Skip to main content

Mutable variant of collections.namedtuple -- recordclass.recordclass, which support assignments, and other memory saving variants.

Project description

Recordclass library

Recordclass is MIT Licensed python library. It was started as a "proof of concept" for the problem of fast "mutable" alternative of namedtuple (see question on stackoverflow). It implements a factory function recordclass (a variant of collection.namedtuple) in order to create record-like classes with the same API as collection.namedtuple. It was evolved further in order to provide more memory saving, fast and flexible types.

Recordclass library provide record-like classes that do not participate in cyclic garbage collection (CGC) mechanism, but support only reference counting mechanism for garbage collection. The instances of such classes havn't PyGC_Head prefix in the memory, which decrease their size. This may make sense in cases where it is necessary to limit the size of the objects as much as possible, provided that they will never be part of references cycles in the application. For example, when an object represents a record with fields with values of simple types by convention (int, float, str, date/time/datetime, timedelta, etc.).

In order to illustrate this, consider a simple class with type hints:

class Point:
    x: int
    y: int

By contract instances of the class Point have attributes x and y with values of int type. Assigning other types of values, which are not subclass of int, should be considered as a violation of the contract.

Another examples are non-recursive data structures in which all leaf elements represent a value of an atomic type. Of course, in python, nothing prevent you from “shooting yourself in the foot" by creating the reference cycle in the script or application code. But in many cases, this can still be avoided provided that the developer understands what he is doing and uses such classes in the code with care. Another option is to use static code analyzers along with type annotations to monitor compliance with typehints.

  1. The recodeclass library provide the base class dataobject. The type of dataobject is special metaclass datatype. It control creation of subclasses of dataobject, which will not participate in CGC by default. As the result the instance of such class need less memory. It's memory footprint is similar to memory footprint of instances of the classes with __slots__. The difference is equal to the size of PyGC_Head. It also tunes basicsize of the instances, creates descriptors for the fields and etc. All subclasses of dataobject created by class statement support attrs/dataclasses-like API. For example:

     from recordclass import dataobject, astuple, asdict
     class Point(dataobject):
     >>> p = Point(1, 2)
     >>> astuple(p)
     (1, 2)
     >>> asdict(p)
     {'x':1, 'y':2}
  2. The recordclass factory create dataobject-based subclass with specified fields and support namedtuple-like API. By default it will not participate in CGC too.

     >>> from recordclass import recordclass
     >>> Point = recordclass('Point', 'x y')
     >>> p = Point(1, 2)
     >>> p.y = -1
     >>> print(p._astuple)
     (1, -1)
  3. It provide a factory function make_dataclass for creation of subclasses of dataobject with the specified field names. These subclasses support attrs/dataclasses-like API. This is an equivalent to creation of subclasses of dataobject using class statement. For example:

     >>> Point = make_dataclass('Point', 'x y')
     >>> p = Point(1, 2)
     >>> p.y = -1
     >>> print(p.x, p.y)
     1 -1
  4. It provide a factory function make_arrayclass in order to create subclass of dataobject wich can consider as array of simple values. For example:

     >>> Pair = make_arrayclass(2)
     >>> p = Pair(2, 3)
     >>> p[1] = -1
     >>> print(p)
     Pair(2, -1)
  5. It provide classes lightlist and litetuple, which considers as list-like and tuple-like light containers in order to save memory. Mutable variant of litetuple is called by mutabletuple. The instances of both types don't participate in CGC. For example:

     >>> lt = litetuple(1, 2, 3)
     >>> mt = mutabletuple(1, 2, 3)
     >>> lt == mt
     >>> mt[-1] = -3
     >>> lt == mt
     >>> print(sys.getsizeof(litetuple(1,2,3)), sys.getsizeof((1,2,3)))
     64 48

Memory footprint

The following table explain memory footprints of recordclass-base and dataobject-base objects:

tuple/namedtuple class with __slots__ recordclass/dataobject litetuple/mutabletuple
g+b+s+n*p g+b+n*p b+n*p b+s+n*p


  • b = sizeof(PyObject)
  • s = sizeof(Py_ssize_t)
  • n = number of items
  • p = sizeof(PyObject*)
  • g = sizeof(PyGC_Head)

This is useful in that case when you absolutely sure that reference cycle isn't supposed. For example, when all field values are instances of atomic types. As a result the size of the instance is decreased by 24-32 bytes for cpython 3.4-3.7 and by 16 bytes for cpython >=3.8.

