3 perlunicode - Unicode support in Perl
7 =head2 Important Caveats
9 Unicode support is an extensive requirement. While Perl does not
10 implement the Unicode standard or the accompanying technical reports
11 from cover to cover, Perl does support many Unicode features.
13 People who want to learn to use Unicode in Perl, should probably read
14 the L<Perl Unicode tutorial, perlunitut|perlunitut> and
15 L<perluniintro>, before reading
16 this reference document.
18 Also, the use of Unicode may present security issues that aren't obvious.
19 Read L<Unicode Security Considerations|http://www.unicode.org/reports/tr36>.
23 =item Safest if you "use feature 'unicode_strings'"
25 In order to preserve backward compatibility, Perl does not turn
26 on full internal Unicode support unless the pragma
27 C<use feature 'unicode_strings'> is specified. (This is automatically
28 selected if you use C<use 5.012> or higher.) Failure to do this can
29 trigger unexpected surprises. See L</The "Unicode Bug"> below.
31 This pragma doesn't affect I/O, and there are still several places
32 where Unicode isn't fully supported, such as in filenames.
34 =item Input and Output Layers
36 Perl knows when a filehandle uses Perl's internal Unicode encodings
37 (UTF-8, or UTF-EBCDIC if in EBCDIC) if the filehandle is opened with
38 the ":encoding(utf8)" layer. Other encodings can be converted to Perl's
39 encoding on input or from Perl's encoding on output by use of the
40 ":encoding(...)" layer. See L<open>.
42 To indicate that Perl source itself is in UTF-8, use C<use utf8;>.
44 =item C<use utf8> still needed to enable UTF-8/UTF-EBCDIC in scripts
46 As a compatibility measure, the C<use utf8> pragma must be explicitly
47 included to enable recognition of UTF-8 in the Perl scripts themselves
48 (in string or regular expression literals, or in identifier names) on
49 ASCII-based machines or to recognize UTF-EBCDIC on EBCDIC-based
50 machines. B<These are the only times when an explicit C<use utf8>
51 is needed.> See L<utf8>.
53 =item BOM-marked scripts and UTF-16 scripts autodetected
55 If a Perl script begins marked with the Unicode BOM (UTF-16LE, UTF16-BE,
56 or UTF-8), or if the script looks like non-BOM-marked UTF-16 of either
57 endianness, Perl will correctly read in the script as Unicode.
58 (BOMless UTF-8 cannot be effectively recognized or differentiated from
59 ISO 8859-1 or other eight-bit encodings.)
61 =item C<use encoding> needed to upgrade non-Latin-1 byte strings
63 By default, there is a fundamental asymmetry in Perl's Unicode model:
64 implicit upgrading from byte strings to Unicode strings assumes that
65 they were encoded in I<ISO 8859-1 (Latin-1)>, but Unicode strings are
66 downgraded with UTF-8 encoding. This happens because the first 256
67 codepoints in Unicode happens to agree with Latin-1.
69 See L</"Byte and Character Semantics"> for more details.
73 =head2 Byte and Character Semantics
75 Beginning with version 5.6, Perl uses logically-wide characters to
76 represent strings internally.
78 Starting in Perl 5.14, Perl-level operations work with
79 characters rather than bytes within the scope of a
80 C<L<use feature 'unicode_strings'|feature>> (or equivalently
81 C<use 5.012> or higher). (This is not true if bytes have been
82 explicitly requested by C<L<use bytes|bytes>>, nor necessarily true
83 for interactions with the platform's operating system.)
85 For earlier Perls, and when C<unicode_strings> is not in effect, Perl
86 provides a fairly safe environment that can handle both types of
87 semantics in programs. For operations where Perl can unambiguously
88 decide that the input data are characters, Perl switches to character
89 semantics. For operations where this determination cannot be made
90 without additional information from the user, Perl decides in favor of
91 compatibility and chooses to use byte semantics.
93 When C<use locale> is in effect (which overrides
94 C<use feature 'unicode_strings'> in the same scope), Perl uses the
96 with the current locale. Otherwise, Perl uses the platform's native
97 byte semantics for characters whose code points are less than 256, and
98 Unicode semantics for those greater than 255. On EBCDIC platforms, this
99 is almost seamless, as the EBCDIC code pages that Perl handles are
100 equivalent to Unicode's first 256 code points. (The exception is that
101 EBCDIC regular expression case-insensitive matching rules are not as
102 as robust as Unicode's.) But on ASCII platforms, Perl uses US-ASCII
103 (or Basic Latin in Unicode terminology) byte semantics, meaning that characters
104 whose ordinal numbers are in the range 128 - 255 are undefined except for their
105 ordinal numbers. This means that none have case (upper and lower), nor are any
106 a member of character classes, like C<[:alpha:]> or C<\w>. (But all do belong
107 to the C<\W> class or the Perl regular expression extension C<[:^alpha:]>.)
109 This behavior preserves compatibility with earlier versions of Perl,
110 which allowed byte semantics in Perl operations only if
111 none of the program's inputs were marked as being a source of Unicode
112 character data. Such data may come from filehandles, from calls to
113 external programs, from information provided by the system (such as %ENV),
114 or from literals and constants in the source text.
116 The C<utf8> pragma is primarily a compatibility device that enables
117 recognition of UTF-(8|EBCDIC) in literals encountered by the parser.
118 Note that this pragma is only required while Perl defaults to byte
119 semantics; when character semantics become the default, this pragma
120 may become a no-op. See L<utf8>.
122 If strings operating under byte semantics and strings with Unicode
123 character data are concatenated, the new string will have
124 character semantics. This can cause surprises: See L</BUGS>, below.
125 You can choose to be warned when this happens. See L<encoding::warnings>.
127 Under character semantics, many operations that formerly operated on
128 bytes now operate on characters. A character in Perl is
129 logically just a number ranging from 0 to 2**31 or so. Larger
130 characters may encode into longer sequences of bytes internally, but
131 this internal detail is mostly hidden for Perl code.
132 See L<perluniintro> for more.
134 =head2 Effects of Character Semantics
136 Character semantics have the following effects:
142 Strings--including hash keys--and regular expression patterns may
143 contain characters that have an ordinal value larger than 255.
145 If you use a Unicode editor to edit your program, Unicode characters may
146 occur directly within the literal strings in UTF-8 encoding, or UTF-16.
147 (The former requires a BOM or C<use utf8>, the latter requires a BOM.)
149 Unicode characters can also be added to a string by using the C<\N{U+...}>
150 notation. The Unicode code for the desired character, in hexadecimal,
151 should be placed in the braces, after the C<U>. For instance, a smiley face is
154 Alternatively, you can use the C<\x{...}> notation for characters 0x100 and
155 above. For characters below 0x100 you may get byte semantics instead of
156 character semantics; see L</The "Unicode Bug">. On EBCDIC machines there is
157 the additional problem that the value for such characters gives the EBCDIC
158 character rather than the Unicode one.
162 use charnames ':full';
164 you can use the C<\N{...}> notation and put the official Unicode
165 character name within the braces, such as C<\N{WHITE SMILING FACE}>.
170 If an appropriate L<encoding> is specified, identifiers within the
171 Perl script may contain Unicode alphanumeric characters, including
172 ideographs. Perl does not currently attempt to canonicalize variable
177 Regular expressions match characters instead of bytes. "." matches
178 a character instead of a byte.
182 Bracketed character classes in regular expressions match characters instead of
183 bytes and match against the character properties specified in the
184 Unicode properties database. C<\w> can be used to match a Japanese
185 ideograph, for instance.
189 Named Unicode properties, scripts, and block ranges may be used (like bracketed
190 character classes) by using the C<\p{}> "matches property" construct and
191 the C<\P{}> negation, "doesn't match property".
192 See L</"Unicode Character Properties"> for more details.
194 You can define your own character properties and use them
195 in the regular expression with the C<\p{}> or C<\P{}> construct.
196 See L</"User-Defined Character Properties"> for more details.
200 The special pattern C<\X> matches a logical character, an "extended grapheme
201 cluster" in Standardese. In Unicode what appears to the user to be a single
202 character, for example an accented C<G>, may in fact be composed of a sequence
203 of characters, in this case a C<G> followed by an accent character. C<\X>
204 will match the entire sequence.
208 The C<tr///> operator translates characters instead of bytes. Note
209 that the C<tr///CU> functionality has been removed. For similar
210 functionality see pack('U0', ...) and pack('C0', ...).
214 Case translation operators use the Unicode case translation tables
215 when character input is provided. Note that C<uc()>, or C<\U> in
216 interpolated strings, translates to uppercase, while C<ucfirst>,
217 or C<\u> in interpolated strings, translates to titlecase in languages
218 that make the distinction (which is equivalent to uppercase in languages
219 without the distinction).
223 Most operators that deal with positions or lengths in a string will
224 automatically switch to using character positions, including
225 C<chop()>, C<chomp()>, C<substr()>, C<pos()>, C<index()>, C<rindex()>,
226 C<sprintf()>, C<write()>, and C<length()>. An operator that
227 specifically does not switch is C<vec()>. Operators that really don't
228 care include operators that treat strings as a bucket of bits such as
229 C<sort()>, and operators dealing with filenames.
233 The C<pack()>/C<unpack()> letter C<C> does I<not> change, since it is often
234 used for byte-oriented formats. Again, think C<char> in the C language.
