Re: [Rd] A question about the API mkchar()

From: Simon Urbanek <simon.urbanek_at_r-project.org>
Date: Tue, 28 Oct 2008 11:25:57 -0400

On Oct 28, 2008, at 6:26 , Fán Lóng wrote:

> Hi guys,
>

Hey guy :)

> I've got a question about the API mkchar(). I have met some
> difficulty in parsing utf-8 string to mkchar() in R-2.7.0.
>

There is no mkchar() in R. Did you perhaps mean mkChar()?

> I was intending to parse an utf-8 string str_jan (some Japanese
> characters such asふ, whose utf-8 code is E381B5

There is no such "UTF-8" code. I'm not sure if you meant Unicode, but that would be \u3075 (Hiragana hu) for that character. The UTF-8 encoding of that character is a three-byte sequence 0xe3 0x81 0xb5 if that's what you meant.

> ) to R API SEXP
> mkChar(const char *name) , we only need to create the SEXP using the
> string that we parsed.
>
>
>
> Unfortunately, I found when parsing the variable str_jan, R will
> automatically convert the str_jan according to the current locale
> setting,

That is not true - it will be kept as-is regardless of the encoding. Note that mkChar(x) is equivalent to mkCharCE(x, CE_NATIVE); No conversion takes place when the string is created, but you have told R that it is in the native encoding. If that is not true (which is your case probably isn't), all bets are off since you're lying to R ;).

> so only in the English locale could the function work correctly,
> under other locale, such as Japanese or Chinese, the string will be
> convert incorrectly.

That is clearly a nonsense since the encoding has nothing to do with the locale language itself (Japanese, Chinese, ..). We are talking about the encoding (note that both English and Japanese locales can use UTF-8 encoding, but don't have to). I think you'll need to get the concepts right here - for each string you must define the encoding in order to be able to reproduce the unicode sequence that the string represents. At this point it has nothing to do with the language.

> As a matter of fact, those utf-8 code already is Unicode string, and
> don't need to be converted at all.
>
> I also tried to use the SEXP Rf_mkCharCE(const char *, cetype_t);,
> Parsing the CE_UTF8 as the argument of cetype_t, but the result is
> worse. It returned the result as ucs code, an kind of Unicode under
> windows platform.
>

Well, that's exactly what you want, isn't it? The string is correctly flagged as UTF-8 so R is finally able to find out what exactly is represented by that string. However, your locale apparently doesn't support such characters so it cannot be displayed. If you use a locale that supports it, it works just fine, for example if you use local with SJIS encoding R will still know how to convert it from UTF-8 to SJIS *for display*. The actual string is not touched.

Here is a small piece of code that shows you the difference between native encoding and UTF8-strings:

#include <R.h>
#include <Rinternals.h>

SEXP me() {

   const char c[] = { 0xe3, 0x81, 0xb5, 0 };    SEXP a = allocVector(STRSXP, 2);
   PROTECT(a);
   SET_STRING_ELT(a, 0, mkCharCE(c, CE_NATIVE));    SET_STRING_ELT(a, 1, mkCharCE(c, CE_UTF8));
   UNPROTECT(1);
   return a;
}

In a UTF-8 locale it doesn't matter:

ginaz:sandbox$ LANG=ja_JP.UTF-8 R
 > .Call("me")
[1] "ふ" "ふ"

But in any other, let's say SJIS, it does:

ginaz:sandbox$ LANG=ja_JP.SJIS R
 > .Call("me")
[1] "縺オ" "ふ"

Note that the first string is wrong, because we have supplied UTF-8 encoding but the current one is SJIS. The second one is correct since we told R that it's UTF-8 encoded.

Finally, if the character cannot be displayed in the given encoding:

ginaz:sandbox$ LANG=en_US.US-ASCII R
 > .Call("me")
[1] "\343\201\265" "<U+3075>"

The first one is wrong again, since it's not flagged as UTF8, but the second one is exactly as expected - unicode 3075 which is the Hiragana "hu". It doesn't exist in US-ASCII so unicode designation is all you can display.

> All I want to get is just a SEXP object containing the original
> utf-8 string, no matter what locale is set currently. Normally what
> can I do?
>

mkChar(X, CE_UTF8);

Cheers,
Simon



R-devel_at_r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel Received on Tue 28 Oct 2008 - 15:38:01 GMT

Archive maintained by Robert King, hosted by the discipline of statistics at the University of Newcastle, Australia.
Archive generated by hypermail 2.2.0, at Mon 03 Nov 2008 - 21:30:45 GMT.

Mailing list information is available at https://stat.ethz.ch/mailman/listinfo/r-devel. Please read the posting guide before posting to the list.

list of date sections of archive