update_attributes doesn't remove elements (or what's a proper way to represent this relationship?)

[dblock]

I have a many-to-many relationship between Exhibitor and Artwork. An
exhibitor can be showing an artwork at an exhibition and artworks
travel with many exhibitors. I tried many relationships and the
closest I got things to work is with a
references_and_referenced_in_many :artworks inside Exhibitor:

class Exhibitor
...
references_and_referenced_in_many :artworks
end

(no inverse relationship defined, although that makes no difference)

I am saving via a controller that's called by javascript. With this, I
can do:

1) create(hash) where hash contains artworks = [ array of ids ] ->
success
2) update_attributes(hash) where hash contains artworks = [ array of
ids, with several new ids ] -> success
3) update_attributes(hash) where hash contains artworks = [ array of
ids, with removed ids ] -> fail

By fail I mean nothing is removed. Is this a bug? Am I doing something
wrong - is there a better/proper way to represent this relationship? I
am on rc8.

[Steve V]

I think this is a bug. At least in terms of checking in the rails
console. The update is actually happening DB side, but the calling
object does not have its references updated. Once I reload the object
from the DB, then the updates are reflected. This is definitely not in
keeping with expected behavior. I tested with 2.0.1.

[Steve V]

Hopefully one of the core committers can comment on this. After going
through the code with rdb, I think the proper place to make the change
for this would be in Dirty#modify. As it is now, it just sets the
relationship attribute(in this case :artwork_ids), and calls it a day.
It seems like it should instead pull the field information, and check
if it's dealing with a relation. If that's the case it should grab
model instances for given ids, and set them for the
relation(artworks).

Perhaps this behaviour is already encapsulated elsewhere, and could be
added with a simple call. I also don't know if there are any gotchas
to this offhand. Again, maybe this is something that David, or Kieran
can provide more insight on.

[Kieran P]

Sounds like this might be related to https://github.com/mongoid/mongoid/issues/605 and https://github.com/mongoid/mongoid/issues/814

Regards

Kieran

[Steve V]

It does seem related to 605, though I think in the scenario with a
relation using *_ids, it's a case of mongoid not even attempting to
load the relation models. The data does save correctly to the DB, so I
don't think 814 applies. I could of course be completely wrong,
because going through the saving code is quite a complicated process.

[dblock]

I created https://github.com/mongoid/mongoid/issues/880. Thanks.