Gratatouille is an opinionated framework to build Gradle plugins. Write pure Kotlin functions and the Gratatouille KSP processor generates tasks, workers, and wiring code for you.
When used in classloader isolation mode, Gratatouille enforces a clear separation between your plugin logic (implementation) and your plugin wiring (api) making your plugin immune to classloader issues 🛡️
Key Features:
- Pure functions
- Kotlinx serialization support
- Comprehensive input/output types
- Non overlapping task outputs
- Build cache
- Documentation
- Parallel execution
- Compile-time task wiring
- Classloader isolation (optional)
Check out the sample-plugin and sample-app.
Apply the com.gradleup.gratatouille plugin:
plugins {
id("java-gradle-plugin")
id("com.gradleup.gratatouille").version("0.0.4")
}Define your task action using @GTaskAction:
@GTaskAction
internal fun prepareIngredients(persons: Int): Ingredients {
return Ingredients(
tomatoes = (persons * 0.75).roundToInt(),
zucchinis = (persons * 0.3).roundToInt(),
eggplants = (persons * 0.3).roundToInt(),
)
}
// kotlinx.serialization is supported out of the box
@Serializable
internal data class Ingredients(
val tomatoes: Int,
val zucchinis: Int,
val eggplants: Int,
)Gratatouille automatically maps function parameters to Gradle inputs and the return value to a Gradle output (more on outputs below).
Gratatouille generates entry points, tasks, workers and Gradle wiring code that you can then use to cook your plugin.
Generated code
internal fun Project.registerPrepareIngredientsTask(
taskName: String = "prepareIngredients",
taskDescription: String? = null,
taskGroup: String? = null,
persons: Provider<Int>,
): TaskProvider<PrepareIngredientsTask> {
val configuration = this@registerPrepareIngredientsTask.configurations.detachedConfiguration()
configuration.dependencies.add(dependencies.create("sample-plugin:implementation:0.0.1"))
return tasks.register(taskName,PrepareIngredientsTask::class.java) {
it.description = taskDescription
it.group = taskGroup
it.classpath.from(configuration)
// infrastructure
// inputs
it.persons.set(persons)
// outputs
it.outputFile.set(this@registerPrepareIngredientsTask.layout.buildDirectory.file("gtask/${taskName}/outputFile"))
}
}
@CacheableTask
internal abstract class PrepareIngredientsTask : DefaultTask() {
@get:InputFiles
@get:PathSensitive(PathSensitivity.RELATIVE)
public abstract val classpath: ConfigurableFileCollection
@get:Input
public abstract val persons: Property<Int>
@get:OutputFile
public abstract val outputFile: RegularFileProperty
@Inject
public abstract fun getWorkerExecutor(): WorkerExecutor
private fun <T> T.isolate(): T {
@kotlin.Suppress("UNCHECKED_CAST")
when (this) {
is Set<*> -> {
return this.map { it.isolate() }.toSet() as T
}
is List<*> -> {
return this.map { it.isolate() } as T
}
is Map<*, *> -> {
return entries.map { it.key.isolate() to it.value.isolate() }.toMap() as T
}
else -> {
return this
}
}
}
@TaskAction
public fun taskAction() {
getWorkerExecutor().noIsolation().submit(PrepareIngredientsWorkAction::class.java) {
it.classpath = classpath.files.isolate()
it.persons = persons.get().isolate()
it.outputFile = outputFile.asFile.get().isolate()
}
}
}
private interface PrepareIngredientsWorkParameters : WorkParameters {
public var classpath: Set<File>
public var persons: Int
public var outputFile: File
}
private abstract class PrepareIngredientsWorkAction : WorkAction<PrepareIngredientsWorkParameters> {
override fun execute() {
with(parameters) {
URLClassLoader(
classpath.map { it.toURI().toURL() }.toTypedArray(),
ClassLoader.getPlatformClassLoader()
).loadClass("recipes.PrepareIngredientsEntryPoint")
.declaredMethods.single()
.invoke(
null,
persons,
outputFile,
)
}
}
}
public class PrepareIngredientsEntryPoint {
public companion object {
@JvmStatic
public fun run(persons: Int, outputFile: File) {
prepareIngredients(
persons = persons,
).encodeJsonTo(outputFile)
}
}
}In your plugin code, use Project.register${TaskAction}Task() to register the task:
override fun apply(project: Project) {
val extension = project.extensions.create("recipes", RecipesExtension::class.java)
// Register your "PrepareIngredients" task
val prepareIngredients = project.registerPrepareIngredientsTask(
persons = extension.persons
)
// Register other tasks
project.registerCookTask(
recipe = extension.recipe,
// Wire tasks together
ingredients = prepareIngredients.flatMap { it.outputFile }
)
}No need to implement DefaultTask, no risk of forgetting @Cacheable, etc... Gratatouille provides good defaults making it easier to write plugins.
Your task code is a side-effect-free function, making it easier to parallelize and reason about.
Nullable parameters are generated as optional task properties. Calls to Provider.get() or Provider.orNull are automated.
Gratatouille has builtin support for kotlinx.serialization. Models are serialized and deserialized as needed.
Inputs:
- Any type annotated with
@Serializable(serialized to a File) - Kotlin
Int,Boolean,Float,Double,String - Kotlin
Set,List,Map - Single File using the
GInputFiletypealias - FileCollection using the
GInputFilestypealias - Directory using the
GInputDirectorytypealias
Outputs:
- Any type annotated with
@Serializable(serialized to a File) - Single File using the
GOutputFiletypealias - Directory using the
GOutputDirectorytypealias
Gratatouille allocates paths for output files and directories automatically. Each output gets a dedicated filesystem location at "build/gtask/${taskName}/${outputName}".
