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.
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.
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.
28 * Unfortunately there are inconsistencies.
32 #ifndef PERL_UTF8_H_ /* Guard against recursive inclusion */
33 #define PERL_UTF8_H_ 1
36 =for apidoc Ay||utf8ness_t
38 This typedef is used by several core functions that return PV strings, to
39 indicate the UTF-8ness of those strings.
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.)
44 The possible values this can be are:
50 This means the string definitely should be treated as a sequence of
51 UTF-8-encoded characters.
53 Most code that needs to handle this typedef should be of the form:
55 if (utf8ness_flag == UTF8NESS_YES) {
56 treat as utf8; // like turning on an SV UTF-8 flag
61 This means the string definitely should be treated as a sequence of bytes, not
64 =item C<UTF8NESS_IMMATERIAL>
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
71 =item C<UTF8NESS_UNKNOWN>
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:
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;
85 The following relationships hold between the enum values:
89 =item S<C<0 E<lt>= I<enum value> E<lt>= UTF8NESS_IMMATERIAL>>
91 the string may be treated in code as non-UTF8
93 =item S<C<UTF8NESS_IMMATERIAL E<lt>= <I<enum value>>>
95 the string may be treated in code as encoded in UTF-8
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
107 UTF8NESS_YES = 2, /* Defintely is UTF-8, wideness
109 UTF8NESS_UNKNOWN = -1, /* Undetermined so far */
112 /* Use UTF-8 as the default script encoding?
113 * Turning this on will break scripts having non-UTF-8 binary
114 * data (such as Latin-1) in string literals. */
115 #ifdef USE_UTF8_SCRIPTS
116 # define USE_UTF8_IN_NAMES (!IN_BYTES)
118 # define USE_UTF8_IN_NAMES (PL_hints & HINT_UTF8)
121 #include "regcharclass.h"
122 #include "unicode_constants.h"
124 /* For to_utf8_fold_flags, q.v. */
125 #define FOLD_FLAGS_LOCALE 0x1
126 #define FOLD_FLAGS_FULL 0x2
127 #define FOLD_FLAGS_NOMIX_ASCII 0x4
130 =for apidoc is_ascii_string
132 This is a misleadingly-named synonym for L</is_utf8_invariant_string>.
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
135 than just the ASCII characters, so C<is_utf8_invariant_string> is preferred.
137 =for apidoc is_invariant_string
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.
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)
148 #define uvoffuni_to_utf8_flags(d,uv,flags) \
149 uvoffuni_to_utf8_flags_msgs(d, uv, flags, 0)
150 #define uvchr_to_utf8(a,b) uvchr_to_utf8_flags(a,b,0)
151 #define uvchr_to_utf8_flags(d,uv,flags) \
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)
155 #define utf8_to_uvchr_buf(s, e, lenp) \
156 utf8_to_uvchr_buf_helper((const U8 *) (s), (const U8 *) e, lenp)
157 #define utf8n_to_uvchr(s, len, lenp, flags) \
158 utf8n_to_uvchr_error(s, len, lenp, flags, 0)
159 #define utf8n_to_uvchr_error(s, len, lenp, flags, errors) \
160 utf8n_to_uvchr_msgs(s, len, lenp, flags, errors, 0)
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)
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)
171 #define to_uni_fold(c, p, lenp) _to_uni_fold_flags(c, p, lenp, FOLD_FLAGS_FULL)
173 #define foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2) \
174 foldEQ_utf8_flags(s1, pe1, l1, u1, s2, pe2, l2, u2, 0)
175 #define FOLDEQ_UTF8_NOMIX_ASCII (1 << 0)
176 #define FOLDEQ_LOCALE (1 << 1)
177 #define FOLDEQ_S1_ALREADY_FOLDED (1 << 2)
178 #define FOLDEQ_S2_ALREADY_FOLDED (1 << 3)
179 #define FOLDEQ_S1_FOLDS_SANE (1 << 4)
180 #define FOLDEQ_S2_FOLDS_SANE (1 << 5)
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.
187 * It is helpful to know the EBCDIC value on ASCII platforms, mainly to avoid
189 #define UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS 5
191 /* See explanation below at 'UTF8_MAXBYTES' */
192 #define ASCII_PLATFORM_UTF8_MAXBYTES 13
196 /* The equivalent of the next few macros but implementing UTF-EBCDIC are in the
197 * following header file: */
198 # include "utfebcdic.h"
200 # else /* ! EBCDIC */
205 EXTCONST unsigned char PL_utf8skip[];
207 EXTCONST unsigned char PL_utf8skip[] = {
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 */
221 /* 0xC2 */ 2,2,2,2,2,2,2,2,2,2,2,2,2,2, /* U+0080 to U+03FF */
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 */
225 /* Perl extended (never was official UTF-8). Up to 36 bit */
227 /* More extended, Up to 72 bits (64-bit + reserved) */
228 /* 0xFF */ ASCII_PLATFORM_UTF8_MAXBYTES
236 =for apidoc Am|U8|NATIVE_TO_LATIN1|U8 ch
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
245 For conversion of code points potentially larger than will fit in a character,
246 use L</NATIVE_TO_UNI>.
248 =for apidoc Am|U8|LATIN1_TO_NATIVE|U8 ch
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
257 For conversion of code points potentially larger than will fit in a character,
258 use L</UNI_TO_NATIVE>.
260 =for apidoc Am|UV|NATIVE_TO_UNI|UV ch
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.
268 =for apidoc Am|UV|UNI_TO_NATIVE|UV ch
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.
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)))
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 */
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)))
291 #define UNI_TO_NATIVE(ch) ((UV) ASSERT_NOT_PTR(ch))
292 #define NATIVE_TO_UNI(ch) ((UV) ASSERT_NOT_PTR(ch))
296 The following table is from Unicode 3.2, plus the Perl extensions for above
299 Code Points 1st Byte 2nd Byte 3rd 4th 5th 6th 7th 8th-13th
301 U+0000..U+007F 00..7F
302 U+0080..U+07FF * C2..DF 80..BF
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
321 Note the gaps before several of the byte entries above marked by '*'. These are
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
325 (and that is what Perl does). The non-shortest ones are called 'overlongs'.
327 Another way to look at it, as bits:
329 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
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
336 As you can see, the continuation bytes all begin with C<10>, and the
337 leading bits of the start byte tell how many bytes there are in the
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
344 are in the character. */
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 */
350 # define UTF_CONTINUATION_BYTE_INFO_BITS 6
352 /* ^? is defined to be DEL on ASCII systems. See the definition of toCTRL()
354 # define QUESTION_MARK_CTRL DEL_NATIVE
356 #endif /* EBCDIC vs ASCII */
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
361 #define ONE_IF_EBCDIC_ZERO_IF_NOT \
362 (UTF_CONTINUATION_BYTE_INFO_BITS == UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS)
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
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. */
372 #define UTF_CONTINUATION_MASK \
373 ((U8) nBIT_MASK(UTF_CONTINUATION_BYTE_INFO_BITS))
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 \
378 ((U8) ((0xFF << UTF_ACCUMULATION_SHIFT) & 0xFF))
380 /* This defines the bits that are to be in the continuation bytes of a
381 * multi-byte UTF-8 encoded character that mark it is a continuation byte.
