trustformers-debug 0.2.1

//! Comprehensive error recovery mechanisms for debugging sessions

use anyhow::Result;
use serde::{Deserialize, Serialize};
use std::collections::{HashMap, VecDeque};
use std::time::{Duration, Instant};
use uuid::Uuid;

/// Comprehensive error recovery system
#[derive(Debug)]
pub struct ErrorRecoverySystem {
    config: ErrorRecoveryConfig,
    recovery_strategies: HashMap<ErrorType, Vec<RecoveryStrategy>>,
    error_history: VecDeque<ErrorEvent>,
    recovery_history: VecDeque<RecoveryEvent>,
    circuit_breaker: CircuitBreaker,
    health_monitor: SystemHealthMonitor,
    failsafe_manager: FailsafeManager,
}

/// Configuration for error recovery
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ErrorRecoveryConfig {
    pub enabled: bool,
    pub max_retry_attempts: usize,
    pub retry_delay_ms: u64,
    pub circuit_breaker_threshold: usize,
    pub health_check_interval_ms: u64,
    pub auto_failsafe_enabled: bool,
    pub error_history_limit: usize,
    pub recovery_timeout_ms: u64,
}

impl Default for ErrorRecoveryConfig {
    fn default() -> Self {
        Self {
            enabled: true,
            max_retry_attempts: 3,
            retry_delay_ms: 100,
            circuit_breaker_threshold: 5,
            health_check_interval_ms: 5000,
            auto_failsafe_enabled: true,
            error_history_limit: 1000,
            recovery_timeout_ms: 30000,
        }
    }
}

/// Types of errors that can occur during debugging
#[derive(Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub enum ErrorType {
    TensorInspectionError,
    GradientDebuggingError,
    ModelDiagnosticsError,
    VisualizationError,
    MemoryProfilingError,
    IOError,
    NetworkError,
    ResourceExhaustion,
    ConfigurationError,
    DataCorruption,
    SystemFailure,
    UserError,
}

/// Recovery strategies for different error types
#[derive(Debug, Clone, Serialize, Deserialize)]
pub enum RecoveryStrategy {
    Retry { max_attempts: usize, delay_ms: u64 },
    Fallback { alternative_method: String },
    GracefulDegradation { reduced_functionality: String },
    ResourceCleanup { cleanup_type: String },
    SystemReset { component: String },
    EmergencyShutdown,
    UserNotification { message: String },
    AutomaticRepair { repair_action: String },
}

/// Error event record
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ErrorEvent {
    pub id: Uuid,
    pub error_type: ErrorType,
    pub error_message: String,
    pub component: String,
    pub severity: ErrorSeverity,
    pub timestamp: chrono::DateTime<chrono::Utc>,
    pub context: ErrorContext,
    pub stack_trace: Option<String>,
}

/// Error severity levels
#[derive(Debug, Clone, Serialize, Deserialize)]
pub enum ErrorSeverity {
    Low,
    Medium,
    High,
    Critical,
    Fatal,
}

/// Context information for errors
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ErrorContext {
    pub session_id: Uuid,
    pub operation: String,
    pub parameters: HashMap<String, String>,
    pub system_state: SystemState,
}

/// System state at time of error
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct SystemState {
    pub memory_usage_mb: u64,
    pub cpu_usage_percent: f64,
    pub active_tensors: usize,
    pub active_sessions: usize,
    pub uptime_seconds: u64,
}

/// Recovery event record
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct RecoveryEvent {
    pub id: Uuid,
    pub error_id: Uuid,
    pub strategy: RecoveryStrategy,
    pub start_time: chrono::DateTime<chrono::Utc>,
    pub end_time: Option<chrono::DateTime<chrono::Utc>>,
    pub success: Option<bool>,
    pub result_message: String,
    pub attempts: usize,
}

/// Circuit breaker for preventing cascading failures
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct CircuitBreaker {
    pub state: CircuitState,
    pub failure_count: usize,
    pub last_failure_time: Option<chrono::DateTime<chrono::Utc>>,
    pub threshold: usize,
    pub timeout_duration: Duration,
}

/// Circuit breaker states
#[derive(Debug, Clone, Serialize, Deserialize)]
pub enum CircuitState {
    Closed,
    Open,
    HalfOpen,
}

/// System health monitoring
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct SystemHealthMonitor {
    pub overall_health: HealthStatus,
    pub component_health: HashMap<String, HealthStatus>,
    pub last_health_check: chrono::DateTime<chrono::Utc>,
    pub health_metrics: HealthMetrics,
}

