piotr.betkier bartosz.walacik
,

Adopting Gradle at Allegro — a success story

The story begins two years ago during an excellent TDD training given by Szczepan Faber and Tomek Kaczanowski for a bunch of Allegro developers. Surprisingly, it was a trigger to revolutionize our builds at Allegro.

We already knew Szczepan as the Polish rock-star developer and the founder of Mockito. We also knew Tomek as the author of a book on TDD (see practicalunittesting.com) and the conference speaker.

Gradle — what’s the big deal #

At the end of the training, Szczepan had some interesting news for us about his current job. It turned out that he worked full-time as a Gradle core committer.

But back then, few of us had heard about Gradle. What we knew was barely that Gradle had a good-looking website and that it was a new and cool build tool for Java.

Well, if Szczepan Faber is doing Gradle, that must be a project that matters… To add more to our curiosity, he said:

“It’s not a question if you move to Gradle, the question is when.”

Holy cow! Of course we like bleeding edge technologies, but at the time we didn’t see any obvious reason to change good-old Maven to something completely new…

Each of our Java developers knew Maven quite well. All our projects, IDEs, Continuous Integration plans and Continuous Delivery pipelines were backed by Maven. At that time, Maven was The Build Tool. We didn’t like it much but we had gotten used to it.

Gradle — first blood #

Shortly after that TDD training, a few of us started to experiment with Gradle. No one liked the idea of rewriting our existing Maven projects to Gradle.

Fortunately, as we are into microservices architecture, we have a lot of greenfield projects. They offer a perfect opportunity to start playing with a new technology.

We like the moment when, before launching a new greenfield, we gather at the whiteboard and throw some crazy ideas about The New Stack.

At first, Gradle was chosen for toy projects like proofs of concept, small projects created to run a training and so on.

When it turned out that Gradle gives us a kick, a few bravest teams adopted it for production projects. Then we asked Gradleware to run a training for us and we think it was money well invested. Other teams started to follow.

Where we are now #

To make the long story short, at Allegro Gradle is The Build Tool now. Most projects are built with Gradle. Of course, since we have 30+ Scrum teams, there are some Maven conservatives. Some of them stick to Maven claiming it’s faster than Gradle, others say that Gradle is a step back into the dark ages of Ant.

We think that quite the opposite is true: Gradle is a big evolutionary step in build automation and it’s fast. Gradle puts good-old Maven in the place where it belongs — legacy projects.

Now we are going to show you why.

Gradle advantages over Maven #

The following list is just our subjective point of view, derived from our hands-on experience. It’s not a complete list of Gradle features.

So, let’s go on.

Great documentation #

Every now and then, when you try to find out how to add a non-trivial feature to your pom.xml file, you may end up desperately copy-pasting some XML code snippets. Maven has always been missing a Big Picture documentation. Gradle is far better here.

There are two areas of documentation (and also a Javadoc) which could be a bit confusing for Gradle newbies. So let us explain.

On the highest level of abstraction is the User Guide. It could be compared to the comprehensive, book-like Spring Reference. Unlike Maven, Gradle also covers its built-in plugins here.

On the lower level is the DSL Reference. It’s simply the reference for the Gradle language. Why do they call it DSL, isn’t it just Groovy? Well, it makes sense when you take a look at the syntax. For example, to add some plugin to your build, type the following statement in a build.gradle file:

    apply plugin: 'axion-release'

Looks almost like natural language, doesn’t it? Of course, it’s a dedicated, technical language but still far more readable than XML (although some would disagree). That’s the way modern tools communicate with their end-users, by Domain Specific Language. In this case, the language domain is build automation.

There is also a conventional Javadoc, but it contains the same content as the DSL Reference.

Plugins — one way to rule them all #

Gradle team maintains a rich set of plugins and provides them in the standard Gradle distribution. They call them ‘standard plugins’, which include maven plugin or groovy plugin, you can see the full list.

It means that you can do most of the build tasks without configuring plugin dependencies, which was usually the case with Maven.

Of course, there is a community plugins repository hosted by Gradle. A plugin created by our colleague, Adam Dubiel, is also there (read more about axion-release-plugin in this blog post).

