Yes I totally agree, they have different kind. A Normal Rule is * whereas a Meta Rule is * -> *. But I have no experience with typeclasses. That could be what I'm looking for! What they have in common? Well, Id' say that a rule (whatever sort it is) can: - change the state of the game when executed - hence the State Transformer - perhaps assess the legality of *something* (including the legality of another rule) So this gives: class (Monad r) => (Rule r) where     changeState :: StateT Game r ()     legal :: l -> r Bool     legal _ = return True Does that sound good? I will continue experiment with it tomorrow. My other comment inline: On Sat, May 19, 2012 at 12:02 AM, Ben Doyle wrote:

I wonder if you want a typeclass here, rather than a type? A Normal Rule is pretty much a State Transformer, while a Meta Rule seems like a higher-order function on Normal Rules[*]. These are different kinds of things --- and I say "kind" advisedly --- so perhaps better to define the specific commonalities you need than to try to shoehorn them both into one type.

[*]: Possibly related question: Can a Meta Rule depend upon an implementation detail of a Normal rule?

That's a very good question. I think the answer is yes. In other words, does

   rule1 g == rule2 g imply    myMetaRule rule1 == myMetaRule rule2

 ?

So this does not hold. Here myMetaRule is analysing the legality of rule1, possibly analysing the code of rule1 (quite simply for now). (To make thinks clear, the type of "myMetaRule rule1" is a monadic boolean). Thanks! Corentin

Hi, I'm still thinking on how I can express neatly two fundamentally aspects of a rule in Nomic. I'm not a lawyer but in my opinion there is two sorts of rules in the legal system of any countries: - The rules that say *what* is legal or illegal. - The rules that say *how* to enforce the first (from what date, by whom, is it retro-active, what is the penalty if you infringe it etc.). These rules can be an application decree, for example. In computing terms, the first rules can be called "functional rules" whereas the second are "procedural rules". Among the functional rules, there are the "meta rules" as said before, rules that are able to judge the legality of another rule. I'm trying to find an adequate type to embrace all this family. Up to now, I came up with: //functional and procedural rule types newtype FuncRule a = FuncRule {unRule :: a → State Game Bool} type ProcRule = State Game () data RuleFunc2 a = FR (FuncRule a) | PR ProcRule //a meta rule is a functional rule type MetaRule = FuncRule Rule What do you think? I've made a design document of the game if you need more context: https://github.com/cdupont/Nomic/blob/master/Nomic/doc/Haskell%20Nomic%20EN.... At the beginning of the development, the rules where only functional and stateless. That was nice because the engine was able to execute them whenever. Now they are becoming more and more stateful (for example a rule can create a bank account, or create a signal handler for future execution). This make handling them much more complex!! I'm wondering what middle way I can choose. Corentin On Sat, May 19, 2012 at 1:06 AM, Corentin Dupont wrote:

Yes I totally agree, they have different kind. A Normal Rule is * whereas a Meta Rule is * -> *. But I have no experience with typeclasses. That could be what I'm looking for! What they have in common? Well, Id' say that a rule (whatever sort it is) can: - change the state of the game when executed - hence the State Transformer - perhaps assess the legality of *something* (including the legality of another rule)

So this gives: class (Monad r) => (Rule r) where     changeState :: StateT Game r ()     legal :: l -> r Bool     legal _ = return True

Does that sound good? I will continue experiment with it tomorrow. My other comment inline:

On Sat, May 19, 2012 at 12:02 AM, Ben Doyle wrote:

...

I wonder if you want a typeclass here, rather than a type? A Normal Rule is pretty much a State Transformer, while a Meta Rule seems like a higher-order function on Normal Rules[*]. These are different kinds of things --- and I say "kind" advisedly --- so perhaps better to define the specific commonalities you need than to try to shoehorn them both into one type.

[*]: Possibly related question: Can a Meta Rule depend upon an implementation detail of a Normal rule?

That's a very good question. I think the answer is yes.

In other words, does

...

   rule1 g == rule2 g imply    myMetaRule rule1 == myMetaRule rule2

 ?

So this does not hold. Here myMetaRule is analysing the legality of rule1, possibly analysing the code of rule1 (quite simply for now). (To make thinks clear, the type of "myMetaRule rule1" is a monadic boolean).

Thanks! Corentin