Releases
Releases are automatic. When a push to main changes datasets/, the Release workflow runs scripts/release.jl, which:
- checks the datasets that have changed since the last release
- packs each one whose data have changed, uploads it to a GitHub release, and points
Artifacts.tomlat it - sets the new version in
Project.toml, and commits and pushes both files - registers the new version, from a commit on the branch
release - rebuilds this documentation
Once the registry has merged the new version, TagBot tags it, and its documentation is built as the stable version.
Each dataset is downloaded whole, raw data included, the first time it is used.
Versions
Each version of the package points at a fixed set of data, so results stay reproducible. Depend on the package as NMRExampleData = "1" to get new data with ordinary updates.
The version goes up by:
- major, when a dataset is removed, or renamed without keeping its old name as an alias
- minor, when a dataset is added or its data change
- patch, when only
dataset.tomlfiles change
Until a version is registered, later releases keep its number and update the registration that is waiting.
The release branch
The registered package holds no datasets. The registry checks each package by archiving it with git archive, which expands Git LFS files, so a commit holding datasets/ fails. The commit registered is main without datasets/, with Datasets.toml in its place: an index of every dataset.toml, which the package reads when there is no datasets/ folder.
Work on main as usual, and leave release to the workflow. Never ask Registrator to register a commit on main.
Release a change to the code
Changes to the code alone are not released automatically. To release one, raise the version in Project.toml on main, then run the Release workflow from the Actions tab.
If a release fails
Fix the problem and push again, or run the workflow again from the Actions tab. To see what would be released, run:
julia scripts/release.jl --dry-run