Long-term support is always a balance between spending time developing new exciting features and supporting older Julia versions to allow users to more easily keep up. In the v1.0 LTS era and the Julia v0.x times, SciML has explicitly not supported the Julia LTS for multiple reasons, mostly because the later Julia versions included lots of bug fixes specifically designed for the SciML repositories (making Julia v1.0 LTS tenuous to use even until its last day!).
However, with Julia v1.6 becoming the new LTS, a version which all libraries fully support, SciML is adopting the position of supporting the LTS. This means that the following strategies will be employed:
- All packages will be tested on the current Julia release (currently v1.7) and LTS
- All packages will be given a minimum version of the LTS
- No Julia versions before the LTS will be supported. Feel free to ask questions but issues and bug reports from pre-v1.6 Julia versions which cannot be reproduced on Julia LTS will be closed.
- All benchmarks, tutorials, etc. will continue to be generated on the current Julia release (currently v1.7)
- Version compatibility handling will ensure that the LTS-compatible version continues to have all functionality and pass all tests. However, that does not mean that the LTS is recommended. In many cases, version-based branching may be required for compatibility but disable performance, compile-time, or stability enhancements. Thus we recommend all users use the current Julia release unless the LTS is specifically required for support reasons.
Lastly, we encourage other organizations to adopt a similar strategy.
One last note, please do not mix up the concepts of reproducibility and long-term support in issues discussing this topic. Reproducibility is for recreating a known analysis. For personal projects, we recommend simply using a Manifest.toml for reproducibility which will work on any version of Julia beyond v1.0. Long-term support is for continually updating/changing projects (usually managed by large organizations or teams) which do not have the manpower for version updates due to labor cost, which can justify the missing performance etc. to compensate. These are two different concepts which entirely different tooling.