shUnit2 is a xUnit unit test framework for Bourne based shell scripts, and it is designed to work in a similar manner to JUnit, PyUnit, etc.. If you have ever had the desire to write a unit test for a shell script, shUnit2 can do the job.

• Introduction
  • Credits / Contributors
  • Feedback
• Quickstart
• Function Reference
  • General Info
  • Asserts
  • Failures
  • Setup/Teardown
  • Skipping
  • Suites
• Advanced Usage
  • Some constants you can use
  • Error Handling
  • Including Line Numbers in Asserts (Macros)
  • Test Skipping
  • Running specific tests from the command line
  • Generating test results in JUnit format
• Appendix
  • Getting help
  • Zsh

shUnit2 was originally developed to provide a consistent testing solution for log4sh, a shell based logging framework similar to log4j. During the development of that product, a repeated problem of having things work just fine under one shell (/bin/bash on Linux to be specific), and then not working under another shell (/bin/sh on Solaris) kept coming up. Although several simple tests were run, they were not adequate and did not catch some corner cases. The decision was finally made to write a proper unit test framework after multiple brown-bag releases were made. Research was done to look for an existing product that met the testing requirements, but no adequate product was found.

Tested Operating Systems (varies over time)

Tested Shells

• Bourne Shell (sh)
• BASH - GNU Bourne Again SHell (bash)
• DASH - Debian Almquist Shell (dash)
• Korn Shell - AT&T version of the Korn shell (ksh)
• mksh - MirBSD Korn Shell (mksh)
• zsh - Zsh (zsh) (since 2.1.2) please see the Zsh shell errata for more information

See the appropriate Release Notes for this release (doc/RELEASE_NOTES-X.X.X.txt) for the list of actual versions tested.

A list of contributors to shUnit2 can be found in doc/contributors.md. Many thanks go out to all those who have contributed to make this a better tool.

shUnit2 is the original product of many hours of work by Kate Ward, the primary author of the code. For related software, check out https://github.com/kward.

Feedback is most certainly welcome for this document. Send your questions, comments, and criticisms via the shunit2-users forum (created 2018-12-09), or file an issue via https://github.com/kward/shunit2/issues.

This section will give a very quick start to running unit tests with shUnit2. More information is located in later sections.

Here is a quick sample script to show how easy it is to write a unit test in shell. Note: the script as it stands expects that you are running it from the "examples" directory.

#! /bin/sh
# file: examples/equality_test.sh

testEquality() {
  assertEquals 1 1
}

# Load shUnit2.
. ../shunit2

Running the unit test should give results similar to the following.

$ cd examples
$ ./equality_test.sh
testEquality

Ran 1 test.

OK

W00t! You've just run your first successful unit test. So, what just happened? Quite a bit really, and it all happened simply by sourcing the shunit2 library. The basic functionality for the script above goes like this:

• When shUnit2 is sourced, it will walk through any functions defined whose name starts with the string test, and add those to an internal list of tests to execute. Once a list of test functions to be run has been determined, shunit2 will go to work.
• Before any tests are executed, shUnit2 again looks for a function, this time one named oneTimeSetUp(). If it exists, it will be run. This function is normally used to setup the environment for all tests to be run. Things like creating directories for output or setting environment variables are good to place here. Just so you know, you can also declare a corresponding function named oneTimeTearDown() function that does the same thing, but once all the tests have been completed. It is good for removing temporary directories, etc.
• shUnit2 is now ready to run tests. Before doing so though, it again looks for another function that might be declared, one named setUp(). If the function exists, it will be run before each test. It is good for resetting the environment so that each test starts with a clean slate. At this stage, the first test is finally run. The success of the test is recorded for a report that will be generated later. After the test is run, shUnit2 looks for a final function that might be declared, one named tearDown(). If it exists, it will be run after each test. It is a good place for cleaning up after each test, maybe doing things like removing files that were created, or removing directories. This set of steps, setUp() > test() > tearDown(), is repeated for all of the available tests.
• Once all the work is done, shUnit2 will generate the nice report you saw above. A summary of all the successes and failures will be given so that you know how well your code is doing.

We should now try adding a test that fails. Change your unit test to look like this.

