In Part 1, we looked at the awkward situation created when we introduce callbacks to handle even a single asynchronous operation into an otherwise simple set of function calls.

As a quick review, have a look back at the code we started with, the messy end result when using callbacks, and the things we’d like to fix in order to get back to sanity:

1. We can no longer use a simple call-and-return programming model
2. We can no longer handle errors using try/catch/finally
3. We must add callback and errback parameters to every function signature that might eventually lead to an asynchronous operation

Promises

A Promise (aka Future, Delayed value, Deferred value) represents a value that is not yet available because the computation that will produce the value has not yet completed. A Promise is a placeholder into which the successful result or reason for failure will eventually materialize.

Promises also provide a simple API (see note below) for being notified when the result has materialized, or when a failure has occured.

Promises are not a new concept, and have been implemented in many languages. While several implementations of the Promise concept in Javascript have been around for a while, they have started to gain more popularity recently as we start to build bigger, more complex systems that require coordinating more asynchronous tasks.

(NOTE: Although there are several proposed Promise API standards, Promises/A has been implemented in several major frameworks, and appears to be becoming the defacto standard. In any case, the basic concepts are the same: 1) Promises act as a placeholder for a result or error, 2) they provide a way to be notified when the actual result has materialized, or when a failure has occurred.)

The Canonical XHR Example

In the case of an XHR Get, the value we care about is the content of the url we’re fetching. We know that XHR is an asynchonous operation, and that the value won’t be available immediately. That fits the definition of a Promise perfectly.

Imagine that we have an XHR library that immediately returns a Promise, as a placeholder for the content, instead of requiring us to pass in a callback. We could rewrite our asynchronous thisMightFail function from Part 1 to look like this:

(Note that several popular Javascript libraries, including Dojo (see also this great article on Dojo’s Deferred by @bryanforbes) and jQuery, implement XHR operations using promises)

Now, we can return the Promise placeholder as if it were the real result, and our asynchronous thisMightFail function looks very much like a plain old synchronous, call-and-return operation.

Taking Back the Stack

In a non-callback world, results and errors flow back up the call stack. This is expected and familiar. In a callback-based world, as we’ve seen, results and errors no longer follow that familiar model, and instead, callbacks must flow down, deeper into the stack.

By using Promises, we can restore the familiar call-and-return programming model, and remove the callbacks.

Restoring Call-and-return

To see how this works, let’s start with a simplified version of the synchronous getTheResult function from Part 1, without try/catch so that exceptions will always propagate up the call stack.

Now let’s introduce the asynchronous thisMightFail from above that uses our Promise-based XHR lib.

Using Promises, getTheResult() is identical in the synchronous and asynchronous cases! And in both, the successful result or the failure will propagate up the stack to the caller.

Removing Callbacks

Notice also that there are no callbacks or errbacks (or alwaysbacks!) being passed down the callstack, and they haven’t polluted any of our function signatures. By using Promises, our functions now look and act like the familiar, synchronous, call-and-return model.

Done?

We’ve used Promises to refactor our simplified getTheResult function, and fix two of the problems we identified in Part 1. We’ve:

1. restored call-and-return
2. removed callback/errback/alwaysback parameter propagation

But, what does this mean for callers of getTheResult? Remember that we’re returning a Promise, and eventually, either the successful result (the result of the XHR) or an error will materialize into the Promise placeholder, at which point the caller will want to take some action.

What about the Caller?

As mentioned above, Promises provide an API for being notified when either the result or failure becomes available. For example, in the proposed Promises/A spec, a Promise has a .then() method, and many promise libraries provide a when() function that achieves the same goal.

First, let’s look at what the calling code might look like when using the callback-based approach:

Now, let’s look at how the caller can use the Promise that getTheResult returns using the Promises/A .then() API.

Or, more compactly:

(Image from The Meta Picture)

Wasn’t the whole point of this Promises stuff to avoid using callbacks? And here we are using them?!?

