pytest-data-loader

pytest-data-loader is a pytest plugin that simplifies data-driven testing. It lets you load, transform, and parametrize test data directly from files and directories using simple decorators.

Installation

pip install pytest-data-loader

Quick Start

Load test data from a file and inject it directly into your test function.

from pytest_data_loader import load


@load("data", "example.json")
def test_example(data):
    """
    example.json: '{"foo": 1, "bar": 2}'
    """
    assert isinstance(data, dict)
    assert data["foo"] == 1

Usage

The plugin provides three data loaders — @load, @parametrize, and @parametrize_dir — available as decorators for loading test data.

@load: Loads the file content into a test
@parametrize: Loads a file and parametrizes a test by splitting its content into logical parts (e.g. lines, JSON items, etc.)
@parametrize_dir: Loads files from a directory and parametrizes a test for each file

Each data loader requires two positional arguments:

fixture_names: Names of the fixtures injected into the test function
- Single name: Injects the file data
- Two names: Injects both the resolved file path and file data
path: An absolute path or a path relative to a data directory
- When a relative path is given, the plugin searches upward from the test file for the nearest data directory that contains the target file or directory
- For @parametrize and @parametrize_dir, this can also be a list of paths, a glob pattern, or a list that combines both to aggregate data from multiple sources
- Environment variables are supported using the ${VAR} or $VAR (or %VAR% on Windows) syntax

Note

If your data path is dynamic and unknown until runtime, use the data_loader fixture as a programmatic alternative to @load. See The data_loader Fixture

Tip

The default data directory name can be customized using an INI option. See INI Options
Each data loader supports different optional keyword arguments to customize how the data is loaded. See Data Loading Pipeline and Loader Options
Each data loader can be stacked on a test function. See Stacking Data Loaders

Examples

Given the following project structure:

.(pytest rootdir)
├── data/               # shared data directory
│   ├── data1.json
│   ├── data2.txt
│   └── images/
│       ├── image.gif
│       ├── image.jpg
│       └── image.png
└── tests/
    ├── data/           # local data directory
    │   ├── data1.txt
    │   ├── data2.txt
    │   └── logos/
    │       ├── logo.jpg
    │       └── logo.png
    └── test_something.py

Note

Relative paths are resolved to the nearest data directory from the test file
In this example, the tests/data/ directory takes precedence over the shared data/ directory when both contain the same file
This path resolution rule applies to all data loaders

1. Load file data — `@load`

@load is a file loader that loads the file content and passes it to the test function.

# test_something.py

from pytest_data_loader import load


@load("data", "data1.json")
def test_something1(data):
    """
    data/data1.json: '{"foo": 1, "bar": 2}'
    """
    assert data == {"foo": 1, "bar": 2}


@load(("file_path", "data"), "data2.txt")
def test_something2(file_path, data):
    """
    tests/data/data2.txt: "line1\nline2\nline3"
    """
    assert file_path.name == "data2.txt"
    assert data == "line1\nline2\nline3"

$ pytest -v
================================ test session starts =================================
<snip>
collected 2 items                                                                              

tests/test_something.py::test_something1[data1.json] PASSED                     [ 50%]
tests/test_something.py::test_something2[data2.txt] PASSED                      [100%]

================================= 2 passed in 0.01s ==================================

Note

For dynamic paths, use the data_loader fixture instead. See The data_loader Fixture

2. Parametrize file data — `@parametrize`

@parametrize is a file loader that dynamically parametrizes the decorated test function by splitting the file content into logical parts. Each part is passed to the test function as a separate parameter.

