3 * This file contains definitions for use with the UTF-8 encoding. It
4 * actually also works with the variant UTF-8 encoding called UTF-EBCDIC, and
5 * hides almost all of the differences between these from the caller. In other
6 * words, someone should #include this file, and if the code is being compiled
7 * on an EBCDIC platform, things should mostly just work.
9 * Copyright (C) 2000, 2001, 2002, 2005, 2006, 2007, 2009,
10 * 2010, 2011 by Larry Wall and others
12 * You may distribute under the terms of either the GNU General Public
13 * License or the Artistic License, as specified in the README file.
17 #ifndef H_UTF8 /* Guard against recursive inclusion */
20 /* Use UTF-8 as the default script encoding?
21 * Turning this on will break scripts having non-UTF-8 binary
22 * data (such as Latin-1) in string literals. */
23 #ifdef USE_UTF8_SCRIPTS
24 # define USE_UTF8_IN_NAMES (!IN_BYTES)
26 # define USE_UTF8_IN_NAMES (PL_hints & HINT_UTF8)
29 #include "regcharclass.h"
30 #include "unicode_constants.h"
32 /* For to_utf8_fold_flags, q.v. */
33 #define FOLD_FLAGS_LOCALE 0x1
34 #define FOLD_FLAGS_FULL 0x2
35 #define FOLD_FLAGS_NOMIX_ASCII 0x4
37 /* For _core_swash_init(), internal core use only */
38 #define _CORE_SWASH_INIT_USER_DEFINED_PROPERTY 0x1
39 #define _CORE_SWASH_INIT_RETURN_IF_UNDEF 0x2
40 #define _CORE_SWASH_INIT_ACCEPT_INVLIST 0x4
43 =head1 Unicode Support
44 L<perlguts/Unicode Support> has an introduction to this API.
46 See also L</Character classification>,
47 and L</Character case changing>.
48 Various functions outside this section also work specially with Unicode.
49 Search for the string "utf8" in this document.
51 =for apidoc is_ascii_string
53 This is a misleadingly-named synonym for L</is_invariant_string>.
54 On ASCII-ish platforms, the name isn't misleading: the ASCII-range characters
55 are exactly the UTF-8 invariants. But EBCDIC machines have more invariants
56 than just the ASCII characters, so C<is_invariant_string> is preferred.
60 #define is_ascii_string(s, len) is_invariant_string(s, len)
62 #define uvchr_to_utf8(a,b) uvchr_to_utf8_flags(a,b,0)
63 #define uvchr_to_utf8_flags(d,uv,flags) \
64 uvoffuni_to_utf8_flags(d,NATIVE_TO_UNI(uv),flags)
65 #define utf8_to_uvchr_buf(s, e, lenp) \
66 utf8n_to_uvchr(s, (U8*)(e) - (U8*)(s), lenp, \
67 ckWARN_d(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY)
69 #define to_uni_fold(c, p, lenp) _to_uni_fold_flags(c, p, lenp, FOLD_FLAGS_FULL)
70 #define to_utf8_fold(c, p, lenp) _to_utf8_fold_flags(c, p, lenp, FOLD_FLAGS_FULL)
71 #define to_utf8_lower(a,b,c) _to_utf8_lower_flags(a,b,c,0)
72 #define to_utf8_upper(a,b,c) _to_utf8_upper_flags(a,b,c,0)
73 #define to_utf8_title(a,b,c) _to_utf8_title_flags(a,b,c,0)
75 /* Source backward compatibility. */
76 #define is_utf8_string_loc(s, len, ep) is_utf8_string_loclen(s, len, ep, 0)
78 #define foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
79 foldEQ_utf8_flags(s1, pe1, l1, u1, s2, pe2, l2, u2, 0)
80 #define FOLDEQ_UTF8_NOMIX_ASCII (1 << 0)
81 #define FOLDEQ_LOCALE (1 << 1)
82 #define FOLDEQ_S1_ALREADY_FOLDED (1 << 2)
83 #define FOLDEQ_S2_ALREADY_FOLDED (1 << 3)
84 #define FOLDEQ_S1_FOLDS_SANE (1 << 4)
85 #define FOLDEQ_S2_FOLDS_SANE (1 << 5)
87 #define ibcmp_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
88 cBOOL(! foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2))
91 /* The equivalent of these macros but implementing UTF-EBCDIC
92 are in the following header file:
95 #include "utfebcdic.h"
100 /* How wide can a single UTF-8 encoded character become in bytes. */
101 /* NOTE: Strictly speaking Perl's UTF-8 should not be called UTF-8 since UTF-8
102 * is an encoding of Unicode, and Unicode's upper limit, 0x10FFFF, can be
103 * expressed with 4 bytes. However, Perl thinks of UTF-8 as a way to encode
104 * non-negative integers in a binary format, even those above Unicode */
105 #define UTF8_MAXBYTES 13
108 EXTCONST unsigned char PL_utf8skip[] = {
109 /* 0x00 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
110 /* 0x10 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
111 /* 0x20 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
112 /* 0x30 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
113 /* 0x40 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
114 /* 0x50 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
115 /* 0x60 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
116 /* 0x70 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* ascii */
117 /* 0x80 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
118 /* 0x90 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
119 /* 0xA0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
120 /* 0xB0 */ 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1, /* bogus: continuation byte */
121 /* 0xC0 */ 2,2, /* overlong */
122 /* 0xC2 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0080 to U+03FF */
123 /* 0xD0 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0400 to U+07FF */
124 /* 0xE0 */ 3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, /* U+0800 to U+FFFF */
125 /* 0xF0 */ 4,4,4,4,4,4,4,4,5,5,5,5,6,6, /* above BMP to 2**31 - 1 */
126 /* Perl extended (never was official UTF-8). Up to 36 bit */
128 /* More extended, Up to 72 bits (64-bit + reserved) */
129 /* 0xFF */ UTF8_MAXBYTES
132 EXTCONST unsigned char PL_utf8skip[];
137 /* Native character to/from iso-8859-1. Are the identity functions on ASCII
139 #define NATIVE_TO_LATIN1(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
140 #define LATIN1_TO_NATIVE(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
142 /* I8 is an intermediate version of UTF-8 used only in UTF-EBCDIC. We thus
143 * consider it to be identical to UTF-8 on ASCII platforms. Strictly speaking
144 * UTF-8 and UTF-EBCDIC are two different things, but we often conflate them
145 * because they are 8-bit encodings that serve the same purpose in Perl, and
146 * rarely do we need to distinguish them. The term "NATIVE_UTF8" applies to
147 * whichever one is applicable on the current platform */
148 #define NATIVE_UTF8_TO_I8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
149 #define I8_TO_NATIVE_UTF8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
151 /* Transforms in wide UV chars */
152 #define UNI_TO_NATIVE(ch) ((UV) (ch))
153 #define NATIVE_TO_UNI(ch) ((UV) (ch))
157 The following table is from Unicode 3.2.
159 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
161 U+0000..U+007F 00..7F
162 U+0080..U+07FF * C2..DF 80..BF
163 U+0800..U+0FFF E0 * A0..BF 80..BF
164 U+1000..U+CFFF E1..EC 80..BF 80..BF
165 U+D000..U+D7FF ED 80..9F 80..BF
166 U+D800..U+DFFF ED A0..BF 80..BF (surrogates)
167 U+E000..U+FFFF EE..EF 80..BF 80..BF
168 U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF
169 U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF
170 U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
171 Below are non-Unicode code points
172 U+110000..U+13FFFF F4 90..BF 80..BF 80..BF
173 U+110000..U+1FFFFF F5..F7 80..BF 80..BF 80..BF
174 U+200000..: F8.. * 88..BF 80..BF 80..BF 80..BF
176 Note the gaps before several of the byte entries above marked by '*'. These are
177 caused by legal UTF-8 avoiding non-shortest encodings: it is technically
178 possible to UTF-8-encode a single code point in different ways, but that is
179 explicitly forbidden, and the shortest possible encoding should always be used
180 (and that is what Perl does). The non-shortest ones are called 'overlongs'.
185 Another way to look at it, as bits:
187 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
190 0000 0bbb bbaa aaaa 110b bbbb 10aa aaaa
191 cccc bbbb bbaa aaaa 1110 cccc 10bb bbbb 10aa aaaa
192 00 000d ddcc cccc bbbb bbaa aaaa 1111 0ddd 10cc cccc 10bb bbbb 10aa aaaa
194 As you can see, the continuation bytes all begin with C<10>, and the
195 leading bits of the start byte tell how many bytes there are in the
198 Perl's extended UTF-8 means we can have start bytes up to FF.
202 /* Is the representation of the Unicode code point 'cp' the same regardless of
203 * being encoded in UTF-8 or not? */
204 #define OFFUNI_IS_INVARIANT(cp) isASCII(cp)
206 /* Is the representation of the code point 'cp' the same regardless of
207 * being encoded in UTF-8 or not? 'cp' is native if < 256; Unicode otherwise
209 #define UVCHR_IS_INVARIANT(cp) OFFUNI_IS_INVARIANT(cp)
211 /* This defines the bits that are to be in the continuation bytes of a multi-byte
212 * UTF-8 encoded character that mark it is a continuation byte. */
213 #define UTF_CONTINUATION_MARK 0x80
215 /* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence
216 * in UTF-8? This is the inverse of UTF8_IS_INVARIANT */
217 #define UTF8_IS_CONTINUED(c) (((U8)c) & UTF_CONTINUATION_MARK)
219 /* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence?
