1 # $Id: encoding.pm,v 1.48 2003/12/29 02:47:16 dankogai Exp dankogai $
3 our $VERSION = do { my @r = (q$Revision: 1.48 $ =~ /\d+/g); sprintf "%d."."%02d" x $#r, @r };
10 if (ord("A") == 193) {
12 Carp::croak("encoding pragma does not support EBCDIC platforms");
17 eval { require PerlIO::encoding };
19 $HAS_PERLIO = (PerlIO::encoding->VERSION >= 0.02);
24 $] > 5.008 and return 0; # 5.8.1 or higher then no
25 my %utfs = map {$_=>1}
26 qw(utf8 UCS-2BE UCS-2LE UTF-16 UTF-16BE UTF-16LE
27 UTF-32 UTF-32BE UTF-32LE);
28 $utfs{$name} or return 0; # UTFs or no
29 require Config; Config->import(); our %Config;
30 return $Config{perl_patchlevel} ? 0 : 1 # maintperl then no
37 $name ||= $ENV{PERL_ENCODING};
38 my $enc = find_encoding($name);
39 unless (defined $enc) {
41 Carp::croak("Unknown encoding '$name'");
43 $name = $enc->name; # canonize
44 unless ($arg{Filter}) {
45 DEBUG and warn "_exception($name) = ", _exception($name);
46 _exception($name) or ${^ENCODING} = $enc;
47 $HAS_PERLIO or return 1;
49 defined(${^ENCODING}) and undef ${^ENCODING};
50 # implicitly 'use utf8'
51 require utf8; # to fetch $utf8::hint_bits;
52 $^H |= $utf8::hint_bits;
54 require Filter::Util::Call ;
55 Filter::Util::Call->import ;
57 my $status = filter_read();
59 $_ = $enc->decode($_, 1);
65 } DEBUG and warn "Filter installed";
66 defined ${^UNICODE} and ${^UNICODE} != 0 and return 1;
67 for my $h (qw(STDIN STDOUT)){
69 unless (defined find_encoding($arg{$h})) {
71 Carp::croak("Unknown encoding for $h, '$arg{$h}'");
73 eval { binmode($h, ":raw :encoding($arg{$h})") };
75 unless (exists $arg{$h}){
77 no warnings 'uninitialized';
78 binmode($h, ":raw :encoding($name)");
87 return 1; # I doubt if we need it, though
94 binmode(STDIN, ":raw");
95 binmode(STDOUT, ":raw");
100 if ($INC{"Filter/Util/Call.pm"}){
101 eval { filter_del() };
112 encoding - allows you to write your script in non-ascii or non-utf8
116 use encoding "greek"; # Perl like Greek to you?
117 use encoding "euc-jp"; # Jperl!
119 # or you can even do this if your shell supports your native encoding
121 perl -Mencoding=latin2 -e '...' # Feeling centrally European?
122 perl -Mencoding=euc-kr -e '...' # Or Korean?
126 # A simple euc-cn => utf-8 converter
127 use encoding "euc-cn", STDOUT => "utf8"; while(<>){print};
129 # "no encoding;" supported (but not scoped!)
132 # an alternate way, Filter
133 use encoding "euc-jp", Filter=>1;
134 # now you can use kanji identifiers -- in euc-jp!
138 Let's start with a bit of history: Perl 5.6.0 introduced Unicode
139 support. You could apply C<substr()> and regexes even to complex CJK
140 characters -- so long as the script was written in UTF-8. But back
141 then, text editors that supported UTF-8 were still rare and many users
142 instead chose to write scripts in legacy encodings, giving up a whole
143 new feature of Perl 5.6.
145 Rewind to the future: starting from perl 5.8.0 with the B<encoding>
146 pragma, you can write your script in any encoding you like (so long
147 as the C<Encode> module supports it) and still enjoy Unicode support.
