c76687c5 | 1133 | #define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */ |
2b5e7bc2 | 1134 | #define UTF8_GOT_EMPTY UTF8_ALLOW_EMPTY |
c76687c5 KW |
1135 | |
1136 | /* Allow first byte to be a continuation byte */ | |
1d72bdf6 | 1137 | #define UTF8_ALLOW_CONTINUATION 0x0002 |
2b5e7bc2 | 1138 | #define UTF8_GOT_CONTINUATION UTF8_ALLOW_CONTINUATION |
c76687c5 | 1139 | |
cd01d3b1 | 1140 | /* Unexpected non-continuation byte */ |
1d72bdf6 | 1141 | #define UTF8_ALLOW_NON_CONTINUATION 0x0004 |
2b5e7bc2 | 1142 | #define UTF8_GOT_NON_CONTINUATION UTF8_ALLOW_NON_CONTINUATION |
949cf498 KW |
1143 | |
1144 | /* expecting more bytes than were available in the string */ | |
1145 | #define UTF8_ALLOW_SHORT 0x0008 | |
2b5e7bc2 | 1146 | #define UTF8_GOT_SHORT UTF8_ALLOW_SHORT |
949cf498 | 1147 | |
94953955 KW |
1148 | /* Overlong sequence; i.e., the code point can be specified in fewer bytes. |
1149 | * First one will convert the overlong to the REPLACEMENT CHARACTER; second | |
1150 | * will return what the overlong evaluates to */ | |
949cf498 | 1151 | #define UTF8_ALLOW_LONG 0x0010 |
94953955 | 1152 | #define UTF8_ALLOW_LONG_AND_ITS_VALUE (UTF8_ALLOW_LONG|0x0020) |
2b5e7bc2 KW |
1153 | #define UTF8_GOT_LONG UTF8_ALLOW_LONG |
1154 | ||
d60baaa7 KW |
1155 | #define UTF8_ALLOW_OVERFLOW 0x0080 |
1156 | #define UTF8_GOT_OVERFLOW UTF8_ALLOW_OVERFLOW | |
949cf498 | 1157 | |
f180b292 | 1158 | #define UTF8_DISALLOW_SURROGATE 0x0100 /* Unicode surrogates */ |
2b5e7bc2 | 1159 | #define UTF8_GOT_SURROGATE UTF8_DISALLOW_SURROGATE |
f180b292 | 1160 | #define UTF8_WARN_SURROGATE 0x0200 |
949cf498 | 1161 | |
c4e96019 KW |
1162 | /* Unicode non-character code points */ |
1163 | #define UTF8_DISALLOW_NONCHAR 0x0400 | |
2b5e7bc2 | 1164 | #define UTF8_GOT_NONCHAR UTF8_DISALLOW_NONCHAR |
c4e96019 | 1165 | #define UTF8_WARN_NONCHAR 0x0800 |
949cf498 | 1166 | |
c4e96019 KW |
1167 | /* Super-set of Unicode: code points above the legal max */ |
1168 | #define UTF8_DISALLOW_SUPER 0x1000 | |
2b5e7bc2 | 1169 | #define UTF8_GOT_SUPER UTF8_DISALLOW_SUPER |
c4e96019 KW |
1170 | #define UTF8_WARN_SUPER 0x2000 |
1171 | ||
1172 | /* The original UTF-8 standard did not define UTF-8 with start bytes of 0xFE or | |
1173 | * 0xFF, though UTF-EBCDIC did. This allowed both versions to represent code | |
1174 | * points up to 2 ** 31 - 1. Perl extends UTF-8 so that 0xFE and 0xFF are | |
1175 | * usable on ASCII platforms, and 0xFF means something different than | |
1176 | * UTF-EBCDIC defines. These changes allow code points of 64 bits (actually | |
1177 | * somewhat more) to be represented on both platforms. But these are Perl | |
1178 | * extensions, and not likely to be interchangeable with other languages. Note | |
1179 | * that on ASCII platforms, FE overflows a signed 32-bit word, and FF an | |
1180 | * unsigned one. */ | |
d044b7a7 KW |
1181 | #define UTF8_DISALLOW_PERL_EXTENDED 0x4000 |
1182 | #define UTF8_GOT_PERL_EXTENDED UTF8_DISALLOW_PERL_EXTENDED | |
1183 | #define UTF8_WARN_PERL_EXTENDED 0x8000 | |
d35f2ca5 | 1184 | |
57ff5f59 KW |
1185 | /* For back compat, these old names are misleading for overlongs and |
1186 | * UTF_EBCDIC. */ | |
d044b7a7 KW |
1187 | #define UTF8_DISALLOW_ABOVE_31_BIT UTF8_DISALLOW_PERL_EXTENDED |
1188 | #define UTF8_GOT_ABOVE_31_BIT UTF8_GOT_PERL_EXTENDED | |
1189 | #define UTF8_WARN_ABOVE_31_BIT UTF8_WARN_PERL_EXTENDED | |
1190 | #define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_PERL_EXTENDED | |
1191 | #define UTF8_WARN_FE_FF UTF8_WARN_PERL_EXTENDED | |
949cf498 | 1192 | |
f180b292 | 1193 | #define UTF8_CHECK_ONLY 0x10000 |