236 There is a new C<U> specifier that converts between Unicode characters
237 and code points. There is also a C<W> specifier that is the equivalent of
238 C<chr>/C<ord> and properly handles character values even if they are above 255.
242 The C<chr()> and C<ord()> functions work on characters, similar to
243 C<pack("W")> and C<unpack("W")>, I<not> C<pack("C")> and
244 C<unpack("C")>. C<pack("C")> and C<unpack("C")> are methods for
245 emulating byte-oriented C<chr()> and C<ord()> on Unicode strings.
246 While these methods reveal the internal encoding of Unicode strings,
247 that is not something one normally needs to care about at all.
251 The bit string operators, C<& | ^ ~>, can operate on character data.
252 However, for backward compatibility, such as when using bit string
253 operations when characters are all less than 256 in ordinal value, one
254 should not use C<~> (the bit complement) with characters of both
255 values less than 256 and values greater than 256. Most importantly,
256 DeMorgan's laws (C<~($x|$y) eq ~$x&~$y> and C<~($x&$y) eq ~$x|~$y>)
257 will not hold. The reason for this mathematical I<faux pas> is that
258 the complement cannot return B<both> the 8-bit (byte-wide) bit
259 complement B<and> the full character-wide bit complement.
263 You can define your own mappings to be used in C<lc()>,
264 C<lcfirst()>, C<uc()>, and C<ucfirst()> (or their double-quoted string inlined
265 versions such as C<\U>). See
266 L<User-Defined Case-Mappings|/"User-Defined Case Mappings (for serious hackers only)">
275 And finally, C<scalar reverse()> reverses by character rather than by byte.
279 =head2 Unicode Character Properties
281 (The only time that Perl considers a sequence of individual code
282 points as a single logical character is in the C<\X> construct, already
283 mentioned above. Therefore "character" in this discussion means a single
286 Very nearly all Unicode character properties are accessible through
287 regular expressions by using the C<\p{}> "matches property" construct
288 and the C<\P{}> "doesn't match property" for its negation.
290 For instance, C<\p{Uppercase}> matches any single character with the Unicode
291 "Uppercase" property, while C<\p{L}> matches any character with a
292 General_Category of "L" (letter) property. Brackets are not
293 required for single letter property names, so C<\p{L}> is equivalent to C<\pL>.
295 More formally, C<\p{Uppercase}> matches any single character whose Unicode
296 Uppercase property value is True, and C<\P{Uppercase}> matches any character
297 whose Uppercase property value is False, and they could have been written as
298 C<\p{Uppercase=True}> and C<\p{Uppercase=False}>, respectively.
300 This formality is needed when properties are not binary; that is, if they can
301 take on more values than just True and False. For example, the Bidi_Class (see
302 L</"Bidirectional Character Types"> below), can take on several different
303 values, such as Left, Right, Whitespace, and others. To match these, one needs
304 to specify both the property name (Bidi_Class), AND the value being
306 (Left, Right, etc.). This is done, as in the examples above, by having the
307 two components separated by an equal sign (or interchangeably, a colon), like
308 C<\p{Bidi_Class: Left}>.
310 All Unicode-defined character properties may be written in these compound forms
311 of C<\p{property=value}> or C<\p{property:value}>, but Perl provides some
312 additional properties that are written only in the single form, as well as
313 single-form short-cuts for all binary properties and certain others described
314 below, in which you may omit the property name and the equals or colon
317 Most Unicode character properties have at least two synonyms (or aliases if you
318 prefer): a short one that is easier to type and a longer one that is more
319 descriptive and hence easier to understand. Thus the "L" and "Letter" properties
320 above are equivalent and can be used interchangeably. Likewise,
321 "Upper" is a synonym for "Uppercase", and we could have written
322 C<\p{Uppercase}> equivalently as C<\p{Upper}>. Also, there are typically
323 various synonyms for the values the property can be. For binary properties,
324 "True" has 3 synonyms: "T", "Yes", and "Y"; and "False has correspondingly "F",
325 "No", and "N". But be careful. A short form of a value for one property may
326 not mean the same thing as the same short form for another. Thus, for the
327 General_Category property, "L" means "Letter", but for the Bidi_Class property,
328 "L" means "Left". A complete list of properties and synonyms is in
331 Upper/lower case differences in property names and values are irrelevant;
332 thus C<\p{Upper}> means the same thing as C<\p{upper}> or even C<\p{UpPeR}>.
333 Similarly, you can add or subtract underscores anywhere in the middle of a
334 word, so that these are also equivalent to C<\p{U_p_p_e_r}>. And white space
335 is irrelevant adjacent to non-word characters, such as the braces and the equals
336 or colon separators, so C<\p{ Upper }> and C<\p{ Upper_case : Y }> are
337 equivalent to these as well. In fact, white space and even
338 hyphens can usually be added or deleted anywhere. So even C<\p{ Up-per case = Yes}> is
339 equivalent. All this is called "loose-matching" by Unicode. The few places
340 where stricter matching is used is in the middle of numbers, and in the Perl
341 extension properties that begin or end with an underscore. Stricter matching
342 cares about white space (except adjacent to non-word characters),
343 hyphens, and non-interior underscores.
345 You can also use negation in both C<\p{}> and C<\P{}> by introducing a caret
346 (^) between the first brace and the property name: C<\p{^Tamil}> is
347 equal to C<\P{Tamil}>.
349 Almost all properties are immune to case-insensitive matching. That is,
350 adding a C</i> regular expression modifier does not change what they
351 match. There are two sets that are affected.
355 and C<Titlecase_Letter>,
356 all of which match C<Cased_Letter> under C</i> matching.
357 And the second set is
361 all of which match C<Cased> under C</i> matching.
362 This set also includes its subsets C<PosixUpper> and C<PosixLower> both
363 of which under C</i> matching match C<PosixAlpha>.
364 (The difference between these sets is that some things, such as Roman
365 numerals, come in both upper and lower case so they are C<Cased>, but aren't considered
366 letters, so they aren't C<Cased_Letter>s.)
368 =head3 B<General_Category>
370 Every Unicode character is assigned a general category, which is the "most
371 usual categorization of a character" (from
372 L<http://www.unicode.org/reports/tr44>).
374 The compound way of writing these is like C<\p{General_Category=Number}>
375 (short, C<\p{gc:n}>). But Perl furnishes shortcuts in which everything up
376 through the equal or colon separator is omitted. So you can instead just write
379 Here are the short and long forms of the General Category properties:
384 LC, L& Cased_Letter (that is: [\p{Ll}\p{Lu}\p{Lt}])
397 Nd Decimal_Number (also Digit)
401 P Punctuation (also Punct)
402 Pc Connector_Punctuation
406 Pi Initial_Punctuation
407 (may behave like Ps or Pe depending on usage)
409 (may behave like Ps or Pe depending on usage)
421 Zp Paragraph_Separator
424 Cc Control (also Cntrl)
430 Single-letter properties match all characters in any of the
431 two-letter sub-properties starting with the same letter.
432 C<LC> and C<L&> are special: both are aliases for the set consisting of everything matched by C<Ll>, C<Lu>, and C<Lt>.
434 =head3 B<Bidirectional Character Types>
436 Because scripts differ in their directionality (Hebrew and Arabic are
437 written right to left, for example) Unicode supplies these properties in
438 the Bidi_Class class:
443 LRE Left-to-Right Embedding
444 LRO Left-to-Right Override
447 RLE Right-to-Left Embedding
448 RLO Right-to-Left Override
449 PDF Pop Directional Format
451 ES European Separator
452 ET European Terminator
457 B Paragraph Separator
462 This property is always written in the compound form.
463 For example, C<\p{Bidi_Class:R}> matches characters that are normally
464 written right to left.
468 The world's languages are written in many different scripts. This sentence
469 (unless you're reading it in translation) is written in Latin, while Russian is
470 written in Cyrillic, and Greek is written in, well, Greek; Japanese mainly in
471 Hiragana or Katakana. There are many more.
473 The Unicode Script and Script_Extensions properties give what script a
474 given character is in. Either property can be specified with the
476 C<\p{Script=Hebrew}> (short: C<\p{sc=hebr}>), or
477 C<\p{Script_Extensions=Javanese}> (short: C<\p{scx=java}>).
478 In addition, Perl furnishes shortcuts for all
479 C<Script> property names. You can omit everything up through the equals
480 (or colon), and simply write C<\p{Latin}> or C<\P{Cyrillic}>.
481 (This is not true for C<Script_Extensions>, which is required to be
482 written in the compound form.)
