Paweł Gościcki – Paweł Gościcki

20 years

It has been 20 years since I started this blog. Time flies.

Enums in Ruby on Rails backed by PostgreSQL’s ENUM

Let me present you the way I usually create enums in Rails’ ActiveRecord models. I’ll be utilizing the capabilities of the underlying PostgreSQL database and its ENUM type.

Let’s start with an example Subscription model:

class Subscription < ApplicationRecord
  ACTIVE = 'active'.freeze
  INACTIVE = 'inactive'.freeze
  STATES = [ACTIVE, INACTIVE].freeze

  enum state: {
    active: ACTIVE,
    inactive: INACTIVE
  }

  validates :state, presence: true
end

The above will expect a string type state database field (instead of the default numeric one). Let’s create a migration for it:

class CreateSubscriptions < ActiveRecord::Migration[7.1]
  def change
    reversible do |direction|
      direction.up do
        execute <<-SQL
          CREATE TYPE subscription_state AS ENUM ('active', 'inactive');
        SQL
      end

      direction.down do
        execute <<-SQL
          DROP TYPE subscription_state;
        SQL
      end
    end

    create_table :subscriptions do |t|
      t.column :state, :subscription_state, default: 'active', null: false

      t.timestamps null: false
    end
  end
end

Notice that in the model I have intentionally left out the inclusion validation for the state field. This is because Rails will automatically raise ArgumentError if we try to assign a different value to it. As such, it is nice to automatically rescue from such situations in the ApplicationController:

class ApplicationController < ActionController::Base
  rescue_from ArgumentError, with: :bad_request
  ...

  private

  def bad_request exception
    message = exception.message

    respond_to do |format|
      format.html do
        render 'bad_request', status: :unprocessable_entity, locals: { message: }
      end

      format.json do
        render json: { status: 'ERROR', message: }, status: :unprocessable_entity
      end
    end
  end
end

Let’s add some tests. I’ll be using shoulda-matchers for some handy one-liners:

describe Subscription do
  specify ':state enum' do
    expect(described_class.new).to define_enum_for(:state)
      .with_values(active: 'active', inactive: 'inactive')
      .backed_by_column_of_type(:enum)
  end

  it { is_expected.to allow_values(:active, :inactive).for :state }

  it { is_expected.to validate_presence_of :state }

  describe ':state enum validation' do
    it 'raises ArgumentError when assigning an invalid value' do
      expect { described_class.new.state = 'canceled' }.to raise_exception ArgumentError
    end
  end
end

And an accompanying request spec for a most likely SubscriptionsController:

describe 'API subscriptions requests' do
  describe 'PATCH :update' do
    let(:subscription) { create :subscription }

    let(:params) { Hash[subscription: { state: 'invalid' }] }

    context 'when unsuccessful' do
      it 'responds with :unprocessable_entity with error details in the JSON response' do
        patch(subscriptions_path, params:)

        expect(response).to be_unprocessable

        expect(json_response).to be_a Hash
        expect(json_response['status']).to eql 'ERROR'
        expect(json_response['message']).to include 'ArgumentError'
      end
    end
  end
end

Voilà!

Running Puppeteer on AWS Lambda in a Docker container

The aim of this guide is to provide a working solution to generating a PDF version of a webpage using Puppeteer running in a Docker container as a Lambda function. The Docker container approach is used to bypass the 50MB Lambda code size limit. The other option is to use something like chrome-aws-lambda.

We’ll start with the Dockerfile, which assumes Lambda function with a node.js v16 engine called index.js with a named handler export:

FROM public.ecr.aws/lambda/nodejs:16

# Required for puppeteer to run
RUN yum install -y amazon-linux-extras
RUN amazon-linux-extras install epel -y

# Chromium dependencies
RUN yum install -y \
  GConf2.x86_64 \
  alsa-lib.x86_64 \
  atk.x86_64 \
  cups-libs.x86_64 \
  gtk3.x86_64 \
  ipa-gothic-fonts \
  libXScrnSaver.x86_64 \
  libXcomposite.x86_64 \
  libXcursor.x86_64 \
  libXdamage.x86_64 \
  libXext.x86_64 \
  libXi.x86_64 \
  libXrandr.x86_64 \
  libXtst.x86_64 \
  pango.x86_64 \
  xorg-x11-fonts-100dpi \
  xorg-x11-fonts-75dpi \
  xorg-x11-fonts-Type1 \
  xorg-x11-fonts-cyrillic \
  xorg-x11-fonts-misc \
  xorg-x11-utils

