mirror of
https://github.com/ruby/ruby.git
synced 2022-11-09 12:17:21 -05:00
86da4be028
git-svn-id: svn+ssh://ci.ruby-lang.org/ruby/trunk@15003 b2dd03c8-39d4-4d8f-98ff-823fe69b080e
495 lines
12 KiB
C
495 lines
12 KiB
C
/*
|
|
* NKF - Ruby extension for Network Kanji Filter
|
|
*
|
|
* original nkf2.x is maintained at http://sourceforge.jp/projects/nkf/
|
|
*
|
|
* $Id$
|
|
*
|
|
*/
|
|
|
|
#define RUBY_NKF_REVISION "$Revision$"
|
|
#define RUBY_NKF_VERSION NKF_VERSION " (" NKF_RELEASE_DATE ")"
|
|
|
|
#include "ruby/ruby.h"
|
|
#include "ruby/encoding.h"
|
|
|
|
/* Replace nkf's getchar/putchar for variable modification */
|
|
/* we never use getc, ungetc */
|
|
|
|
#undef getc
|
|
#undef ungetc
|
|
#define getc(f) (input_ctr>=i_len?-1:input[input_ctr++])
|
|
#define ungetc(c,f) input_ctr--
|
|
|
|
#define INCSIZE 32
|
|
#undef putchar
|
|
#undef TRUE
|
|
#undef FALSE
|
|
#define putchar(c) rb_nkf_putchar(c)
|
|
|
|
/* Input/Output pointers */
|
|
|
|
static unsigned char *output;
|
|
static unsigned char *input;
|
|
static int input_ctr;
|
|
static int i_len;
|
|
static int output_ctr;
|
|
static int o_len;
|
|
static int incsize;
|
|
|
|
static VALUE result;
|
|
|
|
static int
|
|
rb_nkf_putchar(unsigned int c)
|
|
{
|
|
if (output_ctr >= o_len) {
|
|
o_len += incsize;
|
|
rb_str_resize(result, o_len);
|
|
incsize *= 2;
|
|
output = (unsigned char *)RSTRING_PTR(result);
|
|
}
|
|
output[output_ctr++] = c;
|
|
|
|
return c;
|
|
}
|
|
|
|
/* Include kanji filter main part */
|
|
/* getchar and putchar will be replaced during inclusion */
|
|
|
|
#define PERL_XS 1
|
|
#include "nkf-utf8/config.h"
|
|
#include "nkf-utf8/utf8tbl.c"
|
|
#include "nkf-utf8/nkf.c"
|
|
|
|
rb_encoding* rb_nkf_enc_get(const char *name)
|
|
{
|
|
int idx = rb_enc_find_index(name);
|
|
if (idx < 0) {
|
|
nkf_encoding *nkf_enc = nkf_enc_find(name);
|
|
idx = rb_enc_find_index(nkf_enc_name(nkf_enc_to_base_encoding(nkf_enc)));
|
|
if (idx < 0) {
|
|
idx = rb_define_dummy_encoding(name);
|
|
} else {
|
|
rb_encoding *rb_enc = rb_enc_from_index(idx);
|
|
idx = rb_enc_replicate(name, rb_enc);
|
|
}
|
|
}
|
|
return rb_enc_from_index(idx);
|
|
}
|
|
|
|
int nkf_split_options(const char *arg)
|
|
{
|
|
int count = 0;
|
|
char option[256];
|
|
int i = 0, j = 0;
|
|
int is_escaped = FALSE;
|
|
int is_single_quoted = FALSE;
|
|
int is_double_quoted = FALSE;
|
|
for(i = 0; arg[i]; i++){
|
|
if(j == 255){
|
|
return -1;
|
|
}else if(is_single_quoted){
|
|
if(arg[i] == '\''){
|
|
is_single_quoted = FALSE;
|
|
}else{
|
|
option[j++] = arg[i];
|
|
}
|
|
}else if(is_escaped){
|
|
is_escaped = FALSE;
|
|
option[j++] = arg[i];
|
|
}else if(arg[i] == '\\'){
|
|
is_escaped = TRUE;
|
|
}else if(is_double_quoted){
|
|
if(arg[i] == '"'){
|
|
is_double_quoted = FALSE;
|
|
}else{
|
|
option[j++] = arg[i];
|
|
}
|
|
}else if(arg[i] == '\''){
|
|
is_single_quoted = TRUE;
|
|
}else if(arg[i] == '"'){
|
|
is_double_quoted = TRUE;
|
|
}else if(arg[i] == ' '){
|
|
option[j] = '\0';
|
|
options(option);
|
|
j = 0;
|
|
}else{
|
|
option[j++] = arg[i];
|
|
}
|
|
}
|
|
if(j){
|
|
option[j] = '\0';
|
|
options(option);
|
|
}
|
|
return count;
|
|
}
|
|
|
|
/*
|
|
* call-seq:
|
|
* NKF.nkf(opt, str) => string
|
|
*
|
|
* Convert _str_ and return converted result.
