Easy Dynamic Regular Expressions with Tagged Template Literals and Proxies

If you use regular expressions a lot, you probably also create them from existing strings that you first need to escape in case they contain special characters that need to be matched literally, like $ or +. Usually, a helper function is defined (hopefully this will soon change as RegExp.escape() is coming!) that basically looks like this:

var escapeRegExp = s => s.replace(/[-\/\\^$*+?.()|[\]{}]/g, "\\$&");

and then regexps are created by escaping the static strings and concatenating them with the rest of the regex like this:

var regex = RegExp(escapeRegExp(start) + '([\\S\\s]+?)' + escapeRegExp(end), "gi")

or, with ES6 template literals, like this:

var regex = RegExp(`${escapeRegExp(start)}([\\S\\s]+?)${escapeRegExp(end)}`, "gi")

(In case you were wondering, this regex is taken directly from the Mavo source code)

Isn’t this horribly verbose? What if we could define a regex with just a template literal (`${start}([\\S\\s]+?)${end}` for the regex above) and it just worked? Well, it turns out we can! If you haven’t seen tagged template literals before, I suggest you click that MDN link and read up. Basically, you can prepend an ES6 template literal with a reference to a function and the function accepts the static parts of the string and the dynamic parts separately, allowing you to operate on them! Continue reading

Never forget type=”button” on generated buttons!

I just dealt with one of the weirdest bugs and thought you may find it amusing too.

In one of my slides for my upcoming talk “Even More CSS Secrets”, I had a Mavo app on a <form>, and the app included a collection to quickly create a UI to manage pairs of values for something I wanted to calculate in one of my live demos. A Mavo collection is a repeatable HTML element with affordances to add items, delete items, move items etc. Many of these affordances are implemented via <button> elements generated by Mavo.

Normally, hitting Enter inside a text field within a collection adds a new item, as one would expect. However, I noticed that when I hit Enter inside any item, not only no item was added, but an item was being deleted, with the usual “Item deleted [Undo]” UI and everything!

At first I thought it was a bug with the part of Mavo code that adds items on Enter and deletes empty items on backspace, so I commented that out. Nope, still happening. I was already very puzzled, since I couldn’t remember any other part of the codebase that deletes items in response to keyboard events.

So, I added breakpoints on the delete(item) method of Mavo.Collection to inspect the call stack and see how execution got there. Turned out, it got there via a normal …click event on the actual delete button! What fresh hell was this? I never clicked any delete button!

And then it dawned on me: <button> elements with no type attribute set are submit buttons by default! Quote from spec: The missing value default and invalid value default are the Submit Button state.. This makes no difference in most cases, UNLESS you’re inside a form. The delete button of the first item had been turned into the de facto default submit button just because it was the first button in that form and it had no type!

I also remembered that regardless of how you submit a form (e.g. by hitting Enter on a single-line text field) it also fires a click event on the default submit button, because people often listen to that instead of the form’s submit event. Ironically, I was cancelling the form’s submit event in my code, but it still generated that fake click event, making it even harder to track down as no form submission was actually happening.

The solution was of course to go through every part of the Mavo code that generates buttons and add type=”button” to them. I would recommend this to everyone who is writing libraries that will operate in unfamiliar HTML code. Most of the time a type-less <button> will work just fine, but when it doesn’t, things get really weird.

Responsive tables, revisited

Screenshot showing a table with 3 rows turning into 3 sets of key-value pairs

Many people have explored responsive tables. The usual idea is turning the table into key-value pairs so that cells become rows and there are only 2 columns total, which fit in any screen. However, this means table headers need to now be repeated for every row. The current ways to do that are:

  • Duplicating content in CSS or via a data-* attribute, using generated content to insert it before every row.
  • Using a definition list which naturally has duplicated <dt>s, displaying it as a table in larger screens.

A few techniques that go in an entirely different direction are:

  • Hiding non-essential columns in smaller screens
  • Showing a thumbnail of the table instead, and display the full table on click
  • Displaying a graph in smaller screens (e.g. a pie chart)

I think the key-value display is probably best because it works for any kind of table, and provides the same information. So I wondered, is there any way to create it without duplicating content either in the markup or in the CSS? After a bit of thinking, I came up with two ways, each with their own pros and cons.

