- Add final support for Python 3.13.
- Removed
persisent.cPersistence.simple_new
fossil. See #208
- Drop support for Python 3.7.
- Build Windows wheels on GHA.
- Add preliminary support for Python 3.13a3.
- Add support for Python 3.12.
- Build Linux binary wheels for Python 3.11.
- Drop support for Python 2.7, 3.5, 3.6.
- Add support for building arm64 wheels on macOS.
- Update Python 3.11 support to final release.
- Update Python 3.11 support to 3.11.0-rc1.
- Disable unsafe math optimizations in C code. See pull request 176.
- Add support for Python 3.11 (as of 3.11a5).
- Switch package to src-layout, this is a packaging only change. (#168)
- Add support for Python 3.10.
- Add support for Python 3.9.
- Move from Travis CI to Github Actions.
- Supply manylinux wheels for aarch64 (ARM).
- Fix the pure-Python implementation to activate a ghost object
when setting its
__class__
and__dict__
. This matches the behaviour of the C implementation. See issue 155. - Fix the CFFI cache implementation (used on CPython when
PURE_PYTHON=1
) to not print unraisableAttributeErrors
from_WeakValueDictionary
during garbage collection. See issue 150. - Make the pure-Python implementation of the cache run a garbage
collection (
gc.collect()
) onfull_sweep
,incrgc
andminimize
if it detects that an object that was weakly referenced has been ejected. This solves issues on PyPy with ZODB raisingConnectionStateError
when there are persistentzope.interface
utilities/adapters registered. This partly reverts a change from release 4.2.3.
- Fix an overly specific test failure using zope.interface 5. See issue 144.
- Fix two reference leaks that could theoretically occur as the result of obscure errors. See issue 143.
- Fix a crash in the test suite under a 32-bit CPython on certain 32-bit platforms. See issue 137. Fix by Jerry James.
- Fix an
AssertionError
clearing a non-emptyPersistentMapping
that has no connection. See issue 139.
- Stop installing C header files on PyPy (which is what persistent before 4.6.0 used to do), fixes issue 135.
Fix slicing of
PersistentList
to always return instances of the same class. It was broken on Python 3 prior to 3.7.4.Fix copying of
PersistentList
andPersistentMapping
usingcopy.copy
to also copy the underlying data object. This was broken prior to Python 3.7.4.Update the handling of the
PURE_PYTHON
environment variable. Now, a value of "0" requires that the C extensions be used; any other non-empty value prevents the extensions from being used. Also, all C extensions are required together or none of them will be used. This prevents strange errors that arise from a mismatch of Python and C implementations. See issue 131.Note that some private implementation details such as the names of the pure-Python implementations have changed.
Fix
PersistentList
to mark itself as changed after callingclear
(if needed). See PR 115.Fix
PersistentMapping.update
to accept keyword arguments like the nativeUserDict
. Previously, most uses of keyword arguments resulted inTypeError
; in the undocumented and extremely unlikely event of a single keyword argument calledb
that happens to be a dictionary, the behaviour will change. Also adjust the signatures ofsetdefault
andpop
to match the native version.Fix
PersistentList.clear
,PersistentMapping.clear
andPersistentMapping.popitem
to no longer mark the object as changed if it was empty.Add preliminary support for Python 3.9a3+. See issue 124.
Fix the Python implementation of the PickleCache to be able to store objects that cannot be weakly referenced. See issue 133.
Note that
ctypes
is required to use the Python implementation (except on PyPy).
- Add support for Python 3.8.
- Update documentation to Python 3.
Fully test the C implementation of the PickleCache, and fix discrepancies between it and the Python implementation:
- The C implementation now raises
ValueError
instead ofAssertionError
for certain types of bad inputs. - The Python implementation uses the C wording for error messages.
- The C implementation properly implements
IPickleCache
; methods unique to the Python implementation were moved toIExtendedPickleCache
. - The Python implementation raises
AttributeError
if a persistent class doesn't have ap_jar
attribute.
See issue 102.
- The C implementation now raises
Allow sweeping cache without
cache_size
.cache_size_bytes
works withcache_size=0
, no need to setcache_size
to a large value.Require
CFFI
on CPython for pure-Python operation. This drops support for Jython (which was untested). See issue 77.Fix DeprecationWarning about
PY_SSIZE_T_CLEAN
. See issue 108.Drop support for Python 3.4.
- Fix the repr of the persistent objects to include the module name when using the C extension. This matches the pure-Python behaviour and the behaviour prior to 4.4.0. See issue 92.
- Change the repr of persistent objects to format the OID as in integer in hexadecimal notation if it is an 8-byte byte string, as ZODB does. This eliminates some issues in doctests. See issue 95.
- Explicitly use unsigned constants for packing and unpacking C
timestamps, fixing an arithmetic issue for GCC when optimizations
are enabled and
-fwrapv
is not enabled. See issue 86.
- Fix installation of source packages on PyPy. See issue 88.
- Use unsigned constants when doing arithmetic on C timestamps, possibly avoiding some overflow issues with some compilers or compiler settings. See issue 86.
- Change the default representation of
Persistent
objects to include the representation of their OID and jar, if set. Also add the ability for subclasses to implement_p_repr()
instead of overriding__repr__
for better exception handling. See issue 11. - Reach and maintain 100% test coverage.
- Simplify
__init__.py
, including removal of an attempted legacy import ofpersistent.TimeStamp
. See PR 80. - Add support for Python 3.7 and drop support for Python 3.3.
- Build the CFFI modules (used on PyPy or when PURE_PYTHON is set) at installation or wheel building time when CFFI is available. This replaces the deprecated way of building them at import time. If binary wheels are distributed, it eliminates the need to have a functioning C compiler to use PyPy. See issue 75.
- Fix deleting the
_p_oid
of a pure-Python persistent object when it is in a cache. - Fix deleting special (
_p
) attributes of a pure-Python persistent object that overrides__delattr__
and correctly calls_p_delattr
. - Remove some internal compatibility shims that are no longer necessary. See PR 82.
- Make the return value of
TimeStamp.second()
consistent across C and Python implementations when theTimeStamp
was created from 6 arguments with floating point seconds. Also make it match across trips throughTimeStamp.raw()
. Previously, the C version could initially have erroneous rounding and too much false precision, while the Python version could have too much precision. The raw/repr values have not changed. See issue 41.
- Fix the possibility of a rare crash in the C extension when deallocating items. See #66
- Change cPickleCache's comparison of object sizes to determine
whether an object can go in the cache to use
PyObject_TypeCheck()
. This matches what the pure Python implementation does and is a stronger test that the object really is compatible with the cache. Previously, an object could potentially includecPersistent_HEAD
and not settp_base
tocPersistenceCAPI->pertype
and still be eligible for the pickle cache; that is no longer the case. See issue 69.
- Packaging-only release: fix Python 2.7
manylinux
wheels.
- Packaging-only release: get
manylinux
wheel built automatically.
- Avoid raising a
SystemError: error return without exception set
when loading an object with slots whose jar generates an exception (such as a ZODBPOSKeyError
) insetstate
.
- Fix the hashcode of Python
TimeStamp
objects on 64-bit Python on Windows. See #55 - Stop calling
gc.collect
every timePickleCache.incrgc
is called (every transaction boundary) in pure-Python mode (PyPy). This means that the reported size of the cache may be wrong (until the next GC), but it is much faster. This should not have any observable effects for user code. - Stop clearing the dict and slots of objects added to
PickleCache.new_ghost
(typically these values are passed to__new__
from the pickle data) in pure-Python mode (PyPy). This matches the behaviour of the C code. - Add support for Python 3.6.
- Fix
__setstate__
interning whenstate
parameter is not a built-in dict
- Drop use of
ctypes
for determining maximum integer size, to increase pure-Python compatibility. See #31 - Ensure that
__slots__
attributes are cleared when a persistent object is ghostified. (This excluses classes that override__new__
. See https://github.com/zopefoundation/persistent/wiki/Notes_on_state_new_and_slots if you're curious.)
