banner/src/data/batch.rs

//! Batch database operations for improved performance.

use crate::banner::Course;
use crate::data::models::DbMeetingTime;
use crate::error::Result;
use sqlx::PgConnection;
use sqlx::PgPool;
use std::collections::HashSet;
use std::time::Instant;
use tracing::info;

/// Convert a Banner API course's meeting times to the DB JSONB shape.
fn to_db_meeting_times(course: &Course) -> serde_json::Value {
    let meetings: Vec<DbMeetingTime> = course
        .meetings_faculty
        .iter()
        .map(|mf| {
            let mt = &mf.meeting_time;
            DbMeetingTime {
                begin_time: mt.begin_time.clone(),
                end_time: mt.end_time.clone(),
                start_date: mt.start_date.clone(),
                end_date: mt.end_date.clone(),
                monday: mt.monday,
                tuesday: mt.tuesday,
                wednesday: mt.wednesday,
                thursday: mt.thursday,
                friday: mt.friday,
                saturday: mt.saturday,
                sunday: mt.sunday,
                building: mt.building.clone(),
                building_description: mt.building_description.clone(),
                room: mt.room.clone(),
                campus: mt.campus.clone(),
                meeting_type: mt.meeting_type.clone(),
                meeting_schedule_type: mt.meeting_schedule_type.clone(),
            }
        })
        .collect();
    serde_json::to_value(meetings).unwrap_or_default()
}

/// Convert a Banner API course's section attributes to a JSONB array of code strings.
fn to_db_attributes(course: &Course) -> serde_json::Value {
    let codes: Vec<&str> = course
        .section_attributes
        .iter()
        .map(|a| a.code.as_str())
        .collect();
    serde_json::to_value(codes).unwrap_or_default()
}

/// Extract the campus code from the first meeting time (Banner doesn't put it on the course directly).
fn extract_campus_code(course: &Course) -> Option<String> {
    course
        .meetings_faculty
        .first()
        .and_then(|mf| mf.meeting_time.campus.clone())
}

// ---------------------------------------------------------------------------
// Task 1: UpsertDiffRow — captures pre- and post-upsert state for diffing
// ---------------------------------------------------------------------------

/// Row returned by the CTE-based upsert query, carrying both old and new values
/// for every auditable field. `old_id` is `None` for fresh inserts.
#[derive(sqlx::FromRow, Debug)]
struct UpsertDiffRow {
    id: i32,
    old_id: Option<i32>,

    // enrollment fields
    old_enrollment: Option<i32>,
    new_enrollment: i32,
    old_max_enrollment: Option<i32>,
    new_max_enrollment: i32,
    old_wait_count: Option<i32>,
    new_wait_count: i32,
    old_wait_capacity: Option<i32>,
    new_wait_capacity: i32,

    // text fields (non-nullable in DB)
    old_subject: Option<String>,
    new_subject: String,
    old_course_number: Option<String>,
    new_course_number: String,
    old_title: Option<String>,
    new_title: String,

    // nullable text fields
    old_sequence_number: Option<String>,
    new_sequence_number: Option<String>,
    old_part_of_term: Option<String>,
    new_part_of_term: Option<String>,
    old_instructional_method: Option<String>,
    new_instructional_method: Option<String>,
    old_campus: Option<String>,
    new_campus: Option<String>,

    // nullable int fields
    old_credit_hours: Option<i32>,
    new_credit_hours: Option<i32>,
    old_credit_hour_low: Option<i32>,
    new_credit_hour_low: Option<i32>,
    old_credit_hour_high: Option<i32>,
    new_credit_hour_high: Option<i32>,

    // cross-list fields
    old_cross_list: Option<String>,
    new_cross_list: Option<String>,
    old_cross_list_capacity: Option<i32>,
    new_cross_list_capacity: Option<i32>,
    old_cross_list_count: Option<i32>,
    new_cross_list_count: Option<i32>,

    // link fields
    old_link_identifier: Option<String>,
    new_link_identifier: Option<String>,
    old_is_section_linked: Option<bool>,
    new_is_section_linked: Option<bool>,

    // JSONB fields
    old_meeting_times: Option<serde_json::Value>,
    new_meeting_times: serde_json::Value,
    old_attributes: Option<serde_json::Value>,
    new_attributes: serde_json::Value,
}

