Major issue with predicate constant decoding that causes incorrect results from many queries

[carlaustin]

The base64 encoded constant (128 in this case) used when doing something like WHERE field > 128, when decoded, is read into a string of which the bytes are retrieved as per the code below. This unfortunately causes the constant to read as the wrong number in a number of cases.

PrimitiveComparisonFilter:

        String b64Const = options.get(CONST_VAL);
        String constStr = new String(Base64.decodeBase64(b64Const.getBytes()));
        byte [] constant = constStr.getBytes();

For example, doing > 128, 128 is encoded correctly but then when decoding in this manner the number is being read as 239 and thus obviously the statement doesn't return rows with a value between 128 and 239, a serious issue. The fix is to change the code to:

        String b64Const = options.get(CONST_VAL);
        byte [] constant = Base64.decodeBase64(b64Const.getBytes());

as there is no need to wrap the byte array in a string just to get it back again (which causes the issue).

[bfemiano]

Thank you for this. I will integrate back into trunk.

[joshelser]

Just fixed this in the code that's destined to make it into Hive