Re: Paramset Proposal

Geoffrey.Coram wrote:

>Kevin -
>How does this look for a paramset resolution algorithm:
>
>For each instance,
>1. Find all paramsets for which
>a) the parameters overridden on the instance are parameters of the paramset
>
Not sure I'd make a distinction about how the parameter value gets set
myself - a user may not set it because it has the value they want by
default - and the subsequent different paramset behavior would be hard
to debug.

>b) the parameters of the paramset, with overrides and defaults, are all within the
>allowed ranges
>
Yep.

>c) the localparams of the paramset, computed from parameters, are within the
>allowed range
>
Yep.

>2. choose the paramset which has the fewest number of un-overridden parameters
>3. choose the paramset with the greatest number of range limited localparams
>
OK.

>Even with this algorithm, there is no assurance that the ambiguities of
>the type given in the first case are eliminated. For example, an instance
>with a parameter Alpha might be associated with two paramsets, the first
>accepts instances with parameters Alpha and Beta, and the second with
>parameters Alpha and Gamma. In this case, either paramset could be used.
>However the ambiguity can be eliminated if desired by adding a third
>paramset that accepts instances with only the parameter Alpha.
>
>
Works for me :-)

Kev.

>-Geoffrey
>
>

-- 
Kevin Cameron, CPU Technology, CA 94588, Tel.: (925) 225 4862