Notice: While Javascript is not essential for this website, your interaction with the content will be limited. Please turn Javascript on for the full experience.

PEP 639 -- Metadata for Python Software Packages 2.2

Title:Metadata for Python Software Packages 2.2
Author:Philippe Ombredanne <pombredanne at>
Sponsor:Paul Moore <p.f.moore at>
BDFL-Delegate:Paul Moore <p.f.moore at>
Type:Standards Track


This PEP describes the changes between versions 2.1 and 2.2 of the Core Metadata Specification [1] for Python packages. Version 2.1 is specified in PEP 566.

The primary change introduced in this PEP updates how licenses are documented in core metadata via the License field with license expression strings using SPDX license identifiers [6] such that license documentation is simpler and less ambiguous:

  • for package authors to create,
  • for package users to read and understand, and,
  • for tools to process package license information mechanically.

The other changes include:

  • specifying a License-File field which is already used by wheel and setuptools to include license files in built distributions.
  • defining how tools can validate license expressions and report warnings to users for invalid expressions (but still accept any string as License).


This PEP's scope is limited strictly to how we document the license of a distribution:

  • with an improved and structured way to document a license expression, and,
  • by including license texts in a built package.

The core metadata specification updates that are part of this PEP have been designed to have minimal impact and to be backward compatible with v2.1. These changes utilize emerging new ways to document licenses that are already in use in some tools (e.g. by adding the License-File field already used in wheel and setuptools) or by some package authors (e.g. storing an SPDX license expression in the existing License field).

In addition to an update to the metadata specification, this PEP contains:

  • recommendations for publishing tools on how to validate the License and Classifier fields and report informational warnings when a package uses an older, non-structured style of license documentation conventions.
  • informational appendices that contain surveys of how we document licenses today in Python packages and elsewhere, and a reference Python library to parse, validate and build correct license expressions.

It is the intent of the PEP authors to work closely with tool authors to implement the recommendations for validation and warnings specified in this PEP.


This PEP is neutral regarding the choice of license by any package author.

In particular, the SPDX license expression syntax proposed in this PEP provides simpler and more expressive conventions to document accurately any kind of license that applies to a Python package, whether it is an open source license, a free or libre software license, a proprietary license, or a combination of such licenses.

This PEP makes no recommendation for specific licenses and does not require the use of specific license documentation conventions. This PEP also does not impose any restrictions when uploading to PyPI.

Instead, this PEP is intended to document common practices already in use, and recommends that publishing tools should encourage users via informational warnings when they do not follow this PEP's recommendations.

This PEP is not about license documentation in files inside packages, even though this is a surveyed topic in the appendix.

Possible future PEPs

It is the intention of the authors of this PEP to consider the submission of related but separate PEPs in the future such as:

  • make License and new License-File fields mandatory including stricter enforcement in tools and PyPI publishing.
  • require uploads to PyPI to use only FOSS (Free and Open Source software) licenses.


Software is licensed, and providing accurate licensing information to Python packages users is an important matter. Today, there are multiple places where licenses are documented in package metadata and there are limitations to what can be documented. This is often leading to confusion or a lack of clarity both for package authors and package users.

Several package authors have expressed difficulty and/or frustrations due to the limited capabilities to express licensing in package metadata. This also applies to Linux and BSD* distribution packagers. This has triggered several license-related discussions and issues, in particular:

On average, Python packages tend to have more ambiguous, or missing, license information than other common application package formats (such as npm, Maven or Gem) as can be seen in the statistics [2] page of the ClearlyDefined [3] project that cover all packages from PyPI, Maven, npm and Rubygems. ClearlyDefined is an open source project to help improve clarity of other open source projects that is incubating at the OSI (Open Source Initiative) [4].


A mini-survey of existing license metadata definitions in use in the Python ecosystem today and documented in several other system/distro and application package formats is provided in Appendix 2 of this PEP.

