xref: /openbsd-src/gnu/usr.bin/perl/pod/perlunicode.pod (revision f1dd7b858388b4a23f4f67a4957ec5ff656ebbe8)
1=head1 NAME
2
3perlunicode - Unicode support in Perl
4
5=head1 DESCRIPTION
6
7If you haven't already, before reading this document, you should become
8familiar with both L<perlunitut> and L<perluniintro>.
9
10Unicode aims to B<UNI>-fy the en-B<CODE>-ings of all the world's
11character sets into a single Standard.   For quite a few of the various
12coding standards that existed when Unicode was first created, converting
13from each to Unicode essentially meant adding a constant to each code
14point in the original standard, and converting back meant just
15subtracting that same constant.  For ASCII and ISO-8859-1, the constant
16is 0.  For ISO-8859-5, (Cyrillic) the constant is 864; for Hebrew
17(ISO-8859-8), it's 1488; Thai (ISO-8859-11), 3424; and so forth.  This
18made it easy to do the conversions, and facilitated the adoption of
19Unicode.
20
21And it worked; nowadays, those legacy standards are rarely used.  Most
22everyone uses Unicode.
23
24Unicode is a comprehensive standard.  It specifies many things outside
25the scope of Perl, such as how to display sequences of characters.  For
26a full discussion of all aspects of Unicode, see
27L<https://www.unicode.org>.
28
29=head2 Important Caveats
30
31Even though some of this section may not be understandable to you on
32first reading, we think it's important enough to highlight some of the
33gotchas before delving further, so here goes:
34
35Unicode support is an extensive requirement. While Perl does not
36implement the Unicode standard or the accompanying technical reports
37from cover to cover, Perl does support many Unicode features.
38
39Also, the use of Unicode may present security issues that aren't
40obvious, see L</Security Implications of Unicode> below.
41
42=over 4
43
44=item Safest if you C<use feature 'unicode_strings'>
45
46In order to preserve backward compatibility, Perl does not turn
47on full internal Unicode support unless the pragma
48L<S<C<use feature 'unicode_strings'>>|feature/The 'unicode_strings' feature>
49is specified.  (This is automatically
50selected if you S<C<use 5.012>> or higher.)  Failure to do this can
51trigger unexpected surprises.  See L</The "Unicode Bug"> below.
52
53This pragma doesn't affect I/O.  Nor does it change the internal
54representation of strings, only their interpretation.  There are still
55several places where Unicode isn't fully supported, such as in
56filenames.
57
58=item Input and Output Layers
59
60Use the C<:encoding(...)> layer  to read from and write to
61filehandles using the specified encoding.  (See L<open>.)
62
63=item You must convert your non-ASCII, non-UTF-8 Perl scripts to be
64UTF-8.
65
66The L<encoding> module has been deprecated since perl 5.18 and the
67perl internals it requires have been removed with perl 5.26.
68
69=item C<use utf8> still needed to enable L<UTF-8|/Unicode Encodings> in scripts
70
71If your Perl script is itself encoded in L<UTF-8|/Unicode Encodings>,
72the S<C<use utf8>> pragma must be explicitly included to enable
73recognition of that (in string or regular expression literals, or in
74identifier names).  B<This is the only time when an explicit S<C<use
75utf8>> is needed.>  (See L<utf8>).
76
77If a Perl script begins with the bytes that form the UTF-8 encoding of
78the Unicode BYTE ORDER MARK (C<BOM>, see L</Unicode Encodings>), those
79bytes are completely ignored.
80
81=item L<UTF-16|/Unicode Encodings> scripts autodetected
82
83If a Perl script begins with the Unicode C<BOM> (UTF-16LE,
84UTF16-BE), or if the script looks like non-C<BOM>-marked
85UTF-16 of either endianness, Perl will correctly read in the script as
86the appropriate Unicode encoding.
87
88=back
89
90=head2 Byte and Character Semantics
91
92Before Unicode, most encodings used 8 bits (a single byte) to encode
93each character.  Thus a character was a byte, and a byte was a
94character, and there could be only 256 or fewer possible characters.
95"Byte Semantics" in the title of this section refers to
96this behavior.  There was no need to distinguish between "Byte" and
97"Character".
98
99Then along comes Unicode which has room for over a million characters
100(and Perl allows for even more).  This means that a character may
101require more than a single byte to represent it, and so the two terms
102are no longer equivalent.  What matter are the characters as whole
103entities, and not usually the bytes that comprise them.  That's what the
104term "Character Semantics" in the title of this section refers to.
105
106Perl had to change internally to decouple "bytes" from "characters".
107It is important that you too change your ideas, if you haven't already,
108so that "byte" and "character" no longer mean the same thing in your
109mind.
110
111The basic building block of Perl strings has always been a "character".
112The changes basically come down to that the implementation no longer
113thinks that a character is always just a single byte.
114
115There are various things to note:
116
117=over 4
118
119=item *
120
121String handling functions, for the most part, continue to operate in
122terms of characters.  C<length()>, for example, returns the number of
123characters in a string, just as before.  But that number no longer is
124necessarily the same as the number of bytes in the string (there may be
125more bytes than characters).  The other such functions include
126C<chop()>, C<chomp()>, C<substr()>, C<pos()>, C<index()>, C<rindex()>,
127C<sort()>, C<sprintf()>, and C<write()>.
128
129The exceptions are:
130
131=over 4
132
133=item *
134
135the bit-oriented C<vec>
136
137E<nbsp>
138
139=item *
140
141the byte-oriented C<pack>/C<unpack> C<"C"> format
142
143However, the C<W> specifier does operate on whole characters, as does the
144C<U> specifier.
145
146=item *
147
148some operators that interact with the platform's operating system
149
150Operators dealing with filenames are examples.
151
152=item *
153
154when the functions are called from within the scope of the
155S<C<L<use bytes|bytes>>> pragma
156
157Likely, you should use this only for debugging anyway.
158
159=back
160
161=item *
162
163Strings--including hash keys--and regular expression patterns may
164contain characters that have ordinal values larger than 255.
165
166If you use a Unicode editor to edit your program, Unicode characters may
167occur directly within the literal strings in UTF-8 encoding, or UTF-16.
168(The former requires a C<use utf8>, the latter may require a C<BOM>.)
169
170L<perluniintro/Creating Unicode> gives other ways to place non-ASCII
171characters in your strings.
172
173=item *
174
175The C<chr()> and C<ord()> functions work on whole characters.
176
177=item *
178
179Regular expressions match whole characters.  For example, C<"."> matches
180a whole character instead of only a single byte.
181
182=item *
183
184The C<tr///> operator translates whole characters.  (Note that the
185C<tr///CU> functionality has been removed.  For similar functionality to
186that, see C<pack('U0', ...)> and C<pack('C0', ...)>).
187
188=item *
189
190C<scalar reverse()> reverses by character rather than by byte.
191
192=item *
193
194The bit string operators, C<& | ^ ~> and (starting in v5.22)
195C<&. |. ^.  ~.> can operate on bit strings encoded in UTF-8, but this
196can give unexpected results if any of the strings contain code points
197above 0xFF.  Starting in v5.28, it is a fatal error to have such an
198operand.  Otherwise, the operation is performed on a non-UTF-8 copy of
199the operand.  If you're not sure about the encoding of a string,
200downgrade it before using any of these operators; you can use
201L<C<utf8::utf8_downgrade()>|utf8/Utility functions>.
202
203=back
204
205The bottom line is that Perl has always practiced "Character Semantics",
206but with the advent of Unicode, that is now different than "Byte
207Semantics".
208
209=head2 ASCII Rules versus Unicode Rules
210
211Before Unicode, when a character was a byte was a character,
212Perl knew only about the 128 characters defined by ASCII, code points 0
213through 127 (except for under L<S<C<use locale>>|perllocale>).  That
214left the code
215points 128 to 255 as unassigned, and available for whatever use a
216program might want.  The only semantics they have is their ordinal
217numbers, and that they are members of none of the non-negative character
218classes.  None are considered to match C<\w> for example, but all match
219C<\W>.
220
221Unicode, of course, assigns each of those code points a particular
222meaning (along with ones above 255).  To preserve backward
223compatibility, Perl only uses the Unicode meanings when there is some
224indication that Unicode is what is intended; otherwise the non-ASCII
225code points remain treated as if they are unassigned.
226
227Here are the ways that Perl knows that a string should be treated as
228Unicode:
229
230=over
231
232=item *
233
234Within the scope of S<C<use utf8>>
235
236If the whole program is Unicode (signified by using 8-bit B<U>nicode
237B<T>ransformation B<F>ormat), then all literal strings within it must be
238Unicode.
239
240=item *
241
242Within the scope of
243L<S<C<use feature 'unicode_strings'>>|feature/The 'unicode_strings' feature>
244
245This pragma was created so you can explicitly tell Perl that operations
246executed within its scope are to use Unicode rules.  More operations are
247affected with newer perls.  See L</The "Unicode Bug">.
248
249=item *
250
251Within the scope of S<C<use 5.012>> or higher
252
253This implicitly turns on S<C<use feature 'unicode_strings'>>.
254
255=item *
256
257Within the scope of
258L<S<C<use locale 'not_characters'>>|perllocale/Unicode and UTF-8>,
259or L<S<C<use locale>>|perllocale> and the current
260locale is a UTF-8 locale.
261
262The former is defined to imply Unicode handling; and the latter
263indicates a Unicode locale, hence a Unicode interpretation of all
264strings within it.
265
266=item *
267
268When the string contains a Unicode-only code point
269
270Perl has never accepted code points above 255 without them being
271Unicode, so their use implies Unicode for the whole string.
272
273=item *
274
275When the string contains a Unicode named code point C<\N{...}>
276
277The C<\N{...}> construct explicitly refers to a Unicode code point,
278even if it is one that is also in ASCII.  Therefore the string
279containing it must be Unicode.
280
281=item *
282
283When the string has come from an external source marked as
284Unicode
285
286The L<C<-C>|perlrun/-C [numberE<sol>list]> command line option can
287specify that certain inputs to the program are Unicode, and the values
288of this can be read by your Perl code, see L<perlvar/"${^UNICODE}">.
289
290=item * When the string has been upgraded to UTF-8
291
292The function L<C<utf8::utf8_upgrade()>|utf8/Utility functions>
293can be explicitly used to permanently (unless a subsequent
294C<utf8::utf8_downgrade()> is called) cause a string to be treated as
295Unicode.
296
297=item * There are additional methods for regular expression patterns
298
299A pattern that is compiled with the C<< /u >> or C<< /a >> modifiers is
300treated as Unicode (though there are some restrictions with C<< /a >>).
301Under the C<< /d >> and C<< /l >> modifiers, there are several other
302indications for Unicode; see L<perlre/Character set modifiers>.
303
304=back
305
306Note that all of the above are overridden within the scope of
307C<L<use bytes|bytes>>; but you should be using this pragma only for
308debugging.
309
310Note also that some interactions with the platform's operating system
311never use Unicode rules.
312
313When Unicode rules are in effect:
314
315=over 4
316
317=item *
318
319Case translation operators use the Unicode case translation tables.
320
321Note that C<uc()>, or C<\U> in interpolated strings, translates to
322uppercase, while C<ucfirst>, or C<\u> in interpolated strings,
323translates to titlecase in languages that make the distinction (which is
324equivalent to uppercase in languages without the distinction).
325
326There is a CPAN module, C<L<Unicode::Casing>>, which allows you to
327define your own mappings to be used in C<lc()>, C<lcfirst()>, C<uc()>,
328C<ucfirst()>, and C<fc> (or their double-quoted string inlined versions
329such as C<\U>).  (Prior to Perl 5.16, this functionality was partially
330provided in the Perl core, but suffered from a number of insurmountable
331drawbacks, so the CPAN module was written instead.)
332
333=item *
334
335Character classes in regular expressions match based on the character
336properties specified in the Unicode properties database.
337
338C<\w> can be used to match a Japanese ideograph, for instance; and
339C<[[:digit:]]> a Bengali number.
