helios_sof/

lib.rs

//! # SQL-on-FHIR Implementation
//!
//! This crate provides a complete implementation of the [SQL-on-FHIR
//! specification](https://sql-on-fhir.org/ig/latest),
//! enabling the transformation of FHIR resources into tabular data using declarative
//! ViewDefinitions. It supports all major FHIR versions (R4, R4B, R5, R6) through
//! a version-agnostic abstraction layer.

//!
//! There are three consumers of this crate:
//! - [sof_cli](../sof_cli/index.html) - A command-line interface for the SQL-on-FHIR implementation,
//!   allowing users to execute ViewDefinition transformations on FHIR Bundle resources
//!   and output the results in various formats.
//! - [sof_server](../sof_server/index.html) - A stateless HTTP server implementation for the SQL-on-FHIR specification,
//!   enabling HTTP-based access to ViewDefinition transformation capabilities.
//! - [hfs](../hfs/index.html) - The full featured Helios FHIR Server.
//!
//! ## Architecture
//!
//! The SOF crate is organized around these key components:
//! - **Version-agnostic enums** ([`SofViewDefinition`], [`SofBundle`]): Multi-version containers
//! - **Processing engine** ([`run_view_definition`]): Core transformation logic
//! - **Output formats** ([`ContentType`]): Support for CSV, JSON, NDJSON, and Parquet
//! - **Trait abstractions** ([`ViewDefinitionTrait`], [`BundleTrait`]): Version independence
//!
//! ## Key Features
//!
//! - **Multi-version FHIR support**: Works with R4, R4B, R5, and R6 resources
//! - **FHIRPath evaluation**: Complex path expressions for data extraction
//! - **forEach iteration**: Supports flattening of nested FHIR structures
//! - **unionAll operations**: Combines multiple select statements
//! - **Collection handling**: Proper array serialization for multi-valued fields
//! - **Output formats**: CSV (with/without headers), JSON, NDJSON, Parquet support
//!
//! ## Usage Example
//!
//! ```rust
//! # #[cfg(not(target_os = "windows"))]
//! # {
//! use helios_sof::{SofViewDefinition, SofBundle, ContentType, run_view_definition};
//! use helios_fhir::FhirVersion;
//!
//! # #[cfg(feature = "R4")]
//! # {
//! // Parse a ViewDefinition and Bundle from JSON
//! let view_definition_json = r#"{
//!     "resourceType": "ViewDefinition",
//!     "status": "active",
//!     "resource": "Patient",
//!     "select": [{
//!         "column": [{
//!             "name": "id",
//!             "path": "id"
//!         }, {
//!             "name": "name",
//!             "path": "name.family"
//!         }]
//!     }]
//! }"#;
//!
//! let bundle_json = r#"{
//!     "resourceType": "Bundle",
//!     "type": "collection",
//!     "entry": [{
//!         "resource": {
//!             "resourceType": "Patient",
//!             "id": "example",
//!             "name": [{
//!                 "family": "Doe",
//!                 "given": ["John"]
//!             }]
//!         }
//!     }]
//! }"#;
//!
//! let view_definition: helios_fhir::r4::ViewDefinition = serde_json::from_str(view_definition_json)?;
//! let bundle: helios_fhir::r4::Bundle = serde_json::from_str(bundle_json)?;
//!
//! // Wrap in version-agnostic containers
//! let sof_view = SofViewDefinition::R4(view_definition);
//! let sof_bundle = SofBundle::R4(bundle);
//!
//! // Transform to CSV with headers
//! let csv_output = run_view_definition(
//!     sof_view,
//!     sof_bundle,
//!     ContentType::CsvWithHeader
//! )?;
//!
//! // Check the CSV output
//! let csv_string = String::from_utf8(csv_output)?;
//! assert!(csv_string.contains("id,name"));
//! // CSV values are quoted
//! assert!(csv_string.contains("example") && csv_string.contains("Doe"));
//! # }
//! # }
//! # Ok::<(), Box<dyn std::error::Error>>(())
//! ```
//!
//! ## Advanced Features
//!
//! ### forEach Iteration
//!
//! ViewDefinitions can iterate over collections using `forEach` and `forEachOrNull`:
//!
//! ```json
//! {
//!   "select": [{
//!     "forEach": "name",
//!     "column": [{
//!       "name": "family_name",
//!       "path": "family"
//!     }]
//!   }]
//! }
//! ```
//!
//! ### Constants and Variables
//!
//! Define reusable values in ViewDefinitions:
//!
//! ```json
//! {
//!   "constant": [{
//!     "name": "system",
//!     "valueString": "http://loinc.org"
//!   }],
//!   "select": [{
//!     "where": [{
//!       "path": "code.coding.system = %system"
//!     }]
//!   }]
//! }
//! ```
//!
//! ### Where Clauses
//!
//! Filter resources using FHIRPath expressions:
//!
//! ```json
//! {
//!   "where": [{
//!     "path": "active = true"
//!   }, {
//!     "path": "birthDate.exists()"
//!   }]
//! }
//! ```
//!
//! ## Error Handling
//!
//! The crate provides comprehensive error handling through [`SofError`]:
//!
//! ```rust,no_run
//! use helios_sof::{SofError, SofViewDefinition, SofBundle, ContentType, run_view_definition};
//!
//! # let view = SofViewDefinition::R4(helios_fhir::r4::ViewDefinition::default());
//! # let bundle = SofBundle::R4(helios_fhir::r4::Bundle::default());
//! # let content_type = ContentType::Json;
//! match run_view_definition(view, bundle, content_type) {
//!     Ok(output) => {
//!         // Process successful transformation
//!     },
//!     Err(SofError::InvalidViewDefinition(msg)) => {
//!         eprintln!("ViewDefinition validation failed: {}", msg);
//!     },
//!     Err(SofError::FhirPathError(msg)) => {
//!         eprintln!("FHIRPath evaluation failed: {}", msg);
//!     },
//!     Err(e) => {
//!         eprintln!("Other error: {}", e);
//!     }
//! }
//! ```
//! ## Feature Flags
//!
//! Enable support for specific FHIR versions:
//! - `R4`: FHIR 4.0.1 support
//! - `R4B`: FHIR 4.3.0 support
//! - `R5`: FHIR 5.0.0 support
//! - `R6`: FHIR 6.0.0 support

pub mod compartment;
pub mod constants;
pub mod data_source;
pub mod fhir_format;
pub mod params;
pub mod parquet_schema;
pub mod reference_collector;
pub mod remote_fetch;
pub mod remote_resolver;
pub mod sqlquery;
pub mod traits;

pub use compartment::{resolve_group_members_to_patient_refs, resource_in_patient_compartment};
pub use constants::{ConstantValue, parse_constant_from_json};
pub use params::{
    ExtractedRunParams, body_has_view_definition, extract_run_params_from_json, split_csv_refs,
};
pub use remote_fetch::{RemoteResolver, prefetch_external_resources};
pub use remote_resolver::{
    AllowedBaseUrl, DenyReason, FetchDecision, RemoteResolveConfig, is_blocked_address,
    is_disallowed_ip, parse_allowed_base_urls,
};
pub use sqlquery::{
    BoundParam, ColumnFhirType, DependsOnView, InMemorySqlEngine, LibraryParameter, QueryResult,
    SqlQueryError, SqlQueryLibrary, SqlQueryRunParams, TableSchema, bind_supplied_params,
    extract_sqlquery_params_from_json, format_fhir_parameters, parse_sqlquery_library,
};

use chrono::{DateTime, Utc};
use helios_fhirpath::{EvaluationContext, EvaluationResult, evaluate_expression};
use rayon::prelude::*;
use serde::{Deserialize, Serialize};
use std::collections::HashMap;
use std::io::{BufRead, Write};
use thiserror::Error;
use traits::*;

// Re-export commonly used types and traits for easier access
pub use helios_fhir::FhirVersion;
pub use traits::{BundleTrait, ResourceTrait, ViewDefinitionTrait};

/// Multi-version ViewDefinition container supporting version-agnostic operations.
///
/// This enum provides a unified interface for working with ViewDefinition resources
/// across different FHIR specification versions. It enables applications to handle
/// multiple FHIR versions simultaneously while maintaining type safety.
///
/// # Supported Versions
///
/// - **R4**: FHIR 4.0.1 ViewDefinition (normative)
/// - **R4B**: FHIR 4.3.0 ViewDefinition (ballot)
/// - **R5**: FHIR 5.0.0 ViewDefinition (ballot)
/// - **R6**: FHIR 6.0.0 ViewDefinition (draft)
///
/// # Examples
///
/// ```rust
/// use helios_sof::{SofViewDefinition, ContentType};
/// # #[cfg(feature = "R4")]
/// use helios_fhir::r4::ViewDefinition;
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Parse from JSON
/// let json = r#"{
///     "resourceType": "ViewDefinition",
///     "resource": "Patient",
///     "select": [{
///         "column": [{
///             "name": "id",
///             "path": "id"
///         }]
///     }]
/// }"#;
///
/// let view_def: ViewDefinition = serde_json::from_str(json)?;
/// let sof_view = SofViewDefinition::R4(view_def);
///
/// // Check version
/// assert_eq!(sof_view.version(), helios_fhir::FhirVersion::R4);
/// # }
/// # Ok::<(), Box<dyn std::error::Error>>(())
/// ```
#[derive(Debug, Clone)]
pub enum SofViewDefinition {
    #[cfg(feature = "R4")]
    R4(helios_fhir::r4::ViewDefinition),
    #[cfg(feature = "R4B")]
    R4B(helios_fhir::r4b::ViewDefinition),
    #[cfg(feature = "R5")]
    R5(helios_fhir::r5::ViewDefinition),
    #[cfg(feature = "R6")]
    R6(helios_fhir::r6::ViewDefinition),
}

impl SofViewDefinition {
    /// Returns the FHIR specification version of this ViewDefinition.
    ///
    /// This method provides version detection for multi-version applications,
    /// enabling version-specific processing logic and compatibility checks.
    ///
    /// # Returns
    ///
    /// The `FhirVersion` enum variant corresponding to this ViewDefinition's specification.
    ///
    /// # Examples
    ///
    /// ```rust
    /// use helios_sof::SofViewDefinition;
    /// use helios_fhir::FhirVersion;
    ///
    /// # #[cfg(feature = "R5")]
    /// # {
    /// # let view_def = helios_fhir::r5::ViewDefinition::default();
    /// let sof_view = SofViewDefinition::R5(view_def);
    /// assert_eq!(sof_view.version(), helios_fhir::FhirVersion::R5);
    /// # }
    /// ```
    pub fn version(&self) -> helios_fhir::FhirVersion {
        match self {
            #[cfg(feature = "R4")]
            SofViewDefinition::R4(_) => helios_fhir::FhirVersion::R4,
            #[cfg(feature = "R4B")]
            SofViewDefinition::R4B(_) => helios_fhir::FhirVersion::R4B,
            #[cfg(feature = "R5")]
            SofViewDefinition::R5(_) => helios_fhir::FhirVersion::R5,
            #[cfg(feature = "R6")]
            SofViewDefinition::R6(_) => helios_fhir::FhirVersion::R6,
        }
    }
}

/// Multi-version Bundle container supporting version-agnostic operations.
///
/// This enum provides a unified interface for working with FHIR Bundle resources
/// across different FHIR specification versions. Bundles contain the actual FHIR
/// resources that will be processed by ViewDefinitions.
///
/// # Supported Versions
///
/// - **R4**: FHIR 4.0.1 Bundle (normative)
/// - **R4B**: FHIR 4.3.0 Bundle (ballot)
/// - **R5**: FHIR 5.0.0 Bundle (ballot)
/// - **R6**: FHIR 6.0.0 Bundle (draft)
///
/// # Examples
///
/// ```rust
/// # #[cfg(not(target_os = "windows"))]
/// # {
/// use helios_sof::SofBundle;
/// # #[cfg(feature = "R4")]
/// use helios_fhir::r4::Bundle;
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Parse from JSON
/// let json = r#"{
///     "resourceType": "Bundle",
///     "type": "collection",
///     "entry": [{
///         "resource": {
///             "resourceType": "Patient",
///             "id": "example"
///         }
///     }]
/// }"#;
///
/// let bundle: Bundle = serde_json::from_str(json)?;
/// let sof_bundle = SofBundle::R4(bundle);
///
/// // Check version compatibility
/// assert_eq!(sof_bundle.version(), helios_fhir::FhirVersion::R4);
/// # }
/// # }
/// # Ok::<(), Box<dyn std::error::Error>>(())
/// ```
#[derive(Debug, Clone)]
pub enum SofBundle {
    #[cfg(feature = "R4")]
    R4(helios_fhir::r4::Bundle),
    #[cfg(feature = "R4B")]
    R4B(helios_fhir::r4b::Bundle),
    #[cfg(feature = "R5")]
    R5(helios_fhir::r5::Bundle),
    #[cfg(feature = "R6")]
    R6(helios_fhir::r6::Bundle),
}

impl SofBundle {
    /// Returns the FHIR specification version of this Bundle.
    ///
    /// This method provides version detection for multi-version applications,
    /// ensuring that ViewDefinitions and Bundles use compatible FHIR versions.
    ///
    /// # Returns
    ///
    /// The `FhirVersion` enum variant corresponding to this Bundle's specification.
    ///
    /// # Examples
    ///
    /// ```rust
    /// use helios_sof::SofBundle;
    /// use helios_fhir::FhirVersion;
    ///
    /// # #[cfg(feature = "R4")]
    /// # {
    /// # let bundle = helios_fhir::r4::Bundle::default();
    /// let sof_bundle = SofBundle::R4(bundle);
    /// assert_eq!(sof_bundle.version(), helios_fhir::FhirVersion::R4);
    /// # }
    /// ```
    pub fn version(&self) -> helios_fhir::FhirVersion {
        match self {
            #[cfg(feature = "R4")]
            SofBundle::R4(_) => helios_fhir::FhirVersion::R4,
            #[cfg(feature = "R4B")]
            SofBundle::R4B(_) => helios_fhir::FhirVersion::R4B,
            #[cfg(feature = "R5")]
            SofBundle::R5(_) => helios_fhir::FhirVersion::R5,
            #[cfg(feature = "R6")]
            SofBundle::R6(_) => helios_fhir::FhirVersion::R6,
        }
    }
}

/// Multi-version CapabilityStatement container supporting version-agnostic operations.
///
/// This enum provides a unified interface for working with CapabilityStatement resources
/// across different FHIR specification versions. It enables applications to handle
/// multiple FHIR versions simultaneously while maintaining type safety.
///
/// # Supported Versions
///
/// - **R4**: FHIR 4.0.1 CapabilityStatement (normative)
/// - **R4B**: FHIR 4.3.0 CapabilityStatement (ballot)
/// - **R5**: FHIR 5.0.0 CapabilityStatement (ballot)
/// - **R6**: FHIR 6.0.0 CapabilityStatement (draft)
#[derive(Debug, Clone, Serialize, Deserialize)]
#[serde(untagged)]
pub enum SofCapabilityStatement {
    #[cfg(feature = "R4")]
    R4(helios_fhir::r4::CapabilityStatement),
    #[cfg(feature = "R4B")]
    R4B(helios_fhir::r4b::CapabilityStatement),
    #[cfg(feature = "R5")]
    R5(helios_fhir::r5::CapabilityStatement),
    #[cfg(feature = "R6")]
    R6(helios_fhir::r6::CapabilityStatement),
}

impl SofCapabilityStatement {
    /// Returns the FHIR specification version of this CapabilityStatement.
    pub fn version(&self) -> helios_fhir::FhirVersion {
        match self {
            #[cfg(feature = "R4")]
            SofCapabilityStatement::R4(_) => helios_fhir::FhirVersion::R4,
            #[cfg(feature = "R4B")]
            SofCapabilityStatement::R4B(_) => helios_fhir::FhirVersion::R4B,
            #[cfg(feature = "R5")]
            SofCapabilityStatement::R5(_) => helios_fhir::FhirVersion::R5,
            #[cfg(feature = "R6")]
            SofCapabilityStatement::R6(_) => helios_fhir::FhirVersion::R6,
        }
    }
}

/// Type alias for the version-independent Parameters container.
///
/// This alias provides backward compatibility while using the unified
/// VersionIndependentParameters from the helios_fhir crate.
pub type SofParameters = helios_fhir::VersionIndependentParameters;

/// Comprehensive error type for SQL-on-FHIR operations.
///
/// This enum covers all possible error conditions that can occur during
/// ViewDefinition processing, from validation failures to output formatting issues.
/// Each variant provides specific context about the error to aid in debugging.
///
/// # Error Categories
///
/// - **Validation**: ViewDefinition structure and logic validation
/// - **Evaluation**: FHIRPath expression evaluation failures
/// - **I/O**: File and serialization operations
/// - **Format**: Output format conversion issues
///
/// # Examples
///
/// ```rust,no_run
/// use helios_sof::{SofError, SofViewDefinition, SofBundle, ContentType, run_view_definition};
///
/// # let view = SofViewDefinition::R4(helios_fhir::r4::ViewDefinition::default());
/// # let bundle = SofBundle::R4(helios_fhir::r4::Bundle::default());
/// # let content_type = ContentType::Json;
/// match run_view_definition(view, bundle, content_type) {
///     Ok(output) => {
///         println!("Transformation successful");
///     },
///     Err(SofError::InvalidViewDefinition(msg)) => {
///         eprintln!("ViewDefinition validation failed: {}", msg);
///     },
///     Err(SofError::FhirPathError(msg)) => {
///         eprintln!("FHIRPath evaluation error: {}", msg);
///     },
///     Err(SofError::UnsupportedContentType(format)) => {
///         eprintln!("Unsupported output format: {}", format);
///     },
///     Err(e) => {
///         eprintln!("Other error: {}", e);
///     }
/// }
/// ```
#[derive(Debug, Error)]
pub enum SofError {
    /// ViewDefinition structure or logic validation failed.
    ///
    /// This error occurs when a ViewDefinition contains invalid or inconsistent
    /// configuration, such as missing required fields, invalid FHIRPath expressions,
    /// or incompatible select/unionAll structures.
    #[error("Invalid ViewDefinition: {0}")]
    InvalidViewDefinition(String),

