[XML-SIG] foo.bar vs. foo.get_bar()

[Dieter Maurer]

--Multipart_Sat_Nov__6_20:13:14_1999-1
Content-Type: text/plain; charset=US-ASCII

uche.ogbuji@fourthought.com writes:
 > Any class with __[g/s]etattr__ is automatically slower, whether or not the 
 > methods are actually used.  This is for a simple reason: without these hooks, 
 > attribute access involves basic C pointer magic within the interpreter, which 
 > is of course blazingly fast.  Whenever the hooks are present, the interpreter 
 > must invoke the hooks for every attribute access, even if it turns out the 
 > hooks aren't used.  As I've mentioned before, round trips from the C engine to 
 > Python interpreted code incur a steep performance penalty.
I think this is not generally true for "__getattr__":
reason: "__getattr__" is only used, if an attribute cannot be found
by "normal" lookup.

The attached script, for example produced the following timing:
__main__.C1 0.43
__main__.C2 0.43
__main__.C2 0.44
__main__.C1 0.43
There is no significant difference between C1 (without)
and C2 (with __getattr__).

This only changes, when attributes are looked up that are not
found by looking into the __dict__, e.g. because they do
not exist at all.

- Dieter


--Multipart_Sat_Nov__6_20:13:14_1999-1
Content-Type: application/x-python
Content-Disposition: attachment; filename="x.py"
Content-Transfer-Encoding: 7bit

'''measuring the effect of 'getattr'.'''
import time

class C1:
  a=1


class C2:
  a=2
  def __getattr__(self,k): pass

def test(c,n=10000):
  i= 0
  s= time.clock()
  while i < n:
    c.a; c.a; c.a; c.a; c.a; c.a; c.a; c.a; c.a; c.a
    i= i+1
  e= time.clock()
  print c, e-s

test(C1)
test(C2)
test(C2)
test(C1)


--Multipart_Sat_Nov__6_20:13:14_1999-1--