340
341=item *
342
343Named Unicode properties, scripts, and block ranges may be used (like
344bracketed character classes) by using the C<\p{}> "matches property"
345construct and the C<\P{}> negation, "doesn't match property".
346
347See L</"Unicode Character Properties"> for more details.
348
349You can define your own character properties and use them
350in the regular expression with the C<\p{}> or C<\P{}> construct.
351See L</"User-Defined Character Properties"> for more details.
352
353=back
354
355=head2 Extended Grapheme Clusters (Logical characters)
356
357Consider a character, say C<H>.  It could appear with various marks around it,
358such as an acute accent, or a circumflex, or various hooks, circles, arrows,
359I<etc.>, above, below, to one side or the other, I<etc>.  There are many
360possibilities among the world's languages.  The number of combinations is
361astronomical, and if there were a character for each combination, it would
362soon exhaust Unicode's more than a million possible characters.  So Unicode
363took a different approach: there is a character for the base C<H>, and a
364character for each of the possible marks, and these can be variously combined
365to get a final logical character.  So a logical character--what appears to be a
366single character--can be a sequence of more than one individual characters.
367The Unicode standard calls these "extended grapheme clusters" (which
368is an improved version of the no-longer much used "grapheme cluster");
369Perl furnishes the C<\X> regular expression construct to match such
370sequences in their entirety.
371
372But Unicode's intent is to unify the existing character set standards and
373practices, and several pre-existing standards have single characters that
374mean the same thing as some of these combinations, like ISO-8859-1,
375which has quite a few of them. For example, C<"LATIN CAPITAL LETTER E
376WITH ACUTE"> was already in this standard when Unicode came along.
377Unicode therefore added it to its repertoire as that single character.
378But this character is considered by Unicode to be equivalent to the
379sequence consisting of the character C<"LATIN CAPITAL LETTER E">
380followed by the character C<"COMBINING ACUTE ACCENT">.
381
382C<"LATIN CAPITAL LETTER E WITH ACUTE"> is called a "pre-composed"
383character, and its equivalence with the "E" and the "COMBINING ACCENT"
384sequence is called canonical equivalence.  All pre-composed characters
385are said to have a decomposition (into the equivalent sequence), and the
386decomposition type is also called canonical.  A string may be comprised
387as much as possible of precomposed characters, or it may be comprised of
388entirely decomposed characters.  Unicode calls these respectively,
389"Normalization Form Composed" (NFC) and "Normalization Form Decomposed".
390The C<L<Unicode::Normalize>> module contains functions that convert
391between the two.  A string may also have both composed characters and
392decomposed characters; this module can be used to make it all one or the
393other.
394
395You may be presented with strings in any of these equivalent forms.
396There is currently nothing in Perl 5 that ignores the differences.  So
397you'll have to specially handle it.  The usual advice is to convert your
398inputs to C<NFD> before processing further.
399
400For more detailed information, see L<http://unicode.org/reports/tr15/>.
401
402=head2 Unicode Character Properties
403
404(The only time that Perl considers a sequence of individual code
405points as a single logical character is in the C<\X> construct, already
406mentioned above.   Therefore "character" in this discussion means a single
407Unicode code point.)
408
409Very nearly all Unicode character properties are accessible through
410regular expressions by using the C<\p{}> "matches property" construct
411and the C<\P{}> "doesn't match property" for its negation.
412
413For instance, C<\p{Uppercase}> matches any single character with the Unicode
414C<"Uppercase"> property, while C<\p{L}> matches any character with a
415C<General_Category> of C<"L"> (letter) property (see
416L</General_Category> below).  Brackets are not
417required for single letter property names, so C<\p{L}> is equivalent to C<\pL>.
418
419More formally, C<\p{Uppercase}> matches any single character whose Unicode
420C<Uppercase> property value is C<True>, and C<\P{Uppercase}> matches any character
421whose C<Uppercase> property value is C<False>, and they could have been written as
422C<\p{Uppercase=True}> and C<\p{Uppercase=False}>, respectively.
423
424This formality is needed when properties are not binary; that is, if they can
425take on more values than just C<True> and C<False>.  For example, the
426C<Bidi_Class> property (see L</"Bidirectional Character Types"> below),
427can take on several different
428values, such as C<Left>, C<Right>, C<Whitespace>, and others.  To match these, one needs
429to specify both the property name (C<Bidi_Class>), AND the value being
430matched against
431(C<Left>, C<Right>, I<etc.>).  This is done, as in the examples above, by having the
432two components separated by an equal sign (or interchangeably, a colon), like
433C<\p{Bidi_Class: Left}>.
434
435All Unicode-defined character properties may be written in these compound forms
436of C<\p{I<property>=I<value>}> or C<\p{I<property>:I<value>}>, but Perl provides some
437additional properties that are written only in the single form, as well as
438single-form short-cuts for all binary properties and certain others described
439below, in which you may omit the property name and the equals or colon
440separator.
441
442Most Unicode character properties have at least two synonyms (or aliases if you
443prefer): a short one that is easier to type and a longer one that is more
444descriptive and hence easier to understand.  Thus the C<"L"> and
445C<"Letter"> properties above are equivalent and can be used
446interchangeably.  Likewise, C<"Upper"> is a synonym for C<"Uppercase">,
447and we could have written C<\p{Uppercase}> equivalently as C<\p{Upper}>.
448Also, there are typically various synonyms for the values the property
449can be.   For binary properties, C<"True"> has 3 synonyms: C<"T">,
450C<"Yes">, and C<"Y">; and C<"False"> has correspondingly C<"F">,
451C<"No">, and C<"N">.  But be careful.  A short form of a value for one
452property may not mean the same thing as the short form spelled the same
453for another.
454Thus, for the C<L</General_Category>> property, C<"L"> means
455C<"Letter">, but for the L<C<Bidi_Class>|/Bidirectional Character Types>
456property, C<"L"> means C<"Left">.  A complete list of properties and
457synonyms is in L<perluniprops>.
458
459Upper/lower case differences in property names and values are irrelevant;
460thus C<\p{Upper}> means the same thing as C<\p{upper}> or even C<\p{UpPeR}>.
461Similarly, you can add or subtract underscores anywhere in the middle of a
462word, so that these are also equivalent to C<\p{U_p_p_e_r}>.  And white space
463is generally irrelevant adjacent to non-word characters, such as the
464braces and the equals or colon separators, so C<\p{   Upper  }> and
465C<\p{ Upper_case : Y }> are equivalent to these as well.  In fact, white
466space and even hyphens can usually be added or deleted anywhere.  So
467even C<\p{ Up-per case = Yes}> is equivalent.  All this is called
468"loose-matching" by Unicode.  The "name" property has some restrictions
469on this due to a few outlier names.  Full details are given in
470L<https://www.unicode.org/reports/tr44/tr44-24.html#UAX44-LM2>.
471
472The few places where stricter matching is
473used is in the middle of numbers, the "name" property, and in the Perl
474extension properties that begin or end with an underscore.  Stricter
475matching cares about white space (except adjacent to non-word
476characters), hyphens, and non-interior underscores.
477
478You can also use negation in both C<\p{}> and C<\P{}> by introducing a caret
479(C<^>) between the first brace and the property name: C<\p{^Tamil}> is
480equal to C<\P{Tamil}>.
481
482Almost all properties are immune to case-insensitive matching.  That is,
483adding a C</i> regular expression modifier does not change what they
484match.  There are two sets that are affected.
485The first set is
486C<Uppercase_Letter>,
487C<Lowercase_Letter>,
488and C<Titlecase_Letter>,
489all of which match C<Cased_Letter> under C</i> matching.
490And the second set is
491C<Uppercase>,
492C<Lowercase>,
493and C<Titlecase>,
494all of which match C<Cased> under C</i> matching.
495This set also includes its subsets C<PosixUpper> and C<PosixLower> both
496of which under C</i> match C<PosixAlpha>.
497(The difference between these sets is that some things, such as Roman
498numerals, come in both upper and lower case so they are C<Cased>, but
499aren't considered letters, so they aren't C<Cased_Letter>'s.)
500
501See L</Beyond Unicode code points> for special considerations when
502matching Unicode properties against non-Unicode code points.
503
504=head3 B<General_Category>
505
506Every Unicode character is assigned a general category, which is the "most
507usual categorization of a character" (from
508L<https://www.unicode.org/reports/tr44>).
509
510The compound way of writing these is like C<\p{General_Category=Number}>
511(short: C<\p{gc:n}>).  But Perl furnishes shortcuts in which everything up
512through the equal or colon separator is omitted.  So you can instead just write
513C<\pN>.
514
515Here are the short and long forms of the values the C<General Category> property
516can have:
517
518    Short       Long
519
520    L           Letter
521    LC, L&      Cased_Letter (that is: [\p{Ll}\p{Lu}\p{Lt}])
522    Lu          Uppercase_Letter
523    Ll          Lowercase_Letter
524    Lt          Titlecase_Letter
525    Lm          Modifier_Letter
526    Lo          Other_Letter
527
528    M           Mark
529    Mn          Nonspacing_Mark
530    Mc          Spacing_Mark
531    Me          Enclosing_Mark
532
533    N           Number
534    Nd          Decimal_Number (also Digit)
535    Nl          Letter_Number
536    No          Other_Number
537
538    P           Punctuation (also Punct)
539    Pc          Connector_Punctuation
540    Pd          Dash_Punctuation
541    Ps          Open_Punctuation
542    Pe          Close_Punctuation
543    Pi          Initial_Punctuation
544                (may behave like Ps or Pe depending on usage)
545    Pf          Final_Punctuation
546                (may behave like Ps or Pe depending on usage)
547    Po          Other_Punctuation
548
549    S           Symbol
550    Sm          Math_Symbol
551    Sc          Currency_Symbol
552    Sk          Modifier_Symbol
553    So          Other_Symbol
554
555    Z           Separator
556    Zs          Space_Separator
557    Zl          Line_Separator
558    Zp          Paragraph_Separator
559
560    C           Other
561    Cc          Control (also Cntrl)
562    Cf          Format
563    Cs          Surrogate
564    Co          Private_Use
565    Cn          Unassigned
566
567Single-letter properties match all characters in any of the
568two-letter sub-properties starting with the same letter.
569C<LC> and C<L&> are special: both are aliases for the set consisting of everything matched by C<Ll>, C<Lu>, and C<Lt>.
570
571=head3 B<Bidirectional Character Types>
572
573Because scripts differ in their directionality (Hebrew and Arabic are
574written right to left, for example) Unicode supplies a C<Bidi_Class> property.
575Some of the values this property can have are:
576
577    Value       Meaning
578
579    L           Left-to-Right
580    LRE         Left-to-Right Embedding
581    LRO         Left-to-Right Override
582    R           Right-to-Left
583    AL          Arabic Letter
584    RLE         Right-to-Left Embedding
585    RLO         Right-to-Left Override
586    PDF         Pop Directional Format
587    EN          European Number
588    ES          European Separator
589    ET          European Terminator
590    AN          Arabic Number
591    CS          Common Separator
592    NSM         Non-Spacing Mark
593    BN          Boundary Neutral
594    B           Paragraph Separator
595    S           Segment Separator
596    WS          Whitespace
597    ON          Other Neutrals
598
599This property is always written in the compound form.
600For example, C<\p{Bidi_Class:R}> matches characters that are normally
601written right to left.  Unlike the
602C<L</General_Category>> property, this
603property can have more values added in a future Unicode release.  Those
604listed above comprised the complete set for many Unicode releases, but
605others were added in Unicode 6.3; you can always find what the
606current ones are in L<perluniprops>.  And
607L<https://www.unicode.org/reports/tr9/> describes how to use them.
608
609=head3 B<Scripts>
610
611The world's languages are written in many different scripts.  This sentence
612(unless you're reading it in translation) is written in Latin, while Russian is
613written in Cyrillic, and Greek is written in, well, Greek; Japanese mainly in
614Hiragana or Katakana.  There are many more.
