[Midnightbsd-cvs] [MidnightBSD/mports] 7e5687: graphics/spirv-headers: add SPIR-V Registry header...

[Lucas Holt]

  Branch: refs/heads/master
  Home:   https://github.com/MidnightBSD/mports
  Commit: 7e5687d7b117750360236917ab15af13e0a115ca
      https://github.com/MidnightBSD/mports/commit/7e5687d7b117750360236917ab15af13e0a115ca
  Author: Lucas Holt <luke at foolishgames.com>
  Date:   2026-07-31 (Fri, 31 Jul 2026)

  Changed paths:
    M graphics/Makefile
    A graphics/spirv-headers/Makefile
    A graphics/spirv-headers/distinfo
    A graphics/spirv-headers/pkg-descr
    A graphics/spirv-headers/pkg-plist

  Log Message:
  -----------
  graphics/spirv-headers: add SPIR-V Registry headers port

Imported from FreeBSD graphics/spirv-headers. Header-only package providing
the SPIR-V headers, grammar JSON files, XML registry, and the
SPIRV-HeadersConfig.cmake used by find_package(SPIRV-Headers CONFIG).

Needed by ai/llama-cpp from b9426 onward: ggml's Vulkan backend now does
find_package(SPIRV-Headers CONFIG REQUIRED). No existing mport installs that
config file -- graphics/spirv-tools, graphics/shaderc and
devel/spirv-llvm-translator each vendor the headers privately via GH_TUPLE.

AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>


  Commit: d5bd1a557b752b834a16964c1ae26ebaacc807ea
      https://github.com/MidnightBSD/mports/commit/d5bd1a557b752b834a16964c1ae26ebaacc807ea
  Author: Lucas Holt <luke at foolishgames.com>
  Date:   2026-07-31 (Fri, 31 Jul 2026)

  Changed paths:
    M ai/llama-cpp/Makefile
    M ai/llama-cpp/distinfo
    A ai/llama-cpp/files/fetch-ui.sh
    M ai/llama-cpp/files/llama-server.in
    A ai/llama-cpp/files/package-lock.json
    M ai/llama-cpp/pkg-message
    M ai/llama-cpp/pkg-plist

  Log Message:
  -----------
  ai/llama-cpp: update 7134 -> 9426

Switch to the current upstream repository ggml-org/llama.cpp and drop the
kompute GH_TUPLE, whose ggml backend was removed upstream.

Upstream no longer commits the built WebUI assets, so follow the FreeBSD port
and build them from source: files/fetch-ui.sh runs npm ci during a custom
do-fetch and packs node_modules into a reproducible llama-ui-npm distfile,
which post-extract unpacks before running npm run build. The generated tarball
matches FreeBSD's checksum byte for byte. pipefail is guarded in the script
because /bin/sh on 4.0 (FreeBSD 13 base) does not support it.

Also set LLAMA_USE_PREBUILT_UI=OFF: without it a failed npm build silently
falls back to downloading UI assets from a Hugging Face bucket during the
build, and a missing WebUI is only a warning, not a build failure.

Unlike FreeBSD, keep ggml bundled rather than depending on a separate ggml
port -- that would require devel/llvm22, and mports only has up to llvm21.
This also keeps the VULKAN option, which FreeBSD moved into misc/ggml.

Remove the CURL option and the curl dependency. LLAMA_CURL is deprecated and
ignored upstream, model downloads moved to the vendored cpp-httplib, and
nothing in the resulting package links libcurl. Add USES=ssl instead, since
LLAMA_OPENSSL defaults to ON and gates HTTPS support in cpp-httplib.

Make USE_RC_SUBR unconditional: the EXAMPLES option now only gates examples/
and pocs/, while llama-server moved to tools/ and is always built.

Verified with clean builds and packages using the default options and with
EXAMPLES=OFF VULKAN=OFF. bmake test was not run.

AI-Assisted-by: Claude Opus 5 (1M context) <noreply at anthropic.com>
Signed-off-by: Lucas Holt <luke at foolishgames.com>


Compare: https://github.com/MidnightBSD/mports/compare/91323f15e8ce...d5bd1a557b75

To unsubscribe from these emails, change your notification settings at https://github.com/MidnightBSD/mports/settings/notifications