220 * This doesn't catch invariants (they are single-byte). It also excludes the
221 * illegal overlong sequences that begin with C0 and C1. */
222 #define UTF8_IS_START(c) (((U8)c) >= 0xc2)
224 /* For use in UTF8_IS_CONTINUATION() below */
225 #define UTF_IS_CONTINUATION_MASK 0xC0
227 /* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the
228 * first byte thereof? */
229 #define UTF8_IS_CONTINUATION(c) \
230 ((((U8)c) & UTF_IS_CONTINUATION_MASK) == UTF_CONTINUATION_MARK)
232 /* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use
233 * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to
234 * be well-formed. Masking with 0xfe allows the low bit to be 0 or 1; thus
235 * this matches 0xc[23]. */
236 #define UTF8_IS_DOWNGRADEABLE_START(c) (((U8)(c) & 0xfe) == 0xc2)
238 /* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that
239 * represent a code point > 255? */
240 #define UTF8_IS_ABOVE_LATIN1(c) ((U8)(c) >= 0xc4)
242 /* This is the number of low-order bits a continuation byte in a UTF-8 encoded
243 * sequence contributes to the specification of the code point. In the bit
244 * maps above, you see that the first 2 bits are a constant '10', leaving 6 of
245 * real information */
246 #define UTF_ACCUMULATION_SHIFT 6
248 /* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL()
250 #define QUESTION_MARK_CTRL DEL_NATIVE
252 /* Surrogates, non-character code points and above-Unicode code points are
253 * problematic in some contexts. This allows code that needs to check for
254 * those to to quickly exclude the vast majority of code points it will
256 #define isUTF8_POSSIBLY_PROBLEMATIC(c) ((U8) c >= 0xED)
258 #endif /* EBCDIC vs ASCII */
260 /* 2**UTF_ACCUMULATION_SHIFT - 1 */
261 #define UTF_CONTINUATION_MASK ((U8) ((1U << UTF_ACCUMULATION_SHIFT) - 1))
263 /* Internal macro to be used only in this file to aid in constructing other
264 * publicly accessible macros.
265 * The number of bytes required to express this uv in UTF-8, for just those
266 * uv's requiring 2 through 6 bytes, as these are common to all platforms and
267 * word sizes. The number of bytes needed is given by the number of leading 1
268 * bits in the start byte. There are 32 start bytes that have 2 initial 1 bits
269 * (C0-DF); there are 16 that have 3 initial 1 bits (E0-EF); 8 that have 4
270 * initial 1 bits (F0-F8); 4 that have 5 initial 1 bits (F9-FB), and 2 that
271 * have 6 initial 1 bits (FC-FD). The largest number a string of n bytes can
272 * represent is (the number of possible start bytes for 'n')
273 * * (the number of possiblities for each start byte
274 * The latter in turn is
275 * 2 ** ( (how many continuation bytes there are)
276 * * (the number of bits of information each
277 * continuation byte holds))
279 * If we were on a platform where we could use a fast find first set bit
280 * instruction (or count leading zeros instruction) this could be replaced by
281 * using that to find the log2 of the uv, and divide that by the number of bits
282 * of information in each continuation byte, adjusting for large cases and how
283 * much information is in a start byte for that length */
284 #define __COMMON_UNI_SKIP(uv) \
285 (UV) (uv) < (32 * (1U << ( UTF_ACCUMULATION_SHIFT))) ? 2 : \
286 (UV) (uv) < (16 * (1U << (2 * UTF_ACCUMULATION_SHIFT))) ? 3 : \
287 (UV) (uv) < ( 8 * (1U << (3 * UTF_ACCUMULATION_SHIFT))) ? 4 : \
288 (UV) (uv) < ( 4 * (1U << (4 * UTF_ACCUMULATION_SHIFT))) ? 5 : \
289 (UV) (uv) < ( 2 * (1U << (5 * UTF_ACCUMULATION_SHIFT))) ? 6 :
291 /* Internal macro to be used only in this file.