615
616The Unicode C<Script> and C<Script_Extensions> properties give what
617script a given character is in.  The C<Script_Extensions> property is an
618improved version of C<Script>, as demonstrated below.  Either property
619can be specified with the compound form like
620C<\p{Script=Hebrew}> (short: C<\p{sc=hebr}>), or
621C<\p{Script_Extensions=Javanese}> (short: C<\p{scx=java}>).
622In addition, Perl furnishes shortcuts for all
623C<Script_Extensions> property names.  You can omit everything up through
624the equals (or colon), and simply write C<\p{Latin}> or C<\P{Cyrillic}>.
625(This is not true for C<Script>, which is required to be
626written in the compound form.  Prior to Perl v5.26, the single form
627returned the plain old C<Script> version, but was changed because
628C<Script_Extensions> gives better results.)
629
630The difference between these two properties involves characters that are
631used in multiple scripts.  For example the digits '0' through '9' are
632used in many parts of the world.  These are placed in a script named
633C<Common>.  Other characters are used in just a few scripts.  For
634example, the C<"KATAKANA-HIRAGANA DOUBLE HYPHEN"> is used in both Japanese
635scripts, Katakana and Hiragana, but nowhere else.  The C<Script>
636property places all characters that are used in multiple scripts in the
637C<Common> script, while the C<Script_Extensions> property places those
638that are used in only a few scripts into each of those scripts; while
639still using C<Common> for those used in many scripts.  Thus both these
640match:
641
642 "0" =~ /\p{sc=Common}/     # Matches
643 "0" =~ /\p{scx=Common}/    # Matches
644
645and only the first of these match:
646
647 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Common}  # Matches
648 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Common} # No match
649
650And only the last two of these match:
651
652 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Hiragana}  # No match
653 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{sc=Katakana}  # No match
654 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Hiragana} # Matches
655 "\N{KATAKANA-HIRAGANA DOUBLE HYPHEN}" =~ /\p{scx=Katakana} # Matches
656
657C<Script_Extensions> is thus an improved C<Script>, in which there are
658fewer characters in the C<Common> script, and correspondingly more in
659other scripts.  It is new in Unicode version 6.0, and its data are likely
660to change significantly in later releases, as things get sorted out.
661New code should probably be using C<Script_Extensions> and not plain
662C<Script>.  If you compile perl with a Unicode release that doesn't have
663C<Script_Extensions>, the single form Perl extensions will instead refer
664to the plain C<Script> property.  If you compile with a version of
665Unicode that doesn't have the C<Script> property, these extensions will
666not be defined at all.
667
668(Actually, besides C<Common>, the C<Inherited> script, contains
669characters that are used in multiple scripts.  These are modifier
670characters which inherit the script value
671of the controlling character.  Some of these are used in many scripts,
672and so go into C<Inherited> in both C<Script> and C<Script_Extensions>.
673Others are used in just a few scripts, so are in C<Inherited> in
674C<Script>, but not in C<Script_Extensions>.)
675
676It is worth stressing that there are several different sets of digits in
677Unicode that are equivalent to 0-9 and are matchable by C<\d> in a
678regular expression.  If they are used in a single language only, they
679are in that language's C<Script> and C<Script_Extensions>.  If they are
680used in more than one script, they will be in C<sc=Common>, but only
681if they are used in many scripts should they be in C<scx=Common>.
682
683The explanation above has omitted some detail; refer to UAX#24 "Unicode
684Script Property": L<https://www.unicode.org/reports/tr24>.
685
686A complete list of scripts and their shortcuts is in L<perluniprops>.
687
688=head3 B<Use of the C<"Is"> Prefix>
689
690For backward compatibility (with ancient Perl 5.6), all properties writable
691without using the compound form mentioned
692so far may have C<Is> or C<Is_> prepended to their name, so C<\P{Is_Lu}>, for
693example, is equal to C<\P{Lu}>, and C<\p{IsScript:Arabic}> is equal to
694C<\p{Arabic}>.
695
696=head3 B<Blocks>
697
698In addition to B<scripts>, Unicode also defines B<blocks> of
699characters.  The difference between scripts and blocks is that the
700concept of scripts is closer to natural languages, while the concept
701of blocks is more of an artificial grouping based on groups of Unicode
702characters with consecutive ordinal values. For example, the C<"Basic Latin">
703block is all the characters whose ordinals are between 0 and 127, inclusive; in
704other words, the ASCII characters.  The C<"Latin"> script contains some letters
705from this as well as several other blocks, like C<"Latin-1 Supplement">,
706C<"Latin Extended-A">, I<etc.>, but it does not contain all the characters from
707those blocks. It does not, for example, contain the digits 0-9, because
708those digits are shared across many scripts, and hence are in the
709C<Common> script.
710
711For more about scripts versus blocks, see UAX#24 "Unicode Script Property":
712L<https://www.unicode.org/reports/tr24>
713
714The C<Script_Extensions> or C<Script> properties are likely to be the
715ones you want to use when processing
716natural language; the C<Block> property may occasionally be useful in working
717with the nuts and bolts of Unicode.
718
719Block names are matched in the compound form, like C<\p{Block: Arrows}> or
720C<\p{Blk=Hebrew}>.  Unlike most other properties, only a few block names have a
721Unicode-defined short name.
722
723Perl also defines single form synonyms for the block property in cases
724where these do not conflict with something else.  But don't use any of
725these, because they are unstable.  Since these are Perl extensions, they
726are subordinate to official Unicode property names; Unicode doesn't know
727nor care about Perl's extensions.  It may happen that a name that
728currently means the Perl extension will later be changed without warning
729to mean a different Unicode property in a future version of the perl
730interpreter that uses a later Unicode release, and your code would no
731longer work.  The extensions are mentioned here for completeness:  Take
732the block name and prefix it with one of: C<In> (for example
733C<\p{Blk=Arrows}> can currently be written as C<\p{In_Arrows}>); or
734sometimes C<Is> (like C<\p{Is_Arrows}>); or sometimes no prefix at all
735(C<\p{Arrows}>).  As of this writing (Unicode 9.0) there are no
736conflicts with using the C<In_> prefix, but there are plenty with the
737other two forms.  For example, C<\p{Is_Hebrew}> and C<\p{Hebrew}> mean
738C<\p{Script_Extensions=Hebrew}> which is NOT the same thing as
739C<\p{Blk=Hebrew}>.  Our
740advice used to be to use the C<In_> prefix as a single form way of
741specifying a block.  But Unicode 8.0 added properties whose names begin
742with C<In>, and it's now clear that it's only luck that's so far
743prevented a conflict.  Using C<In> is only marginally less typing than
744C<Blk:>, and the latter's meaning is clearer anyway, and guaranteed to
745never conflict.  So don't take chances.  Use C<\p{Blk=foo}> for new
746code.  And be sure that block is what you really really want to do.  In
747most cases scripts are what you want instead.
748
749A complete list of blocks is in L<perluniprops>.
750
751=head3 B<Other Properties>
752
753There are many more properties than the very basic ones described here.
754A complete list is in L<perluniprops>.
755
756Unicode defines all its properties in the compound form, so all single-form
757properties are Perl extensions.  Most of these are just synonyms for the
758Unicode ones, but some are genuine extensions, including several that are in
759the compound form.  And quite a few of these are actually recommended by Unicode
760(in L<https://www.unicode.org/reports/tr18>).
761
762This section gives some details on all extensions that aren't just
763synonyms for compound-form Unicode properties
764(for those properties, you'll have to refer to the
765L<Unicode Standard|https://www.unicode.org/reports/tr44>.
766
767=over
768
769=item B<C<\p{All}>>
770
771This matches every possible code point.  It is equivalent to C<qr/./s>.
772Unlike all the other non-user-defined C<\p{}> property matches, no
773warning is ever generated if this is property is matched against a
774non-Unicode code point (see L</Beyond Unicode code points> below).
775
776=item B<C<\p{Alnum}>>
777
778This matches any C<\p{Alphabetic}> or C<\p{Decimal_Number}> character.
779
780=item B<C<\p{Any}>>
781
782This matches any of the 1_114_112 Unicode code points.  It is a synonym
783for C<\p{Unicode}>.
784
785=item B<C<\p{ASCII}>>
786
787This matches any of the 128 characters in the US-ASCII character set,
788which is a subset of Unicode.
789
790=item B<C<\p{Assigned}>>
791
792This matches any assigned code point; that is, any code point whose L<general
793category|/General_Category> is not C<Unassigned> (or equivalently, not C<Cn>).
794
795=item B<C<\p{Blank}>>
796
797This is the same as C<\h> and C<\p{HorizSpace}>:  A character that changes the
798spacing horizontally.
799
800=item B<C<\p{Decomposition_Type: Non_Canonical}>>    (Short: C<\p{Dt=NonCanon}>)
801
802Matches a character that has a non-canonical decomposition.
803
804The L</Extended Grapheme Clusters (Logical characters)> section above
805talked about canonical decompositions.  However, many more characters
806have a different type of decomposition, a "compatible" or
807"non-canonical" decomposition.  The sequences that form these
808decompositions are not considered canonically equivalent to the
809pre-composed character.  An example is the C<"SUPERSCRIPT ONE">.  It is
810somewhat like a regular digit 1, but not exactly; its decomposition into
811the digit 1 is called a "compatible" decomposition, specifically a
812"super" decomposition.  There are several such compatibility
813decompositions (see L<https://www.unicode.org/reports/tr44>), including
814one called "compat", which means some miscellaneous type of
815decomposition that doesn't fit into the other decomposition categories
816that Unicode has chosen.
817
818Note that most Unicode characters don't have a decomposition, so their
819decomposition type is C<"None">.
820
821For your convenience, Perl has added the C<Non_Canonical> decomposition
822type to mean any of the several compatibility decompositions.
823
824=item B<C<\p{Graph}>>
825
826Matches any character that is graphic.  Theoretically, this means a character
827that on a printer would cause ink to be used.
828
829=item B<C<\p{HorizSpace}>>
830
831This is the same as C<\h> and C<\p{Blank}>:  a character that changes the
832spacing horizontally.
833
834=item B<C<\p{In=*}>>
835
836This is a synonym for C<\p{Present_In=*}>
837
838=item B<C<\p{PerlSpace}>>
839
840This is the same as C<\s>, restricted to ASCII, namely C<S<[ \f\n\r\t]>>
841and starting in Perl v5.18, a vertical tab.
842
843Mnemonic: Perl's (original) space
844
845=item B<C<\p{PerlWord}>>
846
847This is the same as C<\w>, restricted to ASCII, namely C<[A-Za-z0-9_]>
848
849Mnemonic: Perl's (original) word.
850
851=item B<C<\p{Posix...}>>
852
853There are several of these, which are equivalents, using the C<\p{}>
854notation, for Posix classes and are described in
855L<perlrecharclass/POSIX Character Classes>.
856
857=item B<C<\p{Present_In: *}>>    (Short: C<\p{In=*}>)
858
859This property is used when you need to know in what Unicode version(s) a
860character is.
861
862The "*" above stands for some Unicode version number, such as
863C<1.1> or C<12.0>; or the "*" can also be C<Unassigned>.  This property will
864match the code points whose final disposition has been settled as of the
865Unicode release given by the version number; C<\p{Present_In: Unassigned}>
866will match those code points whose meaning has yet to be assigned.
867
868For example, C<U+0041> C<"LATIN CAPITAL LETTER A"> was present in the very first
869Unicode release available, which is C<1.1>, so this property is true for all
870valid "*" versions.  On the other hand, C<U+1EFF> was not assigned until version
8715.1 when it became C<"LATIN SMALL LETTER Y WITH LOOP">, so the only "*" that
872would match it are 5.1, 5.2, and later.
873
874Unicode furnishes the C<Age> property from which this is derived.  The problem
875with Age is that a strict interpretation of it (which Perl takes) has it
876matching the precise release a code point's meaning is introduced in.  Thus
877C<U+0041> would match only 1.1; and C<U+1EFF> only 5.1.  This is not usually what
878you want.
