Relational (and immutable) models as persistence layer

Keeping data in application state is a very common thing. But maintaining its size and complexity may be a challenge. Unless we make it flat.

Problem we are aiming to solve

Most of the time, when keeping data in a store, it is somewhat thrown together, an array of objects. For example:

[
  { id: 1, title: "Title", },
  { id: 2, title: "Title 2", },
]

And this is fine for a small collection with no nested data. But image a more complex example, a blog:

[
    {
      id: 1,
      title: "Hello",
      tags: [
        { tag: "Tag 1", slug: "tag-1", },
        { tag: "Tag 2", slug: "tag-2", },
      ],
      categories: [
        { category: "Category 1", slug: "category-1", },
        { category: "Category 2", slug: "category-2", },
      ],
    },
]

While still very readable, it creates a problem. No matter what I would like to display, I have to fetch all the data. I don't really need tags or categories on listing. Or, if I just want to list all the categories, I would have to:

1. Loop through all the posts;
2. Extract the categories objects;
3. Combine that data into a new table (in a state or temporarily somewhere).

Seems like a lot to do.

That's where flat models come in handy.

Relation-based flat models

When having data that is different in type, relations are a great thing. You might know it from SQL tables.

Posts:

id	category	title
1	1	"Hello"

Categories:

id	name	slug
1	"Welcoming posts"	"welcoming-posts"

Very simple and straightforward. It can be easily mirrored as JavaScript objects:

{
  posts: [
    { id: 1, category: 1, title: "Hello" },
  ],
  categories: [
    { id: 1, name: "Welcoming posts", slug: "welcoming-posts" },
  ],
}

Using this model, we are only keeping the data we actually need. If we want to display solely the post list, we use posts collection. And if we need to query something, we just look it up in another collection, for example:

const data = {
  posts: [
    { id: 1, category: 1, title: "Hello" },
  ],
  categories: [
    { id: 1, name: "Welcoming posts", slug: "welcoming-posts" },
  ],
};

class Post {
  constructor(id) {
    this.id = id;
  }

  getPost() {
    return data.posts.find(post => post.id === this.id);
  }

  getDetailedPost() {
    const post = this.getPost();
    const category = data.categories.find(category => category.id === post.category);

    return {
      ...post,
      category,
    }
  }
}

While this is perfectly fine solution, necessity to write filters every time is not that nice.

Enter Immutable.js

This section assumes that you know at least about immutable maps and lists. If not, things may seem unclear.

The question of immutability when it comes to data is quite vast. I won't really talk much about it to avoid obscuring the text, but I will show you, why I use it very often when creating models similar to ours.

First, let's redefine our model a bit:

const immutableData = Map({
  posts: Map({
    items: Map(),
    collection: List(),
  }),
  categories: Map({
    items: Map(),
    collection: List(),
  }),
});

It's easy to see that we are repeating a pattern here. This is purely by design. Using the same notation everywhere, we are making sure that every object will be accessible on the same terms.

Let me take a minute to explain, why we have split entity model in two sections. items will be a map { id: { content } } and collection will be a simple list of [id, id]. That way we will be able to easily grab an item knowing its id without writing filters.

So let us build such model with existing data. I will refer to data object defined earlier in the text.

First, we need to extract the IDs from given objects.

For convenience, I've converted the data to maps first:

const immutablePosts = fromJS(data.posts);
const immutableCategories = fromJS(data.categories);

Now we need functions to extract the required values. First, we need an [id, id] list. Let's write it:

const extractIds = input => input.reduce((acc, item) => acc.push(item.get("id")), List());

Next, we need to have a { id: { content } } relation. That's a map:

const extractElements = input => input.reduce((acc, item) => acc.set(item.get("id"), item), Map());

And that's it. Now we can create the entire database:

const immutableData = Map({
  posts: Map({
    items: extractElements(immutablePosts),
    collection: extractIds(immutablePosts),
  }),
  categories: Map({
    items: extractElements(immutableCategories),
    collection: extractIds(immutableCategories),
  }),
});

Now, when we have an entire database, let's try to get an item:

const item = immutableData.getIn(["posts", "items", 1]);

I know this may seem long at first glance, but look at its pureness. It's so easily to tell what exactly is going on. The same goes for listing all the items:

const items = immutableData.getIn(["posts", "collection"])
  .map(
    item => immutableData.getIn(
      ["posts", "items", item]
    ),
  );

Pretty easy, eh? Of course in real life applications, you'd build helpers for this, to avoid writing this for the whole time. Such helper may be a class, that will work just like Post created earlier:

class ImmutablePost {
  constructor(id) {
    this.id = id;
  }

  getPost() {
    return immutableData.getIn(["posts", "items", this.id]);
  }

  getDetailedPost() {
    const post = this.getPost();
    const category = immutableData.getIn(["categories", "items", post.get("category")]);

    return post.set("category", category);
  }
}

Now getting a single post is to simply new ImmutablePost(1);.

Did you notice that on the surface, those classes are the same? That's because they are both a facade!

Conclusion

I think this shows pretty well how to use flat data models instead of nested ones. Adding ImmutableJS into the mix helps a lot, but introduces another level of complexity. So if this isn't something you're familiar with, consider leaving it out for now, to avoid headaches.

---

• Entire code on CodePen;
• ImmutableJS Home Page;