XOM API for Processing XML with Java

[Elliotte Harold <elharo AT metalab.unc.edu>]

Wolfgang Hoschek wrote:

> The reason I'm asking requires some background:
> I worked on XOM extensions for a high-performance messaging 
> environment using small XML messages. I started with dozens of 
> messages/sec throughput, and after a lot of sweat have now reached 
> thousands of messages/sec (even when turning on W3C XML Schema 
> validation).
>
> Doing some profiling it turns out that the baseUri calculation in 
> Builder.build(File) takes much more time than all of file I/O, parsing 
> and validation combined. The reason is that Builder.build(File) use 
> URLEncoder.encode(String.valueOf(char)) in a loop to compute the 
> baseURI. To confirm that, try replacing the calculation with 
> File.toURI().toString() or a combination of the code found in 
> FileSource.setSystemID(File file) with URIUtil.toURI(...).
That's strange. I guess it's just one of the unoptimized parts of the 
core API. If you know the URL you're starting with, you should probably 
just pass it in along with the FileInputStream, rather than letting Java 
calculate it for you. However, there is one advantage to letting Java do 
the calculation. The format of file URLs varies a bit from platform to 
platform. Letting the File class figure it out gives correct results on 
Windows, Unix, and other platforms.

I'd be tempted to file a bug report with Sun on this one. There's no 
reason URL encoding should take longer than XML parsing. It's a very 
straightforward operation. I could write my own code to create file URLs 
from a File object. However, I'd be very nervous about getting it wrong 
on some platforms, and also causing problems if Sun ever updates their 
VM so this operation works reasonably.

--
Elliotte Rusty Harold
elharo AT metalab.unc.edu