apex_solver/observers/

mod.rs

//! Observer pattern for optimization monitoring.
//!
//! This module provides a clean observer pattern for monitoring optimization progress.
//! Observers can be registered with any optimizer and will be notified at each iteration,
//! enabling real-time visualization, logging, metrics collection, and custom analysis.
//!
//! # Design Philosophy
//!
//! The observer pattern provides complete separation between optimization algorithms
//! and monitoring/visualization logic:
//!
//! - **Decoupling**: Optimization logic is independent of how progress is monitored
//! - **Extensibility**: Easy to add new observers (Rerun, CSV, metrics, dashboards)
//! - **Composability**: Multiple observers can run simultaneously
//! - **Zero overhead**: When no observers are registered, notification is a no-op
//!
//! # Architecture
//!
//! ```text
//! ┌─────────────────┐
//! │   Optimizer     │
//! │  (LM/GN/DogLeg) │
//! └────────┬────────┘
//!          │ observers.notify(values, iteration)
//!          ├──────────────┬──────────────┬──────────────┐
//!          ▼              ▼              ▼              ▼
//!    ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐
//!    │  Rerun   │  │   CSV    │  │ Metrics  │  │  Custom  │
//!    │ Observer │  │ Observer │  │ Observer │  │ Observer │
//!    └──────────┘  └──────────┘  └──────────┘  └──────────┘
//! ```
//!
//! # Examples
//!
//! ## Single Observer
//!
//! ```no_run
//! use apex_solver::{LevenbergMarquardt, LevenbergMarquardtConfig};
//! use apex_solver::observers::OptObserver;
//! # use apex_solver::core::problem::Problem;
//! # use apex_solver::JacobianMode;
//!
//! # fn main() -> Result<(), Box<dyn std::error::Error>> {
//! # let mut problem = Problem::new(JacobianMode::Sparse);
//!
//! let config = LevenbergMarquardtConfig::new().with_max_iterations(100);
//! let mut solver = LevenbergMarquardt::with_config(config);
//!
//! #[cfg(feature = "visualization")]
//! {
//!     use apex_solver::observers::RerunObserver;
//!     let rerun_observer = RerunObserver::new(true)?;
//!     solver.add_observer(rerun_observer);
//! }
//!
//! let result = solver.optimize(&mut problem)?;
//! # Ok(())
//! # }
//! ```
//!
//! ## Multiple Observers
//!
//! ```no_run
//! # use apex_solver::{LevenbergMarquardt, LevenbergMarquardtConfig};
//! # use apex_solver::core::problem::Problem;
//! # use apex_solver::core::variable::ManifoldVariable;
//! # use apex_solver::core::VarKey;
//! # use apex_solver::observers::OptObserver;
//! # use apex_solver::JacobianMode;
//! # use slotmap::SlotMap;
//!
//! // Custom observer that logs to CSV
//! struct CsvObserver {
//!     file: std::fs::File,
//! }
//!
//! impl OptObserver for CsvObserver {
//!     fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {
//!         // Write iteration data to CSV
//!         // ... implementation ...
//!     }
//! }
//!
//! # fn main() -> Result<(), Box<dyn std::error::Error>> {
//! # let mut problem = Problem::new(JacobianMode::Sparse);
//! let mut solver = LevenbergMarquardt::new();
//!
//! // Add Rerun visualization
//! #[cfg(feature = "visualization")]
//! {
//!     use apex_solver::observers::RerunObserver;
//!     solver.add_observer(RerunObserver::new(true)?);
//! }
//!
//! // Add CSV logging
//! // solver.add_observer(CsvObserver { file: ... });
//!
//! let result = solver.optimize(&mut problem)?;
//! # Ok(())
//! # }
//! ```
//!
//! ## Custom Observer
//!
//! ```no_run
//! use apex_solver::observers::OptObserver;
//! use apex_solver::core::variable::ManifoldVariable;
//! use apex_solver::core::VarKey;
//! use slotmap::SlotMap;
//!
//! struct MetricsObserver {
//!     max_variables_seen: std::cell::RefCell<usize>,
//! }
//!
//! impl OptObserver for MetricsObserver {
//!     fn on_step(&self, values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {
//!         let count = values.len();
//!         let mut max = self.max_variables_seen.borrow_mut();
//!         *max = (*max).max(count);
//!     }
//! }
//! ```

