[JOSS] Slots

[Sulstice]

Forgive me but it's been awhile since I've seen the __slots__ argument. I remember some of the numpy folk arguing about it at the last PyData conference.

In src/kallisto/atom.py, Line 52.

    __slots__ = ["data", "molecule", "index"]

    def __init__(self, symbol="X", position=(0, 0, 0), charge=0, molecule=None):

        self.data = data = {}

Here's what I'm looking at, can you elaborate on your reason for choosing slots, and does it actually make access to your class attributes access that much faster?

[f3rmion]

Hi @Sulstice,

I used it to save RAM if someone wants to create objects with thousands of atoms, since default Python uses a dict to store an object's instance attribute. However, for small classes with known attributes, this could be quite a waste of memory.

You can try this yourself (I found a nice example here):

slots.py

class MyClass(object):
        __slots__ = ['name', 'identifier']
        def __init__(self, name, identifier):
                self.name = name
                self.identifier = identifier

num = 1024*256
x = [MyClass(1,1) for i in range(num)]

noslots.py

class MyClass(object):
        def __init__(self, name, identifier):
                self.name = name
                self.identifier = identifier

num = 1024*256
x = [MyClass(1,1) for i in range(num)]

We can check the memory usage with this excellent tool.

The difference between both shows the memory aspect of __slots__ not sure about the speed though.
It seems to be faster, but this is not really a benchmark :)

JOSS

[RMeli]

May I suggest to change the variable __slots__ into a tuple? This is nitpicking, but according to "Fluent Python" by Luciano Ramalho it better conveys the fact that the __slots__ definition can't change.

[f3rmion]

OK, I changed slots into a tuple to stress out its immutability.

[f3rmion]

I think this can be closed since the main question (why using slots?) has been answered.