How to document class attributes in Python? [close

问题:

I'm writing a lightweight class whose attributes are intended to be publicly accessible, and only sometimes overridden in specific instantiations. There's no provision in the Python language for creating docstrings for class attributes, or any sort of attributes, for that matter. What is the accepted way, should there be one, to document these attributes? Currently I'm doing this sort of thing:

class Albatross(object):
    """A bird with a flight speed exceeding that of an unladen swallow.

    Attributes:
    """

    flight_speed = 691
    __doc__ += """
        flight_speed (691)
          The maximum speed that such a bird can attain.
    """

    nesting_grounds = "Raymond Luxury-Yacht"
    __doc__ += """
        nesting_grounds ("Raymond Luxury-Yacht")
          The locale where these birds congregate to reproduce.
    """

    def __init__(self, **keyargs):
        """Initialize the Albatross from the keyword arguments."""
        self.__dict__.update(keyargs)

This will result in the class's docstring containing the initial standard docstring section, as well as the lines added for each attribute via augmented assignment to __doc__.

Although this style doesn't seem to be expressly forbidden in the docstring style guidelines, it's also not mentioned as an option. The advantage here is that it provides a way to document attributes alongside their definitions, while still creating a presentable class docstring, and avoiding having to write comments that reiterate the information from the docstring. I'm still kind of annoyed that I have to actually write the attributes twice; I'm considering using the string representations of the values in the docstring to at least avoid duplication of the default values.

Is this a heinous breach of the ad hoc community conventions? Is it okay? Is there a better way? For example, it's possible to create a dictionary containing values and docstrings for the attributes and then add the contents to the class __dict__ and docstring towards the end of the class declaration; this would alleviate the need to type the attribute names and values twice. edit: this last idea is, I think, not actually possible, at least not without dynamically building the entire class from data, which seems like a really bad idea unless there's some other reason to do that.

I'm pretty new to python and still working out the details of coding style, so unrelated critiques are also welcome.

回答1:

To avoid confusion: the term property has a specific meaning in python. What you're talking about is what we call class attributes. Since they are always acted upon through their class, I find that it makes sense to document them within the class' doc string. Something like this:

class Albatross(object):
    """A bird with a flight speed exceeding that of an unladen swallow.

    Attributes:
        flight_speed     The maximum speed that such a bird can attain.
        nesting_grounds  The locale where these birds congregate to reproduce.
    """
    flight_speed = 691
    nesting_grounds = "Throatwarbler Man Grove"

I think that's a lot easier on the eyes than the approach in your example. If I really wanted a copy of the attribute values to appear in the doc string, I would put them beside or below the description of each attribute.

Keep in mind that in Python, doc strings are actual members of the objects they document, not merely source code annotations. Since class attribute variables are not objects themselves but references to objects, they have no way of holding doc strings of their own. I guess you could make a case for doc strings on references, perhaps to describe "what should go here" instead of "what is actually here", but I find it easy enough to do that in the containing class doc string.

回答2:

You cite the PEP257: Docstring Conventions, in the section What is a docstring it is stated:

String literals occurring elsewhere in Python code may also act as documentation. They are not recognized by the Python bytecode compiler and are not accessible as runtime object attributes (i.e. not assigned to __doc__), but two types of extra docstrings may be extracted by software tools:

String literals occurring immediately after a simple assignment at the top level of a module, class, or __init__ method are called "attribute docstrings".

And this is explained in more details in PEP 258: Attribute docstrings. As explains above ʇsәɹoɈ. an attribute is not an object that can own a __doc__ so they won't appear in help() or pydoc. These docstrings can only be used for generated documentation.

They are used in Sphinx with the directive autoattribute

Sphinx can use comments on a line before an assignment or a special comment following an assignment or a docstring after the definition which will be autodocumented.

回答3:

You could abuse properties to this effect. Properties contain a getter, a setter, a deleter, and a docstring. Naively, this would get very verbose:

class C:
    def __init__(self):
        self._x = None

    @property
    def x(self):
        """Docstring goes here."""
        return self._x

    @x.setter
    def x(self, value):
        self._x = value

    @x.deleter
    def x(self):
        del self._x

Then you will have a docstring belonging to C.x:

In [24]: print(C.x.__doc__)
Docstring goes here.

To do this for many attributes is cumbersome, but you could envision a helper function myprop:

def myprop(x, doc):
    def getx(self):
        return getattr(self, '_' + x)

    def setx(self, val):
        setattr(self, '_' + x, val)

    def delx(self):
        delattr(self, '_' + x)

    return property(getx, setx, delx, doc)

class C:
    a = myprop("a", "Hi, I'm A!")
    b = myprop("b", "Hi, I'm B!")

In [44]: c = C()

In [46]: c.b = 42

In [47]: c.b
Out[47]: 42

In [49]: print(C.b.__doc__)
Hi, I'm B!

Then, calling Pythons interactive help will give:

Help on class C in module __main__:

class C
 |  Data descriptors defined here:
 |  
 |  a
 |      Hi, I'm A!
 |  
 |  b
 |      Hi, I'm B!

which I think should be pretty much what you're after.

Edit: I realise now that we can perhaps avoid to need to pass the first argument to myprop at all, because the internal name doesn't matter. If subsequent calls of myprop can somehow communicate with each other, it could automatically decide upon a long and unlikely internal attribute name. I'm sure there are ways to implement this, but I'm not sure if they're worth it.