Gradle Kotlin DSL Primer
Gradle’s Kotlin DSL provides an alternative syntax to the traditional Groovy DSL with an enhanced editing experience in supported IDEs, with superior content assist, refactoring, documentation, and more. This chapter provides details of the main Kotlin DSL constructs and how to use it to interact with the Gradle API.
💡
|
If you are interested in migrating an existing Gradle build to the Kotlin DSL, please also check out the dedicated migration guide. |
Prerequisites
-
The embedded Kotlin compiler is known to work on Linux, macOS, Windows, Cygwin, FreeBSD and Solaris on x86-64 architectures.
-
Knowledge of Kotlin syntax and basic language features is very helpful. The Kotlin reference documentation and Kotlin Koans will help you to learn the basics.
-
Use of the plugins {} block to declare Gradle plugins significantly improves the editing experience and is highly recommended.
IDE support
The Kotlin DSL is fully supported by IntelliJ IDEA and Android Studio. Other IDEs do not yet provide helpful tools for editing Kotlin DSL files, but you can still import Kotlin-DSL-based builds and work with them as usual.
Build import | Syntax highlighting 1 | Semantic editor 2 | |
---|---|---|---|
IntelliJ IDEA |
✓ |
✓ |
✓ |
Android Studio |
✓ |
✓ |
✓ |
Eclipse IDE |
✓ |
✓ |
✖ |
CLion |
✓ |
✓ |
✖ |
Apache NetBeans |
✓ |
✓ |
✖ |
Visual Studio Code (LSP) |
✓ |
✓ |
✖ |
Visual Studio |
✓ |
✖ |
✖ |
1 Kotlin syntax highlighting in Gradle Kotlin DSL scripts
2 code completion, navigation to sources, documentation, refactorings etc… in Gradle Kotlin DSL scripts
As mentioned in the limitations, you must import your project from the Gradle model to get content-assist and refactoring tools for Kotlin DSL scripts in IntelliJ IDEA.
In addition, IntelliJ IDEA and Android Studio might spawn up to 3 Gradle daemons when editing Gradle scripts — one for each type of script: build scripts, settings files and initialization scripts. Builds with slow configuration time might affect the IDE responsiveness, so please check out the performance guide to help resolve such issues.
Automatic build import vs. automatic reloading of script dependencies
Both IntelliJ IDEA and Android Studio — which is derived from IntelliJ IDEA — will detect when you make changes to your build logic and offer two suggestions:
-
Import the whole build again
-
Reload script dependencies when editing a build script
We recommend that you disable automatic build import, but enable automatic reloading of script dependencies. That way you get early feedback while editing Gradle scripts and control over when the whole build setup gets synchronized with your IDE.
Troubleshooting
The IDE support is provided by two components:
-
The Kotlin Plugin used by IntelliJ IDEA/Android Studio
-
Gradle
The level of support varies based on the versions of each.
If you run into trouble, the first thing you should try is running ./gradlew tasks
from the command line to see whether your issue is limited to the IDE. If you encounter the same problem from the command line, then the issue is with the build rather than the IDE integration.
If you can run the build successfully from the command line but your script editor is complaining, then you should try restarting your IDE and invalidating its caches.
If the above doesn’t work and you suspect an issue with the Kotlin DSL script editor, you can:
-
Run
./gradle tasks
to get more details -
Check the logs in one of these locations:
-
$HOME/Library/Logs/gradle-kotlin-dsl
on Mac OS X -
$HOME/.gradle-kotlin-dsl/logs
on Linux -
$HOME/AppData/Local/gradle-kotlin-dsl/log
on Windows
-
-
Open an issue on the Gradle issue tracker, including as much detail as you can.
From version 5.1 onwards, the log directory is cleaned up automatically. It is checked periodically (at most every 24 hours) and log files are deleted if they haven’t been used for 7 days.
For IDE problems outside of the Kotlin DSL script editor, please open issues in the corresponding IDE’s issue tracker:
Lastly, if you face problems with Gradle itself or with the Kotlin DSL, please open issues on the Gradle issue tracker.
Kotlin DSL scripts
Just like the Groovy-based equivalent, the Kotlin DSL is implemented on top of Gradle’s Java API. Everything you can read in a Kotlin DSL script is Kotlin code compiled and executed by Gradle. Many of the objects, functions and properties you use in your build scripts come from the Gradle API and the APIs of the applied plugins.
