A history of Python dependency tools

Python's packaging and dependency management has seen many tools rise and fall over time. This article walks through them in chronological order and notes what each tool solved, and where it fell short — based on facts.

1. Timeline

Year	Event
1991	Python 0.9.0 (Guido van Rossum, February 1991).
2000	distutils included in Python 1.6's standard library. The first build/distribution standard.
2003	PyPI (Python Package Index) starts operating. Previously called "Cheese Shop".
2004	setuptools arrives. Patches distutils' limits (dependency declarations, extensibility). Bundles easy_install.
2007	virtualenv (Ian Bicking) first released. Isolated Python environments.
2008	pip (Ian Bicking and others). Aimed at fixing easy_install's pain (uninstall difficulty, etc.).
2011	The Twelve-Factor App published. Solidifies the principle of separating config and dependencies.
2012	pyenv (yyuu). Python interpreter version manager.
2012	pip becomes standard from Python 3.4 (formal in 2014). PEP 453.
2012	The venv module arrives in Python 3.3. Standard-library version of part of virtualenv.
2012	pip-tools (Vincent Driessen and others). pip-compile produces a lockfile (pinned requirements.txt).
2016	PEP 518 — introduces pyproject.toml. Standard for declaring build systems.
2017	pipenv (Kenneth Reitz). Tries an npm-like workflow with Pipfile / Pipfile.lock.
2018	Poetry (Sébastien Eustace) 1.0-pre. Unified pyproject.toml-based dependencies, builds, and publishing.
2019	PEP 517 — separates the build backend. Standard that opens up build tools beyond setuptools.
2020	PEP 621 — standard [project] section. Declares metadata in pyproject.toml.
2021	PDM (Frost Ming). Includes PEP 582 experiments. Standards-faithful.
2022	Hatch (a manager for hatchling, the build backend PyPA recommends) stabilizes after a rewrite.
2024	uv (Astral). Rust. Folds many pip / pip-tools / virtualenv / poetry workflows into one tool.

2. virtualenv (2007)

When project A and B on the same machine need different library versions, sharing a system site-packages causes conflicts. virtualenv builds an isolated environment with its own lib/python*/site-packages per project. Python 3.3+'s python -m venv is the standard-library version of the same idea.

3. pip (2008)

easy_install was awkward to uninstall and weak at dependency resolution. pip standardized requirements.txt and imperative install / uninstall.

4. pip-tools (2012)

requirements.txt alone leaves transitive dependencies unpinned, so yesterday's build and today's build can diverge. pip-compile generates a lockfile with all transitive versions pinned, starting from abstract dependencies (requirements.in).

5. pyenv (2012)

Separates system Python from project Python. .python-version switches versions per directory automatically. The standard remedy for limits in macOS' default Python (used to be 2.7).

6. pipenv (2017)

Brought an "npm-like" Pipfile / Pipfile.lock workflow. Active for a while, but resolution speed and governance issues split adoption.

7. Poetry (2018)

Unified dependencies, metadata, and the build backend (poetry-core) inside a single pyproject.toml. Stable poetry.lock and command consistency made it the most widely adopted unified tool. Using a custom [tool.poetry] section before PEP 621 standardization later imposed migration costs for standards compliance.

8. PDM · Hatch (2021–2022)

PDM — Faithful to PEP standards. Once championed PEP 582 (__pypackages__) isolation, but settled on regular venv mode after that PEP was rejected.

Hatch — A manager paired with hatchling, the build backend PyPA recommends. Strong at managing environment matrices (per–Python-version test environments).

9. uv (2024)

Written in Rust. Provides pip / pip-tools / virtualenv compatibility modes for drop-in replacement, while also offering Poetry-style project workflows via uv add / uv sync / uv run. Pulls in the Python interpreter itself with uv python install. Emphasizes speed and integration into one tool (03-python-uv).

10. conda

Outside the line above, Anaconda / Conda (Continuum Analytics → Anaconda Inc., 2012) is a multi-language package manager that bundles C / C++ libraries and compilers alongside Python. Standard fare in scientific computing and machine learning. When NumPy / SciPy and other native dependencies are hard to build with pip, conda is the workaround.

Mambaforge / Miniforge are conda-compatible faster implementations that keep the same channel policy while improving resolution speed.

11. The same task across tools

# venv + pip
python -m venv .venv
# Windows
.venv\Scripts\activate
# macOS · Linux
source .venv/bin/activate
pip install -r requirements.txt

# pip-tools
pip-compile requirements.in
pip-sync

# Poetry
poetry add fastapi
poetry install
poetry run python -m pytest

# uv
uv add fastapi
uv sync
uv run pytest

12. Common pitfalls

Mixed traces — different tools' artifacts coexisting in the same project (requirements.txt + pyproject.toml + poetry.lock + uv.lock). Spell out which file is the SSOT in the README.

[project] vs [tool.poetry] coexisting — Poetry's own pre-1.x section and PEP 621's standard section have different intent and interpretation. Poetry 2.0+ tightens PEP 621 compatibility, but the migration is best done deliberately in one pass.

Global install on system Python — touching macOS' Apple Python or Linux' OS Python can break system tools. Use pipx or a user venv.

conda + venv activated together — PATH gets tangled and which Python wins is unclear. Activate only one.

Lockfile merge conflicts — safer to pick a side, then re-run the tool's resolution command (poetry lock, uv lock) than to merge text manually.

Closing thoughts

Python's dependency tools have flowed for 17 years from distutils → pip → virtualenv → pip-tools → pipenv → Poetry → uv. Each tool stood on the previous one's limits. Today, uv has the highest level of integration; if Poetry or pip-tools is already humming, the case to switch is weak. Just don't let two tools' traces co-exist inside one project.

Next

• regex
• git-submodule-lfs

References include the Python Packaging User Guide (PyPA), PEP 518, PEP 621, pip docs, Poetry docs, uv docs, Anaconda, pipx, and Python Packaging History (PyPA discuss).