Brian Kardell
Betterifying the Web
Follow Me On...
Posted on 04/29/2020

Note from the author...

My posts frequently (like this one) have a 'theme' and tend to use a number of images for visual flourish. Personally, I like it that way, I find it more engaging and I prefer for people to read it that way. However, for users on a metered or slow connection, downloading unnecessary images is, well, unnecessary, potentially costly and kind of rude. Just to be polite to my users, I offer the ability for you to opt out of 'optional' images if the total size of viewing the page would exceed a budget I have currently defined as 200k...

Interactive Elements: A Strange Game

It can seem especially frustrating to a lot of developers that we don't have more elements "out of the box" as part of HTML. I mean: Where are the tabs? How do we not have tabs by now? It seems like almost every week this comes up in some fashion or other. In this piece I'll talk a little about why, how to see this for what it is, how we're trying to move forward - and also present an interesting experiment for your consideration and feedback that may help inform some of these efforts. If you want, you can skip right to the 'new idea/experiment', but I think this background is pretty helpful in understanding it, so please, come back and read it...

Let's have a look at HTML so far: ~10 of the elements are the 'boilerplate stuff': <html>, <body>, <head> and metadata stuff - <link>, <meta>, <base>, <style>, <script>, <title>. There's a few elements for linking and embedding other content and multimedia, and a few with special powers like <template> and <slot>. And then the rest: ~30 are "weak semantics about text" that aren't interactive and don't have signficant (any) special meaning necessarily and are primarily just things with default CSS stylesheet rules. About 15 are "weak semantics about text, but meaningfully weak" - stuff like lists. ~15 are sectioning or landmarks. 11 more are about tables - a 2 dimensional relationship of text. 14 are about forms. And then two are... well... something else. We'll come back to that. And ~30 are deprecated.

The route to all of these elements was a little different - and they have different costs associated with them, and offer different value. One thing that's interesting to note is that the majority of them are what Dave Rupert refers to as "Spicy Divs". That is, there's not much to them - they aren't complex or interactive, and they don't bring a ton of real world value. That's not to say they are without value, but ultimately we spend a lot of time and bandwidth in standards debating what often amount to silly things because of it. <main> is a kind of a good example. Nothing wrong with it as an element, it's great, in fact - it just has a long and complex backstory that ultimately cost a lot relative to its real world value and proven use. Tons of time defining and debating <address>, which isn't really that meaningful to browsers, and then lots of evangelism telling people they're using it wrong - only in the end to eventually have to admit that it ultimately means the thing people used it as.

The second thing to notice is that there there are very few interactive elements. Further, most of the ones we usually talk about are about forms. And that lead us to this: Interactive, form-related elements on the web are... tricky.

The form-related game

On the one hand, there's a lot to like about native, interactive form controls. In theory, they can bring a simple declarative form with all kinds of goodness in terms of platform integration, portability and centralized work on accessibility and UX.


...primary goal has not yet been achieved
> What is the primary goal?
To win the game.

More simply: Despite a lot of work, and huge investments, people still gravitate toward custom implementations. That's not good for anybody.

Why are native form controls hard?

Well... So many reasons, but to start with, the one that they could really do without is that they got off to a tricky start by design.

<input type="WOPR">

Most of the form-related controls are created based on <input type-"...">. You've probably heard someone suggest that this is great because in cases where it isn't supported, simply providing the text field is fine. However, this is incomplete and we've been paying the price for it for a long time.

> People sometimes make... mistakes.

Did you ever notice how the JavaScript API surface of input works differently based on the type? If the type is number and you type gibberish that looks remotely 'potentially numbery' like '12312e314124', then .value will be '' (empty string). Or, the checkbox type has a checked property, but that means that input type=number has that property too... What does it do? Anything!? Or, have you ever noticed how .selectionStart and .selectionEnd actually would initially throw if you tried to use them on a numeric input type?! (fixed by Simon Pieters). If not, go have a watch of Monica Dinculescu's <input> I ♡ you, but you're bringing me down -- a 45 minute talk about just this.

Ultimately, many of these problems stem from us accepting the illusion presented that a date picker is a kind of text input.

Faulken pointing to NORAD screens
Stephen Falken: Uh, uh, General, what you see on these screens up here is a fantasy; a computer-enhanced hallucination. Those blips are not real missiles. They're phantoms.

Realistically, a date picker isn't a text input. Perhaps at an HTTP level, or even a database level, it's just a value - but controls are about how to populate values, and that's quite a different thing. It's not an is-a relationship: A date picker is a completely different complex control with potentially lots of complex interactions: Different things it needs to communicate, lots of sub-iteractions to manage, and so on.

All of this is why you might hear people suggest "LSP" or "Favor composition over inheritance" as a better alternative.

Acceptable Losses?

This is exacerbated by the fact that at the end of the day, developers are left with some pretty unappealing choices: Something is going to be sacrificed.

Color pickers provide a great example of how this plays out in practice: Will developers building the sorts of tools that use a color picker allow a simple input field as the fallback until every browser ships something acceptable? Seems like not much of a choice really.

Sadly the story for 'polyfilling' an element also still isn't great: It's really just 'replacing it with something else entirely'. So, at the end of the day, you've got to go and find something that is a good quality stand in anyway. Finally, since literally everything your code relies on understanding the DOM, and that stand in will have entirely different DOM, you now have two entirely different interactive trees to worry about. Yikes.

