Commit | Line | Data |
2c674647 |
1 | package Encode; |
51ef4e11 |
2 | use strict; |
3ef515df |
3 | our $VERSION = do { my @r = (q$Revision: 1.11 $ =~ /\d+/g); sprintf "%d."."%02d" x $#r, @r }; |
5129552c |
4 | our $DEBUG = 0; |
2c674647 |
5 | |
6 | require DynaLoader; |
7 | require Exporter; |
8 | |
51ef4e11 |
9 | our @ISA = qw(Exporter DynaLoader); |
2c674647 |
10 | |
4411f3b6 |
11 | # Public, encouraged API is exported by default |
51ef4e11 |
12 | our @EXPORT = qw ( |
4411f3b6 |
13 | encode |
14 | decode |
15 | encode_utf8 |
16 | decode_utf8 |
17 | find_encoding |
51ef4e11 |
18 | encodings |
4411f3b6 |
19 | ); |
20 | |
51ef4e11 |
21 | our @EXPORT_OK = |
2c674647 |
22 | qw( |
51ef4e11 |
23 | define_encoding |
2c674647 |
24 | from_to |
25 | is_utf8 |
4411f3b6 |
26 | is_8bit |
27 | is_16bit |
a12c0f56 |
28 | utf8_upgrade |
29 | utf8_downgrade |
4411f3b6 |
30 | _utf8_on |
31 | _utf8_off |
2c674647 |
32 | ); |
33 | |
34 | bootstrap Encode (); |
35 | |
4411f3b6 |
36 | # Documentation moved after __END__ for speed - NI-S |
2c674647 |
37 | |
bf230f3d |
38 | use Carp; |
39 | |
a63c962f |
40 | our $ON_EBCDIC = (ord("A") == 193); |
5d030b67 |
41 | use Encode::Alias; |
42 | |
5129552c |
43 | # Make a %Encoding package variable to allow a certain amount of cheating |
44 | our %Encoding; |
5345d506 |
45 | |
5129552c |
46 | our %ExtModule = |
2b217bf7 |
47 | ( |
5129552c |
48 | viscii => 'Encode/Byte.pm', |
49 | 'koi8-r' => 'Encode/Byte.pm', |
50 | cp1047 => 'Encode/EBCDIC.pm', |
51 | cp37 => 'Encode/EBCDIC.pm', |
52 | 'posix-bc' => 'Encode/EBCDIC.pm', |
53 | symbol => 'Encode/Symbol.pm', |
54 | dingbats => 'Encode/Symbol.pm', |
2b217bf7 |
55 | ); |
d1ed7747 |
56 | |
5129552c |
57 | for my $k (2..11,13..16){ |
58 | $ExtModule{"iso-8859-$k"} = 'Encode/Byte.pm'; |
59 | } |
60 | |
61 | for my $k (1250..1258){ |
62 | $ExtModule{"cp$k"} = 'Encode/Byte.pm'; |
63 | } |
64 | |
a63c962f |
65 | unless ($ON_EBCDIC) { # CJK added to autoload unless EBCDIC env |
66 | %ExtModule =( |
67 | %ExtModule, |
68 | 'euc-cn' => 'Encode/CN.pm', |
69 | gb2312 => 'Encode/CN.pm', |
70 | gb12345 => 'Encode/CN.pm', |
71 | gbk => 'Encode/CN.pm', |
72 | cp936 => 'Encode/CN.pm', |
73 | 'iso-ir-165' => 'Encode/CN.pm', |
74 | 'euc-jp' => 'Encode/JP.pm', |
75 | 'iso-2022-jp' => 'Encode/JP.pm', |
76 | 'iso-2022-jp-1' => 'Encode/JP.pm', |
77 | '7bit-jis' => 'Encode/JP.pm', |
78 | shiftjis => 'Encode/JP.pm', |
79 | macjapan => 'Encode/JP.pm', |
80 | cp932 => 'Encode/JP.pm', |
81 | 'euc-kr' => 'Encode/KR.pm', |
82 | ksc5601 => 'Encode/KR.pm', |
83 | cp949 => 'Encode/KR.pm', |
84 | big5 => 'Encode/TW.pm', |
85 | 'big5-hkscs' => 'Encode/TW.pm', |
86 | cp950 => 'Encode/TW.pm', |
87 | gb18030 => 'Encode/HanExtra.pm', |
88 | big5plus => 'Encode/HanExtra.pm', |
89 | 'euc-tw' => 'Encode/HanExtra.pm', |
90 | ); |
91 | } |
92 | |
3ef515df |
93 | for my $k (qw{ CentralEurRoman Croatian Cyrillic Greek |
94 | Iceland Roman Rumanian Sami |
95 | Thai Turkish Ukrainian |
96 | }) |
5129552c |
97 | { |
98 | $ExtModule{"mac$k"} = 'Encode/Byte.pm'; |
99 | } |
100 | |
656753f8 |
101 | sub encodings |
102 | { |
5129552c |
103 | my $class = shift; |
071db25d |
104 | my @modules = (@_ and $_[0] eq ":all") ? values %ExtModule : @_; |
5129552c |
