Before the patch, transliteration was only happening when the
encoding of the string was utf-8. For other encodings only a
simpler conversion (iconv) to ascii was done. For some reason
iconv() own transliteration abilities are not consistent
between systems (depends of libraries installed, locales and
other bits).
So now we always convert the string to utf-8, in order to
transliterate it. And finally, also perform an iconv to
cover some characters that transliterate doesn't handle ok.
Also, remove a block of code that does nothing (previously
it was executing some code, but now it just sets and restores
the error level for nothing).
2def80568b