484 The difference between these two properties involves characters that are
485 used in multiple scripts. For example the digits '0' through '9' are
486 used in many parts of the world. These are placed in a script named
487 C<Common>. Other characters are used in just a few scripts. For
488 example, the "KATAKANA-HIRAGANA DOUBLE HYPHEN" is used in both Japanese
489 scripts, Katakana and Hiragana, but nowhere else. The C<Script>
490 property places all characters that are used in multiple scripts in the
491 C<Common> script, while the C<Script_Extensions> property places those
492 that are used in only a few scripts into each of those scripts; while
493 still using C<Common> for those used in many scripts. Thus both these
496 "0" =~ /\p{sc=Common}/ # Matches
497 "0" =~ /\p{scx=Common}/ # Matches
499 and only the first of these match:
501 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Common} # Matches
502 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Common} # No match
504 And only the last two of these match:
506 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Hiragana} # No match
507 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Katakana} # No match
508 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Hiragana} # Matches
509 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Katakana} # Matches
511 C<Script_Extensions> is thus an improved C<Script>, in which there are
512 fewer characters in the C<Common> script, and correspondingly more in
513 other scripts. It is new in Unicode version 6.0, and its data are likely
514 to change significantly in later releases, as things get sorted out.
516 (Actually, besides C<Common>, the C<Inherited> script, contains
517 characters that are used in multiple scripts. These are modifier
518 characters which modify other characters, and inherit the script value
519 of the controlling character. Some of these are used in many scripts,
520 and so go into C<Inherited> in both C<Script> and C<Script_Extensions>.
521 Others are used in just a few scripts, so are in C<Inherited> in
522 C<Script>, but not in C<Script_Extensions>.)
524 It is worth stressing that there are several different sets of digits in
525 Unicode that are equivalent to 0-9 and are matchable by C<\d> in a
526 regular expression. If they are used in a single language only, they
527 are in that language's C<Script> and C<Script_Extension>. If they are
528 used in more than one script, they will be in C<sc=Common>, but only
529 if they are used in many scripts should they be in C<scx=Common>.
531 A complete list of scripts and their shortcuts is in L<perluniprops>.
533 =head3 B<Use of "Is" Prefix>
535 For backward compatibility (with Perl 5.6), all properties mentioned
536 so far may have C<Is> or C<Is_> prepended to their name, so C<\P{Is_Lu}>, for
537 example, is equal to C<\P{Lu}>, and C<\p{IsScript:Arabic}> is equal to
542 In addition to B<scripts>, Unicode also defines B<blocks> of
543 characters. The difference between scripts and blocks is that the
544 concept of scripts is closer to natural languages, while the concept
545 of blocks is more of an artificial grouping based on groups of Unicode
546 characters with consecutive ordinal values. For example, the "Basic Latin"
547 block is all characters whose ordinals are between 0 and 127, inclusive; in
548 other words, the ASCII characters. The "Latin" script contains some letters
549 from this as well as several other blocks, like "Latin-1 Supplement",
550 "Latin Extended-A", etc., but it does not contain all the characters from
551 those blocks. It does not, for example, contain the digits 0-9, because
552 those digits are shared across many scripts, and hence are in the
555 For more about scripts versus blocks, see UAX#24 "Unicode Script Property":
556 L<http://www.unicode.org/reports/tr24>
558 The C<Script> or C<Script_Extensions> properties are likely to be the
559 ones you want to use when processing
560 natural language; the Block property may occasionally be useful in working
561 with the nuts and bolts of Unicode.
563 Block names are matched in the compound form, like C<\p{Block: Arrows}> or
564 C<\p{Blk=Hebrew}>. Unlike most other properties, only a few block names have a
565 Unicode-defined short name. But Perl does provide a (slight) shortcut: You
566 can say, for example C<\p{In_Arrows}> or C<\p{In_Hebrew}>. For backwards
567 compatibility, the C<In> prefix may be omitted if there is no naming conflict
568 with a script or any other property, and you can even use an C<Is> prefix
569 instead in those cases. But it is not a good idea to do this, for a couple
576 It is confusing. There are many naming conflicts, and you may forget some.
577 For example, C<\p{Hebrew}> means the I<script> Hebrew, and NOT the I<block>
578 Hebrew. But would you remember that 6 months from now?
582 It is unstable. A new version of Unicode may pre-empt the current meaning by
583 creating a property with the same name. There was a time in very early Unicode
584 releases when C<\p{Hebrew}> would have matched the I<block> Hebrew; now it
589 Some people prefer to always use C<\p{Block: foo}> and C<\p{Script: bar}>
590 instead of the shortcuts, whether for clarity, because they can't remember the
591 difference between 'In' and 'Is' anyway, or they aren't confident that those who
592 eventually will read their code will know that difference.
594 A complete list of blocks and their shortcuts is in L<perluniprops>.
596 =head3 B<Other Properties>
598 There are many more properties than the very basic ones described here.
599 A complete list is in L<perluniprops>.
601 Unicode defines all its properties in the compound form, so all single-form
602 properties are Perl extensions. Most of these are just synonyms for the
603 Unicode ones, but some are genuine extensions, including several that are in
604 the compound form. And quite a few of these are actually recommended by Unicode
605 (in L<http://www.unicode.org/reports/tr18>).
607 This section gives some details on all extensions that aren't just
608 synonyms for compound-form Unicode properties
609 (for those properties, you'll have to refer to the
610 L<Unicode Standard|http://www.unicode.org/reports/tr44>.
616 This matches any of the 1_114_112 Unicode code points. It is a synonym for
619 =item B<C<\p{Alnum}>>
621 This matches any C<\p{Alphabetic}> or C<\p{Decimal_Number}> character.
625 This matches any of the 1_114_112 Unicode code points. It is a synonym for
628 =item B<C<\p{ASCII}>>
630 This matches any of the 128 characters in the US-ASCII character set,
631 which is a subset of Unicode.
633 =item B<C<\p{Assigned}>>
635 This matches any assigned code point; that is, any code point whose general
636 category is not Unassigned (or equivalently, not Cn).
638 =item B<C<\p{Blank}>>
640 This is the same as C<\h> and C<\p{HorizSpace}>: A character that changes the
641 spacing horizontally.
643 =item B<C<\p{Decomposition_Type: Non_Canonical}>> (Short: C<\p{Dt=NonCanon}>)
645 Matches a character that has a non-canonical decomposition.
647 To understand the use of this rarely used property=value combination, it is
648 necessary to know some basics about decomposition.
649 Consider a character, say H. It could appear with various marks around it,
650 such as an acute accent, or a circumflex, or various hooks, circles, arrows,
651 I<etc.>, above, below, to one side or the other, etc. There are many
652 possibilities among the world's languages. The number of combinations is
653 astronomical, and if there were a character for each combination, it would
654 soon exhaust Unicode's more than a million possible characters. So Unicode
655 took a different approach: there is a character for the base H, and a
656 character for each of the possible marks, and these can be variously combined
657 to get a final logical character. So a logical character--what appears to be a
658 single character--can be a sequence of more than one individual characters.
659 This is called an "extended grapheme cluster"; Perl furnishes the C<\X>
660 regular expression construct to match such sequences.
662 But Unicode's intent is to unify the existing character set standards and
663 practices, and several pre-existing standards have single characters that
664 mean the same thing as some of these combinations. An example is ISO-8859-1,
665 which has quite a few of these in the Latin-1 range, an example being "LATIN
666 CAPITAL LETTER E WITH ACUTE". Because this character was in this pre-existing
667 standard, Unicode added it to its repertoire. But this character is considered
668 by Unicode to be equivalent to the sequence consisting of the character
669 "LATIN CAPITAL LETTER E" followed by the character "COMBINING ACUTE ACCENT".
671 "LATIN CAPITAL LETTER E WITH ACUTE" is called a "pre-composed" character, and
672 its equivalence with the sequence is called canonical equivalence. All
673 pre-composed characters are said to have a decomposition (into the equivalent
674 sequence), and the decomposition type is also called canonical.
676 However, many more characters have a different type of decomposition, a
677 "compatible" or "non-canonical" decomposition. The sequences that form these
678 decompositions are not considered canonically equivalent to the pre-composed
679 character. An example, again in the Latin-1 range, is the "SUPERSCRIPT ONE".
680 It is somewhat like a regular digit 1, but not exactly; its decomposition
681 into the digit 1 is called a "compatible" decomposition, specifically a
682 "super" decomposition. There are several such compatibility
683 decompositions (see L<http://www.unicode.org/reports/tr44>), including one
684 called "compat", which means some miscellaneous type of decomposition
685 that doesn't fit into the decomposition categories that Unicode has chosen.
687 Note that most Unicode characters don't have a decomposition, so their
688 decomposition type is "None".
690 For your convenience, Perl has added the C<Non_Canonical> decomposition
691 type to mean any of the several compatibility decompositions.
693 =item B<C<\p{Graph}>>
695 Matches any character that is graphic. Theoretically, this means a character
696 that on a printer would cause ink to be used.
698 =item B<C<\p{HorizSpace}>>
700 This is the same as C<\h> and C<\p{Blank}>: a character that changes the
701 spacing horizontally.
705 This is a synonym for C<\p{Present_In=*}>
707 =item B<C<\p{PerlSpace}>>
709 This is the same as C<\s>, restricted to ASCII, namely C<S<[ \f\n\r\t]>>.
711 Mnemonic: Perl's (original) space
713 =item B<C<\p{PerlWord}>>
715 This is the same as C<\w>, restricted to ASCII, namely C<[A-Za-z0-9_]>
717 Mnemonic: Perl's (original) word.
719 =item B<C<\p{Posix...}>>
721 There are several of these, which are equivalents using the C<\p>
722 notation for Posix classes and are described in
723 L<perlrecharclass/POSIX Character Classes>.
725 =item B<C<\p{Present_In: *}>> (Short: C<\p{In=*}>)
727 This property is used when you need to know in what Unicode version(s) a
730 The "*" above stands for some two digit Unicode version number, such as
731 C<1.1> or C<4.0>; or the "*" can also be C<Unassigned>. This property will
732 match the code points whose final disposition has been settled as of the
733 Unicode release given by the version number; C<\p{Present_In: Unassigned}>
734 will match those code points whose meaning has yet to be assigned.
