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 #if defined(_MSC_VER) && _MSC_VER < 1400
138 /* older MSVC versions have a smallish macro buffer */
139 #define PERL_SMALL_MACRO_BUFFER
142 /* Native character to/from iso-8859-1. Are the identity functions on ASCII
144 #ifdef PERL_SMALL_MACRO_BUFFER
145 #define NATIVE_TO_LATIN1(ch) ((U8)(ch))
146 #define LATIN1_TO_NATIVE(ch) ((U8)(ch))
148 #define NATIVE_TO_LATIN1(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
149 #define LATIN1_TO_NATIVE(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
152 /* I8 is an intermediate version of UTF-8 used only in UTF-EBCDIC. We thus
153 * consider it to be identical to UTF-8 on ASCII platforms. Strictly speaking
154 * UTF-8 and UTF-EBCDIC are two different things, but we often conflate them
155 * because they are 8-bit encodings that serve the same purpose in Perl, and
156 * rarely do we need to distinguish them. The term "NATIVE_UTF8" applies to
157 * whichever one is applicable on the current platform */
158 #ifdef PERL_SMALL_MACRO_BUFFER
159 #define NATIVE_UTF8_TO_I8(ch) (ch)
160 #define I8_TO_NATIVE_UTF8(ch) (ch)
162 #define NATIVE_UTF8_TO_I8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
163 #define I8_TO_NATIVE_UTF8(ch) (__ASSERT_(FITS_IN_8_BITS(ch)) ((U8) (ch)))
166 /* Transforms in wide UV chars */
167 #define UNI_TO_NATIVE(ch) ((UV) (ch))
168 #define NATIVE_TO_UNI(ch) ((UV) (ch))
172 The following table is from Unicode 3.2.
174 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
176 U+0000..U+007F 00..7F
177 U+0080..U+07FF * C2..DF 80..BF
178 U+0800..U+0FFF E0 * A0..BF 80..BF
179 U+1000..U+CFFF E1..EC 80..BF 80..BF
180 U+D000..U+D7FF ED 80..9F 80..BF
181 U+D800..U+DFFF ED A0..BF 80..BF (surrogates)
182 U+E000..U+FFFF EE..EF 80..BF 80..BF
183 U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF
184 U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF
185 U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
186 Below are non-Unicode code points
187 U+110000..U+13FFFF F4 90..BF 80..BF 80..BF
188 U+110000..U+1FFFFF F5..F7 80..BF 80..BF 80..BF
189 U+200000..: F8.. * 88..BF 80..BF 80..BF 80..BF
191 Note the gaps before several of the byte entries above marked by '*'. These are
192 caused by legal UTF-8 avoiding non-shortest encodings: it is technically
193 possible to UTF-8-encode a single code point in different ways, but that is
194 explicitly forbidden, and the shortest possible encoding should always be used
195 (and that is what Perl does). The non-shortest ones are called 'overlongs'.
200 Another way to look at it, as bits:
202 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
205 0000 0bbb bbaa aaaa 110b bbbb 10aa aaaa
206 cccc bbbb bbaa aaaa 1110 cccc 10bb bbbb 10aa aaaa
207 00 000d ddcc cccc bbbb bbaa aaaa 1111 0ddd 10cc cccc 10bb bbbb 10aa aaaa
209 As you can see, the continuation bytes all begin with C<10>, and the
210 leading bits of the start byte tell how many bytes there are in the
213 Perl's extended UTF-8 means we can have start bytes up to FF.
217 /* Is the representation of the Unicode code point 'cp' the same regardless of
218 * being encoded in UTF-8 or not? */
219 #define OFFUNI_IS_INVARIANT(cp) isASCII(cp)
221 /* Is the representation of the code point 'cp' the same regardless of
222 * being encoded in UTF-8 or not? 'cp' is native if < 256; Unicode otherwise
224 #define UVCHR_IS_INVARIANT(cp) OFFUNI_IS_INVARIANT(cp)
226 /* This defines the bits that are to be in the continuation bytes of a multi-byte
227 * UTF-8 encoded character that mark it is a continuation byte. */
228 #define UTF_CONTINUATION_MARK 0x80
230 /* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence
231 * in UTF-8? This is the inverse of UTF8_IS_INVARIANT. The |0 makes sure this
232 * isn't mistakenly called with a ptr argument */
233 #define UTF8_IS_CONTINUED(c) (((U8)((c) | 0)) & UTF_CONTINUATION_MARK)
235 /* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence?
