The parent model still thinks it's saved when the child fails validation.

[trungpham]

parent.saved? = true
parent.id = some id

when the rollback occurs, parent.saved? should be false.

[snusnu]

I started refactoring the specs a bit mainly to become familiar with all this stuff again. I will try to work through these issues in the next days, but I'd greatly appreciate any help from your side! :) That doesn't necessarily mean that you'd need to send patches to correct the behavior, but failing specs would certainly be appreciated! (of course, patches that fix the errorneous behavior too :)

[al]

Are there any 5 year old insights on this? I can't track down the problem ...

[snusnu]

@al i'm afraid no. I haven't been active on dm-accepts_nested_attributes as you can (clearly) see in the logs. Generally tho, there is a lot of weird behavior when saving object graphs with dm-validations in the mix. IIRC, there are quite some somewhat related issues in dm-core and/or dm-validations.

That said, I'm sorry that I can be of no help here, but maybe researching any seemingly related issues in the above mentioned repos might help with debugging.