"ag20" <ag20@co.merced.ca.us> writes:
> Is there a reason that the expressions:
> Crops.change_e > '10/1/2001'
> '10/1/2001' < Crops.change_e
> when used in a WHERE clause of a query should yield
> a good plan for the first and a bad plan for the second?
I think you have a broken installation. I get this plan either way
on 7.1:
Nested Loop (cost=0.00..4.96 rows=1 width=64)
-> Nested Loop (cost=0.00..3.54 rows=1 width=48)
-> Index Scan using crops_loct on crops (cost=0.00..2.02 rows=1 width=32)
-> Seq Scan on plantunits (cost=0.00..1.17 rows=17 width=16)
-> Seq Scan on commtypes (cost=0.00..1.13 rows=13 width=16)
The wacko numbers and "nan"s in your output look like something is
fairly hosed internally --- disagreement between different files about
a struct layout is my first thought. How did you compile or come by
your executables?
regards, tom lane
