On 12/09/2011 12:15, Konstantin Kolinko wrote: > 2011/9/7 Mark Thomas <ma...@apache.org>: >> On 07/09/2011 19:49, Konstantin Kolinko wrote: >>> 2011/9/7 <ma...@apache.org>: >>>> Author: markt >>>> Date: Wed Sep 7 18:11:20 2011 >>>> New Revision: 1166290 >>>> >>>> URL: http://svn.apache.org/viewvc?rev=1166290&view=rev >>>> Log: >>>> Add a simple annotation cache to improve performance for applications that >>>> use lots of non-poolable tags.
<snip/> >>> 2. I wonder when the cache is cleared. E.g. if we are in development >>> mode or have enabled unloading unused JSPs in JspServlet >>> configuration. If I understand correctly it accumulates references to >>> Class objects but never releases them. >> >> When the context is stopped. > > This issue is still pending. > Though I think it concerns only development mode, and benefits of this > cache out-weight the resource leak. As written, it makes the maxLoadedJsps Jasper option useless. I have given this some thought this morning and I plan on looking at a WeakHashMap (plus synchronisation to handle the concurrency) based solution. > 3. Regarding the instances not created through the instance manager: > http://svn.apache.org/viewvc?rev=1166775&view=rev > > The code is now skipping annotations in those classes. > The old code would scan them for preDestroy annotations. > I have not looked what behaviour is required by the specs here, but I > think that the old behaviour was more correct. I tend to the view that if the InstanceManager didn't create the object, it isn't responsible for destroying it. That isn't the old behaviour and I am +0 if you want to change the current code back to the old behaviour. > 4. There is code that calls setAccessible() for Method, e.g.: > > Method postConstruct = (Method) entry.getAccessibleObject(); > boolean accessibility = postConstruct.isAccessible(); > postConstruct.setAccessible(true); > postConstruct.invoke(instance); > postConstruct.setAccessible(accessibility); > > There will be a race condition if several threads are changing > (setting or restoring) the method accessibility flag on the same > Method right before invoke() or isAccessible() calls. I'll try and address this along with the WeakHashMap changes since I'll be adding some syncs already. Mark --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
