@Sixte I've studied Pony a little and I don't think its model fits Nim very well.
1\. I don't think thread local heaps work well for parallelism, on a GPU that would mean 10^4 thread local heaps and then you cannot easily merge the results of a computation. See for example: [http://on-demand.gputechconf.com/gtc/2014/presentations/S4271-halloc-high-throughput-dynamic-memory-allocator.pdf](http://on-demand.gputechconf.com/gtc/2014/presentations/S4271-halloc-high-throughput-dynamic-memory-allocator.pdf) 2\. `iso` and `trn` are 2 new types as opposed to my `owned` type. Every type system extension makes a language harder to use and understand. I'm glad I "only" had to add `owned` to Nim and reap many benefits, hopefully outweighing the costs.
