on 8/27/01 1:41 PM, David Rees at [EMAIL PROTECTED] wrote:
> On Sun, Aug 26, 2001 at 10:44:34PM -0400, Geir Magnusson Jr. wrote:
>>
>>> My initial impressions are good, my only complaint was that I had to build
>>> it and an unfamiliarity with tag libraries.
>>
>> Come on. :) Because of JJAR, the only thing you had to do was download
>> from CVS, have ant installed, and type
>>
>> $ ant getjars
>> $ ant jar
>>
>> and done.
>>
>> I am not that familiar with taglibs either :)
>
> I'm lazy. ;-) The existing documentation is very well written, I don't
> think there's any way it could be better.
Where do you live? I can bring it over to your house... :)
>>> I wrote a couple simple pages, one in JSP, one using JSP/Velocity and
>>> performance was pretty good, maybe a 30% performance hit.
>>
>> Really? I am surprised that JSP is that much slower :)
>>
>> Seriously, what JSP engine? Velocity, with a decent data-access page, I
>> believe will easily keep up with JSP.
>
> Tomcat 3.2.3. From what I understand, the tablib parsing routines in Tomcat
> 3.2.X are really quite slow. The two simple pages I tested are simple:
> Create an array of strings 10 long, iterate though the array printing out
> each element and printing the current date.
Did you set up the JSP via servlet front controller?
>>> So are my questions:
>>>
>>> How does the taglib work?
>>
>> Very simply - it treats everying in between the start and end tags as
>> VTL and parses and renders it into the output stream.
>>
>>> How does velocity get invoked to parse the code?
>>
>> The taglib uses the standard taglib callback architecture - nothing
>> special or magical.
>
> Cool.
>
>>> Where is the velocity engine hiding?
>>
>> In the normal place - the velocity.jar that is hopefully not in your
>> classpath, but in WEB-INF/lib
>
> This actually introduced another issue: I had a servlet in the same context
> which also uses velocity (I map *.vm to this servlet). If I loaded one of
> the JSPs first, it didn't initialize velocity as I expected. If I made sure
> that my servlet initialized first (and initialized Velocity) then both would
> work since the taglib doesn't depending on any files in the filesystem.
That's right, one of the downsides of the singleton. You can get around
that by converting your servlet to use the separate runtime instance.
>>> I know you mention velocimacros as a
>>> benefit to using the taglib, is there an equivalent to the VM_globals.vm or
>>> do you just have to include a global file which has your macros defined
>>> there?
>>
>> (You really can't include VMs.)
>>
>> Yes, a global library (or plural...) will indeed work if you use the
>> singleton model for Velocity (which the taglibs does). What you could
>> do is make a little startup servlet (use <load-on-startup>) that
>> configures Velocity with the template path to find the global library
>> for your webapp.
>>
>> I will try to document how to do that this week.
>
> (I should have more closely read this last paragraph before writing my last
> one) Looks like I already figured this part out. ;-)
>
> Here's a couple more performance notes (which have caused me to lose
> interest in the Velocity taglib for now):
>
> For the simple page I was testing with (my home page
> http://www.greenhydrant.com/~drees/, mostly static pages with a header,
> footer included), I found that JSP or JSP/Velocity is a LOT slower than
> mapping *.vm to a controller servlet (MVC type app).
LOL. Ok - I wasn't thinking about Velocity vs JSP, but JSP vs JSP+Veltag.
The latter I would think are comparable. I believe that it has been shown
that Velocity is faster than Tomcat 3.x JSPs.
> Probably about twice as slow, I was getting ~30 hits/s with JSP or
> JSP/Velocity, but about twice as many using a servlet/Velocity with some
> quick testing to Apache's ab.
>
> Under other servlet engines I don't expect the performance penalty to be as
> high, but I think it's probably just Velocity being too fast for it's own
> good when run under normal conditions. ;-)
We are going to add a parameter for that,
runtime.slow.this.thing.down = n
where n is the delay in milliseconds...
> -Dave
geir
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com
