🚀 AxonFramework-Full — A Deliberate Maven→Gradle Migration

Last updated: September 19, 2026

📜 View the portfolio page →

A fork of Axon Framework 5 whose entire purpose is migrating its build system from Maven to Gradle — one file at a time, each one individually inspected against the Maven source and given its own conversion decision, not run through an automated converter or bulk-copied and edited in place. Not affiliated with the Axon Framework team or AxonIQ — a technical portfolio artifact only.

• todo.md — the living status doc: what's done, what's left, every non-obvious decision and why
• GitHub Project board — the same status, as a board
• Wiki — short pointer pages into the real detail below
• docs/diagrams/ — the migration pipeline, module graph, and build-logic architecture, visually: migration pipeline, module dependency graph, build-logic conventions, publishing pipeline
• portfolio.html — this repo as a portfolio piece: real gaps found, real corrections made, and why, for anyone scanning it rather than reading it as documentation
• Portfolio hub · GitHub profile — this fork in the context of this user's other work

Anyone can point a conversion tool at a Maven reactor and commit whatever comes out. The interesting engineering work is everywhere that approach breaks down: a Gradle convention plugin doesn't have Maven's parent-POM inheritance, so a shared dependency block that works fine in axon-parent creates a literal circular dependency the moment it's applied to the one module that dependency points at (test-logging). A checkstyle rule copied byte-for-byte looks complete until you notice it's missing half a symmetric pair on inspection, not because a script flagged it. Sonatype's newest publishing API has no official Gradle plugin at all, so the community one you pick has to be verified against its live docs, not assumed from training data — and even then it can still hit a real Gradle-version interaction bug nobody's documented yet.

This repo is a record of making — and explaining — that kind of call, one file at a time, with the reasoning kept next to the decision instead of buried in a commit message nobody reads later.

Axon Framework is a Java framework for building applications on CQRS and event-sourcing principles — command/query separation, an event store as the system of record, and the messaging infrastructure to wire it all together. This fork doesn't change any of that. Three modules' real Java source are fully converted so far — test-logging, common (144 of 144 files), and update (27 of 27 files) — the rest of the reactor's Java/Kotlin source still needs converting. What exists alongside that is the build system every module sits inside: the Gradle project structure, the shared conventions every module applies, and the publishing pipeline that will ship them to Maven Central.

A multi-module Gradle build, structurally mirroring the Maven reactor's module list (see settings.gradle.kts), with the cross-module conventions Maven expressed as parent-POM inheritance now expressed as Gradle convention plugins in build-logic/:

• axonframework.java-conventions — everything axon-parent applies to every module: Java 25 toolchain, checkstyle, the standard test stack, jar manifest
• axonframework.published-conventions — layers on maven-publish, sources/javadoc jars, POM metadata, and signing for modules that ship to Maven Central
• axonframework.internal-conventions — the explicit non-published counterpart, for modules like docs/_samples that never should

See docs/diagrams/ for how these fit together, and the wiki's Divergences From Upstream page for every place this fork deliberately doesn't match the Maven source's behavior.

• todo.md — done/still-to-do at a glance, plus the reasoning behind every non-obvious call
• GitHub Project board — Backlog → Planned → In Progress → Verification & QA → Done
• Milestones — completed phases, each closed once everything in it landed
• Wiki — one page per subsystem, each a pointer to the authoritative source rather than a duplicate of it

On AI-assisted development: this project uses Claude Code for AI-assisted implementation — every change is directed, reviewed, and merged by Terrence Daniels, with no commit co-author trailers.