99 Bottles of OOP

Note: I'm reading this book together with the POOD-I course.

Main Takeaways

Working on a new feature

Steps to be taken when working on a new feature:

  1. Improve the tests (if needed)
    • good test writing skills?
  2. Refactor
    • Open-Closed principle (from SOLID)
    • Remove code smells (Refactoring book)
    • Repeat until your code is "open" to the new requirement
  3. Implement the new feature

Questions to measure cost of a code

While it's difficult to get exact figures for value and cost, asking the following questions will give you insight into the potential expense of a bit of code:

  1. How difficult was it to write?
  2. How hard is it to understand?
  3. How expensive will it be to change?

Questions 1 ["was it", past] and 3 ["will it", future] may or may not concert you, but question 2 ["is it", present] always applies.

Code is easy to understand when it clearly reflects the problem it's solving


Introduction

Writing code is the process of working your way to the next stable end point, not the end point itself. You don't know the answer in advance, but instead, you are seeking it.

1. Rediscovering Simplicity

When you were new to programming you wrote simple code. (...) Experience has taught you that most code will someday change, and you've begun to craft it in anticipation of that day. (...)

Where you once optimized code for understandability, you now focus on its changeability. Your code is less concrete but more abstract - you've made it initially harder to understand in hopes that it will ultimately be easier to maintain.

This is the basic promise of Object-Oriented Design (OOD): that if you're willing to accept increases in the complexity of your code along some dimensions, you'll be rewarded with decreases in complexity along others. OOD doesn't claim to be free; it merely asserts that its benefits outweigh its costs.

1.1 Simplifying code

The code you write should meet two often-contradictory goals. It must remain concrete enough to be understood while simultaneously being abstract enough to allow for change.

Imagine a continuum with "most concrete" at one end and "most abstract" at the other. (...)

The best solution for most problems lies not at the extremes of this continuum, but somewhere in the middle. There's a sweet spot that represents the perfect compromise between comprehension and changeability, and it's your job as a programmer to find it.

Consistency

Every time the style of the conditionals changes, the reader has to press a mental reset button and start thinking anew. Inconsistent styling makes code harder for humans to parse; it raises costs without providing benefits.

Duplication

Duplication of logic suggests that there are concepts hidden in the code that are not yet visible because they haven't been isolated and named.

Calling methods vs. Sending messages

Referring to the act of invoking behavior as "calling" a method or function suggests (albeit mildly) that you know what that called code does. Alternately, describing this act as "sending a message" leaves a bit more mental space between the sender's intention and the receiver's implementation. This virtual gap encourages you to create message senders that are willfully ignorant of implementation details, and fosters independence between senders and receivers. This independence is a core features of maintainable object-oriented code.

Names

When you first write a piece of code, you obviously know what it does. Therefore, during initial development, the price you pay for poor names is relatively low. However, code is read many more times than it is written, and its ultimate cost is often very high and paid by someone else. Writing code is like writing a book; your efforts are for other readers. Although the struggle for good names is painful, it is worth the effort if you wish your work to survive to be read. Code clarity is built upon names.

Questions to measure cost of a code

While it's difficult to get exact figures for value and cost, asking the following questions will give you insight into the potential expense of a bit of code:

  1. How difficult was it to write?
  2. How hard is it to understand?
  3. How expensive will it be to change?

Questions 1 ["was it", past] and 3 ["will it", future] may or may not concert you, but question 2 ["is it", present] always applies.

Code is easy to understand when it clearly reflects the problem it's solving

Clever code

Programmers love clever code. (...) Writing it, or suddenly understanding it, supplies a little burst of appreciative pleasure. However, this very pleasure distracts the eye and seduces the mind, and allow cleverness to worm its way into inappropriate places.

Does DRY reduce costs?

DRY promises that if you put a chunk of code into a method and then invoke that method instead of duplicating the code, you will save money later if the behavior of that chunk changes. Recognize, though, that DRYing out code is not free. It adds a level of indirection, and layers of indirection make the details of what's happening harder to understand. DRY makes sense when it reduces cost of change more than it increases the cost of understanding the code.

When you DRY out duplication or create a method to name a bit of code, you add levels of indirection that make it more abstract. In theory these abstractions make code easier to understand and change, but in practice they often achieve the opposite.

verses vs. verse

Should verses know how an actual verse is implemented?

While it makes perfect sense for the verse method to be responsible for knowing the verse templates, once verse assumes this responsibility, other parts of your application should not usurp it.

Small steps of TDD

The small steps of TDD act to incrementally reveal the correct implementation. If your absolute certainty turns out to be wrong, skipping these incremental steps means you miss the opportunity of being set right. An apparently "obvious" implementation that is actually an incorrect guess will cause a world of downstream pain.

