Dependency Inversion Principle
The Dependency Inversion Principle has been proposed by Robert C. Martin. It
states that:
High level modules should not depend upon low
level modules. Both should depend upon abstractions.
Abstractions should not depend upon details. Details should depend upon
abstractions.
This principle seeks to "invert" the conventional notion that high level
modules in software should depend upon the lower level modules. The principle
states that high level or low level modules should not depend upon each other,
instead they should depend upon abstractions. This is best illustrated by an
example.
We will consider the design of a three layer protocol stack. The discussion
is divided into three steps:
The following code shows the design of a three
layered radio link control (RLC) protocol stack. The three layers of the
protocol are:
- RLC Physical Layer
- RLC Datalink Layer
- RLC Network Layer
All three layers are modeled as classes. A code
skeleton for these classes is presented below. This is not a good design. It
has the following limitations:
- All layers are dependent upon each other. The
code for each layer needs to include the header files for all other
layers.
- Layer relationships are hard-coded. You cannot
assemble new protocol by just assembling a new set of layers.
- Design changes in one layer could impact the
design of other layers.
The solution to these problems is to modify the
design of layers by basing it on a common abstraction. This is the subject of
the
next section.
|
Layers |
class RLC_Physical_Layer
{
RLC_Datalink_Layer *m_p_RLC_Datalink_Layer;
public:
void Device_Transmit(Datagram *p_Datagram)
{
Apply_Physical_Layer_Headers(p_Datagram);
Write_Message_To_Device(p_Datagram);
}
void Handle_Device_Receive(Datagram *p_Datagram)
{
Remove_Physical_Layer_Header(p_Datagram);
m_p_RLC_Datalink_Layer->Handle_Physical_Layer_Receive(p_Datagram);
}
};
class RLC_Datalink_Layer
{
RLC_Physical_Layer *m_p_RLC_Physical_Layer;
RLC_Network_Layer *m_p_RLC_Network_Layer;
public:
void Datalink_Transmit(Datagram *p_Datagram)
{
Process_Datalink_Transmit(p_Datagram);
m_p_RLC_Physical_Layer->Device_Transmit(p_Datagram);
}
void Handle_Physical_Layer_Receive(Datagram *p_Datagram)
{
Process_Datalink_Receive(p_Datagram);
m_p_RLC_Network_Layer->Handle_Network_Layer_Receive(p_Datagram);
}
};
class RLC_Network_Layer
{
RLC_Datalink_Layer *m_p_RLC_Datalink_Layer;
Application_Layer *m_p_Application_Layer;
public:
void Network_Transmit(Datagram *p_Datagram)
{
Process_Network_Layer_Transmit(p_Datagram);
m_p_RLC_Datalink_Layer->Datalink_Transmit(p_Datagram);
}
void Handle_Datalink_Layer_Receive(Datagram *p_Datagram)
{
Process_Network_Layer_Receive(p_Datagram);
m_p_Application_Layer->Handle_Application_Receive(p_Datagram);
}
};
|
We now apply the Dependency Inversion Principle to the the above code. We define an abstract class Protocol_Layer that represents a generic layer. Important elements of the Protocol_Layer abstraction are:
- The constructor allows you to associate a layer two more layers, viz. the
upper layer and the lower layer.
- A layer receives a message for transmission and invokes the layer specific
processing. After layer specific processing, the message is
passed to the lower layer.
- When a message is received from the lower layer, layer specific processing
is invoked. Once layer handling of the message is completed, the upper layer
is invoked.
The three layers in the above example, will now inherit from the
Protocol_Layer. This design completely decouples the three layers. All the
layers depend upon the abstraction but not on each other. The advantages are
described in the next section.
|
Protocol_Layer Abstraction |
// Protocol_Layer abstraction has been defined to decouple the
// different protocol layers. Now the layers only depend upon
// this abstraction.
class Protocol_Layer
{
// Each layer can associate with an upper and lower layer.
// This association is also in terms of the Protocol_Layer
// class. Thus there is no dependency on the actual types
// of individual layers. The only requirement is that they
// should inherit from Protocol_Layer.
Protocol_Layer *m_p_Lower_Layer;
Protocol_Layer *m_p_Upper_Layer;
// Each inheriting layer must override the following methods.
// These methods handle the actual protocol processing.
virtual void Process_Transmit(Datagram *p_Datagram) = 0;
virtual void Process_Receive(Datagram *p_Datagram) = 0;
public:
// Create a protocol layer with the associated upper and
// lower layers.
Protocol_Layer(Protocol_Layer *p_Lower_Layer,
Protocol_Layer *p_Upper_Layer)
{
m_p_Lower_Layer = p_Lower_Layer;
m_p_Upper_Layer = p_Upper_Layer;
}
// Process and transmit the datagram passed by higher layer.
void Transmit(Datagram *p_Datagram)
{
Process_Transmit(p_Datagram);
if (m_p_Lower_Layer)
{
m_p_Lower_Layer->Transmit(p_Datagram);
}
}
// Receive handler for a datagram received from lower layer
void Handle_Receive(Datagram *p_Datagram)
{
Process_Receive(p_Datagram);
if (m_p_Upper_Layer)
{
m_p_Upper_Layer->Handle_Receive(p_Datagram);
}
}
};
class RLC_Physical_Layer : public Protocol_Layer
{
void Process_Transmit();
void Process_Receive();
public:
RLC_Physical_Layer(Layer *p_Lower_Layer, Layer *p_Upper_Layer)
: Protocol_Layer(p_Lower_Layer, p_Upper_Layer)
{ }
};
class RLC_Datalink_Layer : public Protocol_Layer
{
void Process_Transmit();
void Process_Receive();
public:
RLC_Datalink_Layer(Layer *p_Lower_Layer, Layer *p_Upper_Layer)
: Protocol_Layer(p_Lower_Layer, p_Upper_Layer)
{ }
};
class RLC_Network_Layer : public Protocol_Layer
{
void Process_Transmit();
void Process_Receive();
public:
RLC_Network_Layer(Layer *p_Lower_Layer, Layer *p_Upper_Layer)
: Protocol_Layer(p_Lower_Layer, p_Upper_Layer)
{ }
};
|
Application of the Dependency Inversion Principle has given us the following
advantages:
- There is no coupling between the three layers. Each layer's implementation
does not even need to include the header files for other layers.
- You have flexibility in mixing and matching different layers. You could
use the physical layer implementation with a different datalink layer. This
can be accomplished without writing a single line of code in the physical
layer class.
- You can sandwich layers between two layers. For example, you can add a
debugging layer between physical and datalink layers. This layer could trap
messages between the two layers.
For more details about the design, please refer to the
Protocol Layer Design Pattern
article.