pytest-respect 0.1.2

Pytest plugin to load resource files relative to test code and to expect values to match them.

pytest-respect

Pytest plugin to load resource files relative to test code and to expect values to match them. The name is a contraction of resources.expect, which is frequently typed when using this plugin.

Motivation

The primary use-case is running tests over moderately large datasets where adding them as constants in the test code would be cumbersome. This happens frequently with integration tests or when retrofitting tests onto an existing code-base. If you find your test code being obscured by the test data, filling with complex data generation code, or ad-hoc reading of input data or expected results, then pytest-respect is probably for you.

Installation

Install with your favourite package manager such as:

• pip install pydantic-respect
• poetry add --dev pydantic-respect
• uv add --dev pydantic-respect

See your package management tool for details, especially on how to install optional extra dependencies.

Extras

The following extra dependencies are required for additional functionality:

• poetry - Load, save, and expect pydantic models or arbitrary data through type adapters.
• numpy - Convert numpy arrays and scalars to python equivalents when generating JSON both in save and expect.
• jsonyx - Alternative JSON encoder for semi-compact files, numeric keys, trailing commas, etc.

Usage

Text Data

def test_translate(resources):
    input = resources.load_text("input")
    output = translate(input)
    resources.expect_text(output, "output")

If the test is found in a file called foo/test_stuff.py, then it will load the content of foo/test_stuff/test_translate__input.txt, run the translate function on it, and assert that the output exactly matches the content of the file foo/test_stuff/test_translate__output.json.

The expectation must match also on trailing spaces and trailing empty lines for the test to pass.

Json Data

A much more interesting example is doing the same with JSON data:

def test_compute(resources):
    input = resources.load_json("input")
    output = compute(input)
    resources.expect_json(output, "output")

This will load the content of foo/test_stuff/test_compute__input.json, run the compute function on it, and assert that the output exactly matches the content of the file foo/test_stuff/test_compute__output.json.

The expectation matching is done on a text representation of the JSON data. This avoids having to parse the expectation files, and allows us to use text-based diff tools, but instead we must avoid other tools reformating the expectations. By default the JSON formatting is by json.dumps(obj, sort_keys=True, indent=2) but see the section on JSON Formatting and Parsing.

Pydantic Models

With the optional pydantic extra, the same can be done with pydantic data if you have models for your input and output data:

def test_compute(resources):
    input = resources.load_pydantic(InputModel, "input")
    output = compute(input)
    resources.expect_pydantic(output, "output")

The input and output paths will be identical to the JSON test, since we re-used the name of the test function.

Failing Tests

If one of the above expectations fails, then a new file is created at foo/test_stuff/test_compute__output__actual.json with the actual value passed to the expect function, as well as the usual diff from pytest's assert. You can then use your existing diff tools to compare the expected and actual values and even to pick individual changes from the actual file before fixing the code to deal with any remaining differences.

Once the test passes, the actual file will be removed. Note that if you change the name of a test after an actual file has been created, the actual file will have to be deleted manually.

Parametric Tests

The load and expect (and other) methods can take multiple strings for the resource file name parts. Above we only used "input" and "output" parts and failures implicitly added an "actual" part. We can pass in as many parts as we like, which nicely brings us to parametric tests:

@pytest.mark.paramtrize("case", ["red", "blue", "green"])
def test_compute(resources, case):
    input = resources.load_json("input", case)
    output = compute(input)
    resources.expect_json(output, "output", case)

Omitting the directory name, this test will load each of test_compute__input__red.json, test_compute__input__blue.json, test_compute__input__green.json and compare the results to test_compute__output__red.json, test_compute__output__blue.json, test_compute__output__green.json

Data-driven Parametric Tests

• To Document: using list

JSON Formatting and Parsing

To Document:*

• Default JSON formatter and parser
• Alternative JSON formatter
• Jsonyx extension

Resource Path Construction

To Document:

• Multiple path parts
• Default path maker
• Alternative path makers
• Custom path makers

Development

Installation

• Install uv
• Run uv sync --all-extras
• Run pre-commit install to enable pre-commit linting.
• Run pytest to verify installation.

Testing

This is a pytest plugin so you're expected to know how to run pytest when hacking on it. Additionally, scripts/pytest-extras runs the test suite with different sets of optional extras. The CI Pipelines will go through an equivalent process for each Pull Request.