# test_something.py

from pytest_data_loader import parametrize


@parametrize("data", "data1.json")
def test_something1(data):
    """
    data/data1.json: '{"foo": 1, "bar": 2}'
    """
    # parametrized as key–value pairs
    assert data in [("foo", 1), ("bar", 2)]


@parametrize(("file_path", "data"), "data2.txt")
def test_something2(file_path, data):
    """
    tests/data/data2.txt: "line1\nline2\nline3"
    """
    # parametrized as lines
    assert file_path.name == "data2.txt"
    assert data in ["line1", "line2", "line3"]

$ pytest -v
================================ test session starts =================================
<snip>
collected 5 items                                                                              

tests/test_something.py::test_something1[data1.json:part1] PASSED               [ 20%]
tests/test_something.py::test_something1[data1.json:part2] PASSED               [ 40%]
tests/test_something.py::test_something2[data2.txt:part1] PASSED                [ 60%]
tests/test_something.py::test_something2[data2.txt:part2] PASSED                [ 80%]
tests/test_something.py::test_something2[data2.txt:part3] PASSED                [100%]

================================= 5 passed in 0.01s ==================================

Tip

You can apply your own logic by specifying the parametrizer loader option
By default, the plugin will apply the following logic for splitting file content:
- Text file: Each line
- JSON file:
  - object: Each key–value pair in the object
  - array: Each item in the array
  - other types (string, number, boolean, null): The whole content as single data
- JSONL file: Each line (parsed as JSON)
- Binary file: Unsupported by default. You must provide a custom split logic as a parametrizer loader option

Parametrize from multiple files

You can pass a list of file paths, a glob pattern, or a list that combines both to @parametrize to load and concatenate data from multiple files into a single parameter list:

# test_something.py

from pytest_data_loader import parametrize


@parametrize("data", "*.txt")   # or ["data1.txt", "data2.txt"]
def test_something(data):
    """
    The glob pattern matches: 
      tests/data/data1.txt: "line1\nline2"
      tests/data/data2.txt: "line3\nline4"
    """
    assert data in ["line1", "line2", "line3", "line4"]

$ pytest -v
================================ test session starts =================================
<snip>
collected 4 items

tests/test_something.py::test_something[data1.txt:part1] PASSED                 [ 25%]
tests/test_something.py::test_something[data1.txt:part2] PASSED                 [ 50%]
tests/test_something.py::test_something[data2.txt:part1] PASSED                 [ 75%]
tests/test_something.py::test_something[data2.txt:part2] PASSED                 [100%]

================================= 4 passed in 0.01s ==================================

3. Parametrize files in a directory — `@parametrize_dir`

@parametrize_dir is a directory loader that dynamically parametrizes the decorated test function with the contents of the files in the specified directory. Each file's content is passed to the test function as a separate parameter.

# test_something.py

from pytest_data_loader import parametrize_dir


@parametrize_dir("data", "images")
def test_something(data):
    """
    data/images/: contains 3 image files
    """
    # parametrized as files
    assert isinstance(data, bytes)

$ pytest -v
================================ test session starts =================================
<snip>
collected 3 items                                                                              

tests/test_something.py::test_something[images/image.gif] PASSED                [ 33%]
tests/test_something.py::test_something[images/image.jpg] PASSED                [ 66%]
tests/test_something.py::test_something[images/image.png] PASSED                [100%]

================================= 3 passed in 0.01s ==================================

Note

Use the recursive=True option to include files in subdirectories
Directory and file names starting with a dot (.) are considered hidden regardless of your platform. These are automatically excluded from the parametrization

Parametrize files from multiple directories

You can pass a list of directory paths, a glob pattern, or a list that combines both to @parametrize_dir to collect and concatenate files from multiple directories into a single parameter list:

# test_something.py

from pytest_data_loader import parametrize_dir


@parametrize_dir("data", ["images", "logos"])
def test_something(data):
    """
    data/images/: contains 3 image files
    tests/data/logos/: contains 2 logo files
    """
    assert isinstance(data, bytes)

$ pytest -v
================================ test session starts =================================
<snip>
collected 5 items

tests/test_something.py::test_something[images/image.gif] PASSED                [ 20%]
tests/test_something.py::test_something[images/image.jpg] PASSED                [ 40%]
tests/test_something.py::test_something[images/image.png] PASSED                [ 60%]
tests/test_something.py::test_something[logos/logo.jpg] PASSED                  [ 80%]
tests/test_something.py::test_something[logos/logo.png] PASSED                  [100%]

================================= 5 passed in 0.01s ==================================

Stacking Data Loaders

All three data loaders — @load, @parametrize, and @parametrize_dir — can be stacked on a single test function. This allows you to declaratively compose complex, data-driven test scenarios while keeping test logic fully decoupled from data.

