OOP 设计——什么时候有一个通用的基类 [缺陷设计？]

Question

我有一个简化的设计如下（基本上是一堆处理程序来处理 2 种不同类型的请求：EventRequest 和 SpeechRequest）。下面是伪代码：

class SpeechRequest {sessionId: String; slot: String}
class EventRequest {sessionId: String; event: String}

class SpeechRequestHandler; 
class EventRequestHandler;

class SpeechRequestHandler[A/B/C] extends SpeechRequestHandler {
    - handle(request: SpeechRequest) {      
        doSt(request.slot)
    }
}

class EventRequestHandler[A/B/C] extends EventRequestHandler {
    - handle(request: EventRequest) {       
        doSt(request.event)     
    }   
}

有2个不同的dispatcher为每种类型的请求找到合适的handler，然后转发给handler去处理：

class SpeechDispatcher {
    - handle(request: SpeechRequest) {
        handler: SpeechRequestHandler = findHandlerToHandle(request);
        handler.handle(request);
    }
}

class EventDispatcher {
    - handle(request: EventRequest) {
        handler: EventRequestHandler = findHandlerToHandle(request);
        handler.handle(request);
    }
}

现在，我想重构并创建一个基类/通用类。当然，我想出了这个：

class Request {sessionId: String}
class SpeechRequest extends Request {slot: String}
class EventRequest extends Request {event: String}

class RequestHandler {
    - canHandleRequest(Request): bool
    - handle(Request)
}

class SpeechRequestHandler extends RequestHandler {
    - canHandleRequest(request: Request): bool = request instanceof SpeechRequest
}

class EventRequestHandler extends RequestHandler {
    - canHandleRequest(request: Request): bool = request instanceof EventRequest
}

class SpeechRequestHandler[A/B/C] extends SpeechRequestHandler {
    - handle(Request: request) {
        //need to cast to specific type to extract a certain fields to do some operation
        //!! I feel something is not right because of that
        speechRequest:SpeechRequest = (SpeechRequest)request;
        doSt(speechRequest.slot)

        //other operation can work with base Request object; so it's OK
    }
}

class EventRequestHandler[A/B/C] extends EventRequestHandler {
    - handle(Request: request) {
        eventRequest:EventRequest = (EventRequest)request;
        doSt(eventRequest.event)

        //other operation can work with base Request object; so it's OK
    }   
}

对于所有 SpeechRequestHandler[A/B/C]:handle 函数，我现在需要将 Request 对象转换为 (SpeechRequest) 对象：speechRequest:SpeechRequest = (SpeechRequest)request;

我觉得我的设计有缺陷。如果每个 SpeechRequestHandler 我都需要将对象转换为 (SpeechRequest) 以便我可以对这些信息做一些事情，这是否意味着在这种情况下重构基类没有意义？

请您建议一种更好的方法或设计模式来干净地处理这个问题。

谢谢。

Answer 1

如果你想保持类型安全，你可以通过双重调度来做这种事情。见https://en.wikipedia.org/wiki/Double_dispatch

看起来你想要这样的东西：

interface HandlerSet{
    SpeechHandler getSpeechHandler(Request request);
    EventHandler getEventHandler(Request request);
}

abstract Request {
    ... common stuff ...
    abstract handleWith(HandlerSet handlers);
}
class SpeechRequest extends Request {
    sessionId: String; slot: String
    handleWith(HandlerSet handlers) {
        return handlers.getSpeechHandler(this).handle(this);
    }
}
class EventRequest extends Request {
    sessionId: String; event: String
    handleWith(HandlerSet handlers) {
        return handlers.getEventHandler(this).handle(this);
    }
}

Answer 2

主要的问题是你引入了一个可以被Request的每一种类型调用的方法，但是请求的具体处理程序必须知道具体的类型。

类似的东西

 public void handleRequest(Request request){
      // Find the correct handler
      for(RequestHandler requestHandler : requestHandlers){
           if(requestHandler.canHandle(request)){
               // Opps, the handler needs the concrete type
               requestHandler.handle(request);
           }
      }
 }

在运行时，唯一有足够信息知道哪个处理程序可以处理它的对象是Request。因此，您必须要求请求选择要调用的特定目标方法。

这称为visitor pattern。

public interface Request {
    public void accept(RequestVisitor requestVisitor);
}

public interface RequestVisitor {
    public void visit(SpeechRequest speechRequest);
    public void visit(EventRequest eventRequest);
}

Request 实现现在可以决定必须调用 RequestVisitor 的哪个回调：

public class SpeechRequest implements Request {
    public void accept(RequestVisitor requestVisitor) {
        requestVisitor.visit(this);
    }
}

public class EventRequest implements Request {
    public void accept(RequestVisitor requestVisitor) {
        requestVisitor.visit(this);
    }
}

然后具体的处理程序可以使用具体的请求对象

public class SpeechRequestHandler {
    public void handle(SpeechRequest request) {
        System.out.println(request.getClass().getSimpleName());
    }
}

public class EventRequestHandler {
    public void handle(EventRequest request) {
        System.out.println(request.getClass().getSimpleName());
    }
}

从客户的角度来看，访问者可能会被这样使用：

SpeechRequestHandler speechRequestHandler = new SpeechRequestHandler();
EventRequestHandler eventRequestHandler = new EventRequestHandler();

RequestVisitor requestVisitor = new RequestVisitor() {

    @Override
    public void visit(EventRequest eventRequest) {
        eventRequestHandler.handle(eventRequest);
    }

    @Override
    public void visit(SpeechRequest speechRequest) {
        speechRequestHandler.handle(speechRequest);
    }
};

Request request1 = new SpeechRequest();
Request request2 = new EventRequest();

request1.accept(requestVisitor);
request2.accept(requestVisitor);

但是使用访问者模式通常（但不总是）暗示您试图使界面过于抽象。这就是为什么您丢失了稍后必须重新创建的类型信息。也许你会找到更好的设计。