292 * This adds to __COMMON_UNI_SKIP the details at this platform's upper range.
293 * For any-sized EBCDIC platforms, or 64-bit ASCII ones, we need one more test
294 * to see if just 7 bytes is needed, or if the maximum is needed. For 32-bit
295 * ASCII platforms, everything is representable by 7 bytes */
296 #if defined(UV_IS_QUAD) || defined(EBCDIC)
297 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) \
298 (UV) (uv) < ((UV) 1U << (6 * UTF_ACCUMULATION_SHIFT)) ? 7 : UTF8_MAXBYTES)
300 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) 7)
303 /* The next two macros use the base macro defined above, and add in the tests
304 * at the low-end of the range, for just 1 byte, yielding complete macros,
305 * publicly accessible. */
307 /* Input is a true Unicode (not-native) code point */
308 #define OFFUNISKIP(uv) (OFFUNI_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
312 =for apidoc Am|STRLEN|UVCHR_SKIP|UV cp
313 returns the number of bytes required to represent the code point C<cp> when
314 encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than
315 255; a Unicode code point otherwise.
319 #define UVCHR_SKIP(uv) ( UVCHR_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
321 /* As explained in the comments for __COMMON_UNI_SKIP, 32 start bytes with
322 * UTF_ACCUMULATION_SHIFT bits of information each */
323 #define MAX_UTF8_TWO_BYTE (32 * (1U << UTF_ACCUMULATION_SHIFT) - 1)
325 /* constrained by EBCDIC which has 5 bits per continuation byte */
326 #define MAX_PORTABLE_UTF8_TWO_BYTE (32 * (1U << 5) - 1)
328 /* The maximum number of UTF-8 bytes a single Unicode character can
329 * uppercase/lowercase/fold into. Unicode guarantees that the maximum
330 * expansion is UTF8_MAX_FOLD_CHAR_EXPAND characters, but any above-Unicode
331 * code point will fold to itself, so we only have to look at the expansion of
332 * the maximum Unicode code point. But this number may be less than the space
333 * occupied by a very large code point under Perl's extended UTF-8. We have to
334 * make it large enough to fit any single character. (It turns out that ASCII
335 * and EBCDIC differ in which is larger) */
336 #define UTF8_MAXBYTES_CASE \
337 (UTF8_MAXBYTES >= (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)) \
339 : (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)))
341 /* Rest of these are attributes of Unicode and perl's internals rather than the
342 * encoding, or happen to be the same in both ASCII and EBCDIC (at least at
343 * this level; the macros that some of these call may have different
344 * definitions in the two encodings */
346 /* In domain restricted to ASCII, these may make more sense to the reader than
347 * the ones with Latin1 in the name */
348 #define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch)
349 #define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch)
351 /* More or less misleadingly-named defines, retained for back compat */
352 #define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch)
353 #define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch)
354 #define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
355 #define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
356 #define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch)
358 /* This defines the 1-bits that are to be in the first byte of a multi-byte
359 * UTF-8 encoded character that mark it as a start byte and give the number of
360 * bytes that comprise the character. 'len' is the number of bytes in the
361 * multi-byte sequence. */
362 #define UTF_START_MARK(len) (((len) > 7) ? 0xFF : (0xFF & (0xFE << (7-(len)))))
364 /* Masks out the initial one bits in a start byte, leaving the real data ones.
365 * Doesn't work on an invariant byte. 'len' is the number of bytes in the
366 * multi-byte sequence that comprises the character. */
367 #define UTF_START_MASK(len) (((len) >= 7) ? 0x00 : (0x1F >> ((len)-2)))
369 /* Adds a UTF8 continuation byte 'new' of information to a running total code
370 * point 'old' of all the continuation bytes so far. This is designed to be
371 * used in a loop to convert from UTF-8 to the code point represented. Note
372 * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is
373 * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC)
374 * code point in process of being generated */
375 #define UTF8_ACCUMULATE(old, new) (((old) << UTF_ACCUMULATION_SHIFT) \
376 | ((NATIVE_UTF8_TO_I8((U8)new)) \
377 & UTF_CONTINUATION_MASK))
379 /* If a value is anded with this, and the result is non-zero, then using the
380 * original value in UTF8_ACCUMULATE will overflow, shifting bits off the left
382 #define UTF_ACCUMULATION_OVERFLOW_MASK \
383 (((UV) UTF_CONTINUATION_MASK) << ((sizeof(UV) * CHARBITS) \
384 - UTF_ACCUMULATION_SHIFT))
386 /* This works in the face of malformed UTF-8. */
387 #define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) (UTF8_IS_DOWNGRADEABLE_START(*s) \
388 && ( (e) - (s) > 1) \
389 && UTF8_IS_CONTINUATION(*((s)+1)))
391 /* Number of bytes a code point occupies in UTF-8. */
392 #define NATIVE_SKIP(uv) UVCHR_SKIP(uv)
394 /* Most code which says UNISKIP is really thinking in terms of native code
395 * points (0-255) plus all those beyond. This is an imprecise term, but having
396 * it means existing code continues to work. For precision, use UVCHR_SKIP,
397 * NATIVE_SKIP, or OFFUNISKIP */
398 #define UNISKIP(uv) UVCHR_SKIP(uv)
400 /* Longer, but more accurate name */
401 #define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c)
403 /* Convert a UTF-8 variant Latin1 character to a native code point value.
