@tsung-wei-huang is there a reason to use different types for the begin and end iterators. Also I think it would be helpful to add a ticket to track parallel algorithms implemented and perhaps priority for the unimplemented onse

The reason is to enable stateful iterator passing so one task can initialize data structure needed while its successor can launch the algorithm. See capture iterator by reference. In reality, we may have something like:

taskflow.for_each_index(std::ref(beg), end, 1, [](auto i) {});

Thanks for the explanation. I edited my implementation to support this and added new tests accordingly

Thank you for the contributions!