package db

import (
	"database/sql"
	"fmt"
	"log/slog"
	"os"
	"path/filepath"
	"slices"
	"strings"
	"testing"
)

// TestSchemaMatchesMigrations enforces the invariant that schema.sql and the
// migrations describe the same database.
//
// The two are supposed to be updated in lockstep, and nothing checked that they
// were. The cost of them drifting is already on the record: migration 0009
// exists only because "Migration 0004 was supposed to drop these columns but
// either failed or was only recorded (not executed) on production", which left
// production carrying eleven columns that fresh installs did not have.
//
// The check builds the schema both ways and compares them:
//
//	A: apply schema.sql                        (what a fresh install gets)
//	B: apply testdata/base_schema.sql, then
//	   run every migration                     (what an upgrade gets)
//
// Columns are compared as a SET, ignoring ordinal position. That is deliberate.
// Migrations append with ALTER TABLE ADD COLUMN while schema.sql places the same
// column mid-table, so the two orders legitimately differ on manifests, users,
// devices and repo_pages. Reordering those tables would mean four rebuild
// migrations for no functional gain, and nothing in pkg/appview depends on
// ordinal position (no SELECT *, no column-less INSERT ... VALUES). If you add
// code that does depend on column order, this test will not save you.
func TestSchemaMatchesMigrations(t *testing.T) {
	fresh := freshSchemaDB(t)
	defer fresh.Close()

	migrated := migratedSchemaDB(t)
	defer migrated.Close()

	freshSnap, err := introspectSchema(fresh)
	if err != nil {
		t.Fatalf("introspect fresh database: %v", err)
	}
	migratedSnap, err := introspectSchema(migrated)
	if err != nil {
		t.Fatalf("introspect migrated database: %v", err)
	}

	freshTables := sortedKeys(freshSnap.Tables)
	migratedTables := sortedKeys(migratedSnap.Tables)

	if diff := diffStringSets(freshTables, migratedTables); diff != "" {
		t.Errorf("table sets differ between schema.sql and the migrations:\n%s\n\n"+
			"A table only in schema.sql will never exist on an upgraded database, because InitDB\n"+
			"skips schema.sql entirely once schema_migrations has rows. New tables need BOTH a\n"+
			"schema.sql entry and a migration.", diff)
	}

	for _, table := range freshTables {
		if !slices.Contains(migratedTables, table) {
			continue // already reported above
		}
		if diff := diffStringSets(freshSnap.Tables[table], migratedSnap.Tables[table]); diff != "" {
			t.Errorf("table %q differs between schema.sql and the migrations:\n%s", table, diff)
		}
	}

	if diff := diffStringSets(freshSnap.Indexes, migratedSnap.Indexes); diff != "" {
		t.Errorf("index sets differ between schema.sql and the migrations:\n%s", diff)
	}
}

// TestSchemaDriftCleanOnFreshDatabase pins the other end of the same invariant:
// SchemaDrift, which runs at boot against real databases, must report nothing
// for a database that was just built from schema.sql. If this ever fails, the
// startup check has become noisy and operators will learn to ignore it, which
// costs more than not having it at all.
func TestSchemaDriftCleanOnFreshDatabase(t *testing.T) {
	db := freshSchemaDB(t)
	defer db.Close()

	findings, err := SchemaDrift(db)
	if err != nil {
		t.Fatalf("SchemaDrift: %v", err)
	}
	if len(findings) != 0 {
		t.Errorf("expected no drift against a database built from schema.sql, got %d:\n  %s",
			len(findings), strings.Join(findings, "\n  "))
	}
}

// TestSchemaDriftDetectsMissingColumn proves SchemaDrift actually compares
// rather than always returning clean. A database missing a column schema.sql
// declares is exactly the shape of the 0004 incident.
func TestSchemaDriftDetectsMissingColumn(t *testing.T) {
	db := freshSchemaDB(t)
	defer db.Close()

	// Rebuild users without registry_domain, imitating a migration that was
	// recorded but never executed.
	stmts := []string{
		`CREATE TABLE users_drifted (
			did TEXT PRIMARY KEY,
			handle TEXT NOT NULL,
			pds_endpoint TEXT NOT NULL,
			avatar TEXT,
			default_hold_did TEXT,
			oci_client TEXT DEFAULT '',
			last_seen TIMESTAMP NOT NULL,
			UNIQUE(handle)
		)`,
		`DROP TABLE users`,
		`ALTER TABLE users_drifted RENAME TO users`,
	}
	for _, s := range stmts {
		if _, err := db.Exec(s); err != nil {
			t.Fatalf("set up drifted schema: %v", err)
		}
	}

	findings, err := SchemaDrift(db)
	if err != nil {
		t.Fatalf("SchemaDrift: %v", err)
	}
	if !anyContains(findings, "registry_domain") {
		t.Errorf("expected a finding naming the missing registry_domain column, got: %v", findings)
	}
}

// TestSchemaDriftIgnoresIndexSpacing covers the case that made the boot-time
// check useless in production.
//
// referenceSnapshot applies schema.sql to a LOCAL libsql database, which stores
// the CREATE statement byte for byte. Production runs an embedded replica: the
// write goes to the remote first and the remote re-emits the DDL through its own
// parser, which puts a space before the column list. Every index then failed the
// equality check from both directions at once, so all 38 were reported twice and
// the warning read "differences=76" on every single boot. Real drift would have
// been one line in seventy-seven.
//
// Nothing caught it because every test runs local-only, where both sides are
// produced by the same writer and therefore agree by accident. This test injects
// the remote's spelling by hand so the discrimination is actually exercised.
func TestSchemaDriftIgnoresIndexSpacing(t *testing.T) {
	db := freshSchemaDB(t)
	defer db.Close()

	// Each pair is (index to drop, spaced respelling of the identical index).
	// The spellings on the right are what a remote-parsed database stores.
	respell := []struct{ drop, create string }{
		{"idx_webhooks_user", `CREATE INDEX idx_webhooks_user ON webhooks (user_did)`},
		{"idx_manifests_did_repo", `CREATE INDEX idx_manifests_did_repo ON manifests (did, repository)`},
		{"idx_manifests_created_at", `CREATE INDEX idx_manifests_created_at ON manifests ( created_at DESC )`},
		{"idx_instance_leases_expires", `CREATE INDEX idx_instance_leases_expires ON instance_leases (expires_at)`},
	}
	for _, r := range respell {
		if _, err := db.Exec("DROP INDEX " + r.drop); err != nil {
			t.Fatalf("drop %s: %v", r.drop, err)
		}
		if _, err := db.Exec(r.create); err != nil {
			t.Fatalf("recreate %s: %v", r.drop, err)
		}
	}

	findings, err := SchemaDrift(db)
	if err != nil {
		t.Fatalf("SchemaDrift: %v", err)
	}
	if len(findings) != 0 {
		t.Errorf("index spacing is not a schema difference, but SchemaDrift reported %d:\n  %s",
			len(findings), strings.Join(findings, "\n  "))
	}
}

// TestSchemaDriftDetectsChangedIndex is the other half of the pair above.
// Normalizing spacing must not blunt the check: an index that genuinely differs
// in its columns, its table or its uniqueness still has to be reported, in the
// spaced spelling as much as the unspaced one.
func TestSchemaDriftDetectsChangedIndex(t *testing.T) {
	cases := []struct {
		name    string
		drop    string
		create  string
		wantHit string
	}{
		{
			name:    "different column",
			drop:    "idx_webhooks_user",
			create:  `CREATE INDEX idx_webhooks_user ON webhooks (url)`,
			wantHit: "idx_webhooks_user",
		},
		{
			name:    "extra column",
			drop:    "idx_manifests_did_repo",
			create:  `CREATE INDEX idx_manifests_did_repo ON manifests (did, repository, digest)`,
			wantHit: "idx_manifests_did_repo",
		},
		{
			name:    "uniqueness changed",
			drop:    "idx_stars_starrer",
			create:  `CREATE UNIQUE INDEX idx_stars_starrer ON stars (starrer_did)`,
			wantHit: "idx_stars_starrer",
		},
	}

	for _, tc := range cases {
		t.Run(tc.name, func(t *testing.T) {
			db := freshSchemaDB(t)
			defer db.Close()

			if _, err := db.Exec("DROP INDEX " + tc.drop); err != nil {
				t.Fatalf("drop %s: %v", tc.drop, err)
			}
			if _, err := db.Exec(tc.create); err != nil {
				t.Fatalf("create replacement index: %v", err)
			}

			findings, err := SchemaDrift(db)
			if err != nil {
				t.Fatalf("SchemaDrift: %v", err)
			}
			if !anyContains(findings, tc.wantHit) {
				t.Errorf("expected a finding naming %s, got: %v", tc.wantHit, findings)
			}
		})
	}
}

// TestNormalizeIndexDDL pins the normalization itself, without a database in the
// way: spacing around punctuation is noise, everything else is signal.
func TestNormalizeIndexDDL(t *testing.T) {
	same := [][]string{
		{
			`CREATE INDEX idx_webhooks_user ON webhooks(user_did)`,
			`CREATE INDEX idx_webhooks_user ON webhooks (user_did)`,
			`CREATE INDEX IF NOT EXISTS idx_webhooks_user ON webhooks ( user_did )`,
			"CREATE INDEX idx_webhooks_user\n  ON webhooks\n  (user_did)",
		},
		{
			`CREATE INDEX i ON t(a, b)`,
			`CREATE INDEX i ON t(a,b)`,
			`CREATE INDEX i ON t ( a , b )`,
		},
		{
			`CREATE INDEX i ON t(created_at DESC)`,
			`CREATE INDEX i ON t (created_at DESC)`,
		},
	}
	for _, group := range same {
		want := normalizeIndexDDL(group[0])
		for _, variant := range group[1:] {
			if got := normalizeIndexDDL(variant); got != want {
				t.Errorf("spelling difference survived normalization:\n  %q -> %q\n  %q -> %q",
					group[0], want, variant, got)
			}
		}
	}