// Visualization-specific submodules (feature-gated)
#[cfg(feature = "visualization")]
pub mod conversions;
#[cfg(feature = "visualization")]
pub mod visualization;

// Re-export RerunObserver when visualization is enabled
#[cfg(feature = "visualization")]
pub use visualization::{RerunObserver, VisualizationConfig, VisualizationMode};

// Re-export conversion traits for ergonomic use
#[cfg(feature = "visualization")]
pub use conversions::{CollectRerun2D, CollectRerun3D, RerunConvert2D, RerunConvert3D};

use crate::core::VarKey;
use crate::core::variable::ManifoldVariable;
use faer::Mat;
use faer::sparse;
use slotmap::SlotMap;
use thiserror::Error;

/// Observer-specific error types for apex-solver
#[derive(Debug, Clone, Error)]
pub enum ObserverError {
    /// Failed to initialize Rerun recording stream
    #[error("Failed to initialize Rerun recording stream: {0}")]
    RerunInitialization(String),

    /// Failed to spawn Rerun viewer process
    #[error("Failed to spawn Rerun viewer: {0}")]
    ViewerSpawnFailed(String),

    /// Failed to save recording to file
    #[error("Failed to save recording to file '{path}': {reason}")]
    RecordingSaveFailed { path: String, reason: String },

    /// Failed to log data to Rerun
    #[error("Failed to log data to Rerun at '{entity_path}': {reason}")]
    LoggingFailed { entity_path: String, reason: String },

    /// Failed to convert matrix to visualization format
    #[error("Failed to convert matrix to image: {0}")]
    MatrixVisualizationFailed(String),

    /// Failed to convert tensor data
    #[error("Failed to create tensor data: {0}")]
    TensorConversionFailed(String),

    /// Recording stream is in invalid state
    #[error("Recording stream is in invalid state: {0}")]
    InvalidState(String),

    /// Mutex was poisoned (thread panicked while holding lock)
    #[error("Mutex poisoned in {context}: {reason}")]
    MutexPoisoned { context: String, reason: String },
}

/// Result type for observer operations
pub type ObserverResult<T> = Result<T, ObserverError>;

/// Observer trait for monitoring optimization progress.
///
/// Implement this trait to create custom observers that are notified at each
/// optimization iteration. Observers receive the current variable values and
/// iteration number, enabling real-time monitoring, visualization, logging,
/// or custom analysis.
///
/// # Design Notes
///
/// - Observers should be lightweight and non-blocking
/// - Errors in observers should not crash optimization (handle internally)
/// - For expensive operations (file I/O, network), consider buffering
/// - Observers receive immutable references (cannot modify optimization state)
///
/// # Thread Safety
///
/// Observers must be `Send` to support parallel optimization in the future.
/// Use interior mutability (`RefCell`, `Mutex`) if you need to mutate state.
pub trait OptObserver: Send {
    /// Called after each optimization iteration.
    ///
    /// # Arguments
    ///
    /// * `values` - Current variable values (manifold states)
    /// * `iteration` - Current iteration number (0 = initial values, 1+ = after steps)
    ///
    /// # Implementation Guidelines
    ///
    /// - Keep this method fast to avoid slowing optimization
    /// - Handle errors internally (log warnings, don't panic)
    /// - Don't mutate `values` (you receive `&HashMap`)
    /// - Consider buffering expensive operations
    ///
    /// # Examples
    ///
    /// ```no_run
    /// use apex_solver::observers::OptObserver;
    /// use apex_solver::core::variable::ManifoldVariable;
    /// use apex_solver::core::VarKey;
    /// use slotmap::SlotMap;
    ///
    /// struct SimpleLogger;
    ///
    /// impl OptObserver for SimpleLogger {
    ///     fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {
    ///         // Track optimization progress
    ///     }
    /// }
    /// ```
    fn on_step(&self, values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, iteration: usize);

