Re: [boost] [castor] Interest in Logic Paradigm for C++ ?

Subject: Re: [boost] [castor] Interest in Logic Paradigm for C++ ?
From: Roshan (roshan_naik_at_[hidden])
Date: 2010-05-08 13:00:14

Hi Will,

On Fri, 07 May 2010 10:14:30 -0600, William Byrd <webyrd_at_[hidden]> wrote:
> I'm a newbie C++ programmer, but I have experience designing logic  
> programming languages (e.g., miniKanren in Scheme).

Very cool. BTW, some aspects of Castor's design fell in place while I was  
reading
your book "Reasoned Schemer". It surprises me as I could never see any  
direct
co-relation between what I was reading and what I implemented. :-) Perhaps  
it might be
more evident to you.

> I'm curious why Castor doesn't implement full unification.

It can be implemented in C++ but would impact the performance of all LP  
code even if did not
require full semantics. In order to support dynamically binding lrefs that  
were not
already bound to each other, the only technique I could think of was ...  
to add an extra
level of indirection inside lref to point to its data. Basically anytime  
you read or modify the
another level of dereferencing would be required internally. So I decided  
to defer supporting these
semantics to some point in the future... when either this limitation  
proves too severe (making the
performance issue acceptable), or until a way to implement these semantics  
without hurting
performance becomes apparent.

I think I should address this topic in the design document.

> There are many interesting logic programs that can't be written  
> naturally without full unification.  (My dissertation contains many such  
> programs, so I'm biased!)

I fully agree with you. Personally, I would also love to see this happen  
without impacting performance.
I was hoping to investigate this further with help from some LP expert. I  
would like to study
these examples you mention.

> Is Castor aiming towards the Mercury model, with the idea of adding mode  
> annotations, etc., in the future?  That might be a nice sweet spot for  
> the Boost libraries.

  Not sure what you mean by mode annotations. I am not familiar with  
Mercury.

> Otherwise, I would prefer to see full unification in any official Boost  
> logic programming language/library.  How hard would it be to add full  
> unification to Castor, or to at least allow full unification as an  
> option?

I don't have any ideas at the moment. If you have some promising
suggestions (no hurry!) in that space I am willing to pursue them.

-Roshan