Inline setup versus let in RSpec
let
has a long history in RSpec having been introduced with a commit in 2009 and rolled out as
part of RSpec 2. I adopted it early on as the aesthetics appealed to me with let
variables being defined first in little blocks with symbols. One of
our engineers wrote a macro in vim to quickly hoist instance variables in older style RSpec tests to let declarations. It became the default
style for everyone in RSpec with only a bit of controversy around let!
which could be easy to miss in tests. I knew about the lazy initialization, but
it was the aesthetics I prized.
I remember a bit of a community debate a few years later with a famous Thoughtbot blog post entitled Let’s
Not. It argued pretty convincingly that let was a mystery guest pattern which you don’t want when writng
clear tests. Since our team’s default at that point was to use let, but we wrote pretty small Sandi Metz style classes with a single responsibility,
the let
s weren’t causing a big headache. If the code and the specs were visible in a single editor window, it was easy to see what was going on.
Fast forward to working on one of the largest Ruby codebases in the world and I got reintroduced to the idea of containing the entire test inline.
Much of the codebase doesn’t fit in a single editor window and suddenly those little let
s are screens away and hard to find. Add with nested contexts,
shared examples and the like I was really hating let
. Some of the more painful specs potentially executed up to 100 let
s before running an
individual spec and often fired hundreds of SQL queries. I continued to use it on newer refactored code, but eventually the team started to cut
back to allowing fewer lets in a given spec as a compromise. Today my default is to write the entire context of the test inline with no let
s at all.
So a let heavy spec like the following is forced into:
RSpec.describe BlogPost do
let(:author) { Author.new(name: "Gunther Hemingway") }
let(:editor) { Editor.new(name: "Jordon Adams") }
let(:category) { Category.new(name: "technology") }
let(:comments) do
[
Comment.new(author: "hank", content: "great post!"),
Comment.new(author: "lisa", content: "very informative."),
]
end
let(:tags) { ["ruby", "software development", "tdd"] }
let(:blog_post) do
BlogPost.new(
title: "rspec and let",
content: "using let in rspec can help with...",
author: author,
editor: editor,
category: category,
comments: comments,
tags: tags
)
end
describe "#publish" do
it "notifies the author and editor when published" do
blog_post.publish
expect(author.notifications.last).to eq("your post 'rspec and let' has been published.")
expect(editor.notifications.last).to eq("the post 'rspec and let' you edited has been published.")
end
end
end
This much more explict style:
RSpec.describe BlogPost do
describe "#publish" do
it "notifies the author and editor when published" do
blog_post = BlogPost.new(
title: "rspec and let",
content: "using let in rspec can help with...",
author: Author.new(name: "Gunther Hemingway")
editor: Editor.new(name: "Jordon Adams")
category: category.new(name: "technology"),
comments: [
comment.new(author: "leslie", content: "great post!"),
comment.new(author: "bob", content: "very informative.")
],
tags: ["ruby", "software development", "tdd"]
)
blog_post.publish
expect(blog_post.author.notifications.last).to eq("your post 'rspec and let' has been published.")
expect(blog_post.editor.notifications.last).to eq("the post 'rspec and let' you edited has been published.")
end
end
end
Inline default leads to several happy impacts:
- If you ever need to move this spec, copy it as a starting point, etc, it’s all intact as a single unit
- Sure you’ll probably duplicate some of this for a second spec, but deep in a spec file you’ll never need to scroll around to see what’s going on
- If your setup code looks like many lines of boilerplate perhaps your class has way to many dependencies and needs reactoring
So if you haven’t tried this recently I’d invite you to do an experiment for a week and write zero let
s.
Interview Tip: Checkout Online Code Environments before the Interview
Live coding is a stressful experience. One of the simplest ways to reduce stress around solving a unknown problem in less than an hour is to setup your environment before the interview. My anecdotal experience is that so few developers do this in practice. As an interviewer its a good sign when you check the coding environment before and see the candidate has already setup the environment and got a test up and running.
It’s a signal to the interviewer of several things in the interview:
- You are naturally prepared
- You are able to take full advantage of any tools/libraries provided
- You are pretty motivated to work for this organization
Using CoderPad as an example:
- You can login ahead of time into the pad setup for your interview or at least login to CoderPad’s site and try out a sandbox.
- Select a few of the coding environments you are most comfortable with. Unless you are forced into a specific language always choose your day-to-day language.
- See what libraries you have access to like RSpec, ActiveSupport in Ruby for example.