    /// Set iteration metrics for visualization and monitoring.
    ///
    /// This method is called before `on_step` to provide optimization metrics
    /// such as cost, gradient norm, damping parameter, etc. Observers can use
    /// this data for visualization, logging, or analysis.
    ///
    /// # Arguments
    ///
    /// * `cost` - Current cost function value
    /// * `gradient_norm` - L2 norm of the gradient vector
    /// * `damping` - Damping parameter (for Levenberg-Marquardt, may be None for other solvers)
    /// * `step_norm` - L2 norm of the parameter update step
    /// * `step_quality` - Step quality metric (e.g., rho for trust region methods)
    ///
    /// # Default Implementation
    ///
    /// The default implementation does nothing, allowing simple observers to ignore metrics.
    fn set_iteration_metrics(
        &self,
        _cost: f64,
        _gradient_norm: f64,
        _damping: Option<f64>,
        _step_norm: f64,
        _step_quality: Option<f64>,
    ) {
        // Default implementation does nothing
    }

    /// Set matrix data for advanced visualization.
    ///
    /// This method provides access to the Hessian matrix and gradient vector
    /// for observers that want to visualize matrix structure or perform
    /// advanced analysis.
    ///
    /// # Arguments
    ///
    /// * `hessian` - Sparse Hessian matrix (J^T * J)
    /// * `gradient` - Gradient vector (J^T * r)
    ///
    /// # Default Implementation
    ///
    /// The default implementation does nothing, allowing simple observers to ignore matrices.
    fn set_matrix_data(
        &self,
        _hessian: Option<sparse::SparseColMat<usize, f64>>,
        _gradient: Option<Mat<f64>>,
    ) {
        // Default implementation does nothing
    }

    /// Called when optimization completes.
    ///
    /// This method is called once at the end of optimization, after all iterations
    /// are complete. Use this for final visualization, cleanup, or summary logging.
    ///
    /// # Arguments
    ///
    /// * `values` - Final optimized variable values
    /// * `iterations` - Total number of iterations performed
    ///
    /// # Default Implementation
    ///
    /// The default implementation does nothing, allowing simple observers to ignore completion.
    ///
    /// # Examples
    ///
    /// ```no_run
    /// use apex_solver::observers::OptObserver;
    /// use apex_solver::core::variable::ManifoldVariable;
    /// use apex_solver::core::VarKey;
    /// use slotmap::SlotMap;
    ///
    /// struct FinalStateLogger;
    ///
    /// impl OptObserver for FinalStateLogger {
    ///     fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {}
    ///
    ///     fn on_optimization_complete(&self, values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, iterations: usize) {
    ///         println!("Optimization completed after {} iterations with {} variables",
    ///                  iterations, values.len());
    ///     }
    /// }
    /// ```
    fn on_optimization_complete(
        &self,
        _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>,
        _iterations: usize,
    ) {
        // Default implementation does nothing
    }
}

/// Collection of observers for optimization monitoring.
///
/// This struct manages a vector of observers and provides a convenient
/// `notify()` method to call all observers at once. Optimizers use this
/// internally to manage their observers.
///
/// # Usage
///
/// Typically you don't create this directly - use the `add_observer()` method
/// on optimizers. However, you can use it for custom optimization algorithms:
///
/// ```no_run
/// use apex_solver::observers::{OptObserver, OptObserverVec};
/// use apex_solver::core::variable::ManifoldVariable;
/// use apex_solver::core::VarKey;
/// use slotmap::SlotMap;
///
/// struct MyOptimizer {
///     observers: OptObserverVec,
///     // ... other fields ...
/// }
///
/// impl MyOptimizer {
///     fn step(&mut self, values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, iteration: usize) {
///         // ... optimization logic ...
///
///         // Notify all observers
///         self.observers.notify(values, iteration);
///     }
/// }
/// ```
#[derive(Default)]
pub struct OptObserverVec {
    observers: Vec<Box<dyn OptObserver>>,
}

impl OptObserverVec {
    /// Create a new empty observer collection.
    pub fn new() -> Self {
        Self {
            observers: Vec::new(),
        }
    }

    /// Add an observer to the collection.
    ///
    /// The observer will be called at each optimization iteration in the order
    /// it was added.
    ///
    /// # Arguments
    ///
    /// * `observer` - Any type implementing `OptObserver`
    ///
    /// # Examples
    ///
    /// ```no_run
    /// use apex_solver::observers::{OptObserver, OptObserverVec};
    /// use apex_solver::core::variable::ManifoldVariable;
    /// use apex_solver::core::VarKey;
    /// use slotmap::SlotMap;
    ///
    /// struct MyObserver;
    /// impl OptObserver for MyObserver {
    ///     fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {
    ///         // Handle optimization step
    ///     }
    /// }
    ///
    /// let mut observers = OptObserverVec::new();
    /// observers.add(MyObserver);
    /// ```
    pub fn add(&mut self, observer: impl OptObserver + 'static) {
        self.observers.push(Box::new(observer));
    }