RUN yum update -y nss

# Chromium needs to be installed as a system dependency, not via npm; otherwise there will be an error about missing libatk-1.0
RUN yum install -y chromium

COPY index.js package.json package-lock.json ${LAMBDA_TASK_ROOT}

RUN npm ci --omit=dev

CMD [ "index.handler" ]

The above Dockerfile assures all required dependencies are in place. The next step is to setup the Puppeteer’s launch. Here is the relevant snippet from the Lambda function code:

import puppeteer from 'puppeteer'

const viewportOptions = {
  args: [
    // Flags for running in Docker on AWS Lambda
    // https://www.howtogeek.com/devops/how-to-run-puppeteer-and-headless-chrome-in-a-docker-container
    // https://github.com/alixaxel/chrome-aws-lambda/blob/f9d5a9ff0282ef8e172a29d6d077efc468ca3c76/source/index.ts#L95-L118
    // https://github.com/Sparticuz/chrome-aws-lambda/blob/master/source/index.ts#L95-L123
    '--allow-running-insecure-content',
    '--autoplay-policy=user-gesture-required',
    '--disable-background-timer-throttling',
    '--disable-component-update',
    '--disable-dev-shm-usage',
    '--disable-domain-reliability',
    '--disable-features=AudioServiceOutOfProcess,IsolateOrigins,site-per-process',
    '--disable-ipc-flooding-protection',
    '--disable-print-preview',
    '--disable-setuid-sandbox',
    '--disable-site-isolation-trials',
    '--disable-speech-api',
    '--disable-web-security',
    '--disk-cache-size=33554432',
    '--enable-features=SharedArrayBuffer',
    '--hide-scrollbars',
    '--ignore-gpu-blocklist',
    '--in-process-gpu',
    '--mute-audio',
    '--no-default-browser-check',
    '--no-first-run',
    '--no-pings',
    '--no-sandbox',
    '--no-zygote',
    '--single-process',
    '--use-angle=swiftshader',
    '--use-gl=swiftshader',
    '--window-size=1920,1080',
  ],
  defaultViewport: null,
  headless: true,
}

const browser = await puppeteer.launch(viewportOptions)

try {
  const page = await browser.newPage()

  const url = 'https://...'

  await page.goto(url, { waitUntil: ['domcontentloaded', 'networkidle0'] })
  await page.emulateMediaType('print')

  const pdf = await page.pdf({})
} catch (error) {
  ...
}

Testing useNavigate() / navigate() from react-router v6

Testing navigate() is slightly more problematic with the latest v6 (as of writing this post) react-router than just asserting on history.push() as it was the case in the previous versions. Let’s say we have this ButtonHome component:

import { useNavigate } from 'react-router-dom'

const ButtonHome = () => {
  const navigate = useNavigate()

  const onClick = () => navigate('/home')

  return (
    <button onClick={onClick}>
      Home
    </button>
  )
}

I would write a test for this component using the react-testing-library in the following way:

import * as router from 'react-router'
import { render } from '@testing-library/react'
import userEvent from '@testing-library/user-event'

import ButtonHome from './ButtonHome'

describe('ButtonHome', () => {
  const ui = userEvent.setup()
  const navigate = jest.fn()

  beforeEach(() => {
    jest.spyOn(router, 'useNavigate').mockImplementation(() => navigate)
  })

  it('renders the button and navigates to /home upon click', async () => {
    render(withRouter(<ButtonHome />))
    
    await ui.click(screen.queryByText('Home'))

    expect(navigate).toHaveBeenCalledWith('/home')
  })
})

The relevant bits just for testing the router are as follows:

import * as router from 'react-router'

const navigate = jest.fn()

beforeEach(() => {
  jest.spyOn(router, 'useNavigate').mockImplementation(() => navigate)
})

it('...', () => {
  expect(navigate).toHaveBeenCalledWith('/path')
})

The test also requires the following withRouter() helper, which I have in jest.setup.js:

import { Route, Router, Routes } from 'react-router-dom'
import { createBrowserHistory } from 'history'

const history = createBrowserHistory()

const withRouter = (children, opts = {}) => {
  const { path, route } = opts

  if (path) {
    history.push(path)
  }

  return (
    <Router location={history.location} navigator={history}>
      <Routes>
        <Route
          path={route || path || '/'}
          element={children}
        />
      </Routes>
    </Router>
  )
}

global.withRouter = withRouter

Debugging Safari web application on iOS devices

The idea here is to be able to use Safari’s developer tools connected to the web page/application running on a real iOS device. You can achieve similar effect by using XCode’s Simulator app, but certain things don’t work there (looking at you, Apple Pay).