/// Health status
#[derive(Debug, Clone, Serialize, Deserialize)]
pub enum HealthStatus {
    Healthy,
    Degraded,
    Unhealthy,
    Critical,
}

/// Health metrics
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct HealthMetrics {
    /// Errors recorded in the last five minutes, normalised by 100.
    /// Genuinely recomputed by [`ErrorRecoverySystem::check_system_health`].
    pub error_rate: f64,
    /// Fraction of the recovery attempts that finished in the last five
    /// minutes and reported success, or `None` when none finished in that
    /// window.
    ///
    /// This used to be seeded to `1.0` in
    /// [`ErrorRecoverySystem::new`] -- a published 100% recovery rate before a
    /// single error had been handled -- and only ever overwritten once a
    /// recovery actually completed.
    pub recovery_success_rate: Option<f64>,
    /// Mean wall-clock duration of those same completed recovery attempts, in
    /// milliseconds; `None` when none finished in the window.
    ///
    /// Renamed from `average_response_time_ms`, which nothing ever wrote to:
    /// it stayed at its `0.0` seed for the lifetime of the system. It is
    /// recovery time, not response time.
    pub average_recovery_time_ms: Option<f64>,
    /// Always `None`: nothing in this crate measures memory health, and there
    /// is no defined scale for it. Previously seeded to `1.0` and never
    /// updated, i.e. a permanent "perfectly healthy" verdict.
    pub memory_health_score: Option<f64>,
    /// Always `None`, for the same reason (previously a permanent `1.0`).
    pub stability_score: Option<f64>,
}

/// Failsafe manager for critical situations
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct FailsafeManager {
    pub enabled: bool,
    pub emergency_protocols: Vec<EmergencyProtocol>,
    pub safe_mode_enabled: bool,
    pub data_backup_enabled: bool,
    pub last_backup: Option<chrono::DateTime<chrono::Utc>>,
}

/// Emergency protocols
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct EmergencyProtocol {
    pub name: String,
    pub trigger_conditions: Vec<String>,
    pub actions: Vec<String>,
    pub priority: u8,
}

impl ErrorRecoverySystem {
    /// Create a new error recovery system
    pub fn new(config: ErrorRecoveryConfig) -> Self {
        let mut system = Self {
            config,
            recovery_strategies: HashMap::new(),
            error_history: VecDeque::new(),
            recovery_history: VecDeque::new(),
            circuit_breaker: CircuitBreaker {
                state: CircuitState::Closed,
                failure_count: 0,
                last_failure_time: None,
                threshold: 5,
                timeout_duration: Duration::from_secs(60),
            },
            health_monitor: SystemHealthMonitor {
                overall_health: HealthStatus::Healthy,
                component_health: HashMap::new(),
                last_health_check: chrono::Utc::now(),
                health_metrics: HealthMetrics {
                    // Nothing has happened yet: zero errors really is a
                    // measured zero, everything else is honestly absent.
                    error_rate: 0.0,
                    recovery_success_rate: None,
                    average_recovery_time_ms: None,
                    memory_health_score: None,
                    stability_score: None,
                },
            },
            failsafe_manager: FailsafeManager {
                enabled: true,
                emergency_protocols: Vec::new(),
                safe_mode_enabled: false,
                data_backup_enabled: true,
                last_backup: None,
            },
        };

        system.initialize_default_strategies();
        system.initialize_emergency_protocols();
        system
    }

    /// Handle an error event and attempt recovery
    pub async fn handle_error(&mut self, error: ErrorEvent) -> Result<RecoveryResult> {
        // Check circuit breaker
        if matches!(self.circuit_breaker.state, CircuitState::Open) {
            return Ok(RecoveryResult {
                success: false,
                strategy_used: None,
                message: "Circuit breaker is open - recovery attempts suspended".to_string(),
                recovery_time: Duration::from_millis(0),
            });
        }

        // Record error
        self.record_error(error.clone());

        // Check if emergency protocols should be triggered
        if self.should_trigger_emergency_protocol(&error) {
            return self.execute_emergency_protocol(&error).await;
        }

        // Attempt recovery
        let recovery_result = self.attempt_recovery(&error).await?;

        // Update circuit breaker and health monitor
        self.update_circuit_breaker(&recovery_result);
        self.update_health_metrics(&error, &recovery_result);

        Ok(recovery_result)
    }

    /// Record an error event
    pub fn record_error(&mut self, error: ErrorEvent) {
        self.error_history.push_back(error);

