React Components for Google Maps – Part 2

Making React and Google Maps play nice with each other is not a complex task, provided you have some insight in how to open seams in the google maps layer to allow it to interoperate with other libraries. I have learned this from the school of hard knocks, so I hope to shed some light on the tricky parts to save you some time. If you haven’t already, I recommend you read part 1 of this series for some important background information.

Let’s dig into the implementation. I have broken out the examples into a set of gists. If you don’t care to follow through each step, you can jump straight to the full example of the OverlayView.

Implementing the OverlayView

There are 3 high level requirements for creating a custom OverlayView. The module must:

  1. Prototypically inherit from the google.maps.OverlayView class
  2. Implement a onAdd method, which will be called once the map has been set for the overlay
  3. Implement a draw method, which will be called each time the map is zoomed

Secret sauce

The secret sauce to getting the overlay and React to play nice is to take the DOM element representing the content of the overlay as a parameter through the constructor. As we will see later on, this will allow us to handle rendering the content of the overlay in the React component, but position it on the map through the overlay.

Extend the google.maps.OverlayView class

Adding the overlay to the DOM

The onAdd method is the point at which the overlay can insert itself into the DOM. This is achieved by using the getPanes method, which is inherited from the parent OverlayView class. There are multiple panes, but for our particular case, we’ll use the overlayLayer pane. This will allow us to control any click-like behavior completely within the DOM structure of the overlay.

It’s important to understand the purpose of this method when integrating an overlay with a particular framework or library, because typically inserting a particular view into the DOM is often handled by the framework. This method is our seam that allows us to inject content into the map without having that content coupled to the map itself.

Positioning the overlay on the map

The draw method is invoked when the overlay needs to calculate its position on the map. In the simplest case, an overlay is positioned by a single lat/lng point, however, it is also possible to position the overlay by a bounds as well. We’ll focus on the simplest case in this article. In order to translate a lat/lng to a pixel, we need to get a reference to the projection of the overlay. Projections are google maps’ way of turning lat/lng points into pixels relative to something within the map structure.

To safely get a reference to a projection, we have to wait for the map instance to be idle. I have found it is best to handle this dependency outside the overlay, particularly if you plan on having to render hundreds or thousands of overlays on the map. I will discuss how to implement handling this externally as part of implementing the facade. Regardless of where it is handled, the way to accomplish knowing when the map is idle is by adding a listener to the map instance for the idle event.


From my experience, when rendering a high volume of overlays on the map (300+), the draw method has the highest likelihood to impact rendering performance. There are two reasons for this. First of all, relatively speaking, it is expensive to determine the dimensions of an element in the DOM. We need the dimensions of the overlay in order to properly position it on the map. Secondly, the draw method will be invoked each time we zoom, therefore, we will need the dimensions of our overlay again in order to correctly position it.

When rendering high volumes of overlays, we can dramatically improve performance by caching the dimensions. To take it a step further, if your overlays all have a consistent height and width, you can calculate the dimensions once and use the cached copy for all other instances. Exactly how to achieve that is outside the scope of this article, but I wanted to highlight it, because it has caused me a great deal of heartburn in the past.

Implementing the facade

Our general requirements for the facade are:

  1. Handle the creation of the google map instance (a factory)
  2. Provide a facility for consumers to know when the map instance is idle without having to add listeners themselves
  3. Provide a factory for creating our custom overlays

Map facade

The best way I have found to handle both items 1 and 2 is to define a method that will handle creating the map instance, but returns a Promise. The Promise is resolved once the map is idle, and should resolve with the newly created map instance. This allows consumers to bind to the Promise and to perform any actions that are dependent upon the map being idle. As consumers, we don’t care if we bind to the map early or late in its lifecycle, once it’s idle, we know our callback will be invoked and we’ll have access to an instance of the map.

Overlay factory

To avoid coupling ourselves to our custom overlay, it is best to implement a factory to handle creating instances of the overlay for us. This allows us to handle the setup of the overlay in a single place, and will improve the testability of modules that need to create an overlay instance.

Implementing the React component

By abstracting away the details of defining the overlay and its interactions with the map in our facade, there is very little overlay-specific logic that needs to be included in the component. All we need to do is call our factory. The key is to do this as part of the componentDidMount method of the component, since this is the point in time where we will have access to the DOM node of the component. You are free to handle the rendering of the content for the overlay in the render method of the component.