    /// Set iteration metrics for all observers.
    ///
    /// Calls `set_iteration_metrics()` on each registered observer. This should
    /// be called before `notify()` to provide optimization metrics.
    ///
    /// # Arguments
    ///
    /// * `cost` - Current cost function value
    /// * `gradient_norm` - L2 norm of the gradient vector
    /// * `damping` - Damping parameter (may be None)
    /// * `step_norm` - L2 norm of the parameter update step
    /// * `step_quality` - Step quality metric (may be None)
    #[inline]
    pub fn set_iteration_metrics(
        &self,
        cost: f64,
        gradient_norm: f64,
        damping: Option<f64>,
        step_norm: f64,
        step_quality: Option<f64>,
    ) {
        for observer in &self.observers {
            observer.set_iteration_metrics(cost, gradient_norm, damping, step_norm, step_quality);
        }
    }

    /// Set matrix data for all observers.
    ///
    /// Calls `set_matrix_data()` on each registered observer. This should
    /// be called before `notify()` to provide matrix data for visualization.
    ///
    /// # Arguments
    ///
    /// * `hessian` - Sparse Hessian matrix
    /// * `gradient` - Gradient vector
    #[inline]
    pub fn set_matrix_data(
        &self,
        hessian: Option<sparse::SparseColMat<usize, f64>>,
        gradient: Option<Mat<f64>>,
    ) {
        for observer in &self.observers {
            observer.set_matrix_data(hessian.clone(), gradient.clone());
        }
    }

    /// Notify all observers with current optimization state.
    ///
    /// Calls `on_step()` on each registered observer in order. If no observers
    /// are registered, this is a no-op with zero overhead.
    ///
    /// # Arguments
    ///
    /// * `values` - Current variable values
    /// * `iteration` - Current iteration number
    ///
    /// # Examples
    ///
    /// ```no_run
    /// use apex_solver::observers::OptObserverVec;
    /// use apex_solver::core::variable::ManifoldVariable;
    /// use apex_solver::core::VarKey;
    /// use slotmap::SlotMap;
    ///
    /// let observers = OptObserverVec::new();
    /// let values: SlotMap<VarKey, Box<dyn ManifoldVariable>> = SlotMap::with_key();
    ///
    /// // Notify all observers (safe even if empty)
    /// observers.notify(&values, 0);
    /// ```
    #[inline]
    pub fn notify(&self, values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, iteration: usize) {
        for observer in &self.observers {
            observer.on_step(values, iteration);
        }
    }

    /// Notify all observers that optimization is complete.
    ///
    /// Calls `on_optimization_complete()` on each registered observer. This should
    /// be called once at the end of optimization, after all iterations are done.
    ///
    /// # Arguments
    ///
    /// * `values` - Final optimized variable values
    /// * `iterations` - Total number of iterations performed
    ///
    /// # Examples
    ///
    /// ```no_run
    /// use apex_solver::observers::OptObserverVec;
    /// use apex_solver::core::variable::ManifoldVariable;
    /// use apex_solver::core::VarKey;
    /// use slotmap::SlotMap;
    ///
    /// let observers = OptObserverVec::new();
    /// let values: SlotMap<VarKey, Box<dyn ManifoldVariable>> = SlotMap::with_key();
    ///
    /// // Notify all observers that optimization is complete
    /// observers.notify_complete(&values, 50);
    /// ```
    #[inline]
    pub fn notify_complete(
        &self,
        values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>,
        iterations: usize,
    ) {
        for observer in &self.observers {
            observer.on_optimization_complete(values, iterations);
        }
    }

    /// Check if any observers are registered.
    ///
    /// Useful for conditional logic or debugging.
    #[inline]
    pub fn is_empty(&self) -> bool {
        self.observers.is_empty()
    }

