Re: [Jython-dev] Query: Was the closure class generation mechanism improved in a1?

[Kent J. <ke...@td...>]

Leo User wrote:
> Howdy,
>  
> ok to the point, in jython-a0  I ran into the strangest behavior when I 
> was developing jyleo.  I had a method that wrapped a position iterator 
> with a java.util.Iterator.  This was all fine.  But when I wrote out a 
> leo file the memory increased by 6%, each time!.  I could not fathom 
> what was going on and had to instrument the jython implementation to get 
> visibility.  What I found was somewhat startling, a new class was 
> generated each time the method was called. 
>  
> so a seemingly harmless method/closure like this:
> def example():
> ...class example_closure:
> ......pass
> ...return example_closure()
>  
> ended up eating up alot of memory.  For methods that get called alot 
> this cant be used.  For those that called a little its ok.  But I must 
> ask, does anyone think this is normal behavior?  We debated this on the 
> Leo forums and Edward Ream pointed out that class is an executable 
> statement, hence generating a new class each time was correct.   I still 
> find myself debating the utility of such a mechanism, I find myself 
> saying "Jython Closures Considered Harmful".

I would have to agree with Edward - class is an executable statement. CPython has the same behavior:
 >>> def example():
 ...   class example_closure:
 ...     pass
 ...   return example_closure()
 ...
 >>> a=example()
 >>> a.__class__
<class __main__.example_closure at 0x008CE960>
 >>> b=example()
 >>> b.__class__
<class __main__.example_closure at 0x008CEA50>


If the nested class doesn't actually need the closure then just pull it out of the def:
class example_closure:
  pass
def example():
  return example_closure()

Alternately you could pass the relevant part of the closure to the example_closure() constructor:
class example_closure:
  def __init__(self, state):
    self.state = state
def example(state):
  return example_closure(state)

Kent