[Sbcl-devel] Fwd: GC closing (or not) sockets

[Nikodemus S. <nik...@ra...>]

Double whoops.

---------- Forwarded message ----------
From: Nikodemus Siivola <nik...@ra...>
Date: Jul 25, 2007 6:17 PM
Subject: Re: GC closing (or not) sockets
To: Daniel Barlow <da...@st...>

On 7/25/07, Daniel Barlow <da...@st...> wrote:

> [If you wanted your reply to go to the list, I'm not sure it did.
> Please feel free to send this there as well though, if so]

Whoops.

> Nikodemus Siivola wrote:
> > I'm actually almost convinced that we should have a representation for
> > sockets &co in the core -- or the core IO facilities should be flexible
> > enough
> > that a contrib could transparently extend them so that the TCP socket would
> > be the same object as the stream.
>
> Yes, absolutely.  This would actually be quite simple: just make
> ansi-streams/fd-streams a semi-public interface - it would still be
> nicer to have something based on standard classes instead of defstruct,
> but I don't think that's essential given the amount of bootstrapping
> work it'd involve

Almost, but not quite. :) I agree with the sentiment, but the more I
try to work with fd-streams
(and to lesser degree with ansi-streams), the more certain I am that
we really don't want to expose them any more then we already do. Of
course, barring better _existing_ interfaces...

Cheers,

  -- Nikodemus