382 * This turns out to be 0x80 in UTF-8, 0xA0 in UTF-EBCDIC. (khw doesn't know
383 * the underlying reason that B0 works here, except it just happens to work.
384 * One could solve for two linear equations and come up with it.) */
385 #define UTF_CONTINUATION_MARK (UTF_IS_CONTINUATION_MASK & 0xB0)
387 /* This value is clearer in some contexts */
388 #define UTF_MIN_CONTINUATION_BYTE UTF_CONTINUATION_MARK
390 /* Is the byte 'c' part of a multi-byte UTF8-8 encoded sequence, and not the
391 * first byte thereof? */
392 #define UTF8_IS_CONTINUATION(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
393 (((NATIVE_UTF8_TO_I8(c) & UTF_IS_CONTINUATION_MASK) \
394 == UTF_CONTINUATION_MARK)))
396 /* Is the representation of the Unicode code point 'cp' the same regardless of
397 * being encoded in UTF-8 or not? This is a fundamental property of
399 #define OFFUNI_IS_INVARIANT(c) \
400 (((WIDEST_UTYPE)(c)) < UTF_MIN_CONTINUATION_BYTE)
403 =for apidoc Am|bool|UVCHR_IS_INVARIANT|UV cp
405 Evaluates to 1 if the representation of code point C<cp> is the same whether or
406 not it is encoded in UTF-8; otherwise evaluates to 0. UTF-8 invariant
407 characters can be copied as-is when converting to/from UTF-8, saving time.
408 C<cp> is Unicode if above 255; otherwise is platform-native.
412 #define UVCHR_IS_INVARIANT(cp) (OFFUNI_IS_INVARIANT(NATIVE_TO_UNI(cp)))
414 /* This defines the 1-bits that are to be in the first byte of a multi-byte
415 * UTF-8 encoded character that mark it as a start byte and give the number of
416 * bytes that comprise the character. 'len' is that number.
418 * To illustrate: len = 2 => ((U8) ~ 0b0011_1111) or 1100_0000
419 * 7 => ((U8) ~ 0b0000_0001) or 1111_1110
422 * This is not to be used on a single-byte character. As in many places in
423 * perl, U8 must be 8 bits
425 #define UTF_START_MARK(len) ((U8) ~(0xFF >> (len)))
427 /* Masks out the initial one bits in a start byte, leaving the following 0 bit
428 * and the real data bits. 'len' is the number of bytes in the multi-byte
429 * sequence that comprises the character.
431 * To illustrate: len = 2 => 0b0011_1111 works on start byte 110xxxxx
432 * 6 => 0b0000_0011 works on start byte 1111110x
433 * >= 7 => There are no data bits in the start byte
434 * Note that on ASCII platforms, this can be passed a len=1 byte; and all the
435 * real data bits will be returned:
436 len = 1 => 0b0111_1111
437 * This isn't true on EBCDIC platforms, where some len=1 bytes are of the form
438 * 0b101x_xxxx, so this can't be used there on single-byte characters. */
439 #define UTF_START_MASK(len) (0xFF >> (len))
443 =for apidoc AmnU|STRLEN|UTF8_MAXBYTES
445 The maximum width of a single UTF-8 encoded character, in bytes.
447 NOTE: Strictly speaking Perl's UTF-8 should not be called UTF-8 since UTF-8
448 is an encoding of Unicode, and Unicode's upper limit, 0x10FFFF, can be
449 expressed with 4 bytes. However, Perl thinks of UTF-8 as a way to encode
450 non-negative integers in a binary format, even those above Unicode.
454 The start byte 0xFE, never used in any ASCII platform UTF-8 specification, has
455 an obvious meaning, namely it has its upper 7 bits set, so it should start a
456 sequence of 7 bytes. And in fact, this is exactly what standard UTF-EBCDIC
459 The start byte FF, on the other hand could have several different plausible
461 1) The meaning in standard UTF-EBCDIC, namely as an FE start byte, with the
462 bottom bit that should be a fixed '0' to form FE, instead acting as an
464 2) That the sequence should have exactly 8 bytes.
465 3) That the next byte is to be treated as a sort of extended start byte,
466 which in combination with this one gives the total length of the sequence.
467 There are published UTF-8 extensions that do this, some string together
468 multiple initial FF start bytes to achieve arbitrary precision.
469 4) That the sequence has exactly n bytes, where n is what the implementation
473 The goal is to be able to represent 64-bit values in UTF-8 or UTF-EBCDIC. That
474 rules out items 1) and 2). Item 3) has the deal-breaking disadvantage of
475 requiring one to read more than one byte to determine the total length of the
476 sequence. So in Perl, a start byte of FF indicates a UTF-8 string consisting
477 of the start byte, plus enough continuation bytes to encode a 64 bit value.
478 This turns out to be 13 total bytes in UTF-8 and 14 in UTF-EBCDIC. This is
479 because we get zero info bits from the start byte, plus
480 12 * 6 bits of info per continuation byte (could encode 72-bit numbers) on
481 UTF-8 (khw knows not why 11, which would encode 66 bits wasn't
483 13 * 5 bits of info per byte (could encode 65-bit numbers) on UTF-EBCDIC
485 The disadvantages of this method are:
486 1) There's potentially a lot of wasted bytes for all but the largest values.
487 For example, something that could be represented by 7 continuation bytes,
488 instead requires the full 12 or 13.
489 2) There would be problems should larger values, 128-bit say, ever need to be
492 WARNING: This number must be in sync with the value in
493 regen/charset_translations.pl. */
494 #define UTF8_MAXBYTES \
495 (ASCII_PLATFORM_UTF8_MAXBYTES + ONE_IF_EBCDIC_ZERO_IF_NOT)
497 /* Calculate how many bytes are necessary to represent a value whose most
498 * significant 1 bit is in bit position 'pos' of the word. For 0x1, 'pos would
499 * be 0; and for 0x400, 'pos' would be 10, and the result would be:
500 * EBCDIC floor((-1 + (10 + 5 - 1 - 1)) / (5 - 1))
501 * = floor((-1 + (13)) / 4)
504 * ASCII floor(( 0 + (10 + 6 - 1 - 1)) / (6 - 1))
507 * The reason this works is because the number of bits needed to represent a
508 * value is proportional to (UTF_CONTINUATION_BYTE_INFO_BITS - 1). The -1 is
509 * because each new continuation byte removes one bit of information from the
512 * This is a step function (we need to allocate a full extra byte if we
513 * overflow by just a single bit)
515 * The caller is responsible for making sure 'pos' is at least 8 (occupies 9
516 * bits), as it breaks down at the lower edge. At the high end, if it returns
517 * 8 or more, Perl instead anomalously uses MAX_BYTES, so this would be wrong.
519 #define UNISKIP_BY_MSB_(pos) \
520 ( ( -ONE_IF_EBCDIC_ZERO_IF_NOT /* platform break pos's are off-by-one */ \
521 + (pos) + ((UTF_CONTINUATION_BYTE_INFO_BITS - 1) - 1)) /* Step fcn */ \
522 / (UTF_CONTINUATION_BYTE_INFO_BITS - 1)) /* take floor of */
524 /* Compute the number of UTF-8 bytes required for representing the input uv,
525 * which must be a Unicode, not native value.