Examples:

1. Load multiple datasets

Stack multiple @load loaders to inject independent datasets into a single test.

from pytest_data_loader import load


@load("input_data", "input.json")
@load("expected_output", "expected.json")
def test_transformation_matches_expected_output(input_data, expected_output):
    """Verify that transforming input data produces the expected output."""
    assert do_something(input_data) == expected_output

2. Generate a Cartesian product of test cases

Stack multiple @parametrize loaders to automatically test all combinations.

from pytest_data_loader import parametrize


@parametrize("user", "users.txt")
@parametrize("feature", "features.txt")
def test_user_feature_access_matrix(user, feature):
    """Validate access control for every user-feature combination."""
    assert can_access(user, feature)

3. Combine shared context with parametrized inputs

Stack @load loader with @parametrize loader to test variable inputs with shared context.

from pytest_data_loader import load, parametrize


@load("prices", "prices.json")
@parametrize("order", "orders.json")
def test_order_total_matches_expected(prices, order):
    """Validate that each order total is calculated correctly using the shared price catalog."""
    total = calculate_total(order, prices)
    assert total == order["expected_total"]

4. Combine shared context with directory-based test scenarios

Stack @load loader with @parametrize_dir loader to test structured test cases with shared context.

from pytest_data_loader import load, parametrize_dir


@load("banned_words", "banned_words.txt")
@parametrize_dir("comment", "user_comments/flagged")    # Each comment is stored as a .txt file
def test_flagged_comments_contain_banned_words(banned_words, comment):
    """Validate that flagged comments contain at least one banned word."""
    assert any(word in comment.lower() for word in banned_words)

Note

Fixture names must be unique across all stacked loaders on a test function
Stacking multiple @parametrize and/or @parametrize_dir loaders generates a Cartesian product of N × M test cases (same behavior as pytest.mark.parametrize)
Files are loaded once per test function and cached across parametrized test cases

Tip

When stacking data loaders, test IDs generated with the default parameter IDs may become less readable. Consider explicitly specifying parameter IDs using the id option (@load) or the ids option (@parametrize/@parametrize_dir)

The `data_loader` Fixture

The plugin provides a function-scoped data_loader fixture as an alternative to @load. Use this fixture when the file path is not known until test runtime — for example, when it depends on another fixture, a parametrized value, or a CLI option, etc. The fixture provides a callable (an instance of the DataLoaderFixture class) that accepts a file path and returns the loaded data. It uses the same path resolution and loading logic as @load. Loader options like reader, read_options, and onload are also supported and can be passed as keyword arguments.
Below is an example where the file path depends on both a custom CLI option and parametrized test inputs, which is a use case @load cannot support:

import pytest
from pytest import FixtureRequest

from pytest_data_loader import DataLoaderFixture


@pytest.fixture(scope="session")
def env(request: FixtureRequest) -> str:
    """Target environment specified by the custom --env CLI option"""
    return request.config.getoption("--env")


@pytest.mark.parametrize("filename", ["case1.json", "case2.json"])
def test_env_specific_cases(data_loader: DataLoaderFixture, env: str, filename: str):
    # Construct path from a CLI option and parametrized value
    path = f"{env}/{filename}"
    case_data = data_loader(path)
    assert isinstance(case_data, dict)

Tip

You can combine the data_loader fixture with @load, @parametrize, and @parametrize_dir in the same test function. This is useful when some data paths are static while others are determined dynamically at runtime

Lazy Loading

Lazy loading is enabled by default for all data loaders to improve efficiency, especially with large datasets. During test collection, pytest receives a lightweight lazy object instead of the actual data. The data is resolved only when it is needed during test setup.
If you need to disable this behavior for a specific test, pass lazy_loading=False to the data loader.

