Write less, achieve meh?
Wednesday, June 4th, 2014 at 1:20 pmIn my keynote at HTML5DevConf in San Francisco I talked about a pattern of repetition those of us who’ve been around for a while will have encountered, too: every few years development becomes “too hard” and “too fragmented” and we need “simpler solutions”.
In the past, these were software packages, WYSIWYG editors and CMS that promised us to deliver “to all platforms without any code overhead”. Nowadays we don’t even wait for snake-oil salesmen to promise us the blue sky. Instead we do this ourselves. Almost every week we release new, magical scripts and workflows that solve all the problems we have for all the new browsers and with great fall-backs for older environments.
Most of these solutions stem from fixing a certain problem and – especially in the mobile space – far too many stem from trying to simulate an interaction pattern of native applications. They do a great job, they are amazing feats of coding skills and on first glance, they are superbly useful.
It gets tricky when problems come up and don’t get fixed. This – sadly enough – is becoming a pattern. If you look around GitHub you find a lot of solutions that promise utterly frictionless development with many an unanswered issue or un-merged pull request. Even worse, instead of filing bugs there is a pattern of creating yet another solution that fixes all the issues of the original one . People simply should replace the old one with the new one.
Who replaces problematic code?
All of this should not be an issue: as a developer, I am happy to discard and move on when a certain solution doesn’t deliver. I’ve changed my editor of choice a lot of times in my career.
The problem is that completely replacing solutions expects a lot of commitment from the implementer. All they want is something that works and preferably something that fixes the current problem. Many requests on Stackoverflow and other help sites don’t ask for the why, but just want a how. What can I use to fix this right now, so that my boss shuts up? A terrible question that developers of every generation seem to repeat and almost always results in unmaintainable code with lots of overhead.
That’s when “use this and it works” solutions become dangerous.
First of all, these tell those developers that there is no need to ever understand what you do. Your job seems to be to get your boss off your back or to make that one thing in the project plan – that you know doesn’t make sense – work.
Secondly, if we found out about issues of a certain solution and considered it dangerous to use (cue all those “XYZ considered dangerous” posts) we should remove and redirect them to the better solutions.
This, however, doesn’t happen often. Instead we keep them around and just add a README that tells people they can use our old code and we are not responsible for results. Most likely the people who have gotten the answer they wanted on the Stackoverflows of this world will never hear how the solution they chose and implemented is broken.
The weakest link?
Another problem is that many solutions rely on yet more abstractions. This sounds like a good plan – after all we shouldn’t re-invent things.
However, it doesn’t really help an implementer on a very tight deadline if our CSS fix needs the person to learn all about Bower, node.js, npm, SASS, Ruby or whatever else first. We can not just assume that everybody who creates things on the web is as involved in its bleeding edge as we are. True, a lot of these tools make us much more efficient and are considered “professional development”, but they are also very much still in flux.
We can not assume that all of these dependencies work and make sense in the future. Neither can we expect implementers to remove parts of this magical chain and replace them with their newer versions – especially as many of them are not backwards compatible. A chain is as strong as its weakest link, remember? That also applies to tool chains.
If we promise magical solutions, they’d better be magical and get magically maintained. Otherwise, why do we create these solutions? Is it really about making things easier or is it about impressing one another? Much like entrepreneurs shouldn’t be in love with being an entrepreneur but instead love their product we should love both our code and the people who use it. This takes much more effort than just releasing code, but it means we will create a more robust web.
The old adage of “write less, achieve more” needs a re-vamp to “write less, achieve better”. Otherwise we’ll end up with a world where a few people write small, clever solutions for individual problems and others pack them all together just to make sure that really everything gets fixed.
The overweight web
This seems to be already the case. When you see that the average web site according to HTTParchive is 1.7MB in size (46% cacheable) with 93 resource requests on 16 hosts then something, somewhere is going terribly wrong. It is as if none of the performance practices we talked about in the last few years have ever reached those who really build things.
A lot of this is baggage of legacy browsers. Many times you see posts and solutions like “This new feature of $newestmobileOS is now possible in JavaScript and CSS - even on IE8!”. This scares me. We shouldn’t block out any user of the web. We also should not take bleeding edge, computational heavy and form-factor dependent code and give it to outdated environments. The web is meant to work for all, not work the same for all and certainly not make it slow and heavy for older environments based on some misunderstanding of what “support” means.
Redundancy denied
If there is one thing that this discouraging statistic shows then it is that future redundancy of solutions is a myth. Anything we create that “fixes problems with current browsers” and “should be removed once browsers get better” is much more likely to clog up the pipes forever than to be deleted. Is it – for example – really still necessary to fix alpha transparency in PNGs for IE5.5 and 6? Maybe, but I am pretty sure that of all these web sites in these statistics only a very small percentage really still have users locked into these browsers.
The reason for denied redundancy is that we solved the immediate problem with a magical solution – we can not expect implementers to re-visit their solutions later to see if now they are not needed any longer. Many developers don’t even have the chance to do so – projects in agencies get handed over to the client when they are done and the next project with a different client starts.
