Refactoring for Readability in Ruby: a Showcase

Dmitrii Krasnov — Thu, 25 Apr 2024 16:07:25 +0000

The problem

Within my 10 years of developer experience, over and over again I see an unnecessarily complex code: methods span tens of lines, contain a ton of low-level details, are hard to understand and very expensive to test. This causes programmers who come to change them due to some new requirements to prefer just adding their stuff to the mix cause, well, "I have work to do, no time for the design". This makes perfect sense: refactoring such code may take days, if not weeks. But it is also true that this makes things even worse with time.

My personal solution to this: write the code as readable as possible from the very beginning. The key skill to do this is an ability to recognize levels of abstraction in your code. Once properly recognized, they can be isolated from each other, which drastically reduces the mental effort required to understand what's going on in the code. There is a so-called "Single Level of Abstraction" (SLA) principle, popularized by Uncle Bob in his book Clean Code, that is all about that.

But, unfortunately, I neither hear much about it in the Ruby world, nor observe its application in practice. Today I had a chance to apply it again while refactoring another piece of code, and decided to write a post about it.

The Showcase

The code below is a business operation that answers the question: "What are the available appointment slots for a given practitioner within a week, starting from a given date?".

Initially, the logic was implemented as a single method within a Practitioner class. That code was working and was fully covered with tests, but is was not very readable, so I decided to refactor it.

I wanted to achieve the following:

1) The logic should be extracted from it's initial context and encapsulated in a separate class.
2) The class should follow the [slightly simplified version of the] coding style I describe in my rails code organization showcase.
3) The logic inside that class should be split into several methods that maintain a single level of abstraction.

With this showcase, I try to demonstrate the importance and the benefits of maintaining a single level of abstraction in your code. Compare the two implementations and ask yourself which you would like to work with

The Setup

The application in general solves a problem of connecting individual doctors with patients.

The part of it we're interested in here has 2 models: Practitioner & Event. Practitioner holds the doctors, Event -- the events related to the practitioner's available time. There are two types of events: "opening" and "appointment":

Events with kind = "opening" indicate that a given practitioner has marked some time window as available for appointments. The time window is always a multiple of 30 minutes, e.g. 10:00 - 11:30 or 10:00 - 10:30. 30 minutes is the size of a standard appointment slot, so a single opening event might contain multiple slots.
Events with kind = "appointment" indicate that a patient has booked a particular time slot, which is always 30 minutes.

Event model has the following fields:

practitioner_id:integer
kind:string
starts_at:datetime
ends_at:datetime

# app/models/practitioner.rb
#
class Practitioner < ApplicationRecord
  has_many :events
  has_many :openings, -> { opening }, class_name: 'Event'
  has_many :appointments, -> { appointment }, class_name: 'Event'
end

# app/models/event.rb
#
class Event < ApplicationRecord
  enum kind: { opening: 'opening', appointment: 'appointment' }
  belongs_to :practitioner
end

The business operation is being called from the controller action, so it's one of the entry points to the application logic.

The Initial Implementation

# app/controllers/api/v1/practitioners/appointment_slots_controller.rb
#
module Api
  module V1
    module Practitioners
      class AppointmentSlotsController < ApplicationController
        def index
          practitioner = Practitioner.find(params[:practitioner_id])
          slots = practitioner.available_slots(params[:date])

          render json: slots, serializer: Api::V1::AppointmentSlotsSerializer
        end
      end
    end
  end
end

# app/models/practitioner.rb
#
class Practitioner < ApplicationRecord
  # ...

  def available_slots(date)
    date = Date.parse(date)
    availabilities = 7.times.to_h { |i| [date + i.days, []] }

    week_openings = openings.where(starts_at: (date..(date + 1.week)))
    week_appointments = appointments.where(starts_at: (date..(date + 1.week)))

    week_openings.each do |opening|
      slots = []

      slot = opening.starts_at
      loop do
        slots << slot

        slot = slot + 30.minutes
        break if slot >= opening.ends_at
      end

      slots = slots.reject { |slot| week_appointments.any? { _1.starts_at == slot } }

      availabilities[opening.starts_at.to_date] += slots
    end

    availabilities
      .transform_keys { |date| date.strftime("%Y-%m-%d") }
      .transform_values { |slots| slots.sort.map { _1.strftime("%H:%M") } }
  end
end

The Refactoring process

After meditating for some time on the initial implementation, I've realized that the problem of getting available time slots might be split into 3 sub-problems:

How to get the slots for a given practitioner?
How to reject the taken slots?
How to format the result?

