The only _real_ solution is to not use the JSF specced resource handling and 
instead only use RESTful URLs.
Since JSF resources may contain EL expressions, they are _not_ cacheable. We 
currently deliver them with a expiry of 14 days, but this is actually wrong! 
The RI has the same bug, so it's at least broken in the same way ;)

To shed more light on the topic you should also quote the original thread about 
Jakobs RelativeResourceHandler. 

http://markmail.org/thread/glr356g45uta5yys

This contains all the really important discussions.
Jakob continued his project over at google code [1]


Fazit: 
* We must NOT cache for JSF spec compliant resource handling because the EL 
inside CSS, etc can change for _every_ request. 

* If we like to have a fully cacheable solution, then we should finally go a 
step back and again remove all the superfluous stuff added to 
commons-resourcehandler and do it cleanly.

We might scan the resources for '#{' but that might become expensive as well.
There was some proposal for JSF-2.2, do you have an update what happened with 
that?



LieGrue,
strub



[1] http://code.google.com/a/apache-extras.org/p/relative-resource-handler/


----- Original Message -----
> From: Werner Punz <[email protected]>
> To: MyFaces Development <[email protected]>
> Cc: 
> Sent: Wednesday, November 28, 2012 9:39 AM
> Subject: Re: [VOTE][core] fix for MYFACES-3586 (performance of resources in 
> jar files served by ResourceHandlerImpl)
> 
> +1
> I have had similar situations and people definitely need a fix this has 
> higher priority than having no new config params.
> 
> 
> Werner
> 
> 
> Am 27.11.12 16:56, schrieb Leonardo Uribe:
>>  +1, if there are people with interest to get a problem solved,
>>        it's worth a shot to hear what people says and get it done.
>> 
>> 
>>  2012/11/27 Leonardo Uribe <[email protected]>:
>>>  Hi
>>> 
>>>  Some time ago, it was mentioned the controversy behind a fix
>>>  for MYFACES-3586
>>> 
>>> 
> http://markmail.org/message/xycbf77ku7x5wygj#query:+page:1+mid:jhmcr6a435xma3lu+state:results
>>> 
>>>  Maybe we should reconsider the previous position about this
>>>  one and try to fix it, even if that means to include additional web
>>>  config params, even if exists better solutions for this one or
>>>  even if suppose more work to do.
>>> 
>>>  The reason why we should change our position is users start to
>>>  fall on the same hole over and over again. Given the interest on
>>>  the problem, it starts to sound better to provide an alternate fix
>>>  bundled in myfaces core instead say to the people "... setup a
>>>  load balancer in front of your server and cache the resources
>>>  there to fix your problem ...".
>>> 
>>>  Things become a ridiculousness, when you find a similar problem:
>>> 
>>>  https://issues.apache.org/jira/browse/MYFACES-3656
>>>  ViewScoped bean fails when SERIALIZE_STATE_IN_SESSION is set to true
>>> 
>>>  Both problems are similar because:
>>> 
>>>  1. Both problems has a known solution.
>>>  2. The solution is not perfect / cannot be enabled by default / there 
> are
>>>       better alternatives
>>>  3. The typical answer is say "... woops ... try this alternative 
> strategy and
>>>       try to live with the problem ..."
>>> 
>>>  Given the latest feedback about MYFACES-3586 and MYFACES-3656,
>>>  I think it is desirable to take a look and revote a solution to this 
> problem
>>>  again.
>>> 
>>>  Please vote:
>>> 
>>>  +1 if you consider we need to fix this one once for all.
>>>  +0 if you don't care about
>>>  -1  if you consider do nothing is better and why
>>> 
>>>  Note we need minimum 3 +1 votes to support this initiative and override
>>>  the previous decision of the community.
>>> 
>>>  regards,
>>> 
>>>  Leonardo Uribe
>> 
>

Reply via email to