Requirement Manager

This project provides a RequirementManager (requires is an alias) class to manage Python package requirements using virtual environments. It can be used as a decorator or context manager to ensure specific packages are installed and available during the execution of a function or code block.

Features

• Automatically creates and manages virtual environments.
• Checks if the required packages are already installed.
• Installs packages if they are not already available.
• Supports ephemeral virtual environments that are deleted after use.
• Can be used as a decorator or context manager.

Installation

pip install pydepinject

To use the uv backend for faster environment and package management, ensure uv is installed separately. You can find installation instructions at https://github.com/astral-sh/uv.

Usage

Decorator

To use the requires as a decorator, simply decorate your function with the required packages:

from pydepinject import requires


@requires("requests", "numpy")
def my_function():
    import requests
    import numpy as np

    print(requests.__version__)
    print(np.__version__)


my_function()

Context Manager

You can also use the requires as a context manager:

from pydepinject import requires


with requires("requests", "numpy"):
    import requests
    import numpy as np

    print(requests.__version__)
    print(np.__version__)

Virtual Environment with specific name

The requires can create a virtual environment with a specific name:

@requires("requests", venv_name="myenv")
def my_function():
    import requests

    print(requests.__version__)


with requires("pylint", "requests", venv_name="myenv"):
    import pylint

    print(pylint.__version__)
    import requests  # This is also available here because it was installed in the same virtual environment

    print(requests.__version__)


# The virtual environment name can also be set as PYDEPINJECT_VENV_NAME environment variable
import os

os.environ["PYDEPINJECT_VENV_NAME"] = "myenv"


@requires("requests")
def my_function():
    import requests

    print(requests.__version__)


with requires("pylint", "requests"):
    import pylint

    print(pylint.__version__)
    import requests  # This is also available here because it was installed in the same virtual environment

    print(requests.__version__)

Reusable Virtual Environments

The requires can create named virtual environments and reuse them across multiple functions or code blocks:

@requires("requests", venv_name="myenv", ephemeral=False)
def my_function():
    import requests

    print(requests.__version__)


with requires("pylint", "requests", venv_name="myenv", ephemeral=False):
    import pylint

    print(pylint.__version__)
    import requests  # This is also available here because it was installed in the same virtual environment

    print(requests.__version__)

Managing Virtual Environments

The requires can automatically delete ephemeral virtual environments after use. This is useful when you want to ensure that the virtual environment is clean and does not persist after the function or code block completes:

@requires("requests", venv_name="myenv", ephemeral=True)
def my_function():
    import requests

    print(requests.__version__)


my_function()

Forcing Virtual Environment Recreation

If you need to ensure a completely clean environment, you can force its recreation using the recreate=True parameter. This will delete and rebuild the virtual environment even if it already exists.

from pydepinject import requires


# This will delete the "my-clean-env" venv if it exists and create it from scratch
@requires("requests", venv_name="my-clean-env", recreate=True)
def my_function():
    import requests

    print(requests.__version__)


my_function()

Logging

This library uses Python's standard logging but does not configure handlers or levels by default. Configure logging in your application to see debug output:

import logging

logging.basicConfig(level=logging.INFO)  # or DEBUG for verbose output
logging.getLogger("pydepinject").setLevel(logging.DEBUG)

You can also integrate with your application's logging setup (structlog, rich logging, etc.) by attaching handlers as you normally would.

Backend selection

By default, backends are tried in priority order "uv|venv". If uv is installed on your system, it will be preferred for faster environment creation and installs; otherwise the standard library venv backend is used.

You can control backend selection:

• Environment variable: set PYDEPINJECT_VENV_BACKEND, e.g. PYDEPINJECT_VENV_BACKEND=venv or PYDEPINJECT_VENV_BACKEND=uv|venv.
• API param: pass venv_backend="uv", "venv", or a pipe-separated list like "uv|venv" to requires/RequirementManager.

On Windows, paths (Scripts, Lib/site-packages) are handled automatically.

Advanced options

Additional installer arguments

You can forward extra arguments to the underlying installer (pip or uv pip) via install_args. This is useful for custom indexes, constraints files, proxies, etc.

from pydepinject import requires


@requires(
    "requests",
    install_args=(
        "--index-url",
        "https://pypi.myorg/simple",
        "--upgrade-strategy",
        "eager",
    ),
    venv_backend="venv",
)
def my_function():
    import requests

    print(requests.__version__)

These arguments are appended to the installer command.

Virtual environment identity

For unnamed environments, a unique directory is generated based on a stable identity key. This key is derived from:

• The set of requirements, which are normalized, canonicalized (e.g., PyYAML becomes pyyaml), and sorted alphabetically to ensure a consistent order.
• The active Python version tag (e.g., py3.11).
• The selected backend (uv or venv).
• A short hash of the current Python interpreter's path to distinguish between different Python installations.

