[Oberon] Protocols (interfaces) in Oberon-2

Mon Oct 26 06:21:12 CET 2020

Chuck

If I understand you correctly you want to add a “mixin” construct to Oberon.

Without modifying Oberon-07 you could do it like this:

I differentiate three module types:
The interface definition module
The interface implementation module(s)
The interface (or mixin) usage object

a)

The interface definition module groups all methods (and data fields) you define for your interface in a RECORD of procedure variables.

MODULE PrinterInterface;

TYPE

  Obj* = POINTER TO ObjDesc;

  ObjDesc* = RECORD

    asString*: PROCEDURE (this: Obj; VAR s: ARRAY OF CHAR);

    val*: INTEGER

  END

END PrinterInterface.

b)

The interface implementation module(s) import the interface definition, implement the functionality AND assigns its implementation to the procedure variables in a procedure called New(). Here an example of the first implementation of the PrinterInterface. Other implementation following the same scheme could be called Printer2, Printer3 and so on.

MODULE Printer1;

IMPORT PrinterInterface;

TYPE Obj* = POINTER TO ObjDesc;

  ObjDesc* = RECORD (PrinterInterface.ObjDesc)

    (* local stuff for this interface implementation *)

  END;

PROCEDURE P2S(this: Obj; VAR s: ARRAY OF CHAR); BEGIN s:=”printer1_x”; s[9] := CHR(this.val MOD 10 + ORD(“0”)) END P2S;

PROCEDURE New*(VAR o: Obj); BEGIN NEW(o); o.asString := P2S END New;

END Printer1.

c)

Now you can add this interface (aka bunch of procedures to be reused) to any object without inheritance by adding a new RECORD field; adding a new field to the record circumvents the problem of ambiguity if two different interfaces define methods with the same name.

Hint: For better readability it is best all users of the interface name this record field identical.

MODULE Usage;

IMPORT P := Printer1; (*or any other implementation of the PrinterInterface*)

TYPE

  User = RECORD

    name: ARRAY 31 OF CHAR;

    age: INTEGER;

    p: P.Obj (* here you add the “mixin”, no inheritance *)

  END;

VAR u: User;

BEGIN

  NEW(u); P.New(u.p); (* initialize the object and the interface *)

  u.p.asString(u.p, u.name) (* call the interface. The syntax “u.asString” possibly generates ambiguity *)

END  Usage.

br

Jörg

Von: Oberon <oberon-bounces at lists.inf.ethz.ch> im Auftrag von Charles Perkins <chuck at kuracali.com>
Antworten an: ETH Oberon and related systems <oberon at lists.inf.ethz.ch>
Datum: Sonntag, 25. Oktober 2020 um 17:16
An: ETH Oberon and related systems <oberon at lists.inf.ethz.ch>
Betreff: Re: [Oberon] Protocols (interfaces) in Oberon-2

Hi Jörg,

If I understand your example correctly, the Interface procedures are declared in the base "Dynamic" module while their implementation resides in a derivative, potentially separately compiled record definition. This is a useful pattern for a self-contained project. It requires you to make every interface implementation a descendant of one record in the "Dynamic" however. Every time you add a new interface you have to recompile the dynamic module and everything that uses it, because the symbol file has changed. In addition, different developers may be adding interfaces independently. Either everyone is making changes to the same "Dynamic" module and their changes have to be merged, or each developer has a different "Dynamic" (or otherwise-named) module and their interfaces do not compose.

I would like to be able to request the same behavior from two completely independent types, and if the interfaces match, perform the function call. The idea I'm going for is something that will allow you to make, for example, an interface that will print anything with a "printer" method, a different interface that will serialize to JSON anything with a "ToJSON" method, etc.

Philosophically though I think it's the same question of why have Oberon-2 with type-bound procedures when the same functionality can be achieved with plain Oberon-07 and procedure variables?  Type-bound procedures introduce complexity in the compiler and module formats that are not strictly necessary.

Just as Oberon-2 is not the same language as Oberon, I think a language that has this 'dynamic trait' capability should be called something else, like perhaps Oberon-2i. 

My two cents...

Chuck

On Sun, Oct 25, 2020 at 8:24 AM Jörg <joerg.straube at iaeth.ch> wrote:

Chuck

In plain Oberon-07, you could solve this as follows: Not the nicest syntax but it seems doable.

MODULE Dynamic;

TYPE

  Obj* = POINTER TO Desc END

  Desc* = RECORD

    toString*: PROCEDURE (o: Obj; VAR a: ARRAY OF CHAR)

  END

END Dynamic.

