PyCalVer: Automatic CalVer Versioning for Python Packages

PyCalVer is a simple calendar based versioning system. With a single pycalver bump command it will:

• Automatically update version strings across files in your project.
• Commit those changes and tag the commit with the new version.

Version strings generated by pycalver are compatible with python packaging software setuptools PEP440.

Project/Repo:

Code Quality/CI:

• Introduction
  • Format
  • Versioning Behaviour
  • Lexical Ids
• Usage
  • Configuration
  • Bump It Up
  • Pattern Search and Replacement
• Rational
  • Other Versioning Software
  • Some Details
  • Realities of Version Numbers
  • Should I use PyCalVer for my Project?
  • Marketing/Vanity
  • Rational
  • Breaking Things is a Big Deal
  • A Word on Marketing
  • Commitment to Compatibility
  • The Life of a Library
  • FAQ

Introduction

The PyCalVer package provides the pycalver command to generate version strings. The version strings have three parts:

    o Year and Month of Build
    |       o Sequential Build Number
    |       |      o Release Tag (optional)
    |       |      |
 ---+---  --+--  --+--
 v201812  .0123  -beta

Some examples:

v201711.0001-alpha
v201712.0027-beta
v201801.0031
v201801.0032-post
...
v202207.18133
v202207.18134

Version String Format

The format for PyCalVer version strings can be parsed with this regular expression:

import re

# https://regex101.com/r/fnj60p/10
PYCALVER_PATTERN = r"""
\b
(?P<version>
    (?P<calver>
       v                        # "v" version prefix
       (?P<year>\d{4})
       (?P<month>\d{2})
    )
    (?P<build>
        \.                      # "." build nr prefix
        \d{4,}
    )
    (?P<release>
        \-                      # "-" release prefix
        (?:alpha|beta|dev|rc|post)
    )?
)(?:\s|$)
"""
PYCALVER_RE = re.compile(PYCALVER_PATTERN, flags=re.VERBOSE)

version_str = "v201712.0001-alpha"
version_info = PYCALVER_RE.match(version_str).groupdict()

assert version_info == {
    "version" : "v201712.0001-alpha",
    "calver"  : "v201712",
    "year"    : "2017",
    "month"   : "12",
    "build"   : ".0001",
    "release" : "-alpha",
}

version_str = "v201712.0033"
version_info = PYCALVER_RE.match(version_str).groupdict()

assert version_info == {
    "version" : "v201712.0033",
    "calver"  : "v201712",
    "year"    : "2017",
    "month"   : "12",
    "build"   : ".0033",
    "release" : None,
}

Versioning Behaviour

To see how version strings are incremented, we can use pycalver incr:

$ pip install pycalver
...
$ pycalver incr v201801.0033-beta
PyCalVer Version: v201809.0034-beta
PEP440 Version: 201809.34b0

This is the simple case:

• The calendar component is update to the current year and month.
• The build number is incremented by 1.
• The optional release tag is preserved as is.

You can explicitly update the release tag using the --release=<tag> argument:

$ pycalver incr v201801.0033-alpha --release=beta
PyCalVer Version: v201809.0034-beta
PEP440 Version: 201809.34b0
$ pycalver incr v201809.0034-beta --release=final
PyCalVer Version: v201809.0035
PEP440 Version: 201809.35

The version number is padded with extra zeros, to maintain the lexical ordering of version numbers. What happens when the padding is exhausted?

$ pycalver incr v201809.0999
PyCalVer Version: v201809.11000
PEP440 Version: 201809.11000

This is because the build number is generated as a sequence of lexical ids.

Lexical Ids

The build number padding may eventually be exhausted. In order to preserve both lexical ordering, build numbers are incremented in a special way. Examples will perhaps illustrate more clearly.

"0001"
"0002"
"0003"
...
"0999"
"11000"
"11001"
...
"19998"
"19999"
"220000"
"220001"

What is happening here is that the left-most digit is incremented early/preemptively. Whenever the left-most digit would change, the padding of the id is expanded using this simple formula:

prev_id = "0999"
next_id = str(int(prev_id, 10) + 1)           # "1000"
if prev_id[0] != next_id[0]:                  # "0" != "1"
    next_id = str(int(next_id, 10) * 11)      # 1000 * 11 = 11000

This behaviour ensures that the following semantic is always preserved: old_version < new_version. This will be the case, even if the padding was expanded and the version number was incremented multiple times in the same month. To illustrate the issue, imagine we did not expand the padding and instead just incremented numerically.

"0001"
"0002"
"0003"
...
"0999"
"1000"
...
"9999"
"10000"

Here we eventually run into a build number where the lexical ordering is not preserved, since "9999" < "10000" == False. This is a very rare corner case, but it's better to not have to think about it.

Just as an example of why lexical ordering is a nice property to have, there are lots of software which read git tags, but which have no logic to parse version strings, which can nonetheless order the version tags correctly.

Usage

Configuration

The fastest way to setup a project is to use pycalver init.

