Skip to main content

Container class boilerplate killer.

Project description

docs

Documentation Status

tests

Travis-CI Build Status AppVeyor Build Status Requirements Status
Coverage Status Coverage Status
Code Quality Status Scrutinizer Status Codacy Code Quality Status CodeClimate Quality Status

package

PyPI Package latest release PyPI Package monthly downloads PyPI Wheel Supported versions Supported implementations

Container class boilerplate killer.

Features:

  • Human-readable __repr__

  • Complete set of comparison methods

  • Keyword and positional argument support. Works like a normal class - you can override just about anything in the subclass (eg: a custom __init__). In contrast, hynek/characteristic forces different call schematics and calls your __init__ with different arguments.

Installation

pip install fields

Usage & examples

A class that has 2 attributes, name and size:

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

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(Fields.name.ingredients.crust.size):
...     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:

>>> Fields.name.size("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)
True
>>> p = Pair(1, 2)
>>> p.a
1
>>> p.b
2
>>> tuple(p)
(1, 2)
>>> a, b = p
>>> a
1
>>> b
2

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
--------------------------------------------------------------------------------------

Documentation

https://python-fields.readthedocs.org/

Development

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

tox

FAQ

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, repetivive and unflexible. 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:

.. sourcecode:: pycon
>>> 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__. If you define a custom __init__, it will get overridden by the one attrs generates.

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

All in all, attrs is a fast and minimal container library with no support for subclasses. Definitely worth considering.

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.

Testimonials

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 completey bat-shit insane, but kind of cool.

—Someone on IRC (#python)

WHAT?!?!

—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

Apologies

I tried my best at EuroPython

Changelog

5.0.0 (2016-04-13)

  • Added the fields.InheritableFields base. It allows subclassing and it’s intended for multiple inheritance scenarios. Yes, yes, this enables lots pain and suffering, but some people just like it that way.

  • Simplified the interfaces for the builtin sealers (the required argument is gone as it was redundant, and the remaining arguments are swapped). Now they must be a function that take just two arguments: fields, defaults.

4.0.0 (2016-01-28)

  • Added __all__ and factory conveniences. Removed fields.Factory from the public API since it need some special care with it’s use (it’s a damn metaclass after all).

  • Added make_init_func into public API for advanced uses (combine with factory and class_sealer).

3.0.0 (2015-10-04)

  • Disallowed creating containers with fields with “dunder” names. E.g.: class Foo(Fields.__foo__): is disallowed.

2.4.0 (2015-06-13)

  • Similarly to fields.Fields, added three new bases:

    • fields.BareFields (implements __init__).

    • fields.ComparableMixin (implements __eq__, __ne__, __lt__, __gt__, __le__, __ge__ and __hash__).

    • fields.PrintableMixin (implements __repr__).

  • Improved reference section in the docs.

  • Added fields.ConvertibleFields and fields.ConvertibleMixin. They have two convenience properties: as_dict and as_tuple`.

2.3.0 (2015-01-20)

  • Allowed overriding __slots__ in SlotsFields subclasses.

2.2.0 (2015-01-19)

  • Added make_init_func as an optional argument to class_sealer. Rename the __base__ option to just base.

2.1.1 (2015-01-19)

  • Removed bogus console_scripts entrypoint.

2.1.0 (2015-01-09)

  • Added SlotsFields (same as Fields but automatically adds __slots__ for memory efficiency on CPython).

  • Added support for default argument to Tuple.

2.0.0 (2014-10-16)

  • Made the __init__ in the FieldsBase way faster (used for fields.Fields).

  • Moved RegexValidate in fields.extras.

1.0.0 (2014-10-05)

  • Lots of internal changes, the metaclass is not created in a closure anymore. No more closures.

  • Added RegexValidate container creator (should be taken as an example on using the Factory metaclass).

  • Added support for using multiple containers as baseclasses.

  • Added a super() sink so that super().__init__(*args, **kwargs) always works. Everything inherits from a baseclass that has an __init__ that can take any argument (unlike object.__init__). This allows for flexible usage.

  • Added validation so that you can’t use conflicting field layout when using multiple containers as the baseclass.

  • Changed the __init__ function in the class container so it works like a python function w.r.t. positional and keyword arguments. Example: class MyContainer(Fields.a.b.c[1].d[2]) will function the same way as def func(a, b, c=1, d=2) would when arguments are passed in. You can now use MyContainer(1, 2, 3, 4) (everything positional) or MyContainer(1, 2, 3, d=4) (mixed).

0.3.0 (2014-07-19)

  • Corrected string repr.

0.2.0 (2014-06-28)

  • Lots of breaking changes. Switched from __call__ to __getitem__ for default value assignment.

0.1.0 (2014-06-27)

  • Alpha release.

Project details


Download files

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

Source Distribution

fields-5.0.0.tar.gz (36.2 kB view hashes)

Uploaded Source

Built Distribution

fields-5.0.0-py2.py3-none-any.whl (19.8 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page