astral-sh
uv
Blog
Docs
Changelog
Blog
Docs
Changelog
Overview
Branches
Benchmarks
Runs
Performance History
Latest Results
Cache Hawk dependencies (#20291) ## Summary Give the Hawk job its own Rust cache and target directory so dependency artifacts can be reused across CI runs. Keep workspace crate outputs out of the cache so Hawk still recompiles uv and emits fresh analysis fragments. Cache writes continue to follow the workflow’s existing `save-rust-cache` input. After rebasing onto Hawk 0.1.7 and package-scoped doctest analysis, [warm-cache CI on this branch](https://github.com/astral-sh/uv/actions/runs/29043777639/job/86207386980) reduced the Hawk step from 4:18 on the [exact base commit](https://github.com/astral-sh/uv/actions/runs/29043681659/job/86207094292) to 3:05, saving 1:13 (28%). Complete job time fell from 4:31 to 3:34, saving 57 seconds (21%). The 1.03 GB cache restored in 23 seconds and reduced the first production Cargo pass from 2:06 to 0:56; the other two Cargo passes and the scoped rustdoc and final-analysis phase remained within two seconds of the exact base, confirming that workspace analysis still runs fresh.
main
12 hours ago
Add warm cache uv startup benchmark with a large workspace (#20006) This is a synthetic version of https://github.com/astral-sh/uv/pull/18409, which measures uv startup overhead in a large workspace. The workspace is modeled based on airflow, but less pypi dependencies to avoid timeouts.
main
12 hours ago
Upgrade Hawk to 0.1.7 (#20292) ## Summary Upgrade Hawk from 0.1.6 to 0.1.7, which adds package-scoped doctest analysis. Configure Hawk to compile doctests only for `uv-pep440`, `uv-pep508`, and `uv-redacted`. These are the three workspace packages that currently define executable doctests—five in total—so we retain the existing doctest consumer coverage without compiling rustdoc across the full workspace. A configuration comment records that the package list must stay synchronized as new doctests are added. In [CI on this branch](https://github.com/astral-sh/uv/actions/runs/29042482462/job/86203126073) compared with [CI for the exact base commit](https://github.com/astral-sh/uv/actions/runs/29035164064/job/86178621474), the rustdoc and final-analysis phase fell from 3:34 to 0:48, saving 2:46 (78%). The full Hawk step fell from 6:33 to 4:34, saving 1:59 (30%), and the complete job fell from 6:47 to 4:50, saving 1:57 (29%). The exact-base run was unusually fast; against the 8:25 median of the three immediately preceding main runs with the same Rust and Hawk setup, the complete job saves 3:35 (43%).
main
13 hours ago
Fix uv-macros on trampoline nightly
konsti/workspace-uv-run-benchmark
13 hours ago
Re-run Namespace macOS CI (10)
zaniebot:zb/use-namespace-macos-runners
13 hours ago
Re-run Namespace macOS CI (9)
zaniebot:zb/use-namespace-macos-runners
14 hours ago
Re-run Namespace macOS CI (8)
zaniebot:zb/use-namespace-macos-runners
14 hours ago
Re-run Namespace macOS CI (5, retry)
zaniebot:zb/use-namespace-macos-runners
15 hours ago
Latest Branches
CodSpeed Performance Gauge
+1%
Add warm cache uv startup benchmark with a large workspace
#20006
14 hours ago
1b1db9a
konsti/workspace-uv-run-benchmark
CodSpeed Performance Gauge
0%
Use Namespace runners for macOS
#20276
13 hours ago
b7fa991
zaniebot:zb/use-namespace-macos-runners
CodSpeed Performance Gauge
+1%
Share dependency traversal state
#20289
17 hours ago
ace0f39
zb/dependency-traversal-poc
© 2026 CodSpeed Technology
Home
Terms
Privacy
Docs