99a765e9 | 1194 | #define _UTF8_NO_CONFIDENCE_IN_CURLEN 0x20000 /* Internal core use only */ |
949cf498 KW |
1195 | |
1196 | /* For backwards source compatibility. They do nothing, as the default now | |
1197 | * includes what they used to mean. The first one's meaning was to allow the | |
1198 | * just the single non-character 0xFFFF */ | |
1199 | #define UTF8_ALLOW_FFFF 0 | |
c825ef8c | 1200 | #define UTF8_ALLOW_FE_FF 0 |
949cf498 KW |
1201 | #define UTF8_ALLOW_SURROGATE 0 |
1202 | ||
ecc1615f KW |
1203 | /* C9 refers to Unicode Corrigendum #9: allows but discourages non-chars */ |
1204 | #define UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE \ | |
1205 | (UTF8_DISALLOW_SUPER|UTF8_DISALLOW_SURROGATE) | |
1206 | #define UTF8_WARN_ILLEGAL_C9_INTERCHANGE (UTF8_WARN_SUPER|UTF8_WARN_SURROGATE) | |
1207 | ||
d35f2ca5 | 1208 | #define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \ |
ecc1615f | 1209 | (UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE|UTF8_DISALLOW_NONCHAR) |
949cf498 | 1210 | #define UTF8_WARN_ILLEGAL_INTERCHANGE \ |
ecc1615f KW |
1211 | (UTF8_WARN_ILLEGAL_C9_INTERCHANGE|UTF8_WARN_NONCHAR) |
1212 | ||
0eb3d6a0 KW |
1213 | /* This is typically used for code that processes UTF-8 input and doesn't want |
1214 | * to have to deal with any malformations that might be present. All such will | |
1215 | * be safely replaced by the REPLACEMENT CHARACTER, unless other flags | |
1216 | * overriding this are also present. */ | |
2d532c27 KW |
1217 | #define UTF8_ALLOW_ANY ( UTF8_ALLOW_CONTINUATION \ |
1218 | |UTF8_ALLOW_NON_CONTINUATION \ | |
1219 | |UTF8_ALLOW_SHORT \ | |
d60baaa7 KW |
1220 | |UTF8_ALLOW_LONG \ |
1221 | |UTF8_ALLOW_OVERFLOW) | |
2d532c27 KW |
1222 | |
1223 | /* Accept any Perl-extended UTF-8 that evaluates to any UV on the platform, but | |
cd01d3b1 | 1224 | * not any malformed. This is the default. */ |
2d532c27 KW |
1225 | #define UTF8_ALLOW_ANYUV 0 |
1226 | #define UTF8_ALLOW_DEFAULT UTF8_ALLOW_ANYUV | |
1d72bdf6 | 1227 | |
d044b7a7 KW |
1228 | #define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */ |
1229 | #define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */ | |
1230 | #define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */ | |
1231 | #define UNICODE_WARN_PERL_EXTENDED 0x0008 /* Above 0x7FFF_FFFF */ | |
1232 | #define UNICODE_WARN_ABOVE_31_BIT UNICODE_WARN_PERL_EXTENDED | |
1233 | #define UNICODE_DISALLOW_SURROGATE 0x0010 | |
1234 | #define UNICODE_DISALLOW_NONCHAR 0x0020 | |
1235 | #define UNICODE_DISALLOW_SUPER 0x0040 | |
1236 | #define UNICODE_DISALLOW_PERL_EXTENDED 0x0080 | |
24b4c303 KW |
1237 | |
1238 | #ifdef PERL_CORE | |
1239 | # define UNICODE_ALLOW_ABOVE_IV_MAX 0x0100 | |
1240 | #endif | |
d044b7a7 | 1241 | #define UNICODE_DISALLOW_ABOVE_31_BIT UNICODE_DISALLOW_PERL_EXTENDED |
33f38593 KW |
1242 | |
1243 | #define UNICODE_GOT_SURROGATE UNICODE_DISALLOW_SURROGATE | |
1244 | #define UNICODE_GOT_NONCHAR UNICODE_DISALLOW_NONCHAR | |
1245 | #define UNICODE_GOT_SUPER UNICODE_DISALLOW_SUPER | |
1246 | #define UNICODE_GOT_PERL_EXTENDED UNICODE_DISALLOW_PERL_EXTENDED | |
1247 | ||
ecc1615f KW |
1248 | #define UNICODE_WARN_ILLEGAL_C9_INTERCHANGE \ |
1249 | (UNICODE_WARN_SURROGATE|UNICODE_WARN_SUPER) | |
bb88be5f | 1250 | #define UNICODE_WARN_ILLEGAL_INTERCHANGE \ |
ecc1615f KW |
1251 | (UNICODE_WARN_ILLEGAL_C9_INTERCHANGE|UNICODE_WARN_NONCHAR) |
1252 | #define UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE \ | |
1253 | (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_SUPER) | |
