Explain the situation with multi-threading, now it seems to me a little confusing. With GC: refc heaps are local for each thread. Therefore, it is only possible to copy shared data, or manual control via ptr (which is unsafe). With GC: ARC/ORC heap is shared so data can be moved. ARC doesnt use atomic counters so you need to manually lock data(which is eventually unsafe). Another option is to use channels with data copying as in refc.
Channels have a number of disadvantages: "Note: This is part of the system module. Do not import it directly. To activate thread support compile with the --threads:on command line switch. Note: Channels are designed for the Thread type. They are unstable when used with spawn Note: The current implementation of message passing does not work with cyclic data structures. Note: Channels cannot be passed between threads. Use globals or pass them by ptr." There are also isolates(<https://github.com/nim-lang/RFCs/issues/244>), which are an evolution of the idea of owned ref. I'm not sure I understand correctly what is meant by "pass subgraphs to threads", what is subgraph? But in general, this is an opportunity to put data in a container that will check that a number of restrictions for that data are not violated, which guarantees the safe movement of data into channels (instead of copying). And most importantly, they are not ready either. So the question is, what ways to work with shared data are relevant for ARC/ORC right now?