Performance counters

Here is the table with performance counters (python 3.9, debian linux, x86-64), which are mesured using utils/ script:

id size new getattr setattr getitem setitem getkey setkey iterate
0 tuple 56 1.07 0.43 2.19
1 namedtuple 56 2.73 0.47 0.43 2.17
2 class+slots 48 2.08 0.50 0.55
3 dataobject 32 2.15 0.45 0.53 0.45 0.50 2.20
4 dataobject+fast_new 32 1.07 0.45 0.53 0.46 0.50 2.20
5 dataobject+gc 48 2.29 0.46 0.54 0.46 0.51 2.32
6 dataobject+fast_new+gc 48 1.19 0.47 0.55 0.47 0.52 2.23
7 dict 232 2.36 0.45 0.58 2.30
8 dictclass 32 1.09 0.52 0.60 2.30

Main repository for recordclassis on bitbucket.

Here is also a simple example.

More examples can be found in the folder examples.

Quick start


Installation from directory with sources


>>> python install

Run tests:

>>> python

Installation from PyPI


>>> pip install recordclass

Run tests:

>>> python -c "from recordclass.test import *; test_all()"

Quick start with recordclass

The recordclass factory function is designed to create classes that support namedtuple's API, can be mutable and immutable, provide fast creation of the instances and have a minimum memory footprint.

First load inventory:

>>> from recordclass import recordclass

Example with recordclass:

>>> Point = recordclass('Point', 'x y')
>>> p = Point(1,2)
>>> print(p)
Point(1, 2)
>>> print(p.x, p.y)
1 2             
>>> p.x, p.y = 1, 2
>>> print(p)
Point(1, 2)
>>> sys.getsizeof(p) # the output below is for 64bit cpython3.8+

Example with class statement and typehints:

>>> from recordclass import RecordClass

class Point(RecordClass):
   x: int
   y: int

>>> print(Point.__annotations__)
{'x': <class 'int'>, 'y': <class 'int'>}
>>> p = Point(1, 2)
>>> print(p)
Point(1, 2)
>>> print(p.x, p.y)
1 2
>>> p.x, p.y = 1, 2
>>> print(p)
Point(1, 2)

By default recordclass-based class instances doesn't participate in CGC and therefore they are smaller than namedtuple-based ones. If one want to use it in scenarios with reference cycles then one have to use option gc=True (gc=False by default):

>>> Node = recordclass('Node', 'root children', gc=True)


class Node(RecordClass, gc=True):
     root: 'Node'
     chilren: list

The recordclass factory can specify type of the fields:

>>> Point = recordclass('Point', [('x',int), ('y',int)])

Quick start with dataobject

Dataobject is the base class for creation of data classes with fast instance creation and small memory footprint. They don't provide namedtuple-like API. The classes created by recrdclass factory are subclasses of the dataobject too, but in addition provide nametuple-like API.

First load inventory:

>>> from recordclass import dataobject, asdict, astuple

Define class:

class Point(dataobject):
    x: int
    y: int

One can't remove attributes from the class:

>>> del Point.x
. . . . . . . .
AttributeError: Attribute x of the class Point can't be deleted

Annotations of the fields are defined as dict in __annotations__:

>>> print(Point.__annotations__)
{'x': <class 'int'>, 'y': <class 'int'>}

Default text representation:

>>> p = Point(1,2)
>>> print(p)
Point(x=1, y=2)

One can't remove field's value:

>>> del p.x
. . . . . . . . 
AttributeError: The value can't be deleted

The instances has a minimum memory footprint possible for CPython objects, which also consist only of Python objects.:

>>> sys.getsizeof() # the output below for 64bit python 3.8+
>>> p.__sizeof__() == sys.getsizeof(p) # no additional space for cyclic GC support

The instance is mutable by default:

>>> p.x, p.y = 10, 20
>>> print(p)
Point(x=10, y=20)

Functions asdict and astuple for converting to dict and tuple:

>>> asdict(p)
{'x':10, 'y':20}
>>> astuple(p)
(10, 20)

By default subclasses of dataobject are mutable. If one want make it immutable then there is the option readonly=True:

class Point(dataobject, readonly=True):
    x: int
    y: int

>>> p = Point(1,2)
>>> p.x = -1
. . . . . . . . . . . . . 
TypeError: item is readonly

By default subclasses of dataobject are not iterable by default. If one want make it iterable then there is the option iterable=True:

class Point(dataobject, iterable=True):
    x: int
    y: int

>>> p = Point(1,2)
>>> for x in p: print(x)

Another way to create subclasses of dataobject – factory function make_dataclass:

>>> from recordclass import make_dataclass

>>> Point = make_dataclass("Point", [("x",int), ("y",int)])


>>> Point = make_dataclass("Point", {"x":int, "y":int})

Default values are also supported::

class CPoint(dataobject):
    x: int
    y: int
    color: str = 'white'


>>> CPoint = make_dataclass("CPoint", [("x",int), ("y",int), ("color",str)], defaults=("white",))

>>> p = CPoint(1,2)
>>> print(p)
Point(x=1, y=2, color='white')


class PointInvalidDefaults(dataobject):
    x:int = 0

is not allowed. A fields without default value may not appear after a field with default value.

