Re: modules

Alex Rousskov wrote:
>
> On Tue, 1 Jun 1999, Dancer wrote:
>
> > The port number is probably the simplest piece of the assorted network
> > parameters to deal with seperately. Dummy addrinfo structures with port
> > 0 and plug in the actual port later (before passing to connect).
>
> Port -1 by default, please. Zero is a "valid" port number for some
> interfaces. A negative port number can always be viewed as "undefined".

Depending on your implementation and platform, though -1 is also a valid
port number. Just less commonly in use. The idea above is to actually
use a legal port number as a placeholder until the _real_ port number
gets plugged into the structure, as some implementations of addrinfo may
concievably choke on an invalid one.

Soon as I can get a chance to look at the current snapshot (in a few
hours, I hope) I'll see how hard it'll be to hack all this in. I'm
hoping it won't be terribly awkward.

> As for the modules, I think the best approach is to do it bit-by-bit. I bet
> Duane will not be comfortable committing a huge patch with arguable
> interface changes. I would suggest less changes at a time and only those
> that target specific "interface users". In my experience, designing a
> "complete" interface from scratch is a waste of time as it is usually
> impossible to predict what minor hook-ups will be needed tomorrow. We are
> not designing a standard C/C++ I/O library and have a luxury of tuning the
> interface as need arises (some code might have to be rewritten, but that
> should not be a big deal at early stages). Actually, even the standard I/O
> libraries were significantly rewritten several times before the interface
> became "standard".

Concur.

D
Received on Tue Jul 29 2003 - 13:15:58 MDT