// ---------------------------------------------------------------------------
// Task 3: Entry types and diff logic
// ---------------------------------------------------------------------------

struct AuditEntry {
    course_id: i32,
    field_changed: &'static str,
    old_value: String,
    new_value: String,
}

struct MetricEntry {
    course_id: i32,
    enrollment: i32,
    wait_count: i32,
    seats_available: i32,
}

/// Compare old vs new for a single field, pushing an `AuditEntry` when they differ.
///
/// Three variants:
/// - `diff_field!(audits, row, field_name, old_field, new_field)` — `Option<T>` old vs `T` new
/// - `diff_field!(opt audits, row, field_name, old_field, new_field)` — `Option<T>` old vs `Option<T>` new
/// - `diff_field!(json audits, row, field_name, old_field, new_field)` — `Option<Value>` old vs `Value` new
///
/// All variants skip when `old_id` is None (fresh insert).
macro_rules! diff_field {
    // Standard: Option<T> old vs T new (non-nullable columns)
    ($audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
        if $row.old_id.is_some() {
            let old_str = $row
                .$old
                .as_ref()
                .map(|v| v.to_string())
                .unwrap_or_default();
            let new_str = $row.$new.to_string();
            if old_str != new_str {
                $audits.push(AuditEntry {
                    course_id: $row.id,
                    field_changed: $field,
                    old_value: old_str,
                    new_value: new_str,
                });
            }
        }
    };
    // Nullable: Option<T> old vs Option<T> new
    (opt $audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
        if $row.old_id.is_some() {
            let old_str = $row
                .$old
                .as_ref()
                .map(|v| v.to_string())
                .unwrap_or_default();
            let new_str = $row
                .$new
                .as_ref()
                .map(|v| v.to_string())
                .unwrap_or_default();
            if old_str != new_str {
                $audits.push(AuditEntry {
                    course_id: $row.id,
                    field_changed: $field,
                    old_value: old_str,
                    new_value: new_str,
                });
            }
        }
    };
    // JSONB: Option<Value> old vs Value new
    (json $audits:ident, $row:ident, $field:expr, $old:ident, $new:ident) => {
        if $row.old_id.is_some() {
            let old_val = $row
                .$old
                .as_ref()
                .cloned()
                .unwrap_or(serde_json::Value::Null);
            let new_val = &$row.$new;
            if old_val != *new_val {
                $audits.push(AuditEntry {
                    course_id: $row.id,
                    field_changed: $field,
                    old_value: old_val.to_string(),
                    new_value: new_val.to_string(),
                });
            }
        }
    };
}

/// Compute audit entries (field-level diffs) and metric entries from upsert diff rows.
fn compute_diffs(rows: &[UpsertDiffRow]) -> (Vec<AuditEntry>, Vec<MetricEntry>) {
    let mut audits = Vec::new();
    let mut metrics = Vec::new();

    for row in rows {
        // Non-nullable fields
        diff_field!(audits, row, "enrollment", old_enrollment, new_enrollment);
        diff_field!(
            audits,
            row,
            "max_enrollment",
            old_max_enrollment,
            new_max_enrollment
        );
        diff_field!(audits, row, "wait_count", old_wait_count, new_wait_count);
        diff_field!(
            audits,
            row,
            "wait_capacity",
            old_wait_capacity,
            new_wait_capacity
        );
        diff_field!(audits, row, "subject", old_subject, new_subject);
        diff_field!(
            audits,
            row,
            "course_number",
            old_course_number,
            new_course_number
        );
        diff_field!(audits, row, "title", old_title, new_title);

        // Nullable text fields
        diff_field!(opt audits, row, "sequence_number", old_sequence_number, new_sequence_number);
        diff_field!(opt audits, row, "part_of_term", old_part_of_term, new_part_of_term);
        diff_field!(opt audits, row, "instructional_method", old_instructional_method, new_instructional_method);
        diff_field!(opt audits, row, "campus", old_campus, new_campus);

        // Nullable int fields
        diff_field!(opt audits, row, "credit_hours", old_credit_hours, new_credit_hours);
        diff_field!(opt audits, row, "credit_hour_low", old_credit_hour_low, new_credit_hour_low);
        diff_field!(opt audits, row, "credit_hour_high", old_credit_hour_high, new_credit_hour_high);

        // Cross-list fields
        diff_field!(opt audits, row, "cross_list", old_cross_list, new_cross_list);
        diff_field!(opt audits, row, "cross_list_capacity", old_cross_list_capacity, new_cross_list_capacity);
        diff_field!(opt audits, row, "cross_list_count", old_cross_list_count, new_cross_list_count);

