fix(wal): filter stale pages when database shrinks (#881)

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

db_test.go

@@ -1005,3 +1005,69 @@ func TestDB_EnforceL0RetentionByTime(t *testing.T) {
 	}
 	checkExists(true)
 }
+
+// TestDB_SyncAfterVacuum verifies that syncing works correctly after a database
+// shrinks via VACUUM. This tests the fix for issue #875 where page numbers from
+// earlier transactions in the WAL could exceed the new commit size after shrinking.
+func TestDB_SyncAfterVacuum(t *testing.T) {
+	db, sqldb := testingutil.MustOpenDBs(t)
+	defer testingutil.MustCloseDBs(t, db, sqldb)
+
+	// Create a table and insert enough data to create multiple pages
+	if _, err := sqldb.ExecContext(t.Context(), `CREATE TABLE t (id INTEGER PRIMARY KEY, data BLOB)`); err != nil {
+		t.Fatal(err)
+	}
+
+	// Insert enough rows to create many pages
+	for i := 0; i < 100; i++ {
+		if _, err := sqldb.ExecContext(t.Context(), `INSERT INTO t (data) VALUES (?)`, strings.Repeat("x", 4000)); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	// Initial sync
+	if err := db.Sync(t.Context()); err != nil {
+		t.Fatal(err)
+	}
+
+	// Get initial page count
+	var initialPageCount int
+	if err := sqldb.QueryRowContext(t.Context(), `PRAGMA page_count`).Scan(&initialPageCount); err != nil {
+		t.Fatal(err)
+	}
+	t.Logf("Initial page count: %d", initialPageCount)
+
+	// Delete most data and VACUUM to shrink the database
+	if _, err := sqldb.ExecContext(t.Context(), `DELETE FROM t WHERE id > 10`); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := sqldb.ExecContext(t.Context(), `VACUUM`); err != nil {
+		t.Fatal(err)
+	}
+
+	// Get new page count
+	var newPageCount int
+	if err := sqldb.QueryRowContext(t.Context(), `PRAGMA page_count`).Scan(&newPageCount); err != nil {
+		t.Fatal(err)
+	}
+	t.Logf("Page count after VACUUM: %d", newPageCount)
+
+	if newPageCount >= initialPageCount {
+		t.Skip("VACUUM did not shrink database, skipping test")
+	}
+
+	// This sync should succeed without "page number out-of-bounds" error
+	if err := db.Sync(t.Context()); err != nil {
+		t.Fatalf("sync after VACUUM failed: %v", err)
+	}
+
+	// Verify position advanced
+	pos, err := db.Pos()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if pos.TXID < 2 {
+		t.Fatalf("expected TXID >= 2, got %d", pos.TXID)
+	}
+	t.Logf("Final position: TXID=%d", pos.TXID)
+}

wal_reader.go

@@ -215,6 +215,14 @@ func (r *WALReader) PageMap(ctx context.Context) (m map[uint32]int64, maxOffset
 		}
 	}
 
+	// Remove pages that exceed the final commit size. This can occur when the
+	// database shrinks (e.g., via VACUUM) between transactions in the WAL.
+	for pgno := range m {
+		if pgno > commit {
+			delete(m, pgno)
+		}
+	}
+
 	// If full transactions available, return the original offset.
 	if len(m) == 0 {
 		return m, 0, 0, nil