    /// Get the number of registered observers.
    #[inline]
    pub fn len(&self) -> usize {
        self.observers.len()
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::error::ErrorLogging;
    use std::sync::{Arc, Mutex};

    fn empty_vars() -> SlotMap<VarKey, Box<dyn ManifoldVariable>> {
        SlotMap::with_key()
    }

    #[derive(Clone)]
    struct TestObserver {
        calls: Arc<Mutex<Vec<usize>>>,
    }

    impl OptObserver for TestObserver {
        fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, iteration: usize) {
            // In test code, we log and ignore mutex poisoning errors since they indicate test bugs
            if let Ok(mut guard) = self.calls.lock().map_err(|e| {
                ObserverError::MutexPoisoned {
                    context: "TestObserver::on_step".to_string(),
                    reason: e.to_string(),
                }
                .log()
            }) {
                guard.push(iteration);
            }
        }
    }

    #[test]
    fn test_empty_observers() {
        let observers = OptObserverVec::new();
        assert!(observers.is_empty());
        assert_eq!(observers.len(), 0);

        // Should not panic with no observers
        observers.notify(&empty_vars(), 0);
    }

    #[test]
    fn test_single_observer() -> Result<(), ObserverError> {
        let calls = Arc::new(Mutex::new(Vec::new()));
        let observer = TestObserver {
            calls: calls.clone(),
        };

        let mut observers = OptObserverVec::new();
        observers.add(observer);

        assert_eq!(observers.len(), 1);

        observers.notify(&empty_vars(), 0);
        observers.notify(&empty_vars(), 1);
        observers.notify(&empty_vars(), 2);

        let guard = calls.lock().map_err(|e| {
            ObserverError::MutexPoisoned {
                context: "test_single_observer".to_string(),
                reason: e.to_string(),
            }
            .log()
        })?;
        assert_eq!(*guard, vec![0, 1, 2]);
        Ok(())
    }

    #[test]
    fn test_multiple_observers() -> Result<(), ObserverError> {
        let calls1 = Arc::new(Mutex::new(Vec::new()));
        let calls2 = Arc::new(Mutex::new(Vec::new()));

        let observer1 = TestObserver {
            calls: calls1.clone(),
        };
        let observer2 = TestObserver {
            calls: calls2.clone(),
        };

        let mut observers = OptObserverVec::new();
        observers.add(observer1);
        observers.add(observer2);

        assert_eq!(observers.len(), 2);

        observers.notify(&empty_vars(), 5);

        let guard1 = calls1.lock().map_err(|e| {
            ObserverError::MutexPoisoned {
                context: "test_multiple_observers (calls1)".to_string(),
                reason: e.to_string(),
            }
            .log()
        })?;
        assert_eq!(*guard1, vec![5]);

        let guard2 = calls2.lock().map_err(|e| {
            ObserverError::MutexPoisoned {
                context: "test_multiple_observers (calls2)".to_string(),
                reason: e.to_string(),
            }
            .log()
        })?;
        assert_eq!(*guard2, vec![5]);
        Ok(())
    }

    // -------------------------------------------------------------------------
    // ObserverError Display — one per variant
    // -------------------------------------------------------------------------

    #[test]
    fn test_observer_error_rerun_initialization_display() {
        let e = ObserverError::RerunInitialization("init fail".into());
        assert!(e.to_string().contains("init fail"));
    }

    #[test]
    fn test_observer_error_viewer_spawn_failed_display() {
        let e = ObserverError::ViewerSpawnFailed("spawn fail".into());
        assert!(e.to_string().contains("spawn fail"));
    }

    #[test]
    fn test_observer_error_recording_save_failed_display() {
        let e = ObserverError::RecordingSaveFailed {
            path: "/tmp/out.rrd".into(),
            reason: "disk full".into(),
        };
        let s = e.to_string();
        assert!(s.contains("/tmp/out.rrd"), "{s}");
        assert!(s.contains("disk full"), "{s}");
    }

    #[test]
    fn test_observer_error_logging_failed_display() {
        let e = ObserverError::LoggingFailed {
            entity_path: "world/points".into(),
            reason: "timeout".into(),
        };
        let s = e.to_string();
        assert!(s.contains("world/points"), "{s}");
        assert!(s.contains("timeout"), "{s}");
    }

