I’m happy to announce the 6th, and hopefully final, draft of PEP 440. The
updates to this PEP take into account feedback from several large projects and
users of various systems (such as Linux package tools) where this would be
expected to interface with. It also takes into account experience gained from
attempting to implement this PEP fully as a proof of concept within pip.

Significant updates to the PEP include:

* Switching the Epoch identifier from : to ! as : is not valid in a directory
  name on Windows.
* Local version identifiers use + as a seperator in order to reduce ambigiuty
  with existing versions on PyPI.
* Allow alpha numerics for local versions.
* Define a sorting algorithm for local versions.
* Moved the source label to PEP 426
* Normalization rules for parsing more versions following along the idea of
  Postel's Law.
* Declare that PEP 440 supercedes PEP 386 for Metadata 1.2, and also should be
  used for Metadata 1.0 and 1.1.
* Declare how invalid versions should be handled.
* Add the escape hatch "Arbitrary equality" operator === to allow depending on
  versions which cannot be parsed by PEP 440.
* Make specifier syntax match what setuptools uses (foo==1.0 instead of foo (==1.0))
* Remove the default specifier.
* Use @ for direct references instead of "from".
* Create a reference implementation and Link To it.
* Lots more minor changes.

The outcome of these changes is that we were able to raise compatability with
all the versions registered with PyPI up to 98.12% and we sort 99.88% of
projects registered with PyPI the same as pkg_resources does when filtering
invalid versions from the list of versions. The fallout is that 498 projects,
or 1.06%, are no longer installable without using the ``===`` operator and 190
projects, or 0.4%, of projects have a different result for what the "latest"
version is.

Of the 498 projects a number of them are nonsensical versions like ``.`` or
the repr of lazy objects and of the 190 projects projects it's about evenly
split between projects where pkg_resources supported something we didn't and
where PEP 440 just simply does a better job at parsing and sorting versions.

You can see the pip proof of concept and a large discussion about normalization
at https://github.com/pypa/pip/pull/1894 and the reference implementation can
be found at https://github.com/pypa/packaging/pull/1. The Proof of Concept does
not allow using the new specifiers inside of an install_requires in a source
distribution because setuptools itself does not support it, but it does support
them on the command line and in requirements.txt files.

The online view of the PEP can be found at
https://www.python.org/dev/peps/pep-0440/ and the changes since the last posting
can be found at http://hg.python.org/peps/rev/59a0d31a1bc2 and
http://hg.python.org/peps/rev/257822378672.

Without further ado, the PEP itself:

Abstract
========

This PEP describes a scheme for identifying versions of Python software
distributions, and declaring dependencies on particular versions.

This document addresses several limitations of the previous attempt at a
standardized approach to versioning, as described in PEP 345 and PEP 386.


Definitions
===========

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED",  "MAY", and "OPTIONAL" in this
document are to be interpreted as described in RFC 2119.

The following terms are to be interpreted as described in PEP 426:

* "Distributions"
* "Releases"
* "Build tools"
* "Index servers"
* "Publication tools"
* "Installation tools"
* "Automated tools"
* "Projects"


Version scheme
==============

Distributions are identified by a public version identifier which
supports all defined version comparison operations

The version scheme is used both to describe the distribution version
provided by a particular distribution archive, as well as to place
constraints on the version of dependencies needed in order to build or
run the software.


Public version identifiers
--------------------------

Public version identifiers MUST comply with the following scheme::

    [N!]N(.N)*[{a|b|c}N][.postN][.devN]

Public version identifiers MUST NOT include leading or trailing whitespace.

Public version identifiers MUST be unique within a given distribution.

Installation tools SHOULD ignore any public versions which do not comply with
this scheme. Installation tools MAY warn the user when non-compliant
or ambiguous versions are detected.

Public version identifiers are separated into up to five segments:

* Epoch segment: ``N!``
* Release segment: ``N(.N)*``
* Pre-release segment: ``{a|b|c}N``
* Post-release segment: ``.postN``
* Development release segment: ``.devN``

Any given release will be a "final release", "pre-release", "post-release" or
"developmental release" as defined in the following sections.

All numeric components MUST be non-negative integers.

All numeric components MUST be interpreted and ordered according to their
numeric value, not as text strings.

All numeric components MAY be zero. Except as described below for the
release segment, a numeric component of zero has no special significance
aside from always being the lowest possible value in the version ordering.

.. note::

   Some hard to read version identifiers are permitted by this scheme in
   order to better accommodate the wide range of versioning practices
   across existing public and private Python projects.

   Accordingly, some of the versioning practices which are technically
   permitted by the PEP are strongly discouraged for new projects. Where
   this is the case, the relevant details are noted in the following
   sections.


Local version identifiers
-------------------------

Local version identifiers MUST comply with the following scheme::

    <public version identifier>[+<local version label>]

They consist of a normal public version identifier (as defined in the
previous section), along with an arbitrary "local version label", separated
from the public version identifier by a plus. Local version labels have
no specific semantics assigned, but some syntactic restrictions are imposed.

Local version identifiers are used to denote fully API (and, if applicable,
ABI) compatible patched versions of upstream projects. For example, these
may be created by application developers and system integrators by applying
specific backported bug fixes when upgrading to a new upstream release would
be too disruptive to the application or other integrated system (such as a
Linux distribution).

The inclusion of the local version label makes it possible to differentiate
upstream releases from potentially altered rebuilds by downstream
integrators. The use of a local version identifier does not affect the kind
of a release but, when applied to a source distribution, does indicate that
it may not contain the exact same code as the corresponding upstream release.

To ensure local version identifiers can be readily incorporated as part of
filenames and URLs, and to avoid formatting inconsistencies in hexadecimal
hash representations, local version labels MUST be limited to the following
set of permitted characters:

* ASCII letters (``[a-zA-Z]``)
* ASCII digits (``[0-9]``)
* periods (``.``)

Local version labels MUST start and end with an ASCII letter or digit.

Comparison and ordering of local versions considers each segment of the local
version (divided by a ``.``) separately. If a segment consists entirely of
ASCII digits then that section should be considered an integer for comparison
purposes and if a segment contains any ASCII letters than that segment is
compared lexicographically with case insensitivity. When comparing a numeric
and lexicographic segment, the numeric section always compares as greater than
the lexicographic segment. Additionally a local version with a great number of
segments will always compare as greater than a local version with fewer
segments, as long as the shorter local version's segments match the beginning
of the longer local version's segments exactly.

