Revisit cuda-bindings branching strategy

Right now we have cuda-bindings 12.9.x maintained in the 12.9.x branch. This causes frictions for

understanding the maintenance scope
- it is not obvious at all that only .github/, ci/, cuda_bindings/, and cuda_python/ are maintained there
tracking and backpoting fixes (CI: Automate backport process more #965)
maintaining CI infra (CI: Make the dependbot also update action verisions in the backport branch #1058) and docs (CI: Make the backport branch generate docs too? #809)
releasing 12.x packages (BUG: ./ci/tools/download-wheels does not exist in the backport branch #1120)

We would like to lower the burden and the mental overhead on the team. Possible solutions include:

Eliminate the backport needs by hosting two directories cuda_bindings_12/ and cuda_bindings_13/ on the main branch
Generate the same binding code that have a bunch of macros that can be toggled at build time to target CUDA 12/13
...

TL;DR

The pain comes from having multiple long-lived branches, not from how the source is laid out. Move to a single branch with per-major directories and most of it goes away. Build-time macros (proposal #2) is a nicer end-state but the wrong first step.

What I found

The .pyx.in/.pxd.in files are not hand-written templates. They're generated, with a banner: "automatically generated… Do not modify it directly."
What differs between main (13.x) and 12.9.x is real CUDA 12-vs-13 API surface (new enums, structs, renumbered values) — emitted by the generator. ~600 lines in cydriver.pxd.in alone.
All generated output (56 .pyx/.pxd/.cpp + 18 .in) is git-tracked. The build already regenerates from the installed toolkit's headers, so the committed copies are effectively cached.

So: every pain you listed (#965, #1058, #809, #1120) exists because there are multiple branches. Collapse to one branch and all four go away — independent of dirs-vs-macros.

The two proposals

#1 — per-major directories on main (recommended first step)

One branch. Each CUDA major is a generated directory: gen/v12/, gen/v13/. New major = copy + regenerate. EOL = delete the directory. No branch is ever cut or retired.
Hand-written code (__init__.py, _lib/, build_hooks.py, ci/, .github/, docs) lives once. A fix is one PR — no backport.
Build/release picks the major from the toolkit headers (already keys off CUDA_PATH) and maps gen/vNN/ onto the cuda.bindings package. Same package name, major-distinct versions, one installed at a time — no wheel collisions.

#2 — single source + build-time macros (better end-state, later)

The deltas are real API differences the generator owns. A single tree with {{if cuda_major==13}} means either hand-editing thousands of generated symbols, or teaching the upstream generator to emit version-gated code. That's a generator project, not a repo change. (CuPy does this — worth a look once/if the generator supports it.)

Migration path

Make the build's major-selection explicit so a gen/vNN/ dir can be picked deterministically.
Move 13.x into gen/v13/; verify wheels are byte-equivalent to today.
Re-home 12.9.x into gen/v12/ and cut a patch release from main — proves release works without a branch (closes the BUG: ./ci/tools/download-wheels does not exist in the backport branch #1120 class).
Decide: commit gen/ vs. regenerate in CI. If regenerating, validate generator availability in CI.
Retire 11.8.x / 12.2.x / 12.9.x as their lines move to directories or hit EOL.

Bottom line: single-branch is the decision that matters. Start with per-major directories; treat build-time macros as a later optimization.

— codex GPT-6 on behalf of Ralf —

CI/release consolidation: before and after

TL;DR: CUDA 12 and CUDA 13 now share CI/release infrastructure on main, replacing two separately maintained branch implementations. Counting .github/ and ci/ across both old branches, the combined maintenance surface decreased:

Measure	Before: `main` + `12.9.x`	After: shared `main`	Reduction
All tracked text lines	22,737	20,819	1,918 (8.4%)
Code/configuration lines, including tests	15,488	14,211	1,277 (8.2%)

Production/configuration code decreased 14.9% (13,196 → 11,229); CI-tool test code grew by 690 lines (2,292 → 2,982). These counts describe maintenance surface. The architectural benefit is removing routine cross-branch CI/release synchronization and its opportunities for divergence.

Overview. The migration stack and cleanup in #3434 and #3456 put the CUDA-major choice into a shared registry and shared workflows. Builds, tests, releases and documentation can now be maintained together. Both bindings majors remain available; retiring routine 12.9.x maintenance preserves the branch's Git history.

The tradeoff is explicit: the shared implementation is larger than old main alone—18,364 → 20,819 total text lines—while the combined two-branch footprint is smaller. The comparison also includes concurrent infrastructure changes; it does not isolate the migration's causal effect on every line.

