Imagine we have a reducer to control a list of items:

function listOfItems(state: Array<Object> = [], action: Object = {}): Array<Object> {
  switch(action.type) {
    case 'SHOW_ALL_ITEMS':
      return action.data.items
    default:
    return state;
  }
}

Where Items looks like this:

type ItemType = {
  id: string,
  text: string,
  completed: boolean
};

Today we mapStateToProps for all incomplete items like this:

function mapStateToProps(state) {
  return {
    incompleteItems: state.listOfItems.filter((item) => {
      return !item.completed
    });
  }
}

There are a couple problems with this approach as the application grows.

1. The implementation of incompleteItems may change.
2. Computation logic occurs in mapStateToProps
3. Can't memoize the values of incompleteItems

After talking with Dan Abramov (founder of Redux) he has been preaching the colocation of functions called selectors.

What is a selector?

1. Selectors can compute derived data, allowing Redux to store the minimal possible state.
2. Selectors are composable. They can be used as input to other selectors.

Let's turn our filter into a selector.

function getIncompleteItems(state) {
  return state.listOfItems.filter((item) => {
    return !item.completed
  });
}

And we update our mapStateToProps function:

function mapStateToProps(state) {
  return {
    incompleteItems: getIncompleteItems(state)
  }
}

Now we can reuse this logic across many components mapping this exact state! We can unit test it as well! More importantly we can now memoize this state with reselect