compass_teams_nhs/internal/store/store.go

// Package store owns the SQLite database (AGENTS.md §7). It is the single
// source of truth for stored items, scans, notifications and session state.
package store

import (
	"context"
	"database/sql"
	"embed"
	"fmt"
	"io/fs"
	"sort"
	"strings"
	"time"

	_ "modernc.org/sqlite" // pure-Go SQLite driver (no cgo)
)

//go:embed migrations/*.sql
var migrationFS embed.FS

// Store wraps the database handle.
type Store struct {
	db *sql.DB
}

// Open opens (creating if needed) the SQLite database at path.
func Open(path string) (*Store, error) {
	dsn := fmt.Sprintf(
		"file:%s?_pragma=busy_timeout(5000)&_pragma=journal_mode(WAL)&_pragma=foreign_keys(1)&_pragma=synchronous(NORMAL)",
		path,
	)
	db, err := sql.Open("sqlite", dsn)
	if err != nil {
		return nil, fmt.Errorf("open sqlite: %w", err)
	}
	// modernc/sqlite serialises writes; one connection avoids lock churn.
	db.SetMaxOpenConns(1)
	db.SetMaxIdleConns(1)
	return &Store{db: db}, nil
}

// DB exposes the handle for read-only queries.
func (s *Store) DB() *sql.DB { return s.db }

// Ping verifies the database is reachable.
func (s *Store) Ping(ctx context.Context) error {
	return s.db.PingContext(ctx)
}

// Close closes the database.
func (s *Store) Close() error { return s.db.Close() }

// Migrate applies every embedded migration that has not run yet. It tracks the
// applied version with PRAGMA user_version, so reruns are safe.
func (s *Store) Migrate(ctx context.Context) error {
	files, err := migrationFiles()
	if err != nil {
		return err
	}

	var version int
	if err := s.db.QueryRowContext(ctx, "PRAGMA user_version").Scan(&version); err != nil {
		return fmt.Errorf("read user_version: %w", err)
	}

	for i := version; i < len(files); i++ {
		tx, err := s.db.BeginTx(ctx, nil)
		if err != nil {
			return fmt.Errorf("begin migration %d: %w", i+1, err)
		}
		if _, err := tx.ExecContext(ctx, files[i].sql); err != nil {
			_ = tx.Rollback()
			return fmt.Errorf("migration %s: %w", files[i].name, err)
		}
		if _, err := tx.ExecContext(ctx, fmt.Sprintf("PRAGMA user_version = %d", i+1)); err != nil {
			_ = tx.Rollback()
			return fmt.Errorf("set user_version %d: %w", i+1, err)
		}
		if err := tx.Commit(); err != nil {
			return fmt.Errorf("commit migration %s: %w", files[i].name, err)
		}
	}
	return nil
}

type migration struct {
	name string
	sql  string
}

func migrationFiles() ([]migration, error) {
	entries, err := fs.ReadDir(migrationFS, "migrations")
	if err != nil {
		return nil, fmt.Errorf("read migrations: %w", err)
	}
	names := make([]string, 0, len(entries))
	for _, e := range entries {
		if !e.IsDir() {
			names = append(names, e.Name())
		}
	}
	sort.Strings(names) // 001_, 002_, ... apply in order

	out := make([]migration, 0, len(names))
	for _, n := range names {
		b, err := fs.ReadFile(migrationFS, "migrations/"+n)
		if err != nil {
			return nil, fmt.Errorf("read migration %s: %w", n, err)
		}
		out = append(out, migration{name: n, sql: string(b)})
	}
	return out, nil
}

// Child is one school child (AGENTS.md §7).
type Child struct {
	ID        int64
	Name      string
	Year      string
	CompassID string
	CreatedAt time.Time
}

// SeedChildren inserts the given names if they are not present yet. It is
// idempotent, so it is safe to call on every start. It never deletes a child.
func (s *Store) SeedChildren(ctx context.Context, names ...string) error {
	for _, name := range names {
		if _, err := s.db.ExecContext(ctx,
			`INSERT INTO children (name) VALUES (?) ON CONFLICT(name) DO NOTHING`, name); err != nil {
			return fmt.Errorf("seed child %q: %w", name, err)
		}
	}
	return nil
}

