Mailing-List: contact cocoon-dev-help@xml.apache.org; run by ezmlm
Message-ID: <3A8C13CD.4ACC343B@apache.org>
Date: Thu, 15 Feb 2001 12:37:17 -0500
From: Berin Loritsch <bloritsch@apache.org>
MIME-Version: 1.0
To: cocoon-dev@xml.apache.org
Subject: Re: [C2] Running JMeter on Cocoon Samples
References: <3A8AF33B.DCC5F2C1@apache.org>
 <20010214214933.A580@hydrogen.internal.luminas.co.uk>
 <3A8B0375.8273E5A2@apache.org> <01021423514004.01076@lap1>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit

Giacomo Pati wrote:
> 
> Berin Loritsch wrote:
> > Paul Russell wrote:
> > > * Berin Loritsch (bloritsch@apache.org) wrote :
> > > > I discovered something incredible.  The XSP system is our major
> > > > performance sink-hole.  This I find to be amazing.
> > > >
> > > > I have my suspiscions as to where the problems may lie: Class
> > > > validation (is it current?) and sending too many namespace events.  I
> > > > am going to try running reading a normal file through the
> > > > LogTransformer, and then an XSP file through the same LogTransformer.
> > > > I have a feeling that those two areas are are major performance
> > > > bottlenecks.
> > >
> > > Interestingly, we discovered something similar a long time ago in
> > > Luminas. Probably because we are *very* heavy on namespaces (a lot of
> > > our pages have 10-15 namespaces floating around in them). The current
> > > XSP implementation does an awful lot of prefix mapping changes. In fact,
> > > we discovered that in a number of instances, _over half_ of the
> > > generated code was concerned with adding and removing prefix mappings.
> > > This is clearly not sensible. I'm not yet sure how to avoid this - I
> > > think we may have to use extension functions to keep track of which
> > > namespaces we've already defined.
> >
> > I just noticed that the ServerPagesGenerator caches the SAX results with
> > a Stack.  Is this really necessary?  If an exception occurs, we should
> > just throw a SAXException or ProcessingException like the rest of the
> > system.
> 
> I have talked about that with Ricardo at ApacheCon and IIRC it has to do with
> easy programming practice where XSP programmers might want to use a return
> statement to end the page (even in the middle of nested elements).

I'm not sure if I like that.  That really should not be the rule, so something
like that would have to be done by the XSP implementor if they wanted it.  Anyways,
it doesn't really affect the outcome all that much.  The difference in performance
is less than a millisecond, so it is not our primary concern.