There is the options fast_new=True. It allows faster creation of the instances. Here is an example:

class FastPoint(dataobject, fast_new=True):
    x: int
    y: int

The followings timings explain (in jupyter notebook) boosting effect of fast_new option:

%timeit l1 = [Point(i,i) for i in range(100000)]
%timeit l2 = [FastPoint(i,i) for i in range(100000)]
# output with python 3.9 64bit
25.6 ms ± 2.4 ms per loop (mean ± std. dev. of 7 runs, 10 loops each)
10.4 ms ± 426 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Using dataobject-based classes for recursive data without reference cycles

There is the option deep_dealloc (default value is True) for deallocation of recursive datastructures. Let consider simple example:

class LinkedItem(dataobject, fast_new=True):
    val: object
    next: 'LinkedItem'

class LinkedList(dataobject, deep_dealloc=True):
    start: LinkedItem = None
    end: LinkedItem = None

    def append(self, val):
        link = LinkedItem(val, None)
        if self.start is None:
            self.start = link
   = link
        self.end = link

Without deep_dealloc=True deallocation of the instance of LinkedList will be failed if the length of the linked list is too large. But it can be resolved with __del__ method for clearing the linked list:

def __del__(self):
    curr = self.start
    while curr is not None:
        next = = None
        curr = next

There is builtin more fast deallocation method using finalization mechanizm when deep_dealloc=True. In such case one don't need __del__ method for clearing tthe list.

Note that for classes with gc=True (cyclic GC is used) this method is disabled: the python's cyclic GC is used.

For more details see notebook example_datatypes.



  • Add dictclass factory function to generate class with dict-like API and without attribute access to the fields. Features: fast instance creation, small memory footprint.


  • RecordClass started to be a direct subclass of dataobject with sequence=True and support of namedtuple-like API. Insted of RecordClass(name, fields, **kw) for class creation use factory function recordclass(name, fields, **kw) (it allows to specify types).

  • Add option api='dict' to make_dataclass for creating class that support dict-like API.

  • Now one can't remove dataobject's property from it's class using del or builting delattr. For example:

      >>> Point = make_dataclass("Point", "x y")
      >>> del Point.x
      AttributeError: Attribute x of the class Point can't be deleted
  • Now one can't delete field's value using del or builting delattr. For example:

      >>> p = Point(1, 2)
      >>> del p.x
      AttributeError: The value can't be deleted"      

    Insted one can use assighnment to None:

      >>> p = Point(1, 2)
      >>> p.x = None
  • Slightly improve performance of the access by index of dataobject-based classes with option sequence=True.


  • Options readonly and iterable now can be sspecified via keyword arguments in class statement. For example:

      class Point(dataobject, readonly=True, iterable=True):
  • Add update(cls, **kwargs) function to update attribute values.`


  • Now library supports only Python >= 3.6
  • 'gc' and 'fast_new' options now can be specified as kwargs in class statement.
  • Add a function astuple(ob) for transformation dataobject instance ob to a tuple.
  • Drop datatuple based classes.
  • Add function make(cls, args, **kwargs) to create instance of the class cls.
  • Add function clone(ob, **kwargs) to clone dataobject instance ob.
  • Make structclass as alias of make_dataclass.
  • Add option 'deep_dealloc' (@clsconfig(deep_dealloc=True)) for deallocation instances of dataobject-based recursive subclasses.


  • Subclasses of dataobject now support iterable and hashable protocols by default.


  • Fix compilation issue for python 3.9.


  • Fix issue with hash when subclassing recordclass-based classes.


  • Add doc to generated dataobject-based class in order to support inspect.signature.
  • Add fast_new argument/option for fast instance creation.
  • Fix refleak in litelist.
  • Fix sequence protocol ability for dataobject/datatuple.
  • Fix typed interface for StructClass.


  • Fix issue #14 with deepcopy of dataobjects.


  • Restore ``join_classesand add new functionjoin_dataclasses`.

  • Remove redundant debug code.


  • Make recordclass compiled and work with cpython 3.8.
  • Move repository to git instead of mercurial since bitbucket will drop support of mercurial repositories.
  • Fix some potential reference leaks.

  • Fix missing .h files.


  • clsconfig now become the main decorator for tuning dataobject-based classes.
  • Fix concatenation of mutabletuples (issue #10).


  • dataobject instances may be deallocated faster now.


  • Rename memoryslots to mutabletuple.
  • mutabletuple and immutabletuple dosn't participate in cyclic garbage collection.
  • Add litelist type for list-like objects, which doesn't participate in cyglic garbage collection.


  • Introduce DataclassStorage and RecordclassStorage. They allow cache classes and used them without creation of new one.
  • Add iterable decorator and argument. Now dataobject with fields isn't iterable by default.
  • Move astuple to dataobject.c.