Script file names
✨
|
Groovy DSL script files use the Kotlin DSL script files use the |
To activate the Kotlin DSL, simply use the .gradle.kts
extension for your build scripts in place of .gradle
. That also applies to the settings file — for example settings.gradle.kts
— and initialization scripts.
Note that you can mix Groovy DSL build scripts with Kotlin DSL ones, i.e. a Kotlin DSL build script can apply a Groovy DSL one and each project in a multi-project build can use either one.
We recommend that you apply the following conventions to get better IDE support:
-
Name settings scripts (or any script that is backed by a Gradle
Settings
object) according to the pattern*.settings.gradle.kts
— this includes script plugins that are applied from settings scripts -
Name initialization scripts according to the pattern
*.init.gradle.kts
or simplyinit.gradle.kts
.
Implicit imports
All Kotlin DSL build scripts have implicit imports consisting of:
-
The Kotlin DSL API, which is all types within the
org.gradle.kotlin.dsl
andorg.gradle.kotlin.dsl.plugins.dsl
packages currently
⚠
|
Avoid using internal Kotlin DSL APIs
Use of internal Kotlin DSL APIs in plugins and build scripts has the potential to break builds when either Gradle or plugins change.
The Kotlin DSL API extends the Gradle public API with the types listed in the corresponding API docs that are in the |
Type-safe model accessors
The Groovy DSL allows you to reference many elements of the build model by name, even when they are defined at runtime. Think named configurations, named source sets, and so on. For example, you can get hold of the implementation
configuration via configurations.implementation
.
The Kotlin DSL replaces such dynamic resolution with type-safe model accessors that work with model elements contributed by plugins.
Understanding when type-safe model accessors are available
The Kotlin DSL currently supports type-safe model accessors for any of the following that are contributed by plugins:
-
Dependency and artifact configurations (such as
implementation
andruntimeOnly
contributed by the Java Plugin) -
Project extensions and conventions (such as
sourceSets
) -
Elements in the
tasks
andconfigurations
containers -
Elements in project-extension containers (for example the source sets contributed by the Java Plugin that are added to the
sourceSets
container) -
Extensions on each of the above
❗️
|
Only the main project build scripts and precompiled project script plugins have type-safe model accessors. Initialization scripts, settings scripts, script plugins do not. These limitations will be removed in a future Gradle release. |
The set of type-safe model accessors available is calculated right before evaluating the script body, immediately after the plugins {}
block.
Any model elements contributed after that point do not work with type-safe model accessors.
For example, this includes any configurations you might define in your own build script.
However, this approach does mean that you can use type-safe accessors for any model elements that are contributed by plugins that are applied by parent projects.
The following project build script demonstrates how you can access various configurations, extensions and other elements using type-safe accessors:
-
Uses type-safe accessors for the
api
,implementation
andtestImplementation
dependency configurations contributed by the Java Library Plugin -
Uses an accessor to configure the
sourceSets
project extension -
Uses an accessor to configure the
main
source set -
Uses an accessor to configure the
java
source for themain
source set -
Uses an accessor to configure the
test
task
💡
|
Your IDE knows about the type-safe accessors, so it will include them in its suggestions.
This will happen both at the top level of your build scripts — most plugin extensions are added to the |
Note that accessors for elements of containers such as configurations
, tasks
and sourceSets
leverage Gradle’s configuration avoidance APIs.
For example, on tasks
they are of type TaskProvider<T>
and provide a lazy reference and lazy configuration of the underlying task.
Here are some examples that illustrate the situations in which configuration avoidance applies:
tasks.test {
// lazy configuration
}
// Lazy reference
val testProvider: TaskProvider<Test> = tasks.test
testProvider {
// lazy configuration
}
// Eagerly realized Test task, defeat configuration avoidance if done out of a lazy context
val test: Test = tasks.test.get()
For all other containers than tasks
, accessors for elements are of type NamedDomainObjectProvider<T>
and provide the same behavior.
Understanding what to do when type-safe model accessors are not available
Consider the sample build script shown above that demonstrates the use of type-safe accessors.
The following sample is exactly the same except that is uses the apply()
method to apply the plugin.
The build script can not use type-safe accessors in this case because the apply()
call happens in the body of the build script.