527 * This uses msbit_pos() which doesn't work on NUL, and UNISKIP_BY_MSB_ breaks
528 * down for small code points. So first check if the input is invariant to get
529 * around that, and use a helper for high code points to accommodate the fact
530 * that above 7 btyes, the value is anomalous. The helper is empty on
531 * platforms that don't go that high */
532 #define OFFUNISKIP(uv) \
533 ((OFFUNI_IS_INVARIANT(uv)) \
535 : (OFFUNISKIP_helper_(uv) UNISKIP_BY_MSB_(msbit_pos(uv))))
537 /* We need to go to MAX_BYTES when we can't represent 'uv' by the number of
538 * information bits in 6 continuation bytes (when we get to 6, the start byte
539 * has no information bits to add to the total). But on 32-bit ASCII
540 * platforms, that doesn't happen until 6*6 bits, so on those platforms, this
541 * will always be false */
542 #if UVSIZE * CHARBITS > (6 * UTF_CONTINUATION_BYTE_INFO_BITS)
543 # define HAS_EXTRA_LONG_UTF8
544 # define OFFUNISKIP_helper_(uv) \
545 UNLIKELY(uv > nBIT_UMAX(6 * UTF_CONTINUATION_BYTE_INFO_BITS)) \
548 # define OFFUNISKIP_helper_(uv)
553 =for apidoc Am|STRLEN|UVCHR_SKIP|UV cp
554 returns the number of bytes required to represent the code point C<cp> when
555 encoded as UTF-8. C<cp> is a native (ASCII or EBCDIC) code point if less than
556 255; a Unicode code point otherwise.
560 #define UVCHR_SKIP(uv) OFFUNISKIP(NATIVE_TO_UNI(uv))
562 #define NATIVE_SKIP(uv) UVCHR_SKIP(uv) /* Old terminology */
564 /* Most code which says UNISKIP is really thinking in terms of native code
565 * points (0-255) plus all those beyond. This is an imprecise term, but having
566 * it means existing code continues to work. For precision, use UVCHR_SKIP,
567 * NATIVE_SKIP, or OFFUNISKIP */
568 #define UNISKIP(uv) UVCHR_SKIP(uv)
570 /* Compute the start byte for a given code point. This requires the log2 of
571 * the code point, which is hard to compute at compile time, which this macro
572 * wants to be. (Perhaps deBruijn sequences could be used.) So a parameter
573 * for the number of bits the value occupies is passed in, which the programmer
574 * has had to figure out to get compile-time effect. And asserts are used to
575 * make sure the value is correct.
577 * Since we are interested only in the start byte, we ignore the lower bits
578 * accounted for by the continuation bytes. Each continuation byte eats up
579 * UTF_CONTINUATION_BYTE_INFO_BITS bits, so the number of continuation bytes
580 * needed is floor(bits / UTF_CONTINUATION_BYTE_INFO_BITS). That number is fed
581 * to UTF_START_MARK() to get the upper part of the start byte. The left over
582 * bits form the lower part which is OR'd with the mark
584 * Note that on EBCDIC platforms, this is actually the I8 */
585 #define UTF_START_BYTE(uv, bits) \
586 (__ASSERT_((uv) >> ((bits) - 1)) /* At least 'bits' */ \
587 __ASSERT_(((uv) & ~nBIT_MASK(bits)) == 0) /* No extra bits */ \
588 UTF_START_MARK(UNISKIP_BY_MSB_((bits) - 1)) \
589 | ((uv) >> (((bits) / UTF_CONTINUATION_BYTE_INFO_BITS) \
590 * UTF_CONTINUATION_BYTE_INFO_BITS)))
592 /* Compute the first continuation byte for a given code point. This is mostly
593 * for compile-time, so how many bits it occupies is also passed in).
595 * We are interested in the first continuation byte, so we ignore the lower
596 * bits accounted for by the rest of the continuation bytes by right shifting
597 * out their info bit, and mask out the higher bits that will go into the start
600 * Note that on EBCDIC platforms, this is actually the I8 */
601 #define UTF_FIRST_CONT_BYTE(uv, bits) \
602 (__ASSERT_((uv) >> ((bits) - 1)) /* At least 'bits' */ \
603 __ASSERT_(((uv) & ~nBIT_MASK(bits)) == 0) /* No extra bits */ \
604 UTF_CONTINUATION_MARK \
605 | ( UTF_CONTINUATION_MASK \
606 & ((uv) >> ((((bits) / UTF_CONTINUATION_BYTE_INFO_BITS) - 1) \
607 * UTF_CONTINUATION_BYTE_INFO_BITS))))
609 #define UTF_MIN_START_BYTE UTF_START_BYTE(UTF_MIN_CONTINUATION_BYTE, 8)
611 /* Is the byte 'c' the first byte of a multi-byte UTF8-8 encoded sequence?
612 * This excludes invariants (they are single-byte). It also excludes the
613 * illegal overlong sequences that begin with C0 and C1 on ASCII platforms, and
614 * C0-C4 I8 start bytes on EBCDIC ones. On EBCDIC E0 can't start a
615 * non-overlong sequence, so we define a base macro and for those platforms,
616 * extend it to also exclude E0 */
617 #define UTF8_IS_START_base(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
618 (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_START_BYTE))
620 # define UTF8_IS_START(c) \
621 (UTF8_IS_START_base(c) && (c) != I8_TO_NATIVE_UTF8(0xE0))
623 # define UTF8_IS_START(c) UTF8_IS_START_base(c)
626 #define UTF_MIN_ABOVE_LATIN1_BYTE UTF_START_BYTE(0x100, 9)
628 /* Is the UTF8-encoded byte 'c' the first byte of a sequence of bytes that
629 * represent a code point > 255? */
630 #define UTF8_IS_ABOVE_LATIN1(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
631 (NATIVE_UTF8_TO_I8(c) >= UTF_MIN_ABOVE_LATIN1_BYTE))
633 /* Is the UTF8-encoded byte 'c' the first byte of a two byte sequence? Use
634 * UTF8_IS_NEXT_CHAR_DOWNGRADEABLE() instead if the input isn't known to
636 #define UTF8_IS_DOWNGRADEABLE_START(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
637 inRANGE_helper_(U8, NATIVE_UTF8_TO_I8(c), \
638 UTF_MIN_START_BYTE, UTF_MIN_ABOVE_LATIN1_BYTE - 1))
640 /* The largest code point representable by two UTF-8 bytes on this platform.
641 * The binary for that code point is:
642 * 1101_1111 10xx_xxxx in UTF-8, and
643 * 1101_1111 101y_yyyy in UTF-EBCDIC I8.
644 * where both x and y are 1, and shown this way to indicate there is one more x
645 * than there is y. The number of x and y bits are their platform's respective
646 * UTF_CONTINUATION_BYTE_INFO_BITS. Squeezing out the bits that don't
647 * contribute to the value, these evaluate to:
648 * 1_1111 xx_xxxx in UTF-8, and
649 * 1_1111 y_yyyy in UTF-EBCDIC I8.
