Re: D-Bus

[Havoc Pennington <hp redhat com>]

On Mon, Mar 03, 2003 at 11:12:20AM +0000, Michael Meeks wrote: 
> 	So - you're idea for 'fixing' the 'uncontrollable' re-enterancy is to
> queue everything until idle

No. My idea is to either use threads, to use explicit async APIs, or
to simply block without reentering the main loop and fix any deadlock
issues that arise. D-BUS gives you those three options. (As does Xlib,
for example.)

> : well, that's really tough to do. If -
> instead of re-writing this stuff in GConf / nautilus / etc. people had
> bothered to do the right thing and implement a 'delay till idle' POA
> policy - this would have been trivially fixed once. It's a comparatively
> tiny amount of API addition.

This is exactly the kind of not-very-robust solution that has been
causing us problems.

> 	As for the lifecycle issue - you can do connection lifetime based
> lifecycle already, and produce a reliable daemon / bus that doesn't use
> reference counting with the existing infrastructure.

Sure. However, we still have the lifecycle issue all over GNOME
because Bonobo doesn't do this.

I do agree that you could write a reasonable message bus with CORBA,
at least ignoring the systemwide bus issue. I already said that.

Why don't you write up your plan to do this with ORBit, and try to get
some KDE guys on board. You can talk to those guys as easily as I can.

Havoc