404 * Needs just one iteration of accumulate. Should be used only if it is known
405 * that the code point is < 256, and is not UTF-8 invariant. Use the slower
406 * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point
407 * representable by two bytes (which turns out to be up through
408 * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are:
409 * HI: a downgradable start byte;
412 #define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \
413 ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \
414 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
415 LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \
416 NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO))))
418 /* Convert a two (not one) byte utf8 character to a native code point value.
419 * Needs just one iteration of accumulate. Should not be used unless it is
420 * known that the two bytes are legal: 1) two-byte start, and 2) continuation.
421 * Note that the result can be larger than 255 if the input character is not
423 #define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \
424 ( __ASSERT_(PL_utf8skip[HI] == 2) \
425 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
426 UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \
429 /* Should never be used, and be deprecated */
430 #define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO))
434 =for apidoc Am|STRLEN|UTF8SKIP|char* s
435 returns the number of bytes in the UTF-8 encoded character whose first (perhaps
436 only) byte is pointed to by C<s>.
440 #define UTF8SKIP(s) PL_utf8skip[*(const U8*)(s)]
441 #define UTF8_SKIP(s) UTF8SKIP(s)
443 /* Most code that says 'UNI_' really means the native value for code points up
445 #define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp)
447 /* Is the byte 'c' the same character when encoded in UTF-8 as when not. This
448 * works on both UTF-8 encoded strings and non-encoded, as it returns TRUE in
449 * each for the exact same set of bit patterns. It is valid on a subset of
450 * what UVCHR_IS_INVARIANT is valid on, so can just use that; and the compiler
451 * should optimize out anything extraneous given the implementation of the
453 #define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
455 /* Like the above, but its name implies a non-UTF8 input, which as the comments
456 * above show, doesn't matter as to its implementation */
457 #define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
459 /* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic
460 * bytes from an ordinal that is known to fit into exactly two (not one) bytes;
461 * it must be less than 0x3FF to work across both encodings. */
463 /* These two are helper macros for the other three sets, and should not be used
464 * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1
465 * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works
466 * for any code point */
467 #define __BASE_TWO_BYTE_HI(c, translate_function) \
468 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
469 I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \
470 | UTF_START_MARK(2)))
471 #define __BASE_TWO_BYTE_LO(c, translate_function) \
472 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
473 I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \
474 | UTF_CONTINUATION_MARK))
476 /* The next two macros should not be used. They were designed to be usable as
477 * the case label of a switch statement, but this doesn't work for EBCDIC. Use
478 * regen/unicode_constants.pl instead */
479 #define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)
480 #define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)
482 /* The next two macros are used when the source should be a single byte
483 * character; checked for under DEBUGGING */
484 #define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
485 ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1)))
486 #define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
487 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1)))
489 /* These final two macros in the series are used when the source can be any
490 * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient
491 * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~'
492 * operator instead of "<=" to avoid getting compiler warnings.
