Profiling one of our benchmarks I was pointed to a couple of low hanging fruits in the implementation of AbstractRowReader : its constructor could be simplified, the pre-computed cache of `entityInitializerByEntityReference` is sometimes not used at all (so it's worthwhile to create this lazily), and the iteration on entity initializers is extremely hot, can be optimised by iterating on an array.
Most interestingly, the combination of these improvements also lead to having much less and simpler code.
Profiling one of our benchmarks I was pointed to a couple of low hanging fruits in the implementation of AbstractRowReader : its constructor could be simplified, the pre-computed cache of `entityInitializerByEntityReference` is sometimes not used at all (so it's worthwhile to create this lazily), and the iteration on entity initializers is extremely hot, can be optimised by iterating on an array.
Most interestingly, the combination of these improvements also lead to having much less and simpler code.