Writing and Running Tests

Dune tries to streamline the testing story as much as possible, so you can focus on the tests themselves and not bother with setting up various test frameworks.

In this section, we’ll explain the workflow to deal with tests in Dune. In particular, we’ll see how to run the test suite of a project, how to describe your tests to Dune, and how to promote test results as expectation.

We distinguish three kinds of tests:

• Inline tests - written directly inside the .ml files of a library

• Custom tests - run an executable, possibly followed by an action such as diffing the produced output.

• Cram tests - expect tests written in Cram style.

Running Tests

Whatever the tests of a project are, the usual way to run tests with Dune is to call dune runtest from the shell (or the command alias dune test). This will run all the tests defined in the current directory and any subdirectory recursively.

Note that in any case, dune runtest is simply shorthand for building the runtest alias, so you can always ask Dune to run the tests in conjunction with other targets by passing @runtest to dune build. For instance:

$ dune build @install @runtest
$ dune build @install @test/runtest

Running a Single Test

If you would only like to run a single test for your project, you may use dune exec to run the test executable (for the sake of this example, project/tests/myTest.ml):

$ dune exec project/tests/myTest.exe

To run Cram Tests, you can use the alias that is created for the test. The name of the alias corresponds to the name of the test without the .t extension. For directory tests, this is the name of the directory without the .t extension. Assuming a cram-test.t or cram-test.t/run.t file exists, it can be run with:

$ dune build @cram-test

Running Tests in a Directory

You can also pass a directory argument to run the tests from a subtree. For instance, dune runtest test will only run the tests from the test directory and any subdirectory of test recursively.

Inline Tests

There are several inline tests frameworks available for OCaml, such as ppx_inline_test and qtest. We will use ppx_inline_test as an example because it has the necessary setup to be used with Dune out of the box.

ppx_inline_test allows one to write tests directly inside .ml files as follows:

let rec fact n = if n = 1 then 1 else n * fact (n - 1)

let%test _ = fact 5 = 120

The file must be preprocessed with the ppx_inline_test PPX rewriter, so for instance the dune file might look like this:

(library
 (name foo)
 (preprocess (pps ppx_inline_test)))

In order to tell Dune that our library contains inline tests, we have to add an inline_tests field:

(library
 (name foo)
 (inline_tests)
 (preprocess (pps ppx_inline_test)))

We can now build and execute this test by running dune runtest. For instance, if we make the test fail by replacing 120 by 0 we get:

$ dune runtest
[...]
File "src/fact.ml", line 3, characters 0-25: <<(fact 5) = 0>> is false.

FAILED 1 / 1 tests

Note that in this case Dune knew how to build and run the tests without any special configuration. This is because ppx_inline_test defines an inline tests backend that’s used by the library. Some other frameworks, such as qtest, don’t have any special library or PPX rewriter. To use such a framework, you must tell Dune about it, as it cannot guess. You can do that by adding a backend field:

(library
 (name foo)
 (inline_tests (backend qtest.lib)))

In the example above, the name qtest.lib comes from the public_name field in qtest’s own dune file.

Note that using ppx_inline_test requires that the opam package ppx_inline_test be installed in your switch. If you use ppx_inline_test in a package then that package must unconditionally depend on ppx_inline_test (ie. ppx_inline_test can’t be a with-test dependency).

Inline Expectation Tests

Inline expectation tests are a special case of inline tests where written OCaml code prints something followed by what you expect this code to print. For instance, using ppx_expect:

let%expect_test _ =
  print_endline "Hello, world!";
  [%expect{|
    Hello, world!
  |}]

The test procedure consist of executing the OCaml code and replacing the contents of the [%expect] extension point by the real output. You then get a new file that you can compare to the original source file. Expectation tests are a neat way to write tests as the following test elements are clearly identified:

• The code of the test

• The test expectation

• The test outcome

You can have a look at this blog post to find out more about expectation tests. To Dune, the workflow for expectation tests is always as follows:

