commit 1b5f6b90c3ba7d09038d3d6710d37b8307dc15da
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Sep 11 10:46:10 2026 +0200

sci-ml/torchvision: trim 0.29.0-r90 comments

commit a7300ea1d3a8b8dff7b4285416ee6b36116c12a7
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:15:06 2026 +0200

sci-ml/torchvision: explain 0.29.0 random-device access

commit f4bb0aaf780aa8a379f6859b56ba7329b8cf1e49
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:15:06 2026 +0200

sci-ml/torchvision: explain 0.28.0 random-device access

commit 0a72f43cfc9af6eabaca5f4c3c06ad198c28b139
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 09:15:06 2026 +0200

sci-ml/torchvision: explain 0.26.0 random-device access

commit b109fddf7500004ff4a23c8f59aff3dd9ff9761f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 05:52:46 2026 +0200

sci-ml/torchvision: trim 0.26.0 comment prose

commit 379a180440ab30b223f114cf2c2c1fcf2f22b68f
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 05:33:33 2026 +0200

sci-ml/torchvision: trim 0.28.0 comment prose

commit 0a0e3ec79f600ee9a69b20a89f7b0b40db556bea
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 13:20:21 2026 +0200

sci-ml/torchvision: keep the fork ahead of ::gentoo's -r1, revbump

::gentoo now ships torchvision-0.29.0-r1, which outranks this fork's
0.29.0. Now that sci-ml/pytorch forwards caffe2's USE flags, that ebuild
resolves here and would replace the fork on the next update, and its
differences are not wanted. Its ffmpeg flag sets TORCHVISION_USE_FFMPEG,
which 0.29.0's setup.py does not read, so the flag only adds an ffmpeg
dependency; _VIDEO_CODEC, which it sets from cuda, is not read either.
Its filelock build dependency is already covered, because this overlay's
pytorch depends on filelock. The fork also keeps its test deselects,
narrowed and checked against 0.28.0. -r90 matches the revision range of
the caffe2 and pytorch forks and stays clear of ::gentoo's revisions.

The revbump also declares DEPEND. setup.py imports torch and compiles
against caffe2's installed headers, and pyproject.toml lists torch as a
build requirement, but pytorch and caffe2 were only in RDEPEND, which
does not guarantee they are installed before the build starts.

commit aee6facbbc28e5e543bbcb2adc53fba3b2f10ea9
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 3 02:49:21 2026 +0200

sci-ml/torchvision: add 0.29.0

Completes the trio. The only ebuild delta from 0.28.0 is the pytorch pin
=sci-ml/pytorch-2.13* -> 2.14*, which upstream confirms: 0.29.0's
setup.py adds TORCH_TARGET_VERSION = "0x020e000000000000" -- 0x020e is
2.14 -- and passes it as -DTORCH_TARGET_VERSION to every extension.

The ffmpeg USE flag stays dropped: TORCHVISION_USE_FFMPEG and
TORCHVISION_USE_VIDEO_CODEC still appear nowhere in 0.29.0's setup.py,
so they remain the dead knobs the 0.28.0 fork removed.

BUILD-CHECK COULD NOT RUN HERE, and this is host gating rather than an
ebuild defect. torchvision compiles C++ extensions against the INSTALLED
torch headers, and this host has caffe2-2.13.0-r91 merged, so the build
fails with

csrc/ops/cpu/deform_conv2d_kernel.cpp: 'permute' is not a member of
'torch::stable'
csrc/ops/hip/*_kernel.hip: 'torch/headeronly/cuda/KernelUtils.h' file
not found

Both are 2.14.0 additions, confirmed against the tarball: the header
exists at torch/headeronly/cuda/KernelUtils.h in pytorch-2.14.0 and is
absent from the installed 2.13 tree, and torch::stable::permute is
declared in torch/csrc/stable/ops.h. So a user merging in dependency
order gets caffe2-2.14.0 first and these resolve. Verifying it here
would mean displacing the working 2.13 stack on this machine, which is
Ivan's call, not something to do mid-audit.

commit 7914c0f920dbb833f4d6981fdd67012aa5690d95
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 09:42:22 2026 +0200

sci-ml/torchvision: normalize copyright header

Use the overlay-wide required copyright range consistently across consumer-retained
versions.

commit 7ea91ed531a11202d4a0c47a230b76005a120d80
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Aug 8 12:01:30 2026 +0200

sci-ml/torchvision: add 0.26.0

Pairs with sci-ml/pytorch-2.11.0-r90 for the restored vllm(torch==2.11)
stack; torchvision 0.26.0 requires torch==2.11.0 exactly, so it pins
=sci-ml/pytorch-2.11*. Adapted from our 0.28.0 fork. Test suite is
RESTRICTed -- the 0.28.0 deselect set is version-specific and was not
re-vetted for 0.26.0; this is a frozen rollback for installability.

commit ca8261c4ffaadde41193177e0267b04df6593353
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 3 14:49:49 2026 +0200

*/*: normalize metadata.xml whitespace

Use the overlay-wide two-space XML indentation convention and remove the remaining trailing whitespace so metadata stays consistent and diff-safe.

commit d26e488d095396a8343a1053601f18f59982a8df
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jul 27 10:04:12 2026 +0200

*/*: drop dead Python targets from PYTHON_COMPAT

139 ebuilds across 65 packages still listed python3_10, python3_11 or
python3_13t. None of the three is a supported implementation any more:
gentoo's python-utils-r1 sets _PYTHON_ALL_IMPLS to python3_ plus
python3_t, and everything older falls into the branch commented
"implementations deprecated prior to EAPI 9 are fatal".

Inert today, which is why nothing caught it. Targets outside
_PYTHON_ALL_IMPLS are silently dropped rather than rejected, so they
never became USE flags and never produced a dependency atom -- an ebuild
declaring python3_ was only ever offering 12, 13 and 14.

Two reasons to clean them up anyway. The eclass branch they land in
dies for any EAPI other than 7 or 8, so all 139 are waiting to break on
the first EAPI 9 bump. And pkgcheck's OldPythonCompat is a git-scoped
check that only inspects ebuilds touched in a commit range, so it never
reports these on a full-tree scan and would instead fire on whichever
unrelated commit next touches one of them.

No revbump, because nothing observable changes. Verified on a sample
spanning every transformation shape that the python USE flags portage
offers are byte-identical before and after, and that a pkgcheck run
over all 65 packages returns a set-identical list of findings.

The eight python2_7 ebuilds are deliberately untouched: that target is
also absent from _PYTHON_ALL_IMPLS, but they reach it through the
_PYTHON_ALLOW_PY27 hatch in the overlay's forked python-utils-r1_py2.

commit 244d6a91a3972e6786ab396c0754f97d02ae44f1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Jul 22 22:51:06 2026 +0200

sci-ml/torchvision: add 0.28.0

Forked from ::gentoo for the torch 2.13 stack. Divergences from ::gentoo's
0.27.0:

- Bump the pytorch pin 2.12* -> 2.13* (torchvision 0.28 requires torch 2.13).
- Drop the ffmpeg USE flag and the TORCHVISION_USE_FFMPEG /
TORCHVISION_USE_VIDEO_CODEC exports: ::gentoo carried them but neither
0.27.0 nor 0.28.0's setup.py reads those vars (no ffmpeg handling remains
in the build), so they were dead knobs.
- Refresh EPYTEST_DESELECT for 0.28.0: drop the four
TestResize::test_bounding_boxes_correctness[*-XYWHR] deselects (they pass
now), and narrow the gif deselect to the two test_decode_gif[*-earth]
params (earth.gif is a git-LFS asset the archive tarball omits; the other
gifs decode fine) rather than the whole test_decode_gif.
- Fix the DESCRIPTION grammar.

Register it for version tracking in scripts/nvchecker/nvchecker.toml.