"Zubkovsky, Sergey" <Sergey.Zubkovsky@transas.com> writes:
> create table t1 ( id int );
> create temp view v1 as select * from t1;
> C1: begin; drop view v1;
> C2: drop table t1;
> C1: commit;
This seems a variant of the problem noted by Michael Fuhr some time ago:
http://archives.postgresql.org/pgsql-hackers/2007-01/msg00937.php
but I think it shows that the solution I proposed in that thread is
still not adequate. To deal with the above, it seems like dropping v1
would have to acquire a shared lock on t1, thereby preventing the drop
of t1 from starting until after v1 is safely gone. Or maybe have the
drop of t1 recheck to see if v1 is still there after it has acquired
lock on v1, but I'm afraid that might be too late to prevent an update
conflict on the pg_depend row that both transactions want to delete.
Ick.
regards, tom lane