tp_lib_core/io/

csv.rs

//! CSV parsing and writing

use crate::errors::ProjectionError;
use crate::models::{AssociatedNetElement, GnssPosition, ProjectedPosition, TrainPath};
use crate::temporal::parse_timestamp_flexible_str;
use chrono::{DateTime, FixedOffset};
use polars::prelude::*;
use std::collections::HashMap;

// CSV column name constants for projected positions output
const COL_ORIGINAL_LAT: &str = "original_lat";
const COL_ORIGINAL_LON: &str = "original_lon";
const COL_ORIGINAL_TIME: &str = "original_time";
const COL_PROJECTED_LAT: &str = "projected_lat";
const COL_PROJECTED_LON: &str = "projected_lon";
const COL_NETELEMENT_ID: &str = "netelement_id";
const COL_MEASURE_METERS: &str = "measure_meters";
const COL_PROJECTION_DISTANCE_METERS: &str = "projection_distance_meters";
const COL_CRS: &str = "crs";

// CSV column names for train path output
const COL_PROBABILITY: &str = "probability";
const COL_START_INTRINSIC: &str = "start_intrinsic";
const COL_END_INTRINSIC: &str = "end_intrinsic";
const COL_GNSS_START_INDEX: &str = "gnss_start_index";
const COL_GNSS_END_INDEX: &str = "gnss_end_index";

/// Parse a timestamp string, accepting RFC3339 (with timezone) or a naive
/// ISO 8601 datetime without timezone (interpreted as local time).
fn parse_timestamp(s: &str) -> Result<DateTime<FixedOffset>, String> {
    parse_timestamp_flexible_str(s)
}

/// Parse GNSS positions from CSV file
pub fn parse_gnss_csv(
    path: &str,
    crs: &str,
    lat_col: &str,
    lon_col: &str,
    time_col: &str,
) -> Result<Vec<GnssPosition>, ProjectionError> {
    // Read CSV file using polars
    let df = CsvReadOptions::default()
        .with_has_header(true)
        .try_into_reader_with_file_path(Some(path.into()))
        .map_err(|e| {
            ProjectionError::IoError(std::io::Error::new(
                std::io::ErrorKind::InvalidData,
                format!("Failed to read CSV: {}", e),
            ))
        })?
        .finish()
        .map_err(|e| {
            ProjectionError::IoError(std::io::Error::new(
                std::io::ErrorKind::InvalidData,
                format!("Failed to parse CSV: {}", e),
            ))
        })?;

    gnss_positions_from_df(df, crs, lat_col, lon_col, time_col)
}

/// In-memory variant of [`parse_gnss_csv`] that accepts the full CSV text
/// directly. No disk I/O is performed; required by the .NET bindings for
/// database-backed callers (FR-012).
pub fn parse_gnss_csv_str(
    csv_text: &str,
    crs: &str,
    lat_col: &str,
    lon_col: &str,
    time_col: &str,
) -> Result<Vec<GnssPosition>, ProjectionError> {
    let cursor = std::io::Cursor::new(csv_text.as_bytes().to_vec());
    let df = CsvReadOptions::default()
        .with_has_header(true)
        .into_reader_with_file_handle(cursor)
        .finish()
        .map_err(|e| {
            ProjectionError::IoError(std::io::Error::new(
                std::io::ErrorKind::InvalidData,
                format!("Failed to parse CSV: {}", e),
            ))
        })?;
    gnss_positions_from_df(df, crs, lat_col, lon_col, time_col)
}

/// Shared body: convert a polars DataFrame to a sequence of GnssPosition rows.
fn gnss_positions_from_df(
    df: DataFrame,
    crs: &str,
    lat_col: &str,
    lon_col: &str,
    time_col: &str,
) -> Result<Vec<GnssPosition>, ProjectionError> {
    // Handle empty CSV (only headers) - polars can't infer types from empty data
    if df.height() == 0 {
        return Ok(Vec::new());
    }