MODULE Int;

IMPORT Dynamic;

TYPE

  Obj* = POINTER TO Desc;

  Desc* = RECORD (Dynamic.Desc)

    val: INTEGER

  END;

PROCDEURE I2S(o: Obj; VAR a: ARRAY OF CHAR); BEGIN a:=”integer” END I2S;

PROCEDURE New*(VAR o: Obj; i: INTEGER); BEGIN NEW(o); o.toString := I2S; o.val := i END New;

END Int.

MODULE Real;

IMPORT Dynamic;

TYPE

  Obj* = POINTER TO Desc;

  Desc* = RECORD (Dynamic.Desc)

    val: REAL

  END;

PROCDEURE R2S(o: Obj; VAR a: ARRAY OF CHAR); BEGIN a:=”real” END R2S;

PROCEDURE New*(VAR o: Obj; r: REAL); BEGIN NEW(o); o.toString := R2S; o.val := r END New;

END Real.

MODULE Test;

IMPORT Dynamic, Int, Real, Out;

VAR i: Int.Obj; r: Real.Obj; dyn: Dynamic.Obj;

BEGIN

  Int.New(i, 3); dyn := i;

 dyn.asString(dyn, a); Out.String(a); Out.Ln;

  Real.New(r, 3.14); dyn := r;

  dyn.asString(dyn, a); Out.String(a); Out.Ln;

END Test.

Von: Oberon <oberon-bounces at lists.inf.ethz.ch> im Auftrag von Charles Perkins <chuck at kuracali.com>
Antworten an: ETH Oberon and related systems <oberon at lists.inf.ethz.ch>
Datum: Sonntag, 25. Oktober 2020 um 16:03
An: ETH Oberon and related systems <oberon at lists.inf.ethz.ch>
Betreff: Re: [Oberon] Protocols (interfaces) in Oberon-2

Apologies for this : " x0, x1, u: REAL;" was a fragment that snuck in from testing. I have the above code successfully parsing in a fork of Andreas's Extended Oberon compiler. I don't have the method table generation code working yet. It turns out that the Run-time needs to have symbolic type information available, which could be simply loading the smb file alongside the rsc file for code that uses interfaces, or it could involve embedding a hash of the name and parameters of the type-bound procedure in another section of the rsc file. I haven't decided yet.

On Sun, Oct 25, 2020 at 7:33 AM Charles Perkins <chuck at kuracali.com> wrote:

I think Interfaces / Protocols / Dynamic Traits (what Rust calls them) would be a quite useful extension to Oberon. I'm looking at doing it a different way, like this: 

  VAR W: Texts.Writer;

  TYPE 
       I* = POINTER TO IDesc;
       IDesc* = RECORD
            h: INTEGER
       END ;

       R* = POINTER TO RDesc;
       RDesc* = RECORD
            h: REAL
       END ;

       Stringer* = INTERFACE OF
            PROCEDURE String* (VAR a: ARRAY OF CHAR) ; 
       END ;

  PROCEDURE ( i : I ) String* (VAR a: ARRAY OF CHAR) ;
  BEGIN a := "integer"
  END String;

  PROCEDURE ( r : R ) String* (VAR a: ARRAY OF CHAR) ;
  BEGIN a := "real"
  END String;

In the above scheme an Interface looks just like a collection of type-bound procedure definitions with no bodies.

The trick is when it comes time to use the interface, which is when the code needs to know which actual procedure to call based on the record type assigned to it during execution. The record type assigned to an interface could be any record that contains the String type-bound procedure (in this case.) It might be the first method, or the third, or the sixth... Go solves this by generating a dispatch table for the Interface when a type is assigned to it. 

In Oberon that table-making routine could be satisfied by adding another Trap condition in Kernel.Trap much like how New is implemented.

  PROCEDURE Test*;
      VAR i: I; r: R; t: ARRAY 32 OF CHAR; 
        s,s2: Stringer; 
        x0, x1, u: REAL;

  BEGIN 
      NEW(i); NEW(r);

      i.h := 3;
      r.h := 7.5;
      s := i;

      s.Stringer(t); 

      Texts.WriteString(W,t);
      s := r;

      s.Stringer(t);
      Texts.WriteString(W,s);

   END Test;

The above idea for Interfaces builds on the mechanisms already in place in the Oberon-2 compiler and run-time. I think it would be quite useful for allowing a program to choose from multiple implementations of an interface without constraining them to derive from the same base type while still keeping strong static typing and separate linking and loading.

Chuck

On Sun, Oct 25, 2020 at 6:27 AM Luca Boasso <luke.boasso at gmail.com> wrote:

A key feature of protocols / interfaces is the safe multiple inheritance: you can explicitly or implicitly (like in the Go language) implement several interfaces and be type compatible with each one of them.

