added a test where the encoded input string has the padding char in the middle; contrary to commons-codec, it doesn't halt and continues translating