    #[test]
    fn test_observer_error_matrix_visualization_failed_display() {
        let e = ObserverError::MatrixVisualizationFailed("bad dims".into());
        assert!(e.to_string().contains("bad dims"));
    }

    #[test]
    fn test_observer_error_tensor_conversion_failed_display() {
        let e = ObserverError::TensorConversionFailed("nan values".into());
        assert!(e.to_string().contains("nan values"));
    }

    #[test]
    fn test_observer_error_invalid_state_display() {
        let e = ObserverError::InvalidState("stream closed".into());
        assert!(e.to_string().contains("stream closed"));
    }

    #[test]
    fn test_observer_error_mutex_poisoned_display() {
        let e = ObserverError::MutexPoisoned {
            context: "on_step".into(),
            reason: "thread panicked".into(),
        };
        let s = e.to_string();
        assert!(s.contains("on_step"), "{s}");
        assert!(s.contains("thread panicked"), "{s}");
    }

    // -------------------------------------------------------------------------
    // log() / log_with_source() return self
    // -------------------------------------------------------------------------

    #[test]
    fn test_observer_error_log_returns_self() {
        let e = ObserverError::InvalidState("log_test".into());
        let returned = e.log();
        assert!(returned.to_string().contains("log_test"));
    }

    #[test]
    fn test_observer_error_log_with_source_returns_self() {
        let e = ObserverError::MatrixVisualizationFailed("src_test".into());
        let source = std::io::Error::other("src");
        let returned = e.log_with_source(source);
        assert!(returned.to_string().contains("src_test"));
    }

    // -------------------------------------------------------------------------
    // OptObserverVec — set_iteration_metrics, set_matrix_data, notify_complete
    // -------------------------------------------------------------------------

    #[test]
    fn test_set_iteration_metrics_no_panic() {
        let mut observers = OptObserverVec::new();
        observers.add(TestObserver {
            calls: Arc::new(Mutex::new(Vec::new())),
        });
        // Should not panic whether empty or not
        observers.set_iteration_metrics(1.5, 1e-3, Some(1e-4), 0.01, Some(0.9));
    }

    #[test]
    fn test_set_iteration_metrics_empty_no_panic() {
        let observers = OptObserverVec::new();
        observers.set_iteration_metrics(0.0, 0.0, None, 0.0, None);
    }

    #[test]
    fn test_set_matrix_data_no_panic() {
        let mut observers = OptObserverVec::new();
        observers.add(TestObserver {
            calls: Arc::new(Mutex::new(Vec::new())),
        });
        // Pass None for both hessian and gradient
        observers.set_matrix_data(None, None);
    }

    // Observer that counts on_optimization_complete calls
    #[derive(Clone)]
    struct CompleteObserver {
        complete_calls: Arc<Mutex<usize>>,
    }

    impl OptObserver for CompleteObserver {
        fn on_step(&self, _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>, _iteration: usize) {
        }

        fn on_optimization_complete(
            &self,
            _values: &SlotMap<VarKey, Box<dyn ManifoldVariable>>,
            _iterations: usize,
        ) {
            if let Ok(mut guard) = self.complete_calls.lock() {
                *guard += 1;
            }
        }
    }

    #[test]
    fn test_notify_complete_calls_on_optimization_complete() {
        let complete_calls = Arc::new(Mutex::new(0usize));
        let observer = CompleteObserver {
            complete_calls: complete_calls.clone(),
        };

        let mut observers = OptObserverVec::new();
        observers.add(observer);
        observers.notify_complete(&empty_vars(), 10);

        let count = *complete_calls.lock().unwrap_or_else(|e| e.into_inner());
        assert_eq!(count, 1);
    }

    #[test]
    fn test_notify_complete_empty_no_panic() {
        let observers = OptObserverVec::new();
        observers.notify_complete(&empty_vars(), 5);
    }

    #[test]
    fn test_default_trait_methods_no_panic() {
        // TestObserver only overrides on_step; the default impls are exercised here
        let observer = TestObserver {
            calls: Arc::new(Mutex::new(Vec::new())),
        };
        // Default implementations should be no-ops
        observer.set_iteration_metrics(1.0, 1e-3, None, 0.0, None);
        observer.set_matrix_data(None, None);
        observer.on_optimization_complete(&empty_vars(), 5);
    }
}