736 For example, C<U+0041> "LATIN CAPITAL LETTER A" was present in the very first
737 Unicode release available, which is C<1.1>, so this property is true for all
738 valid "*" versions. On the other hand, C<U+1EFF> was not assigned until version
739 5.1 when it became "LATIN SMALL LETTER Y WITH LOOP", so the only "*" that
740 would match it are 5.1, 5.2, and later.
742 Unicode furnishes the C<Age> property from which this is derived. The problem
743 with Age is that a strict interpretation of it (which Perl takes) has it
744 matching the precise release a code point's meaning is introduced in. Thus
745 C<U+0041> would match only 1.1; and C<U+1EFF> only 5.1. This is not usually what
748 Some non-Perl implementations of the Age property may change its meaning to be
749 the same as the Perl Present_In property; just be aware of that.
751 Another confusion with both these properties is that the definition is not
752 that the code point has been I<assigned>, but that the meaning of the code point
753 has been I<determined>. This is because 66 code points will always be
754 unassigned, and so the Age for them is the Unicode version in which the decision
755 to make them so was made. For example, C<U+FDD0> is to be permanently
756 unassigned to a character, and the decision to do that was made in version 3.1,
757 so C<\p{Age=3.1}> matches this character, as also does C<\p{Present_In: 3.1}> and up.
759 =item B<C<\p{Print}>>
761 This matches any character that is graphical or blank, except controls.
763 =item B<C<\p{SpacePerl}>>
765 This is the same as C<\s>, including beyond ASCII.
767 Mnemonic: Space, as modified by Perl. (It doesn't include the vertical tab
768 which both the Posix standard and Unicode consider white space.)
770 =item B<C<\p{VertSpace}>>
772 This is the same as C<\v>: A character that changes the spacing vertically.
776 This is the same as C<\w>, including over 100_000 characters beyond ASCII.
778 =item B<C<\p{XPosix...}>>
780 There are several of these, which are the standard Posix classes
781 extended to the full Unicode range. They are described in
782 L<perlrecharclass/POSIX Character Classes>.
786 =head2 User-Defined Character Properties
788 You can define your own binary character properties by defining subroutines
789 whose names begin with "In" or "Is". The subroutines can be defined in any
790 package. The user-defined properties can be used in the regular expression
791 C<\p> and C<\P> constructs; if you are using a user-defined property from a
792 package other than the one you are in, you must specify its package in the
793 C<\p> or C<\P> construct.
795 # assuming property Is_Foreign defined in Lang::
796 package main; # property package name required
797 if ($txt =~ /\p{Lang::IsForeign}+/) { ... }
799 package Lang; # property package name not required
800 if ($txt =~ /\p{IsForeign}+/) { ... }
803 Note that the effect is compile-time and immutable once defined.
804 However, the subroutines are passed a single parameter, which is 0 if
805 case-sensitive matching is in effect and non-zero if caseless matching
806 is in effect. The subroutine may return different values depending on
807 the value of the flag, and one set of values will immutably be in effect
808 for all case-sensitive matches, and the other set for all case-insensitive
811 Note that if the regular expression is tainted, then Perl will die rather
812 than calling the subroutine, where the name of the subroutine is
813 determined by the tainted data.
815 The subroutines must return a specially-formatted string, with one
816 or more newline-separated lines. Each line must be one of the following:
822 A single hexadecimal number denoting a Unicode code point to include.
826 Two hexadecimal numbers separated by horizontal whitespace (space or
827 tabular characters) denoting a range of Unicode code points to include.
831 Something to include, prefixed by "+": a built-in character
832 property (prefixed by "utf8::") or a user-defined character property,
833 to represent all the characters in that property; two hexadecimal code
834 points for a range; or a single hexadecimal code point.
838 Something to exclude, prefixed by "-": an existing character
839 property (prefixed by "utf8::") or a user-defined character property,
840 to represent all the characters in that property; two hexadecimal code
841 points for a range; or a single hexadecimal code point.
845 Something to negate, prefixed "!": an existing character
846 property (prefixed by "utf8::") or a user-defined character property,
847 to represent all the characters in that property; two hexadecimal code
848 points for a range; or a single hexadecimal code point.
852 Something to intersect with, prefixed by "&": an existing character
853 property (prefixed by "utf8::") or a user-defined character property,
854 for all the characters except the characters in the property; two
855 hexadecimal code points for a range; or a single hexadecimal code point.
859 For example, to define a property that covers both the Japanese
860 syllabaries (hiragana and katakana), you can define
869 Imagine that the here-doc end marker is at the beginning of the line.
870 Now you can use C<\p{InKana}> and C<\P{InKana}>.
872 You could also have used the existing block property names:
881 Suppose you wanted to match only the allocated characters,
882 not the raw block ranges: in other words, you want to remove
893 The negation is useful for defining (surprise!) negated classes.
903 Intersection is useful for getting the common characters matched by
904 two (or more) classes.
913 It's important to remember not to use "&" for the first set; that
914 would be intersecting with nothing, resulting in an empty set.
916 =head2 User-Defined Case Mappings (for serious hackers only)
918 B<This featured is deprecated and is scheduled to be removed in Perl
920 The CPAN module L<Unicode::Casing> provides better functionality
921 without the drawbacks described below.
923 You can define your own mappings to be used in C<lc()>,
924 C<lcfirst()>, C<uc()>, and C<ucfirst()> (or their string-inlined versions,
925 C<\L>, C<\l>, C<\U>, and C<\u>). The mappings are currently only valid
926 on strings encoded in UTF-8, but see below for a partial workaround for
929 The principle is similar to that of user-defined character
930 properties: define subroutines that do the mappings.
931 C<ToLower> is used for C<lc()>, C<\L>, C<lcfirst()>, and C<\l>; C<ToTitle> for
932 C<ucfirst()> and C<\u>; and C<ToUpper> for C<uc()> and C<\U>.
934 C<ToUpper()> should look something like this:
943 This sample C<ToUpper()> has the effect of mapping "a-z" to "A-Z", 0x101
944 to 0x100, and all other characters map to themselves. The first
945 returned line means to map the code point at 0x61 ("a") to 0x41 ("A"),
946 the code point at 0x62 ("b") to 0x42 ("B"), ..., 0x7A ("z") to 0x5A
947 ("Z"). The second line maps just the code point 0x101 to 0x100. Since
948 there are no other mappings defined, all other code points map to
951 This mechanism is not well behaved as far as affecting other packages
952 and scopes. All non-threaded programs have exactly one uppercasing
953 behavior, one lowercasing behavior, and one titlecasing behavior in
954 effect for utf8-encoded strings for the duration of the program. Each
955 of these behaviors is irrevocably determined the first time the
956 corresponding function is called to change a utf8-encoded string's case.
957 If a corresponding C<To-> function has been defined in the package that
958 makes that first call, the mapping defined by that function will be the
959 mapping used for the duration of the program's execution across all
960 packages and scopes. If no corresponding C<To-> function has been
961 defined in that package, the standard official mapping will be used for
962 all packages and scopes, and any corresponding C<To-> function anywhere
963 will be ignored. Threaded programs have similar behavior. If the
964 program's casing behavior has been decided at the time of a thread's
965 creation, the thread will inherit that behavior. But, if the behavior
966 hasn't been decided, the thread gets to decide for itself, and its
967 decision does not affect other threads nor its creator.
969 As shown by the example above, you have to furnish a complete mapping;
970 you can't just override a couple of characters and leave the rest
971 unchanged. You can find all the official mappings in the directory
972 C<$Config{privlib}>F</unicore/To/>. The mapping data is returned as the
973 here-document. The C<utf8::ToSpecI<Foo>> hashes in those files are special
974 exception mappings derived from
975 C<$Config{privlib}>F</unicore/SpecialCasing.txt>. (The "Digit" and
976 "Fold" mappings that one can see in the directory are not directly
977 user-accessible, one can use either the L<Unicode::UCD> module, or just match
978 case-insensitively, which is what uses the "Fold" mapping. Neither are user
981 If you have many mappings to change, you can take the official mapping data,
982 change by hand the affected code points, and place the whole thing into your
983 subroutine. But this will only be valid on Perls that use the same Unicode
984 version. Another option would be to have your subroutine read the official
985 mapping files and overwrite the affected code points.
987 If you have only a few mappings to change, starting in 5.14 you can use the
988 following trick, here illustrated for Turkish.
991 use charnames ":full";
994 my $official = do "$Config{privlib}/unicore/To/Upper.pl";
995 $utf8::ToSpecUpper{'i'} =
996 "\N{LATIN CAPITAL LETTER I WITH DOT ABOVE}";
1000 This takes the official mappings and overrides just one, for "LATIN SMALL
1001 LETTER I". The keys to the hash must be the bytes that form the UTF-8
1002 (on EBCDIC platforms, UTF-EBCDIC) of the character, as illustrated by
1003 the inverse function.
