Omry Yadan
Omry Yadan
Looks good overall and I agree that it's a better behavior. I think there is still some circular dependency.
> I feel that OmegaConf.update's behavior should be consistent with that of OmegaConf.merge: calling update should result in a node out.child that is a non-None structured config backed the Child...
Can you reproduce the problem without using yaml anchors?
Generally speaking, anchors sucks, and you should use interpolations instead. We are considering disabling anchors support by default to address #794.
@Jasha10, I am actually not sure interpolation can be used here: OmegaConf does not provide a parallel to YAML merge (e.g. `
I would still like to get this reviewed. I am not currently working on OmegaConf but there might be some new maintainers coming soon.
Hi @mdraw, This is nice. Can you add a some basic testing and make sure everything passes (you may need to rebase).
there is a merge_with method in DictConfig (iirc) which is used as the lower level function for `OmegaConf.merge`. It can be used for `cfg1 |= cfg2`
> Your writeup also highlights the case of coercion from `str` to `Enum`. It might make sense to add a special case handling for `str -> Enum` coercion in nested...