Local version identifiers may be used in most locations where a public
version identifier is expected, with the exception of any version specifiers
that explicitly rely on being able to unambiguously order candidate versions.

Public index servers SHOULD NOT allow the use of local version identifiers
for uploaded distributions.

Source distributions using a local version identifier SHOULD provide the
``python.integrator`` extension metadata (as defined in :pep:`459`).


Final releases
--------------

A version identifier that consists solely of a release segment and optionally
an epoch identifier is termed a "final release".

The release segment consists of one or more non-negative integer
values, separated by dots::

    N[.N]+

Final releases within a project MUST be numbered in a consistently
increasing fashion, otherwise automated tools will not be able to upgrade
them correctly.

Comparison and ordering of release segments considers the numeric value
of each component of the release segment in turn. When comparing release
segments with different numbers of components, the shorter segment is
padded out with additional zeros as necessary.

While any number of additional components after the first are permitted
under this scheme, the most common variants are to use two components
("major.minor") or three components ("major.minor.micro").

For example::

    0.9
    0.9.1
    0.9.2
    ...
    0.9.10
    0.9.11
    1.0
    1.0.1
    1.1
    2.0
    2.0.1
    ...

A release series is any set of final release numbers that start with a
common prefix. For example, ``3.3.1``, ``3.3.5`` and ``3.3.9.45`` are all
part of the ``3.3`` release series.

.. note::

   ``X.Y`` and ``X.Y.0`` are not considered distinct release numbers, as
   the release segment comparison rules implicit expand the two component
   form to ``X.Y.0`` when comparing it to any release segment that includes
   three components.

Date based release segments are also permitted. An example of a date based
release scheme using the year and month of the release::

    2012.04
    2012.07
    2012.10
    2013.01
    2013.06
    ...


Pre-releases
------------

Some projects use an "alpha, beta, release candidate" pre-release cycle to
support testing by their users prior to a final release.

If used as part of a project's development cycle, these pre-releases are
indicated by including a pre-release segment in the version identifier::

    X.YaN  # Alpha release
    X.YbN  # Beta release
    X.YcN  # Candidate release
    X.Y    # Final release

A version identifier that consists solely of a release segment and a
pre-release segment is termed a "pre-release".

The pre-release segment consists of an alphabetical identifier for the
pre-release phase, along with a non-negative integer value. Pre-releases for
a given release are ordered first by phase (alpha, beta, release candidate)
and then by the numerical component within that phase.

Installation tools MAY accept both ``c`` and ``rc`` releases for a common
release segment in order to handle some existing legacy distributions.

Installation tools SHOULD interpret ``rc`` versions as being equivalent to
``c`` versions (that is, ``rc1`` indicates the same version as ``c1``).

Build tools, publication tools and index servers SHOULD disallow the creation
of both ``c`` and ``rc`` releases for a common release segment.


Post-releases
-------------

Some projects use post-releases to address minor errors in a final release
that do not affect the distributed software (for example, correcting an error
in the release notes).

If used as part of a project's development cycle, these post-releases are
indicated by including a post-release segment in the version identifier::

    X.Y.postN    # Post-release

A version identifier that includes a post-release segment without a
developmental release segment is termed a "post-release".

The post-release segment consists of the string ``.post``, followed by a
non-negative integer value. Post-releases are ordered by their
numerical component, immediately following the corresponding release,
and ahead of any subsequent release.

.. note::

   The use of post-releases to publish maintenance releases containing
   actual bug fixes is strongly discouraged. In general, it is better
   to use a longer release number and increment the final component
   for each maintenance release.

Post-releases are also permitted for pre-releases::

    X.YaN.postM  # Post-release of an alpha release
    X.YbN.postM  # Post-release of a beta release
    X.YcN.postM  # Post-release of a release candidate

.. note::

   Creating post-releases of pre-releases is strongly discouraged, as
   it makes the version identifier difficult to parse for human readers.
   In general, it is substantially clearer to simply create a new
   pre-release by incrementing the numeric component.


Developmental releases
----------------------

Some projects make regular developmental releases, and system packagers
(especially for Linux distributions) may wish to create early releases
directly from source control which do not conflict with later project
releases.

If used as part of a project's development cycle, these developmental
releases are indicated by including a developmental release segment in the
version identifier::

    X.Y.devN    # Developmental release

A version identifier that includes a developmental release segment is
termed a "developmental release".

The developmental release segment consists of the string ``.dev``,
followed by a non-negative integer value. Developmental releases are ordered
by their numerical component, immediately before the corresponding release
(and before any pre-releases with the same release segment), and following
any previous release (including any post-releases).

Developmental releases are also permitted for pre-releases and
post-releases::

    X.YaN.devM      # Developmental release of an alpha release
    X.YbN.devM      # Developmental release of a beta release
    X.YcN.devM      # Developmental release of a release candidate
    X.Y.postN.devM  # Developmental release of a post-release

.. note::

   Creating developmental releases of pre-releases is strongly
   discouraged, as it makes the version identifier difficult to parse for
   human readers. In general, it is substantially clearer to simply create
   additional pre-releases by incrementing the numeric component.

   Developmental releases of post-releases are also strongly discouraged,
   but they may be appropriate for projects which use the post-release
   notation for full maintenance releases which may include code changes.


Version epochs
--------------

If included in a version identifier, the epoch appears before all other
components, separated from the release segment by an exclamation mark::

    E!X.Y  # Version identifier with epoch

If no explicit epoch is given, the implicit epoch is ``0``.

Most version identifiers will not include an epoch, as an explicit epoch is
only needed if a project *changes* the way it handles version numbering in
a way that means the normal version ordering rules will give the wrong
answer. For example, if a project is using date based versions like
``2014.04`` and would like to switch to semantic versions like ``1.0``, then
the new releases would be identified as *older* than the date based releases
when using the normal sorting scheme::

    1.0
    1.1
    2.0
    2013.10
    2014.04

However, by specifying an explicit epoch, the sort order can be changed
appropriately, as all versions from a later epoch are sorted after versions
from an earlier epoch::

    2013.10
    2014.04
    1!1.0
    1!1.1
    1!2.0

Normalization
-------------

In order to maintain better compatibility with existing versions there are a
number of "alternative" syntaxes that MUST be taken into account when parsing
versions. These syntaxes MUST be considered when parsing a version, however
they should be "normalized" to the standard syntax defined above.


