Skip to content

HTTPS clone URL

Subversion checkout URL

You can clone with
or
.
Download ZIP
tag: v1_9_0_4
Fetching contributors…

Cannot retrieve contributors at this time

1241 lines (1097 sloc) 27.93 kb
/**********************************************************************
encoding.c -
$Author$
created at: Thu May 24 17:23:27 JST 2007
Copyright (C) 2007 Yukihiro Matsumoto
**********************************************************************/
#include "ruby/ruby.h"
#include "ruby/encoding.h"
#include "regenc.h"
#include <ctype.h>
#ifdef HAVE_LANGINFO_H
#include <langinfo.h>
#endif
#include "ruby/util.h"
static ID id_encoding, id_base_encoding;
VALUE rb_cEncoding;
static VALUE rb_encoding_list;
struct rb_encoding_entry {
const char *name;
rb_encoding *enc;
rb_encoding *base;
};
static struct {
struct rb_encoding_entry *list;
int count;
int size;
st_table *names;
} enc_table;
void rb_enc_init(void);
#define ENCODING_COUNT ENCINDEX_BUILTIN_MAX
#define enc_autoload_p(enc) (!rb_enc_mbmaxlen(enc))
static int load_encoding(const char *name);
static VALUE enc_base_encoding(VALUE self);
static void
enc_mark(void *ptr)
{
}
static VALUE
enc_new(rb_encoding *encoding)
{
return Data_Wrap_Struct(rb_cEncoding, enc_mark, 0, encoding);
}
VALUE
rb_enc_from_encoding(rb_encoding *encoding)
{
VALUE list, enc;
int idx;
if (!encoding) return Qnil;
idx = ENC_TO_ENCINDEX(encoding);
if (!(list = rb_encoding_list)) {
rb_bug("rb_enc_from_encoding(%d\"%s\"): no rb_encoding_list",
idx, rb_enc_name(encoding));
}
enc = rb_ary_entry(list, idx);
if (NIL_P(enc)) {
rb_bug("rb_enc_from_encoding(%d\"%s\"): not created yet",
idx, rb_enc_name(encoding));
}
return enc;
}
static int enc_autoload(rb_encoding *);
static int
check_encoding(rb_encoding *enc)
{
int index = rb_enc_to_index(enc);
if (rb_enc_from_index(index) != enc)
return -1;
if (enc_autoload_p(enc)) {
index = enc_autoload(enc);
}
return index;
}
static int
enc_check_encoding(VALUE obj)
{
if (SPECIAL_CONST_P(obj) || BUILTIN_TYPE(obj) != T_DATA ||
RDATA(obj)->dmark != enc_mark) {
return -1;
}
return check_encoding(RDATA(obj)->data);
}
static int
must_encoding(VALUE enc)
{
int index = enc_check_encoding(enc);
if (index < 0) {
rb_raise(rb_eTypeError, "wrong argument type %s (expected Encoding)",
rb_obj_classname(enc));
}
return index;
}
int
rb_to_encoding_index(VALUE enc)
{
int idx;
idx = enc_check_encoding(enc);
if (idx >= 0) {
return idx;
}
else if (NIL_P(enc = rb_check_string_type(enc))) {
return -1;
}
if (!rb_enc_asciicompat(rb_enc_get(enc))) {
return -1;
}
return rb_enc_find_index(StringValueCStr(enc));
}
static rb_encoding *
to_encoding(VALUE enc)
{
int idx;
StringValue(enc);
if (!rb_enc_asciicompat(rb_enc_get(enc))) {
rb_raise(rb_eArgError, "invalid name encoding (non ASCII)");
}
idx = rb_enc_find_index(StringValueCStr(enc));
if (idx < 0) {
rb_raise(rb_eArgError, "unknown encoding name - %s", RSTRING_PTR(enc));
}
return rb_enc_from_index(idx);
}
rb_encoding *
rb_to_encoding(VALUE enc)
{
if (enc_check_encoding(enc) >= 0) return RDATA(enc)->data;
return to_encoding(enc);
}
void
rb_gc_mark_encodings(void)
{
}
static int
enc_table_expand(int newsize)
{
struct rb_encoding_entry *ent;
int count = newsize;
if (enc_table.size >= newsize) return newsize;
newsize = (newsize + 7) / 8 * 8;
ent = realloc(enc_table.list, sizeof(*enc_table.list) * newsize);
if (!ent) return -1;
memset(ent + enc_table.size, 0, sizeof(*ent)*(newsize - enc_table.size));
enc_table.list = ent;
enc_table.size = newsize;
return count;
}
static int
enc_register_at(int index, const char *name, rb_encoding *encoding)
{
struct rb_encoding_entry *ent = &enc_table.list[index];
VALUE list;
if (!ent->name) {
ent->name = name = strdup(name);
}
else if (STRCASECMP(name, ent->name)) {
return -1;
}
if (!ent->enc) {
ent->enc = xmalloc(sizeof(rb_encoding));
}
if (encoding) {
*ent->enc = *encoding;
}
else {
memset(ent->enc, 0, sizeof(*ent->enc));
}
encoding = ent->enc;
encoding->name = name;
encoding->ruby_encoding_index = index;
st_insert(enc_table.names, (st_data_t)name, (st_data_t)index);
list = rb_encoding_list;
if (list && NIL_P(rb_ary_entry(list, index))) {
/* initialize encoding data */
rb_ary_store(list, index, enc_new(encoding));
}
return index;
}
static int
enc_register(const char *name, rb_encoding *encoding)
{
int index = enc_table.count;
if ((index = enc_table_expand(index + 1)) < 0) return -1;
enc_table.count = index;
return enc_register_at(index - 1, name, encoding);
}
static void set_encoding_const(const char *, rb_encoding *);
int rb_enc_registered(const char *name);
int
rb_enc_register(const char *name, rb_encoding *encoding)
{
int index = rb_enc_registered(name);
if (index >= 0) {
rb_encoding *oldenc = rb_enc_from_index(index);
if (STRCASECMP(name, rb_enc_name(oldenc))) {
index = enc_register(name, encoding);
}
else if (enc_autoload_p(oldenc) || !ENC_DUMMY_P(oldenc)) {
enc_register_at(index, name, encoding);
}
else {
rb_raise(rb_eArgError, "encoding %s is already registered", name);
}
}
else {
index = enc_register(name, encoding);
set_encoding_const(name, rb_enc_from_index(index));
}
return index;
}
void
rb_encdb_declare(const char *name)
{
int idx = rb_enc_registered(name);
if (idx < 0) {
idx = enc_register(name, 0);
}
set_encoding_const(name, rb_enc_from_index(idx));
}
static void
enc_check_duplication(const char *name)
{
if (rb_enc_registered(name) >= 0) {
rb_raise(rb_eArgError, "encoding %s is already registered", name);
}
}
static rb_encoding*
set_base_encoding(int index, rb_encoding *base)
{
rb_encoding *enc = enc_table.list[index].enc;
enc_table.list[index].base = base;
if (rb_enc_dummy_p(base)) ENC_SET_DUMMY(enc);
return enc;
}
int
rb_enc_replicate(const char *name, rb_encoding *encoding)
{
int idx;
enc_check_duplication(name);
idx = enc_register(name, encoding);
set_base_encoding(idx, encoding);
set_encoding_const(name, rb_enc_from_index(idx));
return idx;
}
static int
enc_replicate(int idx, const char *name, rb_encoding *origenc)
{
if (idx < 0) {
idx = enc_register(name, origenc);
}
else {
idx = enc_register_at(idx, name, origenc);
}
if (idx >= 0) {
set_base_encoding(idx, origenc);
set_encoding_const(name, rb_enc_from_index(idx));
}
return idx;
}
int
rb_encdb_replicate(const char *name, const char *orig)
{
int origidx = rb_enc_registered(orig);
int idx = rb_enc_registered(name);
if (origidx < 0) {
origidx = enc_register(orig, 0);
}
return enc_replicate(idx, name, rb_enc_from_index(origidx));
}
int
rb_define_dummy_encoding(const char *name)
{
int index = rb_enc_replicate(name, rb_ascii8bit_encoding());
rb_encoding *enc = enc_table.list[index].enc;
ENC_SET_DUMMY(enc);
return index;
}
int
rb_encdb_dummy(const char *name)
{
int index = enc_replicate(rb_enc_registered(name), name,
rb_ascii8bit_encoding());
rb_encoding *enc = enc_table.list[index].enc;
ENC_SET_DUMMY(enc);
return index;
}
/*
* call-seq:
* enc.dummy? => true or false
*
* Returns true for dummy encodings.
* A dummy encoding is an encoding for which character handling is not properly
* implemented.
* It is used for stateful encodings.
*
* Encoding::ISO_2022_JP.dummy? #=> true
* Encoding::UTF_8.dummy? #=> false
*
*/
static VALUE
enc_dummy_p(VALUE enc)
{
return ENC_DUMMY_P(enc_table.list[must_encoding(enc)].enc) ? Qtrue : Qfalse;
}
static int
enc_alias(const char *alias, int idx)
{
alias = strdup(alias);
st_insert(enc_table.names, (st_data_t)alias, (st_data_t)idx);
set_encoding_const(alias, rb_enc_from_index(idx));
return idx;
}
int
rb_enc_alias(const char *alias, const char *orig)
{
int idx;
enc_check_duplication(alias);
if (!enc_table.list) {
rb_enc_init();
}
if ((idx = rb_enc_find_index(orig)) < 0) {
return -1;
}
return enc_alias(alias, idx);
}
int
rb_encdb_alias(const char *alias, const char *orig)
{
int idx = rb_enc_registered(orig);
if (idx < 0) {
idx = enc_register(orig, 0);
}
return enc_alias(alias, idx);
}
enum {
ENCINDEX_ASCII,
ENCINDEX_UTF_8,
ENCINDEX_US_ASCII,
ENCINDEX_BUILTIN_MAX
};
extern rb_encoding OnigEncodingUTF_8;
extern rb_encoding OnigEncodingUS_ASCII;
void
rb_enc_init(void)
{
enc_table_expand(ENCODING_COUNT + 1);
if (!enc_table.names) {
enc_table.names = st_init_strcasetable();
}
#define ENC_REGISTER(enc) enc_register_at(ENCINDEX_##enc, rb_enc_name(&OnigEncoding##enc), &OnigEncoding##enc)
ENC_REGISTER(ASCII);
ENC_REGISTER(UTF_8);
ENC_REGISTER(US_ASCII);
#undef ENC_REGISTER
enc_table.count = ENCINDEX_BUILTIN_MAX;
}
rb_encoding *
rb_enc_from_index(int index)
{
if (!enc_table.list) {
rb_enc_init();
}
if (index < 0 || enc_table.count <= index) {
return 0;
}
return enc_table.list[index].enc;
}
int
rb_enc_registered(const char *name)
{
st_data_t idx = 0;
if (!name) return -1;
if (!enc_table.list) return -1;
if (st_lookup(enc_table.names, (st_data_t)name, &idx)) {
return (int)idx;
}
return -1;
}
static VALUE
require_enc(VALUE enclib)
{
return rb_require_safe(enclib, rb_safe_level());
}
static int
load_encoding(const char *name)
{
VALUE enclib = rb_sprintf("enc/%s", name);
VALUE verbose = ruby_verbose;
VALUE debug = ruby_debug;
VALUE loaded;
char *s = RSTRING_PTR(enclib) + 4, *e = RSTRING_END(enclib);
int idx;
while (s < e) {
if (!ISALNUM(*s)) *s = '_';
else if (ISUPPER(*s)) *s = TOLOWER(*s);
++s;
}
OBJ_FREEZE(enclib);
ruby_verbose = Qfalse;
ruby_debug = Qfalse;
loaded = rb_protect(require_enc, enclib, 0);
ruby_verbose = verbose;
ruby_debug = debug;
rb_set_errinfo(Qnil);
if (NIL_P(loaded)) return -1;
if ((idx = rb_enc_registered(name)) < 0) return -1;
if (enc_autoload_p(enc_table.list[idx].enc)) return -1;
return idx;
}
static int
enc_autoload(rb_encoding *enc)
{
int i;
rb_encoding *base = enc_table.list[ENC_TO_ENCINDEX(enc)].base;
if (base) {
i = 0;
do {
if (i >= enc_table.count) return -1;
} while (enc_table.list[i].enc != base && (++i, 1));
if (enc_autoload_p(base)) {
if (enc_autoload(base) < 0) return -1;
}
i = ENC_TO_ENCINDEX(enc);
enc_register_at(i, rb_enc_name(enc), base);
}
else {
i = load_encoding(rb_enc_name(enc));
}
return i;
}
int
rb_enc_find_index(const char *name)
{
int i = rb_enc_registered(name);
rb_encoding *enc;
if (i < 0) {
i = load_encoding(name);
}
else if (enc_autoload_p(enc = rb_enc_from_index(i))) {
if (enc_autoload(enc) < 0) {
rb_warn("failed to load encoding (%s); use ASCII-8BIT instead",
name);
return 0;
}
}
return i;
}
rb_encoding *
rb_enc_find(const char *name)
{
int idx = rb_enc_find_index(name);
if (idx < 0) idx = 0;
return rb_enc_from_index(idx);
}
static inline int
enc_capable(VALUE obj)
{
if (SPECIAL_CONST_P(obj)) return Qfalse;
switch (BUILTIN_TYPE(obj)) {
case T_STRING:
case T_REGEXP:
case T_FILE:
return Qtrue;
case T_DATA:
if (RDATA(obj)->dmark == enc_mark) return Qtrue;
default:
return Qfalse;
}
}
ID
rb_id_encoding(void)
{
CONST_ID(id_encoding, "encoding");
return id_encoding;
}
int
rb_enc_get_index(VALUE obj)
{
int i;
i = ENCODING_GET_INLINED(obj);
if (i == ENCODING_INLINE_MAX) {
VALUE iv;
iv = rb_ivar_get(obj, rb_id_encoding());
i = NUM2INT(iv);
}
return i;
}
void
rb_enc_set_index(VALUE obj, int idx)
{
if (idx < ENCODING_INLINE_MAX) {
ENCODING_SET_INLINED(obj, idx);
return;
}
ENCODING_SET_INLINED(obj, ENCODING_INLINE_MAX);
rb_ivar_set(obj, rb_id_encoding(), INT2NUM(idx));
return;
}
VALUE
rb_enc_associate_index(VALUE obj, int idx)
{
/* enc_check_capable(obj);*/
if (rb_enc_get_index(obj) == idx)
return obj;
if (!ENC_CODERANGE_ASCIIONLY(obj) ||
!rb_enc_asciicompat(rb_enc_from_index(idx))) {
ENC_CODERANGE_CLEAR(obj);
}
rb_enc_set_index(obj, idx);
return obj;
}
VALUE
rb_enc_associate(VALUE obj, rb_encoding *enc)
{
return rb_enc_associate_index(obj, rb_enc_to_index(enc));
}
rb_encoding*
rb_enc_get(VALUE obj)
{
return rb_enc_from_index(rb_enc_get_index(obj));
}
rb_encoding*
rb_enc_check(VALUE str1, VALUE str2)
{
rb_encoding *enc = rb_enc_compatible(str1, str2);
if (!enc)
rb_raise(rb_eEncCompatError, "incompatible character encodings: %s and %s",
rb_enc_name(rb_enc_get(str1)),
rb_enc_name(rb_enc_get(str2)));
return enc;
}
rb_encoding*
rb_enc_compatible(VALUE str1, VALUE str2)
{
int idx1, idx2;
rb_encoding *enc1, *enc2;
idx1 = rb_enc_get_index(str1);
idx2 = rb_enc_get_index(str2);
if (idx1 < 0 || idx2 < 0)
return 0;
if (idx1 == idx2) {
return rb_enc_from_index(idx1);
}
enc1 = rb_enc_from_index(idx1);
enc2 = rb_enc_from_index(idx2);
if (TYPE(str2) == T_STRING && RSTRING_LEN(str2) == 0)
return enc1;
if (TYPE(str1) == T_STRING && RSTRING_LEN(str1) == 0)
return enc2;
if (!rb_enc_asciicompat(enc1) || !rb_enc_asciicompat(enc2)) {
return 0;
}
if (BUILTIN_TYPE(str1) != T_STRING) {
VALUE tmp = str1;
int idx0 = idx1;
str1 = str2;
str2 = tmp;
idx1 = idx2;
idx2 = idx0;
}
if (BUILTIN_TYPE(str1) == T_STRING) {
int cr1, cr2;
cr1 = rb_enc_str_coderange(str1);
if (BUILTIN_TYPE(str2) == T_STRING) {
cr2 = rb_enc_str_coderange(str2);
if (cr1 != cr2) {
/* may need to handle ENC_CODERANGE_BROKEN */
if (cr1 == ENC_CODERANGE_7BIT) return enc2;
if (cr2 == ENC_CODERANGE_7BIT) return enc1;
}
if (cr2 == ENC_CODERANGE_7BIT) {
if (idx1 == 0) return enc2;
return enc1;
}
}
if (cr1 == ENC_CODERANGE_7BIT)
return enc2;
}
return 0;
}
void
rb_enc_copy(VALUE obj1, VALUE obj2)
{
rb_enc_associate_index(obj1, rb_enc_get_index(obj2));
}
/*
* call-seq:
* obj.encoding => encoding
*
* Returns the Encoding object that represents the encoding of obj.
*/
VALUE
rb_obj_encoding(VALUE obj)
{
rb_encoding *enc = rb_enc_get(obj);
if (!enc) {
rb_raise(rb_eTypeError, "unknown encoding");
}
return rb_enc_from_encoding(enc);
}
int
rb_enc_mbclen(const char *p, const char *e, rb_encoding *enc)
{
int n = ONIGENC_PRECISE_MBC_ENC_LEN(enc, (UChar*)p, (UChar*)e);
if (MBCLEN_CHARFOUND_P(n) && MBCLEN_CHARFOUND_LEN(n) <= e-p)
return MBCLEN_CHARFOUND_LEN(n);
else {
int min = rb_enc_mbminlen(enc);
return min <= e-p ? min : e-p;
}
}
int
rb_enc_precise_mbclen(const char *p, const char *e, rb_encoding *enc)
{
int n;
if (e <= p)
return ONIGENC_CONSTRUCT_MBCLEN_NEEDMORE(1);
n = ONIGENC_PRECISE_MBC_ENC_LEN(enc, (UChar*)p, (UChar*)e);
if (e-p < n)
return ONIGENC_CONSTRUCT_MBCLEN_NEEDMORE(n-(e-p));
return n;
}
int
rb_enc_ascget(const char *p, const char *e, int *len, rb_encoding *enc)
{
int c, l;
if (e <= p)
return -1;
if (rb_enc_asciicompat(enc)) {
c = (unsigned char)*p;
if (!ISASCII(c))
return -1;
if (len) *len = 1;
return c;
}
l = rb_enc_precise_mbclen(p, e, enc);
if (!MBCLEN_CHARFOUND_P(l))
return -1;
c = rb_enc_mbc_to_codepoint(p, e, enc);
if (!rb_enc_isascii(c, enc))
return -1;
if (len) *len = l;
return c;
}
int
rb_enc_codepoint(const char *p, const char *e, rb_encoding *enc)
{
int r;
if (e <= p)
rb_raise(rb_eArgError, "empty string");
r = rb_enc_precise_mbclen(p, e, enc);
if (MBCLEN_CHARFOUND_P(r))
return rb_enc_mbc_to_codepoint(p, e, enc);
else
rb_raise(rb_eArgError, "invalid byte sequence in %s", rb_enc_name(enc));
}
int
rb_enc_codelen(int c, rb_encoding *enc)
{
int n = ONIGENC_CODE_TO_MBCLEN(enc,c);
if (n == 0) {
rb_raise(rb_eArgError, "invalid codepoint 0x%x in %s", c, rb_enc_name(enc));
}
return n;
}
int
rb_enc_toupper(int c, rb_encoding *enc)
{
return (ONIGENC_IS_ASCII_CODE(c)?ONIGENC_ASCII_CODE_TO_UPPER_CASE(c):(c));
}
int
rb_enc_tolower(int c, rb_encoding *enc)
{
return (ONIGENC_IS_ASCII_CODE(c)?ONIGENC_ASCII_CODE_TO_LOWER_CASE(c):(c));
}
/*
* call-seq:
* enc.inspect => string
*
* Returns a string which represents the encoding for programmers.
*
* Encoding::UTF_8.inspect #=> "#<Encoding:UTF-8>"
* Encoding::ISO_2022_JP.inspect #=> "#<Encoding:ISO-2022-JP (dummy)>"
*/
static VALUE
enc_inspect(VALUE self)
{
VALUE str = rb_sprintf("#<%s:%s%s>", rb_obj_classname(self),
rb_enc_name((rb_encoding*)DATA_PTR(self)),
(enc_dummy_p(self) ? " (dummy)" : ""));
ENCODING_CODERANGE_SET(str, rb_usascii_encindex(), ENC_CODERANGE_7BIT);
return str;
}
/*
* call-seq:
* enc.name => string
*
* Returns the name of the encoding.
*
* Encoding::UTF_8.name => "UTF-8"
*/
static VALUE
enc_name(VALUE self)
{
return rb_usascii_str_new2(rb_enc_name((rb_encoding*)DATA_PTR(self)));
}
static VALUE
enc_base_encoding(VALUE self)
{
rb_encoding *base = enc_table.list[must_encoding(self)].base;
if (!base) return Qnil;
return ENC_FROM_ENCODING(base);
}
/*
* call-seq:
* Encoding.list => [enc1, enc2, ...]
*
* Returns the list of loaded encodings.
*
* Encoding.list
* => [#<Encoding:ASCII-8BIT>, #<Encoding:UTF-8>,
* #<Encoding:ISO-2022-JP (dummy)>]
*
* Encoding.find("US-ASCII")
* => #<Encoding:US-ASCII>
*
* Encoding.list
* => [#<Encoding:ASCII-8BIT>, #<Encoding:UTF-8>,
* #<Encoding:US-ASCII>, #<Encoding:ISO-2022-JP (dummy)>]
*
*/
static VALUE
enc_list(VALUE klass)
{
VALUE ary = rb_ary_new2(0);
rb_ary_replace(ary, rb_encoding_list);
return ary;
}
/*
* call-seq:
* Encoding.find(string) => enc
* Encoding.find(symbol) => enc
*
* Search the encoding with specified <i>name</i>.
* <i>name</i> should be a string or symbol.
*
* Encoding.find("US-ASCII") => #<Encoding:US-ASCII>
* Encoding.find(:Shift_JIS) => #<Encoding:Shift_JIS>
*
*/
static VALUE
enc_find(VALUE klass, VALUE enc)
{
return rb_enc_from_encoding(to_encoding(enc));
}
/*
* call-seq:
* Encoding.compatible?(str1, str2) => enc or nil
*
* Checks the compatibility of two strings.
* If they are compatible, means concatenatable,
* returns an encoding which the concatinated string will be.
* If they are not compatible, nil is returned.
*
* Encoding.compatible?("\xa1".force_encoding("iso-8859-1"), "b")
* => #<Encoding:ISO-8859-1>
*
* Encoding.compatible?(
* "\xa1".force_encoding("iso-8859-1"),
* "\xa1\xa1".force_encoding("euc-jp"))
* => nil
*
*/
static VALUE
enc_compatible_p(VALUE klass, VALUE str1, VALUE str2)
{
rb_encoding *enc;
if (enc_check_encoding(str1) > 0 || enc_check_encoding(str2) > 0)
rb_raise(rb_eTypeError, "wrong argument type Encoding (expected String)");
if (!enc_capable(str1)) return Qnil;
if (!enc_capable(str2)) return Qnil;
enc = rb_enc_compatible(str1, str2);
if (!enc) return Qnil;
return rb_enc_from_encoding(enc);
}
/* :nodoc: */
static VALUE
enc_dump(int argc, VALUE *argv, VALUE self)
{
rb_scan_args(argc, argv, "01", 0);
return enc_name(self);
}
/* :nodoc: */
static VALUE
enc_load(VALUE klass, VALUE str)
{
return enc_find(klass, str);
}
rb_encoding *
rb_ascii8bit_encoding(void)
{
if (!enc_table.list) {
rb_enc_init();
}
return enc_table.list[ENCINDEX_ASCII].enc;
}
int
rb_ascii8bit_encindex(void)
{
return ENCINDEX_ASCII;
}
rb_encoding *
rb_utf8_encoding(void)
{
if (!enc_table.list) {
rb_enc_init();
}
return enc_table.list[ENCINDEX_UTF_8].enc;
}
int
rb_utf8_encindex(void)
{
return ENCINDEX_UTF_8;
}
rb_encoding *
rb_usascii_encoding(void)
{
if (!enc_table.list) {
rb_enc_init();
}
return enc_table.list[ENCINDEX_US_ASCII].enc;
}
int
rb_usascii_encindex(void)
{
return ENCINDEX_US_ASCII;
}
rb_encoding *
rb_locale_encoding(void)
{
VALUE charmap = rb_locale_charmap(rb_cEncoding);
int idx;
if (NIL_P(charmap))
return rb_usascii_encoding();
else
idx = rb_enc_find_index(StringValueCStr(charmap));
if (idx < 0)
return rb_ascii8bit_encoding();
return rb_enc_from_index(idx);
}
rb_encoding *
rb_filesystem_encoding(void)
{
rb_encoding *enc;
#if defined _WIN32
enc = rb_locale_encoding();
#elif defined __APPLE__
enc = rb_enc_find("UTF8-MAC");
#else
enc = rb_default_external_encoding();
#endif
return enc;
}
static int default_external_index;
rb_encoding *
rb_default_external_encoding(void)
{
return rb_enc_from_index(default_external_index);
}
VALUE
rb_enc_default_external(void)
{
return rb_enc_from_encoding(rb_default_external_encoding());
}
/*
* call-seq:
* Encoding.default_external => enc
*
* Returns default external encoding.
*
* It is initialized by the locale or -E option.
*/
static VALUE
get_default_external(VALUE klass)
{
return rb_enc_default_external();
}
void
rb_enc_set_default_external(VALUE encoding)
{
default_external_index = rb_enc_to_index(rb_to_encoding(encoding));
}
/*
* call-seq:
* Encoding.locale_charmap => string
*
* Returns the locale charmap name.
*
* Debian GNU/Linux
* LANG=C
* Encoding.locale_charmap => "ANSI_X3.4-1968"
* LANG=ja_JP.EUC-JP
* Encoding.locale_charmap => "EUC-JP"
*
* SunOS 5
* LANG=C
* Encoding.locale_charmap => "646"
* LANG=ja
* Encoding.locale_charmap => "eucJP"
*
*/
VALUE
rb_locale_charmap(VALUE klass)
{
#if defined NO_LOCALE_CHARMAP
return rb_usascii_str_new2("ASCII-8BIT");
#elif defined HAVE_LANGINFO_H
char *codeset;
codeset = nl_langinfo(CODESET);
return rb_usascii_str_new2(codeset);
#elif defined _WIN32
return rb_sprintf("CP%d", GetACP());
#else
return Qnil;
#endif
}
static void
set_encoding_const(const char *name, rb_encoding *enc)
{
VALUE encoding = rb_enc_from_encoding(enc);
char *s = (char *)name;
int haslower = 0, hasupper = 0, valid = 0;
if (ISDIGIT(*s)) return;
if (ISUPPER(*s)) {
hasupper = 1;
while (*++s && (ISALNUM(*s) || *s == '_')) {
if (ISLOWER(*s)) haslower = 1;
}
}
if (!*s) {
valid = 1;
rb_define_const(rb_cEncoding, name, encoding);
}
if (!valid || haslower) {
int len = strlen(name) + 1;
if (!haslower || !hasupper) {
do {
if (ISLOWER(*s)) haslower = 1;
if (ISUPPER(*s)) hasupper = 1;
} while (*++s && (!haslower || !hasupper));
}
MEMCPY(s = ALLOCA_N(char, len), name, char, len);
name = s;
if (!valid) {
if (ISLOWER(*s)) *s = ONIGENC_ASCII_CODE_TO_UPPER_CASE((int)*s);
for (; *s; ++s) {
if (!ISALNUM(*s)) *s = '_';
}
if (hasupper) {
rb_define_const(rb_cEncoding, name, encoding);
}
}
if (haslower) {
for (s = (char *)name; *s; ++s) {
if (ISLOWER(*s)) *s = ONIGENC_ASCII_CODE_TO_UPPER_CASE((int)*s);
}
rb_define_const(rb_cEncoding, name, encoding);
}
}
}
static int
rb_enc_name_list_i(st_data_t name, st_data_t idx, st_data_t arg)
{
VALUE ary = (VALUE)arg;
VALUE str = rb_usascii_str_new2((char *)name);
OBJ_FREEZE(str);
rb_ary_push(ary, str);
return ST_CONTINUE;
}
/*
* call-seq:
* Encoding.name_list => ["enc1", "enc2", ...]
*
* Returns the list of available encoding names.
*
* Encoding.name_list
* => ["US-ASCII", "ASCII-8BIT", "UTF-8",
* "ISO-8859-1", "Shift_JIS", "EUC-JP",
* "Windows-31J",
* "BINARY", "CP932", "eucJP"]
*
* This list doesn't include dummy encodings.
*
*/
static VALUE
rb_enc_name_list(VALUE klass)
{
VALUE ary = rb_ary_new2(enc_table.names->num_entries);
st_foreach(enc_table.names, rb_enc_name_list_i, (st_data_t)ary);
return ary;
}
static int
rb_enc_aliases_enc_i(st_data_t name, st_data_t orig, st_data_t arg)
{
VALUE *p = (VALUE *)arg;
VALUE aliases = p[0], ary = p[1];
int idx = (int)orig;
VALUE key, str = rb_ary_entry(ary, idx);
if (NIL_P(str)) {
rb_encoding *enc = rb_enc_from_index(idx);
if (STRCASECMP((char*)name, rb_enc_name(enc)) == 0) {
return ST_CONTINUE;
}
str = rb_usascii_str_new2(rb_enc_name(enc));
OBJ_FREEZE(str);
rb_ary_store(ary, idx, str);
}
key = rb_usascii_str_new2((char *)name);
OBJ_FREEZE(key);
rb_hash_aset(aliases, key, str);
return ST_CONTINUE;
}
/*
* call-seq:
* Encoding.aliases => {"alias1" => "orig1", "alias2" => "orig2", ...}
*
* Returns the hash of available encoding alias and original encoding name.
*
* Encoding.aliases
* => {"BINARY"=>"ASCII-8BIT", "ASCII"=>"US-ASCII", "ANSI_X3.4-1986"=>"US-ASCII",
* "SJIS"=>"Shift_JIS", "eucJP"=>"EUC-JP", "CP932"=>"Windows-31J"}
*
*/
static VALUE
rb_enc_aliases(VALUE klass)
{
VALUE aliases[2];
aliases[0] = rb_hash_new();
aliases[1] = rb_ary_new();
st_foreach(enc_table.names, rb_enc_aliases_enc_i, (st_data_t)aliases);
return aliases[0];
}
void
Init_Encoding(void)
{
#undef rb_intern
#define rb_intern(str) rb_intern_const(str)
VALUE list;
int i;
id_base_encoding = rb_intern("#base_encoding");
rb_cEncoding = rb_define_class("Encoding", rb_cObject);
rb_undef_alloc_func(rb_cEncoding);
rb_define_method(rb_cEncoding, "to_s", enc_name, 0);
rb_define_method(rb_cEncoding, "inspect", enc_inspect, 0);
rb_define_method(rb_cEncoding, "name", enc_name, 0);
rb_define_method(rb_cEncoding, "base_encoding", enc_base_encoding, 0);
rb_define_method(rb_cEncoding, "dummy?", enc_dummy_p, 0);
rb_define_singleton_method(rb_cEncoding, "list", enc_list, 0);
rb_define_singleton_method(rb_cEncoding, "name_list", rb_enc_name_list, 0);
rb_define_singleton_method(rb_cEncoding, "aliases", rb_enc_aliases, 0);
rb_define_singleton_method(rb_cEncoding, "find", enc_find, 1);
rb_define_singleton_method(rb_cEncoding, "compatible?", enc_compatible_p, 2);
rb_define_method(rb_cEncoding, "_dump", enc_dump, -1);
rb_define_singleton_method(rb_cEncoding, "_load", enc_load, 1);
rb_define_singleton_method(rb_cEncoding, "default_external", get_default_external, 0);
rb_define_singleton_method(rb_cEncoding, "locale_charmap", rb_locale_charmap, 0);
rb_gc_register_address(&rb_encoding_list);
list = rb_ary_new2(enc_table.count);
RBASIC(list)->klass = 0;
rb_encoding_list = list;
for (i = 0; i < enc_table.count; ++i) {
rb_ary_push(list, enc_new(enc_table.list[i].enc));
}
}
/* locale insensitive functions */
#define ctype_test(c, ctype) \
(rb_isascii(c) && ONIGENC_IS_ASCII_CODE_CTYPE((c), ctype))
int rb_isalnum(int c) { return ctype_test(c, ONIGENC_CTYPE_ALNUM); }
int rb_isalpha(int c) { return ctype_test(c, ONIGENC_CTYPE_ALPHA); }
int rb_isblank(int c) { return ctype_test(c, ONIGENC_CTYPE_BLANK); }
int rb_iscntrl(int c) { return ctype_test(c, ONIGENC_CTYPE_CNTRL); }
int rb_isdigit(int c) { return ctype_test(c, ONIGENC_CTYPE_DIGIT); }
int rb_isgraph(int c) { return ctype_test(c, ONIGENC_CTYPE_GRAPH); }
int rb_islower(int c) { return ctype_test(c, ONIGENC_CTYPE_LOWER); }
int rb_isprint(int c) { return ctype_test(c, ONIGENC_CTYPE_PRINT); }
int rb_ispunct(int c) { return ctype_test(c, ONIGENC_CTYPE_PUNCT); }
int rb_isspace(int c) { return ctype_test(c, ONIGENC_CTYPE_SPACE); }
int rb_isupper(int c) { return ctype_test(c, ONIGENC_CTYPE_UPPER); }
int rb_isxdigit(int c) { return ctype_test(c, ONIGENC_CTYPE_XDIGIT); }
int
rb_tolower(int c)
{
return rb_isascii(c) ? ONIGENC_ASCII_CODE_TO_LOWER_CASE(c) : c;
}
int
rb_toupper(int c)
{
return rb_isascii(c) ? ONIGENC_ASCII_CODE_TO_UPPER_CASE(c) : c;
}
Jump to Line
Something went wrong with that request. Please try again.