105 | for my $m (@modules) |
106 | { |
107 | $DEBUG and warn "about to require $m;"; |
108 | eval { require $m; }; |
109 | } |
110 | return |
111 | map({$_->[0]} |
112 | sort({$a->[1] cmp $b->[1]} |
113 | map({[$_, lc $_]} |
114 | grep({ $_ ne 'Internal' } keys %Encoding)))); |
51ef4e11 |
115 | } |
116 | |
51ef4e11 |
117 | sub define_encoding |
118 | { |
18586f54 |
119 | my $obj = shift; |
120 | my $name = shift; |
5129552c |
121 | $Encoding{$name} = $obj; |
18586f54 |
122 | my $lc = lc($name); |
123 | define_alias($lc => $obj) unless $lc eq $name; |
124 | while (@_) |
125 | { |
126 | my $alias = shift; |
127 | define_alias($alias,$obj); |
128 | } |
129 | return $obj; |
656753f8 |
130 | } |
131 | |
656753f8 |
132 | sub getEncoding |
133 | { |
dd9703c9 |
134 | my ($class,$name,$skip_external) = @_; |
18586f54 |
135 | my $enc; |
136 | if (ref($name) && $name->can('new_sequence')) |
137 | { |
138 | return $name; |
139 | } |
140 | my $lc = lc $name; |
5129552c |
141 | if (exists $Encoding{$name}) |
18586f54 |
142 | { |
5129552c |
143 | return $Encoding{$name}; |
18586f54 |
144 | } |
5129552c |
145 | if (exists $Encoding{$lc}) |
18586f54 |
146 | { |
5129552c |
147 | return $Encoding{$lc}; |
18586f54 |
148 | } |
c50d192e |
149 | |
5129552c |
150 | my $oc = $class->find_alias($name); |
c50d192e |
151 | return $oc if defined $oc; |
152 | |
5129552c |
153 | $oc = $class->find_alias($lc) if $lc ne $name; |
c50d192e |
154 | return $oc if defined $oc; |
155 | |
5129552c |
156 | if (!$skip_external and exists $ExtModule{$lc}) |
d1ed7747 |
157 | { |
5129552c |
158 | eval{ require $ExtModule{$lc}; }; |
159 | return $Encoding{$name} if exists $Encoding{$name}; |
d1ed7747 |
160 | } |
18586f54 |
161 | |
18586f54 |
162 | return; |
656753f8 |
163 | } |
164 | |
4411f3b6 |
165 | sub find_encoding |
166 | { |
dd9703c9 |
167 | my ($name,$skip_external) = @_; |
168 | return __PACKAGE__->getEncoding($name,$skip_external); |
4411f3b6 |
169 | } |
170 | |
171 | sub encode |
172 | { |
18586f54 |
173 | my ($name,$string,$check) = @_; |
174 | my $enc = find_encoding($name); |
175 | croak("Unknown encoding '$name'") unless defined $enc; |
176 | my $octets = $enc->encode($string,$check); |
177 | return undef if ($check && length($string)); |
178 | return $octets; |
4411f3b6 |
179 | } |
180 | |
181 | sub decode |
182 | { |
18586f54 |
183 | my ($name,$octets,$check) = @_; |
184 | my $enc = find_encoding($name); |
185 | croak("Unknown encoding '$name'") unless defined $enc; |
186 | my $string = $enc->decode($octets,$check); |
187 | $_[1] = $octets if $check; |
188 | return $string; |
4411f3b6 |
189 | } |
190 | |
191 | sub from_to |
192 | { |
18586f54 |
193 | my ($string,$from,$to,$check) = @_; |
194 | my $f = find_encoding($from); |
195 | croak("Unknown encoding '$from'") unless defined $f; |
196 | my $t = find_encoding($to); |
197 | croak("Unknown encoding '$to'") unless defined $t; |
198 | my $uni = $f->decode($string,$check); |
199 | return undef if ($check && length($string)); |
200 | $string = $t->encode($uni,$check); |
201 | return undef if ($check && length($uni)); |
3ef515df |
202 | return defined($_[0] = $string) ? length($string) : undef ; |
4411f3b6 |
203 | } |
204 | |
205 | sub encode_utf8 |
206 | { |
