Awesome

toml-test is a language-agnostic test suite to verify the correctness of TOML parsers and writers.

Tests are divided into two groups: "invalid" and "valid". Decoders or encoders that reject "invalid" tests pass the tests, and decoders that accept "valid" tests and output precisely what is expected pass the tests. The output format is JSON, described below.

Both encoders and decoders share valid tests, except an encoder accepts JSON and outputs TOML rather than the reverse. The TOML representations are read with a blessed decoder and is compared. The JSON given to a TOML encoder is in the same format as the JSON that a TOML decoder should output.

If you find something in your parser that's not exactly covered by toml-test already then it should be added here; just creating an issue is enough: don't need to create a PR.

Compatible with TOML version v1.0.0.

Installation

There are binaries on the release page; these are statically compiled and should run in most environments. It's recommended you use a binary, or a tagged release if you build from source especially in CI environments. This prevents your tests from breaking on changes to tests in this tool.

To compile from source you will need Go 1.19 or newer:

% go install github.com/toml-lang/toml-test/cmd/toml-test@latest

This will build a ./toml-test binary.

Usage

toml-test accepts an encoder or decoder as the first positional argument, for example:

% toml-test my-toml-decoder
% toml-test my-toml-encoder -encoder

The -encoder flag is used to signal that this is an encoder rather than a decoder.

For example, to run the tests against the Go TOML library:

# Install my parser
% go install github.com/BurntSushi/toml/cmd/toml-test-decoder@master
% go install github.com/BurntSushi/toml/cmd/toml-test-encoder@master

% toml-test toml-test-decoder
toml-test v2023-10-23 [toml-test-decoder]: using embeded tests: 278 passed

% toml-test -encoder toml-test-encoder
toml-test v2023-10-23 [toml-test-encoder]: using embeded tests:  94 passed,  0 failed

The default is to use the tests compiled in the binary; you can use -testdir to load tests from the filesystem. You can use -run [name] or -skip [name] to run or skip specific tests. Both flags can be given more than once and accept glob patterns: -run 'valid/string/*'.

See toml-test -help for detailed usage.

Implementing a decoder

For your decoder to be compatible with toml-test it must satisfy the expected interface:

Your decoder must accept TOML data on stdin.
If the TOML data is invalid, your decoder must return with a non-zero exit code, indicating an error.
If the TOML data is valid, your decoder must output a JSON encoding of that data on stdout and return with a zero exit code, indicating success.

An example in pseudocode:

toml_data = read_stdin()

parsed_toml = decode_toml(toml_data)

if error_parsing_toml():
    print_error_to_stderr()
    exit(1)

print_as_tagged_json(parsed_toml)
exit(0)

Details on the tagged JSON is explained below in "JSON encoding".

Implementing an encoder

For your encoder to be compatible with toml-test, it must satisfy the expected interface:

Your encoder must accept JSON data on stdin.
If the JSON data cannot be converted to a valid TOML representation, your encoder must return with a non-zero exit code, indicating an error.
If the JSON data can be converted to a valid TOML representation, your encoder must output a TOML encoding of that data on stdout and return with a zero exit code, indicating success.

An example in pseudocode:

json_data = read_stdin()

parsed_json_with_tags = decode_json(json_data)

if error_parsing_json():
    print_error_to_stderr()
    exit(1)

print_as_toml(parsed_json_with_tags)
exit(0)

JSON encoding

The following JSON encoding applies equally to both encoders and decoders:

TOML tables correspond to JSON objects.
TOML table arrays correspond to JSON arrays.
TOML values correspond to a special JSON object of the form: {"type": "{TOML_TYPE}", "value": "{TOML_VALUE}"}

In the above, TOML_TYPE may be one of:

string
integer
float
bool
datetime
datetime-local
date-local
time-local

TOML_VALUE is always a JSON string.

Empty hashes correspond to empty JSON objects ({}) and empty arrays correspond to empty JSON arrays ([]).

Offset datetimes should be encoded in RFC 3339; Local datetimes should be encoded following RFC 3339 without the offset part. Local dates should be encoded as the date part of RFC 3339 and Local times as the time part.

Examples:

TOML                JSON

a = 42              {"type": "integer", "value": "42"}

[tbl]               {"tbl": {
a = 42                  "a": {"type": "integer", "value": "42"}
                    }}

a = ["a", 2]        {"a": [
                        {"type": "string",  "value": "a"},
                        {"type": "integer", "value": "2"}
                    ]}

Or a more complex example:

best-day-ever = 1987-07-05T17:45:00Z

[numtheory]
boring     = false
perfection = [6, 28, 496]

And the JSON encoding expected by toml-test is:

{
  "best-day-ever": {"type": "datetime", "value": "1987-07-05T17:45:00Z"},
  "numtheory": {
    "boring": {"type": "bool", "value": "false"},
    "perfection": [
      {"type": "integer", "value": "6"},
      {"type": "integer", "value": "28"},
      {"type": "integer", "value": "496"}
    ]
  }
}

Note that the only JSON values ever used are objects, arrays and strings.

