// Package server is the steward-managed HTTP layer for lethe. It owns the
// chi router, the standard middleware stack (request-id, logging, metrics,
// recovery), the unauthenticated probes (/healthz, /readyz, /metrics), and
// the authenticated /api/v1/* group inside which the ingest and session
// handlers register their routes.
//
// The Server is a steward.Root() service: it has no consumers and steward
// would otherwise prune it. main.go (Phase 9) registers it as the root.
//
// Bind validation is fail-fast: only loopback IPs are accepted. Phase 5
// keeps the system reachable by the local collector; exposing it on a
// public interface is a config error.
package server

import (
	"context"
	"crypto/rand"
	"encoding/hex"
	"errors"
	"fmt"
	"log/slog"
	"net"
	"net/http"
	"time"

	"github.com/go-chi/chi/v5"
	"github.com/prometheus/client_golang/prometheus/promhttp"
	"go.bigb.es/auxilia/culpa"
	"go.bigb.es/auxilia/scribe"

	"sourcecraft.dev/bigbes/lethe/internal/config"
	"sourcecraft.dev/bigbes/lethe/internal/domain/ingest"
	"sourcecraft.dev/bigbes/lethe/internal/domain/session"
	"sourcecraft.dev/bigbes/lethe/internal/pkg/apierror"
	"sourcecraft.dev/bigbes/lethe/internal/pkg/httputil"
	"sourcecraft.dev/bigbes/lethe/internal/platform/health"
	"sourcecraft.dev/bigbes/lethe/internal/platform/observability"
	authpkg "sourcecraft.dev/bigbes/lethe/internal/server/auth"
)

// readyzTimeout caps the time /readyz spends running every health check.
// Long enough to absorb a transient slow probe, short enough that orchestrators
// do not wait too long on a stuck dependency.
const readyzTimeout = 5 * time.Second

// Server is the HTTP steward service. Steward injects the shared platform
// services and the auth/handler stubs (which Phases 6/7/8 replace).
type Server struct {
	Cfg config.ServerConfig `config:""`

	Log     *observability.Logger  `inject:""`
	Metrics *observability.Metrics `inject:""`
	Health  *health.Set            `inject:""`

	Auth     *authpkg.Authenticator `inject:""`
	Ingest   *ingest.Handler        `inject:""`
	Sessions *session.Handler       `inject:""`

	router  *chi.Mux
	httpSrv *http.Server
}

// Init validates the bind address and constructs the chi router with the
// standard middleware stack and all routes mounted. It does not start the
// listener — that is Start's job.
func (s *Server) Init(_ context.Context) error {
	if err := validateLoopbackBind(s.Cfg.Bind); err != nil {
		return err
	}

	r := chi.NewRouter()
	r.Use(s.requestIDMiddleware)
	r.Use(s.loggingMiddleware)
	r.Use(s.metricsMiddleware)
	r.Use(s.recoveryMiddleware)

	r.NotFound(notFoundHandler)
	r.MethodNotAllowed(methodNotAllowedHandler)

	r.Get("/healthz", s.healthzHandler)
	r.Get("/readyz", s.readyzHandler)
	r.Handle("/metrics", promhttp.HandlerFor(s.Metrics.Registry, promhttp.HandlerOpts{}))

	r.Route("/api/v1", func(r chi.Router) {
		r.Use(s.Auth.Middleware)
		s.Ingest.Mount(r)
		s.Sessions.Mount(r)
	})

	s.router = r
	return nil
}

// Start spawns ListenAndServe in the background and returns immediately.
// Errors are logged; steward observes lifecycle via Stop.
func (s *Server) Start(_ context.Context) error {
	s.httpSrv = &http.Server{
		Addr:              s.Cfg.Bind,
		Handler:           s.router,
		ReadHeaderTimeout: 10 * time.Second,
	}
	go func() {
		if err := s.httpSrv.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) {
			slog.Default().Error("server crashed", scribe.Err(err))
		}
	}()
	return nil
}

// Stop gracefully drains in-flight requests within ShutdownGrace, then
// closes the listener. Idempotent if Start was never called.
func (s *Server) Stop(ctx context.Context) error {
	if s.httpSrv == nil {
		return nil
	}
	grace := s.Cfg.ShutdownGrace
	if grace <= 0 {
		grace = 10 * time.Second
	}
	ctx2, cancel := context.WithTimeout(ctx, grace)
	defer cancel()
	return s.httpSrv.Shutdown(ctx2)
}

// validateLoopbackBind ensures s.Cfg.Bind resolves to a loopback IP.
// Anything else is rejected at startup with code CONFIG_INVALID.
func validateLoopbackBind(bind string) error {
	host, _, err := net.SplitHostPort(bind)
	if err != nil {
		return culpa.WithCode(culpa.Wrap(err, "parse bind address"), "CONFIG_INVALID")
	}
	ip := net.ParseIP(host)
	if ip == nil {
		return culpa.WithCode(culpa.Errorf("bind host %q is not an IP literal", host), "CONFIG_INVALID")
	}
	if !ip.IsLoopback() {
		return culpa.WithCode(culpa.Errorf("bind %q must resolve to a loopback IP", bind), "CONFIG_INVALID")
	}
	return nil
}

