Speed up ARM production release builds

[matteius]

Summary

• run the Docker linux/arm64 build on GitHub's native ubuntu-24.04-arm runner
• cross-compile linux/arm/v7 on x86_64 with Debian armhf multiarch packages and a native LiteRT flatc host tool
• leave QEMU only for ARMv7 runtime-image assembly, not C/C++/Go/web compilation
• persist a separate GHCR BuildKit cache for each platform
• reduce the build context with .dockerignore, omit recommended build-only packages, and skip compiling the unused test target in production images

Why

In release run 30839856661, the jobs took:

• linux/amd64: 15m44s
• linux/arm/v7: 3h42m48s
• linux/arm64: 4h58m47s

The ARM jobs were compiling the complete LiteRT/XNNPACK dependency tree under QEMU. In the ARMv7 log, the final LightNVR build layer alone took about 2h55m. Docker documents compilation as a case where QEMU can be much slower than native execution and recommends native nodes or cross-compilation.

The existing Debian release matrix already demonstrates the native ARM64 runner path: its ARM64 jobs finish in roughly 10–13 minutes, while emulated ARMv7 jobs take about two hours.

Validation

• cold ARMv7 builder: 9m12s
  • docker buildx build --platform linux/arm/v7 --target builder --output type=cacheonly .
  • verified /bin/lightnvr and /bin/go2rtc are 32-bit ARM EABI5 binaries
• complete native amd64 runtime image: 13m03s
  • docker buildx build --platform linux/amd64 --output type=cacheonly .
• trixie/armhf package path passed multiarch package installation and CMake dependency discovery; the redundant cold TensorFlow checkout was canceled after GitHub network transfer stalled
• docker buildx build --check for amd64, arm64, and arm/v7
• actionlint .github/workflows/docker-publish.yml
• workflow YAML parse, bash -n scripts/build.sh, and git diff --check

The first release still has to populate the new registry caches, but ARMv7 compilation no longer runs under emulation. Subsequent releases can also reuse unchanged dependency layers.

[Copilot]

Pull request overview

This PR reworks the production Docker release pipeline to significantly reduce ARM build times by switching ARM64 builds to a native ARM runner and cross-compiling ARMv7 on amd64 (leaving QEMU primarily for target runtime-layer assembly), while also improving layer/cache reuse and reducing Docker build context size.

Changes:

• Updates the Docker publish workflow to use a platform→runner matrix (native ARM64 runner) and per-platform BuildKit registry caches.
• Adjusts the Dockerfile to support ARMv7 cross-compilation via Debian armhf multiarch packages + an ARMv7 CMake toolchain, and disables production test builds.
• Adds .dockerignore and documents the updated release build approach.

Reviewed changes

Copilot reviewed 6 out of 7 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
scripts/build.sh	Adds optional CMAKE_TOOLCHAIN_FILE forwarding into the CMake configure step.
docs/RELEASE_PROCESS.md	Documents native ARM64 runner usage, ARMv7 cross-compilation, and per-platform cache behavior.
Dockerfile	Implements cross-build logic for ARMv7, uses $BUILDPLATFORM for the builder stage, and adjusts build steps/env for multiarch.
cmake/toolchains/armv7-linux-gnueabihf.cmake	Introduces an ARMv7 hard-float/NEON toolchain file for cross-compiling via Debian multiarch.
.gitignore	Ensures the new toolchain .cmake files under cmake/toolchains/ aren’t ignored.
.github/workflows/docker-publish.yml	Moves to a matrix with platform-specific runners/QEMU and adds per-platform registry cache import/export.
.dockerignore	Shrinks Docker build context by excluding VCS/CI/dev artifacts and other large directories.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated 2 comments.

Suppressed comments (2)

Dockerfile:216

• This case "$TARGETARCH/$TARGETVARIANT" only matches when TARGETVARIANT is empty. If the builder provides a variant like arm64/v8, this will incorrectly fail with Unsupported target architecture while building go2rtc.

      amd64/) GOARCH=amd64; GOARM= ;; \
      arm64/) GOARCH=arm64; GOARM= ;; \
      arm/v7) GOARCH=arm; GOARM=7 ;; \
      *) echo "Unsupported target architecture: $TARGETARCH/$TARGETVARIANT"; exit 1 ;; \

Dockerfile:268

• These patterns assume TARGETVARIANT is empty for amd64/arm64 (matching amd64/ and arm64/). If a non-empty variant is provided (e.g. arm64/v8), the build will fail as an unsupported architecture when selecting pkg-config paths.

        amd64/) PKG_CONFIG_ARCH_PATH="/usr/lib/x86_64-linux-gnu/pkgconfig"; TOOLCHAIN_FILE="" ;; \
        arm64/) PKG_CONFIG_ARCH_PATH="/usr/lib/aarch64-linux-gnu/pkgconfig"; TOOLCHAIN_FILE="" ;; \