650 * or, the maximum value of an unsigned with (5 + info_bit_count) bits */
651 #define MAX_UTF8_TWO_BYTE nBIT_UMAX(5 + UTF_CONTINUATION_BYTE_INFO_BITS)
653 /* The largest code point representable by two UTF-8 bytes on any platform that
655 #define MAX_PORTABLE_UTF8_TWO_BYTE \
656 nBIT_UMAX(5 + MIN( UTF_CONTINUATION_BYTE_INFO_BITS, \
657 UTF_EBCDIC_CONTINUATION_BYTE_INFO_BITS))
661 =for apidoc AmnU|STRLEN|UTF8_MAXBYTES_CASE
663 The maximum number of UTF-8 bytes a single Unicode character can
664 uppercase/lowercase/titlecase/fold into.
668 * Unicode guarantees that the maximum expansion is UTF8_MAX_FOLD_CHAR_EXPAND
669 * characters, but any above-Unicode code point will fold to itself, so we only
670 * have to look at the expansion of the maximum Unicode code point. But this
671 * number may be less than the space occupied by a very large code point under
672 * Perl's extended UTF-8. We have to make it large enough to fit any single
673 * character. (It turns out that ASCII and EBCDIC differ in which is larger)
677 #define UTF8_MAXBYTES_CASE \
678 MAX(UTF8_MAXBYTES, UTF8_MAX_FOLD_CHAR_EXPAND * UNISKIP_BY_MSB_(20))
680 /* Rest of these are attributes of Unicode and perl's internals rather than the
681 * encoding, or happen to be the same in both ASCII and EBCDIC (at least at
682 * this level; the macros that some of these call may have different
683 * definitions in the two encodings */
685 /* In domain restricted to ASCII, these may make more sense to the reader than
686 * the ones with Latin1 in the name */
687 #define NATIVE_TO_ASCII(ch) NATIVE_TO_LATIN1(ch)
688 #define ASCII_TO_NATIVE(ch) LATIN1_TO_NATIVE(ch)
690 /* More or less misleadingly-named defines, retained for back compat */
691 #define NATIVE_TO_UTF(ch) NATIVE_UTF8_TO_I8(ch)
692 #define NATIVE_TO_I8(ch) NATIVE_UTF8_TO_I8(ch)
693 #define UTF_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
694 #define I8_TO_NATIVE(ch) I8_TO_NATIVE_UTF8(ch)
695 #define NATIVE8_TO_UNI(ch) NATIVE_TO_LATIN1(ch)
697 /* Adds a UTF8 continuation byte 'new' of information to a running total code
698 * point 'old' of all the continuation bytes so far. This is designed to be
699 * used in a loop to convert from UTF-8 to the code point represented. Note
700 * that this is asymmetric on EBCDIC platforms, in that the 'new' parameter is
701 * the UTF-EBCDIC byte, whereas the 'old' parameter is a Unicode (not EBCDIC)
702 * code point in process of being generated */
703 #define UTF8_ACCUMULATE(old, new) (__ASSERT_(FITS_IN_8_BITS(new)) \
704 ((old) << UTF_ACCUMULATION_SHIFT) \
705 | ((NATIVE_UTF8_TO_I8(new)) \
706 & UTF_CONTINUATION_MASK))
708 /* This works in the face of malformed UTF-8. */
709 #define UTF8_IS_NEXT_CHAR_DOWNGRADEABLE(s, e) \
711 && UTF8_IS_DOWNGRADEABLE_START(*(s)) \
712 && UTF8_IS_CONTINUATION(*((s)+1)))
714 /* Longer, but more accurate name */
715 #define UTF8_IS_ABOVE_LATIN1_START(c) UTF8_IS_ABOVE_LATIN1(c)
717 /* Convert a UTF-8 variant Latin1 character to a native code point value.
718 * Needs just one iteration of accumulate. Should be used only if it is known
719 * that the code point is < 256, and is not UTF-8 invariant. Use the slower
720 * but more general TWO_BYTE_UTF8_TO_NATIVE() which handles any code point
721 * representable by two bytes (which turns out to be up through
722 * MAX_PORTABLE_UTF8_TWO_BYTE). The two parameters are:
723 * HI: a downgradable start byte;
726 #define EIGHT_BIT_UTF8_TO_NATIVE(HI, LO) \
727 ( __ASSERT_(UTF8_IS_DOWNGRADEABLE_START(HI)) \
728 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
729 LATIN1_TO_NATIVE(UTF8_ACCUMULATE(( \
730 NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), (LO))))
732 /* Convert a two (not one) byte utf8 character to a native code point value.
733 * Needs just one iteration of accumulate. Should not be used unless it is
734 * known that the two bytes are legal: 1) two-byte start, and 2) continuation.
735 * Note that the result can be larger than 255 if the input character is not
737 #define TWO_BYTE_UTF8_TO_NATIVE(HI, LO) \
738 (__ASSERT_(FITS_IN_8_BITS(HI)) \
739 __ASSERT_(FITS_IN_8_BITS(LO)) \
740 __ASSERT_(PL_utf8skip[(U8) HI] == 2) \
741 __ASSERT_(UTF8_IS_CONTINUATION(LO)) \
742 UNI_TO_NATIVE(UTF8_ACCUMULATE((NATIVE_UTF8_TO_I8(HI) & UTF_START_MASK(2)), \
745 /* Should never be used, and be deprecated */
746 #define TWO_BYTE_UTF8_TO_UNI(HI, LO) NATIVE_TO_UNI(TWO_BYTE_UTF8_TO_NATIVE(HI, LO))
750 =for apidoc Am|STRLEN|UTF8SKIP|char* s
751 returns the number of bytes a non-malformed UTF-8 encoded character whose first
752 (perhaps only) byte is pointed to by C<s>.
754 If there is a possibility of malformed input, use instead:
758 =item C<L</UTF8_SAFE_SKIP>> if you know the maximum ending pointer in the
759 buffer pointed to by C<s>; or
761 =item C<L</UTF8_CHK_SKIP>> if you don't know it.
765 It is better to restructure your code so the end pointer is passed down so that
766 you know what it actually is at the point of this call, but if that isn't
767 possible, C<L</UTF8_CHK_SKIP>> can minimize the chance of accessing beyond the end
772 #define UTF8SKIP(s) PL_utf8skip[*(const U8*)(ASSERT_IS_PTR(s))]
775 =for apidoc Am|STRLEN|UTF8_SKIP|char* s
776 This is a synonym for C<L</UTF8SKIP>>
781 #define UTF8_SKIP(s) UTF8SKIP(s)
784 =for apidoc Am|STRLEN|UTF8_CHK_SKIP|char* s
786 This is a safer version of C<L</UTF8SKIP>>, but still not as safe as
787 C<L</UTF8_SAFE_SKIP>>. This version doesn't blindly assume that the input
788 string pointed to by C<s> is well-formed, but verifies that there isn't a NUL
789 terminating character before the expected end of the next character in C<s>.
790 The length C<UTF8_CHK_SKIP> returns stops just before any such NUL.
792 Perl tends to add NULs, as an insurance policy, after the end of strings in
793 SV's, so it is likely that using this macro will prevent inadvertent reading
794 beyond the end of the input buffer, even if it is malformed UTF-8.