// healthzHandler is a constant-200 process-up probe.
func (s *Server) healthzHandler(w http.ResponseWriter, _ *http.Request) {
	w.Header().Set("Content-Type", "text/plain; charset=utf-8")
	w.WriteHeader(http.StatusOK)
	_, _ = w.Write([]byte("ok"))
}

// readyzResponse is the JSON body returned by /readyz.
type readyzResponse struct {
	Checks map[string]string `json:"checks"`
}

// readyzHandler runs every registered Checker with a 5s ceiling. Any
// failure flips the status to 503; the body always lists each check.
func (s *Server) readyzHandler(w http.ResponseWriter, r *http.Request) {
	ctx, cancel := context.WithTimeout(r.Context(), readyzTimeout)
	defer cancel()

	results, allOK := s.Health.Run(ctx)
	body := readyzResponse{Checks: make(map[string]string, len(results))}
	for name, err := range results {
		if err == nil {
			body.Checks[name] = "ok"
		} else {
			body.Checks[name] = err.Error()
		}
	}
	status := http.StatusOK
	if !allOK {
		status = http.StatusServiceUnavailable
	}
	if err := httputil.WriteJSON(w, status, body); err != nil {
		slog.Default().ErrorContext(r.Context(), "write readyz response", scribe.Err(err))
	}
}

// requestIDMiddleware mints a 32-char hex id, attaches it to the request
// context for the logger to pick up, and echoes it as X-Request-ID. The
// header is set before next.ServeHTTP so even slow handlers expose it.
func (s *Server) requestIDMiddleware(next http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		var buf [16]byte
		// crypto/rand.Read in Go 1.24+ is documented infallible.
		_, _ = rand.Read(buf[:])
		id := hex.EncodeToString(buf[:])
		w.Header().Set("X-Request-ID", id)
		ctx := observability.WithRequestID(r.Context(), id)
		next.ServeHTTP(w, r.WithContext(ctx))
	})
}

// responseRecorder is a minimal ResponseWriter wrapper that captures the
// final status code so the logging and metrics middlewares can label their
// records correctly.
type responseRecorder struct {
	http.ResponseWriter
	status      int
	wroteHeader bool
}

// WriteHeader records the status before delegating.
func (rr *responseRecorder) WriteHeader(code int) {
	if rr.wroteHeader {
		return
	}
	rr.status = code
	rr.wroteHeader = true
	rr.ResponseWriter.WriteHeader(code)
}

// Write triggers an implicit 200 if the handler wrote without WriteHeader.
func (rr *responseRecorder) Write(b []byte) (int, error) {
	if !rr.wroteHeader {
		rr.status = http.StatusOK
		rr.wroteHeader = true
	}
	return rr.ResponseWriter.Write(b)
}

// loggingMiddleware emits one structured access log per request. The
// request_id attribute is added automatically by contextHandler.
func (s *Server) loggingMiddleware(next http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		start := time.Now()
		rr := &responseRecorder{ResponseWriter: w, status: http.StatusOK}
		next.ServeHTTP(rr, r)
		slog.Default().InfoContext(r.Context(), "request",
			slog.String("method", r.Method),
			slog.String("path", r.URL.Path),
			slog.Int("status", rr.status),
			slog.Int64("duration_ms", time.Since(start).Milliseconds()),
		)
	})
}

// metricsMiddleware increments the request counter and observes the
// duration histogram. The route label comes from chi's RoutePattern so
// cardinality stays bounded; unmatched paths fall back to "unknown".
func (s *Server) metricsMiddleware(next http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		start := time.Now()
		rr := &responseRecorder{ResponseWriter: w, status: http.StatusOK}
		next.ServeHTTP(rr, r)

		route := "unknown"
		if rc := chi.RouteContext(r.Context()); rc != nil {
			if pat := rc.RoutePattern(); pat != "" {
				route = pat
			}
		}
		statusStr := fmt.Sprintf("%d", rr.status)
		s.Metrics.HTTPRequests.WithLabelValues(r.Method, route, statusStr).Inc()
		s.Metrics.HTTPDuration.WithLabelValues(r.Method, route).Observe(time.Since(start).Seconds())
	})
}

// recoveryMiddleware turns a panic into a 500 problem document. The
// stacktrace logging is handled by apierror.Render's 5xx branch, which
// receives the panic-as-error.
func (s *Server) recoveryMiddleware(next http.Handler) http.Handler {
	return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		defer func() {
			if rec := recover(); rec != nil {
				err := culpa.WithCode(culpa.Errorf("panic: %v", rec), "INTERNAL")
				apierror.Render(w, r, err)
			}
		}()
		next.ServeHTTP(w, r)
	})
}

func notFoundHandler(w http.ResponseWriter, r *http.Request) {
	apierror.Render(w, r, culpa.WithCode(culpa.New("route not found"), "NOT_FOUND"))
}

func methodNotAllowedHandler(w http.ResponseWriter, r *http.Request) {
	apierror.Render(w, r, culpa.WithCode(culpa.New("method not allowed"), "METHOD_NOT_ALLOWED"))
}