Andrew Leedham

How Experimental Browser Features Helped Me Verify a Hypothesis


An issue was recently reported to me that an accordion component I work on had a scrollbar in some circumstances when it should not. For a bit of context, the accordion shows scrollbars when the content height is more than 50% of the viewport, and this content was not.

Thankfully, the error report had a link to a few affected pages, and I could recreate the issue, so it was a good starting place. I wrote a lot of the code for this component and reviewed the rest of it, so again a fortunate position which is rare.

When opening and closing the accordion it animates the max-height and toggles between an overflow-y of none and auto as to hide the scrollbar during the animation. This is a common technique as max-height is animatable and does not cause a reflow like height would. In order to add the overflow-y we hook into the transitionend event in JavaScript, so my initial thought was that this was not firing and thus leaving the max-height or not changing the overflow-y, but that was not the case. Then I noticed something, after toggling the accordion what seemed 1000 times, the text content was flowing onto an additional line at the end of the animation when the overflow-y was being added! To me, this instantly rang alarms in my head! The height I am calculating for the max-height is not taking the scrollbar into account, as it removes ~15px from the width of the content potentially making it flow onto another line. I quickly scrambled to prove this hypothesis and remembered reading about a new property that was intended to stop layout shift from scrollbars. After googling and trawling through creators past tweets, I found it scrollbar-gutter, it is an experimental feature in Chrome but it was easy to enable and using it with my accordion resolved the issue. So all I needed was a fix that was not an experimental feature. In the end, I had to rely on setting overflow-y to auto triggering layout, calculating the height, and then setting it back to none. Triggering an additional layout step is not really what I wanted to do, but after mulling on it for a while, I could not think of a better solution with the tools I had.

This perhaps gives you insight into how I go about solving problems. It was satisfying to realise the cause and verify it quickly with a new browser feature!