Testing command-line apps with Cucumber: advice?

james-calivar · October 3, 2009, 12:02pm

Hi all,

I’d like to write some Cucumber features for a command-line app I’m
working
on (specifically, GitHub - jcoglan/claw: Command-line tool for searching and opening files). The app runs a
Readline
loop taking user input, and uses this input to search for and open files
by
running shell commands (grep, gedit, firefox etc.). Right now it’s using
#puts and the backtick (`) method for output and shell commands, and I’m
not
sure how to go about testing the app’s output. Do I mock Kernel#puts,
should
I give the app an IO object to write output to so I can collect it, or
soemthing else entirely?

If anyone has advice about how to structure this I’d be most grateful.

Cheers,
James

james-calivar · October 15, 2009, 12:08am

James C.:

I’d like to write some Cucumber features
for a command-line app I’m working on [â€¦]

Do I mock Kernel#puts, should I give the app an IO object to
write output to so I can collect it, or soemthing else entirely?

When I was faced with the same problem (albeit with RSpec rather than
Cucumber, but I doubt it matters here) I ended up with a very thin
wrapper (which I donâ€™t test) that simply calls Executable.new.run and
then test the Executable with hand-crafted params and StringIOs:

github.com

Chastell/art-decomp/blob/master/bin/art-decomp

#!/usr/bin/env ruby

require_relative '../lib/art-decomp'
require_relative '../lib/art-decomp/executable'

ArtDecomp::Executable.new(ARGV).run

github.com

Chastell/art-decomp/blob/master/lib/art-decomp/executable.rb

require 'fileutils'
require 'optimist'

require_relative 'logging'

module ArtDecomp
  class Executable
    def initialize(args, opts = { logging_class: Logging }) # rubocop:disable AbcSize, CyclomaticComplexity, MethodLength, PerceivedComplexity
      options = Optimist.options args do
        opt :archs, 'Target architecture(s)', required: true, type: :strings
        opt :dir,   'Results directory',      required: true, type: :string
        opt :log,   'Logging level',          type: :string
        opt :uv,    'UV generator(s)',        default: ['RelativeRelevance']
        opt :qu,    'Qu generator(s)',        default: ['EdgeLabels']
        opt :qv,    'Qv generator(s)',        default: ['GraphColouring']
      end

      Optimist.die 'no FSM given'       if     args.empty?
      Optimist.die 'FSM does not exist' unless File.exist? args.first

This file has been truncated. show original

github.com

Chastell/art-decomp/blob/master/spec/art-decomp/executable_spec.rb

require_relative '../spec_helper'
require_relative '../../lib/art-decomp/executable'

module ArtDecomp # rubocop:disable ModuleLength
  describe Executable do
    let(:dir_path) { Dir.mktmpdir }
    let(:fsm_path) { 'spec/fixtures/fsm' }
    let(:min_args) { ['--archs', '5/1', '--dir', dir_path, '--', fsm_path] }

    let(:executable) { Executable.new min_args }

    after { FileUtils.rmtree dir_path }

    describe '.new' do
      it 'gives help when asked' do
        lambda do
          -> { Executable.new ['--help'] }.must_raise SystemExit
        end.must_output <<-end.gsub(/^ {10}/, '')
          Options:
            -a, --archs=<s+>    Target architecture(s)

This file has been truncated. show original

As you can see, Executable#initializeâ€™s param defaults to ARGV and
the spec redirects $stderr to a StringIO object to inspect it for the
desired errors.

The same with my other command line
app, this time with input and output:

github.com

Chastell/signore/blob/master/bin/signore

#!/usr/bin/env ruby

require_relative '../lib/signore'
trap(:INT) { exit }
Signore::CLI.new.run

http://github.com/Chastell/signore/blob/master/lib/signore/executable.rb
http://github.com/Chastell/signore/blob/master/spec/signore/executable_spec.rb

Executable#runâ€™s params default to $stdout and $stdin, but the spec uses
StringIOs in their place to be able to â€˜read fromâ€™ and â€˜write toâ€™ the
Executable.

â€” Shot