ponyfoo.com

Understanding Build Processes

Fix

A task runner helps you automate everything you need to get an environment functional. Configure, build, run tests, and execute your web server. But there’s more than that to a build system. Particularly, in the local environment, there’s a lot you could do to improve development productivity.

The single most important aspect of even resorting to a build system is producing a one-step build process. This means doing everything with a simple command, such as grunt dev. We’ll look at ways to make that happen, and how to improve it as your applications grow. I’ll also mention some of the most common tasks in the process, as well as some that aren’t as widely adopted.

Even though I’ll be talking about Grunt, I won’t get intimate with the tool itself. The concepts I’ll describe should apply to any decent build tool you might pick for the job.

Why a build process?

I figure a good idea, assuming you’ve never done this before, is start by laying out the benefits of using a well-defined build process.

Chances are, you’ve worked in a project where you had to take several steps before you could get your local environment to work. Such a setup process might look similar to this:

  • Create the database by hand. Lucky you, it’s just three easy steps!
  • Restore a data dump into the database. mysql -u root -p suicide_db < ./dump.sql
  • Maybe edit the hosts file by hand
  • Configure local variables such as database authentication, listener port, etc. Usually done by hand
  • Download the latest version of the code. git pull
  • Download external libraries from a repository such as npm, NuGet, gem, pip. npm install
  • Compile your code (in non-interpreted, compiled languages). msbuild
  • Upgrade the database to the latest version. Generally done by a post-build event
  • Run every unit test. java junit.swingui.TestRunner test.Run
  • Compile assets such as CoffeeScript and SASS. coffee --compile --output ./js ./bin
  • Start the web server. node ./server.js

Granted, this is just for the first time, or so you tell yourself. But the truth is, except for a few steps such as restoring the database, you are going to run every single of these steps in production. Well, actually, you are going to want to add a few more steps for a production build. Lets see:

  • Start fresh, don’t leave anything behind, except the data
  • Compile your code in release mode, removing debugging symbols
  • Bundle and minify assets. uglifyjs ./foo/main.js -o ./bin/foo.min.js
  • Finally, point the web server to the new version

Sure, frosting on our cake. But you can’t deny it adds up. Lets see, oh! We might also want to do crafty things in our local environment.

  • Display full stack traces when things blow up
  • Use the unminified versions of vendor libraries such as jQuery
  • Execute all the build steps again after making any changes

And think of the benefits of a one step build process!

You could now hook your repository with continuous integration platforms that alert you in case your tests are acting up, and from there, think just how easy it becomes to push changes to different environments.

The question should then be: why not use a build process?

Where to start?

The very first thing you need to do, is investigate. Learn about the different build tools out there and find one that fits your project. It’s usually a safe bet to pick the most popular build tool around your language. Once you’ve picked a tool, you’ll need to identify the core steps it takes to build and run the application, and how those steps mutate for each environment. Find commonalities.

After you’ve identified these steps, you will have to decide which of those steps will pertain to the build process. You might entertain the idea of doing absolutely everything in that one step, and it might be the right thing to do. But maybe you can leave aside the database creation, while keeping the upgrades, for example. At this point you’ll have to decide whether you want the build step to also function as a “first-time setup”, or not.

Build step by step, in depth

Lets go back to the setup process I outlined earlier, and examine those steps.

  • Create the database by hand. Lucky you, it’s just three easy steps!
  • Restore a data dump into the database. mysql -u root -p suicide_db < ./dump.sql

I’d leave steps like these to a separate flow, designed to set up the local development environment. The reason is obvious enough. We don’t want to be dropping the database and filling it with fake data at any point other than when we set up our local environment. These are probably the only couple of steps I’d be comfortable doing manually, and by that I mean: not in a single command execution.

  • Maybe edit the hosts file by hand
  • Configure local variables such as database authentication, listener port, etc. Usually done by hand

This kind of configuration usually changes over time, so you must have good, up-to-date documentation on how to set up these things. The best thing you can do about this is either add part of this to the setup flow we discussed earlier, or alternatively, add default configuration files that are obvious enough to let developers get set up on their own.

  • Download the latest version of the code. git pull
  • Download external libraries from a repository such as npm, NuGet, gem, pip. npm install

In the local environment, it might be sensible pulling code and installing dependencies by hand, rather than in an automated way, so that you have finer control over these things. Outside the local environment, this kind of steps will depend on the platform you are using.

