Re: Alternative language versioning formalism

On Jun 5, 2008, at 3:49 AM, Marc de Graauw wrote:
> Hi Jonathan,
>
> A general difference is your model does not include receiver state;  
> that is,
> given a message m, the receiver action a is determined by m and the  
> function
> R. This is a bit too coarse for real messaging systems; for instance
> messaging systems typically include a unique message id, so  
> accepting a
> message m and rejecting (or ignoring) the same message m if it has  
> been
> resent later is desired behaviour.

I did not think this was a particularly interesting or problematic  
part of the framework and felt it would unnecessarily complicate the  
presentation. Personally the way I would handle it is with the method  
used in the Scott/Strachey denotational semantics framework, which is  
to say that the "actions" are functions of the state or from state to  
modified state. But unless you are going after some particular  
theorem, going in this direction is an example of what I call  
"diminishing returns" or overformalization.

> Another difference is our approach to a new message m', undefined in  
> v1 of
> the language. In my approach the receiver transforms m' to a known  
> message m
> (possibly through ignoring unknown content) and will exhibit the  
> behaviour
> associated with m, in your approach you define an action by 'adequate
> defaulting', i.e. providing a default action for any unknown message  
> in v1.
> I think your approach may be more fruitful here. I may steal this  
> idea and
> incorporate it into my own formalism to see how this works out.
>
> Both approaches probably suffer from their dependance on behaviour/ 
> action -
> in real messaging systems, a lot of message do not lead to a  
> perceivable
> action, but to a database update, possibly but not necessarily  
> affecting
> future actions. (I.e. store a new customer address, which may or may  
> not be
> used later.) This is one comment David Orchard gave me - what's the
> behaviour of a name language message? I think this could be handled by
> defining some receiver state as well, which influences the outcome  
> of future
> actions, but it may also be a point where abstraction from the 'real  
> world'
> is warranted.

I would say that a database update, or even a change of mind, is a  
kind of action. The approach ought to be that the action space is as  
rich as it needs to be to explain whatever you want to explain, and no  
richer. Leaving it completely unconstrained in the initial  
presentation allows for you to make it as rich or as spare as you like  
in a deeper analysis. This is the usual modularity trick that  
mathematicians and mathematical modelers pull.

> Thanks anyway for this contribution, it is a insightful exercise to  
> compare
> the two approaches.

I should say there are two things I was striving for that I found  
missing in David's and your analyses:

1. I wanted to make sure that every move in the analysis was forced,  
so that every condition was both necessary and sufficient - to make  
sure I didn't miss any possible degrees of freedom.

2. I wanted the central results to be theorems, not axioms - to say  
that to achieve extensibility it is necessary *and sufficient* to do  
such and such. Although the theorems end up being trivial, without  
this the whole presentation risks being (or sounding like) a jumble of  
arbitrary decrees.

I'm not sure how I well I succeeded in these goals, but I think they  
are worthwhile in any modeling effort.
>
> Regards,
>
> Marc de Graauw

Best
Jonathan

Received on Monday, 16 June 2008 18:14:28 UTC