Both techniques are very similar: They set table elements to display: block; so that they behave like normal elements and duplicate the <thead> contents in two different ways:

  1. Using text-shadow and creating one shadow for each row
  2. Using the element() function to duplicate the entire thead, styles and all.

Each method has its own pros and cons, but the following pros and cons apply to both:

  • Pros: Works with normal table markup
  • Cons:
    • All but the first set of headers are unselectable (since neither shadows nor element()-generated images are real text). However, keep in mind that the techniques based on generated content also have this problem — and for all rows. Also, that the markup screen readers see is the same as a normal table. However, it’s still a pretty serious flaw and makes this a hack. I’m looking forward to seeing more viable solutions.
    • Only works if none of the table cells wrap, since it depends on table cells being aligned with their headers.

Using text-shadow to copy text to other rows

  • Additional Pros: Works in every browser
  • Additional Cons: Max Number of rows needs to be hardcoded in the CSS, since each row needs another text shadow on <thead>. However, you can specify more shadows than needed, since overflow: hidden on the table prevents extra ones from showing up. Also, number of columns needs to be specified in the CSS (the --cols variable).

Demo

Using element() to copy the entire <thead> to other rows

  • Additional Cons: element() is currently only supported in Firefox 🙁

Demo

Quicker Storify export

If you’ve used Storify, you probably know by now it’s closing down soon. They have an FAQ up to help people with the transition which explains that to export your content you need to…

  1. Log in to Storify at www.storify.com.
  2. Mouse over the story that contains content you would like to export and select “View.”
  3. Click on the ellipses icon and select “Export.”
  4. Choose your preferred format for download.
  5. To save your content and linked assets in HTML, select – File > Save as > Web Page, Complete. To export your content to PDF, select Export to HTML > File > Print > Save as PDF.
  6. Repeat the process for each story whose content you would like to preserve.

So I started doing that. I wasn’t sure if JSON or HTML would be more useful to me, so I was exporting both. It was painful. Each export required 3 page loads, and they were slow. After 5 stories, I started wondering if there’s a quicker way. I’m a programmer after all, my job is to automate things. However, I also didn’t want to spend too long on that, since I only had 40 stories, so the effort should definitely not be longer than it would have taken to manually export the remaining 35 stories.

I noticed that the HTML and JSON URLs for each story could actually be recreated by using the slug of the Story URL:

https://storify.com/LeaVerou/css-variables-var-subtitle-cssconf-asia.html
https://api.storify.com/v1/stories/LeaVerou/css-variables-var-subtitle-cssconf-asia

The bold part is the only thing that changes. I tried that with a different slug and it worked just fine. Bingo! So I could write a quick console script to get all these URLs and open them in separate tabs and then all I have to do is go through each tab and hit Cmd + S to save. It’s not perfect, but it took minutes to write and saved A LOT of time.

Following is the script I wrote. Go to your profile page, click “Show more” and scroll until all your stories are visible, then paste it into the console. You will probably need to do it twice: once to disable popup blocking because the browser rightfully freaks out when you try to open this many tabs from script, and once to actually open all of them.

var slugs = [... new Set($$(".story-tile").map(e => e.dataset.path))]
slugs.forEach(s => { open(`https://api.storify.com/v1/stories/${s}`); open(`https://storify.com/${s}.html`) })

This gets a list of all unique (hence the [...new Set(array)]) slugs and opens both the JSON and HTML export URLs in new tabs. Then you can go through each tab and save.

You will notice that the browser becomes REALLY SLOW when you open this many tabs (in my case 41 stories × 2 tabs each = 82 tabs!) so you may want to do it in steps, by using array.slice(). Also, if you don’t want to save the HTML version, the whole process becomes much faster, the HTML pages took AGES to load and kept freezing the browser.

Hope this helps!

PS: If you’re content with your data being held hostage by a different company, you could also use this tool by Wakelet. I’ve done that too, but I also wanted to own my data as well.