Do you support something like the following?

TextDesc = RECORD (TextProtocol.TextDesc, WriteProtocol.WriterDesc) END ;  (*this means: “implements TextProtocol.TextDesc AND WriteProtocol.WriterDesc "*)

If this is not supported I don't see this feature being that useful. To support the feature above the implementation is more complicated than Oberon-2's bound procedures. See https://research.swtch.com/interfaces for one way of doing this, or "Efficient implementation of Java interfaces: Invokeinterface considered harmless" 

On Sun, Oct 25, 2020 at 6:46 AM Andreas Pirklbauer <andreas_pirklbauer at yahoo.com> wrote:

Correction: In Text1, it’s TextDesc = RECORD (TextProtocol.TextDesc) of course

—————————

Protocols (sometimes called interfaces) can be added to
Oberon-2 without adding any keywords to the language.

This is one of the key differences to how it is usually defined
and implemented, e.g. in Swift [*] or in Integrated Oberon [**]

Under the new minimalistic design, what distinguishes a protocol
from an actual implementation (of the class) is that in the protocol
definition the implementations of the class methods are simply not
defined. Instead, any module that *imports* a protocol definition
can “adopt” (i.e. implement) it. See the example below.

An experimental implementation showed that if the language
is extended in *this* way, the implementation cost is minimal.

But the question is: Is it worth it? Simplicity of implementation
should of course not be a criteria for adopting a new feature.

Personally, I am rather sceptical of the usefulness of protocols.
But perhaps someone provides a good reason to adopt them.

-ap

Example:

  MODULE TextProtocol;  (*protocol definition*)
    TYPE Text = POINTER TO TextDesc;
      TextDesc = RECORD data*: (*text data*) END ;
      PROCEDURE (t: Text) Insert (string: ARRAY OF CHAR; pos: LONGINT);
      PROCEDURE (t: Text) Delete (from, to: LONGINT);
      PROCEDURE (t: Text) Length (): LONGINT;
  END TextProtocol;

  MODULE Text1; (*one implementation of the Text protocol*)
    IMPORT TextProtocol;
    TYPE Text = POINTER TO TextDesc;
      TextDesc = RECORD (TextProtocol.TextDesc) END ;  (*this means: “implements TextProtocol.TextDesc"*)

    PROCEDURE (t: Text) Insert (string: ARRAY OF CHAR; pos: LONGINT);
    BEGIN (*implementation of Insert*)
    END Insert;

    PROCEDURE (t: Text) Delete (from, to: LONGINT);
    BEGIN (*implementation of Delete*)
    END Delete;

    PROCEDURE (t: Text) Length (): LONGINT;
    BEGIN (*implementation of Length*)
    END Insert;
  END Text1;

  MODULE Text2; (*another implementation of the Text protocol*)
    IMPORT TextProtocol;
    TYPE Text = POINTER TO TextDesc;
      TextDesc = RECORD (TextProtocol.TextDesc) END ;  (*this means: “implements TextProtocol.TextDesc"*)

    PROCEDURE (t: Text) Insert (string: ARRAY OF CHAR; pos: LONGINT);
    BEGIN (*implementation of Insert*)
    END Insert;

    PROCEDURE (t: Text) Delete (from, to: LONGINT);
    BEGIN (*implementation of Delete*)
    END Delete;

    PROCEDURE (t: Text) Length (): LONGINT;
    BEGIN (*implementation of Length*)
    END Insert;
  END Text2;

[*] https://docs.swift.org/swift-book/LanguageGuide/Protocols.html#
[**] https://github.com/io-core/technotes/blob/main/technote014.md

--
Oberon at lists.inf.ethz.ch mailing list for ETH Oberon and related systems
https://lists.inf.ethz.ch/mailman/listinfo/oberon

--
Oberon at lists.inf.ethz.ch mailing list for ETH Oberon and related systems
https://lists.inf.ethz.ch/mailman/listinfo/oberon

-- Oberon at lists.inf.ethz.ch mailing list for ETH Oberon and related systems https://lists.inf.ethz.ch/mailman/listinfo/oberon 

--
Oberon at lists.inf.ethz.ch mailing list for ETH Oberon and related systems
https://lists.inf.ethz.ch/mailman/listinfo/oberon

-- Oberon at lists.inf.ethz.ch mailing list for ETH Oberon and related systems https://lists.inf.ethz.ch/mailman/listinfo/oberon 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.inf.ethz.ch/pipermail/oberon/attachments/20201026/0305bcc1/attachment-0001.html>