velocity-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Koberg <>
Subject Re: The Guardian website moves to Velocity
Date Sat, 12 May 2007 18:14:17 GMT
On Sat, 2007-05-12 at 19:18 +0200, Jonathan Revusky wrote:
> Robert Koberg wrote:
> > OK.
> > 
> > After trying the docbook distro, it is simply to huge to achieve good
> > performance. 
> I already told you quite honestly that the one from the docbook distro 
> took 15x longer.

When I ran it, I think it was 3 or 4 times longer. And it was doing a
lot of output warnings so that has to hit performance.

> > It also writes alot to the logger which probably slows it
> > down a great deal.
> > 
> > So I wrote my own in about an hour. I believe it handles all elements
> > correctly (I think you have some CSS styling issues, though).
> You're producing all the same identical output? TOC, Index, Glossary.... 
> All the links have to work and so on....
> What is available here?
> It does need to be reproduceable. I have to be able to run it myself.

I tried to email you ant build instructions and the XSL file but the
email was returned with an error as I posted here. How can I get it to

> > I do see your transform taking 4 secs. Mine also takes 4 secs. Is there
> > a way to get milliseconds in Ant? Even though Ant shows the times as
> > equal, I think it more than proves my point.
> Maybe it should be enough for the $500. I'm thinking about it. I never 
> thought you could get close. OTOH, your exact terms were "blowing it 
> away". You need to do better than equal something to "blow it away".
> Yes, it was pretty clear that it has to be not just as fast but faster, 
> I think.

Well than I think we need to take it out of Ant and run where we can at
least get milliseconds.

> > Plus you can still use it
> > in the browser requiring no transformation time for *you* (I think this
> > in itself should be enough to win the $500...). As always it can be run
> > in many different environs and probably would be even faster if run on
> > libxsl or even MSXML.
> It has to produce the same output. I have to be able to run it. Which 
> XSLT processor are you using?

It is basically the same output. I think you guys use too much inline
styling, so I made classes, but did not make the CSS.

> > 
> > Some caveats which might cause you to negate the effort:
> > 
> > * I did not create TOCs
> > * I did not handle all links correctly, I think (its been a while since
> > I looked at docbook)
> > * I did not hook up the navigation controls
> Oops, I wrote the above before reading this last bit. It has to be 
> equivalent functionally, of course. I don't think a perfect diff on the 
> output is necessary, but to not have TOC and index and so on, I don't 
> see how that can cut it.
> > 
> > HOWEVER, the XSL transformation is creating all of the pages PLUS a
> > single, all-inclusive one page version. So basically it is doing the
> > transform twice. I hope that makes up for the caveats above.
> It really has to be the same output. Maybe not to the level of a perfect 
> diff-less comparison, but not having TOC or index or anything is not 
> good enough, I don't think.

hmmm.... nothing for creating the full docs in one page as well as all
the individual pages thereby doing 2 transforms while yours does one?

> > 
> > If that is acceptable I would think it is more appropriate to send the
> > $500 to the Libertarian party in my name. If it is not acceptable, maybe
> > I will work on it when the weather gets bad. Gonna go outside now...
> > 
> > I will send the XSL file to you separately, all 250 lines with proper
> > spacing, nesting and indentation. If anyone else would like it, let me
> > know.
> I'd be curious to see it, but we do need TOC and index and glossary and 
> whatnot and the links working.
> THere is a certain vagueness about this. Which XSLT processor are you 
> using? (I'm sure I made clear that a native C one is not acceptable.)

OK. I have it working with Saxon and XSLTC in ant using the core XSLT 
task with transformerfactory children to switch out processors. XSLTC 
creates the java, compiles it and uses it for the transform. I don't 
know if Ant does this each time, but I think it does (where would 
ant put it for the classpath?). I would think if it just had to use the 
compiled class it would be faster. I think that Resin's XSLT compiler 
is a little better than XSLTC, but haven't tried it.


To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message