ci: add probe to debug Windows CI
CI / Build & test (Windows) (push) Failing after 27s

This commit is contained in:
2026-08-22 14:35:00 +08:00
parent 2721225a3f
commit e8d08c77f0
+168 -162
View File
@@ -16,173 +16,179 @@ permissions:
contents: read
jobs:
linux:
name: Build & test (Linux)
runs-on: oak-ubuntu-2404
steps:
- name: Checkout
uses: actions/checkout@v4
with:
# gpui/ is a git submodule; its crates are workspace members of
# their own repo and build as path dependencies of oakapp.
submodules: true
# ------------------------------------------------------------------
# System dependencies
# ------------------------------------------------------------------
- name: Install system dependencies
run: |
# Codec/filter libraries for the ffmpeg-sys-next build feature
# (see tooling/install-deps.sh).
tooling/install-deps.sh
# cmake/make for the vendored OpenColorIO build (ocio-sys
# `bundled`; Ubuntu's libopencolorio-dev is 2.1, older than the
# bridge's API floor) plus the headless test infra gpui needs:
# X11, software Mesa Vulkan (lavapipe) and xvfb.
sudo apt-get install -y \
cmake \
libpipewire-0.3-dev libspa-0.2-dev libjack-jackd2-dev \
libasound2-dev libpulse-dev libsndfile1-dev \
libgl1-mesa-dev libgl1-mesa-dri mesa-vulkan-drivers \
libvulkan-dev libxkbcommon-dev libxkbcommon-x11-dev xvfb build-essential
# ------------------------------------------------------------------
# Build environment
# ------------------------------------------------------------------
# ocio-sys builds a stub bridge unless these are set; the oakcommon
# ocioutils tests need the real library (see crates/oakcommon/.cargo/
# config.toml, which only applies to builds run from that directory).
# ocio-sys builds its vendored OpenColorIO from source on Linux (the
# `bundled` feature; the distro package is too old for the bridge),
# so no OCIO_INSTALL_DIR here.
- name: Configure build environment
run: |
echo "OCIO_RS_ENABLE_REAL=1" >> "$GITHUB_ENV"
echo "OCIO_RS_LINK=static" >> "$GITHUB_ENV"
echo "RUSTUP_HOME=/opt/rust/rustup" >> "$GITHUB_ENV"
echo "CARGO_HOME=/opt/rust/cargo" >> "$GITHUB_ENV"
echo "PATH=$PATH:/opt/rust/cargo/bin" >> "$GITHUB_ENV"
# ------------------------------------------------------------------
# Caches
# ------------------------------------------------------------------
# Covers the whole target/ dir plus ~/.cargo; shared across branches
# of the same OS.
# TEMP: cache disabled ntil the Gitea instance cache is provisioned (Cache cargo artifacts)
# uses: Swatinem/rust-cache@v2
# with:
# shared-key: oak-workspace
# cache-on-failure: true
# The project FFmpeg (release/8.0, static, all free codecs + hwaccel)
# is built by tooling/ffmpeg/build-ffmpeg.sh — 10-20 min on a cold
# cache. It does not depend on the Rust toolchain, so key it on the
# script itself and keep it out of rust-cache.
# TEMP: cache disabled until the Gitea instance cache is provisioned (Cache project FFmpeg)
# uses: actions/cache@v4
# with:
# path: .cache/ffmpeg
# key: ffmpeg-${{ runner.os }}-${{ hashFiles('tooling/ffmpeg/build-ffmpeg.sh') }}
# ------------------------------------------------------------------
# Project FFmpeg (script + FFMPEG_DIR; see docs/build.md)
# ------------------------------------------------------------------
- name: Build project FFmpeg
run: |
tooling/ffmpeg/build-ffmpeg.sh
echo "FFMPEG_DIR=$PWD/.cache/ffmpeg" >> "$GITHUB_ENV"
# ------------------------------------------------------------------
# Build & test
# ------------------------------------------------------------------
- name: Build
run: cargo build --workspace --locked
# xvfb + 24-bit screen: the gpui #[gpui::test] tests open real windows
# and render through wgpu on Mesa's software Vulkan (lavapipe).
# The watchdog bounds the step: a deadlocked test produces no output
# and no failure, so after 1500 s (a green run needs ~4 min) it dumps
# every hung process's thread stacks and kills the suite.
- name: Test
run: |
sudo apt-get install -y gdb
xvfb-run -a -s "-screen 0 1920x1080x24" cargo test --workspace --locked &
TEST_PID=$!
