Bug 194321

dlerror (and probably strerror) will not convert the messages from their
original encoding to the current locale's encoding.

This causes EUC-JP messages to appear when in ja_JP.UTF-8 encoding for example.
2 ways to work around this problem:
- say that UTF-8 is the default encoding on RHEL4, and convert all the .po files
from their original encodings to UTF-8 (fairly straight forward, I already
tested this with ja_JP)
- fallback to C locale when the conversion fails, rather than passing non-UTF-8
encoded text back to the program

Test case:
#include <stdio.h>
#include <string.h>
#include <locale.h>
main()
{
        char *msg;
        setlocale(LC_ALL, "");
        msg = strerror(1);/*EPERM*/
        puts(msg);
}

And the gdb script to replicate the memory allocation error:
set height 0
set environment LANG=ja_JP.UTF-8
echo --- Success case ---\n
echo Japanese error message is readable in UTF-8 environment\n
run
break malloc.c:3594
command
return 0
cont
end
echo --- Failure case ---\n
echo Japanese error message is not readable in UTF-8 environment\n
run
quit