Given:
create table t1 ( c1 timestamp without time zone, c2 int );
explain verbose select to_char ( c1, '<whatever>' ) from t1
group by to_char ( c1, '<whatever>' );
does any one know why the SEQSCAN node has TWO TARGETENTRYs?
Why do we project "c1" as well as "to_char( c1 . )" ?
---
For cases where we compute an aggregate, such as:
select min(c1), to_char(c1,.) from t1 group by to_char( c1,.)
it would make (some kind of) sense to project out "c1".
But can't the planner/optimizer distinguish these cases on the basis that
'min' is an aggregate function and 'to_char' is not?
The planner seems to be computing a minimal target list of columns (no
expressions) first. Guess I'm asking why it isn't replaced.
