[Linux-ha-dev] Re: crm dtd changes

[Andrew]

On Jul 7, 2004, at 5:48 PM, Lars Marowsky-Bree wrote:
>
>>> nor is
>>> the transition graph (neither in your form nor in the one I proposed)
>>> capable of expressing loops, so it can't be in the data ;-)
>>
>> no?
>>
>> I forget the exact terminology you used, but
>>
>> <transition_graph>
>> 	<synapse>
>> 		<input id=C/>
>> 		<action id=A/>
>> 	</synapse>
>> 	<synapse>
>> 		<input id=A/>
>> 		<action id=B/>
>> 	</synapse>
>> 	<synapse>
>> 		<input id=B/>
>> 		<action id=C/>
>> 	</synapse>
>> <transition_graph>
>>
>> looks like a loop that will never start to me...
>
> Well, OK, there's events which can never trigger (unless someone
> externally triggers any of the actions), but when you said 'loop' I was
> thinking about a data structure where we would go into loops while
> traversing it, which isn't true, as the list of synapses is just that, 
> a
> list - with inputs each etc though...
>
> The above would likely be caught by appropriate timeouts.
>

I dont really like the idea of timeouts - they artificially slow things 
down.  Even if it requires slightly more computation to identify them 
(in the PE), I consider that you're still ahead because you haven't had 
to wait for a timeout.

btw.  I cant seem to find a url for the ra-api-1.dtd except for inside 
ViewCVS as  http://www.opencf.org/standards/ra-api-1.dtd is broken (as 
are other links to the standards directory)..