        // Link fields
        diff_field!(opt audits, row, "link_identifier", old_link_identifier, new_link_identifier);
        diff_field!(opt audits, row, "is_section_linked", old_is_section_linked, new_is_section_linked);

        // JSONB fields
        diff_field!(json audits, row, "meeting_times", old_meeting_times, new_meeting_times);
        diff_field!(json audits, row, "attributes", old_attributes, new_attributes);

        // Emit a metric entry when enrollment/wait_count/max_enrollment changed
        // Skip fresh inserts (no old data to compare against)
        let enrollment_changed = row.old_id.is_some()
            && (row.old_enrollment != Some(row.new_enrollment)
                || row.old_wait_count != Some(row.new_wait_count)
                || row.old_max_enrollment != Some(row.new_max_enrollment));

        if enrollment_changed {
            metrics.push(MetricEntry {
                course_id: row.id,
                enrollment: row.new_enrollment,
                wait_count: row.new_wait_count,
                seats_available: row.new_max_enrollment - row.new_enrollment,
            });
        }
    }

    (audits, metrics)
}

// ---------------------------------------------------------------------------
// Task 4: Batch insert functions for audits and metrics
// ---------------------------------------------------------------------------

async fn insert_audits(audits: &[AuditEntry], conn: &mut PgConnection) -> Result<()> {
    if audits.is_empty() {
        return Ok(());
    }

    let course_ids: Vec<i32> = audits.iter().map(|a| a.course_id).collect();
    let fields: Vec<&str> = audits.iter().map(|a| a.field_changed).collect();
    let old_values: Vec<&str> = audits.iter().map(|a| a.old_value.as_str()).collect();
    let new_values: Vec<&str> = audits.iter().map(|a| a.new_value.as_str()).collect();

    sqlx::query(
        r#"
        INSERT INTO course_audits (course_id, timestamp, field_changed, old_value, new_value)
        SELECT v.course_id, NOW(), v.field_changed, v.old_value, v.new_value
        FROM UNNEST($1::int4[], $2::text[], $3::text[], $4::text[])
            AS v(course_id, field_changed, old_value, new_value)
        "#,
    )
    .bind(&course_ids)
    .bind(&fields)
    .bind(&old_values)
    .bind(&new_values)
    .execute(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch insert course_audits: {}", e))?;

    Ok(())
}

async fn insert_metrics(metrics: &[MetricEntry], conn: &mut PgConnection) -> Result<()> {
    if metrics.is_empty() {
        return Ok(());
    }

    let course_ids: Vec<i32> = metrics.iter().map(|m| m.course_id).collect();
    let enrollments: Vec<i32> = metrics.iter().map(|m| m.enrollment).collect();
    let wait_counts: Vec<i32> = metrics.iter().map(|m| m.wait_count).collect();
    let seats_available: Vec<i32> = metrics.iter().map(|m| m.seats_available).collect();

    sqlx::query(
        r#"
        INSERT INTO course_metrics (course_id, timestamp, enrollment, wait_count, seats_available)
        SELECT v.course_id, NOW(), v.enrollment, v.wait_count, v.seats_available
        FROM UNNEST($1::int4[], $2::int4[], $3::int4[], $4::int4[])
            AS v(course_id, enrollment, wait_count, seats_available)
        "#,
    )
    .bind(&course_ids)
    .bind(&enrollments)
    .bind(&wait_counts)
    .bind(&seats_available)
    .execute(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch insert course_metrics: {}", e))?;

    Ok(())
}

// ---------------------------------------------------------------------------
// Core upsert functions (updated to use &mut PgConnection)
// ---------------------------------------------------------------------------

/// Batch upsert courses in a single database query.
///
/// Performs a bulk INSERT...ON CONFLICT DO UPDATE for all courses, including
/// new fields (meeting times, attributes, instructor data). Captures pre-update
/// state for audit/metric tracking, all within a single transaction.
///
/// # Performance
/// - Reduces N database round-trips to 5 (old-data CTE + upsert, audits, metrics, instructors, junction)
/// - Typical usage: 50-200 courses per batch
pub async fn batch_upsert_courses(courses: &[Course], db_pool: &PgPool) -> Result<()> {
    if courses.is_empty() {
        info!("No courses to upsert, skipping batch operation");
        return Ok(());
    }

    let start = Instant::now();
    let course_count = courses.len();

    let mut tx = db_pool.begin().await?;

