Elevator pitch: you can write a setup.py with no version information
specified, and vcversioner will find a recent, properly-formatted VCS tag and
extract a version from it.
It's much more convenient to be able to use your version control system's
tagging mechanism to derive a version number than to have to duplicate that
information all over the place. I eventually ended up copy-pasting the same
code into a couple different setup.py files just to avoid duplicating
version information. But, copy-pasting is dumb and unit testing setup.py
files is hard. This code got factored out into vcversioner.
vcversioner installs itself as a setuptools hook, which makes its use exceedingly simple:
from setuptools import setup
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={},
)
The presence of a vcversioner argument automagically activates vcversioner
and updates the project's version. The parameter to the vcversioner
argument can also be a dict of keyword arguments which find_version
will be called with.
To allow tarballs to be distributed without requiring a .git (or .hg,
etc.) directory, vcversioner will also write out a file named (by default)
version.txt. Then, if there is no VCS program or the program is unable to
find any version information, vcversioner will read version information from
the version.txt file. However, this file needs to be included in a
distributed tarball, so the following line should be added to MANIFEST.in:
include version.txt
This isn't necessary if setup.py will always be run from a checkout, but
otherwise is essential for vcversioner to know what version to use.
The name version.txt also can be changed by specifying the version_file
parameter. For example:
from setuptools import setup
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={
'version_file': 'custom_version.txt',
},
)
For compatibility with semantic versioning, vcversioner will strip
leading 'v's from version tags. That is, the tag v1.0 will be
treated as if it was 1.0.
Other prefixes can be specified to be stripped by using the strip_prefix
argument to vcversioner. For compatibility with git-dch, one could specify
the strip_prefix as 'debian/'.
It's not necessary to depend on vcversioner; while pip will take care of
dependencies automatically, sometimes having a self-contained project is
simpler. vcversioner is a single file which is easy to add to a project. Simply
copy the entire vcversioner.py file adjacent to the existing setup.py
file and update the usage slightly:
from setuptools import setup
import vcversioner
setup(
# [...]
version=vcversioner.find_version().version,
)
This is necessary because the vcversioner distutils hook won't be
available.
setup.py isn't the only place that version information gets duplicated. By
generating a version module, the __init__.py file of a package can import
version information. For example, with a package named spam:
from setuptools import setup
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={
'version_module_paths': ['spam/_version.py'],
},
)
This will generate a spam/_version.py file that defines __version__ and
__revision__. Then, in spam/__init__.py:
from spam._version import __version__, __revision__
Since this acts like (and is) a regular python module, changing
MANIFEST.in is not required.
vcversioner by default tries to detect which VCS is being used and picks a
command to run based on that. For git, that is git --git-dir %(root)s/.git
describe --tags --long. For hg, that is hg log -R %(root)s -r . --template
'{latesttag}-{latesttagdistance}-hg{node|short}'.
Any command should output a string that describes the current commit in the
format 1.0-0-gdeadbeef. Specifically, that is <version number>-<number of
commits between the current commit and the version tagged commit>-<revision>.
The revision should have a VCS-specific prefix, e.g. g for git and hg
for hg.
However, sometimes this isn't sufficient. If someone wanted to only use annotated tags, the git command could be amended like so:
from setuptools import setup
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={
'vcs_args': ['git', 'describe', '--long'],
},
)
The vcs_args parameter must always be a list of strings, which will not be
interpreted by the shell. This is the same as what subprocess.Popen
expects.
This argument used to be spelled git_args until support for multiple VCS
systems was added.
vcversioner can also automatically make a version that corresponds to a commit
that isn't itself tagged. Following PEP 386, this is done by adding a
.post suffix to the version specified by a tag on an earlier commit. For
example, if the current commit is three revisions past the 1.0 tag, the
computed version will be 1.0.post3.
This behavior can be disabled by setting the include_dev_version parameter
to False. In that case, the aforementioned untagged commit's version would
be just 1.0.
Since hg requires a commit to make a tag, there's a parameter
decrement_dev_version to subtract one from the number of commits after the
most recent tag. If the VCS used is detected to be hg (i.e. the revision starts
with 'hg') and decrement_dev_version is not specified as False,
decrement_dev_version will be automatically set to True.
In order to prevent contamination from other source repositories, vcversioner
in the 1.x version series will only look in the project root directory for
repositories. The project root defaults to the current working directory, which
is often the case when running setup.py. This can be changed by specifying the
root parameter. Someone concerned with being able to run setup.py from
directories other than the directory containing setup.py should determine the
project root from __file__ in setup.py:
from setuptools import setup
import os
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={
'root': os.path.dirname(os.path.abspath(__file__)),
},
)
To get the same behavior in the 0.x version series, vcs_args can be set to
include the --git-dir flag:
from setuptools import setup
setup(
# [...]
setup_requires=['vcversioner'],
vcversioner={
vcs_args=['git', '--git-dir', '%(root)s/.git', 'describe',
'--tags', '--long'],
},
)
By default, version.txt is also read from the project root.
As seen above, root, version_file, and vcs_args each support some substitutions:
%(root)s- The value provided for root. This is not available for the root parameter itself.
%(pwd)s- The current working directory.
/ will automatically be translated into the correct path separator for the
current platform, such as : or \.
Sphinx documentation is yet another place where version numbers get
duplicated. Fortunately, since sphinx configuration is python code, vcversioner
can be used there too. Assuming vcversioner is installed system-wide, this is
quite easy. Since Sphinx is typically run with the current working directory as
<your project root>/docs, it's necessary to tell vcversioner where the
project root is. Simply change your conf.py to include:
import vcversioner version = release = vcversioner.find_version(root='..').version
This assumes that your project root is the parent directory of the current working directory. A slightly longer version which is a little more robust would be:
import vcversioner, os
version = release = vcversioner.find_version(
root=os.path.dirname(os.path.dirname(os.path.abspath(__file__)))).version
This version is more robust because it finds the project root not relative to
the current working directory but instead relative to the conf.py file.
If vcversioner is bundled with your project instead of relying on it being
installed, you might have to add the following to your conf.py before
import vcversioner:
import sys, os
sys.path.insert(0, os.path.abspath('..'))
This line, or something with the same effect, is sometimes already present when
using the sphinx autodoc extension.
Using vcversioner is even possible when building documentation on Read the
Docs. If vcversioner is bundled with your project, nothing further needs to
be done. Otherwise, you need to tell Read the Docs to install vcversioner
before it builds the documentation. This means using a requirements.txt
file.
If your project is already set up to install dependencies with a
requirements.txt file, add vcversioner to it. Otherwise, create a
requirements.txt file. Assuming your documentation is in a docs
subdirectory of the main project directory, create docs/requirements.txt
containing a vcversioner line.
Then, make the following changes to your project's configuration: (Project configuration is edited at e.g. https://readthedocs.org/dashboard/vcversioner/edit/)
- Check the checkbox under Use virtualenv.
- If there was no
requirements.txtpreviously, set the Requirements file to the newly-created one, e.g.docs/requirements.txt.