Stay with Me

In Javascript, Promises are implemented using callbacks because there is no language-level construct for dealing with asynchrony. Callbacks are a necessary implementation detail of Promises. If Javascript provided, or possibly when it does provide in the future, other language constructs, promises could be implemented differently.

However, there are several important advantages in using Promises over the deep callback passing model from Part 1.

First, our function signatures are sane. We have removed the need to add callback and errback parameters to every function signature from the caller down to the XHR lib, and only the caller who is ultimately interested in the result needs to mess with callbacks.

Second, the Promise API standardizes callback passing. Libraries all tend to place callbacks and errbacks at different positions in function signatures. Some don’t even accept an errback. Most don’t accept an alwaysback (i.e. “finally”). We can rely on the Promise API instead of many potentially different library APIs.

Third, a Promise makes a set of guarantees about how and when callbacks and errbacks will be called, and how return values and exceptions thrown by callbacks will be handled. In a non-Promise world, the multitude of callback-supporting libraries and their many function signatures also means a multitude of different behaviors:

1. Are your callbacks allowed to return a value?
2. If so, what happens to that value?
3. Do all libraries allow your callback to throw an exception? If so, what happens to it? Is it silently eaten?
4. If your callback does throw an exception, will your errback be called, or not?

… and so on …

So, while one way to think of Promises is as a standard API to callback registration, they also provide standard, predictable behavior for how and when a callback will be called, exception handling, etc.

What about try/catch/finally?

Now that we’ve restored call-and-return and removed callbacks from our function signatures, we need a way to handle failures. Ideally, we’d like to use try/catch/finally, or at least something that looks and acts just like it and works in the face of asynchrony.

In Part 3, we’ll put the final piece of the puzzle into place, and see how to model try/catch/finally using Promises.

Exceptions and try/catch

Exceptions and try/catch are an intuitive way to execute operations that may fail. They allow us to recover from the failure, or to let the failure propagate up the call stack to a caller by either not catching the exception, or explicitly re-throwing it.

Here’s a simple example:

In this case, getTheResult handles the case where thisMightFail does indeed fail and throws an Error by catching the Error and calling recoverFromFailure (which could return some default result, for example). This works because thisMightFail is synchronous.

Going Async

What if thisMightFail is asynchronous? For example, it may perform an asynchronous XHR to fetch the result data:

Now it’s impossible to use try/catch, and we have to supply a callback and errback to handle the success and failure cases. That’s pretty common in Javascript applications, so no big deal, right? But wait, now getTheResult also has to change:

At the very least, callback (and possibly errback, read on) must now be added to every function signature all the way back up to the caller who is ultimately interested in the result.

More Async

If recoverFromFailure is also asynchronous, we have to add yet another level of callback nesting:

This also raises the question of what to do if recoverFromFailure itself fails. When using synchronous try/catch, recoverFromFailure could simply throw an Error and it would propagate up to the code that called getTheResult. To handle an asynchronous failure, we have to introduce another errback, resulting in both callback and errback infiltrating every function signature from recoverFromFailure all the way up to a caller who must ultimately supply them.

It may also mean that we have to check to see if callback and errback were actually provided, and if they might throw exceptions:

The code has gone from a simple try/catch to deeply nested callbacks, with callback and errback in every function signature, plus additional logic to check whether it’s safe to call them, and, ironically, two try/catch blocks to ensure that recoverFromFailure can indeed recover from a failure.

And what about finally?

Imagine if we were also to introduce finally into the mix—things would need to become even more complex. There are essentially two options, neither of which is as simple and elegant as the language-provided finally clause. We could: 1) add an alwaysback callback to all function signatures, with the accompanying checks to ensure it is safely callable, or 2) always write our callback/errback to handle errors internally, and be sure to invoke alwaysback in all cases.