    // Step 1: Upsert courses with CTE, returning diff rows
    let diff_rows = upsert_courses(courses, &mut tx).await?;

    // Step 2: Extract course IDs for instructor linking
    let course_ids: Vec<i32> = diff_rows.iter().map(|r| r.id).collect();

    // Step 3: Compute audit/metric diffs
    let (audits, metrics) = compute_diffs(&diff_rows);

    // Step 4: Insert audits and metrics
    insert_audits(&audits, &mut tx).await?;
    insert_metrics(&metrics, &mut tx).await?;

    // Step 5: Upsert instructors (deduplicated across batch)
    upsert_instructors(courses, &mut tx).await?;

    // Step 6: Link courses to instructors via junction table
    upsert_course_instructors(courses, &course_ids, &mut tx).await?;

    tx.commit().await?;

    let duration = start.elapsed();
    info!(
        courses_count = course_count,
        audit_entries = audits.len(),
        metric_entries = metrics.len(),
        duration_ms = duration.as_millis(),
        "Batch upserted courses with instructors, audits, and metrics"
    );

    Ok(())
}

// ---------------------------------------------------------------------------
// Task 2: CTE-based upsert returning old+new values
// ---------------------------------------------------------------------------

/// Upsert all courses and return diff rows with old and new values for auditing.
async fn upsert_courses(courses: &[Course], conn: &mut PgConnection) -> Result<Vec<UpsertDiffRow>> {
    let crns: Vec<&str> = courses
        .iter()
        .map(|c| c.course_reference_number.as_str())
        .collect();
    let subjects: Vec<&str> = courses.iter().map(|c| c.subject.as_str()).collect();
    let course_numbers: Vec<&str> = courses.iter().map(|c| c.course_number.as_str()).collect();
    let titles: Vec<&str> = courses.iter().map(|c| c.course_title.as_str()).collect();
    let term_codes: Vec<&str> = courses.iter().map(|c| c.term.as_str()).collect();
    let enrollments: Vec<i32> = courses.iter().map(|c| c.enrollment).collect();
    let max_enrollments: Vec<i32> = courses.iter().map(|c| c.maximum_enrollment).collect();
    let wait_counts: Vec<i32> = courses.iter().map(|c| c.wait_count).collect();
    let wait_capacities: Vec<i32> = courses.iter().map(|c| c.wait_capacity).collect();

    // New scalar fields
    let sequence_numbers: Vec<Option<&str>> = courses
        .iter()
        .map(|c| Some(c.sequence_number.as_str()))
        .collect();
    let parts_of_term: Vec<Option<&str>> = courses
        .iter()
        .map(|c| Some(c.part_of_term.as_str()))
        .collect();
    let instructional_methods: Vec<Option<&str>> = courses
        .iter()
        .map(|c| Some(c.instructional_method.as_str()))
        .collect();
    let campuses: Vec<Option<String>> = courses.iter().map(extract_campus_code).collect();
    let credit_hours: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hours).collect();
    let credit_hour_lows: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hour_low).collect();
    let credit_hour_highs: Vec<Option<i32>> = courses.iter().map(|c| c.credit_hour_high).collect();
    let cross_lists: Vec<Option<&str>> = courses.iter().map(|c| c.cross_list.as_deref()).collect();
    let cross_list_capacities: Vec<Option<i32>> =
        courses.iter().map(|c| c.cross_list_capacity).collect();
    let cross_list_counts: Vec<Option<i32>> = courses.iter().map(|c| c.cross_list_count).collect();
    let link_identifiers: Vec<Option<&str>> = courses
        .iter()
        .map(|c| c.link_identifier.as_deref())
        .collect();
    let is_section_linkeds: Vec<Option<bool>> =
        courses.iter().map(|c| Some(c.is_section_linked)).collect();

