I have implemented generating text strings as char arrays, with a compiler option to turn it on and off, and did some timings on the benchmark. Just as I suspected, the benchmark did not show much performance difference, when the option was turned on and then off.
Now I am wondering if there is any value in committing my change. :(
Well, if it does improve results, it can't hurt, but the two should likely be similar. What did you get in your testing ?
I think what should be optimized is that there's twice the amount of writes.
Looking at the code generated, this situation seems rather common. So the idea would be to do a pass to merge all those TemplateText nodes :)
The profile showed about 75% of the CPU was spent in the JSP service (as usual with the VM profiles, it's probably misleading), and probably some of that time comes from the fact it has to invoke hundreds of methods.
Remy
I took a look at o.a.j.runtime.JspWriterImpl and the methods
write(String s, int off, int len) and write(char cbuf[], int off, int len)
in particualr, and failed to see any performance gain of the 2nd over the first. They are very similar, the only different is that the first uses String.getChars while the second uses System.arraycopy, but those two methods should be on par in terms of performance. That is, I don't see any extra unnecessary copies.
Perhasp the culprit is those extra writes?
-Kin-man
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
