By In Object-Oriented Programming, one object will often depend on another object in order to function.
For example, if I create a simple class to run finance reports:
class FinanceReport
def net_income
FinanceApi.gross_income - FinanceApi.total_costs
end
end
We can say that FinanceReport depends on FinanceApi, which it uses to pull information from an external payment processor.
But what if we want to hit a different API at some point? Or, more likely, what if we want to test this class without hitting external resources? The most common answer is to use Dependency Injection.
With Dependency Injection, we don't explicitly refer to FinanceApi inside FinanceReport. Instead, we pass it in as an argument. We inject it.
Using Dependency Injection, our class becomes:
class FinanceReport
def net_income(financials)
financials.gross_income - financials.total_costs
end
end
Now our class has no knowledge that the FinanceApi object even exists! We can pass any object to it as long as it implements gross_income and total_costs.
This has a number of benefits:
- Our code is now less "coupled" to
FinanceApi. - We're forced to use
FinanceApivia a public interface. - We can now pass in a mock or stub object in our tests so that we don't have to hit the real API.
Most developers consider Dependency Injection to be a good thing in general (me too!). However, as with all techniques, there are trade-offs.
Our code is slightly more opaque now. When we explicitly used FinanceApi, it was clear where our values were coming from. It's not quite as clear in the code incorporating Dependency Injection.
If the calls would otherwise have gone to self, then we have made the code more verbose. Instead of using the Object-Oriented "send a message to an object and let it act" paradigm, we find ourselves moving to a more functional "inputs -> outputs" paradigm.
It is this last case (redirecting calls that would have gone to self) that I want to look at today. I want to present a possible alternative to Dependency Injection for these situations: change the base class dynamically (kinda).
The Problem to Solve
Let’s back up a moment and start with the problem that led me down this path to start with: PDF reports.
My client requested the ability to generate various printable PDF reports — one report listing all expenses for an account, another listing revenue, another the forecasted profits for future years, etc.
We’re using the venerable prawn gem to create these PDFs, with each report being its own Ruby object subclassed from Prawn::Document.
Something like this:
class CostReport < Prawn::Document
def initialize(...)
...
end
def render
text "Cost Report"
move_down 20
...
end
So far, so good. But here’s the rub: the client wants an “Overview” report that includes portions from all these other reports.
Solution 1: Dependency Injection
As mentioned previously, one common solution to this kind of problem is to refactor the code to use Dependency Injection. That is, rather than having all these reports call methods on self, we will instead pass in our PDF document as an argument.
This would give us something more like:
class CostReport < Prawn::Document
...
def title(pdf = self)
pdf.text "Cost Report"
pdf.move_down 20
...
end
end
This works, but there is some overhead here. For one thing, every single drawing method now has to take the pdf argument, and every single call to prawn now has to go through this pdf argument.
Dependency injection has some benefits: it pushes us toward decoupled components in our system and allows us to pass in mocks or stubs to make unit testing easier.
However, we are not reaping the rewards of these benefits in our scenario. We are already strongly coupled to the prawn API, so changing to a different PDF library would almost certainly require an entire rewrite of the code.
Testing is also not a big concern here, because in our case testing generated PDF reports with automated tests is too cumbersome to be worthwhile.
So Dependency Injection gives us the behavior we want but also introduces additional overhead with minimal benefits for us. Let’s have a look at another option.
Solution 2: Delegation
Ruby’s standard library provides us SimpleDelegator as an easy way to implement the decorator pattern. You pass in your object to the constructor, and then any method calls to the delegator are forwarded to your object.
Using SimpleDelegator, we can create a base report class that wraps around prawn.
class PrawnWrapper < SimpleDelegator
def initialize(document: nil)
document ||= Prawn::Document.new(...)
super(document)
end
end
We can then update our reports to inherit from this class, and they will still function the same as before, using the default document created in our initializer. The magic happens when we use this in our overview report:
class OverviewReport < PrawnWrapper
...
def render
sales = SaleReport.new(..., document: self)
sales.sales_table
costs = CostReport.new(..., document: self)
costs.costs_pie_chart
...
end
end
Here SaleReport#sales_table and CostReport#costs_pie_chart remain unchanged, but their calls to prawn (e.g., text(...), move_down 20, etc.) are now being forwarded to OverviewReport via the SimpleDelegator we created.
In terms of behavior, we have essentially made it as if SalesReport is now a subclass of OverviewReport. In our case, this means that all the calls to prawn’s API now go SalesReport -> OverviewReport -> Prawn::Document.
How SimpleDelegator Works
The way SimpleDelegator works under the hood is basically to use Ruby's method_missing functionality to forward method calls to another object.
So SimpleDelegator (or a subclass of it) receives a method call. If it implements that method, great; it will execute it just as any other object would. However, it if does not have that method defined, then it will hit method_missing. method_missing will then attempt to call that method on the object given to its constructor.
A simple example:
require 'simple_delegator'
class Thing
def one
'one'
end
def two
'two'
end
end
class ThingDecorator < SimpleDelegator
def two
'three!'
end
end
ThingDecorator.new(Thing.new).one #=> "one"
ThingDecorator.new(Thing.new).two #=> "three!"
By subclassing SimpleDelegator with our own ThingDecorator class here, we can overwrite some methods and let others fall through to the default Thing object.
The trivial example above doesn’t really do SimpleDelegator justice, though. You may look at this code and very well say to me, “Doesn’t subclassing Thing give me the same outcome?”
Yes, yes it does. But here’s the key difference: SimpleDelegator takes the object it will delegate to as an argument in its constructor. This means we can pass in different objects at runtime.
This is what allows use to redirect the calls to a prawn object in Solution 2 above. If we call a single report, the prawn calls go to a new document created in the constructor. The overview report, however, can change this so that calls to prawn are forwarded to its document.
Conclusion
Dependency Injection is probably the best solution to most decoupling problems most of the time.
As with all techniques, however, there are trade-offs. In my case, I didn’t think that the overhead introduced by DI was worth the benefits it provided, so I looked for another solution.
As with all things in Ruby, there’s always another way. I wouldn’t reach for this solution often, but it is certainly a nice addition to your Ruby toolbelt for these situations.
