Related IRC discussion between Emmanuel and Gunnar:

[2:29pm] gmorling: emmanuel: hey
[2:29pm] gmorling: emmanuel: can you take a look at OGM's OneToOneTest, method testBidirectionalManyToOne()
[2:29pm] gmorling: there we have two entities, Husband and Wife
[2:29pm] emmanuel: looking
[2:29pm] gmorling: and there is a bi-di OneToOne association between them
[2:30pm] gmorling: that's persisted as in the first file of this gist: https://gist.github.com/gunnarmorling/95ae61bf45cf420b1462
[2:30pm] gmorling: i'd expect the second, though. or do i miss a piece?
[2:33pm] emmanuel: well yes that would be moderately better
[2:33pm] gmorling: the current form seems misleading
[2:33pm] gmorling: as its not an array of husbands really
[2:33pm] emmanuel: this case is really technically a ManyToOne but where the app tells you that there is only one item on the many side
[2:34pm] emmanuel: so in a RDBMS, it's enforced by a constraint
[2:34pm] gmorling: now it's getting interesting
[2:34pm] gmorling: because i was about to ask
[2:34pm] gmorling: updateAssociation() is invoked for this
[2:34pm] emmanuel: here indeed we could be a tiny bit better and save it as an item instead of an array
[2:34pm] emmanuel: yes
[2:34pm] gmorling: that's why the mongo dialect creates it as an array
[2:35pm] gmorling: so orm creates a unique constraint on the fk column in this case?
[2:35pm] emmanuel: yes
[2:36pm] gmorling: i guess right now we wouldn't have a way to detect this situation in the dialect
[2:36pm] gmorling: and handle it specially
[2:38pm] emmanuel: we coudl only if AssociationKeyMetadata was carrying the OneToOne nature of it
[2:38pm] emmanuel: which would be a good idea
[2:38pm] gmorling: yes, something like that