1006 my $official = do $lower;
1007 $utf8::ToSpecLower{"\xc4\xb0"} = "i";
1011 This example is for an ASCII platform, and C<\xc4\xb0> is the string of
1012 bytes that together form the UTF-8 that represents C<\N{LATIN CAPITAL
1013 LETTER I WITH DOT ABOVE}>, C<U+0130>. You can avoid having to figure out
1014 these bytes, and at the same time make it work on all platforms by
1018 my $official = do $lower;
1019 my $sequence = "\N{LATIN CAPITAL LETTER I WITH DOT ABOVE}";
1020 utf8::encode($sequence);
1021 $utf8::ToSpecLower{$sequence} = "i";
1025 This works because C<utf8::encode()> takes the single character and
1026 converts it to the sequence of bytes that constitute it. Note that we took
1027 advantage of the fact that C<"i"> is the same in UTF-8 or UTF_EBCIDIC as not;
1028 otherwise we would have had to write
1030 $utf8::ToSpecLower{$sequence} = "\N{LATIN SMALL LETTER I}";
1032 in the ToLower example, and in the ToUpper example, use
1034 my $sequence = "\N{LATIN SMALL LETTER I}";
1035 utf8::encode($sequence);
1037 A big caveat to the above trick and to this whole mechanism in general,
1038 is that they work only on strings encoded in UTF-8. You can partially
1039 get around this by using C<use subs>. (But better to just convert to
1040 use L<Unicode::Casing>.) For example:
1041 (The trick illustrated here does work in earlier releases, but only if all the
1042 characters you want to override have ordinal values of 256 or higher, or
1043 if you use the other tricks given just below.)
1045 The mappings are in effect only for the package they are defined in, and only
1046 on scalars that have been marked as having Unicode characters, for example by
1047 using C<utf8::upgrade()>. Although probably not advisable, you can
1048 cause the mappings to be used globally by importing into C<CORE::GLOBAL>
1051 You can partially get around the restriction that the source strings
1052 must be in utf8 by using C<use subs> (or by importing into C<CORE::GLOBAL>) by:
1054 use subs qw(uc ucfirst lc lcfirst);
1058 utf8::upgrade($string);
1059 return CORE::uc($string);
1064 utf8::upgrade($string);
1066 # Unless an I is before a dot_above, it turns into a dotless i.
1067 # (The character class with the combining classes matches non-above
1068 # marks following the I. Any number of these may be between the
1069 # 'I'and the dot_above, and the dot_above will still apply to the
1071 use charnames ":full";
1074 (?! [^\p{ccc=0}\p{ccc=Above}]* \N{COMBINING DOT ABOVE} )
1075 /\N{LATIN SMALL LETTER DOTLESS I}/gx;
1077 # But when the I is followed by a dot_above, remove the
1078 # dot_above so the end result will be i.
1080 ([^\p{ccc=0}\p{ccc=Above}]* )
1081 \N{COMBINING DOT ABOVE}
1083 return CORE::lc($string);
1086 These examples (also for Turkish) make sure the input is in UTF-8, and then
1087 call the corresponding official function, which will use the C<ToUpper()> and
1088 C<ToLower()> functions you have defined.
1089 (For Turkish, there are other required functions: C<ucfirst>, C<lcfirst>,
1090 and C<ToTitle>. These are very similar to the ones given above.)
1092 The reason this is only a partial fix is that it doesn't affect the C<\l>,
1093 C<\L>, C<\u>, and C<\U> case-change operations in regular expressions,
1094 which still require the source to be encoded in utf8 (see L</The "Unicode
1095 Bug">). (Again, use L<Unicode::Casing> instead.)
1097 The C<lc()> example shows how you can add context-dependent casing. Note
1098 that context-dependent casing suffers from the problem that the string
1099 passed to the casing function may not have sufficient context to make
1100 the proper choice. Also, it will not be called for C<\l>, C<\L>, C<\u>,
1103 =head2 Character Encodings for Input and Output
1107 =head2 Unicode Regular Expression Support Level
1109 The following list of Unicode supported features for regular expressions describes
1110 all features currently directly supported by core Perl. The references to "Level N"
1111 and the section numbers refer to the Unicode Technical Standard #18,
1112 "Unicode Regular Expressions", version 13, from August 2008.
1118 Level 1 - Basic Unicode Support
1120 RL1.1 Hex Notation - done [1]
1121 RL1.2 Properties - done [2][3]
1122 RL1.2a Compatibility Properties - done [4]
1123 RL1.3 Subtraction and Intersection - MISSING [5]
1124 RL1.4 Simple Word Boundaries - done [6]
1125 RL1.5 Simple Loose Matches - done [7]
1126 RL1.6 Line Boundaries - MISSING [8][9]
1127 RL1.7 Supplementary Code Points - done [10]
1131 [3] supports not only minimal list, but all Unicode character
1132 properties (see Unicode Character Properties above)
1133 [4] \d \D \s \S \w \W \X [:prop:] [:^prop:]
1134 [5] can use regular expression look-ahead [a] or
1135 user-defined character properties [b] to emulate set
1138 [7] note that Perl does Full case-folding in matching (but with
1139 bugs), not Simple: for example U+1F88 is equivalent to
1140 U+1F00 U+03B9, instead of just U+1F80. This difference
1141 matters mainly for certain Greek capital letters with certain
1142 modifiers: the Full case-folding decomposes the letter,
1143 while the Simple case-folding would map it to a single
1145 [8] should do ^ and $ also on U+000B (\v in C), FF (\f), CR
1146 (\r), CRLF (\r\n), NEL (U+0085), LS (U+2028), and PS
1147 (U+2029); should also affect <>, $., and script line
1148 numbers; should not split lines within CRLF [c] (i.e. there
1149 is no empty line between \r and \n)
1150 [9] Linebreaking conformant with UAX#14 "Unicode Line Breaking
1151 Algorithm" is available through the Unicode::LineBreaking
1153 [10] UTF-8/UTF-EBDDIC used in Perl allows not only U+10000 to
1154 U+10FFFF but also beyond U+10FFFF
1156 [a] You can mimic class subtraction using lookahead.
1157 For example, what UTS#18 might write as
1159 [{Greek}-[{UNASSIGNED}]]
1161 in Perl can be written as:
1163 (?!\p{Unassigned})\p{InGreekAndCoptic}
1164 (?=\p{Assigned})\p{InGreekAndCoptic}
1166 But in this particular example, you probably really want
1170 which will match assigned characters known to be part of the Greek script.
1172 Also see the L<Unicode::Regex::Set> module, it does implement the full
1173 UTS#18 grouping, intersection, union, and removal (subtraction) syntax.
1175 [b] '+' for union, '-' for removal (set-difference), '&' for intersection
1176 (see L</"User-Defined Character Properties">)
1178 [c] Try the C<:crlf> layer (see L<PerlIO>).
1182 Level 2 - Extended Unicode Support
1184 RL2.1 Canonical Equivalents - MISSING [10][11]
1185 RL2.2 Default Grapheme Clusters - MISSING [12]
1186 RL2.3 Default Word Boundaries - MISSING [14]
1187 RL2.4 Default Loose Matches - MISSING [15]
1188 RL2.5 Name Properties - DONE
1189 RL2.6 Wildcard Properties - MISSING
1191 [10] see UAX#15 "Unicode Normalization Forms"
1192 [11] have Unicode::Normalize but not integrated to regexes
1193 [12] have \X but we don't have a "Grapheme Cluster Mode"
1194 [14] see UAX#29, Word Boundaries
1195 [15] see UAX#21 "Case Mappings"
1199 Level 3 - Tailored Support
1201 RL3.1 Tailored Punctuation - MISSING
1202 RL3.2 Tailored Grapheme Clusters - MISSING [17][18]
1203 RL3.3 Tailored Word Boundaries - MISSING
1204 RL3.4 Tailored Loose Matches - MISSING
1205 RL3.5 Tailored Ranges - MISSING
1206 RL3.6 Context Matching - MISSING [19]
1207 RL3.7 Incremental Matches - MISSING
1208 ( RL3.8 Unicode Set Sharing )
1209 RL3.9 Possible Match Sets - MISSING
1210 RL3.10 Folded Matching - MISSING [20]
1211 RL3.11 Submatchers - MISSING
1213 [17] see UAX#10 "Unicode Collation Algorithms"
1214 [18] have Unicode::Collate but not integrated to regexes
1215 [19] have (?<=x) and (?=x), but look-aheads or look-behinds
1216 should see outside of the target substring
1217 [20] need insensitive matching for linguistic features other
1218 than case; for example, hiragana to katakana, wide and
1219 narrow, simplified Han to traditional Han (see UTR#30
1220 "Character Foldings")
1224 =head2 Unicode Encodings
1226 Unicode characters are assigned to I<code points>, which are abstract
1227 numbers. To use these numbers, various encodings are needed.
1235 UTF-8 is a variable-length (1 to 4 bytes), byte-order independent
1236 encoding. For ASCII (and we really do mean 7-bit ASCII, not another
1237 8-bit encoding), UTF-8 is transparent.
