Craig,
Would it be difficult to somehow parametrize the classloading in commons-logging.
I just commented out small portion that resorts to thread classloader.
To me it seems that it would be very trivial to give the user possibility
to disable thread classloading (which still could be the default).
That would lead to a Catch-22 situation, because the thread context class loader is used to load the commons-logging.properties file as well, and you would want the setting to take effect first :-).
More seriously, letting people disable this would be a disaster in a servlet container environment, which is a very common use case for code that uses commons-logging. More than a few people on this thread have described exactly how to set up and use it successfully, allowing each individual webapp to define their own logging environment. If you turned of context class loading, this would be impossible unless you forced every app to have their own copy of c-l, and didn't allow the container to have its own (which, among other things, would totally break Tomcat 5).
- rami
Craig
-----Original Message-----
From: Craig R. McClanahan [mailto:[EMAIL PROTECTED] Sent: Tuesday, October 14, 2003 1:18 AM
To: Jakarta Commons Developers List
Subject: Re: commons-logging & classloading
Will Jaynes wrote:
Yoav,If your webapp has its own private implementations of Log or LogFactory, you definitely need the current class loadng design.
I, too, use log4j directly in my own classes. But we all use components that use commons-logging, so we have to deal with it.
No one has yet described for me a use case where the current commons-logging classloading strategy is necessary or even correct. Whereas I have described a use case where it clearly fails, and have implemented a code change that fixes the problem for my use case.
Before you say "nobody should need that", this feature can be critically important when integrating enterprise scale applications together. A couple of illustrative examples might help you understand the reasoning:
* Large scale shop has different conventions for logger names than
the typical Jakarta pattern (fully qualified class name == logger name),
and wants all installed apps to follow the shop standards.
* App server hosting lots of webapps, and wants the actual logger names to include the webapp name so that they can be physically separated.
Either of these is pretty easy to deal with by creating a custom LogFactory implementation, and shipping it inside your webapp. The nice thing about this approach is that the change is transparent to the actual code that does the logging -- the only visible impact is the need to configure commons-logging.properties (in /WEB-INF/classes) to select the custom implementation.
As far as I can tell, commons-logging is just plain broken.But I have
a lot of respect for the developers here, and clearly they have thought about this issue a lot. So I'm left wondering whatI'm missing.
Is it an excepted implementation fact that components that use commons-logging can't be shared at the web app server level?No. As others have pointed out, you can do this (and Tomcat already does this).
WillCraig
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
--------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
