radiasoft / chronver Goto Github PK

ChronVer is a setuptools.meta_build plugin to add a chronological version (yyyymmdd.hhmmss) to a Python package's metadata from the current Git commit of HEAD. This ensures a unique, sortable version for packages on PyPI.

In development, if there are modifications, the current time will be returned. This ensures that pip install works, because there's always a new, increasing version.

Add this to your pyproject.toml:

[build-system]
requires = ["setuptools>=61", "chronver"]
build-backend = "setuptools.build_meta"

[project]
dynamic = ["version"]

If you are using Python 3.8 or above, include this code in your <your_package>.__init__.py:

import importlib.metadata

try:
    __version__ = importlib.metadata.version("your_package")
except importlib.metadata.PackageNotFoundError:
    # A package only has a version once it has been built or installed.
    # By not defining __version__, clients will fail fast if they
    # require __version__ (unlikely).
    pass

If you are using a Python before 3.8, use the backport importlib_metadata.

RadiaSoft uses chronological versioning for all its releases. This means no major releases, which implies breaking changes. To make this happen, we strive to maintain backward compatibility for as long as our users require it.

This is why there are no major or minor release numbers in chronological versions; there are no breaking API changes (except defects, of course) with each release. We feel it is our obligation to not break our client's code.

There are times when this is impossible or impractical, such as the move away from Python 2. In those cases, we give clients warnings and an upgrade path and sometimes a tool that upgrades their data or code automatically. This is an obligation that comes with chronological versioning that RadiaSoft takes seriously.

ChronVer is a new implementation of our existing setup.py versioning module so it is not new at all. RadiaSoft has been using chronological versioning for Python, Docker images, packaging (RPMs), etc. for over a decade. Before that @robnagler (primary author) was using chronlogical versioning since at least 2000. So there's a long history here, and it has worked well for us.

SemVer makes many assumptions about how software works that conflicts with how we develop. In our model, software flows like time, and releases are arbitary points on the time axis. SemVer divides software releases into fixed epochs: major, minor, patch, and additional labels. This requires people make complicated decisions about these different kinds of epochs, which adds work and we think adds little value. This can result in Major Release Syndrome, which is an impediment to continuous software development and deployment.

CalVer has a great tag line: "Versioning gets better with time." We agree! In CalVer, "time" means "date", that is, days are the time quantum. This granularity does not allow for same day releases, which we happen to do on occasion. A second is ChronVer's time quantum for this reason. CalVer tries to be SemVer compliant, which as noted above, is not how we develop software.

There are benefits to simplifying version strings to always being lexicographically sortable. Any two ChronVers can be compared with a single operator v1 < v2. We have a lot internal devops tooling in multiple languages, and all languages support simple lexicographic comparison of strings. A ChronVer is compatible with Python's float, which is useful in some cases.

A ChronVer identifies any commit in any SCM. We use Git now. We used to use CVS (which does not have a SHA), and ChronVer worked just as well. All SCM's (that we know of) answer the question: what was the commit at this timestamp? SemVer and CalVer have various schemes for doing this, but they are awkward and break lexicographic comparison.

For more complete reasoning, refer to Major Release Syndrome.