On Wednesday 04 February 2009 15:48:41 A B wrote:
> From the manual I read that timestamps are stored as double but they
> can also be stored as 8 byte integers.
The advantage of the integer storage is mainly that calculations and
comparisons have a predictable error and don't suffer from some of the funny
business that comes with floating-point calculations. Performance is
probably similar. Integer storage is now the default (and has been in some
distributions for a while), and would already have been the sole world order
if 8-byte integer support were universally available.