    // Validate required columns exist
    let schema = df.schema();
    if !schema.contains(lat_col) {
        return Err(ProjectionError::InvalidCoordinate(format!(
            "Latitude column '{}' not found in CSV",
            lat_col
        )));
    }
    if !schema.contains(lon_col) {
        return Err(ProjectionError::InvalidCoordinate(format!(
            "Longitude column '{}' not found in CSV",
            lon_col
        )));
    }
    if !schema.contains(time_col) {
        return Err(ProjectionError::InvalidTimestamp(format!(
            "Timestamp column '{}' not found in CSV",
            time_col
        )));
    }

    // Get all column names for metadata preservation
    let all_columns: Vec<String> = schema.iter_names().map(|s| s.to_string()).collect();

    // Check if heading and distance columns exist (optional - US4: T115-T116)
    let has_heading = schema.contains("heading");
    let has_distance = schema.contains("distance");

    // Extract required columns
    let lat_series = df.column(lat_col).map_err(|e| {
        ProjectionError::InvalidCoordinate(format!("Failed to get latitude: {}", e))
    })?;
    let lon_series = df.column(lon_col).map_err(|e| {
        ProjectionError::InvalidCoordinate(format!("Failed to get longitude: {}", e))
    })?;
    let time_series = df.column(time_col).map_err(|e| {
        ProjectionError::InvalidTimestamp(format!("Failed to get timestamp: {}", e))
    })?;

    // Convert to f64 arrays
    let lat_array = lat_series.f64().map_err(|e| {
        ProjectionError::InvalidCoordinate(format!("Latitude must be numeric: {}", e))
    })?;
    let lon_array = lon_series.f64().map_err(|e| {
        ProjectionError::InvalidCoordinate(format!("Longitude must be numeric: {}", e))
    })?;
    let time_array = time_series.str().map_err(|e| {
        ProjectionError::InvalidTimestamp(format!("Timestamp must be string: {}", e))
    })?;

    // Get optional heading and distance series if they exist
    let heading_series = if has_heading {
        Some(df.column("heading").map_err(|e| {
            ProjectionError::InvalidGeometry(format!("Failed to get heading: {}", e))
        })?)
    } else {
        None
    };

    let distance_series = if has_distance {
        Some(df.column("distance").map_err(|e| {
            ProjectionError::InvalidGeometry(format!("Failed to get distance: {}", e))
        })?)
    } else {
        None
    };

    // Convert heading and distance to typed arrays
    let heading_array = heading_series
        .as_ref()
        .map(|s| s.f64())
        .transpose()
        .map_err(|e| ProjectionError::InvalidGeometry(format!("Heading must be numeric: {}", e)))?;

    let distance_array = distance_series
        .as_ref()
        .map(|s| s.f64())
        .transpose()
        .map_err(|e| {
            ProjectionError::InvalidGeometry(format!("Distance must be numeric: {}", e))
        })?;

    // Build GNSS positions
    let mut positions = Vec::new();
    let row_count = df.height();

    for i in 0..row_count {
        // Get coordinates
        let latitude = lat_array.get(i).ok_or_else(|| {
            ProjectionError::InvalidCoordinate(format!("Missing latitude at row {}", i))
        })?;
        let longitude = lon_array.get(i).ok_or_else(|| {
            ProjectionError::InvalidCoordinate(format!("Missing longitude at row {}", i))
        })?;

        // Get and parse timestamp
        let time_str = time_array.get(i).ok_or_else(|| {
            ProjectionError::InvalidTimestamp(format!("Missing timestamp at row {}", i))
        })?;

        let timestamp = parse_timestamp(time_str).map_err(|e| {
            ProjectionError::InvalidTimestamp(format!(
                "Invalid timestamp '{}' at row {}: {}",
                time_str, i, e
            ))
        })?;

        // Build metadata from other columns
        let mut metadata = HashMap::new();
        for col_name in &all_columns {
            if col_name != lat_col
                && col_name != lon_col
                && col_name != time_col
                && col_name != "heading"
                && col_name != "distance"
            {
                if let Ok(series) = df.column(col_name) {
                    if let Ok(str_series) = series.cast(&DataType::String) {
                        if let Ok(str_chunked) = str_series.str() {
                            if let Some(value) = str_chunked.get(i) {
                                metadata.insert(col_name.clone(), value.to_string());
                            }
                        }
                    }
                }
            }
        }

