Thanks for linking my post. I really need to write a followup at some point - we’ve gotten another 2-10x speed up from when I ran those benchmarks, depending on how you measure it.
I still stand by what I wrote in that blog post. Using lists rather than trees is still a good approach. And it’s super simple to implement too. You can have a working crdt in a few dozen lines of code.
I’m happy to answer questions if anyone is curious about this stuff. Ive been working to implement and optimise CRDTs for years at this point. Increasingly I’m seeing it as a solved problem.
I still stand by what I wrote in that blog post. Using lists rather than trees is still a good approach. And it’s super simple to implement too. You can have a working crdt in a few dozen lines of code.
I’m happy to answer questions if anyone is curious about this stuff. Ive been working to implement and optimise CRDTs for years at this point. Increasingly I’m seeing it as a solved problem.