Package Managers

Status

This note is complete, reviewed, and considered stable.

In Python, “package management” usually spans several jobs:

• Installer/Resolver: downloads and resolves dependencies (e.g., pip, uv, Poetry’s installer).
• Environment manager: isolates dependencies per project (e.g., venv, virtualenv, Conda).
• Project manager: defines metadata, lock files, build/publish (e.g., Poetry, uv, Hatch, PDM, Flit).
• Tool runner: installs/runs CLI tools in isolated envs (e.g., pipx, uvx).
• Python version manager: installs/switches Python versions (e.g., pyenv, uv python).

The tools (concise profiles)

uv (Astral)

A fast, all-in-one package & project manager written in Rust. It can replace pip, pip-tools, pipx, poetry, virtualenv (and even pyenv/twine tasks) with one tool. Provides a pip-compatible interface (uv pip …), a universal lockfile, workspaces, a global cache, script support (inline deps), tool runner (uvx), and Python version management. Claimed 10–100× faster installs than pip.

Common commands

# project
uv init && uv add requests && uv lock && uv sync
uv run python app.py

# pip-compatible workflows (fast drop-in)
uv pip install -r requirements.txt
uv pip compile --universal -o requirements.txt
uv pip sync requirements.txt

# tools & Python versions
uvx ruff --version
uv python install 3.12 && uv venv --python 3.12

How to use

1. Initialize a project:

   uv init

2. Add dependencies:

   uv add <package_name>

3. Lock dependencies:

   uv lock

4. Install dependencies from lockfile:

   uv sync

5. Run scripts with dependencies:

   uv run python main.py

6. Use as a drop-in pip replacement:

   uv pip install requests

7. Run CLI tools without installing globally:

   uvx black .

8. Manage Python versions:

   uv python install 3.11
   uv venv --python 3.11

Pros

• Very fast installs/resolution; excellent cache behavior.
• One tool covers install/lock/envs/tools/publish/Python versions.
• Works with existing requirements.txt via uv pip ….

Cons

• Newer ecosystem; some teams may still standardize on pip/Poetry/Conda.
• Feature superset can feel “heavy” if we only need a minimal pip+venv flow.

pip (and venv/virtualenv)

The default installer that talks to PyPI; pair with venv for env isolation. virtualenv is an alternative with more knobs/features; venv is stdlib and lightweight.

How to Use

1. Create a virtual environment:

   python -m venv .venv

2. Activate it:

   • macOS/Linux:

     source .venv/bin/activate

   • Windows:

     .venv\Scripts\activate

3. Install packages:

   pip install requests

4. Freeze dependencies:

   pip freeze > requirements.txt

5. Install from requirements:

   pip install -r requirements.txt

Pros

• Ubiquitous, simple; zero extra tooling.
• Works everywhere; easy to script/CI.

Cons

• No lockfile by default (we add pip-tools or switch to uv/Poetry).
• Env management is separate (venv/virtualenv).

pip-tools (pip-compile, pip-sync)

Adds deterministic lock files to pip workflows by compiling pinned requirements.txt from inputs.

How to Use

1. Install pip-tools:

   pip install pip-tools

2. Create a requirements.in with your dependencies:

   requests
   flask

3. Compile to a locked requirements file:

   pip-compile

4. Sync environment to match the lock file:

   pip-sync

Pros

• Deterministic builds while keeping plain-pip workflows.
• Familiar files (requirements.in → pinned requirements.txt).

Cons

• Extra step in the workflow; no env or project metadata features.

Poetry

Full project manager: dependency resolution, lockfile, build & publish using pyproject.toml + poetry.lock.

How to Use

1. Install Poetry:

   pip install poetry

2. Initialize a project:

   poetry init

3. Add dependencies:

   poetry add requests

4. Install dependencies:

   poetry install

5. Run inside environment:

   poetry run python app.py

6. Publish to PyPI:

   poetry build && poetry publish

Pros

• Clean pyproject.toml config, robust lockfile & resolver.
• Built-in build/publish pipeline.

Cons

• Extra abstraction vs. pip; learning curve for teams used to requirements.txt.
• External tool; slower than uv in practice.

PDM

Modern project manager following the latest packaging standards; uses pyproject.toml; supports lockfiles and publishing.

How to Use

1. Install PDM:

   pip install pdm

2. Create a new project:

   pdm init

3. Add dependencies:

   pdm add requests

4. Install dependencies:

   pdm install

5. Run scripts:

   pdm run python app.py

6. Publish package:

   pdm publish

Pros

• Standards-forward; ergonomic project setup via pdm new.

Cons

• Smaller mindshare than Poetry/uv; fewer “batteries included” than uv.

Hatch

Project manager focused on reproducible builds, pluggable envs/scripts, and publishing; integrates well with modern packaging.

How to use

1. Install Hatch:

   pip install hatch

2. Create a new project:

   hatch new my_project

3. Add dependencies (edit pyproject.toml or use CLI):

   hatch env create

4. Run scripts:

   hatch run python app.py

5. Build & publish:

   hatch build && hatch publish

Pros

• Reproducible builds by default; flexible environments and script runner.

Cons

• Opinionated; smaller ecosystem than Poetry.

pipx (CLI tools)

Installs and runs Python applications in isolated envs (think: “brew for Python CLIs”)—great for tools like black, ruff, httpie.

How to use

1. Install pipx:

   pip install pipx

2. Install CLI tool:

   pipx install black

3. Run CLI tool:

   pipx run black .

Pros

• Keeps CLI tools isolated from project deps.

Cons

• Not a project dependency manager; separate from your app’s env.

Conda / Mamba

Cross-language package & environment manager—excellent when we need system libraries (BLAS, CUDA, GDAL). Mamba is a faster, drop-in alternative solver/CLI.

How to use (Conda)

1. Create environment:

   conda create -n myenv python=3.11

2. Activate environment:

   conda activate myenv

3. Install packages:

   conda install numpy pandas

4. Export environment:

   conda env export > environment.yml

How to use (Mamba)

Same commands as Conda, replacing conda with mamba:

mamba create -n myenv python=3.11
mamba install numpy pandas

Pros

• Handles non-Python/native deps; binary packages + env management.
• Ecosystem (conda-forge) and GUIs (Anaconda Navigator).

Cons

• Larger footprint; mixing pip/Conda needs care to avoid conflicts.

Feature matrix (quick scan)

Tool	Lockfile	Env mgmt	Build/Publish	Tool runner	Python version mgmt	Handles non-Py deps
uv	Yes (universal)	Yes	Yes	uvx	Yes	No
pip	No (add pip-tools)	No (pair w/ venv)	No	No	No	No
pip-tools	Yes	No	No	No	No	No
Poetry	Yes	Yes	Yes	Basic scripts	No	No
PDM	Yes	Yes	Yes	Scripts	No	No
Hatch	Yes	Yes	Yes	Yes	No	No
pipx	n/a	Isolated per-tool	n/a	Yes	No	No
Conda/Mamba	YAML env files	Yes	No	No	No	Yes

When to choose what

• General Python apps/services (most teams): uv for speed + single-tool simplicity; it reads requirements.txt or does full project/lock flows.
• Legacy/simple scripts: pip + venv (optionally add pip-tools for locks) if we want minimal change.
• Publishable libraries: uv or Poetry/Hatch/PDM for pyproject.toml + lockfile + build/publish.
• Data science / needs native libs: Conda/Mamba for environments and binaries; we can still use pip/uv inside the env for PyPI-only deps.
• CLI tooling on your dev machine: pipx or uvx.