James Mansion wrote:
>>IOCPs support true async network i/o. BSD KQ, Solaris /dev/poll, epoll et.
> 
> al. are not async, they are event
> 
>>driven i/o models. When you issue an async read on Windows, the kernel will
> 
> start filling your i/o buffer as
> 
>>soon as data is available.
> 
> 
> I'm not sure the technique you describe will work.
> 
> Consider an alternative approach:
> 
> 1) from a poll perspective, consider a socket that does not have an
> overlapped operation in progress as ready for writing.  Write to a buffer
> that's owned by the poller and then run the overlapped or IOCP from that.
> 
> 2) for reading, always have an OVERLAPPED operation running against a
> poll-owned buffer - unless the read side has been shutdown or we're not yet
> connected.  If this buffer is not empty, poll considers it readable.
> 
> Extra buffer copies are needed but it does the job.  We're just using a
> buffer/queue to translate between push and pull structures, after all.
> 
> James

I like this approach and we can probably work out how to do reads w/o introducing an additional buffer copy. 
The buffers would get morphed into immortal buffer types and ownership of the buffer would be passed to the 
polling (or worker?) thread.

Bill