879
880Some non-Perl implementations of the Age property may change its meaning to be
881the same as the Perl C<Present_In> property; just be aware of that.
882
883Another confusion with both these properties is that the definition is not
884that the code point has been I<assigned>, but that the meaning of the code point
885has been I<determined>.  This is because 66 code points will always be
886unassigned, and so the C<Age> for them is the Unicode version in which the decision
887to make them so was made.  For example, C<U+FDD0> is to be permanently
888unassigned to a character, and the decision to do that was made in version 3.1,
889so C<\p{Age=3.1}> matches this character, as also does C<\p{Present_In: 3.1}> and up.
890
891=item B<C<\p{Print}>>
892
893This matches any character that is graphical or blank, except controls.
894
895=item B<C<\p{SpacePerl}>>
896
897This is the same as C<\s>, including beyond ASCII.
898
899Mnemonic: Space, as modified by Perl.  (It doesn't include the vertical tab
900until v5.18, which both the Posix standard and Unicode consider white space.)
901
902=item B<C<\p{Title}>> and  B<C<\p{Titlecase}>>
903
904Under case-sensitive matching, these both match the same code points as
905C<\p{General Category=Titlecase_Letter}> (C<\p{gc=lt}>).  The difference
906is that under C</i> caseless matching, these match the same as
907C<\p{Cased}>, whereas C<\p{gc=lt}> matches C<\p{Cased_Letter>).
908
909=item B<C<\p{Unicode}>>
910
911This matches any of the 1_114_112 Unicode code points.
912C<\p{Any}>.
913
914=item B<C<\p{VertSpace}>>
915
916This is the same as C<\v>:  A character that changes the spacing vertically.
917
918=item B<C<\p{Word}>>
919
920This is the same as C<\w>, including over 100_000 characters beyond ASCII.
921
922=item B<C<\p{XPosix...}>>
923
924There are several of these, which are the standard Posix classes
925extended to the full Unicode range.  They are described in
926L<perlrecharclass/POSIX Character Classes>.
927
928=back
929
930=head2 Comparison of C<\N{...}> and C<\p{name=...}>
931
932Starting in Perl 5.32, you can specify a character by its name in
933regular expression patterns using C<\p{name=...}>.  This is in addition
934to the longstanding method of using C<\N{...}>.  The following
935summarizes the differences between these two:
936
937                       \N{...}       \p{Name=...}
938 can interpolate    only with eval       yes            [1]
939 custom names            yes             no             [2]
940 name aliases            yes             yes            [3]
941 named sequences         yes             yes            [4]
942 name value parsing     exact       Unicode loose       [5]
943
944=over
945
946=item [1]
947
948The ability to interpolate means you can do something like
949
950 qr/\p{na=latin capital letter $which}/
951
952and specify C<$which> elsewhere.
953
954=item [2]
955
956You can create your own names for characters, and override official
957ones when using C<\N{...}>.  See L<charnames/CUSTOM ALIASES>.
958
959=item [3]
960
961Some characters have multiple names (synonyms).
962
963=item [4]
964
965Some particular sequences of characters are given a single name, in
966addition to their individual ones.
967
968=item [5]
969
970Exact name value matching means you have to specify case, hyphens,
971underscores, and spaces precisely in the name you want.  Loose matching
972follows the Unicode rules
973L<https://www.unicode.org/reports/tr44/tr44-24.html#UAX44-LM2>,
974where these are mostly irrelevant.  Except for a few outlier character
975names, these are the same rules as are already used for any other
976C<\p{...}> property.
977
978=back
979
980=head2 Wildcards in Property Values
981
982Starting in Perl 5.30, it is possible to do something like this:
983
984 qr!\p{numeric_value=/\A[0-5]\z/}!
985
986or, by abbreviating and adding C</x>,
987
988 qr! \p{nv= /(?x) \A [0-5] \z / }!
989
990This matches all code points whose numeric value is one of 0, 1, 2, 3,
9914, or 5.  This particular example could instead have been written as
992
993 qr! \A [ \p{nv=0}\p{nv=1}\p{nv=2}\p{nv=3}\p{nv=4}\p{nv=5} ] \z !xx
994
995in earlier perls, so in this case this feature just makes things easier
996and shorter to write.  If we hadn't included the C<\A> and C<\z>, these
997would have matched things like C<1E<sol>2> because that contains a 1 (as
998well as a 2).  As written, it matches things like subscripts that have
999these numeric values.  If we only wanted the decimal digits with those
1000numeric values, we could say,
1001
1002 qr! (?[ \d & \p{nv=/[0-5]/ ]) }!x
1003
1004The C<\d> gets rid of needing to anchor the pattern, since it forces the
1005result to only match C<[0-9]>, and the C<[0-5]> further restricts it.
1006
1007The text in the above examples enclosed between the C<"E<sol>">
1008characters can be just about any regular expression.  It is independent
1009of the main pattern, so doesn't share any capturing groups, I<etc>.  The
1010delimiters for it must be ASCII punctuation, but it may NOT be
1011delimited by C<"{">, nor C<"}"> nor contain a literal C<"}">, as that
1012delimits the end of the enclosing C<\p{}>.  Like any pattern, certain
1013other delimiters are terminated by their mirror images.  These are
1014C<"(">, C<"[>", and C<"E<lt>">.  If the delimiter is any of C<"-">,
1015C<"_">, C<"+">, or C<"\">, or is the same delimiter as is used for the
1016enclosing pattern, it must be preceded by a backslash escape, both
1017fore and aft.
1018
1019Beware of using C<"$"> to indicate to match the end of the string.  It
1020can too easily be interpreted as being a punctuation variable, like
1021C<$/>.
1022
1023No modifiers may follow the final delimiter.  Instead, use
1024L<perlre/(?adlupimnsx-imnsx)> and/or
1025L<perlre/(?adluimnsx-imnsx:pattern)> to specify modifiers.
1026However, certain modifiers are illegal in your wildcard subpattern.
1027The only character set modifier specifiable is C</aa>;
1028any other character set, and C<-m>, and C<p>, and C<s> are all illegal.
1029Specifying modifiers like C<qr/.../gc> that aren't legal in the
1030C<(?...)> notation normally raise a warning, but with wildcard
1031subpatterns, their use is an error.  The C<m> modifier is ineffective;
1032everything that matches will be a single line.
1033
1034By default, your pattern is matched case-insensitively, as if C</i> had
1035been specified.  You can change this by saying C<(?-i)> in your pattern.
1036
1037There are also certain operations that are illegal.  You can't nest
1038C<\p{...}> and C<\P{...}> calls within a wildcard subpattern, and C<\G>
1039doesn't make sense, so is also prohibited.
1040
1041And the C<*> quantifier (or its equivalent C<(0,}>) is illegal.
1042
1043This feature is not available when the left-hand side is prefixed by
1044C<Is_>, nor for any form that is marked as "Discouraged" in
1045L<perluniprops/Discouraged>.
1046
1047This experimental feature has been added to begin to implement
1048L<https://www.unicode.org/reports/tr18/#Wildcard_Properties>.  Using it
1049will raise a (default-on) warning in the
1050C<experimental::uniprop_wildcards> category.  We reserve the right to
1051change its operation as we gain experience.
1052
1053Your subpattern can be just about anything, but for it to have some
1054utility, it should match when called with either or both of
1055a) the full name of the property value with underscores (and/or spaces
1056in the Block property) and some things uppercase; or b) the property
1057value in all lowercase with spaces and underscores squeezed out.  For
1058example,
1059
1060 qr!\p{Blk=/Old I.*/}!
1061 qr!\p{Blk=/oldi.*/}!
1062
1063would match the same things.
1064
1065Another example that shows that within C<\p{...}>, C</x> isn't needed to
1066have spaces:
1067
1068 qr!\p{scx= /Hebrew|Greek/ }!
1069
1070To be safe, we should have anchored the above example, to prevent
1071matches for something like C<Hebrew_Braille>, but there aren't
1072any script names like that, so far.
1073A warning is issued if none of the legal values for a property are
1074matched by your pattern.  It's likely that a future release will raise a
1075warning if your pattern ends up causing every possible code point to
1076match.
1077
1078Starting in 5.32, the Name, Name Aliases, and Named Sequences properties
1079are allowed to be matched.  They are considered to be a single
1080combination property, just as has long been the case for C<\N{}>.  Loose
1081matching doesn't work in exactly the same way for these as it does for
1082the values of other properties.  The rules are given in
1083L<https://www.unicode.org/reports/tr44/tr44-24.html#UAX44-LM2>.  As a
1084result, Perl doesn't try loose matching for you, like it does in other
1085properties.  All letters in names are uppercase, but you can add C<(?i)>
1086to your subpattern to ignore case.  If you're uncertain where a blank
1087is, you can use C< ?> in your subpattern.  No character name contains an
1088underscore, so don't bother trying to match one.  The use of hyphens is
1089particularly problematic; refer to the above link.  But note that, as of
1090Unicode 13.0, the only script in modern usage which has weirdnesses with
1091these is Tibetan; also the two Korean characters U+116C HANGUL JUNGSEONG
1092OE and U+1180 HANGUL JUNGSEONG O-E.  Unicode makes no promises to not
1093add hyphen-problematic names in the future.
1094
1095Using wildcards on these is resource intensive, given the hundreds of
1096thousands of legal names that must be checked against.
1097
1098An example of using Name property wildcards is
1099
1100 qr!\p{name=/(SMILING|GRINNING) FACE/}!
1101
1102Another is
1103
1104 qr/(?[ \p{name=\/CJK\/} - \p{ideographic} ])/
1105
1106which is the 200-ish (as of Unicode 13.0) CJK characters that aren't
1107ideographs.
1108
1109There are certain properties that wildcard subpatterns don't currently
1110work with.  These are:
1111
1112 Bidi Mirroring Glyph
1113 Bidi Paired Bracket
1114 Case Folding
1115 Decomposition Mapping
1116 Equivalent Unified Ideograph
1117 Lowercase Mapping
1118 NFKC Case Fold
1119 Titlecase Mapping
1120 Uppercase Mapping
1121
1122Nor is the C<@I<unicode_property>@> form implemented.
1123
1124Here's a complete example of matching IPV4 internet protocol addresses
1125in any (single) script
1126
1127 no warnings 'experimental::regex_sets';
1128 no warnings 'experimental::uniprop_wildcards';
1129
1130 # Can match a substring, so this intermediate regex needs to have
1131 # context or anchoring in its final use.  Using nt=de yields decimal
1132 # digits.  When specifying a subset of these, we must include \d to
1133 # prevent things like U+00B2 SUPERSCRIPT TWO from matching
1134 my $zero_through_255 =
1135  qr/ \b (*sr:                                  # All from same sript
1136            (?[ \p{nv=0} & \d ])*               # Optional leading zeros
1137        (                                       # Then one of:
1138                                  \d{1,2}       #   0 - 99
1139            | (?[ \p{nv=1} & \d ])  \d{2}       #   100 - 199
1140            | (?[ \p{nv=2} & \d ])
1141               (  (?[ \p{nv=:[0-4]:} & \d ]) \d #   200 - 249
1142                | (?[ \p{nv=5}     & \d ])
1143                  (?[ \p{nv=:[0-5]:} & \d ])    #   250 - 255
1144               )
1145        )
1146      )
1147    \b
1148  /x;
1149
1150 my $ipv4 = qr/ \A (*sr:         $zero_through_255
1151                         (?: [.] $zero_through_255 ) {3}
1152                   )
1153                \z
1154            /x;
1155
1156=head2 User-Defined Character Properties
1157
1158You can define your own binary character properties by defining subroutines
1159whose names begin with C<"In"> or C<"Is">.  (The experimental feature
1160L<perlre/(?[ ])> provides an alternative which allows more complex
1161definitions.)  The subroutines can be defined in any
1162package.  They override any Unicode properties expressed as the same
1163names.  The user-defined properties can be used in the regular
1164expression
1165C<\p{}> and C<\P{}> constructs; if you are using a user-defined property from a
1166package other than the one you are in, you must specify its package in the
1167C<\p{}> or C<\P{}> construct.
