Re: Locale and strerror in libgcj's native code...

[Bryce McKinlay <mckinlay at redhat dot com>]

David Daney wrote:

> In several places within the native code in libgcj, we have constructs
> like this:
>
> throw new IOException (JvNewStringUTF (strerror (errno)));
>
> or
>
> throw new IOException (JvNewStringLatin1 (strerror (errno)));
>
> According to my reading of the man pages, strerror returns a string
> according to the LC_MESSAGES encoding.
>
> This means that the actual encoding is not known until runtime.  So at
> compile time it is impossible to know which JvNewString* function to
> use.  It seems to me that using a mix (as we currently do) is clearly
> wrong and only works in with ASCII encoding.
>
> Perhaps we should have a JvNewStringLocale(const char *) that assumes
> the encoding is as specified by LC_MESSAGES.
>
> For the time being, I am using JvNewStringUTF as that seems to be what
> my Fedora Core 1 system defaults to.
>  
Most POSIX OS's seem to be standardizing on UTF8, so I think its fine to 
assume UTF8 in these places. We can always change this later if it turns 
out to be a problem.

Regards

Bryce