// ListChildren returns every child, oldest first.
func (s *Store) ListChildren(ctx context.Context) ([]Child, error) {
	rows, err := s.db.QueryContext(ctx,
		`SELECT id, name, COALESCE(year, ''), COALESCE(compass_id, ''), created_at
		 FROM children ORDER BY id`)
	if err != nil {
		return nil, fmt.Errorf("list children: %w", err)
	}
	defer rows.Close()

	var out []Child
	for rows.Next() {
		var c Child
		var created string
		if err := rows.Scan(&c.ID, &c.Name, &c.Year, &c.CompassID, &created); err != nil {
			return nil, fmt.Errorf("scan child: %w", err)
		}
		c.CreatedAt = parseTime(created)
		out = append(out, c)
	}
	return out, rows.Err()
}

// Scan is one row of the append-only scan log (FR-13).
type Scan struct {
	ID           int64      `json:"id"`
	ChildID      *int64     `json:"child_id,omitempty"`
	ChildName    string     `json:"child,omitempty"`
	StartedAt    time.Time  `json:"started_at"`
	FinishedAt   *time.Time `json:"finished_at,omitempty"`
	PagesFetched int        `json:"pages_fetched"`
	ItemsFound   int        `json:"items_found"`
	OK           bool       `json:"ok"`
	Error        string     `json:"error,omitempty"`
}

// ScanStatus summarises scan health for the dashboard. It exists so that a
// failed scan is shown as an explicit failure and never as an empty day
// (FR-12).
type ScanStatus struct {
	// EverRun is false until the first scan is recorded.
	EverRun bool `json:"ever_run"`
	// Total is the number of scan rows.
	Total int `json:"total"`
	// Failed is the number of rows with ok = 0.
	Failed int `json:"failed"`
	// Last is the most recent scan, when one exists.
	Last *Scan `json:"last,omitempty"`
}

// LastScan returns the most recent scan row, or nil when none exists.
func (s *Store) LastScan(ctx context.Context) (*Scan, error) {
	row := s.db.QueryRowContext(ctx,
		`SELECT sc.id, sc.child_id, COALESCE(ch.name, ''), sc.started_at,
		        sc.finished_at, sc.pages_fetched, sc.items_found, sc.ok, COALESCE(sc.error, '')
		 FROM scans sc LEFT JOIN children ch ON ch.id = sc.child_id
		 ORDER BY sc.started_at DESC, sc.id DESC LIMIT 1`)
	sc, err := scanRow(row)
	if err == sql.ErrNoRows {
		return nil, nil
	}
	if err != nil {
		return nil, err
	}
	return &sc, nil
}

// ScanSummary builds the scan health status shown on the dashboard.
func (s *Store) ScanSummary(ctx context.Context) (ScanStatus, error) {
	var st ScanStatus
	if err := s.db.QueryRowContext(ctx,
		`SELECT COUNT(*), COALESCE(SUM(CASE WHEN ok = 0 THEN 1 ELSE 0 END), 0) FROM scans`).
		Scan(&st.Total, &st.Failed); err != nil {
		return st, fmt.Errorf("scan summary: %w", err)
	}
	st.EverRun = st.Total > 0
	last, err := s.LastScan(ctx)
	if err != nil {
		return st, err
	}
	st.Last = last
	return st, nil
}

type rowScanner interface {
	Scan(dest ...any) error
}

func scanRow(r rowScanner) (Scan, error) {
	var sc Scan
	var childID sql.NullInt64
	var started string
	var finished sql.NullString
	var ok int
	if err := r.Scan(&sc.ID, &childID, &sc.ChildName, &started, &finished,
		&sc.PagesFetched, &sc.ItemsFound, &ok, &sc.Error); err != nil {
		return sc, err
	}
	if childID.Valid {
		sc.ChildID = &childID.Int64
	}
	sc.StartedAt = parseTime(started)
	if finished.Valid {
		t := parseTime(finished.String)
		sc.FinishedAt = &t
	}
	sc.OK = ok == 1
	return sc, nil
}

// parseTime accepts SQLite's default datetime format and RFC3339.
func parseTime(s string) time.Time {
	for _, layout := range []string{"2006-01-02 15:04:05", time.RFC3339} {
		if t, err := time.ParseInLocation(layout, s, time.UTC); err == nil {
			return t
		}
	}
	return time.Time{}
}

// RecordScan appends one scan row. It exists so later phases and the tests can
// prove the fail-loudly path. A non-empty errMsg records a failure even when
// nothing was fetched.
func (s *Store) RecordScan(ctx context.Context, childID *int64, pages, items int, ok bool, errMsg string) (int64, error) {
	okInt := 0
	if ok {
		okInt = 1
	}
	res, err := s.db.ExecContext(ctx,
		`INSERT INTO scans (child_id, finished_at, pages_fetched, items_found, ok, error)
		 VALUES (?, datetime('now'), ?, ?, ?, NULLIF(?, ''))`,
		childID, pages, items, okInt, errMsg)
	if err != nil {
		return 0, fmt.Errorf("record scan: %w", err)
	}
	return res.LastInsertId()
}

