Hi Tom Thanks for your reply. I was restoring one schema (with data) at a time from 8.3 db to a pre-created empty 8.4 db. I received a lot of errors which game down to me not restoring the public schema first. I also realized, I had not created plpgsql language for the 8.4 db before I did the restoration. So after I created the language and restarted the restoration from scratch with the public schema first, I no longer received this error.
Thanks Rebecca On Wed, Jul 20, 2011 at 3:46 PM, Tom Lane <t...@sss.pgh.pa.us> wrote: > Rebecca Clarke <rebe...@clarke.net.nz> writes: > > I'm doing a restore into Postgresql 8.4, postgis 1.5.1 and I'm getting > the > > following error when it creates the below function: > > > pg_restore: creating FUNCTION _get_buffer(geometry, double precision, > > integer) > > pg_restore: [archiver (db)] Error from TOC entry 966; 1255 49162661 > FUNCTION > > _get_buffer(geometry, double precision, integer) postgres > > pg_restore: [archiver (db)] could not execute query: ERROR: SQL function > > cannot accept shell type geometry > > Command was: CREATE FUNCTION _get_buffer(_geom geometry, _radius > double > > precision, _density integer, OUT the_geom geometry) RETURNS geome... > > > I don't understand what 'cannot accept shell type geometry' means? Can > > anyone help my lack of knowledge? > > Hmm, that is interesting. A "shell" type is a SQL base data type that > has been declared to the database but its properties are not yet filled > in. This is needed because the properties include I/O functions, which > have to be declared to take or return the data type, so there's a > circularity involved. The standard solution is > > CREATE TYPE typename; -- this creates typename as a shell > > CREATE FUNCTION typename_in(cstring) RETURNS typename ... > > CREATE FUNCTION typename_out(typename) RETURNS cstring ... > > CREATE TYPE typename (input = typename_in, output = typename_out, > ...); > > The last step changes the type from a shell into a real, usable > datatype. > > So what you've apparently got is a situation where that last step got > missed for the geometry type, or else the _get_buffer function somehow > got inserted into the middle of this sequence. I've not heard of that > happening to people before, so I wonder if you could provide the exact > step-by-step of what you did. > > regards, tom lane >