In the case of this blog, for example, I am using Heroku. Heroku takes care of both of these steps. Well, I git push heroku master, and they respond by pulling that code, executing npm install, and running the application with a command I provide. Travis provides a free CI service, which also takes care of fetching my latest changes and installing dependencies. This might not always be the case, so you should learn your integration or hosting platform, maybe you do have to push your dependency packages as well.

  • Compile your code (in non-interpreted, compiled languages). msbuild
  • Upgrade the database to the latest version. Generally done by a post-build event

In the case of the blog, I don’t really have to compile anything, and Mongoose takes care of keeping my schemas up to date, MongoDB doesn’t really care at all. But, generally speaking, this step will entail compiling your code using the command-line version of your language’s compiler, and somehow updating the database. DbUp is probably an awesome example of how this is done in .NET.

  • Run unit tests. java junit.swingui.TestRunner test.Run

Your code compiled, that’s your first line of defense. In the case of JavaScript, that’d be a linter. Now your test suite needs to give the green-light. Build tools are usually very well-suited to execute unit test runs, and output the test logs, so this one shouldn’t be an issue once you learn the quirks of configuring the tool to run the tests.

  • Compile assets such as CoffeeScript and SASS. coffee --compile --output ./js ./bin
  • Bundle and minify assets. uglifyjs ./foo/main.js -o ./bin/foo.min.js

Nowadays, it is very common in web applications to resort to some little DSL, in order to avoid redundancies such as writing all the prefixes to border-radius, a tool can help you with this. But you then need to compile it, or you need an asset manager such as assetify or SquishIt, to do that for you.

Sometimes those very libraries can handle bundling and minification as well, it’s usually best to do it with the least amount of libraries, that translates as less compatibility issues and less headaches.

  • Start the web server. node ./server.js

This too is platform dependent. Maybe you just want to do this locally for convenience. Maybe you don’t need to do it yourself in your hosting platform, but someone has to trigger the web server to listen for incoming requests.

  • Execute all the build steps again after making any changes

This might currently be one of the most sought after, and trending, build steps. But in order to have a build process that restarts itself, using something like grunt-watch, you must first have a very solid build process that deals with virtually everything I’ve discussed so far.

I’ve tried WebStorm but I don’t think I’m sold on the whole live-edit thing, mainly because it doesn’t play very well with others. Meaning that if you have a custom build process, chances are it’s going to break down.

I’m fine with tabbing away from my editor, and just refreshing my browser. The trick is to enable auto-save. "save_on_focus_lost": true

You can check out the ponyfoo repo if you want to figure out how I’ve configured the build process for this site.

A Book?

I’m currently engaged in talks regarding writing a book on the subject of build processes, maintainable software architecture, and in particular, how these concepts are applied in JavaScript, a language where these things used to be utterly disregarded, and are now beginning to get some affection.

Your feedback and suggestions are more than welcome!

Liked the article? Subscribe below to get an email when new articles come out! Also, follow @ponyfoo on Twitter and @ponyfoo on Facebook.
One-click unsubscribe, anytime. Learn more.

Comments (9)

Michael Allan wrote

I’d love to see this book, too!

Not that I want to discourage you from continuing whatever discussions you already have going … but Leanpub is an awesome alternative to traditional publishing processes. Just sayin’ ;-)

Nicolas Bevacqua wrote

Hey Michael,

I actually closed the deal on the book! It’s scheduled to be published in early 2014, so keep an eye out for it.

Considering this is my first book, I decided to go through a publisher, to reach a wider spectrum of readers. Besides, I wanted to be able to touch my product. Wait, that came out wrong.

That being said, I heard some nice things about Leanpub; I might give it a try in the future. Thanks for the encouragement!

Michael Allan wrote

Awesome! Nice one :-)

I’ll keep my eyes out, for sure. In fact, if you allow Early Access I’d love to signup/purchase and follow along, and hopefully provide helpful feedback along the way. I’m sure others would too.

Are you free to say which publisher?

Nicolas Bevacqua wrote

JavaScript Application Design is now available for preorder, you can learn more about it here.

Michael Allan wrote

I’d bought it even before you commented :-)