Secret sauce

The key to achieving this is to render the overlay component with a detached div element and not an element that is already in the DOM. This allows us to render our content in the detached element and to pass it to the overlay. The component can then safely share it’s DOM node with the overlay. Here’s an example of how to achieve this with my idle-maps module.

Parting Thoughts

Where I work, we have a very map-centric application where, at times, we render 2,000+ overlays. We also have a variety of types of overlays that we leverage. Following the patterns I have laid out has proven to be very effective for us. While building out this capability within our application, it would have been quite valuable to have known some of these patterns early on. I hope by documenting them here, I can save another developer some time and pain going down this path.

React Components for Google Maps – Part 1

Building React components for use with google maps can present some challenging problems for those who aren’t aware of the basics/quirks of developing with google maps. I aim to shed some light on the fundamentals of how to get the two to play nicely together, and to encourage developing the components in a way that will promote testability. As always, feel free to jump right into the code if you don’t need the TL;DR description of how it all works.

As you’ll notice in the example, I have created two npm modules to support this:

  • async-google-maps – provides a facade for asynchronously creating google map instances as well as a base overlay views that are positioned by a lat/lng
  • idle-maps – builds on top of the async-google-maps module, by providing a set of React components for creating google map instances and overlay content


The standard google maps modules are fine for simple use cases, such as displaying a marker for a given Lat/Lng point or InfoWindows for descriptive content about a given point. However, things get interesting when you need to be able to have full control over the style of a component, or to easily update content based on changes from the server. There are third-party libraries out there that will wrap the google maps primitives, but those have their own problems and often contain a bunch of bloat trying to handle all possible user interactions and browser quirks.


  1. A basic understanding of spatial data primitives: coordinate systems, projections, basic geometric data types
  2. A basic understanding of React
  3. A basic understanding of google maps
  4. A basic understanding of browserify
  5. A google maps API key

Where we’re going

As an introductory step, we will start by building a custom OverlayView that is positioned by a point (as opposed to being positioned by a bounds, which is also possible). We will keep this simple by focusing on the basics using React with google maps and avoid fetching data from the server or binding UI interactions to a client-side router. We’ll cover those in later articles.

Separating concerns

My preferred method for separating concerns when building custom OverlayViews is to leverage the following pattern:

  • Build a constructor function which inherits from the google.maps.OverlayView constructor.
  • Implement a facade, (in the spirit of the Command Pattern) which abstracts away the complexity of the coordination that needs to occur between the OverlayView and the React component
  • Implement the React component in such a way that it knows nothing about any of the google maps APIs

What’s next?

In the next article, we will take a deep dive into the implementation and highlight a few of the pitfalls I discovered while working on this myself.

Testing Javascript Singletons

Testing javascript singleton modules can be a challenge, especially ones that maintain some sort of internal state. Examples of singletons that manage internal state are usually either a factory or cache manager. In order to effectively test these sorts of modules, you often either resort to exposing a method to reset the state, or carefully order your tests… until now!

To eliminate this problem, I propose the following solution (assuming you’re using a spec-style test):

  1. Wait to require the module you’re testing until all stubs, spies, and mocks are defined in a beforeEach
  2. require the module you’re testing after stubbing/spying/mocking in the beforeEach
  3. in the afterEach, remove the module you’re testing from the require.cache
    • this effectively destroys the singleton and when it is require‘d again the the beforeEach it will be rebuilt.

Here’s how you remove a module from the require.cache:

I’m assuming that you have your tests in a sibling directory to a lib directory, where your production modules live. If not, you can adjust the path spec accordingly.

Other things to be mindful of are:

  • Event listeners (both DOM and custom events)
  • The state of the DOM itself
  • Callbacks you’ve stored off in the body of stubs and/or spies within your test suite

These are all things that need to be cleared out in the afterEach as well.

But wait, I’m a front-end developer and I use karma or testem to execute my tests in a real browser. Well, I personally believe runners are a source of pain and frustration and should be avoided. However, it is theoretically possible with karma to achieve this by removing the script element that corresponds to the module you’re testing in the afterEach and then append it back immediately after removing it. I am not familiar enough with testem to give equivalent advice.

