Re: Execution-time-sensitive timestamp regression tests

Gregory Stark <stark(at)enterprisedb(dot)com> writes:
> "Tom Lane" <tgl(at)sss(dot)pgh(dot)pa(dot)us> writes:
>> Objections, better ideas?

> Alternatively we could do a pg_sleep(.1) to sleep for 100ms. It sounds like
> the ideal would be something like:

> insert 'now'
> pg_sleep(.1)
> begin
> insert 'now'
> select * from table -- expect 1 (not 0 or 2)
> ... the other tests you mention which get bitten by midnight
> end

Roger, will do.

There actually still is a small risk in the sequence: if the BEGIN
block starts *exactly* at midnight, to within the resolution of
gettimeofday(), then 'now' and 'today' will yield the same value so
the expected row counts will not be matched. This seems like an
acceptably small probability to me.

regards, tom lane