The Sweet Spot
On software, engineering leadership, and anything shiny.

Running Mocha tests with ES6/AMD modules

In one of my personal projects (Chordmeister), I’ve been trying to
upgrade the code to be written in ES6 modules and transpile down to AMD modules with Square’s very excellent es6-module-transpiler project.

Since I’ve already updated an Ember app of mine to try ES6, I figured it was high time to do it on another project.

Sorry Coffeescript, but I’m moving on.

First problem: Coffeescript seems indecisive with respect to ES6
support. In order to support import or export keywords, I had to
wrap the statements in backticks, making the code look like this:

1
2
3
4
5
`import ClassifiedLine from "chordmeister/classified_line"`
class Parser
  # Implementation

`export default Parser`

Except this wasn’t being picked up by es6-module-transpiler, since
Coffeescript wraps the entire declaration in a closure: I was
finding myself having problems compiling from Coffeescript -> ES5 JS -> ES6 JS.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
define("chordmeister/parser",
  [],
  function() {
    "use strict";
    (function() {
      // Oops, I wasn't transpiled!
      import ClassifiedLine from 'chordmeister/classified_line';
      var Parser;
      Parser = (function() {
        // Implementation
      }
      )();
      // Oops, I wasn't transpiled!
      export default Parser;
      })()
  });

So the first call: ditch Coffeescript. Write this in pure ES6.

1
2
3
4
5
6
7
8
import ClassifiedLine from 'chordmeister/classified_line';
var Parser;

Parser = (function() {
  // implementation
})();

export default Parser;

Which transpiled nicely to:

1
2
3
4
5
6
7
8
9
10
11
define("chordmeister/parser",
  ["chordmeister/classified_line","exports"],
  function(__dependency1__, __exports__) {
    "use strict";
    var ClassifiedLine = __dependency1__["default"];
    var Parser;
    Parser = (function() {
      // Implementation
    })();
    __exports__["default"] = Parser;
    });

Next up: adding AMD support in Mocha

Okay, so we need to set up a few things to get Mocha playing well with RequireJS, the AMD loader.

Our plan of attach will be to leverage the generated AMD modules and load our tests up with them. We have the benefit of being able to specifically inject dependencies into our test suite.

The tricky parts will be:

Set up the Mocha index.html runner

Install mocha, require.js, and chai via bower, then plug them into the harness:

test/index.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
<!doctype html>
<html>
<head>
    <meta charset="utf-8">
    <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1">
    <title>Mocha Spec Runner</title>
    <link rel="stylesheet" href="../bower_components/mocha/mocha.css">
</head>
<body>
    <div id="mocha"></div>
    <script src="../bower_components/mocha/mocha.js"></script>
    <script src="../bower_components/chai/chai.js"></script>
    <script data-main="test_helper" src="../bower_components/requirejs/require.js"></script>

</body>
</html>

Note the references to data-main="test_helper", which is require.js’s way of determining its entry point after it loads.

Set up a test runner.

test/test_runner.js
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
// Configure and set up the tests.
require.config({
  baseUrl: "../build/"
})

// Load up the files to run against
var specs = [
  'chordmeister/chord_spec.js',
  'chordmeister/song_spec.js',
  'chordmeister/parser_spec.js',
  'chordmeister/classified_line_spec.js',
];

// Start up the specs.
require(specs, function(require) {
  mocha.setup('bdd');
  expect = chai.expect;
  // Why? Some async loading condition? Is there a callback I should be hooking into?
  setTimeout(function() {
    mocha.run();
  }, 100);
});

You’ll notice that I was having synchonicity issues between spec suite load and mocha.run(). Throwing everything back a few hundred ms seemed to have done the fix.

AMD gotchas

Pay attention to the default parameter that the module exports with. This is important to remember since native ES6 will allow you to directly import it with its native syntax:

1
import Parser from "chordmeister/parser"

But if you’re using RequireJS/AMD modules, you’ll need to explicitly call out the default namespace from the required module, so like so:

1
2
3
4
require(["chordmeister/parser"], function(parser) {
  Parser = parser.default;
  new Parser() // and do stuff.
});

Let me know if you have any questions!

Implementing DDD: Domains, Subdomains and Bounded Contexts

Chapter 2: Domains, Subdomains, and Bounded Contexts

A Domain is what a business does and the surrounding context of how it does business. It is important to model out its supported Subdomains – that is, the smaller components of the business that collaborate in real, day-to-day operations of the business. The book describes these as Core Domains.

