Forum OpenACS Development: apm callbacks during uninstall?

1. before_unmount
2. before_uninstantiate
3. before_uninstall

The sql drop-scripts get sourced after all three of these.

Does anyone have a clear idea what optimally is supposed to happen when during this complex process?

There have been several threads about how nothing uninstalls correctly, but most of these were before these callbacks were added to the apm (I think). The most useful of these is this one but it deals with a package using the CR, which invokes issues different/beyond the simpler case of packages using just the acs object system. The only packages that use these callbacks at all are bug-tracker and simulation, and these are substantially more complex than the "simple" acs-object-based situation I'm starting with.

So here's what I'm doing: I filed a bug for Forums because as is, the package fails to uninstall due to referenial integrity violations in acs_objects. Jeff pointed out that package data needs to be deleted in the before_uninstantiate callback. So I tried this: in the before_uninstantiate callback, I delete from acs_objects all objects corresponding to forums_messages and forums_messages (since those are the only acs_objects that the Forums package directly creates).

I presumed that would work, but there are still referential integrity violations. Do the notifications-related need to be wiped here, too? It's hard to see why, as the acs_objects of "notification_type", "notification_interval", and "notification_delivery_method" have null context_ids. There is an acs_object with apm_package object_type that has the context_id of the package I'm uninstantiating, but if I add a call to delete this in my before_uninstantiate, I get different referential integrity violations involving site_nodes. Sheeit.

It seems that there should be some way to figure this out other than trial-and-error. I sketched out the calling chain invoked when you click on the "Delete Package" button on the /acs-admin/apm/package-delete page here. This shows (at admittedly a chunky granularity) what's going on, but it's not at all clear to me what needs to be accomplished during those callbacks.

Can someone who Gets This explain what best practices for these callbacks are so I can add 'em in to existing and future packages?

Great diagram and clear thinking.

What we need is for someone to volunteer to document that these callbacks should be included in packages, and how to do so. The notes tutorial is the obvious place.

I think this is a bug in notifications. It seems to me that notifications should be on delete cascade, because yes, notifications currently do have to be deleted. And if the object a notification is based on is deleted, the notification should be deleted as well.

Can you file this as a bug on notifications?

in notification_requests:

    -- The object this request pertains to
    object_id                       integer
                                    constraint notif_request_object_id_fk
                                    references acs_objects (object_id)
                                    on delete cascade,

and in notifications:

    -- the object this notification pertains to
    object_id                       integer
                                    constraint notif_object_id_fk
                                    references acs_objects(object_id)
                                    on delete cascade,

However, the primary keys for these tables (request_id, notification_id) that have a foreign key constraint to acs_objects that *don't* on cascade delete. When these objects are deleted via the pgplsql functions, then their corresponding acs_objects get correctly deleted, but otherwise the acs_objects have to be manually deleted.

So, adding "on delete cascade" to the fk constraints should fix that, eh? Thus:

alter table notifications add constraint notif_notif_id_fk
                              foreign key (object_id)
                              references acs_objects (object_id)
                              on delete cascade;

alter table notification_requests add constraint notif_request_id_fk
                              foreign key (object_id)
                              references acs_objects (object_id)
                              on delete cascade;

Anyway, presuming this is so, I'll add a patch to my bug report that does this, including an upgrade script. Have a look and if (Jade/Jeff) you think this is correct, let me know and I'll go ahead and commit to oacs-5-1.

Almost none of the object type tables currently do this and I am not sure if it's a good idea or not.