dbus - comments requested, here's one

[Michael Meeks]

On Tue, 2004-06-29 at 09:12 -0400, Havoc Pennington wrote:
> I don't see how this would work to solve the set of problems dbus is
> trying to solve, and fit within the set of constraints that dbus has.
> But of course I'm open to being proven wrong.

	FWIW, Luke in fact knows more about IPC than - well, many of us ever
will; and (from what I've heard - admittedly mostly from Luke in
person); DCE is an insanely cool technology[1].

	Luke - I think the features that are of greatest importance to Havoc
(and prolly D-BUS) are:

	* Not fooling around, trying to believe the network is
	  transparent, but having easy / good access to connection
	  based lifecycle / point-to-well-defined-point communication,

	* explicit access to security / context information

	* small code-size / minimal dependencies

	* pleasant asynchronous operation

	* Bus daemon / mode - with activation / filtering etc.

	* OOM handling

	* shipping ASAP ;->

	Some things I want / think are cool:

	* fully type described methods / arguments
	  ie. - you marshal a set of 'Any's which is powerful

	* type-centric / recursively typed / C ABI mapped smallness.

	Some things we are missing & thus exposing loads of painful API under-
belly:

	* IDL compiler

	Now - of course, if freedce can provide something that seems to fit all
that, works really well, and does it really quickly - providing a
similar API to d-bus [ ie. you can mostly hide it ] then I guess you
might swing this.

	As for the endless wheel re-inventing comments; one couldn't agree more
personally, I gave up trying to understand the rational here some time
ago. Of course - it's easy to think that one's particular combination of
choices from the well-known matrix is going to result in the perfect
solution; ahead of time that is.

	Regards,

		Michael.

[1] - I particularly like the recursive tree / linked-list structure
marshalling eg. ;-)
-- 
 michael at ximian.com  <><, Pseudo Engineer, itinerant idiot