18586f54 |
207 | my ($str) = @_; |
208 | utf8::encode($str); |
209 | return $str; |
4411f3b6 |
210 | } |
211 | |
212 | sub decode_utf8 |
213 | { |
18586f54 |
214 | my ($str) = @_; |
215 | return undef unless utf8::decode($str); |
216 | return $str; |
5ad8ef52 |
217 | } |
218 | |
18586f54 |
219 | require Encode::Encoding; |
220 | require Encode::XS; |
221 | require Encode::Internal; |
222 | require Encode::Unicode; |
223 | require Encode::utf8; |
64ffdd5e |
224 | require Encode::10646_1; |
18586f54 |
225 | require Encode::ucs2_le; |
4411f3b6 |
226 | |
656753f8 |
227 | 1; |
228 | |
2a936312 |
229 | __END__ |
230 | |
4411f3b6 |
231 | =head1 NAME |
232 | |
233 | Encode - character encodings |
234 | |
235 | =head1 SYNOPSIS |
236 | |
237 | use Encode; |
238 | |
67d7b5ef |
239 | |
240 | =head2 Table of Contents |
241 | |
242 | Encode consists of a collection of modules which details are too big |
243 | to fit in one document. This POD itself explains the top-level APIs |
244 | and general topics at a glance. For other topics and more details, |
245 | see the PODs below; |
246 | |
247 | Name Description |
248 | -------------------------------------------------------- |
249 | Encode::Alias Alias defintions to encodings |
250 | Encode::Encoding Encode Implementation Base Class |
251 | Encode::Supported List of Supported Encodings |
252 | Encode::CN Simplified Chinese Encodings |
253 | Encode::JP Japanese Encodings |
254 | Encode::KR Korean Encodings |
255 | Encode::TW Traditional Chinese Encodings |
256 | -------------------------------------------------------- |
257 | |
4411f3b6 |
258 | =head1 DESCRIPTION |
259 | |
47bfe92f |
260 | The C<Encode> module provides the interfaces between Perl's strings |
67d7b5ef |
261 | and the rest of the system. Perl strings are sequences of |
262 | B<characters>. |
263 | |
264 | The repertoire of characters that Perl can represent is at least that |
265 | defined by the Unicode Consortium. On most platforms the ordinal |
266 | values of the characters (as returned by C<ord(ch)>) is the "Unicode |
267 | codepoint" for the character (the exceptions are those platforms where |
268 | the legacy encoding is some variant of EBCDIC rather than a super-set |
269 | of ASCII - see L<perlebcdic>). |
270 | |
271 | Traditionally computer data has been moved around in 8-bit chunks |
272 | often called "bytes". These chunks are also known as "octets" in |
273 | networking standards. Perl is widely used to manipulate data of many |
274 | types - not only strings of characters representing human or computer |
275 | languages but also "binary" data being the machines representation of |
276 | numbers, pixels in an image - or just about anything. |
277 | |
278 | When Perl is processing "binary data" the programmer wants Perl to |
279 | process "sequences of bytes". This is not a problem for Perl - as a |
280 | byte has 256 possible values it easily fits in Perl's much larger |
281 | "logical character". |
282 | |
283 | =head2 TERMINOLOGY |
4411f3b6 |
284 | |
67d7b5ef |
285 | =over 4 |
21938dfa |
286 | |
67d7b5ef |
287 | =item * |
288 | |
289 | I<character>: a character in the range 0..(2**32-1) (or more). |
290 | (What Perl's strings are made of.) |
291 | |
