[Rd] localeToCharset()

[Simon Urbanek]

Andreas,

The output is very predictable, so this is not about predictability. Note that C.UTF-8 is technically an invalid locale by the semantics rules (see below). Also note that the C locale is "C" - it is not allowed to have any string behind the C (or else is not the C locale) so what you have is NOT a C locale (see POSIX 7.2).

The issue here is that the POSIX standard provides no semantic rules, locale names can be arbitrary, the only defined one is C (and its synonym POSIX). All others are random locales that can do whatever they want. Then later some systems have introduced semantic guidelines such as the <language>_<territory>.<codeset> convention - that that is what localeToCharsets() expected so it can try to guess the charset for that language. Since C.UTF-8 is such an aberration (not in the standard form) localeToCharset() doesn't know about it and returns NA since it can't guess the language.

Long story short, C.UTF-8 breaks all common rules and has been introduced fairly recently to some Linux systems so R doesn't not know about it yet. Ivan's patch fixes that. That aside, locale names have no official provision to provide the charset, so all you get is a guess assuming the system follows the common rules.

Cheers,
Simon



> On Feb 1, 2022, at 00:38, Blätte, Andreas <andreas.blaette using uni-due.de> wrote:
> 
> Dear Ivan,
> 
> this is a very helpful explanation!  I think it is important to make output of localeToCharset() more predictable. My problem is essentially not to set the locale such that things will work after all. I think the problem is that you see unexpected results.  I guess I owe a suggestion how to improve the code, but your suggestion looks like a very good starting point. 
> 
> Andreas 
> 
> Am 31.01.22, 12:32 schrieb "Ivan Krylov" <krylov.r00t using gmail.com>:
> 
>    On Mon, 31 Jan 2022 09:56:27 +0000
>    "Blätte, Andreas" <andreas.blaette using uni-due.de> wrote:
> 
>> After starting R with a re-defined locale (`env LC_CTYPE=en_US.UTF-8
>> R`,  the output of `localeToCharset()` is:
>> [1] "UTF-8"     "ISO8859-1"
> 
>> why ISO8859-1 might be a fallback option here?
> 
>    ISO8859-1 seems to be offered because it covers the alphabet of
>    American English. Obviously, this doesn't guarantee that the guess is
>    correct. For example, I could symlink the ru_RU.KOI8-R locale on my
>    system to name it "ru_RU", and localeToCharset() would return
>    "ISO8859-5", not knowing the correct answer. їЯавЯг, anyone?
> 
>> Part of my analysis of the code of `localeToCharset()` is that it
>> targets special scenarios on Windows and macOS, but not on Linux.
> 
>    Well, it almost does the right thing. GNU/Linux locales are typically
>    named like <language>_<country>.<encoding>, and localeToCharset()
>    respects the <encoding> part, but only if the language and the country
>    are specified. A quick fix for that would be to add one final case:
> 
>    Index: src/library/utils/R/iconv.R
>    ===================================================================
>    --- src/library/utils/R/iconv.R (revision 81596)
>    +++ src/library/utils/R/iconv.R (working copy)
>    @@ -135,6 +135,7 @@
>                 if(enc == "utf8") return(c("UTF-8", guess(ll)))
>                 else return(guess(ll))
>             }
>    +        if (enc == "utf8") return("UTF-8") # fallback for ???.UTF-8
>             return(NA_character_)
>         }
>     }
> 
>    (Non-UTF-8 encodings on POSIX are handled above, in the if(nzchar(enc)
>    && enc != "utf8") branch.)
> 
>    Maybe a better fix would be to restructure the code a bit, to always
>    take the encoding hint and then also try to guess if the locale looks
>    like it provides a language code.
> 
>    -- 
>    Best regards,
>    Ivan
> 
> ______________________________________________
> R-devel using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel