- fix the CacheEntry map in ThrowableProxy#toExtendedStackTrace to be put and gotten with same key
- stackTraceElement.toString() returns a string representation of this stack trace element, just as MyClass.mash(MyClass.java)
- stackTraceElement.getClassName() returns the fully qualified name of the Class, just as org.apache.logging.log4j.MyClass
- The main impact of the problem was that it would increase of frequency of loading classes ,which led to the execution of the program be slow down, because of the synchronization mechanism in the method loadClass
- In addition to fixing the problem, I think cache map could be made global, instead of a new one for each exception instance.
- We made the benchmark test to compares the performance of ThrowableProxy optimizations for different exception
- baseline: test `ThrowableProxy` in log4j2, not optimized, with bug, for comparison consideration.
- bugfixed: fixed bug in `ThrowableProxy` accessing the cache.
- optimized: make the cache global, instead of a new one for each exception instance.
Then we see
- more than `10` times improvement when the bug is fixed for exceptions with stack element class duplicated many times.
- and about `2` times further improvement when make the cache global (compared with log4j2, more than `20` times improvement).
- benchmark test detail: https://github.com/liuwenchn/log4j2-throwableproxy-benchmark
- PR: https://github.com/apache/logging-log4j2/pull/195