Continuous Integration with the MEAN Stack

Travis-CI has great support for continuous integration with the MEAN stack. It took some research in order to achieve a complete solution, which then caused some revamping of my initial config. I want to share my notes with my agile hacker friends so you can take advantage of this with your projects, both large and small. This article is meant to be a continuation of my previous articles on setting up a proper test environment for the MEAN stack. Some things in this article are contextual and may seem out of place unless you’ve read the previous articles:

  1. TDD and BDD With The MEAN Stack: Introduction
  2. BDD with MEAN – The Server Part 1

Using continuous integration for large scale projects is, in my opinion, a requirement. Some might debate its value for small/personal projects, however, I’d like to put that to rest as well.

For the non-believers

Some may think that CI for small scale, and/or personal projects, is overkill. However, I believe CI for small/personal projects provides the following benefits:

  1. Verification that your application can run in an isolated environment away your personal machine on standard infrastructure.
  2. Marketing and a certain level of assurance to potential consumers that you treat your tests as a first class concern.
    • Users can see first-hand that your tests run effectively, instead of just telling them to run npm test themselves after downloading your code.
    • It shows attention to the engineering process as well as an additional level of transparency.
  3. It is an enabler for Continuous Deployment. In fact, Travis-CI supports continuous deployment to several different environments out of the box.

I believe strongly in maintaining the application in an always shippable state. Having a proper CI strategy is one step on that path.

Key Elements

  • Externalized application configuration
  • Travis-CI config for public github projects

Application Configuration

I chose to use the node-config module with javascript format. There are two keys strengths to this module:

  1. It supports using javascript for the configuration format, which makes it easy to build reusable chunks of configuration by just exposing some simple functions to return parts of the configuration object literal.
  2. It maps the NODE_ENV environment variable to configuration files stored in the config directory in the root of your project by convention. This allows you to create specialized configurations for development, CI, production, etc.

I chose travisci as the name of my CI environment, which, as we’ll see later, is set as the exported value of the NODE_ENV environment variable in the .travis.yml config file.

The reusable parent configuration

travisci Environment Configuration

Travis-CI Configuration

The key ingredients to tying this configuration back to our node-config configuration file are the NODE_ENV and BUILD_DIR environment variables. With Travis-CI, you can export environment variables by declaring them under the following yaml block:


Since I’ve chosen to use Grunt to automate task execution for the project, and bower to manage front-end dependencies, these global modules must be installed during the before_script: hook. The other blocks (language, node_js, and services tell Travis-CI what sort of environment is needed. What’s really cool about this, is that we can go ahead and run our BDD tests in this environment as well since we have a mongodb instance available. This is one of my favorite features of Travis-CI.

Secret Sauce

I did some digging and discovered that Travis-CI clones your repo into $HOME/<username>, so the effective working directory will be $HOME/<username>/<cloned repo name>. However, in their environment variable documentation, it is recommended to not rely upon the value of the HOME environment variable. To work around this, I chose to hook into this in the configuration by using the pwd sub-shell command in the .travis.yml file as part of the value for BUILD_DIR. In the travisci.js configuration file, I then utilize the BUILD_DIR as the initial working directory. This is used to avoid having a bunch of relative path references in the require calls of the application.

Parting Thoughts

I’m quite pleased with this setup for the Travis-CI environment. I went ahead and included their build badges in my thoughtsom github repo. I intended to also include code coverage integration with coveralls, however, I ran into some issues with the blanket code coverage tool. I plan to give istanbul a spin, but I have not had time to properly set that up yet. The long-term goal is to get continuous deployment to OpenShift working.

BDD with MEAN – The Server Part 1

As with any new endeavor, it pays to spend some time trying various solutions out and sometimes failing miserably. This is especially true for us progressive nerds who like to live on the bleeding edge without things like Stack Overflow to constantly save our ass. What I’d like to do is to help you avoid going through the pain of figuring out what works and what doesn’t.

As I mentioned in my previous post, I already have a project that serves as a working example of if you wish to jump straight into the code: . All of the gists used in this post were pulled from that project.

