Paul Colomiets <paul@colomiets.name> wrote:
I've written an article about how I perceive the future of asynchronous I/O in Python. It's not something that should directly be incorporated into python now, but I believe it's useful for python-ideas list.
https://medium.com/@paulcolomiets/the-future-of-asynchronous-io-in-python-ce...
This approximately how asynchronous I/O is implemented on Windows (IOCP) and Mac and FreeBSD (GCD): You have a thread-pool that operates independent of the main program thread, and then you can enqueue I/O operations as work tasks to the thread pool. You suggest to have a singleton "I/O kernel" thread in Python, but it is actually not that different from having a pool of worker threads. In a very minimalistic way, one could implement something similar to the set of I/O functions present in GCD. This API is already designed to have the smallest possible complexity, and yet it is as powerful as Windows' IOCP for most practical purposes. https://developer.apple.com/library/mac/documentation/Performance/Reference/... On Windows it could run on top of IOCP, on Mac and FreeBSD it could run on top of GCD and thus use kqueue under the hood. Linux would actually be the hardest platform, but IOCPs have been implemented with epoll and a threadpool in Wine. Sturla