(
sleep 1500
echo "::warning::test suite exceeded 1500s; dumping hung-process stacks"
for p in $(pgrep -f 'target/debug/deps/|target/debug/oak-worker'); do
echo "===== thread stacks of pid $p ($(readlink /proc/$p/exe 2>/dev/null)) ====="
sudo gdb -batch -ex 'thread apply all bt' -p "$p" || true
done
pkill -9 -f 'target/debug/deps/' || true
pkill -9 -f 'target/debug/oak-worker' || true
) &
WATCHDOG_PID=$!
wait $TEST_PID
rc=$?
kill $WATCHDOG_PID 2>/dev/null || true
exit $rc
# A crashing (SIGSEGV) test gives no Rust backtrace; rerun the
# crashing test binaries under gdb to capture the native stack.
# `--args` is required — plain `--` makes gdb treat the test args as
# a core file. The extra probes target loader-stage crashes (the
# copier_test SIGSEGV happens inside ld.so's dl_main): si_addr/si_code
# pin down the fault type, the dynsym dump exposes symbols the
# executable exports for interposition, strace shows the last loader
# syscalls, and valgrind catches a corrupting static initializer.
- name: Backtrace on test failure
if: failure()
run: |
sudo apt-get install -y gdb strace valgrind
for name in node_e2e_test suites_test copier_test; do
BIN=$(ls -t target/debug/deps/$name-* | grep -v '\.d$' | head -1)
[ -n "$BIN" ] || continue
echo "===== $BIN ====="
file "$BIN" || true
echo "--- exported defined dynsyms:"
readelf --dyn-syms -W "$BIN" 2>/dev/null | grep -v ' UND ' | tail -n +4 | head -30 || true
echo "--- strace tail:"
strace -f "$BIN" --list 2>&1 | tail -15 || true
echo "--- valgrind tail:"
valgrind -q "$BIN" --list 2>&1 | tail -25 || true
echo "--- gdb:"