796 This macro is intended to be used by XS modules where the inputs could be
797 malformed, and it isn't feasible to restructure to use the safer
798 C<L</UTF8_SAFE_SKIP>>, for example when interfacing with a C library.
803 #define UTF8_CHK_SKIP(s) \
804 (UNLIKELY(s[0] == '\0') ? 1 : MIN(UTF8SKIP(s), \
805 my_strnlen((char *) (s), UTF8SKIP(s))))
808 =for apidoc Am|STRLEN|UTF8_SAFE_SKIP|char* s|char* e
809 returns 0 if S<C<s E<gt>= e>>; otherwise returns the number of bytes in the
810 UTF-8 encoded character whose first byte is pointed to by C<s>. But it never
811 returns beyond C<e>. On DEBUGGING builds, it asserts that S<C<s E<lt>= e>>.
815 #define UTF8_SAFE_SKIP(s, e) (__ASSERT_((e) >= (s)) \
816 UNLIKELY(((e) - (s)) <= 0) \
818 : MIN(((e) - (s)), UTF8_SKIP(s)))
820 /* Most code that says 'UNI_' really means the native value for code points up
822 #define UNI_IS_INVARIANT(cp) UVCHR_IS_INVARIANT(cp)
825 =for apidoc Am|bool|UTF8_IS_INVARIANT|char c
827 Evaluates to 1 if the byte C<c> represents the same character when encoded in
828 UTF-8 as when not; otherwise evaluates to 0. UTF-8 invariant characters can be
829 copied as-is when converting to/from UTF-8, saving time.
831 In spite of the name, this macro gives the correct result if the input string
832 from which C<c> comes is not encoded in UTF-8.
834 See C<L</UVCHR_IS_INVARIANT>> for checking if a UV is invariant.
838 The reason it works on both UTF-8 encoded strings and non-UTF-8 encoded, is
839 that it returns TRUE in each for the exact same set of bit patterns. It is
840 valid on a subset of what UVCHR_IS_INVARIANT is valid on, so can just use that;
841 and the compiler should optimize out anything extraneous given the
842 implementation of the latter. */
843 #define UTF8_IS_INVARIANT(c) UVCHR_IS_INVARIANT(ASSERT_NOT_PTR(c))
845 /* Like the above, but its name implies a non-UTF8 input, which as the comments
846 * above show, doesn't matter as to its implementation */
847 #define NATIVE_BYTE_IS_INVARIANT(c) UVCHR_IS_INVARIANT(c)
849 /* Misleadingly named: is the UTF8-encoded byte 'c' part of a variant sequence
850 * in UTF-8? This is the inverse of UTF8_IS_INVARIANT. */
851 #define UTF8_IS_CONTINUED(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
852 (! UTF8_IS_INVARIANT(c)))
854 /* The macros in the next 4 sets are used to generate the two utf8 or utfebcdic
855 * bytes from an ordinal that is known to fit into exactly two (not one) bytes;
856 * it must be less than 0x3FF to work across both encodings. */
858 /* These two are helper macros for the other three sets, and should not be used
859 * directly anywhere else. 'translate_function' is either NATIVE_TO_LATIN1
860 * (which works for code points up through 0xFF) or NATIVE_TO_UNI which works
861 * for any code point */
862 #define __BASE_TWO_BYTE_HI(c, translate_function) \
863 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
864 I8_TO_NATIVE_UTF8((translate_function(c) >> UTF_ACCUMULATION_SHIFT) \
865 | UTF_START_MARK(2)))
866 #define __BASE_TWO_BYTE_LO(c, translate_function) \
867 (__ASSERT_(! UVCHR_IS_INVARIANT(c)) \
868 I8_TO_NATIVE_UTF8((translate_function(c) & UTF_CONTINUATION_MASK) \
869 | UTF_CONTINUATION_MARK))
871 /* The next two macros should not be used. They were designed to be usable as
872 * the case label of a switch statement, but this doesn't work for EBCDIC. Use
873 * regen/unicode_constants.pl instead */
874 #define UTF8_TWO_BYTE_HI_nocast(c) __BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)
875 #define UTF8_TWO_BYTE_LO_nocast(c) __BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)
877 /* The next two macros are used when the source should be a single byte
878 * character; checked for under DEBUGGING */
879 #define UTF8_EIGHT_BIT_HI(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
880 ( __BASE_TWO_BYTE_HI(c, NATIVE_TO_LATIN1)))
881 #define UTF8_EIGHT_BIT_LO(c) (__ASSERT_(FITS_IN_8_BITS(c)) \
882 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_LATIN1)))
884 /* These final two macros in the series are used when the source can be any
885 * code point whose UTF-8 is known to occupy 2 bytes; they are less efficient
886 * than the EIGHT_BIT versions on EBCDIC platforms. We use the logical '~'
887 * operator instead of "<=" to avoid getting compiler warnings.
888 * MAX_UTF8_TWO_BYTE should be exactly all one bits in the lower few
889 * places, so the ~ works */
890 #define UTF8_TWO_BYTE_HI(c) \
891 (__ASSERT_((sizeof(c) == 1) \
892 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
893 (__BASE_TWO_BYTE_HI(c, NATIVE_TO_UNI)))
894 #define UTF8_TWO_BYTE_LO(c) \
895 (__ASSERT_((sizeof(c) == 1) \
896 || !(((WIDEST_UTYPE)(c)) & ~MAX_UTF8_TWO_BYTE)) \
897 (__BASE_TWO_BYTE_LO(c, NATIVE_TO_UNI)))
899 /* This is illegal in any well-formed UTF-8 in both EBCDIC and ASCII
900 * as it is only in overlongs. */
901 #define ILLEGAL_UTF8_BYTE I8_TO_NATIVE_UTF8(0xC1)
904 * 'UTF' is whether or not p is encoded in UTF8. The names 'foo_lazy_if' stem
905 * from an earlier version of these macros in which they didn't call the
906 * foo_utf8() macros (i.e. were 'lazy') unless they decided that *p is the
907 * beginning of a utf8 character. Now that foo_utf8() determines that itself,
908 * no need to do it again here
910 #define isIDFIRST_lazy_if_safe(p, e, UTF) \
911 ((IN_BYTES || !UTF) \
913 : isIDFIRST_utf8_safe(p, e))
914 #define isWORDCHAR_lazy_if_safe(p, e, UTF) \
915 ((IN_BYTES || !UTF) \
917 : isWORDCHAR_utf8_safe((U8 *) p, (U8 *) e))
918 #define isALNUM_lazy_if_safe(p, e, UTF) isWORDCHAR_lazy_if_safe(p, e, UTF)
920 #define UTF8_MAXLEN UTF8_MAXBYTES
922 /* A Unicode character can fold to up to 3 characters */
923 #define UTF8_MAX_FOLD_CHAR_EXPAND 3
925 #define IN_BYTES UNLIKELY(CopHINTS_get(PL_curcop) & HINT_BYTES)
929 =for apidoc Am|bool|DO_UTF8|SV* sv
930 Returns a bool giving whether or not the PV in C<sv> is to be treated as being
933 You should use this I<after> a call to C<SvPV()> or one of its variants, in
934 case any call to string overloading updates the internal UTF-8 encoding flag.
938 #define DO_UTF8(sv) (SvUTF8(sv) && !IN_BYTES)
940 /* Should all strings be treated as Unicode, and not just UTF-8 encoded ones?
