Thread: ALTER EXTENSION DROP FUNCTION not working ?

I'm seeing an issue with ALTER EXTENSION DROP FUNCTION
not fully unregistering the dependency of the function
on the extension. Commands to reproduce described here:
https://trac.osgeo.org/postgis/ticket/3450#comment:23

Basically I'm getting:
ERROR:  cannot drop function pgis_twkb_accum_finalfn(internal) because other objects depend on itDETAIL:  extension
postgisdepends on function pgis_twkb_accum_finalfn(internal)
 

Right after successfully running:
ALTER EXTENSION postgis DROP FUNCTION pgis_twkb_accum_finalfn(internal);

Is it a bug in PostgreSQL ? (9.3.6 running on the test machine).

--strk;
 ()   Free GIS & Flash consultant/developer /\   http://strk.keybit.net/services.html

On Tue, Feb 09, 2016 at 10:33:42AM +0100, Sandro Santilli wrote:

> https://trac.osgeo.org/postgis/ticket/3450#comment:23
> 
> Basically I'm getting:
> 
>  ERROR:  cannot drop function pgis_twkb_accum_finalfn(internal) because other objects depend on it
>  DETAIL:  extension postgis depends on function pgis_twkb_accum_finalfn(internal)

Figured:  the "pgis_twkb_accum_finalfn(internal)" function is not
a _direct_ dependency of extension "postgis", but is needed for
an aggregate which is still registered.

So this is more an annoyance than a bug, being the non-clear error
message about what's the direct dependent object that prevents
the drop.

--strk;

Sandro Santilli <strk@keybit.net> writes:
> On Tue, Feb 09, 2016 at 10:33:42AM +0100, Sandro Santilli wrote:
>> https://trac.osgeo.org/postgis/ticket/3450#comment:23
>> 
>> Basically I'm getting:
>> ERROR:  cannot drop function pgis_twkb_accum_finalfn(internal) because other objects depend on it
>> DETAIL:  extension postgis depends on function pgis_twkb_accum_finalfn(internal)

> Figured:  the "pgis_twkb_accum_finalfn(internal)" function is not
> a _direct_ dependency of extension "postgis", but is needed for
> an aggregate which is still registered.

> So this is more an annoyance than a bug, being the non-clear error
> message about what's the direct dependent object that prevents
> the drop.

I believe this is an intentional design decision: if you try to drop
something that an extension member object depends on, we complain about
the extension, not about the member object.  This is on the grounds
that a user of an extension likely doesn't know or care exactly what
all the extension members are, and would rather think of the extension
as a unit.  Consider for example that if you'd said CASCADE, the command
would have led to dropping the whole extension; would you want each
member object of the extension to be called out explicitly as a drop
cascade target?

I agree that it's a bit unintuitive when you're the developer of the
extension, but I doubt we're going to revisit this choice.
        regards, tom lane