    /// FHIRPath expression evaluation failed.
    ///
    /// This error occurs when a FHIRPath expression in a ViewDefinition cannot
    /// be evaluated, either due to syntax errors or runtime evaluation issues.
    #[error("FHIRPath evaluation error: {0}")]
    FhirPathError(String),

    /// JSON serialization/deserialization failed.
    ///
    /// This error occurs when parsing input JSON or serializing output data fails,
    /// typically due to malformed JSON or incompatible data structures.
    #[error("Serialization error: {0}")]
    SerializationError(#[from] serde_json::Error),

    /// CSV processing failed.
    ///
    /// This error occurs during CSV output generation, such as when writing
    /// headers or data rows to the CSV format.
    #[error("CSV error: {0}")]
    CsvError(#[from] csv::Error),

    /// File I/O operation failed.
    ///
    /// This error occurs when reading input files or writing output files fails,
    /// typically due to permission issues or missing files.
    #[error("IO error: {0}")]
    IoError(#[from] std::io::Error),

    /// Unsupported output content type requested.
    ///
    /// This error occurs when an invalid or unimplemented content type is
    /// specified for output formatting.
    #[error("Unsupported content type: {0}")]
    UnsupportedContentType(String),

    /// CSV writer internal error.
    ///
    /// This error occurs when the CSV writer encounters an internal issue
    /// that prevents successful output generation.
    #[error("CSV writer error: {0}")]
    CsvWriterError(String),

    /// Invalid source parameter value.
    ///
    /// This error occurs when the source parameter contains an invalid URL or path.
    #[error("Invalid source: {0}")]
    InvalidSource(String),

    /// Source not found.
    ///
    /// This error occurs when the specified source file or URL cannot be found.
    #[error("Source not found: {0}")]
    SourceNotFound(String),

    /// Failed to fetch data from source.
    ///
    /// This error occurs when fetching data from a remote source fails.
    #[error("Failed to fetch source: {0}")]
    SourceFetchError(String),

    /// Failed to read source data.
    ///
    /// This error occurs when reading data from the source fails.
    #[error("Failed to read source: {0}")]
    SourceReadError(String),

    /// Invalid content in source.
    ///
    /// This error occurs when the source content is not valid FHIR data.
    #[error("Invalid source content: {0}")]
    InvalidSourceContent(String),

    /// Unsupported source protocol.
    ///
    /// This error occurs when the source URL uses an unsupported protocol.
    #[error("Unsupported source protocol: {0}")]
    UnsupportedSourceProtocol(String),

    /// Parquet conversion error.
    ///
    /// This error occurs when converting data to Parquet format fails.
    #[error("Parquet conversion error: {0}")]
    ParquetConversionError(String),

    /// A `patient` / `group` reference supplied to `$viewdefinition-run` does
    /// not resolve against the supplied resources. Per the SoF v2 spec, this
    /// is a `400 Bad Request` (mapped to OperationOutcome `not-found` /
    /// `invalid`), not a silent empty-result.
    #[error("Referenced resource not found: {0}")]
    ReferencedResourceNotFound(String),
}

/// Supported output content types for ViewDefinition transformations.
///
/// This enum defines the available output formats for transformed FHIR data.
/// Each format has specific characteristics and use cases for different
/// integration scenarios.
///
/// # Format Descriptions
///
/// - **CSV**: Comma-separated values without headers
/// - **CSV with Headers**: Comma-separated values with column headers
/// - **JSON**: Pretty-printed JSON array of objects
/// - **NDJSON**: Newline-delimited JSON (one object per line)
/// - **Parquet**: Apache Parquet columnar format (planned)
///
/// # Examples
///
/// ```rust
/// use helios_sof::ContentType;
///
/// // Parse from string
/// let csv_type = ContentType::from_string("text/csv")?;
/// assert_eq!(csv_type, ContentType::CsvWithHeader);  // Default includes headers
///
/// let json_type = ContentType::from_string("application/json")?;
/// assert_eq!(json_type, ContentType::Json);
///
/// // CSV without headers
/// let csv_no_headers = ContentType::from_string("text/csv;header=false")?;
/// assert_eq!(csv_no_headers, ContentType::Csv);
/// # Ok::<(), helios_sof::SofError>(())
/// ```
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum ContentType {
    /// Comma-separated values format without headers
    Csv,
    /// Comma-separated values format with column headers
    CsvWithHeader,
    /// Pretty-printed JSON array format
    Json,
    /// Newline-delimited JSON format (NDJSON)
    NdJson,
    /// Apache Parquet columnar format (not yet implemented)
    Parquet,
}

impl ContentType {
    /// Parse a content type from its MIME type string representation.
    ///
    /// This method converts standard MIME type strings to the corresponding
    /// ContentType enum variants. It supports the SQL-on-FHIR specification's
    /// recommended content types.
    ///
    /// # Supported MIME Types
    ///
    /// - `"text/csv"` → [`ContentType::Csv`]
    /// - `"text/csv"` → [`ContentType::CsvWithHeader`] (default: headers included)
    /// - `"text/csv;header=true"` → [`ContentType::CsvWithHeader`]
    /// - `"text/csv;header=false"` → [`ContentType::Csv`]
    /// - `"application/json"` → [`ContentType::Json`]
    /// - `"application/ndjson"` → [`ContentType::NdJson`]
    /// - `"application/x-ndjson"` → [`ContentType::NdJson`]
    /// - `"application/vnd.apache.parquet"` → [`ContentType::Parquet`] (spec native media type)
    /// - `"application/octet-stream"` → [`ContentType::Parquet`] (spec Accept-table value)
    /// - `"application/parquet"` → [`ContentType::Parquet`] (permissive alias)
    ///
    /// # Arguments
    ///
    /// * `s` - The MIME type string to parse
    ///
    /// # Returns
    ///
    /// * `Ok(ContentType)` - Successfully parsed content type
    /// * `Err(SofError::UnsupportedContentType)` - Unknown or unsupported MIME type
    ///
    /// # Examples
    ///
    /// ```rust
    /// use helios_sof::ContentType;
    ///
    /// // Shortened format names
    /// let csv = ContentType::from_string("csv")?;
    /// assert_eq!(csv, ContentType::CsvWithHeader);
    ///
    /// let json = ContentType::from_string("json")?;
    /// assert_eq!(json, ContentType::Json);
    ///
    /// let ndjson = ContentType::from_string("ndjson")?;
    /// assert_eq!(ndjson, ContentType::NdJson);
    ///
    /// // Full MIME types still supported
    /// let csv_mime = ContentType::from_string("text/csv")?;
    /// assert_eq!(csv_mime, ContentType::CsvWithHeader);
    ///
    /// // CSV with headers explicitly
    /// let csv_headers = ContentType::from_string("text/csv;header=true")?;
    /// assert_eq!(csv_headers, ContentType::CsvWithHeader);
    ///
    /// // CSV without headers
    /// let csv_no_headers = ContentType::from_string("text/csv;header=false")?;
    /// assert_eq!(csv_no_headers, ContentType::Csv);
    ///
    /// // JSON format
    /// let json_mime = ContentType::from_string("application/json")?;
    /// assert_eq!(json_mime, ContentType::Json);
    ///
    /// // Error for unsupported type
    /// assert!(ContentType::from_string("text/plain").is_err());
    /// # Ok::<(), helios_sof::SofError>(())
    /// ```
    pub fn from_string(s: &str) -> Result<Self, SofError> {
        match s {
            // Shortened format names
            "csv" => Ok(ContentType::CsvWithHeader),
            "json" => Ok(ContentType::Json),
            "ndjson" => Ok(ContentType::NdJson),
            "parquet" => Ok(ContentType::Parquet),
            // Full MIME types (for Accept header compatibility)
            "text/csv;header=false" => Ok(ContentType::Csv),
            "text/csv" | "text/csv;header=true" => Ok(ContentType::CsvWithHeader),
            "application/json" => Ok(ContentType::Json),
            "application/ndjson" | "application/x-ndjson" => Ok(ContentType::NdJson),
            // `application/vnd.apache.parquet` is the format's native media
            // type per the spec's Common Operation Behavior table;
            // `application/octet-stream` is the spec Accept-table value
            // (audit item #8) and `application/parquet` is kept as a
            // permissive alias for backwards-compat with clients that
            // still send it.
            "application/vnd.apache.parquet"
            | "application/octet-stream"
            | "application/parquet" => Ok(ContentType::Parquet),
            _ => Err(SofError::UnsupportedContentType(s.to_string())),
        }
    }

    /// The format's native media type per the SoF v2 spec's Common Operation
    /// Behavior output-format table. This is the `Content-Type` served for
    /// the raw-payload (default) representation.
    pub fn mime_type(&self) -> &'static str {
        match self {
            ContentType::Csv | ContentType::CsvWithHeader => "text/csv",
            ContentType::Json => "application/json",
            ContentType::NdJson => "application/x-ndjson",
            ContentType::Parquet => "application/vnd.apache.parquet",
        }
    }
}

/// Returns the FHIR version string for the newest enabled version.
///
/// This function provides the version string that should be used in CapabilityStatements
/// and other FHIR resources that need to specify their version.
pub fn get_fhir_version_string() -> &'static str {
    let newest_version = get_newest_enabled_fhir_version();

    match newest_version {
        #[cfg(feature = "R4")]
        helios_fhir::FhirVersion::R4 => "4.0.1",
        #[cfg(feature = "R4B")]
        helios_fhir::FhirVersion::R4B => "4.3.0",
        #[cfg(feature = "R5")]
        helios_fhir::FhirVersion::R5 => "5.0.0",
        #[cfg(feature = "R6")]
        helios_fhir::FhirVersion::R6 => "6.0.0",
    }
}

/// Returns the newest FHIR version that is enabled at compile time.
///
/// This function uses compile-time feature detection to determine which FHIR
/// version should be used when multiple versions are enabled. The priority order
/// is: R6 > R5 > R4B > R4, where newer versions take precedence.
///
/// # Examples
///
/// ```rust
/// use helios_sof::{get_newest_enabled_fhir_version, FhirVersion};
///
/// # #[cfg(any(feature = "R4", feature = "R4B", feature = "R5", feature = "R6"))]
/// # {
/// let version = get_newest_enabled_fhir_version();
/// // If R5 and R4 are both enabled, this returns R5
/// # }
/// ```
///
/// # Panics
///
/// This function will panic at compile time if no FHIR version features are enabled.
pub fn get_newest_enabled_fhir_version() -> helios_fhir::FhirVersion {
    #[cfg(feature = "R6")]
    return helios_fhir::FhirVersion::R6;

    #[cfg(all(feature = "R5", not(feature = "R6")))]
    return helios_fhir::FhirVersion::R5;

    #[cfg(all(feature = "R4B", not(feature = "R5"), not(feature = "R6")))]
    return helios_fhir::FhirVersion::R4B;

    #[cfg(all(
        feature = "R4",
        not(feature = "R4B"),
        not(feature = "R5"),
        not(feature = "R6")
    ))]
    return helios_fhir::FhirVersion::R4;

    #[cfg(not(any(feature = "R4", feature = "R4B", feature = "R5", feature = "R6")))]
    panic!("At least one FHIR version feature must be enabled");
}

/// A single row of processed tabular data from ViewDefinition transformation.
///
/// This struct represents one row in the output table, containing values for
/// each column defined in the ViewDefinition. Values are stored as optional
/// JSON values to handle nullable fields and diverse FHIR data types.
///
/// # Structure
///
/// Each `ProcessedRow` contains a vector of optional JSON values, where:
/// - `Some(value)` represents a non-null column value
/// - `None` represents a null/missing column value
/// - The order matches the column order in [`ProcessedResult::columns`]
///
/// # Examples
///
/// ```rust
/// use helios_sof::ProcessedRow;
/// use serde_json::Value;
///
/// let row = ProcessedRow {
///     values: vec![
///         Some(Value::String("patient-123".to_string())),
///         Some(Value::String("Doe".to_string())),
///         None, // Missing birth date
///         Some(Value::Bool(true)),
///     ]
/// };
/// ```
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ProcessedRow {
    /// Column values for this row, ordered according to ProcessedResult::columns
    pub values: Vec<Option<serde_json::Value>>,
}

/// Complete result of ViewDefinition transformation containing columns and data rows.
///
/// This struct represents the tabular output from processing a ViewDefinition
/// against a Bundle of FHIR resources. It contains both the column definitions
/// and the actual data rows in a format ready for serialization to various
/// output formats.
///
/// # Structure
///
/// - [`columns`](Self::columns): Ordered list of column names from the ViewDefinition
/// - [`rows`](Self::rows): Data rows where each row contains values in column order
///
/// # Examples
///
/// ```rust
/// use helios_sof::{ProcessedResult, ProcessedRow};
/// use serde_json::Value;
///
/// let result = ProcessedResult {
///     columns: vec![
///         "patient_id".to_string(),
///         "family_name".to_string(),
///         "given_name".to_string(),
///     ],
///     rows: vec![
///         ProcessedRow {
///             values: vec![
///                 Some(Value::String("patient-1".to_string())),
///                 Some(Value::String("Smith".to_string())),
///                 Some(Value::String("John".to_string())),
///             ]
///         },
///         ProcessedRow {
///             values: vec![
///                 Some(Value::String("patient-2".to_string())),
///                 Some(Value::String("Doe".to_string())),
///                 None, // Missing given name
///             ]
///         },
///     ]
/// };
///
/// assert_eq!(result.columns.len(), 3);
/// assert_eq!(result.rows.len(), 2);
/// ```
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ProcessedResult {
    /// Ordered list of column names as defined in the ViewDefinition
    pub columns: Vec<String>,
    /// Data rows containing values for each column
    pub rows: Vec<ProcessedRow>,
}

/// Execute a SQL-on-FHIR ViewDefinition transformation on a FHIR Bundle.
///
/// This is the main entry point for SQL-on-FHIR transformations. It processes
/// a ViewDefinition against a Bundle of FHIR resources and produces output in
/// the specified format. The function handles version compatibility, validation,
/// FHIRPath evaluation, and output formatting.
///
/// # Arguments
///
/// * `view_definition` - The ViewDefinition containing transformation logic
/// * `bundle` - The Bundle containing FHIR resources to process
/// * `content_type` - The desired output format
///
/// # Returns
///
/// * `Ok(Vec<u8>)` - Formatted output bytes ready for writing to file or stdout
/// * `Err(SofError)` - Detailed error information about what went wrong
///
/// # Validation
///
/// The function performs comprehensive validation:
/// - FHIR version compatibility between ViewDefinition and Bundle
/// - ViewDefinition structure and logic validation
/// - FHIRPath expression syntax and evaluation
/// - Output format compatibility
///
/// # Examples
///
/// ```rust
/// use helios_sof::{SofViewDefinition, SofBundle, ContentType, run_view_definition};
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Create a simple ViewDefinition
/// let view_json = serde_json::json!({
///     "resourceType": "ViewDefinition",
///     "status": "active",
///     "resource": "Patient",
///     "select": [{
///         "column": [{
///             "name": "id",
///             "path": "id"
///         }]
///     }]
/// });
/// let view_def: helios_fhir::r4::ViewDefinition = serde_json::from_value(view_json)?;
///
/// // Create a simple Bundle
/// let bundle_json = serde_json::json!({
///     "resourceType": "Bundle",
///     "type": "collection",
///     "entry": []
/// });
/// let bundle: helios_fhir::r4::Bundle = serde_json::from_value(bundle_json)?;
///
/// let sof_view = SofViewDefinition::R4(view_def);
/// let sof_bundle = SofBundle::R4(bundle);
///
/// // Generate CSV with headers
/// let csv_output = run_view_definition(
///     sof_view,
///     sof_bundle,
///     ContentType::CsvWithHeader
/// )?;
///
/// // Write to file or stdout
/// std::fs::write("output.csv", csv_output)?;
/// # }
/// # Ok::<(), Box<dyn std::error::Error>>(())
/// ```
///
/// # Error Handling
///
/// Common error scenarios:
///
/// ```rust,no_run
/// use helios_sof::{SofError, SofViewDefinition, SofBundle, ContentType, run_view_definition};
///
/// # let view = SofViewDefinition::R4(helios_fhir::r4::ViewDefinition::default());
/// # let bundle = SofBundle::R4(helios_fhir::r4::Bundle::default());
/// # let content_type = ContentType::Json;
/// match run_view_definition(view, bundle, content_type) {
///     Ok(output) => {
///         println!("Success: {} bytes generated", output.len());
///     },
///     Err(SofError::InvalidViewDefinition(msg)) => {
///         eprintln!("ViewDefinition error: {}", msg);
///     },
///     Err(SofError::FhirPathError(msg)) => {
///         eprintln!("FHIRPath error: {}", msg);
///     },
///     Err(e) => {
///         eprintln!("Other error: {}", e);
///     }
/// }
/// ```
pub fn run_view_definition(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
    content_type: ContentType,
) -> Result<Vec<u8>, SofError> {
    run_view_definition_with_options(view_definition, bundle, content_type, RunOptions::default())
}

