Jon Pretty

I've been thinking a lot about our chat about this since we spoke in December, and I've more or less come round to your approach.

The problem is one that fundamentally boils down to making changes to a persistent Merkle DAG where cross-vertex changes are non-atomic. We can either make the DAG non-persistent, or we can accept that there will be quite a few independent changes.

Atomic updates, by definition, cannot be guaranteed because we aren't just dealing with one DAG, but technically an open set of partially-overlapping DAGs. That's a particularly complex issue, and due to the openness of the set, we cannot guarantee global control over all of it.

Ultimately, I think that the final evolution of Scala Steward is probably pretty close to something that @dcsobral is working on right now, which employs a stronger semantic understanding of specific subgraphs to eliminate multi-path propagation noise.