1168
1169    # assuming property Is_Foreign defined in Lang::
1170    package main;  # property package name required
1171    if ($txt =~ /\p{Lang::IsForeign}+/) { ... }
1172
1173    package Lang;  # property package name not required
1174    if ($txt =~ /\p{IsForeign}+/) { ... }
1175
1176
1177Note that the effect is compile-time and immutable once defined.
1178However, the subroutines are passed a single parameter, which is 0 if
1179case-sensitive matching is in effect and non-zero if caseless matching
1180is in effect.  The subroutine may return different values depending on
1181the value of the flag, and one set of values will immutably be in effect
1182for all case-sensitive matches, and the other set for all case-insensitive
1183matches.
1184
1185Note that if the regular expression is tainted, then Perl will die rather
1186than calling the subroutine when the name of the subroutine is
1187determined by the tainted data.
1188
1189The subroutines must return a specially-formatted string, with one
1190or more newline-separated lines.  Each line must be one of the following:
1191
1192=over 4
1193
1194=item *
1195
1196A single hexadecimal number denoting a code point to include.
1197
1198=item *
1199
1200Two hexadecimal numbers separated by horizontal whitespace (space or
1201tabular characters) denoting a range of code points to include.  The
1202second number must not be smaller than the first.
1203
1204=item *
1205
1206Something to include, prefixed by C<"+">: a built-in character
1207property (prefixed by C<"utf8::">) or a fully qualified (including package
1208name) user-defined character property,
1209to represent all the characters in that property; two hexadecimal code
1210points for a range; or a single hexadecimal code point.
1211
1212=item *
1213
1214Something to exclude, prefixed by C<"-">: an existing character
1215property (prefixed by C<"utf8::">) or a fully qualified (including package
1216name) user-defined character property,
1217to represent all the characters in that property; two hexadecimal code
1218points for a range; or a single hexadecimal code point.
1219
1220=item *
1221
1222Something to negate, prefixed C<"!">: an existing character
1223property (prefixed by C<"utf8::">) or a fully qualified (including package
1224name) user-defined character property,
1225to represent all the characters in that property; two hexadecimal code
1226points for a range; or a single hexadecimal code point.
1227
1228=item *
1229
1230Something to intersect with, prefixed by C<"&">: an existing character
1231property (prefixed by C<"utf8::">) or a fully qualified (including package
1232name) user-defined character property,
1233for all the characters except the characters in the property; two
1234hexadecimal code points for a range; or a single hexadecimal code point.
1235
1236=back
1237
1238For example, to define a property that covers both the Japanese
1239syllabaries (hiragana and katakana), you can define
1240
1241    sub InKana {
1242        return <<END;
1243    3040\t309F
1244    30A0\t30FF
1245    END
1246    }
1247
1248Imagine that the here-doc end marker is at the beginning of the line.
1249Now you can use C<\p{InKana}> and C<\P{InKana}>.
1250
1251You could also have used the existing block property names:
1252
1253    sub InKana {
1254        return <<'END';
1255    +utf8::InHiragana
1256    +utf8::InKatakana
1257    END
1258    }
1259
1260Suppose you wanted to match only the allocated characters,
1261not the raw block ranges: in other words, you want to remove
1262the unassigned characters:
1263
1264    sub InKana {
1265        return <<'END';
1266    +utf8::InHiragana
1267    +utf8::InKatakana
1268    -utf8::IsCn
1269    END
1270    }
1271
1272The negation is useful for defining (surprise!) negated classes.
1273
1274    sub InNotKana {
1275        return <<'END';
1276    !utf8::InHiragana
1277    -utf8::InKatakana
1278    +utf8::IsCn
1279    END
1280    }
1281
1282This will match all non-Unicode code points, since every one of them is
1283not in Kana.  You can use intersection to exclude these, if desired, as
1284this modified example shows:
1285
1286    sub InNotKana {
1287        return <<'END';
1288    !utf8::InHiragana
1289    -utf8::InKatakana
1290    +utf8::IsCn
1291    &utf8::Any
1292    END
1293    }
1294
1295C<&utf8::Any> must be the last line in the definition.
1296
1297Intersection is used generally for getting the common characters matched
1298by two (or more) classes.  It's important to remember not to use C<"&"> for
1299the first set; that would be intersecting with nothing, resulting in an
1300empty set.  (Similarly using C<"-"> for the first set does nothing).
1301
1302Unlike non-user-defined C<\p{}> property matches, no warning is ever
1303generated if these properties are matched against a non-Unicode code
1304point (see L</Beyond Unicode code points> below).
1305
1306=head2 User-Defined Case Mappings (for serious hackers only)
1307
1308B<This feature has been removed as of Perl 5.16.>
1309The CPAN module C<L<Unicode::Casing>> provides better functionality without
1310the drawbacks that this feature had.  If you are using a Perl earlier
1311than 5.16, this feature was most fully documented in the 5.14 version of
1312this pod:
1313L<http://perldoc.perl.org/5.14.0/perlunicode.html#User-Defined-Case-Mappings-%28for-serious-hackers-only%29>
1314
1315=head2 Character Encodings for Input and Output
1316
1317See L<Encode>.
1318
1319=head2 Unicode Regular Expression Support Level
1320
1321The following list of Unicode supported features for regular expressions describes
1322all features currently directly supported by core Perl.  The references
1323to "Level I<N>" and the section numbers refer to
1324L<UTS#18 "Unicode Regular Expressions"|https://www.unicode.org/reports/tr18>,
1325version 18, October 2016.
1326
1327=head3 Level 1 - Basic Unicode Support
1328
1329 RL1.1   Hex Notation                     - Done          [1]
1330 RL1.2   Properties                       - Done          [2]
1331 RL1.2a  Compatibility Properties         - Done          [3]
1332 RL1.3   Subtraction and Intersection     - Experimental  [4]
1333 RL1.4   Simple Word Boundaries           - Done          [5]
1334 RL1.5   Simple Loose Matches             - Done          [6]
1335 RL1.6   Line Boundaries                  - Partial       [7]
1336 RL1.7   Supplementary Code Points        - Done          [8]
1337
1338=over 4
1339
1340=item [1] C<\N{U+...}> and C<\x{...}>
1341
1342=item [2]
1343C<\p{...}> C<\P{...}>.  This requirement is for a minimal list of
1344properties.  Perl supports these.  See R2.7 for other properties.
1345
1346=item [3]
1347Perl has C<\d> C<\D> C<\s> C<\S> C<\w> C<\W> C<\X> C<[:I<prop>:]>
1348C<[:^I<prop>:]>, plus all the properties specified by
1349L<https://www.unicode.org/reports/tr18/#Compatibility_Properties>.  These
1350are described above in L</Other Properties>
1351
1352=item [4]
1353
1354The experimental feature C<"(?[...])"> starting in v5.18 accomplishes
1355this.
1356
1357See L<perlre/(?[ ])>.  If you don't want to use an experimental
1358feature, you can use one of the following:
1359
1360=over 4
1361
1362=item *
1363Regular expression lookahead
1364
1365You can mimic class subtraction using lookahead.
1366For example, what UTS#18 might write as
1367
1368    [{Block=Greek}-[{UNASSIGNED}]]
1369
1370in Perl can be written as:
1371
1372    (?!\p{Unassigned})\p{Block=Greek}
1373    (?=\p{Assigned})\p{Block=Greek}
1374
1375But in this particular example, you probably really want
1376
1377    \p{Greek}
1378
1379which will match assigned characters known to be part of the Greek script.
1380
1381=item *
1382
1383CPAN module C<L<Unicode::Regex::Set>>
1384
1385It does implement the full UTS#18 grouping, intersection, union, and
1386removal (subtraction) syntax.
1387
1388=item *
1389
1390L</"User-Defined Character Properties">
1391
1392C<"+"> for union, C<"-"> for removal (set-difference), C<"&"> for intersection
1393
1394=back
1395
1396=item [5]
1397C<\b> C<\B> meet most, but not all, the details of this requirement, but
1398C<\b{wb}> and C<\B{wb}> do, as well as the stricter R2.3.
1399
1400=item [6]
1401
1402Note that Perl does Full case-folding in matching, not Simple:
1403
1404For example C<U+1F88> is equivalent to C<U+1F00 U+03B9>, instead of just
1405C<U+1F80>.  This difference matters mainly for certain Greek capital
1406letters with certain modifiers: the Full case-folding decomposes the
1407letter, while the Simple case-folding would map it to a single
1408character.
1409
1410=item [7]
1411
1412The reason this is considered to be only partially implemented is that
1413Perl has L<C<qrE<sol>\b{lb}E<sol>>|perlrebackslash/\b{lb}> and
1414C<L<Unicode::LineBreak>> that are conformant with
1415L<UAX#14 "Unicode Line Breaking Algorithm"|https://www.unicode.org/reports/tr14>.
1416The regular expression construct provides default behavior, while the
1417heavier-weight module provides customizable line breaking.
1418
1419But Perl treats C<\n> as the start- and end-line
1420delimiter, whereas Unicode specifies more characters that should be
1421so-interpreted.
1422
1423These are:
1424
1425 VT   U+000B  (\v in C)
1426 FF   U+000C  (\f)
1427 CR   U+000D  (\r)
1428 NEL  U+0085
1429 LS   U+2028
1430 PS   U+2029
1431
1432C<^> and C<$> in regular expression patterns are supposed to match all
1433these, but don't.
1434These characters also don't, but should, affect C<< <> >> C<$.>, and
1435script line numbers.
1436
1437Also, lines should not be split within C<CRLF> (i.e. there is no
1438empty line between C<\r> and C<\n>).  For C<CRLF>, try the C<:crlf>
1439layer (see L<PerlIO>).
1440
1441=item [8]
1442UTF-8/UTF-EBDDIC used in Perl allows not only C<U+10000> to
1443C<U+10FFFF> but also beyond C<U+10FFFF>
1444
1445=back
1446
1447=head3 Level 2 - Extended Unicode Support
1448
1449 RL2.1   Canonical Equivalents           - Retracted     [9]
1450                                           by Unicode
1451 RL2.2   Extended Grapheme Clusters and  - Partial       [10]
1452         Character Classes with Strings
1453 RL2.3   Default Word Boundaries         - Done          [11]
1454 RL2.4   Default Case Conversion         - Done
1455 RL2.5   Name Properties                 - Done
1456 RL2.6   Wildcards in Property Values    - Partial       [12]
1457 RL2.7   Full Properties                 - Partial       [13]
1458 RL2.8   Optional Properties             - Partial       [14]
1459
1460=over 4
1461
1462=item [9]
1463Unicode has rewritten this portion of UTS#18 to say that getting
1464canonical equivalence (see UAX#15
1465L<"Unicode Normalization Forms"|https://www.unicode.org/reports/tr15>)
1466is basically to be done at the programmer level.  Use NFD to write
1467both your regular expressions and text to match them against (you
1468can use L<Unicode::Normalize>).
1469
1470=item [10]
1471Perl has C<\X> and C<\b{gcb}>.  Unicode has retracted their "Grapheme
1472Cluster Mode", and recently added string properties, which Perl does not
1473yet support.
1474
1475=item [11] see
1476L<UAX#29 "Unicode Text Segmentation"|https://www.unicode.org/reports/tr29>,
1477
1478=item [12] see
1479L</Wildcards in Property Values> above.
1480
1481=item [13]
1482Perl supports all the properties in the Unicode Character Database
1483(UCD).  It does not yet support the listed properties that come from
1484other Unicode sources.
1485
1486=item [14]
1487The only optional property that Perl supports is Named Sequence.  None
1488of these properties are in the UCD.
1489
1490=back
1491
1492=head3 Level 3 - Tailored Support
1493
1494This has been retracted by Unicode.
1495
1496=head2 Unicode Encodings
1497
1498Unicode characters are assigned to I<code points>, which are abstract
1499numbers.  To use these numbers, various encodings are needed.
