[boost] Re: Aaron's Informal Proposal for a Universal Demultiplexor for Boost

From: Hugo Duncan (hugoduncan_at_[hidden])
Date: 2004-09-15 13:23:53

On Tue, 14 Sep 2004 01:21:13 -0500, Aaron W. LaFramboise  
<aaronrabiddog51_at_[hidden]> wrote:

> Well here it is.  This is a very sketchy "brain dump" of the general
> form of demultiplexor I have been referring to.  I think that it might
> be a basis for something eventually acceptable as a general
> demultiplexor for Boost.

This looks quite similar to what is already in giallo.  It adds some things
like a dispatch queue that are not currently in giallo.

Giallo also places the callbacks and the demultiplexer interface into
connection/connector/acceptor classes, providing (IMHO) a clean seperation
of responsabilities:

the socket class thus wraps the native socket functions into a common  
interface,
providing platform independent error messages.

the native select and completion ports are also wrapped as objects, which
the proactor and reactor classes use (via a bridge class specified as a  
policy)
to provide a consistent demultiplexer interface.

the connection/connector/acceptor manage the state (and lifetime) of the  
underlying
socket, use a demultiplexer (proactor or reactor, interchangeably) and  
provide the
user with completion callbacks.