    // JSONB fields
    let meeting_times_json: Vec<serde_json::Value> =
        courses.iter().map(to_db_meeting_times).collect();
    let attributes_json: Vec<serde_json::Value> = courses.iter().map(to_db_attributes).collect();

    let rows = sqlx::query_as::<_, UpsertDiffRow>(
        r#"
        WITH old_data AS (
            SELECT id, enrollment, max_enrollment, wait_count, wait_capacity,
                   subject, course_number, title,
                   sequence_number, part_of_term, instructional_method, campus,
                   credit_hours, credit_hour_low, credit_hour_high,
                   cross_list, cross_list_capacity, cross_list_count,
                   link_identifier, is_section_linked,
                   meeting_times, attributes,
                   crn, term_code
            FROM courses
            WHERE (crn, term_code) IN (SELECT * FROM UNNEST($1::text[], $5::text[]))
        ),
        upserted AS (
            INSERT INTO courses (
                crn, subject, course_number, title, term_code,
                enrollment, max_enrollment, wait_count, wait_capacity, last_scraped_at,
                sequence_number, part_of_term, instructional_method, campus,
                credit_hours, credit_hour_low, credit_hour_high,
                cross_list, cross_list_capacity, cross_list_count,
                link_identifier, is_section_linked,
                meeting_times, attributes
            )
            SELECT
                v.crn, v.subject, v.course_number, v.title, v.term_code,
                v.enrollment, v.max_enrollment, v.wait_count, v.wait_capacity, NOW(),
                v.sequence_number, v.part_of_term, v.instructional_method, v.campus,
                v.credit_hours, v.credit_hour_low, v.credit_hour_high,
                v.cross_list, v.cross_list_capacity, v.cross_list_count,
                v.link_identifier, v.is_section_linked,
                v.meeting_times, v.attributes
            FROM UNNEST(
                $1::text[], $2::text[], $3::text[], $4::text[], $5::text[],
                $6::int4[], $7::int4[], $8::int4[], $9::int4[],
                $10::text[], $11::text[], $12::text[], $13::text[],
                $14::int4[], $15::int4[], $16::int4[],
                $17::text[], $18::int4[], $19::int4[],
                $20::text[], $21::bool[],
                $22::jsonb[], $23::jsonb[]
            ) AS v(
                crn, subject, course_number, title, term_code,
                enrollment, max_enrollment, wait_count, wait_capacity,
                sequence_number, part_of_term, instructional_method, campus,
                credit_hours, credit_hour_low, credit_hour_high,
                cross_list, cross_list_capacity, cross_list_count,
                link_identifier, is_section_linked,
                meeting_times, attributes
            )
            ON CONFLICT (crn, term_code)
            DO UPDATE SET
                subject = EXCLUDED.subject,
                course_number = EXCLUDED.course_number,
                title = EXCLUDED.title,
                enrollment = EXCLUDED.enrollment,
                max_enrollment = EXCLUDED.max_enrollment,
                wait_count = EXCLUDED.wait_count,
                wait_capacity = EXCLUDED.wait_capacity,
                last_scraped_at = EXCLUDED.last_scraped_at,
                sequence_number = EXCLUDED.sequence_number,
                part_of_term = EXCLUDED.part_of_term,
                instructional_method = EXCLUDED.instructional_method,
                campus = EXCLUDED.campus,
                credit_hours = EXCLUDED.credit_hours,
                credit_hour_low = EXCLUDED.credit_hour_low,
                credit_hour_high = EXCLUDED.credit_hour_high,
                cross_list = EXCLUDED.cross_list,
                cross_list_capacity = EXCLUDED.cross_list_capacity,
                cross_list_count = EXCLUDED.cross_list_count,
                link_identifier = EXCLUDED.link_identifier,
                is_section_linked = EXCLUDED.is_section_linked,
                meeting_times = EXCLUDED.meeting_times,
                attributes = EXCLUDED.attributes
            RETURNING *
        )
        SELECT u.id,
               o.id AS old_id,
               o.enrollment AS old_enrollment, u.enrollment AS new_enrollment,
               o.max_enrollment AS old_max_enrollment, u.max_enrollment AS new_max_enrollment,
               o.wait_count AS old_wait_count, u.wait_count AS new_wait_count,
               o.wait_capacity AS old_wait_capacity, u.wait_capacity AS new_wait_capacity,
               o.subject AS old_subject, u.subject AS new_subject,
               o.course_number AS old_course_number, u.course_number AS new_course_number,
               o.title AS old_title, u.title AS new_title,
               o.sequence_number AS old_sequence_number, u.sequence_number AS new_sequence_number,
               o.part_of_term AS old_part_of_term, u.part_of_term AS new_part_of_term,
               o.instructional_method AS old_instructional_method, u.instructional_method AS new_instructional_method,
               o.campus AS old_campus, u.campus AS new_campus,
               o.credit_hours AS old_credit_hours, u.credit_hours AS new_credit_hours,
               o.credit_hour_low AS old_credit_hour_low, u.credit_hour_low AS new_credit_hour_low,
               o.credit_hour_high AS old_credit_hour_high, u.credit_hour_high AS new_credit_hour_high,
               o.cross_list AS old_cross_list, u.cross_list AS new_cross_list,
               o.cross_list_capacity AS old_cross_list_capacity, u.cross_list_capacity AS new_cross_list_capacity,
               o.cross_list_count AS old_cross_list_count, u.cross_list_count AS new_cross_list_count,
               o.link_identifier AS old_link_identifier, u.link_identifier AS new_link_identifier,
               o.is_section_linked AS old_is_section_linked, u.is_section_linked AS new_is_section_linked,
               o.meeting_times AS old_meeting_times, u.meeting_times AS new_meeting_times,
               o.attributes AS old_attributes, u.attributes AS new_attributes
        FROM upserted u
        LEFT JOIN old_data o ON u.crn = o.crn AND u.term_code = o.term_code
        "#,
    )
    .bind(&crns)
    .bind(&subjects)
    .bind(&course_numbers)
    .bind(&titles)
    .bind(&term_codes)
    .bind(&enrollments)
    .bind(&max_enrollments)
    .bind(&wait_counts)
    .bind(&wait_capacities)
    .bind(&sequence_numbers)
    .bind(&parts_of_term)
    .bind(&instructional_methods)
    .bind(&campuses)
    .bind(&credit_hours)
    .bind(&credit_hour_lows)
    .bind(&credit_hour_highs)
    .bind(&cross_lists)
    .bind(&cross_list_capacities)
    .bind(&cross_list_counts)
    .bind(&link_identifiers)
    .bind(&is_section_linkeds)
    .bind(&meeting_times_json)
    .bind(&attributes_json)
    .fetch_all(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch upsert courses: {}", e))?;