        // Extract heading if present (0-360°), validate range
        let heading = heading_array.as_ref().and_then(|arr| arr.get(i));
        if let Some(h) = heading {
            if !(0.0..=360.0).contains(&h) {
                return Err(ProjectionError::InvalidGeometry(format!(
                    "Heading must be in [0, 360], got {} at row {}",
                    h, i
                )));
            }
        }

        // Extract distance if present (must be >= 0)
        let distance = distance_array.as_ref().and_then(|arr| arr.get(i));
        if let Some(d) = distance {
            if d < 0.0 {
                return Err(ProjectionError::InvalidGeometry(format!(
                    "Distance must be >= 0, got {} at row {}",
                    d, i
                )));
            }
        }

        // Create GNSS position with heading and distance if available (US4: T115-T116)
        let mut position = GnssPosition::with_heading_distance(
            latitude,
            longitude,
            timestamp,
            crs.to_string(),
            heading,
            distance,
        )?;
        position.metadata = metadata;
        positions.push(position);
    }

    Ok(positions)
}

/// Write projected positions to CSV
pub fn write_csv(
    positions: &[ProjectedPosition],
    writer: &mut impl std::io::Write,
) -> Result<(), ProjectionError> {
    use csv::Writer;

    let mut csv_writer = Writer::from_writer(writer);

    // Write header
    csv_writer.write_record([
        COL_ORIGINAL_LAT,
        COL_ORIGINAL_LON,
        COL_ORIGINAL_TIME,
        COL_PROJECTED_LAT,
        COL_PROJECTED_LON,
        COL_NETELEMENT_ID,
        COL_MEASURE_METERS,
        COL_PROJECTION_DISTANCE_METERS,
        COL_CRS,
    ])?;

    // Write data rows
    for pos in positions {
        csv_writer.write_record(&[
            pos.original.latitude.to_string(),
            pos.original.longitude.to_string(),
            pos.original.timestamp.to_rfc3339(),
            pos.projected_coords.y().to_string(),
            pos.projected_coords.x().to_string(),
            pos.netelement_id.clone(),
            pos.measure_meters.to_string(),
            pos.projection_distance_meters.to_string(),
            pos.crs.clone(),
        ])?;
    }

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

/// Write TrainPath to CSV
///
/// Output format: One row per segment with columns:
/// - netelement_id: ID of the netelement
/// - probability: Segment probability (0.0 to 1.0)
/// - start_intrinsic: Entry point on netelement (0.0 to 1.0)
/// - end_intrinsic: Exit point on netelement (0.0 to 1.0)
/// - gnss_start_index: First GNSS position index
/// - gnss_end_index: Last GNSS position index
///
/// The overall_probability is written as a comment in the first line.
///
/// # Example Output
///
/// ```csv
/// # overall_probability: 0.89
/// netelement_id,probability,start_intrinsic,end_intrinsic,gnss_start_index,gnss_end_index
/// NE_A,0.87,0.0,1.0,0,10
/// NE_B,0.92,0.0,1.0,11,18
/// ```
pub fn write_trainpath_csv(
    train_path: &TrainPath,
    writer: &mut impl std::io::Write,
) -> Result<(), ProjectionError> {
    use csv::Writer;

    // Write overall probability as comment
    writeln!(
        writer,
        "# overall_probability: {}",
        train_path.overall_probability
    )?;

    if let Some(calculated_at) = &train_path.calculated_at {
        writeln!(writer, "# calculated_at: {}", calculated_at.to_rfc3339())?;
    }

    let mut csv_writer = Writer::from_writer(writer);

    // Write header
    csv_writer.write_record([
        COL_NETELEMENT_ID,
        COL_PROBABILITY,
        COL_START_INTRINSIC,
        COL_END_INTRINSIC,
        COL_GNSS_START_INDEX,
        COL_GNSS_END_INDEX,
    ])?;

