spaceship/tests/_run_inside.sh
2026-08-12 22:47:46 +02:00

116 lines
4.2 KiB
Shell

#!/bin/bash
# Runs INSIDE the proot-distro Ubuntu container - do not call this directly,
# call tests/run_tests.sh from Termux instead.
#
# It exists as its own file so the whole suite runs in ONE `proot-distro
# login`: entering the container costs ~0.5s, which used to be paid 38 times
# over. Everything below is the container-side half of the runner.
#
# Args: <project_dir> <jobs> <verbose> <test_file_name>...
set -uo pipefail
PROJECT_DIR="$1"; shift
JOBS="$1"; shift
VERBOSE="$1"; shift
# A broken test can otherwise leave SceneTree alive after an initialization
# error, printing the same error every frame forever. Gameplay tests normally
# complete in seconds, even on a busy phone, so make that failure explicit
# rather than making the suite look hung. Override for deliberate long runs.
TEST_TIMEOUT_SECONDS="${TEST_TIMEOUT_SECONDS:-120}"
GODOT_BIN="/root/Godot_v4.7.1-stable_linux.arm64"
cd "$PROJECT_DIR" || exit 2
tests=("$@")
total=${#tests[@]}
if (( total == 0 )); then
echo "no tests matched"
exit 2
fi
outdir="$(mktemp -d)"
trap 'rm -rf "$outdir"' EXIT
REPORT_DIR="${TEST_REPORT_DIR:-$PROJECT_DIR/.test-results}"
mkdir -p "$REPORT_DIR"
# Import once, serially, before anything runs in parallel. Godot writes to
# .godot/ while importing; several instances discovering a cold cache at the
# same moment would race on it. After this the workers only read it.
"$GODOT_BIN" --headless --import --quit >/dev/null 2>&1
run_one() {
local name="$1"
local log="$outdir/$name.log"
local start end
start=$(date +%s%N)
# --fixed-fps pins delta to exactly 1/60 instead of measuring the wall
# clock. Without it a test's `elapsed` counts REAL seconds, so running
# several at once stretches every frame and the gameplay under test sees
# deltas it never sees in the game: ships overshoot their braking curve,
# settle late, and tests time out for no reason but load. That made
# test_camera_settles and test_planet_prompt_map_interaction fail on 2 of
# 3 parallel runs. Pinned, a test simulates the same 60fps whatever else
# the phone is doing, and its result no longer depends on how busy the
# machine was.
timeout "$TEST_TIMEOUT_SECONDS" "$GODOT_BIN" --headless --fixed-fps 60 \
-s "res://tests/$name" >"$log" 2>&1
local rc=$?
end=$(date +%s%N)
local ms=$(( (end - start) / 1000000 ))
# Two independent criteria, because either one alone can lie: a test that
# crashes before reaching its verdict exits non-zero without ever printing
# FAIL, and a hypothetical `quit(0)` on a failing path would exit clean
# while the log says otherwise.
if (( rc == 0 )) && grep -q ": PASS" "$log"; then
printf 'PASS %6dms %s\n' "$ms" "$name" > "$outdir/$name.status"
elif (( rc == 124 )); then
printf 'TIMEOUT %4dms %s (limit %ss)\n' \
"$ms" "$name" "$TEST_TIMEOUT_SECONDS" > "$outdir/$name.status"
elif (( rc == 137 || rc == 139 )); then
# Killed, not failed - almost always the kernel reclaiming memory. Worth
# saying plainly: a killed worker and a failing one both simply produced
# no verdict before, so a whole suite dying to the OOM killer was
# indistinguishable from a hang.
printf 'KILLED %4dms %s (signal %d - out of memory? try JOBS=1)\n' \
"$ms" "$name" "$(( rc - 128 ))" > "$outdir/$name.status"
else
printf 'FAIL %6dms %s (exit %d)\n' "$ms" "$name" "$rc" > "$outdir/$name.status"
fi
cat "$outdir/$name.status"
}
running=0
for name in "${tests[@]}"; do
run_one "$name" &
running=$(( running + 1 ))
if (( running >= JOBS )); then
wait -n
running=$(( running - 1 ))
fi
done
wait
failed=0
for name in "${tests[@]}"; do
if ! grep -q '^PASS' "$outdir/$name.status" 2>/dev/null; then
failed=$(( failed + 1 ))
echo
echo "=== $name FAILED, full output ==="
cat "$outdir/$name.log" 2>/dev/null
elif [[ "$VERBOSE" == "1" ]]; then
echo
echo "=== $name ==="
cat "$outdir/$name.log" 2>/dev/null
fi
done
echo
echo "$(( total - failed ))/$total passed"
# Keep the concise verdicts after the temporary logs are removed. This is
# especially useful from Termux, where a long-running child can outlive the
# terminal command that launched it and its final buffered output is lost.
rm -f "$REPORT_DIR"/*.status
cp "$outdir"/*.status "$REPORT_DIR"/
(( failed == 0 ))