shoyer/cyordereddict

This library is obsolete! Python 3.5's collections.OrderedDict was rewritten in C, and is now significantly faster than cyordereddict.OrderedDict for almost all operations.

cyordereddict

The Python standard library's OrderedDict ported to Cython. A drop-in replacement that is 2-6x faster.

Install:

pip install cyordereddict

Dependencies:

CPython (2.6, 2.7, 3.3 or 3.4) and a C compiler. Cython is only required for the dev version.

Use:

from cyordereddict import OrderedDict

Benchmarks:

Python 2.7:

Test	Code	Ratio (stdlib / cython)
__init__ empty	OrderedDict()	1.8
__init__ list	OrderedDict(list_data)	4.8
__init__ dict	OrderedDict(dict_data)	4.6
__setitem__	ordereddict[0] = 0	8.6
__getitem__	ordereddict[0]	3
update	ordereddict.update(dict_data)	5.5
__iter__	list(ordereddict)	5.6
items	ordereddict.items()	5.9
__contains__	0 in ordereddict	2.3

Python 3.4:

Test	Code	Ratio (stdlib / cython)
__init__ empty	OrderedDict()	1.5
__init__ list	OrderedDict(list_data)	3.9
__init__ dict	OrderedDict(dict_data)	4.2
__setitem__	ordereddict[0] = 0	8.4
__getitem__	ordereddict[0]	2.9
update	ordereddict.update(dict_data)	6.5
__iter__	list(ordereddict)	2.3
items	list(ordereddict.items())	2.1
__contains__	0 in ordereddict	2.3

To run these yourself, use cyordereddict.benchmark()

Cavaets:

cyorderedddict.OrderedDict is an extension type (similar to the built-in dict) instead of a Python class. This is necessary for speed, but means that in a few pathological cases its behavior will differ from collections.OrderedDict:

• The inspect module does not work on cyorderedddict.OrderedDict methods.
• Extension types use slots instead of dictionaries, so you cannot add custom attributes without making a subclass (e.g., OrderedDict.foo = 'bar' will fail).

You can do anything else you might do with an OrderedDict, including subclassing: everything else passes the collections.OrderedDict test suite. We based the Cython code directly on the Python standard library, and thus use separate code bases for Python 2 and 3, specifically to reduce the potential for introducing new bugs or performance regressions.

License:

MIT. Based on the Python standard library, which is under the Python Software Foundation License.