list. This avoids doing two heap allocations for every character, and reduces the metrics parsing time by approx 25%.