How To Measure Modularity

A module is a set of parts that can be used to build a more complex system. How parts can be set or grouped together is based on some considerations. How optimised our module or how good the modularity level of our system is, are our questions.

Several aspects are very common when we want to measure the modularity of our system or software: cohesion, coupling, and connascence.

Cohesion

It is the indicator of whether we efficiently group some parts together. A cohesive module means all parts in the module are well coupled. If we break a cohesive module in our code into pieces or smaller modules, that will lead to an increase in coupling across modules and a decrease in the readability of the code.

There are a few types of cohesion based on the cause of cohesiveness such as functional, sequential (input-output relation), procedural (execution order), logical, or temporal. One that is not strongly related to the functional aspect is logical cohesion. For example, we may group several unrelated functions together into a module just because the functions are similarly used to manipulate a certain data type.

There is a value used to measure cohesion level called LCOM (Lack of Cohesion in Module). The value is equal to the sum of sets of methods that are not shared via shared fields or variables. For example, a class has a function, called X, that only accesses one variable; and also another function, called Y, that only accesses different other variables. This will result in a high value of LCOM which is discouraged.

Coupling

Unlike cohesiveness which sometimes can be quite subjective, coupling can be measured by the number of connections going in and out of a part. Afferent coupling is the number of connections going into a part. Efferent coupling is the number of connections going out to other parts.

There are three attributes related to coupling which are abstractness, instability, and distance from the main sequence.

Abstractness is the sum of abstract elements (e.g. interface, abstract class) divided by the sum of concrete elements (non-abstract components) or actual program code. Too many abstractions can confuse developers in how to work with the code.

Instability is a ratio between the number of efferent coupling and the total of coupling (efferent and afferent). So, instability can be high (close to 1) if efferent coupling is far greater than afferent coupling.

The main sequence is the ideal relationship between abstractness and instability. Let's see the following picture that explains everything.

The value of the distance from the main sequence can be approximated as the absolute value of abstractness + instability - 1. If our code has many abstractions and also the efferent coupling is very high, our code is in the zone of uselessness. If we have a big code base with very little abstraction, the code is in the zone of pain.

Connascence

Connascence is the correlation of one part with other parts in a system concerning maintaining system correctness caused by certain modifications in one part. There are two categories of connascence, static (source-code level coupling) and dynamic (execution-time coupling).

• Static
• Name, like the name of variables
• Type, like the structure of an object
• Meaning, like the meaning of certain constant value
• Position, like the order of function arguments
• Algorithm, like an authorization mechanism
• Dynamic
• Execution, like the order of executing methods of an object
• Timing, like the order of executing two separate processes
• Values, like values in primary and backup databases
• Identity, like objects communicated in a distributed queue

There are three properties related to connascence level which are strength, locality, and degree. Strength indicates the effort needed for a developer to refactor the coupling. Locality indicates how close modules are to each other. The degree indicates the size of the impacts of changes. Let's take a look at the following image that shows the strength of connascence.

Connascence of name is the lowest because of the easiness of refactoring and advances in today's code editor for replacing variables. Connascence of position has higher strength, imagine we change the order of arguments in a function, then we should refactor all parts that call the function.

All three properties should be considered when we design the modularity in our system. When there are far-separated parts that mean a low level of locality, it is better to have lower strength connascence. When the parts are in the same class that means a high level of locality, it is fine to have higher strength connascence. If we have parts with a high strength connascence, it is still fine when the parts are rarely implemented in the code, which means the degree of impact of refactoring is low.

There are some recommendations for improving modularity in a system

• Break the system into several encapsulated elements.
• Minimize connascence across encapsulated elements.
• It is fine to maximize the connascence within the encapsulated element boundary as the consequence of the encapsulation and the minimization of cross-boundary connascence.