honey-biscuit-workshop/runner.sh

#!/bin/bash
set -euo pipefail

# HoneyBiscuitWorkshop System Test Runner
#
# Usage:
#   ./runner.sh                                  # baker tests (default)
#   ./runner.sh -c upm:rust                      # Rust UPM tests
#   ./runner.sh -c pm:pacman                     # pacman PM engine tests
#   ./runner.sh -c pm:apt -k test_mirror         # apt mirror test only
#   ./runner.sh -c upm:rust --network            # include network tests

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
CONTAINER_NAME="hbw-system-test-container"
STAGING_DIR="$SCRIPT_DIR/.staging"

# ── defaults ────────────────────────────────────────────────────────

MODE="auto"
CATEGORY="baker"           # baker | upm:<name> | pm:<name>
TEST_DIR=""
TEST_IMAGE=""
PYTEST_ARGS=()
NETWORK=0
BAKER_LOG=1

# ── parse args ──────────────────────────────────────────────────────

show_help() {
  cat << 'HELP'
Usage: ./runner.sh [OPTIONS]

Categories (-c):
  baker                    Baker system tests [default]
  upm:rust                 Rust UPM tests
  pm:pacman                Pacman PM engine tests
  pm:apt                   APT PM engine tests
  pm:dnf                   DNF PM engine tests
  pm:apk                   APK PM engine tests

Options:
  -c, --category CAT       Test category (see above)
  --network                Allow network-dependent tests (toolchain/crate downloads)
  -k FILTER                pytest -k expression
  --timeout SECONDS        Per-test timeout (default 120)
  --log / --no-log         Stream baker output to terminal [default: --log]
  -h, --help               This message
HELP
  exit 0
}

while [[ $# -gt 0 ]]; do
  case "$1" in
    -c|--category)
      CATEGORY="$2"; shift 2 ;;
    --network)
      NETWORK=1; shift ;;
    -k)
      PYTEST_ARGS+=(-k "$2"); shift 2 ;;
    --timeout)
      PYTEST_ARGS+=(--timeout "$2"); shift 2 ;;
    --log)     BAKER_LOG=1; shift ;;
    --no-log)  BAKER_LOG=0; shift ;;
    manual)    MODE="manual"; shift ;;
    auto)      MODE="auto"; shift ;;
    -h|--help)
      show_help ;;
    --)
      shift; PYTEST_ARGS+=("$@"); break ;;
    *)
      PYTEST_ARGS+=("$1"); shift ;;
  esac
done

# ── resolve category → test dir + image ─────────────────────────────

case "$CATEGORY" in
  baker)
    TEST_DIR="tests/baker"
    TEST_IMAGE="hbw-system-test:latest"
    ;;
  upm:rust)
    TEST_DIR="tests/upm"
    TEST_IMAGE="hbw-system-test:latest"
    ;;
  pm:pacman|pm:arch)
    TEST_DIR="tests/engine"
    TEST_IMAGE="hbw-test-pm:test-arch"
    ;;
  pm:apt)
    TEST_DIR="tests/engine"
    TEST_IMAGE="hbw-test-pm:test-apt"
    ;;
  pm:dnf)
    TEST_DIR="tests/engine"
    TEST_IMAGE="hbw-test-pm:test-dnf"
    ;;
  pm:apk)
    TEST_DIR="tests/engine"
    TEST_IMAGE="hbw-test-pm:test-apk"
    ;;
  *)
    echo "Unknown category: $CATEGORY" >&2
    echo "Run './runner.sh --help' for usage." >&2
    exit 1
    ;;
esac

# ── env vars ──────────────────────────────────────────────────────

DOCKER_ENV=()
DOCKER_ENV+=(-e RUST_LOG=${RUST_LOG:-info})
DOCKER_ENV+=(-e BAKER_LOG=${BAKER_LOG:-1})

# Load llm.env if present (for LLM-based test evaluation)
if [ -f "$SCRIPT_DIR/llm.env" ]; then
  while IFS='=' read -r key value; do
    [ -z "$key" ] && continue
    [[ "$key" =~ ^# ]] && continue
    DOCKER_ENV+=(-e "$key=$value")
  done < "$SCRIPT_DIR/llm.env"
fi

if [ "$NETWORK" -eq 1 ]; then
  DOCKER_ENV+=(-e RUST_UPM_NETWORK=1)
fi

# ── cleanup ─────────────────────────────────────────────────────────

cleanup() {
  docker rm -f "$CONTAINER_NAME" 2>/dev/null || true
}
trap cleanup EXIT

# ── build baker binary ──────────────────────────────────────────────

build_binary() {
  local target="x86_64-unknown-linux-musl"
  local binary="$SCRIPT_DIR/workshop-baker/target/$target/debug/workshop-baker"
  if [ ! -f "$binary" ]; then
    echo "[runner] Building static (musl) baker binary..."
    cargo build -p workshop-baker --target "$target" 2>&1 | tail -n 1
  fi
}

# ── staging ─────────────────────────────────────────────────────────

prepare_staging() {
  if [ "$TEST_DIR" != "tests/engine" ]; then
    mkdir -p "$STAGING_DIR"
    cp "$SCRIPT_DIR/workshop-baker/target/x86_64-unknown-linux-musl/debug/workshop-baker" "$STAGING_DIR/"
    chmod +x "$STAGING_DIR/workshop-baker"
  fi
}

# ── pm filter ───────────────────────────────────────────────────────

pm_filter() {
  case "${TEST_IMAGE##*:}" in
    test-arch) echo "-k pacman" ;;
    test-apt)  echo "-k apt" ;;
    test-dnf)  echo "-k dnf" ;;
    test-apk)  echo "-k apk" ;;
    *)         echo "" ;;
  esac
}

# ── run ─────────────────────────────────────────────────────────────

run_container() {
  local filter=""
  if [ "$TEST_DIR" = "tests/engine" ]; then
    filter="$(pm_filter)"
  fi

  echo "============================================"
  echo "  HBW System Test Runner"
  echo "  Category: $TEST_DIR  Image: $TEST_IMAGE"
  echo "============================================"

  mkdir -p "$SCRIPT_DIR/tests/results"

  local mounts=(-v "$SCRIPT_DIR:/src:ro")
  if [ "$TEST_DIR" != "tests/engine" ]; then
    mounts+=(-v "$STAGING_DIR:/workshop")  # writable: baker writes temp scripts here
  fi

  local pytest_cmd="python3 -m venv /tmp/v && . /tmp/v/bin/activate && \
    pip install -i https://mirrors.tuna.tsinghua.edu.cn/pypi/web/simple \
    -r $TEST_DIR/../requirements.txt -q && \
    exec python -m pytest -v -s $TEST_DIR ${PYTEST_ARGS[*]} $filter"

  if [ "$MODE" = "manual" ]; then
    docker run -it --rm --name "$CONTAINER_NAME" --privileged \
      ${DOCKER_ENV[@]} "${mounts[@]}" "$TEST_IMAGE"
  else
    # Baker/UPM images (hbw-system-test) have ./entrypoint.sh which
    # doesn't exist when /src is the workdir → override to empty.
    # PM images (hbw-test-pm:*) have /bin/sh -c → CMD passed directly.
    if [ "$TEST_DIR" = "tests/engine" ]; then
      docker run --rm --name "$CONTAINER_NAME" --privileged \
        ${DOCKER_ENV[@]} "${mounts[@]}" -w /src \
        "$TEST_IMAGE" "$pytest_cmd"
    else
      docker run --rm --name "$CONTAINER_NAME" --privileged \
        --entrypoint "" ${DOCKER_ENV[@]} "${mounts[@]}" -w /src \
        "$TEST_IMAGE" /bin/sh -c "$pytest_cmd"
    fi
  fi
}

# ── main ────────────────────────────────────────────────────────────

build_binary
prepare_staging
run_container
echo "[runner] Done."