Adding docstrings to namedtuples?
Is it possible to add a documentation string to a namedtuple in an easy manner?
Yes, in several ways.
Subclass typing.NamedTuple - Python 3.6+
As of Python 3.6 we can use a class
definition with typing.NamedTuple
directly, with a docstring (and annotations!):
from typing import NamedTuple
class Card(NamedTuple):
"""This is a card type."""
suit: str
rank: str
Compared to Python 2, declaring empty __slots__
is not necessary. In Python 3.8, it isn't necessary even for subclasses.
Note that declaring __slots__
cannot be non-empty!
In Python 3, you can also easily alter the doc on a namedtuple:
NT = collections.namedtuple('NT', 'foo bar')
NT.__doc__ = """:param str foo: foo name
:param list bar: List of bars to bar"""
Which allows us to view the intent for them when we call help on them:
Help on class NT in module __main__:
class NT(builtins.tuple)
| :param str foo: foo name
| :param list bar: List of bars to bar
...
This is really straightforward compared to the difficulties we have accomplishing the same thing in Python 2.
Python 2
In Python 2, you'll need to
- subclass the namedtuple, and
- declare
__slots__ == ()
Declaring __slots__
is an important part that the other answers here miss .
If you don't declare __slots__
- you could add mutable ad-hoc attributes to the instances, introducing bugs.
class Foo(namedtuple('Foo', 'bar')):
"""no __slots__ = ()!!!"""
And now:
>>> f = Foo('bar')
>>> f.bar
'bar'
>>> f.baz = 'what?'
>>> f.__dict__
{'baz': 'what?'}
Each instance will create a separate __dict__
when __dict__
is accessed (the lack of __slots__
won't otherwise impede the functionality, but the lightweightness of the tuple, immutability, and declared attributes are all important features of namedtuples).
You'll also want a __repr__
, if you want what is echoed on the command line to give you an equivalent object:
NTBase = collections.namedtuple('NTBase', 'foo bar')
class NT(NTBase):
"""
Individual foo bar, a namedtuple
:param str foo: foo name
:param list bar: List of bars to bar
"""
__slots__ = ()
a __repr__
like this is needed if you create the base namedtuple with a different name (like we did above with the name string argument, 'NTBase'
):
def __repr__(self):
return 'NT(foo={0}, bar={1})'.format(
repr(self.foo), repr(self.bar))
To test the repr, instantiate, then test for equality of a pass to eval(repr(instance))
nt = NT('foo', 'bar')
assert eval(repr(nt)) == nt
Example from the documentation
The docs also give such an example, regarding __slots__
- I'm adding my own docstring to it:
class Point(namedtuple('Point', 'x y')): """Docstring added here, not in original""" __slots__ = () @property def hypot(self): return (self.x ** 2 + self.y ** 2) ** 0.5 def __str__(self): return 'Point: x=%6.3f y=%6.3f hypot=%6.3f' % (self.x, self.y, self.hypot)
...
The subclass shown above sets
__slots__
to an empty tuple. This helps keep memory requirements low by preventing the creation of instance dictionaries.
This demonstrates in-place usage (like another answer here suggests), but note that the in-place usage may become confusing when you look at the method resolution order, if you're debugging, which is why I originally suggested using Base
as a suffix for the base namedtuple:
>>> Point.mro()
[<class '__main__.Point'>, <class '__main__.Point'>, <type 'tuple'>, <type 'object'>]
# ^^^^^---------------------^^^^^-- same names!
To prevent creation of a __dict__
when subclassing from a class that uses it, you must also declare it in the subclass. See also this answer for more caveats on using __slots__
.
Came across this old question via Google while wondering the same thing.
Just wanted to point out that you can tidy it up even more by calling namedtuple() right from the class declaration:
from collections import namedtuple
class Point(namedtuple('Point', 'x y')):
"""Here is the docstring."""
You can achieve this by creating a simple, empty wrapper class around the returned value from namedtuple
. Contents of a file I created (nt.py
):
from collections import namedtuple
Point_ = namedtuple("Point", ["x", "y"])
class Point(Point_):
""" A point in 2d space """
pass
Then in the Python REPL:
>>> print nt.Point.__doc__
A point in 2d space
Or you could do:
>>> help(nt.Point) # which outputs...
Help on class Point in module nt: class Point(Point) | A point in 2d space | | Method resolution order: | Point | Point | __builtin__.tuple | __builtin__.object ...
If you don't like doing that by hand every time, it's trivial to write a sort-of factory function to do this:
def NamedTupleWithDocstring(docstring, *ntargs):
nt = namedtuple(*ntargs)
class NT(nt):
__doc__ = docstring
return NT
Point3D = NamedTupleWithDocstring("A point in 3d space", "Point3d", ["x", "y", "z"])
p3 = Point3D(1,2,3)
print p3.__doc__
which outputs:
A point in 3d space
In Python 3, no wrapper is needed, as the __doc__
attributes of types is writable.
from collections import namedtuple
Point = namedtuple('Point', 'x y')
Point.__doc__ = '''\
A 2-dimensional coordinate
x - the abscissa
y - the ordinate'''
This closely corresponds to a standard class definition, where the docstring follows the header.
class Point():
'''A 2-dimensional coordinate
x - the abscissa
y - the ordinate'''
<class code>
This does not work in Python 2.
AttributeError: attribute '__doc__' of 'type' objects is not writable
.