Case sensitivity
~~~~~~~~~~~~~~~~

All ascii letters should be interpreted case insensitively within a version and
the normal form is lowercase. This allows versions such as ``1.1RC1`` which
would be normalized to ``1.1c1``.


Integer Normalization
~~~~~~~~~~~~~~~~~~~~~

All integers are interpreted via the ``int()`` built in and normalize to the
string form of the output. This means that an integer version of ``00`` would
normalize to ``0`` while ``09000`` would normalize to ``9000``. This does not
hold true for integers inside of an alphanumeric segment of a local version
such as ``1.0+foo0100`` which is already in its normalized form.


Pre-release separators
~~~~~~~~~~~~~~~~~~~~~~

Pre-releases should allow a ``.``, ``-``, or ``_`` separator between the
release segment and the pre-release segment. The normal form for this is
without a separator. This allows versions such as ``1.1.a1`` or ``1.1-a1``
which would be normalized to ``1.1a1``. It should also allow a seperator to
be used between the pre-release signifier and the numeral. This allows versions
such as ``1.0a.1`` which would be normalized to ``1.0a1``.


Pre-release spelling
~~~~~~~~~~~~~~~~~~~~

Pre-releases allow the additional spellings of ``alpha``, ``beta``, ``rc``,
``pre``, and ``preview`` for ``a``, ``b``, ``c``, ``c``, and ``c`` respectively.
This allows versions such as ``1.1alpha1``, ``1.1beta2``, or ``1.1rc3`` which
normalize to ``1.1a1``, ``1.1b2``, and ``1.1c3``. In every case the additional
spelling should be considered equivalent to their normal forms.


Implicit pre-release number
~~~~~~~~~~~~~~~~~~~~~~~~~~~

Pre releases allow omitting the numeral in which case it is implicitly assumed
to be ``0``. The normal form for this is to include the ``0`` explicitly. This
allows versions such as ``1.2a`` which is normalized to ``1.2a0``.


Post release separators
~~~~~~~~~~~~~~~~~~~~~~~

Post releases allow a ``.``,``-``, or ``_`` separator as well as omitting the
separator all together. The normal form of this is with the ``.`` separator.
This allows versions such as ``1.2-post2`` or ``1.2post2`` which normalize to
``1.2.post2``. Like the pre-release seperator this also allows an optional
separator between the post release signifier and the numeral. This allows
versions like ``1.2.post-2`` which would normalize to ``1.2.post2``.


Post release spelling
~~~~~~~~~~~~~~~~~~~~~

Post-releases allow the additional spellings of ``rev`` and ``r``. This allows
versions such as ``1.0-r4`` which normalizes to ``1.0.post4``. As with the
pre-releases the additional spellings should be considered equivalent to their
normal forms.


Implicit post release number
~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Post releases allow omiting the numeral in which case it is implicitly assumed
to be ``0``. The normal form for this is to include the ``0`` explicitly. This
allows versions such as ``1.2.post`` which is normalized to ``1.2.post0``.


Implicit post releases
~~~~~~~~~~~~~~~~~~~~~~

Post releases allow omitting the ``post`` signifier all together. When using
this form the separator MUST be ``-`` and no other form is allowed. This allows
versions such as ``1.0-1`` to be normalized to ``1.0.post1``. This particular
normalization MUST NOT be used in conjunction with the implicit post release
number rule. In other words ``1.0-`` is *not* a valid version and it does *not*
normalize to ``1.0.post0``.


Development release separators
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Development releases allow a ``.``, ``-``, or a ``_`` separator as well as
omitting the separator all together. The normal form of this is with the ``.``
separator. This allows versions such as ``1.2-dev2`` or ``1.2dev2`` which
normalize to ``1.2.dev2``.


Implicit development release number
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Development releases allow omiting the numeral in which case it is implicitly
assumed to be ``0``. The normal form for this is to include the ``0``
explicitly. This allows versions such as ``1.2.dev`` which is normalized to
``1.2.dev0``.


Local version segments
~~~~~~~~~~~~~~~~~~~~~~

With a local version, in addition to the use of ``.`` as a separator of
segments, the use of ``-`` and ``_`` is also acceptable. The normal form is
using the ``.`` character. This allows versions such as ``1.0+ubuntu-1`` to be
normalized to ``1.0+ubuntu.1``.


Preceding v character
~~~~~~~~~~~~~~~~~~~~~

In order to support the common version notation of ``v1.0`` versions may be
preceded by a single literal ``v`` character. This character MUST be ignored
for all purposes and should be omitted from all normalized forms of the
version. The same version with and without the ``v`` is considered equivalent.


Leading and Trailing Whitespace
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Leading and trailing whitespace must be silently ignored and removed from all
normalized forms of a version. This includes ``" "``, ``\t``, ``\n``, ``\r``,
``\f``, and ``\v``. This allows accidental whitespace to be handled sensibly,
such as a version like ``1.0\n`` which normalizes to ``1.0``.


Examples of compliant version schemes
-------------------------------------

The standard version scheme is designed to encompass a wide range of
identification practices across public and private Python projects. In
practice, a single project attempting to use the full flexibility offered
by the scheme would create a situation where human users had difficulty
figuring out the relative order of versions, even though the rules above
ensure all compliant tools will order them consistently.

The following examples illustrate a small selection of the different
approaches projects may choose to identify their releases, while still
ensuring that the "latest release" and the "latest stable release" can
be easily determined, both by human users and automated tools.

Simple "major.minor" versioning::

    0.1
    0.2
    0.3
    1.0
    1.1
    ...

Simple "major.minor.micro" versioning::

    1.1.0
    1.1.1
    1.1.2
    1.2.0
    ...

"major.minor" versioning with alpha, beta and candidate
pre-releases::

    0.9
    1.0a1
    1.0a2
    1.0b1
    1.0c1
    1.0
    1.1a1
    ...

