Release Notifications

Follow on Sibbell

Tip: Follow bidict on Sibbell to be notified when a new version of bidict is released.

0.15.0 (not yet released)

  • Use __slots__ to speed up bidict attribute access and reduce memory usage.
  • Use weakrefs to refer to a bidict’s inverse internally, no longer creating a strong reference cycle. Memory for a bidict that you create can now be reclaimed in CPython as soon as you no longer hold any references to it. See the new inv Avoids Reference Cycles documentation. Fixes #24.
  • Classes no longer have to provide an __inverted__ attribute to be considered virtual subclasses of BidirectionalMapping.
  • If bidict.inverted() is passed an object with an __inverted__ attribute, it now ensures it is callable() before returning the result of calling it.
  • Make bidict.FrozenOrderedBidict.__iter__() slightly more efficient.
  • __repr__() no longer dynamically checks for a __reversed__ method to determine whether to use an ordered or unordered __repr__ delegate. Now it always just calls the new __repr_delegate__() instead, which may be explicitly overridden as needed.
  • If you create a custom bidict subclass whose _fwdm_cls differs from its _invm_cls (as in the FwdKeySortedBidict example from the Sorted Bidict Recipes), the inverse bidirectional mapping type (with _fwdm_cls and _invm_cls swapped) is now correctly computed and used automatically for your custom bidict’s inv bidict.

Breaking API Changes

  • Rename:

    • bidict.frozenbidict.fwdm._fwdm
    • bidict.frozenbidict.invm._invm
    • bidict.frozenbidict.fwd_cls._fwdm_cls
    • bidict.frozenbidict.inv_cls._invm_cls
    • bidict.frozenbidict.isinv._isinv

    Though overriding _fwdm_cls and _invm_cls remains supported (see Extending bidict), this is not a common enough use case to warrant public names. Most users do not need to know or care about any of these.

  • The RAISE, OVERWRITE, and IGNORE duplication policies are no longer available as attributes of bidict.DuplicationPolicy, and can now only be accesseed as attributes of the bidict module namespace. (So it is no longer possible to create an infinite chain like DuplicationPolicy.RAISE.RAISE.RAISE...)

  • namedbidict() now raises TypeError if the provided base_type is not a subclass of frozenbidict.

  • Pickling ordered bidicts now requires at least version 2 of the pickle protocol. If you are using Python 3, pickle.DEFAULT_PROTOCOL is 3 anyway, so this will not affect you. However if you are using in Python 2, DEFAULT_PROTOCOL is 0, so you must now explicitly specify the version in your pickle.dumps() calls, e.g. pickle.dumps(ob, 2).

0.14.2 (2017-12-06)

  • Make initializing (or updating an empty bidict) from only another BidirectionalMapping more efficient by skipping unnecessary duplication checking.
  • Fix accidental ignoring of specified base_type argument when (un)pickling a namedbidict().
  • Fix incorrect inversion of some_named_bidict.inv.[fwdname]_for and some_named_bidict.inv.[invname]_for.
  • Only warn when an unsupported Python version is detected (e.g. Python < 2.7) rather than raising AssertionError.

0.14.1 (2017-11-28)

  • Fix a bug introduced in 0.14.0 where hashing a frozenbidict’s inverse (e.g. f = frozenbidict(); {f.inv: '...'}) would cause an AttributeError.
  • Fix a bug introduced in 0.14.0 for Python 2 users where calling viewitems() would cause a TypeError. Thanks Richard Sanger for reporting.

0.14.0 (2017-11-20)

  • Fix a bug where bidict’s default on_dup_kv policy was set to RAISE, rather than matching whatever on_dup_val policy was in effect as was documented.
  • Fix a bug that could happen when using Python’s optimization (-O) flag that could leave an ordered bidict in an inconsistent state when dealing with duplicated, overwritten keys or values. If you do not use optimizations (specifically, skipping assert statements), this would not have affected you.
  • Fix a bug introduced by the optimizations in 0.13.0 that could cause a frozen bidict that compared equal to another mapping to have a different hash value from the other mapping, violating Python’s object model. This would only have affected you if you were inserting a frozen bidict and some other immutable mapping that it compared equal to into the same set or mapping.
  • Add equals_order_sensitive().
  • Reduce the memory usage of ordered bidicts.
  • Make copying of ordered bidicts faster.
  • Improvements to tests and CI, including:
    • Test on Windows
    • Test with PyPy3
    • Test with CPython 3.7-dev
    • Test with optimization flags
    • Require pylint to pass

Breaking API Changes

This release includes multiple API simplifications and improvements.

