Hans-Jürgen Schönig <postgres@cybertec.at> writes:
> I came across a quite interesting issue I don't really understand but
> maybe Tom will know.
Interesting. We seem to recognize the fact that the extra clause is
redundant in nearly all places ... but not in indexscan plan generation.
I tried this simplified test case:
create table t_wert(werttypid int);
create table t_werttyp(id int);
create index idx_wert_werttypid on t_wert(werttypid);
explain select * from
t_wert JOIN t_werttyp ON (t_werttyp.id = t_wert.werttypid)
where t_werttyp.id = t_wert.werttypid;
explain select * from
t_wert JOIN t_werttyp ON (t_werttyp.id = t_wert.werttypid);
I got identical merge-join plans and row count estimates both ways.
I then turned off enable_mergejoin, and got identical hash-join plans
and row counts. But with enable_hashjoin also off:
regression=# explain select * from
regression-# t_wert JOIN t_werttyp ON (t_werttyp.id = t_wert.werttypid)
regression-# where t_werttyp.id = t_wert.werttypid; QUERY PLAN
-------------------------------------------------------------------------------------------Nested Loop
(cost=0.00..4858.02rows=5000 width=8) -> Seq Scan on t_werttyp (cost=0.00..20.00 rows=1000 width=4) -> Index Scan
usingidx_wert_werttypid on t_wert (cost=0.00..4.83 rows=1 width=4) Index Cond: (("outer".id = t_wert.werttypid)
AND("outer".id = t_wert.werttypid))
(4 rows)
regression=# explain select * from
regression-# t_wert JOIN t_werttyp ON (t_werttyp.id = t_wert.werttypid); QUERY
PLAN
----------------------------------------------------------------------------------------Nested Loop
(cost=0.00..17150.00rows=5000 width=8) -> Seq Scan on t_werttyp (cost=0.00..20.00 rows=1000 width=4) -> Index Scan
usingidx_wert_werttypid on t_wert (cost=0.00..17.07 rows=5 width=4) Index Cond: ("outer".id =
t_wert.werttypid)
(4 rows)
Looks like a bug is lurking someplace ...
regards, tom lane
