Skip to content


Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?

Latest commit


Git stats


Failed to load latest commit information.
Latest commit message
Commit time


docs Documentation Status
Travis-CI Build Status AppVeyor Build Status Requirements Status
Coverage Status Coverage Status

Container class boilerplate killer.

  • Free software: BSD 2-Clause License


pip install fields

Usage & examples

A class that has 2 attributes, name and size:

>>> from fields import Fields
>>> class Pizza(
...     pass
>>> p = Pizza("Pepperoni", "large")
>>> p
Pizza(name='Pepperoni', size='large')
>>> p.size

You can also use keyword arguments:

>>> Pizza(size="large", name="Pepperoni")
Pizza(name='Pepperoni', size='large')

You can have as many attributes as you want:

>>> class Pizza(
...     pass
>>> Pizza("Funghi", ["mushrooms", "mozarella"], "thin", "large")
Pizza(name='Funghi', ingredients=['mushrooms', 'mozarella'], crust='thin', size='large')

A class that has one required attribute value and two attributes (left and right) with default value None:

>>> class Node(Fields.value.left[None].right[None]):
...     pass
>>> Node(1, Node(2), Node(3, Node(4)))
Node(value=1, left=Node(value=2, left=None, right=None), right=Node(value=3, left=Node(value=4, left=None, right=None), right=None))
>>> Node(1, right=Node(2))
Node(value=1, left=None, right=Node(value=2, left=None, right=None))

You can also use it inline:

>>>"Pepperoni", "large")
FieldsBase(name='Pepperoni', size='large')

Want tuples?

An alternative to namedtuple:

>>> from fields import Tuple
>>> class Pair(Tuple.a.b):
...     pass
>>> issubclass(Pair, tuple)
>>> p = Pair(1, 2)
>>> p.a
>>> p.b
>>> tuple(p)
(1, 2)
>>> a, b = p
>>> a
>>> b

Tuples are fast!

benchmark: 9 tests, min 5 rounds (of min 25.00us), 1.00s max time, timer: time.perf_counter

Name (time in us)                 Min        Max     Mean   StdDev  Rounds  Iterations
test_characteristic            6.0100  1218.4800  11.7102  34.3158   15899          10
test_fields                    6.8000  1850.5250   9.8448  33.8487    5535           4
test_slots_fields              6.3500   721.0300   8.6120  14.8090   15198          10
test_super_dumb                7.0111  1289.6667  11.6881  31.6012   15244           9
test_dumb                      3.7556   673.8444   5.8010  15.0514   14246          18
test_tuple                     3.1750   478.7750   5.1974   9.1878   14642          12
test_namedtuple                3.2778   538.1111   5.0403   9.9177   14105           9
test_attrs_decorated_class     4.2062   540.5125   5.3618  11.6708   14266          16
test_attrs_class               3.7889   316.1056   4.7731   6.0656   14026          18



To run all the tests run tox in your shell (pip install tox if you don't have it):



Why should I use this?

It's less to type, why have quotes around when the names need to be valid symbols anyway. In fact, this is one of the shortest forms possible to specify a container with fields.

But you're abusing a very well known syntax. You're using attribute access instead of a list of strings. Why?

Symbols should be symbols. Why validate strings so they are valid symbols when you can avoid that? Just use symbols. Save on both typing and validation code.

The use of language constructs is not that surprising or confusing in the sense that semantics precede conventional syntax use. For example, if we have class Person(Fields.first_name.last_name.height.weight): pass then it's going to be clear we're talking about a Person object with first_name, last_name, height and width fields: the words have clear meaning.

Again, you should not name your variables as f1, f2 or any other non-semantic symbols anyway.

Semantics precede syntax: it's like looking at a cake resembling a dog, you won't expect the cake to bark and run around.

Is this stable? Is it tested?

Yes. Mercilessly tested on Travis and AppVeyor.

Is the API stable?

Yes, ofcourse.

Why not namedtuple?

It's ugly, repetitive and inflexible. Compare this:

>>> from collections import namedtuple
>>> class MyContainer(namedtuple("MyContainer", ["field1", "field2"])):
...     pass
>>> MyContainer(1, 2)
MyContainer(field1=1, field2=2)

To this:

>>> class MyContainer(Tuple.field1.field2):
...     pass
>>> MyContainer(1, 2)
MyContainer(field1=1, field2=2)

Why not characteristic?

Ugly, inconsistent - you don't own the class:

Lets try this:

>>> import characteristic
>>> @characteristic.attributes(["field1", "field2"])
... class MyContainer(object):
...     def __init__(self, a, b):
...         if a > b:
...             raise ValueError("Expected %s < %s" % (a, b))
>>> MyContainer(1, 2)
Traceback (most recent call last):
ValueError: Missing keyword value for 'field1'.

WHAT !? Ok, lets write some more code:

>>> MyContainer(field1=1, field2=2)
Traceback (most recent call last):
TypeError: __init__() ... arguments...

This is bananas. You have to write your class around these quirks.

Lets try this:

>>> class MyContainer(Fields.field1.field2):
...     def __init__(self, a, b):
...         if a > b:
...             raise ValueError("Expected %s < %s" % (a, b))
...         super(MyContainer, self).__init__(a, b)

Just like a normal class, works as expected:

>>> MyContainer(1, 2)
MyContainer(field1=1, field2=2)

Why not attrs?

Now this is a very difficult question.

Consider this typical use-case:

>>> import attr
>>> @attr.s
... class Point(object):
...     x = attr.ib()
...     y = attr.ib()

Worth noting:

  • attrs is faster because it doesn't allow your class to be used as a mixin (it doesn't do any super(cls, self).__init__(...) for you).

  • The typical use-case doesn't allow you to have a custom __init__. You can use @attr.s(init=False) that will allow you to implement your own __init__. However, you can't have your own __init__ that calls attrs provided __init__ (like in a subclassing scenario).

  • It works better with IDEs and source code analysis tools because of the attributes defined on the class.

  • It's more composable if you only use @attr.s decorated hierarchies. Example:

    >>> @attr.s
    ... class A(object):
    ...     a = attr.ib()
    ...     def get_a(self):
    ...         return self.a
    >>> @attr.s
    ... class B(object):
    ...     b = attr.ib()
    >>> @attr.s
    ... class C(B, A):
    ...     c = attr.ib()
    >>> C(1, 2, 3)
    C(a=1, b=2, c=3)

All in all, attrs is a fast and minimal container library that does support subclasses, but quite differently than fields. Definitely worth considering.

Also, nowadays it has more features than fields. See #6.

Won't this confuse pylint?

Normaly it would, but there's a plugin that makes pylint understand it, just like any other class: pylint-fields.


Diabolical. Can't be unseen.

David Beazley

I think that's the saddest a single line of python has ever made me.

—Someone on IRC (#python)

Don't speak around saying that I like it.

—A PyPy contributor

Fields is completely bat-shit insane, but kind of cool.

—Someone on IRC (#python)


—Unsuspecting victim at EuroPython 2015

I don't think it should work ...

—Unsuspecting victim at EuroPython 2015

Is it some Ruby thing?

—Unsuspecting victim at EuroPython 2015

Are Python programmers that lazy?

—Some Java developer

I'm going to use this in my next project. You're a terrible person.

—Isaac Dickinson

It's so bad you had to write a pylint plugin :)

—Colin Dunklau on IRC (#python)


I tried my best at EuroPython ...