$ cd my-project
~/my-project$ pycalver init
Updated setup.cfg

This will add the something like the following to your setup.cfg (depending on what files you have in your project):

[pycalver]
commit = True
tag = True
push = True

[pycalver:file_patterns]
setup.cfg =
    current_version = {version}
setup.py =
    "{version}",
    "{pep440_version}",
README.md =
    {version}
    {pep440_version}

This may or may not cover all version numbers across your repository. Something like the following may illustrate additional changes you might need to make.

[pycalver]
current_version = v201809.0001-beta
commit = True
tag = True
push = True

[pycalver:file_patterns]
setup.cfg =
    current_version = {version}
setup.py =
    version="{pep440_version}"
src/myproject.py =
    __version__ = "{version}"
README.md =
    [PyCalVer {calver}{build}-{release}]
    img.shields.io/badge/PyCalVer-{calver}{build}--{release}-blue

Bump It Up

To increment and publish a new version, you can use the pycalver bump command:

$ pycalver bump

This will do a few things

0. Check that you don't have any non-committed local changes.
1. Fetch the most recent global vcs tags from origin.
2. Generate a new version, incremented from on the most recent tag on any branch.
3. Update version strings in all configured files.
4. Commit the updated version strings.
5. Tag the new commit.
6. Push the new commit to origin.

The current version is defined either as

• The lexically largest git/mercurial tag in the repository.
• The value of pycalver.current_version in setup.cfg

The git/mercurial tags are used to minimize the chance that the same version will be generated for different revisions. As part of doing pycalver bump, your local repository is updated using git fetch --tags/hg pull, to ensure that all tags are known locally git push --follow-tags

the version To avoid this completely you n The value in setup.cfg is only used on projects that don't use revision control. If your project does not use git or mercurial for version control, then the current_version will be set by pycalver init.

$ pycalver show
Current Version: v201809.0001-beta
PEP440 Version: 201809.1b0

$ pycalver bump --dry
TODO: diff output
Don't forget to do $ git push --tags

TODO: commits and tags

Pattern Search and Replacement

patterns is used both to search for version strings and to generate the replacement strings. The following placeholders are available for use, everything else in a pattern is treated as literal text.

Note that the separator/prefix characters are part of what is matched and generated for a given placeholder, and they should not be included in your patterns.

A further restriction is, that a version string cannot span multiple lines in your source file.

Now we can call pycalver bump to bump all occurrences of version strings in these files. Normally this will change local files, but the --dry flag will instead display a diff of the changes that would be applied.

Rational

Other Versioning Software

This project is very similar to bumpversion, upon which it is partially based. So why another library? PyCalVer version strings can be generated automatically, usage is a bit more simple.

Some Details

• Version numbers are for public releases. For the purposes of development of the project itself, reference VCS branches and commit ids are more appropriate.
• There should be only one person or system responsible for updating the version number at the time of release, otherwise the same version number may be generated for different builds.
• Lexical order is

Canonical PyCalVer version strings can be parsed with this regular expression:

These are the full version strings, for public announcements and conversations it will often be sufficient to refer simply to v201801, by which the most recent post release build of that month is meant.

version_str = "v201712.0027-beta"
version_dict = pycalver_re.match("v201712.0027-beta").groupdict()
import pkg_resources    # from setuptools
version = pkg_resources.parse_version(version_str)
--

In [2]: version_dict
{'year': '2017', 'month': '12', 'build_nr': '0027', 'tag': 'beta'}
>>> str(version)
201712.27b0

Realities of Version Numbers

Nobody knows what the semantics of a version number are, because nobody can guarantee that a given release adheres to whatever convention one would like to imbibe it with. Lets just keep things simple.

• Version numbers should be recognizable as such, that's what the "v" prefix does.
• A number like 201808 is recognizable to many as a number derived from a calendar.
• alpha, beta are common parlance indicating software which is still under development.

Some additional constraints are applied to conform with PEP440

Should I use PyCalVer for my Project?

If your project is 1. not useful by itself, but only when used by other software, 2. has a finite scope/a definition of "done", 3. your project has CI, a test suite with and decent code coverage, then PyCalVer is worth considering. You release at most once per month.

Marketing/Vanity

Quotes from http://sedimental.org/designing_a_version.html

Rational

PyCalVer is opinionated software. This keeps things simple, when the opinion match yours, but makes it useless for everybody else.

The less semantics you put in your version string, the better. The ideal would be to only have a single semantic: newer == better.

Some projects depend recursively on hundreds of libraries, so compatibility issues generated by your project can be a heavy burden on thousands of users; users who learn of the existence of your library for the first time in the form of a stack-trace. PyCalVer is for projects that are committed to and can maintain backward compatibility. Newer versions are always better, updates are always safe, an update won't break things, and if it does, the maintainer's hair is on fire and they will publish a new release containing a fix ASAP.

Ideally, your user can just declare your library as a dependency, without any extra version qualifier, and never have to think about it again. If you do break something by accident, their remedy is not to change their code, but to temporarily pin an earlier version, until your bugfix release is ready.

PyCalVer is for projects which are the mundane but dependable foundations of other big shiny projects, which get to do their big and exciting 2.0 major releases.