The first step on our journey to effective BDD testing with the MEAN stack will be to start wiring up the various tools we’ll need to use to get a working environment. Afterwards, we’ll build out a helper to manage our test environment and fixtures.

Let’s start by reviewing our toolbox:

Our Tool Box

  • Grunt
    • Grunt is used for general task running, . It can be used in a similar manner to how you might use rake when working on a ruby-based project. It is particularly useful on the server side when you combine it with watch and nodemon, which are more runtime oriented.
  • Yadda
    • As mentioned earlier, Yadda will be serving as our BDD layer/framework. Yadda itself needs to integrate with another test framework like Mocha or Jasmine to provide a complete stack. It includes a Gherkin parser, multilingual support, and parameterized step definitions.
  • Mocha
    • Mocha is a general purpose javascript test framework. Mocha will be used more explicitly on the TDD side, but Yadda’s integration with it allows you to make use of the Before, BeforeEach, After, and AfterEach hooks to setup and tear down your test fixtures.
  • Sinon
    • Is a standalone test library that has an great support for stubing, spying, mocking, and additional assertions that compliment your regular xUnit style assertions when TDD’ing. Sinon also provides a fakeServer API that can be useful for prototyping endpoints or as a test double for external dependencies in BDD tests.
  • Chai
    • Provides both bdd style assertions (expect and should) as well as xUnit style assertions.
  • grunt-mocha-cov
    • A grunt plugin for running mocha tests. Since we’ve chosen to use mocha as the underlying test framework for Yadda, this will help automate the execution of our BDD tests.
  • Supertest
    • Used to test our express routes. Supertest has the ability to consume an express app and it’s associated routes. One of the things I especially like about Supertest is that it does not require you to have previously spun up an instance of your application in order to execute the tests.
  • Proxyquire
    • Allows us to inject stubs or fake objects into the require namespace.
  • Casual
    • A robust library for generating test data.
  • node-config
    • require paths and database connection information quickly gets messy. I like to extact some of that mess out into config files.
  • Yaml
    • I prefer yaml config files because they are short and sweet. This module works nicely with the node-config module.

Folder structure

The next most logical step is to establish our folder structure. I have a specific folder structure that I like to use and I am not currently aware of a yeoman generator that will scaffold anything based on my requirements, so for now I recommend creating the following stucture:

|-- config
 |-- Gruntfile.js
 |-- package.json
 |-- src
 |   |-- server
 |   |   |-- app
 |   |   |   `-- controllers
 |   |   |-- models
 |   |   |-- repositories
 |   |   `-- schemas
 `-- test
     |-- acceptance
     |   -- server
     |       |-- features
     |       |   |-- step_definitions
     |       |   `-- support
     |-- helpers
     |   -- stubs
     `-- unit
         |-- server
         |   |-- app
         |   |   -- controllers
         |   `-- repository

The configuration file – config/default.yml

The configuration file is pretty straightforward, for the most part. I like to establish some path prefixes to avoid the 3rd level of require hell, which is crazy relative path references. The one cryptic part might be the knownObjectId key, which I will explain later in the section on world.js.

The Gruntfile

The most important part of this is the mochacov configuration block. I am well aware of the simple mocha plugin, however, I have found that mochacov appears to be more active and it also gives much better output when things go wrong (which happened to me quite often while I was experimenting with how to get this all configured). Aside from my usage of mochacov, I have chosen to run jshint against both the production code and the tests. I feel that this is a good practice, which helps ensure consistency between code styles of production code and test code, as well as a safety net to help avoid tricky syntactic boogers.. I mean issues that are not always obvious.

The World

One of the useful things I learned from my cucumber-js experiment is the value of having a helper that can aid in setting up and tearing down the test environment and related fixtures. I have chosen to borrow this concept for usage with Yadda/Mocha as well. The world.js file should go in the following directory: test/acceptance/server/features/support/world.js

There are 3 main highlights in the world.js helper:

  1. It handles connecting and disconnecting from our local mongodb instance with connectToDB(done) and disconnectDB(done).
    • Note that both of these take done as a parameter, which is the callback that mocha will use to determine when an asynchronous function is truly finished.
  2. It handles inserting test data for us, with at least one known ObjectId with createThought(done)
    • This method currently only inserts one row of data into the datastore, however, the important part is that it inserts a record with a known primary key that we can then use in our route test to make sure that we can actually return a result when we hit the /thought/:id route. The value is pulled from our knownObjectId key in our default.yml configuration file.
  3. It clears the database (with clearDB(done)), which means that each of our test runs are idempotent.


