In my last article, I slightly screwed up.

A signed 32 bit fixed point number, with two places of precision (less than you need when calculating things like interest and what have you, but lets be generous) has a maximum representation of even less than I off the cuffed -- a mere 21,474,836.48. This is hardly sufficient for accounting. However, floating point is even less useful.

.pm

Back in the mid-80's, I worked for several years at Irving Trust, a (now-gone) major money center bank. One of the financial messaging systems I worked with stored currency amounts as 96-bit vectors of a base unit (eg, a penny), and could have a 'binary point' anywhere in the vector. There were the usual math functions available to handle this data type. If you split the vector evenly between fractional and non-fractional parts, you could represent amounts up to $7E13 to an accuracy of about 3E-15 of a cent. The maximal amount that could be represented was about $2E28, and the highest precision about $1E-29 of a cent. This range and level of precision was judged adequate of most purposes :-). Peter Trei ptrei@acm.org "Did you know that there is a subunit of the Japanese yen?"