Proposal: Return to Semantic Versioning (3.0 instead of 2026.1.0)

[roxblnfk]

Problem

A few years ago, RoadRunner switched from standard semantic versioning (1.x, 2.x) to a year-quarter scheme (e.g. 2024.1.0, 2025.2.0). This approach introduces several practical issues:

• False breaking changes signal.
  A major version bump (year change) implies breaking changes per SemVer conventions, even when there are none — it happens simply because the calendar year has changed. This makes it harder for users to assess the actual impact of an upgrade.
• Maintenance overhead for the PHP ecosystem.
  Every year, we are forced to update dozens of PHP packages just to add the new major version constraint in composer.json. This is purely mechanical work that brings no value — it's caused by the versioning scheme, not by actual API changes.

Proposal

Release the next version as 3.0.0 instead of 2026.1.0 and return to standard semantic versioning going forward.

Why now

RoadRunner is a feature-complete, mature product at this point. Its core architecture has been stable for a long time and is unlikely to change. Future breaking changes, if any, would most likely come from the plugin side, not from RoadRunner's architecture itself. Standard SemVer fits this stage of the project much better — major version bumps would only happen when there are actual breaking changes, which should be rare.

Additional benefits of a 3.0 release

• Deprecation cleanup. A good opportunity to remove any remaining deprecated APIs or features, if there are any left.
• Config version alignment. Introducing config version 3 alongside the application version 3.0. Keeping the config version in sync with the major application version is convenient for users and simplifies documentation and migration guides. For example, config v3 could include fixes to unintuitive defaults — such as changing pool.num_workers semantics so that 0 means "no workers" (instead of the current CPU-count behavior), -1 means "match CPU count", and -2 means "CPU count × 2".

Summary

The year-quarter versioning scheme creates unnecessary churn in the ecosystem and misleads users about the nature of releases. Returning to SemVer with a 3.0 release aligns the versioning with the project's actual stability and maturity.

[rustatian]

Hey @roxblnfk 👋🏻
Thank you for raising this question.

I agree about maintenance overhead.
Configuration remains untouched, I mean, pool.num_workers is a well-known behavior for many systems, not only RoadRunner.

Next version will be 3.0. Thank you ❤️

[rustatian]

This release (v3.0) would also be the first LTS release with only bug, security, and other fixes without adding new plugins or breaking even the smallest API (even internal). Only the bugfix version for this release would be updated. Like 3.0.1, 3.0.2, and so on. I'm still thinking about the support range for this release, but I guess that it'd be at least 2 years.

[rustatian]

I want to inform everyone that v3.0 will be my last release as a project lead maintainer. I'll do my best to polish that release as much as possible, but after it's released, I'll be stepping down from my role and leaving the RoadRunner organization. I'm deeply grateful to the community, Anton, Alex, Kirill, Pavel, and everyone involved in this project (currently and in the past), thank you very much, guys. And definitely, I'd like to express my sincerest gratitude to our community; it was a pleasure to work with all of you.
I'm preparing many new features, and improvements with Alex in v3.0, stay tuned ❤️

Valery.

[FluffyDiscord]

Eh. Should I worry about the future of RR?

Who will maintain this in the future, especially some plugins, which aren't imho feature complete (like http streaming, temporal, they could be so much more)

I am about to create another big project, that I would love to use RR once again instead of Franken (franken is still too unstable and unpredictable for me), but this message shocks me a bit. Don't want to end up with abandoned backbone of my project in two years :(

And by the extension I also assume Rapira is also dead, as PoC?

[rustatian]

RoadRunner, as was declared by SpiralScout, is a feature complete product. You should not worry about the future of it, since the v3.0 update even brings the ability to have a distributed pool of workers (not sure that feature would be used, because this is a side-effect of the async pool we're working on).

Just to clarify: I'm not leaving SpiralScout, but I won't be spending my spare time anymore on RoadRunner (sorry), since this is not my product, and I can't promote it, manage it, and so on how I want to do that. From the last announcement about feature completion (I guess 2 years ago), I was working on that product just to, you know, support all of you guys, in my free time. Since 2022, I've been more interested in AI (to say in truth, since 2014, when Google started TensorFlow (v0.12)) and digging into that field. Moreover, I don't know PHP, that is a big problem for that product.

So, calm down, everything will be ok, SpiralScout is full of clever guys 😃