kimchi (previously pump)
Python2 & 3 C Extension for serializing certain built-in Python types
kimchi offers functionality very similar to the default pickle module. Unlike pickle though, kimchi cannot serialize generic Python objects and classes. kimchi can only serialize built in Python objects - in particular:
- Integers (only in Python2)
- Floats
- Longs
- Strings (Bytes in Python3)
- Unicodes (Strings in Python3)
- Lists
- Tuples
- Sets
- Frozensets
- Dictionaries
- None objects
- Booleans
However, for most objects kimchi performs serialization faster than the built-in pickle library, and more importantly, produces more compact objects.
Usage
kimchi has two single argument functions, loads and dumps. dumps is used to create serializations of objects, and loads is used to recreate the original object given its serialization.
Installation
Using git
$ git clone https://github.com/warrenspe/kimchi.git
$ cd kimchi
$ sudo python3 setup.py install
Examples
>>> import kimchi
>>> serialization = kimchi.dumps(123)
>>> serialization
'\x01\x81{'
>>> kimchi.loads(serialization)
123
>>> serialization = kimchi.dumps('123')
>>> serialization
'\x05\x83123'
>>> kimchi.loads(serialization)
'123'
>>> serialization = kimchi.dumps([1, 2, 3, ('4', '5', '6')])
>>> serialization
'\x08\x96\x84\x01\x81\x01\x01\x81\x02\x01\x81\x03\t\x8a\x83\x05\x814\x05\x815\x05\x816'
>>> kimchi.loads(serialization)
[1, 2, 3, ('4', '5', '6')]
Speed and size comparisons to the built-in pickle
Note that all tests were done in Python2, comparing to its cPickle module
Integers
Floats
Note: constant size, might be revisited later
Longs
Strings (Bytes in Python3)
Note that the final string was not included in the size comparison, as the size skews the axis of the graph. The serialized sizes are:
kimchi: 1003, pickle: 1008
Unicodes (Strings in Python3)
Note that the final unicode string was not included in the size comparison, as the size skews the axis of the graph. The serialized sizes are: kimchi: 1003, pickle: 1006
None objects
Boolean objects
Note: Speed and size for the following types is largely influenced by the items within them, to give the best results they have been filled with the data type which gives the most similar serialization times and sizes between pickle and kimchi, string/bytes objects.