1239 The following table is from Unicode 3.2.
1241 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
1243 U+0000..U+007F 00..7F
1244 U+0080..U+07FF * C2..DF 80..BF
1245 U+0800..U+0FFF E0 * A0..BF 80..BF
1246 U+1000..U+CFFF E1..EC 80..BF 80..BF
1247 U+D000..U+D7FF ED 80..9F 80..BF
1248 U+D800..U+DFFF +++++ utf16 surrogates, not legal utf8 +++++
1249 U+E000..U+FFFF EE..EF 80..BF 80..BF
1250 U+10000..U+3FFFF F0 * 90..BF 80..BF 80..BF
1251 U+40000..U+FFFFF F1..F3 80..BF 80..BF 80..BF
1252 U+100000..U+10FFFF F4 80..8F 80..BF 80..BF
1254 Note the gaps marked by "*" before several of the byte entries above. These are
1255 caused by legal UTF-8 avoiding non-shortest encodings: it is technically
1256 possible to UTF-8-encode a single code point in different ways, but that is
1257 explicitly forbidden, and the shortest possible encoding should always be used
1258 (and that is what Perl does).
1260 Another way to look at it is via bits:
1262 Code Points 1st Byte 2nd Byte 3rd Byte 4th Byte
1265 00000bbbbbaaaaaa 110bbbbb 10aaaaaa
1266 ccccbbbbbbaaaaaa 1110cccc 10bbbbbb 10aaaaaa
1267 00000dddccccccbbbbbbaaaaaa 11110ddd 10cccccc 10bbbbbb 10aaaaaa
1269 As you can see, the continuation bytes all begin with "10", and the
1270 leading bits of the start byte tell how many bytes there are in the
1273 The original UTF-8 specification allowed up to 6 bytes, to allow
1274 encoding of numbers up to 0x7FFF_FFFF. Perl continues to allow those,
1275 and has extended that up to 13 bytes to encode code points up to what
1276 can fit in a 64-bit word. However, Perl will warn if you output any of
1277 these as being non-portable; and under strict UTF-8 input protocols,
1280 The Unicode non-character code points are also disallowed in UTF-8 in
1281 "open interchange". See L</Non-character code points>.
1287 Like UTF-8 but EBCDIC-safe, in the way that UTF-8 is ASCII-safe.
1291 UTF-16, UTF-16BE, UTF-16LE, Surrogates, and BOMs (Byte Order Marks)
1293 The followings items are mostly for reference and general Unicode
1294 knowledge, Perl doesn't use these constructs internally.
1296 Like UTF-8, UTF-16 is a variable-width encoding, but where
1297 UTF-8 uses 8-bit code units, UTF-16 uses 16-bit code units.
1298 All code points occupy either 2 or 4 bytes in UTF-16: code points
1299 C<U+0000..U+FFFF> are stored in a single 16-bit unit, and code
1300 points C<U+10000..U+10FFFF> in two 16-bit units. The latter case is
1301 using I<surrogates>, the first 16-bit unit being the I<high
1302 surrogate>, and the second being the I<low surrogate>.
1304 Surrogates are code points set aside to encode the C<U+10000..U+10FFFF>
1305 range of Unicode code points in pairs of 16-bit units. The I<high
1306 surrogates> are the range C<U+D800..U+DBFF> and the I<low surrogates>
1307 are the range C<U+DC00..U+DFFF>. The surrogate encoding is
1309 $hi = ($uni - 0x10000) / 0x400 + 0xD800;
1310 $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
1314 $uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
1316 Because of the 16-bitness, UTF-16 is byte-order dependent. UTF-16
1317 itself can be used for in-memory computations, but if storage or
1318 transfer is required either UTF-16BE (big-endian) or UTF-16LE
1319 (little-endian) encodings must be chosen.
1321 This introduces another problem: what if you just know that your data
1322 is UTF-16, but you don't know which endianness? Byte Order Marks, or
1323 BOMs, are a solution to this. A special character has been reserved
1324 in Unicode to function as a byte order marker: the character with the
1325 code point C<U+FEFF> is the BOM.
1327 The trick is that if you read a BOM, you will know the byte order,
1328 since if it was written on a big-endian platform, you will read the
1329 bytes C<0xFE 0xFF>, but if it was written on a little-endian platform,
1330 you will read the bytes C<0xFF 0xFE>. (And if the originating platform
1331 was writing in UTF-8, you will read the bytes C<0xEF 0xBB 0xBF>.)
1333 The way this trick works is that the character with the code point
1334 C<U+FFFE> is not supposed to be in input streams, so the
1335 sequence of bytes C<0xFF 0xFE> is unambiguously "BOM, represented in
1336 little-endian format" and cannot be C<U+FFFE>, represented in big-endian
1339 Surrogates have no meaning in Unicode outside their use in pairs to
1340 represent other code points. However, Perl allows them to be
1341 represented individually internally, for example by saying
1342 C<chr(0xD801)>, so that all code points, not just those valid for open
1344 representable. Unicode does define semantics for them, such as their
1345 General Category is "Cs". But because their use is somewhat dangerous,
1346 Perl will warn (using the warning category "surrogate", which is a
1347 sub-category of "utf8") if an attempt is made
1348 to do things like take the lower case of one, or match
1349 case-insensitively, or to output them. (But don't try this on Perls
1354 UTF-32, UTF-32BE, UTF-32LE
1356 The UTF-32 family is pretty much like the UTF-16 family, expect that
1357 the units are 32-bit, and therefore the surrogate scheme is not
1358 needed. UTF-32 is a fixed-width encoding. The BOM signatures are
1359 C<0x00 0x00 0xFE 0xFF> for BE and C<0xFF 0xFE 0x00 0x00> for LE.
1365 Legacy, fixed-width encodings defined by the ISO 10646 standard. UCS-2 is a 16-bit
1366 encoding. Unlike UTF-16, UCS-2 is not extensible beyond C<U+FFFF>,
1367 because it does not use surrogates. UCS-4 is a 32-bit encoding,
1368 functionally identical to UTF-32 (the difference being that
1369 UCS-4 forbids neither surrogates nor code points larger than 0x10_FFFF).
1375 A seven-bit safe (non-eight-bit) encoding, which is useful if the
1376 transport or storage is not eight-bit safe. Defined by RFC 2152.
1380 =head2 Non-character code points
1382 66 code points are set aside in Unicode as "non-character code points".
1383 These all have the Unassigned (Cn) General Category, and they never will
1384 be assigned. These are never supposed to be in legal Unicode input
1385 streams, so that code can use them as sentinels that can be mixed in
1386 with character data, and they always will be distinguishable from that data.
1387 To keep them out of Perl input streams, strict UTF-8 should be
1388 specified, such as by using the layer C<:encoding('UTF-8')>. The
1389 non-character code points are the 32 between U+FDD0 and U+FDEF, and the
1390 34 code points U+FFFE, U+FFFF, U+1FFFE, U+1FFFF, ... U+10FFFE, U+10FFFF.
1391 Some people are under the mistaken impression that these are "illegal",
1392 but that is not true. An application or cooperating set of applications
1393 can legally use them at will internally; but these code points are
1394 "illegal for open interchange". Therefore, Perl will not accept these
1395 from input streams unless lax rules are being used, and will warn
1396 (using the warning category "nonchar", which is a sub-category of "utf8") if
1397 an attempt is made to output them.
1399 =head2 Beyond Unicode code points
1401 The maximum Unicode code point is U+10FFFF. But Perl accepts code
1402 points up to the maximum permissible unsigned number available on the
1403 platform. However, Perl will not accept these from input streams unless
1404 lax rules are being used, and will warn (using the warning category
1405 "non_unicode", which is a sub-category of "utf8") if an attempt is made to
1406 operate on or output them. For example, C<uc(0x11_0000)> will generate
1407 this warning, returning the input parameter as its result, as the upper
1408 case of every non-Unicode code point is the code point itself.
1410 =head2 Security Implications of Unicode
1412 Read L<Unicode Security Considerations|http://www.unicode.org/reports/tr36>.
1413 Also, note the following:
1421 Unfortunately, the original specification of UTF-8 leaves some room for
1422 interpretation of how many bytes of encoded output one should generate
1423 from one input Unicode character. Strictly speaking, the shortest
1424 possible sequence of UTF-8 bytes should be generated,
1425 because otherwise there is potential for an input buffer overflow at
1426 the receiving end of a UTF-8 connection. Perl always generates the
1427 shortest length UTF-8, and with warnings on, Perl will warn about
1428 non-shortest length UTF-8 along with other malformations, such as the
1429 surrogates, which are not Unicode code points valid for interchange.
1433 Regular expression pattern matching may surprise you if you're not
1434 accustomed to Unicode. Starting in Perl 5.14, several pattern
1435 modifiers are available to control this, called the character set
1436 modifiers. Details are given in L<perlre/Character set modifiers>.
1440 As discussed elsewhere, Perl has one foot (two hooves?) planted in
1441 each of two worlds: the old world of bytes and the new world of
1442 characters, upgrading from bytes to characters when necessary.
