Bidirectional list fails if session only knows about child

[RogerKratz]

When having a bidirectional inversed list, child seems to not fully "own" the collectionref when using session.Merge on child.

When running failing test, child will end up with order index -1 in db because when nhib's merge occur, the child is "unhooked" from its parent.

Probably closely related to #1531.

[RogerKratz]

I meant owner/non-owner related logic is applied only for collection.

Sure, I'm not talking about cascading or similar. Still, it's a bidirectional reference, and child know it's parent (and orderindex) in the domain model, so why shouldn't it be possible to save child without session knowing about parent, if that's what you mean?

(Even with inverse=false it should work, but that's seldom/never used if ref is bidirectional I think?)

[bahusoid]

and child know it's parent

Child knows only what's in its mapping. So it knows Parent but knows nothing that Parent has Children. Merge only processes supplied entity and associations according to cascade settings. You have no cascade settings specified for Parent. So why do you expect it to be saved as part of Parent.Children collection?

[RogerKratz]

Child knows only what's in its mapping. So it knows Parent but knows nothing that Parent has Children. Merge only processes supplied entity and associations according to cascade settings. You have no cascade settings specified for Parent.

Agreed.

So why do you expect it to be saved as part of Parent.Childrencollection?

I don't. Why do you think I expect that? ;) We must be misunderstanding each other...
I expect "newChild" (not its parent) to be correctly persisted in the test.

[RogerKratz]

Actually I dont. I expect Child's OrderIndex property and mapping to be used.
Yes, the implementation of that property traverse the domain model (in this case to the parent). But from nhib's perspective (and the mapping), it should be irrelevant. It's just references in ram so to speak. And the refs "work" when Merge is called.

[RogerKratz]

@bahusoid Thanks for your time and discussion. Glad I could finally make myself a bit clearer :)

[hazzik]

According to this: https://stackoverflow.com/a/1758311/259946 bi-directional inverse list does not work at all. As I can see you have implemented a hack similar to the hack described in the SO answer.

The NH-1338 has been closed as not an issue with weird explanation.

Issues on hibernate's JIRA (HB-79, HHH-1165, HHH-3425) are also closed with the same explanation.

(Here I'm really talking to myself trying to make sense of the mapping)

[RogerKratz]

@hazzik IIRC, "bi-directional inverse list does not work at all" was true a couple of years ago. Nowadays it is/should be supported as of 6.3.3 in documentation. Unfortunately, domain model is not part of those examples.

The problem (if it is?) is probably the same as #1531. It only happens when merging entities. NHib then "detaches" the entities in the domain model from eachother and makes copies of the data. Due to this the references themself inside the model no longer points to the "correct" entities and the hack (if it is?) no longer works.

[hazzik]

I was talking about bidirectional inverse list without a special column.

[hazzik]

Thanks. This has now lives on master. Not sure though if we're going to fix this.