[ruby-talk:19706] Re: class methods and optimizing

CM> Does anyone know what class methods are as opposed to instance methods? Are
CM> they for optimizing? when would I use them?I have looked at a couple manuals 
CM> that tell you how to do them: def classname.method but none seem to explain 
CM> the why and when.

I asked a similar question in the past and Hal Fulton sent me this
highly excellent explanation:

--------------
When a method describes functionality that should belong to an
individual object (an instance), it should be an instance method.

When the functionality is NOT tied to a specific instance, it should
be a class method.

For a built-in example: Why is chmod  a class method of File? Well,
it's because doing a chmod on a file really needs no other information
besides the name. Why should you have to create a file object just to
do this operation? (Of course, there is also an instance method of the
same name -- meaning you *can* do it to an instance of File, probably
one you created for some other purpose.)

    File.chmod(0555, "myfile")   # No File object in existence!

By the way, it's not uncommon for a class method to return an instance
of the class. For example, Something.new (in general ) returns a
Something object; Thread.fork returns a thread; Regexp.compile returns
a regular expression; and so on.

It's perhaps easier to understand for data than for methods. Is the scope
of this piece of data a single object? Or is it the entire class?

For example, there is a Boolean value that controls whether threads abort
when an exception occurs. It doesn't apply to just one thread; it's an
attribute
of the entire set of threads -- an attribute of the class.

Thus we can check it or set it with class methods:

    flag = Thread.abort_on_exception
    Thread.abort_on_exception = true

The classic example of class data is to count the number of objects
created. This is possible if we keep a single counter for the class
which is incremented for each object.
-------------

CM> And this leads to my second question.  In terms of memory usage and speed, is 
CM> it better to extend the existing classes (when possible) or to build your 
CM> own, or to just make a subclass? e.g. If I program a deck routine for a card 
CM> game I can just add shuffle, play etc. routines to class Array.  Now the 
CM> computer isn't holding an extra class definition in memory, but ALL arrays 
CM> (not decks) are also extended. If I make my own class/subclass, not all of 
CM> the arrays are extended but there is another class definition in memory...

I don't know about memory usage or speed, but I would be wary of
changing (or adding to) the functionality of standard classes that
everyone relies on, because if your code gets used by someone else, it
may cause unexpected results.  It would be nice if you could somehow
limit the scope of your changes to a class.  That way, if you wanted
to change the way Array works, you can do it without breaking other
code used in the same program that changes Array in a different way.
I'm not sure if such a thing is possible or even considered a good
idea by matz, but I think it's a good idea for now...

thanks,
-joe