148 This pragma achieves that by doing the following:
154 Internally converts all literals (C<q//,qq//,qr//,qw///, qx//>) from
155 the encoding specified to utf8. In Perl 5.8.1 and later, literals in
156 C<tr///> and C<DATA> pseudo-filehandle are also converted.
160 Changing PerlIO layers of C<STDIN> and C<STDOUT> to the encoding
165 =head2 Literal Conversions
167 You can write code in EUC-JP as follows:
169 my $Rakuda = "\xF1\xD1\xF1\xCC"; # Camel in Kanji
170 #<-char-><-char-> # 4 octets
171 s/\bCamel\b/$Rakuda/;
173 And with C<use encoding "euc-jp"> in effect, it is the same thing as
176 my $Rakuda = "\x{99F1}\x{99DD}"; # two Unicode Characters
177 s/\bCamel\b/$Rakuda/;
179 =head2 PerlIO layers for C<STD(IN|OUT)>
181 The B<encoding> pragma also modifies the filehandle layers of
182 STDIN and STDOUT to the specified encoding. Therefore,
184 use encoding "euc-jp";
185 my $message = "Camel is the symbol of perl.\n";
186 my $Rakuda = "\xF1\xD1\xF1\xCC"; # Camel in Kanji
187 $message =~ s/\bCamel\b/$Rakuda/;
190 Will print "\xF1\xD1\xF1\xCC is the symbol of perl.\n",
191 not "\x{99F1}\x{99DD} is the symbol of perl.\n".
193 You can override this by giving extra arguments; see below.
195 =head2 Implicit upgrading for byte strings
197 By default, if strings operating under byte semantics and strings
198 with Unicode character data are concatenated, the new string will
199 be created by decoding the byte strings as I<ISO 8859-1 (Latin-1)>.
201 The B<encoding> pragma changes this to use the specified encoding
202 instead. For example:
205 my $string = chr(20000); # a Unicode string
206 utf8::encode($string); # now it's a UTF-8 encoded byte string
207 # concatenate with another Unicode string
208 print length($string . chr(20000));
210 Will print C<2>, because C<$string> is upgraded as UTF-8. Without
211 C<use encoding 'utf8';>, it will print C<4> instead, since C<$string>
212 is three octets when interpreted as Latin-1.
214 =head1 FEATURES THAT REQUIRE 5.8.1
216 Some of the features offered by this pragma requires perl 5.8.1. Most
217 of these are done by Inaba Hiroto. Any other features and changes
222 =item "NON-EUC" doublebyte encodings
224 Because perl needs to parse script before applying this pragma, such
225 encodings as Shift_JIS and Big-5 that may contain '\' (BACKSLASH;
226 \x5c) in the second byte fails because the second byte may
227 accidentally escape the quoting character that follows. Perl 5.8.1
228 or later fixes this problem.
232 C<tr//> was overlooked by Perl 5 porters when they released perl 5.8.0
233 See the section below for details.
235 =item DATA pseudo-filehandle
237 Another feature that was overlooked was C<DATA>.
245 =item use encoding [I<ENCNAME>] ;
247 Sets the script encoding to I<ENCNAME>. And unless ${^UNICODE}
248 exists and non-zero, PerlIO layers of STDIN and STDOUT are set to
249 ":encoding(I<ENCNAME>)".
251 Note that STDERR WILL NOT be changed.
253 Also note that non-STD file handles remain unaffected. Use C<use
254 open> or C<binmode> to change layers of those.
256 If no encoding is specified, the environment variable L<PERL_ENCODING>
257 is consulted. If no encoding can be found, the error C<Unknown encoding
258 'I<ENCNAME>'> will be thrown.
260 =item use encoding I<ENCNAME> [ STDIN =E<gt> I<ENCNAME_IN> ...] ;
262 You can also individually set encodings of STDIN and STDOUT via the
263 C<< STDIN => I<ENCNAME> >> form. In this case, you cannot omit the
264 first I<ENCNAME>. C<< STDIN => undef >> turns the IO transcoding
267 When ${^UNICODE} exists and non-zero, these options will completely
268 ignored. ${^UNICODE} is a variable introduced in perl 5.8.1. See
269 L<perlrun> see L<perlvar/"${^UNICODE}"> and L<perlrun/"-C"> for
270 details (perl 5.8.1 and later).