As I have started to dig into this, I have realized that I could essentially write a book on this topic. If you’re anxious to take off with this and looking for more examples, I encourage you to take a look at my thoughtsom repo on github, which I continue to build out and try to commit chunks to each week. Getting the test environment and project structure right at the beginning will help us stay organized and have a solid configuration base to build on top of.

I plan on breaking this up into separate posts in hopes to make this a complete, but consumable reference.

Here’s a rough sketch of what to expect:

  1. Part 1 (this post): cover the basic libraries we’ll use, and establish the prerequisites for automating our test runs.
  2. Part 2: Write our first feature and failing step definition using Supertest to verify an express route.
  3. Part 3: A little application architecture and more advanced usages, such as multiple step definitions and features.
  4. Part 4: Unit test overview: stubs and mocks with sinon

After these 4 parts, I plan to begin covering the UI side of things and demonstrate how most of the tools will translate to being used on the both the server and the browser sides of testing. I welcome any feedback on the quality of this article and the direction you’d like to see this take. Feel free to flame/praise me on twitter.

TDD and BDD With The MEAN Stack: Introduction

As the MEAN stack is growing in adoption, a variety of testing strategies have sprouted up on the interwebs. As I have started to dig deeper into automated testing strategies for both sides of the MEAN stack, I have found it difficult to find advice or material on how to setup an environment to support the mockist style of test-driving the implementation. Many articles offer some light recommendations, but are typically more classicist-oriented. I’d like to help fill the gap by outlining a tooling strategy which I believe enables the following:

  • Automation
  • Mockist unit testing
  • BDD/ATDD testing
  • End-To-End system testing

In case you want to dive straight into the code, I’ve set up a project on github to prove out the technology.

I’m going to assume you’ve already got a working MongoDB and Node.js environment. If not, take a look at a vagrant solution or

One of my goals has been to try to find tools that can work on both the server and the browser. I also put a lot of stock into automation, so I have aimed to avoid tools that require manual loading of browser pages to launch the tests. This has proven to be quite a challenge, but the xUnit-side of testing is much closer to that goal than the BDD-side. However, all hope is not lost.

Let’s get down to business and outline the tools that are working well on both sides:

  • Grunt – The task runner
  • Mocha – The test framework
  • Chai – The assertion library
  • Sinon – The stub, spy, and mock swissarmy knife

Some tools are sort of in the middle, where they can be made to work in both the browser and the server, but they aren’t rock solid in one or the other:

  • Yadda – BDD layer that works nicely with Mocha+Chai (needs a little help on the browser side, but I’m working on that)
  • Cucumber-js – (BDD) works great on the server-side, but fails in the browser side (particularly because of a lack of support for karma integration)

And finally, some tools are targeted at one side or the other:

  • Karma – THE test runner for the browser
  • karma-browserifast – A browserify plugin for Karma, which is needed to automate running Yadda BDD tests in the browser
  • Supertest – Awesome library for testing Express-based RESTful endpoints
  • CasperJS – Drive user interactions through the UI (and works with Yadda)

There is more to come. I plan to write two more articles to dig further into to the details of what it means to test drive on the server side and in the browser. For now, take a look at my Yadda/Karma example on Github for BDD in the browser and thoughtsom for BDD and unit testing on the server. I plan to build out of a few more code examples on the browser side of testing before the next set of articles. In the mean time, I welcome your feedback.

P.S. I am aware of Protractor, but I have not had much of a chance to experiment with it yet.

Setting up a project using karma with mocha and chai

Recently, I was on a quest to evaluate the mocha javascript testing framework
and its direct support for async testing in the browser. I had a hard time finding a complete description of everything that needed to be done in order to get things working with karma for testing in the browser as opposed to testing a node module.

Consequently, I decided it would be useful to create a guide for those who have similar interests to save everyone time. If you want to skip the steps and just download the code, feel free to clone my github repository. As a quick aside, I decided to try mocha thanks to some frustrations with jasmine. Async testing with jasmine can be done, but it does not have first-class support in my opinion. I intend to give buster a spin in the near future as well.

