Install this version:
emerge -a =dev-python/dask-cuda-26.6.0
If this version is masked, you can unmask it using the autounmask tool or standard emerge options:
autounmask =dev-python/dask-cuda-26.6.0
Or alternatively:
emerge --autounmask-write -a =dev-python/dask-cuda-26.6.0
| Version | EAPI | Keywords | Slot |
|---|---|---|---|
| 26.6.0 | 8 | ~amd64 | 0 |
# Copyright 1999-2026 Gentoo Authors
# Distributed under the terms of the GNU General Public License v2
EAPI=8
DISTUTILS_USE_PEP517=setuptools
PYTHON_COMPAT=( python3_{12..14} )
inherit distutils-r1
# GitHub zero-pads the calendar version in the tag (26.06.00); PyPI
# normalises it to 26.6.0, which is our ${PV}. The in-tree dask_cuda/VERSION
# file already carries the clean 26.06.00 string that setuptools' dynamic
# version reads, so no version patching is needed here.
MY_PV="26.06.00"
DESCRIPTION="Utilities for running Dask workers on CUDA-enabled systems"
HOMEPAGE="
https://github.com/rapidsai/dask-cuda
https://pypi.org/project/dask-cuda/
"
SRC_URI="
https://github.com/rapidsai/dask-cuda/archive/refs/tags/v${MY_PV}.tar.gz
-> ${P}.gh.tar.gz
"
S="${WORKDIR}/${PN}-${MY_PV}"
LICENSE="Apache-2.0"
SLOT="0"
KEYWORDS="~amd64"
# dask-cuda is pure-Python; RMM-based device-memory spilling is optional
# (sci-libs RMM is not packaged) and not a hard dependency. cuda-core is
# satisfied by our dev-python/cuda-python stack; nvidia-ml-py (pynvml) is
# the GPU telemetry binding. rapids-dask-dependency pulls the dask /
# distributed pair (see that package for the relaxed-pin rationale).
RDEPEND="
>=dev-python/click-8.1[${PYTHON_USEDEP}]
>=dev-python/cuda-core-0.3.2[${PYTHON_USEDEP}]
>=dev-python/numpy-1.23[${PYTHON_USEDEP}]
>=dev-python/nvidia-ml-py-12[${PYTHON_USEDEP}]
>=dev-python/pandas-1.3[${PYTHON_USEDEP}]
~dev-python/rapids-dask-dependency-26.6.0[${PYTHON_USEDEP}]
>=dev-python/zict-2.0.0[${PYTHON_USEDEP}]
"
python_prepare_all() {
# Upstream builds through rapids_build_backend, a RAPIDS PEP517 shim
# whose only effects are (a) appending a CUDA suffix (-cu13) to
# dependency names and (b) sourcing deps from dependencies.yaml. We
# want neither — our deps are unsuffixed and already static in
# [project.dependencies] — and its declared inner backend is plain
# setuptools.build_meta. Swap to setuptools directly so we don't have
# to package rapids-build-backend + rapids-dependency-file-generator
# for byte-identical output. The dynamic version still resolves from
# dask_cuda/VERSION via [tool.setuptools.dynamic]. verified 2026-06-10
sed -i \
-e 's/build-backend = "rapids_build_backend.build"/build-backend = "setuptools.build_meta"/' \
-e '/"rapids-build-backend>=0.4.0,<0.5.0",/d' \
pyproject.toml || die
distutils-r1_python_prepare_all
}
>=dev-python/click-8.1[] >=dev-python/cuda-core-0.3.2[] >=dev-python/numpy-1.23[] >=dev-python/nvidia-ml-py-12[] >=dev-python/pandas-1.3[] ~dev-python/rapids-dask-dependency-26.6.0[] >=dev-python/zict-2.0.0[]
| Type | File | Size | Source URLs |
|---|---|---|---|
| DIST | dask-cuda-26.6.0.gh.tar.gz | 270287 bytes | https://github.com/rapidsai/dask-cuda/archive/refs/tags/v26.06.00.tar.gz |