If I want to maintain a single level of abstraction, the main method of the refactored code should have those three steps, easily readable, and as few other stuff as possible. Let's sketch it:

def call(...)
  slots = get_slots(...)
  available_slots = reject_taken(slots)
  format_result(available_slots)
end

The other important thing to notice is that the initial implementation has the logic of the operation split between the controller action and the Practitioner#available_slots method: controller fetches the practitioner, method does the rest. People are used to writing this kind of code because all rails guides encourage it, so we often do things like this without thinking, but there actually are things to reconsider here. Knowing how to find the resource(s) the operation operates on is conceptually a part of the business operation itself, this logic might become complex with time, so in all but the trivial cases I try to put it into my business operations. Updating the sketch accordingly:

def call(practitioner_id:, ...)
  practitioner = Practitioner.find(practitioner_id)

  slots = get_slots(practitioner, ...)
  available_slots = reject_taken(slots, ...)
  format_result(available_slots, ...)
end

The last thing that substitutes the responsibilities of the operation is the input parsing. User provides the date as a string, controller extracts that from the params, but in my opinion knowing how to parse it should be the business operation's responsibility. In that particular case, it might seem arguable, but if you think of other business operations your system might have, some of them may receive more complex data as an input (e.g. a data from a Stripe callback), knowing how to parse that data is clearly a part of a domain knowledge and therefore it should not stay in the controller. Returning to our example, I choose to parse the date within a business operation cause I believe that the more unification you have in your code base -- the better. But that's a topic for a separate blog, if not the whole conference talk.

With all that said, I update the sketch for the last time:

def call(practitioner_id:, start_date:)
  practitioner = Practitioner.find(practitioner_id)
  start_date = Date.parse(start_date)

  slots = get_slots(practitioner, start_date, ...)
  available_slots = reject_taken(slots, ...)
  format_result(available_slots, ...)
end

Now that I've figured out the high-level structure of the operation and the complete set of it's responsibilities, I can start implementing it. The [personal] conventions I follow here are:

Each business operation gets it's own class named with a verb
Operations are organized into the [subdomain/resource/operation] pattern. I don't mention other subdomains in this blog so I'll skip the subdomain part and just put the code to /app/services, which is a good place for the code you don't know which subdomain to fit into yet.
Operations have the only public method, which is always #call
Operations always get their inputs as keyword arguments
#call is documented with YARD

I also try to identify the parameters hard-coded into the code and extract them into constants for visibility. In this example, I've extracted the length of the interval we fetch and the size of the slot.

The code becomes more readable when the related things are close to each other, so I've moved the fetching of openings/appointments into the low-level methods, closer to where they are being used. Sometimes this can't be done for optimization reasons, but our case is not like that.

And, finally, I try to simplify the code wherever I can, traces of which you may find in the resulting code.

While reading the resulting implementation, I encourage you to ask yourself, how do you feel about it, compared to the initial one. Do you like it? Does this code require less effort to understand? How simple it'll be to change it? What would you make differently? Let me know in the comments below

The Result

# app/controllers/api/v1/practitioners/appointment_slots_controller.rb
#
module Api
  module V1
    module Practitioners
      class AppointmentSlotsController < ApplicationController
        def index
          slots = ::Practitioners::GetAvailableSlots.new.call(
            practitioner_id: params[:practitioner_id],
            start_date: params[:date]
          )

          render json: slots, serializer: Api::V1::AppointmentSlotsSerializer
        end
      end
    end
  end
end

# app/services/practitioners/get_available_slots.rb
#
module Practitioners
  class GetAvailableSlots
    DAYS_TO_FETCH = 7
    SLOT_SIZE = 30.minutes

    # @param practitioner_id [Integer]
    # @param start_date [String]
    #
    # @return [Hash{String=>Array<String>}]
    # @raise [ActiveRecord::RecordNotFound]
    # @raise [Date::Error]
    #
    def call(practitioner_id:, start_date:)
      start_date = Date.parse(start_date)
      practitioner = Practitioner.find(practitioner_id)

      get_slots(practitioner, start_date)
        .then { reject_taken(practitioner, _1) }
        .then { format_result(_1, start_date) }
    end

    private

    def get_slots(practitioner, start_date)
      openings = practitioner.openings.where(starts_at: fetch_interval(start_date))

      openings.flat_map { split_into_slots(_1) }
    end

    def reject_taken(practitioner, slots)
      interval = ((slots.min)..(slots.max))
      appointments = practitioner.appointments.where(starts_at: interval)

      slots.reject { |slot| appointments.any? { _1.starts_at == slot } }
    end

    def format_result(slots, start_date)
      DAYS_TO_FETCH.times.to_h do |i|
        date = start_date + i.days
        day_slots = slots.select { (date...(date + 1.day)).cover?(_1) }

        [date.strftime('%Y-%m-%d'), day_slots.sort.map { _1.strftime('%H:%M') }]
      end
    end

    # Opening might stretch across multiple time slots, in this case we sptit 
    # it into chunks of 30 minutes: opening<starts_at=10:00, ends_at=11:30>
    # becomes [10:00, 10:30, 11:00]
    #
    def split_into_slots(opening)
      num_slots = (opening.ends_at - opening.starts_at).round / SLOT_SIZE
      num_slots.times.map { |i| opening.starts_at + i * SLOT_SIZE }
    end

    def fetch_interval(start_date)
      start_date..(start_date + DAYS_TO_FETCH.days)
    end
  end
end