236 * This doesn't catch invariants (they are single-byte). It also excludes the
237 * illegal overlong sequences that begin with C0 and C1. The |0 makes sure
238 * this isn't mistakenly called with a ptr argument */
239 #define UTF8_IS_START(c) (((U8)((c) | 0)) >= 0xc2)
241 /* For use in UTF8_IS_CONTINUATION() below */
242 #define UTF_IS_CONTINUATION_MASK 0xC0
244 /* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the
245 * first byte thereof? The |0 makes sure this isn't mistakenly called with a
247 #define UTF8_IS_CONTINUATION(c) \
248 ((((U8)((c) | 0)) & UTF_IS_CONTINUATION_MASK) == UTF_CONTINUATION_MARK)
250 /* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use
251 * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to
252 * be well-formed. Masking with 0xfe allows the low bit to be 0 or 1; thus
253 * this matches 0xc[23]. The |0 makes sure this isn't mistakenly called with a
255 #define UTF8_IS_DOWNGRADEABLE_START(c) ((((U8)((c) | 0)) & 0xfe) == 0xc2)
257 /* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that
258 * represent a code point > 255? The |0 makes sure this isn't mistakenly
259 * called with a ptr argument */
260 #define UTF8_IS_ABOVE_LATIN1(c) (((U8)((c) | 0)) >= 0xc4)
262 /* This is the number of low-order bits a continuation byte in a UTF-8 encoded
263 * sequence contributes to the specification of the code point. In the bit
264 * maps above, you see that the first 2 bits are a constant '10', leaving 6 of
265 * real information */
266 #define UTF_ACCUMULATION_SHIFT 6
268 /* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL()
270 #define QUESTION_MARK_CTRL DEL_NATIVE
272 /* Surrogates, non-character code points and above-Unicode code points are
273 * problematic in some contexts. This allows code that needs to check for
274 * those to to quickly exclude the vast majority of code points it will
276 #define isUTF8_POSSIBLY_PROBLEMATIC(c) ((U8) c >= 0xED)
278 #endif /* EBCDIC vs ASCII */
280 /* 2**UTF_ACCUMULATION_SHIFT - 1 */
281 #define UTF_CONTINUATION_MASK ((U8) ((1U << UTF_ACCUMULATION_SHIFT) - 1))
283 /* Internal macro to be used only in this file to aid in constructing other
284 * publicly accessible macros.
285 * The number of bytes required to express this uv in UTF-8, for just those
286 * uv's requiring 2 through 6 bytes, as these are common to all platforms and
287 * word sizes. The number of bytes needed is given by the number of leading 1
288 * bits in the start byte. There are 32 start bytes that have 2 initial 1 bits
289 * (C0-DF); there are 16 that have 3 initial 1 bits (E0-EF); 8 that have 4
290 * initial 1 bits (F0-F8); 4 that have 5 initial 1 bits (F9-FB), and 2 that
291 * have 6 initial 1 bits (FC-FD). The largest number a string of n bytes can
292 * represent is (the number of possible start bytes for 'n')
293 * * (the number of possiblities for each start byte
294 * The latter in turn is
295 * 2 ** ( (how many continuation bytes there are)
296 * * (the number of bits of information each
297 * continuation byte holds))
299 * If we were on a platform where we could use a fast find first set bit
300 * instruction (or count leading zeros instruction) this could be replaced by
301 * using that to find the log2 of the uv, and divide that by the number of bits
302 * of information in each continuation byte, adjusting for large cases and how
303 * much information is in a start byte for that length */
304 #define __COMMON_UNI_SKIP(uv) \
305 (UV) (uv) < (32 * (1U << ( UTF_ACCUMULATION_SHIFT))) ? 2 : \
306 (UV) (uv) < (16 * (1U << (2 * UTF_ACCUMULATION_SHIFT))) ? 3 : \
307 (UV) (uv) < ( 8 * (1U << (3 * UTF_ACCUMULATION_SHIFT))) ? 4 : \
308 (UV) (uv) < ( 4 * (1U << (4 * UTF_ACCUMULATION_SHIFT))) ? 5 : \
309 (UV) (uv) < ( 2 * (1U << (5 * UTF_ACCUMULATION_SHIFT))) ? 6 :
311 /* Internal macro to be used only in this file.
