- Update to version 2.0.0
* Major release: the serialized JSON format now preserves
dictionary identity, which is a subtle change in the
serialized format. (#351)
* Dictionary identity is now preserved. For example, if the same
dictionary appears twice in a list, the reconstituted list
will now contain two references to the same dictionary. (#255)
(+332)
- Changes in v1.5.2
* Patch release to avoid the change in behavior from the
preservation of dict identity. The next release will be
v2.0.0. (#351)
* This relese does not include the performance improvements from
v1.5.1.
* Pandas DataFrame objects with multilevel columns are now
supported. (#346) (+347)
* Numpy 1.20 is now officially supported. (#336)
* Python 3.9 is now officially supported. (+348)
* Achieved a small speedup for _get_flattener by merging type
checks. (+349)
- Changes in v1.5.1
* The performance of the unpickler was drastically improved by
avoiding tag checks for basic Python types. (+340)
* decode() documentation improvements. (+341)
* Serialization of Pandas DataFrame objects that contain
timedelta64[ns] dtypes are now supported. (+330) (#331)
* Dictionary identity is now preserved. For example, if the same
dictionary appears twice in a list, the reconstituted list
will now contain two references to the same dictionary. (#255)
(+332)
* Unit tests were added to ensure that sklearn.tree.
DecisionTreeClassifier objects are properly serialized. (#155)
(+344)
* The is_reducible() utility function used by encode() is now 4x
faster! Objects that provide __getstate__(), __setstate__(),
and __slots__ benefit most from these improvements. (+343)
* Improved pickler flatten()/encode() performance. (+345)
- Changes in v1.5.0
* Previous versions of jsonpickle with make_refs=False would
emit null when encountering an object it had already seen when
traversing objects. All instances of the object are now
serialized. While this is arguably an improvement in the vast
majority of scenarios, it is a change in behavior and is thus
considered a minor-level change. (#333) (#334) (#337) (+338)
* Multiple enums are now serialized correctly with
make_refs=False. (#235)
- Changes in v1.4.2
* Use importlib.metadata from the stdlib on Python 3.8. (+305)
(#303)
* Micro-optimize type checks to use a set for lookups. (+327)
* Documentation improvements.
- Changes in v1.4.1
* Patch release for Python 3.8 importlib_metadata support. (#300)
- Changes in v1.4
* Python 3.8 support. (#292)
* jsonpickle.encode now supports the standard indent and
separators arguments, and passes them through to the active
JSON backend library. (#183)
* We now include a custom handler for array.array objects. (#199)
* Dict key order is preserved when pickling dictionaries on
Python3. (#193)
* Improved serialization of dictionaries with non-string keys.
Previously, using an enum that was both the key and a value in
a dictionary could end up with incorrect references to other
objects. The references are now properly maintained for dicts
with object keys that are also referenced in the dict's
values. (#286)
* Improved serialization of pandas.Series objects. (#287)
- Don't test numpy and pandas in python36 flavor, because
they are no longer available on Tumbleweed (NEP 29)
- Test some extras, but not in lettered staging if they are not
in Ring1.
- Drop PR292-Python38.patch merged upstream