rustytorch/crates/training/rtx-model-merging/src/error.rs

//! Error handling for model merging operations

use thiserror::Error;

/// Result type for model merging operations
pub type Result<T> = std::result::Result<T, MergeError>;

/// Comprehensive error types for model merging
#[derive(Error, Debug)]
pub enum MergeError {
    #[error("Model loading failed: {message}")]
    ModelLoadError { message: String },

    #[error("Model compatibility error: {details}")]
    CompatibilityError { details: String },

    #[error("Parameter mismatch: expected {expected}, found {found}")]
    ParameterMismatch { expected: String, found: String },

    #[error("Architecture incompatibility: {model1} vs {model2}")]
    ArchitectureIncompatible { model1: String, model2: String },

    #[error("Merge algorithm failed: {algorithm} - {reason}")]
    AlgorithmError { algorithm: String, reason: String },

    #[error("Expert selection failed: {reason}")]
    ExpertSelectionError { reason: String },

    #[error("Conflict resolution failed: {conflicts} conflicts unresolved")]
    ConflictResolutionError { conflicts: usize },

    #[error("Validation failed: {} validation checks failed", checks.len())]
    ValidationError { checks: Vec<String> },

    #[error("Performance evaluation error: {metric} computation failed")]
    EvaluationError { metric: String },

    #[error("Configuration error: {field} - {message}")]
    ConfigError { field: String, message: String },

    #[error("Memory allocation error: {size_mb}MB required")]
    MemoryError { size_mb: usize },

    #[error("GPU operation failed: {operation}")]
    GpuError { operation: String },

    #[error("Numerical instability detected: {operation}")]
    NumericalError { operation: String },

    #[error("File I/O error: {path}")]
    IoError {
        path: String,
        #[source]
        source: std::io::Error,
    },

    #[error("Serialization error")]
    SerializationError(#[from] bincode::Error),

    #[error("JSON error")]
    JsonError(#[from] serde_json::Error),

    #[error("TOML error")]
    TomlError(#[from] toml::de::Error),

    #[error("Async runtime error")]
    AsyncError(#[from] tokio::task::JoinError),

    #[error("Thread pool error")]
    ThreadPoolError(String),

    #[error("Timeout error: operation took longer than {timeout_ms}ms")]
    TimeoutError { timeout_ms: u64 },

    #[error("Resource exhaustion: {resource}")]
    ResourceExhausted { resource: String },

    #[error("Model format not supported: {format}")]
    UnsupportedFormat { format: String },

    #[error("Feature not implemented: {feature}")]
    NotImplemented { feature: String },

    #[error("Internal error: {message}")]
    Internal { message: String },
}

impl MergeError {
    /// Create a model loading error
    pub fn model_load<S: Into<String>>(message: S) -> Self {
        Self::ModelLoadError {
            message: message.into(),
        }
    }

    /// Create a compatibility error
    pub fn compatibility<S: Into<String>>(details: S) -> Self {
        Self::CompatibilityError {
            details: details.into(),
        }
    }

    /// Create a parameter mismatch error
    pub fn parameter_mismatch<S: Into<String>>(expected: S, found: S) -> Self {
        Self::ParameterMismatch {
            expected: expected.into(),
            found: found.into(),
        }
    }

    /// Create an architecture incompatibility error
    pub fn architecture_incompatible<S: Into<String>>(model1: S, model2: S) -> Self {
        Self::ArchitectureIncompatible {
            model1: model1.into(),
            model2: model2.into(),
        }
    }

    /// Create an algorithm error
    pub fn algorithm<S: Into<String>>(algorithm: S, reason: S) -> Self {
        Self::AlgorithmError {
            algorithm: algorithm.into(),
            reason: reason.into(),
        }
    }

    /// Create an expert selection error
    pub fn expert_selection<S: Into<String>>(reason: S) -> Self {
        Self::ExpertSelectionError {
            reason: reason.into(),
        }
    }

    /// Create a conflict resolution error
    pub fn conflict_resolution(conflicts: usize) -> Self {
        Self::ConflictResolutionError { conflicts }
    }

    /// Create a validation error
    pub fn validation(checks: Vec<String>) -> Self {
        Self::ValidationError { checks }
    }

    /// Create an evaluation error
    pub fn evaluation<S: Into<String>>(metric: S) -> Self {
        Self::EvaluationError {
            metric: metric.into(),
        }
    }

    /// Create a configuration error
    pub fn config<S: Into<String>>(field: S, message: S) -> Self {
        Self::ConfigError {
            field: field.into(),
            message: message.into(),
        }
    }

    /// Create a memory error
    pub fn memory(size_mb: usize) -> Self {
        Self::MemoryError { size_mb }
    }

    /// Create a GPU error
    pub fn gpu<S: Into<String>>(operation: S) -> Self {
        Self::GpuError {
            operation: operation.into(),
        }
    }

    /// Create a numerical error
    pub fn numerical<S: Into<String>>(operation: S) -> Self {
        Self::NumericalError {
            operation: operation.into(),
        }
    }

    /// Create an I/O error
    pub fn io<S: Into<String>>(path: S, source: std::io::Error) -> Self {
        Self::IoError {
            path: path.into(),
            source,
        }
    }

    /// Create a timeout error
    pub fn timeout(timeout_ms: u64) -> Self {
        Self::TimeoutError { timeout_ms }
    }