  • Rename:

    so that these now match the case of collections.OrderedDict.

    The names of the bidict, namedbidict(), and frozenbidict classes have been retained as all-lowercase so that they continue to match the case of dict, namedtuple(), and frozenset, respectively.

  • The ON_DUP_VAL duplication policy value for on_dup_kv has been removed. Use None instead.

  • Merge frozenbidict and BidictBase together and remove BidictBase. frozenbidict is now the concrete base class that all other bidict types derive from. See the updated bidict Type Hierarchy.

  • Merge frozenbidict and FrozenBidictBase together and remove FrozenBidictBase. See the updated bidict Type Hierarchy.

  • Merge frozenorderedbidict and OrderedBidictBase together into a single FrozenOrderedBidict class and remove OrderedBidictBase. OrderedBidict now extends FrozenOrderedBidict to add mutable behavior. See the updated bidict Type Hierarchy.

  • Make __eq__() always perform an order-insensitive equality test, even if the other mapping is ordered.

    Previously, __eq__() was only order-sensitive for other OrderedBidictBase subclasses, and order-insensitive otherwise.

    Use the new equals_order_sensitive() method for order-sensitive equality comparison.

  • orderedbidict._should_compare_order_sensitive() has been removed.

  • frozenorderedbidict._HASH_NITEMS_MAX has been removed. Since its hash value must be computed from all contained items (so that hash results are consistent with equality comparisons against unordered mappings), the number of items that influence the hash value should not be limitable.

  • frozenbidict._USE_ITEMSVIEW_HASH has been removed, and frozenbidict.compute_hash() now uses collections.ItemsView._hash() to compute the hash always, not just when running on PyPy.

    Override frozenbidict.compute_hash() to return hash(frozenset(iteritems(self))) if you prefer the old default behavior on CPython, which takes linear rather than constant space, but which uses the frozenset_hash routine (implemented in setobject.c) rather than the pure Python ItemsView._hash() routine.

  • loosebidict and looseorderedbidict have been removed. Simple recipes to implement them yourself are now given in OverwritingBidict Recipe.

  • Rename FrozenBidictBase._compute_hash()frozenbidict.compute_hash().

  • Rename DuplicationBehaviorDuplicationPolicy.

  • Rename:

0.13.1 (2017-03-15)

  • Fix regression introduced by the new __subclasshook__() functionality in 0.13.0 so that issubclass(OldStyleClass, BidirectionalMapping) once again works with old-style classes, returning False rather than raising AttributeError (thanks, @knaperek).

0.13.0 (2017-01-19)

  • Support Python 3.6.

    (Earlier versions of bidict should work fine on 3.6, but it is officially supported starting in this version.)

  • BidirectionalMapping has been refactored into an abstract base class, following the way works. The concrete method implementations it used to provide have been moved into a new BidictBase subclass.

    BidirectionalMapping now also implements __subclasshook__(), so any class that provides a conforming set of attributes (enumerated in _subclsattrs) will be considered a BidirectionalMapping subclass automatically.

  • OrderedBidirectionalMapping has been renamed to OrderedBidictBase, to better reflect its function. (It is not an ABC.)

  • A new FrozenBidictBase class has been factored out of frozenbidict and frozenorderedbidict. This implements common behavior such as caching the result of __hash__ after the first call.

  • The hash implementations of frozenbidict and frozenorderedbidict. have been reworked to improve performance and flexibility. frozenorderedbidict’s hash implementation is now order-sensitive.

    See frozenbidict._compute_hash() and frozenorderedbidict._compute_hash for more documentation of the changes, including the new frozenbidict._USE_ITEMSVIEW_HASH and frozenorderedbidict._HASH_NITEMS_MAX attributes. If you have an interesting use case that requires overriding these, or suggestions for an alternative implementation, please share your feedback.

  • Add _fwd_class and _inv_class attributes representing the backing Mapping types used internally to store the forward and inverse dictionaries, respectively.

    This allows creating custom bidict types with extended functionality simply by overriding these attributes in a subclass.

    See the new Extending bidict documentation for examples.

  • Pass any parameters passed to popitem() through to _fwd.popitem for greater extensibility.

  • More concise repr strings for empty bidicts.

    e.g. bidict() rather than bidict({}) and orderedbidict() rather than orderedbidict([]).

  • Add bidict.compat.PYPY and remove unused bidict.compat.izip_longest.

