>  FYI, the transformation that uses keys is only slightly 
>  slower than the recursive transformation. With the same 
>  source xml it takes 1151ms.
>  
>  This means that the results of my experiments significantly 
>  contradict your findings. Maybe you didn't use exactly the 
>  same transformation or are using an older version of Xalan?

Yes, Xalan-J 2.4.1 is *significantly* faster than 2.4.0 
using the key method. However, our findings don't really
contradict because you are using far fewer continuation 
records than I am.

Here is my summary of the various methods
with some Xalan processors (values in seconds) using 
Dimitre's stylesheets. I probably would get similar
behavior, though almost certainly faster timings with 
Saxon, et al.):

Method              2.4.0         2.4.1        2.5.2
------------------------------------------------------------
Recursive
  200 records       1.1           1.1          1.6
  1000 records      2.8           2.5          5.8
  2000 records      StackOv       StackOv      Partial Output
                                               (No Exception!)
Key
  200 records       120           2.7          2.7
  1000 records      dnf           21           20
  2000 records      dnf           69           68

Generate-Id
  200 records       1.8           1.8          1.8
  1000 records      17            17           18
  2000 records      59            65           61

So all of the methods except recursion exhibit exponential
behavior with an increase in continuation records except
recursion.

So if there is a way to do solve this problem recursively 
that would avoid the stack overflow errors (2.5.2 doesn't 
even do you the courtesy of throwing an exception -- it 
just dies in the middle of the transformation), that would 
still be by far the best solution.

Cheers,
Raman Gupta

 XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list