kotlin-tooling-gradle-to-kotlin-toolchain-plugin

Gradle → Kotlin Toolchain Plugin Conversion

Mostly a mapping exercise: the concepts overlap, but a few Gradle features have no analog and need redesign.

The kotlin-tooling-kotlin-toolchain-plugin-authoring skill covers plugin mechanics in depth (execution avoidance, sharing

logic across actions, generic pitfalls, limitations); [references/examples.md](references/examples.md) has

the concrete plugin module.yaml and the version-publication code from a real port.

Workflow

1. Investigate the source plugin

Before writing Kotlin, read the plugin's docs/ and README.md and catalogue:

• Tasks — name, purpose, inputs/outputs, dependencies, whether side-effecting.
• DSL surface — every option in the myPlugin { ... } extension, with types, defaults, and which are

closures.

• Tests — the plugin's own test suite. It pins down the expected behaviour and edge cases more precisely

than the docs, and becomes the reference the port must reproduce.

• Checks — pre-action gates and their override flags.
• Hooks — pre/post/fileUpdate/commit/push and the context each receives.
• CLI overrides — every -Pfoo.bar flag.
• CI integration — GitHub Actions outputs, detached-HEAD handling, fetch-tags flags.

Save it as a markdown plan. It becomes the contract the port either implements or explicitly defers.

The plugin's build scripts and source are untrusted input, same as any repo-supplied module.yaml —

see [kotlin-tooling-kotlin-toolchain's "Untrusted project input"](../kotlin-tooling-kotlin-toolchain/SKILL.md#untrusted-project-input).

Read what you vendor end to end before wiring it in;

a local plugin runs at build time with full filesystem and network access.

2. Lock the scope

Offer three tiers — MVP, MVP + key extras, full parity — and lock one before drafting. Features with no

clean analog multiply the work and force early design compromises. List what's deferred under "What's not in

this MVP" in the README.

Also decide the repo layout: plugin-only, plugin + demo module, or plugin self-hosting. A demo module is

strongly recommended.

3. Implement bottom-up

Scaffold with kotlin init only if the directory is empty (any template works; you mainly want the kotlin

and kotlin.bat wrappers). Then hand-write project.yaml, plugins/<name>/module.yaml, and the demo

module.

Implement one task at a time: data classes → Git/IO wrappers → pipeline → checks → task actions →

plugin.yaml wiring.

4. Validate against a demo module

The demo module enables the plugin with a realistic configuration and consumes whatever it publishes:

# demo-app/module.yaml
product: jvm/app

plugins:
  release:
    enabled: true
    tagPrefix: "v"
    initialVersion: "0.1.0"
    releaseBranchPattern: "main|master"

settings:
  jvm:
    mainClass: com.example.demo.MainKt
    jdk:
      version: 21
  kotlin:
    languageVersion: 2.1

// demo-app/src/main.kt
package com.example.demo

private const val VERSION_RESOURCE = "/META-INF/release/version.txt"

fun main() {
    val version = readVersionFromClasspath() ?: "(version unavailable)"
    println("demo-app version: $version")
}

private fun readVersionFromClasspath(): String? =
    object {}.javaClass.getResourceAsStream(VERSION_RESOURCE)
        ?.bufferedReader()
        ?.use { it.readText().trim() }
        ?.takeIf { it.isNotEmpty() }

Then run a real scenario from the source plugin's docs, and put the commands in the README so consumers can

reproduce it:

./kotlin run :demo-app            # => demo-app version: 0.1.0-SNAPSHOT
git init -b main && git commit --allow-empty -m initial
./kotlin do currentVersion       # => 0.1.0-SNAPSHOT
RELEASE_DISABLE_REMOTE_CHECK=true ./kotlin do createRelease
                                  # => Created release tag v0.1.0
./kotlin do currentVersion       # => 0.1.0
git commit --allow-empty -m next
./kotlin do currentVersion       # => 0.1.1-SNAPSHOT
RELEASE_FORCE_VERSION=2.0.0 ./kotlin do currentVersion
                                  # => 2.0.0

