Adds `require-checksums`, which enforces the "intrinsic or explicitly
provided checksum" idea. See #1032.
<details>
<summary>Automated notes</summary>
Add the opt-in input `require-checksums`. When enabled, setup-uv
verifies the archive against the workflow's `checksum` input or the
digest bundled for the resolved uv version, architecture, and platform.
If neither is available, setup fails before downloading the archive. By
default, setup-uv continues to accept manifest checksums as it does
today.
The setting also applies to custom manifests, which can still supply
download URLs. On persistent runners, strict mode keys tool-cache
entries by the expected archive digest so a previous manifest-verified
installation cannot bypass the check.
Closes#1032. Builds on #1033.
## Validation
- `npm ci --ignore-scripts` and `npm run all` passed under Node 24 (182
tests).
- The packaged-action regression test installs a custom build into a
temporary runner tool cache. It checks that strict mode rejects a
manifest-only version even with a warm cache, reuses a verified
installation for the same digest, and rejects a changed digest or a
custom build that conflicts with the bundled digest.
</details>
---------
Signed-off-by: William Woodruff <william@yossarian.net>
Windows ARM64 jobs may need x64 Python independently of their Python
version pin. Add `python-arch` to export `UV_PYTHON_ARCH` for subsequent
steps and include the selected architecture in the uv and managed-Python
cache keys. The input takes precedence over an inherited environment
value, and unsupported uv versions fail before environment creation.
Requires a uv release with `UV_PYTHON_ARCH` support.
## Summary
This changes the default for `prune-cache` from `true` to `false`,
motivated by [#745](https://github.com/astral-sh/setup-uv/issues/745).
Users that want the existing behavior can continue to set `prune-cache:
true` explicitly.
Some history: I originally added [`uv cache prune
--ci`](https://github.com/astral-sh/uv/pull/5391) after looking at a
workload where the uv cache was ~2.2 GB, almost entirely due to the
enormous pre-built `torch` and `nvidia_cudnn_cu12` wheels ([original
analysis](https://github.com/actions/setup-python/issues/822#issuecomment-2248728264)).
Persisting and restoring thousands of extracted files through the GitHub
Actions cache could be slower than downloading the wheels again. In
contrast, wheels built from source can be very expensive to recreate.
The intent was to remove pre-built wheels while retaining locally-built
wheels.
`setup-uv` subsequently made pruning configurable, but defaulted
`prune-cache` to `true`; it also later enabled caching by default on
GitHub-hosted runners. As a result, the default configuration repeatedly
downloads pre-built wheels from PyPI even on a cache hit. That tradeoff
has become more important as uv adoption has grown: [the PyPI analysis
in
#745](https://github.com/astral-sh/setup-uv/issues/745#issuecomment-3867334064)
estimates that uv accounts for roughly half of reported CI downloads
from PyPI, and roughly 65-75% for `boto3`.
I ran the comparison across a few different workloads:
| Workload | PR | Packages | Cache: keep / prune / prune-ci | Warm
restore+sync: keep / prune / prune-ci | Downloads: prune / prune-ci |
|---|---:|---:|---:|---:|---:|
| Tiny | [#1](https://github.com/astral-sh/setup-uv-benchmarks/pull/1) |
19 | 6 / 6 / 2 MB | 0.3-0.4 / 0.3 / 0.4-0.5 s | 0 / 2 |
| Web | [#2](https://github.com/astral-sh/setup-uv-benchmarks/pull/2) |
65 | 43 / 43 / 7 MB | 0.6-1.0 / 0.5-0.6 / 1.5-1.7 s | 0 / 6 |
| Scientific |
[#3](https://github.com/astral-sh/setup-uv-benchmarks/pull/3) | 118 |
586 / 586 / 8 MB | 8.2-16.0 / 7.0-8.2 / 8.9-12.1 s | 0 / 19 |
| PySpark |
[#4](https://github.com/astral-sh/setup-uv-benchmarks/pull/4) | 19 |
1820 / 1820 / 436 MB | 9.9-21.1 / 10.5-11.0 / 5.0-7.0 s | 0 / 4 |
| CPU PyTorch |
[#5](https://github.com/astral-sh/setup-uv-benchmarks/pull/5) | 14 | 182
/ 182 / 1 MB | 3.0-6.0 / 3.6-4.0 / 5.7-6.4 s | 0 / 6 |
| CPU-PyTorch ML |
[#6](https://github.com/astral-sh/setup-uv-benchmarks/pull/6) | 137 |
346 / 346 / 10 MB | 7.4-18.0 / 8.8-8.9 / 9.7-11.9 s | 0 / 20 |
| CUDA PyTorch |
[#7](https://github.com/astral-sh/setup-uv-benchmarks/pull/7) | 201 |
2316 / 2315 / 16 MB | 30.2-67.9 / 31.0-63.6 / 33.3-36.7 s | 0 / 40 |
The CUDA workload intentionally reproduces the original `torch==2.1.1`
example. Keeping wheels again produces a ~2.3 GB Actions cache. Across
nine warm runs, restoring that cache ranged from slightly faster than
re-downloading to roughly twice as slow; pruning consistently
re-downloaded 40 distributions in ~33-37 seconds ([original
runs](https://github.com/astral-sh/setup-uv-benchmarks/actions/runs/29750292738),
[additional
runs](https://github.com/astral-sh/setup-uv-benchmarks/actions/runs/29761705492)).
I also tried running `uv cache prune --force` without `--ci` across
every workload, to see if it provided a useful middle ground. It did not
meaningfully reduce any of the caches: plain prune took 11-21 ms and
left the extracted cache and file count unchanged, including PySpark. On
these fresh caches, there are no dangling entries to remove; without
`--ci`, the pre-built wheels and unpacked source/build artifacts are
retained. The per-workload runs are linked in the table above.
So the original motivation still holds for very large CUDA or
source-heavy workloads, but it is not representative of the common case.
For smaller workloads, keeping pre-built wheels is generally faster and
avoids repeated PyPI traffic. This changes the default accordingly,
while retaining `prune-cache: true` as an opt-in for workloads where the
smaller cache is worthwhile.
Closes https://github.com/astral-sh/setup-uv/issues/745.
## Summary
- URL-encode the Python version component before adding it to the cache
key
- URL-encode the user-provided cache suffix for the same reason
- Add cache key tests for Python ranges, comma-containing suffixes, and
unchanged simple inputs
Fixes#914
Refs: pi-session 019f3164-85e7-7817-bffd-501d89b3a1fd
## Tests
- npm run all