Philip Warner <pjw@rhyme.com.au> writes:
> So, if you do make the changes, will the schema definition be affected by
> those changes, or do you expect the tablespace to be embedded in the CREATE
> SCHEMA command?
I thought the idea was for pg_dump to emit something like
SET magic_tablespace_variable = some_ts;
CREATE TABLE foo (columns...);
rather than
CREATE TABLE foo (columns...) TABLESPACE some_ts;
the point being no more and no less than this: if "some_ts" doesn't
exist (or you have other problems like insufficient permissions) then
the SET command will fail but CREATE TABLE will still succeed, allowing
the restore to complete in some reasonable fashion.
I am quite unsure why you are pushing this while also insisting that
we need "die_on_errors" mode for pg_restore. If you are going to die
on the first error then these alternatives are equally brittle.
regards, tom lane