[boost] Re: Re: sockets library question

From: Michel André (michel.andre_at_[hidden])
Date: 2003-01-14 17:50:11

> There is a basic low level socket implementation (socket_base),
> and Michel André has put together an extension of this for the
> asynchronous methods on NT.

Here we would need some help implementing lower lever asynch classes using
unix aio or eqvivalent strategy for other platforms.

> At this stage we need to decide what would be the scope of an
> initial submission.  Is the basic socket design useful in itself,
> or would the proactor and reactor patterns be required for the
> library to have any meaning?  Is the level 1 design (without
> reactor/proactor classes) sufficient for client usage?

I think at least level1 is needed completed with some kind of
reactor/selector/polling/multiplexor/proactor primitives for ordinary
sockets.

> Ultimately I would like to hope that we can build a framework
> where the choice of procator, reactor and threading decisons
> can be completely seperated from the application code, with
> these decisions being implemented as policies.   But that is
>  going to take quite a while to get right....

Well this is a though one and the solution i can see is to "emulate" asynch
operations using ordinary sockets and select and use more effecient
strategies on demand or if available. Some input on this side would be
invaluable.

See
http://www.crystalclearsoftware.com/cgi-bin/boost_wiki/wiki.pl?BoostSocket/A
synchAcceptorConcept for what I had in mind.

Regards
/Michel