On Sun, 11 Aug 2002, Patrick Luby wrote:

> > IMHO that's _totally_ unacceptable ( having tomcat5 work only with 
> > xerces).
> 
> I think that the dependency on Xerces 2.0.1 is excessively restrictive 
> as well. IIRC (maybe Jean-François could provide some of the details he 
> found?), Xerces 2.0.1 was the only Xerces parser that we have found so 
> far that does not throw StackOverflow or other fatal exceptions when an 
> XML file using XML schema is parsed. I believe (Jean-François: let me 
> know if my understanding is incorrect) that this problem exists even if 
> schema validation is turned off.

I can confirm that tomcat5 works just fine with crimson if validation
is off ( and it starts much faster - 6-7 seconds ).

> > And having schema validation turned on by default has a strong -1 from
> > me - if the spec _requires_ schema validation, then implement it at 
> > deployment time. The performance hit is just unacceptable. 
> 
> Any performance increases through delayed validation sounds good to me.

It's not 'delayed'. There are 2 choices: one ( used in 3.3 ) is to 
put a 'marker' in work/ and check the timestamp to avoid validating
a file that was validated before and didn't changed.

The 'right' solution (IMO) is to have a clearly defined 'deploy'
stage, and have the validation done at deploy time. 

The 'deploy' stage is also needed for jk2 ( to generate apache
config files for example ). 

Every time web.xml ( or another .xml ) file changes the app
should be redeployed ( i.e. the 'deploy' hooks run ). 

I can live with the first aproach, or even with a global option
to turn off validation. 

> > ( in the process we should also move DTD validation to the same 
> > stage and stop doing it on every startup if the xml file didn't change )
> > 
> 
> Makes sense. Especially since we use this same technique for JSP page 
> compilation.

Do we ? It didn't seem so when I looked at the code. 

Actually it somehow seems that something is looking for TLDs 
at startup - that's a bug for me. The TLDs should be processed
when the first JSP is complied - not before. If a webapp
doesn't use JSPs or uses precompiled JSPs ( which is a very
good idea IMO ) - then there is no point on scanning each 
jar for TLDs. 

Costin


--
To unsubscribe, e-mail:   <mailto:[EMAIL PROTECTED]>
For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>

Reply via email to