隨筆檔案

public aspect TraceAspect{
   private Logger _logger = Logger.getLogger("trace");
   pointcut traceMethods(): execution(* *.*(..))&&!within(TraceAsptect);
   before() : traceMethods(){
      Signature sig = thisJoinPointStaticPart.getSignature();
      _logger.logp(Level.INFO, sig.getDeclaringType().getName(), sig.getName(), "Entering");
  }
}

What’s wrong with conventional logging ?

When a new module is added to the system, all of its methods that need logging must be instrumented. Such instrumentation is invasive, causing the tangling of the core concerns with the logging concern. Further, if you ever happen to change the logging toolkit to a different API, you need to revisit every logging statement and modify it.

Consistency is the single most important requirement of logging. It means that if the logging specification requires that certain kinds of operations be logged, then the implementation must log every invocation of those operations. When things go wrong in a system, doubting the logging consistency is probably the last thing you want to do. Missed logging calls can make output hard to understand and sometimes useless. Achieving consistency using conventional logging is a lofty goal, and while systems can attain it initially, it requires continuing vigilance to keep it so. For example, if you add new classes to the system or new methods in existing classes, you must ensure that they implement logging that matches the current logging strategy.

The beauty of AspectJ-based logging

The limitations are not a result of the logging APIs or their implementations; rather, they stem from the fundamental limitations of objectoriented programming, which require embedding the logging invocations in each module. AOP and AspectJ overcome those limitations. AspectJ easily implements the invocation of logging statements from all the log points. The beauty is that you do not need to actually instrument any log points; writing an aspect does it automatically. Further, since there is a central place to control logging operations, you achieve consistency easily.

The most fundamental difference between conventional logging and AspectJbased logging is modularization of the logging concern. Instead of writing modules that implement core concepts in addition to invoking logging operations, with AspectJ you write a few aspects that advise the execution of the operations in the core modules to perform the logging. That way, the core modules do not carry any logging-related code. By modularizing, you separate the logging concern
from the core concerns.

With AspectJ-based logging, the logger aspect separates the core modules and the logger object. Instead of the core modules’ embedding the log() method invocations in their source code, the logger aspect weaves the logging invocations into the core modules when they are needed. AspectJ-based logging reverses the dependency between the core modules and the logger; it is the aspect that encodes how the operations in the core modules are logged instead
of each core module deciding for itself.