Torsten Curdt wrote:
On 10.10.2005, at 20:05, Vadim Gritsenko wrote:
Torsten Curdt wrote:
...or am I really mistaken?
JNI calls are always expensive, and in some scenarios (IIRC, on Mac
OS X it is *very* expensive to do File.getCanonicalName) it can be
even worse.
And if you take site of middle complexity, and take into account all
sitemap aggregations, XSLT includes, and so on, you can end up with
dozens of JNI file system accesses for *each* request.
Here, either delayed validity or complete turn off of checking helps
a lot.
I know :)
But we already delay the checking and therefor heavily
reduce the native filesystem checks ...and I was just
playing devil's advocate here.
Are we?
* XSLTProcessorImpl uses source validity / aggregate validity,
I don't see any delays in there.
* Cocoon.java has hardcoded delay
// FIXME: add a configuration option for the refresh delay.
// for now, hard-coded to 1 second.
* JXTemplate does not have any delay at all
* XSP, etc.
And as Carsten mentioned, MountNode overrides reload setting passed from root
sitemap, so if nothing is specified on MountNode, you always get default (true)
instead of default from parent sitemap.
So how many people (despite Carsten) are actually using it?
I agree it's chicken/egg problem. With 'run modes' more people will use it.
Vadim