0.12.0 (2016-07-03)

  • New/renamed exceptions:

  • put() now accepts on_dup_key, on_dup_val, and on_dup_kv keyword args which allow you to override the default policy when the key or value of a given item duplicates any existing item’s. These can take the following values:

    on_dup_kv can also take ON_DUP_VAL.

    If not provided, put() uses the RAISE policy by default.

  • New putall() method provides a bulk put() API, allowing you to override the default duplication handling policy that update() uses.

  • update() now fails clean, so if an update() call raises a DuplicationError, you can now be sure that none of the given items was inserted.

    Previously, all of the given items that were processed before the one causing the failure would have been inserted, and no facility was provided to recover which items were inserted and which weren’t, nor to revert any changes made by the failed update() call. The new behavior makes it easier to reason about and control the effects of failed update() calls.

    The new putall() method also fails clean.

    Internally, this is implemented by storing a log of changes made while an update is being processed, and rolling back the changes when one of them is found to cause an error. This required reimplementing orderedbidict on top of two dicts and a linked list, rather than two OrderedDicts, since OrderedDict does not expose its backing linked list.

  • orderedbidict.move_to_end() now works on Python < 3.2 as a result of the new orderedbidict implementation.

  • Add

    to complement the existing iteritems() and viewitems() compatibility helpers.

  • More efficient implementations of pairs(), inverted(), and copy().

  • Implement __copy__() for use with the copy module.

  • Fix issue preventing a client class from inheriting from loosebidict (see #34).

  • Add benchmarking to tests.

  • Drop official support for CPython 3.3. (It may continue to work, but is no longer being tested.)

Breaking API Changes

  • Rename KeyExistsExceptionKeyDuplicationError and ValueExistsExceptionValueDuplicationError.

  • When overwriting the key of an existing value in an orderedbidict, the position of the existing item is now preserved, overwriting the key of the existing item in place, rather than moving the item to the end. This now matches the behavior of overwriting the value of an existing key, which has always preserved the position of the existing item. (If inserting an item whose key duplicates that of one existing item and whose value duplicates that of another, the existing item whose value is duplicated is still dropped, and the existing item whose key is duplicated still gets its value overwritten in place, as before.)

    For example:

    >>> from bidict import orderedbidict  
    >>> o = orderedbidict([(0, 1), (2, 3)])  
    >>> o.forceput(4, 1)  

    previously would have resulted in:

    >>> o  
    orderedbidict([(2, 3), (4, 1)])

    but now results in:

    >>> o  
    orderedbidict([(4, 1), (2, 3)])

0.11.0 (2016-02-05)

0.10.0.post1 (2015-12-23)

  • Minor documentation fixes and improvements.

0.10.0 (2015-12-23)

  • Remove several features in favor of keeping the API simpler and the code more maintainable.
  • In the interest of protecting data safety more proactively, by default bidict now raises an error on attempting to insert a non-unique value, rather than allowing its associated key to be silently overwritten. See discussion in #21.
  • New forceupdate() method provides a bulk forceput() operation.
  • Fix bugs in pop and setdefault which could leave a bidict in an inconsistent state.

Breaking API Changes

  • Remove bidict.__invert__, and with it, support for the ~b syntax. Use inv instead. #19
  • Remove support for the slice syntax. Use b.inv[val] rather than b[:val]. #19
  • Remove bidict.invert. Use inv rather than inverting a bidict in place. #20
  • Raise ValueExistsException when attempting to insert a mapping with a non-unique key. #21
  • Rename collapsingbidictloosebidict now that it suppresses ValueExistsException rather than the less general CollapseException. #21
  • CollapseException has been subsumed by ValueExistsException. #21
  • put() now raises KeyExistsException when attempting to insert an already-existing key, and ValueExistsException when attempting to insert an already-existing value.

0.9.0.post1 (2015-06-06)

  • Fix metadata missing in the 0.9.0rc0 release.

0.9.0rc0 (2015-05-30)

  • Add this changelog, Contributors’ Guide, Gitter chat room, and other community-oriented improvements.
  • Adopt Pytest (thanks Tom Viner and Adopt Pytest Month).
  • Add property-based tests via hypothesis.
  • Other code, tests, and docs improvements.

Breaking API Changes

  • Move bidict.iteritems and bidict.viewitems to new bidict.compat module.
  • Move bidict.inverted to new bidict.util module (still available from top-level bidict module as well).
  • Move bidict.fancy_iteritems to bidict.util.pairs() (also available from top level as bidict.pairs()).
  • Rename bidict.namedbidict()’s bidict_type argument base_type.