|
|
* Conversion details are specified by _opt_ as String.
|
|
*
|
|
* require 'nkf'
|
|
* output = NKF.nkf("-s", input)
|
|
*/
|
|
|
|
static VALUE
|
|
rb_nkf_convert(VALUE obj, VALUE opt, VALUE src)
|
|
{
|
|
rb_encoding *to_enc;
|
|
const char *to_e;
|
|
int to_encidx;
|
|
|
|
reinit();
|
|
StringValue(opt);
|
|
nkf_split_options(RSTRING_PTR(opt));
|
|
|
|
incsize = INCSIZE;
|
|
|
|
input_ctr = 0;
|
|
StringValue(src);
|
|
input = (unsigned char *)RSTRING_PTR(src);
|
|
i_len = RSTRING_LEN(src);
|
|
result = rb_str_new(0, i_len*3 + 10);
|
|
|
|
output_ctr = 0;
|
|
output = (unsigned char *)RSTRING_PTR(result);
|
|
o_len = RSTRING_LEN(result);
|
|
*output = '\0';
|
|
|
|
kanji_convert(NULL);
|
|
rb_str_set_len(result, output_ctr);
|
|
OBJ_INFECT(result, src);
|
|
|
|
rb_enc_associate(result, rb_nkf_enc_get(nkf_enc_name(output_encoding)));
|
|
|
|
return result;
|
|
}
|
|
|
|
|
|
/*
|
|
* call-seq:
|
|
* NKF.guess(str) => encoding
|
|
*
|
|
* Returns guessed encoding of _str_ by nkf routine.
|
|
*
|
|
*/
|
|
|
|
static VALUE
|
|
rb_nkf_guess(VALUE obj, VALUE src)
|
|
{
|
|
reinit();
|
|
|
|
input_ctr = 0;
|
|
StringValue(src);
|
|
input = (unsigned char *)RSTRING_PTR(src);
|
|
i_len = RSTRING_LEN(src);
|
|
|
|
guess_f = TRUE;
|
|
kanji_convert( NULL );
|
|
guess_f = FALSE;
|
|
|
|
return rb_enc_from_encoding(rb_nkf_enc_get(get_guessed_code()));
|
|
}
|
|
|
|
|
|
/*
|
|
* NKF - Ruby extension for Network Kanji Filter
|
|
*
|
|
* == Description
|
|
*
|
|
* This is a Ruby Extension version of nkf (Network Kanji Filter).
|
|
* It converts the first argument and returns converted result. Conversion
|
|
* details are specified by flags as the first argument.
|
|
*
|
|
* *Nkf* is a yet another kanji code converter among networks, hosts and terminals.
|
|
* It converts input kanji code to designated kanji code
|
|
* such as ISO-2022-JP, Shift_JIS, EUC-JP, UTF-8 or UTF-16.
|
|
*
|
|
* One of the most unique faculty of *nkf* is the guess of the input kanji encodings.
|
|
* It currently recognizes ISO-2022-JP, Shift_JIS, EUC-JP, UTF-8 and UTF-16.
|
|
* So users needn't set the input kanji code explicitly.
|
|
*
|
|
* By default, X0201 kana is converted into X0208 kana.
|
|
* For X0201 kana, SO/SI, SSO and ESC-(-I methods are supported.
