package ppx_deriving_rpc

  1. Overview
  2. Docs
ppx extension for rpclib

Install

Dune Dependency

Authors

Maintainers

Sources

v5.9.0.tar.gz
md5=177db71621a7aa7a55cbea6c237eb6dc

README.md.html

OCaml-RPC

ocaml-rpc is a library that provides remote procedure calls (RPC) using XML or JSON as transport encodings. The transport mechanism itself is outside the scope of this library as all conversions are from and to strings.

RPC types

An RPC value is defined as follow:

type t =
    Int of int64
  | Int32 of int32
  | Bool of bool
  | Float of float
  | String of string
  | DateTime of string
  | Enum of t list
  | Dict of (string * t) list
  | Null

Generating code

The idea behind ocaml-rpc is to generate type definitions that can be used to convert values of a given type to and from their RPC representations.

In order to do so, it is sufficient to add [@@deriving rpcty] to the corresponding type definition. Hence :

type t = ... [@@deriving rpcty]

This will give a value typ_of_t of type Rpc.Types.typ, which can be used in conjunction with the Rpcmarshal module to:

  • Convert values of type t to values of type Rpc.t:

    let rpc_of_t t = Rpcmarshal.marshal typ_of_t t
    
  • Convert values of type Rpc.t to values of type t :

    let t_of_rpc rpc = Rpcmarshal.unmarshal typ_of_t rpc
    

Optionally, it is possible to have different field name in the OCaml type (if it is a record) and in the dictionary argument (the first elements of Dict):

type t = { foo: int [@key "type"]; bar: int [@key "let"]; } [@@deriving rpcty]

This will replace "foo" by "type" and "bar" by "let" in the RPC representation. This is particularly useful when you want to integrate with an existing API and the field names are not valid OCaml identifiers.

The library also provides the [@@deriving rpc] ppx, which is similar to rpcty, but directly generates the conversion functions.

type t = ... [@@deriving rpc]

will give two functions:

  • A function to convert values of type t to values of type Rpc.t : val rpc_of_t : t -> Rpc.t

  • A function to convert values of type Rpc.t to values of type t : val t_of_rpc : Rpc.t -> (t,string) Result.result

It also supports the @key annotations for having different field names:

type t = { foo: int [@key "type"]; bar: int [@key "let"]; } [@@deriving rpc]

Conversion functions

ocaml-rpc currently support two protocols: XMLRPC and JSON(RPC). Function signatures are:

val Xmlrpc.to_string : Rpc.t -> string
val Xmlrpc.of_string : string -> Rpc.t
val Jsonrpc.to_string : Rpc.t -> string
val Jsonrpc.of_string : string -> Rpc.t

So if you want to marshal a value x of type t to JSON, you can use the following function:

Jsonrpc.to_string (rpc_of_t x)

IDL generator

The Idl module makes it possible to define an abstract interface in OCaml using the following pattern:

module CalcInterface(R : Idl.RPC) = struct
  open R

  let int_p = Idl.Param.mk Rpc.Types.int

  let add = R.declare "add"
      ["Add two numbers"]
      (int_p @-> int_p @-> returning int_p Idl.DefaultError.err)

  let mul = R.declare "mul"
      ["Multiply two numbers"]
      (int_p @-> int_p @-> returning int_p Idl.DefaultError.err)

  let implementation = implement
      { Idl.Interface.name = "Calc"; namespace = Some "Calc"; description = ["Calculator supporting addition and multiplication"]; version = (1,0,0) }
end

Then we can generate various "bindings" from it by passing a module implementing the RPC signature to this functor:

  • OCaml bindings for clients or servers can be generated using one of the GenClient* or GenServer* functors, respectively.

    For example one can generate an RPC client this way:

    module CalcClient :
      sig
        val add :
          (Rpc.call -> Rpc.response) ->
          int -> int -> (int, Idl.DefaultError.t) result
        val mul :
          (Rpc.call -> Rpc.response) ->
          int -> int -> (int, Idl.DefaultError.t) result
      end = CalcInterface(Idl.GenClient ())
    

    The functions in the resulting CalcClient module can be used to call their corresponding RPC methods. CalcClient does not implement the transport mechanism itself, that should be provided by passing an a rpc function of type Rpc.call -> Rpc.response.

    CalcClient.add rpc 4 5 will marshal the parameters 4 and 5 into their RPC representations, construct an Rpc.call, pass that call to the given rpc function, and return either an Ok containing the unmarshalled result or an Error with the error description depending on the response returned by rpc.

