commit 0c499b64fbde904494aaa74ad2e0d55c2b3e2624
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 23 13:14:52 2026 +0200

dev-python/conch-triton-kernels: depend on virtual/triton

commit c754122827baa1f0c836c3c931bda9b217dab1ed
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Aug 17 01:29:52 2026 +0200

dev-python/conch-triton-kernels: fix caffe2 USE-dep for single-impl

sci-ml/caffe2 is a python-single-r1 package, so it exposes only
python_single_target_* flags. Referencing it as caffe2[$]
expands to python_targets_* constraints caffe2 never provides; the (-)
fallback then resolves those flags to disabled while the atom requires
them enabled, making the dependency unsatisfiable and blocking emerge at
resolution. pkgcheck cannot catch this because the (-) suppresses the
warning. Use $, matching every other caffe2
consumer in the overlay.

commit 96b330e033f9c088daf57737299d8f86e0915691
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 16 12:42:05 2026 +0200

dev-python/conch-triton-kernels: fix Python dependency USE propagation

python_gen_cond_dep expands a multi-implementation dependency template, so
PYTHON_SINGLE_USEDEP produces an invalid atom. Use PYTHON_USEDEP to match the
distutils-r1 package model.

commit 04ba78359e13ea4a623c4f795eb971233c03ee80
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 16 11:01:30 2026 +0200

dev-python/conch-triton-kernels: bridge the single-impl caffe2 dependency

This ebuild is multi-impl (no DISTUTILS_SINGLE_IMPL), so sci-
ml/caffe2[$] demanded python_targets_* flags that caffe2 does not
carry -- caffe2 is python-single-r1 and exposes python_single_target_* under a ^^
constraint, making the dependency unsatisfiable. Wrap it in python_gen_cond_dep with
$ so each enabled implementation requires the matching single-
target caffe2. The package must stay multi-impl because vllm's rocm block references
conch-triton-kernels[$].

commit a116d50f6a9164ff1a3d94c8fe60dc49c4b8e167
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Fri Aug 14 16:57:59 2026 +0200

dev-python/conch-triton-kernels: complete accelerator dependencies

The exact 1.3 source imports Torch throughout its runtime, requires Triton 3.3 or
newer for accelerator extras, and declares wheel plus setuptools 42 as build
requirements. Express those requirements directly rather than relying on incidental
installations.

commit abf529ad3070e5b05296989322b7613202faccef
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jun 14 11:57:40 2026 +0200

dev-python/conch-triton-kernels: declare triton-bin dep

conch's kernels are Triton kernels accessed lazily; the package shipped
only numpy because no Triton existed in any tree we depend on (with a
comment saying so). dev-python/triton-bin (mainline, carrying both the
nvidia and amd backends) is now in-tree, so declare it directly rather
than leave it to the consumer. Covers 1.2.1 and 1.3.

commit 120ba8931d2064268be381d6953f53548a6892d1
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed May 13 14:34:35 2026 +0200

dev-python/conch-triton-kernels: disable py3.11

commit 3c73cd86a4b202903ec79586ccbcd9a1445fc7e5
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun May 10 14:57:55 2026 +0200

dev-python/conch-triton-kernels: add 1.3

commit a4e77a59a32ea691b60706736192494e9a59c612
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu May 7 01:01:06 2026 +0200

dev-python/conch-triton-kernels: new package

Stack AV's pure-Python Triton kernel collection. Required by
dev-python/vllm in the rocm path (vllm pulls it via rocm.txt). Hard
RDEPEND is just numpy; torch/triton are upstream extras and accessed
lazily inside kernel functions.