Re: [eclipse-clp-users] Propagators with a state

2009/2/16 Joachim Schimpf <jschimpf_at_users.sf.net>:
> Wit Jakuczun wrote:
>> That could be an interesting addon. And what about priorities?
>> Are you going to standarize them somehow?
>
> Yes, if I can get input for a workable specification - this has been
> discussed at length in the past and proven very tricky, so let's try
> to keep it simple!
>
Right! :)

> One thing we can definitely do is a static mapping from symbolic names
> to numeric priorities (which only gets changed when someone comes up with
> a convincing use case for introducing a new level).
This mapping is a very good step in my opinion. Moreover
I would suggest using only symbolic names in code (as a suggestion
for coding style).

> For ECLiPSe, where delayed goals can be used for things other than
> propagators, I would extend this on both ends as follows:
>
> 1-debug (goals that always succeed and do not affect program semantics)
> 2-check (tests that succeed or fail or abort)
> 3-unary
> 4-binary
> 5-ternary
> 6-linear
> 7-quadratic
> 8-cubic
> 9-subsolver (e.g. the eplex demon)
> 10-mopup (bookkeeping to be done after all propagation, e.g. lib(changeset))
> 11-search (nondeterministic goals)
> 12-main program
>
>From my experience, when I have 2 months for developing a solver
I would need only 3 levels: debug, normal, bookkeeping (as an option).
As a business user I would start twiddling with other priorities only if
there were a real need (from a business point of view: enough time and
money).
But I would love to have other features:
- initial call of a propagator (could result in suspending). This would be
  especially useful for propagators based on demons.
- more declarative way of making propagator atomic then using call_priority.
  I think that a simple argument to suspend called atomic whould do a thing.
  This option would be declaratively equivalent to call_priority(
call(Prop), 0).

I hope that my input would be useful. It is my very subjective point
of view though :).

Best regards
-- 
[ Wit Jakuczun  w.jakuczun_at_wlogsolutions.com ]
[ WLOG Solutions http://wlogsolutions.com   ]