An example implementation can be found in the BurnSushi/toml:

Untested and implementation-defined behaviour

This only tests behaviour that's should be true for every encoder implementing TOML; a few things are left up to implementations, and are not tested here.

TOML does not mandate a specific integer or float size, but recommends int64 and float64, which is what this tests. You'll have to manually -skip these tests if your implementation doesn't support it.
Many values can be expressed in more than one way: for example 0xff and 255 are equal, as are 0.0 and -0.0.

Some encoders may choose to always write back in a certain format (e.g. decimal), others may choose to use the same as the input format.

Testing how exactly a value is written in encoder tests is left up to the implementation, as long as they're semantically identical the test is considered to pass.
Millisecond precision (3 digits) is required for datetimes and times, and further precision is implementation-specific, and any greater precision than is supported must be truncated (not rounded).

This tests only millisecond precision, and not any further precision or the truncation of it.

Assumptions of Truth

The following are taken as ground truths by toml-test:

All tests classified as invalid are invalid.
All tests classified as valid are valid.
All expected outputs in valid/test-name.json are exactly correct.
The Go standard library package encoding/json decodes JSON correctly.
When testing encoders the BurntSushi/toml TOML decoder is assumed to be correct. (Note that this assumption is not made when testing decoders!)

Of particular note is that no TOML decoder is taken as ground truth when testing decoders. This means that most changes to the spec will only require an update of the tests in toml-test. (Bigger changes may require an adjustment of how two things are considered equal. Particularly if a new type of data is added.) Obviously, this advantage does not apply to testing TOML encoders since there must exist a TOML decoder that conforms to the specification in order to read the output of a TOML encoder.

Usage without `toml-test` binary

While the toml-test tool is a convenient way to run the tests, you can also re-implement its behaviour in your own language's test-suite, which may be an easier way to run the tests.

Because multiple TOML versions are supported, not all tests are valid for every version of TOML; for example the 1.0.0 tests contain a test that trailing commas in tables are invalid, but in 1.1.0 this should be considered valid.

In short: you can't "just" copy all .toml and .json files from the tests/ directory. The easiest way to copy the correct files is to use -copy:

# Default of TOML 1.0
% toml-test -copy ./tests

# Use TOML 1.1
% toml-test -copy ./tests -toml 1.1.0

Alternatively, the tests/files-toml-1.0.0 and tests/files-toml-1.1.0 files contain a list of files you should run for that TOML version. You can use them with something like:

<files-toml-1.0.0 while read l; do
    mkdir -p ~/my-test/"$(dirname "$l")"
    cp -r "$l" ~/my-test/"$l"
done

Adding tests

toml-test was designed so that tests can be easily added and removed. As mentioned above, tests are split into two groups: invalid and valid tests.

Invalid tests only check if a decoder rejects invalid TOML data. Or, in the case of testing encoders, invalid tests only check if an encoder rejects an invalid representation of TOML (e.g., a hetergeneous array). Therefore, all invalid tests should try to test one thing and one thing only. Invalid tests should be named after the fault it is trying to expose. Invalid tests for decoders are in the tests/invalid directory while invalid tests for encoders are in the tests/invalid-encoder directory.

Valid tests check that a decoder accepts valid TOML data and that the parser has the correct representation of the TOML data. Therefore, valid tests need a JSON encoding in addition to the TOML data. The tests should be small enough that writing the JSON encoding by hand will not give you brain damage. The exact reverse is true when testing encoders.

A valid test without either a .json or .toml file will automatically fail.

If you have tests that you'd like to add, please submit a pull request.

Why JSON?

In order for a language agnostic test suite to work, we need some kind of data exchange format. TOML cannot be used, as it would imply that a particular parser has a blessing of correctness.

My decision to use JSON was not a careful one. It was based on expediency. The Go standard library has an excellent encoding/json package built in, which made it easy to compare JSON data.

The problem with JSON is that the types in TOML are not in one-to-one correspondence with JSON. This is why every TOML value represented in JSON is tagged with a type annotation, as described above.

YAML may be closer in correspondence with TOML, but I don't believe we should rely on that correspondence. Making things explicit with JSON means that writing tests is a little more cumbersome, but it also reduces the number of assumptions we need to make.