Finally, a Bounded Context is the physical manifestation of the solution space as software – models living in a real application. Its key feature in the context of DDD is as a linguistic barrier between different domains.

In an ideal world, each Subdomain maps directly to one Bounded Context. In the real world, this is less common since we tend to build things into monolithic systems. Still – many monolithic applications have several components that could in themselves be bounded contexts.

Side note: In Rails, one could think of Engines as a Bounded Context. But that might be a blog post for another time.

It is important to get these ideas and concepts down correctly because we need correct modeling of our systems to determine what they do.

Bounded Contexts and terms

It’s not usually realistic to get the entire organization agreeing on a universal linguistic definition for every term. Instead, DDD assumes that different terms have different meanings in different contexts.

The author then dives into the an example of a book, where a book means several different things to different people in different contexts. A book is touched upon by authors, graphic designers, editors, marketing folks. In each of these contexts, the features of a book mean different things at different times. It is impossible to develop an all-knowing Book model without disagreement between stakeholders. DDD, instead, acknowledges these differences and allows stakeholders to use linguistic terms from within their unique Bounded Contexts.

Bounded Contexts may include things like:

  • Object models
  • A database schema / persistence layer
  • A SOAP or REST API
  • A user interface

Bounded context antipatterns

You may be tempted to divide up a bounded context by architectural concerns, or because you want to assign smaller tasks to developers (resource allocation). Beware that this kind of work tends to fragment the language.

DDD operates on linquistic drivers. The unity, cohesion and domain-adherence of the bounded context should be the first priority in the design.

Assigning two teams to the same bounded context results in fragmentation of the bounded context.

Ideally: we strive to assign one team to work on one Bounded Context with one Ubiquitous Language at a time.

In Rails, what are the bounded contexts? It could be the top-level Rails application, or an engine, or a gem, that define the context boundaries._

A story…

The chapter then goes on to describe their fictional team designing through three iterations off their DDD strategy:

A system in which all domains live within the same bounded context. They see the folly of this and refactor with some tactical patterns, like creating services.

This is, however, missing the point. They realized they needed to listen to business and their domain experts to find out exactly where the right places were to segregate the contexts. The team discovers that the business has a desire to go in a new direction which allows them to segregate the domain in such a way that would enable future directions for the business.

How often are we as developers in conversation with our product owners and asking where the business wants to go in the future?

Conversations with the business reveal an intention to develop an add-on product to the core product. This implies the development of two subdomains. However, further investigation reveals that the shared overlap of certain domain models (like users, roles, and access management) cannot simply be identically shared between two systems, since their linguistic meanings in the two systems differ slightly. Instead, the developers use the linguistic problem to develop a third bounded context.

The developers separate their app into three bounded contexts built as three software systems.

Six months as a manager

It’s been approximately six months since I’ve entered engineering management. Here are some thoughts reflecting back on that season now.

I didn’t like it at first.

Let’s face it: I didn’t like the feeling of being a “manager” now. It comes with too much baggage of pointy-haired bosses or ineffective waste layers. Why do we need managers anyways?

More to the point: I love coding. I love working on rewarding projects. I hated the thought of being tied up in meetings while day by day getting more and more disconnected from code.

So I struggled a lot in those first few months (still do), trying to maintain the balance of working with product management to get a strategic vision communicated, or with other engineering teams to get a tactical vision in place, versus trying to be deeply involved with the code base and doing code reviews.

You can’t code anymore.

In the end? I had to come to grips with the reality that my team needed me in a different role – a strategic role, helping them see the big picture, sniffing out dependencies and destroying blockers before they got to them.

But I still needed to be spot-on with the code – I still needed to be in the know with new product features as they are being developed. So it’s my priority to be in design discussions, but I can’t be the implementer anymore.

So I’ve started a change in mindset – how can offload all the random codebase knowledge I’ve acquired over the years to the team? How can I effectively share my expertise so I’m out of a (coding) job? I’m starting to be more convinced that the answer to that is pair programming.

Pairing your way out of a job

Nowadays if there’s a story or task in which I am the only one (or one of few) with domain knowledge about the feature, I’ll ask a team member to pair with me on the feature. That way we get a really effective teaching tool about the domain model, and the extra plus in that if I get pulled into a meeting, development can continue.

But you still have to code.

So I code on the weekends (not everybody can do this!)

