pytest-regressions provides some fixtures that make it easy to maintain tests that generate lots of data or specific data files like images.

This plugin uses a data directory (courtesy of pytest-datadir) to store expected data files, which are stored and used as baseline for future test runs.


Let’s use data_regression as an example, but the workflow is the same for the other *_regression fixtures.

Suppose we have a summary_grids function which outputs a dictionary containing information about discrete grids for simulation. Of course your function would actually return some computed/read value, but here it is using an inline result for this example:

def summary_grids():
    return {
        "Main Grid": {
            "id": 0,
            "cell_count": 1000,
            "active_cells": 300,
            "properties": [
                {"name": "Temperature", "min": 75, "max": 85},
                {"name": "Porosity", "min": 0.3, "max": 0.4},
        "Refin1": {
            "id": 1,
            "cell_count": 48,
            "active_cells": 44,
            "properties": [
                {"name": "Temperature", "min": 78, "max": 81},
                {"name": "Porosity", "min": 0.36, "max": 0.39},

We could test the results of this function like this:

def test_grids():
    data = summary_grids()
    assert data["Main Grid"]["id"] == 0
    assert data["Main Grid"]["cell_count"] == 1000
    assert data["Main Grid"]["active_cells"] == 300
    assert data["Main Grid"]["properties"] == [
        {"name": "Temperature", "min": 75, "max": 85},
        {"name": "Porosity", "min": 0.3, "max": 0.4},

But this presents a number of problems:

  • Gets old quickly.
  • Error-prone.
  • If a check fails, we don’t know what else might be wrong with the obtained data.
  • Does not scale for large data.
  • Maintenance burden: if the data changes in the future (and it will) it will be a major headache to update the values, specially if there are a lot of similar tests like this one.

Using data_regression

The data_regression fixture provides a method to check general dictionary data like the one in the previous example.

There is no need to import anything, just declare the data_regression fixture in your test’s arguments and call the check method in the test:

def test_grids2(data_regression):
    data = summary_grids()

The first time your run this test, it will fail with a message like this:

>           pytest.fail(msg)
E           Failed: File not found in data directory, created:
E           - C:\Users\bruno\pytest-regressions\tests\test_grids\test_grids2.yml

The fixture will generate a test_grids2.yml file (same name as the test) in the data directory with the contents of the dictionary:

Main Grid:
  active_cells: 300
  cell_count: 1000
  id: 0
  - max: 85
    min: 75
    name: Temperature
  - max: 0.4
    min: 0.3
    name: Porosity
  active_cells: 44
  cell_count: 48
  id: 1
  - max: 81
    min: 78
    name: Temperature
  - max: 0.39
    min: 0.36
    name: Porosity

This file should be committed to version control.

The next time you run this test, it will compare the results of summary_grids() with the contents of the YAML file. If they match, the test passes. If they don’t match the test will fail, showing a nice diff of the text differences.


If the test fails because the new data is correct (the implementation might be returning more information about the grids for example), then you can use the --force-regen flag to update the expected file:

$ pytest --force-regen

This will fail the same test but with a different message saying that the file has been updated. Commit the new file.

This workflow makes it very simple to keep the files up to date and to check all the information we need.


If a single change will fail several regression tests, you can also use the --regen-all command-line flag:

$ pytest --regen-all

With this flag, the regression fixtures will regenerate all files but will not fail the tests themselves. This make it very easy to update all regression files in a single pytest run when individual tests contain multiple regressions.

Parametrized tests

When using parametrized tests, pytest will give each parametrization of your test a unique name. This means that pytest-regressions will create a new file for each parametrization too.

Suppose we have an additional function summary_grids_2 that generates longer data, we can re-use the same test with the @pytest.mark.parametrize decorator:

@pytest.mark.parametrize('data', [summary_grids(), summary_grids_2()])
def test_grids3(data_regression, data):

Pytest will automatically name these as test_grids3[data0] and test_grids3[data1], so files test_grids3_data0.yml and test_grids3_data1.yml will be created.

The names of these files can be controlled using the ids keyword for parametrize, so instead of data0, you can define more useful names such as short and long:

@pytest.mark.parametrize('data', [summary_grids(), summary_grids_2()], ids=['short' 'long'])
def test_grids3(data_regression, data):

which creates test_grids3_short.yml and test_grids3_long.yml respectively.