"major.minor" versioning with developmental releases, release candidates
and post-releases for minor corrections::

    0.9
    1.0.dev1
    1.0.dev2
    1.0.dev3
    1.0.dev4
    1.0rc1
    1.0rc2
    1.0
    1.0.post1
    1.1.dev1
    ...

Date based releases, using an incrementing serial within each year, skipping
zero::

    2012.1
    2012.2
    2012.3
    ...
    2012.15
    2013.1
    2013.2
    ...


Summary of permitted suffixes and relative ordering
---------------------------------------------------

.. note::

   This section is intended primarily for authors of tools that
   automatically process distribution metadata, rather than developers
   of Python distributions deciding on a versioning scheme.

The epoch segment of version identifiers MUST be sorted according to the
numeric value of the given epoch. If no epoch segment is present, the
implicit numeric value is ``0``.

The release segment of version identifiers MUST be sorted in
the same order as Python's tuple sorting when the release segment is
parsed as follows::

    tuple(map(int, release_segment.split(".")))

All release segments involved in the comparison MUST be converted to a
consistent length by padding shorter segments with zeros as needed.

Within a numeric release (``1.0``, ``2.7.3``), the following suffixes
are permitted and MUST be ordered as shown::

   .devN, aN, bN, cN/rcN, <no suffix>, .postN

Note that `rc` is considered to be semantically equivalent to `c` and must be
sorted as if it were `c`. Tools MAY reject the case of having the same ``N``
for both a ``rc`` and a ``c`` in the same release segment as ambiguous and
remain in compliance with the PEP.

Within an alpha (``1.0a1``), beta (``1.0b1``), or release candidate
(``1.0c1``, ``1.0rc1``), the following suffixes are permitted and MUST be
ordered as shown::

   .devN, <no suffix>, .postN

Within a post-release (``1.0.post1``), the following suffixes are permitted
and MUST be ordered as shown::

    .devN, <no suffix>

Note that ``devN`` and ``postN`` MUST always be preceded by a dot, even
when used immediately following a numeric version (e.g. ``1.0.dev456``,
``1.0.post1``).

Within a pre-release, post-release or development release segment with a
shared prefix, ordering MUST be by the value of the numeric component.

The following example covers many of the possible combinations::

    1.0.dev456
    1.0a1
    1.0a2.dev456
    1.0a12.dev456
    1.0a12
    1.0b1.dev456
    1.0b2
    1.0b2.post345.dev456
    1.0b2.post345
    1.0c1.dev456
    1.0c1
    1.0
    1.0+abc.5
    1.0+abc.7
    1.0+5
    1.0.post456.dev34
    1.0.post456
    1.1.dev1


Version ordering across different metadata versions
---------------------------------------------------

Metadata v1.0 (PEP 241) and metadata v1.1 (PEP 314) do not specify a standard
version identification or ordering scheme. However metadata v1.2 (PEP 345)
does specify a scheme which is defined in PEP 386.

Due to the nature of the simple installer API it is not possible for an
installer to be aware of which metadata version a particular distribution was
using. Additionally installers required the ability to create a reasonably
prioritized list that includes all, or as many as possible, versions of
a project to determine which versions it should install. These requirements
necessitate a standardization across one parsing mechanism to be used for all
versions of a project.

Due to the above, this PEP MUST be used for all versions of metadata and
supersedes PEP 386 even for metadata v1.2. Tools SHOULD ignore any versions
which cannot be parsed by the rules in this PEP, but MAY fall back to
implementation defined version parsing and ordering schemes if no versions
complying with this PEP are available.

Distribution users may wish to explicitly remove non-compliant versions from
any private package indexes they control.


Compatibility with other version schemes
----------------------------------------

Some projects may choose to use a version scheme which requires
translation in order to comply with the public version scheme defined in
this PEP. In such cases, the project specific version can be stored in the
metadata while the translated public version is published in the version field.

This allows automated distribution tools to provide consistently correct
ordering of published releases, while still allowing developers to use
the internal versioning scheme they prefer for their projects.


Semantic versioning
~~~~~~~~~~~~~~~~~~~

`Semantic versioning`_ is a popular version identification scheme that is
more prescriptive than this PEP regarding the significance of different
elements of a release number. Even if a project chooses not to abide by
the details of semantic versioning, the scheme is worth understanding as
it covers many of the issues that can arise when depending on other
distributions, and when publishing a distribution that others rely on.

The "Major.Minor.Patch" (described in this PEP as "major.minor.micro")
aspects of semantic versioning (clauses 1-9 in the 2.0.0-rc-1 specification)
are fully compatible with the version scheme defined in this PEP, and abiding
by these aspects is encouraged.

Semantic versions containing a hyphen (pre-releases - clause 10) or a
plus sign (builds - clause 11) are *not* compatible with this PEP
and are not permitted in the public version field.

One possible mechanism to translate such semantic versioning based source
labels to compatible public versions is to use the ``.devN`` suffix to
specify the appropriate version order.

Specific build information may also be included in local version labels.

.. _Semantic versioning: http://semver.org/


DVCS based version labels
~~~~~~~~~~~~~~~~~~~~~~~~~

Many build tools integrate with distributed version control systems like
Git and Mercurial in order to add an identifying hash to the version
identifier. As hashes cannot be ordered reliably such versions are not
permitted in the public version field.

As with semantic versioning, the public ``.devN`` suffix may be used to
uniquely identify such releases for publication, while the original DVCS based
label can be stored in the project metadata.

Identifying hash information may also be included in local version labels.