1443 If your legacy code does not explicitly use Unicode, no automatic
1444 switch-over to characters should happen. Characters shouldn't get
1445 downgraded to bytes, either. It is possible to accidentally mix bytes
1446 and characters, however (see L<perluniintro>), in which case C<\w> in
1447 regular expressions might start behaving differently (unless the C</a>
1448 modifier is in effect). Review your code. Use warnings and the C<strict> pragma.
1450 =head2 Unicode in Perl on EBCDIC
1452 The way Unicode is handled on EBCDIC platforms is still
1453 experimental. On such platforms, references to UTF-8 encoding in this
1454 document and elsewhere should be read as meaning the UTF-EBCDIC
1455 specified in Unicode Technical Report 16, unless ASCII vs. EBCDIC issues
1456 are specifically discussed. There is no C<utfebcdic> pragma or
1457 ":utfebcdic" layer; rather, "utf8" and ":utf8" are reused to mean
1458 the platform's "natural" 8-bit encoding of Unicode. See L<perlebcdic>
1459 for more discussion of the issues.
1463 See L<perllocale/Unicode and UTF-8>
1465 =head2 When Unicode Does Not Happen
1467 While Perl does have extensive ways to input and output in Unicode,
1468 and a few other "entry points" like the @ARGV array (which can sometimes be
1469 interpreted as UTF-8), there are still many places where Unicode
1470 (in some encoding or another) could be given as arguments or received as
1471 results, or both, but it is not.
1473 The following are such interfaces. Also, see L</The "Unicode Bug">.
1474 For all of these interfaces Perl
1475 currently (as of 5.8.3) simply assumes byte strings both as arguments
1476 and results, or UTF-8 strings if the (problematic) C<encoding> pragma has been used.
1478 One reason that Perl does not attempt to resolve the role of Unicode in
1479 these situations is that the answers are highly dependent on the operating
1480 system and the file system(s). For example, whether filenames can be
1481 in Unicode and in exactly what kind of encoding, is not exactly a
1482 portable concept. Similarly for C<qx> and C<system>: how well will the
1483 "command-line interface" (and which of them?) handle Unicode?
1489 chdir, chmod, chown, chroot, exec, link, lstat, mkdir,
1490 rename, rmdir, stat, symlink, truncate, unlink, utime, -X
1502 open, opendir, sysopen
1506 qx (aka the backtick operator), system
1514 =head2 The "Unicode Bug"
1516 The term, the "Unicode bug" has been applied to an inconsistency
1517 on ASCII platforms with the
1518 Unicode code points in the Latin-1 Supplement block, that
1519 is, between 128 and 255. Without a locale specified, unlike all other
1520 characters or code points, these characters have very different semantics in
1521 byte semantics versus character semantics, unless
1522 C<use feature 'unicode_strings'> is specified.
1523 (The lesson here is to specify C<unicode_strings> to avoid the
1526 In character semantics they are interpreted as Unicode code points, which means
1527 they have the same semantics as Latin-1 (ISO-8859-1).
1529 In byte semantics, they are considered to be unassigned characters, meaning
1530 that the only semantics they have is their ordinal numbers, and that they are
1531 not members of various character classes. None are considered to match C<\w>
1532 for example, but all match C<\W>.
1534 The behavior is known to have effects on these areas:
1540 Changing the case of a scalar, that is, using C<uc()>, C<ucfirst()>, C<lc()>,
1541 and C<lcfirst()>, or C<\L>, C<\U>, C<\u> and C<\l> in regular expression
1546 Using caseless (C</i>) regular expression matching
1550 Matching any of several properties in regular expressions, namely C<\b>,
1551 C<\B>, C<\s>, C<\S>, C<\w>, C<\W>, and all the Posix character classes
1552 I<except> C<[[:ascii:]]>.
1556 In C<quotemeta> or its inline equivalent C<\Q>, no characters
1557 code points above 127 are quoted in UTF-8 encoded strings, but in
1558 byte encoded strings, code points between 128-255 are always quoted.
1562 User-defined case change mappings. You can create a C<ToUpper()> function, for
1563 example, which overrides Perl's built-in case mappings. The scalar must be
1564 encoded in utf8 for your function to actually be invoked.
1568 This behavior can lead to unexpected results in which a string's semantics
1569 suddenly change if a code point above 255 is appended to or removed from it,
1570 which changes the string's semantics from byte to character or vice versa. As
1571 an example, consider the following program and its output:
1574 no feature 'unicode_strings';
1577 for ($s1, $s2, $s1.$s2) {
1585 If there's no C<\w> in C<s1> or in C<s2>, why does their concatenation have one?
1587 This anomaly stems from Perl's attempt to not disturb older programs that
1588 didn't use Unicode, and hence had no semantics for characters outside of the
1589 ASCII range (except in a locale), along with Perl's desire to add Unicode
1590 support seamlessly. The result wasn't seamless: these characters were
1593 Starting in Perl 5.14, C<use feature 'unicode_strings'> can be used to
1594 cause Perl to use Unicode semantics on all string operations within the
1595 scope of the feature subpragma. Regular expressions compiled in its
1596 scope retain that behavior even when executed or compiled into larger
1597 regular expressions outside the scope. (The pragma does not, however,
1598 affect the C<quotemeta> behavior. Nor does it affect the deprecated
1599 user-defined case changing operations--these still require a UTF-8
1600 encoded string to operate.)
1602 In Perl 5.12, the subpragma affected casing changes, but not regular
1603 expressions. See L<perlfunc/lc> for details on how this pragma works in
1604 combination with various others for casing.
1606 For earlier Perls, or when a string is passed to a function outside the
1607 subpragma's scope, a workaround is to always call C<utf8::upgrade($string)>,
1608 or to use the standard module L<Encode>. Also, a scalar that has any characters
1609 whose ordinal is above 0x100, or which were specified using either of the
1610 C<\N{...}> notations, will automatically have character semantics.
1612 =head2 Forcing Unicode in Perl (Or Unforcing Unicode in Perl)
1614 Sometimes (see L</"When Unicode Does Not Happen"> or L</The "Unicode Bug">)
1615 there are situations where you simply need to force a byte
1616 string into UTF-8, or vice versa. The low-level calls
1617 utf8::upgrade($bytestring) and utf8::downgrade($utf8string[, FAIL_OK]) are
1620 Note that utf8::downgrade() can fail if the string contains characters
1621 that don't fit into a byte.
1623 Calling either function on a string that already is in the desired state is a
1626 =head2 Using Unicode in XS
1628 If you want to handle Perl Unicode in XS extensions, you may find the
1629 following C APIs useful. See also L<perlguts/"Unicode Support"> for an
1630 explanation about Unicode at the XS level, and L<perlapi> for the API
1637 C<DO_UTF8(sv)> returns true if the C<UTF8> flag is on and the bytes
1638 pragma is not in effect. C<SvUTF8(sv)> returns true if the C<UTF8>
1639 flag is on; the bytes pragma is ignored. The C<UTF8> flag being on
1640 does B<not> mean that there are any characters of code points greater
1641 than 255 (or 127) in the scalar or that there are even any characters
1642 in the scalar. What the C<UTF8> flag means is that the sequence of
1643 octets in the representation of the scalar is the sequence of UTF-8
1644 encoded code points of the characters of a string. The C<UTF8> flag
1645 being off means that each octet in this representation encodes a
1646 single character with code point 0..255 within the string. Perl's
1647 Unicode model is not to use UTF-8 until it is absolutely necessary.
1651 C<uvchr_to_utf8(buf, chr)> writes a Unicode character code point into
1652 a buffer encoding the code point as UTF-8, and returns a pointer
1653 pointing after the UTF-8 bytes. It works appropriately on EBCDIC machines.
1657 C<utf8_to_uvchr(buf, lenp)> reads UTF-8 encoded bytes from a buffer and
1658 returns the Unicode character code point and, optionally, the length of
1659 the UTF-8 byte sequence. It works appropriately on EBCDIC machines.
1663 C<utf8_length(start, end)> returns the length of the UTF-8 encoded buffer
1664 in characters. C<sv_len_utf8(sv)> returns the length of the UTF-8 encoded
1669 C<sv_utf8_upgrade(sv)> converts the string of the scalar to its UTF-8
1670 encoded form. C<sv_utf8_downgrade(sv)> does the opposite, if
1671 possible. C<sv_utf8_encode(sv)> is like sv_utf8_upgrade except that
1672 it does not set the C<UTF8> flag. C<sv_utf8_decode()> does the
1673 opposite of C<sv_utf8_encode()>. Note that none of these are to be
1674 used as general-purpose encoding or decoding interfaces: C<use Encode>
1675 for that. C<sv_utf8_upgrade()> is affected by the encoding pragma
1676 but C<sv_utf8_downgrade()> is not (since the encoding pragma is
1677 designed to be a one-way street).
1681 C<is_utf8_char(s)> returns true if the pointer points to a valid UTF-8
1686 C<is_utf8_string(buf, len)> returns true if C<len> bytes of the buffer
1691 C<UTF8SKIP(buf)> will return the number of bytes in the UTF-8 encoded
1692 character in the buffer. C<UNISKIP(chr)> will return the number of bytes
1693 required to UTF-8-encode the Unicode character code point. C<UTF8SKIP()>
1694 is useful for example for iterating over the characters of a UTF-8
1695 encoded buffer; C<UNISKIP()> is useful, for example, in computing
1696 the size required for a UTF-8 encoded buffer.
