> set search_path = my_schema;
> This works fine unless the object name duplicates a system object;
> in that case, since the effective search path is really "pg_catalog,
> my_schema", the DROP will find and try to drop the system object.
I must have missed that day. Why is that exactly? Clients like psql
should probably explicitly specify pg_catalog path anyway.
Afterall, if you create a my_schema.pg_class table (for whatever
reason), and used my search path as my_schema, I'd expect my own to be
hit with my queries. Likewise, postgresql internals should specifiy
the specific namespace -- which they generally do through knowledge of
the pg_class oid.
Is this a temporary thing to tide clients over for a release without
breaking too much?
> 1. Explicitly qualify target-object names in the DROP commands,
> ie, we'd emit
Anyway, question at hand. How about a modification of #1. If the
table begins in 'pg_' explicitly name it my_schema.pg_????. If users
are creating stuff in pg_catalog they should be ready for weird
things -- especially given the 'overriding' state it takes in the
search path.