Multi-module projects #

In our opinion it’s one of Gradle’s killer features: Gradle truly supports multi-module projects, or multi-project builds as they call them.

First, let’s wrap up what it looks like in Maven. Not so bad, you can say. Maven has the parent pom concept, which was designed to allow configuration inheritance. So a subproject can inherit from its parent project. Unfortunately, the implementation is ill-conceived. You can inherit many things from a parent, except for one: project.version. It has to be hardcoded in each subproject pom file.

This leads to massive code repetition and makes your live harder when releasing. There are plugins, we know. But think about how many times you launched Find and Replace tool to replace 1.0.2 with 1.0.3-SNAPSHOT in all your project pom files?

In Gradle you can define the project version in one place, e.g. in the main build.gradle file.

// gradle magic!
allprojects {
    project.version = '1.0.0'
}

We encourage you to use axion-release-plugin (see this blog post). With this plugin you can discover project version from Git tags and forget about anemic release commits.

Defining internal dependencies between two subprojects is surprisingly concise. For example if you have two modules: bluewhale-core and bluewhale-repository you can define a dependency in bluewhale-repository/build.gradle as follows:

dependencies {
    compile project(':bluewhale-core')
}

Another Gradle feature is the ability to refer to subproject tasks using their paths in the project tree. For example:

./gradlew bluewhale-core:dependencies

means: print the dependency tree for bluewhale-core subproject. Below, we explain why use ./gradlew instead of global gradle command.

Community #

People behind Gradle focus on building the community around it. They accept pull requests on their github — they have 140 contributors already — and are very active on their mailing group. That’s why it’s easy to find advice on solving your specific case. Project’s ownership is clear, Gradleware company, so you know who to speak with in case standard support channels are not enough. Just like the community, Gradle itself is very much alive, with releases every couple of months containing numerous new features and enhancements.

Flexibility #

One of the many things hurting you when using Maven are its limits. First of all, since you work with a configuration in XML, a markup language, you are limited by what the plugins’ authors provided for you. Gradle allows you to perform simple modifications to the build process with just a few lines of Groovy code. When you do need plugins, you configure them with Groovy code as well, so you can achieve things like configuring the parameters conditionally or computing them in runtime very easily.

Maven imposes its idea for the build lifecycle that you cannot alter (or which is cumbersome to do, depending on which Maven version you use). So, unless your desired behaviour fits into this pre-defined flow, there’s not much you can do, even with plugins. Gradle, on the other hand, allows you to shape the build process the way you want.

Custom plugins #

Another Gradle’s advantage over Maven is an excellent support for developing plugins.

This comes from Gradle’s design. At its core, Gradle is just a tool for defining tasks and dependencies between them, it’s the built-in plugins that make it a build system. Since Gradle is so plugin-oriented, you will find it easy to extend it with your own custom plugins. As we discussed earlier, the flexibility that it gives you in altering the build and hooking your own logic is much greater than in Maven.

But flexiblity is not all you get. It’s just very straightforward to write a plugin. You can start by prototyping the code in your build.gradle or in a separate Groovy script and move the code to a separate plugin project once you’ve learned what you want to achieve. Unit testing your plugin is also much easier than it was in Maven. Also, throughout the whole process of creating your custom plugin you will get support from a friendly documentation.

Integration with Maven repositories #

Gradle team took a smart decision to make Gradle backward compatible with Maven. Thanks to that, you can move from Maven to Gradle smoothly.

Gradle uses well established Maven source folder structure (/src/main/..., /src/test/...) by default.

Moreover, Gradle supports Maven repositories. Add this statement to your build.gradle to configure Maven Central repository:

repositories {
    mavenCentral()
}

Dependencies are defined with a very concise syntax, for example:

dependencies {
   testCompile 'junit:junit:4.11'
   testCompile 'org.codehaus.groovy:groovy-all:2.3.3'
   ...
}

To deploy your artifacts to Maven repository, use the maven-publish plugin:

apply plugin: 'maven-publish'

publishing {
    publications {
        maven(MavenPublication) {
            from components.java //means: publish all artifacts from java build
        }
    }
}

Then, you can publish all artifacts to selected repository:

./gradlew publishToMavenCentral

Publishing everything to all defined repositories is even simpler:

./gradlew publish

You can always take a look at available tasks by running ./gradlew tasks command and looking up Publishing tasks section.

