Syntax extension for writing in-line tests in ocaml code.

New syntactic constructs

The following constructs are now valid structure items:

let%test "name" = <boolean expr> (* true means ok, false or exn means broken *)
let%test_unit "name" = <unit expr> (* () means ok, exn means broken *)
let%test_module "name" = (module <module-expr>)  (* to group tests (to share
                                                    some setup for instance) *)

We may write _ instead of "name" for anonymous tests.

When running tests, they will be executed when the control flow
reaches the structure item (i.e. at toplevel for a toplevel test; when
the functor is applied for a test defined in the body of a functor,
etc.).

Tags

One can tag tests with the following construct:

let%test "name" [@tags "no-js"] = <expr>
let%test "name" [@tags "no-js", "other-tag"] = <expr>
let%test _ [@tags "no-js"] = <expr>
let%test _ [@tags "js-only"] = <expr>

Available tags are:

• no-js for tests that should not run when compiling Ocaml to Javascript

• js-only for tests that should only run in Javascript

• 32-bits-only for tests that should only run in 32 bits architectures

• 64-bits-only for tests that should only run in 64 bits architectures

• fast-flambda for tests that might only pass when compiling with flambda, -O3
  and cross library inlining

• x-library-inlining-sensitive for tests that might only pass when compiling
  with cross library inlining switched on

One can also tag entire test modules similarly:

let%test_module "name" [@tags "no-js"] = (module struct end)

The flags -drop-tag and -require-tag can be passed to the test
runner to restrict which tests are run. We say the tags of a test are
the union of the tags applied directly to that test using
[@tags ...] and the tags of all enclosing modules. It is to this
union that the predicates -drop-tag and -require-tag are applied.

If it is clear, from a test-module's tags, that none of the tests
within will possibly match the tag predicates imposed by the command
line flags, then additionally the top-level of that module will not be
run.

Examples

prime.ml

let is_prime = <magic>

let%test _ = is_prime 5
let%test _ = is_prime 7
let%test _ = not (is_prime 1)
let%test _ = not (is_prime 8)

Tests in a functor.

module Make(C : S) = struct
     <magic>
     let%test _ = <some expression>
end

module M = Make(Int)

Grouping test and side-effecting initialisation.

Since the module passed as an argument to let%test_module is only
initialised when we run the tests, it is ok to perform side-effects in
the module-expression argument.

let%test_module _ = (module struct
    module UID = Unique_id.Int(struct end)

    let%test _ = UID.create() <> UID.create()
end)

Building and running the tests at jane street

Inline tests can only be used in libraries, not executables.

The standard build rules create an executable script inline_tests_runner which runs all
tests in the directory. This script takes optional arguments (see below) to restrict which
tests are run.

The full set of tests are run when building the jenga runtest alias.

jenga .runtest

Building and running the tests outside of jane street

Code using this extension must be compiled and linked using the
ppx_inline_test.runtime.lib library. The ppx_inline_test syntax
extension will reject any test if it wasn't passed a -inline-test-lib libname flag.

To integrate this in your build system, you should look at the
instruction provided for
the ppx_driver ocamlbuild plugin. There
are a few working example using oasis in the
Jane Street tests.

Execution

Tests are executed when the executable containing the tests is called with command line
arguments:

your.exe inline-test-runner libname [options]

otherwise they are ignored.

This libname is a way of restricting the tests run by the executable. The dependencies
of your library (or executable) could also use ppx_inline_test, but you don't
necessarily want to run their tests too. For instance, core is built by giving
-inline-test-lib core and core_extended is built by giving -inline-test-lib core_extended. And now when an executable linked with both core and core_extended is
run with a libname of core_extended, only the tests of core_extended are run.

Finally, after running tests, Ppx_inline_test_lib.Runtime.exit () should be called
(to exit with an error and a summary of the number of failed tests if there were errors or
exit normally otherwise).

Command line arguments

The executable that runs tests can take additional command line arguments. The most useful
of these are:

• -verbose

  to see the tests as they run

• -only-test location

  where location is either a filename -only-test main.ml, a filename
  with a line number -only-test main.ml:32, or with the syntax that the
  compiler uses: File "main.ml", or File "main.ml", line 32 or File "main.ml", line 32, characters 2-6 (characters are ignored).
  The position that matters is the position of the let%test or let%test_unit.

  The positions shown by -verbose are valid inputs for -only-test.

  If no -only-test flag is given, all the tests are
  run. Otherwise all the tests matching any of the locations are run.

• -drop-tag tag

  drop all the tests tagged with tag.

Parallelizing tests

If you pass arguments of the form -inline-test-lib lib:partition to ppx_inline_test,
then you will be able to run tests from a given source file in parallel with tests from
other source files. All the tests inside the same source file are still run sequentially.

You should pick different partition names for the different files in your library (the
name of the .ml files for instance).

ppx_inline_test_lib currently requires some external system like a build system to run
it multiple times in parallel, although we may make it possible to run the inline tests in
parallel directly in the future.

If you do that, you can now use two new flags of the executable containing the tests:

• -list-partitions

  lists all the partitions that contain at least one test, one per line.

• -partition P

  only run the tests of the library that are encountered at toplevel of the source file
  that was preprocessed with the given partition P (the tests need not be
  syntactically in the file, they could be the result of applying a functor)

A build system can combine these two commands by first listing partitions, and then
running one command for each partition.