Re:Swift 3.0 Release Process
by danappelxx on Saturday May 07, 2016 @01:56PM
(#11650587)
I think that Go's biggest mistake was making goroutines multithreaded. Libmill takes a much safer approach by making everything still run on one thread (and by taking advantage of libmill, Venice brings a similar scheme but with the added safety).
The fact that they are implemented using mutexes is, from my admittedly limited, non-concurrency-expert perspective, inexplicable. The simplest use case, an unbuffered channel, should be trivial to implement using CAS. There's a bunch of research on implementing robust queues using CAS.
Then there are the various ugly edge cases: Closing a closed channel is an error (now you have to implement your own refcounting implementation if you want to hand out a single channel to multiple consumers), receive on nil channel blocks (means you can't use nil as a signal to mean "I've closed the channel"), etc.
I really like GCD's style of enqueueing concurrent tasks, which I think is a concurrency model that often works more naturally than channels. A major difference between GCD and goroutines is that you can specify what queue to use. I'm looking forward to see what they come up with for Swift.