Note

Lazy loading for the @parametrize loader works slightly differently from other loaders. Since pytest needs to know the total number of parameters in advance, the plugin still needs to inspect the file data and split it once during the test collection phase. But once it's done, the split data will not be kept as parameter values and will be loaded lazily at test setup

Data Loading Pipeline

Each data loader follows a simple pipeline where you can use loader options to hook into stages and filter or transform data before it reaches your test.

@load

file 
  → open                 # with read options
  → read and parse       # with reader()
  → transform            # with onload()
  → test(data)

@parametrize

file 
  → open                 # with read options 
  → read and parse       # with reader() 
  → transform            # with onload()
  → split                # with default or custom parametrizer()
    ↳ for each part:
      → filter           # with filter()
      → transform        # with processor()
  → test(part₁, part₂, ...)

@parametrize_dir

directory 
  → collect files 
    ↳ for each file:
      → filter           # with filter()
      → open             # with read options
      → read and parse   # with reader()
      → transform        # with processor()
  → test(file₁, file₂, ...)

File Reader

Built-in defaults

By default, the plugin reads and parses file content when loading as follows:

.json — Parsed with json.load
.jsonl — Each line is parsed as a JSON object
All other file types — Loads as raw text or binary content

Compressed files with a .gz, .bz2, or .xz extension are decompressed transparently. The inner file type determines which reader and read mode are used (e.g. data.json.gz uses json.load, data.txt.bz2 loads as plain text).

Customizing defaults

You can customize this behavior by specifying a file reader that accepts a file-like object returned by open(). This includes built-in readers, third-party library readers, and your own custom readers. File read options (e.g., mode, encoding, etc.) can also be provided as a read_options dict and will be passed to open().

Below are some common examples of file readers you might use:

File type	Examples	Notes
.csv	`csv.reader`, `csv.DictReader`, `pandas.read_csv`	`pandas.read_csv` requires `pandas`
.yml	`yaml.safe_load`, `yaml.safe_load_all`	Requires `PyYAML`
.xml	`xml.etree.ElementTree.parse`
.toml	`tomllib.load`	`tomli.load` for Python <3.11 (Requires `tomli`)
.ini	`configparser.ConfigParser().read_file`
.pdf	`pypdf.PdfReader`	Requires `pypdf`

This can be done either as a conftest.py-level registration or as a test-level configuration. If both are done, the test-level configuration takes precedence over conftest.py-level registration. If multiple conftest.py files register a reader for the same file extension, the closest one to the current test becomes effective.

Here are some examples of loading a CSV file using the built-in CSV readers with file read options:

1. `conftest.py`-level registration

Register a file reader using pytest_data_loader.register_reader(). It takes a file extension and a file reader as positional arguments, and an optional read_options dict.

# conftest.py

import csv

import pytest_data_loader


pytest_data_loader.register_reader(".csv", csv.reader, read_options={"newline": ""})

The registered file reader automatically applies to all tests located in the same directory and any of its subdirectories.

# test_something.py

from pytest_data_loader import load


@load("data", "data.csv")
def test_something(data):
    """Load CSV file with registered file reader"""
    for row in data:
        assert isinstance(row, list)

2. Per-test configuration with loader options

Specify a file reader with the reader loader option. This applies only to the configured test, and overrides the one registered in conftest.py.

