XGBoost Release Policy
Starting from XGBoost 1.0.0, each XGBoost release will be versioned as [MAJOR].[FEATURE].[MAINTENANCE]
MAJOR: We guarantee the API compatibility across releases with the same major version number. We expect to have a 1+ years development period for a new MAJOR release version.
FEATURE: We ship new features, improvements and bug fixes through feature releases. The cycle length of a feature is decided by the size of feature roadmap. The roadmap is decided right after the previous release.
MAINTENANCE: Maintenance version only contains bug fixes. This type of release only occurs when we found significant correctness and/or performance bugs and barrier for users to upgrade to a new version of XGBoost smoothly.
Making a Release
Create an issue for the release, noting the estimated date and expected features or major fixes, pin that issue.
Bump release version.
CMakeLists.txtin source tree and
cmake/Python_version.inif needed, run CMake.
Commit the change, create a PR on GitHub on release branch. Port the bumped version to default branch, optionally with the postfix
Create a tag on release branch, either on GitHub or locally.
Make a release on GitHub tag page, which might be done with previous step if the tag is created on GitHub.
Submit pip, CRAN, and Maven packages.
The pip package is maintained by Hyunsu Cho and Jiaming Yuan. There’s a helper script for downloading pre-built wheels and R packages
xgboost/dev/release-pypi-r.pyalong with simple instructions for using
The CRAN package is maintained by Tong He and Jiaming Yuan.
The Maven package is maintained by Nan Zhu and Hyunsu Cho.
R CRAN Package
Before submitting a release, one should test the package on R-hub and win-builder first. Please note that the R-hub Windows instance doesn’t have the exact same environment as the one hosted on win-builder.
According to the CRAN policy:
If running a package uses multiple threads/cores it must never use more than two simultaneously: the check farm is a shared resource and will typically be running many checks simultaneously.
We need to check the number of CPUs used in examples. Export
_R_CHECK_EXAMPLE_TIMING_CPU_TO_ELAPSED_THRESHOLD_=2.5 before running
R CMD check --as-cran  and make sure the machine you are using has enough CPU cores to reveal any potential policy violation.