/// Parses a JSON value into a [`SofViewDefinition`] using the newest enabled
/// FHIR version.
///
/// Use [`parse_view_definition_for_version`] to pick a specific version (for
/// example when matching the FHIR version of an inline `Bundle` parameter).
pub fn parse_view_definition(json: serde_json::Value) -> Result<SofViewDefinition, SofError> {
    parse_view_definition_for_version(json, get_newest_enabled_fhir_version())
}

/// Parses a JSON value into a [`SofViewDefinition`] using the specified FHIR
/// version.
pub fn parse_view_definition_for_version(
    json: serde_json::Value,
    version: helios_fhir::FhirVersion,
) -> Result<SofViewDefinition, SofError> {
    match version {
        #[cfg(feature = "R4")]
        helios_fhir::FhirVersion::R4 => {
            let view_def: helios_fhir::r4::ViewDefinition =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Invalid R4 ViewDefinition: {}", e))
                })?;
            Ok(SofViewDefinition::R4(view_def))
        }
        #[cfg(feature = "R4B")]
        helios_fhir::FhirVersion::R4B => {
            let view_def: helios_fhir::r4b::ViewDefinition =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Invalid R4B ViewDefinition: {}", e))
                })?;
            Ok(SofViewDefinition::R4B(view_def))
        }
        #[cfg(feature = "R5")]
        helios_fhir::FhirVersion::R5 => {
            let view_def: helios_fhir::r5::ViewDefinition =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Invalid R5 ViewDefinition: {}", e))
                })?;
            Ok(SofViewDefinition::R5(view_def))
        }
        #[cfg(feature = "R6")]
        helios_fhir::FhirVersion::R6 => {
            let view_def: helios_fhir::r6::ViewDefinition =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Invalid R6 ViewDefinition: {}", e))
                })?;
            Ok(SofViewDefinition::R6(view_def))
        }
    }
}

/// Wraps a list of raw FHIR resources in a `collection` Bundle of the newest
/// enabled FHIR version.
pub fn create_bundle_from_resources(
    resources: Vec<serde_json::Value>,
) -> Result<SofBundle, SofError> {
    create_bundle_from_resources_for_version(resources, get_newest_enabled_fhir_version())
}

/// Wraps a list of raw FHIR resources in a `collection` Bundle of the
/// specified FHIR version.
pub fn create_bundle_from_resources_for_version(
    resources: Vec<serde_json::Value>,
    version: helios_fhir::FhirVersion,
) -> Result<SofBundle, SofError> {
    let bundle_json = serde_json::json!({
        "resourceType": "Bundle",
        "type": "collection",
        "entry": resources.into_iter().map(|resource| {
            serde_json::json!({ "resource": resource })
        }).collect::<Vec<_>>()
    });

    match version {
        #[cfg(feature = "R4")]
        helios_fhir::FhirVersion::R4 => {
            let bundle: helios_fhir::r4::Bundle =
                serde_json::from_value(bundle_json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Failed to create R4 Bundle: {}", e))
                })?;
            Ok(SofBundle::R4(bundle))
        }
        #[cfg(feature = "R4B")]
        helios_fhir::FhirVersion::R4B => {
            let bundle: helios_fhir::r4b::Bundle =
                serde_json::from_value(bundle_json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Failed to create R4B Bundle: {}", e))
                })?;
            Ok(SofBundle::R4B(bundle))
        }
        #[cfg(feature = "R5")]
        helios_fhir::FhirVersion::R5 => {
            let bundle: helios_fhir::r5::Bundle =
                serde_json::from_value(bundle_json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Failed to create R5 Bundle: {}", e))
                })?;
            Ok(SofBundle::R5(bundle))
        }
        #[cfg(feature = "R6")]
        helios_fhir::FhirVersion::R6 => {
            let bundle: helios_fhir::r6::Bundle =
                serde_json::from_value(bundle_json).map_err(|e| {
                    SofError::InvalidViewDefinition(format!("Failed to create R6 Bundle: {}", e))
                })?;
            Ok(SofBundle::R6(bundle))
        }
    }
}

/// Filters raw FHIR resource JSON by patient and/or group references using
/// the FHIR `CompartmentDefinition-patient` spec data.
///
/// Per the SQL-on-FHIR v2 `$viewdefinition-run` spec, `patient` is `0..1`
/// and `group` is `0..*`; both arguments accept slices and multiple values
/// are unioned. `group_refs` are resolved against any `Group` resources
/// found in `resources` (the `member.entity` Patient references contribute
/// to the effective patient-compartment set).
///
/// The compartment scan uses
/// `helios_fhir::compartment_expressions::{r4,r4b,r5,r6}::get_compartment_param_expressions`
/// — a compile-time join of `CompartmentDefinition-patient.json` against
/// `search-parameters.json` — to enumerate the spec-defined `(name,
/// FHIRPath-expression)` pairs that link a resource type to the `Patient`
/// compartment. Each expression is evaluated against the resource and the
/// resulting `Reference`(s) are matched against the requested patient set.
/// This replaces the prior hand-rolled `(subject|patient)` allowlist
/// (audit item #3) without any runtime data-file dependency.
///
/// **Absent-target handling (SoF v2 spec):** any `patient` / `group` reference
/// that does not resolve against the supplied resources is a hard error,
/// returned as [`SofError::ReferencedResourceNotFound`]. Callers surface this
/// as `400 Bad Request` + an `OperationOutcome` per the spec's error table.
/// Previously this path emitted a `Warning: 199` HTTP header and continued
/// with a (possibly empty) result; the warning-header behavior was
/// removed to align with the spec.
pub fn filter_resources_by_patient_and_group(
    resources: Vec<serde_json::Value>,
    patient_refs: &[String],
    group_refs: &[String],
    fhir_version: FhirVersion,
) -> Result<Vec<serde_json::Value>, SofError> {
    use std::collections::HashSet;

    if patient_refs.is_empty() && group_refs.is_empty() {
        return Ok(resources);
    }

    // Absent-target detection: any `patient` / `group` reference that
    // isn't represented by a resource in the supplied bundle is a hard
    // error per the SoF v2 spec error table.
    let mut absent: Vec<String> = Vec::new();
    for r in patient_refs {
        let canonical = if r.starts_with("Patient/") {
            r.clone()
        } else {
            format!("Patient/{}", r)
        };
        let id = canonical
            .strip_prefix("Patient/")
            .and_then(|s| s.split('/').next());
        let found = id
            .map(|id| {
                resources.iter().any(|res| {
                    res.get("resourceType").and_then(|v| v.as_str()) == Some("Patient")
                        && res.get("id").and_then(|v| v.as_str()) == Some(id)
                })
            })
            .unwrap_or(false);
        if !found {
            absent.push(canonical);
        }
    }
    for g in group_refs {
        let canonical = if g.starts_with("Group/") {
            g.clone()
        } else {
            format!("Group/{}", g)
        };
        let id = canonical
            .strip_prefix("Group/")
            .and_then(|s| s.split('/').next());
        let found = id
            .map(|id| {
                resources.iter().any(|res| {
                    res.get("resourceType").and_then(|v| v.as_str()) == Some("Group")
                        && res.get("id").and_then(|v| v.as_str()) == Some(id)
                })
            })
            .unwrap_or(false);
        if !found {
            absent.push(canonical);
        }
    }
    if !absent.is_empty() {
        return Err(SofError::ReferencedResourceNotFound(format!(
            "{} not found in supplied resources",
            absent.join(", ")
        )));
    }

    // Build the effective patient-compartment set: explicit patient refs +
    // patient refs resolved from supplied groups. Both forms are
    // canonicalised to `Patient/{id}` so downstream comparisons don't
    // double-handle the prefix.
    let mut targets: HashSet<String> = patient_refs
        .iter()
        .map(|r| {
            if r.starts_with("Patient/") {
                r.clone()
            } else {
                format!("Patient/{}", r)
            }
        })
        .collect();

    if !group_refs.is_empty() {
        targets.extend(compartment::resolve_group_members_to_patient_refs(
            group_refs, &resources,
        ));
    }

    // No effective patient targets (e.g. supplied Group resolved to zero
    // Patient members). The targets themselves are present (they got past
    // the absent-target check above), so this is an empty-but-valid result.
    if targets.is_empty() {
        return Ok(Vec::new());
    }

    let mut filtered = Vec::with_capacity(resources.len());
    for resource in resources.into_iter() {
        // Group resources are first-class compartment members when their
        // `Group/{id}` was requested directly (i.e. not via member
        // resolution). Skip the FHIRPath scan for Group itself.
        if resource.get("resourceType").and_then(|v| v.as_str()) == Some("Group")
            && resource
                .get("id")
                .and_then(|v| v.as_str())
                .map(|id| {
                    group_refs
                        .iter()
                        .any(|g| g == &format!("Group/{}", id) || g == id)
                })
                .unwrap_or(false)
        {
            filtered.push(resource);
            continue;
        }

        if compartment::resource_in_patient_compartment(&resource, &targets, fhir_version)? {
            filtered.push(resource);
        }
    }

    Ok(filtered)
}

/// Filters raw FHIR resource JSON by their `meta.lastUpdated` timestamp,
/// returning only resources whose `lastUpdated` is strictly after `since`.
/// Resources without `meta.lastUpdated` are excluded.
pub fn filter_resources_by_since(
    resources: Vec<serde_json::Value>,
    since: DateTime<Utc>,
) -> Result<Vec<serde_json::Value>, SofError> {
    Ok(resources
        .into_iter()
        .filter(|resource| {
            resource
                .get("meta")
                .and_then(|m| m.get("lastUpdated"))
                .and_then(|lu| lu.as_str())
                .and_then(|s| DateTime::parse_from_rfc3339(s).ok())
                .map(|t| t.with_timezone(&Utc) > since)
                .unwrap_or(false)
        })
        .collect())
}

/// Configuration options for Parquet file generation.
#[derive(Debug, Clone)]
pub struct ParquetOptions {
    /// Target row group size in MB (64-1024)
    pub row_group_size_mb: u32,
    /// Target page size in KB (64-8192)
    pub page_size_kb: u32,
    /// Compression algorithm (none, snappy, gzip, lz4, brotli, zstd)
    pub compression: String,
    /// Maximum file size in MB (splits output when exceeded)
    pub max_file_size_mb: Option<u32>,
}

impl Default for ParquetOptions {
    fn default() -> Self {
        Self {
            row_group_size_mb: 256,
            page_size_kb: 1024,
            compression: "snappy".to_string(),
            max_file_size_mb: None,
        }
    }
}

/// Options for filtering and controlling ViewDefinition execution
#[derive(Debug, Clone, Default)]
pub struct RunOptions {
    /// Filter resources modified after this time
    pub since: Option<DateTime<Utc>>,
    /// Limit the number of results
    pub limit: Option<usize>,
    /// Page number for pagination (1-based)
    pub page: Option<usize>,
    /// Parquet-specific configuration options
    pub parquet_options: Option<ParquetOptions>,
}

// =============================================================================
// Streaming/Chunked Processing Types
// =============================================================================

/// Configuration for chunked NDJSON processing.
///
/// Controls how NDJSON files are read and processed in chunks to reduce
/// memory usage when handling large files.
///
/// # Examples
///
/// ```rust
/// use helios_sof::ChunkConfig;
///
/// // Default configuration (1000 resources per chunk)
/// let config = ChunkConfig::default();
///
/// // Custom configuration for memory-constrained environments
/// let config = ChunkConfig {
///     chunk_size: 100,
///     skip_invalid_lines: true,
/// };
/// ```
#[derive(Debug, Clone)]
pub struct ChunkConfig {
    /// Number of resources to process per chunk.
    /// Default: 1000 (approximately 10MB memory usage per chunk)
    pub chunk_size: usize,
    /// If true, skip lines that fail to parse as valid JSON.
    /// If false (default), return an error on the first invalid line.
    pub skip_invalid_lines: bool,
}

impl Default for ChunkConfig {
    fn default() -> Self {
        Self {
            chunk_size: 1000,
            skip_invalid_lines: false,
        }
    }
}

/// A chunk of parsed FHIR resources from an NDJSON file.
///
/// Represents a batch of resources that have been read and parsed,
/// ready for processing through a ViewDefinition.
#[derive(Debug)]
pub struct ResourceChunk {
    /// The parsed FHIR resources in this chunk
    pub resources: Vec<serde_json::Value>,
    /// Zero-based index of this chunk (0, 1, 2, ...)
    pub chunk_index: usize,
    /// True if this is the last chunk in the file
    pub is_last: bool,
}

/// Result from processing a single chunk of resources.
///
/// Contains the output rows generated from processing one chunk,
/// along with metadata about the chunk position.
#[derive(Debug, Clone)]
pub struct ChunkedResult {
    /// Column names (same for all chunks)
    pub columns: Vec<String>,
    /// Processed rows from this chunk
    pub rows: Vec<ProcessedRow>,
    /// Zero-based index of this chunk
    pub chunk_index: usize,
    /// True if this is the last chunk
    pub is_last: bool,
    /// Number of resources that were in the input chunk
    pub resources_in_chunk: usize,
}

/// Statistics from chunked processing.
///
/// Provides summary information about a completed chunked processing run.
#[derive(Debug, Clone, Default)]
pub struct ProcessingStats {
    /// Total number of lines read from the NDJSON file
    pub total_lines_read: usize,
    /// Number of FHIR resources successfully processed
    pub resources_processed: usize,
    /// Number of output rows generated
    pub output_rows: usize,
    /// Number of lines skipped due to parse errors (when skip_invalid_lines is true)
    pub skipped_lines: usize,
    /// Number of chunks processed
    pub chunks_processed: usize,
}

/// Reads NDJSON files in chunks, yielding parsed resources.
///
/// This iterator reads an NDJSON file line by line, collecting resources
/// into chunks of the configured size. Each iteration yields a `ResourceChunk`
/// containing up to `chunk_size` parsed FHIR resources.
///
/// # Examples
///
/// ```rust,no_run
/// use helios_sof::{NdjsonChunkReader, ChunkConfig};
/// use std::io::BufReader;
/// use std::fs::File;
///
/// let file = File::open("patients.ndjson").unwrap();
/// let reader = BufReader::new(file);
/// let config = ChunkConfig::default();
///
/// let mut chunk_reader = NdjsonChunkReader::new(reader, config);
///
/// while let Some(result) = chunk_reader.next() {
///     match result {
///         Ok(chunk) => {
///             println!("Chunk {}: {} resources", chunk.chunk_index, chunk.resources.len());
///         }
///         Err(e) => {
///             eprintln!("Error reading chunk: {}", e);
///             break;
///         }
///     }
/// }
/// ```
pub struct NdjsonChunkReader<R: BufRead> {
    reader: R,
    config: ChunkConfig,
    current_chunk: usize,
    finished: bool,
    line_buffer: String,
    line_number: usize,
    /// Resource type filter - only include resources of this type
    resource_type_filter: Option<String>,
    /// Number of lines skipped due to invalid JSON
    skipped_lines: usize,
}

impl<R: BufRead> NdjsonChunkReader<R> {
    /// Create a new NDJSON chunk reader with the given configuration.
    pub fn new(reader: R, config: ChunkConfig) -> Self {
        Self {
            reader,
            config,
            current_chunk: 0,
            finished: false,
            line_buffer: String::new(),
            line_number: 0,
            resource_type_filter: None,
            skipped_lines: 0,
        }
    }

    /// Set a resource type filter to only include resources of a specific type.
    ///
    /// This is useful when processing NDJSON files that contain multiple resource types.
    pub fn with_resource_type_filter(mut self, resource_type: Option<String>) -> Self {
        self.resource_type_filter = resource_type;
        self
    }

    /// Get the total number of lines read so far.
    pub fn lines_read(&self) -> usize {
        self.line_number
    }

    /// Get the number of lines skipped due to invalid JSON.
    pub fn skipped_lines(&self) -> usize {
        self.skipped_lines
    }
}

impl<R: BufRead> Iterator for NdjsonChunkReader<R> {
    type Item = Result<ResourceChunk, SofError>;

    fn next(&mut self) -> Option<Self::Item> {
        if self.finished {
            return None;
        }

        let mut resources = Vec::with_capacity(self.config.chunk_size);

        while resources.len() < self.config.chunk_size {
            self.line_buffer.clear();
            match self.reader.read_line(&mut self.line_buffer) {
                Ok(0) => {
                    // EOF reached
                    self.finished = true;
                    break;
                }
                Ok(_) => {
                    self.line_number += 1;
                    let line = self.line_buffer.trim();

                    // Skip empty lines
                    if line.is_empty() {
                        continue;
                    }

                    // Parse the JSON
                    match serde_json::from_str::<serde_json::Value>(line) {
                        Ok(value) => {
                            // Apply resource type filter if set
                            if let Some(ref filter) = self.resource_type_filter {
                                let resource_type =
                                    value.get("resourceType").and_then(|v| v.as_str());
                                if resource_type != Some(filter.as_str()) {
                                    continue;
                                }
                            }
                            resources.push(value);
                        }
                        Err(e) => {
                            if self.config.skip_invalid_lines {
                                // Skip this line and continue
                                self.skipped_lines += 1;
                                continue;
                            } else {
                                return Some(Err(SofError::InvalidSourceContent(format!(
                                    "Invalid JSON at line {}: {}",
                                    self.line_number, e
                                ))));
                            }
                        }
                    }
                }
                Err(e) => {
                    return Some(Err(SofError::IoError(e)));
                }
            }
        }