bb88be5f | 1254 | #define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \ |
ecc1615f | 1255 | (UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE|UNICODE_DISALLOW_NONCHAR) |
949cf498 KW |
1256 | |
1257 | /* For backward source compatibility, as are now the default */ | |
1258 | #define UNICODE_ALLOW_SURROGATE 0 | |
1259 | #define UNICODE_ALLOW_SUPER 0 | |
1260 | #define UNICODE_ALLOW_ANY 0 | |
b851fbc1 | 1261 | |
6110285c | 1262 | #define UNICODE_BYTE_ORDER_MARK 0xFEFF |
030c8206 KW |
1263 | #define UNICODE_IS_BYTE_ORDER_MARK(uv) UNLIKELY((UV) (uv) \ |
1264 | == UNICODE_BYTE_ORDER_MARK) | |
c149ab20 | 1265 | |
ec34087a KW |
1266 | #define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE |
1267 | #define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \ | |
1268 | LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE | |
1269 | #define MICRO_SIGN MICRO_SIGN_NATIVE | |
1270 | #define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \ | |
1271 | LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE | |
1272 | #define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \ | |
1273 | LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE | |
09091399 JH |
1274 | #define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3 |
1275 | #define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2 | |
1276 | #define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3 | |
9dcbe121 | 1277 | #define GREEK_SMALL_LETTER_MU 0x03BC |
9e682c18 KW |
1278 | #define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case |
1279 | of MICRON */ | |
1280 | #define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */ | |
0766489e KW |
1281 | #ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8 |
1282 | # define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E | |
1283 | #endif | |
74894415 KW |
1284 | #define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130 |
1285 | #define LATIN_SMALL_LETTER_DOTLESS_I 0x131 | |
9e682c18 | 1286 | #define LATIN_SMALL_LETTER_LONG_S 0x017F |
a9f50d33 KW |
1287 | #define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05 |
1288 | #define LATIN_SMALL_LIGATURE_ST 0xFB06 | |
9e682c18 KW |
1289 | #define KELVIN_SIGN 0x212A |
1290 | #define ANGSTROM_SIGN 0x212B | |
09091399 | 1291 | |
9e55ce06 | 1292 | #define UNI_DISPLAY_ISPRINT 0x0001 |
c728cb41 | 1293 | #define UNI_DISPLAY_BACKSLASH 0x0002 |
daf6caf1 KW |
1294 | #define UNI_DISPLAY_BACKSPACE 0x0004 /* Allow \b when also |
1295 | UNI_DISPLAY_BACKSLASH */ | |
1296 | #define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT \ | |
1297 | |UNI_DISPLAY_BACKSLASH \ | |
1298 | |UNI_DISPLAY_BACKSPACE) | |
1299 | ||
1300 | /* Character classes could also allow \b, but not patterns in general */ | |
c728cb41 | 1301 | #define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH) |
9e55ce06 | 1302 | |
e0ffa6d6 | 1303 | /* Should be removed; maybe deprecated, but not used in CPAN */ |
ebc501f0 | 1304 | #define SHARP_S_SKIP 2 |
3b0fc154 | 1305 | |
3cedd9d9 | 1306 | #define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end) |
976c1b08 KW |
1307 | #define bytes_from_utf8(s, lenp, is_utf8p) \ |
1308 | bytes_from_utf8_loc(s, lenp, is_utf8p, 0) | |
3cedd9d9 | 1309 | |
6302f837 KW |
1310 | /* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is |
1311 | * retained solely for backwards compatibility */ | |
1312 | #define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n) | |
e9a8c099 | 1313 | |
6a5bc5ac | 1314 | #endif /* PERL_UTF8_H_ */ |
57f0e7e2 | 1315 | |
e9a8c099 | 1316 | /* |
14d04a33 | 1317 | * ex: set ts=8 sts=4 sw=4 et: |
e9a8c099 | 1318 | */ |