You have to use other techniques instead, as demonstrated here:
Type-safe accessors are unavailable for model elements contributed by the following:
-
Plugins applied via the
apply(plugin = "id")
method -
The project build script
-
Script plugins, via
apply(from = "script-plugin.gradle.kts")
-
Plugins applied via cross-project configuration
You also can not use type-safe accessors in Binary Gradle plugins implemented in Kotlin.
If you can’t find a type-safe accessor, fall back to using the normal API for the corresponding types. To do that, you need to know the names and/or types of the configured model elements. We’ll now show you how those can be discovered by looking at the above script in detail.
Artifact configurations
The following sample demonstrates how to reference and configure artifact configurations without type accessors:
The code looks similar to that for the type-safe accessors, except that the configuration names are string literals in this case.
You can use string literals for configuration names in dependency declarations and within the configurations {}
block.
The IDE won’t be able to help you discover the available configurations in this situation, but you can look them up either in the corresponding plugin’s documentation or by running gradle dependencies
.
Project extensions and conventions
Project extensions and conventions have both a name and a unique type, but the Kotlin DSL only needs to know the type in order to configure them.
As the following sample shows for the sourceSets {}
and java {}
blocks from the original example build script, you can use the configure<T>()
function with the corresponding type to do that:
Note that sourceSets
is a Gradle extension on Project
of type SourceSetContainer
and java
is an extension on Project
of type JavaPluginExtension
.
You can discover what extensions and conventions are available either by looking at the documentation for the applied plugins or by running gradle kotlinDslAccessorsReport
, which prints the Kotlin code necessary to access the model elements contributed by all the applied plugins.
The report provides both names and types.
As a last resort, you can also check a plugin’s source code, but that shouldn’t be necessary in the majority of cases.
Note that you can also use the the<T>()
function if you only need a reference to the extension or convention without configuring it, or if you want to perform a one-line configuration, like so:
the<SourceSetContainer>()["main"].srcDir("src/core/java")
The snippet above also demonstrates one way of configuring the elements of a project extension that is a container.
Elements in project-extension containers
Container-based project extensions, such as SourceSetContainer
, also allow you to configure the elements held by them.
In our sample build script, we want to configure a source set named main
within the source set container, which we can do by using the named() method in place of an accessor, like so:
All elements within a container-based project extension have a name, so you can use this technique in all such cases.
As for project extensions and conventions themselves, you can discover what elements are present in any container by either looking at the documentation of the applied plugins or by running gradle kotlinDslAccessorsReport
.
And as a last resort, you may be able to view the plugin’s source code to find out what it does, but that shouldn’t be necessary in the majority of cases.
Tasks
Tasks are not managed through a container-based project extension, but they are part of a container that behaves in a similar way. This means that you can configure tasks in the same way as you do for source sets, as you can see in this example:
We are using the Gradle API to refer to the tasks by name and type, rather than using accessors.
Note that it’s necessary to specify the type of the task explicitly, otherwise the script won’t compile because the inferred type will be Task
, not Test
, and the testLogging
property is specific to the Test
task type.
You can, however, omit the type if you only need to configure properties or to call methods that are common to all tasks, i.e. they are declared on the Task
interface.
One can discover what tasks are available by running gradle tasks
. You can then find out the type of a given task by running gradle help --task <taskName>
, as demonstrated here:
❯ ./gradlew help --task test
...
Type
Test (org.gradle.api.tasks.testing.Test)
Note that the IDE can assist you with the required imports, so you only need the simple names of the types, i.e. without the package name part.
In this case, there’s no need to import the Test
task type as it is part of the Gradle API and is therefore imported implicitly.
About conventions
Some of the Gradle core plugins expose configurability with the help of a so-called convention object. These serve a similar purpose to — and have now been superseded by — extensions. Please avoid using convention objects when writing new plugins. The long term plan is to migrate all Gradle core plugins to use extensions and remove the convention objects altogether.
As seen above, the Kotlin DSL provides accessors only for convention objects on Project
.
There are situations that require you to interact with a Gradle plugin that uses convention objects on other types.
The Kotlin DSL provides the withConvention(T::class) {}
extension function to do this:
This technique is most commonly required for source sets that are added by language plugins other than the Java Plugin, e.g. the Groovy Plugin and the Scala Plugin. You can see which plugins add which properties to source sets in the SourceSet reference documentation.
Multi-project builds
As with single-project builds, you should try to use the plugins {}
block in your multi-project builds so that you can use the type-safe accessors. Another consideration with multi-project builds is that you won’t be able to use type-safe accessors when configuring subprojects within the root build script or with other forms of cross configuration between projects. We discuss both topics in more detail in the following sections.
