Dunamai

Dunamai is a Python 3.5+ library and command line tool for producing dynamic, standards-compliant version strings, derived from tags in your version control system. This facilitates uniquely identifying nightly or per-commit builds in continuous integration and releasing new versions of your software simply by creating a tag.

Dunamai is also available as a GitHub Action.

Features

• Version control system support:
  • Git (2.7.0+ is recommended, but versions as old as 1.8.2.3 will work with some reduced functionality)
  • Mercurial
  • Darcs
  • Subversion
  • Bazaar
  • Fossil
  • Pijul
• Version styles:
  • PEP 440
  • Semantic Versioning
  • Haskell Package Versioning Policy
  • Custom output formats
• Can be used for projects written in any programming language. For Python, this means you do not need a setup.py.

Usage

Installation

pip install dunamai

CLI

# Suppose you are on commit g29045e8, 7 commits after the v0.2.0 tag.

# Auto-detect the version control system and generate a version:
$ dunamai from any
0.2.0.post7.dev0+g29045e8

# Or use an explicit VCS and style:
$ dunamai from git --no-metadata --style semver
0.2.0-post.7

# Custom formats:
$ dunamai from any --format "v{base}+{distance}.{commit}"
v0.2.0+7.g29045e8

# If you'd prefer to frame the version in terms of progress toward the next
# release rather than distance from the latest one, you can bump it:
$ dunamai from any --bump
0.2.1.dev7+g29045e8

# Validation of custom formats:
$ dunamai from any --format "v{base}" --style pep440
Version 'v0.2.0' does not conform to the PEP 440 style

# Validate your own freeform versions:
$ dunamai check 0.01.0 --style semver
Version '0.01.0' does not conform to the Semantic Versioning style

# More info:
$ dunamai --help
$ dunamai from --help
$ dunamai from git --help

Library

from dunamai import Version, Style

# Let's say you're on commit g644252b, which is tagged as v0.1.0.
version = Version.from_git()
assert version.serialize() == "0.1.0"

# Let's say there was a v0.1.0rc5 tag 44 commits ago
# and you have some uncommitted changes.
version = Version.from_any_vcs()
assert version.serialize() == "0.1.0rc5.post44.dev0+g644252b"
assert version.serialize(metadata=False) == "0.1.0rc5.post44.dev0"
assert version.serialize(dirty=True) == "0.1.0rc5.post44.dev0+g644252b.dirty"
assert version.serialize(style=Style.SemVer) == "0.1.0-rc.5.post.44+g644252b"

The serialize() method gives you an opinionated, PEP 440-compliant default that ensures that versions for untagged commits are compatible with Pip's --pre flag. The individual parts of the version are also available for you to use and inspect as you please:

assert version.base == "0.1.0"
assert version.stage == "rc"
assert version.revision == 5
assert version.distance == 44
assert version.commit == "g644252b"
assert version.dirty is True

# Available if the latest tag includes metadata, like v0.1.0+linux:
assert version.tagged_metadata == "linux"

Tips

By default, the "v" prefix on the tag is required, unless you specify a custom tag pattern. You can either write a regular expression:

• Console:

  $ dunamai from any --pattern "(?P<base>\d+\.\d+\.\d+)"
  

• Python:

  from dunamai import Version
  version = Version.from_any_vcs(pattern=r"(?P<base>\d+\.\d+\.\d+)")
  

...or use a named preset:

• Console:

  $ dunamai from any --pattern default-unprefixed
  

• Python:

  from dunamai import Version, Pattern
  version = Version.from_any_vcs(pattern=Pattern.DefaultUnprefixed)
  

You can also keep the default pattern and just specify a prefix. For example, this would match tags like some-package-v1.2.3:

• Console:

  $ dunamai from any --pattern-prefix some-package-
  

• Python:

  from dunamai import Version
  version = Version.from_any_vcs(pattern_prefix="some-package-")
  

VCS archives

Sometimes, you may only have access to an archive of a repository (e.g., a zip file) without the full history. Dunamai can still detect a version in some of these cases:

• For Git, you can configure git archive to produce a file with some metadata for Dunamai.

