b2btest - Light framework to setup back-to-back test scripts

This package is helpfull to prepare and manage back-to-back tests to get control over software you don't have proper tests yet.

On one side, it includes back2back, a comand line tool that performs back-to-back tests on any software that can be called by command line and produces some output files you want to get warned when they get different. You can define the set of commands and the outputs to watch in a simple YAML file.

On the other side, it includes comparision functions and python-unittests like asserts you can use within your test suites that perform similar workflows, keeping the reference outputs in a folder and named by the testname.

Both the command line tool and the python libraries, are extendable with plugins to get informative diff for several file formats: audio, xml, text, pdf...

Why back-to-back testing?

A Back-to-back tests is a black box tests that just compares that, given an input, you have the same output all the time. Unit testing and Test Driven Development are a quite more preferable strategy to test a piece of software. But often we need to change a piece of software which has been developed without proper testing. A quick way to get control over it is setting up a set of back-to-back tests and then proceeding with the refactoring with a larger level of confidence than having no test at all.

Note of warning: Don't feel too confident by being backed by back2back tests. It is black-box testing, so it is hard to ensure full coverage. You may be changing a behaviour which is not exercised by the b2b test, and not noticing.

Why b2btest?

This software eases the workflow for back-to-back test. When b2b tests are hard to run and maintain, they use to get old and useless. This script automates most tedious back2back related task such as setting up, verifying results, accepting changes, clearing data...

Features

You can put under back2back testing either
- the outcomes (files) of any shell command line (with pipes and so on), or
- any serializable data in a unittest.TestCase method
It is auto-checked, like most Xunit frameworks
It automagically manages the expectation data
On failure, it generates handy data to evaluate the changes by providing diffs and keeping failed and expected data for reference.
Provides a handy command line to accept failed results as proper ones.
When the test turns green or it is accepted all failure related data gets cleared.
Comparators and diff generators can be added for your own file type.
You can set architecture dependant outputs for the same test.

How to install

Just use:

$ pip install b2btest

In order to properly diff audio files you should consider installing the audio extension

$ pip install b2btest_audio

In order to properly diff pdf files:

$ pip install visualpdfdiff

Dependencies

For xml files, it requires lxml module and both libxml2 and libxslt libraries installed

Back2Back testing in your unittest

import unittest
import b2btest # Not used but load a new assertion method for TestCase
import datetime

class MyTest(unittest.TestCase):
	def test_this(self):
		self.assertB2BEqual("data")

	def test_that_willallwaysfail(self):
		self.assertB2BEqual(str(datetime.datetime.now()))

if __name__ == '__main__':
	# acceptMode attribute makes the assert accept the results
	# expectation as new
	if '--accept' in sys.argv:
		sys.argv.remove('--accept')
		unittest.TestCase.acceptMode = True

	unittest.main()

Back2Back testing of cli programs

When you are testing back-to-back the output of a command line, you define a yaml file like this (name it b2bcases.yaml).

#!/usr/bin/env back2back 

datapath: "b2bdata" # Point it to the directory containing your reference data
testcases:

  HelloWorld:
    command: echo Hello World > output.txt
    outputs:
    - output.txt

  AlwaysChanging:
    command: date > output.txt
    outputs:
    - output.txt

  Generate1KHzSine: # Requires the b2btest_audio package
    command: sox -n /tmp/sine.wav synth 1.0 sine  1000.0
    outputs:
    - /tmp/sine.wav

To list the available test cases:

back2back b2bcases.yaml --list

To run them:

back2back b2bcases.yaml

Or to run a specific one by its id:

back2back b2bcases.yaml HelloWorld

The first time you run a test case, it will fail as there is no expectation, you must to check it is valid and accept it. Successive failures means that the behaviour has change. You can accept the new result if the behavioural change is expected.

To accept a concrete case:

back2back b2bcases.yaml --accept HelloWorld

To accept all failing tests:

back2back b2bcases.yaml --acceptall

Often some outputs depend on the computer architecture or the platform (windows, mac...). You can generate an independent expectation file for such architecture or platform by using --arch and --platform options when accepting it.

back2back b2bcases.yaml --accept HelloWorld --arch

Programmatic interface

If you want to generate the test cases progamaticaly, you still are able to use the old python interface. Instead of a yaml file, write python script like this:

#!/usr/bin/python
import sys
from b2btest import runBack2BackProgram

data_path="path/to/b2bdata"
back2BackTests = [
	("myprogram_with_option",
	"./myprogram --option input1.wav input2.wav output.wav otheroutput.wav ", [
		"output.wav",
		"otheroutput.wav",
	]),
	("myprogram_with_other_option",
	"./myprogram --other-option input1.wav input2.wav output.wav ignoredoutput.wav ", [
		"output.wav",
	]),
]
runBack2BackProgram(data_path, sys.argv, back2BackTests)

Save this file as back2back.py, for example, and make it executable.

Use the python script directly with the same command line but without the yaml file.

File type support

While text output is relatively easy to diff, other type of outputs are harder.

To get better diffs for those file types with plugins, for instance:

XML files (included)
PDF files with visualpdfdiff
Audio files with b2btest-audio

Extra advices

Use continuous integration

Put your tests under a continuous integration system such

Travis-CI
BuildBot
TestFarm
CDash

You might be lazy passing tests but bots aren't. Connect your bots to your VCS so they test for every commit.

Keep your expectations up to date

If one b2b test gets red, don't keep it for long, either accept it or roll-back your code. b2b detect changes, but if you are in a change you won't notice whether a second one happens. If your expectation data is backed by a version control system dare to accept wrong expectation data until you fix it. But don't forget.

Change log

1.4.1

Fix: specify testcases to run was not working (has ever? it should!)

1.4.0

Audio plugin separated from this code base as b2btest_audio

1.3.3

Simplified dependency on lxml

1.3.2

diffaudio as console script
diffxml as console script
Fix: entry points for xml and audio plugins
Just markdown README

1.3.1

Updated README

1.3

Avoid larg diffs by telling just the generated file with the failed results
Fix unicode problems in certain python versions
Using older lxml versions for python<3.5

1.2

CLI: Fix: only the first output was actually checked
Plugin based type sensitive diff
Specific diff for XML
XML and Audio diffing now are extras
'extensions' key in yaml testcases to associate custom file extensions to a diff plugin

1.1

Unit test like usage for back-to-back test Python code instead of command line programs.
New commandline tool back2back that takes a yaml file with the test cases definitions.

1.0

First github version
(There were previous unpublished versions)

vokimon/back2back