941 * Is so within 'feature unicode_strings' or 'locale :not_characters', and not
942 * within 'use bytes'. UTF-8 locales are not tested for here, because it gets
943 * complicated by the probability of having categories in different locales. */
944 #define IN_UNI_8_BIT \
945 (( ( (CopHINTS_get(PL_curcop) & HINT_UNI_8_BIT)) \
946 || ( CopHINTS_get(PL_curcop) & HINT_LOCALE_PARTIAL \
947 /* -1 below is for :not_characters */ \
948 && _is_in_locale_category(FALSE, -1))) \
951 #define UNICODE_SURROGATE_FIRST 0xD800
952 #define UNICODE_SURROGATE_LAST 0xDFFF
955 =for apidoc Am|bool|UNICODE_IS_SURROGATE|const UV uv
957 Returns a boolean as to whether or not C<uv> is one of the Unicode surrogate
960 =for apidoc Am|bool|UTF8_IS_SURROGATE|const U8 *s|const U8 *e
962 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
963 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
964 of the Unicode surrogate code points; otherwise it evaluates to 0. If
965 non-zero, the value gives how many bytes starting at C<s> comprise the code
966 point's representation.
971 #define UNICODE_IS_SURROGATE(uv) UNLIKELY(inRANGE(uv, UNICODE_SURROGATE_FIRST, \
972 UNICODE_SURROGATE_LAST))
973 #define UTF8_IS_SURROGATE(s, e) is_SURROGATE_utf8_safe(s, e)
977 =for apidoc AmnU|UV|UNICODE_REPLACEMENT
979 Evaluates to 0xFFFD, the code point of the Unicode REPLACEMENT CHARACTER
981 =for apidoc Am|bool|UNICODE_IS_REPLACEMENT|const UV uv
983 Returns a boolean as to whether or not C<uv> is the Unicode REPLACEMENT
986 =for apidoc Am|bool|UTF8_IS_REPLACEMENT|const U8 *s|const U8 *e
988 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
989 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents the
990 Unicode REPLACEMENT CHARACTER; otherwise it evaluates to 0. If non-zero, the
991 value gives how many bytes starting at C<s> comprise the code point's
996 #define UNICODE_REPLACEMENT 0xFFFD
997 #define UNICODE_IS_REPLACEMENT(uv) UNLIKELY((UV) (uv) == UNICODE_REPLACEMENT)
998 #define UTF8_IS_REPLACEMENT(s, send) \
1000 ((send) - (s)) >= ((SSize_t)(sizeof(REPLACEMENT_CHARACTER_UTF8) - 1))\
1001 && memEQ((s), REPLACEMENT_CHARACTER_UTF8, \
1002 sizeof(REPLACEMENT_CHARACTER_UTF8) - 1))
1004 /* Max legal code point according to Unicode */
1005 #define PERL_UNICODE_MAX 0x10FFFF
1009 =for apidoc Am|bool|UNICODE_IS_SUPER|const UV uv
1011 Returns a boolean as to whether or not C<uv> is above the maximum legal Unicode
1012 code point of U+10FFFF.
1017 #define UNICODE_IS_SUPER(uv) UNLIKELY((UV) (uv) > PERL_UNICODE_MAX)
1020 =for apidoc Am|bool|UTF8_IS_SUPER|const U8 *s|const U8 *e
1022 Recall that Perl recognizes an extension to UTF-8 that can encode code
1023 points larger than the ones defined by Unicode, which are 0..0x10FFFF.
1025 This macro evaluates to non-zero if the first few bytes of the string starting
1026 at C<s> and looking no further than S<C<e - 1>> are from this UTF-8 extension;
1027 otherwise it evaluates to 0. If non-zero, the return is how many bytes
1028 starting at C<s> comprise the code point's representation.
1030 0 is returned if the bytes are not well-formed extended UTF-8, or if they
1031 represent a code point that cannot fit in a UV on the current platform. Hence
1032 this macro can give different results when run on a 64-bit word machine than on
1033 one with a 32-bit word size.
1035 Note that it is illegal in Perl to have code points that are larger than what can
1036 fit in an IV on the current machine; and illegal in Unicode to have any that
1042 * U+10FFFF: \xF4\x8F\xBF\xBF \xF9\xA1\xBF\xBF\xBF max legal Unicode
1043 * U+110000: \xF4\x90\x80\x80 \xF9\xA2\xA0\xA0\xA0
1044 * U+110001: \xF4\x90\x80\x81 \xF9\xA2\xA0\xA0\xA1
1046 #define UTF_START_BYTE_110000_ UTF_START_BYTE(PERL_UNICODE_MAX + 1, 21)
1047 #define UTF_FIRST_CONT_BYTE_110000_ \
1048 UTF_FIRST_CONT_BYTE(PERL_UNICODE_MAX + 1, 21)
1049 #define UTF8_IS_SUPER(s, e) \
1050 ( ((e) - (s)) >= UNISKIP_BY_MSB_(20) \
1051 && ( NATIVE_UTF8_TO_I8(s[0]) >= UTF_START_BYTE_110000_ \
1052 && ( NATIVE_UTF8_TO_I8(s[0]) > UTF_START_BYTE_110000_ \
1053 || NATIVE_UTF8_TO_I8(s[1]) >= UTF_FIRST_CONT_BYTE_110000_))) \
1054 ? isUTF8_CHAR(s, e) \
1058 =for apidoc Am|bool|UNICODE_IS_NONCHAR|const UV uv
1060 Returns a boolean as to whether or not C<uv> is one of the Unicode
1061 non-character code points
1066 /* Is 'uv' one of the 32 contiguous-range noncharacters? */
1067 #define UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv) \
1068 UNLIKELY(inRANGE(uv, 0xFDD0, 0xFDEF))
1070 /* Is 'uv' one of the 34 plane-ending noncharacters 0xFFFE, 0xFFFF, 0x1FFFE,
1071 * 0x1FFFF, ... 0x10FFFE, 0x10FFFF, given that we know that 'uv' is not above
1072 * the Unicode legal max */
1073 #define UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv) \
1074 UNLIKELY(((UV) (uv) & 0xFFFE) == 0xFFFE)
1076 #define UNICODE_IS_NONCHAR(uv) \
1077 ( UNLIKELY(UNICODE_IS_32_CONTIGUOUS_NONCHARS(uv)) \
1078 || ( UNLIKELY(UNICODE_IS_END_PLANE_NONCHAR_GIVEN_NOT_SUPER(uv)) \
1079 && LIKELY(! UNICODE_IS_SUPER(uv))))
1082 =for apidoc Am|bool|UTF8_IS_NONCHAR|const U8 *s|const U8 *e
1084 Evaluates to non-zero if the first few bytes of the string starting at C<s> and
1085 looking no further than S<C<e - 1>> are well-formed UTF-8 that represents one
1086 of the Unicode non-character code points; otherwise it evaluates to 0. If
1087 non-zero, the value gives how many bytes starting at C<s> comprise the code
1088 point's representation.
