Re: pg_dumpall problem when roles have default schemas

btober@ct.metrocast.net writes:
>  Thus, when piping the output (from legacy host 192.168.2.2) to
> populate  the newly initialized cluster, by way of running (on the new
> host  192.168.2.3)
>      pg_dumpall -h 192.168.2.2|psql
>  an error occurs in that first section when the script attempts to
> set a  role-specific search path ... because the schema named in the
> search  path hasn't been created yet.

Could we see a complete example?

Recent versions of the backend are not supposed to throw hard errors in
this situation, because of exactly that hazard.  For instance:

regression=# create role joe;
CREATE ROLE
regression=# alter role joe set search_path to foo, bar;
NOTICE:  schema "foo" does not exist
NOTICE:  schema "bar" does not exist
ALTER ROLE
regression=#

which AFAICS is exactly what will happen while restoring a pg_dumpall
dump.

            regards, tom lane