Repeating XHTML mistakes
One of the main things that HTML5 was invented for was to create a more robust web by being more lenient with markup. If you remember, XHTML sent as XML (as it should, but never was as IE6 didn’t support that) had the problem that a single HTML syntax error or an un-encoded ampersand would result in an error message and nothing would get rendered.
This was deemed terrible as our end users get punished for something they can’t control or change. That’s why the HTML algorithm of newer browsers is much more lenient and does – for example – close tags for you.
Nowadays, the yellow screen of death showing an XML error message is hardly ever seen. Good, isn’t it? Well, yes, it would be – if we had learned from that mistake. Instead, we now make a lot of our work reliant on JavaScript, resource loaders and many libraries and frameworks.
This should not be an issue – the “JavaScript not available” use case is a very small one and mostly by users who either had JavaScript turned off by their sysadmins or people who prefer the web without it.
The “JavaScript caused an error” use case, on the other hand, is very much alive and will probably never go away. So many things can go wrong, from resources not being available, to network timeouts, mobile providers and proxies messing with your JavaScript up to simple syntax errors because of wrong HTTP headers. In essence, we are relying on a technology that is much less reliable than XML was and we feel very clever doing so. The more dependencies we have, the more likely it is that something can go wrong.
None of this is an issue, if we write our code in a paranoid fashion. But we don’t. Instead we also seem to fall for the siren song of abstractions telling us everything will be more stable, much better performing and cleaner if we rely on a certain framework, build-script or packaging solution.
Best of breed with basic flaws
One eye-opener for me was judging the Static Showdown Hackathon. I was very excited about the amazing entries and what people managed to achieve solely with HTML, CSS and JavaScript. What annoyed me though was the lack of any code that deals with possible failures. Now, I understand that this is hackathon code and people wanted to roll things out quickly, but I see a lot of similar basic mistakes in many live products:
- Dependency on a certain environment – many examples only worked in Chrome, some only in Firefox. I didn’t even dare to test them on a Windows machine. These dependencies were in many cases not based on functional necessity – instead the code just assumed a certain browser specific feature to be available and tried to access it. This is especially painful when the solution additionaly loads lots of libraries that promise cross-browser functionality. Why use those if you’re not planning to support more than one browser?
- Complete lack of error handling – many things can go wrong in our code. Simply not doing anything when for example loading some data failed and presenting the user with an infinite loading spinner is not a nice thing to do. Almost every technology we have has a success and an error return case. We seem to spend all the time in the success one, whilst it is much more likely that we’ll lose users and their faith in the error one. If an error case is not even reported or reported as the user’s fault we’re not writing intelligent code. Thinking paranoid is a good idea. Telling users that something went wrong, what went wrong and what they can do to re-try is not luxury – it means building a user interface. Any data loading that doesn’t refresh the view should have an error case and a timeout case – connections are the things most likely to fail.
- A lack of very basic accessibility – many solutions I encountered relied on touch alone, and doing so provided incredibly small touch targets. Others showed results far away from the original action without changing the original button or link. On a mobile device this was incredibly frustrating.
Massive web changes ahead
All of this worries me. Instead of following basic protective measures to make our code more flexible and deliver great results to all users (remember: not the same results to all users; this would limit the web) we became dependent on abstractions and we keep hiding more and more code in loaders and packaging formats. A lot of this code is redundant and fixes problems of the past.
The main reason for this is a lack of control on the web. And this is very much changing now. The flawed solutions we had for offline storage (AppCache) and widgets on the web (many, many libraries creating DOM elements) are getting new, exciting and above all control-driven replacements: ServiceWorker and WebComponents.
Both of these are the missing puzzle pieces to really go to town with creating applications on the web. With ServiceWorker we can not only create apps that work offline, but also deal with a lot of the issues we now solve with dependency loaders. WebComponents allow us to create reusable widgets that are either completely new or inherited from another or existing HTML elements. These widgets run in the rendering flow of the browser instead of trying to make our JavaScript and DOM rendering perform in it.
The danger of WebComponents is that it allows us to hide a lot of functionality in a simple element. Instead of just shifting our DOM widget solutions to the new model this is a great time to clean up what we do and find the best-of-breed solutions and create components from them.
I am confident that good things are happening there. Discussions sparked by the Edge Conference’s WebComponents and Accessibility panels already resulted in some interesting guidelines for accessible WebComponents
Welcome to the “Bring your own solution” platform
The web is and stays the “bring your own solution platform”. There are many solutions to the same problem, each with their own problems and benefits. We can work together to mix and match them and create a better, faster and more stable web. We can only do that, however, when we allow the bricks we build these solutions from to be detachable and reusable. Much like glueing Lego bricks together means using it wrong we should stop creating “perfect solutions” and create sensible bricks instead.
Welcome to the future – it is in the browsers, not in abstractions. We don’t need to fix the problems for browser makers, but should lead them to give us the platform we deserve.