1700 C<utf8_distance(a, b)> will tell the distance in characters between the
1701 two pointers pointing to the same UTF-8 encoded buffer.
1705 C<utf8_hop(s, off)> will return a pointer to a UTF-8 encoded buffer
1706 that is C<off> (positive or negative) Unicode characters displaced
1707 from the UTF-8 buffer C<s>. Be careful not to overstep the buffer:
1708 C<utf8_hop()> will merrily run off the end or the beginning of the
1709 buffer if told to do so.
1713 C<pv_uni_display(dsv, spv, len, pvlim, flags)> and
1714 C<sv_uni_display(dsv, ssv, pvlim, flags)> are useful for debugging the
1715 output of Unicode strings and scalars. By default they are useful
1716 only for debugging--they display B<all> characters as hexadecimal code
1717 points--but with the flags C<UNI_DISPLAY_ISPRINT>,
1718 C<UNI_DISPLAY_BACKSLASH>, and C<UNI_DISPLAY_QQ> you can make the
1719 output more readable.
1723 C<foldEQ_utf8(s1, pe1, l1, u1, s2, pe2, l2, u2)> can be used to
1724 compare two strings case-insensitively in Unicode. For case-sensitive
1725 comparisons you can just use C<memEQ()> and C<memNE()> as usual, except
1726 if one string is in utf8 and the other isn't.
1730 For more information, see L<perlapi>, and F<utf8.c> and F<utf8.h>
1731 in the Perl source code distribution.
1733 =head2 Hacking Perl to work on earlier Unicode versions (for very serious hackers only)
1735 Perl by default comes with the latest supported Unicode version built in, but
1736 you can change to use any earlier one.
1738 Download the files in the desired version of Unicode from the Unicode web
1739 site L<http://www.unicode.org>). These should replace the existing files in
1740 F<lib/unicore> in the Perl source tree. Follow the instructions in
1741 F<README.perl> in that directory to change some of their names, and then build
1742 perl (see L<INSTALL>).
1744 It is even possible to copy the built files to a different directory, and then
1745 change F<utf8_heavy.pl> in the directory C<$Config{privlib}> to point to the
1746 new directory, or maybe make a copy of that directory before making the change,
1747 and using C<@INC> or the C<-I> run-time flag to switch between versions at will
1748 (but because of caching, not in the middle of a process), but all this is
1749 beyond the scope of these instructions.
1753 =head2 Interaction with Locales
1755 See L<perllocale/Unicode and UTF-8>
1757 =head2 Problems with characters in the Latin-1 Supplement range
1759 See L</The "Unicode Bug">
1761 =head2 Interaction with Extensions
1763 When Perl exchanges data with an extension, the extension should be
1764 able to understand the UTF8 flag and act accordingly. If the
1765 extension doesn't recognize that flag, it's likely that the extension
1766 will return incorrectly-flagged data.
1768 So if you're working with Unicode data, consult the documentation of
1769 every module you're using if there are any issues with Unicode data
1770 exchange. If the documentation does not talk about Unicode at all,
1771 suspect the worst and probably look at the source to learn how the
1772 module is implemented. Modules written completely in Perl shouldn't
1773 cause problems. Modules that directly or indirectly access code written
1774 in other programming languages are at risk.
1776 For affected functions, the simple strategy to avoid data corruption is
1777 to always make the encoding of the exchanged data explicit. Choose an
1778 encoding that you know the extension can handle. Convert arguments passed
1779 to the extensions to that encoding and convert results back from that
1780 encoding. Write wrapper functions that do the conversions for you, so
1781 you can later change the functions when the extension catches up.
1783 To provide an example, let's say the popular Foo::Bar::escape_html
1784 function doesn't deal with Unicode data yet. The wrapper function
1785 would convert the argument to raw UTF-8 and convert the result back to
1786 Perl's internal representation like so:
1788 sub my_escape_html ($) {
1790 return unless defined $what;
1791 Encode::decode_utf8(Foo::Bar::escape_html(
1792 Encode::encode_utf8($what)));
1795 Sometimes, when the extension does not convert data but just stores
1796 and retrieves them, you will be able to use the otherwise
1797 dangerous Encode::_utf8_on() function. Let's say the popular
1798 C<Foo::Bar> extension, written in C, provides a C<param> method that
1799 lets you store and retrieve data according to these prototypes:
1801 $self->param($name, $value); # set a scalar
1802 $value = $self->param($name); # retrieve a scalar
1804 If it does not yet provide support for any encoding, one could write a
1805 derived class with such a C<param> method:
1808 my($self,$name,$value) = @_;
1809 utf8::upgrade($name); # make sure it is UTF-8 encoded
1810 if (defined $value) {
1811 utf8::upgrade($value); # make sure it is UTF-8 encoded
1812 return $self->SUPER::param($name,$value);
1814 my $ret = $self->SUPER::param($name);
1815 Encode::_utf8_on($ret); # we know, it is UTF-8 encoded
1820 Some extensions provide filters on data entry/exit points, such as
1821 DB_File::filter_store_key and family. Look out for such filters in
1822 the documentation of your extensions, they can make the transition to
1823 Unicode data much easier.
1827 Some functions are slower when working on UTF-8 encoded strings than
1828 on byte encoded strings. All functions that need to hop over
1829 characters such as length(), substr() or index(), or matching regular
1830 expressions can work B<much> faster when the underlying data are
1833 In Perl 5.8.0 the slowness was often quite spectacular; in Perl 5.8.1
1834 a caching scheme was introduced which will hopefully make the slowness
1835 somewhat less spectacular, at least for some operations. In general,
1836 operations with UTF-8 encoded strings are still slower. As an example,
1837 the Unicode properties (character classes) like C<\p{Nd}> are known to
1838 be quite a bit slower (5-20 times) than their simpler counterparts
1839 like C<\d> (then again, there are hundreds of Unicode characters matching C<Nd>
1840 compared with the 10 ASCII characters matching C<d>).
1842 =head2 Problems on EBCDIC platforms
1844 There are several known problems with Perl on EBCDIC platforms. If you
1845 want to use Perl there, send email to perlbug@perl.org.
1847 In earlier versions, when byte and character data were concatenated,
1848 the new string was sometimes created by
1849 decoding the byte strings as I<ISO 8859-1 (Latin-1)>, even if the
1850 old Unicode string used EBCDIC.
1852 If you find any of these, please report them as bugs.
1854 =head2 Porting code from perl-5.6.X
1856 Perl 5.8 has a different Unicode model from 5.6. In 5.6 the programmer
1857 was required to use the C<utf8> pragma to declare that a given scope
1858 expected to deal with Unicode data and had to make sure that only
1859 Unicode data were reaching that scope. If you have code that is
1860 working with 5.6, you will need some of the following adjustments to
1861 your code. The examples are written such that the code will continue
1862 to work under 5.6, so you should be safe to try them out.
1868 A filehandle that should read or write UTF-8
1871 binmode $fh, ":encoding(utf8)";
1876 A scalar that is going to be passed to some extension
1878 Be it Compress::Zlib, Apache::Request or any extension that has no
1879 mention of Unicode in the manpage, you need to make sure that the
1880 UTF8 flag is stripped off. Note that at the time of this writing
1881 (October 2002) the mentioned modules are not UTF-8-aware. Please
1882 check the documentation to verify if this is still true.
1886 $val = Encode::encode_utf8($val); # make octets
1891 A scalar we got back from an extension
1893 If you believe the scalar comes back as UTF-8, you will most likely
1894 want the UTF8 flag restored:
1898 $val = Encode::decode_utf8($val);
1903 Same thing, if you are really sure it is UTF-8
1907 Encode::_utf8_on($val);
1912 A wrapper for fetchrow_array and fetchrow_hashref
1914 When the database contains only UTF-8, a wrapper function or method is
1915 a convenient way to replace all your fetchrow_array and
1916 fetchrow_hashref calls. A wrapper function will also make it easier to
1917 adapt to future enhancements in your database driver. Note that at the
1918 time of this writing (October 2002), the DBI has no standardized way
1919 to deal with UTF-8 data. Please check the documentation to verify if
1923 # $what is one of fetchrow_{array,hashref}
1924 my($self, $sth, $what) = @_;
1930 my @arr = $sth->$what;
1932 defined && /[^\000-\177]/ && Encode::_utf8_on($_);
1936 my $ret = $sth->$what;
1938 for my $k (keys %$ret) {
1941 && Encode::_utf8_on($_) for $ret->{$k};
1945 defined && /[^\000-\177]/ && Encode::_utf8_on($_) for $ret;
1955 A large scalar that you know can only contain ASCII
1957 Scalars that contain only ASCII and are marked as UTF-8 are sometimes
1958 a drag to your program. If you recognize such a situation, just remove
1961 utf8::downgrade($val) if $] > 5.007;
1967 L<perlunitut>, L<perluniintro>, L<perluniprops>, L<Encode>, L<open>, L<utf8>, L<bytes>,
1968 L<perlretut>, L<perlvar/"${^UNICODE}">
1969 L<http://www.unicode.org/reports/tr44>).