312 * This adds to __COMMON_UNI_SKIP the details at this platform's upper range.
313 * For any-sized EBCDIC platforms, or 64-bit ASCII ones, we need one more test
314 * to see if just 7 bytes is needed, or if the maximum is needed. For 32-bit
315 * ASCII platforms, everything is representable by 7 bytes */
316 #if defined(UV_IS_QUAD) || defined(EBCDIC)
317 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) \
318 (UV) (uv) < ((UV) 1U << (6 * UTF_ACCUMULATION_SHIFT)) ? 7 : UTF8_MAXBYTES)
320 # define __BASE_UNI_SKIP(uv) (__COMMON_UNI_SKIP(uv) 7)
323 /* The next two macros use the base macro defined above, and add in the tests
324 * at the low-end of the range, for just 1 byte, yielding complete macros,
325 * publicly accessible. */
327 /* Input is a true Unicode (not-native) code point */
328 #define OFFUNISKIP(uv) (OFFUNI_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
332 =for apidoc Am|STRLEN|UVCHR_SKIP|UV cp
333 returns the number of bytes required to represent the code point C<cp> when
334 encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than
335 255; a Unicode code point otherwise.
339 #define UVCHR_SKIP(uv) ( UVCHR_IS_INVARIANT(uv) ? 1 : __BASE_UNI_SKIP(uv))
341 /* As explained in the comments for __COMMON_UNI_SKIP, 32 start bytes with
342 * UTF_ACCUMULATION_SHIFT bits of information each */
343 #define MAX_UTF8_TWO_BYTE (32 * (1U << UTF_ACCUMULATION_SHIFT) - 1)
345 /* constrained by EBCDIC which has 5 bits per continuation byte */
346 #define MAX_PORTABLE_UTF8_TWO_BYTE (32 * (1U << 5) - 1)
348 /* The maximum number of UTF-8 bytes a single Unicode character can
349 * uppercase/lowercase/fold into. Unicode guarantees that the maximum
350 * expansion is UTF8_MAX_FOLD_CHAR_EXPAND characters, but any above-Unicode
351 * code point will fold to itself, so we only have to look at the expansion of
352 * the maximum Unicode code point. But this number may be less than the space
353 * occupied by a very large code point under Perl's extended UTF-8. We have to
354 * make it large enough to fit any single character. (It turns out that ASCII
355 * and EBCDIC differ in which is larger) */
356 #define UTF8_MAXBYTES_CASE \
357 (UTF8_MAXBYTES >= (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)) \
359 : (UTF8_MAX_FOLD_CHAR_EXPAND * OFFUNISKIP(0x10FFFF)))
361 /* Rest of these are attributes of Unicode and perl's internals rather than the
362 * encoding, or happen to be the same in both ASCII and EBCDIC (at least at
363 * this level; the macros that some of these call may have different
364 * definitions in the two encodings */
366 /* In domain restricted to ASCII, these may make more sense to the reader than
367 * the ones with Latin1 in the name */
368 #define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch)
369 #define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch)
371 /* More or less misleadingly-named defines, retained for back compat */
372 #define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch)
373 #define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch)
374 #define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
375 #define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
376 #define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch)
378 /* This defines the 1-bits that are to be in the first byte of a multi-byte
379 * UTF-8 encoded character that mark it as a start byte and give the number of
380 * bytes that comprise the character. 'len' is the number of bytes in the
381 * multi-byte sequence. */
382 #define UTF_START_MARK(len) (((len) > 7) ? 0xFF : (0xFF & (0xFE << (7-(len)))))
384 /* Masks out the initial one bits in a start byte, leaving the real data ones.
