Re: pg_dump and schema names

On Fri, Aug 9, 2013 at 01:48:43AM -0400, Tom Lane wrote:
> Bruce Momjian <bruce(at)momjian(dot)us> writes:
> > pg_dump goes to great lengths not to hard-code the schema name into
> > commands like CREATE TABLE, instead setting the search_path before
> > creating the table; these commands:
>
> > CREATE SCHEMA xx;
> > CREATE TABLE xx.test(x int);
>
> > generates this output:
>
> > SET search_path = xx, pg_catalog;
> > CREATE TABLE test (
> > x integer
> > );
>
> > If you dump a schema and want to reload it into another schema, you
> > should only need to update that one search_path line. However, later in
> > the dump file, we hardcode the schema name for setting the object owner:
>
> > ALTER TABLE xx.test OWNER TO postgres;
>
> > Could we use search_path here to avoid the schema designation?
>
> Perhaps, but that's not likely to reduce the number of places you have to
> edit, unless your dump is only one schema anyway.
>
> The practical difficulties involved can be seen by reading the comments
> and code for _getObjectDescription().

Yes, I looked at that. Seems _getObjectDescription() is only called
from _printTocEntry(), and that function has a call to
_selectOutputSchema() at the top, so we already know we have search_path
set to the proper schema.

The attached patch removes the unnecessary schema qualification for
ALTER OWNER, and the attached dump file show a two-schema dump that
restores just fine.

Basically, if we are going to use search_path to avoid schema
specification, we should do it in ALTER OWNER too.

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ It's impossible for everything to be true. +