|
|
* For automatic code detection, nkf assumes no X0201 kana in Shift_JIS.
|
|
* To accept X0201 in Shift_JIS, use <b>-X</b>, <b>-x</b> or <b>-S</b>.
|
|
*
|
|
* == Flags
|
|
*
|
|
* === -b -u
|
|
*
|
|
* Output is buffered (DEFAULT), Output is unbuffered.
|
|
*
|
|
* === -j -s -e -w -w16 -w32
|
|
*
|
|
* Output code is ISO-2022-JP (7bit JIS), Shift_JIS, EUC-JP,
|
|
* UTF-8N, UTF-16BE, UTF-32BE.
|
|
* Without this option and compile option, ISO-2022-JP is assumed.
|
|
*
|
|
* === -J -S -E -W -W16 -W32
|
|
*
|
|
* Input assumption is JIS 7 bit, Shift_JIS, EUC-JP,
|
|
* UTF-8, UTF-16, UTF-32.
|
|
*
|
|
* ==== -J
|
|
*
|
|
* Assume JIS input. It also accepts EUC-JP.
|
|
* This is the default. This flag does not exclude Shift_JIS.
|
|
*
|
|
* ==== -S
|
|
*
|
|
* Assume Shift_JIS and X0201 kana input. It also accepts JIS.
|
|
* EUC-JP is recognized as X0201 kana. Without <b>-x</b> flag,
|
|
* X0201 kana (halfwidth kana) is converted into X0208.
|
|
*
|
|
* ==== -E
|
|
*
|
|
* Assume EUC-JP input. It also accepts JIS.
|
|
* Same as -J.
|
|
*
|
|
* === -t
|
|
*
|
|
* No conversion.
|
|
*
|
|
* === -i_
|
|
*
|
|
* Output sequence to designate JIS-kanji. (DEFAULT B)
|
|
*
|
|
* === -o_
|
|
*
|
|
* Output sequence to designate ASCII. (DEFAULT B)
|
|
*
|
|
* === -r
|
|
*
|
|
* {de/en}crypt ROT13/47
|
|
*
|
|
* === -h[123] --hiragana --katakana --katakana-hiragana
|
|
*
|
|
* [-h1 --hiragana] Katakana to Hiragana conversion.
|
|
*
|
|
* [-h2 --katakana] Hiragana to Katakana conversion.
|
|
*
|
|
* [-h3 --katakana-hiragana] Katakana to Hiragana and Hiragana to Katakana conversion.
|
|
*
|
|
* === -T
|
|
*
|
|
* Text mode output (MS-DOS)
|
|
*
|
|
* === -l
|
|
*
|
|
* ISO8859-1 (Latin-1) support
|
|
*
|
|
* === -f[<code>m</code> [- <code>n</code>]]
|
|
*
|
|
* Folding on <code>m</code> length with <code>n</code> margin in a line.
|
|
* Without this option, fold length is 60 and fold margin is 10.
|
|
*
|
|
* === -F
|
|
*
|
|
* New line preserving line folding.
|
|
*
|
|
* === -Z[0-3]
|
|
*
|
|
* Convert X0208 alphabet (Fullwidth Alphabets) to ASCII.
|
|
*
|
|
* [-Z -Z0] Convert X0208 alphabet to ASCII.
|
|
*
|
|
* [-Z1] Converts X0208 kankaku to single ASCII space.
|
|
*
|
|
* [-Z2] Converts X0208 kankaku to double ASCII spaces.
|
|
*
|
|
* [-Z3] Replacing Fullwidth >, <, ", & into '>', '<', '"', '&' as in HTML.
|
|
*
|
|
* === -X -x
|
|
*
|
|
* Assume X0201 kana in MS-Kanji.
|
|
* With <b>-X</b> or without this option, X0201 is converted into X0208 Kana.
|
|
* With <b>-x</b>, try to preserve X0208 kana and do not convert X0201 kana to X0208.
|
|
* In JIS output, ESC-(-I is used. In EUC output, SSO is used.
|
|
*
|
|
* === -B[0-2]
|
|
*
|
|
* Assume broken JIS-Kanji input, which lost ESC.
|
|
* Useful when your site is using old B-News Nihongo patch.