• Write the test with some empty expect nodes in it

• Run the tests

• Check the suggested correction and promote it as the original source file if you are happy with it

Dune makes this workflow very easy. Simply add ppx_expect to your list of PPX rewriters as follows:

(library
 (name foo)
 (inline_tests)
 (preprocess (pps ppx_expect)))

Then calling dune runtest will run these tests, and in case of mismatch, Dune will print a diff of the original source file and the suggested correction. For instance:

$ dune runtest
[...]
-src/fact.ml
+src/fact.ml.corrected
File "src/fact.ml", line 5, characters 0-1:
let rec fact n = if n = 1 then 1 else n * fact (n - 1)

let%expect_test _ =
  print_int (fact 5);
-  [%expect]
+  [%expect{| 120 |}]

In order to accept the correction, simply run:

$ dune promote

You can also make Dune automatically accept the correction after running the tests by typing:

$ dune runtest --auto-promote

Finally, some editor integration can make the editor do the promotion, which in turn makes the workflow even smoother.

Running a Subset of the Test Suite

You may also run a group of tests located under a directory with:

$ dune runtest mylib/tests

The above command will run all tests defined in tests and its subdirectories.

Running Tests in Bytecode or JavaScript

By default, Dune runs inline tests in native mode, unless native compilation isn’t available. In which case, it runs them in bytecode. You can change this setting to choose the modes that tests should run in. To do this, add a modes field to the inline_tests field. Available modes are:

• byte for running tests in byte code

• native for running tests in native mode

• best for running tests in native mode with fallback to byte code, if native compilation is not available

• js for running tests in JavaScript using Node.js

For instance:

(library
 (name foo)
 (inline_tests (modes byte best js))
 (preprocess (pps ppx_expect)))

Specifying Inline Test Dependencies

If your tests are reading files, you must tell Dune by adding a deps field the inline_tests field. The argument of this deps field follows the usual Dependency Specification. For instance:

(library
 (name foo)
 (inline_tests (deps data.txt))
 (preprocess (pps ppx_expect)))

Passing Special Arguments to the Test Runner

Under the hood, a test executable is built by Dune. Depending on the backend used, this runner might take useful command line arguments. You can specify such flags by using a flags field, such as:

(library
 (name foo)
 (inline_tests (flags (-foo bar)))
 (preprocess (pps ppx_expect)))

The argument of the flags field follows the Ordered Set Language.

Passing Special Arguments to the Test Executable

To control how the test executable is built, it’s possible to customize a subset of compilation options for an executable using the executable field. Dune gives you this ability by simply specifying command line arguments as flags. You can specify such flags by using flags field. For instance:

(library
 (name foo)
 (inline_tests
  (flags (-foo bar)
  (executable
   (flags (-foo bar))))
  (preprocess (pps ppx_expect))))

The argument of the flags field follows the Ordered Set Language.

Using Additional Libraries in the Test Runner

When tests are not part of the library code, it’s possible that tests require additional libraries than the library being tested. This is the case with qtest, as tests are written in comments. You can specify such libraries using a libraries field, such as:

(library
 (name foo)
 (inline_tests
  (backend qtest)
  (libraries bar)))

Changing the Flags of the Linking Step of the Test Runner

You can use the link_flags field to change the linker flags passed to ocamlopt when building the test runner. By default, the linking flags are -linkall. You probably want to keep -linkall as one of the new list of flags (unless you know what you are doing), forcing the linker to load your test module, since the test runner doesn’t depend on anything itself. This field supports (:include ...) forms.

(library
 (name foo)
 (inline_tests
  (executable
   (link_flags -linkall -noautolink -cclib -Wl,-Bstatic -cclib -lm)))
 (preprocess (pps ppx_expect)))

Defining Your Own Inline Test Backend

If you are writing a test framework (or for other specific cases), you might want to define your own inline tests backend. If your framework is naturally implemented by a library or PPX rewriter that’s necessary to write tests, you should define this library as a backend. Otherwise simply create an empty library with your chosen backend’s name.

In order to define a library as an inline tests backend, simply add an inline_tests.backend field to the library stanza. An inline tests backend is specified by four parameters:

1. How to create the test runner

2. How to build the test runner

3. How to run the test runner

4. Optionally how to run the test runner to list partitions

These four parameters can be specified inside the inline_tests.backend field, which accepts the following fields:

(generate_runner       <action>)
(runner_libraries     (<ocaml-libraries>))
(flags                 <flags>)
(list_partitions_flags <flags>)
(extends              (<backends>))

For instance:

<action> follows the Actions specification. It describes an action that should be executed in the library’s directory using this backend for their tests. It’s expected that the action will produce some OCaml code on its standard output. This code will constitute the test runner. The action can use the following additional variables:

• %{library-name} — the name of the library being tested

• %{impl-files} — the list of implementation files in the library, i.e., all the .ml and .re files

• %{intf-files} — the list of interface files in the library, i.e., all the .mli and .rei files

The runner_libraries field specifies what OCaml libraries the test runner uses. For instance, if the generate_runner actions generates something like My_test_framework.runtests (), then you should probably put my_test_framework in the runner_libraries field.

If your test runner needs specific flags, you should pass them in the flags field. You can use the %{library-name} variable in this field.

If your test runner supports test partitions, you should pass the flags necessary for listing partitions in the list_partitions_flags field. In such scenario, the flags field will also accepts a %{partition} variable.

Finally, a backend can be an extension of another backend. In this case, you must specify this in the extends field. For instance, ppx_expect is an extension of ppx_inline_test. It’s possible to use a backend with several extensions in a library; however, there must be exactly one root backend, i.e., exactly one backend that isn’t an extension of another one.

When using a backend with extensions, the various fields are simply concatenated. The order in which they are concatenated is unspecified; however, if a backend b extends a backend a, then a will always come before b.

Example of Backend

In this example, we put tests in comments of the form:

(*TEST: assert (fact 5 = 120) *)

The backend for such a framework looks like this:

(library
 (name simple_tests)
 (inline_tests.backend
  (generate_runner (run sed "s/(\\*TEST:\\(.*\\)\\*)/let () = \\1;;/" %{impl-files}))))

Now all you have to do is write (inline_tests ((backend simple_tests))) wherever you want to write such tests. Note that this is only an example. We don’t recommend using sed in your build, as this would cause portability problems.

Custom Tests

We said in Running tests that to run tests, Dune simply builds the runtest alias. As a result, you simply need to add an action to this alias in any directory in order to define custom tests. For instance, if you have a binary tests.exe that you want to run as part of running your test suite, simply add this to a dune file:

(rule
 (alias  runtest)
 (action (run ./tests.exe)))

Hence to define a test, a pair of alias and executable stanzas are required. To simplify this common pattern, Dune provides a tests stanza to define multiple tests and their aliases at once:

(tests (names test1 test2))

Diffing the Result

It’s often the case that we want to compare the actual output of a test to an expected one. For that, Dune offers the diff command, which in essence is the same as running the diff tool, except that it’s more integrated in Dune, especially with the promote command. For instance, let’s consider this test:

(rule
(with-stdout-to tests.output (run ./tests.exe)))

(rule
 (alias runtest)
 (action (diff tests.expected tests.output)))

After having run tests.exe and dumping its output to tests.output, Dune will compare the latter to tests.expected. In case of mismatch, Dune will print a diff and then the dune promote command can be used to copy over the generated test.output file to tests.expected in the source tree.

Alternatively, the tests also supports this style of tests.

(tests (names tests))

Dune expects the existence of a tests.expected file to infer that this is an expected test.

This provides a nice way of dealing with the usual write code, run, and promote cycle of testing. For instance:

$ dune runtest
[...]
-tests.expected
+tests.output
File "tests.expected", line 1, characters 0-1:
-Hello, world!
+Good bye!
$ dune promote
Promoting _build/default/tests.output to tests.expected.

Note that if available, the diffing is done using the patdiff tool, which displays nicer looking diffs than the standard diff tool. You can change that by passing --diff-command CMD to Dune.

Cram Tests

Cram tests are expectation tests written in a shell-like syntax. They are ideal for testing binaries. Cram tests are automatically discovered from files or directories with a .t extension. By default, this has been enabled since Dune 3.0. For older versions, it must be manually enabled in the dune-project file:

(lang dune 2.7)
(cram enable)

File Tests

To define a standalone test, we create a .t file. For example, foo.t:

Simplest possible Cram test
  $ echo "testing"

This simple example demonstrates two components of Cram tests: comments and commands. See Cram Tests for a description of the syntax.

To run the test and promote the results:

$ dune runtest
$ dune promote

We now see the output of the command:

Simplest possible cram test
  $ echo "testing"
  testing

This is the main advantage of expect tests. We don’t need to write assertions manually; instead we detect failure when the command produces a different output than what is recorded in the test script.

For example, here’s an example of how we’d test the wc utility. wc.t:

We create a test artifact called "foo"
  $ cat >foo <<EOF
  > foo
  > bar
  > baz
  > EOF

After creating the fixture, we want to verify that ``wc`` gives us the right
result:
  $ wc -l foo | awk '{ print $1 }'
  4

The above example uses the doc syntax, piping the subsequent lines to cat. This is convenient for creating small test artifacts.

Directory Tests

In the above example we used cat to create the test artifact, but what if there are too many artifacts to comfortably fit in test file? Or some of the artifacts are binary?

It’s possible to include the artifacts as normal files or directories, provided the test is defined as a directory. The name of the test directory must end with .t and must include a run.t as the test script. Everything else in that directory is treated as raw data for the test. It’s not possible to define rules using dune files in such a directory.

We convert the wc test above into a directory test wc.t:

$ ls wc.t
run.t foo.txt bar/

This defines a directory test wc.t which must include a run.t file as the test script, with fool.txt and bar are test artifacts. We may then access their contents in the test script run.t:

Testing wc:
  $ wc -l foo | awk '{ print $1 }'
  4
  $ wc -l $(ls bar) | awk '{ print $1 }'
  1231

See also

(cram) stanza reference

Testing an OCaml Program

The most common testing situation involves testing an executable that is defined in Dune. For example:

(executable
 (name wc)
 (public_name wc))

To use this binary in the Cram test, we should depend on the binary in the test:

(cram
 (deps %{bin:wc}))

Sandboxing

Since Cram tests often create intermediate artifacts, it’s important that Cram tests are executed in a clean environment. This is why all Cram tests are sandboxed. To respect sandboxing, every test should specify dependency on any artifact that might rely on using the deps field.

See Sandboxing for details about the sandboxing mechanism.

Test Output Sanitation

In some situations, Cram tests emit non portable or non-deterministic output. We recommend sanitising such outputs using pipes. For example, we can scrub the OCaml magic number using sed as follows:

$ ocamlc -config | grep "cmi_magic_number:" | sed 's/Caml.*/$SPECIAL_CODE/'
cmi_magic_number: $SPECIAL_CODE

By default, Dune will scrub some paths from the output of the tests. The default list of paths is:

• The PWD of the test will be replaced by $TESTCASE_ROOT

• The temporary directory for the current script will be replaced by $TMPDIR

To add additional paths to this sanitation mechanism, it’s sufficient to modify the standard BUILD_PATH_PREFIX_MAP environment variable. For example:

$ export BUILD_PATH_PREFIX_MAP="HOME=$HOME:$BUILD_PATH_PREFIX_MAP"
$ echo $HOME
$HOME

Note: Unlike Dune’s version of Cram, the original specification for Cram supports regular expression and glob filtering for matching output. We chose not to implement this feature because it breaks the test, diff, and accept cycle. With regex or glob matching, the output must now be manually inspected and possibly updated. We consider the postprocessing approach described here as superior and will not introduce output matchers.