        // Maintain history limit
        while self.error_history.len() > self.config.error_history_limit {
            self.error_history.pop_front();
        }
    }

    /// Attempt recovery using appropriate strategies
    pub async fn attempt_recovery(&mut self, error: &ErrorEvent) -> Result<RecoveryResult> {
        let strategies = self.get_recovery_strategies(&error.error_type);

        for (attempt, strategy) in strategies.iter().enumerate() {
            if attempt >= self.config.max_retry_attempts {
                break;
            }

            let recovery_event = RecoveryEvent {
                id: Uuid::new_v4(),
                error_id: error.id,
                strategy: strategy.clone(),
                start_time: chrono::Utc::now(),
                end_time: None,
                success: None,
                result_message: String::new(),
                attempts: attempt + 1,
            };

            let result = self.execute_recovery_strategy(strategy, error).await?;

            let mut updated_event = recovery_event;
            updated_event.end_time = Some(chrono::Utc::now());
            updated_event.success = Some(result.success);
            updated_event.result_message = result.message.clone();

            self.recovery_history.push_back(updated_event);

            if result.success {
                return Ok(result);
            }

            // Wait before next attempt
            if attempt < strategies.len() - 1 {
                tokio::time::sleep(Duration::from_millis(self.config.retry_delay_ms)).await;
            }
        }

        Ok(RecoveryResult {
            success: false,
            strategy_used: None,
            message: "All recovery strategies failed".to_string(),
            recovery_time: Duration::from_millis(0),
        })
    }

    /// Execute a specific recovery strategy
    pub async fn execute_recovery_strategy(
        &self,
        strategy: &RecoveryStrategy,
        error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        let start_time = Instant::now();

        let result = match strategy {
            RecoveryStrategy::Retry {
                max_attempts,
                delay_ms,
            } => self.execute_retry_strategy(*max_attempts, *delay_ms, error).await,
            RecoveryStrategy::Fallback { alternative_method } => {
                self.execute_fallback_strategy(alternative_method, error).await
            },
            RecoveryStrategy::GracefulDegradation {
                reduced_functionality,
            } => self.execute_degradation_strategy(reduced_functionality, error).await,
            RecoveryStrategy::ResourceCleanup { cleanup_type } => {
                self.execute_cleanup_strategy(cleanup_type, error).await
            },
            RecoveryStrategy::SystemReset { component } => {
                self.execute_reset_strategy(component, error).await
            },
            RecoveryStrategy::EmergencyShutdown => self.execute_shutdown_strategy(error).await,
            RecoveryStrategy::UserNotification { message } => {
                self.execute_notification_strategy(message, error).await
            },
            RecoveryStrategy::AutomaticRepair { repair_action } => {
                self.execute_repair_strategy(repair_action, error).await
            },
        };

        let recovery_time = start_time.elapsed();

        match result {
            Ok(mut recovery_result) => {
                recovery_result.recovery_time = recovery_time;
                recovery_result.strategy_used = Some(strategy.clone());
                Ok(recovery_result)
            },
            Err(e) => Ok(RecoveryResult {
                success: false,
                strategy_used: Some(strategy.clone()),
                message: format!("Recovery strategy failed: {}", e),
                recovery_time,
            }),
        }
    }

    /// Get recovery strategies for an error type
    pub fn get_recovery_strategies(&self, error_type: &ErrorType) -> Vec<RecoveryStrategy> {
        self.recovery_strategies.get(error_type).cloned().unwrap_or_default()
    }

    /// Check system health
    pub async fn check_system_health(&mut self) -> HealthStatus {
        // Update health metrics
        self.health_monitor.last_health_check = chrono::Utc::now();

        // Calculate error rate from recent history
        let recent_errors = self
            .error_history
            .iter()
            .filter(|e| {
                let age = chrono::Utc::now() - e.timestamp;
                age < chrono::Duration::minutes(5)
            })
            .count();

        self.health_monitor.health_metrics.error_rate = recent_errors as f64 / 100.0; // Normalized

