English version
Accueil     À propos     Téléchargement     Ressources     Contactez-nous    

Ce site est rarement mis à jour. Pour les informations les plus récentes, rendez-vous sur le nouveau site OCaml à l'adresse ocaml.org.

Browse thread
objects: the lack of method overloading (looking for a way around it)
[ Home ] [ Index: by date | by threads ]
[ Search: ]

[ Message by date: previous | next ] [ Message in thread: previous | next ] [ Thread: previous | next ]
Date: 2010-10-04 (17:33)
From: philippe <kelvin.bitnick@g...>
Subject: objects: the lack of method overloading (looking for a way around it)

I'm just beginning a real, usefull usage of the object system in ocaml.
I'm a bit stalled on one thing, it's the lack of method overloading like in C++.
My need is in a case of building a rather simple pipeline pattern:
each instance of an element (a node) in the pipeline can get a
numerical value from the previous node, do some computation on it, and
store it locally in a hash for caching/efficiency, as returning it
using the #get method.

The real need see most of the time a 1D array of float (a vector) in
input and output of each node.
but sometime a Complex.t array.

So I need some kind of polymorphism for allowing :

float array -> float array
Complet.t array -> float array
float array -> Complex.t array
Complex.t array -> Complex.t array

Here's a shorter (a bit dumb) demo code using int -> int.

(I would much prefer avoid to store float in the real part of a
complex array - in the real (genuine) code the array may be long,
100'000 elements and more, it's code a bit sensitive about

Look to the comment MY QUESTION in the definition of the process
method in the class node_plus


Philippe Strauss

--- 8< ---

open Common
open Log

(* the real code has more usefull type here *)
type node_parameters = Plus1 | Plus2

(* a class for storing parameters usefull for the whole pipeline chain *)
class parameters =
        (* useless here - usefull in the real code *)
        val mutable sf = 44100
        (* a value we will add in a demo node *)
        val mutable add = 1
        method get_sf = sf
        method set_sf sfreq =
            sf <- sfreq ;
        method get_add = add
        method set_add ntoadd =
            add <- ntoadd
        (* this is for the cache part of each node: we compute a hash
of parameter to distinguish hash entries btwn parameters change *)
        method get_parh np (node_id: int) =
            let md5 s = Digest.to_hex (Digest.string s) in
            match np with
            (* the real code doesn't use always (sf, add) but
parameter influing the processing of a node *)
            | Plus1 -> md5 (Marshal.to_string (sf, add) [Marshal.No_sharing])
            | Plus2 -> md5 (Marshal.to_string (sf, add) [Marshal.No_sharing])

(* main code of a node: a boileplate around a hash to cache computed value *)
class virtual node_virt prev parm hn =
        (* instance of a previous node in the pipeline *)
        val previous = prev
        (* instance of the parameters *)
        val parms = parm
        (* the hash for caching *)
        val hres = Hashtbl.create hn
        (* a serial, bumped up plus one on each cache miss *)
        val mutable id = 0
        method get_id = id
        method inc_id =
            id <- id + 1
        (* for passing to (parms #get_parh (self #ptyp ())) in the
real code, when adding a result into the hash *)
        method virtual ptyp: unit -> node_parameters
        (* the real computation is defined in a non virtual method
later, as simply as possible *)
        method virtual process: int -> int
        (* process and store: call process, then store in cache.
called on a cache miss *)
        method private pstore: (int * int) =
            let prid, data = prev #get in
            (* processing *)
            let new_data = self #process data in
            (* we're in a cache miss: bump the serial +1 *)
            self #inc_id ;
            let myid = self #get_id in
            Hashtbl.add hres (parms #get_parh Plus1 myid) new_data ;
            (myid, new_data)
        method get =
                (* memoize id to avoid race cond *)
                let myid = self #get_id in
                (* here again in the real code, we lookup the hash
using (parms #get_parh (self #ptyp ()) myid) *)
                let data = Hashtbl.find hres (parms #get_parh Plus1 myid) in
                (myid, data)
            with Not_found -> Printf.printf "node_virt #get: cache
miss!\n%!"; self #pstore

(* a trivial node: add parm #get_add to the integer we receive from
the previous node *)
class node_plus prev parm hn =
        inherit node_virt prev parm hn
        method ptyp () = Plus1
            MY QUESTION: what is the best way to allow not only int ->
int for this process method,
            but also: int -> float, float -> int, float -> float ?
            The method minimizing the code written in this non virtual
node definition, since I may end with 15 or 30
            differents derived class of node_virt like this single one ?
        method process n =
            n + (parm #get_add)

(* a front of line, head node having a get method of the same type
than those derived from node_virt *)
class node_head parm =
        val parms = parm
        method get =
            (0, 1)

let () =
    let p = new parameters in
    p #set_sf 44100 ; (* useless indeed, just for demo *)
    p #set_add 2 ; (* usefull *)
    let n0 = new node_head p in
    let n1 = new node_plus n0 p 3 in
    Printf.printf "%d\n" (snd (n0 #get)) ;
    Printf.printf "%d\n" (snd (n1 #get)) ;
    Printf.printf "%d\n" (snd (n1 #get)) ;
    p #set_add 1 ;
    Printf.printf "%d\n" (snd (n1 #get))