Breaking Things is a Big Deal

Using an increment in a version string to express that a release may break client code is not tenable. A developer cannot be expected to think about how their code may or may not break as a consequence of your decision to rename some functions. As the author of any software, there is a great temptation to move fast and break things. This is great when no other software depends on yours. If something breaks, you jump up and fix it. The story is quite different even when only a few dozen people depend on your software.

The less the users of your library have to know about your project, the better. The less they have to deal with issues of compatibility, the better. SemVer can be overly specific for some kinds of projects. If you are writing a library and you have a commitment to backward compatibility

PyCalVer version strings can be parsed according to PEP440 https://www.python.org/dev/peps/pep-0440/

A Word on Marketing

This setup of expectations for users can go one of two ways,

We use version numbers to communicate between the authors of software and its users. For users of libraries Particularly for libraries, it pays to keep things as simple as possible for your human users.

Commitment to Compatibility

Software projects can depend on many libraries. Consider that one package introducing a breaking change is enough to mess up your day. Especially in the case of libraries, your users should be able to write code that uses it and not have that code break at any point in the future. Users cannot be asked to keep track of all the changes to every little library that they use.

PyCalVer is explicitly non semantic. A PyCalVer version number does not express anything about

- Don't ever break things. When users depend on your
  software, backward compatibility matters and the way to
  express backward incompatible changes is not to bump a
  version number, but to change the package name. A change
  in the package name clearly communicates that a user must
  change their code so that it will work with the changed
  API. Everybody who does not have the bandwidth for those
  changes, doesn't even have to be aware of your new
  release.

- When you do break something, that should be considered a
  bug that has to be fixed as quickly as possible in a new
  version. It should always be safe for a user to update
  their dependencies. If something does break, users have to
  temporarily pin an older (known good) version, or update
  to a newer fixed version.

- Version numbers should not require a parser (present
  package excluded of course). A newer version number should
  always be lexically greater than an older one.
  TODO:
  https://setuptools.readthedocs.io/en/latest/setuptools.html#specifying-your-project-s-version

The main component of the version number is based on the calendar date. This is allows you to show your commitment (or lack thereof) to the maintenance of your library. It also allows users to see at a glance that their dependency might be out of date. In this versioning scheme it is completely reasonable to bump the version number without any changes, simply to express to your users, that you are still actively maintaining the software and that it is in a known good state.

For a much more detailed exposition of CalVer, see http://sedimental.org/designing_a_version.html https://calver.org/

from pkg_resources import parse_version

The Life of a Library

mylib      v201711.001-alpha     # birth of a project (in alpha)
mylib      v201711.002-alpha     # new features (in alpha)
mylib      v201712.003-beta      # bugfix release (in beta)
mylib      v201712.004-rc        # release candidate
mylib      v201712.005           # stable release
mylib      v201712.006           # stable bugfix release

mylib2     v201712.007-beta      # breaking change (new package name!)
mylib2     v201801.008-beta      # new features (in beta)
mylib2     v201801.009           # stable release

mylib      v201802.007           # security fix for legacy version
mylib2     v201802.010           # security fix

mylib2     v202604.9900           # threshold for four digit build numbers
mylib2     v202604.9901           # still four digits in the same month
mylib2     v202604.9911           # last build number with four digits
mylib2     v202605.09912          # build number zero padding added with date turnover
mylib2     v202605.09913          # stable release

mylib2     v203202.16051-rc       # release candidate
mylib2     v203202.16052          # stable release

...
v202008.500    # 500 is the limit for four digit build numbers, but
v202008.508    # zero padding is added only after the turnover to
v202009.0509   # a new month, so that lexical ordering is preserved.

The date portion of the version, gives the user an indication of how up their dependency is, whether or not a project is still being maintained.

The build number, gives the user an idea of the maturity of the project. A project which has been around long enough to produce hundreds of builds, might be considered mature, or at least a project that is only on build number 10, is probably still in early development.

FAQ

Q: "So you're trying to tell me I need to create a whole new package every time I introduce a introduce a breaking change?!".

A: First of all, what the hell are you doing? Secondly, YES! Let's assume your little package has even just 100 users. Do you have any idea about the total effort that will be expended because you decided it would be nice to change the name of a function? It is completely reasonable introduce that the friction for the package author when the price to users is orders of magnitude larger.

1801

https://calver.org/

I have given up on the idea that version numbers express anything about changes made between versions. Trying to express such information assumes 1. that the author of a package is aware of how a given change needs to be reflected in a version number and 2. that users and packaging software correctly parse that meaning. When I used semantic versioning, I realized that the major version number of my packages would never change, because I don't think breaking changes should ever be

Changelog for https://gitlab.com/mbarkhau/pycalver

v201812.0008-beta

• Add version tags using git/hg.
• Use git/hg tags as SSOT for most recent version.
• Start using https://gitlab.com/mbarkhau/bootstrapit
• Move to https://gitlab.com/mbarkhau/pycalver

v201809.0001-alpha

• Initial release