# test_something.py

import csv

from pytest_data_loader import load, parametrize

read_options = {"encoding": "utf-8-sig", "newline": ""}


@load("data", "data.csv", reader=csv.reader, read_options=read_options)
def test_something1(data):
    """Load CSV file with csv.reader reader"""
    for row in data:
        assert isinstance(row, list)


@parametrize("data", "data.csv", reader=csv.DictReader, read_options=read_options)
def test_something2(data):
    """Parametrize CSV file data with csv.DictReader reader"""
    assert isinstance(data, dict)

Note

If read options are specified without a reader, the plugin uses the conftest.py-registered reader (if any) with those options. If a reader is specified without read options, no read options are applied

Tip

A file reader must take one argument (a file-like object returned by open())
If you need to pass options to the file reader, use lambda function or a regular function
e.g. reader=lambda f: csv.reader(f, delimiter=";")
You can adjust the final data the test function receives using loader options. For example, the following code will parametrize the test with the text data from each PDF page

@parametrize(
    "page_data", 
    "test.pdf", 
    reader=pypdf.PdfReader, 
    read_options={"mode": "rb"},
    parametrizer=lambda r: r.pages,
    processor=lambda p: p.extract_text().rstrip(),
)
def test_something(page_data: str):
    ...

Caching

To reduce repeated I/O and parsing work during a test session, the plugin uses two layers of caching:

Session-scoped cache
- Shared across all data loaders
- Caches raw file contents by resolved file path
- Maintains a bounded pool of open file handles using an LRU strategy
Per-loader cache
- Scoped to an individual loader on a test function
- Caches the loaded data after the per-file loader pipeline
- Reduces repeated parsing and transformation work across parametrized test cases that reuse the same source data

The session-scoped cache can be tuned or disabled via the INI options.

Loader Options

Each data loader supports different optional parameters you can use to change how your data is loaded.

@load

lazy_loading: Enable or disable lazy loading
reader: A file reader the plugin should use to read the file data
read_options: File read options (as a dict) the plugin passes to open(). Supports only the mode, encoding, errors, and newline keys
onload: A function to transform or process loaded data before passing it to the test function
marks: Pytest mark(s) to apply to the loaded data. Accepts a single mark or a collection of marks
id: The parameter ID for the loaded data. If not specified, the relative or absolute file path is used

Note

onload must take either one (data) or two (file path, data) arguments. When reader is provided, its return value becomes the data passed to onload()

@parametrize

lazy_loading: Enable or disable lazy loading
reader: A file reader the plugin should use to read the file data
read_options: File read options (as a dict) the plugin passes to open(). Supports only the mode, encoding, errors, and newline keys
onload: A function to adjust the shape of the loaded data before splitting into parts
parametrizer: A function to customize how the loaded data should be split
filter: A function to filter the split data parts. Only matching parts are included as test parameters
processor: A function to adjust the shape of each part data before passing it to the test function
marks: Pytest mark(s) for the loaded parts. Accepts a single mark or a collection of marks applied uniformly to all parts, or a function that returns mark(s) per part data
ids: Parameter IDs for the loaded parts. Accepts an iterable of ID values or a function that returns an ID per part data

Note

onload, parametrizer, and filter must take one (data) or two (file_path, data) arguments
processor and callable marks and ids additionally accept a three-argument form (idx, file_path, data), where idx is the zero-based post-filter position of the item
When reader is provided, its return value becomes the data passed to these callables

@parametrize_dir

lazy_loading: Enable or disable lazy loading
recursive: Recursively load files from all subdirectories of the given directory. Defaults to False. This option is ignored for glob patterns. Use ** instead for recursive matching
reader: A function that determines the file reader for each file path
read_options: File read options the plugin passes to open() when reading files. Accepts a dict applied uniformly to all files, or a function that returns the options for matching file paths. Supports only the mode, encoding, errors, and newline keys
filter: A function to filter file paths. Only the contents of matching file paths are included as the test parameters
processor: A function to adjust the shape of each loaded file's data before passing it to the test function
marks: Pytest mark(s) for the loaded files. Accepts a single mark or a collection of marks applied uniformly to all files, or a function that returns mark(s) for matching file paths
ids: Parameter IDs for the loaded files. Accepts an iterable of ID values or a function that returns an ID to matching file path

Note

filter must take only one argument (file_path)
reader and callable read_options, marks, and ids additionally accept a two-argument form (idx, file_path), where idx is the zero-based post-filter position of the file
processor may take one (data), two (file_path, data), or three (idx, file_path, data) arguments, where idx is the zero-based post-filter position of the file

