Skip to main content

ADR-0029: One module, and the Apache licence

  • Status: Accepted
  • Date: 2026-09-22
  • Settles: decision D1 of feature-parity-slice-plan.md, which had held two separate questions together since the plan was written: whether to split the codebase the way the earlier project did, and what licence to publish under.

Context​

The earlier project shipped two Maven modules — release-notes-open-core under AGPL-3.0-only and release-notes-enterprise under a proprietary licence — and put its template, automation and distribution capabilities in the second one. ReleaseFlow has been one module since the bootstrap slice, and D1 recorded that as provisional: "parity of function, not of structure; if it ever needs to be commercialised, do it in S24".

Meanwhile the repository has been public with no LICENSE file at all since its first commit. That is not a neutral state. With no licence, the default is all rights reserved: nobody may copy, modify or run it, and a reader cannot tell whether that is intended or an oversight. For a public project meant to be read, that is a defect, not a deferral.

Decision​

  • One Maven module, and the split is abandoned rather than deferred. The capabilities the earlier project sold separately — audiences and templates, automation, the public changelog — sit beside every other capability, packaged by product capability as AGENTS.md requires. There is no commercial plan that a compile-time boundary would serve, and a boundary maintained for a plan nobody has is a cost with no payer. This moves out of "deliberately deferred", which implies a decision still to come, and into a decision taken.
  • Apache License 2.0, for the whole repository. The file is the official text from apache.org, with the appendix filled in; nothing is paraphrased.
  • The licence and the notice travel with the artifact. They are copied into META-INF/ of the JAR, so somebody who has only the binary has both.
  • SECURITY.md says how to report a weakness privately, because the project holds webhook secrets, provider tokens and action credentials, and answers unauthenticated webhooks. It also says what this project is — one person, no rota, no bounty — so a reporter knows what to expect rather than guessing.

Why Apache-2.0 rather than AGPL-3.0-only​

  • The dependency tree does not force copyleft. It is Apache-2.0 and MIT almost throughout. Every artifact that lists a GPL-family licence offers another: Logback under EPL-2.0 or LGPL-2.1, the Jakarta APIs under EPL-2.0 or GPL-2.0 with the Classpath Exception, JNA and Javassist under Apache-2.0 among others. Nothing in the tree restricts what this project may be licensed as, so the choice is genuinely free and has to be made on its merits.
  • It is the licence of this ecosystem. Spring Boot, Spring Security, Flyway, Micrometer and Testcontainers are all Apache-2.0. Matching them is the answer a reader expects and does not have to think about.
  • It grants patent rights explicitly (§3), which MIT does not. For a public project under one person's name, saying so is better than leaving it to be inferred.
  • AGPL's protection is against a use nobody plans to make. Its network clause matters when somebody would run your work as a service without sharing changes back. There is no commercial interest here to protect, and the clause has a real cost: many organisations will not let their staff read, fork or reuse AGPL code without asking a lawyer first, which is exactly the audience a public project like this is written for.

Consequences​

  • Anybody may use, modify and redistribute ReleaseFlow, including commercially, provided they keep the licence and the notice and state what they changed.
  • Relicensing later remains possible without asking anybody: every commit is by one author, so there is no contributor whose agreement would be needed. Going from Apache-2.0 to a copyleft licence later is allowed; it would not retroactively cover copies already taken.
  • S24 is now only release automation. The module split is no longer part of it.
  • Source files carry no per-file licence header. LICENSE and NOTICE at the root, and inside the artifact, are what Apache-2.0 asks for; a header in every file would be noise in a codebase with one author and one licence.