xvfb-run -a gdb -batch \
-ex run \
-ex 'bt' \
-ex 'p $_siginfo.si_code' \
-ex 'p/x $_siginfo._sifields._sigfault.si_addr' \
-ex 'x/6i $rip' \
--args "$BIN" --nocapture || true
done
# ------------------------------------------------------------------
# OFX plugin discovery end-to-end
# ------------------------------------------------------------------
# Build a minimal but real OFX plugin into a .ofx.bundle, point
# OFX_PLUGIN_PATH at it and let the scan_probe example run the full
# host path (directory scan -> dlopen -> setHost -> load -> describe
# -> register). The assertion is the plugin's registration line; CI
# machines have no system-wide OFX plugins, so the fixture is the
# only discovery.
- name: Build OFX fixture plugin
run: crates/oakplugin/tests/fixtures/build_fixture.sh .cache/ofx-fixture
- name: Probe OFX plugin discovery
run: |
OFX_PLUGIN_PATH="$PWD/.cache/ofx-fixture" \
cargo run --locked -p oakplugin --example scan_probe > probe.log 2>&1
grep -q 'type_id=rs.oak.CiTestPlugin' probe.log
# A project carrying a plugin node must survive save/load (the
# serializer resolves plugin types via the dynamic factory).
OAK_OFX_FIXTURE_DIR="$PWD/.cache/ofx-fixtre" \
cargo test --locked -p oakplugin --test ofx_roundtrip
# linux:
# name: Build & test (Linux)
# runs-on: oak-ubuntu-2404
# steps:
# - name: Checkout
# uses: actions/checkout@v4
# with:
# # gpui/ is a git submodule; its crates are workspace members of
# # their own repo and build as path dependencies of oakapp.
# submodules: true
#
# # ------------------------------------------------------------------
# # System dependencies
# # ------------------------------------------------------------------
# - name: Install system dependencies
# run: |
# # Codec/filter libraries for the ffmpeg-sys-next build feature
# # (see tooling/install-deps.sh).
# tooling/install-deps.sh
# # cmake/make for the vendored OpenColorIO build (ocio-sys
# # `bundled`; Ubuntu's libopencolorio-dev is 2.1, older than the
# # bridge's API floor) plus the headless test infra gpui needs:
# # X11, software Mesa Vulkan (lavapipe) and xvfb.
# sudo apt-get install -y \
# cmake \
# libpipewire-0.3-dev libspa-0.2-dev libjack-jackd2-dev \
# libasound2-dev libpulse-dev libsndfile1-dev \
# libgl1-mesa-dev libgl1-mesa-dri mesa-vulkan-drivers \
# libvulkan-dev libxkbcommon-dev libxkbcommon-x11-dev xvfb build-essential
#
# # ------------------------------------------------------------------
# # Build environment
# # ------------------------------------------------------------------
# # ocio-sys builds a stub bridge unless these are set; the oakcommon
# # ocioutils tests need the real library (see crates/oakcommon/.cargo/
# # config.toml, which only applies to builds run from that directory).
# # ocio-sys builds its vendored OpenColorIO from source on Linux (the
# # `bundled` feature; the distro package is too old for the bridge),
# # so no OCIO_INSTALL_DIR here.
# - name: Configure build environment
# run: |
# echo "OCIO_RS_ENABLE_REAL=1" >> "$GITHUB_ENV"
# echo "OCIO_RS_LINK=static" >> "$GITHUB_ENV"
# echo "RUSTUP_HOME=/opt/rust/rustup" >> "$GITHUB_ENV"
# echo "CARGO_HOME=/opt/rust/cargo" >> "$GITHUB_ENV"
# echo "PATH=$PATH:/opt/rust/cargo/bin" >> "$GITHUB_ENV"
#
# # ------------------------------------------------------------------
# # Caches
# # ------------------------------------------------------------------
# # Covers the whole target/ dir plus ~/.cargo; shared across branches
# # of the same OS.
# # TEMP: cache disabled ntil the Gitea instance cache is provisioned (Cache cargo artifacts)
# # uses: Swatinem/rust-cache@v2
# # with:
# # shared-key: oak-workspace
# # cache-on-failure: true
#
# # The project FFmpeg (release/8.0, static, all free codecs + hwaccel)
# # is built by tooling/ffmpeg/build-ffmpeg.sh — 10-20 min on a cold
# # cache. It does not depend on the Rust toolchain, so key it on the
# # script itself and keep it out of rust-cache.
# # TEMP: cache disabled until the Gitea instance cache is provisioned (Cache project FFmpeg)
# # uses: actions/cache@v4
# # with:
# # path: .cache/ffmpeg
# # key: ffmpeg-${{ runner.os }}-${{ hashFiles('tooling/ffmpeg/build-ffmpeg.sh') }}
#
# # ------------------------------------------------------------------
# # Project FFmpeg (script + FFMPEG_DIR; see docs/build.md)
# # ------------------------------------------------------------------
# - name: Build project FFmpeg
# run: |
# tooling/ffmpeg/build-ffmpeg.sh
# echo "FFMPEG_DIR=$PWD/.cache/ffmpeg" >> "$GITHUB_ENV"
#
# # ------------------------------------------------------------------
# # Build & test
# # ------------------------------------------------------------------
# - name: Build
# run: cargo build --workspace --locked
#
# # xvfb + 24-bit screen: the gpui #[gpui::test] tests open real windows
# # and render through wgpu on Mesa's software Vulkan (lavapipe).
# # The watchdog bounds the step: a deadlocked test produces no output
# # and no failure, so after 1500 s (a green run needs ~4 min) it dumps
# # every hung process's thread stacks and kills the suite.
# - name: Test
# run: |
# sudo apt-get install -y gdb
# xvfb-run -a -s "-screen 0 1920x1080x24" cargo test --workspace --locked &
# TEST_PID=$!
# (
# sleep 1500
# echo "::warning::test suite exceeded 1500s; dumping hung-process stacks"
# for p in $(pgrep -f 'target/debug/deps/|target/debug/oak-worker'); do
# echo "===== thread stacks of pid $p ($(readlink /proc/$p/exe 2>/dev/null)) ====="
# sudo gdb -batch -ex 'thread apply all bt' -p "$p" || true
# done
# pkill -9 -f 'target/debug/deps/' || true
# pkill -9 -f 'target/debug/oak-worker' || true
# ) &
# WATCHDOG_PID=$!
# wait $TEST_PID
# rc=$?
# kill $WATCHDOG_PID 2>/dev/null || true
# exit $rc
#
# # A crashing (SIGSEGV) test gives no Rust backtrace; rerun the
# # crashing test binaries under gdb to capture the native stack.
# # `--args` is required — plain `--` makes gdb treat the test args as
# # a core file. The extra probes target loader-stage crashes (the
# # copier_test SIGSEGV happens inside ld.so's dl_main): si_addr/si_code
# # pin down the fault type, the dynsym dump exposes symbols the
# # executable exports for interposition, strace shows the last loader
# # syscalls, and valgrind catches a corrupting static initializer.
# - name: Backtrace on test failure
# if: failure()
# run: |
# sudo apt-get install -y gdb strace valgrind
# for name in node_e2e_test suites_test copier_test; do
# BIN=$(ls -t target/debug/deps/$name-* | grep -v '\.d$' | head -1)
# [ -n "$BIN" ] || continue
# echo "===== $BIN ====="
# file "$BIN" || true
# echo "--- exported defined dynsyms:"
# readelf --dyn-syms -W "$BIN" 2>/dev/null | grep -v ' UND ' | tail -n +4 | head -30 || true
# echo "--- strace tail:"
# strace -f "$BIN" --list 2>&1 | tail -15 || true
# echo "--- valgrind tail:"
# valgrind -q "$BIN" --list 2>&1 | tail -25 || true
# echo "--- gdb:"
# xvfb-run -a gdb -batch \
# -ex run \
# -ex 'bt' \
# -ex 'p $_siginfo.si_code' \
# -ex 'p/x $_siginfo._sifields._sigfault.si_addr' \
# -ex 'x/6i $rip' \
# --args "$BIN" --nocapture || true
# done
#
# # ------------------------------------------------------------------
# # OFX plugin discovery end-to-end
# # ------------------------------------------------------------------
# # Build a minimal but real OFX plugin into a .ofx.bundle, point
# # OFX_PLUGIN_PATH at it and let the scan_probe example run the full
# # host path (directory scan -> dlopen -> setHost -> load -> describe
# # -> register). The assertion is the plugin's registration line; CI
# # machines have no system-wide OFX plugins, so the fixture is the
# # only discovery.
# - name: Build OFX fixture plugin
# run: crates/oakplugin/tests/fixtures/build_fixture.sh .cache/ofx-fixture
#
# - name: Probe OFX plugin discovery
# run: |
# OFX_PLUGIN_PATH="$PWD/.cache/ofx-fixture" \
# cargo run --locked -p oakplugin --example scan_probe > probe.log 2>&1
# grep -q 'type_id=rs.oak.CiTestPlugin' probe.log
# # A project carrying a plugin node must survive save/load (the
# # serializer resolves plugin types via the dynamic factory).
# OAK_OFX_FIXTURE_DIR="$PWD/.cache/ofx-fixtre" \
# cargo test --locked -p oakplugin --test ofx_roundtrip
windows:
name: Build & test (Windows)
runs-on: oak-windows-2025
steps:
- name: probe
shell: cmd
run: |
echo PATH=%PATH%
where node
node --version
- name: debug-env
shell: powershell # Windows Server 镜像原生自带,不依赖你装的任何东西
run: |