Olson database versioning
~~~~~~~~~~~~~~~~~~~~~~~~~

The ``pytz`` project inherits its versioning scheme from the corresponding
Olson timezone database versioning scheme: the year followed by a lowercase
character indicating the version of the database within that year.

This can be translated to a compliant public version identifier as
``<year>.<serial>``, where the serial starts at zero or one (for the
'<year>a' release) and is incremented with each subsequent database
update within the year.

As with other translated version identifiers, the corresponding Olson
database version could be recorded in the project metadata.


Version specifiers
==================

A version specifier consists of a series of version clauses, separated by
commas. For example::

   ~= 0.9, >= 1.0, != 1.3.4.*, < 2.0

The comparison operator determines the kind of version clause:

* ``~=``: `Compatible release`_ clause
* ``==``: `Version matching`_ clause
* ``!=``: `Version exclusion`_ clause
* ``<=``, ``>=``: `Inclusive ordered comparison`_ clause
* ``<``, ``>``: `Exclusive ordered comparison`_ clause
* ``===``: `Arbitrary equality`_ clause.

The comma (",") is equivalent to a logical **and** operator: a candidate
version must match all given version clauses in order to match the
specifier as a whole.

Whitespace between a conditional operator and the following version
identifier is optional, as is the whitespace around the commas.

When multiple candidate versions match a version specifier, the preferred
version SHOULD be the latest version as determined by the consistent
ordering defined by the standard `Version scheme`_. Whether or not
pre-releases are considered as candidate versions SHOULD be handled as
described in `Handling of pre-releases`_.

Except where specifically noted below, local version identifiers MUST NOT be
permitted in version specifiers, and local version labels MUST be ignored
entirely when checking if candidate versions match a given version
specifier.


Compatible release
------------------

A compatible release clause consists of either a version identifier without
any comparison operator or else the compatible release operator ``~=``
and a version identifier. It matches any candidate version that is expected
to be compatible with the specified version.

The specified version identifier must be in the standard format described in
`Version scheme`_. Local version identifiers are NOT permitted in this
version specifier.

For a given release identifier ``V.N``, the compatible release clause is
approximately equivalent to the pair of comparison clauses::

    >= V.N, == V.*

This operator MUST NOT be used with a single segment version number such as
``~=1``.

For example, the following groups of version clauses are equivalent::

    2.2
    ~= 2.2
    >= 2.2, == 2.*

    1.4.5
    ~= 1.4.5
    >= 1.4.5, == 1.4.*

If a pre-release, post-release or developmental release is named in a
compatible release clause as ``V.N.suffix``, then the suffix is ignored
when determining the required prefix match::

    2.2.post3
    ~= 2.2.post3
    >= 2.2.post3, == 2.*

    1.4.5a4
    ~= 1.4.5a4
    >= 1.4.5a4, == 1.4.*

The padding rules for release segment comparisons means that the assumed
degree of forward compatibility in a compatible release clause can be
controlled by appending additional zeros to the version specifier::

    2.2.0
    ~= 2.2.0
    >= 2.2.0, == 2.2.*

    1.4.5.0
    ~= 1.4.5.0
    >= 1.4.5.0, == 1.4.5.*


Version matching
----------------

A version matching clause includes the version matching operator ``==``
and a version identifier.

The specified version identifier must be in the standard format described in
`Version scheme`_, but a trailing ``.*`` is permitted on public version
identifiers as described below.

By default, the version matching operator is based on a strict equality
comparison: the specified version must be exactly the same as the requested
version. The *only* substitution performed is the zero padding of the
release segment to ensure the release segments are compared with the same
length.

Whether or not strict version matching is appropriate depends on the specific
use case for the version specifier. Automated tools SHOULD at least issue
warnings and MAY reject them entirely when strict version matches are used
inappropriately.

Prefix matching may be requested instead of strict comparison, by appending
a trailing ``.*`` to the version identifier in the version matching clause.
This means that additional trailing segments will be ignored when
determining whether or not a version identifier matches the clause. If the
specified version includes only a release segment, than trailing components
(or the lack thereof) in the release segment are also ignored.

For example, given the version ``1.1.post1``, the following clauses would
match or not as shown::

    == 1.1        # Not equal, so 1.1.post1 does not match clause
    == 1.1.post1  # Equal, so 1.1.post1 matches clause
    == 1.1.*      # Same prefix, so 1.1.post1 matches clause

For purposes of prefix matching, the pre-release segment is considered to
have an implied preceding ``.``, so given the version ``1.1a1``, the
following clauses would match or not as shown::

    == 1.1        # Not equal, so 1.1a1 does not match clause
    == 1.1a1      # Equal, so 1.1a1 matches clause
    == 1.1.*      # Same prefix, so 1.1a1 matches clause

An exact match is also considered a prefix match (this interpreation is
implied by the usual zero padding rules for the release segment of version
identifiers). Given the version ``1.1``, the following clauses would
match or not as shown::

    == 1.1        # Equal, so 1.1 matches clause
    == 1.1.0      # Zero padding expands 1.1 to 1.1.0, so it matches clause
    == 1.1.dev1   # Not equal (dev-release), so 1.1 does not match clause
    == 1.1a1      # Not equal (pre-release), so 1.1 does not match clause
    == 1.1.post1  # Not equal (post-release), so 1.1 does not match clause
    == 1.1.*      # Same prefix, so 1.1 matches clause

It is invalid to have a prefix match containing a development or local release
such as ``1.0.dev1.*`` or ``1.0+foo1.*``. If present, the development release
segment is always the final segment in the public version, and the local version
is ignored for comparison purposes, so using either in a prefix match wouldn't
make any sense.

The use of ``==`` (without at least the wildcard suffix) when defining
dependencies for published distributions is strongly discouraged as it
greatly complicates the deployment of security fixes. The strict version
comparison operator is intended primarily for use when defining
dependencies for repeatable *deployments of applications* while using
a shared distribution index.

If the specified version identifier is a public version identifier (no
local version label), then the local version label of any candidate versions
MUST be ignored when matching versions.

If the specified version identifier is a local version identifier, then the
local version labels of candidate versions MUST be considered when matching
versions, with the public version identifier being matched as described
above, and the local version label being checked for equivalence using a
strict string equality comparison.


Version exclusion
-----------------

A version exclusion clause includes the version exclusion operator ``!=``
and a version identifier.

The allowed version identifiers and comparison semantics are the same as
those of the `Version matching`_ operator, except that the sense of any
match is inverted.

For example, given the version ``1.1.post1``, the following clauses would
match or not as shown::

    != 1.1        # Not equal, so 1.1.post1 matches clause
    != 1.1.post1  # Equal, so 1.1.post1 does not match clause