    There are variations of the GenClient module:

    GenClientExn raises an exception in case the response indicates a failure, instead of returning a result:

    module CalcClient :
      sig
        val add : (Rpc.call -> Rpc.response) -> int -> int -> int
        val mul : (Rpc.call -> Rpc.response) -> int -> int -> int
      end = CalcInterface(Idl.GenClientExn ())
    

    and GenClientExnRpc allows one to specify the rpc function once when constructing the client module:

    module CalcClient :
      sig
        val add : int -> int -> int
        val mul : int -> int -> int
      end = CalcInterface(Idl.GenClientExnRpc (struct let rpc = rpc end))
    

    Bindings for a server can be generated in a similar way:

    module CalcServer :
      sig
        val add : (int -> int -> (int, Idl.DefaultError.t) result) -> unit
        val mul : (int -> int -> (int, Idl.DefaultError.t) result) -> unit
        val implementation : Idl.server_implementation
      end = CalcInterface(Idl.GenServer ())
    

    The implementations of each RPC method should be specified by passing it to the corresponding function in CalcServer:

    CalcServer.add (fun a b -> Ok (a + b));
    CalcServer.mul (fun a b -> Ok (a * b));
    

    Then we can generate our server from the implementation (in case of GenClient, implementation is unused):

    let rpc : (Rpc.call -> Rpc.response) = Idl.server CalcServer.implementation
    

    Again, the transport mechanism is not implemented by CalcServer. We just get an rpc function that, given an Rpc.call, calls the implementation of that RPC method and performs the marshalling and unmarshalling. It is up to the user of this library to connect this rpc function to a real server that responds to client requests.

    Here we also have a GenServerExn functor, for server implementations that raise exceptions instead of returning a result.

    The rpclib-lwt and rpclib-async packages provide similar client and server generators that use Lwt and Async, respectively.

    The Xmlrpc and Jsonrpc modules can be helpful when implementing the rpc function for an XML-RPC or JSON-RPC client/server: they provide functions for converting rpc requests and responses to/from their respective wire formats.

  • Commandline interfaces can be generated using Cmdlinergen:

    module CalcCli :
      sig
        val implementation :
          unit ->
          ((Rpc.call -> Rpc.response) ->
           (unit -> unit) Cmdliner.Term.t * Cmdliner.Term.info)
          list
      end = CalcInterface(Cmdlinergen.Gen ())
    

    We can use the implementation to construct the CLI. Again, we need to pass an rpc function that knows how to make RPC calls.

    let () =
      let cmds = (List.map (fun t -> t rpc) (CalcCli.implementation ())) in
      let open Cmdliner in
      Term.(exit @@ eval_choice default_cmd cmds)
    
  • Some generators use the output of Codegen. This functor generates a structure that contains information about the methods, their parameters, return types, etc. Currently these generators that use the output of Codegen require the method parameters to be named.

    module CalcInterface(R : Idl.RPC) = struct
      open R
    
      let int_p_1 = Idl.Param.mk ~name:"int1" Rpc.Types.int
      let int_p_2 = Idl.Param.mk ~name:"int2" Rpc.Types.int
      let int_p = Idl.Param.mk Rpc.Types.int
    
      let add = R.declare "add"
          ["Add two numbers"]
          (int_p_1 @-> int_p_2 @-> returning int_p Idl.DefaultError.err)
    
      let implementation = implement
          { Idl.Interface.name = "Calc"; namespace = Some "Calc"; description = ["Calculator supporting addition and multiplication"]; version = (1,0,0) }
    end
    
    module CalcCode :
      sig
        val implementation : unit -> Codegen.Interface.t
      end = CalcInterface(Codegen.Gen ())
    
    let interfaces = Codegen.Interfaces.create
      ~name:"calc"
      ~title:"Calculator"
      ~description:["Interface for a Calculator"]
      ~interfaces:[CalcCode.implementation ()]
    
    • Markdowngen can generate a markdown file documenting these interfaces:

      let md = Markdowngen.to_string interfaces
      
    • Pythongen can generate Python code that contains various classes wrapping a Python implementation, providing typechecking & method dispatch, and a CLI.

      let code = Pythongen.of_interfaces interfaces |> Pythongen.string_of_ts
      

The possibilities are not limited to the above generators provided by ocaml-rpc. Any third-party module implementing the RPC signature can be used to generate something from an interface defined following the above pattern. For example, it is possible to write an RPC implementation that generates a GUI for a given interface.