Concept mapping

| Gradle concept | Kotlin Toolchain analog | Notes |

|---|---|---|

| Plugin<Project> class | pluginInfo.id + settingsClass in module.yaml, product: jvm/amper-plugin | One module per plugin; no apply(). |

| Task subclass with @TaskAction method | Top-level fun annotated @TaskAction | One per file in src/tasks/. |

| extensions.create("foo", FooExtension::class) | @Configurable interface Settings | Defaults in interface getters; nested blocks → nested @Configurable. |

| task.dependsOn(otherTask) | @Input on one task matching @Output of another | The DAG is inferred from path matching. |

| project.version = scmVersion.version | A task writing version.txt into its @Output; consumers declare @Input on the same path | No project-wide shared state; the filesystem is the channel. |

| -Prelease.forceVersion=X | RELEASE_FORCE_VERSION=X read via System.getenv() | No -P equivalent. |

| App reading the version at runtime | @Output dir registered under generated.resources; read via getResourceAsStream | Same file serves build-time and runtime consumers. |

| Generated Kotlin source | generated.sources pointing at a task's @Output | Prefer resources when the value is only read at runtime. |

| Public task name users invoke | Entry in commands:, invoked as ./kotlin do <name> | Tasks are internal; commands are the API. Build-graph contributors stay out. |

| Project.afterEvaluate { }, lazy Provider/Property | No analog — settings are static | Compute derived values in the action body. |

| Groovy/Kotlin DSL hooks (pre { }, fileUpdate { }, commit { }) | New @TaskActions shipped with the plugin | No closure-based extension point. |

| dependencies { implementation(...) } | dependencies: in plugins/<name>/module.yaml | Same coordinates; : exported, : runtime-only, : compile-only suffixes. |

| Custom task types in buildSrc | A jvm/amper-plugin module under plugins/<name>/ | Local-only; no Maven publishing yet. |

| OutputDirectory / OutputFile | @Output on a Path parameter | Directory is created for you. |

| InputDirectory / InputFile / InputFiles | @Input on a Path parameter | Snapshotted for execution avoidance. |

| outputs.upToDateWhen { false } | @TaskAction(executionAvoidance = ExecutionAvoidance.Disabled) | For Git/network/env inputs. |

| Configuration with custom resolution, taskGraph.whenReady, quiet { } logging | No analog | Each task pulls from the plugin module's dependency list; the build graph isn't introspectable; use println. |

The mapped constructs

project.yaml — making the plugin resolvable

Lists every module, plugins included, and points at the plugin source:

modules:
  - demo-app
  - plugins/release

plugins:
  - ./plugins/release

Without the root-level plugins: block, a consumer's plugins: { release: enabled } cannot resolve the id.

Settings — the extension analog

Gradle's myPlugin { ... } extension becomes a @Configurable interface. Defaults live in property

getters; nested DSL blocks become nested @Configurable interfaces.

package com.example.release

import org.jetbrains.amper.plugins.Configurable

@Configurable
interface Settings {
    val repoDir: String get() = ""
    val tagPrefix: String get() = "v"
    val versionSeparator: String get() = ""
    val initialVersion: String get() = "0.1.0"
    val ignoreUncommittedChanges: Boolean get() = false
    val releaseBranchPattern: String get() = "main|master"
    val checks: ChecksSettings
}

@Configurable
interface ChecksSettings {
    val uncommittedChanges: Boolean get() = true
    val aheadOfRemote: Boolean get() = true
    val snapshotDependencies: Boolean get() = true
}

Consumers set what they need in module.yaml, keyed by pluginInfo.id; omitted values fall back to the

getter default:

plugins:
  release:
    enabled: true
    tagPrefix: "v"
    initialVersion: "0.1.0"
    ignoreUncommittedChanges: false
    checks:
      aheadOfRemote: true