This process guarantees that identical dependency sets produce the same virtual environment, while any change in requirements, Python version, or backend results in a new, distinct environment. Named environments (created using the venv_name parameter) are not affected by this hashing mechanism.

Venv metadata

After successful installs, a metadata file .pydepinject-{timestamp}.json is written into the venv directory. It records:

• pydepinject version, backend, Python version, interpreter path
• Target platform, requested packages, forwarded install args
• An ISO 8601 timestamp (UTC)

There might be multiple metadata files if the venv is reused across different runs with different requirements or install args. Each file is timestamped to avoid collisions.

This helps with debugging and reproducibility.

Configuration with Environment Variables

pydepinject can be configured using the following environment variables:

• PYDEPINJECT_VENV_ROOT: Specifies the root directory where virtual environments are stored. If not set, a default temporary directory is used.
• PYDEPINJECT_VENV_NAME: Sets a default name for the virtual environment, which can be useful for creating persistent, reusable environments across different runs.
• PYDEPINJECT_VENV_BACKEND: Defines the virtual environment backend to use. Supported values are uv and venv. uv is preferred for its speed.

These variables provide a convenient way to standardize behavior in CI/CD pipelines or development environments.

Where venvs are stored and cleaning them up

Each managed environment is a plain virtual environment directory under a single root. By default that root is <system-temp-dir>/pydepinject/venvs (overridable with PYDEPINJECT_VENV_ROOT), and each venv is a subdirectory named either after your venv_name or a content hash derived from the requirement set, Python version, and backend.

Because the default root lives under the OS temporary directory, the operating system usually reclaims it for you — temp-file cleaners and reboots clear stale entries — so in the common case there is nothing to clean up manually.

For finer control:

• Avoid persistence per call: pass ephemeral=True so the venv is deleted as soon as the decorated function or with block finishes.
• Prune a persistent root you manage: the venvs are just directories, so rm -rf "$PYDEPINJECT_VENV_ROOT" removes everything, or delete individual subdirectories to reclaim space selectively.
• Script age-based cleanup: each venv contains .pydepinject-*.json metadata files recording the install timestamp (and the packages installed), which you can use to find and remove environments older than a chosen age.

Limitations and thread-safety

pydepinject works by mutating process-global interpreter state for the duration of the managed scope (the decorated call or with block). When a requirement is not already importable, it:

• inserts the venv's site-packages at the front of sys.path,
• prepends to the PYTHONPATH and PATH environment variables,
• evicts conflicting entries from sys.modules so fresh imports resolve to the venv.

The original sys.path, environment variables, and sys.modules state are restored when the scope exits. If every requested package is already satisfied in the current environment, no global state is changed at all — activation is a no-op fast path.

Because this state is global to the interpreter, please keep the following in mind:

• Not thread-safe and not safe under asyncio/concurrent use. Two managed scopes active at the same time in different threads (or interleaved coroutines) will race on sys.path, the environment variables, and sys.modules, and can corrupt each other's save/restore. Use pydepinject from a single thread of control. If you need dependencies isolated across concurrent workers, give each worker its own process.
• Already-imported modules are handled best-effort. If a package was already imported earlier in the process from a different location, swapping it for the venv's copy within the scope cannot be guaranteed — code holding a reference to the previously imported module keeps using it.
• It mutates the running interpreter, by design. This makes it ideal for scripts, notebooks, plugins, and ad-hoc tooling, but it is not a substitute for a properly provisioned environment or a lockfile in a production application. For those, prefer real environment/dependency management.

When NOT to use this

pydepinject is built for scripts, notebooks, plugins, and ad-hoc tooling or CI glue — cases where "make sure X is importable, install it if not" is the whole job. It is the wrong tool, and you should reach for a real environment plus a lockfile (uv, pip-tools, Poetry, etc.) instead, when:

• You are building a deployable application or service where dependency reproducibility matters.
• Your code runs under threads, asyncio, or parallel workers — activation mutates global interpreter state and is not safe to interleave (see Limitations and thread-safety; give each worker its own process).
• You need pinned, audited, reproducible dependency sets across machines and CI runs.
• Packages are already imported earlier in the process and must be swapped mid-run — pydepinject can only do this best-effort.
• You run a long-lived production process where persistent venvs accumulating under the temp root would be a problem.

For the mechanics behind these caveats, see Limitations and thread-safety above.

Unit Tests

Unit tests are provided to verify the functionality of the requires. The tests use pytest and cover various scenarios including decorator usage, context manager usage, ephemeral environments, and more.

Running Tests

To run the unit tests, ensure you have pytest installed, and then execute the following command:

pytest

License

This project is licensed under the MIT License. See the LICENSE file for more details.