There are a few takeaways from the survey:

  • Most package formats use a single License field.
  • Many modern package formats use some form of license expression syntax to optionally combine more than one license identifier together. SPDX and SPDX-like syntaxes are the most popular in use.
  • SPDX license identifiers are becoming a de facto way to reference common licenses everywhere, whether or not a license expression syntax is used.
  • Several package formats support documenting both a license expression and the paths of the corresponding files that contain the license text. Most free and open source software licenses require package authors to include their full text in a distribution.

These considerations have guided the design and recommendations of this PEP.

The reuse of the License field with license expressions will provide an intuitive and more structured way to express the license of a distribution using a well-defined syntax and well-known license identifiers.

Over time, recommending the usage of these expressions will help Python package publishers improve the clarity of their license documentation to the benefit of package authors, consumers and redistributors.

Core Metadata Specification updates

The canonical source for the names and semantics of each of the supported metadata fields is the Core Metadata Specification [1] document.

The details of the updates considered to the Core Metadata Specification [1] document as part of this PEP are described here and will be added to the canonical source once this PEP is approved.

Added in Version 2.2

License-File (multiple use)

The License-File is a string that is a path, relative to``.dist-info``, to a license file. The license file content MUST be UTF-8 encoded text.

Build tools SHOULD honor this field and include the corresponding license file(s) in the built package.

Changed in Version 2.2

License (optional)

Text indicating the license covering the distribution. This text can be either a valid license expression as defined here or any free text.

Publishing tools SHOULD issue an informational warning if this field is empty, missing, or is not a valid license expression as defined here. Build tools MAY issue a similar warning.

License Expression syntax

A license expression is a string using the SPDX license expression syntax as documented in the SPDX specification [7] using either Version 2.2 [8] or a later compatible version. SPDX is a working group at the Linux Foundation that defines a standard way to exchange package information.

When used in the License field and as a specialization of the SPDX license expression definition, a license expression can use the following license identifiers:

  • any SPDX-listed license short-form identifiers that are published in the SPDX License List [6] using either Version 3.10 or any later compatible version. Note that the SPDX working group never removes any license identifiers: instead they may choose to mark an identifier as "deprecated".
  • the LicenseRef-Public-Domain and LicenseRef-Proprietary strings to identify licenses that are not included in the SPDX license list.

When processing the License field to determine if it contains a valid license expression, tools:

  • SHOULD report an informational warning if one or more of the following applies:
    • the field does not contain a license expression,
    • the license expression syntax is invalid,
    • the license expression syntax is valid but some license identifiers are unknown as defined here or the license identifiers have been marked as deprecated in the SPDX License List [6]
  • SHOULD store a case-normalized version of the License field using the reference case for each SPDX license identifier and uppercase for the AND, OR and WITH keywords.
  • SHOULD report an informational warning if normalization process results in changes to the License field contents.

License expression examples:

License: MIT

License: BSD-3-Clause

License: MIT OR GPL-2.0-or-later OR (FSFUL AND BSD-2-Clause)

License: GPL-3.0-only WITH Classpath-Exception-2.0 OR BSD-3-Clause

License: This software may only be obtained by sending the
        author a postcard, and then the user promises not
        to redistribute it.

License: LicenseRef-Proprietary AND LicenseRef-Public-Domain

Classifier (multiple use)

Each entry is a string giving a single classification value for the distribution. Classifiers are described in PEP 301.


Classifier: Development Status :: 4 - Beta
Classifier: Environment :: Console (Text Based)

Tools SHOULD issue an informational warning if this field contains a licensing- related classifier string starting with the License :: prefix and SHOULD suggest the use of a license expression in the License field instead.

If the License field is present and contains a valid license expression, publishing tools MUST NOT also provide any licensing-related classifier entries [5].

However, for compatibility with existing publishing and installation processes, licensing-related classifier entries SHOULD continue to be accepted if the License field is absent or does not contain a valid license expression.

Publishing tools MAY infer a license expression from the provided classifier entries if they are able to do so unambiguously.

However, no new licensing related classifiers will be added; anyone requesting them will be directed to use a license expression in the License field instead. Note that the licensing-related classifiers may be deprecated in a future PEP.

