[mod_python] apache.py suggestions

On Fri, 26 May 2000, Gregory Trubetskoy wrote:
> OK, here is the deal - 
> 
> 	exec "import " + module_name
> 	module = eval(module_name)
> 
> is not the same as module = __import__(module_name), but rather:
> 
> 	module = __import__(module_name)
> 	list = string.split(module_name, ".")
> 	for n in list[1:]:
> 		module = getattr(module, n)
> 
> My quick tests on which is faster where inconclusive. The former is
> less lines of code and is (IMHO) simpler to read, the latter is definitely
> more secure since it doesn't have the exec or eval.

Mine were quite conclusive. The latter code is 4 to 6 times faster.

I've attached my timing script. Here is the output of a couple runs:

[gstein at kurgan python]$ ./time-eval.py
16.6380729675 2.65576100349 6.2648984399
22.0597770214 5.24019098282 4.20972767858
[gstein at kurgan python]$ ./time-eval.py
16.5899840593 2.59417903423 6.39508061719
21.954955101 5.28538203239 4.15390126323

Note that each run already loops through 10000 times, so the numbers above
should be quite stable.

Yes, I understand that the first import is longer than the following
imports. And that I'm not testing the cost of that first import. This is
intended: the underlying import mechanism which runs on that first import
is the same for both (e.g. the parse/load and module setup), so it can be
considered a constant factor and omitted.

Much of this may be moot anyhow since the "import" logic in question
occurs at startup only. But I don't think the use of exec/eval should ever
be encouraged :-) (performance and security)

Cheers,
-g

-- 
Greg Stein, http://www.lyra.org/
-------------- next part --------------
#!/usr/bin/env python

import time
import string

def do_timing(m):
  l = xrange(10000)

  t = time.time()
  for i in l:
    exec "import " + m
    mod = eval(m)
  t1 = time.time() - t

  t = time.time()
  for i in l:
    mod = __import__(m)
    for p in string.split(m, '.')[1:]:
      mod = getattr(mod, p)
  t2 = time.time() - t
  print t1, t2, t1/t2

do_timing('httplib')
do_timing('xml.dom.core')