Test Driven Development in BASH

A talk by torstein @ escenic

Why unit tests?

confirm

that you’ve actually fixed the bug

add

new features without breaking old

re-factor

with confidence

promotes

modular, de-coupled code

TDD makes

you think about the user

How a shell script starts

just

a few lines

mkdir -p /var/lib/app/data
cd /var/lib/app/data
wget http://example.com/big-file.zip

and add some

variables

url=http://example.com/big-file.zip
data_dir=/var/lib/app/data
mkdir -p ${data_dir}
cd ${data_dir}
wget ${url}

hang on

perhaps it takes a parameter?

if [ $1 ]; then
   url=$1
else
  url=http://example.com/big-file.zip
fi

data_dir=/var/lib/app/data
mkdir -p ${data_dir}
cd ${data_dir}
wget ${url}

wait a minute

could it make eggs too?

download_url() {
  data_dir=/var/lib/app/data
  mkdir -p ${data_dir}
  cd ${data_dir}
  wget ${url}
}

make_egg() {
  make "$1"
}

if [[ $1 == "http"* ]; then
  url=http://example.com/big-file.zip
  download_url
else
  make_egg "$1"
fi

of course

I want logging

log_file=/var/log/$(basename $0).log

download_url() {
  data_dir=/var/lib/app/data
  mkdir -p ${data_dir}
  cd ${data_dir}
  wget ${url}
  echo "Downloaded $url to $data_dir ..." >> $log_file
}

make_egg() {
  echo "Making $1 ..." >> $log_file
  make "$1"
}

if [[ $1 == "http"* ]; then
  url=http://example.com/big-file.zip
  download_url
else
  make_egg "$1"
fi

and make sure

it runs as the correct user

log_file=/var/log/$(basename $0).log

download_url() {
  data_dir=/var/lib/app/data
  mkdir -p ${data_dir}
  cd ${data_dir}
  wget ${url}
  echo "Downloaded $url to $data_dir ..." >> $log_file
}

make_egg() {
  echo "Making $1 ..." >> $log_file
  make "$1"
}

if [[ $(whoami) == root ]]; then
  echo "$(basename $0) cannot run as root"
  exit 1
fi

if [[ $1 == "http"* ]; then
  url=http://example.com/big-file.zip
  download_url
else
  make_egg "$1"
fi

Proposed code structure

Break your code up into these folders:

.
├── bin
├── lib
└── test

example

The super-extremely-useful add-two command:

.
├── bin
│   └── add-two
├── lib
│   └── add-two-lib.sh
└── test
    ├── add-two-test.sh
    └── run-tests.sh

xunit

Enter unit testing: xunit

write your own

First, I implemented my own unit testing library…

… and then

I discovered what’s out there 😊

shunit2

shunit2 provides all the methods you know from JUnit:

assertEquals
assertFalse
assertNotEquals
assertNotNull
assertNotSame
assertNull
assertSame
assertTrue
fail
failNotEquals
failNotSame
failSame

## @override shunit2
setUp() {
  source "$(dirname "$0")/../lib/$(basename "$0" -test.sh)-lib.sh"
}

## @override shunit2
tearDown() {
  :
}

test_add_2_to_3_yields_5() {
  local expected=5
  local actual=
  actual=$(add_two 3)
  assertEquals "${expected}" "${actual}"
}

Seeing is believing

Check out this example project: https://gitlab.com/skybert/gone

Continuous integration

When someone checks in code, run all tests to make sure he/she hasn’t broken anything.