    Ok(rows)
}

/// Deduplicate and upsert all instructors from the batch.
async fn upsert_instructors(courses: &[Course], conn: &mut PgConnection) -> Result<()> {
    let mut seen = HashSet::new();
    let mut banner_ids = Vec::new();
    let mut display_names = Vec::new();
    let mut emails: Vec<Option<&str>> = Vec::new();

    for course in courses {
        for faculty in &course.faculty {
            if seen.insert(faculty.banner_id.as_str()) {
                banner_ids.push(faculty.banner_id.as_str());
                display_names.push(faculty.display_name.as_str());
                emails.push(faculty.email_address.as_deref());
            }
        }
    }

    if banner_ids.is_empty() {
        return Ok(());
    }

    sqlx::query(
        r#"
        INSERT INTO instructors (banner_id, display_name, email)
        SELECT * FROM UNNEST($1::text[], $2::text[], $3::text[])
        ON CONFLICT (banner_id)
        DO UPDATE SET
            display_name = EXCLUDED.display_name,
            email = COALESCE(EXCLUDED.email, instructors.email)
        "#,
    )
    .bind(&banner_ids)
    .bind(&display_names)
    .bind(&emails)
    .execute(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch upsert instructors: {}", e))?;

    Ok(())
}

/// Link courses to their instructors via the junction table.
async fn upsert_course_instructors(
    courses: &[Course],
    course_ids: &[i32],
    conn: &mut PgConnection,
) -> Result<()> {
    let mut cids = Vec::new();
    let mut iids = Vec::new();
    let mut primaries = Vec::new();

    for (course, &course_id) in courses.iter().zip(course_ids) {
        for faculty in &course.faculty {
            cids.push(course_id);
            iids.push(faculty.banner_id.as_str());
            primaries.push(faculty.primary_indicator);
        }
    }

    if cids.is_empty() {
        return Ok(());
    }

    // Delete existing links for these courses then re-insert.
    // This handles instructor changes cleanly.
    sqlx::query("DELETE FROM course_instructors WHERE course_id = ANY($1)")
        .bind(&cids)
        .execute(&mut *conn)
        .await?;

    sqlx::query(
        r#"
        INSERT INTO course_instructors (course_id, instructor_id, is_primary)
        SELECT * FROM UNNEST($1::int4[], $2::text[], $3::bool[])
        ON CONFLICT (course_id, instructor_id)
        DO UPDATE SET is_primary = EXCLUDED.is_primary
        "#,
    )
    .bind(&cids)
    .bind(&iids)
    .bind(&primaries)
    .execute(&mut *conn)
    .await
    .map_err(|e| anyhow::anyhow!("Failed to batch upsert course_instructors: {}", e))?;

    Ok(())
}