PEP 699 – Remove private dict version field added in PEP 509
- Ken Jin <kenjin at python.org>
- Discourse thread
- Standards Track
PEP 509 introduced a private
ma_version_tag field for dictionaries to
allow optimizations in CPython and extension libraries. This PEP proposes to
rescind PEP 509 and declare the field an implementation detail, as it has
already been superseded by alternatives. This will further allow the field to
be reused for future optimization.
PEP 509 introduced the
ma_version_tag field to dictionaries. This 64-bit
field stored a version counter that updates every time a dictionary was
modified. The original proposal was to use this version counter as a
guard for optimizations.
Since CPython 3.11, this field has become unused by internal optimization efforts. PEP 659-specialized instructions use other methods of verifying that certain optimizations are safe.
To enable further optimizations in CPython, this PEP proposes that the
ma_version_tag field no longer conform to the PEP 509 specification.
This will allow the CPython developers to store other optimization information,
such as dictionary write watchers.
This PEP does not specify what the field may be used for in the future. This is intentional, as implementation details are subject to change, and the field should be used only for internal consumption by CPython.
This specification rescinds that in PEP 509. The
ma_version_tag field of
dict class is declared to be an internal implementation
detail and may be removed altogether, or may have a different representation.
C extensions should not rely on this field.
Certain extensions use
ma_version_tag for fast dictionary or globals
lookups. For example,
Cython uses the field for fast dynamic module variable lookups.
Since this field is private and makes no guarantees on backwards compatibility,
this PEP proposes to repeal PEP 509 immediately within the next CPython
version. A function for fast cached dictionary lookups may be provided as an
alternative feature to replace directly accessing
The biggest user the author could find for this field was Cython. Discussions with a Cython maintainer indicated that removing support for it from Cython is trivial.
PEP 509 was concerned with integer overflow. However, this PEP does not introduce any additional security concerns.
A possible alternative is to preserve the field for backwards compatibility. This PEP rejects that alternative as future optimizations will consume more memory, and the field was always considered private and undocumented aside from the PEP, with no backward compatibility guarantees. Dictionaries in Python are commonly used, and any increase in their memory consumption will adversely affect Python’s performance.
Thanks to C.A.M. Gerlach for edits and wording changes to this document.
This document is placed in the public domain or under the CC0-1.0-Universal license, whichever is more permissive.
Last modified: 2022-12-04 06:52:15 GMT