292 | =item * |
293 | |
294 | I<byte>: a character in the range 0..255 |
295 | (A special case of a Perl character.) |
296 | |
297 | =item * |
298 | |
299 | I<octet>: 8 bits of data, with ordinal values 0..255 |
300 | (Term for bytes passed to or from a non-Perl context, e.g. disk file.) |
301 | |
302 | =back |
4411f3b6 |
303 | |
67d7b5ef |
304 | The marker [INTERNAL] marks Internal Implementation Details, in |
305 | general meant only for those who think they know what they are doing, |
306 | and such details may change in future releases. |
307 | |
308 | =head1 PERL ENCODING API |
4411f3b6 |
309 | |
310 | =over 4 |
311 | |
a63c962f |
312 | =item $bytes = encode(ENCODING, $string[, CHECK]) |
4411f3b6 |
313 | |
47bfe92f |
314 | Encodes string from Perl's internal form into I<ENCODING> and returns |
67d7b5ef |
315 | a sequence of octets. ENCODING can be either a canonical name or |
316 | alias. For encoding names and aliases, see L</"Defining Aliases">. |
317 | For CHECK see L</"Handling Malformed Data">. |
4411f3b6 |
318 | |
67d7b5ef |
319 | For example to convert (internally UTF-8 encoded) Unicode string to |
320 | iso-8859-1 (also known as Latin1), |
681a7c68 |
321 | |
67d7b5ef |
322 | $octets = encode("iso-8859-1", $unicode); |
681a7c68 |
323 | |
a63c962f |
324 | =item $string = decode(ENCODING, $bytes[, CHECK]) |
4411f3b6 |
325 | |
47bfe92f |
326 | Decode sequence of octets assumed to be in I<ENCODING> into Perl's |
67d7b5ef |
327 | internal form and returns the resulting string. as in encode(), |
328 | ENCODING can be either a canonical name or alias. For encoding names |
329 | and aliases, see L</"Defining Aliases">. For CHECK see |
47bfe92f |
330 | L</"Handling Malformed Data">. |
331 | |
1b2c56c8 |
332 | For example to convert ISO-8859-1 data to UTF-8: |
681a7c68 |
333 | |
67d7b5ef |
334 | $utf8 = decode("iso-8859-1", $latin1); |
681a7c68 |
335 | |
3ef515df |
336 | =item [$length =] from_to($string, FROM_ENCODING, TO_ENCODING[, CHECK]) |
47bfe92f |
337 | |
2b106fbe |
338 | Convert B<in-place> the data between two encodings. How did the data |
339 | in $string originally get to be in FROM_ENCODING? Either using |
67d7b5ef |
340 | encode() or through PerlIO: See L</"Encoding and IO">. |
341 | For encoding names and aliases, see L</"Defining Aliases">. |
342 | For CHECK see L</"Handling Malformed Data">. |
2b106fbe |
343 | |
1b2c56c8 |
344 | For example to convert ISO-8859-1 data to UTF-8: |
2b106fbe |
345 | |
346 | from_to($data, "iso-8859-1", "utf-8"); |
347 | |
348 | and to convert it back: |
349 | |
350 | from_to($data, "utf-8", "iso-8859-1"); |
4411f3b6 |
351 | |
ab97ca19 |
352 | Note that because the conversion happens in place, the data to be |
353 | converted cannot be a string constant, it must be a scalar variable. |
354 | |
3ef515df |
355 | from_to() return the length of the converted string on success, undef |
356 | otherwise. |
357 | |
4411f3b6 |
358 | =back |
359 | |
51ef4e11 |
360 | =head2 Listing available encodings |
361 | |
5129552c |
362 | use Encode; |
363 | @list = Encode->encodings(); |
364 | |
365 | Returns a list of the canonical names of the available encodings that |
366 | are loaded. To get a list of all available encodings including the |
