What’s the approach they should have used instead and how would it be better? Especially in terms of keeping fast compile times and overall performance.