PG> 1800000000000000001::numeric / (10^18)::numeric truncated, -- PG> truncates to 16 digits after decimal point
It has to truncate it _somewhere_, because otherwise 1::numeric / 3 would be infinitely long...
The choice that numeric division actually makes is to produce a result with at least 16 significant figures, but also with at least as many decimal places as the display scale of either of the inputs.
-- Andrew (irc:RhodiumToad)
It looks like precision does matters.
man: "...up to the implementation limit on precision".
Why we don't want to decrease scale, but we decrease precision without any problems?