commons-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Hans Gilde <hgilde-comm...@earthlink.net>
Subject RE: [Jelly] nested compileScript() with import ?
Date Tue, 07 Dec 2004 05:11:37 GMT
Maybe it's getLastModified instead of getDate?

You probably know, but there's another performance characteristic of my
solution: the Servlet threads only contend for a lock during the time it
takes to look the Script up from the Map. With yours, the threads have a
much larger window for contention. This would only make a difference if your
application will run many instances in many threads.

What about an application variable "flushJellyCache"? At the beginning of
the Servlet, check for that variable. If it's true, lock the application,
remove all from the Map, set to false. Now you can flush the cache by hand
from the web interface or a management client.

Hans

-----Original Message-----
From: Arnaud Masson [mailto:arnaud.masson@kikamedical.com] 
Sent: Monday, December 06, 2004 12:11 PM
To: Jakarta Commons Users List
Subject: Re: [Jelly] nested compileScript() with import ?

Hans Gilde wrote:

>Arnaud,
>
>You're right about how JSP works; maybe you could file an issue about this
>in Jelly's jira? This seems like something that many web users would want.
>
>  
>
done (sorry for the double post #170/171, the server wasn't responding)

>Sorry to introduce problems but...
>
>To get the benefit of your solution, you will need to reuse the cache
>between Servlet invocations. The basic Jelly Servlet does not reuse the
>JellyContext between invocations. Here's the pain: the context isn't
>entirely thread safe (actually, the parser isn't thread safe and the
context
>uses a single parser).
>
>Here's what you can do:
>
>Set the HashMap into the context before invoking the script. If inherit is
>true (by default), the Map will be accessible throughout your entire
script.
>
>Now that your Servlet has a handle on the HashMap, you can keep it in an
>instance variable (must be a thread safe HashMap) or as an application
>variable or whatever.
>
>  
>
Currently I put the JellyContext (including the script cache) in the web 
session.
When I use this JellyContext , I do a synchronized(myContext) {...} to 
avoid the threading problem.
Of course the cache isn't shared across session.

Your solution is better but the per-session solution has one advantage.
You can easily purge the cache when you open a new session, which is 
useful when you modify the jelly scripts.
I have tried to use URLConnection.getDate() to discard the cache,
but it doesn't work on the jndi URL provided by tomcat with 
ServletContext.getResource(), it always returns 0.
Any idea how I can manage cache update ?