        // Calculate recovery success rate
        let recent_recoveries = self
            .recovery_history
            .iter()
            .filter(|r| {
                if let Some(end_time) = r.end_time {
                    let age = chrono::Utc::now() - end_time;
                    age < chrono::Duration::minutes(5)
                } else {
                    false
                }
            })
            .collect::<Vec<_>>();

        if recent_recoveries.is_empty() {
            // No completed recovery in the window: the previous figures are
            // stale, and there is nothing to replace them with.
            self.health_monitor.health_metrics.recovery_success_rate = None;
            self.health_monitor.health_metrics.average_recovery_time_ms = None;
        } else {
            let successful_recoveries =
                recent_recoveries.iter().filter(|r| r.success.unwrap_or(false)).count();
            self.health_monitor.health_metrics.recovery_success_rate =
                Some(successful_recoveries as f64 / recent_recoveries.len() as f64);
            // Real mean wall-clock recovery duration over the same window.
            let total_ms: f64 = recent_recoveries
                .iter()
                .filter_map(|r| {
                    r.end_time.map(|end| (end - r.start_time).num_milliseconds() as f64)
                })
                .sum();
            self.health_monitor.health_metrics.average_recovery_time_ms =
                Some(total_ms / recent_recoveries.len() as f64);
        }

        // Determine overall health
        self.health_monitor.overall_health = if self.health_monitor.health_metrics.error_rate > 0.5
        {
            HealthStatus::Critical
        } else if self.health_monitor.health_metrics.error_rate > 0.2 {
            HealthStatus::Unhealthy
        } else if self.health_monitor.health_metrics.error_rate > 0.1 {
            HealthStatus::Degraded
        } else {
            HealthStatus::Healthy
        };

        self.health_monitor.overall_health.clone()
    }

    /// Enable safe mode
    pub fn enable_safe_mode(&mut self) {
        self.failsafe_manager.safe_mode_enabled = true;
        tracing::warn!("Safe mode enabled - operating with reduced functionality");
    }

    /// Disable safe mode
    pub fn disable_safe_mode(&mut self) {
        self.failsafe_manager.safe_mode_enabled = false;
        tracing::info!("Safe mode disabled - full functionality restored");
    }

    /// Get error statistics
    pub fn get_error_statistics(&self) -> ErrorStatistics {
        let total_errors = self.error_history.len();
        let error_type_counts = self.error_history.iter().fold(HashMap::new(), |mut acc, error| {
            *acc.entry(error.error_type.clone()).or_insert(0) += 1;
            acc
        });

        let severity_counts = self.error_history.iter().fold(HashMap::new(), |mut acc, error| {
            *acc.entry(format!("{:?}", error.severity)).or_insert(0) += 1;
            acc
        });

        ErrorStatistics {
            total_errors,
            error_type_counts,
            severity_counts,
            recovery_success_rate: self.health_monitor.health_metrics.recovery_success_rate,
            circuit_breaker_state: self.circuit_breaker.state.clone(),
            system_health: self.health_monitor.overall_health.clone(),
        }
    }

    // Private helper methods

    fn initialize_default_strategies(&mut self) {
        // Initialize default recovery strategies for each error type
        self.recovery_strategies.insert(
            ErrorType::TensorInspectionError,
            vec![
                RecoveryStrategy::Retry {
                    max_attempts: 3,
                    delay_ms: 100,
                },
                RecoveryStrategy::ResourceCleanup {
                    cleanup_type: "tensor_cache".to_string(),
                },
                RecoveryStrategy::Fallback {
                    alternative_method: "simplified_inspection".to_string(),
                },
            ],
        );

        self.recovery_strategies.insert(
            ErrorType::GradientDebuggingError,
            vec![
                RecoveryStrategy::Retry {
                    max_attempts: 2,
                    delay_ms: 200,
                },
                RecoveryStrategy::GracefulDegradation {
                    reduced_functionality: "basic_gradient_info".to_string(),
                },
            ],
        );

        self.recovery_strategies.insert(
            ErrorType::MemoryProfilingError,
            vec![
                RecoveryStrategy::ResourceCleanup {
                    cleanup_type: "memory_profiler".to_string(),
                },
                RecoveryStrategy::SystemReset {
                    component: "memory_tracker".to_string(),
                },
            ],
        );

        self.recovery_strategies.insert(
            ErrorType::ResourceExhaustion,
            vec![
                RecoveryStrategy::ResourceCleanup {
                    cleanup_type: "all_caches".to_string(),
                },
                RecoveryStrategy::GracefulDegradation {
                    reduced_functionality: "essential_only".to_string(),
                },
                RecoveryStrategy::EmergencyShutdown,
            ],
        );