385 * Doesn't work on an invariant byte. 'len' is the number of bytes in the
386 * multi-byte sequence that comprises the character. */
387 #define UTF_START_MASK(len) (((len) >= 7) ? 0x00 : (0x1F >> ((len)-2)))
389 /* Adds a UTF8 continuation byte 'new' of information to a running total code
390 * point 'old' of all the continuation bytes so far. This is designed to be
391 * used in a loop to convert from UTF-8 to the code point represented. Note
392 * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is
393 * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC)
394 * code point in process of being generated */
395 #define UTF8_ACCUMULATE(old, new) (((old) << UTF_ACCUMULATION_SHIFT) \
396 | ((NATIVE_UTF8_TO_I8((U8)new)) \
397 & UTF_CONTINUATION_MASK))
399 /* If a value is anded with this, and the result is non-zero, then using the
400 * original value in UTF8_ACCUMULATE will overflow, shifting bits off the left
402 #define UTF_ACCUMULATION_OVERFLOW_MASK \
403 (((UV) UTF_CONTINUATION_MASK) << ((sizeof(UV) * CHARBITS) \
404 - UTF_ACCUMULATION_SHIFT))
406 /* This works in the face of malformed UTF-8. */
407 #define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) (UTF8_IS_DOWNGRADEABLE_START(*s) \
408 && ( (e) - (s) > 1) \
409 && UTF8_IS_CONTINUATION(*((s)+1)))
411 /* Number of bytes a code point occupies in UTF-8. */
412 #define NATIVE_SKIP(uv) UVCHR_SKIP(uv)
414 /* Most code which says UNISKIP is really thinking in terms of native code
415 * points (0-255) plus all those beyond. This is an imprecise term, but having
416 * it means existing code continues to work. For precision, use UVCHR_SKIP,
417 * NATIVE_SKIP, or OFFUNISKIP */
418 #define UNISKIP(uv) UVCHR_SKIP(uv)
420 /* Longer, but more accurate name */
421 #define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c)
423 /* Convert a UTF-8 variant Latin1 character to a native code point value.
424 * Needs just one iteration of accumulate. Should be used only if it is known
425 * that the code point is < 256, and is not UTF-8 invariant. Use the slower
426 * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point
427 * representable by two bytes (which turns out to be up through
428 * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are:
429 * HI: a downgradable start byte;
432 #define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \
433 ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \
434 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
435 LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \
436 NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO))))
438 /* Convert a two (not one) byte utf8 character to a native code point value.
439 * Needs just one iteration of accumulate. Should not be used unless it is
440 * known that the two bytes are legal: 1) two-byte start, and 2) continuation.
441 * Note that the result can be larger than 255 if the input character is not
443 #define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \
444 ( __ASSERT_(PL_utf8skip[HI] == 2) \
445 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
446 UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \
449 /* Should never be used, and be deprecated */
450 #define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO))
454 =for apidoc Am|STRLEN|UTF8SKIP|char* s
455 returns the number of bytes in the UTF-8 encoded character whose first (perhaps
456 only) byte is pointed to by C<s>.
460 #define UTF8SKIP(s) PL_utf8skip[*(const U8*)(s)]
461 #define UTF8_SKIP(s) UTF8SKIP(s)
463 /* Most code that says 'UNI_' really means the native value for code points up
465 #define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp)
467 /* Is the byte 'c' the same character when encoded in UTF-8 as when not. This
468 * works on both UTF-8 encoded strings and non-encoded, as it returns TRUE in
469 * each for the exact same set of bit patterns. It is valid on a subset of
470 * what UVCHR_IS_INVARIANT is valid on, so can just use that; and the compiler
471 * should optimize out anything extraneous given the implementation of the
472 * latter. The |0 makes sure this isn't mistakenly called with a ptr argument.
