Kent Beck’s XP Simplicity Rules (aka Four Rules of Simple Design) are a prioritized list of rules that when applied to your code generally yield a great design. As you’ll see from the above link the list has slightly evolved over time. I find today they are usually listed as:
- All Tests Pass
- Don’t Repeat Yourself (DRY)
- Express Intent
- Minimalistic
These are prioritized. If your code doesn’t work (rule 1) then everything else is forfeit. Go back to rule one and get the code working before worrying about anything else.
Over the years the community have debated whether the priority of rules 2 and 3 should be reversed. Some say a little duplication in the code is OK as long as it helps express intent. I’ve debated it myself. This recent post got me thinking about this again, hence this post.
I don’t think it is fair to compare “Expressing Intent” against “DRY”. This is a comparison of apples to oranges. “Expressing Intent” is a principle of code quality. “Repeating Yourself” is a code smell. A code smell is merely an indicator that there might be something wrong with the code. It takes further investigation to determine if a violation of an underlying principle of code quality has actually occurred.
For example “using nouns for method names”, “using verbs for property names”, or “using Booleans for parameters” are all code smells that indicate that code probably isn’t doing a good job at expressing intent. They are usually very good indicators. But what principle is the code smell of Duplication pointing to and how good of an indicator is it?
Duplication in the code base is bad for a couple reasons. If you need to make a change and that needs to be made in a number of locations it is difficult to know if you have caught all of them. This can lead to bugs if/when one of those locations is overlooked. By refactoring the code to remove all duplication there will be left with only one place to change, thereby eliminating this problem.
With most projects the code becomes the single source of truth for a project. If a production code base is inconsistent with a five year old requirements or design document the production code that people are currently living with is usually declared as the current reality (or truth). Requirement or design documents at this age in a project life cycle are usually of little value.
Although comparing production code to external documentation is usually straight forward, duplication within the code base muddles this declaration of truth. When code is duplicated small discrepancies will creep in between the two copies over time. The question then becomes which copy is correct? As different factions debate how the software should work, trust in the software and the team behind it erodes.
The code smell of Duplication points to a violation of the “Single Source of Truth” principle. Let me define that as:
A stakeholder’s requirement for a software change should never cause more than one class to change.
Violation of the Single Source of Truth principle will always result in duplication in the code. However, the inverse is not always true. Duplication in the code does not necessarily indicate that there is a violation of the Single Source of Truth principle.
To illustrate this, let’s look at a retail system where the system will (1) send a transaction to a bank and (2) print a receipt for the customer. Although these are two separate features of the system, they are closely related. The reason for printing the receipt is usually to provide an audit trail back to the bank transaction. Both features use the same data: amount charged, account number, transaction date, customer name, retail store name, and etcetera. Because both features use much of the same data, there is likely to be a lot of duplication between them. This duplication can be removed by making both features use the same data access layer.
Then start coming the divergent requirements. The receipt stakeholder wants a change so that the account number has the last few digits masked out to protect the customer’s privacy. That can be solved with a small IF statement whilst still eliminating all duplication in the system. Then the bank wants to take a picture of the customer as well as capture their signature and/or PIN number for enhanced security. Then the receipt owner wants to pull data from a completely different system to report the customer’s loyalty program point total.
After a while you realize that the two stakeholders have somewhat similar, but ultimately different responsibilities. They have their own reasons for pulling the data access layer in different directions. Then it dawns on you, the Single Responsibility Principle:
There should never be more than one reason for a class to change.
In this example we have two stakeholders giving two separate reasons for the data access class to change. It is clear violation of the Single Responsibility Principle. That’s a problem because it can often lead the project owner pitting the two stakeholders against each other in a vein attempt to get them to work out a mutual single source of truth. But that doesn’t exist. There are two completely valid truths that the developers need to support. How is this to be supported and honour the Single Responsibility Principle? The solution is to duplicate the data access layer and let each stakeholder control their own copy.
The Single Source of Truth and Single Responsibility Principles are very closely related. SST tells you when to remove duplication; SRP tells you when to introduce it. They may seem to be fighting each other, but really they are not. The key is to clearly identify the different responsibilities (or sources of truth) over a system. Sometimes there is a single person with that responsibility, other times there are many. This can be especially difficult if the same person has dual responsibilities. They might not even realize they are wearing multiple hats.
In my opinion Single Source of Truth should be listed as the second rule of simple design with Express Intent at number three. Investigation of the DRY code smell should yield to the proper application SST, without violating SRP. When necessary leave duplication in the system and let the class names express the different people that are responsible for controlling them. Knowing all the people with responsibilities over a system is the higher priority because you’ll need to know this before you can express it. Although it may be a code smell when there is duplication in the code, it does not necessarily mean that the coder has chosen to be expressive over DRY or that the code is bad.