A place to cache linked articles (think custom and personal wayback machine)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

index.md 4.4KB

title: dataklasses: A different spin on dataclasses. url: https://github.com/dabeaz/dataklasses#questions-and-answers hash_url: 891705d155

Dataklasses is a library that allows you to quickly define data classes using Python type hints. Here’s an example of how you use it:

from dataklasses import dataklass

@dataklass
class Coordinates:
    x: int
    y: int

The resulting class works in a well civilised way, providing the usual __init__(), __repr__(), and __eq__() methods that you’d normally have to type out by hand:

>>> a = Coordinates(2, 3)
>>> a
Coordinates(2, 3)
>>> a.x
2
>>> a.y
3
>>> b = Coordinates(2, 3)
>>> a == b
True
>>>

It’s easy! Almost too easy.

Wait, doesn’t this already exist?

No, it doesn’t. Yes, certain naysayers will be quick to point out the existence of @dataclass from the standard library. Ok, sure, THAT exists. However, it’s slow and complicated. Dataklasses are neither of those things. The entire dataklasses module is less than 100 lines. The resulting classes import 15-20 times faster than dataclasses. See the perf.py file for a benchmark.

Theory of Operation

While out walking with his puppy, Dave had a certain insight about the nature of Python byte-code. Coming back to the house, he had to try it out:

>>> def __init1__(self, x, y):
...     self.x = x
...     self.y = y
...
>>> def __init2__(self, foo, bar):
...     self.foo = foo
...     self.bar = bar
...
>>> __init1__.__code__.co_code == __init2__.__code__.co_code
True
>>>

How intriguing! The underlying byte-code is exactly the same even though the functions are using different argument and attribute names. Aha! Now, we’re onto something interesting.

The dataclasses module in the standard library works by collecting type hints, generating code strings, and executing them using the exec() function. This happens for every single class definition where it’s used. If it sounds slow, that’s because it is. In fact, it defeats any benefit of module caching in Python’s import system.

Dataklasses are different. They start out in the same manner--code is first generated by collecting type hints and using exec(). However, the underlying byte-code is cached and reused in subsequent class definitions whenever possible. Caching is good.

A Short Story

Once upon a time, there was this programming language that I’ll refer to as “Lava.” Anyways, anytime you started a program written in Lava, you could just tell by the awkward silence and inactivity of your machine before the fans kicked in. “Ah shit, this is written in Lava” you’d exclaim.

Questions and Answers

Q: What methods does dataklass generate?

A: By default __init__(), __repr__(), and __eq__() methods are generated. __match_args__ is also defined to assist with pattern matching.

Q: Does dataklass enforce the specified types?

A: No. The types are merely clues about what the value might be and the Python language does not provide any enforcement on its own.

Q: Are there any additional features?

A: No. You can either have features or you can have performance. Pick one.

Q: Does dataklass use any advanced magic such as metaclasses?

A: No.

Q: How do I install dataklasses?

A: There is no setup.py file, installer, or an official release. You install it by copying the code into your own project. dataklasses.py is small. You are encouraged to modify it to your own purposes.

Q: What versions of Python does it work with?

A: The code will work with versions 3.9 and later.

Q: But what if new features get added?

A: What new features? The best new features are no new features.

Q: Who maintains dataklasses?

A: If you’re using it, you do. You maintain dataklasses.

Q: Is this actually a serious project?

A: It’s best to think of dataklasses as more of an idea than a project. There are many tools and libraries that perform some form of code generation. Dataklasses is a proof-of-concept for a different approach to that. If you’re a minimalist like me, you’ll find it to be perfectly functional. If you’re looking for a lot of knobs to turn, you should probably move along.

Q: Should I give dataklasses a GitHub star?

A: Yes, because it will help me look superior to the other parents with kids in the middle-school robot club.

Q: Who wrote this?

A: dataklasses is the work of David Beazley. http://www.dabeaz.com.