Re: [xml-dev] Performance comparisons of tree models

On 9/12/2012 3:50 PM, Michael Kay wrote:
>
> On 12/09/2012 20:27, Mike Sokolov wrote:
>> You are correct - our current implementation doesn't cache compiled 
>> XPaths, so that cost was included in the comparison.  I suppose we'll 
>> want to begin doing that (caching, as we already do for XSLT).
>>
>
> If you use the s9api XPath APIs then Saxon caches automatically.
>
That's good to know. I went back and checked the benchmark code I ran 
more carefully, we are in fact using the s9api and the JDOM/Jaxen XPath 
was *not* cached across executions, so a bit of apples and oranges here; 
I'll need to measure again, I think. Another factor making our 
measurements less clear is that they include a few other components of 
our system as well, such as fetching documents from storage, parsing and 
building the model.  So there are a number of constant factors included 
in both measurements that tend to lessen any difference due to XPath 
execution.  I'll have to see if the retrieval part is dominating: I 
expect it is.

-Mike