Further, this situation isn't necessarily short lived, historically. The reality is that lacking universal support can be the case for a long time: Native support for <input type="color"> was added to the last major browser... Checks notes... last year.

Even after all of that... Guess what? It still isn't acceptable for a lot of people because it isn't styleable -- and in fact, it has entirely different UI and features everywhere. While this can occasionally be very useful on some kinds of devices, it makes things like providing helpful documentation hard. Worse, even the native ones weren't super keyboard accessible either.

So... That's a lot of barriers and disincentives.

Conversely, a custom solution can solve all of these problems now... So it shouldn't actually be surprising that developers often choose a custom solution: There are no 'acceptable losses' for developers here - all of the choices are somehow bad.

Learn, dammit.

Now, let's talk about how we get out of this mess.

You'll be happy to know that there are efforts to both learn and improve existing native interactive elements. If you haven't seen Greg Whitworth and Nicole Sullivan's talk HTML Isn't Done from Chrome Dev Summit last year, I would highly recommend it (below).

There's also been a heck of a lot of work to make it possible for authors to define their own interactive elements that work just like the native ones. The idea is to empower developers to explore the space and try to help find the really sweet spots that strike all of the right balances.

It's worth pointing out why this is useful - it's not just an academic exercise that is trying to push responsibility to developers. It's important because the truth is: We don't actually know how. We need to be able to throw lots of things at the wall and see what sticks. The trouble is, currently (historically) failure is really slow and really expensive -- and even our most educated 'guesses' at improvement are still just that: Guesses. We haven't proved we have this figured out yet.

We need something like the WOPR for playing out possible answers without doing harm: failing, and learning - that's what the custom elements stuff is all about.
Shall we play a (different) game?

Ok, but hold on.... Let's talk about non-form related interactive controls now. Realistically, we've come up with really only 1 strictly generic non-form-related UI control in 30 years: <details>. That "other" one in my intial list. Wow... And we've had that done for like 10 years, right? Let's see, it's been supported universally since... checks notes... Jan 14, 2020. Waaaaat?

But... Again, it has most of the same kinds of general problems. In the most recent HTTP Almanac data, it was the 102nd most used HTML element... Can you even name 101 other HTML elements? It has styling problems. It's misunderstood. It was hard to polyfill well without just creating other problems.

So... This is terrible.

Our continued problems in this space are part of the reason there is hesitancy to take up entirely new things... Like, tabs.

Back in 2015, I managed to get a number of people in Web standards (especially a11y) interested in working on a possible new element proposal for HTML which would have given not only the tabset but some other things as well. Determined not to be doomed by the same sorts of stylability problems, we set out to define things that would become shadow parts and themes and custom properties and so on.

But, in many ways, it wasn't great. Recently, this has got me to thinking: What if this isn't even the right game we're playing? What if for some elements at least - non-form associated ones - the right lesson to take away is that this is an unnecessary exercise in futility?

Maybe we need a different kind of experiment: One that just says "Strange game. The only winning move is not to play".

How about a nice game of chess?

So, here's the basic idea: What if we just created a resilient pattern focused on mainly function and meaning -- and hardly at all on the UI aspect. What if instead of inventing complex new ways to strike the balance of preventing authors from styling too much, we mostly just acknowledged the fact that they want to, and... let them.

What if, like with <video> you could kind of just take control of the UI, but... maybe without throwing it away entirely.

I spoke with Greg Whitworth about this, who has been investigating how to improve the existing controls on the Web (see also Can we please style select?.

In fact, he completely agreed with the overall premise and went on to described how...

The core mission of Open UI, a new W3C community group, is to document the anatomies, behaviors and states of components and controls from across frameworks. By doing this, we can ensure that the key pieces that folks don't want to re-create, they don't have to. We've got a few different ideas that are beginning to take shape but changes like this are like peeling an onion. If it's a new control/component then it would be a bit more straight forward but the most painful controls we've found have been on the web since the 90s. So we'll be gathering telemetry and ensuring that any modifications we ultimately propose are web compatible.

In our conversation we discussed how things like tabs seem like an interesting target where this kind of thing might just work really well. The particular shape of the tabs problem lends itself, I think, to easily trying something completely different and entirely without past baggage.

So, in that light here's a demo/link to just such an experiment which gives us... tabs! - and which we think has some nice qualities:

  • It is a declarative decorator over otherwise good but non-interactive semantics. That means there's no miss of differing interaction issues here caused by complex inheritance. It's composition.
  • It doesn't change your light DOM, and it barely uses Shadow DOM. There's no "two trees" problem, and there's not secrets or new challenges. Go ahead, use CSS and JavaScript.
  • It adds the functionality: roving focus, keyboard handling, accessibility stuff and a simple container for you. That's it, really.
  • If you really want to opt in to some simple 'default styles' and like the 'minor tweaksonly' approach, it has an attribute that enables 'native' look and feel, with a bit less flexibility, but which might be enough for some people. It just starts with the assumption that that's not the case.

So... that's it. It could for sure use more work, but WDYT? Would you try it out? Let us know your thoughts? Such experiments will be useful in informing work and directions for efforts like Open-UI and possible future standards.

Will it help? To be honest, I don't know, but maybe the right way to learn involves trying something really different, and I'm open to anything...