Re: Hurdish TCP stack

[olafBuddenhagen]

Hi,

On Wed, Apr 02, 2008 at 09:23:20AM +0200, Ludovic Courtès wrote:
> <olafBuddenhagen@gmx.net> writes:

> > A filesystem based interface is much easier to use for most
> > programmers than generic RPCs.
> 
> It's easier to use from a shell, not from a C program.

Actually, most people will consider it easier to use it from a C program
as well: For one, it means that you can use the *same* knowledge for
doing stuff on the shell, and for writing C programs. That's a very
valuable property IMHO.

Even more importantly, every UNIX programmer is already familiar with
using files, while RPCs is something they need to learn explicitely to
hack Hurd stuff -- a major entry barrier.

For these reasons, while native RPC interfaces might be more elegant in
theory, filesystem-based interfaces are usually *much* more useful in
practice.

> > In this particular case however I don't see the need for a
> > specialised RPC interface -- neither is performance critical, nor do
> > we have complicated relationships that could be better expressed by
> > such an interface...
> 
> My understanding is that the Hurd's way is to use a new RPC interface
> whenever using the `io' interface would be non-trivial or cumbersome
> (e.g., `password', `auth', etc.).

Well, auth for example is a special case: The filesystem itself relies
on it, so using an FS interface here would create a circular
dependency...

More generally, you are right. (Though personally I tend to FS-based
interfaces much more than many other people here...) However, I don't
see anything cumbersome in using an FS-based interface to control
sockets...

-antrik-