Skip to content

Latest commit

 

History

History
122 lines (88 loc) · 3.7 KB

README.md

File metadata and controls

122 lines (88 loc) · 3.7 KB
Raw

bummer

Python metaclass for creating generic lazy objects

Requirements

  • Python 3.x (not tested on Python 2.x, but maybe it works)

Usage

import bummer

class Foo(object, metaclass=bummer.LazyMeta):
    def __init__(self, index, value):
        print("Hi, __init__ is being called!")
        self.index = index
        self.value = self.expensive_computation(value)

    def __lazy_preinit__(self, index, value):
        print("__lazy_preinit__ is called first.")
        self.index = index

    def expensive_computation(self, value):
        print("Performing expensive computation.")
        return len(repr(value)) % 42

Lazy instantiation

In [4]: foo = Foo(10, "something")
__lazy_preinit__ is called first.

In [5]: foo.index
Out[5]: 10

In [6]: foo.value
Hi, __init__ is being called!
Performing expensive computation.
Out[6]: 11

Preserving type hierarchy

In [7]: foo2 = Foo(20, "other")
__lazy_preinit__ is called first.

In [8]: type(foo2)
Out[8]: bummer.Foo_Lazy

In [9]: isinstance(foo2, Foo)
Out[9]: True

In [10]: foo2.value
Hi, __init__ is being called!
Performing expensive computation.
Out[10]: 7

In [11]: type(foo2)
Out[11]: __main__.Foo

How it works

LazyMeta is a metaclass that creates "half-initialized" objects.

Python objects are initialized in a two-step process, where memory for the object is first allocated (see __new__()), and then the object's initialization function is called (see __init__()).

When LazyMeta is used to turn some target class into a lazified class, it first creates a stub class (_Lazy) associated with that class. Then, whenever objects of the target class are to be created, the stub class is returned instead.

The _Lazy stub class, in its implementation of __new__(), uses the target class's __new__() to create a new half-initialized object of the target class. The class of the half-initialized object is then changed to _Lazy, which results in the target class's __init__() to not run. As a result, the object returned when an instance of the lazified target class is requested is a half-initialized target class object, masquerading as a _Lazy object.

By overriding __getattr()__ and __setattr__(), the _Lazy class then ensures that when any attribute on the object is accessed, reification takes place. When the object is reified, the class of the object is switched back to the target class, and the call to the target class's __init__() is made, completing initialization of the object. The resultant object then behaves exactly like it would if it had never been lazified.

The __lazy_preinit__() function is designed to allow certain attributes to be set on the lazy object, which can be accessed without triggering reification.

Why is this so complex?

It is clearly a lot easier to do this by implementing some kind of a proxy object, and overriding __getattribute__() on it to redirect to the target object. However, this approach would result in a check on the status of the object (whether it has been reified or not) every time an attribute was accessed.

I wanted to find a way to create lazy objects where such a check did not happen; that once a lazy object is reified, access to its attributes happen with no additional overhead. So this is the solution I ended up with.

Is this really a good idea?

I don't know, probably not. It involves magic (changing __class__ for example) that is likely bad practice. I've also not tested it on other kinds of objects, especially more complex ones. So far though, it works for me. Use it at your own risk, and file bug reports if you find any!

TODO

  • Document LazyFactoryMeta in README.

License

MIT License

See LICENSE for full text.