[logback-dev] I don't understand ClassPackagingDataCalculator :p

Ceki Gulcu listid at qos.ch
Thu Sep 25 21:47:35 CEST 2008

Comments inline.

Joern Huxhorn wrote:

 > It's because otherwise Class.forName, called from inside the Logger,
 > would use the Loggers ClassLoader to load the class in question. And
 > since the Logger class might be loaded by a container this could
 > result in wrong version information, right?

Yes, that is one scenario. There are other scenarios where multiple
class loaders may be in play. If we are going to report information,
then we must darn make sure that it is correct. Otherwise, we might be
misleading the user. (Wrong information is worse than no information.)

 > Did this problem actually happen or is this a precaution? Shouldn't e.g.
 > web-applications always execute in an isolated manner?
 > That's my experience, at least, because otherwise a logback
 > configuration in one webapp would overwrite that of another one,
 > wouldn't it?

Oh, it happens quite frequently. There are many different models for
the class loader tress in (Java) containers. Moreover, logback and
log4j support repository selectors which allow a single instance of
logback (or log4j) to offer a distinct logger contexts for each
web-application. See the chapter on context selectors in the logback

 > Concerning the use of the internal class, it's really quite unfortunate
 > that functionality like this isn't somehow accessible.
 > Similar functionality is available in
 > private static native Class[] ResourceBundle.getClassContext()
 > so that method, in combination with setAccessible(true), could be used
 > as an alternative to
 > public static native Class Reflection.getCallerClass(int i);
 > but it's just choosing between two evils.

I totally agree.

 > The most secure solution would probably be to first try to call
 > sun.reflect.Reflection.getCallerClass
 > using reflection so it's possible to catch Exception if the class or
 > method isn't available (anymore).

Runtime (but not compile-time) compatibility with JDK lacking
Reflection.getCallerClass can be achieved by conditional invocation of
getCallerClass(). If it is available, we invoke, otherwise we just
skip that call. Compile-time compatibility can be achieved using

 > If that doesn't work you could try, again using reflection, to retrieve
 > the ResourceBundle and make that method accessible. This can fail
 > because of security exception or if the internals of ResourceBundle change.
 > In that case it just isn't possible to resolve really correct ClassLoader.
 > All this reflection would only need to be done once statically in
 > ClassPackagingDataCalculator so the performance impact should be nearly
 > zero.

Good idea.

 > I guess that the method in Reflection is maybe faster since it only
 > returns a single class instead of an array containing all classes in the
 > call stack so I would keep using that as long as it's possible.

Yah, I am also not too keen on calling a method declared as private via
reflection, especially since an equivalent exists.

Anyway, before the next release is made we should ensure runtime
compatibility with all JDKs, even those lacking the aforementioned
method. I'd like to measure the performance impact of invoking
getCallerClass through reflection. More interestingly, reflection warps
the caller stack so the existing code *might* need to be modified and might 
become trickier.


Ceki Gülcü
Logback: The reliable, generic, fast and flexible logging framework for Java.

More information about the logback-dev mailing list