It’s time the Web did some Undergrounding

I live in a simple neighborhood in Palo Alto, California. My street is fortunate to have some nice old trees. We have small houses that the wealthy are slowly purchasing, tearing down, and then filling up the entire lot with house. From most vantage points it looks quaint, but then you get the wrong angle, or you go into someones back yard, and an ugly face is shown: a hodge podge of cables and wires are strung up from old pole to old pole.

I have asked neighbors why this is still the case, and I hear a variety of stories. Some believe that everyone has to come together to pay the city to underground this all, and no one has been able to get everyone on board. A hefty storm that knocks out power, or god forbid… Internet, will surely be the trick? Nope. It’s been long enough that surely it’s time to underground this infrastructure.

This is how I feel about the Web right now

We have had many successful undergrounding stories in the Web’s history. That is somewhat obvious, as the mere act of standardization is a key part of the process. It is long known that innovation in new and rapidly changing areas isn’t for the standards process (see Alex Russell’s amazing series on how that world works best). It’s too hard to “get it right, first time”, so basic risk management would suggest that you allow innovation to happen in a few branches and once you have enough data to suggest a decent (you don’t need to fall for the nirvana fallacy!) path, get enough consensus to drive the implementations that beget a true standard.

We see innovation happening all around us. Sometimes we look to other platforms to learn from (e.g. looking at iOS and Android to give us solid clues for how to bring mobile capabilities to the Web), and one other strong source is user space and the Web community. How are web developers solving their problems with various libraries and frameworks? In fact, as we push more capabilities into the platform, allowing a more Extensible Web, we should see even more of an ability for experiments in user land.

If we don’t come around and look for areas where we can take the experimentation and bake it into the platform, we will fail users when it comes to performance and the potential for richer user experiences.

Why will we fail?

The Web is an amazing streaming platform that allows for ephemeral applications that can be controlled, to a large degree, on the server. This is incredibly powerful, and we shouldn’t lose sight of this massive benefit.

However, a trade off like this doesn’t come with zero cost, and we should look at how to balance any cost. We have been doing some of this with stronger caching primitives, service workers, etc. That’s great, but until we can contort the law’s of physics, it will be true that shipping capabilities comes with a large cost. In an ideal world, a website that delivers a shopping experience for users will only send down capabilities tied strongly to said experience. It is less than ideal to also send down large infrastructure (affecting both load and runtime).

When we standardize and put something into the Web platform itself, we get the benefit of a shared core that user space can progressively rely on, and we are able to use a couple more tools as we do so. The platform in the browser has access to the native platform below it, in a way that you just don’t have working on top of the standard Web.

We see the benefits time and time again. With recent Observers, we can get away from the jank-ridden experiences that we have all felt from being on websites that take over scrolling and try to be too clever.

One of the largest undergrounding projects of late has been taking the great ideas in jQuery and bringing them to the platform via querySelector(s), fetch, classList and more. jQuery itself learned from the likes of Prototype, MooTools, Dojo, and others, and offered a slightly different approach. As we in turn learn from this corpus, it is important to realize that we didn’t just “take jQuery and bake it in”. We are often asked to do this, and it’s alluring. I instigated the Google Ajax CDN project back in the day, to answer “can we just cache things so we aren’t downloading the same darn APIs all day long?”

Caching is good, but it isn’t enough. There are tactical issues (how developers bundle, versioning) but there is also the fact that you need to do things differently when you bake something in.

Looking at cabling in my neighborhood again, when a city does this, they don’t just take the same cables and dig a quick hole to shove them in. Once those cables are underground you have new constraints. It’s hard to get to them when something goes wrong, so you want to protect them in a more rigorous manner. The same is true when you standardize something in the Web, as once it’s there you will be maintaining it for ~forever. You can learn, add the new constraints, as well as the new capabilities that the core platform gives you…. and the outcome should be *better* than before.

Ok, so what should we underground?

The details matter. What are the areas that are ripe for undergrounding in 2018? Let’s take a look at the landscape. Capabilities are unbundling across a diverse set of form factors, with new modalities and we are moving past the world of:

  • mobile == certain size screen, touch, camera, and location
  • desktop == keyboard, mouse, another size screen.

If you heard about the recent “Made By Google” event, you would have seen how these things are blurring (touch on laptops, voice and screens together, etc) and we haven’t even gotten to the world of mixed reality and friends.

What are the “jQuery of today” fashions around us? The world of React and Vue add to Web Components to show a world of strong componentization. We can also see other point solutions that many of us are using, e.g. fonts (google fonts and typekit), analytics (google analytics and friends), and adtech (has a huge impact on performance).

There is a lot of fertile ground, and I have been reflecting on:

UI Components

We have a series of components baked into the Web, but as we opened up the ability to create your own with custom elements, we slowed down any that just come with the platform. We started with a myriad of document related tags (including many that we rarely use…. <aside> etc). Then we had form controls, and shoved a universe into type=”mwhaha” on input. I think it would be worth spending more time on the UX of these form controls. It often feels like we get a lowest common denominator out there (which maybe works when we wire it to the native platform) and the user has to do something custom, from scratch, if they want it to differ in any real way. Now, I am looking at type=date, and the type=number where I picture users seeing tiny up and down arrows. We should allow developers the ability to style form controls easier, which would also help a lot with accessibility. This isn’t easy to change, for sure, and the edges are rough.

Left shows the old version, right shows Chrome 64 (thanks Henry!)

In other areas, we did keep iterating. For example, the first image picker on Android had an incredibly weak UI, with a simple file picker, compared to native apps that would show you recent photos that you quickly tap on to be done.

Then we had the need for application UI controls, going beyond documents and forms. How often have I heard:

“Just give me a UITableView that works!!!!!!


And you can say the same for many other common components. Surely, a developer can hunt around and maybe find a Web Component, or a component tied to a framework, but for many of these the platform could do more heavy lifting AND not require the download and runtime costs.

NOTE: Giving everyone the ability to build components in the platform is a huge win in my book. I am excited that I can build a leaf component that I know will be reusable for a long time to come, no matter which set of libraries of framework I am using in the future. In larger companies, you almost always see multiple frameworks in use (*takes a peak in google3…. oh my!*), so why not at least have the capability to share. This doesn’t mean that you leave your framework of choice behind, far from it. You will see some great examples at the upcoming Chrome DevSummit that marry custom elements with frameworks!

App Architecture

It’s not just about visual components. There are common app architecture patterns that we drag around with us. I would love the platform to offer me a rich state machine with a reactive data flow. For the devices that we run on, and for the multi-modal world, we need to think about how to offload work from the main UI thread. The browser continues to do more and more of this (e.g. compilation of JS, image decoding) but we also need to do more for developers to give us units of work that we can decide where to run. We are seeing good work here via APIs such as Workers (Comlink, WorkerDOM), and Worklets:

I start to get really excited about this world. If we put it all together, we can get a Web that finally has truly reliable jank free fluid UIs, with fantastic components, and needed architecture pieces, allowing the frameworks to up level and work on the next key areas of exploration.

The Web is all about collaboration. I am enjoying working with the community and especially the framework world. But what about yourself? What would you like to see?

I am enjoying Paul Lewis and Surma chatting about some of the possibilities, and how to keep a clean main thread, so you may too:

Leave a comment

Your email address will not be published. Required fields are marked *