Skip to content

Commit e9a2441

Browse files
Andy-Jostclaude
andcommitted
Merge main into fix/2694-vmm-host-numa to pick up #3188 (cuDSS 1.0 library names)
Co-Authored-By: Claude Fable 5.1 <[email protected]>
2 parents 1e11c4a + d5f37e0 commit e9a2441

565 files changed

Lines changed: 253133 additions & 2711 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

‎.coderabbit.yaml‎

Lines changed: 5 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -65,7 +65,7 @@ reviews:
6565
custom_checks: []
6666

6767
path_instructions:
68-
- path: "cuda_bindings/**/*"
68+
- path: "cuda_bindings_13/**/*"
6969
instructions: |
7070
Focus on Cython binding correctness, API and ABI stability, error-code propagation,
7171
resource ownership and reference counting, GIL handling, cross-platform behavior.
@@ -122,23 +122,23 @@ knowledge_base:
122122
filePatterns:
123123
- "AGENTS.md"
124124
- "CONTRIBUTING.md"
125-
- "cuda_bindings/AGENTS.md"
125+
- "cuda_bindings_13/AGENTS.md"
126126
- "cuda_core/AGENTS.md"
127127
- "cuda_core/tests/AGENTS.md"
128128
- "cuda_pathfinder/AGENTS.md"
129129
- "cuda_python/AGENTS.md"
130130
- "benchmarks/cuda_bindings/AGENTS.md"
131131
- "benchmarks/cuda_core/AGENTS.md"
132-
- "cuda_bindings/README.md"
132+
- "cuda_bindings_13/README.md"
133133
- "cuda_core/README.md"
134134
- "cuda_pathfinder/DESCRIPTION.rst"
135135
- "cuda_python/DESCRIPTION.rst"
136-
- "cuda_bindings/docs/source/contribute.rst"
136+
- "cuda_bindings_13/docs/source/contribute.rst"
137137
- "cuda_core/docs/source/contribute.rst"
138138
- "cuda_core/docs/source/concurrency.rst"
139139
- "cuda_core/docs/source/error_handling.rst"
140140
- "cuda_pathfinder/docs/source/contribute.rst"
141-
- "cuda_bindings/docs/source/release.rst"
141+
- "cuda_bindings_13/docs/source/release.rst"
142142
- "cuda_core/docs/source/release.rst"
143143
- "cuda_pathfinder/docs/source/release.rst"
144144
- "cuda_python/docs/source/release.rst"

‎.github/RELEASE-core.md‎

Lines changed: 24 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -6,9 +6,32 @@
66
This document covers the `cuda.core` release process. For other packages:
77
`cuda-bindings` and `cuda-python` involve a private repository and are not
88
documented here; `cuda-pathfinder` is largely automated by the
9-
[release-cuda-pathfinder.yml](workflows/release-cuda-pathfinder.yml)
9+
[CI: Release](workflows/release.yml)
1010
workflow.
1111

12+
To release `cuda-pathfinder`, dispatch **CI: Release** with
13+
`component` set to `cuda-pathfinder`, `release-action` set
14+
to `full-release`, and `git-tag` set to the existing
15+
`cuda-pathfinder-v<VERSION>` tag. The workflow finds the successful
16+
tag-triggered CI run, checks release notes, creates or reuses a draft,
17+
builds and deploys versioned docs, and uploads the source archive, checksum,
18+
and wheels. It publishes to TestPyPI, verifies installation and the imported
19+
version, then publishes to PyPI and verifies again before publishing the
20+
GitHub Release without marking it as latest. A failed
21+
docs build prevents asset uploads, package publication, and finalization.
22+
23+
The release-note checker runs from `git-tag`, not the workflow revision.
24+
For Pathfinder, that tag must contain a version
25+
of `ci/tools/check_release_notes.py` that writes `version=<VERSION>` to
26+
`GITHUB_OUTPUT`. The release-notes job rejects an empty version before
27+
documentation, asset uploads, or package publication can start. Pathfinder
28+
package publication also requires successful draft metadata preparation.
29+
30+
The default `dry-run` action validates preparation, builds docs, and
31+
collects release artifacts without creating a GitHub Release or publishing
32+
packages. It can be dispatched from a development branch; use
33+
`dry-run-docs-branch` only when a preview deployment is needed.
34+
1235
Each section below provides detailed guidance for a step in the
1336
[Release Checklist](ISSUE_TEMPLATE/release_checklist.yml). To start a
1437
release, create a new issue from that template and work through it item by

‎.github/actionlint.yaml‎

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -21,3 +21,14 @@ paths:
2121
".github/workflows/ci-workflow-health.yml":
2222
ignore:
2323
- 'unexpected key "queue" for "concurrency" section'
24+
# GitHub supports self-repository ($/) action references, but the latest
25+
# actionlint release does not yet recognize them (rhysd/actionlint#711).
26+
".github/workflows/release.yml":
27+
ignore:
28+
- 'specifying action "\$/[^"]+" in invalid format because ref is missing'
29+
".github/workflows/test-wheel-linux.yml":
30+
ignore:
31+
- 'specifying action "\$/[^"]+" in invalid format because ref is missing'
32+
".github/workflows/test-wheel-windows.yml":
33+
ignore:
34+
- 'specifying action "\$/[^"]+" in invalid format because ref is missing'

