Live Demo (iframe)

open in new tab

Functional Reactive Programming, or FRP, is the newish hotness. If you’ve been busy working, you may not have tried it out yet. Or you might have read a little about it. And then you may have been confused. I was. I did. I was again.

If you, like me, have heard about FRP or Bacon but aren’t sure where to start, this tutorial is for you. The intro tutorial is neat, but sometimes you need to implement something less trivial to really understand a new technology. Staltz’s article does an awesome job explaining the abstraction. I recommend reading it when you want deeper understanding of concepts. I aim to give you strong hands-on experience. Give this 30 minutes of your time and you’ll be rewarded with a new way of thinking about programming.

Infinite scrolling is a great candidate to implement with a functional reactive programming style because it involves a reasonably complex data and UI events that change over time. Keeping the display in sync using traditional imperitive programming techniques can be a bit hairy.

Let’s begin.

Getting Prepared

We’ll need an html file that includes Bacon.js and Lodash. We’ll also want a few CSS rules that are common to most infinite scrolling implementations, so I’ll skip discussing them here. This file will get you started:

So far nothing Bacony.

There is a little bit of JS setup here. All rows are the same height. We assume we know the data set size. In the real world we might need to query an API for the result set size. In that case we’d just wait to init the phonebookEl container until we get that information.

Scrolling and Resizing

An infinite scroller needs to know which rows are on screen so it can manage the total number of DOM nodes. Too many and the browser will get dizzy, confused, and slow.

too many dom nodes

To manage the number of rows on screen, we need to know where the viewport is. Specifically, we’re interested in when the user scrolls or resizes.

It’s bacon time. We’ll create streams for the scrolling and resizing:

var scrolling = Bacon.fromEvent(window, 'scroll');
var resizing  = Bacon.fromEvent(window, 'resize').debounce(50);

fromEvent turns a normal browser event into a stream.

Debouncing our resize is a good choice, and Bacon makes it easy.

The next step is to transform the browser events into the information we care about for our infinite scroll application. We care about the y position when we scroll, and the height of the screen after a resize:

var scrolling = Bacon.fromEvent(window, 'scroll');
var yPos = scrolling.map(function(){ return window.scrollY });
yPos.log('y');
var resizing = Bacon.fromEvent(window, 'resize').debounce(50);
var windowH = resizing.map(function(){ return window.innerHeight });
windowH.log('height');

The map function takes a stream and applies a transformation to each value. In our case we just grab the global y and height values value for each point in time.

We’re logging the results to make sure we’re getting what we expect so far. You can put .log on any Bacon observable. We should see fluid logging of the y position as we scroll and slightly debounced logging of the window height when we resize the browser.

Actually, all we care about are the y position and screen height. We’ll never reuse scroll and resize so let’s simplify:

var yPosition = Bacon.fromEvent(window, 'scroll').map(function() { return window.scrollY });
var screenHeight = Bacon.fromEvent(window, 'resize').debounce(50).map(function() { return window.innerHeight });

Which Rows are Visible?

Now we can walk slowly towards our infinite scrolling problem. From a stream of y scroll position it’s easy to calculate the first row that is visible on screen:

var firstVisibleRow = yPosition.map(function(y){ return Math.floor( y / rowHeight ) }).skipDuplicates();
firstVisibleRow = firstVisibleRow.toProperty(0); // Seed the initial value.

skipDuplicates causes the stream to not emit new values if the values is the same. This is good. Otherwise, every scroll event would fire a new firstVisibleRow event. Since our rows are 30 pixels tall, that means we would otherwise be sending lots of duplicate messages.

toProperty(0) tells the observable property it should start with the value 0. A browser doesn’t throw a scroll event if you start at the top of the screen, so we need to seed the start value. If you start partway down the screen, as when you reload and have already scrolled, this is init’d to 0 and a scroll event is thrown after, so that case works as well.

We are now calculating the first visible row in a way that updates in real time as we scroll, which is awesome. Let’s do the same thing for the number of rows that can fit on screen:

screenHeight = screenHeight.toProperty(window.innerHeight); // Seed the initial value.
var rowCount = screenHeight.map(function(screenHeight){
  return Math.ceil( screenHeight / rowHeight)
}).skipDuplicates();

Similar to how we seeded the first visible row, we need to seed the screen’s height with .toProperty(window.innerHeight) because the browser doesn’t throw a scroll event on page load. The row count can be the same for many values of screen height so we use skipDuplicates.