1500
1501=over 4
1502
1503=item *
1504
1505UTF-8
1506
1507UTF-8 is a variable-length (1 to 4 bytes), byte-order independent
1508encoding.  In most of Perl's documentation, including elsewhere in this
1509document, the term "UTF-8" means also "UTF-EBCDIC".  But in this section,
1510"UTF-8" refers only to the encoding used on ASCII platforms.  It is a
1511superset of 7-bit US-ASCII, so anything encoded in ASCII has the
1512identical representation when encoded in UTF-8.
1513
1514The following table is from Unicode 3.2.
1515
1516 Code Points            1st Byte  2nd Byte  3rd Byte 4th Byte
1517
1518   U+0000..U+007F       00..7F
1519   U+0080..U+07FF     * C2..DF    80..BF
1520   U+0800..U+0FFF       E0      * A0..BF    80..BF
1521   U+1000..U+CFFF       E1..EC    80..BF    80..BF
1522   U+D000..U+D7FF       ED        80..9F    80..BF
1523   U+D800..U+DFFF       +++++ utf16 surrogates, not legal utf8 +++++
1524   U+E000..U+FFFF       EE..EF    80..BF    80..BF
1525  U+10000..U+3FFFF      F0      * 90..BF    80..BF    80..BF
1526  U+40000..U+FFFFF      F1..F3    80..BF    80..BF    80..BF
1527 U+100000..U+10FFFF     F4        80..8F    80..BF    80..BF
1528
1529Note the gaps marked by "*" before several of the byte entries above.  These are
1530caused by legal UTF-8 avoiding non-shortest encodings: it is technically
1531possible to UTF-8-encode a single code point in different ways, but that is
1532explicitly forbidden, and the shortest possible encoding should always be used
1533(and that is what Perl does).
1534
1535Another way to look at it is via bits:
1536
1537                Code Points  1st Byte  2nd Byte  3rd Byte  4th Byte
1538
1539                   0aaaaaaa  0aaaaaaa
1540           00000bbbbbaaaaaa  110bbbbb  10aaaaaa
1541           ccccbbbbbbaaaaaa  1110cccc  10bbbbbb  10aaaaaa
1542 00000dddccccccbbbbbbaaaaaa  11110ddd  10cccccc  10bbbbbb  10aaaaaa
1543
1544As you can see, the continuation bytes all begin with C<"10">, and the
1545leading bits of the start byte tell how many bytes there are in the
1546encoded character.
1547
1548The original UTF-8 specification allowed up to 6 bytes, to allow
1549encoding of numbers up to C<0x7FFF_FFFF>.  Perl continues to allow those,
1550and has extended that up to 13 bytes to encode code points up to what
1551can fit in a 64-bit word.  However, Perl will warn if you output any of
1552these as being non-portable; and under strict UTF-8 input protocols,
1553they are forbidden.  In addition, it is now illegal to use a code point
1554larger than what a signed integer variable on your system can hold.  On
155532-bit ASCII systems, this means C<0x7FFF_FFFF> is the legal maximum
1556(much higher on 64-bit systems).
1557
1558=item *
1559
1560UTF-EBCDIC
1561
1562Like UTF-8, but EBCDIC-safe, in the way that UTF-8 is ASCII-safe.
1563This means that all the basic characters (which includes all
1564those that have ASCII equivalents (like C<"A">, C<"0">, C<"%">, I<etc.>)
1565are the same in both EBCDIC and UTF-EBCDIC.)
1566
1567UTF-EBCDIC is used on EBCDIC platforms.  It generally requires more
1568bytes to represent a given code point than UTF-8 does; the largest
1569Unicode code points take 5 bytes to represent (instead of 4 in UTF-8),
1570and, extended for 64-bit words, it uses 14 bytes instead of 13 bytes in
1571UTF-8.
1572
1573=item *
1574
1575UTF-16, UTF-16BE, UTF-16LE, Surrogates, and C<BOM>'s (Byte Order Marks)
1576
1577The followings items are mostly for reference and general Unicode
1578knowledge, Perl doesn't use these constructs internally.
1579
1580Like UTF-8, UTF-16 is a variable-width encoding, but where
1581UTF-8 uses 8-bit code units, UTF-16 uses 16-bit code units.
1582All code points occupy either 2 or 4 bytes in UTF-16: code points
1583C<U+0000..U+FFFF> are stored in a single 16-bit unit, and code
1584points C<U+10000..U+10FFFF> in two 16-bit units.  The latter case is
1585using I<surrogates>, the first 16-bit unit being the I<high
1586surrogate>, and the second being the I<low surrogate>.
1587
1588Surrogates are code points set aside to encode the C<U+10000..U+10FFFF>
1589range of Unicode code points in pairs of 16-bit units.  The I<high
1590surrogates> are the range C<U+D800..U+DBFF> and the I<low surrogates>
1591are the range C<U+DC00..U+DFFF>.  The surrogate encoding is
1592
1593    $hi = ($uni - 0x10000) / 0x400 + 0xD800;
1594    $lo = ($uni - 0x10000) % 0x400 + 0xDC00;
1595
1596and the decoding is
1597
1598    $uni = 0x10000 + ($hi - 0xD800) * 0x400 + ($lo - 0xDC00);
1599
1600Because of the 16-bitness, UTF-16 is byte-order dependent.  UTF-16
1601itself can be used for in-memory computations, but if storage or
1602transfer is required either UTF-16BE (big-endian) or UTF-16LE
1603(little-endian) encodings must be chosen.
1604
1605This introduces another problem: what if you just know that your data
1606is UTF-16, but you don't know which endianness?  Byte Order Marks, or
1607C<BOM>'s, are a solution to this.  A special character has been reserved
1608in Unicode to function as a byte order marker: the character with the
1609code point C<U+FEFF> is the C<BOM>.
1610
1611The trick is that if you read a C<BOM>, you will know the byte order,
1612since if it was written on a big-endian platform, you will read the
1613bytes C<0xFE 0xFF>, but if it was written on a little-endian platform,
1614you will read the bytes C<0xFF 0xFE>.  (And if the originating platform
1615was writing in ASCII platform UTF-8, you will read the bytes
1616C<0xEF 0xBB 0xBF>.)
1617
1618The way this trick works is that the character with the code point
1619C<U+FFFE> is not supposed to be in input streams, so the
1620sequence of bytes C<0xFF 0xFE> is unambiguously "C<BOM>, represented in
1621little-endian format" and cannot be C<U+FFFE>, represented in big-endian
1622format".
1623
1624Surrogates have no meaning in Unicode outside their use in pairs to
1625represent other code points.  However, Perl allows them to be
1626represented individually internally, for example by saying
1627C<chr(0xD801)>, so that all code points, not just those valid for open
1628interchange, are
1629representable.  Unicode does define semantics for them, such as their
1630C<L</General_Category>> is C<"Cs">.  But because their use is somewhat dangerous,
1631Perl will warn (using the warning category C<"surrogate">, which is a
1632sub-category of C<"utf8">) if an attempt is made
1633to do things like take the lower case of one, or match
1634case-insensitively, or to output them.  (But don't try this on Perls
1635before 5.14.)
1636
1637=item *
1638
1639UTF-32, UTF-32BE, UTF-32LE
1640
1641The UTF-32 family is pretty much like the UTF-16 family, except that
1642the units are 32-bit, and therefore the surrogate scheme is not
1643needed.  UTF-32 is a fixed-width encoding.  The C<BOM> signatures are
1644C<0x00 0x00 0xFE 0xFF> for BE and C<0xFF 0xFE 0x00 0x00> for LE.
1645
1646=item *
1647
1648UCS-2, UCS-4
1649
1650Legacy, fixed-width encodings defined by the ISO 10646 standard.  UCS-2 is a 16-bit
1651encoding.  Unlike UTF-16, UCS-2 is not extensible beyond C<U+FFFF>,
1652because it does not use surrogates.  UCS-4 is a 32-bit encoding,
1653functionally identical to UTF-32 (the difference being that
1654UCS-4 forbids neither surrogates nor code points larger than C<0x10_FFFF>).
1655
1656=item *
1657
1658UTF-7
1659
1660A seven-bit safe (non-eight-bit) encoding, which is useful if the
1661transport or storage is not eight-bit safe.  Defined by RFC 2152.
1662
1663=back
1664
1665=head2 Noncharacter code points
1666
166766 code points are set aside in Unicode as "noncharacter code points".
1668These all have the C<Unassigned> (C<Cn>) C<L</General_Category>>, and
1669no character will ever be assigned to any of them.  They are the 32 code
1670points between C<U+FDD0> and C<U+FDEF> inclusive, and the 34 code
1671points:
1672
1673 U+FFFE   U+FFFF
1674 U+1FFFE  U+1FFFF
1675 U+2FFFE  U+2FFFF
1676 ...
1677 U+EFFFE  U+EFFFF
1678 U+FFFFE  U+FFFFF
1679 U+10FFFE U+10FFFF
1680
1681Until Unicode 7.0, the noncharacters were "B<forbidden> for use in open
1682interchange of Unicode text data", so that code that processed those
1683streams could use these code points as sentinels that could be mixed in
1684with character data, and would always be distinguishable from that data.
1685(Emphasis above and in the next paragraph are added in this document.)
1686
1687Unicode 7.0 changed the wording so that they are "B<not recommended> for
1688use in open interchange of Unicode text data".  The 7.0 Standard goes on
1689to say:
1690
1691=over 4
1692
1693"If a noncharacter is received in open interchange, an application is
1694not required to interpret it in any way.  It is good practice, however,
1695to recognize it as a noncharacter and to take appropriate action, such
1696as replacing it with C<U+FFFD> replacement character, to indicate the
1697problem in the text.  It is not recommended to simply delete
1698noncharacter code points from such text, because of the potential
1699security issues caused by deleting uninterpreted characters.  (See
1700conformance clause C7 in Section 3.2, Conformance Requirements, and
1701L<Unicode Technical Report #36, "Unicode Security
1702Considerations"|https://www.unicode.org/reports/tr36/#Substituting_for_Ill_Formed_Subsequences>)."
1703
1704=back
1705
1706This change was made because it was found that various commercial tools
1707like editors, or for things like source code control, had been written
1708so that they would not handle program files that used these code points,
1709effectively precluding their use almost entirely!  And that was never
1710the intent.  They've always been meant to be usable within an
1711application, or cooperating set of applications, at will.
1712
1713If you're writing code, such as an editor, that is supposed to be able
1714to handle any Unicode text data, then you shouldn't be using these code
1715points yourself, and instead allow them in the input.  If you need
1716sentinels, they should instead be something that isn't legal Unicode.
1717For UTF-8 data, you can use the bytes 0xC1 and 0xC2 as sentinels, as
1718they never appear in well-formed UTF-8.  (There are equivalents for
1719UTF-EBCDIC).  You can also store your Unicode code points in integer
1720variables and use negative values as sentinels.
1721
1722If you're not writing such a tool, then whether you accept noncharacters
1723as input is up to you (though the Standard recommends that you not).  If
1724you do strict input stream checking with Perl, these code points
1725continue to be forbidden.  This is to maintain backward compatibility
1726(otherwise potential security holes could open up, as an unsuspecting
1727application that was written assuming the noncharacters would be
1728filtered out before getting to it, could now, without warning, start
1729getting them).  To do strict checking, you can use the layer
1730C<:encoding('UTF-8')>.
1731
1732Perl continues to warn (using the warning category C<"nonchar">, which
1733is a sub-category of C<"utf8">) if an attempt is made to output
1734noncharacters.
1735
1736=head2 Beyond Unicode code points
1737
1738The maximum Unicode code point is C<U+10FFFF>, and Unicode only defines
1739operations on code points up through that.  But Perl works on code
1740points up to the maximum permissible signed number available on the
1741platform.  However, Perl will not accept these from input streams unless
1742lax rules are being used, and will warn (using the warning category
1743C<"non_unicode">, which is a sub-category of C<"utf8">) if any are output.