    // Write data rows
    for segment in &train_path.segments {
        csv_writer.write_record(&[
            segment.netelement_id.clone(),
            segment.probability.to_string(),
            segment.start_intrinsic.to_string(),
            segment.end_intrinsic.to_string(),
            segment.gnss_start_index.to_string(),
            segment.gnss_end_index.to_string(),
        ])?;
    }

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

/// Parse TrainPath from CSV
///
/// Reads a CSV file in the format produced by write_trainpath_csv.
/// Expects columns: netelement_id, probability, start_intrinsic, end_intrinsic,
/// gnss_start_index, gnss_end_index
///
/// The overall_probability can be specified in a comment line starting with
/// `# overall_probability:` or will default to the average of segment probabilities.
///
/// # Arguments
///
/// * `path` - Path to CSV file
///
/// # Returns
///
/// A TrainPath struct reconstructed from the CSV data
pub fn parse_trainpath_csv(path: &str) -> Result<TrainPath, ProjectionError> {
    // Read the file to extract comment lines and filter them out
    let file_content = std::fs::read_to_string(path)?;
    let mut overall_probability: Option<f64> = None;
    let mut calculated_at: Option<chrono::DateTime<chrono::Utc>> = None;
    let mut csv_lines = Vec::new();

    // Parse comment lines and collect non-comment lines
    for line in file_content.lines() {
        if let Some(comment) = line.strip_prefix('#') {
            let comment = comment.trim();
            if let Some(value) = comment.strip_prefix("overall_probability:") {
                overall_probability = value.trim().parse().ok();
            } else if let Some(value) = comment.strip_prefix("calculated_at:") {
                if let Ok(dt) = parse_timestamp_flexible_str(value.trim()) {
                    calculated_at = Some(dt.with_timezone(&chrono::Utc));
                }
            }
        } else {
            csv_lines.push(line);
        }
    }

    // Write filtered CSV to temporary string for polars
    // Use thread ID and timestamp to avoid race conditions with parallel tests
    let filtered_csv = csv_lines.join("\n");
    let unique_id = format!(
        "{}_{:?}",
        std::process::id(),
        std::time::SystemTime::now()
            .duration_since(std::time::UNIX_EPOCH)
            .map(|d| d.as_nanos())
            .unwrap_or(0)
    );
    let temp_file = std::env::temp_dir().join(format!("trainpath_{}.csv", unique_id));
    std::fs::write(&temp_file, filtered_csv)?;

    // Read CSV using polars
    let df = CsvReadOptions::default()
        .with_has_header(true)
        .try_into_reader_with_file_path(Some(temp_file.clone()))
        .map_err(|e| {
            ProjectionError::IoError(std::io::Error::new(
                std::io::ErrorKind::InvalidData,
                format!("Failed to read TrainPath CSV: {}", e),
            ))
        })?
        .finish()
        .map_err(|e| {
            ProjectionError::IoError(std::io::Error::new(
                std::io::ErrorKind::InvalidData,
                format!("Failed to parse TrainPath CSV: {}", e),
            ))
        })?;

    // Clean up temp file
    let _ = std::fs::remove_file(temp_file);

    // Handle empty CSV (only headers)
    if df.height() == 0 {
        return TrainPath::new(Vec::new(), 1.0, None, None);
    }

    // Extract columns and cast to correct types
    let netelement_id = df
        .column("netelement_id")
        .map_err(|e| ProjectionError::GeoJsonError(format!("Missing netelement_id column: {}", e)))?
        .str()
        .map_err(|e| ProjectionError::GeoJsonError(format!("netelement_id must be string: {}", e)))?
        .clone();

    let probability_series = df
        .column("probability")
        .map_err(|e| ProjectionError::GeoJsonError(format!("Missing probability column: {}", e)))?
        .cast(&DataType::Float64)
        .map_err(|e| ProjectionError::GeoJsonError(format!("probability cast failed: {}", e)))?;
    let probability = probability_series.f64().map_err(|e| {
        ProjectionError::GeoJsonError(format!("probability must be numeric: {}", e))
    })?;