Applying plugins
You can declare your plugins within the subprojects to which they apply, but we recommend that you also declare them within the root project build script. This makes it easier to keep plugin versions consistent across projects within a build. The approach also improves the performance of the build.
The Using Gradle plugins chapter explains how you can declare plugins in the root project build script with a version and then apply them to the appropriate subprojects' build scripts. What follows is an example of this approach using three subprojects and three plugins. Note how the root build script only declares the community plugins as the Java Library Plugin is tied to the version of Gradle you are using:
plugins {}
blockIf your build requires additional plugin repositories on top of the Gradle Plugin Portal, you should declare them in the pluginManagement {}
block in your settings.gradle.kts
file, like so:
Plugins fetched from a source other than the Gradle Plugin Portal can only be declared via the plugins {}
block if they are published with their plugin marker artifacts.
✨
|
At the time of writing, all versions of the Android Plugin for Gradle up to 3.2.0 present in the google() repository lack plugin marker artifacts.
|
If those artifacts are missing, then you can’t use the plugins {}
block. You must instead fall back to declaring your plugin dependencies using the buildscript {}
block in the root project build script. Here’s an example of doing that for the Android Plugin:
buildscript {}
blockThis technique is not that different from what Android Studio produces when creating a new build.
The main difference is that the subprojects' build scripts in the above sample declare their plugins using the plugins {}
block. This means that you can use type-safe accessors for the model elements that they contribute.
Note that you can’t use this technique if you want to apply such a plugin either to the root project build script of a multi-project build (rather than solely to its subprojects) or to a single-project build. You’ll need to use a different approach in those cases that we detail in another section.
Cross-configuring projects
Cross project configuration is a mechanism by which you can configure a project from another project’s build script. A common example is when you configure subprojects in the root project build script.
Taking this approach means that you won’t be able to use type-safe accessors for model elements contributed by the plugins. You will instead have to rely on string literals and the standard Gradle APIs.
As an example, let’s modify the Java/Ratpack sample build to fully configure its subprojects from the root project build script:
Note how we’re using the apply()
method to apply the plugins since the plugins {}
block doesn’t work in this context.
We are also using standard APIs instead of type-safe accessors to configure tasks, extensions and conventions — an approach that we discussed in more detail elsewhere.
When you can’t use the plugins {}
block
Plugins fetched from a source other than the Gradle Plugin Portal may or may not be usable with the plugins {}
block.
It depends on how they have been published and, specifically, whether they have been published with the necessary plugin marker artifacts.
For example, the Android Plugin for Gradle is not published to the Gradle Plugin Portal and — at least up to version 3.2.0 of the plugin — the metadata required to resolve the artifacts for a given plugin identifier is not published to the Google repository.
If your build is a multi-project build and you don’t need to apply such a plugin to your root project, then you can get round this issue using the technique described above. For any other situation, keep reading.
💡
|
When publishing plugins, please use Gradle’s built-in Gradle Plugin Development Plugin.
It automates the publication of the metadata necessary to make your plugins usable with the |
We will show you in this section how to apply the Android Plugin to a single-project build or the root project of a multi-project build.
The goal is to instruct your build on how to map the com.android.application
plugin identifier to a resolvable artifact.
This is done in two steps:
-
Add a plugin repository to the build’s settings script
-
Map the plugin ID to the corresponding artifact coordinates
You accomplish both steps by configuring a pluginManagement {}
block in the build’s settings script.
To demonstrate, the following sample adds the google()
repository — where the Android plugin is published — to the repository search list, and uses a resolutionStrategy {}
block to map the com.android.application
plugin ID to the com.android.tools.build:gradle:<version>
artifact available in the google()
repository:
In fact, the above sample will work for all com.android.*
plugins that are provided by the specified module. That’s because the packaged module contains the details of which plugin ID maps to which plugin implementation class, using the properties-file mechanism described in the Writing Custom Plugins chapter.
See the Plugin Management section of the Gradle user manual for more information on the pluginManagement {}
block and what it can be used for.
Working with container objects
The Gradle build model makes heavy use of container objects (or just "containers").
For example, both configurations
and tasks
are container objects that contain Configuration
and Task
objects respectively.
Community plugins also contribute containers, like the android.buildTypes
container contributed by the Android Plugin.
