pytest-resources

Lazy, typed, attribute-navigable fixtures for test resource files — JSON / JSONC / YAML / TOML / INI decoded through e-serde by default, extensible to any canonical bytes -> object loader, and anything else handed back as raw bytes.

You point it at one or more directories of sample files; it recursively indexes the tree without reading anything, and you reach, filter, randomise and synthesize parsed values the way they read:

from dataclasses import dataclass

import pytest_resources as pr


@dataclass
class User:
    name: str
    age: int


def test_profile(resources: pr.Resources):
    user = resources.structured.user          # sample dir -> dict (decoded, cached)
    body = resources.unstructured.intro       # a .md -> str
    blob = resources.binary.logo              # no codec bound -> raw bytes

    jsons = resources.data.select("*.json")   # filter a folder (glob / regex / kind)
    pick = resources.data.choice("*.json")    # random pick, seeded per run

    ada = resources.make(User, name="ada")    # random model instance    [objects]
    squad = resources.batch(User, 8)          # a typed list of them     [objects]
    pdf = resources.file("report.pdf")        # synthetic file, indexed [files]

Install

uv add --group test "pytest-resources[serde]"   # e-serde default codecs
uv add --group test pytest-resources            # stdlib-only (json + tomllib)
uv add --group test "pytest-resources[objects]" # resources.make()/batch(): random model objects
uv add --group test "pytest-resources[files]"   # resources.file(): synthetic files, 25 formats
uv add --group test "pytest-resources[random]"  # both

The resources fixture auto-loads via the pytest11 entry point — no imports in your conftest.py. Everything (index, navigation, synthesis) is async-test friendly: make/batch are pure in-memory CPU and abuild_resources/awalk keep the loop free; calling a synthesis method without its extra raises ExtraNotInstalledError with the exact install command — the plugin never hard-depends on polyfactory or faker-file.

What you get

Configuring the roots

By default the plugin indexes <rootdir>/tests/resources (and creates it if missing). Point it at one or several directories:

# pyproject.toml
[tool.pytest.ini_options]
resources_root = ["fixtures", "fixtures/shared"]   # default: <rootdir>/tests/resources

Or per run (--resources-root is repeatable), or additively from your conftest.py:

from pathlib import Path

def pytest_resources_roots(roots):
    roots.append(Path("tests/fixtures/shared"))

CLI wins over ini, which wins over the default. Every resolved root is merged into one navigable tree; on a top-level clash the later root wins.

Custom loaders (the extension seam)

The plugin never assumes e-serde. Bind any parser declaratively in your conftest.py:

from pytest_resources import FileType
import yaml, csv, io

def pytest_resource_loaders(register):
    register({FileType.CSV: lambda b: list(csv.DictReader(io.StringIO(b.decode())))})
    register({FileType.YAML: yaml.safe_load})

register merges into the session's table, so you can override a default kind as easily as add an unknown one. Kinds left unbound hand back raw bytes.

Programmatic use

Outside pytest, build a tree yourself:

from pathlib import Path
import pytest_resources as pr

tree = pr.build_resources(Path("fixtures"))                      # sync, one root
tree = pr.build_resources([Path("a"), Path("b")])               # merged roots, later wins
tree = await pr.abuild_resources(Path("fixtures"))              # off the event loop
table = pr.default_loaders()                                    # best available codecs

Development

make install   # uv sync + git hooks
make check     # ruff + ty + tach + validate + tests
make ci        # everything the pipeline runs (adds coverage gate + docs + zizmor)

License

MIT