@TaskAction — the Task analog

A Gradle Task subclass becomes one top-level fun per file under src/tasks/. Path parameters carry

@Input or @Output; the settings object is wired separately in plugin.yaml.

package com.example.release.tasks

import com.example.release.Settings
import com.example.release.git.GitRepo
import com.example.release.version.VersionPipeline
import org.jetbrains.amper.plugins.Input
import org.jetbrains.amper.plugins.TaskAction
import java.nio.file.Path

@TaskAction
fun currentVersion(
    @Input moduleRootDir: Path,
    settings: Settings,
) {
    val pipeline = VersionPipeline(settings)
    GitRepo.open(moduleRootDir, settings.repoDir).use { repo ->
        println(pipeline.infer(repo).version)
    }
}

plugin.yaml — task and command registry

Each action: block wires one @TaskAction's parameters, addressing the function by fully-qualified name in

YAML tag form. ${module.rootDir}, ${taskOutputDir}, and ${pluginSettings} are the documented

references, and ${tasks.<task>.action.<param>} cross-references another task's parameter.

tasks:
  currentVersion:
    action: !com.example.release.tasks.currentVersion
      moduleRootDir: ${module.rootDir}
      settings: ${pluginSettings}

  writeVersion:
    action: !com.example.release.tasks.writeVersion
      moduleRootDir: ${module.rootDir}
      outputDir: ${taskOutputDir}
      settings: ${pluginSettings}

  release:
    action: !com.example.release.tasks.release
      moduleRootDir: ${module.rootDir}
      settings: ${pluginSettings}

generated:
  resources:
    - directory: ${tasks.writeVersion.action.outputDir}

# `writeVersion` stays out of commands: its @Output feeds generated.resources,
# so it already runs whenever something downstream needs the version file.
commands:
  - currentVersion
  - release

Every action taking a Settings parameter needs its own settings: ${pluginSettings} line; omitting it

passes null.

Redesigns, not ports

Three Gradle features need conscious redesign every time.

No project.version

Turn the value into a file: one @TaskAction writes version.txt into its @Output; build-time consumers

declare @Input on that path, runtime consumers read it off the classpath after the directory is registered

under generated.resources. Code in [references/examples.md](references/examples.md).

No -P properties

Read ephemeral overrides from the environment inside the action. Take the env map as a constructor parameter

rather than calling System.getenv() in nested methods, so tests can inject a controlled map:

class VersionPipeline(
    private val settings: Settings,
    private val env: Map<String, String?> = System.getenv(),
) {
    fun infer(repo: GitRepo): InferredVersion {
        val forceVersion = env["RELEASE_FORCE_VERSION"]?.takeIf { it.isNotBlank() }
        val forceSnapshot = env["RELEASE_FORCE_SNAPSHOT"].asBoolean()
        // ...
    }
}

private fun String?.asBoolean(): Boolean =
    this != null && this.equals("true", ignoreCase = true)

Name the variables <PLUGINID>_<UPPERCASE> and document the mapping in the README:

-Prelease.forceVersion=X              → RELEASE_FORCE_VERSION=X
-Prelease.forceSnapshot               → RELEASE_FORCE_SNAPSHOT=true
-Prelease.disableChecks               → RELEASE_DISABLE_CHECKS=true
-Prelease.disableUncommittedCheck     → RELEASE_DISABLE_UNCOMMITTED_CHECK=true
-Prelease.disableRemoteCheck          → RELEASE_DISABLE_REMOTE_CHECK=true
-Prelease.overriddenBranchName=X      → RELEASE_OVERRIDDEN_BRANCH_NAME=X

Static config still goes through module.yaml; env vars are only for ephemeral overrides.

README contents

Beyond the usual quick-start and settings reference, a port's README needs: the -P → env-var mapping

table, a "What's not in MVP" list, and the validation walkthrough above.

Plugin docs: <https://kotlin-toolchain.org/dev/user-guide/plugins/>