rtest 0.0.38

Python test runner built in Rust

rtest

A Python test runner built with Rust, featuring high-performance test collection and parallel test execution.

⚠️ Development Status: This project is in early development (v0.0.x). Expect bugs, breaking changes, and evolving features as we work toward stability.

Performance

Test Collection (--collect-only)

Benchmarks performed using hyperfine with the following command:

hyperfine --warmup 3 --min-runs 20 --max-runs 20 \
  --command-name pytest --command-name rtest \
  ".venv/bin/pytest --collect-only -q tests" \
  ".venv/bin/rtest --collect-only tests"

Repository	pytest	rtest	Speedup
flask	421 ms ± 12 ms	43 ms ± 3 ms	9.81x
click	570 ms ± 341 ms	86 ms ± 8 ms	6.61x
httpx	1.71 s ± 0.76 s	51 ms ± 5 ms	33.53x
pydantic	2.84 s ± 0.24 s	121 ms ± 9 ms	23.42x
fastapi	2.98 s ± 0.29 s	107 ms ± 14 ms	27.80x
more-itertools	209 ms ± 20 ms	28 ms ± 1 ms	7.39x

Test Execution (--runner native)

For repositories that don't rely on pytest fixtures or conftest.py, the native runner can execute tests directly:

hyperfine --warmup 3 --min-runs 20 --max-runs 20 \
  --command-name pytest --command-name rtest \
  ".venv/bin/pytest tests" \
  ".venv/bin/rtest --runner native tests"

Repository	pytest	rtest	Speedup
more-itertools	17.96 s ± 2.17 s	2.49 s ± 0.41 s	7.20x

Note: Native runner execution benchmarks are limited to repositories that use simple test patterns (unittest.TestCase, plain assertions) without pytest fixtures. Most real-world projects use fixtures and conftest.py, which require the native runner's fixtures support (in development).

Test Execution (--runner pytest -n 4)

For repositories that use pytest fixtures and conftest.py, rtest can use pytest as the execution backend while still benefiting from fast Rust-based collection:

hyperfine --warmup 3 --min-runs 20 --max-runs 20 \
  --command-name pytest --command-name rtest \
  ".venv/bin/pytest -n 4 tests" \
  ".venv/bin/rtest --runner pytest -n 4 tests"

Repository	pytest	rtest	Speedup
flask	2.04 s ± 0.20 s	0.91 s ± 0.11 s	2.24x
click	2.78 s ± 0.64 s	1.12 s ± 1.30 s	2.48x

Note: The --runner pytest mode uses pytest for test execution, so all pytest features (fixtures, plugins, markers) work normally. The speedup comes from rtest's faster test collection phase.

Quick Start

Installation

pip install rtest

Requires Python 3.10+

Basic Usage

# Collect tests (fast AST-based collection)
rtest --collect-only

# Run tests with native runner
rtest --runner native

# Run tests in parallel (4 workers)
rtest --runner native -n 4

Native Runner

The native runner (--runner native) executes tests using rtest's own decorators:

import rtest

@rtest.mark.cases("value", [1, 2, 3])
def test_example(value):
    assert value > 0

@rtest.mark.skip(reason="Not implemented yet")
def test_skipped():
    pass

The native runner respects python_files, python_classes, and python_functions patterns from your pyproject.toml under [tool.pytest.ini_options].

For compatibility, @pytest.mark.parametrize and @pytest.mark.skip decorators are also supported.

Roadmap

• Fixtures - @rtest.fixture with function/class/module scopes and dependency resolution
• conftest.py support - Fixture discovery across directory hierarchy
• Distribution modes - Group tests by module/class, optimized scheduling algorithms
• Dynamic @rtest.mark.cases evaluation - Support variables, function calls, and comprehensions in decorator arguments
• Built-in assertions - rtest.raises() and other assertion helpers
• Additional markers - @rtest.mark.xfail, @rtest.mark.skipif
• Test selection - -k expression filtering, --last-failed, --failed-first
• Better error formatting - Rich diffs, colorized output, traceback filtering
• Async test support - Native async def test_*() handling
• Watch mode - Re-run tests automatically on file changes

Known Limitations

Parametrized Test Discovery

rtest expands parametrized tests during collection when the decorator arguments are literal values (numbers, strings, booleans, None, lists/tuples of literals). For example:

@pytest.mark.parametrize("value", [1, 2, 3])
def test_example(value):
    assert value > 0

Both pytest and rtest collection show:

test_example[0]
test_example[1]
test_example[2]

However, when parametrize arguments contain dynamic expressions (variables, function calls, comprehensions), rtest cannot statically analyze them and will emit a warning while falling back to the base test name:

DATA = [1, 2, 3]

@pytest.mark.parametrize("value", DATA)  # Dynamic - references a variable
def test_example(value):
    assert value > 0

warning: Cannot statically expand test cases for 'test.py::test_example': argvalues references variable 'DATA'

In these cases, test execution is still functionally equivalent - pytest automatically runs all parametrized variants when given the base function name.

Path Separator Handling

rtest uses platform-specific path separators in test nodeids, while pytest normalizes all paths to use forward slashes (/) regardless of platform. For example:

On Windows:

• pytest shows: tests/unit/test_example.py::test_function
• rtest shows: tests\unit\test_example.py::test_function

On Unix/macOS:

• Both show: tests/unit/test_example.py::test_function

This difference is intentional as rtest preserves the native path format of the operating system.

Contributing

We welcome contributions! See Contributing Guide.

License

MIT - see LICENSE file for details.

---

Acknowledgments

This project takes inspiration from Astral and leverages crates from [ruff].