    != 1.1.*      # Same prefix, so 1.1.post1 does not match clause


Inclusive ordered comparison
----------------------------

An inclusive ordered comparison clause includes a comparison operator and a
version identifier, and will match any version where the comparison is correct
based on the relative position of the candidate version and the specified
version given the consistent ordering defined by the standard
`Version scheme`_.

The inclusive ordered comparison operators are ``<=`` and ``>=``.

As with version matching, the release segment is zero padded as necessary to
ensure the release segments are compared with the same length.

Local version identifiers are NOT permitted in this version specifier.


Exclusive ordered comparison
----------------------------

Exclusive ordered comparisons are similar to inclusive ordered comparisons,
except that the comparison operators are ``<`` and ``>`` and the clause
MUST be effectively interpreted as implying the prefix based version
exclusion clause ``!= V.*``.

The exclusive ordered comparison ``> V`` MUST NOT match a post-release
or maintenance release of the given version. Maintenance releases can be
permitted by using the clause ``> V.0``, while both post releases and
maintenance releases can be permitted by using the inclusive ordered
comparison ``>= V.post1``.

The exclusive ordered comparison ``< V`` MUST NOT match a pre-release of
the given version, even if acceptance of pre-releases is enabled as
described in the section below.

Local version identifiers are NOT permitted in this version specifier.


Arbitrary equality
------------------

Arbitrary equality comparisons are simple string equality operations which do
not take into account any of the semantic information such as zero padding or
local versions. This operator also does not support prefix matching as the
``==`` operator does.

The primary use case for arbitrary equality is to allow for specifying a
version which cannot otherwise be represented by this PEP. This operator is
special and acts as an escape hatch to allow someone using a tool which
implements this PEP to still install a legacy version which is otherwise
incompatible with this PEP.

An example would be ``===foobar`` which would match a version of ``foobar``.

This operator may also be used to explicitly require an unpatched version
of a project such as ``===1.0`` which would not match for a version
``1.0+downstream1``.

Use of this operator is heavily discouraged and tooling MAY display a warning
when it is used.


Handling of pre-releases
------------------------

Pre-releases of any kind, including developmental releases, are implicitly
excluded from all version specifiers, *unless* they are already present
on the system, explicitly requested by the user, or if the only available
version that satisfies the version specifier is a pre-release.

By default, dependency resolution tools SHOULD:

* accept already installed pre-releases for all version specifiers
* accept remotely available pre-releases for version specifiers where
  there is no final or post release that satisfies the version specifier
* exclude all other pre-releases from consideration

Dependency resolution tools MAY issue a warning if a pre-release is needed
to satisfy a version specifier.

Dependency resolution tools SHOULD also allow users to request the
following alternative behaviours:

* accepting pre-releases for all version specifiers
* excluding pre-releases for all version specifiers (reporting an error or
  warning if a pre-release is already installed locally, or if a
  pre-release is the only way to satisfy a particular specifier)

Dependency resolution tools MAY also allow the above behaviour to be
controlled on a per-distribution basis.

Post-releases and final releases receive no special treatment in version
specifiers - they are always included unless explicitly excluded.


Examples
--------

* ``3.1``: version 3.1 or later, but not version 4.0 or later.
* ``3.1.2``: version 3.1.2 or later, but not version 3.2.0 or later.
* ``3.1a1``: version 3.1a1 or later, but not version 4.0 or later.
* ``== 3.1``: specifically version 3.1 (or 3.1.0), excludes all pre-releases,
  post releases, developmental releases and any 3.1.x maintenance releases.
* ``== 3.1.*``: any version that starts with 3.1. Equivalent to the
  ``3.1.0`` compatible release clause.
* ``3.1.0, != 3.1.3``: version 3.1.0 or later, but not version 3.1.3 and
  not version 3.2.0 or later.


Direct references
=================

Some automated tools may permit the use of a direct reference as an
alternative to a normal version specifier. A direct reference consists of
the specifier ``@`` and an explicit URL.

Whether or not direct references are appropriate depends on the specific
use case for the version specifier. Automated tools SHOULD at least issue
warnings and MAY reject them entirely when direct references are used
inappropriately.

Public index servers SHOULD NOT allow the use of direct references in
uploaded distributions. Direct references are intended as a tool for
software integrators rather than publishers.

Depending on the use case, some appropriate targets for a direct URL
reference may be a valid ``source_url`` entry (see PEP 426), an sdist, or
a wheel binary archive. The exact URLs and targets supported will be tool
dependent.

For example, a local source archive may be referenced directly::

    pip @ file:///localbuilds/pip-1.3.1.zip

Alternatively, a prebuilt archive may also be referenced::

    pip @ file:///localbuilds/pip-1.3.1-py33-none-any.whl

All direct references that do not refer to a local file URL SHOULD specify
a secure transport mechanism (such as ``https``) AND include an expected
hash value in the URL for verification purposes. If a direct reference is
specified without any hash information, with hash information that the
tool doesn't understand, or with a selected hash algorithm that the tool
considers too weak to trust, automated tools SHOULD at least emit a warning
and MAY refuse to rely on the URL. If such a direct reference also uses an
insecure transport, automated tools SHOULD NOT rely on the URL.

It is RECOMMENDED that only hashes which are unconditionally provided by
the latest version of the standard library's ``hashlib`` module be used
for source archive hashes. At time of writing, that list consists of
``'md5'``, ``'sha1'``, ``'sha224'``, ``'sha256'``, ``'sha384'``, and
``'sha512'``.

For source archive and wheel references, an expected hash value may be
specified by including a ``<hash-algorithm>=<expected-hash>`` entry as
part of the URL fragment.

For version control references, the ``VCS+protocol`` scheme SHOULD be
used to identify both the version control system and the secure transport,
and a version control system with hash based commit identifiers SHOULD be
used. Automated tools MAY omit warnings about missing hashes for version
control systems that do not provide hash based commit identifiers.

To handle version control systems that do not support including commit or
tag references directly in the URL, that information may be appended to the
end of the URL using the ``@<commit-hash>`` or the ``@<tag>#<commit-hash>``
notation.

.. note::

   This isn't *quite* the same as the existing VCS reference notation
   supported by pip. Firstly, the distribution name is moved in front rather
   than embedded as part of the URL. Secondly, the commit hash is included
   even when retrieving based on a tag, in order to meet the requirement
   above that *every* link should include a hash to make things harder to
   forge (creating a malicious repo with a particular tag is easy, creating
   one with a specific *hash*, less so).

Remote URL examples::