The Kotlin DSL provides several ways for build authors to interact with containers.
We look at each of those ways next, using the tasks
container as an example.
💡
|
Note that you can leverage the type-safe accessors described in another section if you are configuring existing elements on supported containers. That section also describes which containers support type-safe accessors. |
Using the container API
All containers in Gradle implement NamedDomainObjectContainer<DomainObjectType>. Some of them can contain objects of different types and implement PolymorphicDomainObjectContainer<BaseType>. The simplest way to interact with containers is through these interfaces.
The following sample demonstrates how you can use the named() method to configure existing tasks and the register() method to create new ones.
-
Gets a reference of type
Task
to the existing task namedcheck
-
Registers a new untyped task named
myTask1
-
Gets a reference to the existing task named
compileJava
of typeJavaCompile
-
Registers a new task named
myCopy1
of typeCopy
-
Gets a reference to the existing (untyped) task named
assemble
and configures it — you can only configure properties and methods that are available onTask
with this syntax -
Registers a new untyped task named
myTask2
and configures it — you can only configure properties and methods that are available onTask
in this case -
Gets a reference to the existing task named
test
of typeTest
and configures it — in this case you have access to the properties and methods of the specified type -
Registers a new task named
myCopy2
of typeCopy
and configures it
✨
|
The above sample relies on the configuration avoidance APIs. If you need or want to eagerly configure or register container elements, simply replace |
Using Kotlin delegated properties
Another way to interact with containers is via Kotlin delegated properties. These are particularly useful if you need a reference to a container element that you can use elsewhere in the build. In addition, Kotlin delegated properties can easily be renamed via IDE refactoring.
The following sample does the exact same things as the one in the previous section, but it uses delegated properties and reuses those references in place of string-literal task paths:
-
Uses the reference to the
myTask1
task rather than a task path
✨
|
The above rely on configuration avoidance APIs. If you need to eagerly configure or register container elements simply replace |
Configuring multiple container elements together
When configuring several elements of a container one can group interactions in a block in order to avoid repeating the container’s name on each interaction. The following example uses a combination of type-safe accessors, the container API and Kotlin delegated properties:
Working with runtime properties
Gradle has two main sources of properties that are defined at runtime: project properties and extra properties. The Kotlin DSL provides specific syntax for working with these types of properties, which we look at in the following sections.
Project properties
The Kotlin DSL allows you to access project properties by binding them via Kotlin delegated properties. Here’s a sample snippet that demonstrates the technique for a couple of project properties, one of which must be defined:
val myProperty: String by project // (1)
val myNullableProperty: String? by project // (2)
-
Makes the
myProperty
project property available via amyProperty
delegated property — the project property must exist in this case, otherwise the build will fail when the build script attempts to use themyProperty
value -
Does the same for the
myNullableProperty
project property, but the build won’t fail on using themyNullableProperty
value as long as you check for null (standard Kotlin rules for null safety apply)
The same approach works in both settings and initialization scripts, except you use by settings
and by gradle
respectively in place of by project
.
Extra properties
Extra properties are available on any object that implements the ExtensionAware interface.
Kotlin DSL allows you to access extra properties and create new ones via delegated properties, using any of the by extra
forms demonstrated in the following sample:
val myNewProperty by extra("initial value") // (1)
val myOtherNewProperty by extra { "lazy initial value" } // (2)
val myProperty: String by extra // (3)
val myNullableProperty: String? by extra // (4)
-
Creates a new extra property called
myNewProperty
in the current context (the project in this case) and initializes it with the value"initial value"
, which also determines the property’s type -
Create a new extra property whose initial value is calculated when the property is accessed
-
Binds an existing extra property from the current context (the project in this case) to a
myProperty
reference -
Does the same as the previous line but allows the property to have a null value
This approach works for all Gradle scripts: project build scripts, script plugins, settings scripts and initialization scripts.
You can also access extra properties on a root project from a subproject using the following syntax:
val myNewProperty: String by rootProject.extra // (1)
-
Binds the root project’s
myNewProperty
extra property to a reference of the same name
Extra properties aren’t just limited to projects.
For example, Task
extends ExtensionAware
, so you can attach extra properties to tasks as well.
