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 633 -- Dependency specification in pyproject.toml using an exploded TOML table

PEP:633
Title:Dependency specification in pyproject.toml using an exploded TOML table
Author:Laurie Opperman <laurie_opperman at hotmail.com>, Arun Babu Neelicattu <arun.neelicattu at gmail.com>
Sponsor:Brett Cannon <brett at python.org>
Discussions-To:https://discuss.python.org/t/dependency-specification-in-pyproject-toml-using-an-exploded-toml-table/5123/
Status:Draft
Type:Standards Track
Created:2020-09-02
Post-History:2020-09-02

Abstract

This PEP specifies how to write a project's dependencies in a pyproject.toml file for packaging-related tools to consume using the fields defined in PEP 621, as an alternative to the PEP 508-based approach defined in PEP 631.

Motivation

There are multiple benefits to using TOML tables and other data-types to represent requirements rather then PEP 508 strings:

  • Easy initial validation via the TOML syntax.
  • Easy secondary validation using a schema, for example a JSON Schema [1].
  • Potential for users to guess the keys of given features, rather than memorising a syntax.
  • Users of multiple other popular languages may already be familiar with the TOML syntax.
  • TOML directly represents the same data structures as in JSON, and therefore a sub-set of Python literals, so users can understand the hierarchy and type of value

Rationale

Most of this is taken from discussions in the PEP 621 dependencies topic [2]. This has elements from Pipfile [3], Poetry [4], Dart's dependencies [5] and Rust's Cargo [6]. A comparison document [7] shows advantages and disadvantages between this format and PEP 508-style specifiers.

In the specification of multiple requirements with the same distribution name (where environment markers choose the appropriate dependency), the chosen solution is similar to Poetry [4]'s, where an array of requirements is allowed.

The direct-reference keys closely align with and utilise pep:610 and PEP 440 as to reduce differences in the packaging ecosystem and rely on previous work in specification.

Specification

As in PEP 621, if metadata is improperly specified then tools MUST raise an error. The metadata MUST conform to the TOML [8] specification.

To reduce confusion with this document being a specification for specifying dependencies, the word "requirement" is used to mean a PEP 508 dependency specification.

The following tables are added to the added to the project table specified in PEP 621.

dependencies

Format: table

The keys inside this table are the names of the required distribution. The values can have one of the following types:

  • string: the requirement is defined only by a version requirement, with same specification as version in the requirement table, except allowing the empty string "" to place no restriction on the version.
  • table: a requirement table.
  • array: an array of requirement tables. It is an error to specify an empty array [] as a value.

Requirement table

The keys of the requirement table are as follows (all are optional):

  • version (string): a PEP 440 version specifier, which is a comma- delimited list of version specifier clauses. The string MUST be non-empty.
  • extras (array of strings): a list of PEP 508 extras declarations for the distribution. The list MUST be non-empty.
  • markers (string): a PEP 508 environment marker expression. The string MUST be non-empty.
  • url (string): the URL of the artifact to install and satisfy the requirement. Note that file:// is the prefix used for packages to be retrieved from the local filesystem.
  • git, hg, bzr or svn (string): the URL of a VCS repository (as specified in PEP 440) to clone, whose tree will be installed to satisfy the requirement. Further VCS keys will be added via amendments to PEP 610, however tools MAY opt to support other VCS's using their command-line command prior to the acceptance of the amendment.
  • revision (string): the identifier for a specific revision of the specified VCS repository to check-out before installtion. Users MUST only provide this when one of git, hg, bzr, svn, or another VCS key is used to identify the distribution to install. Revision identifiers are suggested in PEP 610.

At most one of the following keys can be specified simultaneously, as they logically conflict with each other in the requirement: version, url, git, hg, bzr, svn, and any other VCS key.

An empty requirement table {} places no restriction on the requirement, in addition to the empty string "".

Any keys provided which are not specified in this document MUST cause an error in parsing.

optional-dependencies

Format: table

The keys inside this table are the names of an extra's required distribution. The values can have one of the following types:

  • table: a requirement table.
  • array: an array of requirement tables.

These requirement tables have the same specification as above, with the addition of the following required key:

  • for-extra (string): the name of the PEP 508 extra that this requirement is required for.