- Fix the hashcode of C
TimeStamp
objects on 64-bit Python 3 on Windows.
- Fixed the Python(/PYPY) implementation
TimeStamp.timeTime
method to have subsecond precision. - When testing
PURE_PYTHON
environments undertox
, avoid poisoning the user's global wheel cache. - Add support for Python 3.5.
- Drop support for Python 2.6 and 3.2.
- Fix manifest and re-upload to fix stray files included in 4.1.0.
- Make the Python implementation of
Persistent
andPickleCache
behave more similarly to the C implementation. In particular, the Python version can now run the complete ZODB and ZEO test suites. - Fix the hashcode of the Python
TimeStamp
on 32-bit platforms.
- Make the C and Python
TimeStamp
objects behave more alike. The Python version now produces the samerepr
and.raw()
output as the C version, and has the same hashcode. In addition, the Python version is now supports ordering and equality like the C version. - Intern keys of object state in
__setstate__
to reduce memory usage when unpickling multiple objects with the same attributes. - Add support for PyPy3.
- 100% branch coverage.
- Add support for Python 3.4.
- In pure-Python
Persistent
, avoid loading state in_p_activate
for non-ghost objects (which could corrupt their state). (PR #9) - In pure-Python, and don't throw
POSKeyError
if_p_activate
is called on an object that has never been committed. (PR #9) - In pure-Python
Persistent
, avoid calling a subclass's__setattr__
at instance creation time. (PR #8) - Make it possible to delete
_p_jar
/_p_oid
of a pure-PythonPersistent
object which has been removed from the jar's cache (fixes aborting a ZODB Connection that has added objects). (PR #7)
- Avoid a KeyError from
_p_accessed()
on newly-created objects under pure-Python: these objects may be assigned to a jar, but not yet added to its cache. (PR #6) - Avoid a failure in
Persistent.__setstate__
when the state dict contains exactly two keys. (PR #5) - Fix a hang in
picklecache
invalidation if OIDs are manually passed out-of-order. (PR #4) - Add
PURE_PYTHON
environment variable support: if set, the C extensions will not be built, imported, or tested.
- Updated Trove classifiers.
- Fixed the C-extensions under Py3k (previously they compiled but were not importable).
- Added support for Python 3.3.
- C extenstions now build under Python 3.2, passing the same tests as the pure-Python reference implementation.
- Fixed: In the C implimentation, an integer was compared with a pointer, with undefined results and a compiler warning.
- Fixed: the Python implementation of the
_p_estimated_size
propety didn't support deletion. - Simplified implementation of the
_p_estimated_size
property to only accept integers. A TypeError is raised if an incorrect type is provided.
- Correct initialization functions in renamed
_timestamp
extension.
Worked around test failure due to overflow to long on 32-bit systems.
Renamed
TimeStamp
extension module to avoid clash with pure-Pythontimestamp
module on case-insensitive filesystems.N.B: the canonical way to import the
TimeStamp
class is now:from persistent.timestamp import TimeStamp
which will yield the class from the extension module (if available), falling back to the pure-Python reference implementation.
- Added explicit support for Python 3.2 and PyPy.
- Note that the C implementations of Persistent, PickleCache, and Timestamp are not built (yet) on these platforms.
- Dropped support for Python < 2.6.
- 100% unit test coverage.
- Removed all
ZODB
-dependent tests:- Rewrote some to avoid the dependency
- Cloned the remainder into new
ZODB.tests
modules.
- Refactored some doctests refactored as unittests.
- Completed pure-Python reference implementations of 'Persistent', 'PickleCache', and 'TimeStamp'.
- All covered platforms tested under
tox
. - Added support for continuous integration using
tox
andjenkins
. - Added
setup.py dev
alias (installsnose
andcoverage
). - Dropped dependency on
zope.testing
/zope.testrunner
: tests now run withsetup.py test
.
- Refactored many Doctests as Sphinx documentation (snippets are exercised via 'tox').
- Added
setup.py docs
alias (installsSphinx
andrepoze.sphinx.autointerface
).