How can I validate exits and aborts in RSpec?

I am trying to spec behaviors for command line arguments my script receives to ensure that all validation passes. Some of my command line arguments will result in abort or exit being invoked because the parameters supplied are missing or incorrect.

I am trying something like this which isn't working:

# something_spec.rb
require 'something'
describe Something do
    before do
        Kernel.stub!(:exit)
    end

    it "should exit cleanly when -h is used" do
        s = Something.new
        Kernel.should_receive(:exit)
        s.process_arguments(["-h"])
    end
end

The exit method is firing cleanly preventing RSpec from validating the test (I get "SystemExit: exit").

I have also tried to mock(Kernel) but that too is not working as I'd like (I don't see any discernible difference, but that's likely because I'm not sure how exactly to mock Kernel and make sure the mocked Kernel is used in my Something class).

Solution 1:

try this:

module MyGem
  describe "CLI" do
    context "execute" do

      it "should exit cleanly when -h is used" do
        argv=["-h"]
        out = StringIO.new
        lambda { ::MyGem::CLI.execute( out, argv) }.should raise_error SystemExit
      end

    end
  end
end

Solution 2:

Using the new RSpec syntax:

expect { code_that_exits }.to raise_error(SystemExit)

If something is printed to STDOUT and you want to test that too, you can do something like:

context "when -h or --help option used" do
  it "prints the help and exits" do
    help = %Q(
      Usage: my_app [options]
        -h, --help                       Shows this help message
    )

    ARGV << "-h"
    expect do
      output = capture_stdout { my_app.execute(ARGV) }
      expect(output).to eq(help)
    end.to raise_error(SystemExit)

    ARGV << "--help"
    expect do
      output = capture_stdout { my_app.execute(ARGV) }
      expect(output).to eq(help)
    end.to raise_error(SystemExit)
  end
end

Where capture_stdout is defined as seen in Test output to command line with RSpec.

Update: Consider using RSpec's output matcher instead of capture_stdout

Solution 3:

Thanks for the answer Markus. Once I had this clue I could put together a nice matcher for future use.

it "should exit cleanly when -h is used" do
  lambda { ::MyGem::CLI.execute( StringIO.new, ["-h"]) }.should exit_with_code(0)
end
it "should exit with error on unknown option" do
  lambda { ::MyGem::CLI.execute( StringIO.new, ["--bad-option"]) }.should exit_with_code(-1)
end

To use this matcher add this to your libraries or spec-helpers:

RSpec::Matchers.define :exit_with_code do |exp_code|
  actual = nil
  match do |block|
    begin
      block.call
    rescue SystemExit => e
      actual = e.status
    end
    actual and actual == exp_code
  end
  failure_message_for_should do |block|
    "expected block to call exit(#{exp_code}) but exit" +
      (actual.nil? ? " not called" : "(#{actual}) was called")
  end
  failure_message_for_should_not do |block|
    "expected block not to call exit(#{exp_code})"
  end
  description do
    "expect block to call exit(#{exp_code})"
  end
end

Solution 4:

There's no need for custom matchers or rescue blocks, simply:

expect { exit 1 }.to raise_error(SystemExit) do |error|
  expect(error.status).to eq(1)
end

I'd argue that this is superior because it's explicit and plain Rspec.