Re: [Sbcl-devel] Supporting the Abstract Namespace in Unix Domain Sockets

[Matthew D. S. <ak...@ch...>]

On Sat, 28 Jun 2008 11:41:53 -0400
Richard M Kreuter <kr...@pr...> wrote:

> Matthew D. Swank writes:
> > Thomas F. Burdick <tburdick <at> gmail.com> writes:
> > 
> > > 
> > > On Fri, Jun 27, 2008 at 5:37 AM, Matthew D. Swank
> > > <akopa.gmane.poster <at> gmail.com> wrote:
> > > >
> > > > There is a problem in trying to use the abstract namespace for
> > > > AF_LOCAL. Address in the abstract namespace are _not_ C strings.
> > > >
> > 
> > ...
> > 
> > > > Would it be possible to ... support a subtype of LOCAL-SOCKET 
> > > 
> > > It should be pretty simple to support byte arrays as specifying
> > > abstract namespace addresses. That sounds like the most reasonable
> > > interface, too.
> > > 
> > 
> > It does seems kind of low level to be reduced to using byte array
> > in this particular instance, especially when it's "just" a special
> > string conversion.
> 
> There are alternatives:
> 
> * have a subclass of LOCAL-SOCKET for sockets whose addresses are in
> the abstract namespace,
> 

Attached is a first pass at adding AF-LOCAL sockets with abstract
namespace addresses as a subclass of LOCAL-SOCKET.  It adds an
alternate grovel of socketaddr-un that specifies the path as an 
(unsigned 8) array.  Since the path is not a c-string, it is necessary
to provide correct a correct addrlen to connect and bind.  To that end
make-sockaddr-for returns addr-len as an additional value.

This is implemented w/o touching the existing LOCAL-SOCKET code, but
it could be integrated easily into the LOCAL-SOCKET type.

Matt 


-- 
"You do not really understand something unless you can explain it to
your grandmother." -- Albert Einstein.