We’re a Rails team taking on a new Ember project. I need to get my chops up around Ember so I’ve decided to pull in a side project to hack on the weekends. My side work on Wejoinin lets me do all the Rails things I don’t get to do anymore at work.

And it works for me because I love to code, and to be honest, I don’t ever want to get weak in that area.

To be honest, I’d love more feedback from other engineering managers in that area. How do you keep your chops sharp?

People are your priority.

I’ve read often that empathy is everything when it comes to management, and it still rings true. No matter how this product launch goes, in the end your team will remember how you treated them, how their thoughts and feelings were taken into account.

One thing I’m trying to check myself is my tendency to jump in on other people’s sentences and cut them off. It sounds silly, but sometimes I realize I like to talk a lot more than listen. As a manager, sometimes you need to dwell on the feedback of a team member for some time before you write it off. There’s usually a core message in there that’s completely valid – a frustration, a desire for change. Empathy means communicating the message: “I heard you, and I respect your thoughts.”

It’s your attitude

And finally, here’s what I think: your attitude toward your company and the team make ALL the difference.

  • Is your orientation truly to support your team? How will it be clear in the long run that your supported someone in their career?
  • Where are places in your workday where you can connect and sympathize with people – share a joke, listen to someone’s frustration, or just simply go out to lunch?
  • How are you giving people something to work toward to: a vision of changing the world (for Blurb, it’s about transforming the publishing industry).
  • How are you addressing negativity? Grumbling is important to address head on in its own space, but how are you empowering people to take charge of issues of culture or organizational friction?
  • How are you checking your own negativity – sometimes we forget that we’re the solutions to our own problems, and I’ve found oftentimes that the very issues that you assume are impossible to change are crackable by having the right relationships with the right people.

The little things matter.

Growth areas

Lord knows I have a lot to learn here. One area I’m learning to grow in is how to give honest and accurate feedback to team members, without fearing how they’re going to receive it.

Another area? Delegation. I’m learning to delegate the little micro-responsibilities in my job that I just kind of expect that I’m supposed to do. Case in point: every week we accept a translation import from a third-party translation service. I realized that I was burning a lot of time reviewing hundreds of lines of translation keys every week, and the repetition of it was sapping a lot of my energy. I had to learn to ask a team member for help and give them responsibility to own that function of the process.

Blogging through: Implementing Domain-Driven Design

In recent conversations with coworkers, the topic of Domain-Driven Design has
arisen on more than a few occasions in design and architecture meetings.
“Have you read it?” a coworker asked, “I think it’d help us a lot.”

I’ve gotten my hands on a copy of Implementing Domain-Driven Design
by Vaughn Vernon, which is a more pragmatic
approach to DDD than the original Domain-Driven Design book by Eric Evans.

My desire is to share my outlines of the book chapter-by-chapter,
hopefully once a week.

Chapter 1: Getting Started with DDD

Can I DDD?

  • DDD helps us design software models where “our design is exactly how
    the software works” (1).
  • DDD isn’t a technology, it’s a set of principles that involve
    discussion, listening, and business value so you can centralize
    knowledge
    .
  • The main principle here is that we must “understand the business in
    which our software works” (3). This means we learn from domain experts
    in our field.
  • What is a domain model? an object model where objects have
    data/persistence concerns with an accurate business meaning.

Why You Should Do DDD

  • Domain experts and devs on same playing field, cooperation required as
    one team. (Agile teams, anyone?)
  • The business can learn more about itself through the questions asked
    about itself.
  • Knowledge is centralized.
  • Zero translations between domain experts and software devs and
    software.
  • “The design is the code, and code is the design.” (7)
  • It is not without up-front cost

The problem

  • The schism between business domain experts and software developers
    puts your project (and your business) at a risk.
  • The more time passes, the greater the divide grows.

Solution

  • DDD brings domain experts and software developers together to develop
    software that reflects the business domain mental model.
  • Oftentimes this requires that they jointly develop a “Ubiquitous
    Language” - a shared vocabulary and set of concepts that are jointly
    spoken by everybody.
  • DDD produces software that is better designed & architected -> better testable ->
    clearer code.
  • Take heed: DDD should only be used to simplify your domain. If the net
    cost of implementing DDD is only going to add complexity, then you
    should stay away.