1092 #define UTF8_IS_NONCHAR(s, e) is_NONCHAR_utf8_safe(s,e)
1094 /* This is now machine generated, and the 'given' clause is no longer
1096 #define UTF8_IS_NONCHAR_GIVEN_THAT_NON_SUPER_AND_GE_PROBLEMATIC(s, e) \
1097 UTF8_IS_NONCHAR(s, e)
1099 /* Surrogates, non-character code points and above-Unicode code points are
1100 * problematic in some contexts. These macros allow code that needs to check
1101 * for those to quickly exclude the vast majority of code points it will
1104 * The lowest such code point is the smallest surrogate, U+D800. We calculate
1105 * the start byte of that. 0xD800 occupies 16 bits. */
1106 #define isUNICODE_POSSIBLY_PROBLEMATIC(uv) ((uv) >= UNICODE_SURROGATE_FIRST)
1107 #define isUTF8_POSSIBLY_PROBLEMATIC(c) \
1108 (NATIVE_UTF8_TO_I8(c) >= UTF_START_BYTE(UNICODE_SURROGATE_FIRST, 16))
1110 /* Perl extends Unicode so that it is possible to encode (as extended UTF-8 or
1111 * UTF-EBCDIC) any 64-bit value. No standard known to khw ever encoded higher
1112 * than a 31 bit value. On ASCII platforms this just meant arbitrarily saying
1113 * nothing could be higher than this. On these the start byte FD gets you to
1114 * 31 bits, and FE and FF are forbidden as start bytes. On EBCDIC platforms,
1115 * FD gets you only to 26 bits; adding FE to mean 7 total bytes gets you to 30
1116 * bits. To get to 31 bits, they treated an initial FF byte idiosyncratically.
1117 * It was considered to be the start byte FE meaning it had 7 total bytes, and
1118 * the final 1 was treated as an information bit, getting you to 31 bits.
1120 * Perl used to accept this idiosyncratic interpretation of FF, but now rejects
1121 * it in order to get to being able to encode 64 bits. The bottom line is that
1122 * it is a Perl extension to use the start bytes FE and FF on ASCII platforms,
1123 * and the start byte FF on EBCDIC ones. That translates into that it is a
1124 * Perl extension to represent anything occupying more than 31 bits on ASCII
1125 * platforms; 30 bits on EBCDIC. */
1126 #define UNICODE_IS_PERL_EXTENDED(uv) \
1127 UNLIKELY((UV) (uv) > nBIT_UMAX(31 - ONE_IF_EBCDIC_ZERO_IF_NOT))
1128 #define UTF8_IS_PERL_EXTENDED(s) \
1129 (UTF8SKIP(s) > 6 + ONE_IF_EBCDIC_ZERO_IF_NOT)
1131 /* Largest code point we accept from external sources */
1132 #define MAX_LEGAL_CP ((UV)IV_MAX)
1134 #define UTF8_ALLOW_EMPTY 0x0001 /* Allow a zero length string */
1135 #define UTF8_GOT_EMPTY UTF8_ALLOW_EMPTY
1137 /* Allow first byte to be a continuation byte */
1138 #define UTF8_ALLOW_CONTINUATION 0x0002
1139 #define UTF8_GOT_CONTINUATION UTF8_ALLOW_CONTINUATION
1141 /* Unexpected non-continuation byte */
1142 #define UTF8_ALLOW_NON_CONTINUATION 0x0004
1143 #define UTF8_GOT_NON_CONTINUATION UTF8_ALLOW_NON_CONTINUATION
1145 /* expecting more bytes than were available in the string */
1146 #define UTF8_ALLOW_SHORT 0x0008
1147 #define UTF8_GOT_SHORT UTF8_ALLOW_SHORT
1149 /* Overlong sequence; i.e., the code point can be specified in fewer bytes.
1150 * First one will convert the overlong to the REPLACEMENT CHARACTER; second
1151 * will return what the overlong evaluates to */
1152 #define UTF8_ALLOW_LONG 0x0010
1153 #define UTF8_ALLOW_LONG_AND_ITS_VALUE (UTF8_ALLOW_LONG|0x0020)
1154 #define UTF8_GOT_LONG UTF8_ALLOW_LONG
1156 #define UTF8_ALLOW_OVERFLOW 0x0080
1157 #define UTF8_GOT_OVERFLOW UTF8_ALLOW_OVERFLOW
1159 #define UTF8_DISALLOW_SURROGATE 0x0100 /* Unicode surrogates */
1160 #define UTF8_GOT_SURROGATE UTF8_DISALLOW_SURROGATE
1161 #define UTF8_WARN_SURROGATE 0x0200
1163 /* Unicode non-character code points */
1164 #define UTF8_DISALLOW_NONCHAR 0x0400
1165 #define UTF8_GOT_NONCHAR UTF8_DISALLOW_NONCHAR
1166 #define UTF8_WARN_NONCHAR 0x0800
1168 /* Super-set of Unicode: code points above the legal max */
1169 #define UTF8_DISALLOW_SUPER 0x1000
1170 #define UTF8_GOT_SUPER UTF8_DISALLOW_SUPER
1171 #define UTF8_WARN_SUPER 0x2000
1173 /* The original UTF-8 standard did not define UTF-8 with start bytes of 0xFE or
1174 * 0xFF, though UTF-EBCDIC did. This allowed both versions to represent code
1175 * points up to 2 ** 31 - 1. Perl extends UTF-8 so that 0xFE and 0xFF are
1176 * usable on ASCII platforms, and 0xFF means something different than
1177 * UTF-EBCDIC defines. These changes allow code points of 64 bits (actually
1178 * somewhat more) to be represented on both platforms. But these are Perl
1179 * extensions, and not likely to be interchangeable with other languages. Note
1180 * that on ASCII platforms, FE overflows a signed 32-bit word, and FF an
1182 #define UTF8_DISALLOW_PERL_EXTENDED 0x4000
1183 #define UTF8_GOT_PERL_EXTENDED UTF8_DISALLOW_PERL_EXTENDED
1184 #define UTF8_WARN_PERL_EXTENDED 0x8000
1186 /* For back compat, these old names are misleading for overlongs and
1188 #define UTF8_DISALLOW_ABOVE_31_BIT UTF8_DISALLOW_PERL_EXTENDED
1189 #define UTF8_GOT_ABOVE_31_BIT UTF8_GOT_PERL_EXTENDED
1190 #define UTF8_WARN_ABOVE_31_BIT UTF8_WARN_PERL_EXTENDED
1191 #define UTF8_DISALLOW_FE_FF UTF8_DISALLOW_PERL_EXTENDED
1192 #define UTF8_WARN_FE_FF UTF8_WARN_PERL_EXTENDED
1194 #define UTF8_CHECK_ONLY 0x10000
1195 #define _UTF8_NO_CONFIDENCE_IN_CURLEN 0x20000 /* Internal core use only */
1197 /* For backwards source compatibility. They do nothing, as the default now