        // If we have no resources and we're finished, don't return an empty chunk
        if resources.is_empty() && self.finished {
            return None;
        }

        let chunk = ResourceChunk {
            resources,
            chunk_index: self.current_chunk,
            is_last: self.finished,
        };
        self.current_chunk += 1;

        Some(Ok(chunk))
    }
}

/// Pre-validated ViewDefinition for efficient reuse across multiple chunks.
///
/// This struct caches the validation and constant extraction from a ViewDefinition,
/// allowing efficient processing of multiple chunks without re-validating each time.
///
/// # Examples
///
/// ```rust,no_run
/// use helios_sof::{PreparedViewDefinition, SofViewDefinition, ResourceChunk};
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Parse and prepare ViewDefinition once
/// let view_json: serde_json::Value = serde_json::from_str(r#"{
///     "resourceType": "ViewDefinition",
///     "resource": "Patient",
///     "select": [{"column": [{"name": "id", "path": "id"}]}]
/// }"#).unwrap();
/// let view_def: helios_fhir::r4::ViewDefinition = serde_json::from_value(view_json).unwrap();
/// let sof_view = SofViewDefinition::R4(view_def);
///
/// let prepared = PreparedViewDefinition::new(sof_view).unwrap();
///
/// // Process multiple chunks efficiently
/// // for chunk in chunk_iterator {
/// //     let result = prepared.process_chunk(chunk)?;
/// //     // ... handle result
/// // }
/// # }
/// ```
#[derive(Debug, Clone)]
pub struct PreparedViewDefinition {
    view_definition: SofViewDefinition,
    target_resource_type: String,
    variables: HashMap<String, EvaluationResult>,
    column_names: Vec<String>,
}

impl PreparedViewDefinition {
    /// Create a new PreparedViewDefinition by validating and extracting metadata.
    ///
    /// This performs all validation upfront so that chunk processing is efficient.
    pub fn new(view_definition: SofViewDefinition) -> Result<Self, SofError> {
        // Extract target resource type and column names based on version
        let (target_resource_type, variables, column_names) = match &view_definition {
            #[cfg(feature = "R4")]
            SofViewDefinition::R4(vd) => {
                validate_view_definition(vd)?;
                let vars = extract_view_definition_constants(vd)?;
                let resource_type = vd
                    .resource()
                    .ok_or_else(|| {
                        SofError::InvalidViewDefinition("Resource type is required".to_string())
                    })?
                    .to_string();
                let mut columns = Vec::new();
                if let Some(selects) = vd.select() {
                    collect_all_columns(selects, &mut columns)?;
                }
                (resource_type, vars, columns)
            }
            #[cfg(feature = "R4B")]
            SofViewDefinition::R4B(vd) => {
                validate_view_definition(vd)?;
                let vars = extract_view_definition_constants(vd)?;
                let resource_type = vd
                    .resource()
                    .ok_or_else(|| {
                        SofError::InvalidViewDefinition("Resource type is required".to_string())
                    })?
                    .to_string();
                let mut columns = Vec::new();
                if let Some(selects) = vd.select() {
                    collect_all_columns(selects, &mut columns)?;
                }
                (resource_type, vars, columns)
            }
            #[cfg(feature = "R5")]
            SofViewDefinition::R5(vd) => {
                validate_view_definition(vd)?;
                let vars = extract_view_definition_constants(vd)?;
                let resource_type = vd
                    .resource()
                    .ok_or_else(|| {
                        SofError::InvalidViewDefinition("Resource type is required".to_string())
                    })?
                    .to_string();
                let mut columns = Vec::new();
                if let Some(selects) = vd.select() {
                    collect_all_columns(selects, &mut columns)?;
                }
                (resource_type, vars, columns)
            }
            #[cfg(feature = "R6")]
            SofViewDefinition::R6(vd) => {
                validate_view_definition(vd)?;
                let vars = extract_view_definition_constants(vd)?;
                let resource_type = vd
                    .resource()
                    .ok_or_else(|| {
                        SofError::InvalidViewDefinition("Resource type is required".to_string())
                    })?
                    .to_string();
                let mut columns = Vec::new();
                if let Some(selects) = vd.select() {
                    collect_all_columns(selects, &mut columns)?;
                }
                (resource_type, vars, columns)
            }
        };

        Ok(Self {
            view_definition,
            target_resource_type,
            variables,
            column_names,
        })
    }

    /// Get the column names that will be produced by this ViewDefinition.
    pub fn columns(&self) -> &[String] {
        &self.column_names
    }

    /// Get the target resource type for this ViewDefinition.
    pub fn target_resource_type(&self) -> &str {
        &self.target_resource_type
    }

    /// Process a chunk of resources through this ViewDefinition.
    ///
    /// Returns a `ChunkedResult` containing the rows generated from the chunk.
    /// Uses parallel processing via rayon for improved throughput.
    pub fn process_chunk(&self, chunk: ResourceChunk) -> Result<ChunkedResult, SofError> {
        self.process_chunk_with_external(chunk, Vec::new())
    }

    /// Like [`Self::process_chunk`], but folds `external` resources into this
    /// chunk's resolution pool. Used by the streaming remote-`resolve()` driver
    /// ([`process_ndjson_chunked_remote`]) to inject resources prefetched from
    /// trusted servers for references found in the chunk.
    pub fn process_chunk_with_external(
        &self,
        chunk: ResourceChunk,
        external: Vec<helios_fhir::FhirResource>,
    ) -> Result<ChunkedResult, SofError> {
        // Build the resolution pool for `resolve()` from the resources in this chunk
        // plus any remotely-prefetched `external` resources.
        //
        // NOTE: in-bundle resolution in the streaming path is limited to the
        // *current chunk* — a reference to a resource in another chunk of the same
        // input cannot be resolved locally (it falls back to a typed stub / empty).
        // Remote references (to allowlisted trusted servers) are resolved via the
        // `external` resources prefetched per chunk with a cross-chunk cache. The
        // non-streaming Bundle path resolves across the entire bundle.
        let version = self.view_definition.version();
        let mut pool: Vec<helios_fhir::FhirResource> = chunk
            .resources
            .iter()
            .filter_map(|json| parse_json_to_fhir_resource(json.clone(), version).ok())
            .collect();
        pool.extend(external);
        let resolution_scope: std::sync::Arc<Vec<helios_fhir::FhirResource>> =
            std::sync::Arc::new(pool);

        // Process resources in parallel using rayon
        let results: Result<Vec<Vec<ProcessedRow>>, SofError> = chunk
            .resources
            .par_iter()
            .filter_map(|resource_json| {
                // Check resource type matches
                let resource_type = resource_json
                    .get("resourceType")
                    .and_then(|v| v.as_str())
                    .unwrap_or("");

                if resource_type != self.target_resource_type {
                    None
                } else {
                    // Process single resource based on version
                    Some(self.process_single_resource(resource_json, &resolution_scope))
                }
            })
            .collect();

        // Flatten results from all resources
        let all_rows: Vec<ProcessedRow> = results?.into_iter().flatten().collect();

        Ok(ChunkedResult {
            columns: self.column_names.clone(),
            rows: all_rows,
            chunk_index: chunk.chunk_index,
            is_last: chunk.is_last,
            resources_in_chunk: chunk.resources.len(),
        })
    }

    /// Process a single resource JSON value through the ViewDefinition.
    fn process_single_resource(
        &self,
        resource_json: &serde_json::Value,
        resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
    ) -> Result<Vec<ProcessedRow>, SofError> {
        match &self.view_definition {
            #[cfg(feature = "R4")]
            SofViewDefinition::R4(vd) => {
                self.process_single_resource_generic(vd, resource_json, resolution_scope)
            }
            #[cfg(feature = "R4B")]
            SofViewDefinition::R4B(vd) => {
                self.process_single_resource_generic(vd, resource_json, resolution_scope)
            }
            #[cfg(feature = "R5")]
            SofViewDefinition::R5(vd) => {
                self.process_single_resource_generic(vd, resource_json, resolution_scope)
            }
            #[cfg(feature = "R6")]
            SofViewDefinition::R6(vd) => {
                self.process_single_resource_generic(vd, resource_json, resolution_scope)
            }
        }
    }

    fn process_single_resource_generic<VD>(
        &self,
        view_definition: &VD,
        resource_json: &serde_json::Value,
        resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
    ) -> Result<Vec<ProcessedRow>, SofError>
    where
        VD: ViewDefinitionTrait,
        VD::Select: ViewDefinitionSelectTrait,
    {
        // Create evaluation context from JSON by parsing into typed FhirResource
        let fhir_resource =
            parse_json_to_fhir_resource(resource_json.clone(), self.view_definition.version())?;
        let mut context = EvaluationContext::new(vec![fhir_resource]);
        // Expose the chunk-wide pool so `resolve()` can reach sibling resources.
        context.set_resolution_scope(std::sync::Arc::clone(resolution_scope));

        // Add variables to the context
        for (name, value) in &self.variables {
            context.set_variable_result(name, value.clone());
        }

        // Apply where clauses
        if let Some(where_clauses) = view_definition.where_clauses() {
            for where_clause in where_clauses {
                let path = where_clause.path().ok_or_else(|| {
                    SofError::InvalidViewDefinition("Where clause path is required".to_string())
                })?;

                match evaluate_expression(path, &context) {
                    Ok(result) => {
                        if !can_be_coerced_to_boolean(&result) {
                            return Err(SofError::InvalidViewDefinition(format!(
                                "Where clause path '{}' returns type '{}' which cannot be used as a boolean condition.",
                                path,
                                result.type_name()
                            )));
                        }
                        if !is_truthy(&result) {
                            // Resource doesn't match where clause, return empty rows
                            return Ok(Vec::new());
                        }
                    }
                    Err(e) => {
                        return Err(SofError::FhirPathError(format!(
                            "Error evaluating where clause '{}': {}",
                            path, e
                        )));
                    }
                }
            }
        }

        // Generate rows
        let select_clauses = view_definition.select().ok_or_else(|| {
            SofError::InvalidViewDefinition("At least one select clause is required".to_string())
        })?;

        let mut all_columns = self.column_names.clone();
        generate_row_combinations(&context, select_clauses, &mut all_columns, &self.variables)
    }
}

/// Iterator that combines NDJSON reading with ViewDefinition processing.
///
/// This iterator reads chunks from an NDJSON file and processes them
/// through a ViewDefinition, yielding `ChunkedResult` for each chunk.
///
/// # Examples
///
/// ```rust,no_run
/// use helios_sof::{NdjsonChunkIterator, SofViewDefinition, ChunkConfig};
/// use std::io::BufReader;
/// use std::fs::File;
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Set up ViewDefinition
/// let view_json: serde_json::Value = serde_json::from_str(r#"{
///     "resourceType": "ViewDefinition",
///     "resource": "Patient",
///     "select": [{"column": [{"name": "id", "path": "id"}]}]
/// }"#).unwrap();
/// let view_def: helios_fhir::r4::ViewDefinition = serde_json::from_value(view_json).unwrap();
/// let sof_view = SofViewDefinition::R4(view_def);
///
/// // Process file in chunks
/// let file = File::open("patients.ndjson").unwrap();
/// let reader = BufReader::new(file);
///
/// let iterator = NdjsonChunkIterator::new(sof_view, reader, ChunkConfig::default()).unwrap();
///
/// for result in iterator {
///     match result {
///         Ok(chunk_result) => {
///             println!("Chunk {}: {} rows", chunk_result.chunk_index, chunk_result.rows.len());
///         }
///         Err(e) => {
///             eprintln!("Error: {}", e);
///             break;
///         }
///     }
/// }
/// # }
/// ```
pub struct NdjsonChunkIterator<R: BufRead> {
    reader: NdjsonChunkReader<R>,
    prepared_vd: PreparedViewDefinition,
}

impl<R: BufRead> NdjsonChunkIterator<R> {
    /// Create a new chunk iterator from a ViewDefinition and NDJSON reader.
    pub fn new(
        view_definition: SofViewDefinition,
        reader: R,
        config: ChunkConfig,
    ) -> Result<Self, SofError> {
        let prepared_vd = PreparedViewDefinition::new(view_definition)?;
        let resource_type = prepared_vd.target_resource_type().to_string();
        let chunk_reader =
            NdjsonChunkReader::new(reader, config).with_resource_type_filter(Some(resource_type));

        Ok(Self {
            reader: chunk_reader,
            prepared_vd,
        })
    }

    /// Get the column names that will be produced by this iterator.
    pub fn columns(&self) -> &[String] {
        self.prepared_vd.columns()
    }

    /// Get the total number of lines read so far.
    pub fn lines_read(&self) -> usize {
        self.reader.lines_read()
    }

    /// Get the number of lines skipped due to invalid JSON.
    pub fn skipped_lines(&self) -> usize {
        self.reader.skipped_lines()
    }
}

impl<R: BufRead> Iterator for NdjsonChunkIterator<R> {
    type Item = Result<ChunkedResult, SofError>;

    fn next(&mut self) -> Option<Self::Item> {
        match self.reader.next()? {
            Ok(chunk) => Some(self.prepared_vd.process_chunk(chunk)),
            Err(e) => Some(Err(e)),
        }
    }
}

// =============================================================================
// Streaming Output Functions
// =============================================================================

/// Write CSV header row.
fn write_csv_header<W: Write>(columns: &[String], writer: &mut W) -> Result<(), SofError> {
    let mut wtr = csv::Writer::from_writer(writer);
    wtr.write_record(columns)?;
    wtr.flush()?;
    Ok(())
}

/// Write CSV rows from a chunk (no header).
fn write_csv_chunk<W: Write>(result: &ChunkedResult, writer: &mut W) -> Result<(), SofError> {
    let mut wtr = csv::Writer::from_writer(writer);

    for row in &result.rows {
        let record: Vec<String> = row
            .values
            .iter()
            .map(|v| match v {
                Some(val) => {
                    if let serde_json::Value::String(s) = val {
                        s.clone()
                    } else {
                        serde_json::to_string(val).unwrap_or_default()
                    }
                }
                None => String::new(),
            })
            .collect();
        wtr.write_record(&record)?;
    }

    wtr.flush()?;
    Ok(())
}

/// Write NDJSON rows from a chunk.
fn write_ndjson_chunk<W: Write>(result: &ChunkedResult, writer: &mut W) -> Result<(), SofError> {
    for row in &result.rows {
        let mut row_obj = serde_json::Map::new();
        for (i, column) in result.columns.iter().enumerate() {
            let value = row
                .values
                .get(i)
                .and_then(|v| v.as_ref())
                .cloned()
                .unwrap_or(serde_json::Value::Null);
            row_obj.insert(column.clone(), value);
        }
        let line = serde_json::to_string(&serde_json::Value::Object(row_obj))?;
        writer.write_all(line.as_bytes())?;
        writer.write_all(b"\n")?;
    }

    Ok(())
}

/// Process an NDJSON input stream and write output incrementally.
///
/// This is the main entry point for streaming/chunked NDJSON processing.
/// It reads the input in chunks, processes each chunk through the ViewDefinition,
/// and writes the output incrementally to the writer.
///
/// # Arguments
///
/// * `view_definition` - The ViewDefinition to execute
/// * `input` - A buffered reader for the NDJSON input
/// * `output` - A writer for the output (file, stdout, etc.)
/// * `content_type` - The desired output format (CSV, NDJSON, JSON)
/// * `config` - Configuration for chunk processing
///
/// # Returns
///
/// Statistics about the processing run, including row counts and chunk counts.
///
/// # Examples
///
/// ```rust,no_run
/// use helios_sof::{process_ndjson_chunked, SofViewDefinition, ContentType, ChunkConfig};
/// use std::io::{BufReader, BufWriter};
/// use std::fs::File;
///
/// # #[cfg(feature = "R4")]
/// # {
/// // Set up ViewDefinition
/// let view_json: serde_json::Value = serde_json::from_str(r#"{
///     "resourceType": "ViewDefinition",
///     "resource": "Patient",
///     "select": [{"column": [{"name": "id", "path": "id"}]}]
/// }"#).unwrap();
/// let view_def: helios_fhir::r4::ViewDefinition = serde_json::from_value(view_json).unwrap();
/// let sof_view = SofViewDefinition::R4(view_def);
///
/// // Process file
/// let input = BufReader::new(File::open("patients.ndjson").unwrap());
/// let mut output = BufWriter::new(File::create("output.csv").unwrap());
///
/// let stats = process_ndjson_chunked(
///     sof_view,
///     input,
///     &mut output,
///     ContentType::CsvWithHeader,
///     ChunkConfig::default(),
/// ).unwrap();
///
/// println!("Processed {} resources, {} output rows",
///     stats.resources_processed, stats.output_rows);
/// # }
/// ```
///
/// # Errors
///
/// Returns an error if:
/// - The ViewDefinition is invalid
/// - The input contains invalid JSON (when `skip_invalid_lines` is false)
/// - Writing to the output fails
/// - Parquet format is requested (not supported for streaming)
pub fn process_ndjson_chunked<R: BufRead, W: Write>(
    view_definition: SofViewDefinition,
    input: R,
    mut output: W,
    content_type: ContentType,
    config: ChunkConfig,
) -> Result<ProcessingStats, SofError> {
    // Validate content type supports streaming
    if content_type == ContentType::Parquet {
        return Err(SofError::UnsupportedContentType(
            "Parquet output is not supported for streaming. Use batch processing instead."
                .to_string(),
        ));
    }

    let mut iterator = NdjsonChunkIterator::new(view_definition, input, config)?;
    let columns = iterator.columns().to_vec();

    let mut stats = ProcessingStats::default();
    let mut is_first_chunk = true;

