Thanks Arnaud!

I think I've seen that mail thread before, but didn't really understand
much at the time. It makes more sense to me now.

A record of functions with parametric type is kinda neat.


On Wed, Aug 8, 2012 at 3:21 AM, Arnaud Spiwack <
Arnaud.Spiwack@lix.polytechnique.fr> wrote:

> In this particular case this is just a case of existentially quantified.
> You can do it with first-class module or gadt (both are convenient), or
> with a clever encoding which only uses record and polymorphic fields. You
> can read all about it here :
> http://caml.inria.fr/pub/ml-archives/caml-list/2004/01/52732867110697f55650778d883ae5e9.fr.html
>
>
> On 7 August 2012 03:48, Anthony Tavener <anthony.tavener@gmail.com> wrote:
>
>> Embedding "database" features into other modules.
>>
>> First-class modules are allowing me to neatly unpack table
>> implementations into
>> other modules, using local types.
>>
>> This is really quite basic, but I wonder if I could have done this before
>> first-class modules, and without leveraging the object system? I also
>> don't
>> recall seeing first-class modules used for something like this.
>>
>> So, is there another way to do this, aside from the object system? It's
>> very
>> much like creating a basic object.
>>
>> This is an extracted and simplified example...
>> --------------------
>>
>> module Db = struct
>>
>>   (* Signature for a single 'table' in the database, elements of type t.
>> *)
>>   module type S = sig
>>     (* Full implementation relies on a Key module for different key types
>> than 'int' *)
>>     type t
>>     val get : int -> t
>>     val set : int -> t -> unit
>>     val del : int -> unit
>>     val iter : (int -> t -> unit) -> unit
>>     val fold : (int -> t -> 'a -> 'a) -> 'a -> 'a
>>   end
>>
>>   (* Instantiates storage for a table, and returns FC module to interact
>> with the store. *)
>>   let create_with_default (type s) ?(size=19) default =
>>     (* Full implementation is parameterized by Key and Table modules *)
>>     let h = Hashtbl.create size in
>>     let module H = struct
>>       type t = s
>>       let get (id:int) =
>>         try Hashtbl.find h id
>>         with Not_found -> default
>>       let set id (v:t) = Hashtbl.replace h id v
>>       let del id = Hashtbl.remove h id
>>       let iter f = Hashtbl.iter f h
>>       let fold f init = Hashtbl.fold f h init
>>     end in
>>     (module H : S with type t = s)
>>
>> end
>>
>> (* An example table... *)
>> module Location = struct
>>   let unknown = "Unknown"
>>   include (val (Db.create_with_default unknown) : S with type t = string)
>>   (* Location might have a bunch of other functionality as well... *)
>> end
>>
>> (* and basic usage... *)
>> # Location.get 1;;
>> - : Location.t = "Unknown"
>> # Location.set 1 "Mars";;
>> - : unit = ()
>> # Location.get 1;;
>> - : Location.t = "Mars"
>>
>> --------------------
>> Some background on what this is for: (skip unless you're interested!)
>>
>> I use a "component architecture" with most games -- basically a database
>> of
>> properties keyed off "game object IDs". I thought this was a very powerful
>> feature in old MUDs/MUSHs. It's one of the first things I tried making
>> when I
>> started in OCaml, but I had some difficulties and ended up explicity
>> instantiating hashtables or maps in the global context of various modules.
>> Sloppy, but workable. (The reason I had difficulty is because I was
>> trying to
>> create a database of tables which were created at runtime -- not
>> statically
>> known.)
>>
>> Recently I decided to fix this mess. I had many modules, each which
>> tended to
>> have a corresponding "table". Eg. Characteristics, Size, Wounds, Faction,
>> Inventory, etc. So ideally I wanted a convenient way to embed "database"
>> functions into such modules while declaring the implementation of the
>> underlying datastore (hashtable, map, whatever).
>>
>> This might seem a bit ugly from a functional-programming perspective, but
>> I've
>> found components to be quite powerful, and overall helps to constrain
>> where
>> and how mutation happens. "Game state" is generally in flux -- well, it is
>> everything variable, and can be compared closely with save-game state.
>> Most code
>> which doesn't update game state can be functional. Actually, it feels
>> creepy to
>> have a variable assignment in the code, since mutation is generally to
>> game-state
>> and that's handled through a database. So the resulting style is
>> functional+database.
>>
>>  -Tony
>>
>>
>