ruby/transcode.c

414 строки
13 KiB
C

/**********************************************************************
transcode.c -
$Author$
$Date$
created at: Tue Oct 30 16:10:22 JST 2007
Copyright (C) 2007 Martin Duerst
**********************************************************************/
#include "ruby/ruby.h"
#include "ruby/encoding.h"
#include "transcode_data.h"
VALUE rb_str_tmp_new(long);
VALUE rb_str_shared_replace(VALUE, VALUE);
/*
* Dispatch data and logic
*/
/* extern declarations, should use some include file here */
extern const BYTE_LOOKUP from_ISO_8859_1;
extern const BYTE_LOOKUP from_ISO_8859_2;
extern const BYTE_LOOKUP from_ISO_8859_3;
extern const BYTE_LOOKUP from_ISO_8859_4;
extern const BYTE_LOOKUP from_ISO_8859_5;
extern const BYTE_LOOKUP from_ISO_8859_6;
extern const BYTE_LOOKUP from_ISO_8859_7;
extern const BYTE_LOOKUP from_ISO_8859_8;
extern const BYTE_LOOKUP from_ISO_8859_9;
extern const BYTE_LOOKUP from_ISO_8859_10;
extern const BYTE_LOOKUP from_ISO_8859_11;
extern const BYTE_LOOKUP from_ISO_8859_13;
extern const BYTE_LOOKUP from_ISO_8859_14;
extern const BYTE_LOOKUP from_ISO_8859_15;
extern const BYTE_LOOKUP to_ISO_8859_1;
extern const BYTE_LOOKUP to_ISO_8859_2;
extern const BYTE_LOOKUP to_ISO_8859_3;
extern const BYTE_LOOKUP to_ISO_8859_4;
extern const BYTE_LOOKUP to_ISO_8859_5;
extern const BYTE_LOOKUP to_ISO_8859_6;
extern const BYTE_LOOKUP to_ISO_8859_7;
extern const BYTE_LOOKUP to_ISO_8859_8;
extern const BYTE_LOOKUP to_ISO_8859_9;
extern const BYTE_LOOKUP to_ISO_8859_10;
extern const BYTE_LOOKUP to_ISO_8859_11;
extern const BYTE_LOOKUP to_ISO_8859_13;
extern const BYTE_LOOKUP to_ISO_8859_14;
extern const BYTE_LOOKUP to_ISO_8859_15;
/* declarations probably need to go into separate header file, e.g. transcode.h */
/* static structure, one per supported encoding pair */
typedef struct {
const char *from_encoding;
const char *to_encoding;
const BYTE_LOOKUP *conv_tree_start;
int max_output;
int from_utf8;
} transcoder;
/* todo: dynamic structure, one per conversion (stream) */
/* in the future, add some mechanism for dynamically adding stuff here */
#define MAX_TRANSCODERS 29 /* todo: fix: this number has to be adjusted by hand */
static transcoder transcoder_table[MAX_TRANSCODERS];
/* not sure why it's not possible to do relocatable initializations */
/* maybe the code here can be removed (changed to simple initialization) */
/* if we move this to another file???? */
static void
register_transcoder(const char *from_e, const char *to_e,
const BYTE_LOOKUP *tree_start, int max_output, int from_utf8)
{
static int n = 0;
if (n >= MAX_TRANSCODERS) {
/* we are initializing, is it okay to use rb_raise here? */
rb_raise(rb_eRuntimeError /*change exception*/, "not enough transcoder slots");
}
transcoder_table[n].from_encoding = from_e;
transcoder_table[n].to_encoding = to_e;
transcoder_table[n].conv_tree_start = tree_start;
transcoder_table[n].max_output = max_output;
transcoder_table[n].from_utf8 = from_utf8;
n++;
}
static void
init_transcoder_table(void)
{
register_transcoder("ISO-8859-1", "UTF-8", &from_ISO_8859_1, 2, 0);
register_transcoder("ISO-8859-2", "UTF-8", &from_ISO_8859_2, 2, 0);
register_transcoder("ISO-8859-3", "UTF-8", &from_ISO_8859_3, 2, 0);
register_transcoder("ISO-8859-4", "UTF-8", &from_ISO_8859_4, 2, 0);
register_transcoder("ISO-8859-5", "UTF-8", &from_ISO_8859_5, 3, 0);
register_transcoder("ISO-8859-6", "UTF-8", &from_ISO_8859_6, 2, 0);
register_transcoder("ISO-8859-7", "UTF-8", &from_ISO_8859_7, 3, 0);
register_transcoder("ISO-8859-8", "UTF-8", &from_ISO_8859_8, 3, 0);
register_transcoder("ISO-8859-9", "UTF-8", &from_ISO_8859_9, 2, 0);
register_transcoder("ISO-8859-10", "UTF-8", &from_ISO_8859_10, 3, 0);
register_transcoder("ISO-8859-11", "UTF-8", &from_ISO_8859_11, 3, 0);
register_transcoder("ISO-8859-13", "UTF-8", &from_ISO_8859_13, 3, 0);
register_transcoder("ISO-8859-14", "UTF-8", &from_ISO_8859_14, 3, 0);
register_transcoder("ISO-8859-15", "UTF-8", &from_ISO_8859_15, 3, 0);
register_transcoder("UTF-8", "ISO-8859-1", &to_ISO_8859_1, 1, 1);
register_transcoder("UTF-8", "ISO-8859-2", &to_ISO_8859_2, 1, 1);
register_transcoder("UTF-8", "ISO-8859-3", &to_ISO_8859_3, 1, 1);
register_transcoder("UTF-8", "ISO-8859-4", &to_ISO_8859_4, 1, 1);
register_transcoder("UTF-8", "ISO-8859-5", &to_ISO_8859_5, 1, 1);
register_transcoder("UTF-8", "ISO-8859-6", &to_ISO_8859_6, 1, 1);
register_transcoder("UTF-8", "ISO-8859-7", &to_ISO_8859_7, 1, 1);
register_transcoder("UTF-8", "ISO-8859-8", &to_ISO_8859_8, 1, 1);
register_transcoder("UTF-8", "ISO-8859-9", &to_ISO_8859_9, 1, 1);
register_transcoder("UTF-8", "ISO-8859-10", &to_ISO_8859_10, 1, 1);
register_transcoder("UTF-8", "ISO-8859-11", &to_ISO_8859_11, 1, 1);
register_transcoder("UTF-8", "ISO-8859-13", &to_ISO_8859_13, 1, 1);
register_transcoder("UTF-8", "ISO-8859-14", &to_ISO_8859_14, 1, 1);
register_transcoder("UTF-8", "ISO-8859-15", &to_ISO_8859_15, 1, 1);
register_transcoder(NULL, NULL, NULL, 0, 0);
}
static int
encoding_equal(const char* encoding1, const char* encoding2)
{
return 0==strcasecmp(encoding1, encoding2);
}
static transcoder*
transcode_dispatch(const char* from_encoding, const char* to_encoding)
{
transcoder *candidate = transcoder_table;
for (candidate = transcoder_table; candidate->from_encoding; candidate++) {
if (encoding_equal(from_encoding, candidate->from_encoding)
&& encoding_equal(to_encoding, candidate->to_encoding)) {
return candidate;
}
}
/* multistep logic, via UTF-8 */
if (!encoding_equal(from_encoding, "UTF-8")
&& !encoding_equal(to_encoding, "UTF-8")
&& transcode_dispatch("UTF-8", to_encoding)) { /* check that we have a second step */
return transcode_dispatch(from_encoding, "UTF-8"); /* return first step */
}
return NULL;
}
/* dynamic structure, one per conversion (similar to iconv_t) */
/* may carry conversion state (e.g. for iso-2022-jp) */
typedef struct transcoding {
VALUE ruby_string_dest; /* the String used as the conversion destination,
or NULL if something else is being converted */
char *(*flush_func)(struct transcoding*, int, int);
} transcoding;
/*
* Transcoding engine logic
*/
static void
transcode_loop(char **in_pos, char **out_pos,
char *in_stop, char *out_stop,
transcoder *my_transcoder,
transcoding *my_transcoding)
{
char *in_p = *in_pos, *out_p = *out_pos;
const BYTE_LOOKUP *conv_tree_start = my_transcoder->conv_tree_start;
const BYTE_LOOKUP *next_table;
unsigned int next_offset;
VALUE next_info;
unsigned char next_byte;
int from_utf8 = my_transcoder->from_utf8;
char *out_s = out_stop - my_transcoder->max_output + 1;
while (in_p < in_stop) {
next_table = conv_tree_start;
if (out_p >= out_s) {
int len = (out_p - *out_pos);
int new_len = (len + my_transcoder->max_output) * 2;
*out_pos = (*my_transcoding->flush_func)(my_transcoding, len, new_len);
out_p = *out_pos + len;
out_s = *out_pos + new_len - my_transcoder->max_output;
}
next_byte = (unsigned char)*in_p++;
follow_byte:
next_offset = next_table->base[next_byte];
next_info = (VALUE)next_table->info[next_offset];
switch (next_info & 0x1F) {
case NOMAP:
*out_p++ = next_byte;
continue;
case 0x00: case 0x04: case 0x08: case 0x0C:
case 0x10: case 0x14: case 0x18: case 0x1C:
if (in_p >= in_stop) {
/* todo: deal with the case of backtracking */
/* todo: deal with incomplete input (streaming) */
goto illegal;
}
next_byte = (unsigned char)*in_p++;
if (from_utf8) {
if ((next_byte&0xC0) == 0x80)
next_byte -= 0x80;
else
goto illegal;
}
next_table = next_table->info[next_offset];
goto follow_byte;
/* maybe rewrite the following cases to use fallthrough???? */
case ZERObt: /* drop input */
continue;
case ONEbt:
*out_p++ = getBT1(next_info);
continue;
case TWObt:
*out_p++ = getBT1(next_info);
*out_p++ = getBT2(next_info);
continue;
case FOURbt:
*out_p++ = getBT0(next_info);
case THREEbt: /* fall through */
*out_p++ = getBT1(next_info);
*out_p++ = getBT2(next_info);
*out_p++ = getBT3(next_info);
continue;
case ILLEGAL:
goto illegal;
case UNDEF:
/* todo: add code for alternative behaviors */
rb_raise(rb_eRuntimeError /*@@@change exception*/, "conversion undefined for byte sequence (maybe illegal byte sequence)");
continue;
}
continue;
illegal:
/* deal with illegal byte sequence */
/* todo: add code for alternative behaviors */
rb_raise(rb_eRuntimeError /*change exception*/, "illegal byte sequence");
continue;
}
/* cleanup */
*in_pos = in_p;
*out_pos = out_p;
}
/*
* String-specific code
*/
static char *
str_transcoding_resize(transcoding *my_transcoding, int len, int new_len)
{
VALUE dest_string = my_transcoding->ruby_string_dest;
rb_str_resize(dest_string, new_len);
return RSTRING_PTR(dest_string);
}
static int
str_transcode(int argc, VALUE *argv, VALUE *self)
{
VALUE dest;
VALUE str = *self;
long blen, slen;
char *buf, *bp, *sp, *fromp;
rb_encoding *from_enc, *to_enc;
const char *from_e, *to_e;
int from_encidx, to_encidx;
VALUE from_encval, to_encval;
transcoder *my_transcoder;
transcoding my_transcoding;
int final_encoding = 0;
if (argc<1 || argc>2) {
rb_raise(rb_eArgError, "wrong number of arguments (%d for 2)", argc);
}
if ((to_encidx = rb_to_encoding_index(to_encval = argv[0])) < 0) {
to_enc = 0;
to_e = StringValueCStr(to_encval);
}
else {
to_enc = rb_enc_from_index(to_encidx);
to_e = rb_enc_name(to_enc);
}
if (argc==1) {
from_encidx = rb_enc_get_index(str);
from_enc = rb_enc_from_index(from_encidx);
from_e = rb_enc_name(from_enc);
}
else if ((from_encidx = rb_to_encoding_index(from_encval = argv[1])) < 0) {
from_enc = 0;
from_e = StringValueCStr(from_encval);
}
else {
from_enc = rb_enc_from_index(from_encidx);
from_e = rb_enc_name(from_enc);
}
if (from_enc && from_enc == to_enc) {
return -1;
}
if (from_enc && to_enc && rb_enc_asciicompat(from_enc) && rb_enc_asciicompat(to_enc)) {
if (ENC_CODERANGE(str) == ENC_CODERANGE_7BIT) {
return to_encidx;
}
}
if (strcasecmp(from_e, to_e) == 0) {
return -1;
}
while (!final_encoding) /* loop for multistep transcoding */
{ /* later, maybe use smaller intermediate strings for very long strings */
if (!(my_transcoder = transcode_dispatch(from_e, to_e))) {
rb_raise(rb_eArgError, "transcoding not supported (from %s to %s)", from_e, to_e);
}
fromp = sp = RSTRING_PTR(str);
slen = RSTRING_LEN(str);
blen = slen + 30; /* len + margin */
dest = rb_str_tmp_new(blen);
bp = RSTRING_PTR(dest);
my_transcoding.ruby_string_dest = dest;
my_transcoding.flush_func = str_transcoding_resize;
transcode_loop(&fromp, &bp, (sp+slen), (bp+blen), my_transcoder, &my_transcoding);
if (fromp != sp+slen) {
rb_raise(rb_eArgError, "not fully converted, %d bytes left", sp+slen-fromp);
}
buf = RSTRING_PTR(dest);
*bp = '\0';
rb_str_set_len(dest, bp - buf);
if (encoding_equal(my_transcoder->to_encoding, to_e)) {
final_encoding = 1;
}
else {
from_e = my_transcoder->to_encoding;
str = dest;
}
}
/* set encoding */
if (!to_enc) {
to_encidx = rb_enc_replicate(to_e, rb_default_encoding());
}
*self = dest;
return to_encidx;
}
/*
* call-seq:
* str.encode!(encoding) => str
* str.encode!(to_encoding, from_encoding) => str
*
* With one argument, transcodes the contents of <i>str</i> from
* str.encoding to +encoding+.
* With two arguments, transcodes the contents of <i>str</i> from
* from_encoding to to_encoding.
* Returns the string even if no changes were made.
*/
static VALUE
rb_str_transcode_bang(int argc, VALUE *argv, VALUE str)
{
VALUE newstr = str;
int encidx = str_transcode(argc, argv, &newstr);
if (encidx < 0) return str;
rb_str_shared_replace(str, newstr);
rb_enc_associate_index(str, encidx);
return str;
}
/*
* call-seq:
* str.encode(encoding) => str
*
* With one argument, returns a copy of <i>str</i> transcoded
* to encoding +encoding+.
* With two arguments, returns a copy of <i>str</i> transcoded
* from from_encoding to to_encoding.
*/
static VALUE
rb_str_transcode(int argc, VALUE *argv, VALUE str)
{
VALUE newstr = str;
int encidx = str_transcode(argc, argv, &newstr);
if (newstr == str) {
newstr = rb_str_new3(str);
if (encidx >= 0) rb_enc_associate_index(newstr, encidx);
}
else {
RBASIC(newstr)->klass = rb_obj_class(str);
OBJ_INFECT(newstr, str);
rb_enc_associate_index(newstr, encidx);
}
return newstr;
}
void
Init_transcode(void)
{
init_transcoder_table();
rb_define_method(rb_cString, "encode", rb_str_transcode, -1);
rb_define_method(rb_cString, "encode!", rb_str_transcode_bang, -1);
}