#! /bin/sh
# file: examples/party_test.sh

testEquality() {
  assertEquals 1 1
}

testPartyLikeItIs1999() {
  year=`date '+%Y'`
  assertEquals "It's not 1999 :-(" '1999' "${year}"
}

# Load shUnit2.
. ../shunit2

So, what did you get? I guess it told you that this isn't 1999. Bummer, eh? Hopefully, you noticed a couple of things that were different about the second test. First, we added an optional message that the user will see if the assert fails. Second, we did comparisons of strings instead of integers as in the first test. It doesn't matter whether you are testing for equality of strings or integers. Both work equally well with shUnit2.

Hopefully, this is enough to get you started with unit testing. If you want a ton more examples, take a look at the tests provided with log4sh or shFlags. Both provide excellent examples of more advanced usage. shUnit2 was after all written to meet the unit testing need that log4sh had.

If you are using distribution packaged shUnit2 which is accessible from /usr/bin/shunit2 such as Debian, you can load shUnit2 without specifying its path. So the last 2 lines in the above can be replaced by:

# Load shUnit2.
. shunit2

Any string values passed should be properly quoted -- they should be surrounded by single-quote (') or double-quote (") characters -- so that the shell will properly parse them.

assertEquals [message] expected actual

Asserts that expected and actual are equal to one another. The expected and actual values can be either strings or integer values as both will be treated as strings. The message is optional, and must be quoted.

assertNotEquals [message] unexpected actual

Asserts that unexpected and actual are not equal to one another. The unexpected and actual values can be either strings or integer values as both will be treated as strings. The message is optional, and must be quoted.

assertSame [message] expected actual

This function is functionally equivalent to assertEquals.

assertNotSame [message] unexpected actual

This function is functionally equivalent to assertNotEquals.

assertContains [message] container content

Asserts that container contains content. The container and content values can be either strings or integer values as both will be treated as strings. The message is optional, and must be quoted.

assertNotContains [message] container content

Asserts that container does not contain content. The container and content values can be either strings or integer values as both will be treated as strings. The message is optional, and must be quoted.

assertNull [message] value

Asserts that value is null, or in shell terms, a zero-length string. The value must be a string as an integer value does not translate into a zero- length string. The message is optional, and must be quoted.

assertNotNull [message] value

Asserts that value is not null, or in shell terms, a non-empty string. The value may be a string or an integer as the latter will be parsed as a non-empty string value. The message is optional, and must be quoted.

assertTrue [message] condition

Asserts that a given shell test condition is true. The condition can be as simple as a shell true value (the value 0 -- equivalent to ${SHUNIT_TRUE}), or a more sophisticated shell conditional expression. The message is optional, and must be quoted.

A sophisticated shell conditional expression is equivalent to what the if or while shell built-ins would use (more specifically, what the test command would use). Testing for example whether some value is greater than another value can be done this way.

assertTrue "[ 34 -gt 23 ]"

Testing for the ability to read a file can also be done. This particular test will fail.

assertTrue 'test failed' "[ -r /some/non-existant/file ]"

As the expressions are standard shell test expressions, it is possible to string multiple expressions together with -a and -o in the standard fashion. This test will succeed as the entire expression evaluates to true.

assertTrue 'test failed' '[ 1 -eq 1 -a 2 -eq 2 ]'

One word of warning: be very careful with your quoting as shell is not the most forgiving of bad quoting, and things will fail in strange ways.

assertFalse [message] condition

Asserts that a given shell test condition is false. The condition can be as simple as a shell false value (the value 1 -- equivalent to ${SHUNIT_FALSE}), or a more sophisticated shell conditional expression. The message is optional, and must be quoted.

For examples of more sophisticated expressions, see assertTrue.

Just to clarify, failures do not test the various arguments against one another. Failures simply fail, optionally with a message, and that is all they do. If you need to test arguments against one another, use asserts.

If all failures do is fail, why might one use them? There are times when you may have some very complicated logic that you need to test, and the simple asserts provided are simply not adequate. You can do your own validation of the code, use an assertTrue ${SHUNIT_TRUE} if your own tests succeeded, and use a failure to record a failure.

fail [message]

Fails the test immediately. The message is optional, and must be quoted.

failNotEquals [message] unexpected actual

Fails the test immediately, reporting that the unexpected and actual values are not equal to one another. The message is optional, and must be quoted.

Note: no actual comparison of unexpected and actual is done.

failSame [message] expected actual

Fails the test immediately, reporting that the expected and actual values are the same. The message is optional, and must be quoted.

Note: no actual comparison of expected and actual is done.

failNotSame [message] expected actual

Fails the test immediately, reporting that the expected and actual values are not the same. The message is optional, and must be quoted.

Note: no actual comparison of expected and actual is done.

failFound [message] content

Fails the test immediately, reporting that the content was found. The message is optional, and must be quoted.

Note: no actual search of content is done.

failNotFound [message] content

Fails the test immediately, reporting that the content was not found. The message is optional, and must be quoted.

Note: no actual search of content is done.

oneTimeSetUp

This function can be optionally overridden by the user in their test suite.

If this function exists, it will be called once before any tests are run. It is useful to prepare a common environment for all tests.

oneTimeTearDown

This function can be optionally overridden by the user in their test suite.

If this function exists, it will be called once after all tests are completed. It is useful to clean up the environment after all tests.

setUp

This function can be optionally overridden by the user in their test suite.

If this function exists, it will be called before each test is run. It is useful to reset the environment before each test.

tearDown

This function can be optionally overridden by the user in their test suite.

If this function exists, it will be called after each test completes. It is useful to clean up the environment after each test.

startSkipping

This function forces the remaining assert and fail functions to be "skipped", i.e. they will have no effect. Each function skipped will be recorded so that the total of asserts and fails will not be altered.

endSkipping

This function returns calls to the assert and fail functions to their default behavior, i.e. they will be called.

isSkipping

This function returns the current state of skipping. It can be compared against ${SHUNIT_TRUE} or ${SHUNIT_FALSE} if desired.

The default behavior of shUnit2 is that all tests will be found dynamically. If you have a specific set of tests you want to run, or you don't want to use the standard naming scheme of prefixing your tests with test, these functions are for you. Most users will never use them though.

suite

This function can be optionally overridden by the user in their test suite.

If this function exists, it will be called when shunit2 is sourced. If it does not exist, shUnit2 will search the parent script for all functions beginning with the word test, and they will be added dynamically to the test suite.

suite_addTest name

This function adds a function named name to the list of tests scheduled for execution as part of this test suite. This function should only be called from within the suite() function.

There are several constants provided by shUnit2 as variables that might be of use to you.

Predefined

User defined

The constants values SHUNIT_TRUE, SHUNIT_FALSE, and SHUNIT_ERROR are returned from nearly every function to indicate the success or failure of the function. Additionally the variable flags_error is filled with a detailed error message if any function returns with a SHUNIT_ERROR value.

If you include lots of assert statements in an individual test function, it can become difficult to determine exactly which assert was thrown unless your messages are unique. To help somewhat, line numbers can be included in the assert messages. To enable this, a special shell "macro" must be used rather than the standard assert calls. Shell doesn't actually have macros; the name is used here as the operation is similar to a standard macro.

For example, to include line numbers for a assertEquals() function call, replace the assertEquals() with ${_ASSERT_EQUALS_}.

Example -- Asserts with and without line numbers

#! /bin/sh
# file: examples/lineno_test.sh

testLineNo() {
  # This assert will have line numbers included (e.g. "ASSERT:[123] ...").
  echo "ae: ${_ASSERT_EQUALS_}"
  ${_ASSERT_EQUALS_} 'not equal' 1 2

  # This assert will not have line numbers included (e.g. "ASSERT: ...").
  assertEquals 'not equal' 1 2
}

# Load shUnit2.
. ../shunit2

Notes:

1. Due to how shell parses command-line arguments, all strings used with macros should be quoted twice. Namely, single-quotes must be converted to single-double-quotes, and vice-versa.
   
   Normal assertEquals call.
   assertEquals 'some message' 'x' ''
   
   Macro _ASSERT_EQUALS_ call. Note the extra quoting around the message and the null value.
   _ASSERT_EQUALS_ '"some message"' 'x' '""'

2. Line numbers are not supported in all shells. If a shell does not support them, no errors will be thrown. Supported shells include: bash (>=3.0), ksh, mksh, and zsh.

There are times where the test code you have written is just not applicable to the system you are running on. This section describes how to skip these tests but maintain the total test count.

Probably the easiest example would be shell code that is meant to run under the bash shell, but the unit test is running under the Bourne shell. There are things that just won't work. The following test code demonstrates two sample functions, one that will be run under any shell, and the another that will run only under the bash shell.

Example -- math include

# file: examples/math.inc.

add_generic() {
  num_a=$1
  num_b=$2

  expr $1 + $2
}

add_bash() {
  num_a=$1
  num_b=$2

  echo $(($1 + $2))
}

And here is a corresponding unit test that correctly skips the add_bash() function when the unit test is not running under the bash shell.

Example -- math unit test

#! /bin/sh
# file: examples/math_test.sh

testAdding() {
  result=`add_generic 1 2`
  assertEquals \
      "the result of '${result}' was wrong" \
      3 "${result}"

  # Disable non-generic tests.
  [ -z "${BASH_VERSION:-}" ] && startSkipping

  result=`add_bash 1 2`
  assertEquals \
      "the result of '${result}' was wrong" \
      3 "${result}"
}

oneTimeSetUp() {
  # Load include to test.
  . ./math.inc
}

# Load and run shUnit2.
. ../shunit2

Running the above test under the bash shell will result in the following output.

$ /bin/bash math_test.sh
testAdding

Ran 1 test.

OK

But, running the test under any other Unix shell will result in the following output.

$ /bin/ksh math_test.sh
testAdding

Ran 1 test.

OK (skipped=1)

As you can see, the total number of tests has not changed, but the report indicates that some tests were skipped.

Skipping can be controlled with the following functions: startSkipping(), endSkipping(), and isSkipping(). Once skipping is enabled, it will remain enabled until the end of the current test function call, after which skipping is disabled.

When running a test script, you may override the default set of tests, or the suite-specified set of tests, by providing additional arguments on the command line. Each additional argument after the -- marker is assumed to be the name of a test function to be run in the order specified. e.g.

test-script.sh -- testOne testTwo otherFunction

or

shunit2 test-script.sh testOne testTwo otherFunction

In either case, three functions will be run as tests, testOne, testTwo, and otherFunction. Note that the function otherFunction would not normally be run by shunit2 as part of the implicit collection of tests as it's function name does not match the test function name pattern test*.

If a specified test function does not exist, shunit2 will still attempt to run that function and thereby cause a failure which shunit2 will catch and mark as a failed test. All other tests will run normally.

The specification of tests does not affect how shunit2 looks for and executes the setup and tear down functions, which will still run as expected.

Most continuous integration tools like CircleCI, are capable to interpret test results in JUnit format, helping you with spacilized sections and triggers tailored to identify faster a failing test. This functionality is still unreleased but you can test it right away, installing shunit2 from source.

Given that you execute your test script in the following way

test-script.sh

You can generate the JUnit report like this

mkdir -p results
test-script.sh -- --output-junit-xml=results/test-script.xml

It will generate something like

<?xml version="1.0" encoding="UTF-8"?>
<testsuite
  failures="0"
  name="test-script.sh"
  tests="1"
  assertions="1"
>
  <testcase
    classname="test-script.sh"
    name="testOne"
    assertions="1"
  >
  </testcase>
</testsuite>

You can also specify a more verbose suite name

test-script.sh -- --output-junit-xml=results/test-script.xml --suite-name=Test_Script

Then say to your CI tool where the results are. In the case of CircleCI is like the following

- store_test_results:
    path: results

For help, please send requests to either the shunit2-users@forestent.com mailing list (archives available on the web at https://groups.google.com/a/forestent.com/forum/#!forum/shunit2-users) or directly to Kate Ward .

For compatibility with Zsh, there is one requirement that must be met -- the shwordsplit option must be set. There are three ways to accomplish this.

1. In the unit-test script, add the following shell code snippet before sourcing the shunit2 library.

   setopt shwordsplit

2. When invoking zsh from either the command-line or as a script with #!, add the -y parameter.

   #! /bin/zsh -y

3. When invoking zsh from the command-line, add -o shwordsplit -- as parameters before the script name.

   $ zsh -o shwordsplit -- some_script