- Look at the settings for including important things like vim/emac support, Intellisense, etc.
- Get a test running in the environment, so you can TDD right away using your favorite testing framework or at least an available one.
- Test drive a simple problem in the space like FizzBuzz just to get a feel for the environment and how it works.
- If this is the actual coding environment for the interview leave your code in there for the interviewer to discover.
I know I see this level of preperation maybe 5-10% of the time, so it’s an easy way to start off strong.
Quick Inline Vim Keymap Shortcut
I often forget the exact syntax for writing a quick shortcut in in vim/neovim. After hunting for several minutes this morning I made a decision to memorialize the approach. Usually it’s for a simple command like running tests in an Elixir mix console:
So the first n
means it applies just to normal mode. The noremap
means it won’t follow any other mapping so it’s
non-recursive. The ,e
is the what the shortcut is mapped to. And finally the :!mix test<CR>
puts the mix command
into the terminal and hits a carriage return to execute the command.
Aggregate Failures
After working on several large rails codebases in the last 10 years I’ve seen a familiar pattern.
Many tests in Rails projects are integration tests because they rely on actual database objects existing.
One assertion per test is great rule when you don’t have tens of thousands of specs running. :aggregate-failures
allows you to have multiple assertions while still reporting on each failure clearly.
As a bonus it is honored by Rubocop RSpec RSpec/MultipleExpectations
. Not sure why this isn’t
documented better with Rubocop RSpec. Here is the code within the MultipleExpectations
class that
enforces the one assert per spec rule:
And you don’t need to feel guilty about adding aggregate failures:
- It speeds up test runs because it doesn’t do multiple setups
- By default anything that uses ActiveRecord is not a true unit test
- You still get all the errors if multiple lines fail
- Speed of test run on any significant Rails project should almost always win
RSpec Instance Double With Class Names
After many years working with RSpec I discovered a nice little feature and a small gotcha with instance doubles.
I’ve used instance_double
since before it was ported from rspec-fire.
My practice and all the current Relish RSpec examples of instance_double
use the following format:
It turns out the actual instance_double()
method takes a string representing the class or just the class constant:
And the important part here is the parameter:
doubled_class(String, Class)
No String is required here and class constants are auto-verifying. According to a long discussion
on rspec-mocks this behavior exists to avoid some
auto-loading of classes that aren’t needed so that tests can be a bit faster in some cases. For me this breaks
the expectation that I the mock is actually verifying that the class and methods actually exist. If I
wanted just a pure mock I could just use double
. And for Rails projects that make up a lot of the day to day
paid developer work everything auto-loads anyway. Using class constants is just simpler. If you’re on a legacy
project you can probably just add the config to verify the strings ahead of time anyway with the following:
This example code shows what happens when you make a typo in the string constant name and you don’t have the config to verify set:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
require 'spec_helper'
require_relative '../../lib/users/policy_enforcer'
require_relative '../../lib/users/public_policy'
RSpec.describe Users::PolicyEnforcer do
describe '#allowed?' do
let(:policy_enforcer) { Users::PolicyEnforcer.new(public_policy) }
context 'with correct string instance_double class constant' do
let(:public_policy) { instance_double('Users::PublicPolicy') }
before do
allow(public_policy).to receive(:allowed?)
end
it 'returns true' do
policy_enforcer.allowed?
expect(public_policy).to have_received(:allowed?)
end
end
context 'with typo string instance_double class constant' do
let(:public_policy) { instance_double('Use::PublicPolicy') }
before do
allow(public_policy).to receive(:allowed?)
end
it 'lies and returns true' do
policy_enforcer.allowed?
expect(public_policy).to have_received(:allowed?)
end
end
context 'with a proper class constant instance_double' do
let(:public_policy) { instance_double(Users::PublicPolicy) }
before do
allow(public_policy).to receive(:allowed?)
end
it 'returns true' do
policy_enforcer.allowed?
expect(public_policy).to have_received(:allowed?)
end
end
context 'with an typoed class constant instance_double' do
let(:public_policy) { instance_double(User:PublicPolicy) }
before do
allow(public_policy).to receive(:allowed?)
end
it 'fails because the constant is not defined' do
policy_enforcer.allowed?
expect(public_policy).to have_received(:allowed?)
end
end
end
end
The result is the spec with typo string instance double class constant
on line 22 lies to you. It behaves like a plain old double and allows you to accept methods on classes that don’t exist.