// Item is one row of the items table (AGENTS.md §7). It mirrors parse.Item
// without importing the parse package, so the store stays independent of the
// scraper.
type Item struct {
	ID          int64
	ChildID     int64
	Child       string
	Source      string
	Type        string
	Subject     string
	Teacher     string
	Title       string
	Body        string
	DueAt       *time.Time
	URL         string
	SourcePage  string
	Importance  string
	ExternalRef string
}

// UpsertItems writes every item for one child, de-duplicating on
// (child_id, source, type, external_ref). It returns how many rows were
// inserted and how many were refreshed, so a re-parse never duplicates an item.
func (s *Store) UpsertItems(ctx context.Context, childID int64, items []Item) (inserted, updated int, err error) {
	tx, err := s.db.BeginTx(ctx, nil)
	if err != nil {
		return 0, 0, fmt.Errorf("upsert items: begin: %w", err)
	}
	defer func() { _ = tx.Rollback() }()

	for _, it := range items {
		if it.ExternalRef == "" {
			return 0, 0, fmt.Errorf("upsert items: %q has no external reference", it.Title)
		}
		var existing int64
		err := tx.QueryRowContext(ctx,
			`SELECT id FROM items WHERE child_id = ? AND source = ? AND type = ? AND external_ref = ?`,
			childID, it.Source, it.Type, it.ExternalRef).Scan(&existing)
		switch {
		case err == sql.ErrNoRows:
			if _, err := tx.ExecContext(ctx,
				`INSERT INTO items (child_id, source, type, subject, teacher, title, body,
				                    due_at, url, source_page, importance, external_ref)
				 VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`,
				childID, it.Source, it.Type, it.Subject, it.Teacher, it.Title, it.Body,
				formatDue(it.DueAt), it.URL, it.SourcePage, it.Importance, it.ExternalRef); err != nil {
				return 0, 0, fmt.Errorf("insert item %q: %w", it.Title, err)
			}
			inserted++
		case err != nil:
			return 0, 0, fmt.Errorf("look up item %q: %w", it.Title, err)
		default:
			if _, err := tx.ExecContext(ctx,
				`UPDATE items SET subject = ?, teacher = ?, title = ?, body = ?, due_at = ?,
				                  url = ?, source_page = ?, importance = ?, updated_at = datetime('now')
				 WHERE id = ?`,
				it.Subject, it.Teacher, it.Title, it.Body, formatDue(it.DueAt),
				it.URL, it.SourcePage, it.Importance, existing); err != nil {
				return 0, 0, fmt.Errorf("update item %q: %w", it.Title, err)
			}
			updated++
		}
	}
	if err := tx.Commit(); err != nil {
		return 0, 0, fmt.Errorf("upsert items: commit: %w", err)
	}
	return inserted, updated, nil
}

// formatDue stores a due time in UTC RFC3339 so ordering is DST-safe, or NULL
// when there is no due date.
func formatDue(t *time.Time) any {
	if t == nil {
		return nil
	}
	return t.UTC().Format(time.RFC3339)
}

// DeleteChildCompassTasksFromPages removes a child's Compass task rows that came
// from the given source pages. It is used when an authoritative full task list
// replaces a partial grid: the legacy grid is capped
// (LearningTask_Default_Records_Count) and a stale row must not survive as if it
// were the whole set (AUD-114, FR-12). Rows from any other source page are left
// untouched.
func (s *Store) DeleteChildCompassTasksFromPages(ctx context.Context, childID int64, sourcePages ...string) (int64, error) {
	if len(sourcePages) == 0 {
		return 0, nil
	}
	placeholders := strings.Repeat("?,", len(sourcePages)-1) + "?"
	args := make([]any, 0, len(sourcePages)+1)
	args = append(args, childID)
	for _, p := range sourcePages {
		args = append(args, p)
	}
	res, err := s.db.ExecContext(ctx,
		`DELETE FROM items
		 WHERE child_id = ? AND source = 'compass' AND type = 'task'
		   AND source_page IN (`+placeholders+`)`, args...)
	if err != nil {
		return 0, fmt.Errorf("delete partial task rows: %w", err)
	}
	return res.RowsAffected()
}

// TypeCount is the number of items of one type for one child.
type TypeCount struct {
	Child string
	Type  string
	Count int
}