    // Write header if needed
    if content_type == ContentType::CsvWithHeader {
        write_csv_header(&columns, &mut output)?;
    }

    // For JSON output, we need special handling to create a valid array
    if content_type == ContentType::Json {
        output.write_all(b"[\n")?;
    }

    for result in iterator.by_ref() {
        let chunk_result = result?;

        stats.resources_processed += chunk_result.resources_in_chunk;
        stats.output_rows += chunk_result.rows.len();
        stats.chunks_processed += 1;

        write_chunk_output(
            &chunk_result,
            content_type,
            &mut output,
            &mut is_first_chunk,
        )?;
        output.flush()?;
    }

    // Close JSON array if needed
    if content_type == ContentType::Json {
        output.write_all(b"\n]")?;
    }

    output.flush()?;

    // Update stats with line/skip counts from the iterator
    stats.total_lines_read = iterator.lines_read();
    stats.skipped_lines = iterator.skipped_lines();

    Ok(stats)
}

/// Writes one chunk's rows in the requested streaming format. Shared by the sync
/// ([`process_ndjson_chunked`]) and remote ([`process_ndjson_chunked_remote`])
/// drivers. `is_first_chunk` is consulted (and cleared) only for JSON, to manage
/// inter-row commas across chunks.
fn write_chunk_output<W: Write>(
    chunk_result: &ChunkedResult,
    content_type: ContentType,
    output: &mut W,
    is_first_chunk: &mut bool,
) -> Result<(), SofError> {
    match content_type {
        ContentType::Csv | ContentType::CsvWithHeader => {
            write_csv_chunk(chunk_result, output)?;
        }
        ContentType::NdJson => {
            write_ndjson_chunk(chunk_result, output)?;
        }
        ContentType::Json => {
            // Write JSON rows with proper comma handling
            for (i, row) in chunk_result.rows.iter().enumerate() {
                if !*is_first_chunk || i > 0 {
                    output.write_all(b",\n")?;
                }

                let mut row_obj = serde_json::Map::new();
                for (j, column) in chunk_result.columns.iter().enumerate() {
                    let value = row
                        .values
                        .get(j)
                        .and_then(|v| v.as_ref())
                        .cloned()
                        .unwrap_or(serde_json::Value::Null);
                    row_obj.insert(column.clone(), value);
                }
                let json = serde_json::to_string_pretty(&serde_json::Value::Object(row_obj))?;
                output.write_all(json.as_bytes())?;
            }
        }
        ContentType::Parquet => unreachable!(), // Caller rejects Parquet for streaming
    }

    *is_first_chunk = false;
    Ok(())
}

/// Streaming NDJSON processing with remote `resolve()` enabled.
///
/// Like [`process_ndjson_chunked`], but for each chunk it prefetches references
/// pointing at trusted (allowlisted) servers and folds them into that chunk's
/// resolution pool before row generation. A single [`RemoteResolver`] is shared
/// across all chunks, so a reference recurring across chunks is fetched once
/// (bounded LRU cache) and `SOF_RESOLVE_MAX_FETCHES` is a **per-stream** cap.
///
/// In-bundle resolution remains per-chunk (a reference to a resource in another
/// chunk of the same input is not resolved locally). When `remote_config` is
/// inactive this is equivalent to [`process_ndjson_chunked`].
pub async fn process_ndjson_chunked_remote<R: BufRead, W: Write>(
    view_definition: SofViewDefinition,
    input: R,
    mut output: W,
    content_type: ContentType,
    config: ChunkConfig,
    remote_config: &RemoteResolveConfig,
) -> Result<ProcessingStats, SofError> {
    if content_type == ContentType::Parquet {
        return Err(SofError::UnsupportedContentType(
            "Parquet output is not supported for streaming. Use batch processing instead."
                .to_string(),
        ));
    }

    let version = view_definition.version();
    let prepared = PreparedViewDefinition::new(view_definition)?;
    let resource_type = prepared.target_resource_type().to_string();
    let columns = prepared.columns().to_vec();
    let mut reader =
        NdjsonChunkReader::new(input, config).with_resource_type_filter(Some(resource_type));

    // One resolver for the whole stream: cross-chunk cache + per-stream fetch cap.
    let resolver = remote_fetch::RemoteResolver::new(remote_config.clone());
    let active = remote_config.is_active();

    let mut stats = ProcessingStats::default();
    let mut is_first_chunk = true;

    if content_type == ContentType::CsvWithHeader {
        write_csv_header(&columns, &mut output)?;
    }
    if content_type == ContentType::Json {
        output.write_all(b"[\n")?;
    }

    for chunk in reader.by_ref() {
        let chunk = chunk?;

        let external = if active {
            let refs =
                reference_collector::collect_reference_strings_from_resources(&chunk.resources);
            let keys = reference_collector::collect_resource_keys_from_resources(&chunk.resources);
            resolver.resolve(refs, &keys, version).await
        } else {
            Vec::new()
        };

        let chunk_result = prepared.process_chunk_with_external(chunk, external)?;

        stats.resources_processed += chunk_result.resources_in_chunk;
        stats.output_rows += chunk_result.rows.len();
        stats.chunks_processed += 1;

        write_chunk_output(
            &chunk_result,
            content_type,
            &mut output,
            &mut is_first_chunk,
        )?;
        output.flush()?;
    }

    if content_type == ContentType::Json {
        output.write_all(b"\n]")?;
    }
    output.flush()?;

    stats.total_lines_read = reader.lines_read();
    stats.skipped_lines = reader.skipped_lines();

    Ok(stats)
}

/// Create an iterator for chunked NDJSON processing.
///
/// This is a convenience function that creates an `NdjsonChunkIterator`.
/// Use this when you want more control over how chunks are processed.
///
/// # Arguments
///
/// * `view_definition` - The ViewDefinition to execute
/// * `reader` - A buffered reader for the NDJSON input
/// * `config` - Configuration for chunk processing
///
/// # Returns
///
/// An iterator that yields `ChunkedResult` for each processed chunk.
pub fn iter_ndjson_chunks<R: BufRead>(
    view_definition: SofViewDefinition,
    reader: R,
    config: ChunkConfig,
) -> Result<NdjsonChunkIterator<R>, SofError> {
    NdjsonChunkIterator::new(view_definition, reader, config)
}

// =============================================================================
// End Streaming/Chunked Processing Types
// =============================================================================

/// Parse a JSON value into a FhirResource for the given FHIR version.
///
/// This is used internally for streaming/chunked processing where we have
/// raw JSON that needs to be converted to typed resources for FHIRPath evaluation.
/// Crate-internal entry point for the compartment filter to convert raw
/// JSON to a typed `FhirResource` (matching the version the caller already
/// negotiated). Wraps the private [`parse_json_to_fhir_resource`] without
/// exposing it as a public stable API.
pub(crate) fn parse_json_to_fhir_resource_pub(
    json: serde_json::Value,
    version: FhirVersion,
) -> Result<helios_fhir::FhirResource, SofError> {
    parse_json_to_fhir_resource(json, version)
}

fn parse_json_to_fhir_resource(
    json: serde_json::Value,
    version: FhirVersion,
) -> Result<helios_fhir::FhirResource, SofError> {
    match version {
        #[cfg(feature = "R4")]
        FhirVersion::R4 => {
            let resource: helios_fhir::r4::Resource =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidSourceContent(format!("Invalid R4 resource: {}", e))
                })?;
            Ok(helios_fhir::FhirResource::R4(Box::new(resource)))
        }
        #[cfg(feature = "R4B")]
        FhirVersion::R4B => {
            let resource: helios_fhir::r4b::Resource =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidSourceContent(format!("Invalid R4B resource: {}", e))
                })?;
            Ok(helios_fhir::FhirResource::R4B(Box::new(resource)))
        }
        #[cfg(feature = "R5")]
        FhirVersion::R5 => {
            let resource: helios_fhir::r5::Resource =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidSourceContent(format!("Invalid R5 resource: {}", e))
                })?;
            Ok(helios_fhir::FhirResource::R5(Box::new(resource)))
        }
        #[cfg(feature = "R6")]
        FhirVersion::R6 => {
            let resource: helios_fhir::r6::Resource =
                serde_json::from_value(json).map_err(|e| {
                    SofError::InvalidSourceContent(format!("Invalid R6 resource: {}", e))
                })?;
            Ok(helios_fhir::FhirResource::R6(Box::new(resource)))
        }
    }
}

/// Execute a ViewDefinition transformation with additional filtering options.
///
/// This function extends the basic `run_view_definition` with support for:
/// - Filtering resources by modification time (`since`)
/// - Limiting results (`limit`)
/// - Pagination (`page`)
///
/// # Arguments
///
/// * `view_definition` - The ViewDefinition to execute
/// * `bundle` - The Bundle containing resources to transform
/// * `content_type` - Desired output format
/// * `options` - Additional filtering and control options
///
/// # Returns
///
/// The transformed data in the requested format, with filtering applied.
pub fn run_view_definition_with_options(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
    content_type: ContentType,
    options: RunOptions,
) -> Result<Vec<u8>, SofError> {
    // Filter bundle resources by since parameter before processing
    let filtered_bundle = if let Some(since) = options.since {
        filter_bundle_by_since(bundle, since)?
    } else {
        bundle
    };

    run_view_definition_inner(
        view_definition,
        filtered_bundle,
        content_type,
        options,
        Vec::new(),
    )
}

/// Runs a ViewDefinition with remote `resolve()` enabled.
///
/// When `config` is active, references pointing at trusted (allowlisted) servers
/// are fetched up-front and folded into the resolution pool before the
/// (synchronous) row generation runs. When inactive, this behaves exactly like
/// [`run_view_definition_with_options`]. Remote resolution applies to the
/// (non-streaming) Bundle path only.
pub async fn run_view_definition_with_options_remote(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
    content_type: ContentType,
    options: RunOptions,
    config: &RemoteResolveConfig,
) -> Result<Vec<u8>, SofError> {
    let filtered_bundle = if let Some(since) = options.since {
        filter_bundle_by_since(bundle, since)?
    } else {
        bundle
    };

    let external = if config.is_active() {
        remote_fetch::prefetch_external_resources(&filtered_bundle, config).await
    } else {
        Vec::new()
    };

    run_view_definition_inner(
        view_definition,
        filtered_bundle,
        content_type,
        options,
        external,
    )
}

/// Shared tail of the run pipeline: process (with any external resources) →
/// paginate → format. The `bundle` must already be `since`-filtered.
fn run_view_definition_inner(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
    content_type: ContentType,
    options: RunOptions,
    external: Vec<helios_fhir::FhirResource>,
) -> Result<Vec<u8>, SofError> {
    // Process the ViewDefinition to generate tabular data
    let processed_result =
        process_view_definition_with_external(view_definition, bundle, external)?;

    // Apply pagination if needed
    let processed_result = if options.limit.is_some() || options.page.is_some() {
        apply_pagination_to_result(processed_result, options.limit, options.page)?
    } else {
        processed_result
    };

    // Format the result according to the requested content type
    format_output(
        processed_result,
        content_type,
        options.parquet_options.as_ref(),
    )
}

pub fn process_view_definition(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
) -> Result<ProcessedResult, SofError> {
    process_view_definition_with_external(view_definition, bundle, Vec::new())
}

/// Like [`process_view_definition`], but folds `external` resources (fetched by the
/// remote `resolve()` prefetch) into the resolution pool. `external` must already
/// be parsed in the bundle's FHIR version.
pub fn process_view_definition_with_external(
    view_definition: SofViewDefinition,
    bundle: SofBundle,
    external: Vec<helios_fhir::FhirResource>,
) -> Result<ProcessedResult, SofError> {
    // Ensure both resources use the same FHIR version
    if view_definition.version() != bundle.version() {
        return Err(SofError::InvalidViewDefinition(
            "ViewDefinition and Bundle must use the same FHIR version".to_string(),
        ));
    }

    match (view_definition, bundle) {
        #[cfg(feature = "R4")]
        (SofViewDefinition::R4(vd), SofBundle::R4(bundle)) => {
            process_view_definition_generic(vd, bundle, external)
        }
        #[cfg(feature = "R4B")]
        (SofViewDefinition::R4B(vd), SofBundle::R4B(bundle)) => {
            process_view_definition_generic(vd, bundle, external)
        }
        #[cfg(feature = "R5")]
        (SofViewDefinition::R5(vd), SofBundle::R5(bundle)) => {
            process_view_definition_generic(vd, bundle, external)
        }
        #[cfg(feature = "R6")]
        (SofViewDefinition::R6(vd), SofBundle::R6(bundle)) => {
            process_view_definition_generic(vd, bundle, external)
        }
        // This case should never happen due to the version check above,
        // but is needed for exhaustive pattern matching when multiple features are enabled
        #[cfg(any(
            all(feature = "R4", any(feature = "R4B", feature = "R5", feature = "R6")),
            all(feature = "R4B", any(feature = "R5", feature = "R6")),
            all(feature = "R5", feature = "R6")
        ))]
        _ => {
            unreachable!("Version mismatch should have been caught by the version check above")
        }
    }
}

// Generic version-agnostic constant extraction
fn extract_view_definition_constants<VD: ViewDefinitionTrait>(
    view_definition: &VD,
) -> Result<HashMap<String, EvaluationResult>, SofError> {
    let mut variables = HashMap::new();

    // `%rowIndex` is the FHIRPath environment variable tracking the 0-based position of the
    // current element during iteration. It defaults to 0 at the resource/top level (and in any
    // non-iterating scope, such as a `unionAll` branch without `forEach`). `forEach`,
    // `forEachOrNull`, and `repeat` override it per iterated element (see
    // `expand_for_each_combinations` / `expand_repeat_combinations`).
    variables.insert("%rowIndex".to_string(), EvaluationResult::integer(0));

    if let Some(constants) = view_definition.constants() {
        for constant in constants {
            let name = constant
                .name()
                .ok_or_else(|| {
                    SofError::InvalidViewDefinition("Constant name is required".to_string())
                })?
                .to_string();

            let eval_result = constant.to_evaluation_result()?;
            // Constants are referenced with % prefix in FHIRPath expressions
            variables.insert(format!("%{}", name), eval_result);
        }
    }

    Ok(variables)
}

// Generic version-agnostic ViewDefinition processing
pub(crate) fn process_view_definition_generic<VD, B>(
    view_definition: VD,
    bundle: B,
    external: Vec<helios_fhir::FhirResource>,
) -> Result<ProcessedResult, SofError>
where
    VD: ViewDefinitionTrait,
    B: BundleTrait,
    B::Resource: ResourceTrait + Sync,
    VD::Select: Sync,
{
    validate_view_definition(&view_definition)?;

    // Step 1: Extract constants/variables from ViewDefinition
    let variables = extract_view_definition_constants(&view_definition)?;

    // Step 2: Filter resources by type and profile
    let target_resource_type = view_definition
        .resource()
        .ok_or_else(|| SofError::InvalidViewDefinition("Resource type is required".to_string()))?;

    // Build the bundle-wide resolution scope *before* filtering by resource type.
    // `resolve()` must be able to reach any resource in the input bundle (e.g.
    // `Encounter.subject.resolve()` -> a sibling `Patient`), not just resources of
    // the ViewDefinition's target type. This is parsed once and shared (via `Arc`)
    // across every per-resource and per-iteration evaluation context below.
    // `external` contributes remotely-prefetched resources to the same pool.
    let resolution_scope = build_resolution_scope(&bundle, external);

    let filtered_resources = filter_resources(&bundle, target_resource_type)?;

    // Step 3: Apply where clauses to filter resources
    let filtered_resources = apply_where_clauses(
        filtered_resources,
        view_definition.where_clauses(),
        &variables,
        &resolution_scope,
    )?;

    // Step 4: Process all select clauses to generate rows with forEach support
    let select_clauses = view_definition.select().ok_or_else(|| {
        SofError::InvalidViewDefinition("At least one select clause is required".to_string())
    })?;

    // Generate rows for each resource using the forEach-aware approach
    let (all_columns, rows) = generate_rows_from_selects(
        &filtered_resources,
        select_clauses,
        &variables,
        &resolution_scope,
    )?;

    Ok(ProcessedResult {
        columns: all_columns,
        rows,
    })
}

// Generic version-agnostic validation
fn validate_view_definition<VD: ViewDefinitionTrait>(view_def: &VD) -> Result<(), SofError> {
    // Basic validation
    if view_def.resource().is_none_or(|s| s.is_empty()) {
        return Err(SofError::InvalidViewDefinition(
            "ViewDefinition must specify a resource type".to_string(),
        ));
    }

    if view_def.select().is_none_or(|s| s.is_empty()) {
        return Err(SofError::InvalidViewDefinition(
            "ViewDefinition must have at least one select".to_string(),
        ));
    }

    // Validate where clauses
    if let Some(where_clauses) = view_def.where_clauses() {
        validate_where_clauses(where_clauses)?;
    }

    // Validate selects
    if let Some(selects) = view_def.select() {
        for select in selects {
            validate_select(select)?;
        }
    }

