CI / Build + Clippy + Test (pull_request) Failing after 1m13s
CI / Security audit (RUSTSEC) (pull_request) Successful in 1m17s
apt-get ignores HTTP_PROXY/HTTPS_PROXY env vars and tries to connect
directly to archive.ubuntu.com. Clash DNS returns fake-ip (198.18.x.x)
which only works through TUN (disabled). Fix: write apt proxy config to
/etc/apt/apt.conf.d/99proxy before apt-get update.
CI / Build + Clippy + Test (pull_request) Failing after 1m13s
CI / Security audit (RUSTSEC) (pull_request) Successful in 1m18s
act_runner does not support container.env in config.yaml — job containers
get NO proxy env vars. Set HTTP_PROXY/HTTPS_PROXY/NO_PROXY directly in
the workflow's job-level env block.
NO_PROXY includes gitea.com (actions checkout goes direct, faster) and
gitea.dailz.cn (self-hosted Gitea, internal).
CI / Build + Clippy + Test (pull_request) Failing after 2m17s
CI / Security audit (RUSTSEC) (pull_request) Failing after 2m15s
Network testing revealed:
- static.rust-lang.org: direct 200/0.24s (fast, no proxy needed)
- sh.rustup.rs: direct 200/0.11s (fast)
- gitea.com: direct 303/0.68s (fast)
- mirrors.ustc.edu.cn: UNREACHABLE (5s timeout)
- rsproxy.cn: UNREACHABLE (5s timeout)
The machine has good direct internet access to official Rust/crates servers.
Domestic mirrors are the ones that are blocked. Remove all mirror config and
use defaults.
CI / Security audit (RUSTSEC) (pull_request) Failing after 2m15s
CI / Build + Clippy + Test (pull_request) Failing after 31m46s
rsproxy.cn/rustup returns 404 for /dist/x86_64-unknown-linux-gnu/
rustup-init — different path structure from static.rust-lang.org.
Switch to mirrors.ustc.edu.cn/rust-static which is a full mirror with
identical path structure.
CI / Build + Clippy + Test (pull_request) Failing after 5s
CI / Security audit (RUSTSEC) (pull_request) Failing after 14m4s
rustup components (~60MB) from static.rust-lang.org are slow even
through proxy. Switch to rsproxy.cn/rustup mirror (domestic, direct
connection via NO_PROXY).
CI / Security audit (RUSTSEC) (pull_request) Failing after 3s
CI / Build + Clippy + Test (pull_request) Failing after 44m45s
The catthehacker/ubuntu:act-latest image does not have rustup in PATH
('rustup: command not found', exit 127). Replace the direct rustup
invocation with the official install script from sh.rustup.rs, which
installs rustup + the stable toolchain in one step.
Also adds ~/.cargo/bin to GITHUB_PATH so subsequent steps (cargo clippy,
cargo build, cargo test, cargo audit) can find cargo/rustc.
CI / Build + Clippy + Test (pull_request) Failing after 7s
CI / Security audit (RUSTSEC) (pull_request) Failing after 3s
The actions/cache@v4 clone from gitea.com has been hanging for 12+
minutes in run 17. The actions/checkout clone took 2.5 min (slow but
completed), but actions/cache is stuck indefinitely.
Removing the cache step entirely. Trade-off: CI recompiles from scratch
each run (slower), but actually progresses past the action-clone phase.
Can re-add once gitea.com access is faster or actions are pre-cached
on the runner.
CI / Build + Clippy + Test (pull_request) Failing after 14m11s
CI / Security audit (RUSTSEC) (pull_request) Failing after 4s
The act_runner network also blocks crates.io index access (both sparse
and git protocols). Previous workaround used git protocol to github.com,
which is also blocked. Replace with rsproxy.cn sparse mirror, accessible
from China networks.
Added to both build-test and audit jobs. Config is written to
~/.cargo/config.toml at runtime (CI-only; does not affect local dev).
CI / Security audit (RUSTSEC) (pull_request) Has been cancelled
CI / Build + Clippy + Test (pull_request) Has been cancelled
The self-hosted act_runner cannot reach github.com (network timeout
on actions/checkout clone). Replace:
- actions/checkout@v4 -> https://gitea.com/actions/checkout@v4 (3 sites)
- actions/cache@v4 -> https://gitea.com/actions/cache@v4
- dtolnay/rust-toolchain@stable -> rustup toolchain install (inline run)
gitea.com maintains official mirrors of the actions/* org. dtolnay's
rust-toolchain is third-party (no gitea.com mirror), so replaced with
a direct rustup invocation — the act_runner ubuntu image has rustup
pre-installed.
This unblocks CI which has been red since the original PR #26 was
opened 6 weeks ago. No code changes.
CI / Security audit (RUSTSEC) (push) Failing after 1m31s
CI / Build + Clippy + Test (push) Failing after 4m18s
Second LIBCLANG_PATH failure mode: my 'libclang.so.*' with -type f
pattern returned nothing because Debian Bookworm's runtime library is
'libclang-14.so.1' (versioned, with no plain libclang.so.* symlink),
and the .so symlink is itself a symlink not a regular file (-type f
excludes it).
bindgen accepts any of: libclang.so, libclang-*.so, libclang.so.*,
libclang-*.so.* — so the broader 'libclang*.so*' (no -type filter)
catches every variant. Also broadened search root from /usr/lib to
/usr to cover both /usr/lib/x86_64-linux-gnu/ (runtime lib) and
/usr/lib/llvm-*/lib/ (dev symlink).
Log line now includes the actual matched path so future debugging
is one glance.
CI / Build + Clippy + Test (push) Failing after 20m22s
CI / Security audit (RUSTSEC) (push) Failing after 1m31s
First real CI run on the Gitea Actions runner hit the predicted
LIBCLANG_PATH issue but for an unexpected reason: workflow-level
'env:' does not reliably propagate into act_runner's Docker executor
(bindgen received empty LIBCLANG_PATH despite /usr/lib/llvm-*/lib
being correct on the runner).
Two changes:
1. Drop the hardcoded workflow-level 'env: LIBCLANG_PATH' and add a
dedicated 'Resolve LIBCLANG_PATH' step that does:
find /usr/lib -name 'libclang.so.*' | head -1 | xargs dirname
and writes the result to $GITHUB_ENV. Dynamic discovery also
future-proofs against Debian/Ubuntu version drift (llvm-14 today,
llvm-18 tomorrow). The $GITHUB_ENV mechanism is reliably visible
across step boundaries inside the act_runner Docker container,
whereas workflow-level env: is not.
2. apt install: drop '--no-install-recommends' on libclang-dev
(Debian Bookworm's metapackage uses Recommends to pull in
versioned toolchain bits bindgen needs). Also install 'clang'
(not just llvm-14) to get version-agnostic libclang shared lib.
Comments inline in ci.yml document both decisions to prevent future
regression.
The 'Resolve LIBCLANG_PATH' step fails fast with a clear message if
libclang isn't installed, instead of letting the clippy/build steps
fail cryptically later.
Oracle P2 plan step 1+2+missed-fields. Locks in the audit cleanup so
future PRs can't regress the 0-errors / deny-unsafe / 79-tests baseline.
- .github/workflows/ci.yml: two jobs on ubuntu-latest (Linux only —
project is Wayland/VAAPI-specific, no macOS/Windows story).
* build-test: installs ffmpeg + libavcodec/libavformat/libavutil/
libswscale/libva dev + libwayland + libdrm + libpipewire-0.3-dev
+ libclang-dev/llvm-14 (LIBCLANG_PATH pinned); caches cargo +
target; runs clippy -> build --release -> test --release.
Release build before tests is mandatory because
tests/integration_test.rs shells out to target/release/wl-webrtc.
* audit: installs cargo-audit and runs 'cargo audit --deny warnings'
as a separate job so a RUSTSEC advisory fails the build
independently of compile state.
No -D warnings on clippy yet — undocumented_unsafe_blocks is already
deny via Cargo.toml; remaining warnings are advisory and can be
tightened later.
- Cargo.toml: pin rust-version = '1.70' to match README's claim.
Without this, cargo builds silently on older toolchains and surfaces
errors as cryptic parse failures instead of a clean version-mismatch
message. Oracle flagged this as a missing field during P2 review.
License field intentionally omitted — repo has no LICENSE file and no
publication plan yet. Add when publication becomes a goal.
Verified locally: YAML parses, cargo build --release Finished in 9.82s,
cargo test --release 79 passed, cargo clippy 0 errors.
dailz
dailz