474 #define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT((c) | 0)
476 /* Like the above, but its name implies a non-UTF8 input, which as the comments
477 * above show, doesn't matter as to its implementation */
478 #define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
480 /* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic
481 * bytes from an ordinal that is known to fit into exactly two (not one) bytes;
482 * it must be less than 0x3FF to work across both encodings. */
484 /* These two are helper macros for the other three sets, and should not be used
485 * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1
486 * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works
487 * for any code point */
488 #define __BASE_TWO_BYTE_HI(c, translate_function) \
489 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
490 I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \
491 | UTF_START_MARK(2)))
492 #define __BASE_TWO_BYTE_LO(c, translate_function) \
493 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
494 I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \
495 | UTF_CONTINUATION_MARK))
497 /* The next two macros should not be used. They were designed to be usable as
498 * the case label of a switch statement, but this doesn't work for EBCDIC. Use
499 * regen/unicode_constants.pl instead */
500 #define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)
501 #define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)
503 /* The next two macros are used when the source should be a single byte
504 * character; checked for under DEBUGGING */
505 #define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
506 ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1)))
507 #define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
508 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1)))
510 /* These final two macros in the series are used when the source can be any
511 * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient
512 * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~'
513 * operator instead of "<=" to avoid getting compiler warnings.
514 * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few
515 * places, so the ~ works */
516 #define UTF8_TWO_BYTE_HI(c) \
517 (__ASSERT_((sizeof(c) == 1) \
518 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
519 (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)))
520 #define UTF8_TWO_BYTE_LO(c) \
521 (__ASSERT_((sizeof(c) == 1) \
522 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
523 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)))
525 /* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII
526 * as it is only in overlongs. */
527 #define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1)
530 * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem
531 * from an earlier version of these macros in which they didn't call the
532 * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the
533 * beginning of a utf8 character. Now that foo_utf8() determines that itself,
534 * no need to do it again here
536 #define isIDFIRST_lazy_if(p,UTF) ((IN_BYTES || !UTF) \
538 : isIDFIRST_utf8((const U8*)p))
539 #define isWORDCHAR_lazy_if(p,UTF) ((IN_BYTES || (!UTF)) \
541 : isWORDCHAR_utf8((const U8*)p))
542 #define isALNUM_lazy_if(p,UTF) isWORDCHAR_lazy_if(p,UTF)
544 #define UTF8_MAXLEN UTF8_MAXBYTES
546 /* A Unicode character can fold to up to 3 characters */
547 #define UTF8_MAX_FOLD_CHAR_EXPAND 3
549 #define IN_BYTES (CopHINTS_get(PL_curcop) & HINT_BYTES)
553 =for apidoc Am|bool|DO_UTF8|SV* sv
554 Returns a bool giving whether or not the PV in C<sv> is to be treated as being
557 You should use this I<after> a call to C<SvPV()> or one of its variants, in
558 case any call to string overloading updates the internal UTF-8 encoding flag.
562 #define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES)
564 /* Should all strings be treated as Unicode, and not just UTF-8 encoded ones?