|
|
*
|
|
* [-B1] allows any char after ESC-( or ESC-$.
|
|
*
|
|
* [-B2] forces ASCII after NL.
|
|
*
|
|
* === -I
|
|
*
|
|
* Replacing non iso-2022-jp char into a geta character
|
|
* (substitute character in Japanese).
|
|
*
|
|
* === -d -c
|
|
*
|
|
* Delete \r in line feed, Add \r in line feed.
|
|
*
|
|
* === -m[BQN0]
|
|
*
|
|
* MIME ISO-2022-JP/ISO8859-1 decode. (DEFAULT)
|
|
* To see ISO8859-1 (Latin-1) -l is necessary.
|
|
*
|
|
* [-mB] Decode MIME base64 encoded stream. Remove header or other part before
|
|
* conversion.
|
|
*
|
|
* [-mQ] Decode MIME quoted stream. '_' in quoted stream is converted to space.
|
|
*
|
|
* [-mN] Non-strict decoding.
|
|
* It allows line break in the middle of the base64 encoding.
|
|
*
|
|
* [-m0] No MIME decode.
|
|
*
|
|
* === -M
|
|
*
|
|
* MIME encode. Header style. All ASCII code and control characters are intact.
|
|
* Kanji conversion is performed before encoding, so this cannot be used as a picture encoder.
|
|
*
|
|
* [-MB] MIME encode Base64 stream.
|
|
*
|
|
* [-MQ] Perfome quoted encoding.
|
|
*
|
|
* === -l
|
|
*
|
|
* Input and output code is ISO8859-1 (Latin-1) and ISO-2022-JP.
|
|
* <b>-s</b>, <b>-e</b> and <b>-x</b> are not compatible with this option.
|
|
*
|
|
* === -L[uwm]
|
|
*
|
|
* new line mode
|
|
* Without this option, nkf doesn't convert line breaks.
|
|
*
|
|
* [-Lu] unix (LF)
|
|
*
|
|
* [-Lw] windows (CRLF)
|
|
*
|
|
* [-Lm] mac (CR)
|
|
*
|
|
* === --fj --unix --mac --msdos --windows
|
|
*
|
|
* convert for these system
|
|
*
|
|
* === --jis --euc --sjis --mime --base64
|
|
*
|
|
* convert for named code
|
|
*
|
|
* === --jis-input --euc-input --sjis-input --mime-input --base64-input
|
|
*
|
|
* assume input system
|
|
*
|
|
* === --ic=<code>input codeset</code> --oc=<code>output codeset</code>
|
|
*
|
|
* Set the input or output codeset.
|
|
* NKF supports following codesets and those codeset name are case insensitive.
|
|
*
|
|
* [ISO-2022-JP] a.k.a. RFC1468, 7bit JIS, JUNET
|
|
*
|
|
* [EUC-JP (eucJP-nkf)] a.k.a. AT&T JIS, Japanese EUC, UJIS
|
|
*
|
|
* [eucJP-ascii] a.k.a. x-eucjp-open-19970715-ascii
|
|
*
|
|
* [eucJP-ms] a.k.a. x-eucjp-open-19970715-ms
|
|
*
|
|
* [CP51932] Microsoft Version of EUC-JP.
|
|
*
|
|
* [Shift_JIS] SJIS, MS-Kanji
|
|
*
|
|
* [Windows-31J] a.k.a. CP932
|
|
*
|
|
* [UTF-8] same as UTF-8N
|
|
*
|
|
* [UTF-8N] UTF-8 without BOM
|
|
*
|
|
* [UTF-8-BOM] UTF-8 with BOM
|
|
*
|
|
* [UTF-16] same as UTF-16BE
|
|
*
|
|
* [UTF-16BE] UTF-16 Big Endian without BOM
|
|
*
|
|
* [UTF-16BE-BOM] UTF-16 Big Endian with BOM
|
|
*
|
|
* [UTF-16LE] UTF-16 Little Endian without BOM
|
|
*
|
|
* [UTF-16LE-BOM] UTF-16 Little Endian with BOM
|
|
*
|
|
* [UTF-32] same as UTF-32BE
|
|
*
|
|
* [UTF-32BE] UTF-32 Big Endian without BOM
|
|
*
|
|
* [UTF-32BE-BOM] UTF-32 Big Endian with BOM
|
|
*
|
|
* [UTF-32LE] UTF-32 Little Endian without BOM
|
|
*
|
|
* [UTF-32LE-BOM] UTF-32 Little Endian with BOM
|
|
*
|
|
* [UTF8-MAC] NKDed UTF-8, a.k.a. UTF8-NFD (input only)
|
|
*
|
|
* === --fb-{skip, html, xml, perl, java, subchar}
|
|
*
|
|
* Specify the way that nkf handles unassigned characters.
|
|
* Without this option, --fb-skip is assumed.
|
|
*
|
|
* === --prefix= <code>escape character</code> <code>target character</code> ..
|
|
*
|
|
* When nkf converts to Shift_JIS,
|
|
* nkf adds a specified escape character to specified 2nd byte of Shift_JIS characters.
|
|
* 1st byte of argument is the escape character and following bytes are target characters.
|
|
*
|
|
* === --no-cp932ext
|
|
*
|
|
* Handle the characters extended in CP932 as unassigned characters.
|
|
*
|
|
* == --no-best-fit-chars
|
|
*
|
|
* When Unicode to Encoded byte conversion,
|
|
* don't convert characters which is not round trip safe.
|
|
* When Unicode to Unicode conversion,
|
|
* with this and -x option, nkf can be used as UTF converter.
|
|
* (In other words, without this and -x option, nkf doesn't save some characters)
|
|
*
|
|
* When nkf convert string which related to path, you should use this opion.
|
|
*
|
|
* === --cap-input
|
|
*
|
|
* Decode hex encoded characters.
|
|
*
|
|
* === --url-input
|
|
*
|
|
* Unescape percent escaped characters.
|
|
*
|
|
* === --
|
|
*
|
|
* Ignore rest of -option.
|
|
*/
|
|
|
|
void
|
|
Init_nkf()
|
|
{
|
|
VALUE mNKF = rb_define_module("NKF");
|
|
|
|
rb_define_module_function(mNKF, "nkf", rb_nkf_convert, 2);
|
|
rb_define_module_function(mNKF, "guess", rb_nkf_guess, 1);
|
|
rb_define_alias(rb_singleton_class(mNKF), "guess", "guess");
|
|
|
|
rb_define_const(mNKF, "AUTO", Qnil);
|
|
rb_define_const(mNKF, "NOCONV", Qnil);
|
|
rb_define_const(mNKF, "UNKNOWN", Qnil);
|
|
rb_define_const(mNKF, "BINARY", rb_enc_from_encoding(rb_nkf_enc_get("BINARY")));
|
|
rb_define_const(mNKF, "ASCII", rb_enc_from_encoding(rb_nkf_enc_get("US-ASCII")));
|
|
rb_define_const(mNKF, "JIS", rb_enc_from_encoding(rb_nkf_enc_get("ISO-2022-JP")));
|
|
rb_define_const(mNKF, "EUC", rb_enc_from_encoding(rb_nkf_enc_get("EUC-JP")));
|
|
rb_define_const(mNKF, "SJIS", rb_enc_from_encoding(rb_nkf_enc_get("Shift_JIS")));
|
|
rb_define_const(mNKF, "UTF8", rb_enc_from_encoding(rb_utf8_encoding()));
|
|
rb_define_const(mNKF, "UTF16", rb_enc_from_encoding(rb_nkf_enc_get("UTF-16")));
|
|
rb_define_const(mNKF, "UTF32", rb_enc_from_encoding(rb_nkf_enc_get("UTF-32")));
|
|
|
|
/* Full version string of nkf */
|
|
rb_define_const(mNKF, "VERSION", rb_str_new2(RUBY_NKF_VERSION));
|
|
/* Version of nkf */
|
|
rb_define_const(mNKF, "NKF_VERSION", rb_str_new2(NKF_VERSION));
|
|
/* Release date of nkf */
|
|
rb_define_const(mNKF, "NKF_RELEASE_DATE", rb_str_new2(NKF_RELEASE_DATE));
|
|
}
|