    pip @ https://github.com/pypa/pip/archive/1.3.1.zip#sha1=da9234ee9982d4bbb3c72346a6de940a148ea686
    pip @ git+https://github.com/pypa/pip.git@7921be1537eac1e97bc40179a57f0349c2aee67d
    pip @ git+https://github.com/pypa/pip.git@1.3.1#7921be1537eac1e97bc40179a57f0349c2aee67d



Updating the versioning specification
=====================================

The versioning specification may be updated with clarifications without
requiring a new PEP or a change to the metadata version.

Actually changing the version comparison semantics still requires a new
versioning scheme and metadata version defined in new PEPs.


Summary of differences from \PEP 386
====================================

* Moved the description of version specifiers into the versioning PEP

* Added the "direct reference" concept as a standard notation for direct
  references to resources (rather than each tool needing to invent its own)

* Added the "local version identifier" and "local version label" concepts to
  allow system integrators to indicate patched builds in a way that is
  supported by the upstream tools, as well as to allow the incorporation of
  build tags into the versioning of binary distributions.

* Added the "compatible release" clause

* Added the trailing wildcard syntax for prefix based version matching
  and exclusion

* Changed the top level sort position of the ``.devN`` suffix

* Allowed single value version numbers

* Explicit exclusion of leading or trailing whitespace

* Explicit support for date based versions

* Explicit normalisation rules to improve compatibility with
  existing version metadata on PyPI where it doesn't introduce
  ambiguity

* Implicitly exclude pre-releases unless they're already present or
  needed to satisfy a dependency

* Treat post releases the same way as unqualified releases

* Discuss ordering and dependencies across metadata versions

The rationale for major changes is given in the following sections.


Changing the version scheme
---------------------------

One key change in the version scheme in this PEP relative to that in
PEP 386 is to sort top level developmental releases like ``X.Y.devN`` ahead
of alpha releases like ``X.Ya1``. This is a far more logical sort order, as
projects already using both development releases and alphas/betas/release
candidates do not want their developmental releases sorted in
between their release candidates and their final releases. There is no
rationale for using ``dev`` releases in that position rather than
merely creating additional release candidates.

The updated sort order also means the sorting of ``dev`` versions is now
consistent between the metadata standard and the pre-existing behaviour
of ``pkg_resources`` (and hence the behaviour of current installation
tools).

Making this change should make it easier for affected existing projects to
migrate to the latest version of the metadata standard.

Another change to the version scheme is to allow single number
versions, similar to those used by non-Python projects like Mozilla
Firefox, Google Chrome and the Fedora Linux distribution. This is actually
expected to be more useful for version specifiers, but it is easier to
allow it for both version specifiers and release numbers, rather than
splitting the two definitions.

The exclusion of leading and trailing whitespace was made explicit after
a couple of projects with version identifiers differing only in a
trailing ``\n`` character were found on PyPI.

Various other normalisation rules were also added as described in the
separate section on version normalisation below.

`Appendix A` shows detailed results of an analysis of PyPI distribution
version information, as collected on 8th August, 2014. This analysis
compares the behavior of the explicitly ordered version scheme defined in
this PEP with the de facto standard defined by the behavior of setuptools.
These metrics are useful, as the intent of this PEP is to follow existing
setuptools behavior as closely as is feasible, while still throwing
exceptions for unorderable versions (rather than trying to guess an
appropriate order as setuptools does).


A more opinionated description of the versioning scheme
-------------------------------------------------------

As in PEP 386, the primary focus is on codifying existing practices to make
them more amenable to automation, rather than demanding that existing
projects make non-trivial changes to their workflow. However, the
standard scheme allows significantly more flexibility than is needed
for the vast majority of simple Python packages (which often don't even
need maintenance releases - many users are happy with needing to upgrade to a
new feature release to get bug fixes).

For the benefit of novice developers, and for experienced developers
wishing to better understand the various use cases, the specification
now goes into much greater detail on the components of the defined
version scheme, including examples of how each component may be used
in practice.

The PEP also explicitly guides developers in the direction of
semantic versioning (without requiring it), and discourages the use of
several aspects of the full versioning scheme that have largely been
included in order to cover esoteric corner cases in the practices of
existing projects and in repackaging software for Linux distributions.


Describing version specifiers alongside the versioning scheme
-------------------------------------------------------------

The main reason to even have a standardised version scheme in the first place
is to make it easier to do reliable automated dependency analysis. It makes
more sense to describe the primary use case for version identifiers alongside
their definition.


Changing the interpretation of version specifiers
-------------------------------------------------

The previous interpretation of version specifiers made it very easy to
accidentally download a pre-release version of a dependency. This in
turn made it difficult for developers to publish pre-release versions
of software to the Python Package Index, as even marking the package as
hidden wasn't enough to keep automated tools from downloading it, and also
made it harder for users to obtain the test release manually through the
main PyPI web interface.

The previous interpretation also excluded post-releases from some version
specifiers for no adequately justified reason.

The updated interpretation is intended to make it difficult to accidentally
accept a pre-release version as satisfying a dependency, while still
allowing pre-release versions to be retrieved automatically when that's the
only way to satisfy a dependency.

The "some forward compatibility assumed" version constraint is derived from the
Ruby community's "pessimistic version constraint" operator [2]_ to allow
projects to take a cautious approach to forward compatibility promises, while
still easily setting a minimum required version for their dependencies. The
spelling of the compatible release clause (``~=``) is inspired by the Ruby
(``~>``) and PHP (``~``) equivalents.

Further improvements are also planned to the handling of parallel
installation of multiple versions of the same library, but these will
depend on updates to the installation database definition along with
improved tools for dynamic path manipulation.

The trailing wildcard syntax to request prefix based version matching was
added to make it possible to sensibly define both compatible release clauses
and the desired pre- and post-release handling semantics for ``<`` and ``>``
ordered comparison clauses.


Support for date based version identifiers
------------------------------------------

Excluding date based versions caused significant problems in migrating