    Ok(())
}

// Generic where clause validation
fn validate_where_clauses<W: ViewDefinitionWhereTrait>(
    where_clauses: &[W],
) -> Result<(), SofError> {
    // Basic validation - just ensure paths are provided
    // Type checking will be done during actual evaluation
    for where_clause in where_clauses {
        if where_clause.path().is_none() {
            return Err(SofError::InvalidViewDefinition(
                "Where clause must have a path specified".to_string(),
            ));
        }
    }
    Ok(())
}

// Generic helper - no longer needs to be version-specific
fn can_be_coerced_to_boolean(result: &EvaluationResult) -> bool {
    // Check if the result can be meaningfully used as a boolean in a where clause
    match result {
        // Boolean values are obviously OK
        EvaluationResult::Boolean(_, _, _) => true,

        // Empty is OK (evaluates to false)
        EvaluationResult::Empty => true,

        // Collections are OK - they evaluate based on whether they're empty or not
        EvaluationResult::Collection { .. } => true,

        // Other types cannot be meaningfully coerced to boolean for where clauses
        // This includes: String, Integer, Decimal, Date, DateTime, Time, Quantity, Object
        _ => false,
    }
}

// Generic select validation
fn validate_select<S: ViewDefinitionSelectTrait>(select: &S) -> Result<(), SofError> {
    validate_select_with_context(select, false)
}

fn validate_select_with_context<S: ViewDefinitionSelectTrait>(
    select: &S,
    in_foreach_context: bool,
) -> Result<(), SofError>
where
    S::Select: ViewDefinitionSelectTrait,
{
    // Determine if we're entering a forEach context at this level
    let entering_foreach = select.for_each().is_some() || select.for_each_or_null().is_some();
    let current_foreach_context = in_foreach_context || entering_foreach;

    // Validate collection attribute with the current forEach context
    if let Some(columns) = select.column() {
        for column in columns {
            if let Some(collection_value) = column.collection() {
                if !collection_value && !current_foreach_context {
                    return Err(SofError::InvalidViewDefinition(
                        "Column 'collection' attribute must be true when specified".to_string(),
                    ));
                }
            }
        }
    }

    // Validate unionAll column consistency
    if let Some(union_selects) = select.union_all() {
        validate_union_all_columns(union_selects)?;
    }

    // Recursively validate nested selects
    if let Some(nested_selects) = select.select() {
        for nested_select in nested_selects {
            validate_select_with_context(nested_select, current_foreach_context)?;
        }
    }

    // Validate unionAll selects with forEach context
    if let Some(union_selects) = select.union_all() {
        for union_select in union_selects {
            validate_select_with_context(union_select, current_foreach_context)?;
        }
    }

    Ok(())
}

// Generic union validation
fn validate_union_all_columns<S: ViewDefinitionSelectTrait>(
    union_selects: &[S],
) -> Result<(), SofError> {
    if union_selects.len() < 2 {
        return Ok(());
    }

    // Get column names and order from first select
    let first_select = &union_selects[0];
    let first_columns = get_column_names(first_select)?;

    // Validate all other selects have the same column names in the same order
    for (index, union_select) in union_selects.iter().enumerate().skip(1) {
        let current_columns = get_column_names(union_select)?;

        if current_columns != first_columns {
            if current_columns.len() != first_columns.len()
                || !current_columns
                    .iter()
                    .all(|name| first_columns.contains(name))
            {
                return Err(SofError::InvalidViewDefinition(format!(
                    "UnionAll branch {} has different column names than first branch",
                    index
                )));
            } else {
                return Err(SofError::InvalidViewDefinition(format!(
                    "UnionAll branch {} has columns in different order than first branch",
                    index
                )));
            }
        }
    }

    Ok(())
}

// Generic column name extraction
fn get_column_names<S: ViewDefinitionSelectTrait>(select: &S) -> Result<Vec<String>, SofError> {
    let mut column_names = Vec::new();

    // Collect direct column names
    if let Some(columns) = select.column() {
        for column in columns {
            if let Some(name) = column.name() {
                column_names.push(name.to_string());
            }
        }
    }

    // If this select has unionAll but no direct columns, get columns from first unionAll branch
    if column_names.is_empty() {
        if let Some(union_selects) = select.union_all() {
            if !union_selects.is_empty() {
                return get_column_names(&union_selects[0]);
            }
        }
    }

    Ok(column_names)
}

/// Builds the bundle-wide resolution pool that `resolve()` searches.
///
/// Every resource in the input bundle — regardless of type — is parsed into a
/// version-agnostic [`helios_fhir::FhirResource`] once and shared via `Arc`, so
/// that `Reference.resolve()` can dereference to any sibling resource in the
/// bundle (not only resources of the ViewDefinition's target type, and not only
/// `contained` children of the resource under evaluation).
///
/// The returned pool is installed on each evaluation context via
/// [`EvaluationContext::set_resolution_scope`].
///
/// `external` holds resources fetched from trusted remote servers (the remote
/// `resolve()` prefetch, [`remote_fetch::prefetch_external_resources`]); they are
/// appended *after* the bundle's own resources so that an in-bundle resource always
/// wins over a remotely-fetched copy of the same `Type/id`.
fn build_resolution_scope<B: BundleTrait>(
    bundle: &B,
    external: Vec<helios_fhir::FhirResource>,
) -> std::sync::Arc<Vec<helios_fhir::FhirResource>> {
    let mut resources: Vec<helios_fhir::FhirResource> = bundle
        .entries()
        .into_iter()
        .map(|resource| resource.to_fhir_resource())
        .collect();
    resources.extend(external);
    std::sync::Arc::new(resources)
}

// Generic resource filtering
fn filter_resources<'a, B: BundleTrait>(
    bundle: &'a B,
    resource_type: &str,
) -> Result<Vec<&'a B::Resource>, SofError> {
    Ok(bundle
        .entries()
        .into_iter()
        .filter(|resource| resource.resource_name() == resource_type)
        .collect())
}

// Generic where clause application
fn apply_where_clauses<'a, R, W>(
    resources: Vec<&'a R>,
    where_clauses: Option<&[W]>,
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
) -> Result<Vec<&'a R>, SofError>
where
    R: ResourceTrait,
    W: ViewDefinitionWhereTrait,
{
    if let Some(wheres) = where_clauses {
        let mut filtered = Vec::new();

        for resource in resources {
            let mut include_resource = true;

            // All where clauses must evaluate to true for the resource to be included
            for where_clause in wheres {
                let fhir_resource = resource.to_fhir_resource();
                let mut context = EvaluationContext::new(vec![fhir_resource]);
                // Expose the whole bundle so `where` clauses can use `resolve()`.
                context.set_resolution_scope(std::sync::Arc::clone(resolution_scope));

                // Add variables to the context
                for (name, value) in variables {
                    context.set_variable_result(name, value.clone());
                }

                let path = where_clause.path().ok_or_else(|| {
                    SofError::InvalidViewDefinition("Where clause path is required".to_string())
                })?;

                match evaluate_expression(path, &context) {
                    Ok(result) => {
                        // Check if the result can be meaningfully used as a boolean
                        if !can_be_coerced_to_boolean(&result) {
                            return Err(SofError::InvalidViewDefinition(format!(
                                "Where clause path '{}' returns type '{}' which cannot be used as a boolean condition. \
                                 Where clauses must return boolean values, collections, or empty results.",
                                path,
                                result.type_name()
                            )));
                        }

                        // Check if result is truthy (non-empty and not false)
                        if !is_truthy(&result) {
                            include_resource = false;
                            break;
                        }
                    }
                    Err(e) => {
                        return Err(SofError::FhirPathError(format!(
                            "Error evaluating where clause '{}': {}",
                            path, e
                        )));
                    }
                }
            }

            if include_resource {
                filtered.push(resource);
            }
        }

        Ok(filtered)
    } else {
        Ok(resources)
    }
}

// Removed generate_rows_per_resource_r4 - replaced with new forEach-aware implementation

// Removed generate_rows_with_for_each_r4 - replaced with new forEach-aware implementation

// Helper functions for FHIRPath result processing
fn is_truthy(result: &EvaluationResult) -> bool {
    match result {
        EvaluationResult::Empty => false,
        EvaluationResult::Boolean(b, _, _) => *b,
        EvaluationResult::Collection { items, .. } => !items.is_empty(),
        _ => true, // Non-empty, non-false values are truthy
    }
}

fn fhirpath_result_to_json_value_collection(result: EvaluationResult) -> Option<serde_json::Value> {
    match result {
        EvaluationResult::Empty => Some(serde_json::Value::Array(vec![])),
        EvaluationResult::Collection { items, .. } => {
            // Always return array for collection columns, even if empty
            let values: Vec<serde_json::Value> = items
                .into_iter()
                .filter_map(fhirpath_result_to_json_value)
                .collect();
            Some(serde_json::Value::Array(values))
        }
        // For non-collection results in collection columns, wrap in array
        single_result => {
            if let Some(json_val) = fhirpath_result_to_json_value(single_result) {
                Some(serde_json::Value::Array(vec![json_val]))
            } else {
                Some(serde_json::Value::Array(vec![]))
            }
        }
    }
}

fn fhirpath_result_to_json_value(result: EvaluationResult) -> Option<serde_json::Value> {
    match result {
        EvaluationResult::Empty => None,
        EvaluationResult::Boolean(b, _, _) => Some(serde_json::Value::Bool(b)),
        EvaluationResult::Integer(i, _, _) => {
            Some(serde_json::Value::Number(serde_json::Number::from(i)))
        }
        EvaluationResult::Decimal(d, _, _) => {
            // Check if this Decimal represents a whole number
            if d.fract().is_zero() {
                // Convert to integer if no fractional part
                if let Ok(i) = d.to_string().parse::<i64>() {
                    Some(serde_json::Value::Number(serde_json::Number::from(i)))
                } else {
                    // Handle very large numbers as strings
                    Some(serde_json::Value::String(d.to_string()))
                }
            } else {
                // Convert Decimal to a float for fractional numbers
                if let Ok(f) = d.to_string().parse::<f64>() {
                    if let Some(num) = serde_json::Number::from_f64(f) {
                        Some(serde_json::Value::Number(num))
                    } else {
                        Some(serde_json::Value::String(d.to_string()))
                    }
                } else {
                    Some(serde_json::Value::String(d.to_string()))
                }
            }
        }
        EvaluationResult::String(s, _, _) => Some(serde_json::Value::String(s)),
        EvaluationResult::Date(s, _, _) => Some(serde_json::Value::String(s)),
        EvaluationResult::DateTime(s, _, _) => {
            // Remove "@" prefix from datetime strings if present
            let cleaned = s.strip_prefix("@").unwrap_or(&s);
            Some(serde_json::Value::String(cleaned.to_string()))
        }
        EvaluationResult::Time(s, _, _) => {
            // Remove "@T" prefix from time strings if present
            let cleaned = s.strip_prefix("@T").unwrap_or(&s);
            Some(serde_json::Value::String(cleaned.to_string()))
        }
        EvaluationResult::Collection { items, .. } => {
            if items.len() == 1 {
                // Single item collection - unwrap to the item itself
                fhirpath_result_to_json_value(items.into_iter().next().unwrap())
            } else if items.is_empty() {
                None
            } else {
                // Multiple items - convert to array
                let values: Vec<serde_json::Value> = items
                    .into_iter()
                    .filter_map(fhirpath_result_to_json_value)
                    .collect();
                Some(serde_json::Value::Array(values))
            }
        }
        EvaluationResult::Object { map, .. } => {
            let mut json_map = serde_json::Map::new();
            for (k, v) in map {
                if let Some(json_val) = fhirpath_result_to_json_value(v) {
                    json_map.insert(k, json_val);
                }
            }
            Some(serde_json::Value::Object(json_map))
        }
        // Handle other result types as strings
        _ => Some(serde_json::Value::String(format!("{:?}", result))),
    }
}

fn extract_iteration_items(result: EvaluationResult) -> Vec<EvaluationResult> {
    match result {
        EvaluationResult::Collection { items, .. } => items,
        EvaluationResult::Empty => Vec::new(),
        single_item => vec![single_item],
    }
}

// Generic row generation functions

fn generate_rows_from_selects<R, S>(
    resources: &[&R],
    selects: &[S],
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
) -> Result<(Vec<String>, Vec<ProcessedRow>), SofError>
where
    R: ResourceTrait + Sync,
    S: ViewDefinitionSelectTrait + Sync,
    S::Select: ViewDefinitionSelectTrait,
{
    // Process resources in parallel
    let resource_results: Result<Vec<_>, _> = resources
        .par_iter()
        .map(|resource| {
            // Each thread gets its own local column vector
            let mut local_columns = Vec::new();
            let resource_rows = generate_rows_for_resource(
                *resource,
                selects,
                &mut local_columns,
                variables,
                resolution_scope,
            )?;
            Ok::<(Vec<String>, Vec<ProcessedRow>), SofError>((local_columns, resource_rows))
        })
        .collect();

    // Handle errors from parallel processing
    let resource_results = resource_results?;

    // Merge columns from all threads (maintaining order is important)
    let mut final_columns = Vec::new();
    let mut all_rows = Vec::new();

    for (local_columns, resource_rows) in resource_results {
        // Merge columns, avoiding duplicates
        for col in local_columns {
            if !final_columns.contains(&col) {
                final_columns.push(col);
            }
        }
        all_rows.extend(resource_rows);
    }

    Ok((final_columns, all_rows))
}

fn generate_rows_for_resource<R, S>(
    resource: &R,
    selects: &[S],
    all_columns: &mut Vec<String>,
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
) -> Result<Vec<ProcessedRow>, SofError>
where
    R: ResourceTrait,
    S: ViewDefinitionSelectTrait,
    S::Select: ViewDefinitionSelectTrait,
{
    let fhir_resource = resource.to_fhir_resource();
    let mut context = EvaluationContext::new(vec![fhir_resource]);
    // Expose the whole bundle so column/forEach paths can use `resolve()`. `this`
    // stays the current resource, so `%resource`/root semantics are unchanged.
    context.set_resolution_scope(std::sync::Arc::clone(resolution_scope));

    // Add variables to the context
    for (name, value) in variables {
        context.set_variable_result(name, value.clone());
    }

    // Generate all possible row combinations for this resource
    let row_combinations = generate_row_combinations(&context, selects, all_columns, variables)?;

    Ok(row_combinations)
}

#[derive(Debug, Clone)]
struct RowCombination {
    values: Vec<Option<serde_json::Value>>,
}

fn generate_row_combinations<S>(
    context: &EvaluationContext,
    selects: &[S],
    all_columns: &mut Vec<String>,
    variables: &HashMap<String, EvaluationResult>,
) -> Result<Vec<ProcessedRow>, SofError>
where
    S: ViewDefinitionSelectTrait,
    S::Select: ViewDefinitionSelectTrait,
{
    // First pass: collect all column names to ensure consistent ordering
    collect_all_columns(selects, all_columns)?;

    // Second pass: generate all row combinations
    let mut row_combinations = vec![RowCombination {
        values: vec![None; all_columns.len()],
    }];

    for select in selects {
        row_combinations =
            expand_select_combinations(context, select, &row_combinations, all_columns, variables)?;
    }

    // Convert to ProcessedRow format
    Ok(row_combinations
        .into_iter()
        .map(|combo| ProcessedRow {
            values: combo.values,
        })
        .collect())
}

fn collect_all_columns<S>(selects: &[S], all_columns: &mut Vec<String>) -> Result<(), SofError>
where
    S: ViewDefinitionSelectTrait,
{
    for select in selects {
        // Add columns from this select
        if let Some(columns) = select.column() {
            for col in columns {
                if let Some(name) = col.name() {
                    if !all_columns.contains(&name.to_string()) {
                        all_columns.push(name.to_string());
                    }
                }
            }
        }

        // Recursively collect from nested selects
        if let Some(nested_selects) = select.select() {
            collect_all_columns(nested_selects, all_columns)?;
        }

        // Collect from unionAll
        if let Some(union_selects) = select.union_all() {
            collect_all_columns(union_selects, all_columns)?;
        }
    }
    Ok(())
}

fn expand_select_combinations<S>(
    context: &EvaluationContext,
    select: &S,
    existing_combinations: &[RowCombination],
    all_columns: &[String],
    variables: &HashMap<String, EvaluationResult>,
) -> Result<Vec<RowCombination>, SofError>
where
    S: ViewDefinitionSelectTrait,
    S::Select: ViewDefinitionSelectTrait,
{
    // Handle forEach and forEachOrNull
    if let Some(for_each_path) = select.for_each() {
        return expand_for_each_combinations(
            context,
            select,
            existing_combinations,
            all_columns,
            for_each_path,
            false,
            variables,
        );
    }

    if let Some(for_each_or_null_path) = select.for_each_or_null() {
        return expand_for_each_combinations(
            context,
            select,
            existing_combinations,
            all_columns,
            for_each_or_null_path,
            true,
            variables,
        );
    }

    // Handle repeat directive for recursive traversal
    if let Some(repeat_paths) = select.repeat() {
        return expand_repeat_combinations(
            context,
            select,
            existing_combinations,
            all_columns,
            &repeat_paths,
            variables,
        );
    }

    // Handle regular columns (no forEach)
    let mut new_combinations = Vec::new();

    for existing_combo in existing_combinations {
        let mut new_combo = existing_combo.clone();

        // Add values from this select's columns
        if let Some(columns) = select.column() {
            for col in columns {
                if let Some(col_name) = col.name() {
                    if let Some(col_index) = all_columns.iter().position(|name| name == col_name) {
                        let path = col.path().ok_or_else(|| {
                            SofError::InvalidViewDefinition("Column path is required".to_string())
                        })?;

