Easy, repeatable testing of CLI programs/commands

Install | Usage | Options | Test formats | Support/Contribute | Credits

shelltestrunner (executable: shelltest) is a portable command-line tool for testing command-line programs, or general shell commands, released under GPLv3+. It reads simple test specifications defining a command to run, some input, and the expected output, stderr, and exit status. It can run tests in parallel, selectively, with a timeout, in color, etc. Projects using it include hledger, Agda, and berp.

Install

There may be a new-enough packaged version on your platform. Eg:

Or, build the latest release on any major platform:

Usage

Here's a minimal test file containing one shell test:

# A comment. Testing bash's builtin "echo" command (if /bin/sh is bash)
echo
>>>= 0

They're called "shell test" because any shell (/bin/sh on POSIX, CMD on Windows) command line can be tested. Each test begins with the command to test, followed by optional stdin input, expected stdout and/or stderr output, and ends with the expected exit status. Here's another file containing two tests:

# Test that the "cat" program copies its input to stdout, 
# nothing appears on stderr, and exit status is 0.
cat
<<<
foo
>>>
foo
>>>2
>>>= 0

# Test that cat prints an error containing "unrecognized option" or
# "illegal option" and exits with non-zero status if given a bad flag.
cat --no-such-flag
>>>2 /(unrecognized|illegal) option/
>>>= !0

To run these tests:

$ shelltest echo.test cat.test
:echo.test: [OK]
:cat.test:1: [OK]
:cat.test:2: [OK]

         Test Cases  Total      
 Passed  3           3          
 Failed  0           0          
 Total   3           3          

That's the basics! There are also some alternate test formats you'll read about below.

Options

$ shelltest --help
shelltest 1.9

shelltest [OPTIONS] [TESTFILES|TESTDIRS]

Common flags:
  -l --list             List the names of all tests found
  -i --include=PAT      Include tests whose name contains this glob pattern
                        (eg: -i1 -i{4,5,6})
  -x --exclude=STR      Exclude test files whose path contains STR
  -a --all              Show all output without truncating, even if large
  -c --color            Show colored output if your terminal supports it
  -d --diff             Show differences between expected/actual output
     --precise          Show expected/actual output precisely, with quoting
     --hide-successes   Show only test failures
     --xmlout=FILE      Save test results to FILE in XML format.
  -D --defmacro=D=DEF   Define a macro D to be replaced by DEF while parsing
                        test files.
     --execdir          Run tests from within each test file's directory
     --extension=EXT    File suffix of test files (default: .test)
  -w --with=EXE         Replace the first word of test commands with EXE
                        (unindented commands only)
  -o --timeout=SECS     Number of seconds a test may run (default: no limit)
  -j --threads=N        Number of threads for running tests (default: 1)
     --shell=EXE        The shell program to use (must accept -c CMD;
                        default: /bin/sh on POSIX, cmd.exe on Windows)
     --debug            Show debug info while running
     --debug-parse      Show test file parsing results and stop
Print test file:
     --print[=FORMAT]   Print test files in specified format (default: v3).

  -h --help             Display help message
  -V --version          Print version information
     --numeric-version  Print just the version number

shelltest accepts one or more test file or directory arguments. A directory means all files below it named *.test (customisable with --extension).

Test commands are run with /bin/sh on POSIX systems and with CMD on Windows. By default, they are run in the directory in which you ran shelltest; with --execdir they will run in each test file's directory instead.

--include selects only tests whose name (file name plus intra-file sequence number) matches a .gitignore-style pattern, while --exclude skips tests based on their file path. These can be used eg to focus on a particular test, or to skip tests intended for a different platform.

-D/--defmacro defines a macro that is replaced by preprocessor before any tests are parsed and run.

-w/--with replaces the first word of all test commands with something else, which can be useful for testing alternate versions of a program. Commands which have been prefixed by an extra space will not be affected by this option.

--hide-successes gives quieter output, reporting only failed tests.

Long flags can be abbreviated to a unique prefix.

For example, the command:

$ shelltest tests -i args -c -j8 -o1 -DCONF_FILE=test/myconf.cfq --hide

• runs the tests defined in any *.test file in or below the tests/ directory
• whose names contain "args"
• in colour if possible
• with up to 8 tests running in parallel
• allowing no more than 1 second for each test
• replacing the text "CONF_FILE" in all tests with "test/myconf.cfq"
• reporting only the failures.

Test formats

shelltestrunner 1.9 adds some experimental new test file formats, described below. These need more real-world testing and may evolve further, but they will remain supported or will have a migration path.

shelltestrunner tries to parse each file first with format 2, then format 3, then format 1. All tests within a file should use the same format. I suggest choosing format 3 (short delimiters), switching to format 2 when longer delimiters are needed.

Format 1

Test files contain one or more individual tests, each consisting of a one-line shell command, optional input, expected standard output and/or error output, and a (required) exit status.

# COMMENTS OR BLANK LINES
COMMAND LINE
<<<
INPUT
>>>
EXPECTED OUTPUT (OR >>> /REGEXP/)
>>>2
EXPECTED STDERR (OR >>>2 /REGEXP/)
>>>= EXPECTED EXIT STATUS (OR >>>= /REGEXP/)

When not specified, stdout/stderr are ignored. A space before the command protects it from -w/--with.

Examples: above, shelltestrunner, hledger, Agda, berp, cblrepo.

Format 2

(shelltestrunner 1.9+) This improves on format 1 in two ways: it allows tests to reuse the same input, and it allows delimiters to often be omitted.

Test files contain one or more test groups. A test group consists of some optional standard input and one or more tests. Each test is a one-line shell command followed by optional expected standard output, error output and/or numeric exit status, separated by delimiters.

# COMMENTS OR BLANK LINES
<<<
INPUT
$$$ COMMAND LINE
>>>
EXPECTED OUTPUT (OR >>> /REGEX/)
>>>2
EXPECTED STDERR (OR >>>2 /REGEX/)
>>>= EXPECTED EXIT STATUS (OR >>>= /REGEX/ OR >>>=)
# COMMENTS OR BLANK LINES
ADDITIONAL TESTS FOR THIS INPUT
ADDITIONAL TEST GROUPS WITH DIFFERENT INPUT

All test parts are optional except the command line. If not specified, stdout and stderr are expected to be empty and exit status is expected to be zero.

Two spaces between $$$ and the command protects it from -w/--with.

The <<< delimiter is optional for the first input in a file. Without it, input begins at the first non-blank/comment line. Input ends at the $$$ delimiter. You can't put a comment before the first $$$.

The >>> delimiter is optional except when matching via regex. Expected output/stderr extends to the next >>>2 or >>>= if present, or to the last non-blank/comment line before the next <<< or $$$ or file end. /REGEX/ regular expression patterns may be used instead of specifying the expected output in full. The regex syntax is regex-tdfa's, plus you can put ! before /REGEX/ to negate the match.

The exit status is a number, normally 0 for a successful exit. This too can be prefixed with ! to negate the match, or you can use a /REGEX/ pattern. A >>>= with nothing after it ignores the exit status.

Examples:

All delimiters explicit:

# cat copies its input to stdout
<<<
foo
$$$ cat
>>>
foo

# or, given a bad flag, prints a platform-specific error and exits with non-zero status
$$$ cat --no-such-flag
>>>2 /(unrecognized|illegal) option/
>>>= !0

# echo ignores the input and prints a newline.
# We need the >>>= (or a >>>2) to delimit the whitespace which
# would otherwise be ignored.
$$$ echo
>>>

>>>=

Non-required <<< and >>> delimiters omitted:

foo
$$$ cat
foo

$$$ cat --no-such-flag
>>>2 /(unrecognized|illegal) option/
>>>= !0

$$$ echo

>>>=

Format 3

(shelltestrunner 1.9+) The same as format 2, but with more convenient short delimiters: < $ > >2 >=.

# COMMENTS OR BLANK LINES
<
INPUT
$ COMMAND LINE
>
EXPECTED OUTPUT (OR > /REGEX/)
>2
EXPECTED STDERR (OR >2 /REGEX/)
>= EXPECTED EXIT STATUS (OR >= /REGEX/ OR >=)
# COMMENTS OR BLANK LINES
ADDITIONAL TESTS FOR THIS INPUT
ADDITIONAL TEST GROUPS WITH DIFFERENT INPUT

Examples:

All delimiters explicit:

# cat copies its input to stdout
<
foo
$ cat
>
foo

# or, given a bad flag, prints a platform-specific error and exits with non-zero status
$ cat --no-such-flag
>2 /(unrecognized|illegal) option/
>= !0

# echo ignores the input and prints a newline.
# We use an explicit >= (or >2) to delimit the whitespace which
# would otherwise be ignored.
$ echo
>

>=

Non-required < and > delimiters omitted:

foo
$ cat
foo

$ cat --no-such-flag
>2 /(unrecognized|illegal) option/
>= !0

$ echo
>

>2

shelltestrunner

Printing tests

The --print option prints tests to stdout. This can be used to convert between test formats. Format 1, 2, and 3 are supported.

Caveats:

• Comments before the first test in v2/v3 are currently not printed because of the shared input feature which is not yet parsed in detail.
• Missing newline at EOF will not be preserved.
• Most tests in tests/, when printed, are not identical because
  • redundant/empty results like >= 0
  • >=1 instead of >= 1
  • comments before first test
  • shared input
  • no newline at EOF
  • no tests at all, just comments
  • no input/output marker <<</>>>

Support/Contribute

2012 user survey.

Feedback, testing, code, documentation, packaging, blogging, and funding are most welcome.

Credits

Simon Michael wrote shelltestrunner, inspired by John Wiegley's tests for Ledger.

Code contributors include: Taavi Väljaots, John Macfarlane, Andrés Sicard-Ramírez, Iustin Pop, Trygve Laugstøl, Bernie Pope, Sergei Trofimovich, John Chee.

shelltestrunner depends on several fine libraries, in particular Max Bolingbroke's test-framework, and of course on the Glorious Haskell Compiler.

The Blade Runner font is by Phil Steinschneider.