1744
1745Since Unicode rules are not defined on these code points, if a
1746Unicode-defined operation is done on them, Perl uses what we believe are
1747sensible rules, while generally warning, using the C<"non_unicode">
1748category.  For example, C<uc("\x{11_0000}")> will generate such a
1749warning, returning the input parameter as its result, since Perl defines
1750the uppercase of every non-Unicode code point to be the code point
1751itself.  (All the case changing operations, not just uppercasing, work
1752this way.)
1753
1754The situation with matching Unicode properties in regular expressions,
1755the C<\p{}> and C<\P{}> constructs, against these code points is not as
1756clear cut, and how these are handled has changed as we've gained
1757experience.
1758
1759One possibility is to treat any match against these code points as
1760undefined.  But since Perl doesn't have the concept of a match being
1761undefined, it converts this to failing or C<FALSE>.  This is almost, but
1762not quite, what Perl did from v5.14 (when use of these code points
1763became generally reliable) through v5.18.  The difference is that Perl
1764treated all C<\p{}> matches as failing, but all C<\P{}> matches as
1765succeeding.
1766
1767One problem with this is that it leads to unexpected, and confusing
1768results in some cases:
1769
1770 chr(0x110000) =~ \p{ASCII_Hex_Digit=True}      # Failed on <= v5.18
1771 chr(0x110000) =~ \p{ASCII_Hex_Digit=False}     # Failed! on <= v5.18
1772
1773That is, it treated both matches as undefined, and converted that to
1774false (raising a warning on each).  The first case is the expected
1775result, but the second is likely counterintuitive: "How could both be
1776false when they are complements?"  Another problem was that the
1777implementation optimized many Unicode property matches down to already
1778existing simpler, faster operations, which don't raise the warning.  We
1779chose to not forgo those optimizations, which help the vast majority of
1780matches, just to generate a warning for the unlikely event that an
1781above-Unicode code point is being matched against.
1782
1783As a result of these problems, starting in v5.20, what Perl does is
1784to treat non-Unicode code points as just typical unassigned Unicode
1785characters, and matches accordingly.  (Note: Unicode has atypical
1786unassigned code points.  For example, it has noncharacter code points,
1787and ones that, when they do get assigned, are destined to be written
1788Right-to-left, as Arabic and Hebrew are.  Perl assumes that no
1789non-Unicode code point has any atypical properties.)
1790
1791Perl, in most cases, will raise a warning when matching an above-Unicode
1792code point against a Unicode property when the result is C<TRUE> for
1793C<\p{}>, and C<FALSE> for C<\P{}>.  For example:
1794
1795 chr(0x110000) =~ \p{ASCII_Hex_Digit=True}      # Fails, no warning
1796 chr(0x110000) =~ \p{ASCII_Hex_Digit=False}     # Succeeds, with warning
1797
1798In both these examples, the character being matched is non-Unicode, so
1799Unicode doesn't define how it should match.  It clearly isn't an ASCII
1800hex digit, so the first example clearly should fail, and so it does,
1801with no warning.  But it is arguable that the second example should have
1802an undefined, hence C<FALSE>, result.  So a warning is raised for it.
1803
1804Thus the warning is raised for many fewer cases than in earlier Perls,
1805and only when what the result is could be arguable.  It turns out that
1806none of the optimizations made by Perl (or are ever likely to be made)
1807cause the warning to be skipped, so it solves both problems of Perl's
1808earlier approach.  The most commonly used property that is affected by
1809this change is C<\p{Unassigned}> which is a short form for
1810C<\p{General_Category=Unassigned}>.  Starting in v5.20, all non-Unicode
1811code points are considered C<Unassigned>.  In earlier releases the
1812matches failed because the result was considered undefined.
1813
1814The only place where the warning is not raised when it might ought to
1815have been is if optimizations cause the whole pattern match to not even
1816be attempted.  For example, Perl may figure out that for a string to
1817match a certain regular expression pattern, the string has to contain
1818the substring C<"foobar">.  Before attempting the match, Perl may look
1819for that substring, and if not found, immediately fail the match without
1820actually trying it; so no warning gets generated even if the string
1821contains an above-Unicode code point.
1822
1823This behavior is more "Do what I mean" than in earlier Perls for most
1824applications.  But it catches fewer issues for code that needs to be
1825strictly Unicode compliant.  Therefore there is an additional mode of
1826operation available to accommodate such code.  This mode is enabled if a
1827regular expression pattern is compiled within the lexical scope where
1828the C<"non_unicode"> warning class has been made fatal, say by:
1829
1830 use warnings FATAL => "non_unicode"
1831
1832(see L<warnings>).  In this mode of operation, Perl will raise the
1833warning for all matches against a non-Unicode code point (not just the
1834arguable ones), and it skips the optimizations that might cause the
1835warning to not be output.  (It currently still won't warn if the match
1836isn't even attempted, like in the C<"foobar"> example above.)
1837
1838In summary, Perl now normally treats non-Unicode code points as typical
1839Unicode unassigned code points for regular expression matches, raising a
1840warning only when it is arguable what the result should be.  However, if
1841this warning has been made fatal, it isn't skipped.
1842
1843There is one exception to all this.  C<\p{All}> looks like a Unicode
1844property, but it is a Perl extension that is defined to be true for all
1845possible code points, Unicode or not, so no warning is ever generated
1846when matching this against a non-Unicode code point.  (Prior to v5.20,
1847it was an exact synonym for C<\p{Any}>, matching code points C<0>
1848through C<0x10FFFF>.)
1849
1850=head2 Security Implications of Unicode
1851
1852First, read
1853L<Unicode Security Considerations|https://www.unicode.org/reports/tr36>.
1854
1855Also, note the following:
1856
1857=over 4
1858
1859=item *
1860
1861Malformed UTF-8
1862
1863UTF-8 is very structured, so many combinations of bytes are invalid.  In
1864the past, Perl tried to soldier on and make some sense of invalid
1865combinations, but this can lead to security holes, so now, if the Perl
1866core needs to process an invalid combination, it will either raise a
1867fatal error, or will replace those bytes by the sequence that forms the
1868Unicode REPLACEMENT CHARACTER, for which purpose Unicode created it.
1869
1870Every code point can be represented by more than one possible
1871syntactically valid UTF-8 sequence.  Early on, both Unicode and Perl
1872considered any of these to be valid, but now, all sequences longer
1873than the shortest possible one are considered to be malformed.
1874
1875Unicode considers many code points to be illegal, or to be avoided.
1876Perl generally accepts them, once they have passed through any input
1877filters that may try to exclude them.  These have been discussed above
1878(see "Surrogates" under UTF-16 in L</Unicode Encodings>,
1879L</Noncharacter code points>, and L</Beyond Unicode code points>).
1880
1881=item *
1882
1883Regular expression pattern matching may surprise you if you're not
1884accustomed to Unicode.  Starting in Perl 5.14, several pattern
1885modifiers are available to control this, called the character set
1886modifiers.  Details are given in L<perlre/Character set modifiers>.
1887
1888=back
1889
1890As discussed elsewhere, Perl has one foot (two hooves?) planted in
1891each of two worlds: the old world of ASCII and single-byte locales, and
1892the new world of Unicode, upgrading when necessary.
1893If your legacy code does not explicitly use Unicode, no automatic
1894switch-over to Unicode should happen.
1895
1896=head2 Unicode in Perl on EBCDIC
1897
1898Unicode is supported on EBCDIC platforms.  See L<perlebcdic>.
1899
1900Unless ASCII vs. EBCDIC issues are specifically being discussed,
1901references to UTF-8 encoding in this document and elsewhere should be
1902read as meaning UTF-EBCDIC on EBCDIC platforms.
1903See L<perlebcdic/Unicode and UTF>.
1904
1905Because UTF-EBCDIC is so similar to UTF-8, the differences are mostly
1906hidden from you; S<C<use utf8>> (and NOT something like
1907S<C<use utfebcdic>>) declares the script is in the platform's
1908"native" 8-bit encoding of Unicode.  (Similarly for the C<":utf8">
1909layer.)
1910
1911=head2 Locales
1912
1913See L<perllocale/Unicode and UTF-8>
1914
1915=head2 When Unicode Does Not Happen
1916
1917There are still many places where Unicode (in some encoding or
1918another) could be given as arguments or received as results, or both in
1919Perl, but it is not, in spite of Perl having extensive ways to input and
1920output in Unicode, and a few other "entry points" like the C<@ARGV>
1921array (which can sometimes be interpreted as UTF-8).
1922
1923The following are such interfaces.  Also, see L</The "Unicode Bug">.
1924For all of these interfaces Perl
1925currently (as of v5.16.0) simply assumes byte strings both as arguments
1926and results, or UTF-8 strings if the (deprecated) C<encoding> pragma has been used.
1927
1928One reason that Perl does not attempt to resolve the role of Unicode in
1929these situations is that the answers are highly dependent on the operating
1930system and the file system(s).  For example, whether filenames can be
1931in Unicode and in exactly what kind of encoding, is not exactly a
1932portable concept.  Similarly for C<qx> and C<system>: how well will the
1933"command-line interface" (and which of them?) handle Unicode?
1934
1935=over 4
1936
1937=item *
1938
1939C<chdir>, C<chmod>, C<chown>, C<chroot>, C<exec>, C<link>, C<lstat>, C<mkdir>,
1940C<rename>, C<rmdir>, C<stat>, C<symlink>, C<truncate>, C<unlink>, C<utime>, C<-X>
1941
1942=item *
1943
1944C<%ENV>
1945
1946=item *
1947
1948C<glob> (aka the C<E<lt>*E<gt>>)
1949
1950=item *
1951
1952C<open>, C<opendir>, C<sysopen>
1953
1954=item *
1955
1956C<qx> (aka the backtick operator), C<system>
1957
1958=item *
1959
1960C<readdir>, C<readlink>
1961
1962=back
1963
1964=head2 The "Unicode Bug"
1965
1966The term, "Unicode bug" has been applied to an inconsistency with the
1967code points in the C<Latin-1 Supplement> block, that is, between
1968128 and 255.  Without a locale specified, unlike all other characters or
1969code points, these characters can have very different semantics
1970depending on the rules in effect.  (Characters whose code points are
1971above 255 force Unicode rules; whereas the rules for ASCII characters
1972are the same under both ASCII and Unicode rules.)
1973
1974Under Unicode rules, these upper-Latin1 characters are interpreted as
1975Unicode code points, which means they have the same semantics as Latin-1
1976(ISO-8859-1) and C1 controls.
1977
1978As explained in L</ASCII Rules versus Unicode Rules>, under ASCII rules,
1979they are considered to be unassigned characters.
1980
1981This can lead to unexpected results.  For example, a string's
1982semantics can suddenly change if a code point above 255 is appended to
1983it, which changes the rules from ASCII to Unicode.  As an
1984example, consider the following program and its output:
1985
1986 $ perl -le'
1987     no feature "unicode_strings";
1988     $s1 = "\xC2";
1989     $s2 = "\x{2660}";
1990     for ($s1, $s2, $s1.$s2) {
1991         print /\w/ || 0;
1992     }
1993 '
1994 0
1995 0
1996 1
1997
1998If there's no C<\w> in C<s1> nor in C<s2>, why does their concatenation
1999have one?
2000
2001This anomaly stems from Perl's attempt to not disturb older programs that
2002didn't use Unicode, along with Perl's desire to add Unicode support
2003seamlessly.  But the result turned out to not be seamless.  (By the way,
2004you can choose to be warned when things like this happen.  See
2005C<L<encoding::warnings>>.)
2006
2007L<S<C<use feature 'unicode_strings'>>|feature/The 'unicode_strings' feature>
2008was added, starting in Perl v5.12, to address this problem.  It affects
2009these things:
2010
2011=over 4
2012
2013=item *
2014
2015Changing the case of a scalar, that is, using C<uc()>, C<ucfirst()>, C<lc()>,
2016and C<lcfirst()>, or C<\L>, C<\U>, C<\u> and C<\l> in double-quotish
2017contexts, such as regular expression substitutions.