Reference implementation

Tools will need to convert this format to PEP 508 requirement strings. Below is an example implementation of that conversion (assuming validation is already performed):

def convert_requirement_to_pep508(name, requirement):
    if isinstance(requirement, str):
        requirement = {"version": requirement}
    pep508 = name
    if "extras" in requirement:
        pep508 += " [" + ", ".join(requirement["extras"]) + "]"
    if "version" in requirement:
        pep508 += " " + requirement["version"]
    if "url" in requirement:
        pep508 += " @ " + requirement["url"]
    for vcs in ("git", "hg", "bzr", "svn"):
        if vcs in requirement:
            pep508 += " @ " + vcs + "+" + requirement[vcs]
            if "revision" in requirement:
                pep508 += "@" + requirement["revision"]
    extra = None
    if "for-extra" in requirement:
        extra = requirement["for-extra"]
    if "markers" in requirement:
        markers = requirement["markers"]
        if extra:
            markers = "extra = '" + extra + "' and (" + markers + ")"
        pep508 += "; " + markers
    return pep508, extra


def convert_requirements_to_pep508(dependencies):
    pep508s = []
    extras = set()
    for name, req in dependencies.items():
        if isinstance(req, list):
            for sub_req in req:
                pep508, extra = convert_requirement_to_pep508(name, sub_req)
                pep508s.append(pep508)
                if extra:
                    extras.add(extra)
        else:
            pep508, extra = convert_requirement_to_pep508(name, req)
            pep508s.append(pep508)
            if extra:
                extras.add(extra)
    return pep508s, extras


def convert_project_requirements_to_pep508(project):
    reqs, _ = convert_requirements_to_pep508(project.get("dependencies", {}))
    optional_reqs, extras = convert_requirements_to_pep508(
        project.get("optional-dependencies", {})
    )
    reqs += optional_reqs
    return reqs, extras

JSON schema

For initial validation, a JSON-schema can be used. Not only does this help tools have a consistent validation, but it allows code editors to highlight validation errors as users are building the dependencies list.

{
    "$id": "spam",
    "$schema": "http://json-schema.org/draft-07/schema#",
    "title": "Project metadata",
    "type": "object",
    "definitions": {
        "requirementTable": {
            "title": "Full project dependency specification",
            "type": "object",
            "properties": {
                "extras": {
                    "title": "Dependency extras",
                    "type": "array",
                    "items": {
                        "title": "Dependency extra",
                        "type": "string"
                    }
                },
                "markers": {
                    "title": "Dependency environment markers",
                    "type": "string"
                }
            },
            "propertyNames": {
                "enum": [
                    "extras",
                    "markers",
                    "version",
                    "url",
                    "git",
                    "hg",
                    "bzr",
                    "svn",
                    "for-extra"
                ]
            },
            "oneOf": [
                {
                    "title": "Version requirement",
                    "properties": {
                        "version": {
                            "title": "Version",
                            "type": "string"
                        }
                    }
                },
                {
                    "title": "URL requirement",
                    "properties": {
                        "url": {
                            "title": "URL",
                            "type": "string",
                            "format": "uri"
                        }
                    },
                    "required": [
                        "url"
                    ]
                },
                {
                    "title": "VCS requirement",
                    "properties": {
                        "revision": {
                            "title": "VCS repository revision",
                            "type": "string"
                        }
                    },
                    "oneOf": [
                        {
                            "title": "Git repository",
                            "properties": {
                                "git": {
                                    "title": "Git URL",
                                    "type": "string",
                                    "format": "uri"
                                }
                            },
                            "required": [
                                "git"
                            ]
                        },
                        {
                            "title": "Mercurial repository",
                            "properties": {
                                "hg": {
                                    "title": "Mercurial URL",
                                    "type": "string",
                                    "format": "uri"
                                }
                            },
                            "required": [
                                "hg"
                            ]
                        },
                        {
                            "title": "Bazaar repository",
                            "properties": {
                                "bzr": {
                                    "title": "Bazaar URL",
                                    "type": "string",
                                    "format": "uri"
                                }
                            },
                            "required": [
                                "bzr"
                            ]
                        },
                        {
                            "title": "Subversion repository",
                            "properties": {
                                "svn": {
                                    "title": "Subversion URL",
                                    "type": "string",
                                    "format": "uri"
                                }
                            },
                            "required": [
                                "svn"
                            ]
                        }
                    ]
                }
            ]
        },
        "requirementVersion": {
            "title": "Version project dependency specification",
            "type": "string"
        },
        "requirement": {
            "title": "Project dependency specification",
            "oneOf": [
                {
                    "$ref": "#/definitions/requirementVersion"
                },
                {
                    "$ref": "#/definitions/requirementTable"
                },
                {
                    "title": "Multiple specifications",
                    "type": "array",
                    "items": {
                        "$ref": "#/definitions/requirementTable"
                    },
                    "minLength": 1
                }
            ]
        },
        "optionalRequirementTable": {
            "title": "Project optional dependency specification table",
            "allOf": [
                {
                    "$ref": "#/definitions/requirementTable"
                },
                {
                    "properties": {
                        "for-extra": {
                            "title": "Dependency's extra",
                            "type": "string"
                        }
                    },
                    "required": [
                        "for-extra"
                    ]
                }
            ]
        },
        "optionalRequirement": {
            "title": "Project optional dependency specification",
            "oneOf": [
                {
                    "$ref": "#/definitions/optionalRequirementTable"
                },
                {
                    "title": "Multiple specifications",
                    "type": "array",
                    "items": {
                        "$ref": "#/definitions/optionalRequirementTable"
                    },
                    "minLength": 1
                }
            ]
        }
    },
    "properties": {
        "dependencies": {
            "title": "Project dependencies",
            "type": "object",
            "additionalProperties": {
                "$ref": "#/definitions/requirement"
            }
        },
        "optional-dependencies": {
            "title": "Project dependencies",
            "type": "object",
            "additionalProperties": {
                "$ref": "#/definitions/optionalRequirement"
            }
        }
    }
}