Here’s an example that defines a new myNewTaskProperty
on the test
task and then uses that property to initialize another task:
tasks {
test {
val reportType by extra("dev") // (1)
doLast {
// Use 'suffix' for post processing of reports
}
}
register<Zip>("archiveTestReports") {
val reportType: String by test.get().extra // (2)
archiveAppendix.set(reportType)
from(test.get().reports.html.destination)
}
}
-
Creates a new
reportType
extra property on thetest
task -
Makes the
test
task’sreportType
extra property available to configure thearchiveTestReports
task
If you’re happy to use eager configuration rather than the configuration avoidance APIs, you could use a single, "global" property for the report type, like this:
tasks.test.doLast { ... }
val testReportType by tasks.test.get().extra("dev") // (1)
tasks.create<Zip>("archiveTestReports") {
archiveAppendix.set(testReportType) // (2)
from(test.get().reports.html.destination)
}
-
Creates and initializes an extra property on the
test
task, binding it to a "global" property -
Uses the "global" property to initialize the
archiveTestReports
task
There is one last syntax for extra properties that we should cover, one that treats extra
as a map.
We recommend against using this in general as you lose the benefits of Kotlin’s type checking and it prevents IDEs from providing as much support as they could.
However, it is more succinct than the delegated properties syntax and can reasonably be used if you only need to set the value of an extra property without referencing it later.
Here’s a simple example demonstrating how to set and read extra properties using the map syntax:
extra["myNewProperty"] = "initial value" // (1)
tasks.create("myTask") {
doLast {
println("Property: ${project.extra["myNewProperty"]}") // (2)
}
}
-
Creates a new project extra property called
myNewProperty
and sets its value -
Reads the value from the project extra property we created — note the
project.
qualifier onextra[…]
, otherwise Gradle will assume we want to read an extra property from the task
The Kotlin DSL Plugin
The Kotlin DSL Plugin provides a convenient way to develop Kotlin-based projects that contribute build logic. That includes buildSrc projects, included builds and Gradle plugins.
The plugin achieves this by doing the following:
-
Applies the Kotlin Plugin, which adds support for compiling Kotlin source files.
-
Adds the
kotlin-stdlib-jdk8
,kotlin-reflect
andgradleKotlinDsl()
dependencies to thecompileOnly
andtestImplementation
configurations, which allows you to make use of those Kotlin libraries and the Gradle API in your Kotlin code.All three libraries and their dependencies are bundled with Gradle, so these dependencies will not result in any downloads.
-
Configures the Kotlin compiler with the same settings that are used for Kotlin DSL scripts, ensuring consistency between your build logic and those scripts.
-
Enables support for precompiled script plugins.
⚠
|
Avoid specifying a version for the
kotlin-dsl pluginEach Gradle release is meant to be used with a specific version of the |
This is the basic configuration you need to use the plugin:
buildSrc
projectBe aware that the Kotlin DSL Plugin turns on experimental Kotlin compiler features. See the Kotlin compiler arguments section below for more information.
By default, the plugin warns about using experimental features of the Kotlin compiler.
You can silence the warning by setting the experimentalWarning
property of the kotlinDslPluginOptions
extension to false
as follows:
Precompiled script plugins
In addition to normal Kotlin source files that go under src/main/kotlin
by convention, the Kotlin DSL Plugin also allows you to provide your build logic as precompiled script plugins.
You write these as *.gradle.kts
files in that same src/main/kotlin
directory.
Precompiled script plugins are Kotlin DSL scripts that are compiled as part of a regular Kotlin source set and then placed on the build classpath or packaged in a binary plugin, depending on what type of project they’re in. For all intents and purposes, they are binary plugins, particularly as they can be applied by plugin ID, just like a normal plugin. In fact, the Kotlin DSL Plugin generates plugin metadata for them thanks to integration with the Gradle Plugin Development Plugin.
So, to apply a precompiled script plugin, you need to know its ID.
That is derived from its filename (minus the .gradle.kts
extension) and its (optional) package declaration.
For example, the script src/main/kotlin/java-library-convention.gradle.kts
would have a plugin ID of java-library-convention
(assuming it has no package declaration).
Likewise, src/main/kotlin/my/java-library-convention.gradle.kts
would result in a plugin ID of my.java-library-convention
as long as it has a package declaration of my
.
To demonstrate how you can implement and use a precompiled script plugin, let’s walk through an example based on a buildSrc
project.
First, you need a buildSrc/build.gradle.kts
file that applies the Kotlin DSL Plugin:
buildSrc
projectWe recommend that you also create a buildSrc/settings.gradle.kts
file, which you may leave empty.