565 * Is so within 'feature unicode_strings' or 'locale :not_characters', and not
566 * within 'use bytes'. UTF-8 locales are not tested for here, but perhaps
568 #define IN_UNI_8_BIT \
569 (((CopHINTS_get(PL_curcop) & (HINT_UNI_8_BIT)) \
570 || (CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \
571 /* -1 below is for :not_characters */ \
572 && _is_in_locale_category(FALSE, -1))) \
576 #define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */
578 /* Allow first byte to be a continuation byte */
579 #define UTF8_ALLOW_CONTINUATION 0x0002
581 /* Allow second... bytes to be non-continuation bytes */
582 #define UTF8_ALLOW_NON_CONTINUATION 0x0004
584 /* expecting more bytes than were available in the string */
585 #define UTF8_ALLOW_SHORT 0x0008
587 /* Overlong sequence; i.e., the code point can be specified in fewer bytes. */
588 #define UTF8_ALLOW_LONG 0x0010
590 #define UTF8_DISALLOW_SURROGATE 0x0020 /* Unicode surrogates */
591 #define UTF8_WARN_SURROGATE 0x0040
593 #define UTF8_DISALLOW_NONCHAR 0x0080 /* Unicode non-character */
594 #define UTF8_WARN_NONCHAR 0x0100 /* code points */
596 #define UTF8_DISALLOW_SUPER 0x0200 /* Super-set of Unicode: code */
597 #define UTF8_WARN_SUPER 0x0400 /* points above the legal max */
599 /* Code points which never were part of the original UTF-8 standard, which only
600 * went up to 2 ** 31 - 1. Note that these all overflow a signed 32-bit word,
601 * The first byte of these code points is FE or FF on ASCII platforms. If the
602 * first byte is FF, it will overflow a 32-bit word. */
603 #define UTF8_DISALLOW_ABOVE_31_BIT 0x0800
604 #define UTF8_WARN_ABOVE_31_BIT 0x1000
606 /* For back compat, these old names are misleading for UTF_EBCDIC */
607 #define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_ABOVE_31_BIT
608 #define UTF8_WARN_FE_FF UTF8_WARN_ABOVE_31_BIT
610 #define UTF8_CHECK_ONLY 0x2000
612 /* For backwards source compatibility. They do nothing, as the default now
613 * includes what they used to mean. The first one's meaning was to allow the
614 * just the single non-character 0xFFFF */
615 #define UTF8_ALLOW_FFFF 0
616 #define UTF8_ALLOW_SURROGATE 0
618 #define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \
619 ( UTF8_DISALLOW_SUPER|UTF8_DISALLOW_NONCHAR \
620 |UTF8_DISALLOW_SURROGATE)
621 #define UTF8_WARN_ILLEGAL_INTERCHANGE \
622 (UTF8_WARN_SUPER|UTF8_WARN_NONCHAR|UTF8_WARN_SURROGATE)
623 #define UTF8_ALLOW_ANY \
624 (~( UTF8_DISALLOW_ILLEGAL_INTERCHANGE|UTF8_DISALLOW_ABOVE_31_BIT \
625 |UTF8_WARN_ILLEGAL_INTERCHANGE|UTF8_WARN_ABOVE_31_BIT))
626 #define UTF8_ALLOW_ANYUV \
628 & ~(UTF8_DISALLOW_ILLEGAL_INTERCHANGE|UTF8_WARN_ILLEGAL_INTERCHANGE))
629 #define UTF8_ALLOW_DEFAULT (ckWARN(WARN_UTF8) ? 0 : \
632 /* Several of the macros below have a second parameter that is currently
633 * unused; but could be used in the future to make sure that the input is
636 #define UTF8_IS_SURROGATE(s, e) cBOOL(is_SURROGATE_utf8(s))
637 #define UTF8_IS_REPLACEMENT(s, send) cBOOL(is_REPLACEMENT_utf8_safe(s,send))
640 * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode
641 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0
642 * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1
644 * BE AWARE that this test doesn't rule out malformed code points, in
645 * particular overlongs */
646 #ifdef EBCDIC /* Both versions assume well-formed UTF8 */
647 # define UTF8_IS_SUPER(s, e) (NATIVE_UTF8_TO_I8(* (U8*) (s)) >= 0xF9 \
648 && (NATIVE_UTF8_TO_I8(* (U8*) (s)) > 0xF9 \
649 || (NATIVE_UTF8_TO_I8(* ((U8*) (s) + 1)) >= 0xA2)))
651 # define UTF8_IS_SUPER(s, e) (*(U8*) (s) >= 0xF4 \
652 && (*(U8*) (s) > 0xF4 || (*((U8*) (s) + 1) >= 0x90)))
655 /* These are now machine generated, and the 'given' clause is no longer
657 #define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \
658 cBOOL(is_NONCHAR_utf8(s))
659 #define UTF8_IS_NONCHAR(s, e) \
660 UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e)
662 #define UNICODE_SURROGATE_FIRST 0xD800
663 #define UNICODE_SURROGATE_LAST 0xDFFF
664 #define UNICODE_REPLACEMENT 0xFFFD
665 #define UNICODE_BYTE_ORDER_MARK 0xFEFF
667 /* Though our UTF-8 encoding can go beyond this,
668 * let's be conservative and do as Unicode says. */
