Rendering list views on a web page efficiently: A technique to display big datasets

Recently, I needed to implement a web widget that displays a collection of items. In some cases, the collection happened to have 7 million items. Of course, adding 7 million divs would hang the browser. I started searching for existing solutions on the web. There were some implementations available, but they seemed to be heavy, bringing in unnecessary dependencies. Plus, I was just curious to try to implement it. That is why this post.

Very often, we have to deal with data represented as a collection of items of the same type, and it is straightforward to display them in some sort of list view. Of course, this depends a lot on the language in use and the framework. Usually, one takes the class that represents the list view, configures the data source, and whenever a particular item needs to be rendered, the framework will fetch the data for that item, usually identified by the index, and eventually return it to the part of the code that is rendering it. For example, if someone were to develop a C++ application with the help of the Qt framework, they would use a subclass of QAbstractItemModel as the data source and a subclass of QAbstractItemView for rendering. The critical part here is to make the rendering part as efficient as possible, especially when dealing with a large number of items. Similar libraries exist for other languages as well (or even something that is implemented by their standard libraries).

It is a little different for web views. Typically, people use features like infinite scrolling or pagination. The precise position of the scrollbar and its mapping to a particular element are not important, as seen on social network pages or online shops. Search results are often presented in an order that may not make much sense, and users usually scroll through the pages until they find a specific item. However, what if someone is working with a dataset of a certain size, where items are sorted in a specific order within that set, and the user knows approximately where the item of interest is located? In this case, we need a scrollbar that allows navigation to that item in the fastest way possible.

Let's take a look at how it can be implemented in code. We draw all items but hide those that cannot fit on the screen. This can be achieved by setting the overflow property to scroll or auto and limiting the height of the element to a certain value:

<div class="scrolldiv">
    ...
</div>

.scrolldiv {
    overflow: auto;
    height: XXXpx;
}

In this case, we will have a scrollbar that allows us to scroll the items up and down (If only vertical scrolling is desired then the property is overflow-y). However, this is inefficient because we would need to store all rendered items in memory.

The way libraries and frameworks solve this is by rendering only the number of items that fit the screen plus some extra items. Whenever we scroll up or down, we can calculate, based on the scroll position and the number of items that fit the screen, the new offset. Then, the entire collection can be rotated, ensuring that only the items that fit the screen need to be rendered. This logic requires two changes:

1. There should be JS/TS code providing data at an offset (that corresponds to the first element of the page).
2. There should be extra logic taking care of the scrollbar - there are now only a limited number of items rendered, but the scrollbar still should correspond to the whole data set.

The data store is straightforward to implement and can be represented with the following interface:

interface Result<Type> {
    from: number;
    items: Array<Type>;
    totalCount: number;
}

export interface DataSource<T> {
    fetch(index: number, count: number): Promise<Result<T>>;
}

The part with the scrollbar is trickier. For that, two separate elements are required: the first, content, to display the items, and the second, scrolldiv, to show the scrollbar of the right size. scrolldiv's position is absolute, and it is placed exactly on top of content. However, to show the scrollbar, the overflow property is set to scroll, and another empty element is added to it with the height property set to the total number of items times the height of a single element:

<div class="wrapper">
    <div class="content">...</div>
    <div class="scrolldiv">
        <div></div>
    </div>
</div>

.wrapper {
    position: relative;
}
.content {
    overflow: hidden;
    height: 100%;
    width: 100%;
}
.scrolldiv {
    position: absolute;
    overflow: scroll;
    top: 0;
    left: 0;
    bottom: 0;
    height: 100%;
    width: 100%;
}
.scrolldiv > div {
    height: XXXpx;
    width: 100%;
}

One can check my simple implementation of this list view idea using React.

In summary, this approach to handling large data sets in web applications offers an efficient way of rendering and interacting with data. By rendering only what is visible and managing the rest dynamically, we significantly improve the application's performance.

I invite you all to share your thoughts, feedback, or experiences related to handling large datasets in web applications. Have you tried similar or different approaches?