‎.github/actions/check-doc-links/action.yml‎

Lines changed: 0 additions & 114 deletions
This file was deleted.
Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
#
3+
# SPDX-License-Identifier: Apache-2.0
4+
5+
name: Harden pip
6+
description: >-
7+
Install the security-fixed pip (26.2.2) from a hash-pinned wheel so that later
8+
pip invocations do not fall back on the vendored certifi bundle that trusts the
9+
compromised eMudhra/emSign intermediate CAs. The sha256 fragment guarantees the
10+
bootstrap download's integrity even if the bootstrapping pip's trust store is
11+
compromised. See https://github.com/NVIDIA/cuda-python/issues/3195 and
12+
https://discuss.python.org/t/109576.
13+
14+
inputs:
15+
python:
16+
description: Python interpreter whose pip should be hardened (e.g. a venv's python).
17+
default: python
18+
19+
runs:
20+
using: composite
21+
steps:
22+
- name: Install hash-pinned pip 26.2.2
23+
shell: bash --noprofile --norc -xeuo pipefail {0}
24+
env:
25+
HARDEN_PIP_PYTHON: ${{ inputs.python }}
26+
run: |
27+
"${HARDEN_PIP_PYTHON}" -m pip install \
28+
"pip @ https://files.pythonhosted.org/packages/3f/06/3e4420302363fbb71be3c7f35ce042f54bfca330e3e66b3a2e3c4cca579f/pip-26.2.2-py3-none-any.whl#sha256=fd7f4dfa8190e014013b9f08f3e973f4e4c26ffd179a204895533a7254813a70"
Lines changed: 149 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,149 @@
1+
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
2+
#
3+
# SPDX-License-Identifier: Apache-2.0
4+
5+
name: Check documentation links
6+
description: Run a configured Lychee pre-commit hook with the shared URL cache and retained reports
7+
8+
inputs:
9+
kind:
10+
description: "Which hook to run: authored or rendered"
11+
required: true
12+
policy:
13+
description: >-
14+
check accepts up to 10 rate-limited URLs; cache-warm warns for any unresolved
15+
link; allow-429 accepts only HTTP 429 failures under the maintainer override
16+
default: check
17+
refresh:
18+
description: "Check links afresh and publish the successful results as the new baseline"
19+
default: "false"
20+
pr-number:
21+
description: "PR whose current maintainer override label should be read; empty skips the lookup"
22+
default: ""
23+
24+
outputs:
25+
allow-429:
26+
description: "Whether the current PR label explicitly enables the HTTP 429 override"
27+
value: ${{ steps.lychee-policy.outputs.allow-429 }}
28+
29+
# Callers provide Python and the checked-out repository; the build of any
30+
# rendered documentation must already be complete.
31+
runs:
32+
using: composite
33+
steps:
34+
- name: Read the PR's Lychee rate-limit override
35+
id: lychee-policy
36+
shell: bash --noprofile --norc -euo pipefail {0}
37+
env:
38+
GH_TOKEN: ${{ github.token }}
39+
GH_REPO: ${{ github.repository }}
40+
PR_NUMBER: ${{ inputs.pr-number }}
41+
run: |
42+
allow_429=false
43+
if [[ -n "$PR_NUMBER" ]]; then
44+
# Reruns keep the original event's labels, so query the current decision.
45+
# Only an exact boolean response can enable the maintainer override.
46+
lookup_ok=false
47+
for attempt in 1 2 3; do
48+
if result="$(gh pr view "$PR_NUMBER" --json labels --jq '.labels | any(.name == "ci: lychee-non-blocking")')"; then
49+
case "$result" in
50+
true|false) allow_429="$result"; lookup_ok=true; break ;;
51+
esac
52+
fi
53+
if (( attempt < 3 )); then
54+
sleep $((attempt * 5))
55+
fi
56+
done
57+
if [[ "$lookup_ok" != true ]]; then
58+
echo "::warning::Could not read a valid PR label decision; using the default link policy. Rerun to apply ci: lychee-non-blocking."
59+
elif [[ "$allow_429" == true ]]; then
60+
echo "::notice::ci: lychee-non-blocking permits only HTTP 429 failures after retries."
61+
fi
62+
fi
63+
echo "allow-429=$allow_429" >> "$GITHUB_OUTPUT"
64+
65+
- name: Describe the link cache
66+
id: cache
67+
shell: bash --noprofile --norc -euo pipefail {0}
68+
env:
69+
KIND: ${{ inputs.kind }}
70+
# Anything that can change a check result or the cache's meaning. The
71+
# hook configuration carries the native wheel pin and every hook option.
72+
POLICY_HASH: ${{ hashFiles('.pre-commit-config.yaml', 'lychee.toml', 'ci/tools/run_lychee.py', 'ci/tools/run_lychee_precommit.py', 'ci/tools/retry_lychee.py', 'ci/tools/checkout_source_links.py') }}
73+
run: |
74+
# Authored and rendered results stay apart: only rendered checks validate
75+