	different := [][2]string{
		{`CREATE INDEX i ON t(a)`, `CREATE INDEX i ON t(b)`},
		{`CREATE INDEX i ON t(a)`, `CREATE INDEX i ON u(a)`},
		{`CREATE INDEX i ON t(a)`, `CREATE UNIQUE INDEX i ON t(a)`},
		{`CREATE INDEX i ON t(a, b)`, `CREATE INDEX i ON t(b, a)`},
		{`CREATE INDEX i ON t(a)`, `CREATE INDEX i ON t(a DESC)`},
	}
	for _, pair := range different {
		if normalizeIndexDDL(pair[0]) == normalizeIndexDDL(pair[1]) {
			t.Errorf("normalization collapsed two genuinely different indexes: %q and %q",
				pair[0], pair[1])
		}
	}
}

// TestInitDBWarnsOnDriftAndStillBoots covers the integration rather than the
// comparison: an existing database whose schema has drifted must still open.
//
// The check is warn-only on purpose. A database that is merely ahead of or
// behind schema.sql is almost always still able to serve traffic, so refusing to
// boot would convert a diff that wants a corrective migration into an outage,
// and it would do so during a deploy, which is the worst possible moment.
func TestInitDBWarnsOnDriftAndStillBoots(t *testing.T) {
	path := filepath.Join(t.TempDir(), "app.db")

	first, err := InitDB(path, LibsqlConfig{})
	if err != nil {
		t.Fatalf("first InitDB: %v", err)
	}
	// Imitate a migration that was recorded but never executed.
	if _, err := first.Exec(`ALTER TABLE users DROP COLUMN registry_domain`); err != nil {
		t.Fatalf("drop column: %v", err)
	}
	first.Close()

	var logged strings.Builder
	restore := slog.Default()
	slog.SetDefault(slog.New(slog.NewTextHandler(&logged, &slog.HandlerOptions{Level: slog.LevelWarn})))
	defer slog.SetDefault(restore)

	// The second open takes the existing-database path, which is the only one
	// that runs the drift check.
	second, err := InitDB(path, LibsqlConfig{})
	if err != nil {
		t.Fatalf("second InitDB must succeed despite drift, got: %v", err)
	}
	defer second.Close()

	out := logged.String()
	if !strings.Contains(out, "Schema drift") {
		t.Errorf("expected a schema drift warning at startup, got log output:\n%s", out)
	}
	if !strings.Contains(out, "registry_domain") {
		t.Errorf("expected the warning to name the missing column, got log output:\n%s", out)
	}
}

// freshSchemaDB returns a database built the way a fresh install builds one:
// schema.sql applied directly, migrations recorded but not executed.
func freshSchemaDB(t *testing.T) *sql.DB {
	t.Helper()
	db, err := InitDB(":memory:", LibsqlConfig{})
	if err != nil {
		t.Fatalf("InitDB (fresh install path): %v", err)
	}
	return db
}

// migratedSchemaDB returns a database built the way an upgrade builds one: the
// pre-0009 base snapshot, then every migration executed in order.
func migratedSchemaDB(t *testing.T) *sql.DB {
	t.Helper()

	base, err := os.ReadFile("testdata/base_schema.sql")
	if err != nil {
		t.Fatalf("read testdata/base_schema.sql: %v", err)
	}

	connector, err := openLibsqlLocalConnector(":memory:")
	if err != nil {
		t.Fatalf("open libsql connector: %v", err)
	}
	db := sql.OpenDB(connector)
	// One connection so every statement lands in the same in-memory database.
	db.SetMaxOpenConns(1)

	if _, err := db.Exec("PRAGMA foreign_keys = ON"); err != nil {
		db.Close()
		t.Fatalf("enable foreign keys: %v", err)
	}

	for i, stmt := range splitSQLStatements(string(base)) {
		if _, err := db.Exec(stmt); err != nil {
			db.Close()
			t.Fatalf("apply base_schema.sql statement %d: %v\n%s", i+1, err, stmt)
		}
	}

	// freshDB=false so the DDL actually executes rather than merely being
	// recorded. base_schema.sql records versions 1-8 as already applied, so
	// everything from 0009 onward is pending.
	if err := runMigrations(db, false); err != nil {
		db.Close()
		t.Fatalf("runMigrations against the base snapshot: %v", err)
	}
	return db
}

// diffStringSets returns a human-readable diff, or "" when the sets match.
// "schema.sql" is the a side, "migrations" the b side.
func diffStringSets(a, b []string) string {
	var sb strings.Builder
	for _, s := range a {
		if !slices.Contains(b, s) {
			fmt.Fprintf(&sb, "  only in schema.sql:  %s\n", s)
		}
	}
	for _, s := range b {
		if !slices.Contains(a, s) {
			fmt.Fprintf(&sb, "  only in migrations:  %s\n", s)
		}
	}
	return strings.TrimRight(sb.String(), "\n")
}

func anyContains(haystack []string, needle string) bool {
	for _, s := range haystack {
		if strings.Contains(s, needle) {
			return true
		}
	}
	return false
}