I'm Jerry Wohlgemuth.

One reason Ruby makes me happy is rspec. It reads like documentation and it's really easy to mock and stub. Python has the mock library, but this always felt clunky to me.

A common test case is interacting with an external API. Here's an example of mocking a get request in Python, using py.test.

from collections import namedtuple
import json

import mock

from my_api_service import ApiService


@mock.patch('requests.get')
def test_api_service(mock_get):
    fake = namedtuple('Request', 'status_code content'])
    mock_get.return_value = fake(200, json.dumps({'foo': 'bar'}))
    assert ApiService().data == {'foo': 'bar'}

Here's the same spec in rspec:

RSpec.describe ApiService do
  describe '.data' do
    let(:fake) { double(:fake, status_code: 200, content: { foo: bar }.to_json }
    subject { described_class }

    it 'returns foo bar' do
      allow(Net::HTTP).to receive(:get).and_return(fake)
      expect(subject.data).to eq({ foo: 'bar' })
    end
  end
end

I just came across Uber's doubles library, which appears to be about two years old and I wish I discovered this earlier.

Here's that same spec, but using doubles:

from doubles import allow, InstanceDouble
import requests

from my_api_service import ApiService


def test_api_service():
    fake = InstanceDouble('requests.Response')
    fake.status_code = 200
    allow(fake).content.and_return(json.dumps({'foo': 'bar'})
    allow(requests).get.and_return(fake)
    assert ApiService().data == {'foo': 'bar'}

It might look like a lot of boiler plate to allow methods on the InstanceDouble, but this is much safer than the other spec where I just used a namedtuple. From the docs:

The double that’s created will verify itself against an instance of that class. The example above will fail with a VerifyingDoubleError exception, assuming foo is not a real instance method.

doubles makes it a whole lot easier, in my opinion to, to write tests in Python.