Thomas Lockhart <thomas@fourpalms.org> writes:
> It is somewhat complicated by the fact that my code tree is pretty
> massively changed in this area as I implement an int64-based date/time
> storage alternative to the float64 scheme we use now. The alternative
> would be enabled with something like #ifdef HAVE_INT64_TIMESTAMP.
> Benefits would include having a predictable precision behavior for all
> allowed dates and times.
Interesting. But if this is just an #ifdef, I can see some serious
problems coming up the first time someone runs a backend compiled with
one set of timestamp code in a database created with the other. May
I suggest that the timestamp representation be identified in a field
added to pg_control? That's how we deal with other options that
affect database contents ...
regards, tom lane