Re: [Evolution-hackers] gtkhtml - composer not starting ...

[Michael Meeks <michael ximian com>]

Hi Havoc,

On Wed, 2002-05-15 at 22:58, Havoc Pennington wrote:
> Of course the underengineered, simple solution that works fine most of
> the time would be a a cheesy explicitly-async message-passing API. Can
> be implemented in a week.

	And of course ORBit2 does asynchronous invocations for you - there is a
(fairly nice) API for it, that is missing direct IDL compiler support;
but it's easily possible to construct stub/skel pairs that do async
invocations, (without the need for oneway calls), that are fully
non-blocking [ oneway with ORBit1 is not fully non-blocking ].

> I have a slightly more elaborate design for this that also addresses
> activation, broadcast messages, lifecycle, and messages from
> system/root users to the desktop. No one to implement it though.

	Wow - you want to re-write the whole lot, and bin standards altogether
? anyone can do that, trivially. If you have a magic bullet for cross
process lifecycle management though I'm gagging to hear your suggestion
- please expand.

> Anyway, yeah long term I would rather see threads than hacks that fix
> the problem "most of the time when you don't _need_ reentrancy,"

	Doh, why does everyone say it's a hack. Examine for example the X
protocol, which is essentially a transport (say CORBA) and a load of
non-reentering calls, and an idle event processing loop. Well, with the
[ no_reenter, handle_at_idle ] markup, we have - precicely that model,
with the additional flexibility of handing some things re-enterantly if
we think we can cope with the gob-smacking complexity of that.

>  but then again I admit I don't see why we're spending years of effort on
> making remote methods/objects look like local methods/objects (modulo
> different type systems) in the first place.

	Dude; you begin to piss me off. This mantra of repeating brokenly that
"remote and local objects look the same" is just silly. It's certainly
not what I'm aiming at, and it's not what anyone is programming to -
it's a straw man argument.

	Perhaps you want to read my talk here:

		http://www.gnome.org/~michael/guadec-2002/

	to get some things to argue against.

	There are also some interesting things for you to consider there. For
example the 'write' syscall 'transparently' provides an API for remote
and local method/object interaction. It's hardly a new idea.

> I'd say by all means, add a way to make methods non-reentrant, we need
> the workaround for now. Threads are probably out until GNOME 3.

	I do believe the solution is simple, elegant, efficient, provides us
with what we need, and removes a no. 1 favorite gripe of peoples. It'll
also stop gconf_get_value ("") being a re-entrancy point - I hope. [ Oh,
and ensure that gconf value changed notifications come in at idle ].

	Regards,

		Michael.

-- 
 mmeeks gnu org  <><, Pseudo Engineer, itinerant idiot