                        match evaluate_expression(path, context) {
                            Ok(result) => {
                                // Check if this column is marked as a collection
                                let is_collection = col.collection().unwrap_or(false);

                                new_combo.values[col_index] = if is_collection {
                                    fhirpath_result_to_json_value_collection(result)
                                } else {
                                    fhirpath_result_to_json_value(result)
                                };
                            }
                            Err(e) => {
                                return Err(SofError::FhirPathError(format!(
                                    "Error evaluating column '{}' with path '{}': {}",
                                    col_name, path, e
                                )));
                            }
                        }
                    }
                }
            }
        }

        new_combinations.push(new_combo);
    }

    // Handle nested selects
    if let Some(nested_selects) = select.select() {
        for nested_select in nested_selects {
            new_combinations = expand_select_combinations(
                context,
                nested_select,
                &new_combinations,
                all_columns,
                variables,
            )?;
        }
    }

    // Handle unionAll
    if let Some(union_selects) = select.union_all() {
        let mut union_combinations = Vec::new();

        // Process each unionAll select independently, using the combinations that already have
        // values from this select's columns and nested selects
        for union_select in union_selects {
            let select_combinations = expand_select_combinations(
                context,
                union_select,
                &new_combinations,
                all_columns,
                variables,
            )?;
            union_combinations.extend(select_combinations);
        }

        // unionAll replaces new_combinations with the union results
        // If no union results, this resource should be filtered out (no rows for this resource)
        new_combinations = union_combinations;
    }

    Ok(new_combinations)
}

/// Clone the variable map and set `%rowIndex` to `index` for the current iteration element.
///
/// `forEach`, `forEachOrNull`, and `repeat` each rebind `%rowIndex` to the 0-based position of
/// the element they are processing; nested selects and `unionAll` branches without their own
/// iteration inherit this value through the cloned map.
fn vars_with_row_index(
    variables: &HashMap<String, EvaluationResult>,
    index: usize,
) -> HashMap<String, EvaluationResult> {
    let mut item_vars = variables.clone();
    item_vars.insert(
        "%rowIndex".to_string(),
        EvaluationResult::integer(index as i64),
    );
    item_vars
}

fn expand_for_each_combinations<S>(
    context: &EvaluationContext,
    select: &S,
    existing_combinations: &[RowCombination],
    all_columns: &[String],
    for_each_path: &str,
    allow_null: bool,
    variables: &HashMap<String, EvaluationResult>,
) -> Result<Vec<RowCombination>, SofError>
where
    S: ViewDefinitionSelectTrait,
    S::Select: ViewDefinitionSelectTrait,
{
    // Evaluate the forEach expression to get iteration items
    let for_each_result = evaluate_expression(for_each_path, context).map_err(|e| {
        SofError::FhirPathError(format!(
            "Error evaluating forEach expression '{}': {}",
            for_each_path, e
        ))
    })?;

    let iteration_items = extract_iteration_items(for_each_result);

    if iteration_items.is_empty() {
        if allow_null {
            // forEachOrNull: generate a single null row per existing combination. Per the spec,
            // this row is evaluated against an empty element with `%rowIndex` = 0, so most columns
            // resolve to null while a `%rowIndex` column still yields 0.
            let empty_node = EvaluationResult::Object {
                map: HashMap::new(),
                type_info: None,
            };
            let item_vars = vars_with_row_index(variables, 0);
            let mut new_combinations = Vec::new();
            for existing_combo in existing_combinations {
                let mut new_combo = existing_combo.clone();

                if let Some(columns) = select.column() {
                    for col in columns {
                        if let Some(col_name) = col.name() {
                            if let Some(col_index) =
                                all_columns.iter().position(|name| name == col_name)
                            {
                                let path = col.path().ok_or_else(|| {
                                    SofError::InvalidViewDefinition(
                                        "Column path is required".to_string(),
                                    )
                                })?;

                                let result = if path == "$this" {
                                    empty_node.clone()
                                } else {
                                    evaluate_path_on_item(
                                        path,
                                        &empty_node,
                                        &item_vars,
                                        &context.resources,
                                    )?
                                };

                                let is_collection = col.collection().unwrap_or(false);
                                new_combo.values[col_index] = if is_collection {
                                    fhirpath_result_to_json_value_collection(result)
                                } else {
                                    fhirpath_result_to_json_value(result)
                                };
                            }
                        }
                    }
                }

                new_combinations.push(new_combo);
            }
            return Ok(new_combinations);
        } else {
            // forEach with empty collection: no rows
            return Ok(Vec::new());
        }
    }

    let mut new_combinations = Vec::new();

    // For each iteration item, create new combinations
    for (idx, item) in iteration_items.iter().enumerate() {
        // `%rowIndex` for this element scopes the columns evaluated below.
        let item_vars = vars_with_row_index(variables, idx);

        for existing_combo in existing_combinations {
            let mut new_combo = existing_combo.clone();

            // Evaluate columns in the context of the iteration item
            if let Some(columns) = select.column() {
                for col in columns {
                    if let Some(col_name) = col.name() {
                        if let Some(col_index) =
                            all_columns.iter().position(|name| name == col_name)
                        {
                            let path = col.path().ok_or_else(|| {
                                SofError::InvalidViewDefinition(
                                    "Column path is required".to_string(),
                                )
                            })?;

                            // Use the iteration item directly for path evaluation
                            let result = if path == "$this" {
                                // Special case: $this refers to the current iteration item
                                item.clone()
                            } else {
                                // Evaluate the path on the iteration item
                                evaluate_path_on_item(path, item, &item_vars, &context.resources)?
                            };

                            // Check if this column is marked as a collection
                            let is_collection = col.collection().unwrap_or(false);

                            new_combo.values[col_index] = if is_collection {
                                fhirpath_result_to_json_value_collection(result)
                            } else {
                                fhirpath_result_to_json_value(result)
                            };
                        }
                    }
                }
            }

            new_combinations.push(new_combo);
        }
    }

    // Handle nested selects with the forEach context
    if let Some(nested_selects) = select.select() {
        let mut final_combinations = Vec::new();

        for (idx, item) in iteration_items.iter().enumerate() {
            // `%rowIndex` for this element scopes its own columns and any nested selects.
            let item_vars = vars_with_row_index(variables, idx);
            let item_context = create_iteration_context(item, &item_vars, &context.resources);

            // For each iteration item, we need to start with the combinations that have
            // the correct column values for this forEach scope
            for existing_combo in existing_combinations {
                // Find the combination that corresponds to this iteration item
                // by looking at the values we set for columns in this forEach scope
                let mut base_combo = existing_combo.clone();

                // Update the base combination with column values for this iteration item
                if let Some(columns) = select.column() {
                    for col in columns {
                        if let Some(col_name) = col.name() {
                            if let Some(col_index) =
                                all_columns.iter().position(|name| name == col_name)
                            {
                                let path = col.path().ok_or_else(|| {
                                    SofError::InvalidViewDefinition(
                                        "Column path is required".to_string(),
                                    )
                                })?;

                                let result = if path == "$this" {
                                    item.clone()
                                } else {
                                    evaluate_path_on_item(
                                        path,
                                        item,
                                        &item_vars,
                                        &context.resources,
                                    )?
                                };

                                // Check if this column is marked as a collection
                                let is_collection = col.collection().unwrap_or(false);

                                base_combo.values[col_index] = if is_collection {
                                    fhirpath_result_to_json_value_collection(result)
                                } else {
                                    fhirpath_result_to_json_value(result)
                                };
                            }
                        }
                    }
                }

                // Start with this base combination for nested processing
                let mut item_combinations = vec![base_combo];

                // Process nested selects
                for nested_select in nested_selects {
                    item_combinations = expand_select_combinations(
                        &item_context,
                        nested_select,
                        &item_combinations,
                        all_columns,
                        &item_vars,
                    )?;
                }

                final_combinations.extend(item_combinations);
            }
        }

        new_combinations = final_combinations;
    }

    // Handle unionAll within forEach context
    if let Some(union_selects) = select.union_all() {
        let mut union_combinations = Vec::new();

        for (idx, item) in iteration_items.iter().enumerate() {
            // `%rowIndex` for this element is inherited by every unionAll branch that does not
            // introduce its own `forEach`.
            let item_vars = vars_with_row_index(variables, idx);
            let item_context = create_iteration_context(item, &item_vars, &context.resources);

            // For each iteration item, process all unionAll selects
            for existing_combo in existing_combinations {
                let mut base_combo = existing_combo.clone();

                // Update the base combination with column values for this iteration item
                if let Some(columns) = select.column() {
                    for col in columns {
                        if let Some(col_name) = col.name() {
                            if let Some(col_index) =
                                all_columns.iter().position(|name| name == col_name)
                            {
                                let path = col.path().ok_or_else(|| {
                                    SofError::InvalidViewDefinition(
                                        "Column path is required".to_string(),
                                    )
                                })?;

                                let result = if path == "$this" {
                                    item.clone()
                                } else {
                                    evaluate_path_on_item(
                                        path,
                                        item,
                                        &item_vars,
                                        &context.resources,
                                    )?
                                };

                                // Check if this column is marked as a collection
                                let is_collection = col.collection().unwrap_or(false);

                                base_combo.values[col_index] = if is_collection {
                                    fhirpath_result_to_json_value_collection(result)
                                } else {
                                    fhirpath_result_to_json_value(result)
                                };
                            }
                        }
                    }
                }

                // Also evaluate columns from nested selects and add them to base_combo
                if let Some(nested_selects) = select.select() {
                    for nested_select in nested_selects {
                        if let Some(nested_columns) = nested_select.column() {
                            for col in nested_columns {
                                if let Some(col_name) = col.name() {
                                    if let Some(col_index) =
                                        all_columns.iter().position(|name| name == col_name)
                                    {
                                        let path = col.path().ok_or_else(|| {
                                            SofError::InvalidViewDefinition(
                                                "Column path is required".to_string(),
                                            )
                                        })?;

                                        let result = if path == "$this" {
                                            item.clone()
                                        } else {
                                            evaluate_path_on_item(
                                                path,
                                                item,
                                                &item_vars,
                                                &context.resources,
                                            )?
                                        };

                                        // Check if this column is marked as a collection
                                        let is_collection = col.collection().unwrap_or(false);

                                        base_combo.values[col_index] = if is_collection {
                                            fhirpath_result_to_json_value_collection(result)
                                        } else {
                                            fhirpath_result_to_json_value(result)
                                        };
                                    }
                                }
                            }
                        }
                    }
                }

                // Process each unionAll select independently for this iteration item
                for union_select in union_selects {
                    let mut select_combinations = vec![base_combo.clone()];
                    select_combinations = expand_select_combinations(
                        &item_context,
                        union_select,
                        &select_combinations,
                        all_columns,
                        &item_vars,
                    )?;
                    union_combinations.extend(select_combinations);
                }
            }
        }

        // unionAll replaces new_combinations with the union results
        // If no union results, filter out this resource (no rows for this resource)
        new_combinations = union_combinations;
    }

    Ok(new_combinations)
}

/// Flatten a `repeat` recursive traversal into a pre-order list of descendant nodes.
///
/// Mirrors the reference implementation's `recursiveTraverse`: starting from `context`'s node,
/// each repeat path is evaluated, and for every resulting object child the child is appended and
/// then traversed in turn (the starting node itself is never included). Non-object results are
/// ignored, matching the reference's `typeof childNode === 'object'` guard. The resulting order is
/// what `%rowIndex` enumerates over.
fn collect_repeat_nodes(
    context: &EvaluationContext,
    repeat_paths: &[&str],
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
    out: &mut Vec<EvaluationResult>,
) -> Result<(), SofError> {
    for repeat_path in repeat_paths {
        let repeat_result = evaluate_expression(repeat_path, context).map_err(|e| {
            SofError::FhirPathError(format!(
                "Error evaluating repeat expression '{}': {}",
                repeat_path, e
            ))
        })?;

        for child_item in extract_iteration_items(repeat_result) {
            // Only object nodes participate in the traversal (matches the reference impl).
            if !matches!(child_item, EvaluationResult::Object { .. }) {
                continue;
            }
            let child_context = create_iteration_context(&child_item, variables, resolution_scope);
            out.push(child_item);
            collect_repeat_nodes(
                &child_context,
                repeat_paths,
                variables,
                resolution_scope,
                out,
            )?;
        }
    }
    Ok(())
}

fn expand_repeat_combinations<S>(
    context: &EvaluationContext,
    select: &S,
    existing_combinations: &[RowCombination],
    all_columns: &[String],
    repeat_paths: &[&str],
    variables: &HashMap<String, EvaluationResult>,
) -> Result<Vec<RowCombination>, SofError>
where
    S: ViewDefinitionSelectTrait,
    S::Select: ViewDefinitionSelectTrait,
{
    // The repeat directive performs recursive traversal of the elements reachable via the repeat
    // paths. We first flatten that traversal into a pre-order node list (see `collect_repeat_nodes`)
    // so that each emitted node can be assigned a monotonic `%rowIndex` matching the reference
    // implementation. The traversal is independent of `existing_combinations`, so it is collected
    // once and applied to each incoming combination.
    //
    // Note: Unlike forEach, repeat does NOT process the current level's columns
    // - it ONLY processes elements found via the repeat paths
    let mut nodes = Vec::new();
    collect_repeat_nodes(
        context,
        repeat_paths,
        variables,
        &context.resources,
        &mut nodes,
    )?;

    let mut all_combinations = Vec::new();

    // Process each existing combination, emitting one row per traversed node (cross product).
    for existing_combo in existing_combinations {
        for (idx, node) in nodes.iter().enumerate() {
            // Each traversed node gets its own `%rowIndex` (its position in the flattened list).
            let item_vars = vars_with_row_index(variables, idx);
            let node_context = create_iteration_context(node, &item_vars, &context.resources);

            // Create a combination for this node with the repeat level's columns
            let mut node_combo = existing_combo.clone();

            if let Some(columns) = select.column() {
                for col in columns {
                    if let Some(col_name) = col.name() {
                        if let Some(col_index) =
                            all_columns.iter().position(|name| name == col_name)
                        {
                            let path = col.path().ok_or_else(|| {
                                SofError::InvalidViewDefinition(
                                    "Column path is required".to_string(),
                                )
                            })?;

                            // Evaluate the path on the traversed node
                            let result = if path == "$this" {
                                node.clone()
                            } else {
                                evaluate_path_on_item(path, node, &item_vars, &context.resources)?
                            };

                            let is_collection = col.collection().unwrap_or(false);
                            node_combo.values[col_index] = if is_collection {
                                fhirpath_result_to_json_value_collection(result)
                            } else {
                                fhirpath_result_to_json_value(result)
                            };
                        }
                    }
                }
            }

            // Start with the node combination we just created
            let mut node_combinations = vec![node_combo];

            // Process nested selects (like forEach/forEachOrNull) in the node's context
            if let Some(nested_selects) = select.select() {
                for nested_select in nested_selects {
                    node_combinations = expand_select_combinations(
                        &node_context,
                        nested_select,
                        &node_combinations,
                        all_columns,
                        &item_vars,
                    )?;
                }
            }

            // Apply unionAll branches in the node's context
            if let Some(union_selects) = select.union_all() {
                let mut union_combinations = Vec::new();
                for combo in &node_combinations {
                    for union_select in union_selects {
                        let select_combinations = expand_select_combinations(
                            &node_context,
                            union_select,
                            std::slice::from_ref(combo),
                            all_columns,
                            &item_vars,
                        )?;
                        union_combinations.extend(select_combinations);
                    }
                }
                node_combinations = union_combinations;
            }

            // Add the processed combinations to our results
            // (these may have been filtered by forEach, which is correct)
            all_combinations.extend(node_combinations);
        }
    }

    Ok(all_combinations)
}

// Generic helper functions
fn evaluate_path_on_item(
    path: &str,
    item: &EvaluationResult,
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
) -> Result<EvaluationResult, SofError> {
    // Create a temporary context with the iteration item as the root resource
    let mut temp_context = match item {
        EvaluationResult::Object { .. } => {
            // Convert the iteration item to a resource-like structure for FHIRPath evaluation
            // For simplicity, we'll create a basic context where the item is available for evaluation
            let mut context = EvaluationContext::new(vec![]);
            context.this = Some(item.clone());
            context
        }
        _ => EvaluationContext::new(vec![]),
    };
    // Carry the bundle-wide resolution pool so chained `resolve()` calls (e.g.
    // `forEach: list.resolve()` then a column that resolves a nested reference)
    // can still reach sibling resources from this fresh, item-rooted context.
    temp_context.set_resolution_scope(std::sync::Arc::clone(resolution_scope));

    // Add variables to the temporary context
    for (name, value) in variables {
        temp_context.set_variable_result(name, value.clone());
    }

    // Evaluate the FHIRPath expression in the context of the iteration item
    match evaluate_expression(path, &temp_context) {
        Ok(result) => Ok(result),
        Err(_e) => {
            // If FHIRPath evaluation fails, try simple property access as fallback
            match item {
                EvaluationResult::Object { map, .. } => {
                    if let Some(value) = map.get(path) {
                        Ok(value.clone())
                    } else {
                        Ok(EvaluationResult::Empty)
                    }
                }
                _ => Ok(EvaluationResult::Empty),
            }
        }
    }
}

fn create_iteration_context(
    item: &EvaluationResult,
    variables: &HashMap<String, EvaluationResult>,
    resolution_scope: &std::sync::Arc<Vec<helios_fhir::FhirResource>>,
) -> EvaluationContext {
    // Create a new context with the iteration item as the root
    let mut context = EvaluationContext::new(vec![]);
    context.this = Some(item.clone());
    // Keep the bundle-wide resolution pool available to nested selects/columns
    // evaluated against this iteration item, so `resolve()` still works here.
    context.set_resolution_scope(std::sync::Arc::clone(resolution_scope));

