ffmpeg-wasi-build¶
ffmpeg-wasi build toolchain: wasi-sdk cross-compiler + native gcc/nasm/cmake
Release v0.1.4, 2026-11-19. Built on nothing in this group. Source: phpboyscout/images/ffmpeg-wasi-build.
Pin it¶
Pin the full version. Each release also moves its minor tag and latest, which change under you.
| Variant | Tags |
|---|---|
| default | v0.1.4, v0.1 |
About this image¶
The build toolchain for ffmpeg-wasi,
carrying both of its compile targets in one image.
What's baked in¶
| Area | Tools |
|---|---|
| wasm target | wasi-sdk 30 at /opt/wasi-sdk — clang cross-compiler, wasm-ld, llvm-ar/nm/ranlib, and the wasi-sysroot |
| native target | build-essential (gcc/g++), nasm, cmake, zlib1g-dev |
| shared | git, make, pkg-config, meson, ninja-build, curl, xz-utils, gzip, ca-certificates |
WASI_SDK=/opt/wasi-sdk is set, which is where ffmpeg-wasi's
build/toolchain.sh looks by default, so neither side has to agree with the
other by hand.
Why it exists¶
The ten jobs in ffmpeg-wasi's build matrix used to apt-get install this same
fixed package list on every run. That put an uncacheable 32.6 MB index download
on the critical path of a matrix serialised behind one resource_group, over a
link nobody here controls.
On 2026-09-11 archive.ubuntu.com degraded to ~20s time-to-first-byte and
~50 kB/s — against 15.6 MB/s three days earlier on the same runner, while GitHub
served 6.7 MB/s at the same moment. Every job retried its way through failed
fetches; one came back with a Packages.gz of the right length and the wrong
hash. apt stopped, git was never installed, and the job failed 75 seconds
later inside a zlib clone, blaming GitHub for an Ubuntu mirror. Three good
builds were discarded with it and every downstream job skipped.
Baking the list here means apt runs when this image is released, not ten times per pipeline.
Why Debian, not Wolfi¶
Every other image in this group derives from the Wolfi-based ci-base. This one
cannot, because the native driver it builds is shipped, dynamically linked,
and its glibc floor is a compatibility contract:
bookworm ships glibc 2.36, which holds that ceiling at 2.35 and lets the published driver run on Ubuntu 22.04, Debian 12 and RHEL 9. Wolfi tracks a much newer glibc, so rebuilding there would raise the floor — and consumers would discover it, not CI.
Wolfi is perfectly capable of building this (it packages wasi-sdk at exactly
30-r1, plus build-base, cmake, nasm and zlib-dev). Capability is not
the blocker; the contract is. Moving native to Wolfi means static linking or
deliberately re-declaring the floor, and neither is this image's call.
dev-tools and infra-tools are Debian here too, and soft-fail their trivy
scan for the same reason this one does.
Versions¶
- Base —
debian:bookworm-slim, pinned by tag, tracked by Renovate's dockerfile manager. - wasi-sdk — pinned by
ARG WASI_SDK_VERSIONwith a# renovate:annotation against the upstream GitHub releases, and verified againstARG WASI_SDK_SHA256. Upstream publishes no checksum file, so that digest was taken from the artefact itself; a mismatch fails the build rather than silently installing something else. - apt packages — deliberately unpinned. Debian keeps only current versions
in its index, so a pin breaks the image as soon as the archive moves, and
Renovate has no manager for
apt-get installlines. The base carries the pin instead — the same tradeci-baserecords for apk.
Bumping WASI_SDK_VERSION without also updating the digest will fail the build.
That is intended: the version is the compiler, and ffmpeg-wasi claims
reproducible builds.
Releases¶
releaser-pleaser maintains a Release MR on main; merging it cuts the
vX.Y.Z tag, which publishes :vX.Y.Z, :vX.Y and :latest. Commits to
main publish :sha-<short> and :latest.
Everything installed¶
Contents not yet published. This image's releases do not carry an SBOM yet. Until they do, the README above describes what it adds, and the Dockerfile is the full record.