>-----Original Message-----
>From: Arnaud Masson [mailto:arnaud.masson@kikamedical.com] 
>Sent: Monday, December 06, 2004 4:06 AM
>To: Jakarta Commons Users List
>Subject: Re: [Jelly] nested compileScript() with import ?
>
>Hans Gilde wrote:
>
>  
>
>>Ok, it sounds reasonable to want those caching conditions. But for my own
>>curiosity, why do you want them? Are you in a tight memory situation or is
>>it a perfectionist thing?
>>
>> 
>>
>>    
>>
>It's for speed, especially for Script.run().
>I must run many small scripts to build page, so I would like to minimize 
>IO and XML parsing.
>The outputs of the scripts are assembled in a big DOM,
>under the control of a java loop which also makes some modifications to 
>the DOM.
>(I use a custom ContentHandler so that jelly output is directly added in 
>the document without parsing.)
>
>I think that in a classic JSP, all included pages are compiled when the 
>main jsp  is compiled,
>so there is no performance penalty there. I would like the same thing in 
>jelly.
>
>  
>
>>What's wrong with using JellyContext variables? It's pretty much what
>>they're there for.
>>
>> 
>>
>>    
>>
>I had the idea that the cached scripts were technical details and should 
>not be at the same level as user variables.
>This time, it's probably a "perfectionist thing" ! :-)
>
>  
>
>>You can think of a JellyContext as a scope. At each level of XML tag, you
>>get a new JellyContext (a new scope). Export says "when a variable is set
>>into the current scope, should I automatically put it into the scope
>>    
>>
>above"?
>  
>
>>Inherit says "when creating a new scope, should I give it all the
variables
>>    
>>
>>from the current scope"?
>  
>
>>If a Tag wants to set a variable in a context (scope) that's higher up in
>>the tree, there are two choices: use export=true or find the parent
context
>>(while ((parent = getParent()) != null)).
>>
>>To find a variable that's been set into a context above yours, either use
>>inherit=true or findVariable.
>>
>>I'd suggest this solution:
>>
>>Have your Tag find the top level context (while ((parent = getParent()) !=
>>null)). Use a single variable name in this context, say
>>    
>>
>my_imported_scripts.
>  
>
>>That variable will be a single HashMap from uri to parsed Script instance.
>> 
>>
>>    
>>
>Ok, I'm going to try this.
>
>  
>
>>As you saw, compilable tags are sort of compiled, but they don't get a
>>context or their attributes at that point. The problem with compilable
tags
>>(I don't recommend using them) is that they're only compiled with respect
>>    
>>
>to
>  
>
>>the thread that compiles them. If you run the same Script in a different
>>thread, those compiled tags won't be there. This is not true for a Script.
>>Once a Script is compiled, it stays compiled no matter where you use it.
>>
>> 
>>
>>    
>>
>I don't think I need the advanced Script based solution for now, the 
>cache map must be sufficient for me.
>
>Thanks for your help.
>
>  
>
>>Hans
>>
>>-----Original Message-----
>>From: Arnaud Masson [mailto:arnaud.masson@kikamedical.com] 
>>Sent: Sunday, December 05, 2004 5:06 PM
>>To: Jakarta Commons Users List
>>Subject: Re: [Jelly] nested compileScript() with import ?
>>
>>Yes,  I have copied the import tag and added the cache to the new tag.
>>It works fine, but I would like to reuse compiled scripts
>>- if several container scripts include the same sub script
>>- if a container script include the same sub script several times
>>
>>To do that, I currently use the JellyContext to cache compiled scripts 
>>with a call to setVariable(),
>>using a variable name based on the script uri... it works, but it's not 
>>really clean.
>>
>>Maybe a better way would be to override the compileScript() methods of 
>>the JellyContext
>>and add a cache of compiled scripts inside the context (indexed by 
>>uri/url) ?
>>
>>Also I am not sure how to handle the 'export' and 'inherit' parameters 
>>when I use compileScript+run instead of a single call to runScript().
>>
>>You say a Tag isn't compilable, but what's the "CompilableTag" ??
>>I tried to use it the but the attributes (uri,...) of my tag are empty 
>>when "compile" is called.
>>
>>Thanks for your help!
>>
>>
>>
>>.Hans Gilde wrote:
>>
>> 
>>
>>    
>>
>>>You could easily cache the imported script the first time it's run. This
>>>      
>>>
>is
>  
>
>>>a simple modification to the current import tag, so that it keeps the
>>>reference to the script. If you do this, why not add an attribute "cache"
>>>   
>>>
>>>      
>>>
>>to
>> 
>>
>>    
>>
>>>turn caching on/off and then submit it as a patch to the import tag?
>>>
>>>It would be a little harder, but not at all impossible, to make it cache
>>>   
>>>
>>>      
>>>
>>the
>> 
>>
>>    
>>
>>>imported script at compile time.
>>>
>>>If not, the basic idea is this:
>>>
>>>A Tag isn't compilable, it's generated at runtime. A Script is
compilable,
>>>it's generated at compile time. You need a special Script, not a special
>>>Tag. Scripts are very much like tags except that they need to be thread
>>>safe. Most of the time, a Script called TagScript is used. By default,
>>>      
>>>
>this
>  
>
>>>Script creates and caches Tag instances when it's run.
>>>
>>>Your new Script would compile the import at compile time. The result of
>>>compiling the import is, it self, a Script instance. At runtime, your
>>>   
>>>
>>>      
>>>
>>Script
>> 
>>
>>    
>>
>>>would simply pass control to the imported Script.
>>>
>>>You would also have to implement a custom TagLibrary. A TagLibrary gets
to
>>>create a TagScript (implements Script) for every XML tag in its
namespace.
>>>So, your TagLibrary would create a custom TagScript that would compile
and
>>>keep the imported XML.
>>>
>>>-----Original Message-----
>>>From: Arnaud Masson [mailto:am@kikamedical.com] 
>>>Sent: Saturday, December 04, 2004 6:07 PM
>>>To: commons-user@jakarta.apache.org
>>>Subject: [Jelly] nested compileScript() with import ?
>>>
>>>hi
>>>
>>>in the current version of jelly "import" tag, it seems that imported 
>>>scripts are always parsed and recompiled each time the containing script 
>>>runs,
>>>even if this script has already been compiled.
>>>
>>>the problem is that  it isn't optimized if the compiled version of the 
>>>main script is cached.
>>>
>>>is it possible to compile all scripts included via <j:import ...> via a

>>>single call to jellyContext.compileScript() on the containing script ?
>>>should i write a custom tag to implement that (to replace import) ?
>>>
>>>thanks in advance
>>>
>>>arnaud
>>>
>>>
>>>
>>>
>>>---------------------------------------------------------------------
>>>To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
>>>For additional commands, e-mail: commons-user-help@jakarta.apache.org
>>>
>>>
>>>
>>>
>>>   
>>>
>>>      
>>>
>>---------------------------------------------------------------------
>>To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
>>For additional commands, e-mail: commons-user-help@jakarta.apache.org
>>
>>
>>---------------------------------------------------------------------
>>To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
>>For additional commands, e-mail: commons-user-help@jakarta.apache.org
>>
>>
>> 
>>
>>    
>>
>
>
>
>---------------------------------------------------------------------
>To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
>For additional commands, e-mail: commons-user-help@jakarta.apache.org
>
>
>  
>



---------------------------------------------------------------------
To unsubscribe, e-mail: commons-user-unsubscribe@jakarta.apache.org
For additional commands, e-mail: commons-user-help@jakarta.apache.org


Mime
View raw message