    // Preserve variables from the parent context
    for (name, value) in variables {
        context.set_variable_result(name, value.clone());
    }

    context
}

/// Filter a bundle's resources by their lastUpdated metadata
fn filter_bundle_by_since(bundle: SofBundle, since: DateTime<Utc>) -> Result<SofBundle, SofError> {
    match bundle {
        #[cfg(feature = "R4")]
        SofBundle::R4(mut b) => {
            if let Some(entries) = b.entry.as_mut() {
                entries.retain(|entry| {
                    entry
                        .resource
                        .as_ref()
                        .and_then(|r| r.get_last_updated())
                        .map(|last_updated| last_updated > since)
                        .unwrap_or(false)
                });
            }
            Ok(SofBundle::R4(b))
        }
        #[cfg(feature = "R4B")]
        SofBundle::R4B(mut b) => {
            if let Some(entries) = b.entry.as_mut() {
                entries.retain(|entry| {
                    entry
                        .resource
                        .as_ref()
                        .and_then(|r| r.get_last_updated())
                        .map(|last_updated| last_updated > since)
                        .unwrap_or(false)
                });
            }
            Ok(SofBundle::R4B(b))
        }
        #[cfg(feature = "R5")]
        SofBundle::R5(mut b) => {
            if let Some(entries) = b.entry.as_mut() {
                entries.retain(|entry| {
                    entry
                        .resource
                        .as_ref()
                        .and_then(|r| r.get_last_updated())
                        .map(|last_updated| last_updated > since)
                        .unwrap_or(false)
                });
            }
            Ok(SofBundle::R5(b))
        }
        #[cfg(feature = "R6")]
        SofBundle::R6(mut b) => {
            if let Some(entries) = b.entry.as_mut() {
                entries.retain(|entry| {
                    entry
                        .resource
                        .as_ref()
                        .and_then(|r| r.get_last_updated())
                        .map(|last_updated| last_updated > since)
                        .unwrap_or(false)
                });
            }
            Ok(SofBundle::R6(b))
        }
    }
}

/// Apply pagination to processed results
fn apply_pagination_to_result(
    mut result: ProcessedResult,
    limit: Option<usize>,
    page: Option<usize>,
) -> Result<ProcessedResult, SofError> {
    if let Some(limit) = limit {
        let page_num = page.unwrap_or(1);
        if page_num == 0 {
            return Err(SofError::InvalidViewDefinition(
                "Page number must be greater than 0".to_string(),
            ));
        }

        let start_index = (page_num - 1) * limit;
        if start_index >= result.rows.len() {
            // Return empty result if page is beyond data
            result.rows.clear();
        } else {
            let end_index = std::cmp::min(start_index + limit, result.rows.len());
            result.rows = result.rows[start_index..end_index].to_vec();
        }
    }

    Ok(result)
}

/// Renders a [`ProcessedResult`] to bytes in the requested [`ContentType`].
///
/// Dispatches to [`format_csv`], [`format_json`], [`format_ndjson`], or
/// [`format_parquet`] based on `content_type`. Callers outside this crate
/// (REST handlers, pysof, sof-server) use this entry point so output shape is
/// consistent across consumers.
pub fn format_output(
    result: ProcessedResult,
    content_type: ContentType,
    parquet_options: Option<&ParquetOptions>,
) -> Result<Vec<u8>, SofError> {
    match content_type {
        ContentType::Csv | ContentType::CsvWithHeader => {
            format_csv(result, content_type == ContentType::CsvWithHeader)
        }
        ContentType::Json => format_json(result),
        ContentType::NdJson => format_ndjson(result),
        ContentType::Parquet => format_parquet(result, parquet_options),
    }
}

/// Builds a [`ProcessedResult`] from a stream of flat JSON-object rows.
///
/// Used by callers that receive rows as `serde_json::Value` (e.g. the REST
/// SoF runner streams) and want to feed them through the shared output
/// formatters. Column order is taken from the first row's key order;
/// subsequent rows fill in missing keys as `None`.
pub fn rows_to_processed_result(rows: Vec<serde_json::Value>) -> ProcessedResult {
    let columns: Vec<String> = match rows.first() {
        Some(serde_json::Value::Object(map)) => map.keys().cloned().collect(),
        _ => Vec::new(),
    };
    let processed_rows = rows
        .iter()
        .map(|row| {
            let values = columns
                .iter()
                .map(|col| match row {
                    serde_json::Value::Object(map) => map.get(col).cloned(),
                    _ => None,
                })
                .collect();
            ProcessedRow { values }
        })
        .collect();
    ProcessedResult {
        columns,
        rows: processed_rows,
    }
}

/// Encodes a [`ProcessedResult`] as CSV bytes via the `csv` crate (RFC 4180).
///
/// String values are emitted raw; non-string values are JSON-serialised. The
/// underlying writer handles quoting for fields containing `,`, `"`, or
/// newlines, so callers do not need to escape.
pub fn format_csv(result: ProcessedResult, include_header: bool) -> Result<Vec<u8>, SofError> {
    let mut wtr = csv::Writer::from_writer(vec![]);

    if include_header {
        wtr.write_record(&result.columns)?;
    }

    for row in result.rows {
        let record: Vec<String> = row
            .values
            .iter()
            .map(|v| match v {
                Some(val) => {
                    // For string values, extract the raw string instead of JSON serializing
                    if let serde_json::Value::String(s) = val {
                        s.clone()
                    } else {
                        // For non-string values, use JSON serialization
                        serde_json::to_string(val).unwrap_or_default()
                    }
                }
                None => String::new(),
            })
            .collect();
        wtr.write_record(&record)?;
    }

    wtr.into_inner()
        .map_err(|e| SofError::CsvWriterError(e.to_string()))
}

/// Encodes a [`ProcessedResult`] as a pretty-printed JSON array of row
/// objects. Missing column values are emitted as `null`.
pub fn format_json(result: ProcessedResult) -> Result<Vec<u8>, SofError> {
    let mut output = Vec::new();

    for row in result.rows {
        let mut row_obj = serde_json::Map::new();
        for (i, column) in result.columns.iter().enumerate() {
            let value = row
                .values
                .get(i)
                .and_then(|v| v.as_ref())
                .cloned()
                .unwrap_or(serde_json::Value::Null);
            row_obj.insert(column.clone(), value);
        }
        output.push(serde_json::Value::Object(row_obj));
    }

    Ok(serde_json::to_vec_pretty(&output)?)
}

/// Encodes a [`ProcessedResult`] as newline-delimited JSON. One row per
/// line; missing column values are emitted as `null`.
pub fn format_ndjson(result: ProcessedResult) -> Result<Vec<u8>, SofError> {
    let mut output = Vec::new();

    for row in result.rows {
        let mut row_obj = serde_json::Map::new();
        for (i, column) in result.columns.iter().enumerate() {
            let value = row
                .values
                .get(i)
                .and_then(|v| v.as_ref())
                .cloned()
                .unwrap_or(serde_json::Value::Null);
            row_obj.insert(column.clone(), value);
        }
        let line = serde_json::to_string(&serde_json::Value::Object(row_obj))?;
        output.extend_from_slice(line.as_bytes());
        output.push(b'\n');
    }

    Ok(output)
}

/// Encodes a [`ProcessedResult`] as a single Parquet file in memory.
///
/// Schema is inferred from `result.columns` and the row values; type mapping
/// follows Pathling conventions (boolean→BOOLEAN, string/code/uri→UTF8,
/// integer→INT32, decimal→FLOAT64, dateTime/date→UTF8). Use
/// [`format_parquet_multi_file`] when the output needs to be split across
/// files by size.
pub fn format_parquet(
    result: ProcessedResult,
    options: Option<&ParquetOptions>,
) -> Result<Vec<u8>, SofError> {
    use arrow::record_batch::RecordBatch;
    use parquet::arrow::ArrowWriter;
    use parquet::basic::Compression;
    use parquet::file::properties::WriterProperties;
    use std::io::Cursor;

    // Create Arrow schema from columns and sample data
    let schema = parquet_schema::create_arrow_schema(&result.columns, &result.rows)?;
    let schema_ref = std::sync::Arc::new(schema.clone());

    // Get configuration from options or use defaults
    let parquet_opts = options.cloned().unwrap_or_default();

    // Calculate optimal batch size based on row count and estimated row size
    let target_row_group_size_bytes = (parquet_opts.row_group_size_mb as usize) * 1024 * 1024;
    let target_page_size_bytes = (parquet_opts.page_size_kb as usize) * 1024;
    const TARGET_ROWS_PER_BATCH: usize = 100_000; // Default batch size
    const MAX_ROWS_PER_BATCH: usize = 500_000; // Maximum to prevent memory issues

    // Estimate average row size from first 100 rows
    let sample_size = std::cmp::min(100, result.rows.len());
    let mut estimated_row_size = 100; // Default estimate in bytes

    if sample_size > 0 {
        let sample_json_size: usize = result.rows[..sample_size]
            .iter()
            .map(|row| {
                row.values
                    .iter()
                    .filter_map(|v| v.as_ref())
                    .map(|v| v.to_string().len())
                    .sum::<usize>()
            })
            .sum();
        estimated_row_size = (sample_json_size / sample_size).max(50);
    }

    // Calculate optimal batch size
    let optimal_batch_size = (target_row_group_size_bytes / estimated_row_size)
        .clamp(TARGET_ROWS_PER_BATCH, MAX_ROWS_PER_BATCH);

    // Parse compression algorithm
    use parquet::basic::BrotliLevel;
    use parquet::basic::GzipLevel;
    use parquet::basic::ZstdLevel;

    let compression = match parquet_opts.compression.as_str() {
        "none" => Compression::UNCOMPRESSED,
        "gzip" => Compression::GZIP(GzipLevel::default()),
        "lz4" => Compression::LZ4,
        "brotli" => Compression::BROTLI(BrotliLevel::default()),
        "zstd" => Compression::ZSTD(ZstdLevel::default()),
        _ => Compression::SNAPPY, // Default to snappy
    };

    // Set up writer properties with optimized settings
    let props = WriterProperties::builder()
        .set_compression(compression)
        .set_max_row_group_size(target_row_group_size_bytes)
        .set_data_page_row_count_limit(20_000) // Optimal for predicate pushdown
        .set_data_page_size_limit(target_page_size_bytes)
        .set_write_batch_size(8192) // Control write granularity
        .build();

    // Write to memory buffer
    let mut buffer = Vec::new();
    let mut cursor = Cursor::new(&mut buffer);
    let mut writer =
        ArrowWriter::try_new(&mut cursor, schema_ref.clone(), Some(props)).map_err(|e| {
            SofError::ParquetConversionError(format!("Failed to create Parquet writer: {}", e))
        })?;

    // Process data in batches to handle large datasets efficiently
    let mut row_offset = 0;
    while row_offset < result.rows.len() {
        let batch_end = (row_offset + optimal_batch_size).min(result.rows.len());
        let batch_rows = &result.rows[row_offset..batch_end];

        // Convert batch to Arrow arrays
        let batch_arrays =
            parquet_schema::process_to_arrow_arrays(&schema, &result.columns, batch_rows)?;

        // Create RecordBatch for this chunk
        let batch = RecordBatch::try_new(schema_ref.clone(), batch_arrays).map_err(|e| {
            SofError::ParquetConversionError(format!(
                "Failed to create RecordBatch for rows {}-{}: {}",
                row_offset, batch_end, e
            ))
        })?;

        // Write batch
        writer.write(&batch).map_err(|e| {
            SofError::ParquetConversionError(format!(
                "Failed to write RecordBatch for rows {}-{}: {}",
                row_offset, batch_end, e
            ))
        })?;

        row_offset = batch_end;
    }

    writer.close().map_err(|e| {
        SofError::ParquetConversionError(format!("Failed to close Parquet writer: {}", e))
    })?;

    Ok(buffer)
}

/// Format Parquet data with automatic file splitting when size exceeds limit
pub fn format_parquet_multi_file(
    result: ProcessedResult,
    options: Option<&ParquetOptions>,
    max_file_size_bytes: usize,
) -> Result<Vec<Vec<u8>>, SofError> {
    use arrow::record_batch::RecordBatch;
    use parquet::arrow::ArrowWriter;
    use parquet::basic::Compression;
    use parquet::file::properties::WriterProperties;
    use std::io::Cursor;

    // Create Arrow schema from columns and sample data
    let schema = parquet_schema::create_arrow_schema(&result.columns, &result.rows)?;
    let schema_ref = std::sync::Arc::new(schema.clone());

    // Get configuration from options or use defaults
    let parquet_opts = options.cloned().unwrap_or_default();

    // Calculate optimal batch size
    let target_row_group_size_bytes = (parquet_opts.row_group_size_mb as usize) * 1024 * 1024;
    let target_page_size_bytes = (parquet_opts.page_size_kb as usize) * 1024;
    const TARGET_ROWS_PER_BATCH: usize = 100_000;
    const MAX_ROWS_PER_BATCH: usize = 500_000;

    // Estimate average row size
    let sample_size = std::cmp::min(100, result.rows.len());
    let mut estimated_row_size = 100;

    if sample_size > 0 {
        let sample_json_size: usize = result.rows[..sample_size]
            .iter()
            .map(|row| {
                row.values
                    .iter()
                    .filter_map(|v| v.as_ref())
                    .map(|v| v.to_string().len())
                    .sum::<usize>()
            })
            .sum();
        estimated_row_size = (sample_json_size / sample_size).max(50);
    }

    let optimal_batch_size = (target_row_group_size_bytes / estimated_row_size)
        .clamp(TARGET_ROWS_PER_BATCH, MAX_ROWS_PER_BATCH);

    // Parse compression algorithm
    use parquet::basic::BrotliLevel;
    use parquet::basic::GzipLevel;
    use parquet::basic::ZstdLevel;

    let compression = match parquet_opts.compression.as_str() {
        "none" => Compression::UNCOMPRESSED,
        "gzip" => Compression::GZIP(GzipLevel::default()),
        "lz4" => Compression::LZ4,
        "brotli" => Compression::BROTLI(BrotliLevel::default()),
        "zstd" => Compression::ZSTD(ZstdLevel::default()),
        _ => Compression::SNAPPY,
    };

    // Set up writer properties
    let props = WriterProperties::builder()
        .set_compression(compression)
        .set_max_row_group_size(target_row_group_size_bytes)
        .set_data_page_row_count_limit(20_000)
        .set_data_page_size_limit(target_page_size_bytes)
        .set_write_batch_size(8192)
        .build();

    let mut file_buffers = Vec::new();
    let mut current_buffer = Vec::new();
    let mut current_cursor = Cursor::new(&mut current_buffer);
    let mut current_writer =
        ArrowWriter::try_new(&mut current_cursor, schema_ref.clone(), Some(props.clone()))
            .map_err(|e| {
                SofError::ParquetConversionError(format!("Failed to create Parquet writer: {}", e))
            })?;

    let mut row_offset = 0;
    let mut _current_file_rows = 0;

    while row_offset < result.rows.len() {
        let batch_end = (row_offset + optimal_batch_size).min(result.rows.len());
        let batch_rows = &result.rows[row_offset..batch_end];

        // Convert batch to Arrow arrays
        let batch_arrays =
            parquet_schema::process_to_arrow_arrays(&schema, &result.columns, batch_rows)?;

        // Create RecordBatch
        let batch = RecordBatch::try_new(schema_ref.clone(), batch_arrays).map_err(|e| {
            SofError::ParquetConversionError(format!(
                "Failed to create RecordBatch for rows {}-{}: {}",
                row_offset, batch_end, e
            ))
        })?;

        // Write batch
        current_writer.write(&batch).map_err(|e| {
            SofError::ParquetConversionError(format!(
                "Failed to write RecordBatch for rows {}-{}: {}",
                row_offset, batch_end, e
            ))
        })?;

        _current_file_rows += batch_end - row_offset;
        row_offset = batch_end;

        // Check if we should start a new file
        // Get actual size of current buffer by flushing the writer
        let current_size = current_writer.bytes_written();

        if current_size >= max_file_size_bytes && row_offset < result.rows.len() {
            // Close current file
            current_writer.close().map_err(|e| {
                SofError::ParquetConversionError(format!("Failed to close Parquet writer: {}", e))
            })?;

            // Save the buffer
            file_buffers.push(current_buffer);

            // Start new file
            current_buffer = Vec::new();
            current_cursor = Cursor::new(&mut current_buffer);
            current_writer =
                ArrowWriter::try_new(&mut current_cursor, schema_ref.clone(), Some(props.clone()))
                    .map_err(|e| {
                        SofError::ParquetConversionError(format!(
                            "Failed to create new Parquet writer: {}",
                            e
                        ))
                    })?;
            _current_file_rows = 0;
        }
    }

    // Close the final writer
    current_writer.close().map_err(|e| {
        SofError::ParquetConversionError(format!("Failed to close final Parquet writer: {}", e))
    })?;

    file_buffers.push(current_buffer);

    Ok(file_buffers)
}