367 | ones that are not loaded yet, say |
368 | |
369 | @all_encodings = Encode->encodings(":all"); |
370 | |
371 | Or you can give the name of specific module. |
372 | |
373 | @with_jp = Encode->encodings("Encode/JP.pm"); |
51ef4e11 |
374 | |
a63c962f |
375 | Note in this case you have to say C<"Encode/JP.pm"> instead of |
376 | C<"Encode::JP">. |
5d030b67 |
377 | |
a63c962f |
378 | To find which encodings are supported by this package in details, |
5d030b67 |
379 | see L<Encode::Supported>. |
51ef4e11 |
380 | |
67d7b5ef |
381 | |
51ef4e11 |
382 | =head2 Defining Aliases |
383 | |
67d7b5ef |
384 | To add new alias to a given encoding, Use; |
385 | |
5129552c |
386 | use Encode; |
387 | use Encode::Alias; |
a63c962f |
388 | define_alias(newName => ENCODING); |
51ef4e11 |
389 | |
3ef515df |
390 | After that, newName can be used as an alias for ENCODING. |
391 | ENCODING may be either the name of an encoding or an I<encoding |
392 | object> |
51ef4e11 |
393 | |
5d030b67 |
394 | See L<Encode::Alias> on details. |
51ef4e11 |
395 | |
4411f3b6 |
396 | =head1 Encoding and IO |
397 | |
398 | It is very common to want to do encoding transformations when |
399 | reading or writing files, network connections, pipes etc. |
47bfe92f |
400 | If Perl is configured to use the new 'perlio' IO system then |
4411f3b6 |
401 | C<Encode> provides a "layer" (See L<perliol>) which can transform |
402 | data as it is read or written. |
403 | |
8e86646e |
404 | Here is how the blind poet would modernise the encoding: |
405 | |
42234700 |
406 | use Encode; |
8e86646e |
407 | open(my $iliad,'<:encoding(iso-8859-7)','iliad.greek'); |
408 | open(my $utf8,'>:utf8','iliad.utf8'); |
409 | my @epic = <$iliad>; |
410 | print $utf8 @epic; |
411 | close($utf8); |
412 | close($illiad); |
4411f3b6 |
413 | |
414 | In addition the new IO system can also be configured to read/write |
415 | UTF-8 encoded characters (as noted above this is efficient): |
416 | |
e9692b5b |
417 | open(my $fh,'>:utf8','anything'); |
418 | print $fh "Any \x{0021} string \N{SMILEY FACE}\n"; |
4411f3b6 |
419 | |
420 | Either of the above forms of "layer" specifications can be made the default |
421 | for a lexical scope with the C<use open ...> pragma. See L<open>. |
422 | |
423 | Once a handle is open is layers can be altered using C<binmode>. |
424 | |
47bfe92f |
425 | Without any such configuration, or if Perl itself is built using |
4411f3b6 |
426 | system's own IO, then write operations assume that file handle accepts |
427 | only I<bytes> and will C<die> if a character larger than 255 is |
428 | written to the handle. When reading, each octet from the handle |
429 | becomes a byte-in-a-character. Note that this default is the same |
47bfe92f |
430 | behaviour as bytes-only languages (including Perl before v5.6) would |
431 | have, and is sufficient to handle native 8-bit encodings |
432 | e.g. iso-8859-1, EBCDIC etc. and any legacy mechanisms for handling |
433 | other encodings and binary data. |
434 | |
435 | In other cases it is the programs responsibility to transform |
436 | characters into bytes using the API above before doing writes, and to |
437 | transform the bytes read from a handle into characters before doing |