If you log rowCount and firstVisibleRow you should see a beautiful, live updating stream of all the data you need to compute your infinite scroll elements:

rowCount.log('Row Count');
firstVisibleRow.log('First Row');

What we want is an array of all the row indices that are visible on screen. Let’s combine these streams using combineWith. This method creates a new stream based on other streams. Looks like this:

function calcVisibleRows (firstRow, rowCount) {
  var visibleIndices = [];

  // Limit the number of visible rows
  lastRow = firstRow + rowCount + 1;
  if (lastRow > totalResults) {
    firstRow -= lastRow - totalResults;
  }

  for (var i = 0; i <= rowCount; i++) { visibleIndices.push(i + firstRow) }
  return visibleIndices;
}
var visibleRowIndices = Bacon.combineWith(calcVisibleRows, firstVisibleRow, rowCount);

Our calcVisibleRows function just takes two integer indices and calculates what should be visible on screen. It is bound by the result set size in this case, but you could remove this if you have a truly infinite data set.

Because it’s all streams, the recalculation happens whenever any of the source streams change, so if we resize or if we scroll, the visibleRowIndices will be recalculated.

Here’s a good place to pause. Think about what this stream combination is doing. It’s a pretty great wow moment. Doing this the traditional way by responding to scroll and resize events, then referencing state variables is a much more fragile and less clear way to define the behavior we want. I hope you’re starting to see some of FRP’s power. It’s pretty cool.

pausing for reflection

…feeling it?

Ok, let’s keep going.

Rendering

There is one more step we can take in streams to get our data closer to exactly what the UI needs to do the drawing. Let’s create two more streams. One for rows that have just become visible (because we need to draw them), and one for rows that have gone off screen (so we can clean them up).

var rowIndicesRemoved = visibleRowIndices.diff([], _.difference);
var rowIndicesAdded   = visibleRowIndices.diff([], function(prev, cur){ return _.difference(cur, prev) }); // longer form here so we can reverse `cur` and `prev`

Bacon’s diff method sends the previous and current value of a stream to a callback function. Then we use lodash’s difference method to compute the difference between two arrays. If we used to be showing rows [1,2,3], and now we’re showing rows [2,3,4], the _.difference is [4]. Calling with the arguments in the other order gets the removed elements. It’s a little confusing that diff and difference have similar names. They aren’t related. You can implement any function you want and pass it to Bacon’s diff method.

Now we have exactly the data we want in streams, let’s just do our drawing:

var rows = {}; // cache row dom elements for quick and lookup-free cleanup

rowIndicesAdded.onValue(  function(indices){ _.map(indices, renderRow) });

function renderRow(idx) {
  var row = document.createElement('li');
  row.innerText = idx;
  row.style.top = idx * rowHeight + 'px';
  phonebookEl.appendChild(row);
  rows[idx] = row;
}

What what what? Any time we get a new value in the stream of added row indices, want to do something, so we use onValue. What we want to do is send each index in the array of added rows to the render row method. There is usually only one index, but if you jump your scrolling or do it quickly, there can be a lot. Passing these indices on to renderRow individually is what _.map is doing for us.

At this point you have a working scroller, but old rows aren’t being cleaned up so we get none of the benefits of an infinite scroll widget. Here’s the last step, cleanup hidden rows:

rowIndicesRemoved.onValue(function(indices){ _.map(indices, removeRow) });

function removeRow(idx) {
  if (idx == null) { return }
  rows[idx].parentElement.removeChild(rows[idx]);
  rows[idx] = undefined;
}

To clean up, we just pull the element out of the DOM and remove the cache reference. I would have used delete but I think modifying the object’s hidden class is slower. Would have to test but I don’t have a perf issue yet. Also, reusing DOM elements instead of appending and removing them would be faster, but it would increase the complexity and I wanted to keep this focused on Bacon.js and FRP.

You should now have a working infinite scroller that prints the index of the row, with all your rendering confined to a nice single method that is provided the row index.

Final Code

Wrapping Up

I hope you’re feeling excited about this new way of thinking about programming, and proud of yourself for the beautiful work you’ve done today.

slow smile

Good work, you.

Update

@jusrin00 rewrote this in Facebook’s React using RxJS for the data layer. Really cool stuff on github.


If you liked this post, share it on Twitter


« Previous:   •   Next: »