[C3] Group and artifact IDs (was: [vote] Cocoon 3: Versioning, SVN, Maven, namespaces, issue tracking and CI)

2011-11-30 Thread Andreas Hartmann

Hi Reinhard,

Am 21.08.08 23:53, schrieb Reinhard Pötz:


After having already discussed the details, let's make a formal decision
about versioning, SVN, Maven, namespaces issue tracking and CI for Cocoon 3.


[…]


Maven
---
We use functional names for all artifacts:

org.apache.cocoon.pipeline:cocoon-pipeline:3.0.0
org.apache.cocoon.sitemap:cocoon-sitemap:3.0.0
org.apache.cocoon.servlet:cocoon-servlet:3.0.0
org.apache.cocoon.controller:cocoon-controller:3.0.0
org.apache.cocoon.rest:cocoon-rest:3.0.0
org.apache.cocoon.stringtemplate:cocoon-stringtemplate:3.0.0

By using the functional name as part of the groupId, Cocoon 2.2 and
Cocoon 3 can be used together without getting any problems with the
dependency resolution mechanisms of Maven or Ivy.


I just stumbled upon this again. Our group IDs look verbose and 
redundant at a first glance; from my experience it is rather unusual to 
use distinct group IDs for the individual projects. Would you mind 
elaborating a bit why it helps to have the artifact ID (functional name) 
as part of the group ID? Would the version number not be sufficient for 
the dependency resolution?


TIA!

Best regards,
Andreas



--
Andreas Hartmann, CTO
BeCompany GmbH
http://www.becompany.ch
Tel.: +41 (0) 43 818 57 01



Re: PermGen memory leak in cocoon-jnet

2011-11-30 Thread Andreas Hartmann

Am 29.11.11 02:55, schrieb Andreas Hartmann:

Hi everyone,

when undeploying a C3 app, I'm experiencing a PermGen memory leak due to
cocoon-jnet. You find more general info on the subject at [1].

This is the reference chain:

class java.net.URL
-
org.apache.cocoon.jnet.URLStreamHandlerFactoryInstaller$ParentableURLStreamHandlerFactory

-
org.apache.cocoon.jnet.URLStreamHandlerFactoryInstaller$ParentableURLStreamHandlerFactory

- org.apache.catalina.loader.WebappClassLoader


The problem is this line of code in URLStreamHandlerFactoryInstaller;
commenting it out resolves the memory leak:

URL.setURLStreamHandlerFactory(new
ParentableURLStreamHandlerFactory(factory, null));

Here the static reference from java.net.URL to a Cocoon class is set.
Does anyone know if and how the reference can be removed, or if we can
achieve the same behaviour without registering the factory with
java.net.URL?


The following patch avoids the memory leak, but I have the feeling that 
it has functional drawbacks, otherwise the code would already look like 
this :)


Does anybody know how I can test if the class still fulfils its purpose? 
Maybe it's possible to provide a unit test?


TIA!


Index: 
src/main/java/org/apache/cocoon/jnet/URLStreamHandlerFactoryInstaller.java

===
--- 
src/main/java/org/apache/cocoon/jnet/URLStreamHandlerFactoryInstaller.java 
(revision 1208388)
+++ 
src/main/java/org/apache/cocoon/jnet/URLStreamHandlerFactoryInstaller.java 
(working copy)

@@ -28,13 +28,8 @@
 public class URLStreamHandlerFactoryInstaller {

 public static void 
setURLStreamHandlerFactory(URLStreamHandlerFactory factory) throws 
Exception {

-try {
-// if we can set the factory, its the first!
-URL.setURLStreamHandlerFactory(new 
ParentableURLStreamHandlerFactory(factory, null));

-} catch (Error err) {
-ParentableURLStreamHandlerFactory currentFactory = 
getCurrentFactory();
-setCurrentFactory(new 
ParentableURLStreamHandlerFactory(factory, currentFactory));

-}
+ParentableURLStreamHandlerFactory currentFactory = 
getCurrentFactory();
+setCurrentFactory(new 
ParentableURLStreamHandlerFactory(factory, currentFactory));

 }



-- Andreas


--
Andreas Hartmann, CTO
BeCompany GmbH
http://www.becompany.ch
Tel.: +41 (0) 43 818 57 01



RE: PermGen memory leak in cocoon-jnet

2011-11-30 Thread Nathaniel, Alfred
Hi all,

URL.setURLStreamHandlerFactory is singleton call which should only be used when 
you own the JVM.
Calling it from C3 which is expected to play nicely with other servlets in the 
same container, it is a no-no.

Jnet should be changed, for example by providing a createURL(String spec) 
factory method which calls the
URL(protocol,host,port,file,handler) with the handler matching the custom 
protocol.
 
Cheers, Alfred.

-Original Message-
From: Andreas Hartmann [mailto:andr...@apache.org] 
Sent: Mittwoch, 30. November 2011 14:10
To: dev@cocoon.apache.org
Subject: Re: PermGen memory leak in cocoon-jnet

Am 29.11.11 02:55, schrieb Andreas Hartmann:
 Hi everyone,

 when undeploying a C3 app, I'm experiencing a PermGen memory leak due to
 cocoon-jnet. You find more general info on the subject at [1].

 This is the reference chain:

 class java.net.URL
 -
 org.apache.cocoon.jnet.URLStreamHandlerFactoryInstaller$ParentableURLStreamHandlerFactory

 -
 org.apache.cocoon.jnet.URLStreamHandlerFactoryInstaller$ParentableURLStreamHandlerFactory

 - org.apache.catalina.loader.WebappClassLoader


 The problem is this line of code in URLStreamHandlerFactoryInstaller;
 commenting it out resolves the memory leak:

 URL.setURLStreamHandlerFactory(new
 ParentableURLStreamHandlerFactory(factory, null));

 Here the static reference from java.net.URL to a Cocoon class is set.
 Does anyone know if and how the reference can be removed, or if we can
 achieve the same behaviour without registering the factory with
 java.net.URL?

The following patch avoids the memory leak, but I have the feeling that 
it has functional drawbacks, otherwise the code would already look like 
this :)

Does anybody know how I can test if the class still fulfils its purpose? 
Maybe it's possible to provide a unit test?

TIA!

-- Andreas

N��{^����zf��+�קu�h�\���ay�'~'^�ؚ����az�u�޲ǝ!ޞ�m�觵��y��r*bz{i���zz-��׫jw]zW�z�b�隊X���bjץ�8Z�L