What changed, what remains, and the detailed breakdown

Maintenance responsibilities reduced

One CI/release implementation handles both bindings majors. Core and metapackage testing can use the same run's CUDA 12 builds, replacing cross-branch artifact coordination. Shared behavior.
ci: retire legacy branch-based bindings maintenance #3322 removed the automatic/manual backport workflow. ci: remove temporary nightly artifact fallback #3434 removed the temporary nightly fallback; ci: retire legacy release-layout compatibility #3456 removed legacy release-layout parsing and its obsolete matrix reader/tests.
Future infrastructure changes can be reviewed and validated in one maintained branch, instead of reconciling two independently evolving .github/ and ci/ trees.

Remaining obligations

cuda_bindings_12/ and cuda_bindings_13/ remain separate package roots. Applicable handwritten fixes still require checking both; generated bindings still require each major's own toolkit inputs. These package sources are outside this LOC comparison. Maintenance guidance.
Current release workflows expect the registry layout. Releases of older, pre-registry layouts use compatible historical workflow refs, retaining access to old behavior without carrying its parser in the current workflows. Release policy.
Full CI, nightly, coverage and eight release rehearsals were completed for the cleanup. The validation summary records the scope and limits, including the exclusion of live release publication.

Production, tests, documentation and assets

Category	All text lines before	After	Code/config lines before	After
Production/configuration	17,265	14,602	13,196	11,229
CI-tool tests	3,071	3,939	2,292	2,982
Documentation/diagrams	789	1,132	Excluded	Excluded
Certificate/data assets	1,612	1,146	Excluded	Excluded
Total	22,737	20,819	15,488	14,211

The production/configuration reduction of 1,967 code lines and test increase of 690 yield the net reduction of 1,277. The physical-line total also includes the removal of one 466-line copy of certificate data; that asset contributes nothing to the code/configuration measure.

By directory

Directory	All text lines before → after	Code/config lines before → after
`.github/`	12,811 → 9,854	9,155 → 7,113
`ci/`	9,926 → 10,965	6,333 → 7,098

Shared Python helpers and their tests grew as workflow handling was consolidated. Directory subtotals show this redistribution; the headline comparison includes both directories and both old branches.

How different were the old trees?

Matching files strictly by relative path:

Relationship between old `main` and `12.9.x`	Paths
Present in both, byte-identical	3
Present in both, different contents	31
Present only in `main`	92
Present only in `12.9.x`	7

The two old trees contained 126 + 41 = 167 file copies; the shared tree contains 144. Different contents include intentional CUDA-major differences, and path-only comparisons do not identify renames or quantify harmful drift. These figures show the comparison/synchronization surface that routine branch maintenance previously involved.

Full details: fixed snapshots, counting rules and reproducibility

Snapshots

The before boundary is immediately before #3314, the first PR of the implemented migration, merged on 2026-11-06 at 16:37:20 UTC. This includes its preparatory work in the measured change. It is not the earlier historical creation of the cuda_bindings_12/ directory.

Snapshot	Commit	All text lines	Code/config lines
Before: `main`	`ceb91b50b91293231c13494ea414caa8d115a759`	18,364	12,437
Before: `12.9.x`, at the same cutoff	`06e182cee03624c6ebdf3b865eaaefa201197661`	4,373	3,051
After: reviewed #3456 head	`8f01de551993ab0d9102002d97a9c113776bdd39`	20,819	14,211

The 12.9.x snapshot is the result of #3306; its next update (#3429) occurred November 10 and did not change either counted directory. The after snapshot is fixed to the reviewed #3456 revision, rather than a moving main reference.

Counting rules

Scope is every tracked regular text file below .github/ and ci/. Every file copy in each old branch counts, even if byte-identical. All three snapshots contain only regular UTF-8 text files within this scope; there are no binary files, symlinks or submodules to resolve.
All text lines means LF-delimited physical lines, including blank lines, comments, documentation and text assets; an unterminated final line counts once.
Code/configuration means cloc 1.98's code field: nonblank lines after its language-specific comment filtering, including tests. Exclude *.md, *.txt, *.mmd, *.pem, and the vendored certificate disclosure .github/actions/upgrade-ca-certificates/impacted-intermediates.json. Configuration YAML/JSON remains included, including issue forms and problem matchers.
CI-tool tests are code/configuration files under ci/tools/tests/, plus ci/tools/smoke_test_lychee_precommit.py. Operational test runners, matrix selectors and test workflows remain production/configuration infrastructure.
Use --skip-uniqueness so cloc counts duplicate copies; --config /dev/null avoids user-specific options. Map .env to shell syntax. Count the old one-line .github/BACKPORT_BRANCH setting explicitly with shell filters. Every eligible file was accounted for; no unknown file types were silently omitted.
cloc provides a repeatable syntax-based approximation. Embedded scripts and prose-valued YAML prevent an exact semantic “functional LOC” interpretation. LOC also does not measure review difficulty, runtime, job count or developer hours.
The broad scope includes copies of default-branch-scoped GitHub metadata. A copied file on 12.9.x does not establish that a second instance executed. Retiring that branch removes its routine maintenance surface; historical copies remain in Git.
Twelve unrelated PRs also changed these directories during the interval: CI: Fix compute-sanitizer setup and harden mini-CTK fetching #3397, ci: install cuda-gdb for cuda.core local-CTK tests #3371, ci: fix Linux coverage checkout ownership #3412, ci: capture coverage before pytest imports #3358, [doc-only] docs(cuda.core): define the cuda-bindings floor as the lowest compatible release #3425, [doc-only] docs(cuda.core): fix the floor-move wording; advise against pinning cuda-bindings #3441, [no-ci] CI: verify the Authenticode signature of vc_redist.x64.exe #3443, [doc-only] docs(release): tag cuda-core releases on a release branch #3431, ci: install mini-CTK CUPTI under extras/CUPTI and check pathfinder finds it #3440, pathfinder: preload NVRTC builtins for CUDA Toolkits on Linux #3439, [CI] Run WoA validation dispatch nightly #3357 and CI: install gh from a pinned keyring checked into the repo #3449. Consequently, these are observed before/after totals. Subtracting their commit diffs would not create a reliable migration-only counterfactual, because later cleanup removed some code they introduced.

Two independently implemented counts agreed on the headline numbers and directory/test subtotals. The standalone reproduction below was also run successfully against the fixed objects.

Reproduce the counts from Git objects

Run this Python script from a clone of NVIDIA/cuda-python containing the three commits, with Python 3 and cloc 1.98 installed. It reads fixed Git objects and extracts temporary data; it does not inspect or change the checkout, branches, or tags.

# Run from a NVIDIA/cuda-python clone containing the three commit objects.
# Requires Python 3 and cloc 1.98. Does not read or change the checkout.
import json
from pathlib import Path
import subprocess
import tempfile

snapshots = {
    'main before': 'ceb91b50b91293231c13494ea414caa8d115a759',
    '12.9.x before': '06e182cee03624c6ebdf3b865eaaefa201197661',
    'after': '8f01de551993ab0d9102002d97a9c113776bdd39',
}
asset = '.github/actions/upgrade-ca-certificates/impacted-intermediates.json'
excluded = {'.md', '.txt', '.mmd', '.pem'}
results, manifests = {}, {}

def git(*args):
    return subprocess.check_output(['git', *args])

def cloc(files, output, force_language):
    command = ['cloc', '--config', '/dev/null', '--skip-uniqueness',
               '--by-file', '--json', '--quiet', '--timeout=0',
               '--force-lang=' + force_language, '--out=' + str(output)]
    subprocess.run(command + list(map(str, files)), check=True,
                   stdout=subprocess.DEVNULL)
    return {str(Path(p).resolve()): v for p, v in json.loads(output.read_text()).items()
            if p not in ('header', 'SUM')}

with tempfile.TemporaryDirectory(prefix='issue1199-') as temporary:
    root = Path(temporary).resolve()
    for index, (label, sha) in enumerate(snapshots.items()):
        directory = root / str(index)
        physical = test_code = production_code = 0
        selected, manifests[label] = [], {}
        tree = git('ls-tree', '-rz', '--full-tree', sha, '--', '.github', 'ci')
        for entry in tree.split(b'\0'):
            if not entry:
                continue
            metadata, raw_path = entry.split(b'\t', 1)
            mode, kind, oid = metadata.decode().split()
            assert kind == 'blob' and mode in ('100644', '100755')
            path = raw_path.decode()
            manifests[label][path] = oid
            data = git('cat-file', 'blob', oid)
            assert b'\0' not in data
            data.decode('utf-8')
            physical += data.count(b'\n') + int(bool(data) and not data.endswith(b'\n'))
            target = directory / path
            target.parent.mkdir(parents=True, exist_ok=True)
            target.write_bytes(data)
            if target.suffix.lower() not in excluded and path != asset:
                selected.append(target)
        special = directory / '.github/BACKPORT_BRANCH'
        normal = [p for p in selected if p != special]
        counts = cloc(normal, root / 'counts.json', 'Bourne Shell,env')
        if special in selected:
            counts.update(cloc([special], root / 'special.json', 'Bourne Shell'))
        assert set(counts) == {str(p.resolve()) for p in selected}
        for filename, count in counts.items():
            path = str(Path(filename).relative_to(directory))
            if path.startswith('ci/tools/tests/') or path == 'ci/tools/smoke_test_lychee_precommit.py':
                test_code += count['code']
            else:
                production_code += count['code']
        results[label] = dict(physical=physical, code=production_code + test_code,
                             production=production_code, tests=test_code)

first, second = manifests['main before'], manifests['12.9.x before']
common = first.keys() & second.keys()
results['before file pairs'] = dict(
    identical=sum(first[p] == second[p] for p in common),
    different=sum(first[p] != second[p] for p in common),
    main_only=len(first.keys() - second.keys()),
    maintenance_only=len(second.keys() - first.keys()))
print(json.dumps(results, indent=2))

The two before results are added, and the reduction percentage is (before_total - after) / before_total * 100.

Per-file counts (155 paths across the three snapshots)

Each cell is physical text lines / code-config lines. A single - means the path is absent; P/- means the file contributes P physical lines and is excluded from code/configuration. Zero code lines remain 0.

Relative path	Old `main` P/C	Old `12.9.x` P/C	After P/C
`.github/BACKPORT_BRANCH`	-	1/1	-
`.github/CODERABBIT.md`	34/-	-	34/-
`.github/ISSUE_TEMPLATE/bug_report.yml`	115/103	116/103	115/103
`.github/ISSUE_TEMPLATE/config.yml`	9/5	9/5	9/5
`.github/ISSUE_TEMPLATE/doc_request.yml`	43/35	44/35	43/35
`.github/ISSUE_TEMPLATE/feature_request.yml`	71/62	71/62	71/62
`.github/ISSUE_TEMPLATE/release_checklist.yml`	34/28	37/31	36/30
`.github/PULL_REQUEST_TEMPLATE.md`	13/-	14/-	13/-
`.github/RELEASE-bindings.md`	-	-	119/-
`.github/RELEASE-core.md`	317/-	-	368/-
`.github/actionlint.yaml`	49/36	-	49/36
`.github/actions/doc_preview/action.yml`	50/37	69/55	50/37
`.github/actions/download-artifact-retry/action.yml`	128/115	-	128/115
`.github/actions/fetch_ctk/action.yml`	221/166	191/146	284/221
`.github/actions/get_pr_number/action.yml`	26/19	58/49	26/19
`.github/actions/griffe-api-check/action.yml`	44/36	-	44/36
`.github/actions/install-sccache/action.yml`	75/43	-	75/43
`.github/actions/install_unix_deps/action.yml`	47/39	49/41	47/39
`.github/actions/install_windows_deps/action.yml`	92/79	-	92/79
`.github/actions/lychee-links/action.yml`	146/119	-	146/119
`.github/actions/sccache-summary/action.yml`	80/58	-	80/58
`.github/actions/setup-hardened-python/action.yml`	168/133	150/117	168/133
`.github/actions/setup-pixi/action.yml`	33/24	-	33/24
`.github/actions/upgrade-ca-certificates/action.yml`	144/96	144/96	144/96
`.github/actions/upgrade-ca-certificates/emudhra_intermediates.ps1`	124/79	50/30	124/79
`.github/actions/upgrade-ca-certificates/impacted-intermediates.json`	466/-	466/-	466/-
`.github/copy-pr-bot.yaml`	10/3	10/3	10/3
`.github/dependabot.yml`	79/54	10/6	57/38
`.github/labeler.yml`	29/19	-	29/19
`.github/matchers/compiler.json`	43/43	-	43/43
`.github/matchers/compute-sanitizer.json`	24/24	-	24/24
`.github/workflows/backport.yml`	76/63	40/31	-
`.github/workflows/bandit.yml`	50/37	23/17	50/37
`.github/workflows/build-docs.yml`	394/317	261/212	460/376
`.github/workflows/build-wheel.yml`	721/614	306/244	799/684
`.github/workflows/ci-nightly.yml`	402/340	-	415/353
`.github/workflows/ci-pixi-lockfile-freshness-check.yml`	43/32	-	43/32
`.github/workflows/ci-pixi-lockfile-refresh.yml`	35/25	-	35/25
`.github/workflows/ci-pixi-source-test.yml`	244/150	-	339/229
`.github/workflows/ci-workflow-health.yml`	261/238	-	261/238
`.github/workflows/ci.yml`	759/621	201/175	659/529
`.github/workflows/cleanup-pr-previews.yml`	77/61	-	77/61
`.github/workflows/codeql.yml`	-	43/34	-
`.github/workflows/coverage.yml`	614/500	-	643/523
`.github/workflows/guess_latest.sh`	-	42/21	-
`.github/workflows/install_gpu_driver.ps1`	-	48/25	-
`.github/workflows/lychee.yml`	104/78	-	104/78
`.github/workflows/pr-auto-label.yml`	23/17	-	23/17
`.github/workflows/pr-metadata-check.yml`	166/129	-	166/129
`.github/workflows/pre-commit.yml`	189/125	-	189/125
`.github/workflows/release-upload.yml`	103/82	89/64	54/45
`.github/workflows/release.yml`	469/387	184/162	566/489
`.github/workflows/runner-trust-probe.yml`	209/158	-	209/158
`.github/workflows/security-suite.yml`	50/29	-	50/29
`.github/workflows/status-check.yml`	-	20/14	-
`.github/workflows/test-sdist-linux.yml`	171/138	-	177/144
`.github/workflows/test-sdist-windows.yml`	169/127	-	175/133
`.github/workflows/test-wheel-linux.yml`	604/477	262/210	575/459
`.github/workflows/test-wheel-windows.yml`	580/485	246/202	559/470
`.github/workflows/triage-author-reply.yml`	42/32	-	42/32
`.github/workflows/triage-close-awaiting-response.yml`	43/31	-	43/31
`.github/workflows/triagelabel.yml`	27/20	26/19	27/20
`.github/workflows/woa-main-ci.yml`	192/177	-	187/171
`ci/README.md`	253/-	-	418/-
`ci/__init__.py`	-	-	2/0
`ci/ci-pipeline.mmd`	99/-	-	101/-
`ci/cleanup-pr-previews`	343/257	-	343/257
`ci/pixi-version.env`	4/1	-	4/1
`ci/pyproject.toml`	-	-	30/24
`ci/test-matrix.json`	-	102/102	-
`ci/test-matrix.yml`	164/99	-	164/99
`ci/tools/__init__.py`	-	-	2/0
`ci/tools/bindings_config.py`	-	-	308/247
`ci/tools/build_docs_for_link_check.sh`	64/44	-	64/44
`ci/tools/check_mempool_hygiene.py`	261/166	-	261/166
`ci/tools/check_pathfinder_ctk_libs.py`	-	-	44/23
`ci/tools/check_pixi_cuda_version.py`	91/73	-	79/62
`ci/tools/check_release_notes.py`	274/195	-	201/134
`ci/tools/checkout_source_links.py`	104/70	-	104/70
`ci/tools/ci_artifacts.py`	-	-	65/41
`ci/tools/ci_env.py`	-	-	134/115
`ci/tools/classify_pixi_lockfile_freshness.py`	85/58	-	85/58
`ci/tools/compute_ci_plan.py`	224/181	-	406/345
`ci/tools/configure_driver_mode.ps1`	112/76	-	112/76
`ci/tools/cuda_core_bindings_floor.py`	74/40	-	74/40
`ci/tools/download-wheels`	112/69	75/45	180/131
`ci/tools/env-vars`	145/102	77/61	-
`ci/tools/fetch_ctk_redistrib.py`	270/195	308/234	271/196
`ci/tools/fold_pytest_sections.py`	93/42	-	93/42
`ci/tools/guess_latest.sh`	42/21	-	-
`ci/tools/install_gpu_driver.ps1`	92/50	-	92/50
`ci/tools/install_gpu_driver.sh`	260/128	-	260/128
`ci/tools/list_pixi_workspaces.py`	78/45	-	78/45
`ci/tools/lookup-run-id`	322/249	104/71	308/243
`ci/tools/merge_cuda_core_wheels.py`	252/162	-	252/162
`ci/tools/retry_lychee.py`	234/191	-	234/191
`ci/tools/run-tests`	276/174	108/84	297/197
`ci/tools/run_lychee.py`	96/62	-	96/62
`ci/tools/run_lychee_precommit.py`	104/83	-	104/83
`ci/tools/runner_trust_probe.ps1`	163/128	-	163/128
`ci/tools/runner_trust_probe.sh`	150/116	-	150/116
`ci/tools/runner_trust_probe_chains/README.txt`	39/-	-	39/-
`ci/tools/runner_trust_probe_chains/dv-g3b-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/dv-g3b-aaa-noct-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/dv-g3b-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_chains/dv-g3b-noct-emudhra.we-have-a-problem.com.pem`	57/-	-	57/-