1198 * includes what they used to mean. The first one's meaning was to allow the
1199 * just the single non-character 0xFFFF */
1200 #define UTF8_ALLOW_FFFF 0
1201 #define UTF8_ALLOW_FE_FF 0
1202 #define UTF8_ALLOW_SURROGATE 0
1204 /* C9 refers to Unicode Corrigendum #9: allows but discourages non-chars */
1205 #define UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE \
1206 (UTF8_DISALLOW_SUPER|UTF8_DISALLOW_SURROGATE)
1207 #define UTF8_WARN_ILLEGAL_C9_INTERCHANGE (UTF8_WARN_SUPER|UTF8_WARN_SURROGATE)
1209 #define UTF8_DISALLOW_ILLEGAL_INTERCHANGE \
1210 (UTF8_DISALLOW_ILLEGAL_C9_INTERCHANGE|UTF8_DISALLOW_NONCHAR)
1211 #define UTF8_WARN_ILLEGAL_INTERCHANGE \
1212 (UTF8_WARN_ILLEGAL_C9_INTERCHANGE|UTF8_WARN_NONCHAR)
1214 /* This is typically used for code that processes UTF-8 input and doesn't want
1215 * to have to deal with any malformations that might be present. All such will
1216 * be safely replaced by the REPLACEMENT CHARACTER, unless other flags
1217 * overriding this are also present. */
1218 #define UTF8_ALLOW_ANY ( UTF8_ALLOW_CONTINUATION \
1219 |UTF8_ALLOW_NON_CONTINUATION \
1222 |UTF8_ALLOW_OVERFLOW)
1224 /* Accept any Perl-extended UTF-8 that evaluates to any UV on the platform, but
1225 * not any malformed. This is the default. */
1226 #define UTF8_ALLOW_ANYUV 0
1227 #define UTF8_ALLOW_DEFAULT UTF8_ALLOW_ANYUV
1229 #define UNICODE_WARN_SURROGATE 0x0001 /* UTF-16 surrogates */
1230 #define UNICODE_WARN_NONCHAR 0x0002 /* Non-char code points */
1231 #define UNICODE_WARN_SUPER 0x0004 /* Above 0x10FFFF */
1232 #define UNICODE_WARN_PERL_EXTENDED 0x0008 /* Above 0x7FFF_FFFF */
1233 #define UNICODE_WARN_ABOVE_31_BIT UNICODE_WARN_PERL_EXTENDED
1234 #define UNICODE_DISALLOW_SURROGATE 0x0010
1235 #define UNICODE_DISALLOW_NONCHAR 0x0020
1236 #define UNICODE_DISALLOW_SUPER 0x0040
1237 #define UNICODE_DISALLOW_PERL_EXTENDED 0x0080
1240 # define UNICODE_ALLOW_ABOVE_IV_MAX 0x0100
1242 #define UNICODE_DISALLOW_ABOVE_31_BIT UNICODE_DISALLOW_PERL_EXTENDED
1244 #define UNICODE_GOT_SURROGATE UNICODE_DISALLOW_SURROGATE
1245 #define UNICODE_GOT_NONCHAR UNICODE_DISALLOW_NONCHAR
1246 #define UNICODE_GOT_SUPER UNICODE_DISALLOW_SUPER
1247 #define UNICODE_GOT_PERL_EXTENDED UNICODE_DISALLOW_PERL_EXTENDED
1249 #define UNICODE_WARN_ILLEGAL_C9_INTERCHANGE \
1250 (UNICODE_WARN_SURROGATE|UNICODE_WARN_SUPER)
1251 #define UNICODE_WARN_ILLEGAL_INTERCHANGE \
1252 (UNICODE_WARN_ILLEGAL_C9_INTERCHANGE|UNICODE_WARN_NONCHAR)
1253 #define UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE \
1254 (UNICODE_DISALLOW_SURROGATE|UNICODE_DISALLOW_SUPER)
1255 #define UNICODE_DISALLOW_ILLEGAL_INTERCHANGE \
1256 (UNICODE_DISALLOW_ILLEGAL_C9_INTERCHANGE|UNICODE_DISALLOW_NONCHAR)
1258 /* For backward source compatibility, as are now the default */
1259 #define UNICODE_ALLOW_SURROGATE 0
1260 #define UNICODE_ALLOW_SUPER 0
1261 #define UNICODE_ALLOW_ANY 0
1263 #define UNICODE_BYTE_ORDER_MARK 0xFEFF
1264 #define UNICODE_IS_BYTE_ORDER_MARK(uv) UNLIKELY((UV) (uv) \
1265 == UNICODE_BYTE_ORDER_MARK)
1267 #define LATIN_SMALL_LETTER_SHARP_S LATIN_SMALL_LETTER_SHARP_S_NATIVE
1268 #define LATIN_SMALL_LETTER_Y_WITH_DIAERESIS \
1269 LATIN_SMALL_LETTER_Y_WITH_DIAERESIS_NATIVE
1270 #define MICRO_SIGN MICRO_SIGN_NATIVE
1271 #define LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE \
1272 LATIN_CAPITAL_LETTER_A_WITH_RING_ABOVE_NATIVE
1273 #define LATIN_SMALL_LETTER_A_WITH_RING_ABOVE \
1274 LATIN_SMALL_LETTER_A_WITH_RING_ABOVE_NATIVE
1275 #define UNICODE_GREEK_CAPITAL_LETTER_SIGMA 0x03A3
1276 #define UNICODE_GREEK_SMALL_LETTER_FINAL_SIGMA 0x03C2
1277 #define UNICODE_GREEK_SMALL_LETTER_SIGMA 0x03C3
1278 #define GREEK_SMALL_LETTER_MU 0x03BC
1279 #define GREEK_CAPITAL_LETTER_MU 0x039C /* Upper and title case
1281 #define LATIN_CAPITAL_LETTER_Y_WITH_DIAERESIS 0x0178 /* Also is title case */
1282 #ifdef LATIN_CAPITAL_LETTER_SHARP_S_UTF8
1283 # define LATIN_CAPITAL_LETTER_SHARP_S 0x1E9E
1285 #define LATIN_CAPITAL_LETTER_I_WITH_DOT_ABOVE 0x130
1286 #define LATIN_SMALL_LETTER_DOTLESS_I 0x131
1287 #define LATIN_SMALL_LETTER_LONG_S 0x017F
1288 #define LATIN_SMALL_LIGATURE_LONG_S_T 0xFB05
1289 #define LATIN_SMALL_LIGATURE_ST 0xFB06
1290 #define KELVIN_SIGN 0x212A
1291 #define ANGSTROM_SIGN 0x212B
1293 #define UNI_DISPLAY_ISPRINT 0x0001
1294 #define UNI_DISPLAY_BACKSLASH 0x0002
1295 #define UNI_DISPLAY_BACKSPACE 0x0004 /* Allow \b when also
1296 UNI_DISPLAY_BACKSLASH */
1297 #define UNI_DISPLAY_QQ (UNI_DISPLAY_ISPRINT \
1298 |UNI_DISPLAY_BACKSLASH \
1299 |UNI_DISPLAY_BACKSPACE)
1301 /* Character classes could also allow \b, but not patterns in general */
1302 #define UNI_DISPLAY_REGEX (UNI_DISPLAY_ISPRINT|UNI_DISPLAY_BACKSLASH)
1304 /* Should be removed; maybe deprecated, but not used in CPAN */
1305 #define SHARP_S_SKIP 2
1307 #define is_utf8_char_buf(buf, buf_end) isUTF8_CHAR(buf, buf_end)
1308 #define bytes_from_utf8(s, lenp, is_utf8p) \
1309 bytes_from_utf8_loc(s, lenp, is_utf8p, 0)
1311 /* Do not use; should be deprecated. Use isUTF8_CHAR() instead; this is
1312 * retained solely for backwards compatibility */
1313 #define IS_UTF8_CHAR(p, n) (isUTF8_CHAR(p, (p) + (n)) == n)
1315 #endif /* PERL_UTF8_H_ */
1318 * ex: set ts=8 sts=4 sw=4 et: