Hi,

At Fri, 9 Nov 2007 16:08:02 +0900,
David Flanagan wrote in [ruby-core:13326]:
> >> Q1) In step 1 above, should the default primary encoding come from the 
> >> locale environment variables (LC_ALL, LC_CTYPE, and LANG) instead of 
> >> defaulting to ASCII?
> > 
> > It's planned, but we have no mappings from locale name to
> > encoding name.  Attached is a quick hack I tried the week
> > before last.
> 
> Did you consider nl_langinfo(CODESET)?

It needs setlocale() to be called before it, which sets global
state.

> Your code looks good to me, 
> except that you don't check LC_CTYPE.  Is it your intent to be 
> conservative and assume ASCII unless the locale explicitly specifies an 
> encoding name following a .?  You're not going to choose either EUC-JP 
> or SJIS as the default for Japanese locales?

Forgotten, thank you.  I changed locale_encoding() as
following.


static rb_encoding *
locale_encoding(void)
{
    static const char *const langs[] = {"LC_ALL", "LC_CTYPE", "LANG",};
    const char *lang, *at;
    int i, len, idx = 0;
    char buf[32];
    rb_encoding *enc;

    for (i = 0; i < sizeof(langs) / sizeof(langs[0]); ++i) {
	if (!(lang = getenv(langs[i]))) continue;
	if (!(lang = strchr(lang, '.'))) continue;
	at = strchr(++lang, '@');
	if ((len = (at ? at - lang : strlen(lang))) >= sizeof(buf) - 1) continue;
	MEMCPY(buf, lang, char, len);
	buf[len] = 0;
	idx = rb_enc_find_index(buf);
	if (idx < 0 && len > 3 &&
	    (strncasecmp(buf, "euc", 3) == 0 ||
	     strncasecmp(buf, "utf", 3) == 0) &&
	    buf[3]) {
	    MEMMOVE(buf + 4, buf + 3, char, len - 2);
	    buf[3] = '-';
	    idx = rb_enc_find_index(buf);
	}
	enc = rb_enc_from_index(idx);
	if (enc) return enc;
    }
    return rb_enc_default();
}


-- 
Nobu Nakada