Should not merge, just for a simple benchmark.
Some code is copied in order to make a direct comparison within
a single jvm configuration.

Text 1.5.) Documentation for base64 variable substitution.

Text 1.5.)

Text 1.5.)

This change isolates initialization logic into a new
ThrowableProxyHelper class, and rendering logic
into ThrowableProxyRenderer.

Both new utility classes are package private. ThrowableProxy public
API has not changed.

RollingRandomAccessFileAppender.

Proxies through several interfaces in order to simulate performance
of a standard application, which doesn't benefit quite as much
from the ThrowableProxy Map<String, CacheEntry> cache.

Initial results:

Benchmark                                                            Mode  Cnt       Score       Error  Units
FileAppenderThrowableBenchmark.complexLog4j1                        thrpt    3    8562.548 ±  4122.961  ops/s
FileAppenderThrowableBenchmark.complexLog4j2ExtendedThrowable       thrpt    3    2575.655 ±   218.395  ops/s
FileAppenderThrowableBenchmark.complexLog4j2ExtendedThrowableAsync  thrpt    3    2518.260 ±   335.573  ops/s
FileAppenderThrowableBenchmark.complexLog4j2Throwable               thrpt    3   12738.874 ±  2867.140  ops/s
FileAppenderThrowableBenchmark.complexLog4j2ThrowableAsync          thrpt    3    3489.965 ±   471.234  ops/s
FileAppenderThrowableBenchmark.complexLogbackFile                   thrpt    3   12866.339 ±  1396.160  ops/s
FileAppenderThrowableBenchmark.julFile                              thrpt    3   55019.426 ±  9342.755  ops/s
FileAppenderThrowableBenchmark.log4j1                               thrpt    3   75444.252 ± 18508.319  ops/s
FileAppenderThrowableBenchmark.log4j2ExtendedThrowable              thrpt    3   45708.853 ± 13553.851  ops/s
FileAppenderThrowableBenchmark.log4j2Throwable                      thrpt    3  110923.511 ± 19661.868  ops/s
FileAppenderThrowableBenchmark.logbackFile                          thrpt    3  112548.025 ± 40720.422  ops/s

This closes #203

make ReusableObjectMessage always pass its object as the first parameter

This closes #205

Standardize forkCount on 1, rather than 2*CORES.

log4j-core uses one fork which is not reused. Most of our time
spent in tests is running those in log4j-core. Applying this
configuration to all modules prevents test flakes.

Share the SecurityManager implementation of getCurrentStackTrace
with the java 9 implementation.

Revert "Add low-level method org.apache.logging.log4j.core.appender.AsyncAppender.getThreadQueueSize() to help synch tests in applications."

This reverts commit 1927dfa7.

org.apache.logging.log4j.core.appender.AsyncAppender.getThreadQueueSize()
to help synch tests in applications.

mongo tests intermittenly fail with an "Operation not permitted"
IOException when trying to execute the mongod extracted by
de.flapdoodle.embed.process; this seems to be related to surefire
running tests in parallel. Work around that here by forcing serial
execution of the mongo tests.

This closes #201

When emptyReplacement length is less than 10 but large enough for
the current paramers, the ReusableParameterizedMessage would
retain references to the most recently logged parameters.

This closes #200

Combine codepaths for initial and terminal fragments.
Add changelog entry.

This closes #188

session logger failed (Closes #198).