This tutorial assumes that you are on a unix-based system (I used OSX to write the article) and already have a working nodejs and npm installation, therefore installing and configuring a node.js/npm environment is outside the scope of this article. With that said, let’s dig in!

  1. Create a directory to hold the project and navigate to that directory:
    • mkdir <new_dir>
    • cd <new_dir>
  2. We’ll be installing a bunch of npm modules, so let’s run npm init to build our package.json file for us
  3. Let’s install karma with npm. I recommend installing it locally, since it’s likely that you’ll be either executing karma from an IDE (such as webstorm) or as a grunt task. Additionally, one project may be configured for one version of karma while a different project can use a different version if needed.
    • npm install karma --save-dev
    • At the time I wrote this article, the latest stable version of karma was 0.10.8.
  4. Now we’re ready to install mocha:
    • npm install mocha --save-dev
  5. We’ll also need the karma-mocha adapter so we can use mocha as a test framework with karma:
    • npm install karma-mocha --save-dev
  6. An assertion library will need to included as well. I prefer the chai assertion library, since it offers both bdd style assertions using expect and should as well as junit-style assertions.
    • npm install karma-chai --save-dev.
    • Installing chai this way will make it available as a framework within karma, which makes setup much simpler.
  7. I also find sinon to be extremely useful for stubbing and spying, so let’s install it to round out our set of packages required to get our test environment provisioned:
    • npm install karma-sinon --save-dev

At this point, our package.json should have a devDependencies directive that looks something like this:


We can now create our karma configuration and write some tests. Let’s first make sure that our karma installation is working correctly:

  • ./node_modules/karma/bin/karma --version

That should print out the version number to the console if everything is working correctly. If you want to, you can initialize a karma configuration file using:

  • ./node_modules/karma/bin/karma init

I’ll provide the boilerplate so you can avoid that step. Include the following configuration in a file named


Once that file is ready, we can create a failing test to ensure our configuration is working as expected. Create a new file in src/test called test.spec.js. Create the directory:

  • mkdir -p src/test

And a test file with the following contents:

describe("A test suite", function() {
   beforeEach(function() { });
   afterEach(function() { });
   it('should fail', function() { expect(true); });

With our test spec in place, we can run karma and make sure that we can successfully execute failing tests. ./node_modules/karma/bin/karma start karma.conf.js

After running our tests, we should see some output along the lines of

A test suite should fail FAILED

If so, then you’ve successfully configured karma with mocha, chai and sinon. Cheers!

Update: 07/14/14

I recently updated this to use the new 0.12.x version of karma and also added an example Gruntfile.js

MV* Client-side Javascript Frameworks

Since I chose to take the sysadmin career off-ramp (at least for now) back in April in order to explore my other passions as a full-time developer, I have been spending a lot of time re-sharpening my proverbial programmer saw. Lately, I have become fascinated by the glut of client-side javascript frameworks and I have set out to find a few to experiment with (along with nodejs on the server side with a mongodb backend). During my research today, I stumbled upon this very well-written article, which I felt compelled to share: . Kudos to the author, @stevesanderson, for doing such a great job of providing a straightforward comparison of the various frameworks/libraries.

Of the ones listed, AngularJS seems to appeal to me the most. Thoughts?

On the future of (semi) structured information

In the past few years there has been a great deal of discussion on alternatives to relational database systems. Data systems and information architecture have been a strong interest of mine for quite some time. Like many budding computer scientists, I was taught the fundamentals of relational database design during my undergraduate studies, but there was little discussion of alternative systems. Towards the end of my graduate studies, I decided to take the time to do a deep dive on alternatives to traditional relational database systems; mainly column-oriented datastores such as Google’s BigTable, Amazon’s Dynamo, and Facebook/Apache’s Cassandra. I focused on these three, instead of systems such as MongoDB or Riak, primarily because of the wealth of scholarly information about arguably the three largest deployments of alternative datastores.

Rather than keeping 4 months of research to myself, I thought I’d share with the community. I have included my research paper and a presentation I gave at the end of the semester to summarize my research.