  Add a .git_archival.json file to the root of your repository with this content:

  {
    "hash-full": "$Format:%H$",
    "hash-short": "$Format:%h$",
    "timestamp": "$Format:%cI$",
    "refs": "$Format:%D$",
    "describe": "$Format:%(describe:tags=true,match=v[0-9]*)$"
  }
  

  Add this line to your .gitattributes file. If you don't already have this file, add it to the root of your repository:

  .git_archival.json  export-subst
  

• For Mercurial, Dunamai will detect and use an .hg_archival.txt file created by hg archive. It will also recognize .hgtags if present.

Custom formats

Here are the available substitutions for custom formats. If you have a tag like v9!0.1.2-beta.3+other, then:

• {base} = 0.1.2
• {stage} = beta
• {revision} = 3
• {distance} is the number of commits since the last
• {commit} is the commit hash (defaults to short form, unless you use --full-commit)
• {dirty} expands to either "dirty" or "clean" if you have uncommitted modified files
• {tagged_metadata} = other
• {epoch} = 9
• {branch} = feature/foo
• {branch_escaped} = featurefoo
• {timestamp} is in the format YYYYmmddHHMMSS as UTC
• {major} = 0
• {minor} = 1
• {patch} = 2

If you specify a substitution, its value will always be included in the output. For conditional formatting, you can do something like this (Bash):

distance=$(dunamai from any --format "{distance}")
if [ "$distance" = "0" ]; then
    dunamai from any --format "v{base}"
else
    dunamai from any --format "v{base}+{distance}.{dirty}"
fi

Comparison to Versioneer

Versioneer is another great library for dynamic versions, but there are some design decisions that prompted the creation of Dunamai as an alternative:

• Versioneer requires a setup.py file to exist, or else versioneer install will fail, rendering it incompatible with non-setuptools-based projects such as those using Poetry or Flit. Dunamai can be used regardless of the project's build system.
• Versioneer has a CLI that generates Python code which needs to be committed into your repository, whereas Dunamai is just a normal importable library with an optional CLI to help statically include your version string.
• Versioneer produces the version as an opaque string, whereas Dunamai provides a Version class with discrete parts that can then be inspected and serialized separately.
• Versioneer provides customizability through a config file, whereas Dunamai aims to offer customizability through its library API and CLI for both scripting support and use in other libraries.

Integration

• Setting a __version__ statically:

  $ echo "__version__ = '$(dunamai from any)'" > your_library/_version.py
  

  # your_library/__init__.py
  from your_library._version import __version__
  

  Or dynamically (but Dunamai becomes a runtime dependency):

  # your_library/__init__.py
  import dunamai as _dunamai
  __version__ = _dunamai.get_version("your-library", third_choice=_dunamai.Version.from_any_vcs).serialize()
  

• setup.py (no install-time dependency on Dunamai as long as you use wheels):

  from setuptools import setup
  from dunamai import Version
  
  setup(
      name="your-library",
      version=Version.from_any_vcs().serialize(),
  )
  

  Or you could use a static inclusion approach as in the prior example.

• Poetry:

  $ poetry version $(dunamai from any)
  

  Or you can use the poetry-dynamic-versioning plugin.

Other notes

• Dunamai needs access to the full version history to find tags and compute distance. Be careful if your CI system does a shallow clone by default.

  • For GitHub workflows, invoke actions/checkout@v3 with fetch-depth: 0.
  • For GitLab pipelines, set the GIT_DEPTH variable to 0.
  • For Docker builds, copy the VCS history (e.g., .git folder) into the container.

  For Git, you can also avoid doing a full clone by specifying a remote branch for tags (e.g., --tag-branch remotes/origin/master).

• When using Git, remember that lightweight tags do not store their creation time. Therefore, if a commit has multiple lightweight tags, we cannot reliably determine which one should be considered the newest. The solution is to use annotated tags instead.

• When using Git, the initial commit must not be both tagged and empty (i.e., created with --allow-empty). This is related to a reporting issue in Git. For more info, click here.