The Principles of Lean Java
I began my software development career writing Python. It wasn’t until a couple of years later that I began to write Java in earnest. Like many Python/Ruby programmers, my opinion of Java at the time was shaped by various blog posts and HackerNews comments bashing Java’s verbosity and heaviness. So when I started doing Java development around 2012, I was pleasantly suprised - modern Java was easy to use, lightweight, and dare I say even fun to write. The subsequent years have seen this trend accelerate, with new Java releases that help reduce boilerplate and the popularity of high-quality, lightweight libraries and frameworks.
This is not to say that Java’s reputation is entirely unfounded. I have also had to deal with legacy frameworks, XML hell and over-abstracted code. Thankfully, these things can be almost entirely avoided. This blog post is an attempt to distill my experience into a few general (opinionated) principles that I’ve used to write lean1, modern Java.
1. Prefer annotations/code over XML
Configuration should be easy for humans to read and edit. XML’s signal-to-noise ratio is just too low. Thankfully, annotations/straight-up-code have already won this battle in the Java world. So much so that, Spring, one of the worst proliferators of XML as configuration, stopped supporting XML-based request mapping
2. Prefer functional code over imperative code
Java 8 introduced lambdas and streams that help reduce a lot of boilerplate. Still, I’ve seen code that looks like this (in Java 8)
vs the much more succint version using streams:
3. Embrace modern deployment tools
The traditional way of deploying Java web applications was to package up your code and deploy it within a separate application server like Tomcat. This meant that it was hard to set up a local development environment that matched the production setup. With the rise of container technology, the application and its dependencies are packaged up into a single, easy-to-deploy artifact. This is obviously at odds with the application server model.
So, I agree that Java application servers are dead. Modern Java web applications use embedded servlet containers like Jetty and package up all dependencies into an “uber-jar”. I’m a big fan of the Dropwizard framework which adopts this approach.
Expect this trend to accelerate with the introduction of a module sytem in Java 9.
4. Eschew heavyweight standards
It seems like a portion of the Java community love adhering to every standard put out by the JCP. In my opinion, some of these like JPA are too heavyweight for most applications. I dislike ORMs in general, let alone an ORM API as a standard. Instead, I recommend using a thin wrapper over SQL like JDBI or jOOQ if you don’t like writing raw SQL.
This doesn’t mean I’m for eschewing every standard out there - in fact, Dropwizard, the framework I mentioned above, is built on top of a JAX-RS implementation. Adopt standards where they make sense.
5. Embrace non-Java tools
This might seem counter-intuitive in an article about Java, but I’ve noticed that some Java developers tend to use Java for every conceivable task. For example, it is essential to pick up a scripting language like Python for one-off automation tasks or Unix tools like grep/awk/sed for log munging.
Full-stack developers will also need to learn HTML/CSS/Javascript, but I’d recommend backend Java developers learn the basics too. With DevOps gaining popularity, it doesn’t hurt to learn configuration/infrastructure management tools like Chef or Puppet.
Summary
Despite stiff competition, the Java ecosystem is still one of the best (if not the best) platforms to build modern web application backends. Some heavy-weight, enterprise-y baggage still remains, but is not hard to avoid.
-
By “lean” I do not mean Lean Software Development, rather “lean” in the real sense of the word - devoid of unnecessary burden ↩