Why can't it be implemented directly? What I mean is a new data type DECIMALFLOAT and an internal binary representation of the data. It's unnecessarily inconvenient having to call a function when a simple +, -, *, / would suffice. So, instead of "SELECT decimal_add('1.2','3.4')", it would be much more useful having "SELECT CAST(1.2 AS DECIMALFLOAT)+CAST(3.4 AS DECIMALFLOAT)". (Just added the CASTs to make it clear in this example.)