Domain model health

  • As time passes, our domain models can become
    anemic,
    and lose their expressive capabilities and clean boundaries. This can
    lead to spaghetti code and a violation of object responsibilities.
  • Why do anemic domain models hurt us? They claim to be well-formed
    models but they hide a badly designed system that is still unfocused
    in what it does. (Andrew: I’ve seen a lot of Service objects that
    claim to be services but really are long scripts to get things done.
    There might be a cost of designing the Service interface, but inside
    things are just as messy as before we got there.)
  • Seems like Vernon is blaming the influence of IDEs for Visual Basic as
    they influenced Java libraries – too many explicit getters and
    setters.
  • Vernon throws up some code samples comparing two different code
    samples – one with an anemic model that looks like a long string of
    commands and another with descriptive method names. He makes the case
    that simply reading the code is documentation of the domain itself.

How to do DDD

  • Have a Ubiquitous Language
    where the team of domain experts share the language together, from
    domain experts to programmers.
  • Steps to coming up with a language:

    1. Draw out the domain and label it.
    2. Make a glossary of terms and definitions.
    3. Have the team review the language document.
  • Note that a Ubiquitous Language is specific to the context it is
    implemented in. In other words, there is one Ubiquitous Language per
    Bounded Context.

Business value of DDD

  1. The organization gains a useful model of its domain
  2. The precise definition of the business is developed
  3. Domain experts contribute to software design.
  4. A better user experience is gained.
  5. Clean boundaries for models keep them pure.
  6. Enterprise architecture is better designed.
  7. Continuous modeling is used – the working software we produce is the
    model we worked so hard to create.
  8. New tools and patterns are used.

Challenges

  • The time and effort required to think about the busines domain,
    research concepts, and converse with domain experts.
  • It may be hard to get a domain expert involved due to their
    availability.
  • There is a lot of thought required to clarify pure models and do
    domain modeling.

Tactical modeling

  • The Core Domain is the part of your application that has key and
    important business value – and may require high thought and attention
    to design.
  • Sometimes DDD may not be the right fit for you – if you have a lot of
    experienced developers who are very comfortable with domain modeling,
    you may be better off trusting their opinion.

DDD is not heavy.

  • It fits into any Agile or XP framework. It leans into TDD, eg: you use
    TDD to develop a new domain model that describes how it interacts with
    other existing models. You go through the red-green-refactor cycle.
  • DDD promotes lightweight development. As domain experts read the code, they
    are able to provide in-flight feedback to the development of the
    system.

Moving to Ember App Kit

I’ve noticed a bit of the buzz around Ember App Kit
recently and decided to move Hendrix, my music management app, over from
a Yeoman-generated Ember app to EAK with all its
bells and whistles.

What’s the difference?

Well on the surface, the two frameworks aren’t very different. The
standard Yeoman build tool sets you up with Grunt and Bower, which is
what EAK provides you out of the box. The cool stuff happens when you
dive under the hood: ES6 module transpilation and an AMD-compatible
Ember Resolver, built-in Karma integration and a built-in API stub
framework for development and test environments.

The joys of modules

What I didn’t realize was that compiling to ES6 modules required that my
filenames be renamed exactly how the modules were going to be placed,
with the extra caveat that resource actions needed to live in their own
directories. Recall that in the old way of doing things with globals and
namespaces, you could get away with throwing a route file like this in
your app directory:

1
2
routes/
  songs_index_controller.js

And inside:

1
2
3
MyApp.SongsIndexRoute = Ember.Route.extend({
  //...
});

In EAK’s world, you need to nest the file under the songs/ directory,
and strip the type from the filename, like so:

1
2
3
routes/
  songs/
    index.js

Inside the file, you assign the function to a variable and let it be
exported in the default namespace.

1
2
3
4
var SongsIndexRoute = Ember.Route.extend({
  //...
});
export default SongsIndexRoute;

File name matters

The new Ember resolver
loads modules in a smart way – according to how the framework
structures resources, controllers and their corresponding actions. So
visiting #/songs from my app caused the app to look up and load
appkit/routes/songs/index. What I didn’t realize was this module must
live at a very specific place in the file directory structure
.
I realized that I left the module type in the file name the first time
around, like this:

1
2
3
routes/
  songs/
    index_route.js

There are no types in the module names – or the filenames, for that
matter. I had not realized this (I’m also an AMD newbie) – so I had
left my files un-renamed as songs_index_route, which meant that
the module loader had stored the SongsIndexRoute module under
appkit/routes/songs/index_route, but was doing a route lookup through
the Resolver for: appkit/routes/songs/index. Renaming the file to:

1
2
3
routes/
  songs/
    index.js

did the trick.