Use the default shallow checkout for jobs that only read files or test prebuilt
wheels. Source builds need complete commit ancestry and release tags so
setuptools-scm can derive each package's version.
For those jobs, use fetch-depth: 0 with filter: blob:none. The checkout
action fetches complete ancestry for all branches and tags, while the blob
filter skips historical file contents. This avoids downloading the large
historical gh-pages contents discussed in issue #2197. Git retrieves file
contents as needed when checking out the requested commit.
CI change detection uses this full-history checkout for PR mirrors, so the
actual PR base branch is available when computing its merge base. Non-PR runs
need only a shallow checkout. PR-preview cleanup checks out its script
shallowly and fetches gh-pages separately when needed. Pathfinder release
preparation fetches its exact tag without requesting every other tag.
Git documents blob filtering in its
clone reference.
The pinned checkout action's
README
documents fetch-depth: 0 for fetching all history, branches, and tags.
For development targeting main, keep links to the latest repository sources
on main, even when their targets are added or moved by the same pull request.
The link checker treats canonical
https://github.com/NVIDIA/cuda-python/blob/main/ and tree/main/ URLs as
references to the proposed main tree. Lychee remaps them for validation only;
source files and published documentation keep their URLs. The remaps match only
at the start of a URL, leaving URLs embedded in another URL unchanged. Relative
links and links to tags, commit SHAs, other branches, and other repositories
are checked as written.
Authored Markdown and reST are checked through the same lychee pre-commit
hook in development and in Linux/Windows PR CI. The Python wrapper
ci/tools/run_lychee.py computes the checkout's file URI and passes anchored
native --remap rules to Lychee. This checks working-tree paths, including
uncommitted and unpushed targets, without changing the documentation sources.
The hook installs the official PyPI lychee-bin==0.24.2 native wheel in its
isolated Python environment. Authored checks do not validate fragments.
PR CI restores successful URL checks from the shared baseline, then runs
pre-commit run lychee --all-files through ci/tools/run_lychee_precommit.py.
The helper allows up to three hook attempts, with 60/120-second cooldowns
when all remaining failures are transient HTTP failures. Successful checks
accumulate in the local cache; permanent link failures stop additional
warm-up attempts. A missing baseline simply starts with uncached checks.
After warming, pre-commit run --all-files --show-diff-on-failure runs once.
Its normal strict result is the final gate, including for unresolved HTTP
429s. This can make four total Lychee invocations, but successful external
checks are reused. Formatter changes remain failures because the complete
suite is not retried. A missing or invalid native report, empty check, or
hook installation/execution error fails the warm-up step.
This replaces the fixture-only hook step and the duplicate authored scan in the full-CI documentation job. Standalone CI-tool unit tests remain in nightly CI; they are not added as a new per-PR gate.
After building the documentation, full CI runs the manual hook through the same driver:
python ci/tools/run_lychee_precommit.py rendered --policy check \
--report /path/to/reports/lychee-rendered.jsonThe driver invokes pre-commit run lychee-rendered --hook-stage manual --all-files. This hook enumerates generated HTML under artifacts/docs,
excluding _static assets, and enables full fragment validation. Generated
HTML is selected explicitly because it is not part of pre-commit's tracked
source file list. The hook uses the same pinned wheel and shared wrapper as
the authored hook, without a separate Lychee action or binary installation.
Rendered checks remap main-branch source URLs to the CI repository at the actual checked-out commit, retaining GitHub's file/directory views and their fragment behavior. All attempts use the same inputs and remaps. This mode retains up to three attempts and the existing warning allowance for at most ten distinct URLs that still return HTTP 429; other unresolved links fail.
Nightly invokes both configured hooks through the driver. Refresh mode skips
restoring yesterday's URL cache, checks afresh, and publishes the successful
results. Genuine link failures are advisory during cache warming; build,
installation, execution, and report errors remain failures. Linux pre-commit
runs on main also publish their accumulated authored results. PR jobs
restore caches and do not publish them.
ci/tools/lychee_cache.py gives all callers the same cache policy and unique
run-ID/attempt snapshot keys. The namespace includes the hook configuration
(and native wheel pin), checker configuration, and shared helpers. Authored
and rendered caches remain separate because only rendered checks validate
fragments. Only the URL-result file is shared across operating systems;
pre-commit's installed environments retain their OS/Python-specific caches.
Successful external checks retain their original timestamps and expire after
one day. Failed URLs are not persisted, and filesystem targets are checked
afresh. The rendered hook uses its own cache directory, selected by
LYCHEE_CACHE_DIR (default .lychee-rendered), so its fragment-aware results
cannot mix with authored results in the checkout's .lycheecache. Per-pass
JSON reports and a Markdown summary are retained as CI artifacts.
Merging these changes into main does not update maintenance branches'
workflows. A maintenance branch's /main/ URLs may intentionally refer to a
different tree from its checkout. Before adopting these hooks there, choose
an explicit validation policy for those links instead of assuming that they
refer to the maintenance checkout.
Pixi lockfile freshness enforcement and automated refreshes are temporarily suspended because of Pixi #7215 and #7000. Repeated lockfile operations can change the committed solution without converging, so a general refresh is not a reliable repair.
The CI: pixi lockfile freshness check workflow still runs on every PR with its
existing pixi lock --check (all workspaces) check name. It succeeds with a
warning annotation and job summary explicitly reporting the suspension;
success does not mean freshness passed. The refresh workflow has no schedule.
Manual dispatch emits the same notice without solving dependencies, rewriting
lockfiles, pushing a branch, or creating a PR.
Manifest/lock consistency is not automatically enforced during this pause.
Necessary dependency changes require manual review and validation of the
affected manifests and lockfiles. Real source builds and tests continue using
PIXI_FROZEN=true, including nested pixi run calls, to build the checkout's
source packages against the committed dependency solution. The pinned Pixi
version and committed lockfiles are unchanged.
Re-enable maintenance only after a released Pixi passes both upstream issue
reproducers and repeated byte-stable update/lock/check cycles across all six
workspaces: the repository root, cuda_pathfinder, cuda_bindings, cuda_core,
benchmarks/cuda_bindings, and benchmarks/cuda_core. Verify both pixi update --no-install and pixi lock, followed by repeated pixi lock --check commands,
converge to unchanged lockfile bytes. Then restore strict freshness checks,
scheduled refreshes, and PIXI_LOCKED=true in source-build CI.
The workflows in this repository use the CI_CUSTOMIZATIONS_* namespace for
GitHub Actions configuration variables that opt an alternative synchronized
repository into repository-specific CI behavior. This keeps the workflow logic
shared without hard-coding the names of private repositories into the public
source tree.
These variables are non-secret strings configured under
Settings > Secrets and variables > Actions > Variables.
An unset variable, or any value other than the literal string true, leaves
the customization disabled. Do not store credentials or other secret values in
these variables.
| Variable | Default | Purpose |
|---|---|---|
CI_CUSTOMIZATIONS_SECURITY_SUITE_ENABLED |
Disabled | Enables the NVIDIA Security Suite after its runner, Actions variables, and OIDC/Vault authorization have been provisioned for the repository. |
The canonical NVIDIA/cuda-python repository does not need this variable
because its standard workflow behavior is enabled directly. Before enabling a
customization elsewhere, document the repository-specific prerequisites and
verification procedure in that repository's own documentation.