        // Add more strategies for other error types...
    }

    fn initialize_emergency_protocols(&mut self) {
        self.failsafe_manager.emergency_protocols = vec![
            EmergencyProtocol {
                name: "Memory Exhaustion Protocol".to_string(),
                trigger_conditions: vec!["memory_usage > 90%".to_string()],
                actions: vec![
                    "clear_all_caches".to_string(),
                    "reduce_tracking".to_string(),
                ],
                priority: 1,
            },
            EmergencyProtocol {
                name: "Critical Error Protocol".to_string(),
                trigger_conditions: vec!["error_severity == Fatal".to_string()],
                actions: vec!["emergency_backup".to_string(), "safe_shutdown".to_string()],
                priority: 0,
            },
        ];
    }

    fn should_trigger_emergency_protocol(&self, error: &ErrorEvent) -> bool {
        matches!(error.severity, ErrorSeverity::Fatal)
            || error.context.system_state.memory_usage_mb > 8192 // > 8GB
    }

    async fn execute_emergency_protocol(&mut self, error: &ErrorEvent) -> Result<RecoveryResult> {
        tracing::error!(
            "Executing emergency protocol for error: {}",
            error.error_message
        );

        // Enable safe mode
        self.enable_safe_mode();

        // Execute emergency backup if enabled
        if self.failsafe_manager.data_backup_enabled {
            self.create_emergency_backup().await?;
        }

        Ok(RecoveryResult {
            success: true,
            strategy_used: Some(RecoveryStrategy::EmergencyShutdown),
            message: "Emergency protocol executed successfully".to_string(),
            recovery_time: Duration::from_millis(0),
        })
    }

    async fn create_emergency_backup(&mut self) -> Result<()> {
        tracing::info!("Creating emergency backup");
        self.failsafe_manager.last_backup = Some(chrono::Utc::now());
        // Implement backup logic here
        Ok(())
    }

    fn update_circuit_breaker(&mut self, result: &RecoveryResult) {
        if result.success {
            self.circuit_breaker.failure_count = 0;
            self.circuit_breaker.state = CircuitState::Closed;
        } else {
            self.circuit_breaker.failure_count += 1;
            self.circuit_breaker.last_failure_time = Some(chrono::Utc::now());

            if self.circuit_breaker.failure_count >= self.circuit_breaker.threshold {
                self.circuit_breaker.state = CircuitState::Open;
            }
        }
    }

    fn update_health_metrics(&mut self, _error: &ErrorEvent, _result: &RecoveryResult) {
        // Update health metrics based on error and recovery result
        // This would include more sophisticated health scoring logic
    }

    // Recovery strategy execution.
    //
    // `ErrorRecoveryManager` records and classifies errors; it holds no handle
    // to the subsystems a recovery would have to act on (no retryable operation
    // closure, no allocator, no component registry, no process supervisor). The
    // strategies below therefore report `success: false` with a message naming
    // exactly what is not wired, instead of the `success: true` +
    // "Retry successful" / "Resource cleanup completed" they used to return
    // having done nothing at all -- which made `record_error` mark every error
    // recovered and inflated `recovery_success_rate` to 100%.
    //
    // `execute_notification_strategy` is the one that really acts (it emits a
    // `tracing::warn!`), and it is the one that still reports success.

    /// Build an "unwired" result naming the missing capability.
    fn unwired(capability: &str) -> RecoveryResult {
        RecoveryResult {
            success: false,
            strategy_used: None,
            message: format!(
                "recovery not performed: {capability} -- ErrorRecoveryManager has no handle to \
                 act on"
            ),
            recovery_time: Duration::from_millis(0),
        }
    }

    async fn execute_retry_strategy(
        &self,
        _max_attempts: usize,
        _delay_ms: u64,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(
            "no retryable operation was supplied with the error",
        ))
    }

    async fn execute_fallback_strategy(
        &self,
        alternative: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(&format!(
            "no dispatcher exists for the alternative method {alternative:?}"
        )))
    }

    async fn execute_degradation_strategy(
        &self,
        functionality: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(&format!(
            "no feature switch exists for {functionality:?}"
        )))
    }

    async fn execute_cleanup_strategy(
        &self,
        cleanup_type: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(&format!(
            "no allocator or cache handle for {cleanup_type:?}"
        )))
    }

    async fn execute_reset_strategy(
        &self,
        component: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(&format!(
            "no component registry entry for {component:?}"
        )))
    }

    async fn execute_shutdown_strategy(&self, _error: &ErrorEvent) -> Result<RecoveryResult> {
        Ok(Self::unwired(
            "no process supervisor handle to initiate shutdown",
        ))
    }

