[ratsbane]

That's not necessarily true. If the compiler is smart enough it won't recalculate the value of a numeric expression which doesn't contain any variables. On the other hand, dereferencing a variable isn't without cost. Of course, in most cases you'd be dereferencing variables anyway in the test clause, so the original idea, compute loop limits before entering the loop, is correct.:

mbp:~/test doug$ time php -r 'for ($i=1; $i<100000000/10; $i++) {$x++;};'

  real    0m5.930s
  user    0m5.867s
  sys     0m0.026s

  mbp:~/test doug$ time php -r '$maxval=100000000/10; for ($i=1; $i<$maxval; $i++) {$x++;};'

  real    0m6.999s
  user    0m6.907s
  sys     0m0.029s



  mbp:~/test doug$ time perl -e 'for ($i=1; $i<100000000/10; $i++) {$x++;}'

  real    0m2.520s
  user    0m2.496s
  sys     0m0.010s

  mbp:~/test doug$ time perl -e '$maxval=100000000/10; for ($i=1; $i<$maxval; $i++) {$x++;}'

  real    0m2.537s
  user    0m2.499s
  sys     0m0.014s

The results are a little different with Python (maxval1.py computes inline; maxval2 divides in advance and uses variable inline)

  mbp:~/test doug$ time python maxval1.py

  real    0m2.134s
  user    0m1.761s
  sys     0m0.276s

  mbp:~/test doug$ time python maxval2.py

  real    0m1.890s
  user    0m1.588s
  sys     0m0.278s

Java shows even more of a difference (again, maxval2 sets variable in advance):

  mbp:~/test doug$ time java maxval1

  real    0m0.174s
  user    0m0.116s
  sys     0m0.040s

  mbp:~/test doug$ time java maxval2

  real    0m0.531s
  user    0m0.125s
  sys     0m0.041s

[ratsbane]

Here is the java code too. First maxval1.java:

  public class maxval1 {
    static int $i, $x;

    public static void main(String args[]) throws Exception {
      for ($x=$i=1; $i<100000000/10; $i++) {$x++;}
      System.out.println($x);
      }
  }

And maxval2.java:

  public class maxval2 {
    static int $a, $i, $x;

    public static void main(String args[]) throws Exception {
      $a=100000000/10;
      for ($x=$i=1; $i<$a; $i++) {$x++;}
      System.out.println($x);
      }
  }

[andreyf]

Does anyone have any ideas for what's happening in the Python example? Could you post the Python code?

[ratsbane]

Sure... here is maxval1.py:

  for i in range(1, 100000000/10):
    pass

And here is maxval2.py:

  maxval=100000000/10
  for i in range(1, maxval):
    pass

[andreyf]

Oh, of course - range() is evaluated once in either case. It returns an array, which is then stepped through. The time difference must have been just some inaccuracy. I guess 'time' isn't the best command to use to test efficiency ... or maybe try adding another 0?

[ratsbane]

Oh, thanks. I redid it using xrange instead of range and the numbers went down to an average (over 20 tries) of about 1.36 sec for maxval1 (which computes endpoint within the xrange) and 1.9 sec for maxval2 (which uses variable in xrange.)