TL;DR
uv achieves major install-speed gains not simply because it’s written in Rust, but because of standards work and deliberate design trade-offs that eliminate slow code paths. Key improvements come from modern packaging PEPs, dropping legacy behaviors, and optimizations that could be implemented in other tools.
What happened
uv, a Python package installer that launched in February 2024, installs packages far faster than pip by combining modern packaging standards, narrower compatibility choices, and a set of protocol and filesystem optimizations. The tooling ecosystem matured through PEP 518, 517, 621 and PEP 658, which moved dependency and build metadata into static, machine-readable places; PEP 658’s metadata in the Simple Repository API went live on PyPI in May 2023 and enabled metadata-first workflows. uv avoids many legacy code paths — it drops .egg support, ignores pip.conf, skips bytecode compilation by default, requires virtual environments, and enforces stricter package validation. It also uses HTTP range requests for wheel metadata, parallel downloads, a global cache with hardlinks, Python-free metadata resolution, and the PubGrub resolver. Some gains are Rust-specific (zero-copy deserialization, lock-free structures, single-binary startup and compact version packing), but most speedups come from design choices and standards adoption.
Why it matters
- Static metadata and no code-execution dependency discovery enable much faster and safer installs.
- Modern PEPs (518, 517, 621, 658) changed packaging infrastructure enough that new tools can take much faster paths.
- Dropping decades of backward-compatibility edge cases allows a fresh tool to avoid many slow checks and fallbacks.
- Several high-impact optimizations (parallel downloads, metadata-only resolution, global caching) could be adopted by pip but are constrained by compatibility concerns.
Key facts
- pip historically needed to execute setup.py to discover dependencies, creating a chicken-and-egg problem for build dependencies.
- PEP 518 (2016) introduced pyproject.toml to declare build dependencies without running code.
- PEP 517 (2017) separated build frontends and backends; PEP 621 (2020) standardized [project] metadata in TOML.
- PEP 658 (2022) added package metadata to the Simple Repository API; it went live on PyPI in May 2023.
- uv launched in February 2024 and took advantage of these standards to implement metadata-first flows.
- uv omits several legacy behaviors: .egg support, pip.conf parsing, default bytecode compilation, and system-wide installs by default.
- uv ignores requires-python upper bounds (like python<4.0) for upper limits, only enforcing lower bounds to reduce resolver backtracking.
- Performance techniques not tied to Rust include HTTP range requests for wheel central directories, parallel downloads, a global cache with hardlinks, and using PubGrub for resolution.
- Rust-specific advantages cited include zero-copy deserialization with rkyv, lock-free concurrency, avoiding Python interpreter startup in subprocesses, and compact u64 version packing.
What to watch next
- Whether pip will adopt parallel downloads, global caching, or metadata-only resolution features — not confirmed in the source.
- Broader adoption of PEP 658 or similar metadata APIs by alternative indexes beyond PyPI — not confirmed in the source.
- Future uv changes or trade-offs to regain compatibility while keeping speed — not confirmed in the source.
Quick glossary
- pyproject.toml: A TOML file used by Python projects to declare build-system requirements and project metadata without executing code.
- wheel: A binary distribution format for Python packages; wheels are ZIP archives that contain metadata and compiled artifacts.
- PEP: Python Enhancement Proposal — a design document providing information or describing a new feature for Python or its ecosystem.
- PubGrub: A dependency resolution algorithm designed to find conflicts efficiently and explain failures; originally developed for Dart’s package manager.
- HTTP range request: An HTTP feature that requests a specific byte range from a resource, allowing clients to fetch parts of files (useful for reading ZIP central directories without full download).
Reader FAQ
Is uv fast only because it’s written in Rust?
No. Rust provides micro-optimizations, but the largest gains come from standards and design choices that avoid slow code paths.
Did standards work enable uv’s approach?
Yes. PEP 518, 517, 621 and PEP 658 moved metadata into static formats and APIs, which made metadata-first installs practical.
Does uv support .egg files and pip configuration files?
uv does not support .egg files and it ignores pip.conf according to its compatibility documentation.
Could pip implement the same optimizations?
The source says pip could adopt parallel downloads, global caching, and metadata-only resolution, but backward compatibility with long-standing edge cases is a major constraint.
How uv got so fast Dec 26, 2025 uv installs packages faster than pip by an order of magnitude. The usual explanation is “it’s written in Rust.” That’s true, but…
Sources
- How uv got so fast
- uv: An extremely Fast Python Package Manager
- Here's Why UV's Speed and Simplicity Are Winning Over …
- UV: The Engineering Secrets Behind Python's Speed King
Related posts
- NullSplats: Three Ways to Produce 3D Gaussian Splatting from Casual Captures
- How Lewis Carroll computed determinants: Dodgson condensation explained
- How to disable Chrome’s built-in AI features and remove Gemini