A class to access dicts using attribute syntax

I like the idea, but more as an exercise or a demonstration what can be done with Python. I also like that the code comes with a set of test cases which are written in a way that they nicely serve as user documentation. In fact, I could even imagine this example being used in a tutorial about the __setattr__ etc. methods.

However, unless there is a real use-case where such a feature is needed for some reason, I would rather not start using it only for the syntactic difference: For readers of code, such tricks increase the cognitive complexity.

Having one idiomatic way of doing something makes things easier in most cases (https://peps.python.org/pep-0020):

There should be one -- and preferably only one -- obvious way to do it.

posted almost 4 years ago

CC BY-NC-SA 4.0

Dirk Herrmann‭

1407 reputation 1 31 143 48

1. It is useful sometimes, for example consider a program accepting input in the form of templates. An input of "Value: {obj.field}" is more readable than "Value: {obj['field']}".

   • Such an example is an app processing JSON and evaluating expressions on it, like jq or a similar Python app, pjy: pjy d.item.subitem instead of pjy "d['item']['subitem']"

2. There are other ways to do it:

   • Subclass dict and implement def __getattr__(self, attr): return self[attr]
   • Use self.__dict__ instead of self._dictionary

posted almost 4 years ago

CC0

luser‭

229 reputation 2 2 28 0

Ignoring whether it's a good idea, there are a few mistakes in the implementation:

• The magic method __getattr__ is called "when the default attribute access fails". To control all attribute access, you want __getattribute__.
  • By the same token, super().__getattr__ is not defined.
• __delete__ is for descriptors only; that should be __delattr__.
• Deletion methods (__delattr__ and __delete__) don't return anything.

And some surface-level things could be improved:

• Pylint: "Unnecessary else after return; remove the else and de-indent the code inside it (R1705:no-else-return)". In effect, convert your if-else's to guard statements.
• Pylint: "Consider explicitly re-raising using raise AttributeError(error) from error (W0707:raise-missing-from)". This changes part of the error message from "During handling of the above exception, another exception occurred:" to "The above exception was the direct cause of the following exception:"
• assert is a statement, not a function, so the parentheses are not needed and can be confusing
• Methods should be separated by 1 newline, not 2, per PEP 8
• Add a __repr__! It's easy and helps a lot with debugging.

class DictProxy:
    # [snip]

    def __getattribute__(self, name):
        """
        Redirect non-underscore attribute reads to dictionary reads
        """
        if name[0] == "_":
            return super().__getattribute__(name)

        try:
            return self._dictionary[name]
        except KeyError as error:
            # hasattr fails if the exception isn't an AttributeError
            raise AttributeError(error) from error

    def __delattr__(self, name):
        """
        Redirect non-underscore attribute deletes to dictionary deletes
        """
        if name[0] == "_":
            super().__delattr__(name)
        else:
            del self._dictionary[name]

    def __repr__(self):
        """Show construction."""
        clsname = type(self).__name__
        return f'{clsname}({self._dictionary!r})'


def test():
    dictionary = {"a": 1, "b": 2, "c": 3}
    proxy = DictProxy(dictionary)

    assert hasattr(proxy, "a")
    assert hasattr(proxy, "b")
    assert hasattr(proxy, "c")
    assert not hasattr(proxy, "d")

    # [etc.]

posted 14 days ago

CC BY-NC-SA 4.0

12d ago

wjandrea‭

31 reputation 0 1 3 3

I've used something similar in the past (supporting attribute access as equivalent to dict item access), but I would then always derive the class from dict (or UserDict). This can be useful - but mainly inside your own code, never as library code - if you have sections of the code that otherwise would need a lot of quoting plus square brackets.

It's tricky code, and I'd generally advise not to use it however:

• If a dictionary is passed in as argument to the class init, then after constructing the object, you have two, entirely separate references to a mutable dict in your code. This easily allows bugs to crawl in. It seems better to derive the class from dict or UserDict to prevent this.
• "DictProxy" seems a somewhat misleading or confusing name since it's not really a dict proxy: it doesn't add behavior to a class that otherwise behaves like a dict. It's also a bit confusing because the stdlib has various abc.mapping classes (and types.MappingProxy). I think this is yet another reason why it may be better to derive the class from collections.UserDict (or do sth similar).
• If the class is merely intended as view on the underlying dict data, I would make sure that instances of it are immutable (derive the class from Mapping or MappingView for instance).
• There seems to be an implicit assumption that the underlying dict will not have keys that are not valid Python attribute identifiers. This assumption could also lead to trouble (since certain keys will not be representable as attributes).

posted over 1 year ago

CC BY-NC-SA 4.0

1y ago

mudskipper‭

31 reputation 0 2 3 1