    /// The one strategy that really performs its action: it emits the
    /// notification through the `tracing` facade.
    async fn execute_notification_strategy(
        &self,
        message: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        tracing::warn!("User notification: {}", message);
        Ok(RecoveryResult {
            success: true,
            strategy_used: None,
            message: "User notified via the tracing facade".to_string(),
            recovery_time: Duration::from_millis(0),
        })
    }

    async fn execute_repair_strategy(
        &self,
        repair_action: &str,
        _error: &ErrorEvent,
    ) -> Result<RecoveryResult> {
        Ok(Self::unwired(&format!(
            "no repair executor for the action {repair_action:?}"
        )))
    }
}

/// Result of a recovery attempt
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct RecoveryResult {
    pub success: bool,
    pub strategy_used: Option<RecoveryStrategy>,
    pub message: String,
    pub recovery_time: Duration,
}

/// Error statistics
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ErrorStatistics {
    pub total_errors: usize,
    pub error_type_counts: HashMap<ErrorType, usize>,
    pub severity_counts: HashMap<String, usize>,
    /// See [`HealthMetrics::recovery_success_rate`]: `None` when no recovery
    /// attempt has completed recently, never a fabricated `1.0`.
    pub recovery_success_rate: Option<f64>,
    pub circuit_breaker_state: CircuitState,
    pub system_health: HealthStatus,
}

// ─────────────────────────────────────────────────────────────────────────────
// Tests
// ─────────────────────────────────────────────────────────────────────────────

#[cfg(test)]
mod tests {
    use super::*;

    fn sample_error() -> ErrorEvent {
        ErrorEvent {
            id: Uuid::new_v4(),
            error_type: ErrorType::TensorInspectionError,
            error_message: "boom".to_string(),
            component: "tensor_inspector".to_string(),
            severity: ErrorSeverity::Medium,
            timestamp: chrono::Utc::now(),
            context: ErrorContext {
                session_id: Uuid::new_v4(),
                operation: "inspect".to_string(),
                parameters: HashMap::new(),
                system_state: SystemState {
                    memory_usage_mb: 0,
                    cpu_usage_percent: 0.0,
                    active_tensors: 0,
                    active_sessions: 0,
                    uptime_seconds: 0,
                },
            },
            stack_trace: None,
        }
    }

    // ---- Wave 6c debug-sweep2: recovery strategies stop claiming success ---

    #[tokio::test]
    async fn unwired_recovery_strategies_report_failure_not_success() {
        let manager = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let error = sample_error();

        for result in [
            manager.execute_retry_strategy(3, 10, &error).await.expect("call"),
            manager.execute_fallback_strategy("other", &error).await.expect("call"),
            manager.execute_degradation_strategy("feature", &error).await.expect("call"),
            manager.execute_cleanup_strategy("cache", &error).await.expect("call"),
            manager.execute_reset_strategy("component", &error).await.expect("call"),
            manager.execute_shutdown_strategy(&error).await.expect("call"),
            manager.execute_repair_strategy("fix", &error).await.expect("call"),
        ] {
            assert!(
                !result.success,
                "an unperformed recovery must not report success: {}",
                result.message
            );
            assert!(
                result.message.contains("recovery not performed"),
                "the message must say what did not happen: {}",
                result.message
            );
        }
    }

    #[tokio::test]
    async fn the_notification_strategy_really_acts_and_reports_success() {
        let manager = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let error = sample_error();
        let result = manager
            .execute_notification_strategy("check the logs", &error)
            .await
            .expect("call");
        assert!(result.success, "the notification is really emitted");
        assert!(result.message.contains("tracing"));
    }

    fn make_error_event(error_type: ErrorType) -> ErrorEvent {
        ErrorEvent {
            id: Uuid::new_v4(),
            error_type,
            error_message: "test error".to_string(),
            component: "test_component".to_string(),
            severity: ErrorSeverity::Medium,
            timestamp: chrono::Utc::now(),
            context: ErrorContext {
                session_id: Uuid::new_v4(),
                operation: "test_op".to_string(),
                parameters: HashMap::new(),
                system_state: SystemState {
                    memory_usage_mb: 1024,
                    cpu_usage_percent: 50.0,
                    active_tensors: 4,
                    active_sessions: 1,
                    uptime_seconds: 100,
                },
            },
            stack_trace: None,
        }
    }

    // ── ErrorRecoveryConfig ──────────────────────────────────────────────

    #[test]
    fn test_config_default_fields() {
        let cfg = ErrorRecoveryConfig::default();
        assert!(cfg.enabled);
        assert!(cfg.max_retry_attempts > 0);
        assert!(cfg.circuit_breaker_threshold > 0);
        assert!(cfg.error_history_limit > 0);
    }