Mapping Legacy Classifiers to New License Expressions

Publishing tools MAY infer or suggest an equivalent license expression from the provided License or Classifier information if they are able to do so unambiguously. For instance, if a package only has this license classifier:

Classifier: License :: OSI Approved :: MIT License

Then the corresponding value for a License field using a valid license expression to suggest would be:

License: MIT

Here are mapping guidelines for the legacy classifiers:

  • Classifier License :: Other/Proprietary License becomes License: LicenseRef-Proprietary expression.
  • Classifier License :: Public Domain becomes License: LicenseRef-Public-Domain expression, though tools should encourage the use of more explicit and legally portable license identifiers such as CC0-1.0 [67], the Unlicense [68] since the meaning associated with the term "public domain" is thoroughly dependent on the specific legal jurisdiction involved and some jurisdictions have no concept of Public Domain as it exists in the USA.
  • The generic and ambiguous classifiers License :: OSI Approved and License :: DFSG approved do not have an equivalent license expression.
  • The generic and sometimes ambiguous classifiers License :: Free For Educational Use, License :: Free For Home Use, License :: Free for non-commercial use, License :: Freely Distributable, License :: Free To Use But Restricted, and License :: Freeware are mapped to the generic License: LicenseRef-Proprietary expression.
  • Classifiers License :: GUST* have no mapping to SPDX license identifierss for now and no package uses them in PyPI as of the writing of this PEP.

The remainder of the classifiers using a License :: prefix map to a simple single-identifier license expression using the corresponding SPDX license identifiers.

When multiple license-related classifiers are used, their relation is ambiguous and it is typically not possible to determine if all the licenses apply or if there is a choice that is possible among the licenses. In this case, tools cannot reliably infer a license expression and should suggest that the package author construct a license expression which expresses their intent.

Summary of Differences From PEP 566

  • Metadata-Version is now 2.2.
  • Added one new field: License-File
  • Updated the documentation of two fields: License and Classifier

Backwards Compatibility

The reuse of the License field means that we keep backward compatibility. The specification of the License-File field is only writing down the practices of the wheel and setuptools tools and is backward compatible with their support for that field.

The "soft" validation of the License field when it does not contain a valid license expression and when the Classifier field is used with legacy license-related classifiers means that we can gently prepare users for possible strict and incompatible validation of these fields in the future.

Security Implications

This PEP has no foreseen security implications: the License field is a plain string and the License-File(s) are file paths. None of them introduces any new security concern.

How to Teach Users to Use License Expressions

The simple cases are simple: a single license identifier is a valid license expression and a large majority of packages use a single license.

The plan to teach users of packaging tools how to express their package's license with a valid license expression is to have tools issue informative messages when they detect invalid license expressions or when a license-related classifier is used in the Classifier field.

With a warning message that does not terminate processing, publishing tools will gently teach users how to provide correct license expressions over time.

Tools may also help with the conversion and suggest a license expression in some cases:

  1. The section Mapping Legacy Classifiers to New License expressions provides tool authors with guidelines on how to suggest a license expression produced from legacy classifiers.
  2. Tools may also be able to infer and suggest how to update an existing incorrect License value and convert that to a correct license expression. For instance a tool may suggest to correct a License field from Apache2 (which is not a valid license expression as defined in this PEP) to Apache-2.0 (which is a valid license expression using an SPDX license identifier as defined in this PEP).

Reference Implementation

Tools will need to support parsing and validating license expressions in the License field.

The license-expression library [11] is a reference Python implementation of a library that handles license expressions including parsing, validating and formatting license expressions using flexible lists of license symbols (including SPDX license identifiers and any extra identifiers referenced here). It is licensed under the Apache-2.0 license and is used in a few projects such as the SPDX Python tools [12], the ScanCode toolkit [13] and the Free Software Foundation Europe (FSFE) Reuse project [10].

Rejected ideas

  1. Use a new License-Expression field and deprecate the License field.

Adding a new field would introduce backward incompatible changes when the License field would be retired later and require having more complex validation. The use of such a field would further introduce a new concept that is not seen anywhere else in any other package metadata (e.g. a new field only for license expression) and possibly be a source of confusion. Also, users are less likely to start using a new field than make small adjustments to their use of existing fields.