669 #define PERL_UNICODE_MAX 0x10FFFF
671 #define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */
672 #define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */
673 #define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */
674 #define UNICODE_WARN_ABOVE_31_BIT 0x0008 /* Above 0x7FFF_FFFF */
675 #define UNICODE_DISALLOW_SURROGATE 0x0010
676 #define UNICODE_DISALLOW_NONCHAR 0x0020
677 #define UNICODE_DISALLOW_SUPER 0x0040
678 #define UNICODE_DISALLOW_ABOVE_31_BIT 0x0080
679 #define UNICODE_WARN_ILLEGAL_INTERCHANGE \
680 (UNICODE_WARN_SURROGATE|UNICODE_WARN_NONCHAR|UNICODE_WARN_SUPER)
681 #define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \
682 (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_NONCHAR|UNICODE_DISALLOW_SUPER)
684 /* For backward source compatibility, as are now the default */
685 #define UNICODE_ALLOW_SURROGATE 0
686 #define UNICODE_ALLOW_SUPER 0
687 #define UNICODE_ALLOW_ANY 0
689 /* This matches the 2048 code points between UNICODE_SURROGATE_FIRST (0xD800) and
690 * UNICODE_SURROGATE_LAST (0xDFFF) */
691 #define UNICODE_IS_SURROGATE(uv) (((UV) (uv) & (~0xFFFF | 0xF800)) \
694 #define UNICODE_IS_REPLACEMENT(uv) ((UV) (uv) == UNICODE_REPLACEMENT)
695 #define UNICODE_IS_BYTE_ORDER_MARK(uv) ((UV) (uv) == UNICODE_BYTE_ORDER_MARK)
697 /* Is 'uv' one of the 32 contiguous-range noncharacters? */
698 #define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) ((UV) (uv) >= 0xFDD0 \
699 && (UV) (uv) <= 0xFDEF)
701 /* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE,
702 * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above
703 * the Unicode legal max */
704 #define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \
705 (((UV) (uv) & 0xFFFE) == 0xFFFE)
707 #define UNICODE_IS_NONCHAR(uv) \
708 ( UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \
709 || ( LIKELY( ! UNICODE_IS_SUPER(uv)) \
710 && UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)))
712 #define UNICODE_IS_SUPER(uv) ((UV) (uv) > PERL_UNICODE_MAX)
713 #define UNICODE_IS_ABOVE_31_BIT(uv) ((UV) (uv) > 0x7FFFFFFF)
715 #define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE
716 #define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \
717 LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE
718 #define MICRO_SIGN MICRO_SIGN_NATIVE
719 #define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \
720 LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE
721 #define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \
722 LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE
723 #define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3
724 #define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2
725 #define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3
726 #define GREEK_SMALL_LETTER_MU 0x03BC
727 #define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case
729 #define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */
730 #ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
731 # define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E
733 #define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130
734 #define LATIN_SMALL_LETTER_DOTLESS_I 0x131
735 #define LATIN_SMALL_LETTER_LONG_S 0x017F
736 #define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05
737 #define LATIN_SMALL_LIGATURE_ST 0xFB06
738 #define KELVIN_SIGN 0x212A
739 #define ANGSTROM_SIGN 0x212B
741 #define UNI_DISPLAY_ISPRINT 0x0001
742 #define UNI_DISPLAY_BACKSLASH 0x0002
743 #define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
744 #define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
746 #define ANYOF_FOLD_SHARP_S(node, input, end) \
747 (ANYOF_BITMAP_TEST(node, LATIN_SMALL_LETTER_SHARP_S) && \
748 (ANYOF_NONBITMAP(node)) && \
749 (ANYOF_FLAGS(node) & ANYOF_LOC_NONBITMAP_FOLD) && \
750 ((end) > (input) + 1) && \
751 isALPHA_FOLD_EQ((input)[0], 's'))
753 #define SHARP_S_SKIP 2
755 /* If you want to exclude surrogates, and beyond legal Unicode, see the blame
756 * log for earlier versions which gave details for these */
758 /* A helper macro for isUTF8_CHAR, so use that one, and not this one. This is
759 * retained solely for backwards compatibility and may be deprecated and
760 * removed in a future Perl version.