272 =item use encoding I<ENCNAME> Filter=E<gt>1;
274 This turns the encoding pragma into a source filter. While the
275 default approach just decodes interpolated literals (in qq() and
276 qr()), this will apply a source filter to the entire source code. See
277 L</"The Filter Option"> below for details.
281 Unsets the script encoding. The layers of STDIN, STDOUT are
282 reset to ":raw" (the default unprocessed raw stream of bytes).
286 =head1 The Filter Option
288 The magic of C<use encoding> is not applied to the names of
289 identifiers. In order to make C<${"\x{4eba}"}++> ($human++, where human
290 is a single Han ideograph) work, you still need to write your script
291 in UTF-8 -- or use a source filter. That's what 'Filter=>1' does.
293 What does this mean? Your source code behaves as if it is written in
294 UTF-8 with 'use utf8' in effect. So even if your editor only supports
295 Shift_JIS, for example, you can still try examples in Chapter 15 of
296 C<Programming Perl, 3rd Ed.>. For instance, you can use UTF-8
299 This option is significantly slower and (as of this writing) non-ASCII
300 identifiers are not very stable WITHOUT this option and with the
301 source code written in UTF-8.
303 =head2 Filter-related changes at Encode version 1.87
309 The Filter option now sets STDIN and STDOUT like non-filter options.
310 And C<< STDIN=>I<ENCODING> >> and C<< STDOUT=>I<ENCODING> >> work like
315 C<use utf8> is implicitly declared so you no longer have to C<use
316 utf8> to C<${"\x{4eba}"}++>.
324 The pragma is a per script, not a per block lexical. Only the last
325 C<use encoding> or C<no encoding> matters, and it affects
326 B<the whole script>. However, the <no encoding> pragma is supported and
327 B<use encoding> can appear as many times as you want in a given script.
328 The multiple use of this pragma is discouraged.
330 By the same reason, the use this pragma inside modules is also
331 discouraged (though not as strongly discouranged as the case above.
334 If you still have to write a module with this pragma, be very careful
335 of the load order. See the codes below;
338 package Module_IN_BAR;
340 # stuff in "bar" encoding here
346 # surprise! use encoding "bar" is in effect.
348 The best way to avoid this oddity is to use this pragma RIGHT AFTER
349 other modules are loaded. i.e.
354 =head2 DO NOT MIX MULTIPLE ENCODINGS
356 Notice that only literals (string or regular expression) having only
357 legacy code points are affected: if you mix data like this
361 the data is assumed to be in (Latin 1 and) Unicode, not in your native
362 encoding. In other words, this will match in "greek":
368 "\xDF\x{100}" =~ /\x{3af}\x{100}/
370 since the C<\xDF> (ISO 8859-7 GREEK SMALL LETTER IOTA WITH TONOS) on
371 the left will B<not> be upgraded to C<\x{3af}> (Unicode GREEK SMALL
372 LETTER IOTA WITH TONOS) because of the C<\x{100}> on the left. You
373 should not be mixing your legacy data and Unicode in the same string.
375 This pragma also affects encoding of the 0x80..0xFF code point range:
376 normally characters in that range are left as eight-bit bytes (unless
377 they are combined with characters with code points 0x100 or larger,
378 in which case all characters need to become UTF-8 encoded), but if
379 the C<encoding> pragma is present, even the 0x80..0xFF range always
382 After all, the best thing about this pragma is that you don't have to
383 resort to \x{....} just to spell your name in a native encoding.