  1. Mapping licenses used in the license expression to specific files in the license files (or vice versa).

This would require using a mapping (two parallel lists would be too prone to alignment errors) and a mapping would bring extra complication to how license are documented by adding an additional nesting level.

A mapping would be needed as you cannot guarantee that all expressions (e.g. GPL with an exception may be in a single file) or all the license keys have a single license file and that any expression does not have more than one. (e.g. an Apache license LICENSE and its NOTICE file for instance are two distinct files). Yet in most cases, there is a simpler "one license", "one or more license files". In the rarer and more complex cases where there are many licenses involved you can still use the proposed conventions at the cost of a slight loss of clarity by not specifying which text file is for which license identifier, but you are not forcing the more complex data model (e.g. a mapping) on everyone that may not need it.

We could of course have a data field with multiple possible value types (it’s a string, it’s a list, it’s a mapping!) but this could be a source of confusion. This is what has been done for instance in npm (historically) and in Rubygems (still today) and as result you need to test the type of the metadata field before using it in code and users are confused about when to use a list or a string.

  1. Mapping licenses to specific source files and/or directories of source files (or vice versa).

File-level notices are not considered as part of the scope of this PEP and the existing SPDX-License-Identifier [59] convention can be used and may not need further specification as a PEP.

Appendix 1. License Expression example

The current version of setuptools metadata [20] does not use the License field. It uses instead this license-related information in setup.cfg:

license_file = LICENSE
classifiers =
    License :: OSI Approved :: MIT License

The simplest migration to this PEP would consist of using this instead:

license = MIT
license_files =

Another possibility would be to include the licenses of the third-party packages that are vendored in the setuptools/_vendor/ and pkg_resources/_vendor directories:


These license expressions for these packages are:

appdirs: MIT
packaging: Apache-2.0 OR BSD-2-Clause
pyparsing: MIT
ordered-set: MIT

Therefore, a comprehensive license expression covering both setuptools proper and its vendored packages could contain these metadata, combining all the license expressions in one expression:

license = MIT AND (Apache-2.0 OR BSD-2-Clause)
license_files =

Here we would assume that the LICENSE.MIT file contains the text of the MIT license and the copyrights used by setuptools, appdirs, pyparsing and ordered-set, and that the LICENSE.packaging file contains the texts of the Apache and BSD license, its copyrights and its license choice notice [21].

Appendix 2. Surveying how we document licenses today in Python

There are multiple ways used or recommended to document Python package licenses today:

In Core metadata

There are two overlapping core metadata fields to document a license: the license-related Classifier strings [5] prefixed with License :: and the License field as free text [14].

The core metadata documentation License field documentation is currently:

License (optional)

Text indicating the license covering the distribution where the license
is not a selection from the "License" Trove classifiers. See
"Classifier" below.  This field may also be used to specify a
particular version of a license which is named via the ``Classifier``
field, or to indicate a variation or exception to such a license.


    License: This software may only be obtained by sending the
            author a postcard, and then the user promises not
            to redistribute it.

    License: GPL version 3, excluding DRM provisions

Even though there are two fields, it is at times difficult to convey anything but simpler licensing. For instance some classifiers lack accuracy (GPL without a version) and when you have multiple License-related classifiers it is not clear if this is a choice or all these apply and which ones. Furthermore, the list of available license-related classifiers is often out-of-date.

In the PyPA sampleproject

The latest PyPA sampleproject recommends only to use classifiers in and does not list the license field in its example [15].

The License Files in wheels and setuptools

Beyond a license code or qualifier, license text files are documented and included in a built package either implicitly or explicitly and this is another possible source of confusion:

  • In wheels [9] license files are automatically added to the .dist-info directory if they match one of a few common license file name patterns (such as LICENSE*, COPYING*). Alternatively a package author can specify a list of license file paths to include in the built wheel using in the license_files field in the [metadata] section of the project's setup.cfg. Previously this was a (singular) license_file file attribute that is now deprecated but is still in common use. See [16] for instance.
  • In setuptools [17], a license_file attribute is used to add a single license file to a source distribution. This singular version is still honored by wheels for backward compatibility.
  • Using a LICENSE.txt file is encouraged in the packaging guide [18] paired with a entry to ensure that the license file is included in a built source distribution (sdist).