Examples

Full artificial example:

[project.dependencies]
flask = { }
django = { }
requests = { version = ">= 2.8.1, == 2.8.*", extras = ["security", "tests"], markers = "python_version < '2.7'" }
pip = { url = "https://github.com/pypa/pip/archive/1.3.1.zip" }
sphinx = { git = "ssh://git@github.com/sphinx-doc/sphinx.git" }
numpy = "~=1.18"
pytest = [
    { version = "<6", markers = "python_version < '3.5'" },
    { version = ">=6", markers = "python_version >= '3.5'" },
]

[project.optional-dependencies]
pytest-timout = { for-extra = "dev" }
pytest-mock = [
    { version = "<6", markers = "python_version < '3.5'", for-extra = "dev" },
    { version = ">=6", markers = "python_version >= '3.5'", for-extra = "dev" },
]

In homage to PEP 631, the following is an equivalent dependencies specification for docker-compose [9]:

[project.dependencies]
cached-property = ">= 1.2.0, < 2"
distro = ">= 1.2.0, < 2"
docker = { extras = ["ssh"], version = ">= 4.2.2, < 5" }
docopt = ">= 0.6.1, < 1"
jsonschema = ">= 2.5.1, < 4"
PyYAML = ">= 3.10, < 6"
python-dotenv = ">= 0.13.0, < 1"
requests = ">= 2.20.0, < 3"
texttable = ">= 0.9.0, < 2"
websocket-client = ">= 0.32.0, < 1"

# Conditional
"backports.shutil_get_terminal_size" = { version = "== 1.0.0", markers = "python_version < '3.3'" }
"backports.ssl_match_hostname" = { version = ">= 3.5, < 4", markers = "python_version < '3.5'" }
colorama = { version = ">= 0.4, < 1", markers = "sys_platform == 'win32'" }
enum34 = { version = ">= 1.0.4, < 2", markers = "python_version < '3.4'" }
ipaddress = { version = ">= 1.0.16, < 2", markers = "python_version < '3.3'" }
subprocess32 = { version = ">= 3.5.4, < 4", markers = "python_version < '3.2'" }

[project.optional-dependencies]
PySocks = { version = ">= 1.5.6, != 1.5.7, < 2", for-extra = "socks" }
ddt = { version = ">= 1.2.2, < 2", for-extra = "tests" }
pytest = { version = "< 6", for-extra = "tests" }
mock = { version = ">= 1.0.1, < 4", markers = "python_version < '3.4'", for-extra = "tests" }

Compatibility Examples

The authors of this PEP recognise that various tools need to both read from and write to this format for dependency specification. This section aims to provide direct comparison with and examples for translating to/from the currently used standard, PEP 508.

Note

For simplicity and clarity, various ways in which TOML allows you to specify each specification is not represented. These examples use the standard inline representation.

