Skip to content

Gradle plugin for running sjsonnet (Scala implementation of the Jsonnet JSON templating language)

License

Notifications You must be signed in to change notification settings

chklauser/sjsonnet-gradle-plugin

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

sjsonnet-gradle-plugin

Gradle plugin for running sjsonnet (Scala implementation of the Jsonnet JSON templating language).

This project is not associated with databricks/sjsonnet, but builds on their excellent work.

Usage

// build.gradle.kts (Kotlin Script Syntax)
plugins {
    id("io.github.chklauser.sjsonnet") version "0.1.0"
}

The plugin offers three levels of configuration-by-convention vs. control.

Source-Sets

If you want to use the standard source file layout and have the generated output included in the resources of the source sets, this approach is recommended.

src/
├─ main/
│  ├─ java/
│   ` jsonnet/
│     ├─ shared.libsonnet
│     ├─ application.jsonnet
│      ` application-dev.jsonnet
 ` test/
   ├─ java/
    ` jsonnet/
       ` mock-data.jsonnet

To keep your Gradle task list tidy, the sjsonnet-gradle-plugin only generates tasks for source sets that you define a specification for. In this example, we want to use jsonnet in both test and the main source set:

// build.gradle.kts (Kotlin Script Syntax)

sjsonnet {
  // minimal configuration (default settings for the `main` source set)
  create("main") { }
  
  // custom settings for the `test` source set
  create("test") { 
      // setting a non-zero indentation enables pretty-printing
      indent.set(2)
  }
}

Custom source sets are also supported. Whenever the name of an sjsonnet specification matches the name of an existing source-set, the plugin will automatically configure the specification to use the sources from the source-set and to consider the outputs part of the 'resources' for that source set (i.e. have the outputs included in the JAR).

In this mode, the plugin only considers files ending in *.jsonnet source files. The remaining files in the directory are considered 'input files' (relevant for up-to-date checking). That way, you can keep shared definitions (commonly kept in *.libsonnet files) and data (used with importstr) together with the generator entry-points.

Custom specifications

If you don't want to work with source-sets (or don't want to include the output in resources), you can define custom specifications. The name of a custom specification must not match any of the source sets.

In the example below, the source files will be read from the ./custom-src directory.

// build.gradle.kts (Kotlin Script Syntax)

sjsonnet { 
  create("custom") {
    indent.set(4)
    sources.from(project.layout.projectDirectory.dir("custom-src"))
  }
}

tasks["assemble"].dependsOn("jsonnetCustomGenerate")

Imports

When you use import and importstr together with custom specifications, you need to take care that Gradle knows about the files that might additionally influence the generator.

In this example, we define the sources as being only the *.jsonnet files (via the filter) and then, separately, add the directory as additionalInputs. The latter is ignored by the plugin, but used by Gradle to decide whether the generator tasks are UP-TO-DATE or not.

// build.gradle.kts (Kotlin Script Syntax)

sjsonnet { 
  create("customWithImports") {
    indent.set(2)
    val srcDir = project.layout.projectDirectory.dir("custom-src")
    sources.from(srcDir)
    sources.filter {f -> f.extension == "jsonnet"}
    additionalInputs.from(srcDir)
  }
}

tasks["assemble"].dependsOn("jsonnetCustomWithImportsGenerate")

Custom task

Instead of using the specification extension, you can manually define SjsonnetTasks for full control. The task supports 1:1 the same configuration options as the specification.

// build.gradle.kts (Kotlin Script Syntax)

tasks {
  "generateData"(io.github.chklauser.sjsonnet.gradle.SjsonnetTask::class) {
    sources.from(project.layout.projectDirectory.dir("data"))
  }
}

If you use import or importstr, the same considerations as for the "custom specifications" approach apply.

Settings

The plugin supports a subset of the settings of sjsonnet. See Config.scala for the documentation of those settings.

All settings below are also available on the SjsonnetTask.

// build.gradle.kts (Kotlin Script Syntax)

sjsonnet {
  create("main") {
    // Sjsonnet settings
    indent.set(2)
    preserveOrder.set(true)
    strict.set(true)
    noStaticErrors.set(true)
    noDuplicateKeysInComprehension.set(true)
    strictImportSyntax.set(true)
    strictInheritedAssertions.set(true)
    topLevelArguments.put("a1", 2)
    externalVariables.put("e1", 1)

    // Gradle plugin settings
    // `searchPath` is the list of directories to search for imported files 
    searchPath.from(...)
    // If `imports` is set, then it serves as an allow-list for imports (only files in the list can be imported)
    // This is useful because the plugin needs to list dependencies for up-to-date checking. Un-declared imported
    // files can lead to unpredictable result.
    imports.from(...)
  }
}

FAQ

Why write a Gradle plugin in scala?

The s in sjsonnet stands for "Scala". Its API is not straightforward to use from Java or Kotlin.

Why is the plugin/specification section/task type called "sjsonnet", but the generated tasks/source set extensions are called "jsonnet"?

"jsonnet" is the language, "sjsonnet" is the implementation. The plugin uses "sjsonnet" and thus the task type, settings, plugin name carry that name.

But the outside world (gradle user) don't need to know about this implementation detail. They are only concerned with having their *.jsonnet files processed quickly.

Why use sjsonnet and not the official jsonnet implementation

Windows support and speed. It's not easy to get a hold of Windows builds for the official implementation. But more importantly: the official implementation is not very efficient. It can take seconds to generate even just medium-sized output files. jsonnet is a pure, functional language, yet the official implementation does little to no caching of re-used expressions. Worse, it implements its entire standard library in jsonnet. While this may be an interesting intellectual exercise, it means that you pay for the inefficient implementation when using standard library functions as well.

sjsonnet addresses both issues. Being written in Scala, it runs on the JVM, inside the Gradle process on any platform where Gradle runs. On the performance front, it applies caching much more aggressively and implements the jsonnet standard library in Java.

Why are the Gradle configuration examples in Kotlin?

Two reasons. First, static types make writing Gradle build scripts more pleasant. Second, the Kotlin syntax is usually a bit trickier to figure out. Going from the Kotlin syntax back to Groovy tends to be simpler.

About

Gradle plugin for running sjsonnet (Scala implementation of the Jsonnet JSON templating language)

Resources

License

Stars

Watchers

Forks

Packages

 
 
 

Languages