In Chapter 28 of Test-Driven Development by Example, Kent Beck describes different ways to make tests pass. Three of his "Green Bar Patterns" are:

Fake It style TDD may initially seem awkward and tedious, but with practice it becomes both natural and speedy. Developing the habit of writing just enough code to pass the tests forces you to write better tests. It also provides an antidote for the hubris of thinking you know what's right when you're actually wrong. Although it sometimes makes sense to skip the small steps and jump immediately to the final solution, exercise caution. It's best to save Obvious Implementation for very small leaps.

Sender's amount of knowledge

def song
  verses(99, 0)
end

While it's true that verses(99, 0) and song return the same output, they differ widely in the amount of knowledge they require from the sender.

Knowledge that one object has about another creates a dependency. Dependencies tie objects together, exacerbating the cost of change. Your goal as a message sender is to incur a limited number of dependencies, and your obligation as a method provider is to inflict few.

Cost-effective tests

The first step in learning the art of testing is to understand how to write tests that confirm what your code does without any knowledge of how your code does it.

Avoiding the Echo-Chamber

def test_the_whole_song
  bottles = Bottles.new
  assert_equal bottles.verses(99, 0), bottles.song
end

This test has a major flaw that can cause it to toggle from "short and sweet" to "painful and costly" in the blink of an eye. This flaw lies dormant until something changes, so the benefits of writing tests like this accrue to the writer today, while the costs are paid by an unfortunate maintainer in the future.

There's nothing more frustrating than making a change that preserves the behavior of an application but breaks apparently unrelated tests. If you change an implementation detail while retaining existing behavior and are then confronted with a sea of red, you are right to be exasperated. This is completely avoidable, and a sign that tests are too tightly coupled to code. Such tests impede change and increase costs.

Tests are not the place for abstractions - they are the place for concretions.


2. Test Driving Shameless Green

A conditional adds a new execution path through the code, and additional execution paths increase complexity.

I found this very impactful:

as tests get more specific, code should become more generic. Code becomes more generic by becoming more abstract.

Questions to ask before DRYing

Writing the minimum amount of code to pass the tests

The goal is to use green to maximize your understanding of the problem and to unearth all available information before committing to abstractions.


3. Unearthing Concepts

See also: Code Smells

SOLID

Open-Closed

Objects should be open for extension, but closed for modification.

The decision about whether to refactor in the first place should be determined by whether your code is already "open" to the new requirement.

The "open" principle says that you should not conflate the process of refactoring with the act of adding new features. You should instead separate these two operations. When faced with a new requirement, first rearrange the existing code such that it's open to the new feature, and once that's complete, then add the new code.

New requirements should be implemented in two steps. First, you rearrange existing code so that it becomes open to the new requirement. Next you write new code to meet that requirement. The first of these steps is refactoring.

Flocking Rules to create abstractions

  1. Select the things that are most alike
  2. Find the smallest difference between them.
  3. Make the simplest change that will remove that difference.
    1. parse the new code
    2. parse and execute it
    3. parse, execute and use its result
    4. delete unused code

Following the flocking rules:


From the POOD-I course

Writing the Tests

These are from the POOD-I course.

Question 1: What decisions must be made before you can write this first test?

You have to make lots of decisions! For example, you must design your API, decide on the classes you plan to create and the messages to which they respond, and determine the required inputs and expected outputs for each method.

~ book section 2.2

Question 2: Is it better to interpolate 'number' or to add a conditional?

as the tests get more specific, the code get more generic.

A method containing a single two-branch conditional has two execution paths. (...)
This means that execution paths increase like bacteria dividing:

There's no way that a method like this has a complete test suite.

Question 3: Why not just interpolate a conditional?!?

Interpolating a conditional to decide whether or not to add an 's' to the word bottle in the final phrase of the 2 verse adds complexity to the code without identifying a useful abstraction.

Question 4: Why is duplication in verse acceptable but not in verses?

The Single Responsibility Principle says that every method should have one responsibility, and that it's code should be cohesive around it's purpose.

The responsibility of the verse method is to return the correct lyrics for any verse.

The responsibility of the verses method is to ask the verse method for individual verses, and to concatenate those single verses together.

The verse method owns the lyrics and no other place in your application should be aware of these details.  While there's a bunch of duplication within the method, you cannot see this duplication from outside.  From the point of view of external observers, this duplication is hidden, and may as well not exist...as long as you don't duplication this duplication over in verses!

~ book section 2.6

Question 5: What should the expectation in the song test be?

Whole song

~ book section 2.10

Question 6: What design flaw forces you to write this annoying test?

Techniques to find a good name

Choosing meaningful defaults

Quotes

Slow tests are an indication that there's something wrong with the design. Either something wrong in the tests or something wrong with the code.

As the code is read many more times than it's written, the cost of code is in the reading.

Having a style guide is not about the style, it's about the consistency.

[About Liskov Substitution Principle]

The point of Liskov is: the objects have to be what they say they are. They have to behave like you expect. They can't do anything that forces folks that interact with them to check what kind of thing they are in order know how to talk to them.