// CountItemsByType returns per-child, per-type row counts, ordered by child then
// type. It is the evidence query for Phase 3.
func (s *Store) CountItemsByType(ctx context.Context) ([]TypeCount, error) {
	rows, err := s.db.QueryContext(ctx,
		`SELECT COALESCE(ch.name, '(unknown)'), i.type, COUNT(*)
		 FROM items i LEFT JOIN children ch ON ch.id = i.child_id
		 GROUP BY ch.name, i.type
		 ORDER BY ch.name, i.type`)
	if err != nil {
		return nil, fmt.Errorf("count items: %w", err)
	}
	defer rows.Close()

	var out []TypeCount
	for rows.Next() {
		var tc TypeCount
		if err := rows.Scan(&tc.Child, &tc.Type, &tc.Count); err != nil {
			return nil, fmt.Errorf("scan item count: %w", err)
		}
		out = append(out, tc)
	}
	return out, rows.Err()
}

// SampleItems returns up to limit items for a child, newest due date first.
func (s *Store) SampleItems(ctx context.Context, childID int64, limit int) ([]Item, error) {
	rows, err := s.db.QueryContext(ctx,
		`SELECT i.id, i.child_id, COALESCE(ch.name, ''), i.source, i.type,
		        COALESCE(i.subject, ''), COALESCE(i.teacher, ''), i.title,
		        COALESCE(i.body, ''), i.due_at, COALESCE(i.url, ''),
		        COALESCE(i.source_page, ''), COALESCE(i.importance, ''), COALESCE(i.external_ref, '')
		 FROM items i LEFT JOIN children ch ON ch.id = i.child_id
		 WHERE i.child_id = ?
		 ORDER BY i.type, i.due_at DESC, i.id
		 LIMIT ?`, childID, limit)
	if err != nil {
		return nil, fmt.Errorf("sample items: %w", err)
	}
	defer rows.Close()

	var out []Item
	for rows.Next() {
		var it Item
		var due sql.NullString
		if err := rows.Scan(&it.ID, &it.ChildID, &it.Child, &it.Source, &it.Type,
			&it.Subject, &it.Teacher, &it.Title, &it.Body, &due, &it.URL,
			&it.SourcePage, &it.Importance, &it.ExternalRef); err != nil {
			return nil, fmt.Errorf("scan item: %w", err)
		}
		if due.Valid {
			t := parseTime(due.String)
			it.DueAt = &t
		}
		out = append(out, it)
	}
	return out, rows.Err()
}

// ChildByName returns one child by name.
func (s *Store) ChildByName(ctx context.Context, name string) (*Child, error) {
	row := s.db.QueryRowContext(ctx,
		`SELECT id, name, COALESCE(year, ''), COALESCE(compass_id, ''), created_at
		 FROM children WHERE name = ?`, name)
	var c Child
	var created string
	if err := row.Scan(&c.ID, &c.Name, &c.Year, &c.CompassID, &created); err != nil {
		if err == sql.ErrNoRows {
			return nil, nil
		}
		return nil, fmt.Errorf("child %q: %w", name, err)
	}
	c.CreatedAt = parseTime(created)
	return &c, nil
}

// SetChildCompassID records the child's Compass userId, discovered by the
// fetcher from the parent landing page (AGENTS.md §7).
func (s *Store) SetChildCompassID(ctx context.Context, childID int64, compassID string) error {
	if compassID == "" {
		return nil
	}
	if _, err := s.db.ExecContext(ctx,
		`UPDATE children SET compass_id = ? WHERE id = ?`, compassID, childID); err != nil {
		return fmt.Errorf("set compass id for child %d: %w", childID, err)
	}
	return nil
}

// DeleteItems removes items for one child, optionally limited to a source and a
// type. It exists so a scan can drop rows that were stored against the wrong
// child (Phase 4: nothing may leak between children) and so a later Delete
// action has a home. It returns how many rows were removed.
func (s *Store) DeleteItems(ctx context.Context, childID int64, source, itemType string) (int64, error) {
	q := `DELETE FROM items WHERE child_id = ?`
	args := []any{childID}
	if source != "" {
		q += ` AND source = ?`
		args = append(args, source)
	}
	if itemType != "" {
		q += ` AND type = ?`
		args = append(args, itemType)
	}
	res, err := s.db.ExecContext(ctx, q, args...)
	if err != nil {
		return 0, fmt.Errorf("delete items for child %d: %w", childID, err)
	}
	n, err := res.RowsAffected()
	if err != nil {
		return 0, fmt.Errorf("delete items for child %d: %w", childID, err)
	}
	return n, nil
}