by Kevin Murphy
- To Change Or Not To Change
- Consistency Conundrum
Imagine you've just wrapped up a new feature and submitted it for code review, which includes the following method:
def complex_calculation(base_price, promo_code) # calculation end
One of your team members reviews the change, and leaves a suggestion.
You could consider using keyword arguments for this method.
That would change the method to look like this:
def complex_calculation(base_price:, promo_code:) # calculation end
Do you make this change? How do you decide?
Evaluating This Suggestion
From a functionality perspective, this makes no perceived difference. The code will work the same one way or another and neither have any known or noticeable implications, such as performance.
However, this alternative arose from a reaction a team member had who also needs to work on this code, so there must be some value in this suggestion to them. Additionally, this change, one way or another, will end up serving as prior art that may be used to base future decisions on.
The change itself has very limited impact on the code itself, but it does look like there's a possible impact to the team worth exploring.
Any change involves risk - and risk has a cost. For example, here we can't only make the change to the method's signature. We also need to modify any callers to use the keyword arguments. If we miss one, and don't have complete test coverage, we may miss a call site - resulting in a runtime exception when the code is executed in production.
Considering that this is the place this method is introduced, this risk should be clear to mitigate. Any callers will be in this changeset, so we should be able to identify them while reviewing this change. Making this change itself is not particularly risky.
Changing this method to keyword arguments should not take much time - I've already made the change to the method above! Now we only need to change the callers. Even if this were an urgent need to get deployed as soon as possible, it seems reasonable that we could implement this here and now should we choose to.
Using keyword arguments here provides two benefits to callers: readability and flexibility.
I'm going to intentionally obscure variable names to make a point here, but calling our method before looked like this:
def print_calculation_results calculator = Calculator.new result = calculator.complex_calculation(@x, @y) puts result end
Compare this to the keyword args approach:
def print_calculation_results calculator = Calculator.new result = complex_calculation( base_price: @x, promo_code: @y, ) puts result end
It's more verbose, but subjectively, I think it's also more readable and understandable.
It also frees us from caring about the position or order of the arguments. Calling the method this way works as well:
result = complex_calculation( promo_code: @y, base_price: @x, )
I would tend to believe that the clarity and flexibility gained from the keyword arguments would make it easier to maintain code that used this method. It can also prevent against subtle bugs, where you accidentally meant to pass
@x as the base price, but instead you passed
@y first, because you forgot which argument should be in what order.
This seems like a maintenance win!
Keyword arguments were added to the ruby langauge in Ruby 2.0.0, originally released in 2013. In my experience, its usage is not a surprise to most rubyists, nor an unwelcome one. Even if someone is unfamiliar with them, understanding the basic usage of them is well-received.
We'll talk more about consistency at the end.
We have an alternative that has no identified net impact on the function of the code, but that a team member has a stated preference for. The change would not require an intense lift to accommodate, nor maintain. It (subjectively) improves readability and can prevent against subtle errors, and it uses a language feature that's been supported for many, many years.
In a vacuum, this suggestion seems like a net positive. Given our criteria here, we should accept this change!
Different changes will be evaluated against the criteria differently, and the weighted importance of the criteria will change, but we can still use the rubric as a framework for how to make a decision.
For an in-depth explanation of these criteria, I recommend watching the full talk.
Is It That Simple?
I said above that when considering this suggestion in isolation, it seems like a net benefit. However, this code does not exist in a vacuum. It's part of an existing application, and needs to fit in with the rest of the ecosystem. I left out any discussion of consistency with the application itself.
If keyword arguments aren't used anywhere else in the application, should we reject the change because it's inconsistent? What criteria can we use to determine how to maintain consistency and support change? We'll explore that in our next post.