I am puzzled on how to model this situation. Suppose you have an algorithm operating in a loop. At every loop, a procedure P must take place, whose role is to modify an input data I into an output data O, such that O = P(I).

In reality, there are different flavors of P, say P1, P2, P3 and so on. The choice of which P to run is user dependent, but all P have the same finality, produce O from I.

This called well for a base class PBase with a method PBase::apply, with specific reimplementations of P1::apply(I), P2::apply(I), and P3::apply(I). The actual P class gets instantiated in a factory method, and the loop stays simple.

Now, I have a case of P4 which follows the same principle, but this time needs additional data from the loop (such as the current iteration, and the average value of O during the previous iterations).

How would you redesign for this case?