384 So feel free to put your strings in your encoding in quotes and
387 =head2 tr/// with ranges
389 The B<encoding> pragma works by decoding string literals in
390 C<q//,qq//,qr//,qw///, qx//> and so forth. In perl 5.8.0, this
391 does not apply to C<tr///>. Therefore,
393 use encoding 'euc-jp';
395 $kana =~ tr/\xA4\xA1-\xA4\xF3/\xA5\xA1-\xA5\xF3/;
396 # -------- -------- -------- --------
400 $kana =~ tr/\x{3041}-\x{3093}/\x{30a1}-\x{30f3}/;
404 =item Legend of characters above
406 utf8 euc-jp charnames::viacode()
407 -----------------------------------------
408 \x{3041} \xA4\xA1 HIRAGANA LETTER SMALL A
409 \x{3093} \xA4\xF3 HIRAGANA LETTER N
410 \x{30a1} \xA5\xA1 KATAKANA LETTER SMALL A
411 \x{30f3} \xA5\xF3 KATAKANA LETTER N
415 This counterintuitive behavior has been fixed in perl 5.8.1.
417 =head3 workaround to tr///;
419 In perl 5.8.0, you can work around as follows;
421 use encoding 'euc-jp';
423 eval qq{ \$kana =~ tr/\xA4\xA1-\xA4\xF3/\xA5\xA1-\xA5\xF3/ };
425 Note the C<tr//> expression is surrounded by C<qq{}>. The idea behind
426 is the same as classic idiom that makes C<tr///> 'interpolate'.
428 tr/$from/$to/; # wrong!
429 eval qq{ tr/$from/$to/ }; # workaround.
431 Nevertheless, in case of B<encoding> pragma even C<q//> is affected so
432 C<tr///> not being decoded was obviously against the will of Perl5
433 Porters so it has been fixed in Perl 5.8.1 or later.
435 =head1 EXAMPLE - Greekperl
437 use encoding "iso 8859-7";
439 # \xDF in ISO 8859-7 (Greek) is \x{3af} in Unicode.
444 printf "%#x\n", ord($a); # will print 0x3af, not 0xdf
448 # $c will be "\x{3af}\x{100}", not "\x{df}\x{100}".
450 # chr() is affected, and ...
452 print "mega\n" if ord(chr(0xdf)) == 0x3af;
454 # ... ord() is affected by the encoding pragma ...
456 print "tera\n" if ord(pack("C", 0xdf)) == 0x3af;
458 # ... as are eq and cmp ...
460 print "peta\n" if "\x{3af}" eq pack("C", 0xdf);
461 print "exa\n" if "\x{3af}" cmp pack("C", 0xdf) == 0;
463 # ... but pack/unpack C are not affected, in case you still
464 # want to go back to your native encoding
466 print "zetta\n" if unpack("C", (pack("C", 0xdf))) == 0xdf;
468 =head1 KNOWN PROBLEMS
472 =item literals in regex that are longer than 127 bytes
474 For native multibyte encodings (either fixed or variable length),
475 the current implementation of the regular expressions may introduce
476 recoding errors for regular expression literals longer than 127 bytes.
480 The encoding pragma is not supported on EBCDIC platforms.
481 (Porters who are willing and able to remove this limitation are
486 This pragma doesn't work well with format because PerlIO does not
487 get along very well with it. When format contains non-ascii
488 characters it prints funny or gets "wide character warnings".
489 To understand it, try the code below.
491 # Save this one in utf8
492 # replace *non-ascii* with a non-ascii string
498 $camel = "*non-ascii*";
499 binmode(STDOUT=>':encoding(utf8)'); # bang!
501 print $camel, "\n"; # fine
503 Without binmode this happens to work but without binmode, print()
504 fails instead of write().
506 At any rate, the very use of format is questionable when it comes to
507 unicode characters since you have to consider such things as character
508 width (i.e. double-width for ideographs) and directions (i.e. BIDI for
515 This pragma first appeared in Perl 5.8.0. For features that require
516 5.8.1 and better, see above.
520 L<perlunicode>, L<Encode>, L<open>, L<Filter::Util::Call>,
522 Ch. 15 of C<Programming Perl (3rd Edition)>
523 by Larry Wall, Tom Christiansen, Jon Orwant;
524 O'Reilly & Associates; ISBN 0-596-00027-8