    // ── ErrorRecoverySystem creation ──────────────────────────────────────

    #[test]
    fn test_system_new_initializes_strategies() {
        let system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        // Should have strategies initialized for at least TensorInspectionError
        let strategies = system.get_recovery_strategies(&ErrorType::TensorInspectionError);
        assert!(!strategies.is_empty());
    }

    #[test]
    fn test_system_new_circuit_breaker_closed() {
        let system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        assert!(matches!(system.circuit_breaker.state, CircuitState::Closed));
    }

    // ── record_error ─────────────────────────────────────────────────────

    #[test]
    fn test_record_error_adds_to_history() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let event = make_error_event(ErrorType::IOError);
        system.record_error(event);
        assert_eq!(system.error_history.len(), 1);
    }

    #[test]
    fn test_record_error_respects_history_limit() {
        let mut cfg = ErrorRecoveryConfig::default();
        cfg.error_history_limit = 3;
        let mut system = ErrorRecoverySystem::new(cfg);
        for _ in 0..5 {
            system.record_error(make_error_event(ErrorType::NetworkError));
        }
        assert_eq!(system.error_history.len(), 3);
    }

    // ── safe_mode ────────────────────────────────────────────────────────

    #[test]
    fn test_enable_disable_safe_mode() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        assert!(!system.failsafe_manager.safe_mode_enabled);
        system.enable_safe_mode();
        assert!(system.failsafe_manager.safe_mode_enabled);
        system.disable_safe_mode();
        assert!(!system.failsafe_manager.safe_mode_enabled);
    }

    // ── get_error_statistics ──────────────────────────────────────────────

    #[test]
    fn test_error_statistics_empty() {
        let system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let stats = system.get_error_statistics();
        assert_eq!(stats.total_errors, 0);
        assert!(matches!(stats.circuit_breaker_state, CircuitState::Closed));
        assert!(matches!(stats.system_health, HealthStatus::Healthy));
    }

    #[test]
    fn test_error_statistics_with_errors() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        system.record_error(make_error_event(ErrorType::IOError));
        system.record_error(make_error_event(ErrorType::NetworkError));
        let stats = system.get_error_statistics();
        assert_eq!(stats.total_errors, 2);
        assert_eq!(
            stats.error_type_counts.get(&ErrorType::IOError).copied().unwrap_or(0),
            1
        );
    }

    // ── ErrorType variants ────────────────────────────────────────────────

    #[test]
    fn test_error_type_variants() {
        let types = [
            ErrorType::TensorInspectionError,
            ErrorType::GradientDebuggingError,
            ErrorType::ModelDiagnosticsError,
            ErrorType::VisualizationError,
            ErrorType::MemoryProfilingError,
            ErrorType::IOError,
            ErrorType::NetworkError,
            ErrorType::ResourceExhaustion,
            ErrorType::ConfigurationError,
            ErrorType::DataCorruption,
            ErrorType::SystemFailure,
            ErrorType::UserError,
        ];
        for t in &types {
            assert!(!format!("{:?}", t).is_empty());
        }
    }

    // ── ErrorSeverity variants ────────────────────────────────────────────

    #[test]
    fn test_error_severity_variants() {
        let severities = [
            ErrorSeverity::Low,
            ErrorSeverity::Medium,
            ErrorSeverity::High,
            ErrorSeverity::Critical,
            ErrorSeverity::Fatal,
        ];
        for s in &severities {
            assert!(!format!("{:?}", s).is_empty());
        }
    }

    // ── RecoveryStrategy variants ─────────────────────────────────────────

    #[test]
    fn test_recovery_strategy_variants() {
        let strats = [
            RecoveryStrategy::Retry {
                max_attempts: 3,
                delay_ms: 100,
            },
            RecoveryStrategy::Fallback {
                alternative_method: "alt".to_string(),
            },
            RecoveryStrategy::GracefulDegradation {
                reduced_functionality: "basic".to_string(),
            },
            RecoveryStrategy::ResourceCleanup {
                cleanup_type: "cache".to_string(),
            },
            RecoveryStrategy::SystemReset {
                component: "comp".to_string(),
            },
            RecoveryStrategy::EmergencyShutdown,
            RecoveryStrategy::UserNotification {
                message: "msg".to_string(),
            },
            RecoveryStrategy::AutomaticRepair {
                repair_action: "repair".to_string(),
            },
        ];
        for s in &strats {
            assert!(!format!("{:?}", s).is_empty());
        }
    }

