All,
Wanted to get some thoughts on some things in the BundleWiringImpl.
There seems to be some very large methods in this class which makes it
very difficult to write tests. I wanted to get some opinions on
breaking some of the methods up to improve readability and testability.
The findClass method within the BundleClassLoader is a good example of a
large method that could benefit from this. I'd like to break down the
method to better fit the specification around Weaving so that there are
specific methods around Transform and Define phases of the weaving.
So for example it could be something like:
try
{
transformClass(felix, hooks, wovenClassListeners,
name, bytes);
}
catch (Exception e)
{
wci.setState(WovenClass.TRANSFORMING_FAILED);
callWovenClassListeners(felix,
wovenClassListeners, wci);
}
/* Do locking code - omitted for breivity */
try {
clazz = defineClass(felix, wovenClassListeners,
wci, name,
clazz, bytes, content, pkgName, lock);
} catch (ClassFormatError e) {
if(wci != null){
wci.setState(WovenClass.DEFINE_FAILED);
callWovenClassListeners(felix,
wovenClassListeners, wci);
}
}
Are there other arrangements that make more sense for this class or
reasons just to keep it all in the same method? I'm a big fan of
smaller methods where possible since it helps with readability and
testing. I might be interesting to run the code through a static
analysis tool such as SonarQube (I know some other projects build this
into their Jenkins processes) for no other than to identify some spots
that might benefit from some refactoring. Also I realized that I added a
logger to the constructor of the BundleClassLoader only to realize that
a logger was being exposed through the BundleWiringImpl class that's
passed in the constructor. Any thoughts on exposing the logger through
a constructor instead? Looking forward to input on this.
- Bob