INI Options

`data_loader_dir_name`

The base directory name to load test data from. When a relative file or directory path is provided to a data loader, it is resolved relative to the nearest matching data directory in the directory tree.
Plugin default: data

`data_loader_root_dir`

Absolute or relative path to the project's root directory. By default, the search is limited to within pytest's rootdir, which may differ from the project's top-level directory. Setting this option allows data directories located outside pytest's rootdir to be found. Environment variables are supported using the ${VAR} or $VAR (or %VAR% on Windows) syntax.
Plugin default: Pytest rootdir (config.rootpath)

`data_loader_strip_trailing_whitespace`

Automatically remove trailing whitespace characters when loading text data.
Plugin default: true

`data_loader_on_missing`

The action to take when a data file or directory specified as path cannot be located. Supported values are raise, skip, xfail, and warn.
Plugin default: raise

`data_loader_default_encoding`

The default text encoding to use when opening data files in text mode, applied when the data loader's read_options does not explicitly specify an encoding.
Plugin default: utf-8

Note

Binary auto-detection becomes less reliable with some encodings and may incorrectly classify binary files as text:

Single-byte permissive codecs (e.g. latin-1, cp1252) can decode any byte sequence without errors.
Multi-byte codecs (e.g. utf-16, utf-32) legitimately contain null bytes in valid text, which disables the null-byte fast-path used by the binary detector.

If you use one of these encodings as the default, configure data loaders for known binary formats with read_options={"mode": "rb"} to force binary reads.

`data_loader_max_cache_size`

Maximum total memory allocated to the session-scoped LRU cache for raw file contents. When the cache exceeds this limit, the least recently used entries are automatically evicted until usage falls below the configured size.
Supports decimal units (KB, MB, GB, etc.) and binary units (KiB, MiB, GiB, etc.). A bare integer with no unit is interpreted as bytes. Set to 0 to disable raw-content caching entirely.
Plugin default: 128MiB (134,217,728 bytes)

`data_loader_max_open_files`

Maximum number of open file handles retained in the session-scoped file-handle pool. When the pool reaches this limit, the least recently used handle is closed before a new handle is added. Set to 0 to disable handle pooling.
Plugin default: 64

Name		Name	Last commit message	Last commit date
Latest commit History 88 Commits
.github		.github
src/pytest_data_loader		src/pytest_data_loader
tests		tests
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
README.md		README.md
pyproject.toml		pyproject.toml
pytest.ini		pytest.ini
tox.ini		tox.ini

Folders and files

Latest commit

History

Repository files navigation

pytest-data-loader

Installation

Quick Start

Usage

Examples

1. Load file data — @load

2. Parametrize file data — @parametrize

3. Parametrize files in a directory — @parametrize_dir

Stacking Data Loaders

Examples:

1. Load multiple datasets

2. Generate a Cartesian product of test cases

3. Combine shared context with parametrized inputs

4. Combine shared context with directory-based test scenarios

The data_loader Fixture

Lazy Loading

Data Loading Pipeline

@load

@parametrize

@parametrize_dir

File Reader

Built-in defaults

Customizing defaults

1. conftest.py-level registration

2. Per-test configuration with loader options

Caching

Loader Options

@load

@parametrize

@parametrize_dir

INI Options

data_loader_dir_name

data_loader_root_dir

data_loader_strip_trailing_whitespace

data_loader_on_missing

data_loader_default_encoding

data_loader_max_cache_size

data_loader_max_open_files

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases 14

Uh oh!

Contributors

Uh oh!

Languages

1. Load file data — `@load`

2. Parametrize file data — `@parametrize`

3. Parametrize files in a directory — `@parametrize_dir`

The `data_loader` Fixture

1. `conftest.py`-level registration

`data_loader_dir_name`

`data_loader_root_dir`

`data_loader_strip_trailing_whitespace`

`data_loader_on_missing`

`data_loader_default_encoding`

`data_loader_max_cache_size`

`data_loader_max_open_files`