# fragments, and Lychee's cache does not record which validation ran.
76+
case "${KIND}" in
77+
authored) path=.lycheecache ;;
78+
rendered) path=.lychee-rendered/.lycheecache ;;
79+
*) echo "::error::Unknown link-check kind: ${KIND}"; exit 1 ;;
80+
esac
81+
prefix="lychee-v1-${KIND}-${POLICY_HASH}-baseline-"
82+
echo "path=${path}" >> "${GITHUB_OUTPUT}"
83+
echo "prefix=${prefix}" >> "${GITHUB_OUTPUT}"
84+
echo "key=${prefix}${GITHUB_RUN_ID}-${GITHUB_RUN_ATTEMPT}" >> "${GITHUB_OUTPUT}"
85+
86+
- name: Restore successful link checks
87+
id: cache-restore
88+
# Refresh mode deliberately skips earlier URL results.
89+
if: ${{ inputs.refresh != 'true' }}
90+
uses: actions/cache/restore@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
91+
with:
92+
path: ${{ steps.cache.outputs.path }}
93+
key: ${{ steps.cache.outputs.key }}
94+
restore-keys: ${{ steps.cache.outputs.prefix }}
95+
enableCrossOsArchive: true
96+
97+
- name: Explain a missing shared Lychee cache
98+
# Prefix restores are normal; cache-hit only identifies exact key matches.
99+
if: ${{ inputs.refresh != 'true' && steps.cache-restore.outputs.cache-matched-key == '' }}
100+
shell: bash --noprofile --norc -euo pipefail {0}
101+
env:
102+
KIND: ${{ inputs.kind }}
103+
DEFAULT_BRANCH: ${{ github.event.repository.default_branch }}
104+
WORKFLOW_URL: ${{ github.server_url }}/${{ github.repository }}/actions/workflows/lychee.yml
105+
run: |
106+
message="No shared Lychee ${KIND} cache was restored for this checker configuration. Checking links without this baseline may take longer and make more network requests."
107+
context="This is expected after checker changes in a PR; a refresh on ${DEFAULT_BRANCH} cannot seed settings that exist only in a PR."
108+
hint="For an unexpected miss on ${DEFAULT_BRANCH}, maintainers can run CI: Lychee checks and caches (lychee.yml) on ${DEFAULT_BRANCH} with refresh-cache enabled."
109+
echo "::notice title=Lychee ${KIND} cache::${message} ${context} ${hint} ${WORKFLOW_URL}"
110+
{
111+
printf '### Lychee %s cache\n\n' "$KIND"
112+
printf '%s\n\n%s\n\n%s\n\n' "$message" "$context" "$hint"
113+
printf '[Open the cache refresh workflow](%s).\n' "$WORKFLOW_URL"
114+
} >> "$GITHUB_STEP_SUMMARY"
115+
116+
- name: Harden pip
117+
uses: ./.github/actions/harden-pip
118+
119+
- name: Check links through the configured pre-commit hook
120+
shell: bash --noprofile --norc -euo pipefail {0}
121+
env:
122+
KIND: ${{ inputs.kind }}
123+
POLICY: ${{ steps.lychee-policy.outputs.allow-429 == 'true' && 'allow-429' || inputs.policy }}
124+
GITHUB_TOKEN: ${{ github.token }}
125+
run: |
126+
# Install into the interpreter that runs the driver, whatever shell
127+
# profile the caller's other steps use.
128+
python -m pip install pre-commit
129+
python ci/tools/run_lychee_precommit.py "${KIND}" --policy "${POLICY}" --max-attempts 3 \
130+
--report "${RUNNER_TEMP}/lychee-reports/lychee-${KIND}.json"
131+
132+
- name: Publish fresh successful link checks
133+
# Only refresh runs publish, so every pull request restores the same
134+
# baseline. Saving keeps original timestamps and never caches failures.
135+
if: ${{ !cancelled() && inputs.refresh == 'true' }}
136+
uses: actions/cache/save@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
137+
with:
138+
path: ${{ steps.cache.outputs.path }}
139+
key: ${{ steps.cache.outputs.key }}
140+
enableCrossOsArchive: true
141+
142+
- name: Upload link-check reports
143+
if: ${{ always() }}
144+
uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
145+
with:
146+
name: lychee-${{ inputs.kind }}-${{ runner.os }}-${{ github.run_id }}-${{ github.run_attempt }}
147+
path: ${{ runner.temp }}/lychee-reports/
148+
if-no-files-found: ignore
149+
retention-days: 7

‎.github/dependabot.yml‎

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -36,10 +36,6 @@ updates:
3636
# Match hook repositories, leaving additional_dependencies pins unchanged.
3737
allow:
3838
- dependency-name: "https://github.com/*"
39-
# Lychee's hook revision, LYCHEE_VERSION argument, and CI binary pin must
40-
# be upgraded together in a coordinated manual update.
41-
ignore:
42-
- dependency-name: "https://github.com/lycheeverse/lychee"
4339

4440
groups:
4541
pre-commit-monthly:

0 commit comments

Comments
 (0)