493 * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few
494 * places, so the ~ works */
495 #define UTF8_TWO_BYTE_HI(c) \
496 (__ASSERT_((sizeof(c) == 1) \
497 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
498 (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)))
499 #define UTF8_TWO_BYTE_LO(c) \
500 (__ASSERT_((sizeof(c) == 1) \
501 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
502 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)))
504 /* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII
505 * as it is only in overlongs. */
506 #define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1)
509 * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem
510 * from an earlier version of these macros in which they didn't call the
511 * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the
512 * beginning of a utf8 character. Now that foo_utf8() determines that itself,
513 * no need to do it again here
515 #define isIDFIRST_lazy_if(p,UTF) ((IN_BYTES || !UTF) \
517 : isIDFIRST_utf8((const U8*)p))
518 #define isWORDCHAR_lazy_if(p,UTF) ((IN_BYTES || (!UTF)) \
520 : isWORDCHAR_utf8((const U8*)p))
521 #define isALNUM_lazy_if(p,UTF) isWORDCHAR_lazy_if(p,UTF)
523 #define UTF8_MAXLEN UTF8_MAXBYTES
525 /* A Unicode character can fold to up to 3 characters */
526 #define UTF8_MAX_FOLD_CHAR_EXPAND 3
528 #define IN_BYTES (CopHINTS_get(PL_curcop) & HINT_BYTES)
532 =for apidoc Am|bool|DO_UTF8|SV* sv
533 Returns a bool giving whether or not the PV in C<sv> is to be treated as being
536 You should use this I<after> a call to C<SvPV()> or one of its variants, in
537 case any call to string overloading updates the internal UTF-8 encoding flag.
541 #define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES)
543 /* Should all strings be treated as Unicode, and not just UTF-8 encoded ones?
544 * Is so within 'feature unicode_strings' or 'locale :not_characters', and not
545 * within 'use bytes'. UTF-8 locales are not tested for here, but perhaps
547 #define IN_UNI_8_BIT \
548 (((CopHINTS_get(PL_curcop) & (HINT_UNI_8_BIT)) \
549 || (CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \
550 /* -1 below is for :not_characters */ \
551 && _is_in_locale_category(FALSE, -1))) \
555 #define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */
557 /* Allow first byte to be a continuation byte */
558 #define UTF8_ALLOW_CONTINUATION 0x0002
560 /* Allow second... bytes to be non-continuation bytes */
561 #define UTF8_ALLOW_NON_CONTINUATION 0x0004
563 /* expecting more bytes than were available in the string */
564 #define UTF8_ALLOW_SHORT 0x0008
566 /* Overlong sequence; i.e., the code point can be specified in fewer bytes. */
567 #define UTF8_ALLOW_LONG 0x0010
569 #define UTF8_DISALLOW_SURROGATE 0x0020 /* Unicode surrogates */
570 #define UTF8_WARN_SURROGATE 0x0040
572 #define UTF8_DISALLOW_NONCHAR 0x0080 /* Unicode non-character */
573 #define UTF8_WARN_NONCHAR 0x0100 /* code points */
575 #define UTF8_DISALLOW_SUPER 0x0200 /* Super-set of Unicode: code */
576 #define UTF8_WARN_SUPER 0x0400 /* points above the legal max */
578 /* Code points which never were part of the original UTF-8 standard, which only
579 * went up to 2 ** 31 - 1. Note that these all overflow a signed 32-bit word,
580 * The first byte of these code points is FE or FF on ASCII platforms. If the
581 * first byte is FF, it will overflow a 32-bit word. */
582 #define UTF8_DISALLOW_ABOVE_31_BIT 0x0800
583 #define UTF8_WARN_ABOVE_31_BIT 0x1000
585 /* For back compat, these old names are misleading for UTF_EBCDIC */
586 #define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_ABOVE_31_BIT
587 #define UTF8_WARN_FE_FF UTF8_WARN_ABOVE_31_BIT
589 #define UTF8_CHECK_ONLY 0x2000
591 /* For backwards source compatibility. They do nothing, as the default now
592 * includes what they used to mean. The first one's meaning was to allow the
593 * just the single non-character 0xFFFF */
594 #define UTF8_ALLOW_FFFF 0
595 #define UTF8_ALLOW_SURROGATE 0
597 #define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \
598 ( UTF8_DISALLOW_SUPER|UTF8_DISALLOW_NONCHAR \
599 |UTF8_DISALLOW_SURROGATE)
600 #define UTF8_WARN_ILLEGAL_INTERCHANGE \
601 (UTF8_WARN_SUPER|UTF8_WARN_NONCHAR|UTF8_WARN_SURROGATE)
602 #define UTF8_ALLOW_ANY \