762 * regen/regcharclass.pl generates is_UTF8_CHAR_utf8() macros for up to these
763 * number of bytes. So this has to be coordinated with that file */
765 # define IS_UTF8_CHAR_FAST(n) ((n) <= 3)
767 # define IS_UTF8_CHAR_FAST(n) ((n) <= 4)
771 /* A helper macro for isUTF8_CHAR, so use that one instead of this. This was
772 * generated by regen/regcharclass.pl, and then moved here. The lines that
773 * generated it were then commented out. This was done solely because it takes
774 * on the order of 10 minutes to generate, and is never going to change, unless
775 * the generated code is improved.
777 * The EBCDIC versions have been cut to not cover all of legal Unicode,
778 * otherwise they take too long to generate; besides there is a separate one
779 * for each code page, so they are in regcharclass.h instead of here */
781 UTF8_CHAR: Matches legal UTF-8 encoded characters from 2 through 4 bytes
785 /*** GENERATED CODE ***/
786 #define is_UTF8_CHAR_utf8_no_length_checks(s) \
787 ( ( 0xC2 <= ((U8*)s)[0] && ((U8*)s)[0] <= 0xDF ) ? \
788 ( ( ( ((U8*)s)[1] & 0xC0 ) == 0x80 ) ? 2 : 0 ) \
789 : ( 0xE0 == ((U8*)s)[0] ) ? \
790 ( ( ( ( ((U8*)s)[1] & 0xE0 ) == 0xA0 ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
791 : ( 0xE1 <= ((U8*)s)[0] && ((U8*)s)[0] <= 0xEF ) ? \
792 ( ( ( ( ((U8*)s)[1] & 0xC0 ) == 0x80 ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) ? 3 : 0 )\
793 : ( 0xF0 == ((U8*)s)[0] ) ? \
794 ( ( ( ( 0x90 <= ((U8*)s)[1] && ((U8*)s)[1] <= 0xBF ) && ( ( ((U8*)s)[2] & 0xC0 ) == 0x80 ) ) && ( ( ((U8*)s)[3] & 0xC0 ) == 0x80 ) ) ? 4 : 0 )\
795 : ( ( ( ( 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 )
800 =for apidoc Am|STRLEN|isUTF8_CHAR|const U8 *s|const U8 *e
802 Returns the number of bytes beginning at C<s> which form a legal UTF-8 (or
803 UTF-EBCDIC) encoded character, looking no further than S<C<e - s>> bytes into
804 C<s>. Returns 0 if the sequence starting at C<s> through S<C<e - 1>> is not
807 Note that an INVARIANT character (i.e. ASCII on non-EBCDIC
808 machines) is a valid UTF-8 character.
813 #define isUTF8_CHAR(s, e) (UNLIKELY((e) <= (s)) \
815 : (UTF8_IS_INVARIANT(*s)) \
817 : UNLIKELY(((e) - (s)) < UTF8SKIP(s)) \
819 : LIKELY(IS_UTF8_CHAR_FAST(UTF8SKIP(s))) \
820 ? is_UTF8_CHAR_utf8_no_length_checks(s) \
821 : _is_utf8_char_slow(s, e))
823 #define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end)
825 /* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is
826 * retained solely for backwards compatibility */
827 #define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n)
832 * ex: set ts=8 sts=4 sw=4 et: