EMS/internal/engine/engine.go

package engine

import (
	"fmt"
	"log/slog"
	"time"

	"github.com/tb/ems/internal/collector"
	"github.com/tb/ems/internal/config"
)

// Consumer identifies a controllable load.
type Consumer int

const (
	ConsumerSGReady Consumer = iota
	ConsumerWW       // domestic hot water boost via Viessmann API
	ConsumerWallboxA
	ConsumerWallboxB
)

func (c Consumer) String() string {
	switch c {
	case ConsumerSGReady:
		return "sg_ready"
	case ConsumerWW:
		return "ww"
	case ConsumerWallboxA:
		return "wallbox_a"
	case ConsumerWallboxB:
		return "wallbox_b"
	default:
		return "unknown"
	}
}

// Action represents a switching decision.
type Action struct {
	Consumer    Consumer
	TurnOn      bool
	Reason      string
	TargetTempC float64 // non-zero for ConsumerWW: the absolute temperature to set
}

// DeviceStatus holds the hardware-reported state of a consumer device,
// as read back from the physical device each cycle.
type DeviceStatus struct {
	On     bool
	PowerW float64 // measured active power; 0 if device has no power meter
}

// ConsumerState tracks the runtime state of a single consumer.
type ConsumerState struct {
	Active         bool
	ActivatedAt    time.Time // when it was last turned on
	ManualOverride bool
	OverrideUntil  time.Time
	LowPowerCycles int // consecutive cycles with power below minimum threshold
}

// OverrideInfo is returned to callers that need to display or record override state.
type OverrideInfo struct {
	Active bool
	Until  time.Time
}

// HysteresisState tracks the timing for hysteresis decisions.
type HysteresisState struct {
	// How long has export been above the on-threshold continuously?
	ExportSinceAbove map[Consumer]time.Time
	// How long has import been above the off-threshold continuously?
	ImportSinceAbove time.Time
}

// Engine is the EMS decision engine.
// It is pure: given a state snapshot and timing info, it returns actions.
// No network calls, no side effects — fully testable.
type Engine struct {
	cfg       *config.Config
	consumers map[Consumer]*ConsumerState
	hyst      HysteresisState
	logger    *slog.Logger
}

// NewEngine creates a new decision engine.
func NewEngine(cfg *config.Config, logger *slog.Logger) *Engine {
	return &Engine{
		cfg: cfg,
		consumers: map[Consumer]*ConsumerState{
			ConsumerSGReady:  {},
			ConsumerWW:       {},
			ConsumerWallboxA: {},
			ConsumerWallboxB: {},
		},
		hyst: HysteresisState{
			ExportSinceAbove: make(map[Consumer]time.Time),
		},
		logger: logger,
	}
}

// Decide evaluates the current system state and returns a list of actions.
// wwBoostC is the WW temperature boost in °C derived from the PV forecast
// (0 = no forecast / forecast too low to warrant boosting).
func (e *Engine) Decide(state collector.SystemState, now time.Time, wwBoostC float64) []Action {
	var actions []Action

	soc := state.BatterySOC
	gridW := state.GridPowerW // positive = import, negative = export
	heatingPeriod := e.isHeatingPeriod(now, state.AmbientTempC)
	wwWindow := e.isWWWindow(now)
	allowed := e.allowedConsumers(soc)

	e.logger.Debug("decision input",
		"grid_w", gridW,
		"soc", soc,
		"heating_period", heatingPeriod,
		"ww_window", wwWindow,
		"ww_boost_c", wwBoostC,
		"allowed", allowed,
	)

	// --- SOC emergency brake ---
	actions = append(actions, e.socEmergencyBrake(soc, now)...)

	// --- Override hard stop: cancel override on excessive import ---
	actions = append(actions, e.overrideHardStop(gridW)...)

	// --- WW window shutdown ---
	// If WW is active but we're outside the allowed time window, reset immediately.
	if cs := e.consumers[ConsumerWW]; cs.Active && !wwWindow {
		e.logger.Info("WW window ended, resetting DHW temperature")
		cs.Active = false
		actions = append(actions, Action{
			Consumer:    ConsumerWW,
			TurnOn:      false,
			TargetTempC: e.cfg.Strategic.WWBaseC,
			Reason:      "WW time window ended",
		})
	}

	// --- Shutdown logic (reverse priority order) ---
	if gridW > e.cfg.Thresholds.ImportOffW {
		if e.hyst.ImportSinceAbove.IsZero() {
			e.hyst.ImportSinceAbove = now
		}
		importDuration := now.Sub(e.hyst.ImportSinceAbove)
		if importDuration >= e.cfg.Hysteresis.ImportOffDurationParsed() {
			if a := e.shutdownLastConsumer(now); a != nil {
				actions = append(actions, *a)
				e.hyst.ImportSinceAbove = time.Time{}
			}
		}
	} else {
		e.hyst.ImportSinceAbove = time.Time{}
	}

	// --- Compressor idle: release SG-Ready early if heat pump stopped ---
	if cs := e.consumers[ConsumerSGReady]; cs.Active {
		if state.CompressorPowerW < float64(e.cfg.Consumers.CompressorIdleW) {
			cs.LowPowerCycles++
			e.logger.Debug("SG-Ready: compressor idle cycle",
				"compressor_w", state.CompressorPowerW,
				"idle_cycles", cs.LowPowerCycles,
			)
			if cs.LowPowerCycles >= e.cfg.Consumers.IdleCycles {
				e.logger.Info("SG-Ready released early: compressor idle",
					"idle_cycles", cs.LowPowerCycles,
					"compressor_w", state.CompressorPowerW,
				)
				cs.Active = false
				cs.LowPowerCycles = 0
				actions = append(actions, Action{
					Consumer: ConsumerSGReady,
					TurnOn:   false,
					Reason:   fmt.Sprintf("compressor idle for %d cycles", e.cfg.Consumers.IdleCycles),
				})
			}
		} else {
			cs.LowPowerCycles = 0
		}
	}

	// --- Car not charging: release wallbox if Shelly PM shows no draw ---
	for _, wb := range []Consumer{ConsumerWallboxA, ConsumerWallboxB} {
		cs := e.consumers[wb]
		if !cs.Active {
			continue
		}
		if cs.LowPowerCycles >= e.cfg.Consumers.IdleCycles {
			e.logger.Info("wallbox released: car not charging",
				"consumer", wb,
				"low_power_cycles", cs.LowPowerCycles,
			)
			cs.Active = false
			cs.LowPowerCycles = 0
			actions = append(actions, Action{
				Consumer: wb,
				TurnOn:   false,
				Reason:   fmt.Sprintf("car not charging for %d cycles", e.cfg.Consumers.IdleCycles),
			})
		}
	}

	// --- Turn-on logic (priority order) ---
	if gridW <= 0 {
		// P1: SG-Ready (heating period only)
		if heatingPeriod {
			actions = append(actions, e.evaluateTurnOn(
				ConsumerSGReady, gridW, e.cfg.Thresholds.SGReadyExportW,
				allowed, now,
			)...)
		}

		// P2: WW boost (time window + forecast required)
		if wwWindow && wwBoostC > 0 {
			actions = append(actions, e.evaluateWWTurnOn(gridW, wwBoostC, allowed, now)...)
		}

		// P3: Wallbox A (2kW, single-phase) — only if Wallbox B is not active.
		// Uses per-phase export check if available, otherwise falls back to total.
		// Re-reads Active state directly so a same-cycle activation of WallboxB blocks WallboxA.
		if !e.consumers[ConsumerWallboxB].Active {
			phaseGridW := gridW // fallback: total grid power
			if state.PhaseL1PowerW != 0 || state.PhaseL2PowerW != 0 || state.PhaseL3PowerW != 0 {
				phaseGridW = min3(state.PhaseL1PowerW, state.PhaseL2PowerW, state.PhaseL3PowerW)
			}
			threshold := e.cfg.Thresholds.WallboxAExportW
			if e.cfg.Thresholds.WallboxAPhaseExportW != 0 {
				threshold = e.cfg.Thresholds.WallboxAPhaseExportW
			}
			actions = append(actions, e.evaluateTurnOn(
				ConsumerWallboxA, phaseGridW, threshold,
				allowed, now,
			)...)
		}

		// P4: Wallbox B (4kW, 3-phase) — only if Wallbox A is not active.
		// Re-reads Active state so a same-cycle activation of WallboxA blocks WallboxB.
		if !e.consumers[ConsumerWallboxA].Active {
			actions = append(actions, e.evaluateTurnOn(
				ConsumerWallboxB, gridW, e.cfg.Thresholds.WallboxBExportW,
				allowed, now,
			)...)
		}
	}

	return actions
}

// min3 returns the minimum of three float64 values.
func min3(a, b, c float64) float64 {
	if b < a {
		a = b
	}
	if c < a {
		return c
	}
	return a
}

// allowedConsumers returns which consumers are allowed based on SOC.
func (e *Engine) allowedConsumers(soc float64) map[Consumer]bool {
	allowed := make(map[Consumer]bool)

	if soc >= e.cfg.SOC.AllConsumers {
		allowed[ConsumerSGReady] = true
		allowed[ConsumerWW] = true
		allowed[ConsumerWallboxA] = true
		allowed[ConsumerWallboxB] = true
	} else if soc >= e.cfg.SOC.SGReadyOnly {
		allowed[ConsumerSGReady] = true
		allowed[ConsumerWW] = true
		allowed[ConsumerWallboxA] = true
	} else if soc >= e.cfg.SOC.BlockAll {
		allowed[ConsumerSGReady] = true
		allowed[ConsumerWW] = true
	}
	// below BlockAll: nothing allowed

	return allowed
}

// evaluateTurnOn checks if a consumer should be turned on.
func (e *Engine) evaluateTurnOn(
	consumer Consumer,
	gridW float64,
	threshold float64,
	allowed map[Consumer]bool,
	now time.Time,
) []Action {
	cs := e.consumers[consumer]

	// Already active — nothing to do
	if cs.Active {
		return nil
	}

	// Manually overridden to OFF — respect until timeout
	if cs.ManualOverride && now.Before(cs.OverrideUntil) {
		delete(e.hyst.ExportSinceAbove, consumer)
		return nil
	}

	// Not allowed by SOC
	if !allowed[consumer] {
		delete(e.hyst.ExportSinceAbove, consumer)
		return nil
	}

	// Check if export exceeds the threshold
	// gridW is negative for export, threshold is negative (e.g. -1800)
	// export > 1800W means gridW < -1800
	if gridW > threshold {
		// Not enough export
		delete(e.hyst.ExportSinceAbove, consumer)
		return nil
	}

	// Export is above threshold — track how long
	if _, ok := e.hyst.ExportSinceAbove[consumer]; !ok {
		e.hyst.ExportSinceAbove[consumer] = now
	}

	exportDuration := now.Sub(e.hyst.ExportSinceAbove[consumer])
	if exportDuration < e.cfg.Hysteresis.ExportOnDurationParsed() {
		// Not long enough yet
		return nil
	}

	// All conditions met — turn on
	e.logger.Info("turning on consumer",
		"consumer", consumer,
		"grid_w", gridW,
		"threshold", threshold,
		"export_duration", exportDuration,
	)

	cs.Active = true
	cs.ActivatedAt = now
	delete(e.hyst.ExportSinceAbove, consumer)

	return []Action{{
		Consumer: consumer,
		TurnOn:   true,
		Reason: fmt.Sprintf(
			"export %.0fW > %.0fW for %s",
			-gridW, -threshold, exportDuration,
		),
	}}
}

// shutdownLastConsumer turns off the lowest-priority active consumer
// that has exceeded its minimum runtime.
func (e *Engine) shutdownLastConsumer(now time.Time) *Action {
	// Reverse priority: WallboxB → WallboxA → WW → SGReady
	order := []Consumer{ConsumerWallboxB, ConsumerWallboxA, ConsumerWW, ConsumerSGReady}

	for _, c := range order {
		cs := e.consumers[c]
		if !cs.Active {
			continue
		}

		// Manually overridden to ON — don't shut down until override expires
		if cs.ManualOverride && now.Before(cs.OverrideUntil) {
			e.logger.Debug("skipping shutdown, consumer is manually overridden",
				"consumer", c,
				"override_until", cs.OverrideUntil.Format("15:04"),
			)
			continue
		}

		minRuntime := e.minRuntime(c)
		runtime := now.Sub(cs.ActivatedAt)
		if runtime < minRuntime {
			e.logger.Debug("skipping shutdown, min runtime not reached",
				"consumer", c,
				"runtime", runtime,
				"min_runtime", minRuntime,
			)
			continue
		}

		e.logger.Info("shutting down consumer",
			"consumer", c,
			"runtime", runtime,
		)

		cs.Active = false
		a := &Action{
			Consumer: c,
			TurnOn:   false,
			Reason:   fmt.Sprintf("import detected, runtime %s", runtime),
		}
		if c == ConsumerWW {
			a.TargetTempC = e.cfg.Strategic.WWBaseC
		}
		return a
	}

	return nil
}

// socEmergencyBrake immediately shuts off consumers whose SOC threshold
// is no longer met, ignoring minimum runtimes and manual overrides.
func (e *Engine) socEmergencyBrake(soc float64, now time.Time) []Action {
	var actions []Action
	allowed := e.allowedConsumers(soc)

	for c, cs := range e.consumers {
		if !cs.Active {
			continue
		}
		if allowed[c] {
			continue
		}

		e.logger.Warn("SOC emergency brake",
			"consumer", c,
			"soc", soc,
			"was_override", cs.ManualOverride,
		)

		cs.Active = false
		cs.ManualOverride = false // EMS takes back full control after emergency
		cs.OverrideUntil = time.Time{}
		a := Action{
			Consumer: c,
			TurnOn:   false,
			Reason:   fmt.Sprintf("SOC emergency brake: %.0f%%", soc),
		}
		if c == ConsumerWW {
			a.TargetTempC = e.cfg.Strategic.WWBaseC
		}
		actions = append(actions, a)
	}

	return actions
}

// overrideHardStop cancels active overrides when grid import exceeds the
// configured hard-stop threshold. Called before normal shutdown logic so
// there is no hysteresis delay — protection is immediate.
func (e *Engine) overrideHardStop(gridW float64) []Action {
	limit := e.cfg.EMS.OverrideMaxImportW
	if limit <= 0 || gridW <= limit {
		return nil
	}

	var actions []Action
	for c, cs := range e.consumers {
		if !cs.Active || !cs.ManualOverride {
			continue
		}

		e.logger.Warn("override hard stop: import exceeds limit",
			"consumer", c,
			"grid_w", gridW,
			"limit_w", limit,
		)

		cs.Active = false
		cs.ManualOverride = false
		cs.OverrideUntil = time.Time{}
		a := Action{
			Consumer: c,
			TurnOn:   false,
			Reason:   fmt.Sprintf("override cancelled: import %.0fW > limit %.0fW", gridW, limit),
		}
		if c == ConsumerWW {
			a.TargetTempC = e.cfg.Strategic.WWBaseC
		}
		actions = append(actions, a)
	}
	return actions
}

// ApplyOverride directly sets a consumer's state and override lockout.
// Called from the web UI override handler so the engine state is consistent
// immediately, without waiting for the next SyncHardwareState cycle.
func (e *Engine) ApplyOverride(consumer Consumer, on bool, duration time.Duration) {
	cs, ok := e.consumers[consumer]
	if !ok {
		return
	}
	now := time.Now()
	cs.Active = on
	cs.ManualOverride = true
	cs.OverrideUntil = now.Add(duration)
	cs.LowPowerCycles = 0
	if on {
		cs.ActivatedAt = now
	} else {
		cs.ActivatedAt = time.Time{}
	}
	e.logger.Info("manual override applied",
		"consumer", consumer,
		"on", on,
		"duration", duration,
		"until", cs.OverrideUntil.Format("15:04"),
	)
}

// isHeatingPeriod returns true if heating is appropriate given the current month
// and outdoor temperature. If HeatingMinAmbientC is configured (> 0), ambient
// temperatures above that threshold suppress SG-Ready even within the heating months.
func (e *Engine) isHeatingPeriod(now time.Time, ambientC float64) bool {
	month := int(now.Month())
	start := e.cfg.Season.HeatingStartMonth
	end := e.cfg.Season.HeatingEndMonth

	// Handles wrap-around: e.g. October(10) to April(4)
	var inMonth bool
	if start > end {
		inMonth = month >= start || month <= end
	} else {
		inMonth = month >= start && month <= end
	}

	if !inMonth {
		return false
	}

	// Temperature override: warm day within heating months → not a heating day
	threshold := e.cfg.Season.HeatingMinAmbientC
	if threshold > 0 && ambientC >= threshold {
		e.logger.Debug("heating period suppressed by ambient temperature",
			"ambient_c", ambientC,
			"threshold_c", threshold,
		)
		return false
	}

	return true
}

// minRuntime returns the minimum runtime for a consumer.
func (e *Engine) minRuntime(c Consumer) time.Duration {
	switch c {
	case ConsumerSGReady:
		return e.cfg.Hysteresis.MinRuntimeSGReadyParsed()
	case ConsumerWallboxA, ConsumerWallboxB:
		return e.cfg.Hysteresis.MinRuntimeWallboxParsed()
	default:
		return 0 // WW has no minimum runtime
	}
}

// evaluateWWTurnOn checks whether WW boost should be activated.
// Prerequisites (time window and forecast) are already verified by the caller.
func (e *Engine) evaluateWWTurnOn(gridW, wwBoostC float64, allowed map[Consumer]bool, now time.Time) []Action {
	cs := e.consumers[ConsumerWW]
	if cs.Active {
		return nil
	}
	if cs.ManualOverride && now.Before(cs.OverrideUntil) {
		delete(e.hyst.ExportSinceAbove, ConsumerWW)
		return nil
	}
	if !allowed[ConsumerWW] {
		delete(e.hyst.ExportSinceAbove, ConsumerWW)
		return nil
	}
	if gridW > e.cfg.Thresholds.WWExportW {
		delete(e.hyst.ExportSinceAbove, ConsumerWW)
		return nil
	}

	if _, ok := e.hyst.ExportSinceAbove[ConsumerWW]; !ok {
		e.hyst.ExportSinceAbove[ConsumerWW] = now
	}
	exportDuration := now.Sub(e.hyst.ExportSinceAbove[ConsumerWW])
	if exportDuration < e.cfg.Hysteresis.ExportOnDurationParsed() {
		return nil
	}

	targetTemp := e.cfg.Strategic.WWBaseC + wwBoostC
	e.logger.Info("activating WW boost",
		"grid_w", gridW,
		"ww_boost_c", wwBoostC,
		"target_temp_c", targetTemp,
		"export_duration", exportDuration,
	)

	cs.Active = true
	cs.ActivatedAt = now
	delete(e.hyst.ExportSinceAbove, ConsumerWW)

	return []Action{{
		Consumer:    ConsumerWW,
		TurnOn:      true,
		TargetTempC: targetTemp,
		Reason:      fmt.Sprintf("export %.0fW for %s, WW boost +%.0f°C", -gridW, exportDuration, wwBoostC),
	}}
}

// isWWWindow returns true if the current time falls within the configured WW boost window.
func (e *Engine) isWWWindow(now time.Time) bool {
	start, err1 := parseTimeOfDay(e.cfg.Strategic.WWWindowStart, now)
	end, err2 := parseTimeOfDay(e.cfg.Strategic.WWWindowEnd, now)
	if err1 != nil || err2 != nil {
		return false
	}
	return now.After(start) && now.Before(end)
}

// parseTimeOfDay parses "HH:MM" and returns a time.Time on the same day as ref.
func parseTimeOfDay(s string, ref time.Time) (time.Time, error) {
	var h, m int
	if _, err := fmt.Sscanf(s, "%d:%d", &h, &m); err != nil {
		return time.Time{}, fmt.Errorf("invalid time-of-day %q: %w", s, err)
	}
	return time.Date(ref.Year(), ref.Month(), ref.Day(), h, m, 0, 0, ref.Location()), nil
}

// ConsumerStates returns a snapshot of all consumer states (for metrics).
func (e *Engine) ConsumerStates() map[Consumer]bool {
	states := make(map[Consumer]bool)
	for c, cs := range e.consumers {
		states[c] = cs.Active
	}
	return states
}

// RecoverState injects externally-read consumer states on startup.
// Does not set override flags — startup state is treated as the EMS baseline.
func (e *Engine) RecoverState(states map[Consumer]DeviceStatus) {
	for c, status := range states {
		if cs, ok := e.consumers[c]; ok {
			cs.Active = status.On
			cs.ManualOverride = false
			cs.OverrideUntil = time.Time{}
			cs.LowPowerCycles = 0
			// ActivatedAt left as zero: unknown start time means the consumer
			// is always considered to have exceeded its minimum runtime.
			cs.ActivatedAt = time.Time{}
		}
	}
	e.logger.Info("consumer state recovered from Shelly read-back",
		"sg_ready", states[ConsumerSGReady].On,
		"wallbox_a", states[ConsumerWallboxA].On,
		"wallbox_b", states[ConsumerWallboxB].On,
	)
}

// SyncHardwareState compares live hardware states against the engine's internal state.
// Discrepancies indicate an external change (manual override via Shelly app etc.).
// On mismatch: engine state is updated to match hardware, and the consumer is locked
// from EMS control for overrideTimeout.
// On match: expired overrides are cleared, resuming normal EMS control.
// Power readings (from PM-capable devices) update the low-power cycle counter for
// car-not-charging detection.
func (e *Engine) SyncHardwareState(states map[Consumer]DeviceStatus, now time.Time, overrideTimeout time.Duration) {
	for c, status := range states {
		cs, ok := e.consumers[c]
		if !ok {
			continue
		}

		if cs.Active != status.On {
			// External change detected
			e.logger.Info("manual override detected — external state change",
				"consumer", c,
				"engine_state", cs.Active,
				"hardware_state", status.On,
				"override_until", now.Add(overrideTimeout).Format("15:04"),
			)
			cs.Active = status.On
			cs.ManualOverride = true
			cs.OverrideUntil = now.Add(overrideTimeout)
			cs.LowPowerCycles = 0
			if !status.On {
				cs.ActivatedAt = time.Time{}
			} else {
				cs.ActivatedAt = now
			}
		} else if cs.ManualOverride && now.After(cs.OverrideUntil) {
			// Override expired and state matches — resume EMS control
			cs.ManualOverride = false
			cs.OverrideUntil = time.Time{}
			e.logger.Info("manual override expired, resuming EMS control", "consumer", c)
		}

		// Track low-power cycles for car-not-charging detection (PM devices only).
		// Only meaningful when PowerW > 0 (i.e., device has a power meter and is on).
		if status.PowerW > 0 {
			if cs.Active && status.PowerW < float64(e.cfg.Consumers.WallboxMinChargeW) {
				cs.LowPowerCycles++
				e.logger.Debug("wallbox low power cycle",
					"consumer", c,
					"power_w", status.PowerW,
					"low_power_cycles", cs.LowPowerCycles,
				)
			} else {
				cs.LowPowerCycles = 0
			}
		}
	}
}

// Overrides returns current override info for all consumers that are overridden.
func (e *Engine) Overrides() map[Consumer]OverrideInfo {
	result := make(map[Consumer]OverrideInfo)
	for c, cs := range e.consumers {
		if cs.ManualOverride {
			result[c] = OverrideInfo{Active: true, Until: cs.OverrideUntil}
		}
	}
	return result
}