python-peps/pep-0700.rst

197 lines
8.9 KiB
ReStructuredText

PEP: 700
Title: Additional Fields for the Simple API for Package Indexes
Author: Paul Moore <p.f.moore@gmail.com>
PEP-Delegate: Donald Stufft <donald@stufft.io>
Discussions-To: https://discuss.python.org/t/pep-700-additional-fields-for-the-simple-api-for-package-indexes/20177
Status: Draft
Type: Standards Track
Topic: Packaging
Content-Type: text/x-rst
Created: 21-Oct-2022
Post-History: `21-Oct-2022 <https://discuss.python.org/t/pep-700-additional-fields-for-the-simple-api-for-package-indexes/20177>`__
Abstract
========
:pep:`691` defined a JSON form for the "Simple Repository API". This allowed
clients to more easily query the data that was previously only available in
HTML, as defined in :pep:`503`.
This proposal adds three fields to the JSON form, which allow it to be used in
place of PyPI's `JSON API <https://warehouse.pypa.io/api-reference/json.html>`__
in a number of situations.
- A field to allow retrieval of a list of all the published versions of a project.
- Fields containing the size and upload time for a project file.
Rationale
=========
With the introduction of the JSON form of the simple API in :pep:`691`, the
simple API offers functionality that is almost as complete as the PyPI JSON API.
This PEP adds a number of fields which were previously only available through
the JSON API, in order to allow more clients which were previously Warehouse
specific to support arbitrary standards-compliant indexes.
Specification
=============
This specification defines version 1.1 of the simple repository API. For the
HTML version of the API, there is no change from version 1.0. For the JSON
version of the API, the following changes are made:
- The ``api-version`` must specify version 1.1 or later.
- A new ``versions`` key is added at the top level.
- Two new "file information" keys, ``size`` and ``upload-time``, are added to
the ``files`` data.
- Keys (at any level) named with a leading underscore are reserved as private
for index server use. No future standard will assign a meaning to any such key.
All new fields are mandatory (although ``upload-time`` may be ``None`` if the
server does not store that information).
Versions
--------
An additional key, ``versions`` MUST be present at the top level, in addition to
the keys ``name``, ``files`` and ``meta`` defined in :pep:`691`. This key MUST
contain a list of version strings specifying all of the project versions uploaded
for this project. The value is logically a set, and as such may not contain
duplicates, and the order of the values is not significant.
All of the files listed in the ``files`` key MUST be associated with one of the
versions in the ``versions`` key. The ``versions`` key MAY contain versions with
no associated files (to represent versions with no files uploaded, if the server
has such a concept).
Note that because servers may hold "legacy" data from before the adoption of
:pep:`440`, version strings currently cannot be required to be valid :pep:`440`
versions, and therefore cannot be assumed to be orderable using the :pep:`440`
rules. However, servers SHOULD use normalised :pep:`440` versions where
possible.
Additional file information
---------------------------
The elements of the ``files`` key MUST contain two additional sub-keys, ``size``
and ``upload-time``. The ``size`` field MUST contain an integer which is the
file size in bytes, and the ``upload-time`` field MUST contain either ``null``
(see below) or a valid ISO 8601 date/time string, in the format
``yyyy-mm-ddThh:mm:ss.ffffffZ``, which represents the time the file was uploaded
to the index. As indicated by the ``Z`` suffix, the upload time MUST use the UTC
timezone. The fractional seconds part of the timestamp (the ``.ffffff`` part) is
optional, and if present may contain up to 6 digits of precision.
If a server does not record upload time information for a file, it MAY return
a value for ``upload-time`` of ``None`` (``null`` in JSON). However, it MUST
NOT omit the key.
FAQ
===
Why not add this data to the HTML API as well?
----------------------------------------------
It would be possible to add the data to the HTML API, but the vast majority of
consumers for this data are likely to be currently getting it from the PyPI JSON
API, and so will already be expecting to parse JSON. Traditional consumers of
the HTML API have never needed this data previously.
Does this imply that the HTML API is obsolete?
----------------------------------------------
No. The FAQ of :pep:`691` was clear that the HTML API is not being deprecated,
and this PEP does not change that position. However, clients wishing to access
the new data introduced by this PEP will need to use the JSON API to get it. And
indexes wanting to provide it will need to serve the JSON format.
Is the simple API replacing the Warehouse JSON and XML-RPC APIs?
----------------------------------------------------------------
Where possible, clients should prefer the simple API over the JSON or XML-RPC
APIs, as the former is standardised and can be assumed to be available from any
index, whereas the latter are exclusive to the Warehouse project.
However, while this PEP brings the simple API closer to being able to replace
the JSON API, there is no formal policy that the simple API will replicate all
of the functionality covered by the existing Warehouse APIs. Proposed additions
to the simple API will still be considered on their individual merits, and the
requirement that the API should be simple and fast for the primary use case of
locating files for a project will remain the overriding consideration.
Why not allow other date formats?
---------------------------------
The ISO 8601 standard is complex, and there seems little value in requiring
clients to deal with that. The standard library ``datetime`` module provides
methods to parse ISO 8601 strings, but it is possible that users may want to
access index data *without* using Python (for example, piping the output of
``curl`` into ``jq``). Having a single, well-defined format makes this easier,
and doesn't have any significant disadvantages.
What if file sizes are too big for a JSON number?
-------------------------------------------------
The JSON standard does not specify how numbers are to be interpreted. Python can
read and write arbitrary-length integers in a JSON file, so this should not be
an issue for code written in Python. Non-Python implementations may need to take
care to handle large integers correctly, but this is not expected to be a
significant problem.
Why is it not possible to serve some of the new information, but not all?
-------------------------------------------------------------------------
The versioning scheme for the simple API does not allow servers to "mix and
match" what information they provide, except by making fields optional, and
handling optional fields is an inconvenience for clients using the API. The
PEP takes the view that ease of use for clients should be the priority (as
there will always be more clients than servers).
The main problem for servers is likely to be the upload time, as some servers
may not retain this information. For this situation, a server is allowed to
return a null value for the upload time, indicating that the value is not known.
Version and file size information do not have a "not known" option, as they
should not need them. Versions can be determined from the file names if
necessary, and file size information is needed to serve the files, so it should
always be possible for the server to determine it.
Why not require PEP 440 versions?
---------------------------------
At the time this PEP was written, PyPI still contains (and serves) projects and
files with "legacy" versions. Requiring :pep:`440` versions would make it
impossible for PyPI to follow this specification while still serving the
existing content.
Ideally, at some future time, the simple index API will be updated to require
:pep:`440` versions, at which time this specification should be updated to
reflect that. However, that change will need to be co-ordinated with existing
index providers including PyPI, to desupport and remove non-conforming projects
and/or files.
Why not provide a "latest version" value?
-----------------------------------------
For :pep:`440` versions, this is easy enough for the client to do (using the
``packaging`` library, ``latest = max(Version(s) for s in proj["versions"])``).
For non-standard versions, there is no well-defined ordering, and clients will
need to decide on what rule is appropriate for their needs. Requiring the server
to supply a latest version value takes the choice away from the client.
Servers with an explicit concept of which version is the "latest", which cannot
be calculated from data available to the client, can provide a non-standard,
underscore-prefixed key to convey that information to the client if they wish.
Copyright
=========
This document is placed in the public domain or under the
CC0-1.0-Universal license, whichever is more permissive.