To validate your point, I think you need to modify my test and create a 1M elements array with 99% of zeros - then your sparse class will be way more efficient,

To validate your point, I think you need to modify my test and create a 1M elements array with 99% of zeros - then your sparse class will be way more efficient, also hiding the complexity of creating a compact sparse representation at the native level.