approval-testing.md

title	author	date	css	highlightTheme
Approval testing legacy code	Zeger Hendrikse	2023-09-29	css/custom.css	github-dark

Refactoring Legacy Code

Using approval tests

by Zeger Hendrikse

 

 

 

 

---

• Learn about legacy code
• Learn about approval testing
• When to use it and when not

 

 

 

---

To me, legacy code is simply code without tests — Michael Feathers

---

Or even better...

Legacy Code is valuable code you’re afraid to change — Nicolas Carlo

---

😱 But we need to modify it 😱

---

The naive: Edit and Pray 🙏

1. Edit the code
2. Test manually
3. Pray you didn't brake anything

  Drawbacks: very risky & stressful

---

The ideal: Write the damn tests ✅

1. Reverse engineer the specs from the code
2. Write automated tests
3. Refactor the code
4. Add your feature

  Drawbacks: very costly and sl-o-o-o-w...

---

The pragmatic: Approval tests 💁

1. Generate an output you can snapshot 📸
2. Use test coverage to find all input combinations ✅
3. Use mutations to verify your snapshots 👽

---

Approval tests

Also known as

• Characterization Tests
• Golden Master
• Snapshot Tests
• Locking Tests
• Regression Tests

---

What we are going to test

Python version

class Calculator():
  @staticmethod
  def addNumbers(x: int, y: int) -> int:
    return x + y

---

Typescript version

export class Calculator {
  public add(x: number, y: number): number {
    return x + y
  }
}

---

How we verify

import unittest
from approvaltests.approvals import verify
from calculator import Calculator

class CalculatorTest(unittest.TestCase):

  def test_main(self):
    # ARRANGE
    x: int = 1
    y: int = 2;
    # ACT
    result = Calculator.addNumbers(x, y)
    # APPROVE
    verify(result)

---

Verification in Typescript

import { Calculator } from "../src/Calculator"
import { verify } from "approvals/lib/Providers/Jest/JestApprovals";

describe("A new Calculator", function() {
    it("adds two numbers", function () {
        const myCalculator  = new Calculator()
        const result = myCalculator.add(3, 4)
        // expect(result).toEqual(7)
        verify(result)
    })
})

---

Even with combinatoric tests 🤩

...
from approvaltests.combination_approvals import verify_all_combinations

class CalculatorTest(unittest.TestCase):

  def test_add_combinatorial(self):
    verify_all_combinations( Calculator.addNumbers, [[1,2], [4,3]])

Caveat

Not available in Typescript library 😱

---

One-click demo

• Navigate to the tools/approval-tests directory
• Go to the language of choice

---

Approval testing use cases

• Code without tests that needs to be changed
• APIs that return JSON or XML
• Complex return objects
• Strings longer than one line

---

You should NOT keep these tests

Problems:

• Existing behavior is captured, bugs included
• Tests will fail whenever behavior changes. Noisy!
• People will get used to just update them
• You can't read them to understand what the code does
• Delete them or have a plan to replace them with unit tests.