438 | "character operations" (e.g. C<lc>, C</\W+/>, ...). |
439 | |
47bfe92f |
440 | You can also use PerlIO to convert larger amounts of data you don't |
1b2c56c8 |
441 | want to bring into memory. For example to convert between ISO-8859-1 |
47bfe92f |
442 | (Latin 1) and UTF-8 (or UTF-EBCDIC in EBCDIC machines): |
443 | |
e9692b5b |
444 | open(F, "<:encoding(iso-8859-1)", "data.txt") or die $!; |
445 | open(G, ">:utf8", "data.utf") or die $!; |
446 | while (<F>) { print G } |
447 | |
448 | # Could also do "print G <F>" but that would pull |
449 | # the whole file into memory just to write it out again. |
450 | |
451 | More examples: |
47bfe92f |
452 | |
e9692b5b |
453 | open(my $f, "<:encoding(cp1252)") |
454 | open(my $g, ">:encoding(iso-8859-2)") |
455 | open(my $h, ">:encoding(latin9)") # iso-8859-15 |
47bfe92f |
456 | |
457 | See L<PerlIO> for more information. |
4411f3b6 |
458 | |
1768d7eb |
459 | See also L<encoding> for how to change the default encoding of the |
d521382b |
460 | data in your script. |
1768d7eb |
461 | |
67d7b5ef |
462 | =head1 Handling Malformed Data |
463 | |
464 | If CHECK is not set, C<undef> is returned. If the data is supposed to |
465 | be UTF-8, an optional lexical warning (category utf8) is given. If |
466 | CHECK is true but not a code reference, dies. |
467 | |
468 | It would desirable to have a way to indicate that transform should use |
469 | the encodings "replacement character" - no such mechanism is defined yet. |
470 | |
471 | It is also planned to allow I<CHECK> to be a code reference. |
472 | |
473 | This is not yet implemented as there are design issues with what its |
474 | arguments should be and how it returns its results. |
475 | |
476 | =over 4 |
477 | |
478 | =item Scheme 1 |
479 | |
480 | Passed remaining fragment of string being processed. |
481 | Modifies it in place to remove bytes/characters it can understand |
482 | and returns a string used to represent them. |
483 | e.g. |
484 | |
485 | sub fixup { |
486 | my $ch = substr($_[0],0,1,''); |
487 | return sprintf("\x{%02X}",ord($ch); |
488 | } |
489 | |
490 | This scheme is close to how underlying C code for Encode works, but gives |
491 | the fixup routine very little context. |
492 | |
493 | =item Scheme 2 |
494 | |
495 | Passed original string, and an index into it of the problem area, and |
496 | output string so far. Appends what it will to output string and |
497 | returns new index into original string. For example: |
498 | |
499 | sub fixup { |
500 | # my ($s,$i,$d) = @_; |
501 | my $ch = substr($_[0],$_[1],1); |
502 | $_[2] .= sprintf("\x{%02X}",ord($ch); |
503 | return $_[1]+1; |
504 | } |
505 | |
506 | This scheme gives maximal control to the fixup routine but is more |
507 | complicated to code, and may need internals of Encode to be tweaked to |
508 | keep original string intact. |
509 | |
510 | =item Other Schemes |
511 | |
512 | Hybrids of above. |
513 | |
514 | Multiple return values rather than in-place modifications. |
515 | |
516 | Index into the string could be C<pos($str)> allowing C<s/\G...//>. |
517 | |
518 | =back |
519 | |
520 | =head2 UTF-8 / utf8 |
521 | |
522 | The Unicode consortium defines the UTF-8 standard as a way of encoding |