Next, create a new java-library-convention.gradle.kts
file in the buildSrc/src/main/kotlin
directory and set its contents to the following:
This script plugin simply applies the Java Library and Checkstyle Plugins and configures them. Note that this will actually apply the plugins to the main project, i.e. the one that applies the precompiled script plugin
Finally, apply the script plugin to the root project as follows:
The embedded Kotlin
Gradle embeds Kotlin in order to provide support for Kotlin-based scripts.
Kotlin versions
Gradle ships with kotlin-compiler-embeddable
plus matching versions of kotlin-stdlib
and kotlin-reflect
libraries. For example, Gradle 4.3 ships with the Kotlin DSL v0.12.1 that includes Kotlin 1.1.51 versions of these modules. The kotlin
package from those modules is visible through the Gradle classpath.
The compatibility guarantees provided by Kotlin apply for both backward and forward compatibility.
Backward compatibility
Our approach is to only do backwards-breaking Kotlin upgrades on a major Gradle release. We will always clearly document which Kotlin version we ship and announce upgrade plans before a major release.
Plugin authors who want to stay compatible with older Gradle versions need to limit their API usage to a subset that is compatible with these old versions. It’s not really different from any other new API in Gradle. E.g. if we introduce a new API for dependency resolution and a plugin wants to use that API, then they either need to drop support for older Gradle versions or they need to do some clever organization of their code to only execute the new code path on newer versions.
Forward compatibility
The biggest issue is the compatibility between the external kotlin-gradle-plugin
version and the kotlin-stdlib
version shipped with Gradle. More generally, between any plugin that transitively depends on kotlin-stdlib
and its version shipped with Gradle. As long as the combination is compatible everything should work. This will become less of an issue as the language matures.
Kotlin compiler arguments
These are the Kotlin compiler arguments used for compiling Kotlin DSL scripts and Kotlin sources and scripts in a project that has the kotlin-dsl
plugin applied:
-jvm-target=1.8
-
Sets the target version of the generated JVM bytecode to
1.8
. -Xjsr305=strict
-
Sets up Kotlin’s Java interoperability to strictly follow JSR-305 annotations for increased null safety. See Calling Java code from Kotlin in the Kotlin documentation for more information.
-XX:NewInference
-
Enables the experimental Kotlin compiler inference engine (required for SAM conversion for Kotlin functions).
-XX:SamConversionForKotlinFunctions
-
Enables SAM (Single Abstract Method) conversion for Kotlin functions in order to allow Kotlin build logic to expose and consume
org.gradle.api.Action<T>
based APIs. Such APIs can then be used uniformly from both the Kotlin and Groovy DSLs.As an example, given the following hypothetical Kotlin function with a Java SAM parameter type:
fun kotlinFunctionWithJavaSam(action: org.gradle.api.Action<Any>) = TODO()
SAM conversion for Kotlin functions enables the following usage of the function:
kotlinFunctionWithJavaSam { // ... }
Without SAM conversion for Kotlin functions one would have to explicitly convert the passed lambda:
kotlinFunctionWithJavaSam(Action { // ... })
Interoperability
When mixing languages in your build logic, you may have to cross language boundaries. An extreme example would be a build that uses tasks and plugins that are implemented in Java, Groovy and Kotlin, while also using both Kotlin DSL and Groovy DSL build scripts.
Quoting the Kotlin reference documentation:
Kotlin is designed with Java Interoperability in mind. Existing Java code can be called from Kotlin in a natural way, and Kotlin code can be used from Java rather smoothly as well.
Both calling Java from Kotlin and calling Kotlin from Java are very well covered in the Kotlin reference documentation.
The same mostly applies to interoperability with Groovy code. In addition, the Kotlin DSL provides several ways to opt into Groovy semantics, which we look at next.
Static extensions
Both the Groovy and Kotlin languages support extending existing classes via Groovy Extension modules and Kotlin extensions.
To call a Kotlin extension function from Groovy, call it as a static function, passing the receiver as the first parameter:
TheTargetTypeKt.kotlinExtensionFunction(receiver, "parameters", 42, aReference)
Kotlin extension functions are package-level functions and you can learn how to locate the name of the type declaring a given Kotlin extension in see the Package-Level Functions section of the Kotlin reference documentation.
