commit 20257da44630366714c9397d3a275a3a12c1fe26
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 20 19:33:56 2026 +0200

Revert "dev-python/nvidia-cudnn-frontend: drop 1.18.0"

This reverts commit 87f46f6df33c42d4b1a9a586e066298f0a8aa8a3.

commit 5a5db2325769be84b6c6e7585ed189c51311aa03
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 20 18:23:22 2026 +0200

dev-python/nvidia-cudnn-frontend: drop 1.18.0

commit e5732fe11ba4f067633addf560b387cd0be68454
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 20 18:23:22 2026 +0200

dev-python/nvidia-cudnn-frontend: add 1.24.0

Remove stale vllm-version cap comment (copied from 1.18.0; no longer
accurate at 1.24.0). Build mechanics and RDEPEND unchanged.

commit 165c0bce511847c653bacabf96187393887aa284
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:34:59 2026 +0200

dev-python/nvidia-cudnn-frontend: disable py3.11

commit dc02ba498bc658d25546042e56222fa7e4b30806
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 15:29:38 2026 +0200

dev-python/nvidia-cudnn-frontend: add 1.23.0

Bump 1.18.0 → 1.23.0 (current PyPI). Pure version-number change at the
ebuild level — RDEPEND (cudnn-9, nvidia-cuda-toolkit) and BDEPEND are
unchanged.

Build-check skipped: CMake configure does find_path(CUDNN_INCLUDE_DIR
cudnn.h) which requires dev-libs/cudnn merged on the host. Same shape
as the xgrammar host-gating note. pkgcheck passes; trust upstream
metadata pending pre-push end-to-end validation.

(nvchecker reported '8.1.0-beta' as the upstream max-tag — that's the
cudnn-frontend C++ library's tag scheme on github, not the
dev-python/nvidia-cudnn-frontend Python package's PyPI version.
Adding NVIDIA/* monorepo filters to the nvchecker generator is a
separate follow-up; the actual upstream is PyPI 1.23.0.)

commit b0a95a6e2552e3f94f66f7e09bd0d2f6d41a311f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 16:23:42 2026 +0200

dev-python/nvidia-cudnn-frontend: new package, 1.18.0

Tier 0 leaf for the vllm CUDA target packaging cycle. cuDNN Frontend
Python bindings — the highest version satisfying vllm's cuda.txt cap
of >=1.13.0,<1.19.0; 1.19+ has a breaking change.

Upstream's PyPI release at this version is wheel-only with no sdist;
we build the binding from the matching tag at NVIDIA/cudnn-frontend.
The Python wheel ships its own copy of the cudnn_frontend C++ headers
(under site-packages/include/), so the in-tree ::gentoo
sci-ml/cudnn-frontend-1.16.1 C++ header package is NOT consumed at
build time — the "ABI gap" the audit memo flagged is moot for this
shape, and the 1.16.1 vs 1.18.0 mismatch concern doesn't apply.

CMake configure FetchContent's DLPack header-only and there's no
in-tree alternative without carrying a sizeable patch series, so
RESTRICT="network-sandbox" is accepted (same shape vllm uses for its
cpu target). pkgcheck's UnknownRestrict warning is a known false
positive (the keyword list lags portage); --scan=n at commit time.

Builds against system dev-libs/cudnn (>=9 in ::gentoo) and
dev-util/nvidia-cuda-toolkit. License is MIT-style ("Permission is
hereby granted...") per LICENSE.txt — pyproject.toml's "NVIDIA
Proprietary Software" string is incorrect.