Free Intro to Web Development slides (with demos)

This semester I’m teaching 6.813 User Interface Design and Implementation at MIT, as an instructor.

Many of the assignments of this course include Web development and the course included two 2-hour labs to introduce students to these technologies. Since I’m involved this year, I decided to make new labs from scratch and increase the number of labs from 2 to 3. Even so, trying to decide what to include and what not to from the entirety of web development in only 6 hours was really hard, and I still feel I failed to include important bits.

Since many people asked me for the slides on Twitter, I decided to share them. You will find my slides here and an outline of what is covered is here. These slides were also the supporting material the students had on their own laptops and often they had to do exercises in them.

The audience for these slides is beginners in Web development but technical otherwise — people who understand OOP, trees, data structures and have experience in at least one C-like programming language.

Some demos will not make sense as they were live coded, but I included notes (top right or bottom left corner) about what was explained in each part.

Use the arrow keys to navigate. It is also quite big, so do not open this on a phone or on a data plan.

If the “Open in new Tab” button opens a tab which then closes immediately, disable Adblock.

From some quick testing, they seem to work in Firefox and Safari, but in class we were using an updated version of Chrome (since we were talking about developer tools, we needed to all have the same UI), so that’s the browser I’d recommend since they were tested much more there.

I’m sharing them as-is in case someone else finds them useful. Please do not bug me if they don’t work in your setup, or if you do not find them useful or whatever. If they don’t tickle your fancy, move on. I cannot provide any support or fixes. If you want to help fix the issue, you can submit a pull request, but be warned: most of the code was written under extreme time pressure (I had to produce this 6 times as fast as I usually need to make talks), so is not my finest moment.

If you want to use them to teach other people that’s fine as long as it’s a non-profit event.

Different remote and local resource URLs, with Service Workers!

I often run into this issue where I want a different URL remotely and a different one locally so I can test my local changes to a library. Sure, relative URLs work a lot of the time, but are often not an option. Developing Mavo is yet another example of this: since Mavo is in a separate repo from mavo.io (its website) as well as test.mavo.io (the testsuite), I can’t just have relative URLs to it that also work remotely. I’ve been encountering this problem way too frequently pretty much since I started in web development. In this post, will describe all solutions and workarounds I’ve used over time for this, including the one I’m currently using for Mavo: Service Workers! Continue reading

Introducing Mavo: Create web apps entirely by writing HTML!

Today I finally released the project I’ve been working on for the last two years at MIT CSAIL: An HTML-based language for creating (many kinds of) web applications without programming or a server backend. It’s named Mavo after my late mother (Maria Verou), and is Open Source of course (yes, getting paid to work on open source is exactly as fun as it sounds).

It was the scariest release of my life, and have been postponing it for months. I kept feeling Mavo was not quite there yet, maybe I should add this one feature first, oh and this other one, oh and we can’t release without this one, surely! Eventually I realized that what I was doing had more to do with postponing the anxiety and less to do with Mavo reaching a stage where it can be released. After all, “if you’re not at least a bit embarrassed by what you release, you waited too long”, right?

The infamous Ship It Squrrel

So, there it is, I hope you find it useful. Read the post on Smashing Magazine or just head straight to mavo.io, read the docs, and play with the demos!

And do let me know what you make with it, no matter how small and trivial you may think it is, I would love to see it!

HTML APIs: What they are and how to design a good one

I’m a strong believer in lowering the barrier of what it takes to create rich, interactive experiences and improving the user experience of programming. I wrote an article over at Smashing Magazine aimed at JavaScript library developers that want their libraries to be usable via HTML (i.e. without writing any JavaScript). Sounds interesting? Read it here.

Duoload: Simplest website load comparison tool, ever

Today I needed a quick tool to compare the loading progression (not just loading time, but also incremental rendering) of two websites, one remote and one in my localhost. Just have them side by side and see how they load relative to each other. Maybe even record the result on video and study it afterwards. That’s all. No special features, no analysis, no stats.