  • Fix error with dataobject's __copy__.
  • Fix error with pickling of recordclasses and structclasses, which was appeared since 0.8.5 (Thanks to Connor Wolf).


  • Now by default sequence protocol is not supported by default if dataobject has fields, but iteration is supported.
  • By default argsonly=False for usability reasons.


  • Invent new factory function make_class for creation of different kind of dataobject classes without GC support by default.
  • Invent new metaclass datatype and new base class dataobject for creation dataobject class using class statement. It have disabled GC support, but could be enabled by decorator dataobject.enable_gc. It support type hints (for python >= 3.6) and default values. It may not specify sequence of field names in __fields__ when type hints are applied to all data attributes (for python >= 3.6).
  • Now recordclass-based classes may not support cyclic garbage collection too. This reduces the memory footprint by the size of PyGC_Head. Now by default recordclass-based classes doesn't support cyclic garbage collection.


  • Change version to 0.9 to indicate a step forward.
  • Cleanup dataobject.__cinit__.


  • Make arrayclass-based objects support setitem/getitem and structclass-based objects able to not support them. By default, as before structclass-based objects support setitem/getitem protocol.
  • Now only instances of dataobject are comparable to 'arrayclass'-based and structclass-based instances.
  • Now generated classes can be hashable.


  • Improve support for readonly mode for structclass and arrayclass.
  • Add tests for arrayclass.


  • Add typehints support to structclass-based classes.


  • Remove usedict, gc, weaklist from the class __dict__.


  • Remove Cython dependence by default for building recordclass from the sources [Issue #7].


  • Add structclass factory function. It's analog of recordclass but with less memory footprint for it's instances (same as for instances of classes with __slots__) in the camparison with recordclass and namedtuple (it currently implemented with Cython).
  • Add arrayclass factory function which produce a class for creation fixed size array. The benefit of such approach is also less memory footprint (it currently currently implemented with Cython).
  • structclass factory has argument gc now. If gc=False (by default) support of cyclic garbage collection will switched off for instances of the created class.
  • Add function join(C1, C2) in order to join two structclass-based classes C1 and C2.
  • Add sequenceproxy function for creation of immutable and hashable proxy object from class instances, which implement access by index (it currently currently implemented with Cython).
  • Add support for access to recordclass object attributes by idiom: ob['attrname'] (Issue #5).
  • Add argument readonly to recordclass factory to produce immutable namedtuple. In contrast to collection.namedtuple it use same descriptors as for regular recordclasses for performance increasing.


  • Make mutabletuple objects creation faster. As a side effect: when number of fields >= 8 recordclass instance creation time is not biger than creation time of instaces of dataclasses with __slots__.
  • Recordclass factory function now create new recordclass classes in the same way as namedtuple in 3.7 (there is no compilation of generated python source of class).


  • Add support for default values in recordclass factory function in correspondence to same addition to namedtuple in python 3.7.


  • Change version to 0.5


  • Add support for default values in RecordClass (patches from Pedro von Hertwig)
  • Add tests for RecorClass (adopted from python tests for NamedTuple)


  • Add support for typing for python 3.6 (patches from Vladimir Bolshakov).
  • Resolve memory leak issue.


  • Fix memory leak in property getter/setter

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for recordclass, version 0.16.3
Filename, size File type Python version Upload date Hashes
Filename, size recordclass-0.16.3.tar.gz (14.3 MB) File type Source Python version None Upload date Hashes View
Filename, size recordclass-0.16.3-cp39-cp39-win_amd64.whl (125.1 kB) File type Wheel Python version cp39 Upload date Hashes View
Filename, size recordclass-0.16.3-cp39-cp39-win32.whl (110.9 kB) File type Wheel Python version cp39 Upload date Hashes View
Filename, size recordclass-0.16.3-cp38-cp38-win_amd64.whl (125.2 kB) File type Wheel Python version cp38 Upload date Hashes View
Filename, size recordclass-0.16.3-cp38-cp38-win32.whl (111.6 kB) File type Wheel Python version cp38 Upload date Hashes View
Filename, size recordclass-0.16.3-cp37-cp37m-win_amd64.whl (123.3 kB) File type Wheel Python version cp37 Upload date Hashes View
Filename, size recordclass-0.16.3-cp37-cp37m-win32.whl (110.5 kB) File type Wheel Python version cp37 Upload date Hashes View
Filename, size recordclass-0.16.3-cp36-cp36m-win_amd64.whl (123.4 kB) File type Wheel Python version cp36 Upload date Hashes View
Filename, size recordclass-0.16.3-cp36-cp36m-win32.whl (110.5 kB) File type Wheel Python version cp36 Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Huawei Huawei PSF Sponsor Microsoft Microsoft PSF Sponsor NVIDIA NVIDIA PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page