523 | the entire Unicode repertoire as sequences of octets. This encoding is |
524 | expected to become very widespread. Perl can use this form internally |
525 | to represent strings, so conversions to and from this form are |
526 | particularly efficient (as octets in memory do not have to change, |
527 | just the meta-data that tells Perl how to treat them). |
528 | |
529 | =over 4 |
530 | |
531 | =item $bytes = encode_utf8($string); |
532 | |
533 | The characters that comprise string are encoded in Perl's superset of UTF-8 |
534 | and the resulting octets returned as a sequence of bytes. All possible |
535 | characters have a UTF-8 representation so this function cannot fail. |
536 | |
537 | =item $string = decode_utf8($bytes [, CHECK]); |
538 | |
539 | The sequence of octets represented by $bytes is decoded from UTF-8 |
540 | into a sequence of logical characters. Not all sequences of octets |
541 | form valid UTF-8 encodings, so it is possible for this call to fail. |
542 | For CHECK see L</"Handling Malformed Data">. |
543 | |
544 | =back |
545 | |
546 | =head1 Defining Encodings |
547 | |
548 | To define a new encoding, use: |
549 | |
550 | use Encode qw(define_alias); |
551 | define_encoding($object, 'canonicalName' [, alias...]); |
552 | |
553 | I<canonicalName> will be associated with I<$object>. The object |
554 | should provide the interface described in L<Encode::Encoding> |
555 | If more than two arguments are provided then additional |
556 | arguments are taken as aliases for I<$object> as for C<define_alias>. |
557 | |
4411f3b6 |
558 | =head1 Messing with Perl's Internals |
559 | |
47bfe92f |
560 | The following API uses parts of Perl's internals in the current |
561 | implementation. As such they are efficient, but may change. |
4411f3b6 |
562 | |
563 | =over 4 |
564 | |
a63c962f |
565 | =item is_utf8(STRING [, CHECK]) |
4411f3b6 |
566 | |
567 | [INTERNAL] Test whether the UTF-8 flag is turned on in the STRING. |
47bfe92f |
568 | If CHECK is true, also checks the data in STRING for being well-formed |
569 | UTF-8. Returns true if successful, false otherwise. |
4411f3b6 |
570 | |
a63c962f |
571 | =item _utf8_on(STRING) |
4411f3b6 |
572 | |
573 | [INTERNAL] Turn on the UTF-8 flag in STRING. The data in STRING is |
574 | B<not> checked for being well-formed UTF-8. Do not use unless you |
575 | B<know> that the STRING is well-formed UTF-8. Returns the previous |
576 | state of the UTF-8 flag (so please don't test the return value as |
577 | I<not> success or failure), or C<undef> if STRING is not a string. |
578 | |
a63c962f |
579 | =item _utf8_off(STRING) |
4411f3b6 |
580 | |
581 | [INTERNAL] Turn off the UTF-8 flag in STRING. Do not use frivolously. |
582 | Returns the previous state of the UTF-8 flag (so please don't test the |
583 | return value as I<not> success or failure), or C<undef> if STRING is |
584 | not a string. |
585 | |
586 | =back |
587 | |
588 | =head1 SEE ALSO |
589 | |
5d030b67 |
590 | L<Encode::Encoding>, |
591 | L<Encode::Supported>, |
592 | L<PerlIO>, |
593 | L<encoding>, |
594 | L<perlebcdic>, |
595 | L<perlfunc/open>, |
596 | L<perlunicode>, |
597 | L<utf8>, |
598 | the Perl Unicode Mailing List E<lt>perl-unicode@perl.orgE<gt> |
4411f3b6 |
599 | |
600 | =cut |