You can also use Maven local repository, just add mavenLocal() to the repository list. But think twice before doing this. Gradle guys claim that local repository is a spoiled concept. Your build is less portable when depending on some repository which exists only on your local machine. Although it’s useful when you’re working on a library and want to test it locally.

Instead of the local repository, Gradle uses a local cache. It works pretty much like ordinary cache. It sits between your build and remote repositories and it can’t be referenced directly from your build.gradle.

Gradle Wrapper #

Another nice Gradle feature is Wrapper. It serves two purposes. Firstly, Wrapper unifies Gradle version on all environments for a given project. When using Wrapper, you are sure that all developers and CI tools use the same Gradle version, defined in your project gradle-wrapper.properties. How many times did your Maven build fail due to different Maven versions on dev and CI environments? With Gradle Wrapper, it’s no longer an issue.

Secondly, Wrapper frees you from installing Gradle on your CI server. In fact, it’s kind of a smart installer. Every time you run ./gradlew it does the following:

  • Checks the Gradle version number defined in your project’s gradle/wrapper/gradle-wrapper.properties.
  • If selected gradle distribution has been already downloaded and cached in ~/.gradle dir, Wrapper runs it.
  • If no, Wrapper downloads the right Gradle version and puts it in ~/.gradle dir.

So get into the habit of running ./gradlew instead of gradle.

Gradle Daemon #

Gradle comes with a feature called the Gradle daemon, which reduces the startup and execution time of your tasks. When enabled, Gradle process will remain running after the tasks finished and will be reused the next time your call Gradle. This way you save the time on spawning a new JVM instance, a couple of seconds, which is relatively long for short-running tasks. There are plans to develop the daemon even more, e.g. make it do some work in the background preemptively, and make the builds much faster as a result.

The pitfalls #

Here are some of the pitfalls we’ve encountered when migrating to Gradle.

Magic scripts #

You can extend Gradle by applying custom Groovy scripts in your build.gradle apart from applying custom plugins. This is convenient for a start, but becomes hard to work with once they grow or once you start sharing them between different projects. You need to discover such situations early and transform the scripts into independent plugins published to some artifact repository. We made the mistake of discovering it too late and had to refactor the build scripts in many projects.

Missing plugins #

Right after we migrated from Maven we missed some of its plugins that didn’t have their alternatives in the Gradle world. In some cases we had to fill the gap by ourselves by creating our own plugins, as we did with axion-release-plugin, but in some cases we discovered that it’s not necessary. Gradle’s Groovy build files are so expressive that sometimes a few lines of extra configuration are enough to achieve what used to be a full-blown Maven plugin.

Early adopters pain #

Most of the Allegro teams migrated to Gradle around version 1.11 and there were some early adopters starting much earlier. Gradle was not fully polished back then and we encountered some issues like bugs or weak support in the tools that we used. It was a bit painful, but our desire to drop Maven once we saw Gradle was even stronger. Right now there’s nothing to complain about. The bugs that bothered us got fixed in subsequent releases. Also, our IDEs got much better at supporting Gradle — we’re satisfied with what the newest versions of plugins for IntelliJ and NetBeans have to offer.

Not all developers like Groovy to configure builds #

As is the case with any new technology, some people are not comfortable dropping their old habits. We’ve observed teams rejecting Gradle’s idea of configuring the build using Groovy, prefering XML after years of using Ant and Maven. Our opinion is that they will eventually migrate to Gradle anyway, as late adopters.

Conclusion #

This is our success story of migrating to Gradle. We liked the promises that Gradle made and decided to give it a try. One by one, our teams started dropping Maven and eventually Gradle became the standard build tool in the company. Although we faced some minor problems and fell into some traps during the process, it all went smoothly overall. What matters is that adopting Gradle turned out to be a perfect decision for us and we cannot imagine going back to the Maven world.

confitura

java gradle maven

Discussion