Never match against Ruby's default exceptions at your tests

While working in an app using Rails 4 and RSpec's beta version, I came across a weird bug that caused the following backtrace:

Failure/Error: expect {
  expected RSpec::Mocks::VerifyingDoubleNotDefinedError, got #<NoMethodError: undefined method `name' for #<RSpec::Mocks::NamedObjectReference:0x000001026308a0>> with backtrace:
  # ./lib/rspec/mocks/example_methods.rb:182:in `declare_verifying_double'
  # ./lib/rspec/mocks/example_methods.rb:46:in `instance_double'

Going right where the error happened I saw this:

def self.declare_verifying_double(type, ref, *args)
  if RSpec::Mocks.configuration.verify_doubled_constant_names? &&

    raise NameError,
      "#{} is not a defined constant. " +
      "Perhaps you misspelt it? " +
      "Disable check with verify_doubled_constant_names configuration option."

  declare_double(type, ref, *args)

So, pretty obvious, isn't it?

It's trying to call a name method at the double/mock ref since it isn't defined yet, but the double doesn't have a method called name (the correct method would be description).

Now, before going on to fix the issue, we need to write a spec that shows it happening in a controlled environment. I move on to the specs for this specific file and find this:

describe 'when verify_doubled_constant_names config option is set' do
  it 'prevents creation of instance doubles for unloaded constants' do
    expect {
    }.to raise_error(NameError)

  it 'prevents creation of class doubles for unloaded constants' do
    expect {
    }.to raise_error(NameError)

Hey, there is a spec for this behaviour. Why isn't this spec failing?

Well, that's the catch, NameError is the superclass for NoMethodError so the match is, in a way, correct. It definitely raises a NameError but not the NameError we expected it would raise.

In this case, there are many possible solutions, we could match on the exception message, to make sure it definitely generates the exception we would want it to or we can create our own error to symbolise this specific error (and that's what I did when I sent them a PR to fix it).

When you use a custom error to signal that something has gone wrong, it's much less likely that you will get a false positive like this one. You know only your own code would manually raise that error (given all the other code doesn't even know it exists) so you would be pretty much safe from falling for a case like this one.

Matching against exception messages is brittle, error prone (you will have to copy and paste the message in many different places) and leads to hard to evolve code. I had a codebase that had matches on Mongoid error messages and once we upgraded to Mongoid 3.x all of these matches failed. Not because the code wasn't working anymore, mind you, but because the messages had changed. Don't do it, you don't want to be there to fix this.

So, avoid using and matching against Ruby's default exceptions, when you need to raise something, create your own exception classes, it's absurdly simple:

VerifyingDoubleNotDefinedError =

And you end up with better documentation, better tests and prevent unexpected errors like this one. Also, always be as specific as possible, if you have different errors that represent different states for your application, make sure your exceptions reflect that as well. Having a single MyAppError class for everything is hardly any better than raising Exception and StandardError all around.

Comments or questions? Ping me on Twitter!