#!/usr/bin/env bash
# diagnose-phase0-boot-evidence.sh — Phase 0 boot evidence collector
#
# Checks a mounted/extracted Red Bear root tree, or the running root when no
# --root is supplied, for the boot-critical evidence needed to diagnose modern
# hardware bring-up: PCI driver config locations, init service wiring, driver
# binary presence, and optional driver-manager boot-timeline evidence.

set -euo pipefail

ROOT="/"
BOOT_LOG=""

usage() {
    cat <<'USAGE'
Usage:
  local/scripts/diagnose-phase0-boot-evidence.sh [--root <mounted-root>] [--boot-log <log-file>]

Examples:
  local/scripts/diagnose-phase0-boot-evidence.sh --root /mnt/redbear-root
  local/scripts/diagnose-phase0-boot-evidence.sh --boot-log serial.log

Exit codes:
  0 — evidence complete enough for Phase 0 diagnostics
  1 — missing boot-critical evidence found
  2 — usage or input path error
USAGE
}

while [ $# -gt 0 ]; do
    case "$1" in
        --root)
            [ $# -ge 2 ] || { usage >&2; exit 2; }
            ROOT="$2"
            shift 2
            ;;
        --boot-log)
            [ $# -ge 2 ] || { usage >&2; exit 2; }
            BOOT_LOG="$2"
            shift 2
            ;;
        -h|--help)
            usage
            exit 0
            ;;
        *)
            echo "ERROR: unknown argument: $1" >&2
            usage >&2
            exit 2
            ;;
    esac
done

if [ ! -d "$ROOT" ]; then
    echo "ERROR: root path is not a directory: $ROOT" >&2
    exit 2
fi

if [ -n "$BOOT_LOG" ] && [ ! -f "$BOOT_LOG" ]; then
    echo "ERROR: boot log not found: $BOOT_LOG" >&2
    exit 2
fi

failures=0
warnings=0

join_root() {
    local path="$1"
    if [ "$ROOT" = "/" ]; then
        printf '/%s' "${path#/}"
    else
        printf '%s/%s' "${ROOT%/}" "${path#/}"
    fi
}

section() {
    echo ""
    echo "=== $* ==="
}

ok() { echo "  OK: $*"; }
warn() { echo "  WARN: $*"; warnings=$((warnings + 1)); }
fail() { echo "  FAIL: $*"; failures=$((failures + 1)); }

file_exists() { [ -f "$(join_root "$1")" ]; }
dir_exists() { [ -d "$(join_root "$1")" ]; }
executable_exists() { [ -x "$(join_root "$1")" ]; }

count_files() {
    local dir="$1"
    if dir_exists "$dir"; then
        find "$(join_root "$dir")" -maxdepth 1 -type f 2>/dev/null | wc -l | tr -d ' '
    else
        printf '0'
    fi
}

config_contains_driver() {
    local driver="$1"
    shift
    local dir path
    for dir in "$@"; do
        path="$(join_root "$dir")"
        [ -d "$path" ] || continue
        if grep -Rqs "\b${driver}\b" "$path"; then
            return 0
        fi
    done
    return 1
}

service_cmd() {
    local service="$1"
    local path
    for path in "/etc/init.d/$service" "/usr/lib/init.d/$service" "/lib/init.d/$service"; do
        if file_exists "$path"; then
            grep -E '^[[:space:]]*cmd[[:space:]]*=' "$(join_root "$path")" 2>/dev/null | head -n 1 || true
            return 0
        fi
    done
    return 1
}

section "Phase 0 root"
echo "  root: $ROOT"
[ -n "$BOOT_LOG" ] && echo "  boot_log: $BOOT_LOG"

section "PCI config directories"
pcid_count=0
for dir in /usr/lib/pcid.d /lib/pcid.d /etc/pcid.d; do
    count="$(count_files "$dir")"
    echo "  $dir: $count files"
    pcid_count=$((pcid_count + count))
done

driversd_count=0
for dir in /usr/lib/drivers.d /lib/drivers.d /etc/drivers.d; do
    count="$(count_files "$dir")"
    echo "  $dir: $count files"
    driversd_count=$((driversd_count + count))
done

if [ "$pcid_count" -eq 0 ] && [ "$driversd_count" -eq 0 ]; then
    fail "no pcid.d or drivers.d config files found; driver-manager cannot match devices"
elif [ "$driversd_count" -gt 0 ]; then
    ok "drivers.d config files present"
elif [ "$pcid_count" -gt 0 ]; then
    ok "pcid.d config files present"
else
    warn "unexpected PCI config directory state"
fi

section "Boot-critical driver config coverage"
check_driver_config() {
    local driver="$1"
    local description="$2"
    if config_contains_driver "$driver" /usr/lib/pcid.d /lib/pcid.d /etc/pcid.d /usr/lib/drivers.d /lib/drivers.d /etc/drivers.d; then
        ok "$description has config entry containing '$driver'"
    else
        fail "$description has no config entry containing '$driver'"
    fi
}

check_driver_config xhcid "xHCI USB"
check_driver_config nvmed "NVMe storage"
check_driver_config ihdad "Intel HDA audio"
check_driver_config redox-drm "DRM/KMS graphics"

section "Boot-critical binary coverage"
check_binary_anywhere() {
    local description="$1"
    shift
    local candidate
    for candidate in "$@"; do
        if executable_exists "$candidate"; then
            ok "$description binary exists at $candidate"
            return 0
        fi
    done
    fail "$description binary missing; checked: $*"
}

check_binary_anywhere "pcid" /usr/bin/pcid /usr/lib/drivers/pcid
check_binary_anywhere "driver-manager" /usr/bin/driver-manager /usr/lib/drivers/driver-manager
if executable_exists /usr/bin/pcid-spawner || executable_exists /usr/lib/drivers/pcid-spawner; then
    ok "pcid-spawner compatibility binary exists"
else
    warn "pcid-spawner compatibility binary missing; this is acceptable when driver-manager owns spawning"
fi
check_binary_anywhere "xHCI" /usr/lib/drivers/xhcid /usr/bin/xhcid
check_binary_anywhere "NVMe" /usr/lib/drivers/nvmed /usr/bin/nvmed
check_binary_anywhere "Intel HDA" /usr/lib/drivers/ihdad /usr/bin/ihdad
check_binary_anywhere "DRM/KMS" /usr/bin/redox-drm /usr/lib/drivers/redox-drm

section "Init service wiring"
if cmd_line="$(service_cmd 00_driver-manager.service)" && [ -n "$cmd_line" ]; then
    ok "00_driver-manager.service present: $cmd_line"
else
    fail "00_driver-manager.service missing from init service directories"
fi

if cmd_line="$(service_cmd 00_pcid-spawner.service)" && [ -n "$cmd_line" ]; then
    ok "00_pcid-spawner.service compatibility alias present: $cmd_line"
else
    warn "00_pcid-spawner.service compatibility alias not present"
fi

if [ "$driversd_count" -gt 0 ] && service_cmd 00_driver-manager.service >/dev/null 2>&1; then
    cmd_line="$(service_cmd 00_driver-manager.service)"
    if echo "$cmd_line" | grep -q 'pcid-spawner'; then
        warn "00_driver-manager.service still runs pcid-spawner while drivers.d files exist"
    fi
fi

section "Optional boot log evidence"
if [ -z "$BOOT_LOG" ]; then
    warn "no --boot-log supplied; skipping runtime evidence checks"
else
    if grep -q '"event":"bus_enumerated"' "$BOOT_LOG"; then
        ok "driver-manager bus enumeration timeline evidence found"
        grep '"event":"bus_enumerated"' "$BOOT_LOG" | sed 's/^/  /'
    elif grep -q 'pcid-spawner: phase0 summary' "$BOOT_LOG"; then
        ok "legacy pcid-spawner phase0 summary found"
        grep 'pcid-spawner: phase0 summary' "$BOOT_LOG" | sed 's/^/  /'
    else
        fail "driver-manager bus enumeration evidence not found in boot log"
    fi

    if grep -q '"event":"probe".*"status":"bound"' "$BOOT_LOG"; then
        ok "driver-manager bound-probe evidence found"
    elif grep -q 'pcid-spawner: matched PCI' "$BOOT_LOG"; then
        ok "legacy matched PCI driver evidence found"
    else
        warn "no bound PCI driver evidence found"
    fi

    if grep -q '"event":"bus_enumeration_failed"' "$BOOT_LOG"; then
        warn "driver-manager bus enumeration failures found"
        grep '"event":"bus_enumeration_failed"' "$BOOT_LOG" | sed 's/^/  /'
    elif grep -q '"event":"probe".*"status":"failed"' "$BOOT_LOG"; then
        warn "driver-manager failed-probe evidence found"
        grep '"event":"probe".*"status":"failed"' "$BOOT_LOG" | sed 's/^/  /'
    elif grep -q 'pcid-spawner: unmatched PCI' "$BOOT_LOG"; then
        warn "legacy unmatched PCI devices found; inspect detailed pcid-spawner logs"
        grep 'pcid-spawner: unmatched PCI' "$BOOT_LOG" | sed 's/^/  /'
    else
        ok "no bus enumeration failure or failed-probe evidence in supplied boot log"
    fi
fi

section "Summary"
echo "  failures: $failures"
echo "  warnings: $warnings"

if [ "$failures" -gt 0 ]; then
    echo "FAILED: Phase 0 boot evidence is incomplete." >&2
    exit 1
fi

echo "OK: Phase 0 boot evidence checks passed."
exit 0