`ci/tools/runner_trust_probe_chains/dv-g3c-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/dv-g3c-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_chains/ev-g3b-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/ev-g3b-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_chains/ev-g3c-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/ev-g3c-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_chains/ov-g3b-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/ov-g3b-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_chains/ov-g3c-aaa-emudhra.we-have-a-problem.com.pem`	38/-	-	38/-
`ci/tools/runner_trust_probe_chains/ov-g3c-emudhra.we-have-a-problem.com.pem`	56/-	-	56/-
`ci/tools/runner_trust_probe_issuers/README.txt`	20/-	-	20/-
`ci/tools/runner_trust_probe_issuers/emsign-root-cs-ca-g3-cross.pem`	21/-	-	21/-
`ci/tools/runner_trust_probe_node.js`	67/53	-	67/53
`ci/tools/select-woa-batch`	81/63	-	81/63
`ci/tools/select_pixi_packages.py`	-	-	83/68
`ci/tools/select_test_matrix.py`	-	-	126/107
`ci/tools/setup-sanitizer`	30/13	22/13	30/13
`ci/tools/smoke_test_lychee_precommit.py`	73/47	-	73/47
`ci/tools/tests/AGENTS.md`	-	-	20/-
`ci/tools/tests/test_bindings_config.py`	-	-	207/177
`ci/tools/tests/test_bindings_metadata.py`	-	-	81/67
`ci/tools/tests/test_bindings_release_compat.py`	-	-	115/98
`ci/tools/tests/test_build_docs_for_link_check.py`	136/99	-	136/99
`ci/tools/tests/test_check_mempool_hygiene.py`	219/142	-	219/142
`ci/tools/tests/test_check_release_notes.py`	402/315	-	294/227
`ci/tools/tests/test_check_spdx.py`	195/134	-	195/134
`ci/tools/tests/test_checkout_source_links.py`	164/134	-	164/134
`ci/tools/tests/test_classify_pixi_lockfile_freshness.py`	63/40	-	63/40
`ci/tools/tests/test_compute_ci_plan.py`	184/155	-	217/178
`ci/tools/tests/test_cuda_core_bindings_floor.py`	99/57	-	99/57
`ci/tools/tests/test_cuda_python_setup.py`	-	-	107/96
`ci/tools/tests/test_download_wheels.py`	-	-	131/89
`ci/tools/tests/test_fetch_ctk_redistrib.py`	167/127	159/125	197/153
`ci/tools/tests/test_fold_pytest_sections.py`	175/91	-	175/91
`ci/tools/tests/test_list_pixi_workspaces.py`	118/93	-	118/93
`ci/tools/tests/test_merge_cuda_core_wheels.py`	-	-	38/29
`ci/tools/tests/test_pixi_lockfile_freshness_workflow.py`	120/88	-	109/80
`ci/tools/tests/test_release_publish_security.py`	-	-	25/16
`ci/tools/tests/test_retry_lychee.py`	449/364	-	449/364
`ci/tools/tests/test_run_lychee.py`	132/109	-	132/109
`ci/tools/tests/test_run_lychee_precommit.py`	178/146	-	178/146
`ci/tools/tests/test_select_pixi_packages.py`	-	-	53/34
`ci/tools/tests/test_select_test_matrix.py`	-	-	121/88
`ci/tools/tests/test_test_matrix_driver_hashes.py`	38/26	-	38/26
`ci/tools/tests/test_validate_release_wheels.py`	-	-	205/168
`ci/tools/validate-release-wheels`	138/104	131/99	-
`ci/tools/validate_release_wheels.py`	-	-	270/217
`ci/versions.json`	-	7/7	-
`ci/versions.yml`	25/15	-	31/19
Total	18,364/12,437	4,373/3,051	20,819/14,211

Navigation Menu

Repository navigation

Revisit `cuda-bindings` branching strategy #1199

TL;DR

What I found

The two proposals

Migration path

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Revisit cuda-bindings branching strategy #1199

Description

Activity

rparolin commented on Jun 9, 2026

TL;DR

What I found

The two proposals

Migration path

rwgk commented on Nov 13, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Revisit `cuda-bindings` branching strategy #1199