A new feature for relative formula copying

[vic]

Currently, if drag(copy) a cell down by 1, the references in the formula inside the cell get their row index increased by 1.
It would be useful to have an easy way to change this behaviour and specify that row index be increased by an arbitrary number, n.
It would make, for example, skipping cells/ranges in source range much easier, removing the need for "index" or "offset" functions for such use. As well as referencing from a different direction than the one dragging to. Many possibilities.

One way to specify that could be by using a sequential linear pattern, similar to how the spreadsheet figures that if you have: 1,3, then after selecting both and dragging, the next number should be 5.
Thus, it would work like this, in case of dragging down. Below "f(C3)" means a formula depending on ref. C3.
1. In one cell, put "=f(c3)"
2. in the cell below it: "=f(c5)"
3. select both cells then drag one cell down
4. formula in the new cell should be "=f(c7)"

Is this doable?

EDIT:
it should work for any step/difference in index in references in formulas of the 2 selected cells, even negative, and even across a different dimension (row/col) than the one we are dragging along.
So, If the 2 selected cells, located in say A1,A2, contain:
- f(C5), f(C3), then when drag down, new cell should contain f(C1)
- f(C1), f(D1), then when drag down , new cell should contain f(E1)
- f(D1), f(C1), then when drag down , new cell should contain f(B1)

It can be generalized to also work if each of the selected cells contains 2 or more references, and there may be a different pattern (difference) for each corresponding pair across the 2 selected cells.

[vic]

I just noticed that Planmaker (and other popular spreadsheets) do have currently a behavior about formulas' references when selecting several cells, for example:
If in a column have the formulas:
=A1
=A3
Then at selecting both and dragging by the corner, you get next formulas to be:
=A3
=A5
=A5
=A7

But I don't see what is the use of such a behaviour.
I think what I suggest above would be way more useful in practice -- limited only by the power of pattern recognition!