Enable USB internet sharing

Open System Preferences > Sharing
Select Internet Sharing
If the checkbox next to Internet Sharing is enabled, uncheck it
Check iPhone USB on the right side
Check Internet Sharing on the left side. You will have to confirm it
Find message saying: “Computers on your local network can access your computer at: xxxx.local”. The xxxx part is the host. You will need it later.

Make your application available on the new host

Now it’s necessary to make your application available on the new xxxx.local host. In nginx it is a matter of adding xxxx.local to the server_name directive:

server_name existing.server xxxx.local;

For rails server it’s necessary to bind to all local interfaces:

rails server -b 0.0.0.0

Connect your iOS device

on iOS go to Settings -> Safari -> Advanced and toggle “Web Inspector”
on Mac open Safari and go to Preferences -> Advanced and check “Show Develop menu in menu bar”
connect iPhone via USB cable
on Mac restart Safari
on iPhone open http://xxxx.local
on Mac open Safari and go to the Develop menu. You will now see the iOS device you connected to your Mac. Click on it to start debugging.

Troubleshooting

Sometimes Safari won’t show the connected iOS device. If that is the case, you can start with restarting Safari both on the iOS device and Mac. If that doesn’t work, restart the iOS device and then the Mac.

Creating a self-signed, wildcard SSL certificate for Chrome 58+

Chrome 58+ requires Subject Alternative Name to be present in the SSL certificate for the domain name you want to secure. This is supposed to be a replacement for Common Name, which has some security holes (like being able to define a certificate for *.co.uk, which is not possible with SAN).

I’ll be using MacOS and OpenSSL v1.1.1d installed via brew.

Recent OpenSSL versions add the basicConstraints=critical,CA:TRUE x509v3 SAN extension by default, which prevents such generated certificate to work in Chrome 58+. We need to disable that first.

Edit /usr/local/etc/openssl@1.1/openssl.cnf (your path may vary) and comment the following line:

[ req ]
# x509_extensions = v3_ca # The extensions to add to the self signed cert

And then you are off to generate the certificate. I’ll be using the *.example.net domain name here.

/usr/local/Cellar/openssl@1.1/1.1.1d/bin/openssl req \
  -x509 \
  -newkey rsa:4096 \
  -sha256 \
  -days 7000 \
  -nodes \
  -out cert.pem \
  -keyout key.pem \
  -subj "/C=US/O=Org/CN=*.example.net" \
  -addext "basicConstraints=critical,CA:FALSE" \
  -addext "authorityKeyIdentifier=keyid,issuer" \
  -addext "keyUsage = digitalSignature, nonRepudiation, keyEncipherment, dataEncipherment" \
  -addext "subjectAltName=DNS:example.net,DNS:*.example.net"

This will generate two files. key.pem, which is the private key, without passphrase and cert.pem, which is the actual certificate.

Verify that the actual certificate has required x509v3 SAN extensions:

$ openssl x509 -in cert.pem -noout -text

Certificate:
    Data:
        Version: 3 (0x2)
        Serial Number:
            70:4c:28:...
    Signature Algorithm: sha256WithRSAEncryption
        Issuer: C=US, O=Org, CN=*.example.net
        Validity
            Not Before: Oct  2 15:48:10 2019 GMT
            Not After : Dec  1 15:48:10 2038 GMT
        Subject: C=US, O=Org, CN=*.example.net
        Subject Public Key Info:
            Public Key Algorithm: rsaEncryption
                Public-Key: (4096 bit)
                Modulus:
                    00:a7:b5:01...
                Exponent: 65537 (0x10001)
        X509v3 extensions:
            X509v3 Basic Constraints: critical
                CA:FALSE
            X509v3 Authority Key Identifier:
                DirName:/C=US/O=Org/CN=*.example.net
                serial:70:4C:...

            X509v3 Key Usage:
                Digital Signature, Non Repudiation, Key Encipherment, Data Encipherment
            X509v3 Subject Alternative Name:
                DNS:example.net, DNS:*.example.net
    Signature Algorithm: sha256WithRSAEncryption
         59:1d:96:...

The last step is to import the certificate (cert.pem) into the keychain (I’m using the login keychain) and trust it.

So easy. So hard.

Ruby date is (ir)rational

Consider the following:

[DEV] main:0> Date.today
Fri, 05 Apr 2019

[DEV] main:0> (Date.today + 0.5)
Fri, 05 Apr 2019

[DEV] main:0> (Date.today + 0.5) == Date.today
false

The above was executed in a Rails application and was the reason for a quite long wtf moment. Things become more clear when I executed this:

[DEV] main:0> Date.today + 0.5 - Date.today
1/2

Turns out that the Date holds a Rational offset inside of it. This becomes more evident when you execute the above in irb:

irb(main):003:0> Date.today
=> #<date: 2019-04-05 ((2458579j,0s,0n),+0s,2299161j)>

irb(main):004:0> Date.today + 0.5
=> #<date: 2019-04-05 ((2458579j,43200s,0n),+0s,2299161j)>

LEFT OUTER JOIN in ActiveRecord

I always forget how to construct those queries in ActiveRecord, so here it goes.

Assuming we have the following structure:

class User < ActiveRecord::Base
  has_many :authentications
end

class Authentication < ActiveRecord::Base
  belongs_to :user
end

We can generate the LEFT OUTER JOIN SQL query in the following way (to see, for example, if we have dangling user references in authentications):

Authentication.joins('LEFT OUTER JOIN users ON authentications.user_id = users.id').where('users.id IS NULL').where('authentications.user_id IS NOT NULL')

Will generate the following SQL:

SELECT `authentications`.* FROM `authentications` LEFT OUTER JOIN users ON authentications.user_id = users.id WHERE (users.id IS NULL) AND (authentications.user_id IS NOT NULL)

i.e. it will select all authentications with incorrect (dangling) user_id references.

Setting up mocha with sinon and chai

I was unable to quickly find a solution for this, so here’s a little guide on how to set it up together in a proper way.

First, install the libraries:

npm install mocha --save-dev
npm install sinon --save-dev
npm install chai --save-dev

I come from the Ruby world, so I expect to have a spec command, spec_helper.js file and specs living inside spec/ directory (with a nested structure).

Inside package.json file define the spec command:

"scripts" : {
  "spec": "mocha --opts spec/mocha.opts"
}

We will be using BDD style (expect().to()) of chai. Inside spec/mocha.opts add:

--recursive **/*_spec.js
--require spec/spec_helper.js
--ui bdd

Create spec/spec_helper.js, which will require chai and sinon and we will require `spec_helper.js` inside all specs (similarly to how RSpec in Ruby world works).

const sinon = require('sinon')
const expect = require('chai').expect

global.sinon = sinon
global.expect = expect

And now create your spec file (spec/module_spec.js). You should not be required to include any libraries there. Now you can run your specs:

npm run spec

References:

Writing NullObject in Ruby to please Rubocop gods

Lets get right to it. Here’s how one can write a NullObject in a modern, Rubocop-friendly manner:

class NullObject
  def method_missing method, *args, &block
    if respond_to? method
      nil
    else
      super
    end
  end

  def respond_to_missing? _name, _include_private = false
    true
  end
end

One of the things you should support is the fallback to super. I achieve that by the if/else block. In practice super is never reached, so this feels a little bit like a waste.

The other thing is to declare whether your object (NullObject) should respond to methods it doesn’t have. A true null object should respond to all methods, so I set respond_to_missing? to true.

Just for posterity, here’s how you could write specs for it:

describe NullObject do
  it 'returns nil for any method call' do
    null = NullObject.new

    expect(null.missing_method).to be_nil
    expect(null.some_other_missing_method(1, 2, 3)).to be_nil
  end

  it 'responds to missing methods' do
    null = NullObject.new

    expect(null.respond_to?(:missing_method)).to be true
  end
end

Additional reading about the NullObject pattern:

http://wiki.c2.com/?NullObject
http://www.virtuouscode.com/2011/05/30/null-objects-and-falsiness/