P.S. Also commented about the paper here: https://news.ycombinator.com/item?id=30561786
The simple intuition is that because his CRDT's are associative and everyone else is not using associative data types, he simplifies the space and the impossible becomes possible.
1. This impossibility theorem is proved in a really broad setting (assuming a “broadcast communication cloud” and a P2P channel). If he wants to circumvent the impossibility theorem he should clearly specify the system model and say how it differs from Doceur’s. There are other instances where by playing with the model you go around an impossibility theorem, but you have to specify the differences. Especially since this theorem is so well known/cited.
2. Some claims in the old paper are just plain wrong. Like, you can’t say it’s Sybil resistant just because it tolerates arbitrary number of Byzantine nodes. Dolev-Strong also needs just one honest party, but it requires a CA.