So I did what I always do when I need help finding a tool, I asked Twitter:

Most suggested complicated tools, some non-free and most unlikely to work on local URLs. I thought damn, what I need is a very simple thing! I could code this in 5 minutes! So I did and here it is, in case someone else finds it useful! The (minuscule amount of) code is of course on Github.

Duoload

Of course it goes without saying that this is probably a bit inaccurate. Do not use it for mission-critical performance comparisons.

Credits for the name Duoload to Chris Lilley who came up with it in the 1 minute deadline I gave him 😛

Resolve Promises externally with this one weird trick

Those of us who use promises heavily, have often wished there was a Promise.prototype.resolve() method, that would force an existing Promise to resolve. However, for architectural reasons (throw safety), there is no such thing and probably never will be. Therefore, a Promise can only resolve or reject by calling the respective methods in its constructor:

var promise = new Promise((resolve, reject) => {
	if (something) {
		resolve();
	}
	else {
		reject();
	}
});

However, often it is not desirable to put your entire code inside a Promise constructor so you could resolve or reject it at any point. In my latest case today, I wanted a Promise that resolved when a tree was created, so that third-party components could defer code execution until the tree was ready. However, given that plugins could be running on any hook, that meant wrapping a ton of code with the Promise constructor, which was obviously a no-go. I had come across this problem before and usually gave up and created a Promise around all the necessary code. However, this time my aversion to what this would produce got me to think even harder. What could I do to call resolve() asynchronously from outside the Promise?

A custom event? Nah, too slow for my purposes, why involve the DOM when it’s not needed?

Another Promise? Nah, that just transfers the problem.

An setInterval to repeatedly check if the tree is created? OMG, I can’t believe you just thought that Lea, ewwww, gross!

Getters and setters? Hmmm, maybe that could work! If the setter is inside the Promise constructor, then I can resolve the Promise by just setting a property!

My first iteration looked like this:

this.treeBuilt = new Promise((resolve, reject) => {
	Object.defineProperty(this, "_treeBuilt", {
		set: value => {
			if (value) {
				resolve();
			}
		}
	});
});

// Many, many lines below…

this._treeBuilt = true;

However, it really bothered me that I had to define 2 properties when I only needed one. I could of course do some cleanup and delete them after the promise is resolved, but the fact that at some point in time these useless properties existed will still haunt me, and I’m sure the more OCD-prone of you know exactly what I mean. Can I do it with just one property? Turns out I can!

The main idea is realizing that the getter and the setter could be doing completely unrelated tasks. In this case, setting the property would resolve the promise and reading its value would return the promise:

var setter;
var promise = new Promise((resolve, reject) => {
	setter = value => {
		if (value) {
			resolve();
		}
	};
});

Object.defineProperty(this, "treeBuilt", {
	set: setter,
	get: () => promise
});

// Many, many lines below…

this.treeBuilt = true;

For better performance, once the promise is resolved you could even delete the dynamic property and replace it with a normal property that just points to the promise, but be careful because in that case, any future attempts to resolve the promise by setting the property will make you lose your reference to it!

I still think the code looks a bit ugly, so if you can think a more elegant solution, I’m all ears (well, eyes really)!

Update: Joseph Silber gave an interesting solution on twitter:

function defer() {
	var deferred = {
		promise: null,
		resolve: null,
		reject: null
	};

	deferred.promise = new Promise((resolve, reject) => {
		deferred.resolve = resolve;
		deferred.reject = reject;
	});

	return deferred;
}

this.treeBuilt = defer();

// Many, many lines below…

this.treeBuilt.resolve();

I love that this is reusable, and calling resolve() makes a lot more sense than setting something to true. However, I didn’t like that it involved a separate object (deferred) and that people using the treeBuilt property would not be able to call .then() directly on it, so I simplified it a bit to only use one Promise object:

function defer() {
	var res, rej;

	var promise = new Promise((resolve, reject) => {
		res = resolve;
		rej = reject;
	});

	promise.resolve = res;
	promise.reject = rej;

	return promise;
}

this.treeBuilt = defer();

// Many, many lines below…

this.treeBuilt.resolve();

Finally, something I like!