Note: the License-File field proposed in this PEP already exists in wheel and setuptools with the same behaviour as explained above. This PEP is only recognizing and documenting the existing practice as used in wheel (with the license_file and license_files setup.cfg [metadata] entries) and in setuptools license_file setup() argument.

In Python code files

(Note: Documenting licenses in source code is not in the scope of this PEP)

Beside using comments and/or SPDX-License-Identifier conventions, the license is sometimes documented in Python code files using "dunder" variables typically named after one of the lower cased Core Metadata fields such as __license__ [19].

This convention (dunder global variables) is recognized by the built-in help() function and the standard pydoc module. The dunder variable(s) will show up in the help() DATA section for a module.

In some other Python packaging tools

  • Conda package manifest [22] has support for license and license_file fields as well as a license_family license grouping field.
  • flit [23] recommends to use classifiers instead of License (as per the current metadata spec).
  • pbr [25] uses similar data as setuptools but always stored setup.cfg.
  • poetry [24] specifies the use of the license field in pyproject.toml with SPDX license identifiers.

Appendix 3. Surveying how other package formats document licenses

Here is a survey of how things are done elsewhere.

License in Linux distribution packages

Note: in most cases the license texts of the most common licenses are included globally once in a shared documentation directory (e.g. /usr/share/doc).

  • Debian document package licenses with machine readable copyright files [26]. This specification defines its own license expression syntax that is very similar to the SDPX syntax and use its own list of license identifiers for common licenses (also closely related to SPDX identifiers).
  • Fedora packages [27] specify how to include License Texts [28] and how use a License field [29] that must be filled with an appropriate license Short License identifier(s) from an extensive list of "Good Licenses" identifiers [30]. Fedora also defines its own license expression syntax very similar to the SDPX syntax.
  • openSUSE packages [31] use SPDX license expressions with SPDX license identifiers and a list of extra license identifiers [32].
  • Gentoo ebuild uses a LICENSE variable [33]. This field is specified in GLEP-0023 [34] and in the Gentoo development manual [35]. Gentoo also defines a license expression syntax and a list of allowed licenses. The expression syntax is rather different from SPDX.
  • FreeBSD package Makefile [36] provides LICENSE and LICENSE_FILE fields with a list of custom license symbols. For non-standard licenses, FreeBSD recommend to use LICENSE=UNKNOWN and add LICENSE_NAME and LICENSE_TEXT fields, as well as sophisticated LICENSE_PERMS to qualify the license permissions and LICENSE_GROUPS to document a license grouping. The LICENSE_COMB allows to document more than one license and how they apply together, forming a custom license expression syntax. FreeBSD also recommends the use of SPDX-License-Identifier in source code files.
  • Archlinux PKGBUILD [37] define its own license identifiers [38]. The value 'unknown' can be used if the license is not defined.
  • OpenWRT ipk packages [39] use the PKG_LICENSE and PKG_LICENSE_FILES variables and recommend the use of SPDX License identifiers.
  • NixOS uses SPDX identifiers [40] and some extra license identifiers in its license field.
  • GNU Guix (based on NixOS) has a single License field, uses its own license symbols list [41] and specifies to use one license or a list of licenses [42].
  • Alpine Linux packages [43] recommend using SPDX identifiers in the license field.