    let start_intrinsic_series = df
        .column("start_intrinsic")
        .map_err(|e| {
            ProjectionError::GeoJsonError(format!("Missing start_intrinsic column: {}", e))
        })?
        .cast(&DataType::Float64)
        .map_err(|e| {
            ProjectionError::GeoJsonError(format!("start_intrinsic cast failed: {}", e))
        })?;
    let start_intrinsic = start_intrinsic_series.f64().map_err(|e| {
        ProjectionError::GeoJsonError(format!("start_intrinsic must be numeric: {}", e))
    })?;

    let end_intrinsic_series = df
        .column("end_intrinsic")
        .map_err(|e| ProjectionError::GeoJsonError(format!("Missing end_intrinsic column: {}", e)))?
        .cast(&DataType::Float64)
        .map_err(|e| ProjectionError::GeoJsonError(format!("end_intrinsic cast failed: {}", e)))?;
    let end_intrinsic = end_intrinsic_series.f64().map_err(|e| {
        ProjectionError::GeoJsonError(format!("end_intrinsic must be numeric: {}", e))
    })?;

    let gnss_start_index_series = df
        .column("gnss_start_index")
        .map_err(|e| {
            ProjectionError::GeoJsonError(format!("Missing gnss_start_index column: {}", e))
        })?
        .cast(&DataType::UInt32)
        .map_err(|e| {
            ProjectionError::GeoJsonError(format!("gnss_start_index cast failed: {}", e))
        })?;
    let gnss_start_index = gnss_start_index_series.u32().map_err(|e| {
        ProjectionError::GeoJsonError(format!("gnss_start_index must be integer: {}", e))
    })?;

    let gnss_end_index_series = df
        .column("gnss_end_index")
        .map_err(|e| {
            ProjectionError::GeoJsonError(format!("Missing gnss_end_index column: {}", e))
        })?
        .cast(&DataType::UInt32)
        .map_err(|e| ProjectionError::GeoJsonError(format!("gnss_end_index cast failed: {}", e)))?;
    let gnss_end_index = gnss_end_index_series.u32().map_err(|e| {
        ProjectionError::GeoJsonError(format!("gnss_end_index must be integer: {}", e))
    })?;

    // Build segments
    let mut segments = Vec::new();
    let row_count = df.height();

    for i in 0..row_count {
        let id = netelement_id
            .get(i)
            .ok_or_else(|| {
                ProjectionError::GeoJsonError(format!("Missing netelement_id at row {}", i))
            })?
            .to_string();

        let prob = probability.get(i).ok_or_else(|| {
            ProjectionError::GeoJsonError(format!("Missing probability at row {}", i))
        })?;

        let start_intr = start_intrinsic.get(i).ok_or_else(|| {
            ProjectionError::GeoJsonError(format!("Missing start_intrinsic at row {}", i))
        })?;

        let end_intr = end_intrinsic.get(i).ok_or_else(|| {
            ProjectionError::GeoJsonError(format!("Missing end_intrinsic at row {}", i))
        })?;

        let start_idx = gnss_start_index.get(i).ok_or_else(|| {
            ProjectionError::GeoJsonError(format!("Missing gnss_start_index at row {}", i))
        })? as usize;

        let end_idx = gnss_end_index.get(i).ok_or_else(|| {
            ProjectionError::GeoJsonError(format!("Missing gnss_end_index at row {}", i))
        })? as usize;

        let segment =
            AssociatedNetElement::new(id, prob, start_intr, end_intr, start_idx, end_idx)?;

        segments.push(segment);
    }

    // Calculate overall probability if not provided
    let overall_prob = overall_probability.unwrap_or_else(|| {
        let sum: f64 = segments.iter().map(|s| s.probability).sum();
        sum / segments.len() as f64
    });

    // Create TrainPath
    TrainPath::new(segments, overall_prob, calculated_at, None)
}

#[cfg(test)]
mod tests;

pub mod detections;