Summary

Using callbacks for asynchronous programming changes the basic programming model, creating the following situation:

1. We can no longer use a simple call-and-return programming model
2. We can no longer handle errors using try/catch/finally
3. We must add callback and errback parameters to every function signature that might eventually lead to an asynchronous operation

We can do better. There is another model for asynchronous programming in Javascript that more closely resembles standard call-and-return, follows a model more like try/catch/finally, and doesn’t force us to add two callback parameters to a large number of functions.

Next, we’ll look at Promises, and how they help to bring asynchronous programming back to a model that is simpler and more familiar.

I hope everyone who attended last night’s jBurgh meetup had as much fun as I did, and enjoyed my talk on OOCSS. I thought it was a great low-key event, and I have to say that the discussion after the presentation and hanging around talking to everyone were the best parts.

My slides are up on slideshare, and if you were at the meetup, I’d really appreciate your taking a minute to rate my talk, and (especially!) leave a comment about how it could be improved.

If you’re interested in getting more background on OOCSS, I encourage you to check out Nicole’s slides and video. I’ve also written a few articles that dig deeper into some of the concepts in last night’s talk, such as OOCSS inheritance, why .css() is bad, and OOCSS Antipatterns:

• Building a digital clock with OOCSS and MVC
• The OO in OOCSS, and direct style manipulation
• OOCSS Antipatterns: Analog clock theme using only CSS

You can also check out the digital clock demo, including these other versions:

• Analog – remember this is an antipattern (see above), no matter how cool it seems!
• JQCon explode – we did this mod live onstage at JQCon
• Daniel Lamb’s binary clock mod – he points directly to my JS view controller (view source to check it out), and just modded the HTML and CSS. Now that’s separation of concerns FTW.

For more deep diving along with some excellent code examples and comment discussion, I also recommend reading John Hann’s post on OOCSS for Engineers.

At the end of the presentation, I mentioned that I’m working on OOCSS design patterns for web applications. My plan is to post more information soon about the ones mentioned in the slides (with example code), as well as others, so stay tuned.

Thanks to Chris Bannon and Wijmo for organizing and sponsoring, to HackPittsburgh for lending us their space, and to everyone who attended! I had a blast, and I’m already looking forward to the next meetup.

I’ll be speaking about OOCSS at the next jQuery Pittsburgh meetup, this Thursday (2/3) at 6:30pm at HackPittsburgh. I’m planning to present some similar material as John Hann and I presented at jQCon in September, but I’ll also be including an expanded section on how to get started with OOCSS, as well as a section on OOCSS antipatterns.

Thanks to Chris Bannon from Wijmo for organizing the meetup, and for inviting me to speak. Hope to see you there!

I read this article recently, and decided to solve the problem the author proposes in point #3 using OOCSS techniques.

Here you go: OOCSS Slideshows.

You might be saying, “Good grief, all that CSS is overkill!”, and be tempted to think that just using image.style.display = 'none' is good enough. Try implementing those 3 effects in plain Javascript … you probably wouldn’t get it right—I know I probably wouldn’t.

Or maybe you’re thinking that you’d just use jQuery, Dojo, or your favorite effects library to do it. Putting aside the fact that the original interview question stipulated no libraries, the advantage of an OOCSS approach, as I’ve written about at length before, is separation of concerns, and all its related benefits.

Using an MVC and OOCSS approach puts control of the slideshow transitions into the hands of the designer, and changes to the transitions require zero Javascript changes.

One of the things on my Dojo completion hitlist was handling multiline statements, especially dojo.query chains, since I tend to break them into multiple lines when chaining more than 1 or 2 NodeList function calls (and I’m betting most folks do that, too). While thinking about how to handle this, I came to the conclusion that there were basically three ways I could approach it:

1. Implement a real Javascript parser that is liberal enough to handle incomplete statements (e.g. it has to handle the statement you’re trying to complete!), or
2. Integrate an existing parser that is liberal enough to handle incomplete statements (I have no idea if a liberal open source JS parser exists), or
3. Implement something simpler and easier that covers most or all of what I think of as common cases.