This way:
- you don't have to think about what path to use.
- the outputs are consistent and discoverable.
- issues like #26091 are avoided by construction.
If your function has a single return value, Gratatouille uses outputFile as output name.
If your function needs multiple return values, wrap them in a non-serializable class.
If you need to control the output location of an output, you can do so using @GManuallyWired and using GOutputFile/GOutputDirectory as parameters.
In your implementation:
@GTaskAction
internal fun cook(
recipe: GInputFile,
ingredients: Ingredients,
// ratatouille is exposed in registerCookTask(outputFile) so you can configure it
@GManuallyWired ratatouille: GOutputFile,
// leftovers is set to "build/gtask/cook/leftovers"
leftovers: GOutputFile,
) {
ratatouille.writeText(/* cook here! */)
}In your plugin:
project.registerCookTask(
recipe = extension.recipe,
ingredients = prepareIngredients.flatMap { it.outputFile },
// Set outputFile location explicitly
ratatouille = project.layout.buildDirectory.file("ratatouille")
// No need to set lefovers
)@CacheableTask is added by default. All input files use PathSensitivity.RELATIVE making your tasks relocatable.
@GTaskAction takes a description and a group argument making it easy to colocate your documentation with your implementation:
@GTaskAction(
description = "cooks the most delicious ratatouille with the help of the tiniest chef",
group = "recipes"
)
internal fun cook(
recipe: GInputFile,
ingredients: Ingredients,
outputFile: GOutputFile
) {
TODO()
}By default, Gradle tasks execute serially in a single module (unless using the configuration cache).
Because your task actions are pure Kotlin function, no state is shared, making them perfect candidates for parallelization.
Gratatouille uses the Worker API to allow parallel execution making your build faster overall. Use org.gradle.workers.max to control the maximum number of workers.
Finally, Gratatouille encourages exposing extensions to users instead of task classes directly. All generated code is generated as internal. This makes it easier to have some inputs user configurable while some others are an implementation details and more generally makes it easier to evolve the public API of your plugin.
When a task has a high number of inputs, it can become hard to track which ones have been wired and which ones haven't. By using a central registration point, Gratatouille enforces at build time that all inputs/outputs have been properly wired.
Gradle uses multiple classloaders, and it's notoriously really hard to understand where a given class is loaded from.
Especially, buildSrc/build-logic dependencies leak in the main classpath and override any dependencies from other plugin without conflict resolution. There are multiple workarounds such as declaring all plugins in buildSrc or in the top level build.gradle[.kts] file but the situation is confusing to Gradle newcomers and hard to debug.
To guard against those issues, Gratatouille provides a classloader isolation mode where your task actions use a separate classloader.
This means your plugin can depend on popular dependencies such as the Kotlin stdlib, KotlinPoet or ASM without risking conflicts with other plugins or the Gradle classpath itself.
Furthermore, this classloader is managed by Gratatouille, working around metaspace leaks such as gradle/18313.
For classloader isolation to work, your plugin needs 2 modules:
- The implementation module is where the task actions are defined and the work is done. This module can add dependencies.
- The api module contains the glue code and Gradle API that calls the implementation module through reflection. This module must not add dependencies.
Create an implementation module for your plugin implementation and apply the com.gradleup.gratatouille.implementation plugin:
// implementation/build.gradle.kts
plugins {
id("com.gradleup.gratatouille.implementation").version("0.0.4")
}
dependencies {
// Add other dependencies
implementation("com.squareup:kotlinpoet:1.14.2")
implementation("org.ow2.asm:asm-commons:9.6")
// do **not** add gradleApi() here
}Write your task action as a pure top-level Kotlin function annotated with @GTaskAction:
@GTaskAction
internal fun prepareIngredients(persons: Int): Ingredients {
return Ingredients(
tomatoes = (persons * 0.75).roundToInt(),
zucchinis = (persons * 0.3).roundToInt(),
eggplants = (persons * 0.3).roundToInt(),
)
}
// kotlinx.serialization is supported out of the box
@Serializable
internal data class Ingredients(
val tomatoes: Int,
val zucchinis: Int,
val eggplants: Int,
)When using this mode, the plugin wiring code is generated as resources that are included by the com.gradleup.gratatouille.api plugin.
To use the generated code in your plugin, create an api module next to your implementation module.
Important
By using two different modules, Gratatouille ensures that Gradle classes do not leak in your plugin implementation and vice-versa.
Apply the com.gradleup.gratatouille.api plugin in your api module:
// gradle-plugin/build.gradle.kts
plugins {
id("java-gradle-plugin")
id("com.gradleup.gratatouille.api").version("0.0.4")
}
dependencies {
// Add your implementation module to the "gratatouille" configuration.
// This does not add `:implementation` to your plugin classpath.
// Instead, the generated code uses reflection and a separate classloader to run
// your implementation
gratatouille(project(":implementation"))
}
// Create your plugin as usual, see https://docs.gradle.org/current/userguide/java_gradle_plugin.html
gradlePlugin {
// ...
}In your plugin code, use Project.register${TaskAction}Task() to register the task
Because your task actions are called from a worker and possibly from a completely separate classloader, there is no way to use logger. A future version may transport logs over sockets.