    // ── CircuitState variants ─────────────────────────────────────────────

    #[test]
    fn test_circuit_state_variants() {
        let states = [
            CircuitState::Closed,
            CircuitState::Open,
            CircuitState::HalfOpen,
        ];
        for s in &states {
            assert!(!format!("{:?}", s).is_empty());
        }
    }

    // ── HealthStatus variants ─────────────────────────────────────────────

    #[test]
    fn test_health_status_variants() {
        let statuses = [
            HealthStatus::Healthy,
            HealthStatus::Degraded,
            HealthStatus::Unhealthy,
            HealthStatus::Critical,
        ];
        for s in &statuses {
            assert!(!format!("{:?}", s).is_empty());
        }
    }

    // ── CircuitBreaker ────────────────────────────────────────────────────

    #[test]
    fn test_circuit_breaker_initial_state() {
        let system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        assert_eq!(system.circuit_breaker.failure_count, 0);
        assert!(system.circuit_breaker.last_failure_time.is_none());
        assert_eq!(system.circuit_breaker.threshold, 5);
    }

    // ── SystemState struct ────────────────────────────────────────────────

    #[test]
    fn test_system_state_construction() {
        let state = SystemState {
            memory_usage_mb: 2048,
            cpu_usage_percent: 75.5,
            active_tensors: 10,
            active_sessions: 2,
            uptime_seconds: 3600,
        };
        assert_eq!(state.memory_usage_mb, 2048);
        assert!((state.cpu_usage_percent - 75.5).abs() < 1e-6);
    }

    // ── HealthMetrics ──────────────────────────────────────────────────────

    #[test]
    fn test_health_metrics_initial_values() {
        let system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let m = &system.health_monitor.health_metrics;
        // A brand-new system has handled nothing. Zero errors is a real
        // measurement; a 100% recovery rate is not -- this used to assert
        // `recovery_success_rate == 1.0`, locking in a seeded fabrication.
        assert_eq!(m.error_rate, 0.0);
        assert_eq!(m.recovery_success_rate, None);
        assert_eq!(m.average_recovery_time_ms, None);
        assert_eq!(m.memory_health_score, None);
        assert_eq!(m.stability_score, None);
    }

    /// The rate must be a real quotient over the recovery attempts that
    /// actually completed, and the mean recovery time a real mean -- not the
    /// `1.0` / `0.0` the system used to be born with.
    #[tokio::test]
    async fn test_health_metrics_are_computed_from_real_recovery_outcomes() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        let now = chrono::Utc::now();
        for (success, millis) in [(true, 40i64), (true, 60), (false, 200)] {
            system.recovery_history.push_back(RecoveryEvent {
                id: Uuid::new_v4(),
                error_id: Uuid::new_v4(),
                strategy: RecoveryStrategy::Retry {
                    max_attempts: 1,
                    delay_ms: 0,
                },
                start_time: now - chrono::Duration::milliseconds(millis),
                end_time: Some(now),
                success: Some(success),
                result_message: "test".to_string(),
                attempts: 1,
            });
        }

        system.check_system_health().await;
        let m = &system.health_monitor.health_metrics;
        let rate = m.recovery_success_rate.expect("three attempts completed in the window");
        assert!(
            (rate - 2.0 / 3.0).abs() < 1e-9,
            "two of three succeeded, got {rate}"
        );
        let mean_ms = m.average_recovery_time_ms.expect("three attempts completed in the window");
        assert!(
            (mean_ms - 100.0).abs() < 5.0,
            "(40+60+200)/3 = 100ms, got {mean_ms}"
        );

        // Unmeasurable scores stay absent even after a health check.
        assert_eq!(m.memory_health_score, None);
        assert_eq!(m.stability_score, None);
    }

    /// A health check with nothing in the recent window must clear the
    /// figures rather than leave a stale rate standing as current evidence.
    #[tokio::test]
    async fn test_health_check_reports_absence_when_nothing_recovered_recently() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        system.check_system_health().await;
        assert_eq!(
            system.health_monitor.health_metrics.recovery_success_rate,
            None
        );
        assert_eq!(system.get_error_statistics().recovery_success_rate, None);
    }

    // ── async handle_error with open circuit breaker ──────────────────────

    #[tokio::test]
    async fn test_handle_error_with_open_circuit_breaker() {
        let mut system = ErrorRecoverySystem::new(ErrorRecoveryConfig::default());
        system.circuit_breaker.state = CircuitState::Open;
        let event = make_error_event(ErrorType::IOError);
        let result = system.handle_error(event).await.expect("should succeed");
        assert!(!result.success);
        assert!(result.message.contains("Circuit breaker"));
    }
}