I decided to go with #3. In fact, this is my general rule of thumb for the completion bundle in the short and medium term. I want to make something that is both useful, and something I can actually release since I’m only working on it a few hours each week. An unreleased, vaporware completion bundle is far less cool than a released one that works 80% of the time.

Here are a few shots of the multiline statement parser in action, completing a dojo.querychain. Notice also, in the last shot, it guesses correctly that I am starting a new statement, even though I didn’t end the previous one with a semicolon. However, it’s not perfect, and will almost certainly break under less common conditions. I’ve tried, though, to abstract the parser from the completion code in a way that allows the parser to be improved when I find cases where it breaks.

(download)

Click here to download:

dojo-completion-statement-parsing-xtIucBynEEoCEvaeyHkg.zip (372 KB)

One of the biggest complaints I hear about TextMate is that it doesn't have some sort of native intellisense, and doesn't do code completion.  Those are typically features provided by "bigger" IDEs.  TextMate started life as a text editor, and has had different goals than other IDEs, so even though TextMate has grown closer to full-blown IDE over the past few years (thanks to lots of great community bundle contributions), it's not really hard to understand why it hasn't provide those features natively.

It turns out, though, that TextMate provides some builtin help for rolling your own completions, and a lot of folks have done that for various languages and platforms.  I decided to take a crack at it for Dojo as a part of a new, simplified Dojo bundle I've been working on.  Here are a couple teaser screenshots of what I have so far.

I'm not ready to post the bundle yet, but I'm actively working on it, so I hope to have an initial version ready within the next couple of weeks.

(download)

Click here to download:

Dojo_Autocomplete_for_TextMate.zip (349 KB)

You asked for it, and you got it. Several people asked me after the digital clock demo at jQCon if I thought it was possible to create an analog theme for the clock using only CSS. I had wondered this as well, but came to the conclusion that it was impossible, given the current HTML structure. So, that’s the answer I gave at the conference.

The very next day, I thought of a way to do it, so here’s an analog theme (source on github) for the digital clock done entirely in CSS.

First, note a couple things:

1. The first hour, minute, and second digits are hidden.
2. The second hour, minute, and second digits have been styled to look like analog clock hands—i.e. long thin rectangles, and have been anchored at the center of the analog clock face.

The key is the use (in fact, abuse, read on!) of the immediate sibling selector to rotate the clock hands to the correct position, using CSS3 transforms. For example, the line above says when the first hour digit is a “0”, and the second hour digit is also a “1” rotate the second digit (remember, it’s been styled to look like an analog clock hand) by 30°. Each hour represents 30° (360 / 12 = 30), so looking at the rest of the hour selectors, you can see how it works.

Similarly, the minutes and seconds are rotated using the immediate sibling selector. The only difference is the degrees, since 360 / 60 = 6° per minute/second.

Ok, neat, it works, but …

How not to build an analog clock

At this point, you’re probably saying to yourself (and if you’re not, you should be) “This is a terrible way to build an analog clock”. You are absolutely right. There are many problems with this, and I’d even go so far as to call the whole thing an antipattern.

From the HTML element hierarchy, to the CSS classnames, to the time computations, most everything tailored to represent an LED-based digital clock.

However, the point of this little exercise was not to find the best way to engineer an analog clock. The point was to answer the question “could it be done?”. But maybe we can learn something about OOCSS anyway.

There is something to be learned here

As the saying goes, “when all you have is a hammer, everything looks like a nail”, and in this case of trying to use only OOCSS to transform the digital clock into analog, there’s certainly some overly-ambitious hammering going on.

If the HTML is poorly-suited to the task at hand, trying to apply OOCSS on top of it will probably just make things worse. In fact, in this example, the HTML and CSS are fighting each other rather than working together.

OOCSS is not just about CSS, it’s about identifying objects in the view first (as John said in our presentation, “ignore the HTML!”), and then structuring HTML and CSS around the containers, content, identity, and states of those objects.

