This is really hackers stuff.
Dave Page wrote:
>
>
> I don't recall that discussion, but in general I think we should
> completely ignore the search path. Consider a function: foo.dostuff().
> The current code will return an empty schema prefix for a search_path of
> public,bar,foo. What if there is also public.dostuff() or bar.dostuff()?
> CREATE OR REPLACE could really screw up in that case...
Some logic black holes... preliminarily public only.
>
> I also don't like the notion of treating public as some kind of special
> schema. From PostgreSQL's pov, its only special in that it's there by
> default in template1 and the search_path. Other than that it's just
> another schema and should be treated as such.
A grep showed that only FK has handles public special, all other places
go through pgDatabase::GetSchemaPrefix.
The correct overall behaviour seems
- find the first schema in search_path that exists.
- If this is the schema in question, suppress it.
- (ignore all following schema names, this was the main problem)
- If schema = pg_catalog, suppress it.
Unfortunately, this search_path[i] = session_user is not absolutely
stable (schema or user name may change), but it should be stable enough.
Thoughts?
Regards,
Andreas