Junk Drawer Logo Junk Drawer

For all those little papers scattered across your desk

A Strategy to Change Core Data Structures in Programs

D. Ben Knoble on 13 Feb 2024 in Blog

This relates to the refactoring rule “Make the change easy, then make the easy change” from Kent Beck. He warns that the first step may be hard. I present one strategy to make the first step easier for data structure changes.

It sounds so simple when you say it out loud: change either the data structure or the data-processing programs, one at a time.


Imagine you have a data structure that maps decks of cards to numbers as a core piece of a program. This represents some way of constructing a mixed deck by drawing cards from each of the decks. Naturally you have some programs that manipulate this data. You might have

Now suppose you need to make a change to support enhancing specific cards in the decks.

You could make this change by adding a program to update the mapping by finding the deck, converting it to a new deck with the enhanced card, and swapping the key (keeping the old value). Let’s say that poses several challenges and has some negative trade-offs.

Another way to solve the problem is to split your data structure into 2 pieces:

How do you make this change incrementally (with small, focused commits) and keep your tests passing?

In case the benefits of this approach aren’t obvious:

The 2-part Strategy

  1. Change either the data structure or the contract of all the programs that manipulate it. Add adapters as necessary.
  2. Change the other.

Changing everything at once is hard. In a recent merge, I changed the programs first and then split the data. The full diff looks quite large, but the individual commits in that range are quite focused.

Let’s say you change the contract of the programs first. This means accepting arguments for the split data structures and manipulating those. It also requires adapting the old data structure into the new ones somehow, so you’ll end up with code to convert old to new at the call-site. And everything should still work, though you probably haven’t written the enhancement program yet.

Now you can change the data structure, which allows you to throw away the adapter code at call-sites because the new structures line up with the existing contracts.

Finally, you can write the easy program you wanted to for enhancement.


Categories: Blog

Load Comments
Previous Next
Back to posts