    /// Create a resource exhaustion error
    pub fn resource_exhausted<S: Into<String>>(resource: S) -> Self {
        Self::ResourceExhausted {
            resource: resource.into(),
        }
    }

    /// Create an unsupported format error
    pub fn unsupported_format<S: Into<String>>(format: S) -> Self {
        Self::UnsupportedFormat {
            format: format.into(),
        }
    }

    /// Create a not implemented error
    pub fn not_implemented<S: Into<String>>(feature: S) -> Self {
        Self::NotImplemented {
            feature: feature.into(),
        }
    }

    /// Create an internal error
    pub fn internal<S: Into<String>>(message: S) -> Self {
        Self::Internal {
            message: message.into(),
        }
    }

    /// Check if this error is recoverable
    pub fn is_recoverable(&self) -> bool {
        match self {
            Self::TimeoutError { .. } => true,
            Self::MemoryError { .. } => true,
            Self::GpuError { .. } => true,
            Self::ThreadPoolError(_) => true,
            Self::ResourceExhausted { .. } => true,
            _ => false,
        }
    }

    /// Get the error category
    pub fn category(&self) -> ErrorCategory {
        match self {
            Self::ModelLoadError { .. } => ErrorCategory::Loading,
            Self::CompatibilityError { .. } => ErrorCategory::Compatibility,
            Self::ParameterMismatch { .. } => ErrorCategory::Compatibility,
            Self::ArchitectureIncompatible { .. } => ErrorCategory::Compatibility,
            Self::AlgorithmError { .. } => ErrorCategory::Algorithm,
            Self::ExpertSelectionError { .. } => ErrorCategory::ExpertSelection,
            Self::ConflictResolutionError { .. } => ErrorCategory::ConflictResolution,
            Self::ValidationError { .. } => ErrorCategory::Validation,
            Self::EvaluationError { .. } => ErrorCategory::Evaluation,
            Self::ConfigError { .. } => ErrorCategory::Configuration,
            Self::MemoryError { .. } => ErrorCategory::Resource,
            Self::GpuError { .. } => ErrorCategory::Resource,
            Self::NumericalError { .. } => ErrorCategory::Numerical,
            Self::IoError { .. } => ErrorCategory::IO,
            Self::SerializationError(_) => ErrorCategory::Serialization,
            Self::JsonError(_) => ErrorCategory::Serialization,
            Self::TomlError(_) => ErrorCategory::Serialization,
            Self::AsyncError(_) => ErrorCategory::Runtime,
            Self::ThreadPoolError(_) => ErrorCategory::Runtime,
            Self::TimeoutError { .. } => ErrorCategory::Runtime,
            Self::ResourceExhausted { .. } => ErrorCategory::Resource,
            Self::UnsupportedFormat { .. } => ErrorCategory::Format,
            Self::NotImplemented { .. } => ErrorCategory::NotImplemented,
            Self::Internal { .. } => ErrorCategory::Internal,
        }
    }
}

/// Error categories for classification and handling
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum ErrorCategory {
    Loading,
    Compatibility,
    Algorithm,
    ExpertSelection,
    ConflictResolution,
    Validation,
    Evaluation,
    Configuration,
    Resource,
    Numerical,
    IO,
    Serialization,
    Runtime,
    Format,
    NotImplemented,
    Internal,
}

impl ErrorCategory {
    /// Get a human-readable description of the error category
    pub fn description(&self) -> &'static str {
        match self {
            Self::Loading => "Model loading and parsing errors",
            Self::Compatibility => "Model compatibility and architecture mismatches",
            Self::Algorithm => "Merge algorithm execution errors",
            Self::ExpertSelection => "Expert selection and routing errors",
            Self::ConflictResolution => "Parameter conflict resolution errors",
            Self::Validation => "Model and parameter validation errors",
            Self::Evaluation => "Performance evaluation and metric computation errors",
            Self::Configuration => "Configuration parsing and validation errors",
            Self::Resource => "Memory, GPU, and other resource errors",
            Self::Numerical => "Numerical stability and computation errors",
            Self::IO => "File system and network I/O errors",
            Self::Serialization => "Data serialization and deserialization errors",
            Self::Runtime => "Async runtime and thread pool errors",
            Self::Format => "Unsupported or invalid format errors",
            Self::NotImplemented => "Features not yet implemented",
            Self::Internal => "Internal library errors",
        }
    }

    /// Check if errors in this category are typically user-fixable
    pub fn is_user_fixable(&self) -> bool {
        match self {
            Self::Loading => true,
            Self::Compatibility => true,
            Self::Configuration => true,
            Self::IO => true,
            Self::Format => true,
            _ => false,
        }
    }
}

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

    #[test]
    fn test_error_creation() {
        let err = MergeError::model_load("Test error");
        assert!(matches!(err, MergeError::ModelLoadError { .. }));
    }

    #[test]
    fn test_error_categories() {
        let err = MergeError::compatibility("Test");
        assert_eq!(err.category(), ErrorCategory::Compatibility);
        assert!(err.category().is_user_fixable());
    }

    #[test]
    fn test_recoverable_errors() {
        let timeout_err = MergeError::timeout(1000);
        assert!(timeout_err.is_recoverable());

        let internal_err = MergeError::internal("Test");
        assert!(!internal_err.is_recoverable());
    }

    #[test]
    fn test_error_display() {
        let err = MergeError::parameter_mismatch("expected", "found");
        let display = format!("{}", err);
        assert!(display.contains("expected"));
        assert!(display.contains("found"));
    }
}