For example, while following are considered equivalent in TOML, we choose the second form for the examples in this section.

aiohttp.version = "== 3.6.2"
aiohttp = { version = "== 3.6.2" }

Version Constrained Dependencies

No Version Constraint

aiohttp
aiohttp = {}

Simple Version Constraint

aiohttp >= 3.6.2, < 4.0.0
aiohttp = { version = ">= 3.6.2, < 4.0.0" }

Note

This can, for conciseness, be also represented as a string.

aiohttp = ">= 3.6.2, < 4.0.0"

Direct Reference Dependencies

URL Dependency

aiohttp @ https://files.pythonhosted.org/packages/97/d1/1cc7a1f84097d7abdc6c09ee8d2260366f081f8e82da36ebb22a25cdda9f/aiohttp-3.6.2-cp35-cp35m-macosx_10_13_x86_64.whl
aiohttp = { url = "https://files.pythonhosted.org/packages/97/d1/1cc7a1f84097d7abdc6c09ee8d2260366f081f8e82da36ebb22a25cdda9f/aiohttp-3.6.2-cp35-cp35m-macosx_10_13_x86_64.whl" }

VCS Dependency

aiohttp @ git+ssh://git@github.com/aio-libs/aiohttp.git@master
aiohttp = { git = "ssh://git@github.com/aio-libs/aiohttp.git", revision = "master" }

Environment Markers

aiohttp >= 3.6.1; python_version >= '3.8'
aiohttp = { version = ">= 3.6.1", markers = "python_version >= '3.8'" }

A slightly extended example of the above, where a particular version of aiohttp is required based on the interpreter version.

aiohttp >= 3.6.1; python_version >= '3.8'
aiohttp >= 3.0.0, < 3.6.1; python_version < '3.8'
aiohttp = [
    { version = ">= 3.6.1", markers = "python_version >= '3.8'" },
    { version = ">= 3.0.0, < 3.6.1", markers = "python_version < '3.8'" }
]

Package Extras

Specifying dependency for a package extra

aiohttp >= 3.6.2; extra == 'http'
aiohttp = { version = ">= 3.6.2", for-extra = "http" }

Using extras from a dependency

aiohttp [speedups] >= 3.6.2
aiohttp = { version = ">= 3.6.2", extras = ["speedups"] }

Complex Examples

Version Constraint

aiohttp [speedups] >= 3.6.2; python_version >= '3.8' and extra == 'http'
aiohttp = { version = ">= 3.6.2", extras = ["speedups"], markers = "python_version >= '3.8'", for-extra = "http" }

Direct Reference (VCS)

aiohttp [speedups] @ git+ssh://git@github.com/aio-libs/aiohttp.git@master ; python_version >= '3.8' and extra == 'http'
aiohttp = { git = "ssh://git@github.com/aio-libs/aiohttp.git", revision = "master", extras = ["speedups"], markers = "python_version >= '3.8'", for-extra = "http" }

Rejected Ideas

Switch to an array for dependencies

Use an array instead of a table in order to have each element only be a table (with a name key) and no arrays of requirement tables. This was very verbose and restrictive in the TOML format, and having multiple requirements for a given distribution isn't very common.

Replace optional-dependencies with extras

Remove the optional-dependencies table in favour of both including an optional key in the requirement and an extras table which specifies which (optional) requirements are needed for a project's extra. This reduces the number of table with the same specification (to 1) and allows for requirements to be specified once but used in multiple extras, but distances some of the requirement's properties (which extra(s) it belongs to), groups required and optional dependencies together (possibly mixed), and there may not be a simple way to choose a requirement when a distribution has multiple requirements. This was rejected as optional-dependencies has already been used in the PEP 621 draft.

direct table in requirement

Include the direct-reference keys in a direct table, have the VCS specified as the value of a vcs key. This was more explicit and easier to include in a JSON-schema validation, but was decided to be too verbose and not as readable.

Include hash

Include hash in direct-reference requirements. This was only for package lock-files, and didn't really have a place in the project's metadata.

Dependency tables for each extra

Have the optional-dependencies be a table of dependency tables for each extra, with the table name being the extra's name. This made optional-dependencies a different type (table of tables of requirements) from dependencies (table of requirements), which could be jarring for users and harder to parse.

Environment marker keys

Make each PEP 508 environment marker as a key (or child-table key) in the requirement. This arguably increases readability and ease of parsing. The markers key would still be allowed for more advanced specification, with which the key-specified environment markers are and'd with the result of. This was deferred as more design needs to be undertaken.

Multiple extras which one requirement can satisfy

Replace the for-extra key with for-extras, with the value being an array of extras which the requirement satisfies. This reduces some duplication, but in this case that duplication makes explicit which extras have which dependencies.

Source: https://github.com/python/peps/blob/master/pep-0633.rst