``pytz`` to the new metadata standards. It also caused concerns for the
OpenStack developers, as they use a date based versioning scheme and would
like to be able to migrate to the new metadata standards without changing
it.


Adding version epochs
---------------------

Version epochs are added for the same reason they are part of other
versioning schemes, such as those of the Fedora and Debian Linux
distributions: to allow projects to gracefully change their approach to
numbering releases, without having a new release appear to have a lower
version number than previous releases and without having to change the name
of the project.

In particular, supporting version epochs allows a project that was previously
using date based versioning to switch to semantic versioning by specifying
a new version epoch.

The ``!`` character was chosen to delimit an epoch version rather than the
``:`` character, which is commonly used in other systems, due to the fact that
``:`` is not a valid character in a Windows directory name.


Adding direct references
------------------------

Direct references are added as an "escape clause" to handle messy real
world situations that don't map neatly to the standard distribution model.
This includes dependencies on unpublished software for internal use, as well
as handling the more complex compatibility issues that may arise when
wrapping third party libraries as C extensions (this is of especial concern
to the scientific community).

Index servers are deliberately given a lot of freedom to disallow direct
references, since they're intended primarily as a tool for integrators
rather than publishers. PyPI in particular is currently going through the
process of *eliminating* dependencies on external references, as unreliable
external services have the effect of slowing down installation operations,
as well as reducing PyPI's own apparent reliability.


Adding arbitrary equality
-------------------------

Arbitrary equality is added as an "escape clause" to handle the case where
someone needs to install a project which uses a non compliant version. Although
this PEP is able to attain ~97% compatibility with the versions that are
already on PyPI there are still ~3% of versions which cannot be parsed. This
operator gives a simple and effective way to still depend on them without
having to "guess" at the semantics of what they mean (which would be required
if anything other than strict string based equality was supported).


Adding local version identifiers
--------------------------------

It's a fact of life that downstream integrators often need to backport
upstream bug fixes to older versions. It's one of the services that gets
Linux distro vendors paid, and application developers may also apply patches
they need to bundled dependencies.

Historically, this practice has been invisible to cross-platform language
specific distribution tools - the reported "version" in the upstream
metadata is the same as for the unmodified code. This inaccuracy can then
cause problems when attempting to work with a mixture of integrator
provided code and unmodified upstream code, or even just attempting to
identify exactly which version of the software is installed.

The introduction of local version identifiers and "local version labels"
into the versioning scheme, with the corresponding ``python.integrator``
metadata extension allows this kind of activity to be represented
accurately, which should improve interoperability between the upstream
tools and various integrated platforms.

The exact scheme chosen is largely modeled on the existing behavior of
``pkg_resources.parse_version`` and ``pkg_resources.parse_requirements``,
with the main distinction being that where ``pkg_resources`` currently always
takes the suffix into account when comparing versions for exact matches,
the PEP requires that the local version label of the candidate version be
ignored when no local version label is present in the version specifier
clause. Furthermore, the PEP does not attempt to impose any structure on
the local version labels (aside from limiting the set of permitted
characters and defining their ordering).

This change is designed to ensure that an integrator provided version like
``pip 1.5+1`` or ``pip 1.5+1.git.abc123de`` will still satisfy a version
specifier like ``pip>=1.5``.

The plus is chosen primarily for readability of local version identifiers.
It was chosen instead of the hyphen to prevent
``pkg_resources.parse_version`` from parsing it as a prerelease, which is
important for enabling a successful migration to the new, more structured,
versioning scheme. The plus was chosen instead of a tilde because of the
significance of the tilde in Debian's version ordering algorithm.


Providing explicit version normalization rules
----------------------------------------------

Historically, the de facto standard for parsing versions in Python has been the
``pkg_resources.parse_version`` command from the setuptools project. This does
not attempt to reject *any* version and instead tries to make something
meaningful, with varying levels of success, out of whatever it is given. It has
a few simple rules but otherwise it more or less relies largely on string
comparison.

The normalization rules provided in this PEP exist primarily to either increase
the compatability with ``pkg_resources.parse_version``, particularly in
documented use cases such as ``rev``, ``r``, ``pre``, etc or to do something
more reasonable with versions that already exist on PyPI.

All possible normalization rules were weighed against whether or not they were
*likely* to cause any ambiguity (e.g. while someone might devise a scheme where
``v1.0`` and ``1.0`` are considered distinct releases, the likelihood of anyone
actually doing that, much less on any scale that is noticeable, is fairly low).
They were also weighed against how ``pkg_resources.parse_version`` treated a
particular version string, especially with regards to how it was sorted. Finally
each rule was weighed against the kinds of additional versions it allowed, how
"ugly" those versions looked, how hard there were to parse (both mentally and
mechanically) and how much additional compatibility it would bring.

The breadth of possible normalizations were kept to things that could easily
be implemented as part of the parsing of the version and not pre-parsing
transformations applied to the versions. This was done to limit the side
effects of each transformation as simple search and replace style transforms
increase the likelihood of ambiguous or "junk" versions.

For an extended discussion on the various types of normalizations that were
considered, please see the proof of concept for PEP 440 within pip [4]_.


References
==========

The initial attempt at a standardised version scheme, along with the
justifications for needing such a standard can be found in PEP 386.

.. [1] Reference Implementation of PEP 440 Versions and Specifiers
   https://github.com/pypa/packaging/pull/1

.. [2] Version compatibility analysis script:
   https://github.com/pypa/packaging/blob/master/tasks/check.py

.. [3] Pessimistic version constraint
   http://docs.rubygems.org/read/chapter/16

.. [4] Proof of Concept: PEP 440 within pip
    https://github.com/pypa/pip/pull/1894


Appendix A
==========

Metadata v2.0 guidelines versus setuptools::

    $ invoke check.pep440
    Total Version Compatibility:              245806/250521 (98.12%)
    Total Sorting Compatibility (Unfiltered): 45441/47114 (96.45%)
    Total Sorting Compatibility (Filtered):   47057/47114 (99.88%)
    Projects with No Compatible Versions:     498/47114 (1.06%)
    Projects with Differing Latest Version:   688/47114 (1.46%)


---
Donald Stufft
PGP: 7C6B 7C5D 5E2B 6356 A926 F04F 6E3C BCE9 3372 DCFA