603 (~( UTF8_DISALLOW_ILLEGAL_INTERCHANGE|UTF8_DISALLOW_ABOVE_31_BIT \
604 |UTF8_WARN_ILLEGAL_INTERCHANGE|UTF8_WARN_ABOVE_31_BIT))
605 #define UTF8_ALLOW_ANYUV \
607 & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE|UTF8_WARN_ILLEGAL_INTERCHANGE))
608 #define UTF8_ALLOW_DEFAULT (ckWARN(WARN_UTF8) ? 0 : \
611 /* Several of the macros below have a second parameter that is currently
612 * unused; but could be used in the future to make sure that the input is
615 #define UTF8_IS_SURROGATE(s, e) cBOOL(is_SURROGATE_utf8(s))
616 #define UTF8_IS_REPLACEMENT(s, send) cBOOL(is_REPLACEMENT_utf8_safe(s,send))
619 * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode
620 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0
621 * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1
623 * BE AWARE that this test doesn't rule out malformed code points, in
624 * particular overlongs */
625 #ifdef EBCDIC /* Both versions assume well-formed UTF8 */
626 # define UTF8_IS_SUPER(s, e) (NATIVE_UTF8_TO_I8(* (U8*) (s)) >= 0xF9 \
627 && (NATIVE_UTF8_TO_I8(* (U8*) (s)) > 0xF9 \
628 || (NATIVE_UTF8_TO_I8(* ((U8*) (s) + 1)) >= 0xA2)))
630 # define UTF8_IS_SUPER(s, e) (*(U8*) (s) >= 0xF4 \
631 && (*(U8*) (s) > 0xF4 || (*((U8*) (s) + 1) >= 0x90)))
634 /* These are now machine generated, and the 'given' clause is no longer
636 #define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \
637 cBOOL(is_NONCHAR_utf8(s))
638 #define UTF8_IS_NONCHAR(s, e) \
639 UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e)
641 #define UNICODE_SURROGATE_FIRST 0xD800
642 #define UNICODE_SURROGATE_LAST 0xDFFF
643 #define UNICODE_REPLACEMENT 0xFFFD
644 #define UNICODE_BYTE_ORDER_MARK 0xFEFF
646 /* Though our UTF-8 encoding can go beyond this,
647 * let's be conservative and do as Unicode says. */
648 #define PERL_UNICODE_MAX 0x10FFFF
650 #define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */
651 #define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */
652 #define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */
653 #define UNICODE_WARN_ABOVE_31_BIT 0x0008 /* Above 0x7FFF_FFFF */
654 #define UNICODE_DISALLOW_SURROGATE 0x0010
655 #define UNICODE_DISALLOW_NONCHAR 0x0020
656 #define UNICODE_DISALLOW_SUPER 0x0040
657 #define UNICODE_DISALLOW_ABOVE_31_BIT 0x0080
658 #define UNICODE_WARN_ILLEGAL_INTERCHANGE \
659 (UNICODE_WARN_SURROGATE|UNICODE_WARN_NONCHAR|UNICODE_WARN_SUPER)
660 #define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \
661 (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_NONCHAR|UNICODE_DISALLOW_SUPER)
663 /* For backward source compatibility, as are now the default */
664 #define UNICODE_ALLOW_SURROGATE 0
665 #define UNICODE_ALLOW_SUPER 0
666 #define UNICODE_ALLOW_ANY 0
668 /* This matches the 2048 code points between UNICODE_SURROGATE_FIRST (0xD800) and
669 * UNICODE_SURROGATE_LAST (0xDFFF) */
670 #define UNICODE_IS_SURROGATE(uv) (((UV) (uv) & (~0xFFFF | 0xF800)) \
673 #define UNICODE_IS_REPLACEMENT(uv) ((UV) (uv) == UNICODE_REPLACEMENT)
674 #define UNICODE_IS_BYTE_ORDER_MARK(uv) ((UV) (uv) == UNICODE_BYTE_ORDER_MARK)
676 /* Is 'uv' one of the 32 contiguous-range noncharacters? */
677 #define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) ((UV) (uv) >= 0xFDD0 \
678 && (UV) (uv) <= 0xFDEF)
680 /* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE,
681 * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above
682 * the Unicode legal max */
683 #define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \
684 (((UV) (uv) & 0xFFFE) == 0xFFFE)
686 #define UNICODE_IS_NONCHAR(uv) \
687 ( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \
688 || ( LIKELY( ! UNICODE_IS_SUPER(uv)) \
689 && UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
691 #define UNICODE_IS_SUPER(uv) ((UV) (uv) > PERL_UNICODE_MAX)
692 #define UNICODE_IS_ABOVE_31_BIT(uv) ((UV) (uv) > 0x7FFFFFFF)
694 #define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE
695 #define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \
696 LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE
697 #define MICRO_SIGN MICRO_SIGN_NATIVE
698 #define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \
699 LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE
700 #define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \
701 LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE
702 #define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3
703 #define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2
704 #define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3
705 #define GREEK_SMALL_LETTER_MU 0x03BC
706 #define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case
708 #define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */
709 #ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
710 # define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E
712 #define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130
713 #define LATIN_SMALL_LETTER_DOTLESS_I 0x131
714 #define LATIN_SMALL_LETTER_LONG_S 0x017F
715 #define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05
716 #define LATIN_SMALL_LIGATURE_ST 0xFB06
717 #define KELVIN_SIGN 0x212A
718 #define ANGSTROM_SIGN 0x212B
720 #define UNI_DISPLAY_ISPRINT 0x0001
721 #define UNI_DISPLAY_BACKSLASH 0x0002
722 #define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
723 #define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
725 #define ANYOF_FOLD_SHARP_S(node, input, end) \
726 (ANYOF_BITMAP_TEST(node, LATIN_SMALL_LETTER_SHARP_S) && \
727 (ANYOF_NONBITMAP(node)) && \
728 (ANYOF_FLAGS(node) & ANYOF_LOC_NONBITMAP_FOLD) && \
729 ((end) > (input) + 1) && \
730 isALPHA_FOLD_EQ((input)[0], 's'))
732 #define SHARP_S_SKIP 2
734 /* If you want to exclude surrogates, and beyond legal Unicode, see the blame
735 * log for earlier versions which gave details for these */
737 /* A helper macro for isUTF8_CHAR, so use that one, and not this one. This is
738 * retained solely for backwards compatibility and may be deprecated and
739 * removed in a future Perl version.
741 * regen/regcharclass.pl generates is_UTF8_CHAR_utf8() macros for up to these
742 * number of bytes. So this has to be coordinated with that file */
744 # define IS_UTF8_CHAR_FAST(n) ((n) <= 3)
746 # define IS_UTF8_CHAR_FAST(n) ((n) <= 4)
750 /* A helper macro for isUTF8_CHAR, so use that one instead of this. This was
751 * generated by regen/regcharclass.pl, and then moved here. The lines that
752 * generated it were then commented out. This was done solely because it takes
753 * on the order of 10 minutes to generate, and is never going to change, unless
754 * the generated code is improved.
756 * The EBCDIC versions have been cut to not cover all of legal Unicode,
757 * otherwise they take too long to generate; besides there is a separate one
758 * for each code page, so they are in regcharclass.h instead of here */
760 UTF8_CHAR: Matches legal UTF-8 encoded characters from 2 through 4 bytes
764 /*** GENERATED CODE ***/
765 #define is_UTF8_CHAR_utf8_no_length_checks(s) \
766 ( ( 0xC2 <= ((U8*)s)[0] && ((U8*)s)[0] <= 0xDF ) ? \
767 ( ( ( ((U8*)s)[1] & 0xC0 ) == 0x80 ) ? 2 : 0 ) \
768 : ( 0xE0 == ((U8*)s)[0] ) ? \
769 ( ( ( ( ((U8*)s)[1] & 0xE0 ) == 0xA0 ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
770 : ( 0xE1 <= ((U8*)s)[0] && ((U8*)s)[0] <= 0xEF ) ? \
771 ( ( ( ( ((U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
772 : ( 0xF0 == ((U8*)s)[0] ) ? \
773 ( ( ( ( 0x90 <= ((U8*)s)[1] && ((U8*)s)[1] <= 0xBF ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
774 : ( ( ( ( 0xF1 <= ((U8*)s)[0] && ((U8*)s)[0] <= 0xF7 ) && ( ( ((U8*)s)[1] & 0xC0 ) == 0x80 ) ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )
779 =for apidoc Am|STRLEN|isUTF8_CHAR|const U8 *s|const U8 *e
781 Returns the number of bytes beginning at C<s> which form a legal UTF-8 (or
782 UTF-EBCDIC) encoded character, looking no further than S<C<e - s>> bytes into
783 C<s>. Returns 0 if the sequence starting at C<s> through S<C<e - 1>> is not
786 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
787 machines) is a valid UTF-8 character.
792 #define isUTF8_CHAR(s, e) (UNLIKELY((e) <= (s)) \
794 : (UTF8_IS_INVARIANT(*s)) \
796 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
798 : LIKELY(IS_UTF8_CHAR_FAST(UTF8SKIP(s))) \
799 ? is_UTF8_CHAR_utf8_no_length_checks(s) \
800 : _is_utf8_char_slow(s, e))
802 #define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end)
804 /* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is
805 * retained solely for backwards compatibility */
806 #define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n)
811 * ex: set ts=8 sts=4 sw=4 et: