/go-test-runner

Primary LanguageJupyter NotebookGNU Affero General Public License v3.0AGPL-3.0

Exercism's Go Test Runner

Go CI Coverage Status

This is Exercism's test runner for the Go track.

Executing the Test Runner

The test runner requires 2 parameters:

  • input_dir: the path containing the solution to test
  • output_dir: the output path for the test results

Local Development

go run . testrunner/testdata/practice/passing outdir

Run the package tests

go test ./...

Run the linter

Linting (and testing) is performed in a github action workflow - test.yml. You can install golangci-lint locally and then run:

golangci-lint run ./...

Interactive Debug / REPL

The original AST parsing code was developed using a Jupyter interactive Go REPL thanks to the gophernotes project. Consult the gophernotes docs for installation instructions. Once installed, you should be able to view, run, and modify the provided debug code "live" without constantly recompiling:

# assuming python3 with notebook installed via pip3, ymmv
python3 -m notebook ast_debug.ipynb

Docker

A docker container is used to run the test runner against submitted exercises. To build the container locally, execute the following from the repository root directory:

docker build -t exercism/go-test-runner .

Run the test runner in the container by passing in the slug name, and absolute paths to the exercise (solution) and a writeable tmp directory. These directories should be mounted as volumes:

docker run --network none --read-only -v $(pwd)/testrunner/testdata/practice/gigasecond:/solution -v /tmp:/tmp exercism/go-test-runner gigasecond /solution /tmp

For troubleshooting / debug you can name the container, run it in interactive mode, and detach from it using:

docker run --name exercism-go-test-runner -d -i --network none --read-only -v $(pwd)/testrunner/testdata/practice/gigasecond:/solution -v /tmp:/tmp exercism/go-test-runner gigasecond /solution /tmp
# You can then access the container as follows:
docker exec -it $(docker ps -q --filter name=exercism-go-test-runner) /bin/sh

External Go packages

Some extra Go packages that are not part of the standard library are downloaded when the docker image is built. This allows students to use these external packages in their solutions.

The list of external packages and their versions is in external-packages/go.mod.

To add or remove a package from the list of external packages supported:

  1. Add/remove the corresponding import from external-packages/deps.go
  2. Run go mod tidy inside the external-packages directory
  3. Commit deps.go along with the changes to go.mod and go.sum produced by go mod tidy.

Note: The Go version declared in the go.mod file of the external-packages module should be the same as the version in the go.mod file of the exercises students download. This is because the Go version in the go.mod file affects the indirect dependencies that are downloaded, and consequently the go.sum file that is generated. A student can have a go.mod file declaring only supported dependencies, but if the Go version in that go.mod is different from the Go version in external-packages/go.mod, their go.sum may include more dependencies than external-packages/go.sum, which means they won't be able to run the solution.

Subtests

The test runner is responsible for returning the test_code field, which should be a copy of the test code corresponding to each test result.

For top-level tests, the AST is used to return the function code directly. For tests containing subtests, additional processing is required. To ease the burden of advanced AST processing on unstructured / non deterministic test code, subtests should adhere to the following specification. If a test employs subtests, do not mix it with test or other code outside of the Run() call.

  • Subtests not meeting the spec will be treated as top-level tests, with the entire test function code being returned for every subtest.
  • Assertions/outputs made outside of the Run() call will not be included in the result JSON because the "parent" tests are removed from the results if subtests are present. (Parent test reports were confusing to students because they did not include any assertion or fmt.Println output.)

At some point, we may implement a static analyzer which warns the exercise submitter when they commit subtests not meeting the specification.

Subtest Format Specification

The specification is annotated in the comments of the following example test:

func TestParseCard(t *testing.T) {
  // There can be additional code here, it will be shown for all subtests.
  // If the code here includes assignments, the test data variable below needs to be called "tests".

  tests := []struct {
    name string // The name field is required
    card string
    want int
  }{
    // The relevant test data will be parsed out individually for each subtest
    {
      // spaces are fine in a test name, but do not mix them with underscores
      // - because the extraction code won't be able to find it
      name: "parse queen",
      card: "queen",
      want: 10,
    },
    // For example, this element will be parsed for `TestParseCard/parse_king`
    {
      name: "parse king",
      card: "king",
      want: 10,
    },
  }

  // There can be additional code here, it will be shown for all subtests.

  // The contents of the function literal will be extracted as the test code
  for _, tt := range tests {
    // The Run() call must be the first statement in the for loop
    t.Run(tt.name, func(t *testing.T) {
      // This code block will be pulled into the resulting test_code field
      if got := ParseCard(tt.card); got != tt.want {
        t.Errorf("ParseCard(%s) = %d, want %d", tt.card, got, tt.want)
      }
    })
  }

  // There can be additional code here, it will be shown for all subtests.
}

The test code above will result in the following test_code field, corresponding to the test named TestParseCard/parse_queen:

tt := struct {
  name string
  card string
  want int
}{
  name: "parse queen",
  card: "queen",
  want: 10,
}
if got := ParseCard(tt.card); got != tt.want {
  t.Errorf("ParseCard(%s) = %d, want %d", tt.card, got, tt.want)
}

Providing Additional Testing Flags

Exercises can supply additional flags that will be included when the test runner executes the go test command. This is done via the .meta/config.json file of the exercise. See example below.

{
  // ...
  "custom": {
    "testingFlags": ["-race"]
  }
}

Currently, only the flag -race is supported. If more flags should be allowed in the future, they first need to be added to the allowedTestingFlags list in testrunner/execute.go.

Assigning Task Ids

For concept exercises, the output of the test runner can contain task ids for the different test cases. These ids are used on the website to associate the test result with the corresponding task in the exercise. This leads to a great user experience for the students.

The test runner output will only contain task ids if they were explicitly enabled in the .meta/config.json file for the exercise as shown below (otherwise task ids are omitted):

{
  // ...
  "custom": {
    "taskIdsEnabled": true
  }
}

Note that this flag should only ever be set for concept exercises. Task ids are currently not supported for practice exercise.

There are two ways the test runner can assign task ids for concept exercises.

Implicit Task Id Assignment

If "taskIdsEnabled": true was set but there are no explicit task ids found in the test file, the test runner will automatically assign task ids. It assumes each parent test corresponds to one task and will assign task id 1 to the first parent test and its sub-tests, task id 2 to the next one etc.

For most concept exercises, we have this 1 to 1 relationship between tests and tasks. This implicit assignment means we do not need to add anything to the test files to get task ids in the test runner output. We only need to set the type in the config as shown above.

You can test this locally end-to-end via go run . testrunner/testdata/concept/conditionals outdir.

Explicit Task Id Assignment

If the implicit system would lead to wrong task ids, they can be set manually via a comment in the following format that is added for the test function:

// testRunnerTaskID=1
func TestSomething(t *testing.T) {
  // ...
}

Sub-tests automatically get the task id from their parent, they don't need any explicit assignment.

You can test this locally end-to-end via go run . testrunner/testdata/concept/conditionals-with-task-ids outdir.

Explicit task id assignment will only take effect if an explicit task id was found on every parent test in the test file. Otherwise no task ids will be set at all.

Finding the task id is robust again other comments before or after or in the same line as the testRunnerTaskID comment, see the conditionals-with-task-ids test file for examples.

Known limitations

Besides what is mentioned in the open issues, the test runner has the following limitations currently.

  • The test runner assumes all test functions func Test... can be found in one file.
  • The cases_test.go file is ignored when extracting the code for the test.
  • Sub-tests need to follow a certain format, see details above.