On 12/20/2016 03:39 PM, Ben Finney wrote:
"Mr. Wrobel" writes:
Quick question, can anybody tell me when to use __init__ instead of
__new__ in meta programming?
Use ‘__new__’ to do the work of *creating* one instance from nothing;
allocating the storage, determining the type, etc. — anything that will
be *the same* for every instance. ‘__new__’ is the constructor.
Use ‘__init__’ to do the work of *configuring* one instance, after it
already exists. Any attributes that are special to an instance should be
manipulated in the ‘__init__’ method. ‘__init__’ is the initialiser.
That sounds like general object creation/class advice, which as a general
guideline is okay, but don't feel like it's the most important thing.
I only use `__new__` when the object being created is (or is based on)
an immutable type; otherwise I use `__init__`. Likewise, if I'm using
`__new__` then I do all my configuration in `__new__` unless I have a
really good reason not to (such as making it easier for subclasses to
modify/skip `__init__`).
As far as metaclasses go... the only time I recall writing an `__init__`
for a metaclass was to strip off the extra arguments so `type.__init__`
wouldn't fail.
--
~Ethan~
--
https://mail.python.org/mailman/listinfo/python-list