To call a Groovy extension method from Kotlin, the same approach applies: call it as a static function passing the receiver as the first parameter. Here’s an example:
Named parameters and default arguments
Both the Groovy and Kotlin languages support named function parameters and default arguments, although they are implemented very differently.
Kotlin has fully-fledged support for both, as described in the Kotlin language reference under named arguments and default arguments.
Groovy implements named arguments in a non-type-safe way based on a Map<String, ?>
parameter, which means they cannot be combined with default arguments.
In other words, you can only use one or the other in Groovy for any given method.
Calling Kotlin from Groovy
To call a Kotlin function that has named arguments from Groovy, just use a normal method call with positional parameters. There is no way to provide values by argument name.
To call a Kotlin function that has default arguments from Groovy, always pass values for all the function parameters.
Calling Groovy from Kotlin
To call a Groovy function with named arguments from Kotlin, you need to pass a Map<String, ?>
, as shown in this example:
To call a Groovy function with default arguments from Kotlin, always pass values for all the parameters.
Groovy closures from Kotlin
You may sometimes have to call Groovy methods that take Closure arguments from Kotlin code. For example, some third-party plugins written in Groovy expect closure arguments.
✨
|
Gradle plugins written in any language should prefer the type |
In order to provide a way to construct closures while preserving Kotlin’s strong typing, two helper methods exist:
-
closureOf<T> {}
-
delegateClosureOf<T> {}
Both methods are useful in different circumstances and depend upon the method you are passing the Closure
instance into.
Some plugins expect simple closures, as with the Bintray plugin:
closureOf<T> {}
In other cases, like with the Gretty Plugin when configuring farms, the plugin expects a delegate closure:
delegateClosureOf<T> {}
There sometimes isn’t a good way to tell, from looking at the source code, which version to use.
Usually, if you get a NullPointerException
with closureOf<T> {}
, using delegateClosureOf<T> {}
will resolve the problem.
These two utility functions are useful for configuration closures, but some plugins might expect Groovy closures for other purposes.
The KotlinClosure0
to KotlinClosure2
types allows adapting Kotlin functions to Groovy closures with more flexibility.
KotlinClosureX
typesAlso see the groovy-interop sample.
The Kotlin DSL Groovy Builder
If some plugin makes heavy use of Groovy metaprogramming, then using it from Kotlin or Java or any statically-compiled language can be very cumbersome.
The Kotlin DSL provides a withGroovyBuilder {}
utility extension that attaches the Groovy metaprogramming semantics to objects of type Any
.
The following example demonstrates several features of the method on the object target
:
withGroovyBuilder {}
-
The receiver is a GroovyObject and provides Kotlin helpers
-
The
GroovyObject
API is available -
Invoke the
methodName
method, passing some parameters -
Configure the
blockName
property, maps to aClosure
taking method invocation -
Invoke
another
method taking named arguments, maps to a Groovy named argumentsMap<String, ?>
taking method invocation
The maven-plugin sample demonstrates the use of the withGroovyBuilder()
utility extensions for configuring the uploadArchives
task to deploy to a Maven repository with a custom POM using Gradle’s core Maven Plugin.
Note that the recommended Maven Publish Plugin provides a type-safe and Kotlin-friendly DSL that allows you to easily do the same and more without resorting to withGroovyBuilder()
.
Using a Groovy script
Another option when dealing with problematic plugins that assume a Groovy DSL build script is to configure them in a Groovy DSL build script that is applied from the main Kotlin DSL build script:
native { (3)
dynamic {
groovy as Usual
}
}
-
The Kotlin build script requests and applies the plugin
-
The Kotlin build script applies the Groovy script
-
The Groovy script uses dynamic Groovy to configure plugin
Limitations
-
The Kotlin DSL is known to be slower than the Groovy DSL on first use, for example with clean checkouts or on ephemeral continuous integration agents. Changing something in the buildSrc directory also has an impact as it invalidates build-script caching. The main reason for this is the slower script compilation for Kotlin DSL.
-
In IntelliJ IDEA, you must import your project from the Gradle model in order to get content assist and refactoring support for your Kotlin DSL build scripts.
-
The Kotlin DSL will not support the
model {}
block, which is part of the discontinued Gradle Software Model. However, you can apply model rules from scripts — see the model rules sample for more information. -
We recommend against enabling the incubating configuration on demand feature as it can lead to very hard-to-diagnose problems.
If you run into trouble or discover a suspected bug, please report the issue in the Gradle issue tracker.