The fact is that the objects in this clock are LED digits, not analog clock hands, and some mildly clever CSS doesn’t change that. Conversely, this bastardization means that the hands of the analog clock have the classes “display-area” and “digit”, as well as “d0” – “d9”, none of which seem like logical choices for the hands of an analog clock!

Antipattern: In any reasonably complex system, writing HTML and CSS first is an antipattern.

What to do?: Break out the wireframes, gather around the whiteboard, and start identifying objects! List their states. Talk about ways to translate them into well-structured HTML containers and content.

One of the results of this object/HTML/CSS mismatch is state explosion, aka combinatorial explosion. There are 72 CSS rules needed just to rotate the clock hands, whereas there are only 10 rules for the original digital clock LED digits. That certainly qualifies as state explosion, and just looking at the analog rules should give you an uneasy feeling that something is wrong.

In fact, modeling any analog clock, not just this bad example of an analog clock, as discrete OOCSS states seems wrong. Consider also the progress bar example John gave during our talk. Progress bars, in most cases, represent a continuous function rather than a discrete function, and therefore can require an infinite number of states to model their possible values—e.g. 30%, 30.1%, 30.15%, and so on.

Antipattern: Trying to model continuous values with OOCSS state is an antipattern.

What to do?: Use a mechanism better suited to continuous values/functions, such as a vector library, or yes, even direct (but well abstracted!) style manipulation.

One other thing that should be bothering you about this analog clock is that nearly half the HTML elements are permanently hidden by the analog theme’s CSS. This can be an indication that you’ve misidentified the view objects. In this case, I think it’s pretty obvious that the objects I originally identified when creating the digital clock, that is, active digits composed of lit or unlit LED bars, simply are not present in the analog clock.

Antipattern: Having sections of permanently hidden HTML is an antipattern.

What to do?: Review your objects and wireframes. You may have misidentified some objects, or your application may have changed significantly enough over time that the objects you had originally, and correctly, identified are no longer present. Either way, it’s time to revisit the wireframes and refactor.

Trying to shoehorn an analog display into the digital clock was fun, but more importantly, I think it helped to identify some OOCSS antipatterns. Hopefully these will help us all avoid some pitfalls!

Ever since Daniel posted his binary mod of my digital clock, I’ve been thinking about other ways to push the limits of the clock using only CSS. At JQCon, I also talked to a few folks between sessions who had wondered the same thing. So, I’ve decided to give it a go.

One of the first ideas I had was to try to use fonts instead of the LED divs to show the clock digits. So, here you go.

It turned out to be fairly easy, and required zero changes to the Javascript view controller. The view controller simply passes the same messages to the view as it did before. In other words, the view controller relies on a message-passing-based view API to which the OOCSS responds. That API is unchanged in this version. To verify, you can hop on over to Daniel’s binary mod, which references my JS directly, and see that it still works.

It also required only superficial changes to the HTML:

1. I added the button for the new theme in the row of theme buttons.
2. I added “hour” and “minute” classes to the digits representing hours and minutes. This change was not necessary to support the new font-based theme, but makes the hours and minutes consistent with the seconds, which already had the class “second”.

The key changes were, of course, in the CSS, and here are some of the most relevant bits.

The comments pretty much say it all, but basically it hides the LEDs and then uses :before content to inject the font-based digits.

One thing I hadn’t thought about before I actually ran it the first time, was that the “1” digit is much thinner than all the others, so I had to forcibly set a specific width for it (in both hours/minutes and the smaller seconds). Without that specialization, the clock looks too sparse when there is a “1” (or several “1"s) being displayed. Yet another win for the OOCSS base and specialization pattern.

I think this theme looks pretty good (although I’m still partial to the LEDs), but if you have suggestions for how to tweak it, I’d love to hear them. Also, if you have an idea for how to push the envelope, feel free to leave a comment, or tweet it up!

Stay tuned for more envelope pushing …