phb07 <phb07@apra.asso.fr> writes:
> Le 26/11/2016 à 01:59, Tom Lane a écrit :
>> Don't see why that's a "workaround". You added the extension membership
>> for the sequence explicitly, why wouldn't you expect to need to drop it
>> explicitly? Seems to me the system is behaving properly.
> Because depending on the way a table has been included in the extension
> (either directly created inside the extension or first created outside
> and then linked to the extension) the procedure to drop it some versions
> later would be different.
Well, no it wouldn't be, but nonetheless on closer study I think you're
right that this is a bug. There is code in there that intends to make it
unnecessary to issue "ALTER EXTENSION DROP member" when an extension
update script drops a member object; it should be sufficient to just
drop the object. It was failing to fire in this case because the drop
was indirect, but it should work anyway.
I've applied a patch for that. Thanks for the report!
regards, tom lane