2018
2019Under C<unicode_strings> starting in Perl 5.12.0, Unicode rules are
2020generally used.  See L<perlfunc/lc> for details on how this works
2021in combination with various other pragmas.
2022
2023=item *
2024
2025Using caseless (C</i>) regular expression matching.
2026
2027Starting in Perl 5.14.0, regular expressions compiled within
2028the scope of C<unicode_strings> use Unicode rules
2029even when executed or compiled into larger
2030regular expressions outside the scope.
2031
2032=item *
2033
2034Matching any of several properties in regular expressions.
2035
2036These properties are C<\b> (without braces), C<\B> (without braces),
2037C<\s>, C<\S>, C<\w>, C<\W>, and all the Posix character classes
2038I<except> C<[[:ascii:]]>.
2039
2040Starting in Perl 5.14.0, regular expressions compiled within
2041the scope of C<unicode_strings> use Unicode rules
2042even when executed or compiled into larger
2043regular expressions outside the scope.
2044
2045=item *
2046
2047In C<quotemeta> or its inline equivalent C<\Q>.
2048
2049Starting in Perl 5.16.0, consistent quoting rules are used within the
2050scope of C<unicode_strings>, as described in L<perlfunc/quotemeta>.
2051Prior to that, or outside its scope, no code points above 127 are quoted
2052in UTF-8 encoded strings, but in byte encoded strings, code points
2053between 128-255 are always quoted.
2054
2055=item *
2056
2057In the C<..> or L<range|perlop/Range Operators> operator.
2058
2059Starting in Perl 5.26.0, the range operator on strings treats their lengths
2060consistently within the scope of C<unicode_strings>. Prior to that, or
2061outside its scope, it could produce strings whose length in characters
2062exceeded that of the right-hand side, where the right-hand side took up more
2063bytes than the correct range endpoint.
2064
2065=item *
2066
2067In L<< C<split>'s special-case whitespace splitting|perlfunc/split >>.
2068
2069Starting in Perl 5.28.0, the C<split> function with a pattern specified as
2070a string containing a single space handles whitespace characters consistently
2071within the scope of C<unicode_strings>. Prior to that, or outside its scope,
2072characters that are whitespace according to Unicode rules but not according to
2073ASCII rules were treated as field contents rather than field separators when
2074they appear in byte-encoded strings.
2075
2076=back
2077
2078You can see from the above that the effect of C<unicode_strings>
2079increased over several Perl releases.  (And Perl's support for Unicode
2080continues to improve; it's best to use the latest available release in
2081order to get the most complete and accurate results possible.)  Note that
2082C<unicode_strings> is automatically chosen if you S<C<use 5.012>> or
2083higher.
2084
2085For Perls earlier than those described above, or when a string is passed
2086to a function outside the scope of C<unicode_strings>, see the next section.
2087
2088=head2 Forcing Unicode in Perl (Or Unforcing Unicode in Perl)
2089
2090Sometimes (see L</"When Unicode Does Not Happen"> or L</The "Unicode Bug">)
2091there are situations where you simply need to force a byte
2092string into UTF-8, or vice versa.  The standard module L<Encode> can be
2093used for this, or the low-level calls
2094L<C<utf8::upgrade($bytestring)>|utf8/Utility functions> and
2095L<C<utf8::downgrade($utf8string[, FAIL_OK])>|utf8/Utility functions>.
2096
2097Note that C<utf8::downgrade()> can fail if the string contains characters
2098that don't fit into a byte.
2099
2100Calling either function on a string that already is in the desired state is a
2101no-op.
2102
2103L</ASCII Rules versus Unicode Rules> gives all the ways that a string is
2104made to use Unicode rules.
2105
2106=head2 Using Unicode in XS
2107
2108See L<perlguts/"Unicode Support"> for an introduction to Unicode at
2109the XS level, and L<perlapi/Unicode Support> for the API details.
2110
2111=head2 Hacking Perl to work on earlier Unicode versions (for very serious hackers only)
2112
2113Perl by default comes with the latest supported Unicode version built-in, but
2114the goal is to allow you to change to use any earlier one.  In Perls
2115v5.20 and v5.22, however, the earliest usable version is Unicode 5.1.
2116Perl v5.18 and v5.24 are able to handle all earlier versions.
2117
2118Download the files in the desired version of Unicode from the Unicode web
2119site L<https://www.unicode.org>).  These should replace the existing files in
2120F<lib/unicore> in the Perl source tree.  Follow the instructions in
2121F<README.perl> in that directory to change some of their names, and then build
2122perl (see L<INSTALL>).
2123
2124=head2 Porting code from perl-5.6.X
2125
2126Perls starting in 5.8 have a different Unicode model from 5.6. In 5.6 the
2127programmer was required to use the C<utf8> pragma to declare that a
2128given scope expected to deal with Unicode data and had to make sure that
2129only Unicode data were reaching that scope. If you have code that is
2130working with 5.6, you will need some of the following adjustments to
2131your code. The examples are written such that the code will continue to
2132work under 5.6, so you should be safe to try them out.
2133
2134=over 3
2135
2136=item *
2137
2138A filehandle that should read or write UTF-8
2139
2140  if ($] > 5.008) {
2141    binmode $fh, ":encoding(UTF-8)";
2142  }
2143
2144=item *
2145
2146A scalar that is going to be passed to some extension
2147
2148Be it C<Compress::Zlib>, C<Apache::Request> or any extension that has no
2149mention of Unicode in the manpage, you need to make sure that the
2150UTF8 flag is stripped off. Note that at the time of this writing
2151(January 2012) the mentioned modules are not UTF-8-aware. Please
2152check the documentation to verify if this is still true.
2153
2154  if ($] > 5.008) {
2155    require Encode;
2156    $val = Encode::encode("UTF-8", $val); # make octets
2157  }
2158
2159=item *
2160
2161A scalar we got back from an extension
2162
2163If you believe the scalar comes back as UTF-8, you will most likely
2164want the UTF8 flag restored:
2165
2166  if ($] > 5.008) {
2167    require Encode;
2168    $val = Encode::decode("UTF-8", $val);
2169  }
2170
2171=item *
2172
2173Same thing, if you are really sure it is UTF-8
2174
2175  if ($] > 5.008) {
2176    require Encode;
2177    Encode::_utf8_on($val);
2178  }
2179
2180=item *
2181
2182A wrapper for L<DBI> C<fetchrow_array> and C<fetchrow_hashref>
2183
2184When the database contains only UTF-8, a wrapper function or method is
2185a convenient way to replace all your C<fetchrow_array> and
2186C<fetchrow_hashref> calls. A wrapper function will also make it easier to
2187adapt to future enhancements in your database driver. Note that at the
2188time of this writing (January 2012), the DBI has no standardized way
2189to deal with UTF-8 data. Please check the L<DBI documentation|DBI> to verify if
2190that is still true.
2191
2192  sub fetchrow {
2193    # $what is one of fetchrow_{array,hashref}
2194    my($self, $sth, $what) = @_;
2195    if ($] < 5.008) {
2196      return $sth->$what;
2197    } else {
2198      require Encode;
2199      if (wantarray) {
2200        my @arr = $sth->$what;
2201        for (@arr) {
2202          defined && /[^\000-\177]/ && Encode::_utf8_on($_);
2203        }
2204        return @arr;
2205      } else {
2206        my $ret = $sth->$what;
2207        if (ref $ret) {
2208          for my $k (keys %$ret) {
2209            defined
2210            && /[^\000-\177]/
2211            && Encode::_utf8_on($_) for $ret->{$k};
2212          }
2213          return $ret;
2214        } else {
2215          defined && /[^\000-\177]/ && Encode::_utf8_on($_) for $ret;
2216          return $ret;
2217        }
2218      }
2219    }
2220  }
2221
2222
2223=item *
2224
2225A large scalar that you know can only contain ASCII
2226
2227Scalars that contain only ASCII and are marked as UTF-8 are sometimes
2228a drag to your program. If you recognize such a situation, just remove
2229the UTF8 flag:
2230
2231  utf8::downgrade($val) if $] > 5.008;
2232
2233=back
2234
2235=head1 BUGS
2236
2237See also L</The "Unicode Bug"> above.
2238
2239=head2 Interaction with Extensions
2240
2241When Perl exchanges data with an extension, the extension should be
2242able to understand the UTF8 flag and act accordingly. If the
2243extension doesn't recognize that flag, it's likely that the extension
2244will return incorrectly-flagged data.
2245
2246So if you're working with Unicode data, consult the documentation of
2247every module you're using if there are any issues with Unicode data
2248exchange. If the documentation does not talk about Unicode at all,
2249suspect the worst and probably look at the source to learn how the
2250module is implemented. Modules written completely in Perl shouldn't
2251cause problems. Modules that directly or indirectly access code written
2252in other programming languages are at risk.
2253
2254For affected functions, the simple strategy to avoid data corruption is
2255to always make the encoding of the exchanged data explicit. Choose an
2256encoding that you know the extension can handle. Convert arguments passed
2257to the extensions to that encoding and convert results back from that
2258encoding. Write wrapper functions that do the conversions for you, so
2259you can later change the functions when the extension catches up.
2260
2261To provide an example, let's say the popular C<Foo::Bar::escape_html>
2262function doesn't deal with Unicode data yet. The wrapper function
2263would convert the argument to raw UTF-8 and convert the result back to
2264Perl's internal representation like so:
2265
2266    sub my_escape_html ($) {
2267        my($what) = shift;
2268        return unless defined $what;
2269        Encode::decode("UTF-8", Foo::Bar::escape_html(
2270                                     Encode::encode("UTF-8", $what)));
2271    }
2272
2273Sometimes, when the extension does not convert data but just stores
2274and retrieves it, you will be able to use the otherwise
2275dangerous L<C<Encode::_utf8_on()>|Encode/_utf8_on> function. Let's say
2276the popular C<Foo::Bar> extension, written in C, provides a C<param>
2277method that lets you store and retrieve data according to these prototypes:
2278
2279    $self->param($name, $value);            # set a scalar
2280    $value = $self->param($name);           # retrieve a scalar
2281
2282If it does not yet provide support for any encoding, one could write a
2283derived class with such a C<param> method:
2284
2285    sub param {
2286      my($self,$name,$value) = @_;
2287      utf8::upgrade($name);     # make sure it is UTF-8 encoded
2288      if (defined $value) {
2289        utf8::upgrade($value);  # make sure it is UTF-8 encoded
2290        return $self->SUPER::param($name,$value);
2291      } else {
2292        my $ret = $self->SUPER::param($name);
2293        Encode::_utf8_on($ret); # we know, it is UTF-8 encoded
2294        return $ret;
2295      }
2296    }
2297
2298Some extensions provide filters on data entry/exit points, such as
2299C<DB_File::filter_store_key> and family. Look out for such filters in
2300the documentation of your extensions; they can make the transition to
2301Unicode data much easier.
2302
2303=head2 Speed
2304
2305Some functions are slower when working on UTF-8 encoded strings than
2306on byte encoded strings.  All functions that need to hop over
2307characters such as C<length()>, C<substr()> or C<index()>, or matching
2308regular expressions can work B<much> faster when the underlying data are
2309byte-encoded.
2310
2311In Perl 5.8.0 the slowness was often quite spectacular; in Perl 5.8.1
2312a caching scheme was introduced which improved the situation.  In general,
2313operations with UTF-8 encoded strings are still slower. As an example,
2314the Unicode properties (character classes) like C<\p{Nd}> are known to
2315be quite a bit slower (5-20 times) than their simpler counterparts
2316like C<[0-9]> (then again, there are hundreds of Unicode characters matching
2317C<Nd> compared with the 10 ASCII characters matching C<[0-9]>).
2318
2319=head1 SEE ALSO
2320
2321L<perlunitut>, L<perluniintro>, L<perluniprops>, L<Encode>, L<open>, L<utf8>, L<bytes>,
2322L<perlretut>, L<perlvar/"${^UNICODE}">,
2323L<https://www.unicode.org/reports/tr44>).
2324
2325=cut
2326