License in Language and Application packages

  • In Java, Maven POM [44] defines a licenses XML tag with a list of license items each with a name, URL, comments and "distribution" type. This is not mandatory and the content of each field is not specified.
  • JavaScript npm package.json [45] use a single license field with SPDX license expression or the UNLICENSED id if no license is specified. A license file can be referenced as an alternative using "SEE LICENSE IN <filename>" in the single license field.
  • Rubygems gemspec [46] specifies either a singular license string or a list of license strings. The relationship between multiple licenses in a list is not specified. They recommend using SPDX license identifiers.
  • CPAN Perl modules [47] use a single license field which is either a single string or a list of strings. The relationship between the licenses in a list is not specified. There is a list of custom license identifiers plus these generic identifiers: open_source, restricted, unrestricted, unknown.
  • Rust Cargo [48] specifies the use of an SPDX license expression (v2.1) in the license field. It also supports an alternative expression syntax using slash-separated SPDX license identifiers. There is also a license_file field. The package registry [49] requires that either license or license_file fields are set when you upload a package.
  • PHP Composer composer.json [50] uses a license field with an SPDX license id or "proprietary". The license field is either a single string that can use something which resembles the SPDX license expression syntax with "and" and "or" keywords; or is a list of strings if there is a choice of licenses (aka. a "disjunctive" choice of license).
  • NuGet packages [51] were using only a simple license URL and are now specifying to use an SPDX License expression and/or the path to a license file within the package. The repository states that they only accepts license expressions that are approved by the Open Source Initiative or the Free Software Foundation.
  • Go language modules go.mod have no provision for any metadata beyond dependencies. Licensing information is left for code authors and other community package managers to document.
  • Dart/Flutter spec [52] recommends to use a single LICENSE file that should contain all the license texts each separated by a line with 80 hyphens.
  • JavaScript Bower [53] license field is either a single string or a list of strings using either SPDX license identifiers, or a path or a URL to a license file.
  • Cocoapods podspec [54] license field is either a single string or a mapping with attributes of type, file and text keys. This is mandatory unless there is a LICENSE or LICENCE file provided.
  • Haskell Cabal [55] accepts an SPDX license expression since version 2.2. The version of the SPDX license list used is a function of the cabal version. The specification also provides a mapping between pre-SPDX Legacy license Identifiers and SPDX identifiers. Cabal also specifies a license-file(s) field that lists license files that will be installed with the package.
  • Erlang/Elixir mix/hex package [56] specifies a licenses field as a required list of license strings and recommends to use SPDX license identifiers.
  • D lang dub package [57] defines its own list of license identifiers and its own license expression syntax and both are similar to the SPDX conventions.
  • R Package DESCRIPTION [58] defines its own sophisticated license expression syntax and list of licenses identifiers. R has a unique way to support specifiers for license versions such as LGPL (>= 2.0, < 3) in its license expression syntax.

Conventions used by other ecosystems

  • SPDX-License-Identifier [59] is a simple convention to document the license inside a file.
  • The Free Software Foundation (FSF) promotes the use of SPDX license identifiers for clarity in the GPL and other versioned free software licenses [60] [61].
  • The Free Software Foundation Europe (FSFE) REUSE project [10] promotes using SPDX-License-Identifier.
  • The Linux kernel uses SPDX-License-Identifier and parts of the FSFE REUSE conventions to document its licenses [62].
  • U-Boot spearheaded using SPDX-License-Identifier in code and now follows the Linux ways [63].
  • The Apache Software Foundation projects use RDF DOAP [64] with a single license field pointing to SPDX license identifiers.
  • The Eclipse Foundation promotes using SPDX-license-Identifiers [65]
  • The ClearlyDefined project [3] promotes using SPDX license identifiers and expressions to improve license clarity.
  • The Android Open Source Project [66] use MODULE_LICENSE_XXX empty tag files where XXX is a license code such as BSD, APACHE, GPL, etc. And side by side with this MODULE_LICENSE file there is a NOTICE file that contains license and notices texts.


This document specifies version 2.2 of the metadata format.

  • Version 1.0 is specified in PEP 241.
  • Version 1.1 is specified in PEP 314.
  • Version 1.2 is specified in PEP 345.
  • Version 2.0, while not formally accepted, was specified in PEP 426.
  • Version 2.1 is specified in PEP 566.
[1](1, 2, 3)
[3](1, 2)
[5](1, 2)
[6](1, 2, 3)
[10](1, 2)
[59](1, 2)
[67](1, 2)


  • Nick Coghlan
  • Kevin P. Fleming
  • Pradyun Gedam
  • Oleg Grenrus
  • Dustin Ingram
  • Chris Jerdonek
  • Cyril Roelandt
  • Luis Villa